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
28 #define MAX_PERSONALITY 10UL
30 #define LEVEL_MULTIPATH (-4)
31 #define LEVEL_LINEAR (-1)
33 #define MaxSector (~(sector_t)0)
34 #define MD_THREAD_NAME_MAX 14
36 static inline int pers_to_level (int pers)
39 case MULTIPATH: return LEVEL_MULTIPATH;
41 case TRANSLUCENT: return -2;
42 case LINEAR: return LEVEL_LINEAR;
47 case RAID10: return 10;
53 static inline int level_to_pers (int level)
56 case LEVEL_MULTIPATH: return MULTIPATH;
58 case -2: return TRANSLUCENT;
59 case LEVEL_LINEAR: return LINEAR;
65 case 10: return RAID10;
70 typedef struct mddev_s mddev_t;
71 typedef struct mdk_rdev_s mdk_rdev_t;
73 #define MAX_MD_DEVS 256 /* Max number of md dev */
76 * options passed in raidrun:
79 #define MAX_CHUNK_SIZE (4096*1024)
85 static inline int disk_faulty(mdp_disk_t * d)
87 return d->state & (1 << MD_DISK_FAULTY);
90 static inline int disk_active(mdp_disk_t * d)
92 return d->state & (1 << MD_DISK_ACTIVE);
95 static inline int disk_sync(mdp_disk_t * d)
97 return d->state & (1 << MD_DISK_SYNC);
100 static inline int disk_spare(mdp_disk_t * d)
102 return !disk_sync(d) && !disk_active(d) && !disk_faulty(d);
105 static inline int disk_removed(mdp_disk_t * d)
107 return d->state & (1 << MD_DISK_REMOVED);
110 static inline void mark_disk_faulty(mdp_disk_t * d)
112 d->state |= (1 << MD_DISK_FAULTY);
115 static inline void mark_disk_active(mdp_disk_t * d)
117 d->state |= (1 << MD_DISK_ACTIVE);
120 static inline void mark_disk_sync(mdp_disk_t * d)
122 d->state |= (1 << MD_DISK_SYNC);
125 static inline void mark_disk_spare(mdp_disk_t * d)
130 static inline void mark_disk_removed(mdp_disk_t * d)
132 d->state = (1 << MD_DISK_FAULTY) | (1 << MD_DISK_REMOVED);
135 static inline void mark_disk_inactive(mdp_disk_t * d)
137 d->state &= ~(1 << MD_DISK_ACTIVE);
140 static inline void mark_disk_nonsync(mdp_disk_t * d)
142 d->state &= ~(1 << MD_DISK_SYNC);
146 * MD's 'extended' device
150 struct list_head same_set; /* RAID devices within the same set */
152 sector_t size; /* Device size (in blocks) */
153 mddev_t *mddev; /* RAID array if running */
154 unsigned long last_events; /* IO event timestamp */
156 struct block_device *bdev; /* block device handle */
158 struct page *sb_page;
160 sector_t data_offset; /* start of data in array */
162 int preferred_minor; /* autorun support */
164 /* A device can be in one of three states based on two flags:
165 * Not working: faulty==1 in_sync==0
166 * Fully working: faulty==0 in_sync==1
169 * faulty==0 in_sync==0
171 * It can never have faulty==1, in_sync==1
172 * This reduces the burden of testing multiple flags in many cases
174 int faulty; /* if faulty do not issue IO requests */
175 int in_sync; /* device is a full member of the array */
177 int desc_nr; /* descriptor index in the superblock */
178 int raid_disk; /* role of device in array */
180 atomic_t nr_pending; /* number of pending requests.
181 * only maintained for arrays that
182 * support hot removal
186 typedef struct mdk_personality_s mdk_personality_t;
191 mdk_personality_t *pers;
194 struct list_head disks;
198 struct gendisk *gendisk;
200 /* Superblock information */
210 sector_t size; /* used size of component devices */
211 sector_t array_size; /* exported array size */
216 struct mdk_thread_s *thread; /* management thread */
217 struct mdk_thread_s *sync_thread; /* doing resync or reconstruct */
218 sector_t curr_resync; /* blocks scheduled */
219 unsigned long resync_mark; /* a recent timestamp */
220 sector_t resync_mark_cnt;/* blocks written at resync_mark */
222 sector_t resync_max_sectors; /* may be set by personality */
223 /* recovery/resync flags
224 * NEEDED: we might need to start a resync/recover
225 * RUNNING: a thread is running, or about to be started
226 * SYNC: actually doing a resync, not a recovery
227 * ERR: and IO error was detected - abort the resync/recovery
228 * INTR: someone requested a (clean) early abort.
229 * DONE: thread is done and is waiting to be reaped
231 #define MD_RECOVERY_RUNNING 0
232 #define MD_RECOVERY_SYNC 1
233 #define MD_RECOVERY_ERR 2
234 #define MD_RECOVERY_INTR 3
235 #define MD_RECOVERY_DONE 4
236 #define MD_RECOVERY_NEEDED 5
237 unsigned long recovery;
239 int in_sync; /* know to not need resync */
240 struct semaphore reconfig_sem;
243 int changed; /* true if we might need to reread partition info */
244 int degraded; /* whether md should consider
248 atomic_t recovery_active; /* blocks scheduled, but not written */
249 wait_queue_head_t recovery_wait;
250 sector_t recovery_cp;
251 unsigned int safemode; /* if set, update "clean" superblock
252 * when no writes pending.
254 unsigned int safemode_delay;
255 struct timer_list safemode_timer;
256 atomic_t writes_pending;
257 request_queue_t *queue; /* for plugging ... */
259 struct list_head all_mddevs;
263 static inline void rdev_dec_pending(mdk_rdev_t *rdev, mddev_t *mddev)
265 int faulty = rdev->faulty;
266 if (atomic_dec_and_test(&rdev->nr_pending) && faulty)
267 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
270 static inline void md_sync_acct(struct block_device *bdev, unsigned long nr_sectors)
272 atomic_add(nr_sectors, &bdev->bd_contains->bd_disk->sync_io);
275 struct mdk_personality_s
278 struct module *owner;
279 int (*make_request)(request_queue_t *q, struct bio *bio);
280 int (*run)(mddev_t *mddev);
281 int (*stop)(mddev_t *mddev);
282 void (*status)(struct seq_file *seq, mddev_t *mddev);
283 /* error_handler must set ->faulty and clear ->in_sync
284 * if appropriate, and should abort recovery if needed
286 void (*error_handler)(mddev_t *mddev, mdk_rdev_t *rdev);
287 int (*hot_add_disk) (mddev_t *mddev, mdk_rdev_t *rdev);
288 int (*hot_remove_disk) (mddev_t *mddev, int number);
289 int (*spare_active) (mddev_t *mddev);
290 int (*sync_request)(mddev_t *mddev, sector_t sector_nr, int go_faster);
291 int (*resize) (mddev_t *mddev, sector_t sectors);
292 int (*reshape) (mddev_t *mddev, int raid_disks);
296 static inline char * mdname (mddev_t * mddev)
298 return mddev->gendisk ? mddev->gendisk->disk_name : "mdX";
301 extern mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr);
304 * iterates through some rdev ringlist. It's safe to remove the
305 * current 'rdev'. Dont touch 'tmp' though.
307 #define ITERATE_RDEV_GENERIC(head,rdev,tmp) \
309 for ((tmp) = (head).next; \
310 (rdev) = (list_entry((tmp), mdk_rdev_t, same_set)), \
311 (tmp) = (tmp)->next, (tmp)->prev != &(head) \
314 * iterates through the 'same array disks' ringlist
316 #define ITERATE_RDEV(mddev,rdev,tmp) \
317 ITERATE_RDEV_GENERIC((mddev)->disks,rdev,tmp)
320 * Iterates through 'pending RAID disks'
322 #define ITERATE_RDEV_PENDING(rdev,tmp) \
323 ITERATE_RDEV_GENERIC(pending_raid_disks,rdev,tmp)
325 typedef struct mdk_thread_s {
326 void (*run) (mddev_t *mddev);
328 wait_queue_head_t wqueue;
330 struct completion *event;
331 struct task_struct *tsk;
335 #define THREAD_WAKEUP 0
337 #define __wait_event_lock_irq(wq, condition, lock, cmd) \
339 wait_queue_t __wait; \
340 init_waitqueue_entry(&__wait, current); \
342 add_wait_queue(&wq, &__wait); \
344 set_current_state(TASK_UNINTERRUPTIBLE); \
347 spin_unlock_irq(&lock); \
350 spin_lock_irq(&lock); \
352 current->state = TASK_RUNNING; \
353 remove_wait_queue(&wq, &__wait); \
356 #define wait_event_lock_irq(wq, condition, lock, cmd) \
360 __wait_event_lock_irq(wq, condition, lock, cmd); \