xref: /linux/drivers/md/md.h (revision 95f28190aa012b18eab14799b905b6db3cf31529)
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/backing-dev.h>
20 #include <linux/badblocks.h>
21 #include <linux/kobject.h>
22 #include <linux/list.h>
23 #include <linux/mm.h>
24 #include <linux/mutex.h>
25 #include <linux/timer.h>
26 #include <linux/wait.h>
27 #include <linux/workqueue.h>
28 #include "md-cluster.h"
29 
30 #define MaxSector (~(sector_t)0)
31 
32 /*
33  * These flags should really be called "NO_RETRY" rather than
34  * "FAILFAST" because they don't make any promise about time lapse,
35  * only about the number of retries, which will be zero.
36  * REQ_FAILFAST_DRIVER is not included because
37  * Commit: 4a27446f3e39 ("[SCSI] modify scsi to handle new fail fast flags.")
38  * seems to suggest that the errors it avoids retrying should usually
39  * be retried.
40  */
41 #define	MD_FAILFAST	(REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT)
42 /*
43  * MD's 'extended' device
44  */
45 struct md_rdev {
46 	struct list_head same_set;	/* RAID devices within the same set */
47 
48 	sector_t sectors;		/* Device size (in 512bytes sectors) */
49 	struct mddev *mddev;		/* RAID array if running */
50 	int last_events;		/* IO event timestamp */
51 
52 	/*
53 	 * If meta_bdev is non-NULL, it means that a separate device is
54 	 * being used to store the metadata (superblock/bitmap) which
55 	 * would otherwise be contained on the same device as the data (bdev).
56 	 */
57 	struct block_device *meta_bdev;
58 	struct block_device *bdev;	/* block device handle */
59 
60 	struct page	*sb_page, *bb_page;
61 	int		sb_loaded;
62 	__u64		sb_events;
63 	sector_t	data_offset;	/* start of data in array */
64 	sector_t	new_data_offset;/* only relevant while reshaping */
65 	sector_t	sb_start;	/* offset of the super block (in 512byte sectors) */
66 	int		sb_size;	/* bytes in the superblock */
67 	int		preferred_minor;	/* autorun support */
68 
69 	struct kobject	kobj;
70 
71 	/* A device can be in one of three states based on two flags:
72 	 * Not working:   faulty==1 in_sync==0
73 	 * Fully working: faulty==0 in_sync==1
74 	 * Working, but not
75 	 * in sync with array
76 	 *                faulty==0 in_sync==0
77 	 *
78 	 * It can never have faulty==1, in_sync==1
79 	 * This reduces the burden of testing multiple flags in many cases
80 	 */
81 
82 	unsigned long	flags;	/* bit set of 'enum flag_bits' bits. */
83 	wait_queue_head_t blocked_wait;
84 
85 	int desc_nr;			/* descriptor index in the superblock */
86 	int raid_disk;			/* role of device in array */
87 	int new_raid_disk;		/* role that the device will have in
88 					 * the array after a level-change completes.
89 					 */
90 	int saved_raid_disk;		/* role that device used to have in the
91 					 * array and could again if we did a partial
92 					 * resync from the bitmap
93 					 */
94 	union {
95 		sector_t recovery_offset;/* If this device has been partially
96 					 * recovered, this is where we were
97 					 * up to.
98 					 */
99 		sector_t journal_tail;	/* If this device is a journal device,
100 					 * this is the journal tail (journal
101 					 * recovery start point)
102 					 */
103 	};
104 
105 	atomic_t	nr_pending;	/* number of pending requests.
106 					 * only maintained for arrays that
107 					 * support hot removal
108 					 */
109 	atomic_t	read_errors;	/* number of consecutive read errors that
110 					 * we have tried to ignore.
111 					 */
112 	time64_t	last_read_error;	/* monotonic time since our
113 						 * last read error
114 						 */
115 	atomic_t	corrected_errors; /* number of corrected read errors,
116 					   * for reporting to userspace and storing
117 					   * in superblock.
118 					   */
119 	struct work_struct del_work;	/* used for delayed sysfs removal */
120 
121 	struct kernfs_node *sysfs_state; /* handle for 'state'
122 					   * sysfs entry */
123 
124 	struct badblocks badblocks;
125 
126 	struct {
127 		short offset;	/* Offset from superblock to start of PPL.
128 				 * Not used by external metadata. */
129 		unsigned int size;	/* Size in sectors of the PPL space */
130 		sector_t sector;	/* First sector of the PPL space */
131 	} ppl;
132 };
133 enum flag_bits {
134 	Faulty,			/* device is known to have a fault */
135 	In_sync,		/* device is in_sync with rest of array */
136 	Bitmap_sync,		/* ..actually, not quite In_sync.  Need a
137 				 * bitmap-based recovery to get fully in sync.
138 				 * The bit is only meaningful before device
139 				 * has been passed to pers->hot_add_disk.
140 				 */
141 	WriteMostly,		/* Avoid reading if at all possible */
142 	AutoDetected,		/* added by auto-detect */
143 	Blocked,		/* An error occurred but has not yet
144 				 * been acknowledged by the metadata
145 				 * handler, so don't allow writes
146 				 * until it is cleared */
147 	WriteErrorSeen,		/* A write error has been seen on this
148 				 * device
149 				 */
150 	FaultRecorded,		/* Intermediate state for clearing
151 				 * Blocked.  The Fault is/will-be
152 				 * recorded in the metadata, but that
153 				 * metadata hasn't been stored safely
154 				 * on disk yet.
155 				 */
156 	BlockedBadBlocks,	/* A writer is blocked because they
157 				 * found an unacknowledged bad-block.
158 				 * This can safely be cleared at any
159 				 * time, and the writer will re-check.
160 				 * It may be set at any time, and at
161 				 * worst the writer will timeout and
162 				 * re-check.  So setting it as
163 				 * accurately as possible is good, but
164 				 * not absolutely critical.
165 				 */
166 	WantReplacement,	/* This device is a candidate to be
167 				 * hot-replaced, either because it has
168 				 * reported some faults, or because
169 				 * of explicit request.
170 				 */
171 	Replacement,		/* This device is a replacement for
172 				 * a want_replacement device with same
173 				 * raid_disk number.
174 				 */
175 	Candidate,		/* For clustered environments only:
176 				 * This device is seen locally but not
177 				 * by the whole cluster
178 				 */
179 	Journal,		/* This device is used as journal for
180 				 * raid-5/6.
181 				 * Usually, this device should be faster
182 				 * than other devices in the array
183 				 */
184 	ClusterRemove,
185 	RemoveSynchronized,	/* synchronize_rcu() was called after
186 				 * this device was known to be faulty,
187 				 * so it is safe to remove without
188 				 * another synchronize_rcu() call.
189 				 */
190 	ExternalBbl,            /* External metadata provides bad
191 				 * block management for a disk
192 				 */
193 	FailFast,		/* Minimal retries should be attempted on
194 				 * this device, so use REQ_FAILFAST_DEV.
195 				 * Also don't try to repair failed reads.
196 				 * It is expects that no bad block log
197 				 * is present.
198 				 */
199 	LastDev,		/* Seems to be the last working dev as
200 				 * it didn't fail, so don't use FailFast
201 				 * any more for metadata
202 				 */
203 };
204 
205 static inline int is_badblock(struct md_rdev *rdev, sector_t s, int sectors,
206 			      sector_t *first_bad, int *bad_sectors)
207 {
208 	if (unlikely(rdev->badblocks.count)) {
209 		int rv = badblocks_check(&rdev->badblocks, rdev->data_offset + s,
210 					sectors,
211 					first_bad, bad_sectors);
212 		if (rv)
213 			*first_bad -= rdev->data_offset;
214 		return rv;
215 	}
216 	return 0;
217 }
218 extern int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
219 			      int is_new);
220 extern int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
221 				int is_new);
222 struct md_cluster_info;
223 
224 /* change UNSUPPORTED_MDDEV_FLAGS for each array type if new flag is added */
225 enum mddev_flags {
226 	MD_ARRAY_FIRST_USE,	/* First use of array, needs initialization */
227 	MD_CLOSING,		/* If set, we are closing the array, do not open
228 				 * it then */
229 	MD_JOURNAL_CLEAN,	/* A raid with journal is already clean */
230 	MD_HAS_JOURNAL,		/* The raid array has journal feature set */
231 	MD_CLUSTER_RESYNC_LOCKED, /* cluster raid only, which means node
232 				   * already took resync lock, need to
233 				   * release the lock */
234 	MD_FAILFAST_SUPPORTED,	/* Using MD_FAILFAST on metadata writes is
235 				 * supported as calls to md_error() will
236 				 * never cause the array to become failed.
237 				 */
238 	MD_HAS_PPL,		/* The raid array has PPL feature set */
239 	MD_HAS_MULTIPLE_PPLS,	/* The raid array has multiple PPLs feature set */
240 	MD_ALLOW_SB_UPDATE,	/* md_check_recovery is allowed to update
241 				 * the metadata without taking reconfig_mutex.
242 				 */
243 	MD_UPDATING_SB,		/* md_check_recovery is updating the metadata
244 				 * without explicitly holding reconfig_mutex.
245 				 */
246 };
247 
248 enum mddev_sb_flags {
249 	MD_SB_CHANGE_DEVS,		/* Some device status has changed */
250 	MD_SB_CHANGE_CLEAN,	/* transition to or from 'clean' */
251 	MD_SB_CHANGE_PENDING,	/* switch from 'clean' to 'active' in progress */
252 	MD_SB_NEED_REWRITE,	/* metadata write needs to be repeated */
253 };
254 
255 struct mddev {
256 	void				*private;
257 	struct md_personality		*pers;
258 	dev_t				unit;
259 	int				md_minor;
260 	struct list_head		disks;
261 	unsigned long			flags;
262 	unsigned long			sb_flags;
263 
264 	int				suspended;
265 	atomic_t			active_io;
266 	int				ro;
267 	int				sysfs_active; /* set when sysfs deletes
268 						       * are happening, so run/
269 						       * takeover/stop are not safe
270 						       */
271 	struct gendisk			*gendisk;
272 
273 	struct kobject			kobj;
274 	int				hold_active;
275 #define	UNTIL_IOCTL	1
276 #define	UNTIL_STOP	2
277 
278 	/* Superblock information */
279 	int				major_version,
280 					minor_version,
281 					patch_version;
282 	int				persistent;
283 	int				external;	/* metadata is
284 							 * managed externally */
285 	char				metadata_type[17]; /* externally set*/
286 	int				chunk_sectors;
287 	time64_t			ctime, utime;
288 	int				level, layout;
289 	char				clevel[16];
290 	int				raid_disks;
291 	int				max_disks;
292 	sector_t			dev_sectors;	/* used size of
293 							 * component devices */
294 	sector_t			array_sectors; /* exported array size */
295 	int				external_size; /* size managed
296 							* externally */
297 	__u64				events;
298 	/* If the last 'event' was simply a clean->dirty transition, and
299 	 * we didn't write it to the spares, then it is safe and simple
300 	 * to just decrement the event count on a dirty->clean transition.
301 	 * So we record that possibility here.
302 	 */
303 	int				can_decrease_events;
304 
305 	char				uuid[16];
306 
307 	/* If the array is being reshaped, we need to record the
308 	 * new shape and an indication of where we are up to.
309 	 * This is written to the superblock.
310 	 * If reshape_position is MaxSector, then no reshape is happening (yet).
311 	 */
312 	sector_t			reshape_position;
313 	int				delta_disks, new_level, new_layout;
314 	int				new_chunk_sectors;
315 	int				reshape_backwards;
316 
317 	struct md_thread		*thread;	/* management thread */
318 	struct md_thread		*sync_thread;	/* doing resync or reconstruct */
319 
320 	/* 'last_sync_action' is initialized to "none".  It is set when a
321 	 * sync operation (i.e "data-check", "requested-resync", "resync",
322 	 * "recovery", or "reshape") is started.  It holds this value even
323 	 * when the sync thread is "frozen" (interrupted) or "idle" (stopped
324 	 * or finished).  It is overwritten when a new sync operation is begun.
325 	 */
326 	char				*last_sync_action;
327 	sector_t			curr_resync;	/* last block scheduled */
328 	/* As resync requests can complete out of order, we cannot easily track
329 	 * how much resync has been completed.  So we occasionally pause until
330 	 * everything completes, then set curr_resync_completed to curr_resync.
331 	 * As such it may be well behind the real resync mark, but it is a value
332 	 * we are certain of.
333 	 */
334 	sector_t			curr_resync_completed;
335 	unsigned long			resync_mark;	/* a recent timestamp */
336 	sector_t			resync_mark_cnt;/* blocks written at resync_mark */
337 	sector_t			curr_mark_cnt; /* blocks scheduled now */
338 
339 	sector_t			resync_max_sectors; /* may be set by personality */
340 
341 	atomic64_t			resync_mismatches; /* count of sectors where
342 							    * parity/replica mismatch found
343 							    */
344 
345 	/* allow user-space to request suspension of IO to regions of the array */
346 	sector_t			suspend_lo;
347 	sector_t			suspend_hi;
348 	/* if zero, use the system-wide default */
349 	int				sync_speed_min;
350 	int				sync_speed_max;
351 
352 	/* resync even though the same disks are shared among md-devices */
353 	int				parallel_resync;
354 
355 	int				ok_start_degraded;
356 
357 	unsigned long			recovery;
358 	/* If a RAID personality determines that recovery (of a particular
359 	 * device) will fail due to a read error on the source device, it
360 	 * takes a copy of this number and does not attempt recovery again
361 	 * until this number changes.
362 	 */
363 	int				recovery_disabled;
364 
365 	int				in_sync;	/* know to not need resync */
366 	/* 'open_mutex' avoids races between 'md_open' and 'do_md_stop', so
367 	 * that we are never stopping an array while it is open.
368 	 * 'reconfig_mutex' protects all other reconfiguration.
369 	 * These locks are separate due to conflicting interactions
370 	 * with bdev->bd_mutex.
371 	 * Lock ordering is:
372 	 *  reconfig_mutex -> bd_mutex : e.g. do_md_run -> revalidate_disk
373 	 *  bd_mutex -> open_mutex:  e.g. __blkdev_get -> md_open
374 	 */
375 	struct mutex			open_mutex;
376 	struct mutex			reconfig_mutex;
377 	atomic_t			active;		/* general refcount */
378 	atomic_t			openers;	/* number of active opens */
379 
380 	int				changed;	/* True if we might need to
381 							 * reread partition info */
382 	int				degraded;	/* whether md should consider
383 							 * adding a spare
384 							 */
385 
386 	atomic_t			recovery_active; /* blocks scheduled, but not written */
387 	wait_queue_head_t		recovery_wait;
388 	sector_t			recovery_cp;
389 	sector_t			resync_min;	/* user requested sync
390 							 * starts here */
391 	sector_t			resync_max;	/* resync should pause
392 							 * when it gets here */
393 
394 	struct kernfs_node		*sysfs_state;	/* handle for 'array_state'
395 							 * file in sysfs.
396 							 */
397 	struct kernfs_node		*sysfs_action;  /* handle for 'sync_action' */
398 
399 	struct work_struct del_work;	/* used for delayed sysfs removal */
400 
401 	/* "lock" protects:
402 	 *   flush_bio transition from NULL to !NULL
403 	 *   rdev superblocks, events
404 	 *   clearing MD_CHANGE_*
405 	 *   in_sync - and related safemode and MD_CHANGE changes
406 	 *   pers (also protected by reconfig_mutex and pending IO).
407 	 *   clearing ->bitmap
408 	 *   clearing ->bitmap_info.file
409 	 *   changing ->resync_{min,max}
410 	 *   setting MD_RECOVERY_RUNNING (which interacts with resync_{min,max})
411 	 */
412 	spinlock_t			lock;
413 	wait_queue_head_t		sb_wait;	/* for waiting on superblock updates */
414 	atomic_t			pending_writes;	/* number of active superblock writes */
415 
416 	unsigned int			safemode;	/* if set, update "clean" superblock
417 							 * when no writes pending.
418 							 */
419 	unsigned int			safemode_delay;
420 	struct timer_list		safemode_timer;
421 	struct percpu_ref		writes_pending;
422 	int				sync_checkers;	/* # of threads checking writes_pending */
423 	struct request_queue		*queue;	/* for plugging ... */
424 
425 	struct bitmap			*bitmap; /* the bitmap for the device */
426 	struct {
427 		struct file		*file; /* the bitmap file */
428 		loff_t			offset; /* offset from superblock of
429 						 * start of bitmap. May be
430 						 * negative, but not '0'
431 						 * For external metadata, offset
432 						 * from start of device.
433 						 */
434 		unsigned long		space; /* space available at this offset */
435 		loff_t			default_offset; /* this is the offset to use when
436 							 * hot-adding a bitmap.  It should
437 							 * eventually be settable by sysfs.
438 							 */
439 		unsigned long		default_space; /* space available at
440 							* default offset */
441 		struct mutex		mutex;
442 		unsigned long		chunksize;
443 		unsigned long		daemon_sleep; /* how many jiffies between updates? */
444 		unsigned long		max_write_behind; /* write-behind mode */
445 		int			external;
446 		int			nodes; /* Maximum number of nodes in the cluster */
447 		char                    cluster_name[64]; /* Name of the cluster */
448 	} bitmap_info;
449 
450 	atomic_t			max_corr_read_errors; /* max read retries */
451 	struct list_head		all_mddevs;
452 
453 	struct attribute_group		*to_remove;
454 
455 	struct bio_set			*bio_set;
456 	struct bio_set			*sync_set; /* for sync operations like
457 						   * metadata and bitmap writes
458 						   */
459 
460 	/* Generic flush handling.
461 	 * The last to finish preflush schedules a worker to submit
462 	 * the rest of the request (without the REQ_PREFLUSH flag).
463 	 */
464 	struct bio *flush_bio;
465 	atomic_t flush_pending;
466 	struct work_struct flush_work;
467 	struct work_struct event_work;	/* used by dm to report failure event */
468 	void (*sync_super)(struct mddev *mddev, struct md_rdev *rdev);
469 	struct md_cluster_info		*cluster_info;
470 	unsigned int			good_device_nr;	/* good device num within cluster raid */
471 };
472 
473 enum recovery_flags {
474 	/*
475 	 * If neither SYNC or RESHAPE are set, then it is a recovery.
476 	 */
477 	MD_RECOVERY_RUNNING,	/* a thread is running, or about to be started */
478 	MD_RECOVERY_SYNC,	/* actually doing a resync, not a recovery */
479 	MD_RECOVERY_RECOVER,	/* doing recovery, or need to try it. */
480 	MD_RECOVERY_INTR,	/* resync needs to be aborted for some reason */
481 	MD_RECOVERY_DONE,	/* thread is done and is waiting to be reaped */
482 	MD_RECOVERY_NEEDED,	/* we might need to start a resync/recover */
483 	MD_RECOVERY_REQUESTED,	/* user-space has requested a sync (used with SYNC) */
484 	MD_RECOVERY_CHECK,	/* user-space request for check-only, no repair */
485 	MD_RECOVERY_RESHAPE,	/* A reshape is happening */
486 	MD_RECOVERY_FROZEN,	/* User request to abort, and not restart, any action */
487 	MD_RECOVERY_ERROR,	/* sync-action interrupted because io-error */
488 };
489 
490 static inline int __must_check mddev_lock(struct mddev *mddev)
491 {
492 	return mutex_lock_interruptible(&mddev->reconfig_mutex);
493 }
494 
495 /* Sometimes we need to take the lock in a situation where
496  * failure due to interrupts is not acceptable.
497  */
498 static inline void mddev_lock_nointr(struct mddev *mddev)
499 {
500 	mutex_lock(&mddev->reconfig_mutex);
501 }
502 
503 static inline int mddev_trylock(struct mddev *mddev)
504 {
505 	return mutex_trylock(&mddev->reconfig_mutex);
506 }
507 extern void mddev_unlock(struct mddev *mddev);
508 
509 static inline void md_sync_acct(struct block_device *bdev, unsigned long nr_sectors)
510 {
511 	atomic_add(nr_sectors, &bdev->bd_contains->bd_disk->sync_io);
512 }
513 
514 static inline void md_sync_acct_bio(struct bio *bio, unsigned long nr_sectors)
515 {
516 	atomic_add(nr_sectors, &bio->bi_disk->sync_io);
517 }
518 
519 struct md_personality
520 {
521 	char *name;
522 	int level;
523 	struct list_head list;
524 	struct module *owner;
525 	bool (*make_request)(struct mddev *mddev, struct bio *bio);
526 	int (*run)(struct mddev *mddev);
527 	void (*free)(struct mddev *mddev, void *priv);
528 	void (*status)(struct seq_file *seq, struct mddev *mddev);
529 	/* error_handler must set ->faulty and clear ->in_sync
530 	 * if appropriate, and should abort recovery if needed
531 	 */
532 	void (*error_handler)(struct mddev *mddev, struct md_rdev *rdev);
533 	int (*hot_add_disk) (struct mddev *mddev, struct md_rdev *rdev);
534 	int (*hot_remove_disk) (struct mddev *mddev, struct md_rdev *rdev);
535 	int (*spare_active) (struct mddev *mddev);
536 	sector_t (*sync_request)(struct mddev *mddev, sector_t sector_nr, int *skipped);
537 	int (*resize) (struct mddev *mddev, sector_t sectors);
538 	sector_t (*size) (struct mddev *mddev, sector_t sectors, int raid_disks);
539 	int (*check_reshape) (struct mddev *mddev);
540 	int (*start_reshape) (struct mddev *mddev);
541 	void (*finish_reshape) (struct mddev *mddev);
542 	/* quiesce suspends or resumes internal processing.
543 	 * 1 - stop new actions and wait for action io to complete
544 	 * 0 - return to normal behaviour
545 	 */
546 	void (*quiesce) (struct mddev *mddev, int quiesce);
547 	/* takeover is used to transition an array from one
548 	 * personality to another.  The new personality must be able
549 	 * to handle the data in the current layout.
550 	 * e.g. 2drive raid1 -> 2drive raid5
551 	 *      ndrive raid5 -> degraded n+1drive raid6 with special layout
552 	 * If the takeover succeeds, a new 'private' structure is returned.
553 	 * This needs to be installed and then ->run used to activate the
554 	 * array.
555 	 */
556 	void *(*takeover) (struct mddev *mddev);
557 	/* congested implements bdi.congested_fn().
558 	 * Will not be called while array is 'suspended' */
559 	int (*congested)(struct mddev *mddev, int bits);
560 	/* Changes the consistency policy of an active array. */
561 	int (*change_consistency_policy)(struct mddev *mddev, const char *buf);
562 };
563 
564 struct md_sysfs_entry {
565 	struct attribute attr;
566 	ssize_t (*show)(struct mddev *, char *);
567 	ssize_t (*store)(struct mddev *, const char *, size_t);
568 };
569 extern struct attribute_group md_bitmap_group;
570 
571 static inline struct kernfs_node *sysfs_get_dirent_safe(struct kernfs_node *sd, char *name)
572 {
573 	if (sd)
574 		return sysfs_get_dirent(sd, name);
575 	return sd;
576 }
577 static inline void sysfs_notify_dirent_safe(struct kernfs_node *sd)
578 {
579 	if (sd)
580 		sysfs_notify_dirent(sd);
581 }
582 
583 static inline char * mdname (struct mddev * mddev)
584 {
585 	return mddev->gendisk ? mddev->gendisk->disk_name : "mdX";
586 }
587 
588 static inline int sysfs_link_rdev(struct mddev *mddev, struct md_rdev *rdev)
589 {
590 	char nm[20];
591 	if (!test_bit(Replacement, &rdev->flags) &&
592 	    !test_bit(Journal, &rdev->flags) &&
593 	    mddev->kobj.sd) {
594 		sprintf(nm, "rd%d", rdev->raid_disk);
595 		return sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
596 	} else
597 		return 0;
598 }
599 
600 static inline void sysfs_unlink_rdev(struct mddev *mddev, struct md_rdev *rdev)
601 {
602 	char nm[20];
603 	if (!test_bit(Replacement, &rdev->flags) &&
604 	    !test_bit(Journal, &rdev->flags) &&
605 	    mddev->kobj.sd) {
606 		sprintf(nm, "rd%d", rdev->raid_disk);
607 		sysfs_remove_link(&mddev->kobj, nm);
608 	}
609 }
610 
611 /*
612  * iterates through some rdev ringlist. It's safe to remove the
613  * current 'rdev'. Dont touch 'tmp' though.
614  */
615 #define rdev_for_each_list(rdev, tmp, head)				\
616 	list_for_each_entry_safe(rdev, tmp, head, same_set)
617 
618 /*
619  * iterates through the 'same array disks' ringlist
620  */
621 #define rdev_for_each(rdev, mddev)				\
622 	list_for_each_entry(rdev, &((mddev)->disks), same_set)
623 
624 #define rdev_for_each_safe(rdev, tmp, mddev)				\
625 	list_for_each_entry_safe(rdev, tmp, &((mddev)->disks), same_set)
626 
627 #define rdev_for_each_rcu(rdev, mddev)				\
628 	list_for_each_entry_rcu(rdev, &((mddev)->disks), same_set)
629 
630 struct md_thread {
631 	void			(*run) (struct md_thread *thread);
632 	struct mddev		*mddev;
633 	wait_queue_head_t	wqueue;
634 	unsigned long		flags;
635 	struct task_struct	*tsk;
636 	unsigned long		timeout;
637 	void			*private;
638 };
639 
640 #define THREAD_WAKEUP  0
641 
642 static inline void safe_put_page(struct page *p)
643 {
644 	if (p) put_page(p);
645 }
646 
647 extern int register_md_personality(struct md_personality *p);
648 extern int unregister_md_personality(struct md_personality *p);
649 extern int register_md_cluster_operations(struct md_cluster_operations *ops,
650 		struct module *module);
651 extern int unregister_md_cluster_operations(void);
652 extern int md_setup_cluster(struct mddev *mddev, int nodes);
653 extern void md_cluster_stop(struct mddev *mddev);
654 extern struct md_thread *md_register_thread(
655 	void (*run)(struct md_thread *thread),
656 	struct mddev *mddev,
657 	const char *name);
658 extern void md_unregister_thread(struct md_thread **threadp);
659 extern void md_wakeup_thread(struct md_thread *thread);
660 extern void md_check_recovery(struct mddev *mddev);
661 extern void md_reap_sync_thread(struct mddev *mddev);
662 extern int mddev_init_writes_pending(struct mddev *mddev);
663 extern bool md_write_start(struct mddev *mddev, struct bio *bi);
664 extern void md_write_inc(struct mddev *mddev, struct bio *bi);
665 extern void md_write_end(struct mddev *mddev);
666 extern void md_done_sync(struct mddev *mddev, int blocks, int ok);
667 extern void md_error(struct mddev *mddev, struct md_rdev *rdev);
668 extern void md_finish_reshape(struct mddev *mddev);
669 
670 extern int mddev_congested(struct mddev *mddev, int bits);
671 extern void md_flush_request(struct mddev *mddev, struct bio *bio);
672 extern void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
673 			   sector_t sector, int size, struct page *page);
674 extern int md_super_wait(struct mddev *mddev);
675 extern int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
676 			struct page *page, int op, int op_flags,
677 			bool metadata_op);
678 extern void md_do_sync(struct md_thread *thread);
679 extern void md_new_event(struct mddev *mddev);
680 extern void md_allow_write(struct mddev *mddev);
681 extern void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev);
682 extern void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors);
683 extern int md_check_no_bitmap(struct mddev *mddev);
684 extern int md_integrity_register(struct mddev *mddev);
685 extern int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev);
686 extern int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale);
687 
688 extern void mddev_init(struct mddev *mddev);
689 extern int md_run(struct mddev *mddev);
690 extern void md_stop(struct mddev *mddev);
691 extern void md_stop_writes(struct mddev *mddev);
692 extern int md_rdev_init(struct md_rdev *rdev);
693 extern void md_rdev_clear(struct md_rdev *rdev);
694 
695 extern void md_handle_request(struct mddev *mddev, struct bio *bio);
696 extern void mddev_suspend(struct mddev *mddev);
697 extern void mddev_resume(struct mddev *mddev);
698 extern struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
699 				   struct mddev *mddev);
700 
701 extern void md_reload_sb(struct mddev *mddev, int raid_disk);
702 extern void md_update_sb(struct mddev *mddev, int force);
703 extern void md_kick_rdev_from_array(struct md_rdev * rdev);
704 struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr);
705 
706 static inline void rdev_dec_pending(struct md_rdev *rdev, struct mddev *mddev)
707 {
708 	int faulty = test_bit(Faulty, &rdev->flags);
709 	if (atomic_dec_and_test(&rdev->nr_pending) && faulty) {
710 		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
711 		md_wakeup_thread(mddev->thread);
712 	}
713 }
714 
715 extern struct md_cluster_operations *md_cluster_ops;
716 static inline int mddev_is_clustered(struct mddev *mddev)
717 {
718 	return mddev->cluster_info && mddev->bitmap_info.nodes > 1;
719 }
720 
721 /* clear unsupported mddev_flags */
722 static inline void mddev_clear_unsupported_flags(struct mddev *mddev,
723 	unsigned long unsupported_flags)
724 {
725 	mddev->flags &= ~unsupported_flags;
726 }
727 
728 static inline void mddev_check_writesame(struct mddev *mddev, struct bio *bio)
729 {
730 	if (bio_op(bio) == REQ_OP_WRITE_SAME &&
731 	    !bio->bi_disk->queue->limits.max_write_same_sectors)
732 		mddev->queue->limits.max_write_same_sectors = 0;
733 }
734 
735 static inline void mddev_check_write_zeroes(struct mddev *mddev, struct bio *bio)
736 {
737 	if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
738 	    !bio->bi_disk->queue->limits.max_write_zeroes_sectors)
739 		mddev->queue->limits.max_write_zeroes_sectors = 0;
740 }
741 #endif /* _MD_MD_H */
742