md: move lots of #include lines out of .h files and into .c
[linux-2.6.git] / include / linux / raid / md_k.h
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 #ifdef CONFIG_BLOCK
19
20 #define MaxSector (~(sector_t)0)
21
22 typedef struct mddev_s mddev_t;
23 typedef struct mdk_rdev_s mdk_rdev_t;
24
25 /*
26  * options passed in raidrun:
27  */
28
29 /* Currently this must fit in an 'int' */
30 #define MAX_CHUNK_SIZE (1<<30)
31
32 /*
33  * MD's 'extended' device
34  */
35 struct mdk_rdev_s
36 {
37         struct list_head same_set;      /* RAID devices within the same set */
38
39         sector_t size;                  /* Device size (in blocks) */
40         mddev_t *mddev;                 /* RAID array if running */
41         int last_events;                /* IO event timestamp */
42
43         struct block_device *bdev;      /* block device handle */
44
45         struct page     *sb_page;
46         int             sb_loaded;
47         __u64           sb_events;
48         sector_t        data_offset;    /* start of data in array */
49         sector_t        sb_start;       /* offset of the super block (in 512byte sectors) */
50         int             sb_size;        /* bytes in the superblock */
51         int             preferred_minor;        /* autorun support */
52
53         struct kobject  kobj;
54
55         /* A device can be in one of three states based on two flags:
56          * Not working:   faulty==1 in_sync==0
57          * Fully working: faulty==0 in_sync==1
58          * Working, but not
59          * in sync with array
60          *                faulty==0 in_sync==0
61          *
62          * It can never have faulty==1, in_sync==1
63          * This reduces the burden of testing multiple flags in many cases
64          */
65
66         unsigned long   flags;
67 #define Faulty          1               /* device is known to have a fault */
68 #define In_sync         2               /* device is in_sync with rest of array */
69 #define WriteMostly     4               /* Avoid reading if at all possible */
70 #define BarriersNotsupp 5               /* BIO_RW_BARRIER is not supported */
71 #define AllReserved     6               /* If whole device is reserved for
72                                          * one array */
73 #define AutoDetected    7               /* added by auto-detect */
74 #define Blocked         8               /* An error occured on an externally
75                                          * managed array, don't allow writes
76                                          * until it is cleared */
77 #define StateChanged    9               /* Faulty or Blocked has changed during
78                                          * interrupt, so it needs to be
79                                          * notified by the thread */
80         wait_queue_head_t blocked_wait;
81
82         int desc_nr;                    /* descriptor index in the superblock */
83         int raid_disk;                  /* role of device in array */
84         int saved_raid_disk;            /* role that device used to have in the
85                                          * array and could again if we did a partial
86                                          * resync from the bitmap
87                                          */
88         sector_t        recovery_offset;/* If this device has been partially
89                                          * recovered, this is where we were
90                                          * up to.
91                                          */
92
93         atomic_t        nr_pending;     /* number of pending requests.
94                                          * only maintained for arrays that
95                                          * support hot removal
96                                          */
97         atomic_t        read_errors;    /* number of consecutive read errors that
98                                          * we have tried to ignore.
99                                          */
100         atomic_t        corrected_errors; /* number of corrected read errors,
101                                            * for reporting to userspace and storing
102                                            * in superblock.
103                                            */
104         struct work_struct del_work;    /* used for delayed sysfs removal */
105
106         struct sysfs_dirent *sysfs_state; /* handle for 'state'
107                                            * sysfs entry */
108 };
109
110 struct mddev_s
111 {
112         void                            *private;
113         struct mdk_personality          *pers;
114         dev_t                           unit;
115         int                             md_minor;
116         struct list_head                disks;
117         unsigned long                   flags;
118 #define MD_CHANGE_DEVS  0       /* Some device status has changed */
119 #define MD_CHANGE_CLEAN 1       /* transition to or from 'clean' */
120 #define MD_CHANGE_PENDING 2     /* superblock update in progress */
121
122         int                             ro;
123
124         struct gendisk                  *gendisk;
125
126         struct kobject                  kobj;
127         int                             hold_active;
128 #define UNTIL_IOCTL     1
129 #define UNTIL_STOP      2
130
131         /* Superblock information */
132         int                             major_version,
133                                         minor_version,
134                                         patch_version;
135         int                             persistent;
136         int                             external;       /* metadata is
137                                                          * managed externally */
138         char                            metadata_type[17]; /* externally set*/
139         int                             chunk_size;
140         time_t                          ctime, utime;
141         int                             level, layout;
142         char                            clevel[16];
143         int                             raid_disks;
144         int                             max_disks;
145         sector_t                        size; /* used size of component devices */
146         sector_t                        array_sectors; /* exported array size */
147         __u64                           events;
148
149         char                            uuid[16];
150
151         /* If the array is being reshaped, we need to record the
152          * new shape and an indication of where we are up to.
153          * This is written to the superblock.
154          * If reshape_position is MaxSector, then no reshape is happening (yet).
155          */
156         sector_t                        reshape_position;
157         int                             delta_disks, new_level, new_layout, new_chunk;
158
159         struct mdk_thread_s             *thread;        /* management thread */
160         struct mdk_thread_s             *sync_thread;   /* doing resync or reconstruct */
161         sector_t                        curr_resync;    /* last block scheduled */
162         unsigned long                   resync_mark;    /* a recent timestamp */
163         sector_t                        resync_mark_cnt;/* blocks written at resync_mark */
164         sector_t                        curr_mark_cnt; /* blocks scheduled now */
165
166         sector_t                        resync_max_sectors; /* may be set by personality */
167
168         sector_t                        resync_mismatches; /* count of sectors where
169                                                             * parity/replica mismatch found
170                                                             */
171
172         /* allow user-space to request suspension of IO to regions of the array */
173         sector_t                        suspend_lo;
174         sector_t                        suspend_hi;
175         /* if zero, use the system-wide default */
176         int                             sync_speed_min;
177         int                             sync_speed_max;
178
179         /* resync even though the same disks are shared among md-devices */
180         int                             parallel_resync;
181
182         int                             ok_start_degraded;
183         /* recovery/resync flags 
184          * NEEDED:   we might need to start a resync/recover
185          * RUNNING:  a thread is running, or about to be started
186          * SYNC:     actually doing a resync, not a recovery
187          * RECOVER:  doing recovery, or need to try it.
188          * INTR:     resync needs to be aborted for some reason
189          * DONE:     thread is done and is waiting to be reaped
190          * REQUEST:  user-space has requested a sync (used with SYNC)
191          * CHECK:    user-space request for for check-only, no repair
192          * RESHAPE:  A reshape is happening
193          *
194          * If neither SYNC or RESHAPE are set, then it is a recovery.
195          */
196 #define MD_RECOVERY_RUNNING     0
197 #define MD_RECOVERY_SYNC        1
198 #define MD_RECOVERY_RECOVER     2
199 #define MD_RECOVERY_INTR        3
200 #define MD_RECOVERY_DONE        4
201 #define MD_RECOVERY_NEEDED      5
202 #define MD_RECOVERY_REQUESTED   6
203 #define MD_RECOVERY_CHECK       7
204 #define MD_RECOVERY_RESHAPE     8
205 #define MD_RECOVERY_FROZEN      9
206
207         unsigned long                   recovery;
208         int                             recovery_disabled; /* if we detect that recovery
209                                                             * will always fail, set this
210                                                             * so we don't loop trying */
211
212         int                             in_sync;        /* know to not need resync */
213         struct mutex                    reconfig_mutex;
214         atomic_t                        active;         /* general refcount */
215         atomic_t                        openers;        /* number of active opens */
216
217         int                             changed;        /* true if we might need to reread partition info */
218         int                             degraded;       /* whether md should consider
219                                                          * adding a spare
220                                                          */
221         int                             barriers_work;  /* initialised to true, cleared as soon
222                                                          * as a barrier request to slave
223                                                          * fails.  Only supported
224                                                          */
225         struct bio                      *biolist;       /* bios that need to be retried
226                                                          * because BIO_RW_BARRIER is not supported
227                                                          */
228
229         atomic_t                        recovery_active; /* blocks scheduled, but not written */
230         wait_queue_head_t               recovery_wait;
231         sector_t                        recovery_cp;
232         sector_t                        resync_min;     /* user requested sync
233                                                          * starts here */
234         sector_t                        resync_max;     /* resync should pause
235                                                          * when it gets here */
236
237         struct sysfs_dirent             *sysfs_state;   /* handle for 'array_state'
238                                                          * file in sysfs.
239                                                          */
240         struct sysfs_dirent             *sysfs_action;  /* handle for 'sync_action' */
241
242         struct work_struct del_work;    /* used for delayed sysfs removal */
243
244         spinlock_t                      write_lock;
245         wait_queue_head_t               sb_wait;        /* for waiting on superblock updates */
246         atomic_t                        pending_writes; /* number of active superblock writes */
247
248         unsigned int                    safemode;       /* if set, update "clean" superblock
249                                                          * when no writes pending.
250                                                          */ 
251         unsigned int                    safemode_delay;
252         struct timer_list               safemode_timer;
253         atomic_t                        writes_pending; 
254         struct request_queue            *queue; /* for plugging ... */
255
256         atomic_t                        write_behind; /* outstanding async IO */
257         unsigned int                    max_write_behind; /* 0 = sync */
258
259         struct bitmap                   *bitmap; /* the bitmap for the device */
260         struct file                     *bitmap_file; /* the bitmap file */
261         long                            bitmap_offset; /* offset from superblock of
262                                                         * start of bitmap. May be
263                                                         * negative, but not '0'
264                                                         */
265         long                            default_bitmap_offset; /* this is the offset to use when
266                                                                 * hot-adding a bitmap.  It should
267                                                                 * eventually be settable by sysfs.
268                                                                 */
269
270         struct list_head                all_mddevs;
271 };
272
273
274 static inline void rdev_dec_pending(mdk_rdev_t *rdev, mddev_t *mddev)
275 {
276         int faulty = test_bit(Faulty, &rdev->flags);
277         if (atomic_dec_and_test(&rdev->nr_pending) && faulty)
278                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
279 }
280
281 static inline void md_sync_acct(struct block_device *bdev, unsigned long nr_sectors)
282 {
283         atomic_add(nr_sectors, &bdev->bd_contains->bd_disk->sync_io);
284 }
285
286 struct mdk_personality
287 {
288         char *name;
289         int level;
290         struct list_head list;
291         struct module *owner;
292         int (*make_request)(struct request_queue *q, struct bio *bio);
293         int (*run)(mddev_t *mddev);
294         int (*stop)(mddev_t *mddev);
295         void (*status)(struct seq_file *seq, mddev_t *mddev);
296         /* error_handler must set ->faulty and clear ->in_sync
297          * if appropriate, and should abort recovery if needed 
298          */
299         void (*error_handler)(mddev_t *mddev, mdk_rdev_t *rdev);
300         int (*hot_add_disk) (mddev_t *mddev, mdk_rdev_t *rdev);
301         int (*hot_remove_disk) (mddev_t *mddev, int number);
302         int (*spare_active) (mddev_t *mddev);
303         sector_t (*sync_request)(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster);
304         int (*resize) (mddev_t *mddev, sector_t sectors);
305         int (*check_reshape) (mddev_t *mddev);
306         int (*start_reshape) (mddev_t *mddev);
307         int (*reconfig) (mddev_t *mddev, int layout, int chunk_size);
308         /* quiesce moves between quiescence states
309          * 0 - fully active
310          * 1 - no new requests allowed
311          * others - reserved
312          */
313         void (*quiesce) (mddev_t *mddev, int state);
314 };
315
316
317 struct md_sysfs_entry {
318         struct attribute attr;
319         ssize_t (*show)(mddev_t *, char *);
320         ssize_t (*store)(mddev_t *, const char *, size_t);
321 };
322
323
324 static inline char * mdname (mddev_t * mddev)
325 {
326         return mddev->gendisk ? mddev->gendisk->disk_name : "mdX";
327 }
328
329 /*
330  * iterates through some rdev ringlist. It's safe to remove the
331  * current 'rdev'. Dont touch 'tmp' though.
332  */
333 #define rdev_for_each_list(rdev, tmp, head)                             \
334         list_for_each_entry_safe(rdev, tmp, head, same_set)
335
336 /*
337  * iterates through the 'same array disks' ringlist
338  */
339 #define rdev_for_each(rdev, tmp, mddev)                         \
340         list_for_each_entry_safe(rdev, tmp, &((mddev)->disks), same_set)
341
342 #define rdev_for_each_rcu(rdev, mddev)                          \
343         list_for_each_entry_rcu(rdev, &((mddev)->disks), same_set)
344
345 typedef struct mdk_thread_s {
346         void                    (*run) (mddev_t *mddev);
347         mddev_t                 *mddev;
348         wait_queue_head_t       wqueue;
349         unsigned long           flags;
350         struct task_struct      *tsk;
351         unsigned long           timeout;
352 } mdk_thread_t;
353
354 #define THREAD_WAKEUP  0
355
356 #define __wait_event_lock_irq(wq, condition, lock, cmd)                 \
357 do {                                                                    \
358         wait_queue_t __wait;                                            \
359         init_waitqueue_entry(&__wait, current);                         \
360                                                                         \
361         add_wait_queue(&wq, &__wait);                                   \
362         for (;;) {                                                      \
363                 set_current_state(TASK_UNINTERRUPTIBLE);                \
364                 if (condition)                                          \
365                         break;                                          \
366                 spin_unlock_irq(&lock);                                 \
367                 cmd;                                                    \
368                 schedule();                                             \
369                 spin_lock_irq(&lock);                                   \
370         }                                                               \
371         current->state = TASK_RUNNING;                                  \
372         remove_wait_queue(&wq, &__wait);                                \
373 } while (0)
374
375 #define wait_event_lock_irq(wq, condition, lock, cmd)                   \
376 do {                                                                    \
377         if (condition)                                                  \
378                 break;                                                  \
379         __wait_event_lock_irq(wq, condition, lock, cmd);                \
380 } while (0)
381
382 static inline void safe_put_page(struct page *p)
383 {
384         if (p) put_page(p);
385 }
386
387 #endif /* CONFIG_BLOCK */
388 #endif
389
390
391 extern int register_md_personality(struct mdk_personality *p);
392 extern int unregister_md_personality(struct mdk_personality *p);
393 extern mdk_thread_t * md_register_thread(void (*run) (mddev_t *mddev),
394                                 mddev_t *mddev, const char *name);
395 extern void md_unregister_thread(mdk_thread_t *thread);
396 extern void md_wakeup_thread(mdk_thread_t *thread);
397 extern void md_check_recovery(mddev_t *mddev);
398 extern void md_write_start(mddev_t *mddev, struct bio *bi);
399 extern void md_write_end(mddev_t *mddev);
400 extern void md_done_sync(mddev_t *mddev, int blocks, int ok);
401 extern void md_error(mddev_t *mddev, mdk_rdev_t *rdev);
402
403 extern void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
404                            sector_t sector, int size, struct page *page);
405 extern void md_super_wait(mddev_t *mddev);
406 extern int sync_page_io(struct block_device *bdev, sector_t sector, int size,
407                         struct page *page, int rw);
408 extern void md_do_sync(mddev_t *mddev);
409 extern void md_new_event(mddev_t *mddev);
410 extern int md_allow_write(mddev_t *mddev);
411 extern void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev);