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