xref: /linux/drivers/md/md.h (revision af873fcecef567abf8a3468b06dd4e4aab46da6d)
1 /* SPDX-License-Identifier: GPL-2.0-or-later */
2 /*
3    md.h : kernel internal structure of the Linux MD driver
4           Copyright (C) 1996-98 Ingo Molnar, Gadi Oxman
5 
6 */
7 
8 #ifndef _MD_MD_H
9 #define _MD_MD_H
10 
11 #include <linux/blkdev.h>
12 #include <linux/backing-dev.h>
13 #include <linux/badblocks.h>
14 #include <linux/kobject.h>
15 #include <linux/list.h>
16 #include <linux/mm.h>
17 #include <linux/mutex.h>
18 #include <linux/timer.h>
19 #include <linux/wait.h>
20 #include <linux/workqueue.h>
21 #include "md-cluster.h"
22 
23 #define MaxSector (~(sector_t)0)
24 
25 /*
26  * These flags should really be called "NO_RETRY" rather than
27  * "FAILFAST" because they don't make any promise about time lapse,
28  * only about the number of retries, which will be zero.
29  * REQ_FAILFAST_DRIVER is not included because
30  * Commit: 4a27446f3e39 ("[SCSI] modify scsi to handle new fail fast flags.")
31  * seems to suggest that the errors it avoids retrying should usually
32  * be retried.
33  */
34 #define	MD_FAILFAST	(REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT)
35 /*
36  * MD's 'extended' device
37  */
38 struct md_rdev {
39 	struct list_head same_set;	/* RAID devices within the same set */
40 
41 	sector_t sectors;		/* Device size (in 512bytes sectors) */
42 	struct mddev *mddev;		/* RAID array if running */
43 	int last_events;		/* IO event timestamp */
44 
45 	/*
46 	 * If meta_bdev is non-NULL, it means that a separate device is
47 	 * being used to store the metadata (superblock/bitmap) which
48 	 * would otherwise be contained on the same device as the data (bdev).
49 	 */
50 	struct block_device *meta_bdev;
51 	struct block_device *bdev;	/* block device handle */
52 
53 	struct page	*sb_page, *bb_page;
54 	int		sb_loaded;
55 	__u64		sb_events;
56 	sector_t	data_offset;	/* start of data in array */
57 	sector_t	new_data_offset;/* only relevant while reshaping */
58 	sector_t	sb_start;	/* offset of the super block (in 512byte sectors) */
59 	int		sb_size;	/* bytes in the superblock */
60 	int		preferred_minor;	/* autorun support */
61 
62 	struct kobject	kobj;
63 
64 	/* A device can be in one of three states based on two flags:
65 	 * Not working:   faulty==1 in_sync==0
66 	 * Fully working: faulty==0 in_sync==1
67 	 * Working, but not
68 	 * in sync with array
69 	 *                faulty==0 in_sync==0
70 	 *
71 	 * It can never have faulty==1, in_sync==1
72 	 * This reduces the burden of testing multiple flags in many cases
73 	 */
74 
75 	unsigned long	flags;	/* bit set of 'enum flag_bits' bits. */
76 	wait_queue_head_t blocked_wait;
77 
78 	int desc_nr;			/* descriptor index in the superblock */
79 	int raid_disk;			/* role of device in array */
80 	int new_raid_disk;		/* role that the device will have in
81 					 * the array after a level-change completes.
82 					 */
83 	int saved_raid_disk;		/* role that device used to have in the
84 					 * array and could again if we did a partial
85 					 * resync from the bitmap
86 					 */
87 	union {
88 		sector_t recovery_offset;/* If this device has been partially
89 					 * recovered, this is where we were
90 					 * up to.
91 					 */
92 		sector_t journal_tail;	/* If this device is a journal device,
93 					 * this is the journal tail (journal
94 					 * recovery start point)
95 					 */
96 	};
97 
98 	atomic_t	nr_pending;	/* number of pending requests.
99 					 * only maintained for arrays that
100 					 * support hot removal
101 					 */
102 	atomic_t	read_errors;	/* number of consecutive read errors that
103 					 * we have tried to ignore.
104 					 */
105 	time64_t	last_read_error;	/* monotonic time since our
106 						 * last read error
107 						 */
108 	atomic_t	corrected_errors; /* number of corrected read errors,
109 					   * for reporting to userspace and storing
110 					   * in superblock.
111 					   */
112 	struct work_struct del_work;	/* used for delayed sysfs removal */
113 
114 	struct kernfs_node *sysfs_state; /* handle for 'state'
115 					   * sysfs entry */
116 
117 	struct badblocks badblocks;
118 
119 	struct {
120 		short offset;	/* Offset from superblock to start of PPL.
121 				 * Not used by external metadata. */
122 		unsigned int size;	/* Size in sectors of the PPL space */
123 		sector_t sector;	/* First sector of the PPL space */
124 	} ppl;
125 };
126 enum flag_bits {
127 	Faulty,			/* device is known to have a fault */
128 	In_sync,		/* device is in_sync with rest of array */
129 	Bitmap_sync,		/* ..actually, not quite In_sync.  Need a
130 				 * bitmap-based recovery to get fully in sync.
131 				 * The bit is only meaningful before device
132 				 * has been passed to pers->hot_add_disk.
133 				 */
134 	WriteMostly,		/* Avoid reading if at all possible */
135 	AutoDetected,		/* added by auto-detect */
136 	Blocked,		/* An error occurred but has not yet
137 				 * been acknowledged by the metadata
138 				 * handler, so don't allow writes
139 				 * until it is cleared */
140 	WriteErrorSeen,		/* A write error has been seen on this
141 				 * device
142 				 */
143 	FaultRecorded,		/* Intermediate state for clearing
144 				 * Blocked.  The Fault is/will-be
145 				 * recorded in the metadata, but that
146 				 * metadata hasn't been stored safely
147 				 * on disk yet.
148 				 */
149 	BlockedBadBlocks,	/* A writer is blocked because they
150 				 * found an unacknowledged bad-block.
151 				 * This can safely be cleared at any
152 				 * time, and the writer will re-check.
153 				 * It may be set at any time, and at
154 				 * worst the writer will timeout and
155 				 * re-check.  So setting it as
156 				 * accurately as possible is good, but
157 				 * not absolutely critical.
158 				 */
159 	WantReplacement,	/* This device is a candidate to be
160 				 * hot-replaced, either because it has
161 				 * reported some faults, or because
162 				 * of explicit request.
163 				 */
164 	Replacement,		/* This device is a replacement for
165 				 * a want_replacement device with same
166 				 * raid_disk number.
167 				 */
168 	Candidate,		/* For clustered environments only:
169 				 * This device is seen locally but not
170 				 * by the whole cluster
171 				 */
172 	Journal,		/* This device is used as journal for
173 				 * raid-5/6.
174 				 * Usually, this device should be faster
175 				 * than other devices in the array
176 				 */
177 	ClusterRemove,
178 	RemoveSynchronized,	/* synchronize_rcu() was called after
179 				 * this device was known to be faulty,
180 				 * so it is safe to remove without
181 				 * another synchronize_rcu() call.
182 				 */
183 	ExternalBbl,            /* External metadata provides bad
184 				 * block management for a disk
185 				 */
186 	FailFast,		/* Minimal retries should be attempted on
187 				 * this device, so use REQ_FAILFAST_DEV.
188 				 * Also don't try to repair failed reads.
189 				 * It is expects that no bad block log
190 				 * is present.
191 				 */
192 	LastDev,		/* Seems to be the last working dev as
193 				 * it didn't fail, so don't use FailFast
194 				 * any more for metadata
195 				 */
196 };
197 
198 static inline int is_badblock(struct md_rdev *rdev, sector_t s, int sectors,
199 			      sector_t *first_bad, int *bad_sectors)
200 {
201 	if (unlikely(rdev->badblocks.count)) {
202 		int rv = badblocks_check(&rdev->badblocks, rdev->data_offset + s,
203 					sectors,
204 					first_bad, bad_sectors);
205 		if (rv)
206 			*first_bad -= rdev->data_offset;
207 		return rv;
208 	}
209 	return 0;
210 }
211 extern int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
212 			      int is_new);
213 extern int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
214 				int is_new);
215 struct md_cluster_info;
216 
217 /* change UNSUPPORTED_MDDEV_FLAGS for each array type if new flag is added */
218 enum mddev_flags {
219 	MD_ARRAY_FIRST_USE,	/* First use of array, needs initialization */
220 	MD_CLOSING,		/* If set, we are closing the array, do not open
221 				 * it then */
222 	MD_JOURNAL_CLEAN,	/* A raid with journal is already clean */
223 	MD_HAS_JOURNAL,		/* The raid array has journal feature set */
224 	MD_CLUSTER_RESYNC_LOCKED, /* cluster raid only, which means node
225 				   * already took resync lock, need to
226 				   * release the lock */
227 	MD_FAILFAST_SUPPORTED,	/* Using MD_FAILFAST on metadata writes is
228 				 * supported as calls to md_error() will
229 				 * never cause the array to become failed.
230 				 */
231 	MD_HAS_PPL,		/* The raid array has PPL feature set */
232 	MD_HAS_MULTIPLE_PPLS,	/* The raid array has multiple PPLs feature set */
233 	MD_ALLOW_SB_UPDATE,	/* md_check_recovery is allowed to update
234 				 * the metadata without taking reconfig_mutex.
235 				 */
236 	MD_UPDATING_SB,		/* md_check_recovery is updating the metadata
237 				 * without explicitly holding reconfig_mutex.
238 				 */
239 };
240 
241 enum mddev_sb_flags {
242 	MD_SB_CHANGE_DEVS,		/* Some device status has changed */
243 	MD_SB_CHANGE_CLEAN,	/* transition to or from 'clean' */
244 	MD_SB_CHANGE_PENDING,	/* switch from 'clean' to 'active' in progress */
245 	MD_SB_NEED_REWRITE,	/* metadata write needs to be repeated */
246 };
247 
248 struct mddev {
249 	void				*private;
250 	struct md_personality		*pers;
251 	dev_t				unit;
252 	int				md_minor;
253 	struct list_head		disks;
254 	unsigned long			flags;
255 	unsigned long			sb_flags;
256 
257 	int				suspended;
258 	atomic_t			active_io;
259 	int				ro;
260 	int				sysfs_active; /* set when sysfs deletes
261 						       * are happening, so run/
262 						       * takeover/stop are not safe
263 						       */
264 	struct gendisk			*gendisk;
265 
266 	struct kobject			kobj;
267 	int				hold_active;
268 #define	UNTIL_IOCTL	1
269 #define	UNTIL_STOP	2
270 
271 	/* Superblock information */
272 	int				major_version,
273 					minor_version,
274 					patch_version;
275 	int				persistent;
276 	int				external;	/* metadata is
277 							 * managed externally */
278 	char				metadata_type[17]; /* externally set*/
279 	int				chunk_sectors;
280 	time64_t			ctime, utime;
281 	int				level, layout;
282 	char				clevel[16];
283 	int				raid_disks;
284 	int				max_disks;
285 	sector_t			dev_sectors;	/* used size of
286 							 * component devices */
287 	sector_t			array_sectors; /* exported array size */
288 	int				external_size; /* size managed
289 							* externally */
290 	__u64				events;
291 	/* If the last 'event' was simply a clean->dirty transition, and
292 	 * we didn't write it to the spares, then it is safe and simple
293 	 * to just decrement the event count on a dirty->clean transition.
294 	 * So we record that possibility here.
295 	 */
296 	int				can_decrease_events;
297 
298 	char				uuid[16];
299 
300 	/* If the array is being reshaped, we need to record the
301 	 * new shape and an indication of where we are up to.
302 	 * This is written to the superblock.
303 	 * If reshape_position is MaxSector, then no reshape is happening (yet).
304 	 */
305 	sector_t			reshape_position;
306 	int				delta_disks, new_level, new_layout;
307 	int				new_chunk_sectors;
308 	int				reshape_backwards;
309 
310 	struct md_thread		*thread;	/* management thread */
311 	struct md_thread		*sync_thread;	/* doing resync or reconstruct */
312 
313 	/* 'last_sync_action' is initialized to "none".  It is set when a
314 	 * sync operation (i.e "data-check", "requested-resync", "resync",
315 	 * "recovery", or "reshape") is started.  It holds this value even
316 	 * when the sync thread is "frozen" (interrupted) or "idle" (stopped
317 	 * or finished).  It is overwritten when a new sync operation is begun.
318 	 */
319 	char				*last_sync_action;
320 	sector_t			curr_resync;	/* last block scheduled */
321 	/* As resync requests can complete out of order, we cannot easily track
322 	 * how much resync has been completed.  So we occasionally pause until
323 	 * everything completes, then set curr_resync_completed to curr_resync.
324 	 * As such it may be well behind the real resync mark, but it is a value
325 	 * we are certain of.
326 	 */
327 	sector_t			curr_resync_completed;
328 	unsigned long			resync_mark;	/* a recent timestamp */
329 	sector_t			resync_mark_cnt;/* blocks written at resync_mark */
330 	sector_t			curr_mark_cnt; /* blocks scheduled now */
331 
332 	sector_t			resync_max_sectors; /* may be set by personality */
333 
334 	atomic64_t			resync_mismatches; /* count of sectors where
335 							    * parity/replica mismatch found
336 							    */
337 
338 	/* allow user-space to request suspension of IO to regions of the array */
339 	sector_t			suspend_lo;
340 	sector_t			suspend_hi;
341 	/* if zero, use the system-wide default */
342 	int				sync_speed_min;
343 	int				sync_speed_max;
344 
345 	/* resync even though the same disks are shared among md-devices */
346 	int				parallel_resync;
347 
348 	int				ok_start_degraded;
349 
350 	unsigned long			recovery;
351 	/* If a RAID personality determines that recovery (of a particular
352 	 * device) will fail due to a read error on the source device, it
353 	 * takes a copy of this number and does not attempt recovery again
354 	 * until this number changes.
355 	 */
356 	int				recovery_disabled;
357 
358 	int				in_sync;	/* know to not need resync */
359 	/* 'open_mutex' avoids races between 'md_open' and 'do_md_stop', so
360 	 * that we are never stopping an array while it is open.
361 	 * 'reconfig_mutex' protects all other reconfiguration.
362 	 * These locks are separate due to conflicting interactions
363 	 * with bdev->bd_mutex.
364 	 * Lock ordering is:
365 	 *  reconfig_mutex -> bd_mutex : e.g. do_md_run -> revalidate_disk
366 	 *  bd_mutex -> open_mutex:  e.g. __blkdev_get -> md_open
367 	 */
368 	struct mutex			open_mutex;
369 	struct mutex			reconfig_mutex;
370 	atomic_t			active;		/* general refcount */
371 	atomic_t			openers;	/* number of active opens */
372 
373 	int				changed;	/* True if we might need to
374 							 * reread partition info */
375 	int				degraded;	/* whether md should consider
376 							 * adding a spare
377 							 */
378 
379 	atomic_t			recovery_active; /* blocks scheduled, but not written */
380 	wait_queue_head_t		recovery_wait;
381 	sector_t			recovery_cp;
382 	sector_t			resync_min;	/* user requested sync
383 							 * starts here */
384 	sector_t			resync_max;	/* resync should pause
385 							 * when it gets here */
386 
387 	struct kernfs_node		*sysfs_state;	/* handle for 'array_state'
388 							 * file in sysfs.
389 							 */
390 	struct kernfs_node		*sysfs_action;  /* handle for 'sync_action' */
391 
392 	struct work_struct del_work;	/* used for delayed sysfs removal */
393 
394 	/* "lock" protects:
395 	 *   flush_bio transition from NULL to !NULL
396 	 *   rdev superblocks, events
397 	 *   clearing MD_CHANGE_*
398 	 *   in_sync - and related safemode and MD_CHANGE changes
399 	 *   pers (also protected by reconfig_mutex and pending IO).
400 	 *   clearing ->bitmap
401 	 *   clearing ->bitmap_info.file
402 	 *   changing ->resync_{min,max}
403 	 *   setting MD_RECOVERY_RUNNING (which interacts with resync_{min,max})
404 	 */
405 	spinlock_t			lock;
406 	wait_queue_head_t		sb_wait;	/* for waiting on superblock updates */
407 	atomic_t			pending_writes;	/* number of active superblock writes */
408 
409 	unsigned int			safemode;	/* if set, update "clean" superblock
410 							 * when no writes pending.
411 							 */
412 	unsigned int			safemode_delay;
413 	struct timer_list		safemode_timer;
414 	struct percpu_ref		writes_pending;
415 	int				sync_checkers;	/* # of threads checking writes_pending */
416 	struct request_queue		*queue;	/* for plugging ... */
417 
418 	struct bitmap			*bitmap; /* the bitmap for the device */
419 	struct {
420 		struct file		*file; /* the bitmap file */
421 		loff_t			offset; /* offset from superblock of
422 						 * start of bitmap. May be
423 						 * negative, but not '0'
424 						 * For external metadata, offset
425 						 * from start of device.
426 						 */
427 		unsigned long		space; /* space available at this offset */
428 		loff_t			default_offset; /* this is the offset to use when
429 							 * hot-adding a bitmap.  It should
430 							 * eventually be settable by sysfs.
431 							 */
432 		unsigned long		default_space; /* space available at
433 							* default offset */
434 		struct mutex		mutex;
435 		unsigned long		chunksize;
436 		unsigned long		daemon_sleep; /* how many jiffies between updates? */
437 		unsigned long		max_write_behind; /* write-behind mode */
438 		int			external;
439 		int			nodes; /* Maximum number of nodes in the cluster */
440 		char                    cluster_name[64]; /* Name of the cluster */
441 	} bitmap_info;
442 
443 	atomic_t			max_corr_read_errors; /* max read retries */
444 	struct list_head		all_mddevs;
445 
446 	struct attribute_group		*to_remove;
447 
448 	struct bio_set			bio_set;
449 	struct bio_set			sync_set; /* for sync operations like
450 						   * metadata and bitmap writes
451 						   */
452 
453 	/* Generic flush handling.
454 	 * The last to finish preflush schedules a worker to submit
455 	 * the rest of the request (without the REQ_PREFLUSH flag).
456 	 */
457 	struct bio *flush_bio;
458 	atomic_t flush_pending;
459 	ktime_t start_flush, last_flush; /* last_flush is when the last completed
460 					  * flush was started.
461 					  */
462 	struct work_struct flush_work;
463 	struct work_struct event_work;	/* used by dm to report failure event */
464 	void (*sync_super)(struct mddev *mddev, struct md_rdev *rdev);
465 	struct md_cluster_info		*cluster_info;
466 	unsigned int			good_device_nr;	/* good device num within cluster raid */
467 
468 	bool	has_superblocks:1;
469 };
470 
471 enum recovery_flags {
472 	/*
473 	 * If neither SYNC or RESHAPE are set, then it is a recovery.
474 	 */
475 	MD_RECOVERY_RUNNING,	/* a thread is running, or about to be started */
476 	MD_RECOVERY_SYNC,	/* actually doing a resync, not a recovery */
477 	MD_RECOVERY_RECOVER,	/* doing recovery, or need to try it. */
478 	MD_RECOVERY_INTR,	/* resync needs to be aborted for some reason */
479 	MD_RECOVERY_DONE,	/* thread is done and is waiting to be reaped */
480 	MD_RECOVERY_NEEDED,	/* we might need to start a resync/recover */
481 	MD_RECOVERY_REQUESTED,	/* user-space has requested a sync (used with SYNC) */
482 	MD_RECOVERY_CHECK,	/* user-space request for check-only, no repair */
483 	MD_RECOVERY_RESHAPE,	/* A reshape is happening */
484 	MD_RECOVERY_FROZEN,	/* User request to abort, and not restart, any action */
485 	MD_RECOVERY_ERROR,	/* sync-action interrupted because io-error */
486 	MD_RECOVERY_WAIT,	/* waiting for pers->start() to finish */
487 	MD_RESYNCING_REMOTE,	/* remote node is running resync thread */
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 	/*
527 	 * start up works that do NOT require md_thread. tasks that
528 	 * requires md_thread should go into start()
529 	 */
530 	int (*run)(struct mddev *mddev);
531 	/* start up works that require md threads */
532 	int (*start)(struct mddev *mddev);
533 	void (*free)(struct mddev *mddev, void *priv);
534 	void (*status)(struct seq_file *seq, struct mddev *mddev);
535 	/* error_handler must set ->faulty and clear ->in_sync
536 	 * if appropriate, and should abort recovery if needed
537 	 */
538 	void (*error_handler)(struct mddev *mddev, struct md_rdev *rdev);
539 	int (*hot_add_disk) (struct mddev *mddev, struct md_rdev *rdev);
540 	int (*hot_remove_disk) (struct mddev *mddev, struct md_rdev *rdev);
541 	int (*spare_active) (struct mddev *mddev);
542 	sector_t (*sync_request)(struct mddev *mddev, sector_t sector_nr, int *skipped);
543 	int (*resize) (struct mddev *mddev, sector_t sectors);
544 	sector_t (*size) (struct mddev *mddev, sector_t sectors, int raid_disks);
545 	int (*check_reshape) (struct mddev *mddev);
546 	int (*start_reshape) (struct mddev *mddev);
547 	void (*finish_reshape) (struct mddev *mddev);
548 	void (*update_reshape_pos) (struct mddev *mddev);
549 	/* quiesce suspends or resumes internal processing.
550 	 * 1 - stop new actions and wait for action io to complete
551 	 * 0 - return to normal behaviour
552 	 */
553 	void (*quiesce) (struct mddev *mddev, int quiesce);
554 	/* takeover is used to transition an array from one
555 	 * personality to another.  The new personality must be able
556 	 * to handle the data in the current layout.
557 	 * e.g. 2drive raid1 -> 2drive raid5
558 	 *      ndrive raid5 -> degraded n+1drive raid6 with special layout
559 	 * If the takeover succeeds, a new 'private' structure is returned.
560 	 * This needs to be installed and then ->run used to activate the
561 	 * array.
562 	 */
563 	void *(*takeover) (struct mddev *mddev);
564 	/* congested implements bdi.congested_fn().
565 	 * Will not be called while array is 'suspended' */
566 	int (*congested)(struct mddev *mddev, int bits);
567 	/* Changes the consistency policy of an active array. */
568 	int (*change_consistency_policy)(struct mddev *mddev, const char *buf);
569 };
570 
571 struct md_sysfs_entry {
572 	struct attribute attr;
573 	ssize_t (*show)(struct mddev *, char *);
574 	ssize_t (*store)(struct mddev *, const char *, size_t);
575 };
576 extern struct attribute_group md_bitmap_group;
577 
578 static inline struct kernfs_node *sysfs_get_dirent_safe(struct kernfs_node *sd, char *name)
579 {
580 	if (sd)
581 		return sysfs_get_dirent(sd, name);
582 	return sd;
583 }
584 static inline void sysfs_notify_dirent_safe(struct kernfs_node *sd)
585 {
586 	if (sd)
587 		sysfs_notify_dirent(sd);
588 }
589 
590 static inline char * mdname (struct mddev * mddev)
591 {
592 	return mddev->gendisk ? mddev->gendisk->disk_name : "mdX";
593 }
594 
595 static inline int sysfs_link_rdev(struct mddev *mddev, struct md_rdev *rdev)
596 {
597 	char nm[20];
598 	if (!test_bit(Replacement, &rdev->flags) &&
599 	    !test_bit(Journal, &rdev->flags) &&
600 	    mddev->kobj.sd) {
601 		sprintf(nm, "rd%d", rdev->raid_disk);
602 		return sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
603 	} else
604 		return 0;
605 }
606 
607 static inline void sysfs_unlink_rdev(struct mddev *mddev, struct md_rdev *rdev)
608 {
609 	char nm[20];
610 	if (!test_bit(Replacement, &rdev->flags) &&
611 	    !test_bit(Journal, &rdev->flags) &&
612 	    mddev->kobj.sd) {
613 		sprintf(nm, "rd%d", rdev->raid_disk);
614 		sysfs_remove_link(&mddev->kobj, nm);
615 	}
616 }
617 
618 /*
619  * iterates through some rdev ringlist. It's safe to remove the
620  * current 'rdev'. Dont touch 'tmp' though.
621  */
622 #define rdev_for_each_list(rdev, tmp, head)				\
623 	list_for_each_entry_safe(rdev, tmp, head, same_set)
624 
625 /*
626  * iterates through the 'same array disks' ringlist
627  */
628 #define rdev_for_each(rdev, mddev)				\
629 	list_for_each_entry(rdev, &((mddev)->disks), same_set)
630 
631 #define rdev_for_each_safe(rdev, tmp, mddev)				\
632 	list_for_each_entry_safe(rdev, tmp, &((mddev)->disks), same_set)
633 
634 #define rdev_for_each_rcu(rdev, mddev)				\
635 	list_for_each_entry_rcu(rdev, &((mddev)->disks), same_set)
636 
637 struct md_thread {
638 	void			(*run) (struct md_thread *thread);
639 	struct mddev		*mddev;
640 	wait_queue_head_t	wqueue;
641 	unsigned long		flags;
642 	struct task_struct	*tsk;
643 	unsigned long		timeout;
644 	void			*private;
645 };
646 
647 #define THREAD_WAKEUP  0
648 
649 static inline void safe_put_page(struct page *p)
650 {
651 	if (p) put_page(p);
652 }
653 
654 extern int register_md_personality(struct md_personality *p);
655 extern int unregister_md_personality(struct md_personality *p);
656 extern int register_md_cluster_operations(struct md_cluster_operations *ops,
657 		struct module *module);
658 extern int unregister_md_cluster_operations(void);
659 extern int md_setup_cluster(struct mddev *mddev, int nodes);
660 extern void md_cluster_stop(struct mddev *mddev);
661 extern struct md_thread *md_register_thread(
662 	void (*run)(struct md_thread *thread),
663 	struct mddev *mddev,
664 	const char *name);
665 extern void md_unregister_thread(struct md_thread **threadp);
666 extern void md_wakeup_thread(struct md_thread *thread);
667 extern void md_check_recovery(struct mddev *mddev);
668 extern void md_reap_sync_thread(struct mddev *mddev);
669 extern int mddev_init_writes_pending(struct mddev *mddev);
670 extern bool md_write_start(struct mddev *mddev, struct bio *bi);
671 extern void md_write_inc(struct mddev *mddev, struct bio *bi);
672 extern void md_write_end(struct mddev *mddev);
673 extern void md_done_sync(struct mddev *mddev, int blocks, int ok);
674 extern void md_error(struct mddev *mddev, struct md_rdev *rdev);
675 extern void md_finish_reshape(struct mddev *mddev);
676 
677 extern int mddev_congested(struct mddev *mddev, int bits);
678 extern void md_flush_request(struct mddev *mddev, struct bio *bio);
679 extern void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
680 			   sector_t sector, int size, struct page *page);
681 extern int md_super_wait(struct mddev *mddev);
682 extern int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
683 			struct page *page, int op, int op_flags,
684 			bool metadata_op);
685 extern void md_do_sync(struct md_thread *thread);
686 extern void md_new_event(struct mddev *mddev);
687 extern void md_allow_write(struct mddev *mddev);
688 extern void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev);
689 extern void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors);
690 extern int md_check_no_bitmap(struct mddev *mddev);
691 extern int md_integrity_register(struct mddev *mddev);
692 extern int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev);
693 extern int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale);
694 
695 extern void mddev_init(struct mddev *mddev);
696 extern int md_run(struct mddev *mddev);
697 extern int md_start(struct mddev *mddev);
698 extern void md_stop(struct mddev *mddev);
699 extern void md_stop_writes(struct mddev *mddev);
700 extern int md_rdev_init(struct md_rdev *rdev);
701 extern void md_rdev_clear(struct md_rdev *rdev);
702 
703 extern void md_handle_request(struct mddev *mddev, struct bio *bio);
704 extern void mddev_suspend(struct mddev *mddev);
705 extern void mddev_resume(struct mddev *mddev);
706 extern struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
707 				   struct mddev *mddev);
708 
709 extern void md_reload_sb(struct mddev *mddev, int raid_disk);
710 extern void md_update_sb(struct mddev *mddev, int force);
711 extern void md_kick_rdev_from_array(struct md_rdev * rdev);
712 struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr);
713 struct md_rdev *md_find_rdev_rcu(struct mddev *mddev, dev_t dev);
714 
715 static inline void rdev_dec_pending(struct md_rdev *rdev, struct mddev *mddev)
716 {
717 	int faulty = test_bit(Faulty, &rdev->flags);
718 	if (atomic_dec_and_test(&rdev->nr_pending) && faulty) {
719 		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
720 		md_wakeup_thread(mddev->thread);
721 	}
722 }
723 
724 extern struct md_cluster_operations *md_cluster_ops;
725 static inline int mddev_is_clustered(struct mddev *mddev)
726 {
727 	return mddev->cluster_info && mddev->bitmap_info.nodes > 1;
728 }
729 
730 /* clear unsupported mddev_flags */
731 static inline void mddev_clear_unsupported_flags(struct mddev *mddev,
732 	unsigned long unsupported_flags)
733 {
734 	mddev->flags &= ~unsupported_flags;
735 }
736 
737 static inline void mddev_check_writesame(struct mddev *mddev, struct bio *bio)
738 {
739 	if (bio_op(bio) == REQ_OP_WRITE_SAME &&
740 	    !bio->bi_disk->queue->limits.max_write_same_sectors)
741 		mddev->queue->limits.max_write_same_sectors = 0;
742 }
743 
744 static inline void mddev_check_write_zeroes(struct mddev *mddev, struct bio *bio)
745 {
746 	if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
747 	    !bio->bi_disk->queue->limits.max_write_zeroes_sectors)
748 		mddev->queue->limits.max_write_zeroes_sectors = 0;
749 }
750 #endif /* _MD_MD_H */
751