7bb6d1abf71e85944bc20e82243cc88df8353e71
[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 /* 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_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
88         int desc_nr;                    /* descriptor index in the superblock */
89         int raid_disk;                  /* role of device in array */
90         int saved_raid_disk;            /* role that device used to have in the
91                                          * array and could again if we did a partial
92                                          * resync from the bitmap
93                                          */
94         sector_t        recovery_offset;/* If this device has been partially
95                                          * recovered, this is where we were
96                                          * up to.
97                                          */
98
99         atomic_t        nr_pending;     /* number of pending requests.
100                                          * only maintained for arrays that
101                                          * support hot removal
102                                          */
103         atomic_t        read_errors;    /* number of consecutive read errors that
104                                          * we have tried to ignore.
105                                          */
106         atomic_t        corrected_errors; /* number of corrected read errors,
107                                            * for reporting to userspace and storing
108                                            * in superblock.
109                                            */
110         struct work_struct del_work;    /* used for delayed sysfs removal */
111 };
112
113 struct mddev_s
114 {
115         void                            *private;
116         struct mdk_personality          *pers;
117         dev_t                           unit;
118         int                             md_minor;
119         struct list_head                disks;
120         unsigned long                   flags;
121 #define MD_CHANGE_DEVS  0       /* Some device status has changed */
122 #define MD_CHANGE_CLEAN 1       /* transition to or from 'clean' */
123 #define MD_CHANGE_PENDING 2     /* superblock update in progress */
124
125         int                             ro;
126
127         struct gendisk                  *gendisk;
128
129         struct kobject                  kobj;
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_size; /* 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         int                             ok_start_degraded;
180         /* recovery/resync flags 
181          * NEEDED:   we might need to start a resync/recover
182          * RUNNING:  a thread is running, or about to be started
183          * SYNC:     actually doing a resync, not a recovery
184          * ERR:      and IO error was detected - abort the resync/recovery
185          * INTR:     someone requested a (clean) early abort.
186          * DONE:     thread is done and is waiting to be reaped
187          * REQUEST:  user-space has requested a sync (used with SYNC)
188          * CHECK:    user-space request for for check-only, no repair
189          * RESHAPE:  A reshape is happening
190          *
191          * If neither SYNC or RESHAPE are set, then it is a recovery.
192          */
193 #define MD_RECOVERY_RUNNING     0
194 #define MD_RECOVERY_SYNC        1
195 #define MD_RECOVERY_ERR         2
196 #define MD_RECOVERY_INTR        3
197 #define MD_RECOVERY_DONE        4
198 #define MD_RECOVERY_NEEDED      5
199 #define MD_RECOVERY_REQUESTED   6
200 #define MD_RECOVERY_CHECK       7
201 #define MD_RECOVERY_RESHAPE     8
202 #define MD_RECOVERY_FROZEN      9
203
204         unsigned long                   recovery;
205
206         int                             in_sync;        /* know to not need resync */
207         struct mutex                    reconfig_mutex;
208         atomic_t                        active;
209
210         int                             changed;        /* true if we might need to reread partition info */
211         int                             degraded;       /* whether md should consider
212                                                          * adding a spare
213                                                          */
214         int                             barriers_work;  /* initialised to true, cleared as soon
215                                                          * as a barrier request to slave
216                                                          * fails.  Only supported
217                                                          */
218         struct bio                      *biolist;       /* bios that need to be retried
219                                                          * because BIO_RW_BARRIER is not supported
220                                                          */
221
222         atomic_t                        recovery_active; /* blocks scheduled, but not written */
223         wait_queue_head_t               recovery_wait;
224         sector_t                        recovery_cp;
225         sector_t                        resync_max;     /* resync should pause
226                                                          * when it gets here */
227
228         spinlock_t                      write_lock;
229         wait_queue_head_t               sb_wait;        /* for waiting on superblock updates */
230         atomic_t                        pending_writes; /* number of active superblock writes */
231
232         unsigned int                    safemode;       /* if set, update "clean" superblock
233                                                          * when no writes pending.
234                                                          */ 
235         unsigned int                    safemode_delay;
236         struct timer_list               safemode_timer;
237         atomic_t                        writes_pending; 
238         struct request_queue            *queue; /* for plugging ... */
239
240         atomic_t                        write_behind; /* outstanding async IO */
241         unsigned int                    max_write_behind; /* 0 = sync */
242
243         struct bitmap                   *bitmap; /* the bitmap for the device */
244         struct file                     *bitmap_file; /* the bitmap file */
245         long                            bitmap_offset; /* offset from superblock of
246                                                         * start of bitmap. May be
247                                                         * negative, but not '0'
248                                                         */
249         long                            default_bitmap_offset; /* this is the offset to use when
250                                                                 * hot-adding a bitmap.  It should
251                                                                 * eventually be settable by sysfs.
252                                                                 */
253
254         struct list_head                all_mddevs;
255 };
256
257
258 static inline void rdev_dec_pending(mdk_rdev_t *rdev, mddev_t *mddev)
259 {
260         int faulty = test_bit(Faulty, &rdev->flags);
261         if (atomic_dec_and_test(&rdev->nr_pending) && faulty)
262                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
263 }
264
265 static inline void md_sync_acct(struct block_device *bdev, unsigned long nr_sectors)
266 {
267         atomic_add(nr_sectors, &bdev->bd_contains->bd_disk->sync_io);
268 }
269
270 struct mdk_personality
271 {
272         char *name;
273         int level;
274         struct list_head list;
275         struct module *owner;
276         int (*make_request)(struct request_queue *q, struct bio *bio);
277         int (*run)(mddev_t *mddev);
278         int (*stop)(mddev_t *mddev);
279         void (*status)(struct seq_file *seq, mddev_t *mddev);
280         /* error_handler must set ->faulty and clear ->in_sync
281          * if appropriate, and should abort recovery if needed 
282          */
283         void (*error_handler)(mddev_t *mddev, mdk_rdev_t *rdev);
284         int (*hot_add_disk) (mddev_t *mddev, mdk_rdev_t *rdev);
285         int (*hot_remove_disk) (mddev_t *mddev, int number);
286         int (*spare_active) (mddev_t *mddev);
287         sector_t (*sync_request)(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster);
288         int (*resize) (mddev_t *mddev, sector_t sectors);
289         int (*check_reshape) (mddev_t *mddev);
290         int (*start_reshape) (mddev_t *mddev);
291         int (*reconfig) (mddev_t *mddev, int layout, int chunk_size);
292         /* quiesce moves between quiescence states
293          * 0 - fully active
294          * 1 - no new requests allowed
295          * others - reserved
296          */
297         void (*quiesce) (mddev_t *mddev, int state);
298 };
299
300
301 struct md_sysfs_entry {
302         struct attribute attr;
303         ssize_t (*show)(mddev_t *, char *);
304         ssize_t (*store)(mddev_t *, const char *, size_t);
305 };
306
307
308 static inline char * mdname (mddev_t * mddev)
309 {
310         return mddev->gendisk ? mddev->gendisk->disk_name : "mdX";
311 }
312
313 /*
314  * iterates through some rdev ringlist. It's safe to remove the
315  * current 'rdev'. Dont touch 'tmp' though.
316  */
317 #define rdev_for_each_list(rdev, tmp, list)                             \
318                                                                         \
319         for ((tmp) = (list).next;                                       \
320                 (rdev) = (list_entry((tmp), mdk_rdev_t, same_set)),     \
321                         (tmp) = (tmp)->next, (tmp)->prev != &(list)     \
322                 ; )
323 /*
324  * iterates through the 'same array disks' ringlist
325  */
326 #define rdev_for_each(rdev, tmp, mddev)                         \
327         rdev_for_each_list(rdev, tmp, (mddev)->disks)
328
329 typedef struct mdk_thread_s {
330         void                    (*run) (mddev_t *mddev);
331         mddev_t                 *mddev;
332         wait_queue_head_t       wqueue;
333         unsigned long           flags;
334         struct task_struct      *tsk;
335         unsigned long           timeout;
336 } mdk_thread_t;
337
338 #define THREAD_WAKEUP  0
339
340 #define __wait_event_lock_irq(wq, condition, lock, cmd)                 \
341 do {                                                                    \
342         wait_queue_t __wait;                                            \
343         init_waitqueue_entry(&__wait, current);                         \
344                                                                         \
345         add_wait_queue(&wq, &__wait);                                   \
346         for (;;) {                                                      \
347                 set_current_state(TASK_UNINTERRUPTIBLE);                \
348                 if (condition)                                          \
349                         break;                                          \
350                 spin_unlock_irq(&lock);                                 \
351                 cmd;                                                    \
352                 schedule();                                             \
353                 spin_lock_irq(&lock);                                   \
354         }                                                               \
355         current->state = TASK_RUNNING;                                  \
356         remove_wait_queue(&wq, &__wait);                                \
357 } while (0)
358
359 #define wait_event_lock_irq(wq, condition, lock, cmd)                   \
360 do {                                                                    \
361         if (condition)                                                  \
362                 break;                                                  \
363         __wait_event_lock_irq(wq, condition, lock, cmd);                \
364 } while (0)
365
366 static inline void safe_put_page(struct page *p)
367 {
368         if (p) put_page(p);
369 }
370
371 #endif /* CONFIG_BLOCK */
372 #endif
373