xref: /linux/drivers/md/md.h (revision 6e11664f148454a127dd89e8698c3e3e80e5f62f)
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, /* TODO */
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,	/* TODO */
42 	ID_LLBITMAP,	/* TODO */
43 };
44 
45 struct md_submodule_head {
46 	enum md_submodule_type	type;
47 	enum md_submodule_id	id;
48 	const char		*name;
49 	struct module		*owner;
50 };
51 
52 /*
53  * These flags should really be called "NO_RETRY" rather than
54  * "FAILFAST" because they don't make any promise about time lapse,
55  * only about the number of retries, which will be zero.
56  * REQ_FAILFAST_DRIVER is not included because
57  * Commit: 4a27446f3e39 ("[SCSI] modify scsi to handle new fail fast flags.")
58  * seems to suggest that the errors it avoids retrying should usually
59  * be retried.
60  */
61 #define	MD_FAILFAST	(REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT)
62 
63 /* Status of sync thread. */
64 enum sync_action {
65 	/*
66 	 * Represent by MD_RECOVERY_SYNC, start when:
67 	 * 1) after assemble, sync data from first rdev to other copies, this
68 	 * must be done first before other sync actions and will only execute
69 	 * once;
70 	 * 2) resize the array(notice that this is not reshape), sync data for
71 	 * the new range;
72 	 */
73 	ACTION_RESYNC,
74 	/*
75 	 * Represent by MD_RECOVERY_RECOVER, start when:
76 	 * 1) for new replacement, sync data based on the replace rdev or
77 	 * available copies from other rdev;
78 	 * 2) for new member disk while the array is degraded, sync data from
79 	 * other rdev;
80 	 * 3) reassemble after power failure or re-add a hot removed rdev, sync
81 	 * data from first rdev to other copies based on bitmap;
82 	 */
83 	ACTION_RECOVER,
84 	/*
85 	 * Represent by MD_RECOVERY_SYNC | MD_RECOVERY_REQUESTED |
86 	 * MD_RECOVERY_CHECK, start when user echo "check" to sysfs api
87 	 * sync_action, used to check if data copies from differenct rdev are
88 	 * the same. The number of mismatch sectors will be exported to user
89 	 * by sysfs api mismatch_cnt;
90 	 */
91 	ACTION_CHECK,
92 	/*
93 	 * Represent by MD_RECOVERY_SYNC | MD_RECOVERY_REQUESTED, start when
94 	 * user echo "repair" to sysfs api sync_action, usually paired with
95 	 * ACTION_CHECK, used to force syncing data once user found that there
96 	 * are inconsistent data,
97 	 */
98 	ACTION_REPAIR,
99 	/*
100 	 * Represent by MD_RECOVERY_RESHAPE, start when new member disk is added
101 	 * to the conf, notice that this is different from spares or
102 	 * replacement;
103 	 */
104 	ACTION_RESHAPE,
105 	/*
106 	 * Represent by MD_RECOVERY_FROZEN, can be set by sysfs api sync_action
107 	 * or internal usage like setting the array read-only, will forbid above
108 	 * actions.
109 	 */
110 	ACTION_FROZEN,
111 	/*
112 	 * All above actions don't match.
113 	 */
114 	ACTION_IDLE,
115 	NR_SYNC_ACTIONS,
116 };
117 
118 /*
119  * The struct embedded in rdev is used to serialize IO.
120  */
121 struct serial_in_rdev {
122 	struct rb_root_cached serial_rb;
123 	spinlock_t serial_lock;
124 	wait_queue_head_t serial_io_wait;
125 };
126 
127 /*
128  * MD's 'extended' device
129  */
130 struct md_rdev {
131 	struct list_head same_set;	/* RAID devices within the same set */
132 
133 	sector_t sectors;		/* Device size (in 512bytes sectors) */
134 	struct mddev *mddev;		/* RAID array if running */
135 	unsigned long last_events;	/* IO event timestamp */
136 
137 	/*
138 	 * If meta_bdev is non-NULL, it means that a separate device is
139 	 * being used to store the metadata (superblock/bitmap) which
140 	 * would otherwise be contained on the same device as the data (bdev).
141 	 */
142 	struct block_device *meta_bdev;
143 	struct block_device *bdev;	/* block device handle */
144 	struct file *bdev_file;		/* Handle from open for bdev */
145 
146 	struct page	*sb_page, *bb_page;
147 	int		sb_loaded;
148 	__u64		sb_events;
149 	sector_t	data_offset;	/* start of data in array */
150 	sector_t	new_data_offset;/* only relevant while reshaping */
151 	sector_t	sb_start;	/* offset of the super block (in 512byte sectors) */
152 	int		sb_size;	/* bytes in the superblock */
153 	int		preferred_minor;	/* autorun support */
154 
155 	struct kobject	kobj;
156 
157 	/* A device can be in one of three states based on two flags:
158 	 * Not working:   faulty==1 in_sync==0
159 	 * Fully working: faulty==0 in_sync==1
160 	 * Working, but not
161 	 * in sync with array
162 	 *                faulty==0 in_sync==0
163 	 *
164 	 * It can never have faulty==1, in_sync==1
165 	 * This reduces the burden of testing multiple flags in many cases
166 	 */
167 
168 	unsigned long	flags;	/* bit set of 'enum flag_bits' bits. */
169 	wait_queue_head_t blocked_wait;
170 
171 	int desc_nr;			/* descriptor index in the superblock */
172 	int raid_disk;			/* role of device in array */
173 	int new_raid_disk;		/* role that the device will have in
174 					 * the array after a level-change completes.
175 					 */
176 	int saved_raid_disk;		/* role that device used to have in the
177 					 * array and could again if we did a partial
178 					 * resync from the bitmap
179 					 */
180 	union {
181 		sector_t recovery_offset;/* If this device has been partially
182 					 * recovered, this is where we were
183 					 * up to.
184 					 */
185 		sector_t journal_tail;	/* If this device is a journal device,
186 					 * this is the journal tail (journal
187 					 * recovery start point)
188 					 */
189 	};
190 
191 	atomic_t	nr_pending;	/* number of pending requests.
192 					 * only maintained for arrays that
193 					 * support hot removal
194 					 */
195 	atomic_t	read_errors;	/* number of consecutive read errors that
196 					 * we have tried to ignore.
197 					 */
198 	time64_t	last_read_error;	/* monotonic time since our
199 						 * last read error
200 						 */
201 	atomic_t	corrected_errors; /* number of corrected read errors,
202 					   * for reporting to userspace and storing
203 					   * in superblock.
204 					   */
205 
206 	struct serial_in_rdev *serial;  /* used for raid1 io serialization */
207 
208 	struct kernfs_node *sysfs_state; /* handle for 'state'
209 					   * sysfs entry */
210 	/* handle for 'unacknowledged_bad_blocks' sysfs dentry */
211 	struct kernfs_node *sysfs_unack_badblocks;
212 	/* handle for 'bad_blocks' sysfs dentry */
213 	struct kernfs_node *sysfs_badblocks;
214 	struct badblocks badblocks;
215 
216 	struct {
217 		short offset;	/* Offset from superblock to start of PPL.
218 				 * Not used by external metadata. */
219 		unsigned int size;	/* Size in sectors of the PPL space */
220 		sector_t sector;	/* First sector of the PPL space */
221 	} ppl;
222 };
223 enum flag_bits {
224 	Faulty,			/* device is known to have a fault */
225 	In_sync,		/* device is in_sync with rest of array */
226 	Bitmap_sync,		/* ..actually, not quite In_sync.  Need a
227 				 * bitmap-based recovery to get fully in sync.
228 				 * The bit is only meaningful before device
229 				 * has been passed to pers->hot_add_disk.
230 				 */
231 	WriteMostly,		/* Avoid reading if at all possible */
232 	AutoDetected,		/* added by auto-detect */
233 	Blocked,		/* An error occurred but has not yet
234 				 * been acknowledged by the metadata
235 				 * handler, so don't allow writes
236 				 * until it is cleared */
237 	WriteErrorSeen,		/* A write error has been seen on this
238 				 * device
239 				 */
240 	FaultRecorded,		/* Intermediate state for clearing
241 				 * Blocked.  The Fault is/will-be
242 				 * recorded in the metadata, but that
243 				 * metadata hasn't been stored safely
244 				 * on disk yet.
245 				 */
246 	BlockedBadBlocks,	/* A writer is blocked because they
247 				 * found an unacknowledged bad-block.
248 				 * This can safely be cleared at any
249 				 * time, and the writer will re-check.
250 				 * It may be set at any time, and at
251 				 * worst the writer will timeout and
252 				 * re-check.  So setting it as
253 				 * accurately as possible is good, but
254 				 * not absolutely critical.
255 				 */
256 	WantReplacement,	/* This device is a candidate to be
257 				 * hot-replaced, either because it has
258 				 * reported some faults, or because
259 				 * of explicit request.
260 				 */
261 	Replacement,		/* This device is a replacement for
262 				 * a want_replacement device with same
263 				 * raid_disk number.
264 				 */
265 	Candidate,		/* For clustered environments only:
266 				 * This device is seen locally but not
267 				 * by the whole cluster
268 				 */
269 	Journal,		/* This device is used as journal for
270 				 * raid-5/6.
271 				 * Usually, this device should be faster
272 				 * than other devices in the array
273 				 */
274 	ClusterRemove,
275 	ExternalBbl,            /* External metadata provides bad
276 				 * block management for a disk
277 				 */
278 	FailFast,		/* Minimal retries should be attempted on
279 				 * this device, so use REQ_FAILFAST_DEV.
280 				 * Also don't try to repair failed reads.
281 				 * It is expects that no bad block log
282 				 * is present.
283 				 */
284 	LastDev,		/* Seems to be the last working dev as
285 				 * it didn't fail, so don't use FailFast
286 				 * any more for metadata
287 				 */
288 	CollisionCheck,		/*
289 				 * check if there is collision between raid1
290 				 * serial bios.
291 				 */
292 	Nonrot,			/* non-rotational device (SSD) */
293 };
294 
is_badblock(struct md_rdev * rdev,sector_t s,sector_t sectors,sector_t * first_bad,sector_t * bad_sectors)295 static inline int is_badblock(struct md_rdev *rdev, sector_t s, sector_t sectors,
296 			      sector_t *first_bad, sector_t *bad_sectors)
297 {
298 	if (unlikely(rdev->badblocks.count)) {
299 		int rv = badblocks_check(&rdev->badblocks, rdev->data_offset + s,
300 					sectors,
301 					first_bad, bad_sectors);
302 		if (rv)
303 			*first_bad -= rdev->data_offset;
304 		return rv;
305 	}
306 	return 0;
307 }
308 
rdev_has_badblock(struct md_rdev * rdev,sector_t s,int sectors)309 static inline int rdev_has_badblock(struct md_rdev *rdev, sector_t s,
310 				    int sectors)
311 {
312 	sector_t first_bad;
313 	sector_t bad_sectors;
314 
315 	return is_badblock(rdev, s, sectors, &first_bad, &bad_sectors);
316 }
317 
318 extern bool rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
319 			       int is_new);
320 extern void rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
321 				 int is_new);
322 struct md_cluster_info;
323 struct md_cluster_operations;
324 
325 /**
326  * enum mddev_flags - md device flags.
327  * @MD_ARRAY_FIRST_USE: First use of array, needs initialization.
328  * @MD_CLOSING: If set, we are closing the array, do not open it then.
329  * @MD_JOURNAL_CLEAN: A raid with journal is already clean.
330  * @MD_HAS_JOURNAL: The raid array has journal feature set.
331  * @MD_CLUSTER_RESYNC_LOCKED: cluster raid only, which means node, already took
332  *			       resync lock, need to release the lock.
333  * @MD_FAILFAST_SUPPORTED: Using MD_FAILFAST on metadata writes is supported as
334  *			    calls to md_error() will never cause the array to
335  *			    become failed.
336  * @MD_HAS_PPL:  The raid array has PPL feature set.
337  * @MD_HAS_MULTIPLE_PPLS: The raid array has multiple PPLs feature set.
338  * @MD_NOT_READY: do_md_run() is active, so 'array_state', ust not report that
339  *		   array is ready yet.
340  * @MD_BROKEN: This is used to stop writes and mark array as failed.
341  * @MD_DELETED: This device is being deleted
342  *
343  * change UNSUPPORTED_MDDEV_FLAGS for each array type if new flag is added
344  */
345 enum mddev_flags {
346 	MD_ARRAY_FIRST_USE,
347 	MD_CLOSING,
348 	MD_JOURNAL_CLEAN,
349 	MD_HAS_JOURNAL,
350 	MD_CLUSTER_RESYNC_LOCKED,
351 	MD_FAILFAST_SUPPORTED,
352 	MD_HAS_PPL,
353 	MD_HAS_MULTIPLE_PPLS,
354 	MD_NOT_READY,
355 	MD_BROKEN,
356 	MD_DELETED,
357 };
358 
359 enum mddev_sb_flags {
360 	MD_SB_CHANGE_DEVS,		/* Some device status has changed */
361 	MD_SB_CHANGE_CLEAN,	/* transition to or from 'clean' */
362 	MD_SB_CHANGE_PENDING,	/* switch from 'clean' to 'active' in progress */
363 	MD_SB_NEED_REWRITE,	/* metadata write needs to be repeated */
364 };
365 
366 #define NR_SERIAL_INFOS		8
367 /* record current range of serialize IOs */
368 struct serial_info {
369 	struct rb_node node;
370 	sector_t start;		/* start sector of rb node */
371 	sector_t last;		/* end sector of rb node */
372 	sector_t _subtree_last; /* highest sector in subtree of rb node */
373 };
374 
375 /*
376  * mddev->curr_resync stores the current sector of the resync but
377  * also has some overloaded values.
378  */
379 enum {
380 	/* No resync in progress */
381 	MD_RESYNC_NONE = 0,
382 	/* Yielded to allow another conflicting resync to commence */
383 	MD_RESYNC_YIELDED = 1,
384 	/* Delayed to check that there is no conflict with another sync */
385 	MD_RESYNC_DELAYED = 2,
386 	/* Any value greater than or equal to this is in an active resync */
387 	MD_RESYNC_ACTIVE = 3,
388 };
389 
390 struct mddev {
391 	void				*private;
392 	struct md_personality		*pers;
393 	dev_t				unit;
394 	int				md_minor;
395 	struct list_head		disks;
396 	unsigned long			flags;
397 	unsigned long			sb_flags;
398 
399 	int				suspended;
400 	struct mutex			suspend_mutex;
401 	struct percpu_ref		active_io;
402 	int				ro;
403 	int				sysfs_active; /* set when sysfs deletes
404 						       * are happening, so run/
405 						       * takeover/stop are not safe
406 						       */
407 	struct gendisk			*gendisk;    /* mdraid gendisk */
408 	struct gendisk			*dm_gendisk; /* dm-raid gendisk */
409 
410 	struct kobject			kobj;
411 	int				hold_active;
412 #define	UNTIL_IOCTL	1
413 #define	UNTIL_STOP	2
414 
415 	/* Superblock information */
416 	int				major_version,
417 					minor_version,
418 					patch_version;
419 	int				persistent;
420 	int				external;	/* metadata is
421 							 * managed externally */
422 	char				metadata_type[17]; /* externally set*/
423 	int				chunk_sectors;
424 	time64_t			ctime, utime;
425 	int				level, layout;
426 	char				clevel[16];
427 	int				raid_disks;
428 	int				max_disks;
429 	sector_t			dev_sectors;	/* used size of
430 							 * component devices */
431 	sector_t			array_sectors; /* exported array size */
432 	int				external_size; /* size managed
433 							* externally */
434 	__u64				events;
435 	/* If the last 'event' was simply a clean->dirty transition, and
436 	 * we didn't write it to the spares, then it is safe and simple
437 	 * to just decrement the event count on a dirty->clean transition.
438 	 * So we record that possibility here.
439 	 */
440 	int				can_decrease_events;
441 
442 	char				uuid[16];
443 
444 	/* If the array is being reshaped, we need to record the
445 	 * new shape and an indication of where we are up to.
446 	 * This is written to the superblock.
447 	 * If reshape_position is MaxSector, then no reshape is happening (yet).
448 	 */
449 	sector_t			reshape_position;
450 	int				delta_disks, new_level, new_layout;
451 	int				new_chunk_sectors;
452 	int				reshape_backwards;
453 
454 	struct md_thread __rcu		*thread;	/* management thread */
455 	struct md_thread __rcu		*sync_thread;	/* doing resync or reconstruct */
456 
457 	/*
458 	 * Set when a sync operation is started. It holds this value even
459 	 * when the sync thread is "frozen" (interrupted) or "idle" (stopped
460 	 * or finished). It is overwritten when a new sync operation is begun.
461 	 */
462 	enum sync_action		last_sync_action;
463 	sector_t			curr_resync;	/* last block scheduled */
464 	/* As resync requests can complete out of order, we cannot easily track
465 	 * how much resync has been completed.  So we occasionally pause until
466 	 * everything completes, then set curr_resync_completed to curr_resync.
467 	 * As such it may be well behind the real resync mark, but it is a value
468 	 * we are certain of.
469 	 */
470 	sector_t			curr_resync_completed;
471 	unsigned long			resync_mark;	/* a recent timestamp */
472 	sector_t			resync_mark_cnt;/* blocks written at resync_mark */
473 	sector_t			curr_mark_cnt; /* blocks scheduled now */
474 
475 	sector_t			resync_max_sectors; /* may be set by personality */
476 
477 	atomic64_t			resync_mismatches; /* count of sectors where
478 							    * parity/replica mismatch found
479 							    */
480 
481 	/* allow user-space to request suspension of IO to regions of the array */
482 	sector_t			suspend_lo;
483 	sector_t			suspend_hi;
484 	/* if zero, use the system-wide default */
485 	int				sync_speed_min;
486 	int				sync_speed_max;
487 	int				sync_io_depth;
488 
489 	/* resync even though the same disks are shared among md-devices */
490 	int				parallel_resync;
491 
492 	int				ok_start_degraded;
493 
494 	unsigned long			recovery;
495 	/* If a RAID personality determines that recovery (of a particular
496 	 * device) will fail due to a read error on the source device, it
497 	 * takes a copy of this number and does not attempt recovery again
498 	 * until this number changes.
499 	 */
500 	int				recovery_disabled;
501 
502 	int				in_sync;	/* know to not need resync */
503 	/* 'open_mutex' avoids races between 'md_open' and 'do_md_stop', so
504 	 * that we are never stopping an array while it is open.
505 	 * 'reconfig_mutex' protects all other reconfiguration.
506 	 * These locks are separate due to conflicting interactions
507 	 * with disk->open_mutex.
508 	 * Lock ordering is:
509 	 *  reconfig_mutex -> disk->open_mutex
510 	 *  disk->open_mutex -> open_mutex:  e.g. __blkdev_get -> md_open
511 	 */
512 	struct mutex			open_mutex;
513 	struct mutex			reconfig_mutex;
514 	atomic_t			active;		/* general refcount */
515 	atomic_t			openers;	/* number of active opens */
516 
517 	int				changed;	/* True if we might need to
518 							 * reread partition info */
519 	int				degraded;	/* whether md should consider
520 							 * adding a spare
521 							 */
522 
523 	unsigned long			normal_io_events; /* IO event timestamp */
524 	atomic_t			recovery_active; /* blocks scheduled, but not written */
525 	wait_queue_head_t		recovery_wait;
526 	sector_t			recovery_cp;
527 	sector_t			resync_min;	/* user requested sync
528 							 * starts here */
529 	sector_t			resync_max;	/* resync should pause
530 							 * when it gets here */
531 
532 	struct kernfs_node		*sysfs_state;	/* handle for 'array_state'
533 							 * file in sysfs.
534 							 */
535 	struct kernfs_node		*sysfs_action;  /* handle for 'sync_action' */
536 	struct kernfs_node		*sysfs_completed;	/*handle for 'sync_completed' */
537 	struct kernfs_node		*sysfs_degraded;	/*handle for 'degraded' */
538 	struct kernfs_node		*sysfs_level;		/*handle for 'level' */
539 
540 	/* used for delayed sysfs removal */
541 	struct work_struct del_work;
542 	/* used for register new sync thread */
543 	struct work_struct sync_work;
544 
545 	/* "lock" protects:
546 	 *   flush_bio transition from NULL to !NULL
547 	 *   rdev superblocks, events
548 	 *   clearing MD_CHANGE_*
549 	 *   in_sync - and related safemode and MD_CHANGE changes
550 	 *   pers (also protected by reconfig_mutex and pending IO).
551 	 *   clearing ->bitmap
552 	 *   clearing ->bitmap_info.file
553 	 *   changing ->resync_{min,max}
554 	 *   setting MD_RECOVERY_RUNNING (which interacts with resync_{min,max})
555 	 */
556 	spinlock_t			lock;
557 	wait_queue_head_t		sb_wait;	/* for waiting on superblock updates */
558 	atomic_t			pending_writes;	/* number of active superblock writes */
559 
560 	unsigned int			safemode;	/* if set, update "clean" superblock
561 							 * when no writes pending.
562 							 */
563 	unsigned int			safemode_delay;
564 	struct timer_list		safemode_timer;
565 	struct percpu_ref		writes_pending;
566 	int				sync_checkers;	/* # of threads checking writes_pending */
567 
568 	void				*bitmap; /* the bitmap for the device */
569 	struct bitmap_operations	*bitmap_ops;
570 	struct {
571 		struct file		*file; /* the bitmap file */
572 		loff_t			offset; /* offset from superblock of
573 						 * start of bitmap. May be
574 						 * negative, but not '0'
575 						 * For external metadata, offset
576 						 * from start of device.
577 						 */
578 		unsigned long		space; /* space available at this offset */
579 		loff_t			default_offset; /* this is the offset to use when
580 							 * hot-adding a bitmap.  It should
581 							 * eventually be settable by sysfs.
582 							 */
583 		unsigned long		default_space; /* space available at
584 							* default offset */
585 		struct mutex		mutex;
586 		unsigned long		chunksize;
587 		unsigned long		daemon_sleep; /* how many jiffies between updates? */
588 		unsigned long		max_write_behind; /* write-behind mode */
589 		int			external;
590 		int			nodes; /* Maximum number of nodes in the cluster */
591 		char                    cluster_name[64]; /* Name of the cluster */
592 	} bitmap_info;
593 
594 	atomic_t			max_corr_read_errors; /* max read retries */
595 	struct list_head		all_mddevs;
596 
597 	const struct attribute_group	*to_remove;
598 
599 	struct bio_set			bio_set;
600 	struct bio_set			sync_set; /* for sync operations like
601 						   * metadata and bitmap writes
602 						   */
603 	struct bio_set			io_clone_set;
604 
605 	struct work_struct event_work;	/* used by dm to report failure event */
606 	mempool_t *serial_info_pool;
607 	void (*sync_super)(struct mddev *mddev, struct md_rdev *rdev);
608 	struct md_cluster_info		*cluster_info;
609 	struct md_cluster_operations *cluster_ops;
610 	unsigned int			good_device_nr;	/* good device num within cluster raid */
611 	unsigned int			noio_flag; /* for memalloc scope API */
612 
613 	/*
614 	 * Temporarily store rdev that will be finally removed when
615 	 * reconfig_mutex is unlocked, protected by reconfig_mutex.
616 	 */
617 	struct list_head		deleting;
618 
619 	/* The sequence number for sync thread */
620 	atomic_t sync_seq;
621 
622 	bool	has_superblocks:1;
623 	bool	fail_last_dev:1;
624 	bool	serialize_policy:1;
625 };
626 
627 enum recovery_flags {
628 	/* flags for sync thread running status */
629 
630 	/*
631 	 * set when one of sync action is set and new sync thread need to be
632 	 * registered, or just add/remove spares from conf.
633 	 */
634 	MD_RECOVERY_NEEDED,
635 	/* sync thread is running, or about to be started */
636 	MD_RECOVERY_RUNNING,
637 	/* sync thread needs to be aborted for some reason */
638 	MD_RECOVERY_INTR,
639 	/* sync thread is done and is waiting to be unregistered */
640 	MD_RECOVERY_DONE,
641 	/* running sync thread must abort immediately, and not restart */
642 	MD_RECOVERY_FROZEN,
643 	/* waiting for pers->start() to finish */
644 	MD_RECOVERY_WAIT,
645 	/* interrupted because io-error */
646 	MD_RECOVERY_ERROR,
647 
648 	/* flags determines sync action, see details in enum sync_action */
649 
650 	/* if just this flag is set, action is resync. */
651 	MD_RECOVERY_SYNC,
652 	/*
653 	 * paired with MD_RECOVERY_SYNC, if MD_RECOVERY_CHECK is not set,
654 	 * action is repair, means user requested resync.
655 	 */
656 	MD_RECOVERY_REQUESTED,
657 	/*
658 	 * paired with MD_RECOVERY_SYNC and MD_RECOVERY_REQUESTED, action is
659 	 * check.
660 	 */
661 	MD_RECOVERY_CHECK,
662 	/* recovery, or need to try it */
663 	MD_RECOVERY_RECOVER,
664 	/* reshape */
665 	MD_RECOVERY_RESHAPE,
666 	/* remote node is running resync thread */
667 	MD_RESYNCING_REMOTE,
668 };
669 
670 enum md_ro_state {
671 	MD_RDWR,
672 	MD_RDONLY,
673 	MD_AUTO_READ,
674 	MD_MAX_STATE
675 };
676 
md_is_rdwr(struct mddev * mddev)677 static inline bool md_is_rdwr(struct mddev *mddev)
678 {
679 	return (mddev->ro == MD_RDWR);
680 }
681 
reshape_interrupted(struct mddev * mddev)682 static inline bool reshape_interrupted(struct mddev *mddev)
683 {
684 	/* reshape never start */
685 	if (mddev->reshape_position == MaxSector)
686 		return false;
687 
688 	/* interrupted */
689 	if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
690 		return true;
691 
692 	/* running reshape will be interrupted soon. */
693 	if (test_bit(MD_RECOVERY_WAIT, &mddev->recovery) ||
694 	    test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
695 	    test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
696 		return true;
697 
698 	return false;
699 }
700 
mddev_lock(struct mddev * mddev)701 static inline int __must_check mddev_lock(struct mddev *mddev)
702 {
703 	int ret;
704 
705 	ret = mutex_lock_interruptible(&mddev->reconfig_mutex);
706 
707 	/* MD_DELETED is set in do_md_stop with reconfig_mutex.
708 	 * So check it here.
709 	 */
710 	if (!ret && test_bit(MD_DELETED, &mddev->flags)) {
711 		ret = -ENODEV;
712 		mutex_unlock(&mddev->reconfig_mutex);
713 	}
714 
715 	return ret;
716 }
717 
718 /* Sometimes we need to take the lock in a situation where
719  * failure due to interrupts is not acceptable.
720  * It doesn't need to check MD_DELETED here, the owner which
721  * holds the lock here can't be stopped. And all paths can't
722  * call this function after do_md_stop.
723  */
mddev_lock_nointr(struct mddev * mddev)724 static inline void mddev_lock_nointr(struct mddev *mddev)
725 {
726 	mutex_lock(&mddev->reconfig_mutex);
727 }
728 
mddev_trylock(struct mddev * mddev)729 static inline int mddev_trylock(struct mddev *mddev)
730 {
731 	int ret;
732 
733 	ret = mutex_trylock(&mddev->reconfig_mutex);
734 	if (!ret && test_bit(MD_DELETED, &mddev->flags)) {
735 		ret = -ENODEV;
736 		mutex_unlock(&mddev->reconfig_mutex);
737 	}
738 	return ret;
739 }
740 extern void mddev_unlock(struct mddev *mddev);
741 
742 struct md_personality
743 {
744 	struct md_submodule_head head;
745 
746 	bool __must_check (*make_request)(struct mddev *mddev, struct bio *bio);
747 	/*
748 	 * start up works that do NOT require md_thread. tasks that
749 	 * requires md_thread should go into start()
750 	 */
751 	int (*run)(struct mddev *mddev);
752 	/* start up works that require md threads */
753 	int (*start)(struct mddev *mddev);
754 	void (*free)(struct mddev *mddev, void *priv);
755 	void (*status)(struct seq_file *seq, struct mddev *mddev);
756 	/* error_handler must set ->faulty and clear ->in_sync
757 	 * if appropriate, and should abort recovery if needed
758 	 */
759 	void (*error_handler)(struct mddev *mddev, struct md_rdev *rdev);
760 	int (*hot_add_disk) (struct mddev *mddev, struct md_rdev *rdev);
761 	int (*hot_remove_disk) (struct mddev *mddev, struct md_rdev *rdev);
762 	int (*spare_active) (struct mddev *mddev);
763 	sector_t (*sync_request)(struct mddev *mddev, sector_t sector_nr,
764 				 sector_t max_sector, int *skipped);
765 	int (*resize) (struct mddev *mddev, sector_t sectors);
766 	sector_t (*size) (struct mddev *mddev, sector_t sectors, int raid_disks);
767 	int (*check_reshape) (struct mddev *mddev);
768 	int (*start_reshape) (struct mddev *mddev);
769 	void (*finish_reshape) (struct mddev *mddev);
770 	void (*update_reshape_pos) (struct mddev *mddev);
771 	void (*prepare_suspend) (struct mddev *mddev);
772 	/* quiesce suspends or resumes internal processing.
773 	 * 1 - stop new actions and wait for action io to complete
774 	 * 0 - return to normal behaviour
775 	 */
776 	void (*quiesce) (struct mddev *mddev, int quiesce);
777 	/* takeover is used to transition an array from one
778 	 * personality to another.  The new personality must be able
779 	 * to handle the data in the current layout.
780 	 * e.g. 2drive raid1 -> 2drive raid5
781 	 *      ndrive raid5 -> degraded n+1drive raid6 with special layout
782 	 * If the takeover succeeds, a new 'private' structure is returned.
783 	 * This needs to be installed and then ->run used to activate the
784 	 * array.
785 	 */
786 	void *(*takeover) (struct mddev *mddev);
787 	/* Changes the consistency policy of an active array. */
788 	int (*change_consistency_policy)(struct mddev *mddev, const char *buf);
789 	/* convert io ranges from array to bitmap */
790 	void (*bitmap_sector)(struct mddev *mddev, sector_t *offset,
791 			      unsigned long *sectors);
792 };
793 
794 struct md_sysfs_entry {
795 	struct attribute attr;
796 	ssize_t (*show)(struct mddev *, char *);
797 	ssize_t (*store)(struct mddev *, const char *, size_t);
798 };
799 extern const struct attribute_group md_bitmap_group;
800 
sysfs_get_dirent_safe(struct kernfs_node * sd,char * name)801 static inline struct kernfs_node *sysfs_get_dirent_safe(struct kernfs_node *sd, char *name)
802 {
803 	if (sd)
804 		return sysfs_get_dirent(sd, name);
805 	return sd;
806 }
sysfs_notify_dirent_safe(struct kernfs_node * sd)807 static inline void sysfs_notify_dirent_safe(struct kernfs_node *sd)
808 {
809 	if (sd)
810 		sysfs_notify_dirent(sd);
811 }
812 
mdname(struct mddev * mddev)813 static inline char * mdname (struct mddev * mddev)
814 {
815 	return mddev->gendisk ? mddev->gendisk->disk_name : "mdX";
816 }
817 
sysfs_link_rdev(struct mddev * mddev,struct md_rdev * rdev)818 static inline int sysfs_link_rdev(struct mddev *mddev, struct md_rdev *rdev)
819 {
820 	char nm[20];
821 	if (!test_bit(Replacement, &rdev->flags) &&
822 	    !test_bit(Journal, &rdev->flags) &&
823 	    mddev->kobj.sd) {
824 		sprintf(nm, "rd%d", rdev->raid_disk);
825 		return sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
826 	} else
827 		return 0;
828 }
829 
sysfs_unlink_rdev(struct mddev * mddev,struct md_rdev * rdev)830 static inline void sysfs_unlink_rdev(struct mddev *mddev, struct md_rdev *rdev)
831 {
832 	char nm[20];
833 	if (!test_bit(Replacement, &rdev->flags) &&
834 	    !test_bit(Journal, &rdev->flags) &&
835 	    mddev->kobj.sd) {
836 		sprintf(nm, "rd%d", rdev->raid_disk);
837 		sysfs_remove_link(&mddev->kobj, nm);
838 	}
839 }
840 
841 /*
842  * iterates through some rdev ringlist. It's safe to remove the
843  * current 'rdev'. Dont touch 'tmp' though.
844  */
845 #define rdev_for_each_list(rdev, tmp, head)				\
846 	list_for_each_entry_safe(rdev, tmp, head, same_set)
847 
848 /*
849  * iterates through the 'same array disks' ringlist
850  */
851 #define rdev_for_each(rdev, mddev)				\
852 	list_for_each_entry(rdev, &((mddev)->disks), same_set)
853 
854 #define rdev_for_each_safe(rdev, tmp, mddev)				\
855 	list_for_each_entry_safe(rdev, tmp, &((mddev)->disks), same_set)
856 
857 #define rdev_for_each_rcu(rdev, mddev)				\
858 	list_for_each_entry_rcu(rdev, &((mddev)->disks), same_set)
859 
860 struct md_thread {
861 	void			(*run) (struct md_thread *thread);
862 	struct mddev		*mddev;
863 	wait_queue_head_t	wqueue;
864 	unsigned long		flags;
865 	struct task_struct	*tsk;
866 	unsigned long		timeout;
867 	void			*private;
868 };
869 
870 struct md_io_clone {
871 	struct mddev	*mddev;
872 	struct bio	*orig_bio;
873 	unsigned long	start_time;
874 	sector_t	offset;
875 	unsigned long	sectors;
876 	struct bio	bio_clone;
877 };
878 
879 #define THREAD_WAKEUP  0
880 
safe_put_page(struct page * p)881 static inline void safe_put_page(struct page *p)
882 {
883 	if (p) put_page(p);
884 }
885 
886 int register_md_submodule(struct md_submodule_head *msh);
887 void unregister_md_submodule(struct md_submodule_head *msh);
888 
889 extern struct md_thread *md_register_thread(
890 	void (*run)(struct md_thread *thread),
891 	struct mddev *mddev,
892 	const char *name);
893 extern void md_unregister_thread(struct mddev *mddev, struct md_thread __rcu **threadp);
894 extern void md_wakeup_thread(struct md_thread __rcu *thread);
895 extern void md_check_recovery(struct mddev *mddev);
896 extern void md_reap_sync_thread(struct mddev *mddev);
897 extern enum sync_action md_sync_action(struct mddev *mddev);
898 extern enum sync_action md_sync_action_by_name(const char *page);
899 extern const char *md_sync_action_name(enum sync_action action);
900 extern void md_write_start(struct mddev *mddev, struct bio *bi);
901 extern void md_write_inc(struct mddev *mddev, struct bio *bi);
902 extern void md_write_end(struct mddev *mddev);
903 extern void md_done_sync(struct mddev *mddev, int blocks, int ok);
904 extern void md_error(struct mddev *mddev, struct md_rdev *rdev);
905 extern void md_finish_reshape(struct mddev *mddev);
906 void md_submit_discard_bio(struct mddev *mddev, struct md_rdev *rdev,
907 			struct bio *bio, sector_t start, sector_t size);
908 void md_account_bio(struct mddev *mddev, struct bio **bio);
909 void md_free_cloned_bio(struct bio *bio);
910 
911 extern bool __must_check md_flush_request(struct mddev *mddev, struct bio *bio);
912 extern void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
913 			   sector_t sector, int size, struct page *page);
914 extern int md_super_wait(struct mddev *mddev);
915 extern int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
916 		struct page *page, blk_opf_t opf, bool metadata_op);
917 extern void md_do_sync(struct md_thread *thread);
918 extern void md_new_event(void);
919 extern void md_allow_write(struct mddev *mddev);
920 extern void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev);
921 extern void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors);
922 extern int md_check_no_bitmap(struct mddev *mddev);
923 extern int md_integrity_register(struct mddev *mddev);
924 extern int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale);
925 
926 extern int mddev_init(struct mddev *mddev);
927 extern void mddev_destroy(struct mddev *mddev);
928 void md_init_stacking_limits(struct queue_limits *lim);
929 struct mddev *md_alloc(dev_t dev, char *name);
930 void mddev_put(struct mddev *mddev);
931 extern int md_run(struct mddev *mddev);
932 extern int md_start(struct mddev *mddev);
933 extern void md_stop(struct mddev *mddev);
934 extern void md_stop_writes(struct mddev *mddev);
935 extern int md_rdev_init(struct md_rdev *rdev);
936 extern void md_rdev_clear(struct md_rdev *rdev);
937 
938 extern bool md_handle_request(struct mddev *mddev, struct bio *bio);
939 extern int mddev_suspend(struct mddev *mddev, bool interruptible);
940 extern void mddev_resume(struct mddev *mddev);
941 extern void md_idle_sync_thread(struct mddev *mddev);
942 extern void md_frozen_sync_thread(struct mddev *mddev);
943 extern void md_unfrozen_sync_thread(struct mddev *mddev);
944 
945 extern void md_update_sb(struct mddev *mddev, int force);
946 extern void mddev_create_serial_pool(struct mddev *mddev, struct md_rdev *rdev);
947 extern void mddev_destroy_serial_pool(struct mddev *mddev,
948 				      struct md_rdev *rdev);
949 struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr);
950 struct md_rdev *md_find_rdev_rcu(struct mddev *mddev, dev_t dev);
951 
is_rdev_broken(struct md_rdev * rdev)952 static inline bool is_rdev_broken(struct md_rdev *rdev)
953 {
954 	return !disk_live(rdev->bdev->bd_disk);
955 }
956 
rdev_dec_pending(struct md_rdev * rdev,struct mddev * mddev)957 static inline void rdev_dec_pending(struct md_rdev *rdev, struct mddev *mddev)
958 {
959 	int faulty = test_bit(Faulty, &rdev->flags);
960 	if (atomic_dec_and_test(&rdev->nr_pending) && faulty) {
961 		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
962 		md_wakeup_thread(mddev->thread);
963 	}
964 }
965 
mddev_is_clustered(struct mddev * mddev)966 static inline int mddev_is_clustered(struct mddev *mddev)
967 {
968 	return mddev->cluster_info && mddev->bitmap_info.nodes > 1;
969 }
970 
971 /* clear unsupported mddev_flags */
mddev_clear_unsupported_flags(struct mddev * mddev,unsigned long unsupported_flags)972 static inline void mddev_clear_unsupported_flags(struct mddev *mddev,
973 	unsigned long unsupported_flags)
974 {
975 	mddev->flags &= ~unsupported_flags;
976 }
977 
mddev_check_write_zeroes(struct mddev * mddev,struct bio * bio)978 static inline void mddev_check_write_zeroes(struct mddev *mddev, struct bio *bio)
979 {
980 	if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
981 	    !bio->bi_bdev->bd_disk->queue->limits.max_write_zeroes_sectors)
982 		mddev->gendisk->queue->limits.max_write_zeroes_sectors = 0;
983 }
984 
mddev_suspend_and_lock(struct mddev * mddev)985 static inline int mddev_suspend_and_lock(struct mddev *mddev)
986 {
987 	int ret;
988 
989 	ret = mddev_suspend(mddev, true);
990 	if (ret)
991 		return ret;
992 
993 	ret = mddev_lock(mddev);
994 	if (ret)
995 		mddev_resume(mddev);
996 
997 	return ret;
998 }
999 
mddev_suspend_and_lock_nointr(struct mddev * mddev)1000 static inline void mddev_suspend_and_lock_nointr(struct mddev *mddev)
1001 {
1002 	mddev_suspend(mddev, false);
1003 	mutex_lock(&mddev->reconfig_mutex);
1004 }
1005 
mddev_unlock_and_resume(struct mddev * mddev)1006 static inline void mddev_unlock_and_resume(struct mddev *mddev)
1007 {
1008 	mddev_unlock(mddev);
1009 	mddev_resume(mddev);
1010 }
1011 
1012 struct mdu_array_info_s;
1013 struct mdu_disk_info_s;
1014 
1015 extern int mdp_major;
1016 extern struct workqueue_struct *md_bitmap_wq;
1017 void md_autostart_arrays(int part);
1018 int md_set_array_info(struct mddev *mddev, struct mdu_array_info_s *info);
1019 int md_add_new_disk(struct mddev *mddev, struct mdu_disk_info_s *info);
1020 int do_md_run(struct mddev *mddev);
1021 #define MDDEV_STACK_INTEGRITY	(1u << 0)
1022 int mddev_stack_rdev_limits(struct mddev *mddev, struct queue_limits *lim,
1023 		unsigned int flags);
1024 int mddev_stack_new_rdev(struct mddev *mddev, struct md_rdev *rdev);
1025 void mddev_update_io_opt(struct mddev *mddev, unsigned int nr_stripes);
1026 
1027 extern const struct block_device_operations md_fops;
1028 
1029 /*
1030  * MD devices can be used undeneath by DM, in which case ->gendisk is NULL.
1031  */
mddev_is_dm(struct mddev * mddev)1032 static inline bool mddev_is_dm(struct mddev *mddev)
1033 {
1034 	return !mddev->gendisk;
1035 }
1036 
mddev_trace_remap(struct mddev * mddev,struct bio * bio,sector_t sector)1037 static inline void mddev_trace_remap(struct mddev *mddev, struct bio *bio,
1038 		sector_t sector)
1039 {
1040 	if (!mddev_is_dm(mddev))
1041 		trace_block_bio_remap(bio, disk_devt(mddev->gendisk), sector);
1042 }
1043 
rdev_blocked(struct md_rdev * rdev)1044 static inline bool rdev_blocked(struct md_rdev *rdev)
1045 {
1046 	/*
1047 	 * Blocked will be set by error handler and cleared by daemon after
1048 	 * updating superblock, meanwhile write IO should be blocked to prevent
1049 	 * reading old data after power failure.
1050 	 */
1051 	if (test_bit(Blocked, &rdev->flags))
1052 		return true;
1053 
1054 	/*
1055 	 * Faulty device should not be accessed anymore, there is no need to
1056 	 * wait for bad block to be acknowledged.
1057 	 */
1058 	if (test_bit(Faulty, &rdev->flags))
1059 		return false;
1060 
1061 	/* rdev is blocked by badblocks. */
1062 	if (test_bit(BlockedBadBlocks, &rdev->flags))
1063 		return true;
1064 
1065 	return false;
1066 }
1067 
1068 #define mddev_add_trace_msg(mddev, fmt, args...)			\
1069 do {									\
1070 	if (!mddev_is_dm(mddev))					\
1071 		blk_add_trace_msg((mddev)->gendisk->queue, fmt, ##args); \
1072 } while (0)
1073 
1074 #endif /* _MD_MD_H */
1075