Merge branch 'akpm'
[linux-2.6/next.git] / drivers / md / md.h
blob1e586bb4452e3011a71cc59ae31e5e96ed63488e
1 /*
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)
8 any later version.
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.
15 #ifndef _MD_MD_H
16 #define _MD_MD_H
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>
27 #define MaxSector (~(sector_t)0)
29 typedef struct mddev_s mddev_t;
30 typedef struct mdk_rdev_s mdk_rdev_t;
32 /* Bad block numbers are stored sorted in a single page.
33 * 64bits is used for each block or extent.
34 * 54 bits are sector number, 9 bits are extent size,
35 * 1 bit is an 'acknowledged' flag.
37 #define MD_MAX_BADBLOCKS (PAGE_SIZE/8)
40 * MD's 'extended' device
42 struct mdk_rdev_s
44 struct list_head same_set; /* RAID devices within the same set */
46 sector_t sectors; /* Device size (in 512bytes sectors) */
47 mddev_t *mddev; /* RAID array if running */
48 int last_events; /* IO event timestamp */
51 * If meta_bdev is non-NULL, it means that a separate device is
52 * being used to store the metadata (superblock/bitmap) which
53 * would otherwise be contained on the same device as the data (bdev).
55 struct block_device *meta_bdev;
56 struct block_device *bdev; /* block device handle */
58 struct page *sb_page, *bb_page;
59 int sb_loaded;
60 __u64 sb_events;
61 sector_t data_offset; /* start of data in array */
62 sector_t sb_start; /* offset of the super block (in 512byte sectors) */
63 int sb_size; /* bytes in the superblock */
64 int preferred_minor; /* autorun support */
66 struct kobject kobj;
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
75 * It can never have faulty==1, in_sync==1
76 * This reduces the burden of testing multiple flags in many cases
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 AutoDetected 7 /* added by auto-detect */
84 #define Blocked 8 /* An error occurred but has not yet
85 * been acknowledged by the metadata
86 * handler, so don't allow writes
87 * until it is cleared */
88 #define WriteErrorSeen 9 /* A write error has been seen on this
89 * device
91 #define FaultRecorded 10 /* Intermediate state for clearing
92 * Blocked. The Fault is/will-be
93 * recorded in the metadata, but that
94 * metadata hasn't been stored safely
95 * on disk yet.
97 #define BlockedBadBlocks 11 /* A writer is blocked because they
98 * found an unacknowledged bad-block.
99 * This can safely be cleared at any
100 * time, and the writer will re-check.
101 * It may be set at any time, and at
102 * worst the writer will timeout and
103 * re-check. So setting it as
104 * accurately as possible is good, but
105 * not absolutely critical.
107 wait_queue_head_t blocked_wait;
109 int desc_nr; /* descriptor index in the superblock */
110 int raid_disk; /* role of device in array */
111 int new_raid_disk; /* role that the device will have in
112 * the array after a level-change completes.
114 int saved_raid_disk; /* role that device used to have in the
115 * array and could again if we did a partial
116 * resync from the bitmap
118 sector_t recovery_offset;/* If this device has been partially
119 * recovered, this is where we were
120 * up to.
123 atomic_t nr_pending; /* number of pending requests.
124 * only maintained for arrays that
125 * support hot removal
127 atomic_t read_errors; /* number of consecutive read errors that
128 * we have tried to ignore.
130 struct timespec last_read_error; /* monotonic time since our
131 * last read error
133 atomic_t corrected_errors; /* number of corrected read errors,
134 * for reporting to userspace and storing
135 * in superblock.
137 struct work_struct del_work; /* used for delayed sysfs removal */
139 struct sysfs_dirent *sysfs_state; /* handle for 'state'
140 * sysfs entry */
142 struct badblocks {
143 int count; /* count of bad blocks */
144 int unacked_exist; /* there probably are unacknowledged
145 * bad blocks. This is only cleared
146 * when a read discovers none
148 int shift; /* shift from sectors to block size
149 * a -ve shift means badblocks are
150 * disabled.*/
151 u64 *page; /* badblock list */
152 int changed;
153 seqlock_t lock;
155 sector_t sector;
156 sector_t size; /* in sectors */
157 } badblocks;
160 #define BB_LEN_MASK (0x00000000000001FFULL)
161 #define BB_OFFSET_MASK (0x7FFFFFFFFFFFFE00ULL)
162 #define BB_ACK_MASK (0x8000000000000000ULL)
163 #define BB_MAX_LEN 512
164 #define BB_OFFSET(x) (((x) & BB_OFFSET_MASK) >> 9)
165 #define BB_LEN(x) (((x) & BB_LEN_MASK) + 1)
166 #define BB_ACK(x) (!!((x) & BB_ACK_MASK))
167 #define BB_MAKE(a, l, ack) (((a)<<9) | ((l)-1) | ((u64)(!!(ack)) << 63))
169 extern int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
170 sector_t *first_bad, int *bad_sectors);
171 static inline int is_badblock(mdk_rdev_t *rdev, sector_t s, int sectors,
172 sector_t *first_bad, int *bad_sectors)
174 if (unlikely(rdev->badblocks.count)) {
175 int rv = md_is_badblock(&rdev->badblocks, rdev->data_offset + s,
176 sectors,
177 first_bad, bad_sectors);
178 if (rv)
179 *first_bad -= rdev->data_offset;
180 return rv;
182 return 0;
184 extern int rdev_set_badblocks(mdk_rdev_t *rdev, sector_t s, int sectors,
185 int acknowledged);
186 extern int rdev_clear_badblocks(mdk_rdev_t *rdev, sector_t s, int sectors);
187 extern void md_ack_all_badblocks(struct badblocks *bb);
189 struct mddev_s
191 void *private;
192 struct mdk_personality *pers;
193 dev_t unit;
194 int md_minor;
195 struct list_head disks;
196 unsigned long flags;
197 #define MD_CHANGE_DEVS 0 /* Some device status has changed */
198 #define MD_CHANGE_CLEAN 1 /* transition to or from 'clean' */
199 #define MD_CHANGE_PENDING 2 /* switch from 'clean' to 'active' in progress */
200 #define MD_ARRAY_FIRST_USE 3 /* First use of array, needs initialization */
202 int suspended;
203 atomic_t active_io;
204 int ro;
205 int sysfs_active; /* set when sysfs deletes
206 * are happening, so run/
207 * takeover/stop are not safe
209 int ready; /* See when safe to pass
210 * IO requests down */
211 struct gendisk *gendisk;
213 struct kobject kobj;
214 int hold_active;
215 #define UNTIL_IOCTL 1
216 #define UNTIL_STOP 2
218 /* Superblock information */
219 int major_version,
220 minor_version,
221 patch_version;
222 int persistent;
223 int external; /* metadata is
224 * managed externally */
225 char metadata_type[17]; /* externally set*/
226 int chunk_sectors;
227 time_t ctime, utime;
228 int level, layout;
229 char clevel[16];
230 int raid_disks;
231 int max_disks;
232 sector_t dev_sectors; /* used size of
233 * component devices */
234 sector_t array_sectors; /* exported array size */
235 int external_size; /* size managed
236 * externally */
237 __u64 events;
238 /* If the last 'event' was simply a clean->dirty transition, and
239 * we didn't write it to the spares, then it is safe and simple
240 * to just decrement the event count on a dirty->clean transition.
241 * So we record that possibility here.
243 int can_decrease_events;
245 char uuid[16];
247 /* If the array is being reshaped, we need to record the
248 * new shape and an indication of where we are up to.
249 * This is written to the superblock.
250 * If reshape_position is MaxSector, then no reshape is happening (yet).
252 sector_t reshape_position;
253 int delta_disks, new_level, new_layout;
254 int new_chunk_sectors;
256 atomic_t plug_cnt; /* If device is expecting
257 * more bios soon.
259 struct mdk_thread_s *thread; /* management thread */
260 struct mdk_thread_s *sync_thread; /* doing resync or reconstruct */
261 sector_t curr_resync; /* last block scheduled */
262 /* As resync requests can complete out of order, we cannot easily track
263 * how much resync has been completed. So we occasionally pause until
264 * everything completes, then set curr_resync_completed to curr_resync.
265 * As such it may be well behind the real resync mark, but it is a value
266 * we are certain of.
268 sector_t curr_resync_completed;
269 unsigned long resync_mark; /* a recent timestamp */
270 sector_t resync_mark_cnt;/* blocks written at resync_mark */
271 sector_t curr_mark_cnt; /* blocks scheduled now */
273 sector_t resync_max_sectors; /* may be set by personality */
275 sector_t resync_mismatches; /* count of sectors where
276 * parity/replica mismatch found
279 /* allow user-space to request suspension of IO to regions of the array */
280 sector_t suspend_lo;
281 sector_t suspend_hi;
282 /* if zero, use the system-wide default */
283 int sync_speed_min;
284 int sync_speed_max;
286 /* resync even though the same disks are shared among md-devices */
287 int parallel_resync;
289 int ok_start_degraded;
290 /* recovery/resync flags
291 * NEEDED: we might need to start a resync/recover
292 * RUNNING: a thread is running, or about to be started
293 * SYNC: actually doing a resync, not a recovery
294 * RECOVER: doing recovery, or need to try it.
295 * INTR: resync needs to be aborted for some reason
296 * DONE: thread is done and is waiting to be reaped
297 * REQUEST: user-space has requested a sync (used with SYNC)
298 * CHECK: user-space request for check-only, no repair
299 * RESHAPE: A reshape is happening
301 * If neither SYNC or RESHAPE are set, then it is a recovery.
303 #define MD_RECOVERY_RUNNING 0
304 #define MD_RECOVERY_SYNC 1
305 #define MD_RECOVERY_RECOVER 2
306 #define MD_RECOVERY_INTR 3
307 #define MD_RECOVERY_DONE 4
308 #define MD_RECOVERY_NEEDED 5
309 #define MD_RECOVERY_REQUESTED 6
310 #define MD_RECOVERY_CHECK 7
311 #define MD_RECOVERY_RESHAPE 8
312 #define MD_RECOVERY_FROZEN 9
314 unsigned long recovery;
315 /* If a RAID personality determines that recovery (of a particular
316 * device) will fail due to a read error on the source device, it
317 * takes a copy of this number and does not attempt recovery again
318 * until this number changes.
320 int recovery_disabled;
322 int in_sync; /* know to not need resync */
323 /* 'open_mutex' avoids races between 'md_open' and 'do_md_stop', so
324 * that we are never stopping an array while it is open.
325 * 'reconfig_mutex' protects all other reconfiguration.
326 * These locks are separate due to conflicting interactions
327 * with bdev->bd_mutex.
328 * Lock ordering is:
329 * reconfig_mutex -> bd_mutex : e.g. do_md_run -> revalidate_disk
330 * bd_mutex -> open_mutex: e.g. __blkdev_get -> md_open
332 struct mutex open_mutex;
333 struct mutex reconfig_mutex;
334 atomic_t active; /* general refcount */
335 atomic_t openers; /* number of active opens */
337 int changed; /* True if we might need to
338 * reread partition info */
339 int degraded; /* whether md should consider
340 * adding a spare
343 atomic_t recovery_active; /* blocks scheduled, but not written */
344 wait_queue_head_t recovery_wait;
345 sector_t recovery_cp;
346 sector_t resync_min; /* user requested sync
347 * starts here */
348 sector_t resync_max; /* resync should pause
349 * when it gets here */
351 struct sysfs_dirent *sysfs_state; /* handle for 'array_state'
352 * file in sysfs.
354 struct sysfs_dirent *sysfs_action; /* handle for 'sync_action' */
356 struct work_struct del_work; /* used for delayed sysfs removal */
358 spinlock_t write_lock;
359 wait_queue_head_t sb_wait; /* for waiting on superblock updates */
360 atomic_t pending_writes; /* number of active superblock writes */
362 unsigned int safemode; /* if set, update "clean" superblock
363 * when no writes pending.
365 unsigned int safemode_delay;
366 struct timer_list safemode_timer;
367 atomic_t writes_pending;
368 struct request_queue *queue; /* for plugging ... */
370 struct bitmap *bitmap; /* the bitmap for the device */
371 struct {
372 struct file *file; /* the bitmap file */
373 loff_t offset; /* offset from superblock of
374 * start of bitmap. May be
375 * negative, but not '0'
376 * For external metadata, offset
377 * from start of device.
379 loff_t default_offset; /* this is the offset to use when
380 * hot-adding a bitmap. It should
381 * eventually be settable by sysfs.
383 struct mutex mutex;
384 unsigned long chunksize;
385 unsigned long daemon_sleep; /* how many jiffies between updates? */
386 unsigned long max_write_behind; /* write-behind mode */
387 int external;
388 } bitmap_info;
390 atomic_t max_corr_read_errors; /* max read retries */
391 struct list_head all_mddevs;
393 struct attribute_group *to_remove;
395 struct bio_set *bio_set;
397 /* Generic flush handling.
398 * The last to finish preflush schedules a worker to submit
399 * the rest of the request (without the REQ_FLUSH flag).
401 struct bio *flush_bio;
402 atomic_t flush_pending;
403 struct work_struct flush_work;
404 struct work_struct event_work; /* used by dm to report failure event */
405 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
409 static inline void rdev_dec_pending(mdk_rdev_t *rdev, mddev_t *mddev)
411 int faulty = test_bit(Faulty, &rdev->flags);
412 if (atomic_dec_and_test(&rdev->nr_pending) && faulty)
413 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
416 static inline void md_sync_acct(struct block_device *bdev, unsigned long nr_sectors)
418 atomic_add(nr_sectors, &bdev->bd_contains->bd_disk->sync_io);
421 struct mdk_personality
423 char *name;
424 int level;
425 struct list_head list;
426 struct module *owner;
427 int (*make_request)(mddev_t *mddev, struct bio *bio);
428 int (*run)(mddev_t *mddev);
429 int (*stop)(mddev_t *mddev);
430 void (*status)(struct seq_file *seq, mddev_t *mddev);
431 /* error_handler must set ->faulty and clear ->in_sync
432 * if appropriate, and should abort recovery if needed
434 void (*error_handler)(mddev_t *mddev, mdk_rdev_t *rdev);
435 int (*hot_add_disk) (mddev_t *mddev, mdk_rdev_t *rdev);
436 int (*hot_remove_disk) (mddev_t *mddev, int number);
437 int (*spare_active) (mddev_t *mddev);
438 sector_t (*sync_request)(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster);
439 int (*resize) (mddev_t *mddev, sector_t sectors);
440 sector_t (*size) (mddev_t *mddev, sector_t sectors, int raid_disks);
441 int (*check_reshape) (mddev_t *mddev);
442 int (*start_reshape) (mddev_t *mddev);
443 void (*finish_reshape) (mddev_t *mddev);
444 /* quiesce moves between quiescence states
445 * 0 - fully active
446 * 1 - no new requests allowed
447 * others - reserved
449 void (*quiesce) (mddev_t *mddev, int state);
450 /* takeover is used to transition an array from one
451 * personality to another. The new personality must be able
452 * to handle the data in the current layout.
453 * e.g. 2drive raid1 -> 2drive raid5
454 * ndrive raid5 -> degraded n+1drive raid6 with special layout
455 * If the takeover succeeds, a new 'private' structure is returned.
456 * This needs to be installed and then ->run used to activate the
457 * array.
459 void *(*takeover) (mddev_t *mddev);
463 struct md_sysfs_entry {
464 struct attribute attr;
465 ssize_t (*show)(mddev_t *, char *);
466 ssize_t (*store)(mddev_t *, const char *, size_t);
468 extern struct attribute_group md_bitmap_group;
470 static inline struct sysfs_dirent *sysfs_get_dirent_safe(struct sysfs_dirent *sd, char *name)
472 if (sd)
473 return sysfs_get_dirent(sd, NULL, name);
474 return sd;
476 static inline void sysfs_notify_dirent_safe(struct sysfs_dirent *sd)
478 if (sd)
479 sysfs_notify_dirent(sd);
482 static inline char * mdname (mddev_t * mddev)
484 return mddev->gendisk ? mddev->gendisk->disk_name : "mdX";
487 static inline int sysfs_link_rdev(mddev_t *mddev, mdk_rdev_t *rdev)
489 char nm[20];
490 sprintf(nm, "rd%d", rdev->raid_disk);
491 return sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
494 static inline void sysfs_unlink_rdev(mddev_t *mddev, mdk_rdev_t *rdev)
496 char nm[20];
497 sprintf(nm, "rd%d", rdev->raid_disk);
498 sysfs_remove_link(&mddev->kobj, nm);
502 * iterates through some rdev ringlist. It's safe to remove the
503 * current 'rdev'. Dont touch 'tmp' though.
505 #define rdev_for_each_list(rdev, tmp, head) \
506 list_for_each_entry_safe(rdev, tmp, head, same_set)
509 * iterates through the 'same array disks' ringlist
511 #define rdev_for_each(rdev, tmp, mddev) \
512 list_for_each_entry_safe(rdev, tmp, &((mddev)->disks), same_set)
514 #define rdev_for_each_rcu(rdev, mddev) \
515 list_for_each_entry_rcu(rdev, &((mddev)->disks), same_set)
517 typedef struct mdk_thread_s {
518 void (*run) (mddev_t *mddev);
519 mddev_t *mddev;
520 wait_queue_head_t wqueue;
521 unsigned long flags;
522 struct task_struct *tsk;
523 unsigned long timeout;
524 } mdk_thread_t;
526 #define THREAD_WAKEUP 0
528 #define __wait_event_lock_irq(wq, condition, lock, cmd) \
529 do { \
530 wait_queue_t __wait; \
531 init_waitqueue_entry(&__wait, current); \
533 add_wait_queue(&wq, &__wait); \
534 for (;;) { \
535 set_current_state(TASK_UNINTERRUPTIBLE); \
536 if (condition) \
537 break; \
538 spin_unlock_irq(&lock); \
539 cmd; \
540 schedule(); \
541 spin_lock_irq(&lock); \
543 current->state = TASK_RUNNING; \
544 remove_wait_queue(&wq, &__wait); \
545 } while (0)
547 #define wait_event_lock_irq(wq, condition, lock, cmd) \
548 do { \
549 if (condition) \
550 break; \
551 __wait_event_lock_irq(wq, condition, lock, cmd); \
552 } while (0)
554 static inline void safe_put_page(struct page *p)
556 if (p) put_page(p);
559 extern int register_md_personality(struct mdk_personality *p);
560 extern int unregister_md_personality(struct mdk_personality *p);
561 extern mdk_thread_t * md_register_thread(void (*run) (mddev_t *mddev),
562 mddev_t *mddev, const char *name);
563 extern void md_unregister_thread(mdk_thread_t *thread);
564 extern void md_wakeup_thread(mdk_thread_t *thread);
565 extern void md_check_recovery(mddev_t *mddev);
566 extern void md_write_start(mddev_t *mddev, struct bio *bi);
567 extern void md_write_end(mddev_t *mddev);
568 extern void md_done_sync(mddev_t *mddev, int blocks, int ok);
569 extern void md_error(mddev_t *mddev, mdk_rdev_t *rdev);
571 extern int mddev_congested(mddev_t *mddev, int bits);
572 extern void md_flush_request(mddev_t *mddev, struct bio *bio);
573 extern void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
574 sector_t sector, int size, struct page *page);
575 extern void md_super_wait(mddev_t *mddev);
576 extern int sync_page_io(mdk_rdev_t *rdev, sector_t sector, int size,
577 struct page *page, int rw, bool metadata_op);
578 extern void md_do_sync(mddev_t *mddev);
579 extern void md_new_event(mddev_t *mddev);
580 extern int md_allow_write(mddev_t *mddev);
581 extern void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev);
582 extern void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors);
583 extern int md_check_no_bitmap(mddev_t *mddev);
584 extern int md_integrity_register(mddev_t *mddev);
585 extern void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev);
586 extern int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale);
587 extern void restore_bitmap_write_access(struct file *file);
589 extern void mddev_init(mddev_t *mddev);
590 extern int md_run(mddev_t *mddev);
591 extern void md_stop(mddev_t *mddev);
592 extern void md_stop_writes(mddev_t *mddev);
593 extern int md_rdev_init(mdk_rdev_t *rdev);
595 extern void mddev_suspend(mddev_t *mddev);
596 extern void mddev_resume(mddev_t *mddev);
597 extern struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
598 mddev_t *mddev);
599 extern struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
600 mddev_t *mddev);
601 extern int mddev_check_plugged(mddev_t *mddev);
602 extern void md_trim_bio(struct bio *bio, int offset, int size);
603 #endif /* _MD_MD_H */