md: restart recovery cleanly after device failure.
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / include / linux / raid / md_k.h
... / ...
CommitLineData
1/*
2 md_k.h : kernel internal structure of the Linux MD driver
3 Copyright (C) 1996-98 Ingo Molnar, Gadi Oxman
4
5 This program is free software; you can redistribute it and/or modify
6 it under the terms of the GNU General Public License as published by
7 the Free Software Foundation; either version 2, or (at your option)
8 any later version.
9
10 You should have received a copy of the GNU General Public License
11 (for example /usr/src/linux/COPYING); if not, write to the Free
12 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
13*/
14
15#ifndef _MD_K_H
16#define _MD_K_H
17
18/* and dm-bio-list.h is not under include/linux because.... ??? */
19#include "../../../drivers/md/dm-bio-list.h"
20
21#ifdef CONFIG_BLOCK
22
23#define LEVEL_MULTIPATH (-4)
24#define LEVEL_LINEAR (-1)
25#define LEVEL_FAULTY (-5)
26
27/* we need a value for 'no level specified' and 0
28 * means 'raid0', so we need something else. This is
29 * for internal use only
30 */
31#define LEVEL_NONE (-1000000)
32
33#define MaxSector (~(sector_t)0)
34
35typedef struct mddev_s mddev_t;
36typedef struct mdk_rdev_s mdk_rdev_t;
37
38/*
39 * options passed in raidrun:
40 */
41
42/* Currently this must fit in an 'int' */
43#define MAX_CHUNK_SIZE (1<<30)
44
45/*
46 * MD's 'extended' device
47 */
48struct mdk_rdev_s
49{
50 struct list_head same_set; /* RAID devices within the same set */
51
52 sector_t size; /* Device size (in blocks) */
53 mddev_t *mddev; /* RAID array if running */
54 long last_events; /* IO event timestamp */
55
56 struct block_device *bdev; /* block device handle */
57
58 struct page *sb_page;
59 int sb_loaded;
60 __u64 sb_events;
61 sector_t data_offset; /* start of data in array */
62 sector_t sb_offset;
63 int sb_size; /* bytes in the superblock */
64 int preferred_minor; /* autorun support */
65
66 struct kobject kobj;
67
68 /* A device can be in one of three states based on two flags:
69 * Not working: faulty==1 in_sync==0
70 * Fully working: faulty==0 in_sync==1
71 * Working, but not
72 * in sync with array
73 * faulty==0 in_sync==0
74 *
75 * It can never have faulty==1, in_sync==1
76 * This reduces the burden of testing multiple flags in many cases
77 */
78
79 unsigned long flags;
80#define Faulty 1 /* device is known to have a fault */
81#define In_sync 2 /* device is in_sync with rest of array */
82#define WriteMostly 4 /* Avoid reading if at all possible */
83#define BarriersNotsupp 5 /* BIO_RW_BARRIER is not supported */
84#define AllReserved 6 /* If whole device is reserved for
85 * one array */
86#define AutoDetected 7 /* added by auto-detect */
87#define Blocked 8 /* An error occured on an externally
88 * managed array, don't allow writes
89 * until it is cleared */
90 wait_queue_head_t blocked_wait;
91
92 int desc_nr; /* descriptor index in the superblock */
93 int raid_disk; /* role of device in array */
94 int saved_raid_disk; /* role that device used to have in the
95 * array and could again if we did a partial
96 * resync from the bitmap
97 */
98 sector_t recovery_offset;/* If this device has been partially
99 * recovered, this is where we were
100 * up to.
101 */
102
103 atomic_t nr_pending; /* number of pending requests.
104 * only maintained for arrays that
105 * support hot removal
106 */
107 atomic_t read_errors; /* number of consecutive read errors that
108 * we have tried to ignore.
109 */
110 atomic_t corrected_errors; /* number of corrected read errors,
111 * for reporting to userspace and storing
112 * in superblock.
113 */
114 struct work_struct del_work; /* used for delayed sysfs removal */
115};
116
117struct mddev_s
118{
119 void *private;
120 struct mdk_personality *pers;
121 dev_t unit;
122 int md_minor;
123 struct list_head disks;
124 unsigned long flags;
125#define MD_CHANGE_DEVS 0 /* Some device status has changed */
126#define MD_CHANGE_CLEAN 1 /* transition to or from 'clean' */
127#define MD_CHANGE_PENDING 2 /* superblock update in progress */
128
129 int ro;
130
131 struct gendisk *gendisk;
132
133 struct kobject kobj;
134
135 /* Superblock information */
136 int major_version,
137 minor_version,
138 patch_version;
139 int persistent;
140 int external; /* metadata is
141 * managed externally */
142 char metadata_type[17]; /* externally set*/
143 int chunk_size;
144 time_t ctime, utime;
145 int level, layout;
146 char clevel[16];
147 int raid_disks;
148 int max_disks;
149 sector_t size; /* used size of component devices */
150 sector_t array_size; /* exported array size */
151 __u64 events;
152
153 char uuid[16];
154
155 /* If the array is being reshaped, we need to record the
156 * new shape and an indication of where we are up to.
157 * This is written to the superblock.
158 * If reshape_position is MaxSector, then no reshape is happening (yet).
159 */
160 sector_t reshape_position;
161 int delta_disks, new_level, new_layout, new_chunk;
162
163 struct mdk_thread_s *thread; /* management thread */
164 struct mdk_thread_s *sync_thread; /* doing resync or reconstruct */
165 sector_t curr_resync; /* last block scheduled */
166 unsigned long resync_mark; /* a recent timestamp */
167 sector_t resync_mark_cnt;/* blocks written at resync_mark */
168 sector_t curr_mark_cnt; /* blocks scheduled now */
169
170 sector_t resync_max_sectors; /* may be set by personality */
171
172 sector_t resync_mismatches; /* count of sectors where
173 * parity/replica mismatch found
174 */
175
176 /* allow user-space to request suspension of IO to regions of the array */
177 sector_t suspend_lo;
178 sector_t suspend_hi;
179 /* if zero, use the system-wide default */
180 int sync_speed_min;
181 int sync_speed_max;
182
183 /* resync even though the same disks are shared among md-devices */
184 int parallel_resync;
185
186 int ok_start_degraded;
187 /* recovery/resync flags
188 * NEEDED: we might need to start a resync/recover
189 * RUNNING: a thread is running, or about to be started
190 * SYNC: actually doing a resync, not a recovery
191 * INTR: resync needs to be aborted for some reason
192 * DONE: thread is done and is waiting to be reaped
193 * REQUEST: user-space has requested a sync (used with SYNC)
194 * CHECK: user-space request for for check-only, no repair
195 * RESHAPE: A reshape is happening
196 *
197 * If neither SYNC or RESHAPE are set, then it is a recovery.
198 */
199#define MD_RECOVERY_RUNNING 0
200#define MD_RECOVERY_SYNC 1
201#define MD_RECOVERY_INTR 3
202#define MD_RECOVERY_DONE 4
203#define MD_RECOVERY_NEEDED 5
204#define MD_RECOVERY_REQUESTED 6
205#define MD_RECOVERY_CHECK 7
206#define MD_RECOVERY_RESHAPE 8
207#define MD_RECOVERY_FROZEN 9
208
209 unsigned long recovery;
210
211 int in_sync; /* know to not need resync */
212 struct mutex reconfig_mutex;
213 atomic_t active;
214
215 int changed; /* true if we might need to reread partition info */
216 int degraded; /* whether md should consider
217 * adding a spare
218 */
219 int barriers_work; /* initialised to true, cleared as soon
220 * as a barrier request to slave
221 * fails. Only supported
222 */
223 struct bio *biolist; /* bios that need to be retried
224 * because BIO_RW_BARRIER is not supported
225 */
226
227 atomic_t recovery_active; /* blocks scheduled, but not written */
228 wait_queue_head_t recovery_wait;
229 sector_t recovery_cp;
230 sector_t resync_max; /* resync should pause
231 * when it gets here */
232
233 spinlock_t write_lock;
234 wait_queue_head_t sb_wait; /* for waiting on superblock updates */
235 atomic_t pending_writes; /* number of active superblock writes */
236
237 unsigned int safemode; /* if set, update "clean" superblock
238 * when no writes pending.
239 */
240 unsigned int safemode_delay;
241 struct timer_list safemode_timer;
242 atomic_t writes_pending;
243 struct request_queue *queue; /* for plugging ... */
244
245 atomic_t write_behind; /* outstanding async IO */
246 unsigned int max_write_behind; /* 0 = sync */
247
248 struct bitmap *bitmap; /* the bitmap for the device */
249 struct file *bitmap_file; /* the bitmap file */
250 long bitmap_offset; /* offset from superblock of
251 * start of bitmap. May be
252 * negative, but not '0'
253 */
254 long default_bitmap_offset; /* this is the offset to use when
255 * hot-adding a bitmap. It should
256 * eventually be settable by sysfs.
257 */
258
259 struct list_head all_mddevs;
260};
261
262
263static inline void rdev_dec_pending(mdk_rdev_t *rdev, mddev_t *mddev)
264{
265 int faulty = test_bit(Faulty, &rdev->flags);
266 if (atomic_dec_and_test(&rdev->nr_pending) && faulty)
267 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
268}
269
270static inline void md_sync_acct(struct block_device *bdev, unsigned long nr_sectors)
271{
272 atomic_add(nr_sectors, &bdev->bd_contains->bd_disk->sync_io);
273}
274
275struct mdk_personality
276{
277 char *name;
278 int level;
279 struct list_head list;
280 struct module *owner;
281 int (*make_request)(struct request_queue *q, struct bio *bio);
282 int (*run)(mddev_t *mddev);
283 int (*stop)(mddev_t *mddev);
284 void (*status)(struct seq_file *seq, mddev_t *mddev);
285 /* error_handler must set ->faulty and clear ->in_sync
286 * if appropriate, and should abort recovery if needed
287 */
288 void (*error_handler)(mddev_t *mddev, mdk_rdev_t *rdev);
289 int (*hot_add_disk) (mddev_t *mddev, mdk_rdev_t *rdev);
290 int (*hot_remove_disk) (mddev_t *mddev, int number);
291 int (*spare_active) (mddev_t *mddev);
292 sector_t (*sync_request)(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster);
293 int (*resize) (mddev_t *mddev, sector_t sectors);
294 int (*check_reshape) (mddev_t *mddev);
295 int (*start_reshape) (mddev_t *mddev);
296 int (*reconfig) (mddev_t *mddev, int layout, int chunk_size);
297 /* quiesce moves between quiescence states
298 * 0 - fully active
299 * 1 - no new requests allowed
300 * others - reserved
301 */
302 void (*quiesce) (mddev_t *mddev, int state);
303};
304
305
306struct md_sysfs_entry {
307 struct attribute attr;
308 ssize_t (*show)(mddev_t *, char *);
309 ssize_t (*store)(mddev_t *, const char *, size_t);
310};
311
312
313static inline char * mdname (mddev_t * mddev)
314{
315 return mddev->gendisk ? mddev->gendisk->disk_name : "mdX";
316}
317
318/*
319 * iterates through some rdev ringlist. It's safe to remove the
320 * current 'rdev'. Dont touch 'tmp' though.
321 */
322#define rdev_for_each_list(rdev, tmp, list) \
323 \
324 for ((tmp) = (list).next; \
325 (rdev) = (list_entry((tmp), mdk_rdev_t, same_set)), \
326 (tmp) = (tmp)->next, (tmp)->prev != &(list) \
327 ; )
328/*
329 * iterates through the 'same array disks' ringlist
330 */
331#define rdev_for_each(rdev, tmp, mddev) \
332 rdev_for_each_list(rdev, tmp, (mddev)->disks)
333
334typedef struct mdk_thread_s {
335 void (*run) (mddev_t *mddev);
336 mddev_t *mddev;
337 wait_queue_head_t wqueue;
338 unsigned long flags;
339 struct task_struct *tsk;
340 unsigned long timeout;
341} mdk_thread_t;
342
343#define THREAD_WAKEUP 0
344
345#define __wait_event_lock_irq(wq, condition, lock, cmd) \
346do { \
347 wait_queue_t __wait; \
348 init_waitqueue_entry(&__wait, current); \
349 \
350 add_wait_queue(&wq, &__wait); \
351 for (;;) { \
352 set_current_state(TASK_UNINTERRUPTIBLE); \
353 if (condition) \
354 break; \
355 spin_unlock_irq(&lock); \
356 cmd; \
357 schedule(); \
358 spin_lock_irq(&lock); \
359 } \
360 current->state = TASK_RUNNING; \
361 remove_wait_queue(&wq, &__wait); \
362} while (0)
363
364#define wait_event_lock_irq(wq, condition, lock, cmd) \
365do { \
366 if (condition) \
367 break; \
368 __wait_event_lock_irq(wq, condition, lock, cmd); \
369} while (0)
370
371static inline void safe_put_page(struct page *p)
372{
373 if (p) put_page(p);
374}
375
376#endif /* CONFIG_BLOCK */
377#endif
378