xref: /linux/drivers/md/md.h (revision 9b960d8cd6f712cb2c03e2bdd4d5ca058238037f)
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 	int 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;
408 
409 	struct kobject			kobj;
410 	int				hold_active;
411 #define	UNTIL_IOCTL	1
412 #define	UNTIL_STOP	2
413 
414 	/* Superblock information */
415 	int				major_version,
416 					minor_version,
417 					patch_version;
418 	int				persistent;
419 	int				external;	/* metadata is
420 							 * managed externally */
421 	char				metadata_type[17]; /* externally set*/
422 	int				chunk_sectors;
423 	time64_t			ctime, utime;
424 	int				level, layout;
425 	char				clevel[16];
426 	int				raid_disks;
427 	int				max_disks;
428 	sector_t			dev_sectors;	/* used size of
429 							 * component devices */
430 	sector_t			array_sectors; /* exported array size */
431 	int				external_size; /* size managed
432 							* externally */
433 	__u64				events;
434 	/* If the last 'event' was simply a clean->dirty transition, and
435 	 * we didn't write it to the spares, then it is safe and simple
436 	 * to just decrement the event count on a dirty->clean transition.
437 	 * So we record that possibility here.
438 	 */
439 	int				can_decrease_events;
440 
441 	char				uuid[16];
442 
443 	/* If the array is being reshaped, we need to record the
444 	 * new shape and an indication of where we are up to.
445 	 * This is written to the superblock.
446 	 * If reshape_position is MaxSector, then no reshape is happening (yet).
447 	 */
448 	sector_t			reshape_position;
449 	int				delta_disks, new_level, new_layout;
450 	int				new_chunk_sectors;
451 	int				reshape_backwards;
452 
453 	struct md_thread __rcu		*thread;	/* management thread */
454 	struct md_thread __rcu		*sync_thread;	/* doing resync or reconstruct */
455 
456 	/*
457 	 * Set when a sync operation is started. It holds this value even
458 	 * when the sync thread is "frozen" (interrupted) or "idle" (stopped
459 	 * or finished). It is overwritten when a new sync operation is begun.
460 	 */
461 	enum sync_action		last_sync_action;
462 	sector_t			curr_resync;	/* last block scheduled */
463 	/* As resync requests can complete out of order, we cannot easily track
464 	 * how much resync has been completed.  So we occasionally pause until
465 	 * everything completes, then set curr_resync_completed to curr_resync.
466 	 * As such it may be well behind the real resync mark, but it is a value
467 	 * we are certain of.
468 	 */
469 	sector_t			curr_resync_completed;
470 	unsigned long			resync_mark;	/* a recent timestamp */
471 	sector_t			resync_mark_cnt;/* blocks written at resync_mark */
472 	sector_t			curr_mark_cnt; /* blocks scheduled now */
473 
474 	sector_t			resync_max_sectors; /* may be set by personality */
475 
476 	atomic64_t			resync_mismatches; /* count of sectors where
477 							    * parity/replica mismatch found
478 							    */
479 
480 	/* allow user-space to request suspension of IO to regions of the array */
481 	sector_t			suspend_lo;
482 	sector_t			suspend_hi;
483 	/* if zero, use the system-wide default */
484 	int				sync_speed_min;
485 	int				sync_speed_max;
486 
487 	/* resync even though the same disks are shared among md-devices */
488 	int				parallel_resync;
489 
490 	int				ok_start_degraded;
491 
492 	unsigned long			recovery;
493 	/* If a RAID personality determines that recovery (of a particular
494 	 * device) will fail due to a read error on the source device, it
495 	 * takes a copy of this number and does not attempt recovery again
496 	 * until this number changes.
497 	 */
498 	int				recovery_disabled;
499 
500 	int				in_sync;	/* know to not need resync */
501 	/* 'open_mutex' avoids races between 'md_open' and 'do_md_stop', so
502 	 * that we are never stopping an array while it is open.
503 	 * 'reconfig_mutex' protects all other reconfiguration.
504 	 * These locks are separate due to conflicting interactions
505 	 * with disk->open_mutex.
506 	 * Lock ordering is:
507 	 *  reconfig_mutex -> disk->open_mutex
508 	 *  disk->open_mutex -> open_mutex:  e.g. __blkdev_get -> md_open
509 	 */
510 	struct mutex			open_mutex;
511 	struct mutex			reconfig_mutex;
512 	atomic_t			active;		/* general refcount */
513 	atomic_t			openers;	/* number of active opens */
514 
515 	int				changed;	/* True if we might need to
516 							 * reread partition info */
517 	int				degraded;	/* whether md should consider
518 							 * adding a spare
519 							 */
520 
521 	atomic_t			recovery_active; /* blocks scheduled, but not written */
522 	wait_queue_head_t		recovery_wait;
523 	sector_t			recovery_cp;
524 	sector_t			resync_min;	/* user requested sync
525 							 * starts here */
526 	sector_t			resync_max;	/* resync should pause
527 							 * when it gets here */
528 
529 	struct kernfs_node		*sysfs_state;	/* handle for 'array_state'
530 							 * file in sysfs.
531 							 */
532 	struct kernfs_node		*sysfs_action;  /* handle for 'sync_action' */
533 	struct kernfs_node		*sysfs_completed;	/*handle for 'sync_completed' */
534 	struct kernfs_node		*sysfs_degraded;	/*handle for 'degraded' */
535 	struct kernfs_node		*sysfs_level;		/*handle for 'level' */
536 
537 	/* used for delayed sysfs removal */
538 	struct work_struct del_work;
539 	/* used for register new sync thread */
540 	struct work_struct sync_work;
541 
542 	/* "lock" protects:
543 	 *   flush_bio transition from NULL to !NULL
544 	 *   rdev superblocks, events
545 	 *   clearing MD_CHANGE_*
546 	 *   in_sync - and related safemode and MD_CHANGE changes
547 	 *   pers (also protected by reconfig_mutex and pending IO).
548 	 *   clearing ->bitmap
549 	 *   clearing ->bitmap_info.file
550 	 *   changing ->resync_{min,max}
551 	 *   setting MD_RECOVERY_RUNNING (which interacts with resync_{min,max})
552 	 */
553 	spinlock_t			lock;
554 	wait_queue_head_t		sb_wait;	/* for waiting on superblock updates */
555 	atomic_t			pending_writes;	/* number of active superblock writes */
556 
557 	unsigned int			safemode;	/* if set, update "clean" superblock
558 							 * when no writes pending.
559 							 */
560 	unsigned int			safemode_delay;
561 	struct timer_list		safemode_timer;
562 	struct percpu_ref		writes_pending;
563 	int				sync_checkers;	/* # of threads checking writes_pending */
564 
565 	void				*bitmap; /* the bitmap for the device */
566 	struct bitmap_operations	*bitmap_ops;
567 	struct {
568 		struct file		*file; /* the bitmap file */
569 		loff_t			offset; /* offset from superblock of
570 						 * start of bitmap. May be
571 						 * negative, but not '0'
572 						 * For external metadata, offset
573 						 * from start of device.
574 						 */
575 		unsigned long		space; /* space available at this offset */
576 		loff_t			default_offset; /* this is the offset to use when
577 							 * hot-adding a bitmap.  It should
578 							 * eventually be settable by sysfs.
579 							 */
580 		unsigned long		default_space; /* space available at
581 							* default offset */
582 		struct mutex		mutex;
583 		unsigned long		chunksize;
584 		unsigned long		daemon_sleep; /* how many jiffies between updates? */
585 		unsigned long		max_write_behind; /* write-behind mode */
586 		int			external;
587 		int			nodes; /* Maximum number of nodes in the cluster */
588 		char                    cluster_name[64]; /* Name of the cluster */
589 	} bitmap_info;
590 
591 	atomic_t			max_corr_read_errors; /* max read retries */
592 	struct list_head		all_mddevs;
593 
594 	const struct attribute_group	*to_remove;
595 
596 	struct bio_set			bio_set;
597 	struct bio_set			sync_set; /* for sync operations like
598 						   * metadata and bitmap writes
599 						   */
600 	struct bio_set			io_clone_set;
601 
602 	struct work_struct event_work;	/* used by dm to report failure event */
603 	mempool_t *serial_info_pool;
604 	void (*sync_super)(struct mddev *mddev, struct md_rdev *rdev);
605 	struct md_cluster_info		*cluster_info;
606 	struct md_cluster_operations *cluster_ops;
607 	unsigned int			good_device_nr;	/* good device num within cluster raid */
608 	unsigned int			noio_flag; /* for memalloc scope API */
609 
610 	/*
611 	 * Temporarily store rdev that will be finally removed when
612 	 * reconfig_mutex is unlocked, protected by reconfig_mutex.
613 	 */
614 	struct list_head		deleting;
615 
616 	/* The sequence number for sync thread */
617 	atomic_t sync_seq;
618 
619 	bool	has_superblocks:1;
620 	bool	fail_last_dev:1;
621 	bool	serialize_policy:1;
622 };
623 
624 enum recovery_flags {
625 	/* flags for sync thread running status */
626 
627 	/*
628 	 * set when one of sync action is set and new sync thread need to be
629 	 * registered, or just add/remove spares from conf.
630 	 */
631 	MD_RECOVERY_NEEDED,
632 	/* sync thread is running, or about to be started */
633 	MD_RECOVERY_RUNNING,
634 	/* sync thread needs to be aborted for some reason */
635 	MD_RECOVERY_INTR,
636 	/* sync thread is done and is waiting to be unregistered */
637 	MD_RECOVERY_DONE,
638 	/* running sync thread must abort immediately, and not restart */
639 	MD_RECOVERY_FROZEN,
640 	/* waiting for pers->start() to finish */
641 	MD_RECOVERY_WAIT,
642 	/* interrupted because io-error */
643 	MD_RECOVERY_ERROR,
644 
645 	/* flags determines sync action, see details in enum sync_action */
646 
647 	/* if just this flag is set, action is resync. */
648 	MD_RECOVERY_SYNC,
649 	/*
650 	 * paired with MD_RECOVERY_SYNC, if MD_RECOVERY_CHECK is not set,
651 	 * action is repair, means user requested resync.
652 	 */
653 	MD_RECOVERY_REQUESTED,
654 	/*
655 	 * paired with MD_RECOVERY_SYNC and MD_RECOVERY_REQUESTED, action is
656 	 * check.
657 	 */
658 	MD_RECOVERY_CHECK,
659 	/* recovery, or need to try it */
660 	MD_RECOVERY_RECOVER,
661 	/* reshape */
662 	MD_RECOVERY_RESHAPE,
663 	/* remote node is running resync thread */
664 	MD_RESYNCING_REMOTE,
665 };
666 
667 enum md_ro_state {
668 	MD_RDWR,
669 	MD_RDONLY,
670 	MD_AUTO_READ,
671 	MD_MAX_STATE
672 };
673 
md_is_rdwr(struct mddev * mddev)674 static inline bool md_is_rdwr(struct mddev *mddev)
675 {
676 	return (mddev->ro == MD_RDWR);
677 }
678 
reshape_interrupted(struct mddev * mddev)679 static inline bool reshape_interrupted(struct mddev *mddev)
680 {
681 	/* reshape never start */
682 	if (mddev->reshape_position == MaxSector)
683 		return false;
684 
685 	/* interrupted */
686 	if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
687 		return true;
688 
689 	/* running reshape will be interrupted soon. */
690 	if (test_bit(MD_RECOVERY_WAIT, &mddev->recovery) ||
691 	    test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
692 	    test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
693 		return true;
694 
695 	return false;
696 }
697 
mddev_lock(struct mddev * mddev)698 static inline int __must_check mddev_lock(struct mddev *mddev)
699 {
700 	return mutex_lock_interruptible(&mddev->reconfig_mutex);
701 }
702 
703 /* Sometimes we need to take the lock in a situation where
704  * failure due to interrupts is not acceptable.
705  */
mddev_lock_nointr(struct mddev * mddev)706 static inline void mddev_lock_nointr(struct mddev *mddev)
707 {
708 	mutex_lock(&mddev->reconfig_mutex);
709 }
710 
mddev_trylock(struct mddev * mddev)711 static inline int mddev_trylock(struct mddev *mddev)
712 {
713 	return mutex_trylock(&mddev->reconfig_mutex);
714 }
715 extern void mddev_unlock(struct mddev *mddev);
716 
md_sync_acct(struct block_device * bdev,unsigned long nr_sectors)717 static inline void md_sync_acct(struct block_device *bdev, unsigned long nr_sectors)
718 {
719 	if (blk_queue_io_stat(bdev->bd_disk->queue))
720 		atomic_add(nr_sectors, &bdev->bd_disk->sync_io);
721 }
722 
md_sync_acct_bio(struct bio * bio,unsigned long nr_sectors)723 static inline void md_sync_acct_bio(struct bio *bio, unsigned long nr_sectors)
724 {
725 	md_sync_acct(bio->bi_bdev, nr_sectors);
726 }
727 
728 struct md_personality
729 {
730 	struct md_submodule_head head;
731 
732 	bool __must_check (*make_request)(struct mddev *mddev, struct bio *bio);
733 	/*
734 	 * start up works that do NOT require md_thread. tasks that
735 	 * requires md_thread should go into start()
736 	 */
737 	int (*run)(struct mddev *mddev);
738 	/* start up works that require md threads */
739 	int (*start)(struct mddev *mddev);
740 	void (*free)(struct mddev *mddev, void *priv);
741 	void (*status)(struct seq_file *seq, struct mddev *mddev);
742 	/* error_handler must set ->faulty and clear ->in_sync
743 	 * if appropriate, and should abort recovery if needed
744 	 */
745 	void (*error_handler)(struct mddev *mddev, struct md_rdev *rdev);
746 	int (*hot_add_disk) (struct mddev *mddev, struct md_rdev *rdev);
747 	int (*hot_remove_disk) (struct mddev *mddev, struct md_rdev *rdev);
748 	int (*spare_active) (struct mddev *mddev);
749 	sector_t (*sync_request)(struct mddev *mddev, sector_t sector_nr,
750 				 sector_t max_sector, int *skipped);
751 	int (*resize) (struct mddev *mddev, sector_t sectors);
752 	sector_t (*size) (struct mddev *mddev, sector_t sectors, int raid_disks);
753 	int (*check_reshape) (struct mddev *mddev);
754 	int (*start_reshape) (struct mddev *mddev);
755 	void (*finish_reshape) (struct mddev *mddev);
756 	void (*update_reshape_pos) (struct mddev *mddev);
757 	void (*prepare_suspend) (struct mddev *mddev);
758 	/* quiesce suspends or resumes internal processing.
759 	 * 1 - stop new actions and wait for action io to complete
760 	 * 0 - return to normal behaviour
761 	 */
762 	void (*quiesce) (struct mddev *mddev, int quiesce);
763 	/* takeover is used to transition an array from one
764 	 * personality to another.  The new personality must be able
765 	 * to handle the data in the current layout.
766 	 * e.g. 2drive raid1 -> 2drive raid5
767 	 *      ndrive raid5 -> degraded n+1drive raid6 with special layout
768 	 * If the takeover succeeds, a new 'private' structure is returned.
769 	 * This needs to be installed and then ->run used to activate the
770 	 * array.
771 	 */
772 	void *(*takeover) (struct mddev *mddev);
773 	/* Changes the consistency policy of an active array. */
774 	int (*change_consistency_policy)(struct mddev *mddev, const char *buf);
775 	/* convert io ranges from array to bitmap */
776 	void (*bitmap_sector)(struct mddev *mddev, sector_t *offset,
777 			      unsigned long *sectors);
778 };
779 
780 struct md_sysfs_entry {
781 	struct attribute attr;
782 	ssize_t (*show)(struct mddev *, char *);
783 	ssize_t (*store)(struct mddev *, const char *, size_t);
784 };
785 extern const struct attribute_group md_bitmap_group;
786 
sysfs_get_dirent_safe(struct kernfs_node * sd,char * name)787 static inline struct kernfs_node *sysfs_get_dirent_safe(struct kernfs_node *sd, char *name)
788 {
789 	if (sd)
790 		return sysfs_get_dirent(sd, name);
791 	return sd;
792 }
sysfs_notify_dirent_safe(struct kernfs_node * sd)793 static inline void sysfs_notify_dirent_safe(struct kernfs_node *sd)
794 {
795 	if (sd)
796 		sysfs_notify_dirent(sd);
797 }
798 
mdname(struct mddev * mddev)799 static inline char * mdname (struct mddev * mddev)
800 {
801 	return mddev->gendisk ? mddev->gendisk->disk_name : "mdX";
802 }
803 
sysfs_link_rdev(struct mddev * mddev,struct md_rdev * rdev)804 static inline int sysfs_link_rdev(struct mddev *mddev, struct md_rdev *rdev)
805 {
806 	char nm[20];
807 	if (!test_bit(Replacement, &rdev->flags) &&
808 	    !test_bit(Journal, &rdev->flags) &&
809 	    mddev->kobj.sd) {
810 		sprintf(nm, "rd%d", rdev->raid_disk);
811 		return sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
812 	} else
813 		return 0;
814 }
815 
sysfs_unlink_rdev(struct mddev * mddev,struct md_rdev * rdev)816 static inline void sysfs_unlink_rdev(struct mddev *mddev, struct md_rdev *rdev)
817 {
818 	char nm[20];
819 	if (!test_bit(Replacement, &rdev->flags) &&
820 	    !test_bit(Journal, &rdev->flags) &&
821 	    mddev->kobj.sd) {
822 		sprintf(nm, "rd%d", rdev->raid_disk);
823 		sysfs_remove_link(&mddev->kobj, nm);
824 	}
825 }
826 
827 /*
828  * iterates through some rdev ringlist. It's safe to remove the
829  * current 'rdev'. Dont touch 'tmp' though.
830  */
831 #define rdev_for_each_list(rdev, tmp, head)				\
832 	list_for_each_entry_safe(rdev, tmp, head, same_set)
833 
834 /*
835  * iterates through the 'same array disks' ringlist
836  */
837 #define rdev_for_each(rdev, mddev)				\
838 	list_for_each_entry(rdev, &((mddev)->disks), same_set)
839 
840 #define rdev_for_each_safe(rdev, tmp, mddev)				\
841 	list_for_each_entry_safe(rdev, tmp, &((mddev)->disks), same_set)
842 
843 #define rdev_for_each_rcu(rdev, mddev)				\
844 	list_for_each_entry_rcu(rdev, &((mddev)->disks), same_set)
845 
846 struct md_thread {
847 	void			(*run) (struct md_thread *thread);
848 	struct mddev		*mddev;
849 	wait_queue_head_t	wqueue;
850 	unsigned long		flags;
851 	struct task_struct	*tsk;
852 	unsigned long		timeout;
853 	void			*private;
854 };
855 
856 struct md_io_clone {
857 	struct mddev	*mddev;
858 	struct bio	*orig_bio;
859 	unsigned long	start_time;
860 	sector_t	offset;
861 	unsigned long	sectors;
862 	struct bio	bio_clone;
863 };
864 
865 #define THREAD_WAKEUP  0
866 
safe_put_page(struct page * p)867 static inline void safe_put_page(struct page *p)
868 {
869 	if (p) put_page(p);
870 }
871 
872 int register_md_submodule(struct md_submodule_head *msh);
873 void unregister_md_submodule(struct md_submodule_head *msh);
874 
875 extern struct md_thread *md_register_thread(
876 	void (*run)(struct md_thread *thread),
877 	struct mddev *mddev,
878 	const char *name);
879 extern void md_unregister_thread(struct mddev *mddev, struct md_thread __rcu **threadp);
880 extern void md_wakeup_thread(struct md_thread __rcu *thread);
881 extern void md_check_recovery(struct mddev *mddev);
882 extern void md_reap_sync_thread(struct mddev *mddev);
883 extern enum sync_action md_sync_action(struct mddev *mddev);
884 extern enum sync_action md_sync_action_by_name(const char *page);
885 extern const char *md_sync_action_name(enum sync_action action);
886 extern void md_write_start(struct mddev *mddev, struct bio *bi);
887 extern void md_write_inc(struct mddev *mddev, struct bio *bi);
888 extern void md_write_end(struct mddev *mddev);
889 extern void md_done_sync(struct mddev *mddev, int blocks, int ok);
890 extern void md_error(struct mddev *mddev, struct md_rdev *rdev);
891 extern void md_finish_reshape(struct mddev *mddev);
892 void md_submit_discard_bio(struct mddev *mddev, struct md_rdev *rdev,
893 			struct bio *bio, sector_t start, sector_t size);
894 void md_account_bio(struct mddev *mddev, struct bio **bio);
895 void md_free_cloned_bio(struct bio *bio);
896 
897 extern bool __must_check md_flush_request(struct mddev *mddev, struct bio *bio);
898 extern void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
899 			   sector_t sector, int size, struct page *page);
900 extern int md_super_wait(struct mddev *mddev);
901 extern int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
902 		struct page *page, blk_opf_t opf, bool metadata_op);
903 extern void md_do_sync(struct md_thread *thread);
904 extern void md_new_event(void);
905 extern void md_allow_write(struct mddev *mddev);
906 extern void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev);
907 extern void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors);
908 extern int md_check_no_bitmap(struct mddev *mddev);
909 extern int md_integrity_register(struct mddev *mddev);
910 extern int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale);
911 
912 extern int mddev_init(struct mddev *mddev);
913 extern void mddev_destroy(struct mddev *mddev);
914 void md_init_stacking_limits(struct queue_limits *lim);
915 struct mddev *md_alloc(dev_t dev, char *name);
916 void mddev_put(struct mddev *mddev);
917 extern int md_run(struct mddev *mddev);
918 extern int md_start(struct mddev *mddev);
919 extern void md_stop(struct mddev *mddev);
920 extern void md_stop_writes(struct mddev *mddev);
921 extern int md_rdev_init(struct md_rdev *rdev);
922 extern void md_rdev_clear(struct md_rdev *rdev);
923 
924 extern bool md_handle_request(struct mddev *mddev, struct bio *bio);
925 extern int mddev_suspend(struct mddev *mddev, bool interruptible);
926 extern void mddev_resume(struct mddev *mddev);
927 extern void md_idle_sync_thread(struct mddev *mddev);
928 extern void md_frozen_sync_thread(struct mddev *mddev);
929 extern void md_unfrozen_sync_thread(struct mddev *mddev);
930 
931 extern void md_update_sb(struct mddev *mddev, int force);
932 extern void mddev_create_serial_pool(struct mddev *mddev, struct md_rdev *rdev);
933 extern void mddev_destroy_serial_pool(struct mddev *mddev,
934 				      struct md_rdev *rdev);
935 struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr);
936 struct md_rdev *md_find_rdev_rcu(struct mddev *mddev, dev_t dev);
937 
is_rdev_broken(struct md_rdev * rdev)938 static inline bool is_rdev_broken(struct md_rdev *rdev)
939 {
940 	return !disk_live(rdev->bdev->bd_disk);
941 }
942 
rdev_dec_pending(struct md_rdev * rdev,struct mddev * mddev)943 static inline void rdev_dec_pending(struct md_rdev *rdev, struct mddev *mddev)
944 {
945 	int faulty = test_bit(Faulty, &rdev->flags);
946 	if (atomic_dec_and_test(&rdev->nr_pending) && faulty) {
947 		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
948 		md_wakeup_thread(mddev->thread);
949 	}
950 }
951 
mddev_is_clustered(struct mddev * mddev)952 static inline int mddev_is_clustered(struct mddev *mddev)
953 {
954 	return mddev->cluster_info && mddev->bitmap_info.nodes > 1;
955 }
956 
957 /* clear unsupported mddev_flags */
mddev_clear_unsupported_flags(struct mddev * mddev,unsigned long unsupported_flags)958 static inline void mddev_clear_unsupported_flags(struct mddev *mddev,
959 	unsigned long unsupported_flags)
960 {
961 	mddev->flags &= ~unsupported_flags;
962 }
963 
mddev_check_write_zeroes(struct mddev * mddev,struct bio * bio)964 static inline void mddev_check_write_zeroes(struct mddev *mddev, struct bio *bio)
965 {
966 	if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
967 	    !bio->bi_bdev->bd_disk->queue->limits.max_write_zeroes_sectors)
968 		mddev->gendisk->queue->limits.max_write_zeroes_sectors = 0;
969 }
970 
mddev_suspend_and_lock(struct mddev * mddev)971 static inline int mddev_suspend_and_lock(struct mddev *mddev)
972 {
973 	int ret;
974 
975 	ret = mddev_suspend(mddev, true);
976 	if (ret)
977 		return ret;
978 
979 	ret = mddev_lock(mddev);
980 	if (ret)
981 		mddev_resume(mddev);
982 
983 	return ret;
984 }
985 
mddev_suspend_and_lock_nointr(struct mddev * mddev)986 static inline void mddev_suspend_and_lock_nointr(struct mddev *mddev)
987 {
988 	mddev_suspend(mddev, false);
989 	mutex_lock(&mddev->reconfig_mutex);
990 }
991 
mddev_unlock_and_resume(struct mddev * mddev)992 static inline void mddev_unlock_and_resume(struct mddev *mddev)
993 {
994 	mddev_unlock(mddev);
995 	mddev_resume(mddev);
996 }
997 
998 struct mdu_array_info_s;
999 struct mdu_disk_info_s;
1000 
1001 extern int mdp_major;
1002 extern struct workqueue_struct *md_bitmap_wq;
1003 void md_autostart_arrays(int part);
1004 int md_set_array_info(struct mddev *mddev, struct mdu_array_info_s *info);
1005 int md_add_new_disk(struct mddev *mddev, struct mdu_disk_info_s *info);
1006 int do_md_run(struct mddev *mddev);
1007 #define MDDEV_STACK_INTEGRITY	(1u << 0)
1008 int mddev_stack_rdev_limits(struct mddev *mddev, struct queue_limits *lim,
1009 		unsigned int flags);
1010 int mddev_stack_new_rdev(struct mddev *mddev, struct md_rdev *rdev);
1011 void mddev_update_io_opt(struct mddev *mddev, unsigned int nr_stripes);
1012 
1013 extern const struct block_device_operations md_fops;
1014 
1015 /*
1016  * MD devices can be used undeneath by DM, in which case ->gendisk is NULL.
1017  */
mddev_is_dm(struct mddev * mddev)1018 static inline bool mddev_is_dm(struct mddev *mddev)
1019 {
1020 	return !mddev->gendisk;
1021 }
1022 
mddev_trace_remap(struct mddev * mddev,struct bio * bio,sector_t sector)1023 static inline void mddev_trace_remap(struct mddev *mddev, struct bio *bio,
1024 		sector_t sector)
1025 {
1026 	if (!mddev_is_dm(mddev))
1027 		trace_block_bio_remap(bio, disk_devt(mddev->gendisk), sector);
1028 }
1029 
rdev_blocked(struct md_rdev * rdev)1030 static inline bool rdev_blocked(struct md_rdev *rdev)
1031 {
1032 	/*
1033 	 * Blocked will be set by error handler and cleared by daemon after
1034 	 * updating superblock, meanwhile write IO should be blocked to prevent
1035 	 * reading old data after power failure.
1036 	 */
1037 	if (test_bit(Blocked, &rdev->flags))
1038 		return true;
1039 
1040 	/*
1041 	 * Faulty device should not be accessed anymore, there is no need to
1042 	 * wait for bad block to be acknowledged.
1043 	 */
1044 	if (test_bit(Faulty, &rdev->flags))
1045 		return false;
1046 
1047 	/* rdev is blocked by badblocks. */
1048 	if (test_bit(BlockedBadBlocks, &rdev->flags))
1049 		return true;
1050 
1051 	return false;
1052 }
1053 
1054 #define mddev_add_trace_msg(mddev, fmt, args...)			\
1055 do {									\
1056 	if (!mddev_is_dm(mddev))					\
1057 		blk_add_trace_msg((mddev)->gendisk->queue, fmt, ##args); \
1058 } while (0)
1059 
1060 #endif /* _MD_MD_H */
1061