md: collect bitmap-specific fields into one structure.
[linux-2.6.git] / drivers / md / md.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_MD_H
16 #define _MD_MD_H
17
18 #include <linux/blkdev.h>
19 #include <linux/kobject.h>
20 #include <linux/list.h>
21 #include <linux/mm.h>
22 #include <linux/mutex.h>
23 #include <linux/timer.h>
24 #include <linux/wait.h>
25 #include <linux/workqueue.h>
26
27 #define MaxSector (~(sector_t)0)
28
29 typedef struct mddev_s mddev_t;
30 typedef struct mdk_rdev_s mdk_rdev_t;
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 sectors;               /* Device size (in 512bytes sectors) */
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                             suspended;
123         atomic_t                        active_io;
124         int                             ro;
125
126         struct gendisk                  *gendisk;
127
128         struct kobject                  kobj;
129         int                             hold_active;
130 #define UNTIL_IOCTL     1
131 #define UNTIL_STOP      2
132
133         /* Superblock information */
134         int                             major_version,
135                                         minor_version,
136                                         patch_version;
137         int                             persistent;
138         int                             external;       /* metadata is
139                                                          * managed externally */
140         char                            metadata_type[17]; /* externally set*/
141         int                             chunk_sectors;
142         time_t                          ctime, utime;
143         int                             level, layout;
144         char                            clevel[16];
145         int                             raid_disks;
146         int                             max_disks;
147         sector_t                        dev_sectors;    /* used size of
148                                                          * component devices */
149         sector_t                        array_sectors; /* exported array size */
150         int                             external_size; /* size managed
151                                                         * externally */
152         __u64                           events;
153
154         char                            uuid[16];
155
156         /* If the array is being reshaped, we need to record the
157          * new shape and an indication of where we are up to.
158          * This is written to the superblock.
159          * If reshape_position is MaxSector, then no reshape is happening (yet).
160          */
161         sector_t                        reshape_position;
162         int                             delta_disks, new_level, new_layout;
163         int                             new_chunk_sectors;
164
165         struct mdk_thread_s             *thread;        /* management thread */
166         struct mdk_thread_s             *sync_thread;   /* doing resync or reconstruct */
167         sector_t                        curr_resync;    /* last block scheduled */
168         /* As resync requests can complete out of order, we cannot easily track
169          * how much resync has been completed.  So we occasionally pause until
170          * everything completes, then set curr_resync_completed to curr_resync.
171          * As such it may be well behind the real resync mark, but it is a value
172          * we are certain of.
173          */
174         sector_t                        curr_resync_completed;
175         unsigned long                   resync_mark;    /* a recent timestamp */
176         sector_t                        resync_mark_cnt;/* blocks written at resync_mark */
177         sector_t                        curr_mark_cnt; /* blocks scheduled now */
178
179         sector_t                        resync_max_sectors; /* may be set by personality */
180
181         sector_t                        resync_mismatches; /* count of sectors where
182                                                             * parity/replica mismatch found
183                                                             */
184
185         /* allow user-space to request suspension of IO to regions of the array */
186         sector_t                        suspend_lo;
187         sector_t                        suspend_hi;
188         /* if zero, use the system-wide default */
189         int                             sync_speed_min;
190         int                             sync_speed_max;
191
192         /* resync even though the same disks are shared among md-devices */
193         int                             parallel_resync;
194
195         int                             ok_start_degraded;
196         /* recovery/resync flags 
197          * NEEDED:   we might need to start a resync/recover
198          * RUNNING:  a thread is running, or about to be started
199          * SYNC:     actually doing a resync, not a recovery
200          * RECOVER:  doing recovery, or need to try it.
201          * INTR:     resync needs to be aborted for some reason
202          * DONE:     thread is done and is waiting to be reaped
203          * REQUEST:  user-space has requested a sync (used with SYNC)
204          * CHECK:    user-space request for check-only, no repair
205          * RESHAPE:  A reshape is happening
206          *
207          * If neither SYNC or RESHAPE are set, then it is a recovery.
208          */
209 #define MD_RECOVERY_RUNNING     0
210 #define MD_RECOVERY_SYNC        1
211 #define MD_RECOVERY_RECOVER     2
212 #define MD_RECOVERY_INTR        3
213 #define MD_RECOVERY_DONE        4
214 #define MD_RECOVERY_NEEDED      5
215 #define MD_RECOVERY_REQUESTED   6
216 #define MD_RECOVERY_CHECK       7
217 #define MD_RECOVERY_RESHAPE     8
218 #define MD_RECOVERY_FROZEN      9
219
220         unsigned long                   recovery;
221         int                             recovery_disabled; /* if we detect that recovery
222                                                             * will always fail, set this
223                                                             * so we don't loop trying */
224
225         int                             in_sync;        /* know to not need resync */
226         /* 'open_mutex' avoids races between 'md_open' and 'do_md_stop', so
227          * that we are never stopping an array while it is open.
228          * 'reconfig_mutex' protects all other reconfiguration.
229          * These locks are separate due to conflicting interactions
230          * with bdev->bd_mutex.
231          * Lock ordering is:
232          *  reconfig_mutex -> bd_mutex : e.g. do_md_run -> revalidate_disk
233          *  bd_mutex -> open_mutex:  e.g. __blkdev_get -> md_open
234          */
235         struct mutex                    open_mutex;
236         struct mutex                    reconfig_mutex;
237         atomic_t                        active;         /* general refcount */
238         atomic_t                        openers;        /* number of active opens */
239
240         int                             changed;        /* true if we might need to reread partition info */
241         int                             degraded;       /* whether md should consider
242                                                          * adding a spare
243                                                          */
244         int                             barriers_work;  /* initialised to true, cleared as soon
245                                                          * as a barrier request to slave
246                                                          * fails.  Only supported
247                                                          */
248         struct bio                      *biolist;       /* bios that need to be retried
249                                                          * because BIO_RW_BARRIER is not supported
250                                                          */
251
252         atomic_t                        recovery_active; /* blocks scheduled, but not written */
253         wait_queue_head_t               recovery_wait;
254         sector_t                        recovery_cp;
255         sector_t                        resync_min;     /* user requested sync
256                                                          * starts here */
257         sector_t                        resync_max;     /* resync should pause
258                                                          * when it gets here */
259
260         struct sysfs_dirent             *sysfs_state;   /* handle for 'array_state'
261                                                          * file in sysfs.
262                                                          */
263         struct sysfs_dirent             *sysfs_action;  /* handle for 'sync_action' */
264
265         struct work_struct del_work;    /* used for delayed sysfs removal */
266
267         spinlock_t                      write_lock;
268         wait_queue_head_t               sb_wait;        /* for waiting on superblock updates */
269         atomic_t                        pending_writes; /* number of active superblock writes */
270
271         unsigned int                    safemode;       /* if set, update "clean" superblock
272                                                          * when no writes pending.
273                                                          */ 
274         unsigned int                    safemode_delay;
275         struct timer_list               safemode_timer;
276         atomic_t                        writes_pending; 
277         struct request_queue            *queue; /* for plugging ... */
278
279         atomic_t                        write_behind; /* outstanding async IO */
280         unsigned int                    max_write_behind; /* 0 = sync */
281
282         struct bitmap                   *bitmap; /* the bitmap for the device */
283         struct {
284                 struct file             *file; /* the bitmap file */
285                 long                    offset; /* offset from superblock of
286                                                  * start of bitmap. May be
287                                                  * negative, but not '0'
288                                                  */
289                 long                    default_offset; /* this is the offset to use when
290                                                          * hot-adding a bitmap.  It should
291                                                          * eventually be settable by sysfs.
292                                                          */
293                 struct mutex            mutex;
294         } bitmap_info;
295
296         struct list_head                all_mddevs;
297
298         /* Generic barrier handling.
299          * If there is a pending barrier request, all other
300          * writes are blocked while the devices are flushed.
301          * The last to finish a flush schedules a worker to
302          * submit the barrier request (without the barrier flag),
303          * then submit more flush requests.
304          */
305         struct bio *barrier;
306         atomic_t flush_pending;
307         struct work_struct barrier_work;
308 };
309
310
311 static inline void rdev_dec_pending(mdk_rdev_t *rdev, mddev_t *mddev)
312 {
313         int faulty = test_bit(Faulty, &rdev->flags);
314         if (atomic_dec_and_test(&rdev->nr_pending) && faulty)
315                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
316 }
317
318 static inline void md_sync_acct(struct block_device *bdev, unsigned long nr_sectors)
319 {
320         atomic_add(nr_sectors, &bdev->bd_contains->bd_disk->sync_io);
321 }
322
323 struct mdk_personality
324 {
325         char *name;
326         int level;
327         struct list_head list;
328         struct module *owner;
329         int (*make_request)(struct request_queue *q, struct bio *bio);
330         int (*run)(mddev_t *mddev);
331         int (*stop)(mddev_t *mddev);
332         void (*status)(struct seq_file *seq, mddev_t *mddev);
333         /* error_handler must set ->faulty and clear ->in_sync
334          * if appropriate, and should abort recovery if needed 
335          */
336         void (*error_handler)(mddev_t *mddev, mdk_rdev_t *rdev);
337         int (*hot_add_disk) (mddev_t *mddev, mdk_rdev_t *rdev);
338         int (*hot_remove_disk) (mddev_t *mddev, int number);
339         int (*spare_active) (mddev_t *mddev);
340         sector_t (*sync_request)(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster);
341         int (*resize) (mddev_t *mddev, sector_t sectors);
342         sector_t (*size) (mddev_t *mddev, sector_t sectors, int raid_disks);
343         int (*check_reshape) (mddev_t *mddev);
344         int (*start_reshape) (mddev_t *mddev);
345         void (*finish_reshape) (mddev_t *mddev);
346         /* quiesce moves between quiescence states
347          * 0 - fully active
348          * 1 - no new requests allowed
349          * others - reserved
350          */
351         void (*quiesce) (mddev_t *mddev, int state);
352         /* takeover is used to transition an array from one
353          * personality to another.  The new personality must be able
354          * to handle the data in the current layout.
355          * e.g. 2drive raid1 -> 2drive raid5
356          *      ndrive raid5 -> degraded n+1drive raid6 with special layout
357          * If the takeover succeeds, a new 'private' structure is returned.
358          * This needs to be installed and then ->run used to activate the
359          * array.
360          */
361         void *(*takeover) (mddev_t *mddev);
362 };
363
364
365 struct md_sysfs_entry {
366         struct attribute attr;
367         ssize_t (*show)(mddev_t *, char *);
368         ssize_t (*store)(mddev_t *, const char *, size_t);
369 };
370
371
372 static inline char * mdname (mddev_t * mddev)
373 {
374         return mddev->gendisk ? mddev->gendisk->disk_name : "mdX";
375 }
376
377 /*
378  * iterates through some rdev ringlist. It's safe to remove the
379  * current 'rdev'. Dont touch 'tmp' though.
380  */
381 #define rdev_for_each_list(rdev, tmp, head)                             \
382         list_for_each_entry_safe(rdev, tmp, head, same_set)
383
384 /*
385  * iterates through the 'same array disks' ringlist
386  */
387 #define rdev_for_each(rdev, tmp, mddev)                         \
388         list_for_each_entry_safe(rdev, tmp, &((mddev)->disks), same_set)
389
390 #define rdev_for_each_rcu(rdev, mddev)                          \
391         list_for_each_entry_rcu(rdev, &((mddev)->disks), same_set)
392
393 typedef struct mdk_thread_s {
394         void                    (*run) (mddev_t *mddev);
395         mddev_t                 *mddev;
396         wait_queue_head_t       wqueue;
397         unsigned long           flags;
398         struct task_struct      *tsk;
399         unsigned long           timeout;
400 } mdk_thread_t;
401
402 #define THREAD_WAKEUP  0
403
404 #define __wait_event_lock_irq(wq, condition, lock, cmd)                 \
405 do {                                                                    \
406         wait_queue_t __wait;                                            \
407         init_waitqueue_entry(&__wait, current);                         \
408                                                                         \
409         add_wait_queue(&wq, &__wait);                                   \
410         for (;;) {                                                      \
411                 set_current_state(TASK_UNINTERRUPTIBLE);                \
412                 if (condition)                                          \
413                         break;                                          \
414                 spin_unlock_irq(&lock);                                 \
415                 cmd;                                                    \
416                 schedule();                                             \
417                 spin_lock_irq(&lock);                                   \
418         }                                                               \
419         current->state = TASK_RUNNING;                                  \
420         remove_wait_queue(&wq, &__wait);                                \
421 } while (0)
422
423 #define wait_event_lock_irq(wq, condition, lock, cmd)                   \
424 do {                                                                    \
425         if (condition)                                                  \
426                 break;                                                  \
427         __wait_event_lock_irq(wq, condition, lock, cmd);                \
428 } while (0)
429
430 static inline void safe_put_page(struct page *p)
431 {
432         if (p) put_page(p);
433 }
434
435 extern int register_md_personality(struct mdk_personality *p);
436 extern int unregister_md_personality(struct mdk_personality *p);
437 extern mdk_thread_t * md_register_thread(void (*run) (mddev_t *mddev),
438                                 mddev_t *mddev, const char *name);
439 extern void md_unregister_thread(mdk_thread_t *thread);
440 extern void md_wakeup_thread(mdk_thread_t *thread);
441 extern void md_check_recovery(mddev_t *mddev);
442 extern void md_write_start(mddev_t *mddev, struct bio *bi);
443 extern void md_write_end(mddev_t *mddev);
444 extern void md_done_sync(mddev_t *mddev, int blocks, int ok);
445 extern void md_error(mddev_t *mddev, mdk_rdev_t *rdev);
446
447 extern int mddev_congested(mddev_t *mddev, int bits);
448 extern void md_barrier_request(mddev_t *mddev, struct bio *bio);
449 extern void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
450                            sector_t sector, int size, struct page *page);
451 extern void md_super_wait(mddev_t *mddev);
452 extern int sync_page_io(struct block_device *bdev, sector_t sector, int size,
453                         struct page *page, int rw);
454 extern void md_do_sync(mddev_t *mddev);
455 extern void md_new_event(mddev_t *mddev);
456 extern int md_allow_write(mddev_t *mddev);
457 extern void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev);
458 extern void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors);
459 extern int md_check_no_bitmap(mddev_t *mddev);
460 extern int md_integrity_register(mddev_t *mddev);
461 void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev);
462
463 #endif /* _MD_MD_H */