spa_impl.h revision 43466aae47bfcd2ad9bf501faec8e75c08095e4f
2fa9e406ahrens * CDDL HEADER START
3fa9e406ahrens *
4fa9e406ahrens * The contents of this file are subject to the terms of the
5ea8dc4beschrock * Common Development and Distribution License (the "License").
6ea8dc4beschrock * You may not use this file except in compliance with the License.
7fa9e406ahrens *
8fa9e406ahrens * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9fa9e406ahrens * or
10fa9e406ahrens * See the License for the specific language governing permissions
11fa9e406ahrens * and limitations under the License.
12fa9e406ahrens *
13fa9e406ahrens * When distributing Covered Code, include this CDDL HEADER in each
14fa9e406ahrens * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15fa9e406ahrens * If applicable, add the following below this CDDL HEADER, with the
16fa9e406ahrens * fields enclosed by brackets "[]" replaced with your own identifying
17fa9e406ahrens * information: Portions Copyright [yyyy] [name of copyright owner]
18fa9e406ahrens *
19fa9e406ahrens * CDDL HEADER END
20fa9e406ahrens */
223f9d6adLin Ling * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
2369962b5Matthew Ahrens * Copyright (c) 2013 by Delphix. All rights reserved.
24e9103aaGarrett D'Amore * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
25fa9e406ahrens */
27fa9e406ahrens#ifndef _SYS_SPA_IMPL_H
28fa9e406ahrens#define	_SYS_SPA_IMPL_H
30fa9e406ahrens#include <sys/spa.h>
31fa9e406ahrens#include <sys/vdev.h>
32fa9e406ahrens#include <sys/metaslab.h>
33fa9e406ahrens#include <sys/dmu.h>
34fa9e406ahrens#include <sys/dsl_pool.h>
35fa9e406ahrens#include <sys/uberblock_impl.h>
36fa9e406ahrens#include <sys/zfs_context.h>
37fa9e406ahrens#include <sys/avl.h>
38fa9e406ahrens#include <sys/refcount.h>
39fa9e406ahrens#include <sys/bplist.h>
40cde58dbMatthew Ahrens#include <sys/bpobj.h>
4143466aaMax Grossman#include <zfeature_common.h>
43fa9e406ahrens#ifdef	__cplusplus
44fa9e406ahrensextern "C" {
47ea8dc4beschrocktypedef struct spa_error_entry {
48ea8dc4beschrock	zbookmark_t	se_bookmark;
49ea8dc4beschrock	char		*se_name;
50ea8dc4beschrock	avl_node_t	se_avl;
51ea8dc4beschrock} spa_error_entry_t;
5306eeb2aektypedef struct spa_history_phys {
5406eeb2aek	uint64_t sh_pool_create_len;	/* ending offset of zpool create */
5506eeb2aek	uint64_t sh_phys_max_off;	/* physical EOF */
5606eeb2aek	uint64_t sh_bof;		/* logical BOF */
5706eeb2aek	uint64_t sh_eof;		/* logical EOF */
5806eeb2aek	uint64_t sh_records_lost;	/* num of records overwritten */
5906eeb2aek} spa_history_phys_t;
61fa94a07brendanstruct spa_aux_vdev {
62fa94a07brendan	uint64_t	sav_object;		/* MOS object for device list */
63fa94a07brendan	nvlist_t	*sav_config;		/* cached device config */
64fa94a07brendan	vdev_t		**sav_vdevs;		/* devices */
65fa94a07brendan	int		sav_count;		/* number devices */
66fa94a07brendan	boolean_t	sav_sync;		/* sync the device list */
67fa94a07brendan	nvlist_t	**sav_pending;		/* pending device additions */
68fa94a07brendan	uint_t		sav_npending;		/* # pending devices */
71e05725bbonwicktypedef struct spa_config_lock {
72e05725bbonwick	kmutex_t	scl_lock;
73e05725bbonwick	kthread_t	*scl_writer;
74e14bb32Jeff Bonwick	int		scl_write_wanted;
75e05725bbonwick	kcondvar_t	scl_cv;
76e05725bbonwick	refcount_t	scl_count;
77e05725bbonwick} spa_config_lock_t;
79c5904d1eschrocktypedef struct spa_config_dirent {
80c5904d1eschrock	list_node_t	scd_link;
81c5904d1eschrock	char		*scd_path;
82c5904d1eschrock} spa_config_dirent_t;
84ec94d32Adam Leventhaltypedef enum zio_taskq_type {
85e14bb32Jeff Bonwick	ZIO_TASKQ_ISSUE = 0,
8680eb36fGeorge Wilson	ZIO_TASKQ_ISSUE_HIGH,
87e14bb32Jeff Bonwick	ZIO_TASKQ_INTERRUPT,
8880eb36fGeorge Wilson	ZIO_TASKQ_INTERRUPT_HIGH,
89e14bb32Jeff Bonwick	ZIO_TASKQ_TYPES
90ec94d32Adam Leventhal} zio_taskq_type_t;
91e14bb32Jeff Bonwick
9235a5a35Jonathan Adams/*
93ec94d32Adam Leventhal * State machine for the zpool-poolname process.  The states transitions
9435a5a35Jonathan Adams * are done as follows:
9535a5a35Jonathan Adams *
9635a5a35Jonathan Adams *	From		   To			Routine
9735a5a35Jonathan Adams *	PROC_NONE	-> PROC_CREATED		spa_activate()
9835a5a35Jonathan Adams *	PROC_CREATED	-> PROC_ACTIVE		spa_thread()
9935a5a35Jonathan Adams *	PROC_ACTIVE	-> PROC_DEACTIVATE	spa_deactivate()
10035a5a35Jonathan Adams *	PROC_DEACTIVATE	-> PROC_GONE		spa_thread()
10135a5a35Jonathan Adams *	PROC_GONE	-> PROC_NONE		spa_deactivate()
10235a5a35Jonathan Adams */
10335a5a35Jonathan Adamstypedef enum spa_proc_state {
10435a5a35Jonathan Adams	SPA_PROC_NONE,		/* spa_proc = &p0, no process created */
10535a5a35Jonathan Adams	SPA_PROC_CREATED,	/* spa_activate() has proc, is waiting */
10635a5a35Jonathan Adams	SPA_PROC_ACTIVE,	/* taskqs created, spa_proc set */
10735a5a35Jonathan Adams	SPA_PROC_DEACTIVATE,	/* spa_deactivate() requests process exit */
10835a5a35Jonathan Adams	SPA_PROC_GONE		/* spa_thread() is exiting, spa_proc = &p0 */
10935a5a35Jonathan Adams} spa_proc_state_t;
11035a5a35Jonathan Adams
111ec94d32Adam Leventhaltypedef struct spa_taskqs {
112ec94d32Adam Leventhal	uint_t stqs_count;
113ec94d32Adam Leventhal	taskq_t **stqs_taskq;
114ec94d32Adam Leventhal} spa_taskqs_t;
115ec94d32Adam Leventhal
116fa9e406ahrensstruct spa {
117fa9e406ahrens	/*
118fa9e406ahrens	 * Fields protected by spa_namespace_lock.
119fa9e406ahrens	 */
120e14bb32Jeff Bonwick	char		spa_name[MAXNAMELEN];	/* pool name */
1218704186Dan McDonald	char		*spa_comment;		/* comment */
1220373e76bonwick	avl_node_t	spa_avl;		/* node in spa_namespace_avl */
1230373e76bonwick	nvlist_t	*spa_config;		/* last synced config */
1240373e76bonwick	nvlist_t	*spa_config_syncing;	/* currently syncing config */
1251195e68Mark J Musante	nvlist_t	*spa_config_splitting;	/* config for splitting */
1264b964adGeorge Wilson	nvlist_t	*spa_load_info;		/* info and errors from load */
127fa9e406ahrens	uint64_t	spa_config_txg;		/* txg of last config change */
128fa9e406ahrens	int		spa_sync_pass;		/* iterate-to-convergence */
12988b7b0fMatthew Ahrens	pool_state_t	spa_state;		/* pool state */
130ea8dc4beschrock	int		spa_inject_ref;		/* injection references */
131ea8dc4beschrock	uint8_t		spa_sync_on;		/* sync threads are running */
132ea8dc4beschrock	spa_load_state_t spa_load_state;	/* current load operation */
1334b964adGeorge Wilson	uint64_t	spa_import_flags;	/* import specific flags */
134ec94d32Adam Leventhal	spa_taskqs_t	spa_zio_taskq[ZIO_TYPES][ZIO_TASKQ_TYPES];
135fa9e406ahrens	dsl_pool_t	*spa_dsl_pool;
136ad135b5Christopher Siden	boolean_t	spa_is_initializing;	/* true while opening pool */
137fa9e406ahrens	metaslab_class_t *spa_normal_class;	/* normal data class */
1388654d02perrin	metaslab_class_t *spa_log_class;	/* intent log data class */
139fa9e406ahrens	uint64_t	spa_first_txg;		/* first txg after spa_open() */
1400373e76bonwick	uint64_t	spa_final_txg;		/* txg of export/destroy */
141fa9e406ahrens	uint64_t	spa_freeze_txg;		/* freeze pool at this txg */
142468c413Tim Haley	uint64_t	spa_load_max_txg;	/* best initial ub_txg */
143b24ab67Jeff Bonwick	uint64_t	spa_claim_max_txg;	/* highest claimed birth txg */
14411027bcTim Haley	timespec_t	spa_loaded_ts;		/* 1st successful open time */
145fa9e406ahrens	objset_t	*spa_meta_objset;	/* copy of dp->dp_meta_objset */
146fa9e406ahrens	txg_list_t	spa_vdev_txg_list;	/* per-txg dirty vdev list */
147fa9e406ahrens	vdev_t		*spa_root_vdev;		/* top-level vdev container */
148e9103aaGarrett D'Amore	uint64_t	spa_config_guid;	/* config pool guid */
149e9103aaGarrett D'Amore	uint64_t	spa_load_guid;		/* spa_load initialized guid */
150dfbb943George Wilson	uint64_t	spa_last_synced_guid;	/* last synced guid */
151e14bb32Jeff Bonwick	list_t		spa_config_dirty_list;	/* vdevs with dirty config */
152e14bb32Jeff Bonwick	list_t		spa_state_dirty_list;	/* vdevs with dirty state */
153fa94a07brendan	spa_aux_vdev_t	spa_spares;		/* hot spares */
154fa94a07brendan	spa_aux_vdev_t	spa_l2cache;		/* L2ARC cache devices */
155ad135b5Christopher Siden	nvlist_t	*spa_label_features;	/* Features for reading MOS */
156fa9e406ahrens	uint64_t	spa_config_object;	/* MOS object for pool config */
1578f18d1fGeorge Wilson	uint64_t	spa_config_generation;	/* config generation number */
158fa9e406ahrens	uint64_t	spa_syncing_txg;	/* txg currently syncing */
159cde58dbMatthew Ahrens	bpobj_t		spa_deferred_bpobj;	/* deferred-free bplist */
160b24ab67Jeff Bonwick	bplist_t	spa_free_bplist[TXG_SIZE]; /* bplist of stuff to free */
161fa9e406ahrens	uberblock_t	spa_ubsync;		/* last synced uberblock */
162fa9e406ahrens	uberblock_t	spa_uberblock;		/* current uberblock */
163468c413Tim Haley	boolean_t	spa_extreme_rewind;	/* rewind past deferred frees */
16444ecc53George Wilson	uint64_t	spa_last_io;		/* lbolt of last non-scan I/O */
165fa9e406ahrens	kmutex_t	spa_scrub_lock;		/* resilver/scrub lock */
166fa9e406ahrens	uint64_t	spa_scrub_inflight;	/* in-flight scrub I/Os */
167fa9e406ahrens	kcondvar_t	spa_scrub_io_cv;	/* scrub I/O completion */
168fa9e406ahrens	uint8_t		spa_scrub_active;	/* active or suspended? */
169fa9e406ahrens	uint8_t		spa_scrub_type;		/* type of scrub we're doing */
1700373e76bonwick	uint8_t		spa_scrub_finished;	/* indicator to rotate logs */
171088f389ahrens	uint8_t		spa_scrub_started;	/* started since last boot */
172088f389ahrens	uint8_t		spa_scrub_reopen;	/* scrub doing vdev_reopen */
1733f9d6adLin Ling	uint64_t	spa_scan_pass_start;	/* start time per pass/reboot */
1743f9d6adLin Ling	uint64_t	spa_scan_pass_exam;	/* examined bytes per pass */
175ea8dc4beschrock	kmutex_t	spa_async_lock;		/* protect async state */
176ea8dc4beschrock	kthread_t	*spa_async_thread;	/* thread doing async task */
177ea8dc4beschrock	int		spa_async_suspended;	/* async tasks suspended */
178ea8dc4beschrock	kcondvar_t	spa_async_cv;		/* wait for thread_exit() */
179ea8dc4beschrock	uint16_t	spa_async_tasks;	/* async task mask */
180fa9e406ahrens	char		*spa_root;		/* alternate root directory */
181ea8dc4beschrock	uint64_t	spa_ena;		/* spa-wide ereport ENA */
182468c413Tim Haley	int		spa_last_open_failed;	/* error if last open failed */
183468c413Tim Haley	uint64_t	spa_last_ubsync_txg;	/* "best" uberblock txg */
184468c413Tim Haley	uint64_t	spa_last_ubsync_txg_ts;	/* timestamp from that ub */
185468c413Tim Haley	uint64_t	spa_load_txg;		/* ub txg that loaded */
186468c413Tim Haley	uint64_t	spa_load_txg_ts;	/* timestamp from that ub */
187468c413Tim Haley	uint64_t	spa_load_meta_errors;	/* verify metadata err count */
188468c413Tim Haley	uint64_t	spa_load_data_errors;	/* verify data err count */
189468c413Tim Haley	uint64_t	spa_verify_min_txg;	/* start txg of verify scrub */
190ea8dc4beschrock	kmutex_t	spa_errlog_lock;	/* error log lock */
191ea8dc4beschrock	uint64_t	spa_errlog_last;	/* last error log object */
192ea8dc4beschrock	uint64_t	spa_errlog_scrub;	/* scrub error log object */
193ea8dc4beschrock	kmutex_t	spa_errlist_lock;	/* error list/ereport lock */
194ea8dc4beschrock	avl_tree_t	spa_errlist_last;	/* last error list */
195ea8dc4beschrock	avl_tree_t	spa_errlist_scrub;	/* scrub error list */
19699653d4eschrock	uint64_t	spa_deflate;		/* should we deflate? */
19706eeb2aek	uint64_t	spa_history;		/* history object */
19806eeb2aek	kmutex_t	spa_history_lock;	/* history lock */
19939c2341eschrock	vdev_t		*spa_pending_vdev;	/* pending vdev additions */
200b1b8ab3lling	kmutex_t	spa_props_lock;		/* property lock */
201b1b8ab3lling	uint64_t	spa_pool_props_object;	/* object for properties */
202b1b8ab3lling	uint64_t	spa_bootfs;		/* default boot filesystem */
203e14bb32Jeff Bonwick	uint64_t	spa_failmode;		/* failure mode for the pool */
204e14bb32Jeff Bonwick	uint64_t	spa_delegation;		/* delegation on/off */
205c5904d1eschrock	list_t		spa_config_list;	/* previous cache file(s) */
20654d692bGeorge Wilson	zio_t		*spa_async_zio_root;	/* root of all async I/O */
207e14bb32Jeff Bonwick	zio_t		*spa_suspend_zio_root;	/* root of all suspended I/O */
208e14bb32Jeff Bonwick	kmutex_t	spa_suspend_lock;	/* protects suspend_zio_root */
209e14bb32Jeff Bonwick	kcondvar_t	spa_suspend_cv;		/* notification of resume */
210e14bb32Jeff Bonwick	uint8_t		spa_suspended;		/* pool is suspended */
211b24ab67Jeff Bonwick	uint8_t		spa_claiming;		/* pool is doing zil_claim() */
21209c9d37George Wilson	boolean_t	spa_debug;		/* debug enabled? */
213bf82a41eschrock	boolean_t	spa_is_root;		/* pool is root */
214088f389ahrens	int		spa_minref;		/* num refs when first opened */
2158ad4d6dJeff Bonwick	int		spa_mode;		/* FREAD | FWRITE */
216b87f3afperrin	spa_log_state_t spa_log_state;		/* log state */
217573ca77George Wilson	uint64_t	spa_autoexpand;		/* lun expansion on/off */
218b24ab67Jeff Bonwick	ddt_t		*spa_ddt[ZIO_CHECKSUM_FUNCTIONS]; /* in-core DDTs */
219b24ab67Jeff Bonwick	uint64_t	spa_ddt_stat_object;	/* DDT statistics */
220b24ab67Jeff Bonwick	uint64_t	spa_dedup_ditto;	/* dedup ditto threshold */
221b24ab67Jeff Bonwick	uint64_t	spa_dedup_checksum;	/* default dedup checksum */
222485bbbfGeorge Wilson	uint64_t	spa_dspace;		/* dspace in normal class */
223a152156Jeff Bonwick	kmutex_t	spa_vdev_top_lock;	/* dueling offline/remove */
22435a5a35Jonathan Adams	kmutex_t	spa_proc_lock;		/* protects spa_proc* */
22535a5a35Jonathan Adams	kcondvar_t	spa_proc_cv;		/* spa_proc_state transitions */
22635a5a35Jonathan Adams	spa_proc_state_t spa_proc_state;	/* see definition */
22735a5a35Jonathan Adams	struct proc	*spa_proc;		/* "zpool-poolname" process */
22835a5a35Jonathan Adams	uint64_t	spa_did;		/* if procp != p0, did of t1 */
229b693757Eric Schrock	boolean_t	spa_autoreplace;	/* autoreplace set in open */
2308f18d1fGeorge Wilson	int		spa_vdev_locks;		/* locks grabbed */
2313f9d6adLin Ling	uint64_t	spa_creation_version;	/* version at pool creation */
232ad135b5Christopher Siden	uint64_t	spa_prev_software_version; /* See ub_software_version */
233ad135b5Christopher Siden	uint64_t	spa_feat_for_write_obj;	/* required to write to pool */
234ad135b5Christopher Siden	uint64_t	spa_feat_for_read_obj;	/* required to read from pool */
235ad135b5Christopher Siden	uint64_t	spa_feat_desc_obj;	/* Feature descriptions */
23643466aaMax Grossman	uint64_t	spa_feat_enabled_txg_obj; /* Feature enabled txg */
23743466aaMax Grossman	/* cache feature refcounts */
23843466aaMax Grossman	uint64_t	spa_feat_refcount_cache[SPA_FEATURES];
239283b846George.Wilson	cyclic_id_t	spa_deadman_cycid;	/* cyclic id */
240283b846George.Wilson	uint64_t	spa_deadman_calls;	/* number of deadman calls */
24169962b5Matthew Ahrens	hrtime_t	spa_sync_starttime;	/* starting time fo spa_sync */
242283b846George.Wilson	uint64_t	spa_deadman_synctime;	/* deadman expiration timer */
24369962b5Matthew Ahrens
24469962b5Matthew Ahrens	/*
24569962b5Matthew Ahrens	 * spa_iokstat_lock protects spa_iokstat and
24669962b5Matthew Ahrens	 * spa_queue_stats[].
24769962b5Matthew Ahrens	 */
24869962b5Matthew Ahrens	kmutex_t	spa_iokstat_lock;
249c3a6601Matthew Ahrens	struct kstat	*spa_iokstat;		/* kstat of io to this pool */
25069962b5Matthew Ahrens	struct {
25169962b5Matthew Ahrens		int spa_active;
25269962b5Matthew Ahrens		int spa_queued;
25369962b5Matthew Ahrens	} spa_queue_stats[ZIO_PRIORITY_NUM_QUEUEABLE];
25469962b5Matthew Ahrens
2553cb69f7Will Andrews	hrtime_t	spa_ccw_fail_time;	/* Conf cache write fail time */
25669962b5Matthew Ahrens
257fa9e406ahrens	/*
258f717074Will Andrews	 * spa_refcount & spa_config_lock must be the last elements
25991ebeefahrens	 * because refcount_t changes size based on compilation options.
26091ebeefahrens	 * In order for the MDB module to function correctly, the other
26191ebeefahrens	 * fields must remain in the same location.
262fa9e406ahrens	 */
263e14bb32Jeff Bonwick	spa_config_lock_t spa_config_lock[SCL_LOCKS]; /* config changes */
264fa9e406ahrens	refcount_t	spa_refcount;		/* number of opens */
267c5904d1eschrockextern const char *spa_config_path;
269ec94d32Adam Leventhalextern void spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
270ec94d32Adam Leventhal    task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent);
271ec94d32Adam Leventhal
272fa9e406ahrens#ifdef	__cplusplus
276fa9e406ahrens#endif	/* _SYS_SPA_IMPL_H */