xref: /linux/drivers/md/md.h (revision 6dfafbd0299a60bfb5d5e277fdf100037c7ded07)
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 <linux/raid/md_u.h>
22 #include <trace/events/block.h>
23 
24 #define MaxSector (~(sector_t)0)
25 
26 enum md_submodule_type {
27 	MD_PERSONALITY = 0,
28 	MD_CLUSTER,
29 	MD_BITMAP,
30 };
31 
32 enum md_submodule_id {
33 	ID_LINEAR	= LEVEL_LINEAR,
34 	ID_RAID0	= 0,
35 	ID_RAID1	= 1,
36 	ID_RAID4	= 4,
37 	ID_RAID5	= 5,
38 	ID_RAID6	= 6,
39 	ID_RAID10	= 10,
40 	ID_CLUSTER,
41 	ID_BITMAP,
42 	ID_LLBITMAP,
43 	ID_BITMAP_NONE,
44 };
45 
46 struct md_submodule_head {
47 	enum md_submodule_type	type;
48 	enum md_submodule_id	id;
49 	const char		*name;
50 	struct module		*owner;
51 };
52 
53 /*
54  * These flags should really be called "NO_RETRY" rather than
55  * "FAILFAST" because they don't make any promise about time lapse,
56  * only about the number of retries, which will be zero.
57  * REQ_FAILFAST_DRIVER is not included because
58  * Commit: 4a27446f3e39 ("[SCSI] modify scsi to handle new fail fast flags.")
59  * seems to suggest that the errors it avoids retrying should usually
60  * be retried.
61  */
62 #define	MD_FAILFAST	(REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT)
63 
64 /* Status of sync thread. */
65 enum sync_action {
66 	/*
67 	 * Represent by MD_RECOVERY_SYNC, start when:
68 	 * 1) after assemble, sync data from first rdev to other copies, this
69 	 * must be done first before other sync actions and will only execute
70 	 * once;
71 	 * 2) resize the array(notice that this is not reshape), sync data for
72 	 * the new range;
73 	 */
74 	ACTION_RESYNC,
75 	/*
76 	 * Represent by MD_RECOVERY_RECOVER, start when:
77 	 * 1) for new replacement, sync data based on the replace rdev or
78 	 * available copies from other rdev;
79 	 * 2) for new member disk while the array is degraded, sync data from
80 	 * other rdev;
81 	 * 3) reassemble after power failure or re-add a hot removed rdev, sync
82 	 * data from first rdev to other copies based on bitmap;
83 	 */
84 	ACTION_RECOVER,
85 	/*
86 	 * Represent by MD_RECOVERY_SYNC | MD_RECOVERY_REQUESTED |
87 	 * MD_RECOVERY_CHECK, start when user echo "check" to sysfs api
88 	 * sync_action, used to check if data copies from differenct rdev are
89 	 * the same. The number of mismatch sectors will be exported to user
90 	 * by sysfs api mismatch_cnt;
91 	 */
92 	ACTION_CHECK,
93 	/*
94 	 * Represent by MD_RECOVERY_SYNC | MD_RECOVERY_REQUESTED, start when
95 	 * user echo "repair" to sysfs api sync_action, usually paired with
96 	 * ACTION_CHECK, used to force syncing data once user found that there
97 	 * are inconsistent data,
98 	 */
99 	ACTION_REPAIR,
100 	/*
101 	 * Represent by MD_RECOVERY_RESHAPE, start when new member disk is added
102 	 * to the conf, notice that this is different from spares or
103 	 * replacement;
104 	 */
105 	ACTION_RESHAPE,
106 	/*
107 	 * Represent by MD_RECOVERY_FROZEN, can be set by sysfs api sync_action
108 	 * or internal usage like setting the array read-only, will forbid above
109 	 * actions.
110 	 */
111 	ACTION_FROZEN,
112 	/*
113 	 * All above actions don't match.
114 	 */
115 	ACTION_IDLE,
116 	NR_SYNC_ACTIONS,
117 };
118 
119 /*
120  * The struct embedded in rdev is used to serialize IO.
121  */
122 struct serial_in_rdev {
123 	struct rb_root_cached serial_rb;
124 	spinlock_t serial_lock;
125 	wait_queue_head_t serial_io_wait;
126 };
127 
128 /*
129  * MD's 'extended' device
130  */
131 struct md_rdev {
132 	struct list_head same_set;	/* RAID devices within the same set */
133 
134 	sector_t sectors;		/* Device size (in 512bytes sectors) */
135 	struct mddev *mddev;		/* RAID array if running */
136 	unsigned long last_events;	/* IO event timestamp */
137 
138 	/*
139 	 * If meta_bdev is non-NULL, it means that a separate device is
140 	 * being used to store the metadata (superblock/bitmap) which
141 	 * would otherwise be contained on the same device as the data (bdev).
142 	 */
143 	struct block_device *meta_bdev;
144 	struct block_device *bdev;	/* block device handle */
145 	struct file *bdev_file;		/* Handle from open for bdev */
146 
147 	struct page	*sb_page, *bb_page;
148 	int		sb_loaded;
149 	__u64		sb_events;
150 	sector_t	data_offset;	/* start of data in array */
151 	sector_t	new_data_offset;/* only relevant while reshaping */
152 	sector_t	sb_start;	/* offset of the super block (in 512byte sectors) */
153 	int		sb_size;	/* bytes in the superblock */
154 	int		preferred_minor;	/* autorun support */
155 
156 	struct kobject	kobj;
157 
158 	/* A device can be in one of three states based on two flags:
159 	 * Not working:   faulty==1 in_sync==0
160 	 * Fully working: faulty==0 in_sync==1
161 	 * Working, but not
162 	 * in sync with array
163 	 *                faulty==0 in_sync==0
164 	 *
165 	 * It can never have faulty==1, in_sync==1
166 	 * This reduces the burden of testing multiple flags in many cases
167 	 */
168 
169 	unsigned long	flags;	/* bit set of 'enum flag_bits' bits. */
170 	wait_queue_head_t blocked_wait;
171 
172 	int desc_nr;			/* descriptor index in the superblock */
173 	int raid_disk;			/* role of device in array */
174 	int new_raid_disk;		/* role that the device will have in
175 					 * the array after a level-change completes.
176 					 */
177 	int saved_raid_disk;		/* role that device used to have in the
178 					 * array and could again if we did a partial
179 					 * resync from the bitmap
180 					 */
181 	union {
182 		sector_t recovery_offset;/* If this device has been partially
183 					 * recovered, this is where we were
184 					 * up to.
185 					 */
186 		sector_t journal_tail;	/* If this device is a journal device,
187 					 * this is the journal tail (journal
188 					 * recovery start point)
189 					 */
190 	};
191 
192 	atomic_t	nr_pending;	/* number of pending requests.
193 					 * only maintained for arrays that
194 					 * support hot removal
195 					 */
196 	atomic_t	read_errors;	/* number of consecutive read errors that
197 					 * we have tried to ignore.
198 					 */
199 	time64_t	last_read_error;	/* monotonic time since our
200 						 * last read error
201 						 */
202 	atomic_t	corrected_errors; /* number of corrected read errors,
203 					   * for reporting to userspace and storing
204 					   * in superblock.
205 					   */
206 
207 	struct serial_in_rdev *serial;  /* used for raid1 io serialization */
208 
209 	struct kernfs_node *sysfs_state; /* handle for 'state'
210 					   * sysfs entry */
211 	/* handle for 'unacknowledged_bad_blocks' sysfs dentry */
212 	struct kernfs_node *sysfs_unack_badblocks;
213 	/* handle for 'bad_blocks' sysfs dentry */
214 	struct kernfs_node *sysfs_badblocks;
215 	struct badblocks badblocks;
216 
217 	struct {
218 		short offset;	/* Offset from superblock to start of PPL.
219 				 * Not used by external metadata. */
220 		unsigned int size;	/* Size in sectors of the PPL space */
221 		sector_t sector;	/* First sector of the PPL space */
222 	} ppl;
223 };
224 enum flag_bits {
225 	Faulty,			/* device is known to have a fault */
226 	In_sync,		/* device is in_sync with rest of array */
227 	Bitmap_sync,		/* ..actually, not quite In_sync.  Need a
228 				 * bitmap-based recovery to get fully in sync.
229 				 * The bit is only meaningful before device
230 				 * has been passed to pers->hot_add_disk.
231 				 */
232 	WriteMostly,		/* Avoid reading if at all possible */
233 	AutoDetected,		/* added by auto-detect */
234 	Blocked,		/* An error occurred but has not yet
235 				 * been acknowledged by the metadata
236 				 * handler, so don't allow writes
237 				 * until it is cleared */
238 	WriteErrorSeen,		/* A write error has been seen on this
239 				 * device
240 				 */
241 	FaultRecorded,		/* Intermediate state for clearing
242 				 * Blocked.  The Fault is/will-be
243 				 * recorded in the metadata, but that
244 				 * metadata hasn't been stored safely
245 				 * on disk yet.
246 				 */
247 	BlockedBadBlocks,	/* A writer is blocked because they
248 				 * found an unacknowledged bad-block.
249 				 * This can safely be cleared at any
250 				 * time, and the writer will re-check.
251 				 * It may be set at any time, and at
252 				 * worst the writer will timeout and
253 				 * re-check.  So setting it as
254 				 * accurately as possible is good, but
255 				 * not absolutely critical.
256 				 */
257 	WantReplacement,	/* This device is a candidate to be
258 				 * hot-replaced, either because it has
259 				 * reported some faults, or because
260 				 * of explicit request.
261 				 */
262 	Replacement,		/* This device is a replacement for
263 				 * a want_replacement device with same
264 				 * raid_disk number.
265 				 */
266 	Candidate,		/* For clustered environments only:
267 				 * This device is seen locally but not
268 				 * by the whole cluster
269 				 */
270 	Journal,		/* This device is used as journal for
271 				 * raid-5/6.
272 				 * Usually, this device should be faster
273 				 * than other devices in the array
274 				 */
275 	ClusterRemove,
276 	ExternalBbl,            /* External metadata provides bad
277 				 * block management for a disk
278 				 */
279 	FailFast,		/* Minimal retries should be attempted on
280 				 * this device, so use REQ_FAILFAST_DEV.
281 				 * Also don't try to repair failed reads.
282 				 * It is expects that no bad block log
283 				 * is present.
284 				 */
285 	LastDev,		/* Seems to be the last working dev as
286 				 * it didn't fail, so don't use FailFast
287 				 * any more for metadata
288 				 */
289 	CollisionCheck,		/*
290 				 * check if there is collision between raid1
291 				 * serial bios.
292 				 */
293 	Nonrot,			/* non-rotational device (SSD) */
294 };
295 
296 static inline int is_badblock(struct md_rdev *rdev, sector_t s, sector_t sectors,
297 			      sector_t *first_bad, sector_t *bad_sectors)
298 {
299 	if (unlikely(rdev->badblocks.count)) {
300 		int rv = badblocks_check(&rdev->badblocks, rdev->data_offset + s,
301 					sectors,
302 					first_bad, bad_sectors);
303 		if (rv)
304 			*first_bad -= rdev->data_offset;
305 		return rv;
306 	}
307 	return 0;
308 }
309 
310 static inline int rdev_has_badblock(struct md_rdev *rdev, sector_t s,
311 				    int sectors)
312 {
313 	sector_t first_bad;
314 	sector_t bad_sectors;
315 
316 	return is_badblock(rdev, s, sectors, &first_bad, &bad_sectors);
317 }
318 
319 extern bool rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
320 			       int is_new);
321 extern void rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
322 				 int is_new);
323 struct md_cluster_info;
324 struct md_cluster_operations;
325 
326 /**
327  * enum mddev_flags - md device flags.
328  * @MD_ARRAY_FIRST_USE: First use of array, needs initialization.
329  * @MD_CLOSING: If set, we are closing the array, do not open it then.
330  * @MD_JOURNAL_CLEAN: A raid with journal is already clean.
331  * @MD_HAS_JOURNAL: The raid array has journal feature set.
332  * @MD_CLUSTER_RESYNC_LOCKED: cluster raid only, which means node, already took
333  *			       resync lock, need to release the lock.
334  * @MD_FAILFAST_SUPPORTED: Using MD_FAILFAST on metadata writes is supported as
335  *			    calls to md_error() will never cause the array to
336  *			    become failed.
337  * @MD_HAS_PPL:  The raid array has PPL feature set.
338  * @MD_HAS_MULTIPLE_PPLS: The raid array has multiple PPLs feature set.
339  * @MD_NOT_READY: do_md_run() is active, so 'array_state', ust not report that
340  *		   array is ready yet.
341  * @MD_BROKEN: This is used to stop writes and mark array as failed.
342  * @MD_DELETED: This device is being deleted
343  *
344  * change UNSUPPORTED_MDDEV_FLAGS for each array type if new flag is added
345  */
346 enum mddev_flags {
347 	MD_ARRAY_FIRST_USE,
348 	MD_CLOSING,
349 	MD_JOURNAL_CLEAN,
350 	MD_HAS_JOURNAL,
351 	MD_CLUSTER_RESYNC_LOCKED,
352 	MD_FAILFAST_SUPPORTED,
353 	MD_HAS_PPL,
354 	MD_HAS_MULTIPLE_PPLS,
355 	MD_NOT_READY,
356 	MD_BROKEN,
357 	MD_DO_DELETE,
358 	MD_DELETED,
359 };
360 
361 enum mddev_sb_flags {
362 	MD_SB_CHANGE_DEVS,		/* Some device status has changed */
363 	MD_SB_CHANGE_CLEAN,	/* transition to or from 'clean' */
364 	MD_SB_CHANGE_PENDING,	/* switch from 'clean' to 'active' in progress */
365 	MD_SB_NEED_REWRITE,	/* metadata write needs to be repeated */
366 };
367 
368 #define NR_SERIAL_INFOS		8
369 /* record current range of serialize IOs */
370 struct serial_info {
371 	struct rb_node node;
372 	sector_t start;		/* start sector of rb node */
373 	sector_t last;		/* end sector of rb node */
374 	sector_t _subtree_last; /* highest sector in subtree of rb node */
375 };
376 
377 /*
378  * mddev->curr_resync stores the current sector of the resync but
379  * also has some overloaded values.
380  */
381 enum {
382 	/* No resync in progress */
383 	MD_RESYNC_NONE = 0,
384 	/* Yielded to allow another conflicting resync to commence */
385 	MD_RESYNC_YIELDED = 1,
386 	/* Delayed to check that there is no conflict with another sync */
387 	MD_RESYNC_DELAYED = 2,
388 	/* Any value greater than or equal to this is in an active resync */
389 	MD_RESYNC_ACTIVE = 3,
390 };
391 
392 struct mddev {
393 	void				*private;
394 	struct md_personality		*pers;
395 	dev_t				unit;
396 	int				md_minor;
397 	struct list_head		disks;
398 	unsigned long			flags;
399 	unsigned long			sb_flags;
400 
401 	int				suspended;
402 	struct mutex			suspend_mutex;
403 	struct percpu_ref		active_io;
404 	int				ro;
405 	int				sysfs_active; /* set when sysfs deletes
406 						       * are happening, so run/
407 						       * takeover/stop are not safe
408 						       */
409 	struct gendisk			*gendisk;    /* mdraid gendisk */
410 	struct gendisk			*dm_gendisk; /* dm-raid gendisk */
411 
412 	struct kobject			kobj;
413 	int				hold_active;
414 #define	UNTIL_IOCTL	1
415 #define	UNTIL_STOP	2
416 
417 	/* Superblock information */
418 	int				major_version,
419 					minor_version,
420 					patch_version;
421 	int				persistent;
422 	int				external;	/* metadata is
423 							 * managed externally */
424 	char				metadata_type[17]; /* externally set*/
425 	int				chunk_sectors;
426 	time64_t			ctime, utime;
427 	int				level, layout;
428 	char				clevel[16];
429 	int				raid_disks;
430 	int				max_disks;
431 	sector_t			dev_sectors;	/* used size of
432 							 * component devices */
433 	sector_t			array_sectors; /* exported array size */
434 	int				external_size; /* size managed
435 							* externally */
436 	unsigned int			logical_block_size;
437 	__u64				events;
438 	/* If the last 'event' was simply a clean->dirty transition, and
439 	 * we didn't write it to the spares, then it is safe and simple
440 	 * to just decrement the event count on a dirty->clean transition.
441 	 * So we record that possibility here.
442 	 */
443 	int				can_decrease_events;
444 
445 	char				uuid[16];
446 
447 	/* If the array is being reshaped, we need to record the
448 	 * new shape and an indication of where we are up to.
449 	 * This is written to the superblock.
450 	 * If reshape_position is MaxSector, then no reshape is happening (yet).
451 	 */
452 	sector_t			reshape_position;
453 	int				delta_disks, new_level, new_layout;
454 	int				new_chunk_sectors;
455 	int				reshape_backwards;
456 
457 	struct md_thread __rcu		*thread;	/* management thread */
458 	struct md_thread __rcu		*sync_thread;	/* doing resync or reconstruct */
459 
460 	/*
461 	 * Set when a sync operation is started. It holds this value even
462 	 * when the sync thread is "frozen" (interrupted) or "idle" (stopped
463 	 * or finished). It is overwritten when a new sync operation is begun.
464 	 */
465 	enum sync_action		last_sync_action;
466 	sector_t			curr_resync;	/* last block scheduled */
467 	/* As resync requests can complete out of order, we cannot easily track
468 	 * how much resync has been completed.  So we occasionally pause until
469 	 * everything completes, then set curr_resync_completed to curr_resync.
470 	 * As such it may be well behind the real resync mark, but it is a value
471 	 * we are certain of.
472 	 */
473 	sector_t			curr_resync_completed;
474 	unsigned long			resync_mark;	/* a recent timestamp */
475 	sector_t			resync_mark_cnt;/* blocks written at resync_mark */
476 	sector_t			curr_mark_cnt; /* blocks scheduled now */
477 
478 	sector_t			resync_max_sectors; /* may be set by personality */
479 
480 	atomic64_t			resync_mismatches; /* count of sectors where
481 							    * parity/replica mismatch found
482 							    */
483 
484 	/* allow user-space to request suspension of IO to regions of the array */
485 	sector_t			suspend_lo;
486 	sector_t			suspend_hi;
487 	/* if zero, use the system-wide default */
488 	int				sync_speed_min;
489 	int				sync_speed_max;
490 	int				sync_io_depth;
491 
492 	/* resync even though the same disks are shared among md-devices */
493 	int				parallel_resync;
494 
495 	int				ok_start_degraded;
496 
497 	unsigned long			recovery;
498 	/* If a RAID personality determines that recovery (of a particular
499 	 * device) will fail due to a read error on the source device, it
500 	 * takes a copy of this number and does not attempt recovery again
501 	 * until this number changes.
502 	 */
503 	int				recovery_disabled;
504 
505 	int				in_sync;	/* know to not need resync */
506 	/* 'open_mutex' avoids races between 'md_open' and 'do_md_stop', so
507 	 * that we are never stopping an array while it is open.
508 	 * 'reconfig_mutex' protects all other reconfiguration.
509 	 * These locks are separate due to conflicting interactions
510 	 * with disk->open_mutex.
511 	 * Lock ordering is:
512 	 *  reconfig_mutex -> disk->open_mutex
513 	 *  disk->open_mutex -> open_mutex:  e.g. __blkdev_get -> md_open
514 	 */
515 	struct mutex			open_mutex;
516 	struct mutex			reconfig_mutex;
517 	atomic_t			active;		/* general refcount */
518 	atomic_t			openers;	/* number of active opens */
519 
520 	int				changed;	/* True if we might need to
521 							 * reread partition info */
522 	int				degraded;	/* whether md should consider
523 							 * adding a spare
524 							 */
525 
526 	unsigned long			normal_io_events; /* IO event timestamp */
527 	atomic_t			recovery_active; /* blocks scheduled, but not written */
528 	wait_queue_head_t		recovery_wait;
529 	sector_t			resync_offset;
530 	sector_t			resync_min;	/* user requested sync
531 							 * starts here */
532 	sector_t			resync_max;	/* resync should pause
533 							 * when it gets here */
534 
535 	struct kernfs_node		*sysfs_state;	/* handle for 'array_state'
536 							 * file in sysfs.
537 							 */
538 	struct kernfs_node		*sysfs_action;  /* handle for 'sync_action' */
539 	struct kernfs_node		*sysfs_completed;	/*handle for 'sync_completed' */
540 	struct kernfs_node		*sysfs_degraded;	/*handle for 'degraded' */
541 	struct kernfs_node		*sysfs_level;		/*handle for 'level' */
542 
543 	/* used for delayed sysfs removal */
544 	struct work_struct del_work;
545 	/* used for register new sync thread */
546 	struct work_struct sync_work;
547 
548 	/* "lock" protects:
549 	 *   flush_bio transition from NULL to !NULL
550 	 *   rdev superblocks, events
551 	 *   clearing MD_CHANGE_*
552 	 *   in_sync - and related safemode and MD_CHANGE changes
553 	 *   pers (also protected by reconfig_mutex and pending IO).
554 	 *   clearing ->bitmap
555 	 *   clearing ->bitmap_info.file
556 	 *   changing ->resync_{min,max}
557 	 *   setting MD_RECOVERY_RUNNING (which interacts with resync_{min,max})
558 	 */
559 	spinlock_t			lock;
560 	wait_queue_head_t		sb_wait;	/* for waiting on superblock updates */
561 	atomic_t			pending_writes;	/* number of active superblock writes */
562 
563 	unsigned int			safemode;	/* if set, update "clean" superblock
564 							 * when no writes pending.
565 							 */
566 	unsigned int			safemode_delay;
567 	struct timer_list		safemode_timer;
568 	struct percpu_ref		writes_pending;
569 	int				sync_checkers;	/* # of threads checking writes_pending */
570 
571 	enum md_submodule_id		bitmap_id;
572 	void				*bitmap; /* the bitmap for the device */
573 	struct bitmap_operations	*bitmap_ops;
574 	struct {
575 		struct file		*file; /* the bitmap file */
576 		loff_t			offset; /* offset from superblock of
577 						 * start of bitmap. May be
578 						 * negative, but not '0'
579 						 * For external metadata, offset
580 						 * from start of device.
581 						 */
582 		unsigned long		space; /* space available at this offset */
583 		loff_t			default_offset; /* this is the offset to use when
584 							 * hot-adding a bitmap.  It should
585 							 * eventually be settable by sysfs.
586 							 */
587 		unsigned long		default_space; /* space available at
588 							* default offset */
589 		struct mutex		mutex;
590 		unsigned long		chunksize;
591 		unsigned long		daemon_sleep; /* how many jiffies between updates? */
592 		unsigned long		max_write_behind; /* write-behind mode */
593 		int			external;
594 		int			nodes; /* Maximum number of nodes in the cluster */
595 		char                    cluster_name[64]; /* Name of the cluster */
596 	} bitmap_info;
597 
598 	atomic_t			max_corr_read_errors; /* max read retries */
599 	struct list_head		all_mddevs;
600 
601 	const struct attribute_group	*to_remove;
602 
603 	struct bio_set			bio_set;
604 	struct bio_set			sync_set; /* for sync operations like
605 						   * metadata and bitmap writes
606 						   */
607 	struct bio_set			io_clone_set;
608 
609 	struct work_struct event_work;	/* used by dm to report failure event */
610 	mempool_t *serial_info_pool;
611 	void (*sync_super)(struct mddev *mddev, struct md_rdev *rdev);
612 	struct md_cluster_info		*cluster_info;
613 	struct md_cluster_operations *cluster_ops;
614 	unsigned int			good_device_nr;	/* good device num within cluster raid */
615 	unsigned int			noio_flag; /* for memalloc scope API */
616 
617 	/*
618 	 * Temporarily store rdev that will be finally removed when
619 	 * reconfig_mutex is unlocked, protected by reconfig_mutex.
620 	 */
621 	struct list_head		deleting;
622 
623 	/* The sequence number for sync thread */
624 	atomic_t sync_seq;
625 
626 	bool	has_superblocks:1;
627 	bool	fail_last_dev:1;
628 	bool	serialize_policy:1;
629 };
630 
631 enum recovery_flags {
632 	/* flags for sync thread running status */
633 
634 	/*
635 	 * set when one of sync action is set and new sync thread need to be
636 	 * registered, or just add/remove spares from conf.
637 	 */
638 	MD_RECOVERY_NEEDED,
639 	/* sync thread is running, or about to be started */
640 	MD_RECOVERY_RUNNING,
641 	/* sync thread needs to be aborted for some reason */
642 	MD_RECOVERY_INTR,
643 	/* sync thread is done and is waiting to be unregistered */
644 	MD_RECOVERY_DONE,
645 	/* running sync thread must abort immediately, and not restart */
646 	MD_RECOVERY_FROZEN,
647 	/* waiting for pers->start() to finish */
648 	MD_RECOVERY_WAIT,
649 	/* interrupted because io-error */
650 	MD_RECOVERY_ERROR,
651 
652 	/* flags determines sync action, see details in enum sync_action */
653 
654 	/* if just this flag is set, action is resync. */
655 	MD_RECOVERY_SYNC,
656 	/*
657 	 * paired with MD_RECOVERY_SYNC, if MD_RECOVERY_CHECK is not set,
658 	 * action is repair, means user requested resync.
659 	 */
660 	MD_RECOVERY_REQUESTED,
661 	/*
662 	 * paired with MD_RECOVERY_SYNC and MD_RECOVERY_REQUESTED, action is
663 	 * check.
664 	 */
665 	MD_RECOVERY_CHECK,
666 	/* recovery, or need to try it */
667 	MD_RECOVERY_RECOVER,
668 	/* reshape */
669 	MD_RECOVERY_RESHAPE,
670 	/* remote node is running resync thread */
671 	MD_RESYNCING_REMOTE,
672 	/* raid456 lazy initial recover */
673 	MD_RECOVERY_LAZY_RECOVER,
674 };
675 
676 enum md_ro_state {
677 	MD_RDWR,
678 	MD_RDONLY,
679 	MD_AUTO_READ,
680 	MD_MAX_STATE
681 };
682 
683 static inline bool md_is_rdwr(struct mddev *mddev)
684 {
685 	return (mddev->ro == MD_RDWR);
686 }
687 
688 static inline bool reshape_interrupted(struct mddev *mddev)
689 {
690 	/* reshape never start */
691 	if (mddev->reshape_position == MaxSector)
692 		return false;
693 
694 	/* interrupted */
695 	if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
696 		return true;
697 
698 	/* running reshape will be interrupted soon. */
699 	if (test_bit(MD_RECOVERY_WAIT, &mddev->recovery) ||
700 	    test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
701 	    test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
702 		return true;
703 
704 	return false;
705 }
706 
707 static inline int __must_check mddev_lock(struct mddev *mddev)
708 {
709 	int ret;
710 
711 	ret = mutex_lock_interruptible(&mddev->reconfig_mutex);
712 
713 	/* MD_DELETED is set in do_md_stop with reconfig_mutex.
714 	 * So check it here.
715 	 */
716 	if (!ret && test_bit(MD_DELETED, &mddev->flags)) {
717 		ret = -ENODEV;
718 		mutex_unlock(&mddev->reconfig_mutex);
719 	}
720 
721 	return ret;
722 }
723 
724 /* Sometimes we need to take the lock in a situation where
725  * failure due to interrupts is not acceptable.
726  * It doesn't need to check MD_DELETED here, the owner which
727  * holds the lock here can't be stopped. And all paths can't
728  * call this function after do_md_stop.
729  */
730 static inline void mddev_lock_nointr(struct mddev *mddev)
731 {
732 	mutex_lock(&mddev->reconfig_mutex);
733 }
734 
735 static inline int mddev_trylock(struct mddev *mddev)
736 {
737 	int ret;
738 
739 	ret = mutex_trylock(&mddev->reconfig_mutex);
740 	if (!ret && test_bit(MD_DELETED, &mddev->flags)) {
741 		ret = -ENODEV;
742 		mutex_unlock(&mddev->reconfig_mutex);
743 	}
744 	return ret;
745 }
746 extern void mddev_unlock(struct mddev *mddev);
747 
748 struct md_personality
749 {
750 	struct md_submodule_head head;
751 
752 	bool __must_check (*make_request)(struct mddev *mddev, struct bio *bio);
753 	/*
754 	 * start up works that do NOT require md_thread. tasks that
755 	 * requires md_thread should go into start()
756 	 */
757 	int (*run)(struct mddev *mddev);
758 	/* start up works that require md threads */
759 	int (*start)(struct mddev *mddev);
760 	void (*free)(struct mddev *mddev, void *priv);
761 	void (*status)(struct seq_file *seq, struct mddev *mddev);
762 	/* error_handler must set ->faulty and clear ->in_sync
763 	 * if appropriate, and should abort recovery if needed
764 	 */
765 	void (*error_handler)(struct mddev *mddev, struct md_rdev *rdev);
766 	int (*hot_add_disk) (struct mddev *mddev, struct md_rdev *rdev);
767 	int (*hot_remove_disk) (struct mddev *mddev, struct md_rdev *rdev);
768 	int (*spare_active) (struct mddev *mddev);
769 	sector_t (*sync_request)(struct mddev *mddev, sector_t sector_nr,
770 				 sector_t max_sector, int *skipped);
771 	int (*resize) (struct mddev *mddev, sector_t sectors);
772 	sector_t (*size) (struct mddev *mddev, sector_t sectors, int raid_disks);
773 	int (*check_reshape) (struct mddev *mddev);
774 	int (*start_reshape) (struct mddev *mddev);
775 	void (*finish_reshape) (struct mddev *mddev);
776 	void (*update_reshape_pos) (struct mddev *mddev);
777 	void (*prepare_suspend) (struct mddev *mddev);
778 	/* quiesce suspends or resumes internal processing.
779 	 * 1 - stop new actions and wait for action io to complete
780 	 * 0 - return to normal behaviour
781 	 */
782 	void (*quiesce) (struct mddev *mddev, int quiesce);
783 	/* takeover is used to transition an array from one
784 	 * personality to another.  The new personality must be able
785 	 * to handle the data in the current layout.
786 	 * e.g. 2drive raid1 -> 2drive raid5
787 	 *      ndrive raid5 -> degraded n+1drive raid6 with special layout
788 	 * If the takeover succeeds, a new 'private' structure is returned.
789 	 * This needs to be installed and then ->run used to activate the
790 	 * array.
791 	 */
792 	void *(*takeover) (struct mddev *mddev);
793 	/* Changes the consistency policy of an active array. */
794 	int (*change_consistency_policy)(struct mddev *mddev, const char *buf);
795 	/* convert io ranges from array to bitmap */
796 	void (*bitmap_sector)(struct mddev *mddev, sector_t *offset,
797 			      unsigned long *sectors);
798 };
799 
800 struct md_sysfs_entry {
801 	struct attribute attr;
802 	ssize_t (*show)(struct mddev *, char *);
803 	ssize_t (*store)(struct mddev *, const char *, size_t);
804 };
805 
806 static inline struct kernfs_node *sysfs_get_dirent_safe(struct kernfs_node *sd, char *name)
807 {
808 	if (sd)
809 		return sysfs_get_dirent(sd, name);
810 	return sd;
811 }
812 static inline void sysfs_notify_dirent_safe(struct kernfs_node *sd)
813 {
814 	if (sd)
815 		sysfs_notify_dirent(sd);
816 }
817 
818 static inline char * mdname (struct mddev * mddev)
819 {
820 	return mddev->gendisk ? mddev->gendisk->disk_name : "mdX";
821 }
822 
823 static inline int sysfs_link_rdev(struct mddev *mddev, struct md_rdev *rdev)
824 {
825 	char nm[20];
826 	if (!test_bit(Replacement, &rdev->flags) &&
827 	    !test_bit(Journal, &rdev->flags) &&
828 	    mddev->kobj.sd) {
829 		sprintf(nm, "rd%d", rdev->raid_disk);
830 		return sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
831 	} else
832 		return 0;
833 }
834 
835 static inline void sysfs_unlink_rdev(struct mddev *mddev, struct md_rdev *rdev)
836 {
837 	char nm[20];
838 	if (!test_bit(Replacement, &rdev->flags) &&
839 	    !test_bit(Journal, &rdev->flags) &&
840 	    mddev->kobj.sd) {
841 		sprintf(nm, "rd%d", rdev->raid_disk);
842 		sysfs_remove_link(&mddev->kobj, nm);
843 	}
844 }
845 
846 /*
847  * iterates through some rdev ringlist. It's safe to remove the
848  * current 'rdev'. Dont touch 'tmp' though.
849  */
850 #define rdev_for_each_list(rdev, tmp, head)				\
851 	list_for_each_entry_safe(rdev, tmp, head, same_set)
852 
853 /*
854  * iterates through the 'same array disks' ringlist
855  */
856 #define rdev_for_each(rdev, mddev)				\
857 	list_for_each_entry(rdev, &((mddev)->disks), same_set)
858 
859 #define rdev_for_each_safe(rdev, tmp, mddev)				\
860 	list_for_each_entry_safe(rdev, tmp, &((mddev)->disks), same_set)
861 
862 #define rdev_for_each_rcu(rdev, mddev)				\
863 	list_for_each_entry_rcu(rdev, &((mddev)->disks), same_set)
864 
865 struct md_thread {
866 	void			(*run) (struct md_thread *thread);
867 	struct mddev		*mddev;
868 	wait_queue_head_t	wqueue;
869 	unsigned long		flags;
870 	struct task_struct	*tsk;
871 	unsigned long		timeout;
872 	void			*private;
873 };
874 
875 struct md_io_clone {
876 	struct mddev	*mddev;
877 	struct bio	*orig_bio;
878 	unsigned long	start_time;
879 	sector_t	offset;
880 	unsigned long	sectors;
881 	enum stat_group	rw;
882 	struct bio	bio_clone;
883 };
884 
885 #define THREAD_WAKEUP  0
886 
887 #define md_wakeup_thread(thread) do {   \
888 	rcu_read_lock();                    \
889 	__md_wakeup_thread(thread);         \
890 	rcu_read_unlock();                  \
891 } while (0)
892 
893 static inline void safe_put_page(struct page *p)
894 {
895 	if (p) put_page(p);
896 }
897 
898 int register_md_submodule(struct md_submodule_head *msh);
899 void unregister_md_submodule(struct md_submodule_head *msh);
900 
901 extern struct md_thread *md_register_thread(
902 	void (*run)(struct md_thread *thread),
903 	struct mddev *mddev,
904 	const char *name);
905 extern void md_unregister_thread(struct mddev *mddev, struct md_thread __rcu **threadp);
906 extern void __md_wakeup_thread(struct md_thread __rcu *thread);
907 extern void md_check_recovery(struct mddev *mddev);
908 extern void md_reap_sync_thread(struct mddev *mddev);
909 extern enum sync_action md_sync_action(struct mddev *mddev);
910 extern enum sync_action md_sync_action_by_name(const char *page);
911 extern const char *md_sync_action_name(enum sync_action action);
912 extern void md_write_start(struct mddev *mddev, struct bio *bi);
913 extern void md_write_inc(struct mddev *mddev, struct bio *bi);
914 extern void md_write_end(struct mddev *mddev);
915 extern void md_done_sync(struct mddev *mddev, int blocks, int ok);
916 extern void md_error(struct mddev *mddev, struct md_rdev *rdev);
917 extern void md_finish_reshape(struct mddev *mddev);
918 void md_submit_discard_bio(struct mddev *mddev, struct md_rdev *rdev,
919 			struct bio *bio, sector_t start, sector_t size);
920 void md_account_bio(struct mddev *mddev, struct bio **bio);
921 void md_free_cloned_bio(struct bio *bio);
922 
923 extern bool __must_check md_flush_request(struct mddev *mddev, struct bio *bio);
924 void md_write_metadata(struct mddev *mddev, struct md_rdev *rdev,
925 		       sector_t sector, int size, struct page *page,
926 		       unsigned int offset);
927 extern int md_super_wait(struct mddev *mddev);
928 extern int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
929 		struct page *page, blk_opf_t opf, bool metadata_op);
930 extern void md_do_sync(struct md_thread *thread);
931 extern void md_new_event(void);
932 extern void md_allow_write(struct mddev *mddev);
933 extern void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev);
934 extern void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors);
935 extern int md_check_no_bitmap(struct mddev *mddev);
936 extern int md_integrity_register(struct mddev *mddev);
937 extern int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale);
938 
939 extern int mddev_init(struct mddev *mddev);
940 extern void mddev_destroy(struct mddev *mddev);
941 void md_init_stacking_limits(struct queue_limits *lim);
942 struct mddev *md_alloc(dev_t dev, char *name);
943 void mddev_put(struct mddev *mddev);
944 extern int md_run(struct mddev *mddev);
945 extern int md_start(struct mddev *mddev);
946 extern void md_stop(struct mddev *mddev);
947 extern void md_stop_writes(struct mddev *mddev);
948 extern int md_rdev_init(struct md_rdev *rdev);
949 extern void md_rdev_clear(struct md_rdev *rdev);
950 
951 extern bool md_handle_request(struct mddev *mddev, struct bio *bio);
952 extern int mddev_suspend(struct mddev *mddev, bool interruptible);
953 extern void mddev_resume(struct mddev *mddev);
954 extern void md_idle_sync_thread(struct mddev *mddev);
955 extern void md_frozen_sync_thread(struct mddev *mddev);
956 extern void md_unfrozen_sync_thread(struct mddev *mddev);
957 
958 extern void md_update_sb(struct mddev *mddev, int force);
959 extern void mddev_create_serial_pool(struct mddev *mddev, struct md_rdev *rdev);
960 extern void mddev_destroy_serial_pool(struct mddev *mddev,
961 				      struct md_rdev *rdev);
962 struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr);
963 struct md_rdev *md_find_rdev_rcu(struct mddev *mddev, dev_t dev);
964 
965 static inline bool is_rdev_broken(struct md_rdev *rdev)
966 {
967 	return !disk_live(rdev->bdev->bd_disk);
968 }
969 
970 static inline void rdev_dec_pending(struct md_rdev *rdev, struct mddev *mddev)
971 {
972 	int faulty = test_bit(Faulty, &rdev->flags);
973 	if (atomic_dec_and_test(&rdev->nr_pending) && faulty) {
974 		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
975 		md_wakeup_thread(mddev->thread);
976 	}
977 }
978 
979 static inline int mddev_is_clustered(struct mddev *mddev)
980 {
981 	return mddev->cluster_info && mddev->bitmap_info.nodes > 1;
982 }
983 
984 /* clear unsupported mddev_flags */
985 static inline void mddev_clear_unsupported_flags(struct mddev *mddev,
986 	unsigned long unsupported_flags)
987 {
988 	mddev->flags &= ~unsupported_flags;
989 }
990 
991 static inline void mddev_check_write_zeroes(struct mddev *mddev, struct bio *bio)
992 {
993 	if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
994 	    !bio->bi_bdev->bd_disk->queue->limits.max_write_zeroes_sectors)
995 		mddev->gendisk->queue->limits.max_write_zeroes_sectors = 0;
996 }
997 
998 static inline int mddev_suspend_and_lock(struct mddev *mddev)
999 {
1000 	int ret;
1001 
1002 	ret = mddev_suspend(mddev, true);
1003 	if (ret)
1004 		return ret;
1005 
1006 	ret = mddev_lock(mddev);
1007 	if (ret)
1008 		mddev_resume(mddev);
1009 
1010 	return ret;
1011 }
1012 
1013 static inline void mddev_suspend_and_lock_nointr(struct mddev *mddev)
1014 {
1015 	mddev_suspend(mddev, false);
1016 	mutex_lock(&mddev->reconfig_mutex);
1017 }
1018 
1019 static inline void mddev_unlock_and_resume(struct mddev *mddev)
1020 {
1021 	mddev_unlock(mddev);
1022 	mddev_resume(mddev);
1023 }
1024 
1025 struct mdu_array_info_s;
1026 struct mdu_disk_info_s;
1027 
1028 extern int mdp_major;
1029 void md_autostart_arrays(int part);
1030 int md_set_array_info(struct mddev *mddev, struct mdu_array_info_s *info);
1031 int md_add_new_disk(struct mddev *mddev, struct mdu_disk_info_s *info);
1032 int do_md_run(struct mddev *mddev);
1033 #define MDDEV_STACK_INTEGRITY	(1u << 0)
1034 int mddev_stack_rdev_limits(struct mddev *mddev, struct queue_limits *lim,
1035 		unsigned int flags);
1036 int mddev_stack_new_rdev(struct mddev *mddev, struct md_rdev *rdev);
1037 void mddev_update_io_opt(struct mddev *mddev, unsigned int nr_stripes);
1038 
1039 extern const struct block_device_operations md_fops;
1040 
1041 /*
1042  * MD devices can be used undeneath by DM, in which case ->gendisk is NULL.
1043  */
1044 static inline bool mddev_is_dm(struct mddev *mddev)
1045 {
1046 	return !mddev->gendisk;
1047 }
1048 
1049 static inline bool raid_is_456(struct mddev *mddev)
1050 {
1051 	return mddev->level == ID_RAID4 || mddev->level == ID_RAID5 ||
1052 	       mddev->level == ID_RAID6;
1053 }
1054 
1055 static inline void mddev_trace_remap(struct mddev *mddev, struct bio *bio,
1056 		sector_t sector)
1057 {
1058 	if (!mddev_is_dm(mddev))
1059 		trace_block_bio_remap(bio, disk_devt(mddev->gendisk), sector);
1060 }
1061 
1062 static inline bool rdev_blocked(struct md_rdev *rdev)
1063 {
1064 	/*
1065 	 * Blocked will be set by error handler and cleared by daemon after
1066 	 * updating superblock, meanwhile write IO should be blocked to prevent
1067 	 * reading old data after power failure.
1068 	 */
1069 	if (test_bit(Blocked, &rdev->flags))
1070 		return true;
1071 
1072 	/*
1073 	 * Faulty device should not be accessed anymore, there is no need to
1074 	 * wait for bad block to be acknowledged.
1075 	 */
1076 	if (test_bit(Faulty, &rdev->flags))
1077 		return false;
1078 
1079 	/* rdev is blocked by badblocks. */
1080 	if (test_bit(BlockedBadBlocks, &rdev->flags))
1081 		return true;
1082 
1083 	return false;
1084 }
1085 
1086 #define mddev_add_trace_msg(mddev, fmt, args...)			\
1087 do {									\
1088 	if (!mddev_is_dm(mddev))					\
1089 		blk_add_trace_msg((mddev)->gendisk->queue, fmt, ##args); \
1090 } while (0)
1091 
1092 #endif /* _MD_MD_H */
1093