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