xref: /linux/drivers/md/md.h (revision 019b383d1132e4051de0d2e43254454b86538cf4)
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 /*
37  * The struct embedded in rdev is used to serialize IO.
38  */
39 struct serial_in_rdev {
40 	struct rb_root_cached serial_rb;
41 	spinlock_t serial_lock;
42 	wait_queue_head_t serial_io_wait;
43 };
44 
45 /*
46  * MD's 'extended' device
47  */
48 struct md_rdev {
49 	struct list_head same_set;	/* RAID devices within the same set */
50 
51 	sector_t sectors;		/* Device size (in 512bytes sectors) */
52 	struct mddev *mddev;		/* RAID array if running */
53 	int last_events;		/* IO event timestamp */
54 
55 	/*
56 	 * If meta_bdev is non-NULL, it means that a separate device is
57 	 * being used to store the metadata (superblock/bitmap) which
58 	 * would otherwise be contained on the same device as the data (bdev).
59 	 */
60 	struct block_device *meta_bdev;
61 	struct block_device *bdev;	/* block device handle */
62 
63 	struct page	*sb_page, *bb_page;
64 	int		sb_loaded;
65 	__u64		sb_events;
66 	sector_t	data_offset;	/* start of data in array */
67 	sector_t	new_data_offset;/* only relevant while reshaping */
68 	sector_t	sb_start;	/* offset of the super block (in 512byte sectors) */
69 	int		sb_size;	/* bytes in the superblock */
70 	int		preferred_minor;	/* autorun support */
71 
72 	struct kobject	kobj;
73 
74 	/* A device can be in one of three states based on two flags:
75 	 * Not working:   faulty==1 in_sync==0
76 	 * Fully working: faulty==0 in_sync==1
77 	 * Working, but not
78 	 * in sync with array
79 	 *                faulty==0 in_sync==0
80 	 *
81 	 * It can never have faulty==1, in_sync==1
82 	 * This reduces the burden of testing multiple flags in many cases
83 	 */
84 
85 	unsigned long	flags;	/* bit set of 'enum flag_bits' bits. */
86 	wait_queue_head_t blocked_wait;
87 
88 	int desc_nr;			/* descriptor index in the superblock */
89 	int raid_disk;			/* role of device in array */
90 	int new_raid_disk;		/* role that the device will have in
91 					 * the array after a level-change completes.
92 					 */
93 	int saved_raid_disk;		/* role that device used to have in the
94 					 * array and could again if we did a partial
95 					 * resync from the bitmap
96 					 */
97 	union {
98 		sector_t recovery_offset;/* If this device has been partially
99 					 * recovered, this is where we were
100 					 * up to.
101 					 */
102 		sector_t journal_tail;	/* If this device is a journal device,
103 					 * this is the journal tail (journal
104 					 * recovery start point)
105 					 */
106 	};
107 
108 	atomic_t	nr_pending;	/* number of pending requests.
109 					 * only maintained for arrays that
110 					 * support hot removal
111 					 */
112 	atomic_t	read_errors;	/* number of consecutive read errors that
113 					 * we have tried to ignore.
114 					 */
115 	time64_t	last_read_error;	/* monotonic time since our
116 						 * last read error
117 						 */
118 	atomic_t	corrected_errors; /* number of corrected read errors,
119 					   * for reporting to userspace and storing
120 					   * in superblock.
121 					   */
122 
123 	struct serial_in_rdev *serial;  /* used for raid1 io serialization */
124 
125 	struct kernfs_node *sysfs_state; /* handle for 'state'
126 					   * sysfs entry */
127 	/* handle for 'unacknowledged_bad_blocks' sysfs dentry */
128 	struct kernfs_node *sysfs_unack_badblocks;
129 	/* handle for 'bad_blocks' sysfs dentry */
130 	struct kernfs_node *sysfs_badblocks;
131 	struct badblocks badblocks;
132 
133 	struct {
134 		short offset;	/* Offset from superblock to start of PPL.
135 				 * Not used by external metadata. */
136 		unsigned int size;	/* Size in sectors of the PPL space */
137 		sector_t sector;	/* First sector of the PPL space */
138 	} ppl;
139 };
140 enum flag_bits {
141 	Faulty,			/* device is known to have a fault */
142 	In_sync,		/* device is in_sync with rest of array */
143 	Bitmap_sync,		/* ..actually, not quite In_sync.  Need a
144 				 * bitmap-based recovery to get fully in sync.
145 				 * The bit is only meaningful before device
146 				 * has been passed to pers->hot_add_disk.
147 				 */
148 	WriteMostly,		/* Avoid reading if at all possible */
149 	AutoDetected,		/* added by auto-detect */
150 	Blocked,		/* An error occurred but has not yet
151 				 * been acknowledged by the metadata
152 				 * handler, so don't allow writes
153 				 * until it is cleared */
154 	WriteErrorSeen,		/* A write error has been seen on this
155 				 * device
156 				 */
157 	FaultRecorded,		/* Intermediate state for clearing
158 				 * Blocked.  The Fault is/will-be
159 				 * recorded in the metadata, but that
160 				 * metadata hasn't been stored safely
161 				 * on disk yet.
162 				 */
163 	BlockedBadBlocks,	/* A writer is blocked because they
164 				 * found an unacknowledged bad-block.
165 				 * This can safely be cleared at any
166 				 * time, and the writer will re-check.
167 				 * It may be set at any time, and at
168 				 * worst the writer will timeout and
169 				 * re-check.  So setting it as
170 				 * accurately as possible is good, but
171 				 * not absolutely critical.
172 				 */
173 	WantReplacement,	/* This device is a candidate to be
174 				 * hot-replaced, either because it has
175 				 * reported some faults, or because
176 				 * of explicit request.
177 				 */
178 	Replacement,		/* This device is a replacement for
179 				 * a want_replacement device with same
180 				 * raid_disk number.
181 				 */
182 	Candidate,		/* For clustered environments only:
183 				 * This device is seen locally but not
184 				 * by the whole cluster
185 				 */
186 	Journal,		/* This device is used as journal for
187 				 * raid-5/6.
188 				 * Usually, this device should be faster
189 				 * than other devices in the array
190 				 */
191 	ClusterRemove,
192 	RemoveSynchronized,	/* synchronize_rcu() was called after
193 				 * this device was known to be faulty,
194 				 * so it is safe to remove without
195 				 * another synchronize_rcu() call.
196 				 */
197 	ExternalBbl,            /* External metadata provides bad
198 				 * block management for a disk
199 				 */
200 	FailFast,		/* Minimal retries should be attempted on
201 				 * this device, so use REQ_FAILFAST_DEV.
202 				 * Also don't try to repair failed reads.
203 				 * It is expects that no bad block log
204 				 * is present.
205 				 */
206 	LastDev,		/* Seems to be the last working dev as
207 				 * it didn't fail, so don't use FailFast
208 				 * any more for metadata
209 				 */
210 	CollisionCheck,		/*
211 				 * check if there is collision between raid1
212 				 * serial bios.
213 				 */
214 };
215 
216 static inline int is_badblock(struct md_rdev *rdev, sector_t s, int sectors,
217 			      sector_t *first_bad, int *bad_sectors)
218 {
219 	if (unlikely(rdev->badblocks.count)) {
220 		int rv = badblocks_check(&rdev->badblocks, rdev->data_offset + s,
221 					sectors,
222 					first_bad, bad_sectors);
223 		if (rv)
224 			*first_bad -= rdev->data_offset;
225 		return rv;
226 	}
227 	return 0;
228 }
229 extern int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
230 			      int is_new);
231 extern int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
232 				int is_new);
233 struct md_cluster_info;
234 
235 /**
236  * enum mddev_flags - md device flags.
237  * @MD_ARRAY_FIRST_USE: First use of array, needs initialization.
238  * @MD_CLOSING: If set, we are closing the array, do not open it then.
239  * @MD_JOURNAL_CLEAN: A raid with journal is already clean.
240  * @MD_HAS_JOURNAL: The raid array has journal feature set.
241  * @MD_CLUSTER_RESYNC_LOCKED: cluster raid only, which means node, already took
242  *			       resync lock, need to release the lock.
243  * @MD_FAILFAST_SUPPORTED: Using MD_FAILFAST on metadata writes is supported as
244  *			    calls to md_error() will never cause the array to
245  *			    become failed.
246  * @MD_HAS_PPL:  The raid array has PPL feature set.
247  * @MD_HAS_MULTIPLE_PPLS: The raid array has multiple PPLs feature set.
248  * @MD_ALLOW_SB_UPDATE: md_check_recovery is allowed to update the metadata
249  *			 without taking reconfig_mutex.
250  * @MD_UPDATING_SB: md_check_recovery is updating the metadata without
251  *		     explicitly holding reconfig_mutex.
252  * @MD_NOT_READY: do_md_run() is active, so 'array_state', ust not report that
253  *		   array is ready yet.
254  * @MD_BROKEN: This is used to stop writes and mark array as failed.
255  * @MD_DELETED: This device is being deleted
256  *
257  * change UNSUPPORTED_MDDEV_FLAGS for each array type if new flag is added
258  */
259 enum mddev_flags {
260 	MD_ARRAY_FIRST_USE,
261 	MD_CLOSING,
262 	MD_JOURNAL_CLEAN,
263 	MD_HAS_JOURNAL,
264 	MD_CLUSTER_RESYNC_LOCKED,
265 	MD_FAILFAST_SUPPORTED,
266 	MD_HAS_PPL,
267 	MD_HAS_MULTIPLE_PPLS,
268 	MD_ALLOW_SB_UPDATE,
269 	MD_UPDATING_SB,
270 	MD_NOT_READY,
271 	MD_BROKEN,
272 	MD_DELETED,
273 };
274 
275 enum mddev_sb_flags {
276 	MD_SB_CHANGE_DEVS,		/* Some device status has changed */
277 	MD_SB_CHANGE_CLEAN,	/* transition to or from 'clean' */
278 	MD_SB_CHANGE_PENDING,	/* switch from 'clean' to 'active' in progress */
279 	MD_SB_NEED_REWRITE,	/* metadata write needs to be repeated */
280 };
281 
282 #define NR_SERIAL_INFOS		8
283 /* record current range of serialize IOs */
284 struct serial_info {
285 	struct rb_node node;
286 	sector_t start;		/* start sector of rb node */
287 	sector_t last;		/* end sector of rb node */
288 	sector_t _subtree_last; /* highest sector in subtree of rb node */
289 };
290 
291 /*
292  * mddev->curr_resync stores the current sector of the resync but
293  * also has some overloaded values.
294  */
295 enum {
296 	/* No resync in progress */
297 	MD_RESYNC_NONE = 0,
298 	/* Yielded to allow another conflicting resync to commence */
299 	MD_RESYNC_YIELDED = 1,
300 	/* Delayed to check that there is no conflict with another sync */
301 	MD_RESYNC_DELAYED = 2,
302 	/* Any value greater than or equal to this is in an active resync */
303 	MD_RESYNC_ACTIVE = 3,
304 };
305 
306 struct mddev {
307 	void				*private;
308 	struct md_personality		*pers;
309 	dev_t				unit;
310 	int				md_minor;
311 	struct list_head		disks;
312 	unsigned long			flags;
313 	unsigned long			sb_flags;
314 
315 	int				suspended;
316 	struct percpu_ref		active_io;
317 	int				ro;
318 	int				sysfs_active; /* set when sysfs deletes
319 						       * are happening, so run/
320 						       * takeover/stop are not safe
321 						       */
322 	struct gendisk			*gendisk;
323 
324 	struct kobject			kobj;
325 	int				hold_active;
326 #define	UNTIL_IOCTL	1
327 #define	UNTIL_STOP	2
328 
329 	/* Superblock information */
330 	int				major_version,
331 					minor_version,
332 					patch_version;
333 	int				persistent;
334 	int				external;	/* metadata is
335 							 * managed externally */
336 	char				metadata_type[17]; /* externally set*/
337 	int				chunk_sectors;
338 	time64_t			ctime, utime;
339 	int				level, layout;
340 	char				clevel[16];
341 	int				raid_disks;
342 	int				max_disks;
343 	sector_t			dev_sectors;	/* used size of
344 							 * component devices */
345 	sector_t			array_sectors; /* exported array size */
346 	int				external_size; /* size managed
347 							* externally */
348 	__u64				events;
349 	/* If the last 'event' was simply a clean->dirty transition, and
350 	 * we didn't write it to the spares, then it is safe and simple
351 	 * to just decrement the event count on a dirty->clean transition.
352 	 * So we record that possibility here.
353 	 */
354 	int				can_decrease_events;
355 
356 	char				uuid[16];
357 
358 	/* If the array is being reshaped, we need to record the
359 	 * new shape and an indication of where we are up to.
360 	 * This is written to the superblock.
361 	 * If reshape_position is MaxSector, then no reshape is happening (yet).
362 	 */
363 	sector_t			reshape_position;
364 	int				delta_disks, new_level, new_layout;
365 	int				new_chunk_sectors;
366 	int				reshape_backwards;
367 
368 	struct md_thread __rcu		*thread;	/* management thread */
369 	struct md_thread __rcu		*sync_thread;	/* doing resync or reconstruct */
370 
371 	/* 'last_sync_action' is initialized to "none".  It is set when a
372 	 * sync operation (i.e "data-check", "requested-resync", "resync",
373 	 * "recovery", or "reshape") is started.  It holds this value even
374 	 * when the sync thread is "frozen" (interrupted) or "idle" (stopped
375 	 * or finished).  It is overwritten when a new sync operation is begun.
376 	 */
377 	char				*last_sync_action;
378 	sector_t			curr_resync;	/* last block scheduled */
379 	/* As resync requests can complete out of order, we cannot easily track
380 	 * how much resync has been completed.  So we occasionally pause until
381 	 * everything completes, then set curr_resync_completed to curr_resync.
382 	 * As such it may be well behind the real resync mark, but it is a value
383 	 * we are certain of.
384 	 */
385 	sector_t			curr_resync_completed;
386 	unsigned long			resync_mark;	/* a recent timestamp */
387 	sector_t			resync_mark_cnt;/* blocks written at resync_mark */
388 	sector_t			curr_mark_cnt; /* blocks scheduled now */
389 
390 	sector_t			resync_max_sectors; /* may be set by personality */
391 
392 	atomic64_t			resync_mismatches; /* count of sectors where
393 							    * parity/replica mismatch found
394 							    */
395 
396 	/* allow user-space to request suspension of IO to regions of the array */
397 	sector_t			suspend_lo;
398 	sector_t			suspend_hi;
399 	/* if zero, use the system-wide default */
400 	int				sync_speed_min;
401 	int				sync_speed_max;
402 
403 	/* resync even though the same disks are shared among md-devices */
404 	int				parallel_resync;
405 
406 	int				ok_start_degraded;
407 
408 	unsigned long			recovery;
409 	/* If a RAID personality determines that recovery (of a particular
410 	 * device) will fail due to a read error on the source device, it
411 	 * takes a copy of this number and does not attempt recovery again
412 	 * until this number changes.
413 	 */
414 	int				recovery_disabled;
415 
416 	int				in_sync;	/* know to not need resync */
417 	/* 'open_mutex' avoids races between 'md_open' and 'do_md_stop', so
418 	 * that we are never stopping an array while it is open.
419 	 * 'reconfig_mutex' protects all other reconfiguration.
420 	 * These locks are separate due to conflicting interactions
421 	 * with disk->open_mutex.
422 	 * Lock ordering is:
423 	 *  reconfig_mutex -> disk->open_mutex
424 	 *  disk->open_mutex -> open_mutex:  e.g. __blkdev_get -> md_open
425 	 */
426 	struct mutex			open_mutex;
427 	struct mutex			reconfig_mutex;
428 	atomic_t			active;		/* general refcount */
429 	atomic_t			openers;	/* number of active opens */
430 
431 	int				changed;	/* True if we might need to
432 							 * reread partition info */
433 	int				degraded;	/* whether md should consider
434 							 * adding a spare
435 							 */
436 
437 	atomic_t			recovery_active; /* blocks scheduled, but not written */
438 	wait_queue_head_t		recovery_wait;
439 	sector_t			recovery_cp;
440 	sector_t			resync_min;	/* user requested sync
441 							 * starts here */
442 	sector_t			resync_max;	/* resync should pause
443 							 * when it gets here */
444 
445 	struct kernfs_node		*sysfs_state;	/* handle for 'array_state'
446 							 * file in sysfs.
447 							 */
448 	struct kernfs_node		*sysfs_action;  /* handle for 'sync_action' */
449 	struct kernfs_node		*sysfs_completed;	/*handle for 'sync_completed' */
450 	struct kernfs_node		*sysfs_degraded;	/*handle for 'degraded' */
451 	struct kernfs_node		*sysfs_level;		/*handle for 'level' */
452 
453 	struct work_struct del_work;	/* used for delayed sysfs removal */
454 
455 	/* "lock" protects:
456 	 *   flush_bio transition from NULL to !NULL
457 	 *   rdev superblocks, events
458 	 *   clearing MD_CHANGE_*
459 	 *   in_sync - and related safemode and MD_CHANGE changes
460 	 *   pers (also protected by reconfig_mutex and pending IO).
461 	 *   clearing ->bitmap
462 	 *   clearing ->bitmap_info.file
463 	 *   changing ->resync_{min,max}
464 	 *   setting MD_RECOVERY_RUNNING (which interacts with resync_{min,max})
465 	 */
466 	spinlock_t			lock;
467 	wait_queue_head_t		sb_wait;	/* for waiting on superblock updates */
468 	atomic_t			pending_writes;	/* number of active superblock writes */
469 
470 	unsigned int			safemode;	/* if set, update "clean" superblock
471 							 * when no writes pending.
472 							 */
473 	unsigned int			safemode_delay;
474 	struct timer_list		safemode_timer;
475 	struct percpu_ref		writes_pending;
476 	int				sync_checkers;	/* # of threads checking writes_pending */
477 	struct request_queue		*queue;	/* for plugging ... */
478 
479 	struct bitmap			*bitmap; /* the bitmap for the device */
480 	struct {
481 		struct file		*file; /* the bitmap file */
482 		loff_t			offset; /* offset from superblock of
483 						 * start of bitmap. May be
484 						 * negative, but not '0'
485 						 * For external metadata, offset
486 						 * from start of device.
487 						 */
488 		unsigned long		space; /* space available at this offset */
489 		loff_t			default_offset; /* this is the offset to use when
490 							 * hot-adding a bitmap.  It should
491 							 * eventually be settable by sysfs.
492 							 */
493 		unsigned long		default_space; /* space available at
494 							* default offset */
495 		struct mutex		mutex;
496 		unsigned long		chunksize;
497 		unsigned long		daemon_sleep; /* how many jiffies between updates? */
498 		unsigned long		max_write_behind; /* write-behind mode */
499 		int			external;
500 		int			nodes; /* Maximum number of nodes in the cluster */
501 		char                    cluster_name[64]; /* Name of the cluster */
502 	} bitmap_info;
503 
504 	atomic_t			max_corr_read_errors; /* max read retries */
505 	struct list_head		all_mddevs;
506 
507 	const struct attribute_group	*to_remove;
508 
509 	struct bio_set			bio_set;
510 	struct bio_set			sync_set; /* for sync operations like
511 						   * metadata and bitmap writes
512 						   */
513 	struct bio_set			io_clone_set;
514 
515 	/* Generic flush handling.
516 	 * The last to finish preflush schedules a worker to submit
517 	 * the rest of the request (without the REQ_PREFLUSH flag).
518 	 */
519 	struct bio *flush_bio;
520 	atomic_t flush_pending;
521 	ktime_t start_flush, prev_flush_start; /* prev_flush_start is when the previous completed
522 						* flush was started.
523 						*/
524 	struct work_struct flush_work;
525 	struct work_struct event_work;	/* used by dm to report failure event */
526 	mempool_t *serial_info_pool;
527 	void (*sync_super)(struct mddev *mddev, struct md_rdev *rdev);
528 	struct md_cluster_info		*cluster_info;
529 	unsigned int			good_device_nr;	/* good device num within cluster raid */
530 	unsigned int			noio_flag; /* for memalloc scope API */
531 
532 	/*
533 	 * Temporarily store rdev that will be finally removed when
534 	 * reconfig_mutex is unlocked, protected by reconfig_mutex.
535 	 */
536 	struct list_head		deleting;
537 
538 	/* Used to synchronize idle and frozen for action_store() */
539 	struct mutex			sync_mutex;
540 	/* The sequence number for sync thread */
541 	atomic_t sync_seq;
542 
543 	bool	has_superblocks:1;
544 	bool	fail_last_dev:1;
545 	bool	serialize_policy:1;
546 };
547 
548 enum recovery_flags {
549 	/*
550 	 * If neither SYNC or RESHAPE are set, then it is a recovery.
551 	 */
552 	MD_RECOVERY_RUNNING,	/* a thread is running, or about to be started */
553 	MD_RECOVERY_SYNC,	/* actually doing a resync, not a recovery */
554 	MD_RECOVERY_RECOVER,	/* doing recovery, or need to try it. */
555 	MD_RECOVERY_INTR,	/* resync needs to be aborted for some reason */
556 	MD_RECOVERY_DONE,	/* thread is done and is waiting to be reaped */
557 	MD_RECOVERY_NEEDED,	/* we might need to start a resync/recover */
558 	MD_RECOVERY_REQUESTED,	/* user-space has requested a sync (used with SYNC) */
559 	MD_RECOVERY_CHECK,	/* user-space request for check-only, no repair */
560 	MD_RECOVERY_RESHAPE,	/* A reshape is happening */
561 	MD_RECOVERY_FROZEN,	/* User request to abort, and not restart, any action */
562 	MD_RECOVERY_ERROR,	/* sync-action interrupted because io-error */
563 	MD_RECOVERY_WAIT,	/* waiting for pers->start() to finish */
564 	MD_RESYNCING_REMOTE,	/* remote node is running resync thread */
565 };
566 
567 enum md_ro_state {
568 	MD_RDWR,
569 	MD_RDONLY,
570 	MD_AUTO_READ,
571 	MD_MAX_STATE
572 };
573 
574 static inline bool md_is_rdwr(struct mddev *mddev)
575 {
576 	return (mddev->ro == MD_RDWR);
577 }
578 
579 static inline bool is_md_suspended(struct mddev *mddev)
580 {
581 	return percpu_ref_is_dying(&mddev->active_io);
582 }
583 
584 static inline int __must_check mddev_lock(struct mddev *mddev)
585 {
586 	return mutex_lock_interruptible(&mddev->reconfig_mutex);
587 }
588 
589 /* Sometimes we need to take the lock in a situation where
590  * failure due to interrupts is not acceptable.
591  */
592 static inline void mddev_lock_nointr(struct mddev *mddev)
593 {
594 	mutex_lock(&mddev->reconfig_mutex);
595 }
596 
597 static inline int mddev_trylock(struct mddev *mddev)
598 {
599 	return mutex_trylock(&mddev->reconfig_mutex);
600 }
601 extern void mddev_unlock(struct mddev *mddev);
602 
603 static inline void md_sync_acct(struct block_device *bdev, unsigned long nr_sectors)
604 {
605 	atomic_add(nr_sectors, &bdev->bd_disk->sync_io);
606 }
607 
608 static inline void md_sync_acct_bio(struct bio *bio, unsigned long nr_sectors)
609 {
610 	md_sync_acct(bio->bi_bdev, nr_sectors);
611 }
612 
613 struct md_personality
614 {
615 	char *name;
616 	int level;
617 	struct list_head list;
618 	struct module *owner;
619 	bool __must_check (*make_request)(struct mddev *mddev, struct bio *bio);
620 	/*
621 	 * start up works that do NOT require md_thread. tasks that
622 	 * requires md_thread should go into start()
623 	 */
624 	int (*run)(struct mddev *mddev);
625 	/* start up works that require md threads */
626 	int (*start)(struct mddev *mddev);
627 	void (*free)(struct mddev *mddev, void *priv);
628 	void (*status)(struct seq_file *seq, struct mddev *mddev);
629 	/* error_handler must set ->faulty and clear ->in_sync
630 	 * if appropriate, and should abort recovery if needed
631 	 */
632 	void (*error_handler)(struct mddev *mddev, struct md_rdev *rdev);
633 	int (*hot_add_disk) (struct mddev *mddev, struct md_rdev *rdev);
634 	int (*hot_remove_disk) (struct mddev *mddev, struct md_rdev *rdev);
635 	int (*spare_active) (struct mddev *mddev);
636 	sector_t (*sync_request)(struct mddev *mddev, sector_t sector_nr, int *skipped);
637 	int (*resize) (struct mddev *mddev, sector_t sectors);
638 	sector_t (*size) (struct mddev *mddev, sector_t sectors, int raid_disks);
639 	int (*check_reshape) (struct mddev *mddev);
640 	int (*start_reshape) (struct mddev *mddev);
641 	void (*finish_reshape) (struct mddev *mddev);
642 	void (*update_reshape_pos) (struct mddev *mddev);
643 	void (*prepare_suspend) (struct mddev *mddev);
644 	/* quiesce suspends or resumes internal processing.
645 	 * 1 - stop new actions and wait for action io to complete
646 	 * 0 - return to normal behaviour
647 	 */
648 	void (*quiesce) (struct mddev *mddev, int quiesce);
649 	/* takeover is used to transition an array from one
650 	 * personality to another.  The new personality must be able
651 	 * to handle the data in the current layout.
652 	 * e.g. 2drive raid1 -> 2drive raid5
653 	 *      ndrive raid5 -> degraded n+1drive raid6 with special layout
654 	 * If the takeover succeeds, a new 'private' structure is returned.
655 	 * This needs to be installed and then ->run used to activate the
656 	 * array.
657 	 */
658 	void *(*takeover) (struct mddev *mddev);
659 	/* Changes the consistency policy of an active array. */
660 	int (*change_consistency_policy)(struct mddev *mddev, const char *buf);
661 };
662 
663 struct md_sysfs_entry {
664 	struct attribute attr;
665 	ssize_t (*show)(struct mddev *, char *);
666 	ssize_t (*store)(struct mddev *, const char *, size_t);
667 };
668 extern const struct attribute_group md_bitmap_group;
669 
670 static inline struct kernfs_node *sysfs_get_dirent_safe(struct kernfs_node *sd, char *name)
671 {
672 	if (sd)
673 		return sysfs_get_dirent(sd, name);
674 	return sd;
675 }
676 static inline void sysfs_notify_dirent_safe(struct kernfs_node *sd)
677 {
678 	if (sd)
679 		sysfs_notify_dirent(sd);
680 }
681 
682 static inline char * mdname (struct mddev * mddev)
683 {
684 	return mddev->gendisk ? mddev->gendisk->disk_name : "mdX";
685 }
686 
687 static inline int sysfs_link_rdev(struct mddev *mddev, struct md_rdev *rdev)
688 {
689 	char nm[20];
690 	if (!test_bit(Replacement, &rdev->flags) &&
691 	    !test_bit(Journal, &rdev->flags) &&
692 	    mddev->kobj.sd) {
693 		sprintf(nm, "rd%d", rdev->raid_disk);
694 		return sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
695 	} else
696 		return 0;
697 }
698 
699 static inline void sysfs_unlink_rdev(struct mddev *mddev, struct md_rdev *rdev)
700 {
701 	char nm[20];
702 	if (!test_bit(Replacement, &rdev->flags) &&
703 	    !test_bit(Journal, &rdev->flags) &&
704 	    mddev->kobj.sd) {
705 		sprintf(nm, "rd%d", rdev->raid_disk);
706 		sysfs_remove_link(&mddev->kobj, nm);
707 	}
708 }
709 
710 /*
711  * iterates through some rdev ringlist. It's safe to remove the
712  * current 'rdev'. Dont touch 'tmp' though.
713  */
714 #define rdev_for_each_list(rdev, tmp, head)				\
715 	list_for_each_entry_safe(rdev, tmp, head, same_set)
716 
717 /*
718  * iterates through the 'same array disks' ringlist
719  */
720 #define rdev_for_each(rdev, mddev)				\
721 	list_for_each_entry(rdev, &((mddev)->disks), same_set)
722 
723 #define rdev_for_each_safe(rdev, tmp, mddev)				\
724 	list_for_each_entry_safe(rdev, tmp, &((mddev)->disks), same_set)
725 
726 #define rdev_for_each_rcu(rdev, mddev)				\
727 	list_for_each_entry_rcu(rdev, &((mddev)->disks), same_set)
728 
729 struct md_thread {
730 	void			(*run) (struct md_thread *thread);
731 	struct mddev		*mddev;
732 	wait_queue_head_t	wqueue;
733 	unsigned long		flags;
734 	struct task_struct	*tsk;
735 	unsigned long		timeout;
736 	void			*private;
737 };
738 
739 struct md_io_clone {
740 	struct mddev	*mddev;
741 	struct bio	*orig_bio;
742 	unsigned long	start_time;
743 	struct bio	bio_clone;
744 };
745 
746 #define THREAD_WAKEUP  0
747 
748 static inline void safe_put_page(struct page *p)
749 {
750 	if (p) put_page(p);
751 }
752 
753 extern int register_md_personality(struct md_personality *p);
754 extern int unregister_md_personality(struct md_personality *p);
755 extern int register_md_cluster_operations(struct md_cluster_operations *ops,
756 		struct module *module);
757 extern int unregister_md_cluster_operations(void);
758 extern int md_setup_cluster(struct mddev *mddev, int nodes);
759 extern void md_cluster_stop(struct mddev *mddev);
760 extern struct md_thread *md_register_thread(
761 	void (*run)(struct md_thread *thread),
762 	struct mddev *mddev,
763 	const char *name);
764 extern void md_unregister_thread(struct mddev *mddev, struct md_thread __rcu **threadp);
765 extern void md_wakeup_thread(struct md_thread __rcu *thread);
766 extern void md_check_recovery(struct mddev *mddev);
767 extern void md_reap_sync_thread(struct mddev *mddev);
768 extern int mddev_init_writes_pending(struct mddev *mddev);
769 extern bool md_write_start(struct mddev *mddev, struct bio *bi);
770 extern void md_write_inc(struct mddev *mddev, struct bio *bi);
771 extern void md_write_end(struct mddev *mddev);
772 extern void md_done_sync(struct mddev *mddev, int blocks, int ok);
773 extern void md_error(struct mddev *mddev, struct md_rdev *rdev);
774 extern void md_finish_reshape(struct mddev *mddev);
775 void md_submit_discard_bio(struct mddev *mddev, struct md_rdev *rdev,
776 			struct bio *bio, sector_t start, sector_t size);
777 void md_account_bio(struct mddev *mddev, struct bio **bio);
778 
779 extern bool __must_check md_flush_request(struct mddev *mddev, struct bio *bio);
780 extern void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
781 			   sector_t sector, int size, struct page *page);
782 extern int md_super_wait(struct mddev *mddev);
783 extern int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
784 		struct page *page, blk_opf_t opf, bool metadata_op);
785 extern void md_do_sync(struct md_thread *thread);
786 extern void md_new_event(void);
787 extern void md_allow_write(struct mddev *mddev);
788 extern void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev);
789 extern void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors);
790 extern int md_check_no_bitmap(struct mddev *mddev);
791 extern int md_integrity_register(struct mddev *mddev);
792 extern int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev);
793 extern int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale);
794 
795 extern void mddev_init(struct mddev *mddev);
796 struct mddev *md_alloc(dev_t dev, char *name);
797 void mddev_put(struct mddev *mddev);
798 extern int md_run(struct mddev *mddev);
799 extern int md_start(struct mddev *mddev);
800 extern void md_stop(struct mddev *mddev);
801 extern void md_stop_writes(struct mddev *mddev);
802 extern int md_rdev_init(struct md_rdev *rdev);
803 extern void md_rdev_clear(struct md_rdev *rdev);
804 
805 extern void md_handle_request(struct mddev *mddev, struct bio *bio);
806 extern void mddev_suspend(struct mddev *mddev);
807 extern void mddev_resume(struct mddev *mddev);
808 
809 extern void md_reload_sb(struct mddev *mddev, int raid_disk);
810 extern void md_update_sb(struct mddev *mddev, int force);
811 extern void mddev_create_serial_pool(struct mddev *mddev, struct md_rdev *rdev,
812 				     bool is_suspend);
813 extern void mddev_destroy_serial_pool(struct mddev *mddev, struct md_rdev *rdev,
814 				      bool is_suspend);
815 struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr);
816 struct md_rdev *md_find_rdev_rcu(struct mddev *mddev, dev_t dev);
817 
818 static inline bool is_rdev_broken(struct md_rdev *rdev)
819 {
820 	return !disk_live(rdev->bdev->bd_disk);
821 }
822 
823 static inline void rdev_dec_pending(struct md_rdev *rdev, struct mddev *mddev)
824 {
825 	int faulty = test_bit(Faulty, &rdev->flags);
826 	if (atomic_dec_and_test(&rdev->nr_pending) && faulty) {
827 		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
828 		md_wakeup_thread(mddev->thread);
829 	}
830 }
831 
832 extern struct md_cluster_operations *md_cluster_ops;
833 static inline int mddev_is_clustered(struct mddev *mddev)
834 {
835 	return mddev->cluster_info && mddev->bitmap_info.nodes > 1;
836 }
837 
838 /* clear unsupported mddev_flags */
839 static inline void mddev_clear_unsupported_flags(struct mddev *mddev,
840 	unsigned long unsupported_flags)
841 {
842 	mddev->flags &= ~unsupported_flags;
843 }
844 
845 static inline void mddev_check_write_zeroes(struct mddev *mddev, struct bio *bio)
846 {
847 	if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
848 	    !bio->bi_bdev->bd_disk->queue->limits.max_write_zeroes_sectors)
849 		mddev->queue->limits.max_write_zeroes_sectors = 0;
850 }
851 
852 struct mdu_array_info_s;
853 struct mdu_disk_info_s;
854 
855 extern int mdp_major;
856 extern struct workqueue_struct *md_bitmap_wq;
857 void md_autostart_arrays(int part);
858 int md_set_array_info(struct mddev *mddev, struct mdu_array_info_s *info);
859 int md_add_new_disk(struct mddev *mddev, struct mdu_disk_info_s *info);
860 int do_md_run(struct mddev *mddev);
861 
862 extern const struct block_device_operations md_fops;
863 
864 #endif /* _MD_MD_H */
865