4c5e2d00ff5e8f95e783988c01ca1f3b955d933d
[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 MaxSector (~(sector_t)0)
24
25 typedef struct mddev_s mddev_t;
26 typedef struct mdk_rdev_s mdk_rdev_t;
27
28 /*
29  * options passed in raidrun:
30  */
31
32 /* Currently this must fit in an 'int' */
33 #define MAX_CHUNK_SIZE (1<<30)
34
35 /*
36  * MD's 'extended' device
37  */
38 struct mdk_rdev_s
39 {
40         struct list_head same_set;      /* RAID devices within the same set */
41
42         sector_t size;                  /* Device size (in blocks) */
43         mddev_t *mddev;                 /* RAID array if running */
44         int last_events;                /* IO event timestamp */
45
46         struct block_device *bdev;      /* block device handle */
47
48         struct page     *sb_page;
49         int             sb_loaded;
50         __u64           sb_events;
51         sector_t        data_offset;    /* start of data in array */
52         sector_t        sb_start;       /* offset of the super block (in 512byte sectors) */
53         int             sb_size;        /* bytes in the superblock */
54         int             preferred_minor;        /* autorun support */
55
56         struct kobject  kobj;
57
58         /* A device can be in one of three states based on two flags:
59          * Not working:   faulty==1 in_sync==0
60          * Fully working: faulty==0 in_sync==1
61          * Working, but not
62          * in sync with array
63          *                faulty==0 in_sync==0
64          *
65          * It can never have faulty==1, in_sync==1
66          * This reduces the burden of testing multiple flags in many cases
67          */
68
69         unsigned long   flags;
70 #define Faulty          1               /* device is known to have a fault */
71 #define In_sync         2               /* device is in_sync with rest of array */
72 #define WriteMostly     4               /* Avoid reading if at all possible */
73 #define BarriersNotsupp 5               /* BIO_RW_BARRIER is not supported */
74 #define AllReserved     6               /* If whole device is reserved for
75                                          * one array */
76 #define AutoDetected    7               /* added by auto-detect */
77 #define Blocked         8               /* An error occured on an externally
78                                          * managed array, don't allow writes
79                                          * until it is cleared */
80 #define StateChanged    9               /* Faulty or Blocked has changed during
81                                          * interrupt, so it needs to be
82                                          * notified by the thread */
83         wait_queue_head_t blocked_wait;
84
85         int desc_nr;                    /* descriptor index in the superblock */
86         int raid_disk;                  /* role of device in array */
87         int saved_raid_disk;            /* role that device used to have in the
88                                          * array and could again if we did a partial
89                                          * resync from the bitmap
90                                          */
91         sector_t        recovery_offset;/* If this device has been partially
92                                          * recovered, this is where we were
93                                          * up to.
94                                          */
95
96         atomic_t        nr_pending;     /* number of pending requests.
97                                          * only maintained for arrays that
98                                          * support hot removal
99                                          */
100         atomic_t        read_errors;    /* number of consecutive read errors that
101                                          * we have tried to ignore.
102                                          */
103         atomic_t        corrected_errors; /* number of corrected read errors,
104                                            * for reporting to userspace and storing
105                                            * in superblock.
106                                            */
107         struct work_struct del_work;    /* used for delayed sysfs removal */
108
109         struct sysfs_dirent *sysfs_state; /* handle for 'state'
110                                            * sysfs entry */
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         int                             hold_active;
131 #define UNTIL_IOCTL     1
132 #define UNTIL_STOP      2
133
134         /* Superblock information */
135         int                             major_version,
136                                         minor_version,
137                                         patch_version;
138         int                             persistent;
139         int                             external;       /* metadata is
140                                                          * managed externally */
141         char                            metadata_type[17]; /* externally set*/
142         int                             chunk_size;
143         time_t                          ctime, utime;
144         int                             level, layout;
145         char                            clevel[16];
146         int                             raid_disks;
147         int                             max_disks;
148         sector_t                        size; /* used size of component devices */
149         sector_t                        array_sectors; /* exported array size */
150         __u64                           events;
151
152         char                            uuid[16];
153
154         /* If the array is being reshaped, we need to record the
155          * new shape and an indication of where we are up to.
156          * This is written to the superblock.
157          * If reshape_position is MaxSector, then no reshape is happening (yet).
158          */
159         sector_t                        reshape_position;
160         int                             delta_disks, new_level, new_layout, new_chunk;
161
162         struct mdk_thread_s             *thread;        /* management thread */
163         struct mdk_thread_s             *sync_thread;   /* doing resync or reconstruct */
164         sector_t                        curr_resync;    /* last block scheduled */
165         unsigned long                   resync_mark;    /* a recent timestamp */
166         sector_t                        resync_mark_cnt;/* blocks written at resync_mark */
167         sector_t                        curr_mark_cnt; /* blocks scheduled now */
168
169         sector_t                        resync_max_sectors; /* may be set by personality */
170
171         sector_t                        resync_mismatches; /* count of sectors where
172                                                             * parity/replica mismatch found
173                                                             */
174
175         /* allow user-space to request suspension of IO to regions of the array */
176         sector_t                        suspend_lo;
177         sector_t                        suspend_hi;
178         /* if zero, use the system-wide default */
179         int                             sync_speed_min;
180         int                             sync_speed_max;
181
182         /* resync even though the same disks are shared among md-devices */
183         int                             parallel_resync;
184
185         int                             ok_start_degraded;
186         /* recovery/resync flags 
187          * NEEDED:   we might need to start a resync/recover
188          * RUNNING:  a thread is running, or about to be started
189          * SYNC:     actually doing a resync, not a recovery
190          * RECOVER:  doing recovery, or need to try it.
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_RECOVER     2
202 #define MD_RECOVERY_INTR        3
203 #define MD_RECOVERY_DONE        4
204 #define MD_RECOVERY_NEEDED      5
205 #define MD_RECOVERY_REQUESTED   6
206 #define MD_RECOVERY_CHECK       7
207 #define MD_RECOVERY_RESHAPE     8
208 #define MD_RECOVERY_FROZEN      9
209
210         unsigned long                   recovery;
211         int                             recovery_disabled; /* if we detect that recovery
212                                                             * will always fail, set this
213                                                             * so we don't loop trying */
214
215         int                             in_sync;        /* know to not need resync */
216         struct mutex                    reconfig_mutex;
217         atomic_t                        active;         /* general refcount */
218         atomic_t                        openers;        /* number of active opens */
219
220         int                             changed;        /* true if we might need to reread partition info */
221         int                             degraded;       /* whether md should consider
222                                                          * adding a spare
223                                                          */
224         int                             barriers_work;  /* initialised to true, cleared as soon
225                                                          * as a barrier request to slave
226                                                          * fails.  Only supported
227                                                          */
228         struct bio                      *biolist;       /* bios that need to be retried
229                                                          * because BIO_RW_BARRIER is not supported
230                                                          */
231
232         atomic_t                        recovery_active; /* blocks scheduled, but not written */
233         wait_queue_head_t               recovery_wait;
234         sector_t                        recovery_cp;
235         sector_t                        resync_min;     /* user requested sync
236                                                          * starts here */
237         sector_t                        resync_max;     /* resync should pause
238                                                          * when it gets here */
239
240         struct sysfs_dirent             *sysfs_state;   /* handle for 'array_state'
241                                                          * file in sysfs.
242                                                          */
243         struct sysfs_dirent             *sysfs_action;  /* handle for 'sync_action' */
244
245         struct work_struct del_work;    /* used for delayed sysfs removal */
246
247         spinlock_t                      write_lock;
248         wait_queue_head_t               sb_wait;        /* for waiting on superblock updates */
249         atomic_t                        pending_writes; /* number of active superblock writes */
250
251         unsigned int                    safemode;       /* if set, update "clean" superblock
252                                                          * when no writes pending.
253                                                          */ 
254         unsigned int                    safemode_delay;
255         struct timer_list               safemode_timer;
256         atomic_t                        writes_pending; 
257         struct request_queue            *queue; /* for plugging ... */
258
259         atomic_t                        write_behind; /* outstanding async IO */
260         unsigned int                    max_write_behind; /* 0 = sync */
261
262         struct bitmap                   *bitmap; /* the bitmap for the device */
263         struct file                     *bitmap_file; /* the bitmap file */
264         long                            bitmap_offset; /* offset from superblock of
265                                                         * start of bitmap. May be
266                                                         * negative, but not '0'
267                                                         */
268         long                            default_bitmap_offset; /* this is the offset to use when
269                                                                 * hot-adding a bitmap.  It should
270                                                                 * eventually be settable by sysfs.
271                                                                 */
272
273         struct list_head                all_mddevs;
274 };
275
276
277 static inline void rdev_dec_pending(mdk_rdev_t *rdev, mddev_t *mddev)
278 {
279         int faulty = test_bit(Faulty, &rdev->flags);
280         if (atomic_dec_and_test(&rdev->nr_pending) && faulty)
281                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
282 }
283
284 static inline void md_sync_acct(struct block_device *bdev, unsigned long nr_sectors)
285 {
286         atomic_add(nr_sectors, &bdev->bd_contains->bd_disk->sync_io);
287 }
288
289 struct mdk_personality
290 {
291         char *name;
292         int level;
293         struct list_head list;
294         struct module *owner;
295         int (*make_request)(struct request_queue *q, struct bio *bio);
296         int (*run)(mddev_t *mddev);
297         int (*stop)(mddev_t *mddev);
298         void (*status)(struct seq_file *seq, mddev_t *mddev);
299         /* error_handler must set ->faulty and clear ->in_sync
300          * if appropriate, and should abort recovery if needed 
301          */
302         void (*error_handler)(mddev_t *mddev, mdk_rdev_t *rdev);
303         int (*hot_add_disk) (mddev_t *mddev, mdk_rdev_t *rdev);
304         int (*hot_remove_disk) (mddev_t *mddev, int number);
305         int (*spare_active) (mddev_t *mddev);
306         sector_t (*sync_request)(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster);
307         int (*resize) (mddev_t *mddev, sector_t sectors);
308         int (*check_reshape) (mddev_t *mddev);
309         int (*start_reshape) (mddev_t *mddev);
310         int (*reconfig) (mddev_t *mddev, int layout, int chunk_size);
311         /* quiesce moves between quiescence states
312          * 0 - fully active
313          * 1 - no new requests allowed
314          * others - reserved
315          */
316         void (*quiesce) (mddev_t *mddev, int state);
317 };
318
319
320 struct md_sysfs_entry {
321         struct attribute attr;
322         ssize_t (*show)(mddev_t *, char *);
323         ssize_t (*store)(mddev_t *, const char *, size_t);
324 };
325
326
327 static inline char * mdname (mddev_t * mddev)
328 {
329         return mddev->gendisk ? mddev->gendisk->disk_name : "mdX";
330 }
331
332 /*
333  * iterates through some rdev ringlist. It's safe to remove the
334  * current 'rdev'. Dont touch 'tmp' though.
335  */
336 #define rdev_for_each_list(rdev, tmp, head)                             \
337         list_for_each_entry_safe(rdev, tmp, head, same_set)
338
339 /*
340  * iterates through the 'same array disks' ringlist
341  */
342 #define rdev_for_each(rdev, tmp, mddev)                         \
343         list_for_each_entry_safe(rdev, tmp, &((mddev)->disks), same_set)
344
345 #define rdev_for_each_rcu(rdev, mddev)                          \
346         list_for_each_entry_rcu(rdev, &((mddev)->disks), same_set)
347
348 typedef struct mdk_thread_s {
349         void                    (*run) (mddev_t *mddev);
350         mddev_t                 *mddev;
351         wait_queue_head_t       wqueue;
352         unsigned long           flags;
353         struct task_struct      *tsk;
354         unsigned long           timeout;
355 } mdk_thread_t;
356
357 #define THREAD_WAKEUP  0
358
359 #define __wait_event_lock_irq(wq, condition, lock, cmd)                 \
360 do {                                                                    \
361         wait_queue_t __wait;                                            \
362         init_waitqueue_entry(&__wait, current);                         \
363                                                                         \
364         add_wait_queue(&wq, &__wait);                                   \
365         for (;;) {                                                      \
366                 set_current_state(TASK_UNINTERRUPTIBLE);                \
367                 if (condition)                                          \
368                         break;                                          \
369                 spin_unlock_irq(&lock);                                 \
370                 cmd;                                                    \
371                 schedule();                                             \
372                 spin_lock_irq(&lock);                                   \
373         }                                                               \
374         current->state = TASK_RUNNING;                                  \
375         remove_wait_queue(&wq, &__wait);                                \
376 } while (0)
377
378 #define wait_event_lock_irq(wq, condition, lock, cmd)                   \
379 do {                                                                    \
380         if (condition)                                                  \
381                 break;                                                  \
382         __wait_event_lock_irq(wq, condition, lock, cmd);                \
383 } while (0)
384
385 static inline void safe_put_page(struct page *p)
386 {
387         if (p) put_page(p);
388 }
389
390 #endif /* CONFIG_BLOCK */
391 #endif
392
393
394 extern int register_md_personality(struct mdk_personality *p);
395 extern int unregister_md_personality(struct mdk_personality *p);
396 extern mdk_thread_t * md_register_thread(void (*run) (mddev_t *mddev),
397                                 mddev_t *mddev, const char *name);
398 extern void md_unregister_thread(mdk_thread_t *thread);
399 extern void md_wakeup_thread(mdk_thread_t *thread);
400 extern void md_check_recovery(mddev_t *mddev);
401 extern void md_write_start(mddev_t *mddev, struct bio *bi);
402 extern void md_write_end(mddev_t *mddev);
403 extern void md_done_sync(mddev_t *mddev, int blocks, int ok);
404 extern void md_error(mddev_t *mddev, mdk_rdev_t *rdev);
405
406 extern void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
407                            sector_t sector, int size, struct page *page);
408 extern void md_super_wait(mddev_t *mddev);
409 extern int sync_page_io(struct block_device *bdev, sector_t sector, int size,
410                         struct page *page, int rw);
411 extern void md_do_sync(mddev_t *mddev);
412 extern void md_new_event(mddev_t *mddev);
413 extern int md_allow_write(mddev_t *mddev);
414 extern void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev);