xref: /openbmc/linux/drivers/md/md.h (revision adb80490)
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 
113 	/*
114 	 * The members for check collision of write behind IOs.
115 	 */
116 	struct list_head wb_list;
117 	spinlock_t wb_list_lock;
118 	wait_queue_head_t wb_io_wait;
119 
120 	struct work_struct del_work;	/* used for delayed sysfs removal */
121 
122 	struct kernfs_node *sysfs_state; /* handle for 'state'
123 					   * sysfs entry */
124 
125 	struct badblocks badblocks;
126 
127 	struct {
128 		short offset;	/* Offset from superblock to start of PPL.
129 				 * Not used by external metadata. */
130 		unsigned int size;	/* Size in sectors of the PPL space */
131 		sector_t sector;	/* First sector of the PPL space */
132 	} ppl;
133 };
134 enum flag_bits {
135 	Faulty,			/* device is known to have a fault */
136 	In_sync,		/* device is in_sync with rest of array */
137 	Bitmap_sync,		/* ..actually, not quite In_sync.  Need a
138 				 * bitmap-based recovery to get fully in sync.
139 				 * The bit is only meaningful before device
140 				 * has been passed to pers->hot_add_disk.
141 				 */
142 	WriteMostly,		/* Avoid reading if at all possible */
143 	AutoDetected,		/* added by auto-detect */
144 	Blocked,		/* An error occurred but has not yet
145 				 * been acknowledged by the metadata
146 				 * handler, so don't allow writes
147 				 * until it is cleared */
148 	WriteErrorSeen,		/* A write error has been seen on this
149 				 * device
150 				 */
151 	FaultRecorded,		/* Intermediate state for clearing
152 				 * Blocked.  The Fault is/will-be
153 				 * recorded in the metadata, but that
154 				 * metadata hasn't been stored safely
155 				 * on disk yet.
156 				 */
157 	BlockedBadBlocks,	/* A writer is blocked because they
158 				 * found an unacknowledged bad-block.
159 				 * This can safely be cleared at any
160 				 * time, and the writer will re-check.
161 				 * It may be set at any time, and at
162 				 * worst the writer will timeout and
163 				 * re-check.  So setting it as
164 				 * accurately as possible is good, but
165 				 * not absolutely critical.
166 				 */
167 	WantReplacement,	/* This device is a candidate to be
168 				 * hot-replaced, either because it has
169 				 * reported some faults, or because
170 				 * of explicit request.
171 				 */
172 	Replacement,		/* This device is a replacement for
173 				 * a want_replacement device with same
174 				 * raid_disk number.
175 				 */
176 	Candidate,		/* For clustered environments only:
177 				 * This device is seen locally but not
178 				 * by the whole cluster
179 				 */
180 	Journal,		/* This device is used as journal for
181 				 * raid-5/6.
182 				 * Usually, this device should be faster
183 				 * than other devices in the array
184 				 */
185 	ClusterRemove,
186 	RemoveSynchronized,	/* synchronize_rcu() was called after
187 				 * this device was known to be faulty,
188 				 * so it is safe to remove without
189 				 * another synchronize_rcu() call.
190 				 */
191 	ExternalBbl,            /* External metadata provides bad
192 				 * block management for a disk
193 				 */
194 	FailFast,		/* Minimal retries should be attempted on
195 				 * this device, so use REQ_FAILFAST_DEV.
196 				 * Also don't try to repair failed reads.
197 				 * It is expects that no bad block log
198 				 * is present.
199 				 */
200 	LastDev,		/* Seems to be the last working dev as
201 				 * it didn't fail, so don't use FailFast
202 				 * any more for metadata
203 				 */
204 	WBCollisionCheck,	/*
205 				 * multiqueue device should check if there
206 				 * is collision between write behind bios.
207 				 */
208 };
209 
210 static inline int is_badblock(struct md_rdev *rdev, sector_t s, int sectors,
211 			      sector_t *first_bad, int *bad_sectors)
212 {
213 	if (unlikely(rdev->badblocks.count)) {
214 		int rv = badblocks_check(&rdev->badblocks, rdev->data_offset + s,
215 					sectors,
216 					first_bad, bad_sectors);
217 		if (rv)
218 			*first_bad -= rdev->data_offset;
219 		return rv;
220 	}
221 	return 0;
222 }
223 extern int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
224 			      int is_new);
225 extern int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
226 				int is_new);
227 struct md_cluster_info;
228 
229 /* change UNSUPPORTED_MDDEV_FLAGS for each array type if new flag is added */
230 enum mddev_flags {
231 	MD_ARRAY_FIRST_USE,	/* First use of array, needs initialization */
232 	MD_CLOSING,		/* If set, we are closing the array, do not open
233 				 * it then */
234 	MD_JOURNAL_CLEAN,	/* A raid with journal is already clean */
235 	MD_HAS_JOURNAL,		/* The raid array has journal feature set */
236 	MD_CLUSTER_RESYNC_LOCKED, /* cluster raid only, which means node
237 				   * already took resync lock, need to
238 				   * release the lock */
239 	MD_FAILFAST_SUPPORTED,	/* Using MD_FAILFAST on metadata writes is
240 				 * supported as calls to md_error() will
241 				 * never cause the array to become failed.
242 				 */
243 	MD_HAS_PPL,		/* The raid array has PPL feature set */
244 	MD_HAS_MULTIPLE_PPLS,	/* The raid array has multiple PPLs feature set */
245 	MD_ALLOW_SB_UPDATE,	/* md_check_recovery is allowed to update
246 				 * the metadata without taking reconfig_mutex.
247 				 */
248 	MD_UPDATING_SB,		/* md_check_recovery is updating the metadata
249 				 * without explicitly holding reconfig_mutex.
250 				 */
251 };
252 
253 enum mddev_sb_flags {
254 	MD_SB_CHANGE_DEVS,		/* Some device status has changed */
255 	MD_SB_CHANGE_CLEAN,	/* transition to or from 'clean' */
256 	MD_SB_CHANGE_PENDING,	/* switch from 'clean' to 'active' in progress */
257 	MD_SB_NEED_REWRITE,	/* metadata write needs to be repeated */
258 };
259 
260 #define NR_WB_INFOS	8
261 /* record current range of write behind IOs */
262 struct wb_info {
263 	sector_t lo;
264 	sector_t hi;
265 	struct list_head list;
266 };
267 
268 struct mddev {
269 	void				*private;
270 	struct md_personality		*pers;
271 	dev_t				unit;
272 	int				md_minor;
273 	struct list_head		disks;
274 	unsigned long			flags;
275 	unsigned long			sb_flags;
276 
277 	int				suspended;
278 	atomic_t			active_io;
279 	int				ro;
280 	int				sysfs_active; /* set when sysfs deletes
281 						       * are happening, so run/
282 						       * takeover/stop are not safe
283 						       */
284 	struct gendisk			*gendisk;
285 
286 	struct kobject			kobj;
287 	int				hold_active;
288 #define	UNTIL_IOCTL	1
289 #define	UNTIL_STOP	2
290 
291 	/* Superblock information */
292 	int				major_version,
293 					minor_version,
294 					patch_version;
295 	int				persistent;
296 	int				external;	/* metadata is
297 							 * managed externally */
298 	char				metadata_type[17]; /* externally set*/
299 	int				chunk_sectors;
300 	time64_t			ctime, utime;
301 	int				level, layout;
302 	char				clevel[16];
303 	int				raid_disks;
304 	int				max_disks;
305 	sector_t			dev_sectors;	/* used size of
306 							 * component devices */
307 	sector_t			array_sectors; /* exported array size */
308 	int				external_size; /* size managed
309 							* externally */
310 	__u64				events;
311 	/* If the last 'event' was simply a clean->dirty transition, and
312 	 * we didn't write it to the spares, then it is safe and simple
313 	 * to just decrement the event count on a dirty->clean transition.
314 	 * So we record that possibility here.
315 	 */
316 	int				can_decrease_events;
317 
318 	char				uuid[16];
319 
320 	/* If the array is being reshaped, we need to record the
321 	 * new shape and an indication of where we are up to.
322 	 * This is written to the superblock.
323 	 * If reshape_position is MaxSector, then no reshape is happening (yet).
324 	 */
325 	sector_t			reshape_position;
326 	int				delta_disks, new_level, new_layout;
327 	int				new_chunk_sectors;
328 	int				reshape_backwards;
329 
330 	struct md_thread		*thread;	/* management thread */
331 	struct md_thread		*sync_thread;	/* doing resync or reconstruct */
332 
333 	/* 'last_sync_action' is initialized to "none".  It is set when a
334 	 * sync operation (i.e "data-check", "requested-resync", "resync",
335 	 * "recovery", or "reshape") is started.  It holds this value even
336 	 * when the sync thread is "frozen" (interrupted) or "idle" (stopped
337 	 * or finished).  It is overwritten when a new sync operation is begun.
338 	 */
339 	char				*last_sync_action;
340 	sector_t			curr_resync;	/* last block scheduled */
341 	/* As resync requests can complete out of order, we cannot easily track
342 	 * how much resync has been completed.  So we occasionally pause until
343 	 * everything completes, then set curr_resync_completed to curr_resync.
344 	 * As such it may be well behind the real resync mark, but it is a value
345 	 * we are certain of.
346 	 */
347 	sector_t			curr_resync_completed;
348 	unsigned long			resync_mark;	/* a recent timestamp */
349 	sector_t			resync_mark_cnt;/* blocks written at resync_mark */
350 	sector_t			curr_mark_cnt; /* blocks scheduled now */
351 
352 	sector_t			resync_max_sectors; /* may be set by personality */
353 
354 	atomic64_t			resync_mismatches; /* count of sectors where
355 							    * parity/replica mismatch found
356 							    */
357 
358 	/* allow user-space to request suspension of IO to regions of the array */
359 	sector_t			suspend_lo;
360 	sector_t			suspend_hi;
361 	/* if zero, use the system-wide default */
362 	int				sync_speed_min;
363 	int				sync_speed_max;
364 
365 	/* resync even though the same disks are shared among md-devices */
366 	int				parallel_resync;
367 
368 	int				ok_start_degraded;
369 
370 	unsigned long			recovery;
371 	/* If a RAID personality determines that recovery (of a particular
372 	 * device) will fail due to a read error on the source device, it
373 	 * takes a copy of this number and does not attempt recovery again
374 	 * until this number changes.
375 	 */
376 	int				recovery_disabled;
377 
378 	int				in_sync;	/* know to not need resync */
379 	/* 'open_mutex' avoids races between 'md_open' and 'do_md_stop', so
380 	 * that we are never stopping an array while it is open.
381 	 * 'reconfig_mutex' protects all other reconfiguration.
382 	 * These locks are separate due to conflicting interactions
383 	 * with bdev->bd_mutex.
384 	 * Lock ordering is:
385 	 *  reconfig_mutex -> bd_mutex : e.g. do_md_run -> revalidate_disk
386 	 *  bd_mutex -> open_mutex:  e.g. __blkdev_get -> md_open
387 	 */
388 	struct mutex			open_mutex;
389 	struct mutex			reconfig_mutex;
390 	atomic_t			active;		/* general refcount */
391 	atomic_t			openers;	/* number of active opens */
392 
393 	int				changed;	/* True if we might need to
394 							 * reread partition info */
395 	int				degraded;	/* whether md should consider
396 							 * adding a spare
397 							 */
398 
399 	atomic_t			recovery_active; /* blocks scheduled, but not written */
400 	wait_queue_head_t		recovery_wait;
401 	sector_t			recovery_cp;
402 	sector_t			resync_min;	/* user requested sync
403 							 * starts here */
404 	sector_t			resync_max;	/* resync should pause
405 							 * when it gets here */
406 
407 	struct kernfs_node		*sysfs_state;	/* handle for 'array_state'
408 							 * file in sysfs.
409 							 */
410 	struct kernfs_node		*sysfs_action;  /* handle for 'sync_action' */
411 
412 	struct work_struct del_work;	/* used for delayed sysfs removal */
413 
414 	/* "lock" protects:
415 	 *   flush_bio transition from NULL to !NULL
416 	 *   rdev superblocks, events
417 	 *   clearing MD_CHANGE_*
418 	 *   in_sync - and related safemode and MD_CHANGE changes
419 	 *   pers (also protected by reconfig_mutex and pending IO).
420 	 *   clearing ->bitmap
421 	 *   clearing ->bitmap_info.file
422 	 *   changing ->resync_{min,max}
423 	 *   setting MD_RECOVERY_RUNNING (which interacts with resync_{min,max})
424 	 */
425 	spinlock_t			lock;
426 	wait_queue_head_t		sb_wait;	/* for waiting on superblock updates */
427 	atomic_t			pending_writes;	/* number of active superblock writes */
428 
429 	unsigned int			safemode;	/* if set, update "clean" superblock
430 							 * when no writes pending.
431 							 */
432 	unsigned int			safemode_delay;
433 	struct timer_list		safemode_timer;
434 	struct percpu_ref		writes_pending;
435 	int				sync_checkers;	/* # of threads checking writes_pending */
436 	struct request_queue		*queue;	/* for plugging ... */
437 
438 	struct bitmap			*bitmap; /* the bitmap for the device */
439 	struct {
440 		struct file		*file; /* the bitmap file */
441 		loff_t			offset; /* offset from superblock of
442 						 * start of bitmap. May be
443 						 * negative, but not '0'
444 						 * For external metadata, offset
445 						 * from start of device.
446 						 */
447 		unsigned long		space; /* space available at this offset */
448 		loff_t			default_offset; /* this is the offset to use when
449 							 * hot-adding a bitmap.  It should
450 							 * eventually be settable by sysfs.
451 							 */
452 		unsigned long		default_space; /* space available at
453 							* default offset */
454 		struct mutex		mutex;
455 		unsigned long		chunksize;
456 		unsigned long		daemon_sleep; /* how many jiffies between updates? */
457 		unsigned long		max_write_behind; /* write-behind mode */
458 		int			external;
459 		int			nodes; /* Maximum number of nodes in the cluster */
460 		char                    cluster_name[64]; /* Name of the cluster */
461 	} bitmap_info;
462 
463 	atomic_t			max_corr_read_errors; /* max read retries */
464 	struct list_head		all_mddevs;
465 
466 	struct attribute_group		*to_remove;
467 
468 	struct bio_set			bio_set;
469 	struct bio_set			sync_set; /* for sync operations like
470 						   * metadata and bitmap writes
471 						   */
472 
473 	/* Generic flush handling.
474 	 * The last to finish preflush schedules a worker to submit
475 	 * the rest of the request (without the REQ_PREFLUSH flag).
476 	 */
477 	struct bio *flush_bio;
478 	atomic_t flush_pending;
479 	ktime_t start_flush, last_flush; /* last_flush is when the last completed
480 					  * flush was started.
481 					  */
482 	struct work_struct flush_work;
483 	struct work_struct event_work;	/* used by dm to report failure event */
484 	mempool_t *wb_info_pool;
485 	void (*sync_super)(struct mddev *mddev, struct md_rdev *rdev);
486 	struct md_cluster_info		*cluster_info;
487 	unsigned int			good_device_nr;	/* good device num within cluster raid */
488 
489 	bool	has_superblocks:1;
490 };
491 
492 enum recovery_flags {
493 	/*
494 	 * If neither SYNC or RESHAPE are set, then it is a recovery.
495 	 */
496 	MD_RECOVERY_RUNNING,	/* a thread is running, or about to be started */
497 	MD_RECOVERY_SYNC,	/* actually doing a resync, not a recovery */
498 	MD_RECOVERY_RECOVER,	/* doing recovery, or need to try it. */
499 	MD_RECOVERY_INTR,	/* resync needs to be aborted for some reason */
500 	MD_RECOVERY_DONE,	/* thread is done and is waiting to be reaped */
501 	MD_RECOVERY_NEEDED,	/* we might need to start a resync/recover */
502 	MD_RECOVERY_REQUESTED,	/* user-space has requested a sync (used with SYNC) */
503 	MD_RECOVERY_CHECK,	/* user-space request for check-only, no repair */
504 	MD_RECOVERY_RESHAPE,	/* A reshape is happening */
505 	MD_RECOVERY_FROZEN,	/* User request to abort, and not restart, any action */
506 	MD_RECOVERY_ERROR,	/* sync-action interrupted because io-error */
507 	MD_RECOVERY_WAIT,	/* waiting for pers->start() to finish */
508 	MD_RESYNCING_REMOTE,	/* remote node is running resync thread */
509 };
510 
511 static inline int __must_check mddev_lock(struct mddev *mddev)
512 {
513 	return mutex_lock_interruptible(&mddev->reconfig_mutex);
514 }
515 
516 /* Sometimes we need to take the lock in a situation where
517  * failure due to interrupts is not acceptable.
518  */
519 static inline void mddev_lock_nointr(struct mddev *mddev)
520 {
521 	mutex_lock(&mddev->reconfig_mutex);
522 }
523 
524 static inline int mddev_trylock(struct mddev *mddev)
525 {
526 	return mutex_trylock(&mddev->reconfig_mutex);
527 }
528 extern void mddev_unlock(struct mddev *mddev);
529 
530 static inline void md_sync_acct(struct block_device *bdev, unsigned long nr_sectors)
531 {
532 	atomic_add(nr_sectors, &bdev->bd_contains->bd_disk->sync_io);
533 }
534 
535 static inline void md_sync_acct_bio(struct bio *bio, unsigned long nr_sectors)
536 {
537 	atomic_add(nr_sectors, &bio->bi_disk->sync_io);
538 }
539 
540 struct md_personality
541 {
542 	char *name;
543 	int level;
544 	struct list_head list;
545 	struct module *owner;
546 	bool (*make_request)(struct mddev *mddev, struct bio *bio);
547 	/*
548 	 * start up works that do NOT require md_thread. tasks that
549 	 * requires md_thread should go into start()
550 	 */
551 	int (*run)(struct mddev *mddev);
552 	/* start up works that require md threads */
553 	int (*start)(struct mddev *mddev);
554 	void (*free)(struct mddev *mddev, void *priv);
555 	void (*status)(struct seq_file *seq, struct mddev *mddev);
556 	/* error_handler must set ->faulty and clear ->in_sync
557 	 * if appropriate, and should abort recovery if needed
558 	 */
559 	void (*error_handler)(struct mddev *mddev, struct md_rdev *rdev);
560 	int (*hot_add_disk) (struct mddev *mddev, struct md_rdev *rdev);
561 	int (*hot_remove_disk) (struct mddev *mddev, struct md_rdev *rdev);
562 	int (*spare_active) (struct mddev *mddev);
563 	sector_t (*sync_request)(struct mddev *mddev, sector_t sector_nr, int *skipped);
564 	int (*resize) (struct mddev *mddev, sector_t sectors);
565 	sector_t (*size) (struct mddev *mddev, sector_t sectors, int raid_disks);
566 	int (*check_reshape) (struct mddev *mddev);
567 	int (*start_reshape) (struct mddev *mddev);
568 	void (*finish_reshape) (struct mddev *mddev);
569 	void (*update_reshape_pos) (struct mddev *mddev);
570 	/* quiesce suspends or resumes internal processing.
571 	 * 1 - stop new actions and wait for action io to complete
572 	 * 0 - return to normal behaviour
573 	 */
574 	void (*quiesce) (struct mddev *mddev, int quiesce);
575 	/* takeover is used to transition an array from one
576 	 * personality to another.  The new personality must be able
577 	 * to handle the data in the current layout.
578 	 * e.g. 2drive raid1 -> 2drive raid5
579 	 *      ndrive raid5 -> degraded n+1drive raid6 with special layout
580 	 * If the takeover succeeds, a new 'private' structure is returned.
581 	 * This needs to be installed and then ->run used to activate the
582 	 * array.
583 	 */
584 	void *(*takeover) (struct mddev *mddev);
585 	/* congested implements bdi.congested_fn().
586 	 * Will not be called while array is 'suspended' */
587 	int (*congested)(struct mddev *mddev, int bits);
588 	/* Changes the consistency policy of an active array. */
589 	int (*change_consistency_policy)(struct mddev *mddev, const char *buf);
590 };
591 
592 struct md_sysfs_entry {
593 	struct attribute attr;
594 	ssize_t (*show)(struct mddev *, char *);
595 	ssize_t (*store)(struct mddev *, const char *, size_t);
596 };
597 extern struct attribute_group md_bitmap_group;
598 
599 static inline struct kernfs_node *sysfs_get_dirent_safe(struct kernfs_node *sd, char *name)
600 {
601 	if (sd)
602 		return sysfs_get_dirent(sd, name);
603 	return sd;
604 }
605 static inline void sysfs_notify_dirent_safe(struct kernfs_node *sd)
606 {
607 	if (sd)
608 		sysfs_notify_dirent(sd);
609 }
610 
611 static inline char * mdname (struct mddev * mddev)
612 {
613 	return mddev->gendisk ? mddev->gendisk->disk_name : "mdX";
614 }
615 
616 static inline int sysfs_link_rdev(struct mddev *mddev, struct md_rdev *rdev)
617 {
618 	char nm[20];
619 	if (!test_bit(Replacement, &rdev->flags) &&
620 	    !test_bit(Journal, &rdev->flags) &&
621 	    mddev->kobj.sd) {
622 		sprintf(nm, "rd%d", rdev->raid_disk);
623 		return sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
624 	} else
625 		return 0;
626 }
627 
628 static inline void sysfs_unlink_rdev(struct mddev *mddev, struct md_rdev *rdev)
629 {
630 	char nm[20];
631 	if (!test_bit(Replacement, &rdev->flags) &&
632 	    !test_bit(Journal, &rdev->flags) &&
633 	    mddev->kobj.sd) {
634 		sprintf(nm, "rd%d", rdev->raid_disk);
635 		sysfs_remove_link(&mddev->kobj, nm);
636 	}
637 }
638 
639 /*
640  * iterates through some rdev ringlist. It's safe to remove the
641  * current 'rdev'. Dont touch 'tmp' though.
642  */
643 #define rdev_for_each_list(rdev, tmp, head)				\
644 	list_for_each_entry_safe(rdev, tmp, head, same_set)
645 
646 /*
647  * iterates through the 'same array disks' ringlist
648  */
649 #define rdev_for_each(rdev, mddev)				\
650 	list_for_each_entry(rdev, &((mddev)->disks), same_set)
651 
652 #define rdev_for_each_safe(rdev, tmp, mddev)				\
653 	list_for_each_entry_safe(rdev, tmp, &((mddev)->disks), same_set)
654 
655 #define rdev_for_each_rcu(rdev, mddev)				\
656 	list_for_each_entry_rcu(rdev, &((mddev)->disks), same_set)
657 
658 struct md_thread {
659 	void			(*run) (struct md_thread *thread);
660 	struct mddev		*mddev;
661 	wait_queue_head_t	wqueue;
662 	unsigned long		flags;
663 	struct task_struct	*tsk;
664 	unsigned long		timeout;
665 	void			*private;
666 };
667 
668 #define THREAD_WAKEUP  0
669 
670 static inline void safe_put_page(struct page *p)
671 {
672 	if (p) put_page(p);
673 }
674 
675 extern int register_md_personality(struct md_personality *p);
676 extern int unregister_md_personality(struct md_personality *p);
677 extern int register_md_cluster_operations(struct md_cluster_operations *ops,
678 		struct module *module);
679 extern int unregister_md_cluster_operations(void);
680 extern int md_setup_cluster(struct mddev *mddev, int nodes);
681 extern void md_cluster_stop(struct mddev *mddev);
682 extern struct md_thread *md_register_thread(
683 	void (*run)(struct md_thread *thread),
684 	struct mddev *mddev,
685 	const char *name);
686 extern void md_unregister_thread(struct md_thread **threadp);
687 extern void md_wakeup_thread(struct md_thread *thread);
688 extern void md_check_recovery(struct mddev *mddev);
689 extern void md_reap_sync_thread(struct mddev *mddev);
690 extern int mddev_init_writes_pending(struct mddev *mddev);
691 extern bool md_write_start(struct mddev *mddev, struct bio *bi);
692 extern void md_write_inc(struct mddev *mddev, struct bio *bi);
693 extern void md_write_end(struct mddev *mddev);
694 extern void md_done_sync(struct mddev *mddev, int blocks, int ok);
695 extern void md_error(struct mddev *mddev, struct md_rdev *rdev);
696 extern void md_finish_reshape(struct mddev *mddev);
697 
698 extern int mddev_congested(struct mddev *mddev, int bits);
699 extern void md_flush_request(struct mddev *mddev, struct bio *bio);
700 extern void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
701 			   sector_t sector, int size, struct page *page);
702 extern int md_super_wait(struct mddev *mddev);
703 extern int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
704 			struct page *page, int op, int op_flags,
705 			bool metadata_op);
706 extern void md_do_sync(struct md_thread *thread);
707 extern void md_new_event(struct mddev *mddev);
708 extern void md_allow_write(struct mddev *mddev);
709 extern void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev);
710 extern void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors);
711 extern int md_check_no_bitmap(struct mddev *mddev);
712 extern int md_integrity_register(struct mddev *mddev);
713 extern int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev);
714 extern int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale);
715 
716 extern void mddev_init(struct mddev *mddev);
717 extern int md_run(struct mddev *mddev);
718 extern int md_start(struct mddev *mddev);
719 extern void md_stop(struct mddev *mddev);
720 extern void md_stop_writes(struct mddev *mddev);
721 extern int md_rdev_init(struct md_rdev *rdev);
722 extern void md_rdev_clear(struct md_rdev *rdev);
723 
724 extern void md_handle_request(struct mddev *mddev, struct bio *bio);
725 extern void mddev_suspend(struct mddev *mddev);
726 extern void mddev_resume(struct mddev *mddev);
727 extern struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
728 				   struct mddev *mddev);
729 
730 extern void md_reload_sb(struct mddev *mddev, int raid_disk);
731 extern void md_update_sb(struct mddev *mddev, int force);
732 extern void md_kick_rdev_from_array(struct md_rdev * rdev);
733 extern void mddev_create_wb_pool(struct mddev *mddev, struct md_rdev *rdev,
734 				 bool is_suspend);
735 struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr);
736 struct md_rdev *md_find_rdev_rcu(struct mddev *mddev, dev_t dev);
737 
738 static inline void rdev_dec_pending(struct md_rdev *rdev, struct mddev *mddev)
739 {
740 	int faulty = test_bit(Faulty, &rdev->flags);
741 	if (atomic_dec_and_test(&rdev->nr_pending) && faulty) {
742 		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
743 		md_wakeup_thread(mddev->thread);
744 	}
745 }
746 
747 extern struct md_cluster_operations *md_cluster_ops;
748 static inline int mddev_is_clustered(struct mddev *mddev)
749 {
750 	return mddev->cluster_info && mddev->bitmap_info.nodes > 1;
751 }
752 
753 /* clear unsupported mddev_flags */
754 static inline void mddev_clear_unsupported_flags(struct mddev *mddev,
755 	unsigned long unsupported_flags)
756 {
757 	mddev->flags &= ~unsupported_flags;
758 }
759 
760 static inline void mddev_check_writesame(struct mddev *mddev, struct bio *bio)
761 {
762 	if (bio_op(bio) == REQ_OP_WRITE_SAME &&
763 	    !bio->bi_disk->queue->limits.max_write_same_sectors)
764 		mddev->queue->limits.max_write_same_sectors = 0;
765 }
766 
767 static inline void mddev_check_write_zeroes(struct mddev *mddev, struct bio *bio)
768 {
769 	if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
770 	    !bio->bi_disk->queue->limits.max_write_zeroes_sectors)
771 		mddev->queue->limits.max_write_zeroes_sectors = 0;
772 }
773 #endif /* _MD_MD_H */
774