xref: /openbmc/linux/drivers/md/md.h (revision 65ee8aeb)
1 /*
2    md.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_MD_H
16 #define _MD_MD_H
17 
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>
26 #include "md-cluster.h"
27 
28 #define MaxSector (~(sector_t)0)
29 
30 /* Bad block numbers are stored sorted in a single page.
31  * 64bits is used for each block or extent.
32  * 54 bits are sector number, 9 bits are extent size,
33  * 1 bit is an 'acknowledged' flag.
34  */
35 #define MD_MAX_BADBLOCKS	(PAGE_SIZE/8)
36 
37 /*
38  * MD's 'extended' device
39  */
40 struct md_rdev {
41 	struct list_head same_set;	/* RAID devices within the same set */
42 
43 	sector_t sectors;		/* Device size (in 512bytes sectors) */
44 	struct mddev *mddev;		/* RAID array if running */
45 	int last_events;		/* IO event timestamp */
46 
47 	/*
48 	 * If meta_bdev is non-NULL, it means that a separate device is
49 	 * being used to store the metadata (superblock/bitmap) which
50 	 * would otherwise be contained on the same device as the data (bdev).
51 	 */
52 	struct block_device *meta_bdev;
53 	struct block_device *bdev;	/* block device handle */
54 
55 	struct page	*sb_page, *bb_page;
56 	int		sb_loaded;
57 	__u64		sb_events;
58 	sector_t	data_offset;	/* start of data in array */
59 	sector_t	new_data_offset;/* only relevant while reshaping */
60 	sector_t	sb_start;	/* offset of the super block (in 512byte sectors) */
61 	int		sb_size;	/* bytes in the superblock */
62 	int		preferred_minor;	/* autorun support */
63 
64 	struct kobject	kobj;
65 
66 	/* A device can be in one of three states based on two flags:
67 	 * Not working:   faulty==1 in_sync==0
68 	 * Fully working: faulty==0 in_sync==1
69 	 * Working, but not
70 	 * in sync with array
71 	 *                faulty==0 in_sync==0
72 	 *
73 	 * It can never have faulty==1, in_sync==1
74 	 * This reduces the burden of testing multiple flags in many cases
75 	 */
76 
77 	unsigned long	flags;	/* bit set of 'enum flag_bits' bits. */
78 	wait_queue_head_t blocked_wait;
79 
80 	int desc_nr;			/* descriptor index in the superblock */
81 	int raid_disk;			/* role of device in array */
82 	int new_raid_disk;		/* role that the device will have in
83 					 * the array after a level-change completes.
84 					 */
85 	int saved_raid_disk;		/* role that device used to have in the
86 					 * array and could again if we did a partial
87 					 * resync from the bitmap
88 					 */
89 	sector_t	recovery_offset;/* If this device has been partially
90 					 * recovered, this is where we were
91 					 * up to.
92 					 */
93 
94 	atomic_t	nr_pending;	/* number of pending requests.
95 					 * only maintained for arrays that
96 					 * support hot removal
97 					 */
98 	atomic_t	read_errors;	/* number of consecutive read errors that
99 					 * we have tried to ignore.
100 					 */
101 	struct timespec last_read_error;	/* monotonic time since our
102 						 * last read error
103 						 */
104 	atomic_t	corrected_errors; /* number of corrected read errors,
105 					   * for reporting to userspace and storing
106 					   * in superblock.
107 					   */
108 	struct work_struct del_work;	/* used for delayed sysfs removal */
109 
110 	struct kernfs_node *sysfs_state; /* handle for 'state'
111 					   * sysfs entry */
112 
113 	struct badblocks {
114 		int	count;		/* count of bad blocks */
115 		int	unacked_exist;	/* there probably are unacknowledged
116 					 * bad blocks.  This is only cleared
117 					 * when a read discovers none
118 					 */
119 		int	shift;		/* shift from sectors to block size
120 					 * a -ve shift means badblocks are
121 					 * disabled.*/
122 		u64	*page;		/* badblock list */
123 		int	changed;
124 		seqlock_t lock;
125 
126 		sector_t sector;
127 		sector_t size;		/* in sectors */
128 	} badblocks;
129 };
130 enum flag_bits {
131 	Faulty,			/* device is known to have a fault */
132 	In_sync,		/* device is in_sync with rest of array */
133 	Bitmap_sync,		/* ..actually, not quite In_sync.  Need a
134 				 * bitmap-based recovery to get fully in sync
135 				 */
136 	Unmerged,		/* device is being added to array and should
137 				 * be considerred for bvec_merge_fn but not
138 				 * yet for actual IO
139 				 */
140 	WriteMostly,		/* Avoid reading if at all possible */
141 	AutoDetected,		/* added by auto-detect */
142 	Blocked,		/* An error occurred but has not yet
143 				 * been acknowledged by the metadata
144 				 * handler, so don't allow writes
145 				 * until it is cleared */
146 	WriteErrorSeen,		/* A write error has been seen on this
147 				 * device
148 				 */
149 	FaultRecorded,		/* Intermediate state for clearing
150 				 * Blocked.  The Fault is/will-be
151 				 * recorded in the metadata, but that
152 				 * metadata hasn't been stored safely
153 				 * on disk yet.
154 				 */
155 	BlockedBadBlocks,	/* A writer is blocked because they
156 				 * found an unacknowledged bad-block.
157 				 * This can safely be cleared at any
158 				 * time, and the writer will re-check.
159 				 * It may be set at any time, and at
160 				 * worst the writer will timeout and
161 				 * re-check.  So setting it as
162 				 * accurately as possible is good, but
163 				 * not absolutely critical.
164 				 */
165 	WantReplacement,	/* This device is a candidate to be
166 				 * hot-replaced, either because it has
167 				 * reported some faults, or because
168 				 * of explicit request.
169 				 */
170 	Replacement,		/* This device is a replacement for
171 				 * a want_replacement device with same
172 				 * raid_disk number.
173 				 */
174 	Candidate,		/* For clustered environments only:
175 				 * This device is seen locally but not
176 				 * by the whole cluster
177 				 */
178 };
179 
180 #define BB_LEN_MASK	(0x00000000000001FFULL)
181 #define BB_OFFSET_MASK	(0x7FFFFFFFFFFFFE00ULL)
182 #define BB_ACK_MASK	(0x8000000000000000ULL)
183 #define BB_MAX_LEN	512
184 #define BB_OFFSET(x)	(((x) & BB_OFFSET_MASK) >> 9)
185 #define BB_LEN(x)	(((x) & BB_LEN_MASK) + 1)
186 #define BB_ACK(x)	(!!((x) & BB_ACK_MASK))
187 #define BB_MAKE(a, l, ack) (((a)<<9) | ((l)-1) | ((u64)(!!(ack)) << 63))
188 
189 extern int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
190 			  sector_t *first_bad, int *bad_sectors);
191 static inline int is_badblock(struct md_rdev *rdev, sector_t s, int sectors,
192 			      sector_t *first_bad, int *bad_sectors)
193 {
194 	if (unlikely(rdev->badblocks.count)) {
195 		int rv = md_is_badblock(&rdev->badblocks, rdev->data_offset + s,
196 					sectors,
197 					first_bad, bad_sectors);
198 		if (rv)
199 			*first_bad -= rdev->data_offset;
200 		return rv;
201 	}
202 	return 0;
203 }
204 extern int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
205 			      int is_new);
206 extern int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
207 				int is_new);
208 extern void md_ack_all_badblocks(struct badblocks *bb);
209 
210 struct md_cluster_info;
211 
212 struct mddev {
213 	void				*private;
214 	struct md_personality		*pers;
215 	dev_t				unit;
216 	int				md_minor;
217 	struct list_head		disks;
218 	unsigned long			flags;
219 #define MD_CHANGE_DEVS	0	/* Some device status has changed */
220 #define MD_CHANGE_CLEAN 1	/* transition to or from 'clean' */
221 #define MD_CHANGE_PENDING 2	/* switch from 'clean' to 'active' in progress */
222 #define MD_UPDATE_SB_FLAGS (1 | 2 | 4)	/* If these are set, md_update_sb needed */
223 #define MD_ARRAY_FIRST_USE 3    /* First use of array, needs initialization */
224 #define MD_STILL_CLOSED	4	/* If set, then array has not been opened since
225 				 * md_ioctl checked on it.
226 				 */
227 
228 	int				suspended;
229 	atomic_t			active_io;
230 	int				ro;
231 	int				sysfs_active; /* set when sysfs deletes
232 						       * are happening, so run/
233 						       * takeover/stop are not safe
234 						       */
235 	int				ready; /* See when safe to pass
236 						* IO requests down */
237 	struct gendisk			*gendisk;
238 
239 	struct kobject			kobj;
240 	int				hold_active;
241 #define	UNTIL_IOCTL	1
242 #define	UNTIL_STOP	2
243 
244 	/* Superblock information */
245 	int				major_version,
246 					minor_version,
247 					patch_version;
248 	int				persistent;
249 	int				external;	/* metadata is
250 							 * managed externally */
251 	char				metadata_type[17]; /* externally set*/
252 	int				chunk_sectors;
253 	time_t				ctime, utime;
254 	int				level, layout;
255 	char				clevel[16];
256 	int				raid_disks;
257 	int				max_disks;
258 	sector_t			dev_sectors;	/* used size of
259 							 * component devices */
260 	sector_t			array_sectors; /* exported array size */
261 	int				external_size; /* size managed
262 							* externally */
263 	__u64				events;
264 	/* If the last 'event' was simply a clean->dirty transition, and
265 	 * we didn't write it to the spares, then it is safe and simple
266 	 * to just decrement the event count on a dirty->clean transition.
267 	 * So we record that possibility here.
268 	 */
269 	int				can_decrease_events;
270 
271 	char				uuid[16];
272 
273 	/* If the array is being reshaped, we need to record the
274 	 * new shape and an indication of where we are up to.
275 	 * This is written to the superblock.
276 	 * If reshape_position is MaxSector, then no reshape is happening (yet).
277 	 */
278 	sector_t			reshape_position;
279 	int				delta_disks, new_level, new_layout;
280 	int				new_chunk_sectors;
281 	int				reshape_backwards;
282 
283 	struct md_thread		*thread;	/* management thread */
284 	struct md_thread		*sync_thread;	/* doing resync or reconstruct */
285 
286 	/* 'last_sync_action' is initialized to "none".  It is set when a
287 	 * sync operation (i.e "data-check", "requested-resync", "resync",
288 	 * "recovery", or "reshape") is started.  It holds this value even
289 	 * when the sync thread is "frozen" (interrupted) or "idle" (stopped
290 	 * or finished).  It is overwritten when a new sync operation is begun.
291 	 */
292 	char				*last_sync_action;
293 	sector_t			curr_resync;	/* last block scheduled */
294 	/* As resync requests can complete out of order, we cannot easily track
295 	 * how much resync has been completed.  So we occasionally pause until
296 	 * everything completes, then set curr_resync_completed to curr_resync.
297 	 * As such it may be well behind the real resync mark, but it is a value
298 	 * we are certain of.
299 	 */
300 	sector_t			curr_resync_completed;
301 	unsigned long			resync_mark;	/* a recent timestamp */
302 	sector_t			resync_mark_cnt;/* blocks written at resync_mark */
303 	sector_t			curr_mark_cnt; /* blocks scheduled now */
304 
305 	sector_t			resync_max_sectors; /* may be set by personality */
306 
307 	atomic64_t			resync_mismatches; /* count of sectors where
308 							    * parity/replica mismatch found
309 							    */
310 
311 	/* allow user-space to request suspension of IO to regions of the array */
312 	sector_t			suspend_lo;
313 	sector_t			suspend_hi;
314 	/* if zero, use the system-wide default */
315 	int				sync_speed_min;
316 	int				sync_speed_max;
317 
318 	/* resync even though the same disks are shared among md-devices */
319 	int				parallel_resync;
320 
321 	int				ok_start_degraded;
322 	/* recovery/resync flags
323 	 * NEEDED:   we might need to start a resync/recover
324 	 * RUNNING:  a thread is running, or about to be started
325 	 * SYNC:     actually doing a resync, not a recovery
326 	 * RECOVER:  doing recovery, or need to try it.
327 	 * INTR:     resync needs to be aborted for some reason
328 	 * DONE:     thread is done and is waiting to be reaped
329 	 * REQUEST:  user-space has requested a sync (used with SYNC)
330 	 * CHECK:    user-space request for check-only, no repair
331 	 * RESHAPE:  A reshape is happening
332 	 * ERROR:    sync-action interrupted because io-error
333 	 *
334 	 * If neither SYNC or RESHAPE are set, then it is a recovery.
335 	 */
336 #define	MD_RECOVERY_RUNNING	0
337 #define	MD_RECOVERY_SYNC	1
338 #define	MD_RECOVERY_RECOVER	2
339 #define	MD_RECOVERY_INTR	3
340 #define	MD_RECOVERY_DONE	4
341 #define	MD_RECOVERY_NEEDED	5
342 #define	MD_RECOVERY_REQUESTED	6
343 #define	MD_RECOVERY_CHECK	7
344 #define MD_RECOVERY_RESHAPE	8
345 #define	MD_RECOVERY_FROZEN	9
346 #define	MD_RECOVERY_ERROR	10
347 
348 	unsigned long			recovery;
349 	/* If a RAID personality determines that recovery (of a particular
350 	 * device) will fail due to a read error on the source device, it
351 	 * takes a copy of this number and does not attempt recovery again
352 	 * until this number changes.
353 	 */
354 	int				recovery_disabled;
355 
356 	int				in_sync;	/* know to not need resync */
357 	/* 'open_mutex' avoids races between 'md_open' and 'do_md_stop', so
358 	 * that we are never stopping an array while it is open.
359 	 * 'reconfig_mutex' protects all other reconfiguration.
360 	 * These locks are separate due to conflicting interactions
361 	 * with bdev->bd_mutex.
362 	 * Lock ordering is:
363 	 *  reconfig_mutex -> bd_mutex : e.g. do_md_run -> revalidate_disk
364 	 *  bd_mutex -> open_mutex:  e.g. __blkdev_get -> md_open
365 	 */
366 	struct mutex			open_mutex;
367 	struct mutex			reconfig_mutex;
368 	atomic_t			active;		/* general refcount */
369 	atomic_t			openers;	/* number of active opens */
370 
371 	int				changed;	/* True if we might need to
372 							 * reread partition info */
373 	int				degraded;	/* whether md should consider
374 							 * adding a spare
375 							 */
376 	int				merge_check_needed; /* at least one
377 							     * member device
378 							     * has a
379 							     * merge_bvec_fn */
380 
381 	atomic_t			recovery_active; /* blocks scheduled, but not written */
382 	wait_queue_head_t		recovery_wait;
383 	sector_t			recovery_cp;
384 	sector_t			resync_min;	/* user requested sync
385 							 * starts here */
386 	sector_t			resync_max;	/* resync should pause
387 							 * when it gets here */
388 
389 	struct kernfs_node		*sysfs_state;	/* handle for 'array_state'
390 							 * file in sysfs.
391 							 */
392 	struct kernfs_node		*sysfs_action;  /* handle for 'sync_action' */
393 
394 	struct work_struct del_work;	/* used for delayed sysfs removal */
395 
396 	/* "lock" protects:
397 	 *   flush_bio transition from NULL to !NULL
398 	 *   rdev superblocks, events
399 	 *   clearing MD_CHANGE_*
400 	 *   in_sync - and related safemode and MD_CHANGE changes
401 	 *   pers (also protected by reconfig_mutex and pending IO).
402 	 *   clearing ->bitmap
403 	 *   clearing ->bitmap_info.file
404 	 *   changing ->resync_{min,max}
405 	 *   setting MD_RECOVERY_RUNNING (which interacts with resync_{min,max})
406 	 */
407 	spinlock_t			lock;
408 	wait_queue_head_t		sb_wait;	/* for waiting on superblock updates */
409 	atomic_t			pending_writes;	/* number of active superblock writes */
410 
411 	unsigned int			safemode;	/* if set, update "clean" superblock
412 							 * when no writes pending.
413 							 */
414 	unsigned int			safemode_delay;
415 	struct timer_list		safemode_timer;
416 	atomic_t			writes_pending;
417 	struct request_queue		*queue;	/* for plugging ... */
418 
419 	struct bitmap			*bitmap; /* the bitmap for the device */
420 	struct {
421 		struct file		*file; /* the bitmap file */
422 		loff_t			offset; /* offset from superblock of
423 						 * start of bitmap. May be
424 						 * negative, but not '0'
425 						 * For external metadata, offset
426 						 * from start of device.
427 						 */
428 		unsigned long		space; /* space available at this offset */
429 		loff_t			default_offset; /* this is the offset to use when
430 							 * hot-adding a bitmap.  It should
431 							 * eventually be settable by sysfs.
432 							 */
433 		unsigned long		default_space; /* space available at
434 							* default offset */
435 		struct mutex		mutex;
436 		unsigned long		chunksize;
437 		unsigned long		daemon_sleep; /* how many jiffies between updates? */
438 		unsigned long		max_write_behind; /* write-behind mode */
439 		int			external;
440 		int			nodes; /* Maximum number of nodes in the cluster */
441 		char                    cluster_name[64]; /* Name of the cluster */
442 	} bitmap_info;
443 
444 	atomic_t			max_corr_read_errors; /* max read retries */
445 	struct list_head		all_mddevs;
446 
447 	struct attribute_group		*to_remove;
448 
449 	struct bio_set			*bio_set;
450 
451 	/* Generic flush handling.
452 	 * The last to finish preflush schedules a worker to submit
453 	 * the rest of the request (without the REQ_FLUSH flag).
454 	 */
455 	struct bio *flush_bio;
456 	atomic_t flush_pending;
457 	struct work_struct flush_work;
458 	struct work_struct event_work;	/* used by dm to report failure event */
459 	void (*sync_super)(struct mddev *mddev, struct md_rdev *rdev);
460 	struct md_cluster_info		*cluster_info;
461 };
462 
463 static inline int __must_check mddev_lock(struct mddev *mddev)
464 {
465 	return mutex_lock_interruptible(&mddev->reconfig_mutex);
466 }
467 
468 /* Sometimes we need to take the lock in a situation where
469  * failure due to interrupts is not acceptable.
470  */
471 static inline void mddev_lock_nointr(struct mddev *mddev)
472 {
473 	mutex_lock(&mddev->reconfig_mutex);
474 }
475 
476 static inline int mddev_is_locked(struct mddev *mddev)
477 {
478 	return mutex_is_locked(&mddev->reconfig_mutex);
479 }
480 
481 static inline int mddev_trylock(struct mddev *mddev)
482 {
483 	return mutex_trylock(&mddev->reconfig_mutex);
484 }
485 extern void mddev_unlock(struct mddev *mddev);
486 
487 static inline void md_sync_acct(struct block_device *bdev, unsigned long nr_sectors)
488 {
489 	atomic_add(nr_sectors, &bdev->bd_contains->bd_disk->sync_io);
490 }
491 
492 struct md_personality
493 {
494 	char *name;
495 	int level;
496 	struct list_head list;
497 	struct module *owner;
498 	void (*make_request)(struct mddev *mddev, struct bio *bio);
499 	int (*run)(struct mddev *mddev);
500 	void (*free)(struct mddev *mddev, void *priv);
501 	void (*status)(struct seq_file *seq, struct mddev *mddev);
502 	/* error_handler must set ->faulty and clear ->in_sync
503 	 * if appropriate, and should abort recovery if needed
504 	 */
505 	void (*error_handler)(struct mddev *mddev, struct md_rdev *rdev);
506 	int (*hot_add_disk) (struct mddev *mddev, struct md_rdev *rdev);
507 	int (*hot_remove_disk) (struct mddev *mddev, struct md_rdev *rdev);
508 	int (*spare_active) (struct mddev *mddev);
509 	sector_t (*sync_request)(struct mddev *mddev, sector_t sector_nr, int *skipped);
510 	int (*resize) (struct mddev *mddev, sector_t sectors);
511 	sector_t (*size) (struct mddev *mddev, sector_t sectors, int raid_disks);
512 	int (*check_reshape) (struct mddev *mddev);
513 	int (*start_reshape) (struct mddev *mddev);
514 	void (*finish_reshape) (struct mddev *mddev);
515 	/* quiesce moves between quiescence states
516 	 * 0 - fully active
517 	 * 1 - no new requests allowed
518 	 * others - reserved
519 	 */
520 	void (*quiesce) (struct mddev *mddev, int state);
521 	/* takeover is used to transition an array from one
522 	 * personality to another.  The new personality must be able
523 	 * to handle the data in the current layout.
524 	 * e.g. 2drive raid1 -> 2drive raid5
525 	 *      ndrive raid5 -> degraded n+1drive raid6 with special layout
526 	 * If the takeover succeeds, a new 'private' structure is returned.
527 	 * This needs to be installed and then ->run used to activate the
528 	 * array.
529 	 */
530 	void *(*takeover) (struct mddev *mddev);
531 	/* congested implements bdi.congested_fn().
532 	 * Will not be called while array is 'suspended' */
533 	int (*congested)(struct mddev *mddev, int bits);
534 	/* mergeable_bvec is use to implement ->merge_bvec_fn */
535 	int (*mergeable_bvec)(struct mddev *mddev,
536 			      struct bvec_merge_data *bvm,
537 			      struct bio_vec *biovec);
538 };
539 
540 struct md_sysfs_entry {
541 	struct attribute attr;
542 	ssize_t (*show)(struct mddev *, char *);
543 	ssize_t (*store)(struct mddev *, const char *, size_t);
544 };
545 extern struct attribute_group md_bitmap_group;
546 
547 static inline struct kernfs_node *sysfs_get_dirent_safe(struct kernfs_node *sd, char *name)
548 {
549 	if (sd)
550 		return sysfs_get_dirent(sd, name);
551 	return sd;
552 }
553 static inline void sysfs_notify_dirent_safe(struct kernfs_node *sd)
554 {
555 	if (sd)
556 		sysfs_notify_dirent(sd);
557 }
558 
559 static inline char * mdname (struct mddev * mddev)
560 {
561 	return mddev->gendisk ? mddev->gendisk->disk_name : "mdX";
562 }
563 
564 static inline int sysfs_link_rdev(struct mddev *mddev, struct md_rdev *rdev)
565 {
566 	char nm[20];
567 	if (!test_bit(Replacement, &rdev->flags) && mddev->kobj.sd) {
568 		sprintf(nm, "rd%d", rdev->raid_disk);
569 		return sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
570 	} else
571 		return 0;
572 }
573 
574 static inline void sysfs_unlink_rdev(struct mddev *mddev, struct md_rdev *rdev)
575 {
576 	char nm[20];
577 	if (!test_bit(Replacement, &rdev->flags) && mddev->kobj.sd) {
578 		sprintf(nm, "rd%d", rdev->raid_disk);
579 		sysfs_remove_link(&mddev->kobj, nm);
580 	}
581 }
582 
583 /*
584  * iterates through some rdev ringlist. It's safe to remove the
585  * current 'rdev'. Dont touch 'tmp' though.
586  */
587 #define rdev_for_each_list(rdev, tmp, head)				\
588 	list_for_each_entry_safe(rdev, tmp, head, same_set)
589 
590 /*
591  * iterates through the 'same array disks' ringlist
592  */
593 #define rdev_for_each(rdev, mddev)				\
594 	list_for_each_entry(rdev, &((mddev)->disks), same_set)
595 
596 #define rdev_for_each_safe(rdev, tmp, mddev)				\
597 	list_for_each_entry_safe(rdev, tmp, &((mddev)->disks), same_set)
598 
599 #define rdev_for_each_rcu(rdev, mddev)				\
600 	list_for_each_entry_rcu(rdev, &((mddev)->disks), same_set)
601 
602 struct md_thread {
603 	void			(*run) (struct md_thread *thread);
604 	struct mddev		*mddev;
605 	wait_queue_head_t	wqueue;
606 	unsigned long		flags;
607 	struct task_struct	*tsk;
608 	unsigned long		timeout;
609 	void			*private;
610 };
611 
612 #define THREAD_WAKEUP  0
613 
614 static inline void safe_put_page(struct page *p)
615 {
616 	if (p) put_page(p);
617 }
618 
619 extern int register_md_personality(struct md_personality *p);
620 extern int unregister_md_personality(struct md_personality *p);
621 extern int register_md_cluster_operations(struct md_cluster_operations *ops,
622 		struct module *module);
623 extern int unregister_md_cluster_operations(void);
624 extern int md_setup_cluster(struct mddev *mddev, int nodes);
625 extern void md_cluster_stop(struct mddev *mddev);
626 extern struct md_thread *md_register_thread(
627 	void (*run)(struct md_thread *thread),
628 	struct mddev *mddev,
629 	const char *name);
630 extern void md_unregister_thread(struct md_thread **threadp);
631 extern void md_wakeup_thread(struct md_thread *thread);
632 extern void md_check_recovery(struct mddev *mddev);
633 extern void md_reap_sync_thread(struct mddev *mddev);
634 extern void md_write_start(struct mddev *mddev, struct bio *bi);
635 extern void md_write_end(struct mddev *mddev);
636 extern void md_done_sync(struct mddev *mddev, int blocks, int ok);
637 extern void md_error(struct mddev *mddev, struct md_rdev *rdev);
638 extern void md_finish_reshape(struct mddev *mddev);
639 
640 extern int mddev_congested(struct mddev *mddev, int bits);
641 extern void md_flush_request(struct mddev *mddev, struct bio *bio);
642 extern void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
643 			   sector_t sector, int size, struct page *page);
644 extern void md_super_wait(struct mddev *mddev);
645 extern int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
646 			struct page *page, int rw, bool metadata_op);
647 extern void md_do_sync(struct md_thread *thread);
648 extern void md_new_event(struct mddev *mddev);
649 extern int md_allow_write(struct mddev *mddev);
650 extern void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev);
651 extern void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors);
652 extern int md_check_no_bitmap(struct mddev *mddev);
653 extern int md_integrity_register(struct mddev *mddev);
654 extern void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev);
655 extern int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale);
656 
657 extern void mddev_init(struct mddev *mddev);
658 extern int md_run(struct mddev *mddev);
659 extern void md_stop(struct mddev *mddev);
660 extern void md_stop_writes(struct mddev *mddev);
661 extern int md_rdev_init(struct md_rdev *rdev);
662 extern void md_rdev_clear(struct md_rdev *rdev);
663 
664 extern void mddev_suspend(struct mddev *mddev);
665 extern void mddev_resume(struct mddev *mddev);
666 extern struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
667 				   struct mddev *mddev);
668 extern struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
669 				   struct mddev *mddev);
670 
671 extern void md_unplug(struct blk_plug_cb *cb, bool from_schedule);
672 extern void md_reload_sb(struct mddev *mddev);
673 extern void md_update_sb(struct mddev *mddev, int force);
674 extern void md_kick_rdev_from_array(struct md_rdev * rdev);
675 struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr);
676 static inline int mddev_check_plugged(struct mddev *mddev)
677 {
678 	return !!blk_check_plugged(md_unplug, mddev,
679 				   sizeof(struct blk_plug_cb));
680 }
681 
682 static inline void rdev_dec_pending(struct md_rdev *rdev, struct mddev *mddev)
683 {
684 	int faulty = test_bit(Faulty, &rdev->flags);
685 	if (atomic_dec_and_test(&rdev->nr_pending) && faulty) {
686 		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
687 		md_wakeup_thread(mddev->thread);
688 	}
689 }
690 
691 extern struct md_cluster_operations *md_cluster_ops;
692 static inline int mddev_is_clustered(struct mddev *mddev)
693 {
694 	return mddev->cluster_info && mddev->bitmap_info.nodes > 1;
695 }
696 #endif /* _MD_MD_H */
697