xref: /linux/drivers/md/md.h (revision bf36793fa260cb68cc817f311f1f683788261796)
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 <trace/events/block.h>
22 #include "md-cluster.h"
23 
24 #define MaxSector (~(sector_t)0)
25 
26 /*
27  * These flags should really be called "NO_RETRY" rather than
28  * "FAILFAST" because they don't make any promise about time lapse,
29  * only about the number of retries, which will be zero.
30  * REQ_FAILFAST_DRIVER is not included because
31  * Commit: 4a27446f3e39 ("[SCSI] modify scsi to handle new fail fast flags.")
32  * seems to suggest that the errors it avoids retrying should usually
33  * be retried.
34  */
35 #define	MD_FAILFAST	(REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT)
36 
37 /* Status of sync thread. */
38 enum sync_action {
39 	/*
40 	 * Represent by MD_RECOVERY_SYNC, start when:
41 	 * 1) after assemble, sync data from first rdev to other copies, this
42 	 * must be done first before other sync actions and will only execute
43 	 * once;
44 	 * 2) resize the array(notice that this is not reshape), sync data for
45 	 * the new range;
46 	 */
47 	ACTION_RESYNC,
48 	/*
49 	 * Represent by MD_RECOVERY_RECOVER, start when:
50 	 * 1) for new replacement, sync data based on the replace rdev or
51 	 * available copies from other rdev;
52 	 * 2) for new member disk while the array is degraded, sync data from
53 	 * other rdev;
54 	 * 3) reassemble after power failure or re-add a hot removed rdev, sync
55 	 * data from first rdev to other copies based on bitmap;
56 	 */
57 	ACTION_RECOVER,
58 	/*
59 	 * Represent by MD_RECOVERY_SYNC | MD_RECOVERY_REQUESTED |
60 	 * MD_RECOVERY_CHECK, start when user echo "check" to sysfs api
61 	 * sync_action, used to check if data copies from differenct rdev are
62 	 * the same. The number of mismatch sectors will be exported to user
63 	 * by sysfs api mismatch_cnt;
64 	 */
65 	ACTION_CHECK,
66 	/*
67 	 * Represent by MD_RECOVERY_SYNC | MD_RECOVERY_REQUESTED, start when
68 	 * user echo "repair" to sysfs api sync_action, usually paired with
69 	 * ACTION_CHECK, used to force syncing data once user found that there
70 	 * are inconsistent data,
71 	 */
72 	ACTION_REPAIR,
73 	/*
74 	 * Represent by MD_RECOVERY_RESHAPE, start when new member disk is added
75 	 * to the conf, notice that this is different from spares or
76 	 * replacement;
77 	 */
78 	ACTION_RESHAPE,
79 	/*
80 	 * Represent by MD_RECOVERY_FROZEN, can be set by sysfs api sync_action
81 	 * or internal usage like setting the array read-only, will forbid above
82 	 * actions.
83 	 */
84 	ACTION_FROZEN,
85 	/*
86 	 * All above actions don't match.
87 	 */
88 	ACTION_IDLE,
89 	NR_SYNC_ACTIONS,
90 };
91 
92 /*
93  * The struct embedded in rdev is used to serialize IO.
94  */
95 struct serial_in_rdev {
96 	struct rb_root_cached serial_rb;
97 	spinlock_t serial_lock;
98 	wait_queue_head_t serial_io_wait;
99 };
100 
101 /*
102  * MD's 'extended' device
103  */
104 struct md_rdev {
105 	struct list_head same_set;	/* RAID devices within the same set */
106 
107 	sector_t sectors;		/* Device size (in 512bytes sectors) */
108 	struct mddev *mddev;		/* RAID array if running */
109 	int last_events;		/* IO event timestamp */
110 
111 	/*
112 	 * If meta_bdev is non-NULL, it means that a separate device is
113 	 * being used to store the metadata (superblock/bitmap) which
114 	 * would otherwise be contained on the same device as the data (bdev).
115 	 */
116 	struct block_device *meta_bdev;
117 	struct block_device *bdev;	/* block device handle */
118 	struct file *bdev_file;		/* Handle from open for bdev */
119 
120 	struct page	*sb_page, *bb_page;
121 	int		sb_loaded;
122 	__u64		sb_events;
123 	sector_t	data_offset;	/* start of data in array */
124 	sector_t	new_data_offset;/* only relevant while reshaping */
125 	sector_t	sb_start;	/* offset of the super block (in 512byte sectors) */
126 	int		sb_size;	/* bytes in the superblock */
127 	int		preferred_minor;	/* autorun support */
128 
129 	struct kobject	kobj;
130 
131 	/* A device can be in one of three states based on two flags:
132 	 * Not working:   faulty==1 in_sync==0
133 	 * Fully working: faulty==0 in_sync==1
134 	 * Working, but not
135 	 * in sync with array
136 	 *                faulty==0 in_sync==0
137 	 *
138 	 * It can never have faulty==1, in_sync==1
139 	 * This reduces the burden of testing multiple flags in many cases
140 	 */
141 
142 	unsigned long	flags;	/* bit set of 'enum flag_bits' bits. */
143 	wait_queue_head_t blocked_wait;
144 
145 	int desc_nr;			/* descriptor index in the superblock */
146 	int raid_disk;			/* role of device in array */
147 	int new_raid_disk;		/* role that the device will have in
148 					 * the array after a level-change completes.
149 					 */
150 	int saved_raid_disk;		/* role that device used to have in the
151 					 * array and could again if we did a partial
152 					 * resync from the bitmap
153 					 */
154 	union {
155 		sector_t recovery_offset;/* If this device has been partially
156 					 * recovered, this is where we were
157 					 * up to.
158 					 */
159 		sector_t journal_tail;	/* If this device is a journal device,
160 					 * this is the journal tail (journal
161 					 * recovery start point)
162 					 */
163 	};
164 
165 	atomic_t	nr_pending;	/* number of pending requests.
166 					 * only maintained for arrays that
167 					 * support hot removal
168 					 */
169 	atomic_t	read_errors;	/* number of consecutive read errors that
170 					 * we have tried to ignore.
171 					 */
172 	time64_t	last_read_error;	/* monotonic time since our
173 						 * last read error
174 						 */
175 	atomic_t	corrected_errors; /* number of corrected read errors,
176 					   * for reporting to userspace and storing
177 					   * in superblock.
178 					   */
179 
180 	struct serial_in_rdev *serial;  /* used for raid1 io serialization */
181 
182 	struct kernfs_node *sysfs_state; /* handle for 'state'
183 					   * sysfs entry */
184 	/* handle for 'unacknowledged_bad_blocks' sysfs dentry */
185 	struct kernfs_node *sysfs_unack_badblocks;
186 	/* handle for 'bad_blocks' sysfs dentry */
187 	struct kernfs_node *sysfs_badblocks;
188 	struct badblocks badblocks;
189 
190 	struct {
191 		short offset;	/* Offset from superblock to start of PPL.
192 				 * Not used by external metadata. */
193 		unsigned int size;	/* Size in sectors of the PPL space */
194 		sector_t sector;	/* First sector of the PPL space */
195 	} ppl;
196 };
197 enum flag_bits {
198 	Faulty,			/* device is known to have a fault */
199 	In_sync,		/* device is in_sync with rest of array */
200 	Bitmap_sync,		/* ..actually, not quite In_sync.  Need a
201 				 * bitmap-based recovery to get fully in sync.
202 				 * The bit is only meaningful before device
203 				 * has been passed to pers->hot_add_disk.
204 				 */
205 	WriteMostly,		/* Avoid reading if at all possible */
206 	AutoDetected,		/* added by auto-detect */
207 	Blocked,		/* An error occurred but has not yet
208 				 * been acknowledged by the metadata
209 				 * handler, so don't allow writes
210 				 * until it is cleared */
211 	WriteErrorSeen,		/* A write error has been seen on this
212 				 * device
213 				 */
214 	FaultRecorded,		/* Intermediate state for clearing
215 				 * Blocked.  The Fault is/will-be
216 				 * recorded in the metadata, but that
217 				 * metadata hasn't been stored safely
218 				 * on disk yet.
219 				 */
220 	BlockedBadBlocks,	/* A writer is blocked because they
221 				 * found an unacknowledged bad-block.
222 				 * This can safely be cleared at any
223 				 * time, and the writer will re-check.
224 				 * It may be set at any time, and at
225 				 * worst the writer will timeout and
226 				 * re-check.  So setting it as
227 				 * accurately as possible is good, but
228 				 * not absolutely critical.
229 				 */
230 	WantReplacement,	/* This device is a candidate to be
231 				 * hot-replaced, either because it has
232 				 * reported some faults, or because
233 				 * of explicit request.
234 				 */
235 	Replacement,		/* This device is a replacement for
236 				 * a want_replacement device with same
237 				 * raid_disk number.
238 				 */
239 	Candidate,		/* For clustered environments only:
240 				 * This device is seen locally but not
241 				 * by the whole cluster
242 				 */
243 	Journal,		/* This device is used as journal for
244 				 * raid-5/6.
245 				 * Usually, this device should be faster
246 				 * than other devices in the array
247 				 */
248 	ClusterRemove,
249 	ExternalBbl,            /* External metadata provides bad
250 				 * block management for a disk
251 				 */
252 	FailFast,		/* Minimal retries should be attempted on
253 				 * this device, so use REQ_FAILFAST_DEV.
254 				 * Also don't try to repair failed reads.
255 				 * It is expects that no bad block log
256 				 * is present.
257 				 */
258 	LastDev,		/* Seems to be the last working dev as
259 				 * it didn't fail, so don't use FailFast
260 				 * any more for metadata
261 				 */
262 	CollisionCheck,		/*
263 				 * check if there is collision between raid1
264 				 * serial bios.
265 				 */
266 	Nonrot,			/* non-rotational device (SSD) */
267 };
268 
269 static inline int is_badblock(struct md_rdev *rdev, sector_t s, int sectors,
270 			      sector_t *first_bad, int *bad_sectors)
271 {
272 	if (unlikely(rdev->badblocks.count)) {
273 		int rv = badblocks_check(&rdev->badblocks, rdev->data_offset + s,
274 					sectors,
275 					first_bad, bad_sectors);
276 		if (rv)
277 			*first_bad -= rdev->data_offset;
278 		return rv;
279 	}
280 	return 0;
281 }
282 
283 static inline int rdev_has_badblock(struct md_rdev *rdev, sector_t s,
284 				    int sectors)
285 {
286 	sector_t first_bad;
287 	int bad_sectors;
288 
289 	return is_badblock(rdev, s, sectors, &first_bad, &bad_sectors);
290 }
291 
292 extern int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
293 			      int is_new);
294 extern int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
295 				int is_new);
296 struct md_cluster_info;
297 
298 /**
299  * enum mddev_flags - md device flags.
300  * @MD_ARRAY_FIRST_USE: First use of array, needs initialization.
301  * @MD_CLOSING: If set, we are closing the array, do not open it then.
302  * @MD_JOURNAL_CLEAN: A raid with journal is already clean.
303  * @MD_HAS_JOURNAL: The raid array has journal feature set.
304  * @MD_CLUSTER_RESYNC_LOCKED: cluster raid only, which means node, already took
305  *			       resync lock, need to release the lock.
306  * @MD_FAILFAST_SUPPORTED: Using MD_FAILFAST on metadata writes is supported as
307  *			    calls to md_error() will never cause the array to
308  *			    become failed.
309  * @MD_HAS_PPL:  The raid array has PPL feature set.
310  * @MD_HAS_MULTIPLE_PPLS: The raid array has multiple PPLs feature set.
311  * @MD_NOT_READY: do_md_run() is active, so 'array_state', ust not report that
312  *		   array is ready yet.
313  * @MD_BROKEN: This is used to stop writes and mark array as failed.
314  * @MD_DELETED: This device is being deleted
315  *
316  * change UNSUPPORTED_MDDEV_FLAGS for each array type if new flag is added
317  */
318 enum mddev_flags {
319 	MD_ARRAY_FIRST_USE,
320 	MD_CLOSING,
321 	MD_JOURNAL_CLEAN,
322 	MD_HAS_JOURNAL,
323 	MD_CLUSTER_RESYNC_LOCKED,
324 	MD_FAILFAST_SUPPORTED,
325 	MD_HAS_PPL,
326 	MD_HAS_MULTIPLE_PPLS,
327 	MD_NOT_READY,
328 	MD_BROKEN,
329 	MD_DELETED,
330 };
331 
332 enum mddev_sb_flags {
333 	MD_SB_CHANGE_DEVS,		/* Some device status has changed */
334 	MD_SB_CHANGE_CLEAN,	/* transition to or from 'clean' */
335 	MD_SB_CHANGE_PENDING,	/* switch from 'clean' to 'active' in progress */
336 	MD_SB_NEED_REWRITE,	/* metadata write needs to be repeated */
337 };
338 
339 #define NR_SERIAL_INFOS		8
340 /* record current range of serialize IOs */
341 struct serial_info {
342 	struct rb_node node;
343 	sector_t start;		/* start sector of rb node */
344 	sector_t last;		/* end sector of rb node */
345 	sector_t _subtree_last; /* highest sector in subtree of rb node */
346 };
347 
348 /*
349  * mddev->curr_resync stores the current sector of the resync but
350  * also has some overloaded values.
351  */
352 enum {
353 	/* No resync in progress */
354 	MD_RESYNC_NONE = 0,
355 	/* Yielded to allow another conflicting resync to commence */
356 	MD_RESYNC_YIELDED = 1,
357 	/* Delayed to check that there is no conflict with another sync */
358 	MD_RESYNC_DELAYED = 2,
359 	/* Any value greater than or equal to this is in an active resync */
360 	MD_RESYNC_ACTIVE = 3,
361 };
362 
363 struct mddev {
364 	void				*private;
365 	struct md_personality		*pers;
366 	dev_t				unit;
367 	int				md_minor;
368 	struct list_head		disks;
369 	unsigned long			flags;
370 	unsigned long			sb_flags;
371 
372 	int				suspended;
373 	struct mutex			suspend_mutex;
374 	struct percpu_ref		active_io;
375 	int				ro;
376 	int				sysfs_active; /* set when sysfs deletes
377 						       * are happening, so run/
378 						       * takeover/stop are not safe
379 						       */
380 	struct gendisk			*gendisk;
381 
382 	struct kobject			kobj;
383 	int				hold_active;
384 #define	UNTIL_IOCTL	1
385 #define	UNTIL_STOP	2
386 
387 	/* Superblock information */
388 	int				major_version,
389 					minor_version,
390 					patch_version;
391 	int				persistent;
392 	int				external;	/* metadata is
393 							 * managed externally */
394 	char				metadata_type[17]; /* externally set*/
395 	int				chunk_sectors;
396 	time64_t			ctime, utime;
397 	int				level, layout;
398 	char				clevel[16];
399 	int				raid_disks;
400 	int				max_disks;
401 	sector_t			dev_sectors;	/* used size of
402 							 * component devices */
403 	sector_t			array_sectors; /* exported array size */
404 	int				external_size; /* size managed
405 							* externally */
406 	__u64				events;
407 	/* If the last 'event' was simply a clean->dirty transition, and
408 	 * we didn't write it to the spares, then it is safe and simple
409 	 * to just decrement the event count on a dirty->clean transition.
410 	 * So we record that possibility here.
411 	 */
412 	int				can_decrease_events;
413 
414 	char				uuid[16];
415 
416 	/* If the array is being reshaped, we need to record the
417 	 * new shape and an indication of where we are up to.
418 	 * This is written to the superblock.
419 	 * If reshape_position is MaxSector, then no reshape is happening (yet).
420 	 */
421 	sector_t			reshape_position;
422 	int				delta_disks, new_level, new_layout;
423 	int				new_chunk_sectors;
424 	int				reshape_backwards;
425 
426 	struct md_thread __rcu		*thread;	/* management thread */
427 	struct md_thread __rcu		*sync_thread;	/* doing resync or reconstruct */
428 
429 	/*
430 	 * Set when a sync operation is started. It holds this value even
431 	 * when the sync thread is "frozen" (interrupted) or "idle" (stopped
432 	 * or finished). It is overwritten when a new sync operation is begun.
433 	 */
434 	enum sync_action		last_sync_action;
435 	sector_t			curr_resync;	/* last block scheduled */
436 	/* As resync requests can complete out of order, we cannot easily track
437 	 * how much resync has been completed.  So we occasionally pause until
438 	 * everything completes, then set curr_resync_completed to curr_resync.
439 	 * As such it may be well behind the real resync mark, but it is a value
440 	 * we are certain of.
441 	 */
442 	sector_t			curr_resync_completed;
443 	unsigned long			resync_mark;	/* a recent timestamp */
444 	sector_t			resync_mark_cnt;/* blocks written at resync_mark */
445 	sector_t			curr_mark_cnt; /* blocks scheduled now */
446 
447 	sector_t			resync_max_sectors; /* may be set by personality */
448 
449 	atomic64_t			resync_mismatches; /* count of sectors where
450 							    * parity/replica mismatch found
451 							    */
452 
453 	/* allow user-space to request suspension of IO to regions of the array */
454 	sector_t			suspend_lo;
455 	sector_t			suspend_hi;
456 	/* if zero, use the system-wide default */
457 	int				sync_speed_min;
458 	int				sync_speed_max;
459 
460 	/* resync even though the same disks are shared among md-devices */
461 	int				parallel_resync;
462 
463 	int				ok_start_degraded;
464 
465 	unsigned long			recovery;
466 	/* If a RAID personality determines that recovery (of a particular
467 	 * device) will fail due to a read error on the source device, it
468 	 * takes a copy of this number and does not attempt recovery again
469 	 * until this number changes.
470 	 */
471 	int				recovery_disabled;
472 
473 	int				in_sync;	/* know to not need resync */
474 	/* 'open_mutex' avoids races between 'md_open' and 'do_md_stop', so
475 	 * that we are never stopping an array while it is open.
476 	 * 'reconfig_mutex' protects all other reconfiguration.
477 	 * These locks are separate due to conflicting interactions
478 	 * with disk->open_mutex.
479 	 * Lock ordering is:
480 	 *  reconfig_mutex -> disk->open_mutex
481 	 *  disk->open_mutex -> open_mutex:  e.g. __blkdev_get -> md_open
482 	 */
483 	struct mutex			open_mutex;
484 	struct mutex			reconfig_mutex;
485 	atomic_t			active;		/* general refcount */
486 	atomic_t			openers;	/* number of active opens */
487 
488 	int				changed;	/* True if we might need to
489 							 * reread partition info */
490 	int				degraded;	/* whether md should consider
491 							 * adding a spare
492 							 */
493 
494 	atomic_t			recovery_active; /* blocks scheduled, but not written */
495 	wait_queue_head_t		recovery_wait;
496 	sector_t			recovery_cp;
497 	sector_t			resync_min;	/* user requested sync
498 							 * starts here */
499 	sector_t			resync_max;	/* resync should pause
500 							 * when it gets here */
501 
502 	struct kernfs_node		*sysfs_state;	/* handle for 'array_state'
503 							 * file in sysfs.
504 							 */
505 	struct kernfs_node		*sysfs_action;  /* handle for 'sync_action' */
506 	struct kernfs_node		*sysfs_completed;	/*handle for 'sync_completed' */
507 	struct kernfs_node		*sysfs_degraded;	/*handle for 'degraded' */
508 	struct kernfs_node		*sysfs_level;		/*handle for 'level' */
509 
510 	/* used for delayed sysfs removal */
511 	struct work_struct del_work;
512 	/* used for register new sync thread */
513 	struct work_struct sync_work;
514 
515 	/* "lock" protects:
516 	 *   flush_bio transition from NULL to !NULL
517 	 *   rdev superblocks, events
518 	 *   clearing MD_CHANGE_*
519 	 *   in_sync - and related safemode and MD_CHANGE changes
520 	 *   pers (also protected by reconfig_mutex and pending IO).
521 	 *   clearing ->bitmap
522 	 *   clearing ->bitmap_info.file
523 	 *   changing ->resync_{min,max}
524 	 *   setting MD_RECOVERY_RUNNING (which interacts with resync_{min,max})
525 	 */
526 	spinlock_t			lock;
527 	wait_queue_head_t		sb_wait;	/* for waiting on superblock updates */
528 	atomic_t			pending_writes;	/* number of active superblock writes */
529 
530 	unsigned int			safemode;	/* if set, update "clean" superblock
531 							 * when no writes pending.
532 							 */
533 	unsigned int			safemode_delay;
534 	struct timer_list		safemode_timer;
535 	struct percpu_ref		writes_pending;
536 	int				sync_checkers;	/* # of threads checking writes_pending */
537 
538 	struct bitmap			*bitmap; /* the bitmap for the device */
539 	struct {
540 		struct file		*file; /* the bitmap file */
541 		loff_t			offset; /* offset from superblock of
542 						 * start of bitmap. May be
543 						 * negative, but not '0'
544 						 * For external metadata, offset
545 						 * from start of device.
546 						 */
547 		unsigned long		space; /* space available at this offset */
548 		loff_t			default_offset; /* this is the offset to use when
549 							 * hot-adding a bitmap.  It should
550 							 * eventually be settable by sysfs.
551 							 */
552 		unsigned long		default_space; /* space available at
553 							* default offset */
554 		struct mutex		mutex;
555 		unsigned long		chunksize;
556 		unsigned long		daemon_sleep; /* how many jiffies between updates? */
557 		unsigned long		max_write_behind; /* write-behind mode */
558 		int			external;
559 		int			nodes; /* Maximum number of nodes in the cluster */
560 		char                    cluster_name[64]; /* Name of the cluster */
561 	} bitmap_info;
562 
563 	atomic_t			max_corr_read_errors; /* max read retries */
564 	struct list_head		all_mddevs;
565 
566 	const struct attribute_group	*to_remove;
567 
568 	struct bio_set			bio_set;
569 	struct bio_set			sync_set; /* for sync operations like
570 						   * metadata and bitmap writes
571 						   */
572 	struct bio_set			io_clone_set;
573 
574 	/* Generic flush handling.
575 	 * The last to finish preflush schedules a worker to submit
576 	 * the rest of the request (without the REQ_PREFLUSH flag).
577 	 */
578 	struct bio *flush_bio;
579 	atomic_t flush_pending;
580 	ktime_t start_flush, prev_flush_start; /* prev_flush_start is when the previous completed
581 						* flush was started.
582 						*/
583 	struct work_struct flush_work;
584 	struct work_struct event_work;	/* used by dm to report failure event */
585 	mempool_t *serial_info_pool;
586 	void (*sync_super)(struct mddev *mddev, struct md_rdev *rdev);
587 	struct md_cluster_info		*cluster_info;
588 	unsigned int			good_device_nr;	/* good device num within cluster raid */
589 	unsigned int			noio_flag; /* for memalloc scope API */
590 
591 	/*
592 	 * Temporarily store rdev that will be finally removed when
593 	 * reconfig_mutex is unlocked, protected by reconfig_mutex.
594 	 */
595 	struct list_head		deleting;
596 
597 	/* The sequence number for sync thread */
598 	atomic_t sync_seq;
599 
600 	bool	has_superblocks:1;
601 	bool	fail_last_dev:1;
602 	bool	serialize_policy:1;
603 };
604 
605 enum recovery_flags {
606 	/* flags for sync thread running status */
607 
608 	/*
609 	 * set when one of sync action is set and new sync thread need to be
610 	 * registered, or just add/remove spares from conf.
611 	 */
612 	MD_RECOVERY_NEEDED,
613 	/* sync thread is running, or about to be started */
614 	MD_RECOVERY_RUNNING,
615 	/* sync thread needs to be aborted for some reason */
616 	MD_RECOVERY_INTR,
617 	/* sync thread is done and is waiting to be unregistered */
618 	MD_RECOVERY_DONE,
619 	/* running sync thread must abort immediately, and not restart */
620 	MD_RECOVERY_FROZEN,
621 	/* waiting for pers->start() to finish */
622 	MD_RECOVERY_WAIT,
623 	/* interrupted because io-error */
624 	MD_RECOVERY_ERROR,
625 
626 	/* flags determines sync action, see details in enum sync_action */
627 
628 	/* if just this flag is set, action is resync. */
629 	MD_RECOVERY_SYNC,
630 	/*
631 	 * paired with MD_RECOVERY_SYNC, if MD_RECOVERY_CHECK is not set,
632 	 * action is repair, means user requested resync.
633 	 */
634 	MD_RECOVERY_REQUESTED,
635 	/*
636 	 * paired with MD_RECOVERY_SYNC and MD_RECOVERY_REQUESTED, action is
637 	 * check.
638 	 */
639 	MD_RECOVERY_CHECK,
640 	/* recovery, or need to try it */
641 	MD_RECOVERY_RECOVER,
642 	/* reshape */
643 	MD_RECOVERY_RESHAPE,
644 	/* remote node is running resync thread */
645 	MD_RESYNCING_REMOTE,
646 };
647 
648 enum md_ro_state {
649 	MD_RDWR,
650 	MD_RDONLY,
651 	MD_AUTO_READ,
652 	MD_MAX_STATE
653 };
654 
655 static inline bool md_is_rdwr(struct mddev *mddev)
656 {
657 	return (mddev->ro == MD_RDWR);
658 }
659 
660 static inline bool reshape_interrupted(struct mddev *mddev)
661 {
662 	/* reshape never start */
663 	if (mddev->reshape_position == MaxSector)
664 		return false;
665 
666 	/* interrupted */
667 	if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
668 		return true;
669 
670 	/* running reshape will be interrupted soon. */
671 	if (test_bit(MD_RECOVERY_WAIT, &mddev->recovery) ||
672 	    test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
673 	    test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
674 		return true;
675 
676 	return false;
677 }
678 
679 static inline int __must_check mddev_lock(struct mddev *mddev)
680 {
681 	return mutex_lock_interruptible(&mddev->reconfig_mutex);
682 }
683 
684 /* Sometimes we need to take the lock in a situation where
685  * failure due to interrupts is not acceptable.
686  */
687 static inline void mddev_lock_nointr(struct mddev *mddev)
688 {
689 	mutex_lock(&mddev->reconfig_mutex);
690 }
691 
692 static inline int mddev_trylock(struct mddev *mddev)
693 {
694 	return mutex_trylock(&mddev->reconfig_mutex);
695 }
696 extern void mddev_unlock(struct mddev *mddev);
697 
698 static inline void md_sync_acct(struct block_device *bdev, unsigned long nr_sectors)
699 {
700 	if (blk_queue_io_stat(bdev->bd_disk->queue))
701 		atomic_add(nr_sectors, &bdev->bd_disk->sync_io);
702 }
703 
704 static inline void md_sync_acct_bio(struct bio *bio, unsigned long nr_sectors)
705 {
706 	md_sync_acct(bio->bi_bdev, nr_sectors);
707 }
708 
709 struct md_personality
710 {
711 	char *name;
712 	int level;
713 	struct list_head list;
714 	struct module *owner;
715 	bool __must_check (*make_request)(struct mddev *mddev, struct bio *bio);
716 	/*
717 	 * start up works that do NOT require md_thread. tasks that
718 	 * requires md_thread should go into start()
719 	 */
720 	int (*run)(struct mddev *mddev);
721 	/* start up works that require md threads */
722 	int (*start)(struct mddev *mddev);
723 	void (*free)(struct mddev *mddev, void *priv);
724 	void (*status)(struct seq_file *seq, struct mddev *mddev);
725 	/* error_handler must set ->faulty and clear ->in_sync
726 	 * if appropriate, and should abort recovery if needed
727 	 */
728 	void (*error_handler)(struct mddev *mddev, struct md_rdev *rdev);
729 	int (*hot_add_disk) (struct mddev *mddev, struct md_rdev *rdev);
730 	int (*hot_remove_disk) (struct mddev *mddev, struct md_rdev *rdev);
731 	int (*spare_active) (struct mddev *mddev);
732 	sector_t (*sync_request)(struct mddev *mddev, sector_t sector_nr,
733 				 sector_t max_sector, int *skipped);
734 	int (*resize) (struct mddev *mddev, sector_t sectors);
735 	sector_t (*size) (struct mddev *mddev, sector_t sectors, int raid_disks);
736 	int (*check_reshape) (struct mddev *mddev);
737 	int (*start_reshape) (struct mddev *mddev);
738 	void (*finish_reshape) (struct mddev *mddev);
739 	void (*update_reshape_pos) (struct mddev *mddev);
740 	void (*prepare_suspend) (struct mddev *mddev);
741 	/* quiesce suspends or resumes internal processing.
742 	 * 1 - stop new actions and wait for action io to complete
743 	 * 0 - return to normal behaviour
744 	 */
745 	void (*quiesce) (struct mddev *mddev, int quiesce);
746 	/* takeover is used to transition an array from one
747 	 * personality to another.  The new personality must be able
748 	 * to handle the data in the current layout.
749 	 * e.g. 2drive raid1 -> 2drive raid5
750 	 *      ndrive raid5 -> degraded n+1drive raid6 with special layout
751 	 * If the takeover succeeds, a new 'private' structure is returned.
752 	 * This needs to be installed and then ->run used to activate the
753 	 * array.
754 	 */
755 	void *(*takeover) (struct mddev *mddev);
756 	/* Changes the consistency policy of an active array. */
757 	int (*change_consistency_policy)(struct mddev *mddev, const char *buf);
758 };
759 
760 struct md_sysfs_entry {
761 	struct attribute attr;
762 	ssize_t (*show)(struct mddev *, char *);
763 	ssize_t (*store)(struct mddev *, const char *, size_t);
764 };
765 extern const struct attribute_group md_bitmap_group;
766 
767 static inline struct kernfs_node *sysfs_get_dirent_safe(struct kernfs_node *sd, char *name)
768 {
769 	if (sd)
770 		return sysfs_get_dirent(sd, name);
771 	return sd;
772 }
773 static inline void sysfs_notify_dirent_safe(struct kernfs_node *sd)
774 {
775 	if (sd)
776 		sysfs_notify_dirent(sd);
777 }
778 
779 static inline char * mdname (struct mddev * mddev)
780 {
781 	return mddev->gendisk ? mddev->gendisk->disk_name : "mdX";
782 }
783 
784 static inline int sysfs_link_rdev(struct mddev *mddev, struct md_rdev *rdev)
785 {
786 	char nm[20];
787 	if (!test_bit(Replacement, &rdev->flags) &&
788 	    !test_bit(Journal, &rdev->flags) &&
789 	    mddev->kobj.sd) {
790 		sprintf(nm, "rd%d", rdev->raid_disk);
791 		return sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
792 	} else
793 		return 0;
794 }
795 
796 static inline void sysfs_unlink_rdev(struct mddev *mddev, struct md_rdev *rdev)
797 {
798 	char nm[20];
799 	if (!test_bit(Replacement, &rdev->flags) &&
800 	    !test_bit(Journal, &rdev->flags) &&
801 	    mddev->kobj.sd) {
802 		sprintf(nm, "rd%d", rdev->raid_disk);
803 		sysfs_remove_link(&mddev->kobj, nm);
804 	}
805 }
806 
807 /*
808  * iterates through some rdev ringlist. It's safe to remove the
809  * current 'rdev'. Dont touch 'tmp' though.
810  */
811 #define rdev_for_each_list(rdev, tmp, head)				\
812 	list_for_each_entry_safe(rdev, tmp, head, same_set)
813 
814 /*
815  * iterates through the 'same array disks' ringlist
816  */
817 #define rdev_for_each(rdev, mddev)				\
818 	list_for_each_entry(rdev, &((mddev)->disks), same_set)
819 
820 #define rdev_for_each_safe(rdev, tmp, mddev)				\
821 	list_for_each_entry_safe(rdev, tmp, &((mddev)->disks), same_set)
822 
823 #define rdev_for_each_rcu(rdev, mddev)				\
824 	list_for_each_entry_rcu(rdev, &((mddev)->disks), same_set)
825 
826 struct md_thread {
827 	void			(*run) (struct md_thread *thread);
828 	struct mddev		*mddev;
829 	wait_queue_head_t	wqueue;
830 	unsigned long		flags;
831 	struct task_struct	*tsk;
832 	unsigned long		timeout;
833 	void			*private;
834 };
835 
836 struct md_io_clone {
837 	struct mddev	*mddev;
838 	struct bio	*orig_bio;
839 	unsigned long	start_time;
840 	struct bio	bio_clone;
841 };
842 
843 #define THREAD_WAKEUP  0
844 
845 static inline void safe_put_page(struct page *p)
846 {
847 	if (p) put_page(p);
848 }
849 
850 extern int register_md_personality(struct md_personality *p);
851 extern int unregister_md_personality(struct md_personality *p);
852 extern int register_md_cluster_operations(const struct md_cluster_operations *ops,
853 		struct module *module);
854 extern int unregister_md_cluster_operations(void);
855 extern int md_setup_cluster(struct mddev *mddev, int nodes);
856 extern void md_cluster_stop(struct mddev *mddev);
857 extern struct md_thread *md_register_thread(
858 	void (*run)(struct md_thread *thread),
859 	struct mddev *mddev,
860 	const char *name);
861 extern void md_unregister_thread(struct mddev *mddev, struct md_thread __rcu **threadp);
862 extern void md_wakeup_thread(struct md_thread __rcu *thread);
863 extern void md_check_recovery(struct mddev *mddev);
864 extern void md_reap_sync_thread(struct mddev *mddev);
865 extern enum sync_action md_sync_action(struct mddev *mddev);
866 extern enum sync_action md_sync_action_by_name(const char *page);
867 extern const char *md_sync_action_name(enum sync_action action);
868 extern void md_write_start(struct mddev *mddev, struct bio *bi);
869 extern void md_write_inc(struct mddev *mddev, struct bio *bi);
870 extern void md_write_end(struct mddev *mddev);
871 extern void md_done_sync(struct mddev *mddev, int blocks, int ok);
872 extern void md_error(struct mddev *mddev, struct md_rdev *rdev);
873 extern void md_finish_reshape(struct mddev *mddev);
874 void md_submit_discard_bio(struct mddev *mddev, struct md_rdev *rdev,
875 			struct bio *bio, sector_t start, sector_t size);
876 void md_account_bio(struct mddev *mddev, struct bio **bio);
877 void md_free_cloned_bio(struct bio *bio);
878 
879 extern bool __must_check md_flush_request(struct mddev *mddev, struct bio *bio);
880 extern void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
881 			   sector_t sector, int size, struct page *page);
882 extern int md_super_wait(struct mddev *mddev);
883 extern int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
884 		struct page *page, blk_opf_t opf, bool metadata_op);
885 extern void md_do_sync(struct md_thread *thread);
886 extern void md_new_event(void);
887 extern void md_allow_write(struct mddev *mddev);
888 extern void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev);
889 extern void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors);
890 extern int md_check_no_bitmap(struct mddev *mddev);
891 extern int md_integrity_register(struct mddev *mddev);
892 extern int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale);
893 
894 extern int mddev_init(struct mddev *mddev);
895 extern void mddev_destroy(struct mddev *mddev);
896 void md_init_stacking_limits(struct queue_limits *lim);
897 struct mddev *md_alloc(dev_t dev, char *name);
898 void mddev_put(struct mddev *mddev);
899 extern int md_run(struct mddev *mddev);
900 extern int md_start(struct mddev *mddev);
901 extern void md_stop(struct mddev *mddev);
902 extern void md_stop_writes(struct mddev *mddev);
903 extern int md_rdev_init(struct md_rdev *rdev);
904 extern void md_rdev_clear(struct md_rdev *rdev);
905 
906 extern bool md_handle_request(struct mddev *mddev, struct bio *bio);
907 extern int mddev_suspend(struct mddev *mddev, bool interruptible);
908 extern void mddev_resume(struct mddev *mddev);
909 extern void md_idle_sync_thread(struct mddev *mddev);
910 extern void md_frozen_sync_thread(struct mddev *mddev);
911 extern void md_unfrozen_sync_thread(struct mddev *mddev);
912 
913 extern void md_reload_sb(struct mddev *mddev, int raid_disk);
914 extern void md_update_sb(struct mddev *mddev, int force);
915 extern void mddev_create_serial_pool(struct mddev *mddev, struct md_rdev *rdev);
916 extern void mddev_destroy_serial_pool(struct mddev *mddev,
917 				      struct md_rdev *rdev);
918 struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr);
919 struct md_rdev *md_find_rdev_rcu(struct mddev *mddev, dev_t dev);
920 
921 static inline bool is_rdev_broken(struct md_rdev *rdev)
922 {
923 	return !disk_live(rdev->bdev->bd_disk);
924 }
925 
926 static inline void rdev_dec_pending(struct md_rdev *rdev, struct mddev *mddev)
927 {
928 	int faulty = test_bit(Faulty, &rdev->flags);
929 	if (atomic_dec_and_test(&rdev->nr_pending) && faulty) {
930 		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
931 		md_wakeup_thread(mddev->thread);
932 	}
933 }
934 
935 extern const struct md_cluster_operations *md_cluster_ops;
936 static inline int mddev_is_clustered(struct mddev *mddev)
937 {
938 	return mddev->cluster_info && mddev->bitmap_info.nodes > 1;
939 }
940 
941 /* clear unsupported mddev_flags */
942 static inline void mddev_clear_unsupported_flags(struct mddev *mddev,
943 	unsigned long unsupported_flags)
944 {
945 	mddev->flags &= ~unsupported_flags;
946 }
947 
948 static inline void mddev_check_write_zeroes(struct mddev *mddev, struct bio *bio)
949 {
950 	if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
951 	    !bio->bi_bdev->bd_disk->queue->limits.max_write_zeroes_sectors)
952 		mddev->gendisk->queue->limits.max_write_zeroes_sectors = 0;
953 }
954 
955 static inline int mddev_suspend_and_lock(struct mddev *mddev)
956 {
957 	int ret;
958 
959 	ret = mddev_suspend(mddev, true);
960 	if (ret)
961 		return ret;
962 
963 	ret = mddev_lock(mddev);
964 	if (ret)
965 		mddev_resume(mddev);
966 
967 	return ret;
968 }
969 
970 static inline void mddev_suspend_and_lock_nointr(struct mddev *mddev)
971 {
972 	mddev_suspend(mddev, false);
973 	mutex_lock(&mddev->reconfig_mutex);
974 }
975 
976 static inline void mddev_unlock_and_resume(struct mddev *mddev)
977 {
978 	mddev_unlock(mddev);
979 	mddev_resume(mddev);
980 }
981 
982 struct mdu_array_info_s;
983 struct mdu_disk_info_s;
984 
985 extern int mdp_major;
986 extern struct workqueue_struct *md_bitmap_wq;
987 void md_autostart_arrays(int part);
988 int md_set_array_info(struct mddev *mddev, struct mdu_array_info_s *info);
989 int md_add_new_disk(struct mddev *mddev, struct mdu_disk_info_s *info);
990 int do_md_run(struct mddev *mddev);
991 #define MDDEV_STACK_INTEGRITY	(1u << 0)
992 int mddev_stack_rdev_limits(struct mddev *mddev, struct queue_limits *lim,
993 		unsigned int flags);
994 int mddev_stack_new_rdev(struct mddev *mddev, struct md_rdev *rdev);
995 void mddev_update_io_opt(struct mddev *mddev, unsigned int nr_stripes);
996 
997 extern const struct block_device_operations md_fops;
998 
999 /*
1000  * MD devices can be used undeneath by DM, in which case ->gendisk is NULL.
1001  */
1002 static inline bool mddev_is_dm(struct mddev *mddev)
1003 {
1004 	return !mddev->gendisk;
1005 }
1006 
1007 static inline void mddev_trace_remap(struct mddev *mddev, struct bio *bio,
1008 		sector_t sector)
1009 {
1010 	if (!mddev_is_dm(mddev))
1011 		trace_block_bio_remap(bio, disk_devt(mddev->gendisk), sector);
1012 }
1013 
1014 #define mddev_add_trace_msg(mddev, fmt, args...)			\
1015 do {									\
1016 	if (!mddev_is_dm(mddev))					\
1017 		blk_add_trace_msg((mddev)->gendisk->queue, fmt, ##args); \
1018 } while (0)
1019 
1020 #endif /* _MD_MD_H */
1021