md: use sysfs_notify_dirent to notify changes to md/dev-xxx/state
[pandora-kernel.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 /* 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
35 typedef struct mddev_s mddev_t;
36 typedef 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  */
48 struct 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_start;       /* offset of the super block (in 512byte sectors) */
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 #define StateChanged    9               /* Faulty or Blocked has changed during
91                                          * interrupt, so it needs to be
92                                          * notified by the thread */
93         wait_queue_head_t blocked_wait;
94
95         int desc_nr;                    /* descriptor index in the superblock */
96         int raid_disk;                  /* role of device in array */
97         int saved_raid_disk;            /* role that device used to have in the
98                                          * array and could again if we did a partial
99                                          * resync from the bitmap
100                                          */
101         sector_t        recovery_offset;/* If this device has been partially
102                                          * recovered, this is where we were
103                                          * up to.
104                                          */
105
106         atomic_t        nr_pending;     /* number of pending requests.
107                                          * only maintained for arrays that
108                                          * support hot removal
109                                          */
110         atomic_t        read_errors;    /* number of consecutive read errors that
111                                          * we have tried to ignore.
112                                          */
113         atomic_t        corrected_errors; /* number of corrected read errors,
114                                            * for reporting to userspace and storing
115                                            * in superblock.
116                                            */
117         struct work_struct del_work;    /* used for delayed sysfs removal */
118
119         struct sysfs_dirent *sysfs_state; /* handle for 'state'
120                                            * sysfs entry */
121 };
122
123 struct mddev_s
124 {
125         void                            *private;
126         struct mdk_personality          *pers;
127         dev_t                           unit;
128         int                             md_minor;
129         struct list_head                disks;
130         unsigned long                   flags;
131 #define MD_CHANGE_DEVS  0       /* Some device status has changed */
132 #define MD_CHANGE_CLEAN 1       /* transition to or from 'clean' */
133 #define MD_CHANGE_PENDING 2     /* superblock update in progress */
134
135         int                             ro;
136
137         struct gendisk                  *gendisk;
138
139         struct kobject                  kobj;
140
141         /* Superblock information */
142         int                             major_version,
143                                         minor_version,
144                                         patch_version;
145         int                             persistent;
146         int                             external;       /* metadata is
147                                                          * managed externally */
148         char                            metadata_type[17]; /* externally set*/
149         int                             chunk_size;
150         time_t                          ctime, utime;
151         int                             level, layout;
152         char                            clevel[16];
153         int                             raid_disks;
154         int                             max_disks;
155         sector_t                        size; /* used size of component devices */
156         sector_t                        array_sectors; /* exported array size */
157         __u64                           events;
158
159         char                            uuid[16];
160
161         /* If the array is being reshaped, we need to record the
162          * new shape and an indication of where we are up to.
163          * This is written to the superblock.
164          * If reshape_position is MaxSector, then no reshape is happening (yet).
165          */
166         sector_t                        reshape_position;
167         int                             delta_disks, new_level, new_layout, new_chunk;
168
169         struct mdk_thread_s             *thread;        /* management thread */
170         struct mdk_thread_s             *sync_thread;   /* doing resync or reconstruct */
171         sector_t                        curr_resync;    /* last block scheduled */
172         unsigned long                   resync_mark;    /* a recent timestamp */
173         sector_t                        resync_mark_cnt;/* blocks written at resync_mark */
174         sector_t                        curr_mark_cnt; /* blocks scheduled now */
175
176         sector_t                        resync_max_sectors; /* may be set by personality */
177
178         sector_t                        resync_mismatches; /* count of sectors where
179                                                             * parity/replica mismatch found
180                                                             */
181
182         /* allow user-space to request suspension of IO to regions of the array */
183         sector_t                        suspend_lo;
184         sector_t                        suspend_hi;
185         /* if zero, use the system-wide default */
186         int                             sync_speed_min;
187         int                             sync_speed_max;
188
189         /* resync even though the same disks are shared among md-devices */
190         int                             parallel_resync;
191
192         int                             ok_start_degraded;
193         /* recovery/resync flags 
194          * NEEDED:   we might need to start a resync/recover
195          * RUNNING:  a thread is running, or about to be started
196          * SYNC:     actually doing a resync, not a recovery
197          * RECOVER:  doing recovery, or need to try it.
198          * INTR:     resync needs to be aborted for some reason
199          * DONE:     thread is done and is waiting to be reaped
200          * REQUEST:  user-space has requested a sync (used with SYNC)
201          * CHECK:    user-space request for for check-only, no repair
202          * RESHAPE:  A reshape is happening
203          *
204          * If neither SYNC or RESHAPE are set, then it is a recovery.
205          */
206 #define MD_RECOVERY_RUNNING     0
207 #define MD_RECOVERY_SYNC        1
208 #define MD_RECOVERY_RECOVER     2
209 #define MD_RECOVERY_INTR        3
210 #define MD_RECOVERY_DONE        4
211 #define MD_RECOVERY_NEEDED      5
212 #define MD_RECOVERY_REQUESTED   6
213 #define MD_RECOVERY_CHECK       7
214 #define MD_RECOVERY_RESHAPE     8
215 #define MD_RECOVERY_FROZEN      9
216
217         unsigned long                   recovery;
218
219         int                             in_sync;        /* know to not need resync */
220         struct mutex                    reconfig_mutex;
221         atomic_t                        active;         /* general refcount */
222         atomic_t                        openers;        /* number of active opens */
223
224         int                             changed;        /* true if we might need to reread partition info */
225         int                             degraded;       /* whether md should consider
226                                                          * adding a spare
227                                                          */
228         int                             barriers_work;  /* initialised to true, cleared as soon
229                                                          * as a barrier request to slave
230                                                          * fails.  Only supported
231                                                          */
232         struct bio                      *biolist;       /* bios that need to be retried
233                                                          * because BIO_RW_BARRIER is not supported
234                                                          */
235
236         atomic_t                        recovery_active; /* blocks scheduled, but not written */
237         wait_queue_head_t               recovery_wait;
238         sector_t                        recovery_cp;
239         sector_t                        resync_min;     /* user requested sync
240                                                          * starts here */
241         sector_t                        resync_max;     /* resync should pause
242                                                          * when it gets here */
243
244         struct sysfs_dirent             *sysfs_state;   /* handle for 'array_state'
245                                                          * file in sysfs.
246                                                          */
247
248         spinlock_t                      write_lock;
249         wait_queue_head_t               sb_wait;        /* for waiting on superblock updates */
250         atomic_t                        pending_writes; /* number of active superblock writes */
251
252         unsigned int                    safemode;       /* if set, update "clean" superblock
253                                                          * when no writes pending.
254                                                          */ 
255         unsigned int                    safemode_delay;
256         struct timer_list               safemode_timer;
257         atomic_t                        writes_pending; 
258         struct request_queue            *queue; /* for plugging ... */
259
260         atomic_t                        write_behind; /* outstanding async IO */
261         unsigned int                    max_write_behind; /* 0 = sync */
262
263         struct bitmap                   *bitmap; /* the bitmap for the device */
264         struct file                     *bitmap_file; /* the bitmap file */
265         long                            bitmap_offset; /* offset from superblock of
266                                                         * start of bitmap. May be
267                                                         * negative, but not '0'
268                                                         */
269         long                            default_bitmap_offset; /* this is the offset to use when
270                                                                 * hot-adding a bitmap.  It should
271                                                                 * eventually be settable by sysfs.
272                                                                 */
273
274         struct list_head                all_mddevs;
275 };
276
277
278 static inline void rdev_dec_pending(mdk_rdev_t *rdev, mddev_t *mddev)
279 {
280         int faulty = test_bit(Faulty, &rdev->flags);
281         if (atomic_dec_and_test(&rdev->nr_pending) && faulty)
282                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
283 }
284
285 static inline void md_sync_acct(struct block_device *bdev, unsigned long nr_sectors)
286 {
287         atomic_add(nr_sectors, &bdev->bd_contains->bd_disk->sync_io);
288 }
289
290 struct mdk_personality
291 {
292         char *name;
293         int level;
294         struct list_head list;
295         struct module *owner;
296         int (*make_request)(struct request_queue *q, struct bio *bio);
297         int (*run)(mddev_t *mddev);
298         int (*stop)(mddev_t *mddev);
299         void (*status)(struct seq_file *seq, mddev_t *mddev);
300         /* error_handler must set ->faulty and clear ->in_sync
301          * if appropriate, and should abort recovery if needed 
302          */
303         void (*error_handler)(mddev_t *mddev, mdk_rdev_t *rdev);
304         int (*hot_add_disk) (mddev_t *mddev, mdk_rdev_t *rdev);
305         int (*hot_remove_disk) (mddev_t *mddev, int number);
306         int (*spare_active) (mddev_t *mddev);
307         sector_t (*sync_request)(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster);
308         int (*resize) (mddev_t *mddev, sector_t sectors);
309         int (*check_reshape) (mddev_t *mddev);
310         int (*start_reshape) (mddev_t *mddev);
311         int (*reconfig) (mddev_t *mddev, int layout, int chunk_size);
312         /* quiesce moves between quiescence states
313          * 0 - fully active
314          * 1 - no new requests allowed
315          * others - reserved
316          */
317         void (*quiesce) (mddev_t *mddev, int state);
318 };
319
320
321 struct md_sysfs_entry {
322         struct attribute attr;
323         ssize_t (*show)(mddev_t *, char *);
324         ssize_t (*store)(mddev_t *, const char *, size_t);
325 };
326
327
328 static inline char * mdname (mddev_t * mddev)
329 {
330         return mddev->gendisk ? mddev->gendisk->disk_name : "mdX";
331 }
332
333 /*
334  * iterates through some rdev ringlist. It's safe to remove the
335  * current 'rdev'. Dont touch 'tmp' though.
336  */
337 #define rdev_for_each_list(rdev, tmp, list)                             \
338                                                                         \
339         for ((tmp) = (list).next;                                       \
340                 (rdev) = (list_entry((tmp), mdk_rdev_t, same_set)),     \
341                         (tmp) = (tmp)->next, (tmp)->prev != &(list)     \
342                 ; )
343 /*
344  * iterates through the 'same array disks' ringlist
345  */
346 #define rdev_for_each(rdev, tmp, mddev)                         \
347         rdev_for_each_list(rdev, tmp, (mddev)->disks)
348
349 #define rdev_for_each_rcu(rdev, mddev)                          \
350         list_for_each_entry_rcu(rdev, &((mddev)->disks), same_set)
351
352 typedef struct mdk_thread_s {
353         void                    (*run) (mddev_t *mddev);
354         mddev_t                 *mddev;
355         wait_queue_head_t       wqueue;
356         unsigned long           flags;
357         struct task_struct      *tsk;
358         unsigned long           timeout;
359 } mdk_thread_t;
360
361 #define THREAD_WAKEUP  0
362
363 #define __wait_event_lock_irq(wq, condition, lock, cmd)                 \
364 do {                                                                    \
365         wait_queue_t __wait;                                            \
366         init_waitqueue_entry(&__wait, current);                         \
367                                                                         \
368         add_wait_queue(&wq, &__wait);                                   \
369         for (;;) {                                                      \
370                 set_current_state(TASK_UNINTERRUPTIBLE);                \
371                 if (condition)                                          \
372                         break;                                          \
373                 spin_unlock_irq(&lock);                                 \
374                 cmd;                                                    \
375                 schedule();                                             \
376                 spin_lock_irq(&lock);                                   \
377         }                                                               \
378         current->state = TASK_RUNNING;                                  \
379         remove_wait_queue(&wq, &__wait);                                \
380 } while (0)
381
382 #define wait_event_lock_irq(wq, condition, lock, cmd)                   \
383 do {                                                                    \
384         if (condition)                                                  \
385                 break;                                                  \
386         __wait_event_lock_irq(wq, condition, lock, cmd);                \
387 } while (0)
388
389 static inline void safe_put_page(struct page *p)
390 {
391         if (p) put_page(p);
392 }
393
394 #endif /* CONFIG_BLOCK */
395 #endif
396