spa_impl.h revision 45818ee124adeaaf947698996b4f4c722afc6d1f
1fa9e406ahrens/*
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 http://www.opensolaris.org/os/licensing.
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 */
21fa9e406ahrens/*
223f9d6adLin Ling * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
2381cd5c5Matthew Ahrens * Copyright (c) 2011, 2015 by Delphix. All rights reserved.
24e9103aaGarrett D'Amore * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
25bc9014eJustin Gibbs * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
2645818eeMatthew Ahrens * Copyright 2013 Saso Kiselkov. All rights reserved.
27fa9e406ahrens */
28fa9e406ahrens
29fa9e406ahrens#ifndef _SYS_SPA_IMPL_H
30fa9e406ahrens#define	_SYS_SPA_IMPL_H
31fa9e406ahrens
32fa9e406ahrens#include <sys/spa.h>
33fa9e406ahrens#include <sys/vdev.h>
34fa9e406ahrens#include <sys/metaslab.h>
35fa9e406ahrens#include <sys/dmu.h>
36fa9e406ahrens#include <sys/dsl_pool.h>
37fa9e406ahrens#include <sys/uberblock_impl.h>
38fa9e406ahrens#include <sys/zfs_context.h>
39fa9e406ahrens#include <sys/avl.h>
40fa9e406ahrens#include <sys/refcount.h>
41fa9e406ahrens#include <sys/bplist.h>
42cde58dbMatthew Ahrens#include <sys/bpobj.h>
435d7b4d4Matthew Ahrens#include <sys/zfeature.h>
4443466aaMax Grossman#include <zfeature_common.h>
45fa9e406ahrens
46fa9e406ahrens#ifdef	__cplusplus
47fa9e406ahrensextern "C" {
48fa9e406ahrens#endif
49fa9e406ahrens
50ea8dc4beschrocktypedef struct spa_error_entry {
517802d7bMatthew Ahrens	zbookmark_phys_t	se_bookmark;
527802d7bMatthew Ahrens	char			*se_name;
537802d7bMatthew Ahrens	avl_node_t		se_avl;
54ea8dc4beschrock} spa_error_entry_t;
55ea8dc4beschrock
5606eeb2aektypedef struct spa_history_phys {
5706eeb2aek	uint64_t sh_pool_create_len;	/* ending offset of zpool create */
5806eeb2aek	uint64_t sh_phys_max_off;	/* physical EOF */
5906eeb2aek	uint64_t sh_bof;		/* logical BOF */
6006eeb2aek	uint64_t sh_eof;		/* logical EOF */
6106eeb2aek	uint64_t sh_records_lost;	/* num of records overwritten */
6206eeb2aek} spa_history_phys_t;
6306eeb2aek
64fa94a07brendanstruct spa_aux_vdev {
65fa94a07brendan	uint64_t	sav_object;		/* MOS object for device list */
66fa94a07brendan	nvlist_t	*sav_config;		/* cached device config */
67fa94a07brendan	vdev_t		**sav_vdevs;		/* devices */
68fa94a07brendan	int		sav_count;		/* number devices */
69fa94a07brendan	boolean_t	sav_sync;		/* sync the device list */
70fa94a07brendan	nvlist_t	**sav_pending;		/* pending device additions */
71fa94a07brendan	uint_t		sav_npending;		/* # pending devices */
72fa94a07brendan};
73fa94a07brendan
74e05725bbonwicktypedef struct spa_config_lock {
75e05725bbonwick	kmutex_t	scl_lock;
76e05725bbonwick	kthread_t	*scl_writer;
77e14bb32Jeff Bonwick	int		scl_write_wanted;
78e05725bbonwick	kcondvar_t	scl_cv;
79e05725bbonwick	refcount_t	scl_count;
80e05725bbonwick} spa_config_lock_t;
81e05725bbonwick
82c5904d1eschrocktypedef struct spa_config_dirent {
83c5904d1eschrock	list_node_t	scd_link;
84c5904d1eschrock	char		*scd_path;
85c5904d1eschrock} spa_config_dirent_t;
86c5904d1eschrock
87ec94d32Adam Leventhaltypedef enum zio_taskq_type {
88e14bb32Jeff Bonwick	ZIO_TASKQ_ISSUE = 0,
8980eb36fGeorge Wilson	ZIO_TASKQ_ISSUE_HIGH,
90e14bb32Jeff Bonwick	ZIO_TASKQ_INTERRUPT,
9180eb36fGeorge Wilson	ZIO_TASKQ_INTERRUPT_HIGH,
92e14bb32Jeff Bonwick	ZIO_TASKQ_TYPES
93ec94d32Adam Leventhal} zio_taskq_type_t;
94e14bb32Jeff Bonwick
9535a5a35Jonathan Adams/*
96ec94d32Adam Leventhal * State machine for the zpool-poolname process.  The states transitions
9735a5a35Jonathan Adams * are done as follows:
9835a5a35Jonathan Adams *
9935a5a35Jonathan Adams *	From		   To			Routine
10035a5a35Jonathan Adams *	PROC_NONE	-> PROC_CREATED		spa_activate()
10135a5a35Jonathan Adams *	PROC_CREATED	-> PROC_ACTIVE		spa_thread()
10235a5a35Jonathan Adams *	PROC_ACTIVE	-> PROC_DEACTIVATE	spa_deactivate()
10335a5a35Jonathan Adams *	PROC_DEACTIVATE	-> PROC_GONE		spa_thread()
10435a5a35Jonathan Adams *	PROC_GONE	-> PROC_NONE		spa_deactivate()
10535a5a35Jonathan Adams */
10635a5a35Jonathan Adamstypedef enum spa_proc_state {
10735a5a35Jonathan Adams	SPA_PROC_NONE,		/* spa_proc = &p0, no process created */
10835a5a35Jonathan Adams	SPA_PROC_CREATED,	/* spa_activate() has proc, is waiting */
10935a5a35Jonathan Adams	SPA_PROC_ACTIVE,	/* taskqs created, spa_proc set */
11035a5a35Jonathan Adams	SPA_PROC_DEACTIVATE,	/* spa_deactivate() requests process exit */
11135a5a35Jonathan Adams	SPA_PROC_GONE		/* spa_thread() is exiting, spa_proc = &p0 */
11235a5a35Jonathan Adams} spa_proc_state_t;
11335a5a35Jonathan Adams
114ec94d32Adam Leventhaltypedef struct spa_taskqs {
115ec94d32Adam Leventhal	uint_t stqs_count;
116ec94d32Adam Leventhal	taskq_t **stqs_taskq;
117ec94d32Adam Leventhal} spa_taskqs_t;
118ec94d32Adam Leventhal
119fa9e406ahrensstruct spa {
120fa9e406ahrens	/*
121fa9e406ahrens	 * Fields protected by spa_namespace_lock.
122fa9e406ahrens	 */
123e14bb32Jeff Bonwick	char		spa_name[MAXNAMELEN];	/* pool name */
1248704186Dan McDonald	char		*spa_comment;		/* comment */
1250373e76bonwick	avl_node_t	spa_avl;		/* node in spa_namespace_avl */
1260373e76bonwick	nvlist_t	*spa_config;		/* last synced config */
1270373e76bonwick	nvlist_t	*spa_config_syncing;	/* currently syncing config */
1281195e68Mark J Musante	nvlist_t	*spa_config_splitting;	/* config for splitting */
1294b964adGeorge Wilson	nvlist_t	*spa_load_info;		/* info and errors from load */
130fa9e406ahrens	uint64_t	spa_config_txg;		/* txg of last config change */
131fa9e406ahrens	int		spa_sync_pass;		/* iterate-to-convergence */
13288b7b0fMatthew Ahrens	pool_state_t	spa_state;		/* pool state */
133ea8dc4beschrock	int		spa_inject_ref;		/* injection references */
134ea8dc4beschrock	uint8_t		spa_sync_on;		/* sync threads are running */
135ea8dc4beschrock	spa_load_state_t spa_load_state;	/* current load operation */
1364b964adGeorge Wilson	uint64_t	spa_import_flags;	/* import specific flags */
137ec94d32Adam Leventhal	spa_taskqs_t	spa_zio_taskq[ZIO_TYPES][ZIO_TASKQ_TYPES];
138fa9e406ahrens	dsl_pool_t	*spa_dsl_pool;
139ad135b5Christopher Siden	boolean_t	spa_is_initializing;	/* true while opening pool */
140fa9e406ahrens	metaslab_class_t *spa_normal_class;	/* normal data class */
1418654d02perrin	metaslab_class_t *spa_log_class;	/* intent log data class */
142fa9e406ahrens	uint64_t	spa_first_txg;		/* first txg after spa_open() */
1430373e76bonwick	uint64_t	spa_final_txg;		/* txg of export/destroy */
144fa9e406ahrens	uint64_t	spa_freeze_txg;		/* freeze pool at this txg */
145468c413Tim Haley	uint64_t	spa_load_max_txg;	/* best initial ub_txg */
146b24ab67Jeff Bonwick	uint64_t	spa_claim_max_txg;	/* highest claimed birth txg */
14711027bcTim Haley	timespec_t	spa_loaded_ts;		/* 1st successful open time */
148fa9e406ahrens	objset_t	*spa_meta_objset;	/* copy of dp->dp_meta_objset */
149bc9014eJustin Gibbs	kmutex_t	spa_evicting_os_lock;	/* Evicting objset list lock */
150bc9014eJustin Gibbs	list_t		spa_evicting_os_list;	/* Objsets being evicted. */
151bc9014eJustin Gibbs	kcondvar_t	spa_evicting_os_cv;	/* Objset Eviction Completion */
152fa9e406ahrens	txg_list_t	spa_vdev_txg_list;	/* per-txg dirty vdev list */
153fa9e406ahrens	vdev_t		*spa_root_vdev;		/* top-level vdev container */
15481cd5c5Matthew Ahrens	int		spa_min_ashift;		/* of vdevs in normal class */
15581cd5c5Matthew Ahrens	int		spa_max_ashift;		/* of vdevs in normal class */
156e9103aaGarrett D'Amore	uint64_t	spa_config_guid;	/* config pool guid */
157e9103aaGarrett D'Amore	uint64_t	spa_load_guid;		/* spa_load initialized guid */
158dfbb943George Wilson	uint64_t	spa_last_synced_guid;	/* last synced guid */
159e14bb32Jeff Bonwick	list_t		spa_config_dirty_list;	/* vdevs with dirty config */
160e14bb32Jeff Bonwick	list_t		spa_state_dirty_list;	/* vdevs with dirty state */
161fa94a07brendan	spa_aux_vdev_t	spa_spares;		/* hot spares */
162fa94a07brendan	spa_aux_vdev_t	spa_l2cache;		/* L2ARC cache devices */
163ad135b5Christopher Siden	nvlist_t	*spa_label_features;	/* Features for reading MOS */
164fa9e406ahrens	uint64_t	spa_config_object;	/* MOS object for pool config */
1658f18d1fGeorge Wilson	uint64_t	spa_config_generation;	/* config generation number */
166fa9e406ahrens	uint64_t	spa_syncing_txg;	/* txg currently syncing */
167cde58dbMatthew Ahrens	bpobj_t		spa_deferred_bpobj;	/* deferred-free bplist */
168b24ab67Jeff Bonwick	bplist_t	spa_free_bplist[TXG_SIZE]; /* bplist of stuff to free */
16945818eeMatthew Ahrens	zio_cksum_salt_t spa_cksum_salt;	/* secret salt for cksum */
17045818eeMatthew Ahrens	/* checksum context templates */
17145818eeMatthew Ahrens	kmutex_t	spa_cksum_tmpls_lock;
17245818eeMatthew Ahrens	void		*spa_cksum_tmpls[ZIO_CHECKSUM_FUNCTIONS];
173fa9e406ahrens	uberblock_t	spa_ubsync;		/* last synced uberblock */
174fa9e406ahrens	uberblock_t	spa_uberblock;		/* current uberblock */
175468c413Tim Haley	boolean_t	spa_extreme_rewind;	/* rewind past deferred frees */
17644ecc53George Wilson	uint64_t	spa_last_io;		/* lbolt of last non-scan I/O */
177fa9e406ahrens	kmutex_t	spa_scrub_lock;		/* resilver/scrub lock */
178fa9e406ahrens	uint64_t	spa_scrub_inflight;	/* in-flight scrub I/Os */
179fa9e406ahrens	kcondvar_t	spa_scrub_io_cv;	/* scrub I/O completion */
180fa9e406ahrens	uint8_t		spa_scrub_active;	/* active or suspended? */
181fa9e406ahrens	uint8_t		spa_scrub_type;		/* type of scrub we're doing */
1820373e76bonwick	uint8_t		spa_scrub_finished;	/* indicator to rotate logs */
183088f389ahrens	uint8_t		spa_scrub_started;	/* started since last boot */
184088f389ahrens	uint8_t		spa_scrub_reopen;	/* scrub doing vdev_reopen */
1853f9d6adLin Ling	uint64_t	spa_scan_pass_start;	/* start time per pass/reboot */
1863f9d6adLin Ling	uint64_t	spa_scan_pass_exam;	/* examined bytes per pass */
187ea8dc4beschrock	kmutex_t	spa_async_lock;		/* protect async state */
188ea8dc4beschrock	kthread_t	*spa_async_thread;	/* thread doing async task */
189ea8dc4beschrock	int		spa_async_suspended;	/* async tasks suspended */
190ea8dc4beschrock	kcondvar_t	spa_async_cv;		/* wait for thread_exit() */
191ea8dc4beschrock	uint16_t	spa_async_tasks;	/* async task mask */
192fa9e406ahrens	char		*spa_root;		/* alternate root directory */
193ea8dc4beschrock	uint64_t	spa_ena;		/* spa-wide ereport ENA */
194468c413Tim Haley	int		spa_last_open_failed;	/* error if last open failed */
195468c413Tim Haley	uint64_t	spa_last_ubsync_txg;	/* "best" uberblock txg */
196468c413Tim Haley	uint64_t	spa_last_ubsync_txg_ts;	/* timestamp from that ub */
197468c413Tim Haley	uint64_t	spa_load_txg;		/* ub txg that loaded */
198468c413Tim Haley	uint64_t	spa_load_txg_ts;	/* timestamp from that ub */
199468c413Tim Haley	uint64_t	spa_load_meta_errors;	/* verify metadata err count */
200468c413Tim Haley	uint64_t	spa_load_data_errors;	/* verify data err count */
201468c413Tim Haley	uint64_t	spa_verify_min_txg;	/* start txg of verify scrub */
202ea8dc4beschrock	kmutex_t	spa_errlog_lock;	/* error log lock */
203ea8dc4beschrock	uint64_t	spa_errlog_last;	/* last error log object */
204ea8dc4beschrock	uint64_t	spa_errlog_scrub;	/* scrub error log object */
205ea8dc4beschrock	kmutex_t	spa_errlist_lock;	/* error list/ereport lock */
206ea8dc4beschrock	avl_tree_t	spa_errlist_last;	/* last error list */
207ea8dc4beschrock	avl_tree_t	spa_errlist_scrub;	/* scrub error list */
20899653d4eschrock	uint64_t	spa_deflate;		/* should we deflate? */
20906eeb2aek	uint64_t	spa_history;		/* history object */
21006eeb2aek	kmutex_t	spa_history_lock;	/* history lock */
21139c2341eschrock	vdev_t		*spa_pending_vdev;	/* pending vdev additions */
212b1b8ab3lling	kmutex_t	spa_props_lock;		/* property lock */
213b1b8ab3lling	uint64_t	spa_pool_props_object;	/* object for properties */
214b1b8ab3lling	uint64_t	spa_bootfs;		/* default boot filesystem */
215e14bb32Jeff Bonwick	uint64_t	spa_failmode;		/* failure mode for the pool */
216e14bb32Jeff Bonwick	uint64_t	spa_delegation;		/* delegation on/off */
217c5904d1eschrock	list_t		spa_config_list;	/* previous cache file(s) */
2186f834bcMatthew Ahrens	/* per-CPU array of root of async I/O: */
2196f834bcMatthew Ahrens	zio_t		**spa_async_zio_root;
220e14bb32Jeff Bonwick	zio_t		*spa_suspend_zio_root;	/* root of all suspended I/O */
221e14bb32Jeff Bonwick	kmutex_t	spa_suspend_lock;	/* protects suspend_zio_root */
222e14bb32Jeff Bonwick	kcondvar_t	spa_suspend_cv;		/* notification of resume */
223e14bb32Jeff Bonwick	uint8_t		spa_suspended;		/* pool is suspended */
224b24ab67Jeff Bonwick	uint8_t		spa_claiming;		/* pool is doing zil_claim() */
22509c9d37George Wilson	boolean_t	spa_debug;		/* debug enabled? */
226bf82a41eschrock	boolean_t	spa_is_root;		/* pool is root */
227088f389ahrens	int		spa_minref;		/* num refs when first opened */
2288ad4d6dJeff Bonwick	int		spa_mode;		/* FREAD | FWRITE */
229b87f3afperrin	spa_log_state_t spa_log_state;		/* log state */
230573ca77George Wilson	uint64_t	spa_autoexpand;		/* lun expansion on/off */
231b24ab67Jeff Bonwick	ddt_t		*spa_ddt[ZIO_CHECKSUM_FUNCTIONS]; /* in-core DDTs */
232b24ab67Jeff Bonwick	uint64_t	spa_ddt_stat_object;	/* DDT statistics */
233b24ab67Jeff Bonwick	uint64_t	spa_dedup_ditto;	/* dedup ditto threshold */
234b24ab67Jeff Bonwick	uint64_t	spa_dedup_checksum;	/* default dedup checksum */
235485bbbfGeorge Wilson	uint64_t	spa_dspace;		/* dspace in normal class */
236a152156Jeff Bonwick	kmutex_t	spa_vdev_top_lock;	/* dueling offline/remove */
23735a5a35Jonathan Adams	kmutex_t	spa_proc_lock;		/* protects spa_proc* */
23835a5a35Jonathan Adams	kcondvar_t	spa_proc_cv;		/* spa_proc_state transitions */
23935a5a35Jonathan Adams	spa_proc_state_t spa_proc_state;	/* see definition */
24035a5a35Jonathan Adams	struct proc	*spa_proc;		/* "zpool-poolname" process */
24135a5a35Jonathan Adams	uint64_t	spa_did;		/* if procp != p0, did of t1 */
242b693757Eric Schrock	boolean_t	spa_autoreplace;	/* autoreplace set in open */
2438f18d1fGeorge Wilson	int		spa_vdev_locks;		/* locks grabbed */
2443f9d6adLin Ling	uint64_t	spa_creation_version;	/* version at pool creation */
245ad135b5Christopher Siden	uint64_t	spa_prev_software_version; /* See ub_software_version */
246ad135b5Christopher Siden	uint64_t	spa_feat_for_write_obj;	/* required to write to pool */
247ad135b5Christopher Siden	uint64_t	spa_feat_for_read_obj;	/* required to read from pool */
248ad135b5Christopher Siden	uint64_t	spa_feat_desc_obj;	/* Feature descriptions */
24943466aaMax Grossman	uint64_t	spa_feat_enabled_txg_obj; /* Feature enabled txg */
25043466aaMax Grossman	/* cache feature refcounts */
25143466aaMax Grossman	uint64_t	spa_feat_refcount_cache[SPA_FEATURES];
252283b846George.Wilson	cyclic_id_t	spa_deadman_cycid;	/* cyclic id */
253283b846George.Wilson	uint64_t	spa_deadman_calls;	/* number of deadman calls */
25469962b5Matthew Ahrens	hrtime_t	spa_sync_starttime;	/* starting time fo spa_sync */
255283b846George.Wilson	uint64_t	spa_deadman_synctime;	/* deadman expiration timer */
25669962b5Matthew Ahrens
25769962b5Matthew Ahrens	/*
25869962b5Matthew Ahrens	 * spa_iokstat_lock protects spa_iokstat and
25969962b5Matthew Ahrens	 * spa_queue_stats[].
26069962b5Matthew Ahrens	 */
26169962b5Matthew Ahrens	kmutex_t	spa_iokstat_lock;
262c3a6601Matthew Ahrens	struct kstat	*spa_iokstat;		/* kstat of io to this pool */
26369962b5Matthew Ahrens	struct {
26469962b5Matthew Ahrens		int spa_active;
26569962b5Matthew Ahrens		int spa_queued;
26669962b5Matthew Ahrens	} spa_queue_stats[ZIO_PRIORITY_NUM_QUEUEABLE];
26769962b5Matthew Ahrens
2683cb69f7Will Andrews	hrtime_t	spa_ccw_fail_time;	/* Conf cache write fail time */
26969962b5Matthew Ahrens
270fa9e406ahrens	/*
271f717074Will Andrews	 * spa_refcount & spa_config_lock must be the last elements
27291ebeefahrens	 * because refcount_t changes size based on compilation options.
27391ebeefahrens	 * In order for the MDB module to function correctly, the other
27491ebeefahrens	 * fields must remain in the same location.
275fa9e406ahrens	 */
276e14bb32Jeff Bonwick	spa_config_lock_t spa_config_lock[SCL_LOCKS]; /* config changes */
277fa9e406ahrens	refcount_t	spa_refcount;		/* number of opens */
278fa9e406ahrens};
279fa9e406ahrens
280c5904d1eschrockextern const char *spa_config_path;
281fa9e406ahrens
282ec94d32Adam Leventhalextern void spa_taskq_dispatch_ent(spa_t *spa, zio_type_t t, zio_taskq_type_t q,
283ec94d32Adam Leventhal    task_func_t *func, void *arg, uint_t flags, taskq_ent_t *ent);
284ec94d32Adam Leventhal
285fa9e406ahrens#ifdef	__cplusplus
286fa9e406ahrens}
287fa9e406ahrens#endif
288fa9e406ahrens
289fa9e406ahrens#endif	/* _SYS_SPA_IMPL_H */
290