2 md_k.h : kernel internal structure of the Linux MD driver
3 Copyright (C) 1996-98 Ingo Molnar, Gadi Oxman
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)
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.
18 #define MD_RESERVED 0UL
23 #define TRANSLUCENT 5UL
27 #define MAX_PERSONALITY 9UL
29 #define LEVEL_MULTIPATH (-4)
30 #define LEVEL_LINEAR (-1)
32 #define MaxSector (~(sector_t)0)
33 #define MD_THREAD_NAME_MAX 14
35 static inline int pers_to_level (int pers)
38 case MULTIPATH: return LEVEL_MULTIPATH;
40 case TRANSLUCENT: return -2;
41 case LINEAR: return LEVEL_LINEAR;
51 static inline int level_to_pers (int level)
54 case LEVEL_MULTIPATH: return MULTIPATH;
56 case -2: return TRANSLUCENT;
57 case LEVEL_LINEAR: return LINEAR;
67 typedef struct mddev_s mddev_t;
68 typedef struct mdk_rdev_s mdk_rdev_t;
70 #define MAX_MD_DEVS 256 /* Max number of md dev */
73 * options passed in raidrun:
76 #define MAX_CHUNK_SIZE (4096*1024)
82 static inline int disk_faulty(mdp_disk_t * d)
84 return d->state & (1 << MD_DISK_FAULTY);
87 static inline int disk_active(mdp_disk_t * d)
89 return d->state & (1 << MD_DISK_ACTIVE);
92 static inline int disk_sync(mdp_disk_t * d)
94 return d->state & (1 << MD_DISK_SYNC);
97 static inline int disk_spare(mdp_disk_t * d)
99 return !disk_sync(d) && !disk_active(d) && !disk_faulty(d);
102 static inline int disk_removed(mdp_disk_t * d)
104 return d->state & (1 << MD_DISK_REMOVED);
107 static inline void mark_disk_faulty(mdp_disk_t * d)
109 d->state |= (1 << MD_DISK_FAULTY);
112 static inline void mark_disk_active(mdp_disk_t * d)
114 d->state |= (1 << MD_DISK_ACTIVE);
117 static inline void mark_disk_sync(mdp_disk_t * d)
119 d->state |= (1 << MD_DISK_SYNC);
122 static inline void mark_disk_spare(mdp_disk_t * d)
127 static inline void mark_disk_removed(mdp_disk_t * d)
129 d->state = (1 << MD_DISK_FAULTY) | (1 << MD_DISK_REMOVED);
132 static inline void mark_disk_inactive(mdp_disk_t * d)
134 d->state &= ~(1 << MD_DISK_ACTIVE);
137 static inline void mark_disk_nonsync(mdp_disk_t * d)
139 d->state &= ~(1 << MD_DISK_SYNC);
143 * MD's 'extended' device
147 struct list_head same_set; /* RAID devices within the same set */
149 sector_t size; /* Device size (in blocks) */
150 mddev_t *mddev; /* RAID array if running */
151 unsigned long last_events; /* IO event timestamp */
153 struct block_device *bdev; /* block device handle */
155 struct page *sb_page;
157 sector_t data_offset; /* start of data in array */
159 int preferred_minor; /* autorun support */
161 /* A device can be in one of three states based on two flags:
162 * Not working: faulty==1 in_sync==0
163 * Fully working: faulty==0 in_sync==1
166 * faulty==0 in_sync==0
168 * It can never have faulty==1, in_sync==1
169 * This reduces the burden of testing multiple flags in many cases
171 int faulty; /* if faulty do not issue IO requests */
172 int in_sync; /* device is a full member of the array */
174 int desc_nr; /* descriptor index in the superblock */
175 int raid_disk; /* role of device in array */
177 atomic_t nr_pending; /* number of pending requests.
178 * only maintained for arrays that
179 * support hot removal
183 typedef struct mdk_personality_s mdk_personality_t;
188 mdk_personality_t *pers;
191 struct list_head disks;
195 struct gendisk *gendisk;
197 /* Superblock information */
207 sector_t size; /* used size of component devices */
208 sector_t array_size; /* exported array size */
213 struct mdk_thread_s *thread; /* management thread */
214 struct mdk_thread_s *sync_thread; /* doing resync or reconstruct */
215 sector_t curr_resync; /* blocks scheduled */
216 unsigned long resync_mark; /* a recent timestamp */
217 sector_t resync_mark_cnt;/* blocks written at resync_mark */
219 /* recovery/resync flags
220 * NEEDED: we might need to start a resync/recover
221 * RUNNING: a thread is running, or about to be started
222 * SYNC: actually doing a resync, not a recovery
223 * ERR: and IO error was detected - abort the resync/recovery
224 * INTR: someone requested a (clean) early abort.
225 * DONE: thread is done and is waiting to be reaped
227 #define MD_RECOVERY_RUNNING 0
228 #define MD_RECOVERY_SYNC 1
229 #define MD_RECOVERY_ERR 2
230 #define MD_RECOVERY_INTR 3
231 #define MD_RECOVERY_DONE 4
232 #define MD_RECOVERY_NEEDED 5
233 unsigned long recovery;
235 int in_sync; /* know to not need resync */
236 struct semaphore reconfig_sem;
239 int changed; /* true if we might need to reread partition info */
240 int degraded; /* whether md should consider
244 atomic_t recovery_active; /* blocks scheduled, but not written */
245 wait_queue_head_t recovery_wait;
246 sector_t recovery_cp;
247 unsigned int safemode; /* if set, update "clean" superblock
248 * when no writes pending.
250 unsigned int safemode_delay;
251 struct timer_list safemode_timer;
252 atomic_t writes_pending;
253 request_queue_t *queue; /* for plugging ... */
255 struct list_head all_mddevs;
259 static inline void rdev_dec_pending(mdk_rdev_t *rdev, mddev_t *mddev)
261 int faulty = rdev->faulty;
262 if (atomic_dec_and_test(&rdev->nr_pending) && faulty)
263 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
266 struct mdk_personality_s
269 struct module *owner;
270 int (*make_request)(request_queue_t *q, struct bio *bio);
271 int (*run)(mddev_t *mddev);
272 int (*stop)(mddev_t *mddev);
273 void (*status)(struct seq_file *seq, mddev_t *mddev);
274 /* error_handler must set ->faulty and clear ->in_sync
275 * if appropriate, and should abort recovery if needed
277 void (*error_handler)(mddev_t *mddev, mdk_rdev_t *rdev);
278 int (*hot_add_disk) (mddev_t *mddev, mdk_rdev_t *rdev);
279 int (*hot_remove_disk) (mddev_t *mddev, int number);
280 int (*spare_active) (mddev_t *mddev);
281 int (*sync_request)(mddev_t *mddev, sector_t sector_nr, int go_faster);
282 int (*resize) (mddev_t *mddev, sector_t sectors);
283 int (*reshape) (mddev_t *mddev, int raid_disks);
287 static inline char * mdname (mddev_t * mddev)
289 return mddev->gendisk ? mddev->gendisk->disk_name : "mdX";
292 extern mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr);
295 * iterates through some rdev ringlist. It's safe to remove the
296 * current 'rdev'. Dont touch 'tmp' though.
298 #define ITERATE_RDEV_GENERIC(head,rdev,tmp) \
300 for ((tmp) = (head).next; \
301 (rdev) = (list_entry((tmp), mdk_rdev_t, same_set)), \
302 (tmp) = (tmp)->next, (tmp)->prev != &(head) \
305 * iterates through the 'same array disks' ringlist
307 #define ITERATE_RDEV(mddev,rdev,tmp) \
308 ITERATE_RDEV_GENERIC((mddev)->disks,rdev,tmp)
311 * Iterates through 'pending RAID disks'
313 #define ITERATE_RDEV_PENDING(rdev,tmp) \
314 ITERATE_RDEV_GENERIC(pending_raid_disks,rdev,tmp)
316 typedef struct mdk_thread_s {
317 void (*run) (mddev_t *mddev);
319 wait_queue_head_t wqueue;
321 struct completion *event;
322 struct task_struct *tsk;
326 #define THREAD_WAKEUP 0
328 #define __wait_event_lock_irq(wq, condition, lock, cmd) \
330 wait_queue_t __wait; \
331 init_waitqueue_entry(&__wait, current); \
333 add_wait_queue(&wq, &__wait); \
335 set_current_state(TASK_UNINTERRUPTIBLE); \
338 spin_unlock_irq(&lock); \
341 spin_lock_irq(&lock); \
343 current->state = TASK_RUNNING; \
344 remove_wait_queue(&wq, &__wait); \
347 #define wait_event_lock_irq(wq, condition, lock, cmd) \
351 __wait_event_lock_irq(wq, condition, lock, cmd); \