xref: /linux/drivers/md/md.h (revision 69050f8d6d075dc01af7a5f2f550a8067510366f)
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  * Number of guaranteed raid bios in case of extreme VM load:
27  */
28 #define	NR_RAID_BIOS 256
29 
30 enum md_submodule_type {
31 	MD_PERSONALITY = 0,
32 	MD_CLUSTER,
33 	MD_BITMAP,
34 };
35 
36 enum md_submodule_id {
37 	ID_LINEAR	= LEVEL_LINEAR,
38 	ID_RAID0	= 0,
39 	ID_RAID1	= 1,
40 	ID_RAID4	= 4,
41 	ID_RAID5	= 5,
42 	ID_RAID6	= 6,
43 	ID_RAID10	= 10,
44 	ID_CLUSTER,
45 	ID_BITMAP,
46 	ID_LLBITMAP,
47 	ID_BITMAP_NONE,
48 };
49 
50 struct md_submodule_head {
51 	enum md_submodule_type	type;
52 	enum md_submodule_id	id;
53 	const char		*name;
54 	struct module		*owner;
55 };
56 
57 /*
58  * These flags should really be called "NO_RETRY" rather than
59  * "FAILFAST" because they don't make any promise about time lapse,
60  * only about the number of retries, which will be zero.
61  * REQ_FAILFAST_DRIVER is not included because
62  * Commit: 4a27446f3e39 ("[SCSI] modify scsi to handle new fail fast flags.")
63  * seems to suggest that the errors it avoids retrying should usually
64  * be retried.
65  */
66 #define	MD_FAILFAST	(REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT)
67 
68 /* Status of sync thread. */
69 enum sync_action {
70 	/*
71 	 * Represent by MD_RECOVERY_SYNC, start when:
72 	 * 1) after assemble, sync data from first rdev to other copies, this
73 	 * must be done first before other sync actions and will only execute
74 	 * once;
75 	 * 2) resize the array(notice that this is not reshape), sync data for
76 	 * the new range;
77 	 */
78 	ACTION_RESYNC,
79 	/*
80 	 * Represent by MD_RECOVERY_RECOVER, start when:
81 	 * 1) for new replacement, sync data based on the replace rdev or
82 	 * available copies from other rdev;
83 	 * 2) for new member disk while the array is degraded, sync data from
84 	 * other rdev;
85 	 * 3) reassemble after power failure or re-add a hot removed rdev, sync
86 	 * data from first rdev to other copies based on bitmap;
87 	 */
88 	ACTION_RECOVER,
89 	/*
90 	 * Represent by MD_RECOVERY_SYNC | MD_RECOVERY_REQUESTED |
91 	 * MD_RECOVERY_CHECK, start when user echo "check" to sysfs api
92 	 * sync_action, used to check if data copies from differenct rdev are
93 	 * the same. The number of mismatch sectors will be exported to user
94 	 * by sysfs api mismatch_cnt;
95 	 */
96 	ACTION_CHECK,
97 	/*
98 	 * Represent by MD_RECOVERY_SYNC | MD_RECOVERY_REQUESTED, start when
99 	 * user echo "repair" to sysfs api sync_action, usually paired with
100 	 * ACTION_CHECK, used to force syncing data once user found that there
101 	 * are inconsistent data,
102 	 */
103 	ACTION_REPAIR,
104 	/*
105 	 * Represent by MD_RECOVERY_RESHAPE, start when new member disk is added
106 	 * to the conf, notice that this is different from spares or
107 	 * replacement;
108 	 */
109 	ACTION_RESHAPE,
110 	/*
111 	 * Represent by MD_RECOVERY_FROZEN, can be set by sysfs api sync_action
112 	 * or internal usage like setting the array read-only, will forbid above
113 	 * actions.
114 	 */
115 	ACTION_FROZEN,
116 	/*
117 	 * All above actions don't match.
118 	 */
119 	ACTION_IDLE,
120 	NR_SYNC_ACTIONS,
121 };
122 
123 /*
124  * The struct embedded in rdev is used to serialize IO.
125  */
126 struct serial_in_rdev {
127 	struct rb_root_cached serial_rb;
128 	spinlock_t serial_lock;
129 	wait_queue_head_t serial_io_wait;
130 };
131 
132 /*
133  * MD's 'extended' device
134  */
135 struct md_rdev {
136 	struct list_head same_set;	/* RAID devices within the same set */
137 
138 	sector_t sectors;		/* Device size (in 512bytes sectors) */
139 	struct mddev *mddev;		/* RAID array if running */
140 	unsigned long last_events;	/* IO event timestamp */
141 
142 	/*
143 	 * If meta_bdev is non-NULL, it means that a separate device is
144 	 * being used to store the metadata (superblock/bitmap) which
145 	 * would otherwise be contained on the same device as the data (bdev).
146 	 */
147 	struct block_device *meta_bdev;
148 	struct block_device *bdev;	/* block device handle */
149 	struct file *bdev_file;		/* Handle from open for bdev */
150 
151 	struct page	*sb_page, *bb_page;
152 	int		sb_loaded;
153 	__u64		sb_events;
154 	sector_t	data_offset;	/* start of data in array */
155 	sector_t	new_data_offset;/* only relevant while reshaping */
156 	sector_t	sb_start;	/* offset of the super block (in 512byte sectors) */
157 	int		sb_size;	/* bytes in the superblock */
158 	int		preferred_minor;	/* autorun support */
159 
160 	struct kobject	kobj;
161 
162 	/* A device can be in one of three states based on two flags:
163 	 * Not working:   faulty==1 in_sync==0
164 	 * Fully working: faulty==0 in_sync==1
165 	 * Working, but not
166 	 * in sync with array
167 	 *                faulty==0 in_sync==0
168 	 *
169 	 * It can never have faulty==1, in_sync==1
170 	 * This reduces the burden of testing multiple flags in many cases
171 	 */
172 
173 	unsigned long	flags;	/* bit set of 'enum flag_bits' bits. */
174 	wait_queue_head_t blocked_wait;
175 
176 	int desc_nr;			/* descriptor index in the superblock */
177 	int raid_disk;			/* role of device in array */
178 	int new_raid_disk;		/* role that the device will have in
179 					 * the array after a level-change completes.
180 					 */
181 	int saved_raid_disk;		/* role that device used to have in the
182 					 * array and could again if we did a partial
183 					 * resync from the bitmap
184 					 */
185 	union {
186 		sector_t recovery_offset;/* If this device has been partially
187 					 * recovered, this is where we were
188 					 * up to.
189 					 */
190 		sector_t journal_tail;	/* If this device is a journal device,
191 					 * this is the journal tail (journal
192 					 * recovery start point)
193 					 */
194 	};
195 
196 	atomic_t	nr_pending;	/* number of pending requests.
197 					 * only maintained for arrays that
198 					 * support hot removal
199 					 */
200 	atomic_t	read_errors;	/* number of consecutive read errors that
201 					 * we have tried to ignore.
202 					 */
203 	time64_t	last_read_error;	/* monotonic time since our
204 						 * last read error
205 						 */
206 	atomic_t	corrected_errors; /* number of corrected read errors,
207 					   * for reporting to userspace and storing
208 					   * in superblock.
209 					   */
210 
211 	struct serial_in_rdev *serial;  /* used for raid1 io serialization */
212 
213 	struct kernfs_node *sysfs_state; /* handle for 'state'
214 					   * sysfs entry */
215 	/* handle for 'unacknowledged_bad_blocks' sysfs dentry */
216 	struct kernfs_node *sysfs_unack_badblocks;
217 	/* handle for 'bad_blocks' sysfs dentry */
218 	struct kernfs_node *sysfs_badblocks;
219 	struct badblocks badblocks;
220 
221 	struct {
222 		short offset;	/* Offset from superblock to start of PPL.
223 				 * Not used by external metadata. */
224 		unsigned int size;	/* Size in sectors of the PPL space */
225 		sector_t sector;	/* First sector of the PPL space */
226 	} ppl;
227 };
228 enum flag_bits {
229 	Faulty,			/* device is known to have a fault */
230 	In_sync,		/* device is in_sync with rest of array */
231 	Bitmap_sync,		/* ..actually, not quite In_sync.  Need a
232 				 * bitmap-based recovery to get fully in sync.
233 				 * The bit is only meaningful before device
234 				 * has been passed to pers->hot_add_disk.
235 				 */
236 	WriteMostly,		/* Avoid reading if at all possible */
237 	AutoDetected,		/* added by auto-detect */
238 	Blocked,		/* An error occurred but has not yet
239 				 * been acknowledged by the metadata
240 				 * handler, so don't allow writes
241 				 * until it is cleared */
242 	WriteErrorSeen,		/* A write error has been seen on this
243 				 * device
244 				 */
245 	FaultRecorded,		/* Intermediate state for clearing
246 				 * Blocked.  The Fault is/will-be
247 				 * recorded in the metadata, but that
248 				 * metadata hasn't been stored safely
249 				 * on disk yet.
250 				 */
251 	BlockedBadBlocks,	/* A writer is blocked because they
252 				 * found an unacknowledged bad-block.
253 				 * This can safely be cleared at any
254 				 * time, and the writer will re-check.
255 				 * It may be set at any time, and at
256 				 * worst the writer will timeout and
257 				 * re-check.  So setting it as
258 				 * accurately as possible is good, but
259 				 * not absolutely critical.
260 				 */
261 	WantReplacement,	/* This device is a candidate to be
262 				 * hot-replaced, either because it has
263 				 * reported some faults, or because
264 				 * of explicit request.
265 				 */
266 	Replacement,		/* This device is a replacement for
267 				 * a want_replacement device with same
268 				 * raid_disk number.
269 				 */
270 	Candidate,		/* For clustered environments only:
271 				 * This device is seen locally but not
272 				 * by the whole cluster
273 				 */
274 	Journal,		/* This device is used as journal for
275 				 * raid-5/6.
276 				 * Usually, this device should be faster
277 				 * than other devices in the array
278 				 */
279 	ClusterRemove,
280 	ExternalBbl,            /* External metadata provides bad
281 				 * block management for a disk
282 				 */
283 	FailFast,		/* Minimal retries should be attempted on
284 				 * this device, so use REQ_FAILFAST_DEV.
285 				 * Also don't try to repair failed reads.
286 				 * It is expects that no bad block log
287 				 * is present.
288 				 */
289 	LastDev,		/* Seems to be the last working dev as
290 				 * it didn't fail, so don't use FailFast
291 				 * any more for metadata
292 				 */
293 	CollisionCheck,		/*
294 				 * check if there is collision between raid1
295 				 * serial bios.
296 				 */
297 	Nonrot,			/* non-rotational device (SSD) */
298 };
299 
300 static inline int is_badblock(struct md_rdev *rdev, sector_t s, sector_t sectors,
301 			      sector_t *first_bad, sector_t *bad_sectors)
302 {
303 	if (unlikely(rdev->badblocks.count)) {
304 		int rv = badblocks_check(&rdev->badblocks, rdev->data_offset + s,
305 					sectors,
306 					first_bad, bad_sectors);
307 		if (rv)
308 			*first_bad -= rdev->data_offset;
309 		return rv;
310 	}
311 	return 0;
312 }
313 
314 static inline int rdev_has_badblock(struct md_rdev *rdev, sector_t s,
315 				    int sectors)
316 {
317 	sector_t first_bad;
318 	sector_t bad_sectors;
319 
320 	return is_badblock(rdev, s, sectors, &first_bad, &bad_sectors);
321 }
322 
323 extern bool rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
324 			       int is_new);
325 extern void rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
326 				 int is_new);
327 struct md_cluster_info;
328 struct md_cluster_operations;
329 
330 /**
331  * enum mddev_flags - md device flags.
332  * @MD_ARRAY_FIRST_USE: First use of array, needs initialization.
333  * @MD_CLOSING: If set, we are closing the array, do not open it then.
334  * @MD_JOURNAL_CLEAN: A raid with journal is already clean.
335  * @MD_HAS_JOURNAL: The raid array has journal feature set.
336  * @MD_CLUSTER_RESYNC_LOCKED: cluster raid only, which means node, already took
337  *			       resync lock, need to release the lock.
338  * @MD_FAILFAST_SUPPORTED: Using MD_FAILFAST on metadata writes is supported as
339  *			    calls to md_error() will never cause the array to
340  *			    become failed.
341  * @MD_HAS_PPL:  The raid array has PPL feature set.
342  * @MD_HAS_MULTIPLE_PPLS: The raid array has multiple PPLs feature set.
343  * @MD_NOT_READY: do_md_run() is active, so 'array_state', ust not report that
344  *		   array is ready yet.
345  * @MD_BROKEN: This is used to stop writes and mark array as failed.
346  * @MD_DELETED: This device is being deleted
347  * @MD_HAS_SUPERBLOCK: There is persistence sb in member disks.
348  * @MD_FAILLAST_DEV: Allow last rdev to be removed.
349  * @MD_SERIALIZE_POLICY: Enforce write IO is not reordered, just used by raid1.
350  *
351  * change UNSUPPORTED_MDDEV_FLAGS for each array type if new flag is added
352  */
353 enum mddev_flags {
354 	MD_ARRAY_FIRST_USE,
355 	MD_CLOSING,
356 	MD_JOURNAL_CLEAN,
357 	MD_HAS_JOURNAL,
358 	MD_CLUSTER_RESYNC_LOCKED,
359 	MD_FAILFAST_SUPPORTED,
360 	MD_HAS_PPL,
361 	MD_HAS_MULTIPLE_PPLS,
362 	MD_NOT_READY,
363 	MD_BROKEN,
364 	MD_DO_DELETE,
365 	MD_DELETED,
366 	MD_HAS_SUPERBLOCK,
367 	MD_FAILLAST_DEV,
368 	MD_SERIALIZE_POLICY,
369 };
370 
371 enum mddev_sb_flags {
372 	MD_SB_CHANGE_DEVS,		/* Some device status has changed */
373 	MD_SB_CHANGE_CLEAN,	/* transition to or from 'clean' */
374 	MD_SB_CHANGE_PENDING,	/* switch from 'clean' to 'active' in progress */
375 	MD_SB_NEED_REWRITE,	/* metadata write needs to be repeated */
376 };
377 
378 #define NR_SERIAL_INFOS		8
379 /* record current range of serialize IOs */
380 struct serial_info {
381 	struct rb_node node;
382 	sector_t start;		/* start sector of rb node */
383 	sector_t last;		/* end sector of rb node */
384 	sector_t _subtree_last; /* highest sector in subtree of rb node */
385 };
386 
387 /*
388  * mddev->curr_resync stores the current sector of the resync but
389  * also has some overloaded values.
390  */
391 enum {
392 	/* No resync in progress */
393 	MD_RESYNC_NONE = 0,
394 	/* Yielded to allow another conflicting resync to commence */
395 	MD_RESYNC_YIELDED = 1,
396 	/* Delayed to check that there is no conflict with another sync */
397 	MD_RESYNC_DELAYED = 2,
398 	/* Any value greater than or equal to this is in an active resync */
399 	MD_RESYNC_ACTIVE = 3,
400 };
401 
402 struct mddev {
403 	void				*private;
404 	struct md_personality		*pers;
405 	dev_t				unit;
406 	int				md_minor;
407 	struct list_head		disks;
408 	unsigned long			flags;
409 	unsigned long			sb_flags;
410 
411 	int				suspended;
412 	struct mutex			suspend_mutex;
413 	struct percpu_ref		active_io;
414 	int				ro;
415 	int				sysfs_active; /* set when sysfs deletes
416 						       * are happening, so run/
417 						       * takeover/stop are not safe
418 						       */
419 	struct gendisk			*gendisk;    /* mdraid gendisk */
420 	struct gendisk			*dm_gendisk; /* dm-raid gendisk */
421 
422 	struct kobject			kobj;
423 	int				hold_active;
424 #define	UNTIL_IOCTL	1
425 #define	UNTIL_STOP	2
426 
427 	/* Superblock information */
428 	int				major_version,
429 					minor_version,
430 					patch_version;
431 	int				persistent;
432 	int				external;	/* metadata is
433 							 * managed externally */
434 	char				metadata_type[17]; /* externally set*/
435 	int				chunk_sectors;
436 	time64_t			ctime, utime;
437 	int				level, layout;
438 	char				clevel[16];
439 	int				raid_disks;
440 	int				max_disks;
441 	sector_t			dev_sectors;	/* used size of
442 							 * component devices */
443 	sector_t			array_sectors; /* exported array size */
444 	int				external_size; /* size managed
445 							* externally */
446 	unsigned int			logical_block_size;
447 	__u64				events;
448 	/* If the last 'event' was simply a clean->dirty transition, and
449 	 * we didn't write it to the spares, then it is safe and simple
450 	 * to just decrement the event count on a dirty->clean transition.
451 	 * So we record that possibility here.
452 	 */
453 	int				can_decrease_events;
454 
455 	char				uuid[16];
456 
457 	/* If the array is being reshaped, we need to record the
458 	 * new shape and an indication of where we are up to.
459 	 * This is written to the superblock.
460 	 * If reshape_position is MaxSector, then no reshape is happening (yet).
461 	 */
462 	sector_t			reshape_position;
463 	int				delta_disks, new_level, new_layout;
464 	int				new_chunk_sectors;
465 	int				reshape_backwards;
466 
467 	struct md_thread __rcu		*thread;	/* management thread */
468 	struct md_thread __rcu		*sync_thread;	/* doing resync or reconstruct */
469 
470 	/*
471 	 * Set when a sync operation is started. It holds this value even
472 	 * when the sync thread is "frozen" (interrupted) or "idle" (stopped
473 	 * or finished). It is overwritten when a new sync operation is begun.
474 	 */
475 	enum sync_action		last_sync_action;
476 	sector_t			curr_resync;	/* last block scheduled */
477 	/* As resync requests can complete out of order, we cannot easily track
478 	 * how much resync has been completed.  So we occasionally pause until
479 	 * everything completes, then set curr_resync_completed to curr_resync.
480 	 * As such it may be well behind the real resync mark, but it is a value
481 	 * we are certain of.
482 	 */
483 	sector_t			curr_resync_completed;
484 	unsigned long			resync_mark;	/* a recent timestamp */
485 	sector_t			resync_mark_cnt;/* blocks written at resync_mark */
486 	sector_t			curr_mark_cnt; /* blocks scheduled now */
487 
488 	sector_t			resync_max_sectors; /* may be set by personality */
489 
490 	atomic64_t			resync_mismatches; /* count of sectors where
491 							    * parity/replica mismatch found
492 							    */
493 
494 	/* allow user-space to request suspension of IO to regions of the array */
495 	sector_t			suspend_lo;
496 	sector_t			suspend_hi;
497 	/* if zero, use the system-wide default */
498 	int				sync_speed_min;
499 	int				sync_speed_max;
500 	int				sync_io_depth;
501 
502 	/* resync even though the same disks are shared among md-devices */
503 	int				parallel_resync;
504 
505 	int				ok_start_degraded;
506 
507 	unsigned long			recovery;
508 
509 	int				in_sync;	/* know to not need resync */
510 	/* 'open_mutex' avoids races between 'md_open' and 'do_md_stop', so
511 	 * that we are never stopping an array while it is open.
512 	 * 'reconfig_mutex' protects all other reconfiguration.
513 	 * These locks are separate due to conflicting interactions
514 	 * with disk->open_mutex.
515 	 * Lock ordering is:
516 	 *  reconfig_mutex -> disk->open_mutex
517 	 *  disk->open_mutex -> open_mutex:  e.g. __blkdev_get -> md_open
518 	 */
519 	struct mutex			open_mutex;
520 	struct mutex			reconfig_mutex;
521 	atomic_t			active;		/* general refcount */
522 	atomic_t			openers;	/* number of active opens */
523 
524 	int				changed;	/* True if we might need to
525 							 * reread partition info */
526 	int				degraded;	/* whether md should consider
527 							 * adding a spare
528 							 */
529 
530 	unsigned long			normal_io_events; /* IO event timestamp */
531 	atomic_t			recovery_active; /* blocks scheduled, but not written */
532 	wait_queue_head_t		recovery_wait;
533 	sector_t			resync_offset;
534 	sector_t			resync_min;	/* user requested sync
535 							 * starts here */
536 	sector_t			resync_max;	/* resync should pause
537 							 * when it gets here */
538 
539 	struct kernfs_node		*sysfs_state;	/* handle for 'array_state'
540 							 * file in sysfs.
541 							 */
542 	struct kernfs_node		*sysfs_action;  /* handle for 'sync_action' */
543 	struct kernfs_node		*sysfs_completed;	/*handle for 'sync_completed' */
544 	struct kernfs_node		*sysfs_degraded;	/*handle for 'degraded' */
545 	struct kernfs_node		*sysfs_level;		/*handle for 'level' */
546 
547 	/* used for delayed sysfs removal */
548 	struct work_struct del_work;
549 	/* used for register new sync thread */
550 	struct work_struct sync_work;
551 
552 	/* "lock" protects:
553 	 *   flush_bio transition from NULL to !NULL
554 	 *   rdev superblocks, events
555 	 *   clearing MD_CHANGE_*
556 	 *   in_sync - and related safemode and MD_CHANGE changes
557 	 *   pers (also protected by reconfig_mutex and pending IO).
558 	 *   clearing ->bitmap
559 	 *   clearing ->bitmap_info.file
560 	 *   changing ->resync_{min,max}
561 	 *   setting MD_RECOVERY_RUNNING (which interacts with resync_{min,max})
562 	 */
563 	spinlock_t			lock;
564 	wait_queue_head_t		sb_wait;	/* for waiting on superblock updates */
565 	atomic_t			pending_writes;	/* number of active superblock writes */
566 
567 	unsigned int			safemode;	/* if set, update "clean" superblock
568 							 * when no writes pending.
569 							 */
570 	unsigned int			safemode_delay;
571 	struct timer_list		safemode_timer;
572 	struct percpu_ref		writes_pending;
573 	int				sync_checkers;	/* # of threads checking writes_pending */
574 
575 	enum md_submodule_id		bitmap_id;
576 	void				*bitmap; /* the bitmap for the device */
577 	struct bitmap_operations	*bitmap_ops;
578 	struct {
579 		struct file		*file; /* the bitmap file */
580 		loff_t			offset; /* offset from superblock of
581 						 * start of bitmap. May be
582 						 * negative, but not '0'
583 						 * For external metadata, offset
584 						 * from start of device.
585 						 */
586 		unsigned long		space; /* space available at this offset */
587 		loff_t			default_offset; /* this is the offset to use when
588 							 * hot-adding a bitmap.  It should
589 							 * eventually be settable by sysfs.
590 							 */
591 		unsigned long		default_space; /* space available at
592 							* default offset */
593 		struct mutex		mutex;
594 		unsigned long		chunksize;
595 		unsigned long		daemon_sleep; /* how many jiffies between updates? */
596 		unsigned long		max_write_behind; /* write-behind mode */
597 		int			external;
598 		int			nodes; /* Maximum number of nodes in the cluster */
599 		char                    cluster_name[64]; /* Name of the cluster */
600 	} bitmap_info;
601 
602 	atomic_t			max_corr_read_errors; /* max read retries */
603 	struct list_head		all_mddevs;
604 
605 	const struct attribute_group	*to_remove;
606 
607 	struct bio_set			bio_set;
608 	struct bio_set			sync_set; /* for sync operations like
609 						   * metadata and bitmap writes
610 						   */
611 	struct bio_set			io_clone_set;
612 
613 	struct work_struct event_work;	/* used by dm to report failure event */
614 	mempool_t *serial_info_pool;
615 	void (*sync_super)(struct mddev *mddev, struct md_rdev *rdev);
616 	struct md_cluster_info		*cluster_info;
617 	struct md_cluster_operations *cluster_ops;
618 	unsigned int			good_device_nr;	/* good device num within cluster raid */
619 	unsigned int			noio_flag; /* for memalloc scope API */
620 
621 	/*
622 	 * Temporarily store rdev that will be finally removed when
623 	 * reconfig_mutex is unlocked, protected by reconfig_mutex.
624 	 */
625 	struct list_head		deleting;
626 
627 	/* The sequence number for sync thread */
628 	atomic_t sync_seq;
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 
650 	/* flags determines sync action, see details in enum sync_action */
651 
652 	/* if just this flag is set, action is resync. */
653 	MD_RECOVERY_SYNC,
654 	/*
655 	 * paired with MD_RECOVERY_SYNC, if MD_RECOVERY_CHECK is not set,
656 	 * action is repair, means user requested resync.
657 	 */
658 	MD_RECOVERY_REQUESTED,
659 	/*
660 	 * paired with MD_RECOVERY_SYNC and MD_RECOVERY_REQUESTED, action is
661 	 * check.
662 	 */
663 	MD_RECOVERY_CHECK,
664 	/* recovery, or need to try it */
665 	MD_RECOVERY_RECOVER,
666 	/* reshape */
667 	MD_RECOVERY_RESHAPE,
668 	/* remote node is running resync thread */
669 	MD_RESYNCING_REMOTE,
670 	/* raid456 lazy initial recover */
671 	MD_RECOVERY_LAZY_RECOVER,
672 };
673 
674 enum md_ro_state {
675 	MD_RDWR,
676 	MD_RDONLY,
677 	MD_AUTO_READ,
678 	MD_MAX_STATE
679 };
680 
681 static inline bool md_is_rdwr(struct mddev *mddev)
682 {
683 	return (mddev->ro == MD_RDWR);
684 }
685 
686 static inline bool reshape_interrupted(struct mddev *mddev)
687 {
688 	/* reshape never start */
689 	if (mddev->reshape_position == MaxSector)
690 		return false;
691 
692 	/* interrupted */
693 	if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
694 		return true;
695 
696 	/* running reshape will be interrupted soon. */
697 	if (test_bit(MD_RECOVERY_WAIT, &mddev->recovery) ||
698 	    test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
699 	    test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
700 		return true;
701 
702 	return false;
703 }
704 
705 static inline int __must_check mddev_lock(struct mddev *mddev)
706 {
707 	int ret;
708 
709 	ret = mutex_lock_interruptible(&mddev->reconfig_mutex);
710 
711 	/* MD_DELETED is set in do_md_stop with reconfig_mutex.
712 	 * So check it here.
713 	 */
714 	if (!ret && test_bit(MD_DELETED, &mddev->flags)) {
715 		ret = -ENODEV;
716 		mutex_unlock(&mddev->reconfig_mutex);
717 	}
718 
719 	return ret;
720 }
721 
722 /* Sometimes we need to take the lock in a situation where
723  * failure due to interrupts is not acceptable.
724  * It doesn't need to check MD_DELETED here, the owner which
725  * holds the lock here can't be stopped. And all paths can't
726  * call this function after do_md_stop.
727  */
728 static inline void mddev_lock_nointr(struct mddev *mddev)
729 {
730 	mutex_lock(&mddev->reconfig_mutex);
731 }
732 
733 static inline int mddev_trylock(struct mddev *mddev)
734 {
735 	int ret;
736 
737 	ret = mutex_trylock(&mddev->reconfig_mutex);
738 	if (ret && test_bit(MD_DELETED, &mddev->flags)) {
739 		ret = 0;
740 		mutex_unlock(&mddev->reconfig_mutex);
741 	}
742 	return ret;
743 }
744 extern void mddev_unlock(struct mddev *mddev);
745 
746 struct md_personality
747 {
748 	struct md_submodule_head head;
749 
750 	bool __must_check (*make_request)(struct mddev *mddev, struct bio *bio);
751 	/*
752 	 * start up works that do NOT require md_thread. tasks that
753 	 * requires md_thread should go into start()
754 	 */
755 	int (*run)(struct mddev *mddev);
756 	/* start up works that require md threads */
757 	int (*start)(struct mddev *mddev);
758 	void (*free)(struct mddev *mddev, void *priv);
759 	void (*status)(struct seq_file *seq, struct mddev *mddev);
760 	/* error_handler must set ->faulty and clear ->in_sync
761 	 * if appropriate, and should abort recovery if needed
762 	 */
763 	void (*error_handler)(struct mddev *mddev, struct md_rdev *rdev);
764 	int (*hot_add_disk) (struct mddev *mddev, struct md_rdev *rdev);
765 	int (*hot_remove_disk) (struct mddev *mddev, struct md_rdev *rdev);
766 	int (*spare_active) (struct mddev *mddev);
767 	sector_t (*sync_request)(struct mddev *mddev, sector_t sector_nr,
768 				 sector_t max_sector, int *skipped);
769 	int (*resize) (struct mddev *mddev, sector_t sectors);
770 	sector_t (*size) (struct mddev *mddev, sector_t sectors, int raid_disks);
771 	int (*check_reshape) (struct mddev *mddev);
772 	int (*start_reshape) (struct mddev *mddev);
773 	void (*finish_reshape) (struct mddev *mddev);
774 	void (*update_reshape_pos) (struct mddev *mddev);
775 	void (*prepare_suspend) (struct mddev *mddev);
776 	/* quiesce suspends or resumes internal processing.
777 	 * 1 - stop new actions and wait for action io to complete
778 	 * 0 - return to normal behaviour
779 	 */
780 	void (*quiesce) (struct mddev *mddev, int quiesce);
781 	/* takeover is used to transition an array from one
782 	 * personality to another.  The new personality must be able
783 	 * to handle the data in the current layout.
784 	 * e.g. 2drive raid1 -> 2drive raid5
785 	 *      ndrive raid5 -> degraded n+1drive raid6 with special layout
786 	 * If the takeover succeeds, a new 'private' structure is returned.
787 	 * This needs to be installed and then ->run used to activate the
788 	 * array.
789 	 */
790 	void *(*takeover) (struct mddev *mddev);
791 	/* Changes the consistency policy of an active array. */
792 	int (*change_consistency_policy)(struct mddev *mddev, const char *buf);
793 	/* convert io ranges from array to bitmap */
794 	void (*bitmap_sector)(struct mddev *mddev, sector_t *offset,
795 			      unsigned long *sectors);
796 };
797 
798 struct md_sysfs_entry {
799 	struct attribute attr;
800 	ssize_t (*show)(struct mddev *, char *);
801 	ssize_t (*store)(struct mddev *, const char *, size_t);
802 };
803 
804 static inline struct kernfs_node *sysfs_get_dirent_safe(struct kernfs_node *sd, char *name)
805 {
806 	if (sd)
807 		return sysfs_get_dirent(sd, name);
808 	return sd;
809 }
810 static inline void sysfs_notify_dirent_safe(struct kernfs_node *sd)
811 {
812 	if (sd)
813 		sysfs_notify_dirent(sd);
814 }
815 
816 static inline char * mdname (struct mddev * mddev)
817 {
818 	return mddev->gendisk ? mddev->gendisk->disk_name : "mdX";
819 }
820 
821 static inline int sysfs_link_rdev(struct mddev *mddev, struct md_rdev *rdev)
822 {
823 	char nm[20];
824 	if (!test_bit(Replacement, &rdev->flags) &&
825 	    !test_bit(Journal, &rdev->flags) &&
826 	    mddev->kobj.sd) {
827 		sprintf(nm, "rd%d", rdev->raid_disk);
828 		return sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
829 	} else
830 		return 0;
831 }
832 
833 static inline void sysfs_unlink_rdev(struct mddev *mddev, struct md_rdev *rdev)
834 {
835 	char nm[20];
836 	if (!test_bit(Replacement, &rdev->flags) &&
837 	    !test_bit(Journal, &rdev->flags) &&
838 	    mddev->kobj.sd) {
839 		sprintf(nm, "rd%d", rdev->raid_disk);
840 		sysfs_remove_link(&mddev->kobj, nm);
841 	}
842 }
843 
844 /*
845  * iterates through some rdev ringlist. It's safe to remove the
846  * current 'rdev'. Dont touch 'tmp' though.
847  */
848 #define rdev_for_each_list(rdev, tmp, head)				\
849 	list_for_each_entry_safe(rdev, tmp, head, same_set)
850 
851 /*
852  * iterates through the 'same array disks' ringlist
853  */
854 #define rdev_for_each(rdev, mddev)				\
855 	list_for_each_entry(rdev, &((mddev)->disks), same_set)
856 
857 #define rdev_for_each_safe(rdev, tmp, mddev)				\
858 	list_for_each_entry_safe(rdev, tmp, &((mddev)->disks), same_set)
859 
860 #define rdev_for_each_rcu(rdev, mddev)				\
861 	list_for_each_entry_rcu(rdev, &((mddev)->disks), same_set)
862 
863 struct md_thread {
864 	void			(*run) (struct md_thread *thread);
865 	struct mddev		*mddev;
866 	wait_queue_head_t	wqueue;
867 	unsigned long		flags;
868 	struct task_struct	*tsk;
869 	unsigned long		timeout;
870 	void			*private;
871 };
872 
873 struct md_io_clone {
874 	struct mddev	*mddev;
875 	struct bio	*orig_bio;
876 	unsigned long	start_time;
877 	sector_t	offset;
878 	unsigned long	sectors;
879 	enum stat_group	rw;
880 	struct bio	bio_clone;
881 };
882 
883 #define THREAD_WAKEUP  0
884 
885 #define md_wakeup_thread(thread) do {   \
886 	rcu_read_lock();                    \
887 	__md_wakeup_thread(thread);         \
888 	rcu_read_unlock();                  \
889 } while (0)
890 
891 static inline void safe_put_page(struct page *p)
892 {
893 	if (p) put_page(p);
894 }
895 
896 int register_md_submodule(struct md_submodule_head *msh);
897 void unregister_md_submodule(struct md_submodule_head *msh);
898 
899 extern struct md_thread *md_register_thread(
900 	void (*run)(struct md_thread *thread),
901 	struct mddev *mddev,
902 	const char *name);
903 extern void md_unregister_thread(struct mddev *mddev, struct md_thread __rcu **threadp);
904 extern void __md_wakeup_thread(struct md_thread __rcu *thread);
905 extern void md_check_recovery(struct mddev *mddev);
906 extern void md_reap_sync_thread(struct mddev *mddev);
907 extern enum sync_action md_sync_action(struct mddev *mddev);
908 extern enum sync_action md_sync_action_by_name(const char *page);
909 extern const char *md_sync_action_name(enum sync_action action);
910 extern void md_write_start(struct mddev *mddev, struct bio *bi);
911 extern void md_write_inc(struct mddev *mddev, struct bio *bi);
912 extern void md_write_end(struct mddev *mddev);
913 extern void md_done_sync(struct mddev *mddev, int blocks);
914 extern void md_sync_error(struct mddev *mddev);
915 extern void md_error(struct mddev *mddev, struct md_rdev *rdev);
916 extern void md_finish_reshape(struct mddev *mddev);
917 void md_submit_discard_bio(struct mddev *mddev, struct md_rdev *rdev,
918 			struct bio *bio, sector_t start, sector_t size);
919 void md_account_bio(struct mddev *mddev, struct bio **bio);
920 void md_free_cloned_bio(struct bio *bio);
921 
922 extern bool __must_check md_flush_request(struct mddev *mddev, struct bio *bio);
923 void md_write_metadata(struct mddev *mddev, struct md_rdev *rdev,
924 		       sector_t sector, int size, struct page *page,
925 		       unsigned int offset);
926 extern int md_super_wait(struct mddev *mddev);
927 extern int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
928 		struct page *page, blk_opf_t opf, bool metadata_op);
929 extern void md_do_sync(struct md_thread *thread);
930 extern void md_new_event(void);
931 extern void md_allow_write(struct mddev *mddev);
932 extern void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev);
933 extern void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors);
934 extern int md_check_no_bitmap(struct mddev *mddev);
935 extern int md_integrity_register(struct mddev *mddev);
936 extern int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale);
937 
938 extern int mddev_init(struct mddev *mddev);
939 extern void mddev_destroy(struct mddev *mddev);
940 void md_init_stacking_limits(struct queue_limits *lim);
941 struct mddev *md_alloc(dev_t dev, char *name);
942 void mddev_put(struct mddev *mddev);
943 extern int md_run(struct mddev *mddev);
944 extern int md_start(struct mddev *mddev);
945 extern void md_stop(struct mddev *mddev);
946 extern void md_stop_writes(struct mddev *mddev);
947 extern int md_rdev_init(struct md_rdev *rdev);
948 extern void md_rdev_clear(struct md_rdev *rdev);
949 
950 extern bool md_handle_request(struct mddev *mddev, struct bio *bio);
951 extern int mddev_suspend(struct mddev *mddev, bool interruptible);
952 extern void mddev_resume(struct mddev *mddev);
953 extern void md_idle_sync_thread(struct mddev *mddev);
954 extern void md_frozen_sync_thread(struct mddev *mddev);
955 extern void md_unfrozen_sync_thread(struct mddev *mddev);
956 
957 extern void md_update_sb(struct mddev *mddev, int force);
958 extern void mddev_create_serial_pool(struct mddev *mddev, struct md_rdev *rdev);
959 extern void mddev_destroy_serial_pool(struct mddev *mddev,
960 				      struct md_rdev *rdev);
961 struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr);
962 struct md_rdev *md_find_rdev_rcu(struct mddev *mddev, dev_t dev);
963 
964 static inline bool is_rdev_broken(struct md_rdev *rdev)
965 {
966 	return !disk_live(rdev->bdev->bd_disk);
967 }
968 
969 static inline void rdev_dec_pending(struct md_rdev *rdev, struct mddev *mddev)
970 {
971 	int faulty = test_bit(Faulty, &rdev->flags);
972 	if (atomic_dec_and_test(&rdev->nr_pending) && faulty) {
973 		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
974 		md_wakeup_thread(mddev->thread);
975 	}
976 }
977 
978 static inline int mddev_is_clustered(struct mddev *mddev)
979 {
980 	return mddev->cluster_info && mddev->bitmap_info.nodes > 1;
981 }
982 
983 /* clear unsupported mddev_flags */
984 static inline void mddev_clear_unsupported_flags(struct mddev *mddev,
985 	unsigned long unsupported_flags)
986 {
987 	mddev->flags &= ~unsupported_flags;
988 }
989 
990 static inline void mddev_check_write_zeroes(struct mddev *mddev, struct bio *bio)
991 {
992 	if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
993 	    !bio->bi_bdev->bd_disk->queue->limits.max_write_zeroes_sectors)
994 		mddev->gendisk->queue->limits.max_write_zeroes_sectors = 0;
995 }
996 
997 static inline int mddev_suspend_and_lock(struct mddev *mddev)
998 {
999 	int ret;
1000 
1001 	ret = mddev_suspend(mddev, true);
1002 	if (ret)
1003 		return ret;
1004 
1005 	ret = mddev_lock(mddev);
1006 	if (ret)
1007 		mddev_resume(mddev);
1008 
1009 	return ret;
1010 }
1011 
1012 static inline void mddev_suspend_and_lock_nointr(struct mddev *mddev)
1013 {
1014 	mddev_suspend(mddev, false);
1015 	mutex_lock(&mddev->reconfig_mutex);
1016 }
1017 
1018 static inline void mddev_unlock_and_resume(struct mddev *mddev)
1019 {
1020 	mddev_unlock(mddev);
1021 	mddev_resume(mddev);
1022 }
1023 
1024 struct mdu_array_info_s;
1025 struct mdu_disk_info_s;
1026 
1027 extern int mdp_major;
1028 void md_autostart_arrays(int part);
1029 int md_set_array_info(struct mddev *mddev, struct mdu_array_info_s *info);
1030 int md_add_new_disk(struct mddev *mddev, struct mdu_disk_info_s *info);
1031 int do_md_run(struct mddev *mddev);
1032 #define MDDEV_STACK_INTEGRITY	(1u << 0)
1033 int mddev_stack_rdev_limits(struct mddev *mddev, struct queue_limits *lim,
1034 		unsigned int flags);
1035 int mddev_stack_new_rdev(struct mddev *mddev, struct md_rdev *rdev);
1036 void mddev_update_io_opt(struct mddev *mddev, unsigned int nr_stripes);
1037 
1038 extern const struct block_device_operations md_fops;
1039 
1040 /*
1041  * MD devices can be used undeneath by DM, in which case ->gendisk is NULL.
1042  */
1043 static inline bool mddev_is_dm(struct mddev *mddev)
1044 {
1045 	return !mddev->gendisk;
1046 }
1047 
1048 static inline bool raid_is_456(struct mddev *mddev)
1049 {
1050 	return mddev->level == ID_RAID4 || mddev->level == ID_RAID5 ||
1051 	       mddev->level == ID_RAID6;
1052 }
1053 
1054 static inline void mddev_trace_remap(struct mddev *mddev, struct bio *bio,
1055 		sector_t sector)
1056 {
1057 	if (!mddev_is_dm(mddev))
1058 		trace_block_bio_remap(bio, disk_devt(mddev->gendisk), sector);
1059 }
1060 
1061 static inline bool rdev_blocked(struct md_rdev *rdev)
1062 {
1063 	/*
1064 	 * Blocked will be set by error handler and cleared by daemon after
1065 	 * updating superblock, meanwhile write IO should be blocked to prevent
1066 	 * reading old data after power failure.
1067 	 */
1068 	if (test_bit(Blocked, &rdev->flags))
1069 		return true;
1070 
1071 	/*
1072 	 * Faulty device should not be accessed anymore, there is no need to
1073 	 * wait for bad block to be acknowledged.
1074 	 */
1075 	if (test_bit(Faulty, &rdev->flags))
1076 		return false;
1077 
1078 	/* rdev is blocked by badblocks. */
1079 	if (test_bit(BlockedBadBlocks, &rdev->flags))
1080 		return true;
1081 
1082 	return false;
1083 }
1084 
1085 #define mddev_add_trace_msg(mddev, fmt, args...)			\
1086 do {									\
1087 	if (!mddev_is_dm(mddev))					\
1088 		blk_add_trace_msg((mddev)->gendisk->queue, fmt, ##args); \
1089 } while (0)
1090 
1091 #endif /* _MD_MD_H */
1092