Merge branch 'upstream-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mfashe...
[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
120 struct mddev_s
121 {
122         void                            *private;
123         struct mdk_personality          *pers;
124         dev_t                           unit;
125         int                             md_minor;
126         struct list_head                disks;
127         unsigned long                   flags;
128 #define MD_CHANGE_DEVS  0       /* Some device status has changed */
129 #define MD_CHANGE_CLEAN 1       /* transition to or from 'clean' */
130 #define MD_CHANGE_PENDING 2     /* superblock update in progress */
131
132         int                             ro;
133
134         struct gendisk                  *gendisk;
135
136         struct kobject                  kobj;
137
138         /* Superblock information */
139         int                             major_version,
140                                         minor_version,
141                                         patch_version;
142         int                             persistent;
143         int                             external;       /* metadata is
144                                                          * managed externally */
145         char                            metadata_type[17]; /* externally set*/
146         int                             chunk_size;
147         time_t                          ctime, utime;
148         int                             level, layout;
149         char                            clevel[16];
150         int                             raid_disks;
151         int                             max_disks;
152         sector_t                        size; /* used size of component devices */
153         sector_t                        array_sectors; /* exported array size */
154         __u64                           events;
155
156         char                            uuid[16];
157
158         /* If the array is being reshaped, we need to record the
159          * new shape and an indication of where we are up to.
160          * This is written to the superblock.
161          * If reshape_position is MaxSector, then no reshape is happening (yet).
162          */
163         sector_t                        reshape_position;
164         int                             delta_disks, new_level, new_layout, new_chunk;
165
166         struct mdk_thread_s             *thread;        /* management thread */
167         struct mdk_thread_s             *sync_thread;   /* doing resync or reconstruct */
168         sector_t                        curr_resync;    /* last block scheduled */
169         unsigned long                   resync_mark;    /* a recent timestamp */
170         sector_t                        resync_mark_cnt;/* blocks written at resync_mark */
171         sector_t                        curr_mark_cnt; /* blocks scheduled now */
172
173         sector_t                        resync_max_sectors; /* may be set by personality */
174
175         sector_t                        resync_mismatches; /* count of sectors where
176                                                             * parity/replica mismatch found
177                                                             */
178
179         /* allow user-space to request suspension of IO to regions of the array */
180         sector_t                        suspend_lo;
181         sector_t                        suspend_hi;
182         /* if zero, use the system-wide default */
183         int                             sync_speed_min;
184         int                             sync_speed_max;
185
186         /* resync even though the same disks are shared among md-devices */
187         int                             parallel_resync;
188
189         int                             ok_start_degraded;
190         /* recovery/resync flags 
191          * NEEDED:   we might need to start a resync/recover
192          * RUNNING:  a thread is running, or about to be started
193          * SYNC:     actually doing a resync, not a recovery
194          * RECOVER:  doing recovery, or need to try it.
195          * INTR:     resync needs to be aborted for some reason
196          * DONE:     thread is done and is waiting to be reaped
197          * REQUEST:  user-space has requested a sync (used with SYNC)
198          * CHECK:    user-space request for for check-only, no repair
199          * RESHAPE:  A reshape is happening
200          *
201          * If neither SYNC or RESHAPE are set, then it is a recovery.
202          */
203 #define MD_RECOVERY_RUNNING     0
204 #define MD_RECOVERY_SYNC        1
205 #define MD_RECOVERY_RECOVER     2
206 #define MD_RECOVERY_INTR        3
207 #define MD_RECOVERY_DONE        4
208 #define MD_RECOVERY_NEEDED      5
209 #define MD_RECOVERY_REQUESTED   6
210 #define MD_RECOVERY_CHECK       7
211 #define MD_RECOVERY_RESHAPE     8
212 #define MD_RECOVERY_FROZEN      9
213
214         unsigned long                   recovery;
215
216         int                             in_sync;        /* know to not need resync */
217         struct mutex                    reconfig_mutex;
218         atomic_t                        active;         /* general refcount */
219         atomic_t                        openers;        /* number of active opens */
220
221         int                             changed;        /* true if we might need to reread partition info */
222         int                             degraded;       /* whether md should consider
223                                                          * adding a spare
224                                                          */
225         int                             barriers_work;  /* initialised to true, cleared as soon
226                                                          * as a barrier request to slave
227                                                          * fails.  Only supported
228                                                          */
229         struct bio                      *biolist;       /* bios that need to be retried
230                                                          * because BIO_RW_BARRIER is not supported
231                                                          */
232
233         atomic_t                        recovery_active; /* blocks scheduled, but not written */
234         wait_queue_head_t               recovery_wait;
235         sector_t                        recovery_cp;
236         sector_t                        resync_min;     /* user requested sync
237                                                          * starts here */
238         sector_t                        resync_max;     /* resync should pause
239                                                          * when it gets here */
240
241         spinlock_t                      write_lock;
242         wait_queue_head_t               sb_wait;        /* for waiting on superblock updates */
243         atomic_t                        pending_writes; /* number of active superblock writes */
244
245         unsigned int                    safemode;       /* if set, update "clean" superblock
246                                                          * when no writes pending.
247                                                          */ 
248         unsigned int                    safemode_delay;
249         struct timer_list               safemode_timer;
250         atomic_t                        writes_pending; 
251         struct request_queue            *queue; /* for plugging ... */
252
253         atomic_t                        write_behind; /* outstanding async IO */
254         unsigned int                    max_write_behind; /* 0 = sync */
255
256         struct bitmap                   *bitmap; /* the bitmap for the device */
257         struct file                     *bitmap_file; /* the bitmap file */
258         long                            bitmap_offset; /* offset from superblock of
259                                                         * start of bitmap. May be
260                                                         * negative, but not '0'
261                                                         */
262         long                            default_bitmap_offset; /* this is the offset to use when
263                                                                 * hot-adding a bitmap.  It should
264                                                                 * eventually be settable by sysfs.
265                                                                 */
266
267         struct list_head                all_mddevs;
268 };
269
270
271 static inline void rdev_dec_pending(mdk_rdev_t *rdev, mddev_t *mddev)
272 {
273         int faulty = test_bit(Faulty, &rdev->flags);
274         if (atomic_dec_and_test(&rdev->nr_pending) && faulty)
275                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
276 }
277
278 static inline void md_sync_acct(struct block_device *bdev, unsigned long nr_sectors)
279 {
280         atomic_add(nr_sectors, &bdev->bd_contains->bd_disk->sync_io);
281 }
282
283 struct mdk_personality
284 {
285         char *name;
286         int level;
287         struct list_head list;
288         struct module *owner;
289         int (*make_request)(struct request_queue *q, struct bio *bio);
290         int (*run)(mddev_t *mddev);
291         int (*stop)(mddev_t *mddev);
292         void (*status)(struct seq_file *seq, mddev_t *mddev);
293         /* error_handler must set ->faulty and clear ->in_sync
294          * if appropriate, and should abort recovery if needed 
295          */
296         void (*error_handler)(mddev_t *mddev, mdk_rdev_t *rdev);
297         int (*hot_add_disk) (mddev_t *mddev, mdk_rdev_t *rdev);
298         int (*hot_remove_disk) (mddev_t *mddev, int number);
299         int (*spare_active) (mddev_t *mddev);
300         sector_t (*sync_request)(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster);
301         int (*resize) (mddev_t *mddev, sector_t sectors);
302         int (*check_reshape) (mddev_t *mddev);
303         int (*start_reshape) (mddev_t *mddev);
304         int (*reconfig) (mddev_t *mddev, int layout, int chunk_size);
305         /* quiesce moves between quiescence states
306          * 0 - fully active
307          * 1 - no new requests allowed
308          * others - reserved
309          */
310         void (*quiesce) (mddev_t *mddev, int state);
311 };
312
313
314 struct md_sysfs_entry {
315         struct attribute attr;
316         ssize_t (*show)(mddev_t *, char *);
317         ssize_t (*store)(mddev_t *, const char *, size_t);
318 };
319
320
321 static inline char * mdname (mddev_t * mddev)
322 {
323         return mddev->gendisk ? mddev->gendisk->disk_name : "mdX";
324 }
325
326 /*
327  * iterates through some rdev ringlist. It's safe to remove the
328  * current 'rdev'. Dont touch 'tmp' though.
329  */
330 #define rdev_for_each_list(rdev, tmp, list)                             \
331                                                                         \
332         for ((tmp) = (list).next;                                       \
333                 (rdev) = (list_entry((tmp), mdk_rdev_t, same_set)),     \
334                         (tmp) = (tmp)->next, (tmp)->prev != &(list)     \
335                 ; )
336 /*
337  * iterates through the 'same array disks' ringlist
338  */
339 #define rdev_for_each(rdev, tmp, mddev)                         \
340         rdev_for_each_list(rdev, tmp, (mddev)->disks)
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