xref: /linux/drivers/md/md.h (revision ee8287e068a3995b0f8001dd6931e221dfb7c530)
1 /* SPDX-License-Identifier: GPL-2.0-or-later */
2 /*
3    md.h : kernel internal structure of the Linux MD driver
4           Copyright (C) 1996-98 Ingo Molnar, Gadi Oxman
5 
6 */
7 
8 #ifndef _MD_MD_H
9 #define _MD_MD_H
10 
11 #include <linux/blkdev.h>
12 #include <linux/backing-dev.h>
13 #include <linux/badblocks.h>
14 #include <linux/kobject.h>
15 #include <linux/list.h>
16 #include <linux/mm.h>
17 #include <linux/mutex.h>
18 #include <linux/timer.h>
19 #include <linux/wait.h>
20 #include <linux/workqueue.h>
21 #include <trace/events/block.h>
22 #include "md-cluster.h"
23 
24 #define MaxSector (~(sector_t)0)
25 
26 /*
27  * These flags should really be called "NO_RETRY" rather than
28  * "FAILFAST" because they don't make any promise about time lapse,
29  * only about the number of retries, which will be zero.
30  * REQ_FAILFAST_DRIVER is not included because
31  * Commit: 4a27446f3e39 ("[SCSI] modify scsi to handle new fail fast flags.")
32  * seems to suggest that the errors it avoids retrying should usually
33  * be retried.
34  */
35 #define	MD_FAILFAST	(REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT)
36 
37 /*
38  * The struct embedded in rdev is used to serialize IO.
39  */
40 struct serial_in_rdev {
41 	struct rb_root_cached serial_rb;
42 	spinlock_t serial_lock;
43 	wait_queue_head_t serial_io_wait;
44 };
45 
46 /*
47  * MD's 'extended' device
48  */
49 struct md_rdev {
50 	struct list_head same_set;	/* RAID devices within the same set */
51 
52 	sector_t sectors;		/* Device size (in 512bytes sectors) */
53 	struct mddev *mddev;		/* RAID array if running */
54 	int last_events;		/* IO event timestamp */
55 
56 	/*
57 	 * If meta_bdev is non-NULL, it means that a separate device is
58 	 * being used to store the metadata (superblock/bitmap) which
59 	 * would otherwise be contained on the same device as the data (bdev).
60 	 */
61 	struct block_device *meta_bdev;
62 	struct block_device *bdev;	/* block device handle */
63 	struct file *bdev_file;		/* Handle from open for bdev */
64 
65 	struct page	*sb_page, *bb_page;
66 	int		sb_loaded;
67 	__u64		sb_events;
68 	sector_t	data_offset;	/* start of data in array */
69 	sector_t	new_data_offset;/* only relevant while reshaping */
70 	sector_t	sb_start;	/* offset of the super block (in 512byte sectors) */
71 	int		sb_size;	/* bytes in the superblock */
72 	int		preferred_minor;	/* autorun support */
73 
74 	struct kobject	kobj;
75 
76 	/* A device can be in one of three states based on two flags:
77 	 * Not working:   faulty==1 in_sync==0
78 	 * Fully working: faulty==0 in_sync==1
79 	 * Working, but not
80 	 * in sync with array
81 	 *                faulty==0 in_sync==0
82 	 *
83 	 * It can never have faulty==1, in_sync==1
84 	 * This reduces the burden of testing multiple flags in many cases
85 	 */
86 
87 	unsigned long	flags;	/* bit set of 'enum flag_bits' bits. */
88 	wait_queue_head_t blocked_wait;
89 
90 	int desc_nr;			/* descriptor index in the superblock */
91 	int raid_disk;			/* role of device in array */
92 	int new_raid_disk;		/* role that the device will have in
93 					 * the array after a level-change completes.
94 					 */
95 	int saved_raid_disk;		/* role that device used to have in the
96 					 * array and could again if we did a partial
97 					 * resync from the bitmap
98 					 */
99 	union {
100 		sector_t recovery_offset;/* If this device has been partially
101 					 * recovered, this is where we were
102 					 * up to.
103 					 */
104 		sector_t journal_tail;	/* If this device is a journal device,
105 					 * this is the journal tail (journal
106 					 * recovery start point)
107 					 */
108 	};
109 
110 	atomic_t	nr_pending;	/* number of pending requests.
111 					 * only maintained for arrays that
112 					 * support hot removal
113 					 */
114 	atomic_t	read_errors;	/* number of consecutive read errors that
115 					 * we have tried to ignore.
116 					 */
117 	time64_t	last_read_error;	/* monotonic time since our
118 						 * last read error
119 						 */
120 	atomic_t	corrected_errors; /* number of corrected read errors,
121 					   * for reporting to userspace and storing
122 					   * in superblock.
123 					   */
124 
125 	struct serial_in_rdev *serial;  /* used for raid1 io serialization */
126 
127 	struct kernfs_node *sysfs_state; /* handle for 'state'
128 					   * sysfs entry */
129 	/* handle for 'unacknowledged_bad_blocks' sysfs dentry */
130 	struct kernfs_node *sysfs_unack_badblocks;
131 	/* handle for 'bad_blocks' sysfs dentry */
132 	struct kernfs_node *sysfs_badblocks;
133 	struct badblocks badblocks;
134 
135 	struct {
136 		short offset;	/* Offset from superblock to start of PPL.
137 				 * Not used by external metadata. */
138 		unsigned int size;	/* Size in sectors of the PPL space */
139 		sector_t sector;	/* First sector of the PPL space */
140 	} ppl;
141 };
142 enum flag_bits {
143 	Faulty,			/* device is known to have a fault */
144 	In_sync,		/* device is in_sync with rest of array */
145 	Bitmap_sync,		/* ..actually, not quite In_sync.  Need a
146 				 * bitmap-based recovery to get fully in sync.
147 				 * The bit is only meaningful before device
148 				 * has been passed to pers->hot_add_disk.
149 				 */
150 	WriteMostly,		/* Avoid reading if at all possible */
151 	AutoDetected,		/* added by auto-detect */
152 	Blocked,		/* An error occurred but has not yet
153 				 * been acknowledged by the metadata
154 				 * handler, so don't allow writes
155 				 * until it is cleared */
156 	WriteErrorSeen,		/* A write error has been seen on this
157 				 * device
158 				 */
159 	FaultRecorded,		/* Intermediate state for clearing
160 				 * Blocked.  The Fault is/will-be
161 				 * recorded in the metadata, but that
162 				 * metadata hasn't been stored safely
163 				 * on disk yet.
164 				 */
165 	BlockedBadBlocks,	/* A writer is blocked because they
166 				 * found an unacknowledged bad-block.
167 				 * This can safely be cleared at any
168 				 * time, and the writer will re-check.
169 				 * It may be set at any time, and at
170 				 * worst the writer will timeout and
171 				 * re-check.  So setting it as
172 				 * accurately as possible is good, but
173 				 * not absolutely critical.
174 				 */
175 	WantReplacement,	/* This device is a candidate to be
176 				 * hot-replaced, either because it has
177 				 * reported some faults, or because
178 				 * of explicit request.
179 				 */
180 	Replacement,		/* This device is a replacement for
181 				 * a want_replacement device with same
182 				 * raid_disk number.
183 				 */
184 	Candidate,		/* For clustered environments only:
185 				 * This device is seen locally but not
186 				 * by the whole cluster
187 				 */
188 	Journal,		/* This device is used as journal for
189 				 * raid-5/6.
190 				 * Usually, this device should be faster
191 				 * than other devices in the array
192 				 */
193 	ClusterRemove,
194 	ExternalBbl,            /* External metadata provides bad
195 				 * block management for a disk
196 				 */
197 	FailFast,		/* Minimal retries should be attempted on
198 				 * this device, so use REQ_FAILFAST_DEV.
199 				 * Also don't try to repair failed reads.
200 				 * It is expects that no bad block log
201 				 * is present.
202 				 */
203 	LastDev,		/* Seems to be the last working dev as
204 				 * it didn't fail, so don't use FailFast
205 				 * any more for metadata
206 				 */
207 	CollisionCheck,		/*
208 				 * check if there is collision between raid1
209 				 * serial bios.
210 				 */
211 	Nonrot,			/* non-rotational device (SSD) */
212 };
213 
214 static inline int is_badblock(struct md_rdev *rdev, sector_t s, int sectors,
215 			      sector_t *first_bad, int *bad_sectors)
216 {
217 	if (unlikely(rdev->badblocks.count)) {
218 		int rv = badblocks_check(&rdev->badblocks, rdev->data_offset + s,
219 					sectors,
220 					first_bad, bad_sectors);
221 		if (rv)
222 			*first_bad -= rdev->data_offset;
223 		return rv;
224 	}
225 	return 0;
226 }
227 
228 static inline int rdev_has_badblock(struct md_rdev *rdev, sector_t s,
229 				    int sectors)
230 {
231 	sector_t first_bad;
232 	int bad_sectors;
233 
234 	return is_badblock(rdev, s, sectors, &first_bad, &bad_sectors);
235 }
236 
237 extern int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
238 			      int is_new);
239 extern int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
240 				int is_new);
241 struct md_cluster_info;
242 
243 /**
244  * enum mddev_flags - md device flags.
245  * @MD_ARRAY_FIRST_USE: First use of array, needs initialization.
246  * @MD_CLOSING: If set, we are closing the array, do not open it then.
247  * @MD_JOURNAL_CLEAN: A raid with journal is already clean.
248  * @MD_HAS_JOURNAL: The raid array has journal feature set.
249  * @MD_CLUSTER_RESYNC_LOCKED: cluster raid only, which means node, already took
250  *			       resync lock, need to release the lock.
251  * @MD_FAILFAST_SUPPORTED: Using MD_FAILFAST on metadata writes is supported as
252  *			    calls to md_error() will never cause the array to
253  *			    become failed.
254  * @MD_HAS_PPL:  The raid array has PPL feature set.
255  * @MD_HAS_MULTIPLE_PPLS: The raid array has multiple PPLs feature set.
256  * @MD_NOT_READY: do_md_run() is active, so 'array_state', ust not report that
257  *		   array is ready yet.
258  * @MD_BROKEN: This is used to stop writes and mark array as failed.
259  * @MD_DELETED: This device is being deleted
260  *
261  * change UNSUPPORTED_MDDEV_FLAGS for each array type if new flag is added
262  */
263 enum mddev_flags {
264 	MD_ARRAY_FIRST_USE,
265 	MD_CLOSING,
266 	MD_JOURNAL_CLEAN,
267 	MD_HAS_JOURNAL,
268 	MD_CLUSTER_RESYNC_LOCKED,
269 	MD_FAILFAST_SUPPORTED,
270 	MD_HAS_PPL,
271 	MD_HAS_MULTIPLE_PPLS,
272 	MD_NOT_READY,
273 	MD_BROKEN,
274 	MD_DELETED,
275 };
276 
277 enum mddev_sb_flags {
278 	MD_SB_CHANGE_DEVS,		/* Some device status has changed */
279 	MD_SB_CHANGE_CLEAN,	/* transition to or from 'clean' */
280 	MD_SB_CHANGE_PENDING,	/* switch from 'clean' to 'active' in progress */
281 	MD_SB_NEED_REWRITE,	/* metadata write needs to be repeated */
282 };
283 
284 #define NR_SERIAL_INFOS		8
285 /* record current range of serialize IOs */
286 struct serial_info {
287 	struct rb_node node;
288 	sector_t start;		/* start sector of rb node */
289 	sector_t last;		/* end sector of rb node */
290 	sector_t _subtree_last; /* highest sector in subtree of rb node */
291 };
292 
293 /*
294  * mddev->curr_resync stores the current sector of the resync but
295  * also has some overloaded values.
296  */
297 enum {
298 	/* No resync in progress */
299 	MD_RESYNC_NONE = 0,
300 	/* Yielded to allow another conflicting resync to commence */
301 	MD_RESYNC_YIELDED = 1,
302 	/* Delayed to check that there is no conflict with another sync */
303 	MD_RESYNC_DELAYED = 2,
304 	/* Any value greater than or equal to this is in an active resync */
305 	MD_RESYNC_ACTIVE = 3,
306 };
307 
308 struct mddev {
309 	void				*private;
310 	struct md_personality		*pers;
311 	dev_t				unit;
312 	int				md_minor;
313 	struct list_head		disks;
314 	unsigned long			flags;
315 	unsigned long			sb_flags;
316 
317 	int				suspended;
318 	struct mutex			suspend_mutex;
319 	struct percpu_ref		active_io;
320 	int				ro;
321 	int				sysfs_active; /* set when sysfs deletes
322 						       * are happening, so run/
323 						       * takeover/stop are not safe
324 						       */
325 	struct gendisk			*gendisk;
326 
327 	struct kobject			kobj;
328 	int				hold_active;
329 #define	UNTIL_IOCTL	1
330 #define	UNTIL_STOP	2
331 
332 	/* Superblock information */
333 	int				major_version,
334 					minor_version,
335 					patch_version;
336 	int				persistent;
337 	int				external;	/* metadata is
338 							 * managed externally */
339 	char				metadata_type[17]; /* externally set*/
340 	int				chunk_sectors;
341 	time64_t			ctime, utime;
342 	int				level, layout;
343 	char				clevel[16];
344 	int				raid_disks;
345 	int				max_disks;
346 	sector_t			dev_sectors;	/* used size of
347 							 * component devices */
348 	sector_t			array_sectors; /* exported array size */
349 	int				external_size; /* size managed
350 							* externally */
351 	__u64				events;
352 	/* If the last 'event' was simply a clean->dirty transition, and
353 	 * we didn't write it to the spares, then it is safe and simple
354 	 * to just decrement the event count on a dirty->clean transition.
355 	 * So we record that possibility here.
356 	 */
357 	int				can_decrease_events;
358 
359 	char				uuid[16];
360 
361 	/* If the array is being reshaped, we need to record the
362 	 * new shape and an indication of where we are up to.
363 	 * This is written to the superblock.
364 	 * If reshape_position is MaxSector, then no reshape is happening (yet).
365 	 */
366 	sector_t			reshape_position;
367 	int				delta_disks, new_level, new_layout;
368 	int				new_chunk_sectors;
369 	int				reshape_backwards;
370 
371 	struct md_thread __rcu		*thread;	/* management thread */
372 	struct md_thread __rcu		*sync_thread;	/* doing resync or reconstruct */
373 
374 	/* 'last_sync_action' is initialized to "none".  It is set when a
375 	 * sync operation (i.e "data-check", "requested-resync", "resync",
376 	 * "recovery", or "reshape") is started.  It holds this value even
377 	 * when the sync thread is "frozen" (interrupted) or "idle" (stopped
378 	 * or finished).  It is overwritten when a new sync operation is begun.
379 	 */
380 	char				*last_sync_action;
381 	sector_t			curr_resync;	/* last block scheduled */
382 	/* As resync requests can complete out of order, we cannot easily track
383 	 * how much resync has been completed.  So we occasionally pause until
384 	 * everything completes, then set curr_resync_completed to curr_resync.
385 	 * As such it may be well behind the real resync mark, but it is a value
386 	 * we are certain of.
387 	 */
388 	sector_t			curr_resync_completed;
389 	unsigned long			resync_mark;	/* a recent timestamp */
390 	sector_t			resync_mark_cnt;/* blocks written at resync_mark */
391 	sector_t			curr_mark_cnt; /* blocks scheduled now */
392 
393 	sector_t			resync_max_sectors; /* may be set by personality */
394 
395 	atomic64_t			resync_mismatches; /* count of sectors where
396 							    * parity/replica mismatch found
397 							    */
398 
399 	/* allow user-space to request suspension of IO to regions of the array */
400 	sector_t			suspend_lo;
401 	sector_t			suspend_hi;
402 	/* if zero, use the system-wide default */
403 	int				sync_speed_min;
404 	int				sync_speed_max;
405 
406 	/* resync even though the same disks are shared among md-devices */
407 	int				parallel_resync;
408 
409 	int				ok_start_degraded;
410 
411 	unsigned long			recovery;
412 	/* If a RAID personality determines that recovery (of a particular
413 	 * device) will fail due to a read error on the source device, it
414 	 * takes a copy of this number and does not attempt recovery again
415 	 * until this number changes.
416 	 */
417 	int				recovery_disabled;
418 
419 	int				in_sync;	/* know to not need resync */
420 	/* 'open_mutex' avoids races between 'md_open' and 'do_md_stop', so
421 	 * that we are never stopping an array while it is open.
422 	 * 'reconfig_mutex' protects all other reconfiguration.
423 	 * These locks are separate due to conflicting interactions
424 	 * with disk->open_mutex.
425 	 * Lock ordering is:
426 	 *  reconfig_mutex -> disk->open_mutex
427 	 *  disk->open_mutex -> open_mutex:  e.g. __blkdev_get -> md_open
428 	 */
429 	struct mutex			open_mutex;
430 	struct mutex			reconfig_mutex;
431 	atomic_t			active;		/* general refcount */
432 	atomic_t			openers;	/* number of active opens */
433 
434 	int				changed;	/* True if we might need to
435 							 * reread partition info */
436 	int				degraded;	/* whether md should consider
437 							 * adding a spare
438 							 */
439 
440 	atomic_t			recovery_active; /* blocks scheduled, but not written */
441 	wait_queue_head_t		recovery_wait;
442 	sector_t			recovery_cp;
443 	sector_t			resync_min;	/* user requested sync
444 							 * starts here */
445 	sector_t			resync_max;	/* resync should pause
446 							 * when it gets here */
447 
448 	struct kernfs_node		*sysfs_state;	/* handle for 'array_state'
449 							 * file in sysfs.
450 							 */
451 	struct kernfs_node		*sysfs_action;  /* handle for 'sync_action' */
452 	struct kernfs_node		*sysfs_completed;	/*handle for 'sync_completed' */
453 	struct kernfs_node		*sysfs_degraded;	/*handle for 'degraded' */
454 	struct kernfs_node		*sysfs_level;		/*handle for 'level' */
455 
456 	/* used for delayed sysfs removal */
457 	struct work_struct del_work;
458 	/* used for register new sync thread */
459 	struct work_struct sync_work;
460 
461 	/* "lock" protects:
462 	 *   flush_bio transition from NULL to !NULL
463 	 *   rdev superblocks, events
464 	 *   clearing MD_CHANGE_*
465 	 *   in_sync - and related safemode and MD_CHANGE changes
466 	 *   pers (also protected by reconfig_mutex and pending IO).
467 	 *   clearing ->bitmap
468 	 *   clearing ->bitmap_info.file
469 	 *   changing ->resync_{min,max}
470 	 *   setting MD_RECOVERY_RUNNING (which interacts with resync_{min,max})
471 	 */
472 	spinlock_t			lock;
473 	wait_queue_head_t		sb_wait;	/* for waiting on superblock updates */
474 	atomic_t			pending_writes;	/* number of active superblock writes */
475 
476 	unsigned int			safemode;	/* if set, update "clean" superblock
477 							 * when no writes pending.
478 							 */
479 	unsigned int			safemode_delay;
480 	struct timer_list		safemode_timer;
481 	struct percpu_ref		writes_pending;
482 	int				sync_checkers;	/* # of threads checking writes_pending */
483 
484 	struct bitmap			*bitmap; /* the bitmap for the device */
485 	struct {
486 		struct file		*file; /* the bitmap file */
487 		loff_t			offset; /* offset from superblock of
488 						 * start of bitmap. May be
489 						 * negative, but not '0'
490 						 * For external metadata, offset
491 						 * from start of device.
492 						 */
493 		unsigned long		space; /* space available at this offset */
494 		loff_t			default_offset; /* this is the offset to use when
495 							 * hot-adding a bitmap.  It should
496 							 * eventually be settable by sysfs.
497 							 */
498 		unsigned long		default_space; /* space available at
499 							* default offset */
500 		struct mutex		mutex;
501 		unsigned long		chunksize;
502 		unsigned long		daemon_sleep; /* how many jiffies between updates? */
503 		unsigned long		max_write_behind; /* write-behind mode */
504 		int			external;
505 		int			nodes; /* Maximum number of nodes in the cluster */
506 		char                    cluster_name[64]; /* Name of the cluster */
507 	} bitmap_info;
508 
509 	atomic_t			max_corr_read_errors; /* max read retries */
510 	struct list_head		all_mddevs;
511 
512 	const struct attribute_group	*to_remove;
513 
514 	struct bio_set			bio_set;
515 	struct bio_set			sync_set; /* for sync operations like
516 						   * metadata and bitmap writes
517 						   */
518 	struct bio_set			io_clone_set;
519 
520 	/* Generic flush handling.
521 	 * The last to finish preflush schedules a worker to submit
522 	 * the rest of the request (without the REQ_PREFLUSH flag).
523 	 */
524 	struct bio *flush_bio;
525 	atomic_t flush_pending;
526 	ktime_t start_flush, prev_flush_start; /* prev_flush_start is when the previous completed
527 						* flush was started.
528 						*/
529 	struct work_struct flush_work;
530 	struct work_struct event_work;	/* used by dm to report failure event */
531 	mempool_t *serial_info_pool;
532 	void (*sync_super)(struct mddev *mddev, struct md_rdev *rdev);
533 	struct md_cluster_info		*cluster_info;
534 	unsigned int			good_device_nr;	/* good device num within cluster raid */
535 	unsigned int			noio_flag; /* for memalloc scope API */
536 
537 	/*
538 	 * Temporarily store rdev that will be finally removed when
539 	 * reconfig_mutex is unlocked, protected by reconfig_mutex.
540 	 */
541 	struct list_head		deleting;
542 
543 	/* Used to synchronize idle and frozen for action_store() */
544 	struct mutex			sync_mutex;
545 	/* The sequence number for sync thread */
546 	atomic_t sync_seq;
547 
548 	bool	has_superblocks:1;
549 	bool	fail_last_dev:1;
550 	bool	serialize_policy:1;
551 };
552 
553 enum recovery_flags {
554 	/*
555 	 * If neither SYNC or RESHAPE are set, then it is a recovery.
556 	 */
557 	MD_RECOVERY_RUNNING,	/* a thread is running, or about to be started */
558 	MD_RECOVERY_SYNC,	/* actually doing a resync, not a recovery */
559 	MD_RECOVERY_RECOVER,	/* doing recovery, or need to try it. */
560 	MD_RECOVERY_INTR,	/* resync needs to be aborted for some reason */
561 	MD_RECOVERY_DONE,	/* thread is done and is waiting to be reaped */
562 	MD_RECOVERY_NEEDED,	/* we might need to start a resync/recover */
563 	MD_RECOVERY_REQUESTED,	/* user-space has requested a sync (used with SYNC) */
564 	MD_RECOVERY_CHECK,	/* user-space request for check-only, no repair */
565 	MD_RECOVERY_RESHAPE,	/* A reshape is happening */
566 	MD_RECOVERY_FROZEN,	/* User request to abort, and not restart, any action */
567 	MD_RECOVERY_ERROR,	/* sync-action interrupted because io-error */
568 	MD_RECOVERY_WAIT,	/* waiting for pers->start() to finish */
569 	MD_RESYNCING_REMOTE,	/* remote node is running resync thread */
570 };
571 
572 enum md_ro_state {
573 	MD_RDWR,
574 	MD_RDONLY,
575 	MD_AUTO_READ,
576 	MD_MAX_STATE
577 };
578 
579 static inline bool md_is_rdwr(struct mddev *mddev)
580 {
581 	return (mddev->ro == MD_RDWR);
582 }
583 
584 static inline bool reshape_interrupted(struct mddev *mddev)
585 {
586 	/* reshape never start */
587 	if (mddev->reshape_position == MaxSector)
588 		return false;
589 
590 	/* interrupted */
591 	if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
592 		return true;
593 
594 	/* running reshape will be interrupted soon. */
595 	if (test_bit(MD_RECOVERY_WAIT, &mddev->recovery) ||
596 	    test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
597 	    test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
598 		return true;
599 
600 	return false;
601 }
602 
603 static inline int __must_check mddev_lock(struct mddev *mddev)
604 {
605 	return mutex_lock_interruptible(&mddev->reconfig_mutex);
606 }
607 
608 /* Sometimes we need to take the lock in a situation where
609  * failure due to interrupts is not acceptable.
610  */
611 static inline void mddev_lock_nointr(struct mddev *mddev)
612 {
613 	mutex_lock(&mddev->reconfig_mutex);
614 }
615 
616 static inline int mddev_trylock(struct mddev *mddev)
617 {
618 	return mutex_trylock(&mddev->reconfig_mutex);
619 }
620 extern void mddev_unlock(struct mddev *mddev);
621 
622 static inline void md_sync_acct(struct block_device *bdev, unsigned long nr_sectors)
623 {
624 	if (blk_queue_io_stat(bdev->bd_disk->queue))
625 		atomic_add(nr_sectors, &bdev->bd_disk->sync_io);
626 }
627 
628 static inline void md_sync_acct_bio(struct bio *bio, unsigned long nr_sectors)
629 {
630 	md_sync_acct(bio->bi_bdev, nr_sectors);
631 }
632 
633 struct md_personality
634 {
635 	char *name;
636 	int level;
637 	struct list_head list;
638 	struct module *owner;
639 	bool __must_check (*make_request)(struct mddev *mddev, struct bio *bio);
640 	/*
641 	 * start up works that do NOT require md_thread. tasks that
642 	 * requires md_thread should go into start()
643 	 */
644 	int (*run)(struct mddev *mddev);
645 	/* start up works that require md threads */
646 	int (*start)(struct mddev *mddev);
647 	void (*free)(struct mddev *mddev, void *priv);
648 	void (*status)(struct seq_file *seq, struct mddev *mddev);
649 	/* error_handler must set ->faulty and clear ->in_sync
650 	 * if appropriate, and should abort recovery if needed
651 	 */
652 	void (*error_handler)(struct mddev *mddev, struct md_rdev *rdev);
653 	int (*hot_add_disk) (struct mddev *mddev, struct md_rdev *rdev);
654 	int (*hot_remove_disk) (struct mddev *mddev, struct md_rdev *rdev);
655 	int (*spare_active) (struct mddev *mddev);
656 	sector_t (*sync_request)(struct mddev *mddev, sector_t sector_nr, int *skipped);
657 	int (*resize) (struct mddev *mddev, sector_t sectors);
658 	sector_t (*size) (struct mddev *mddev, sector_t sectors, int raid_disks);
659 	int (*check_reshape) (struct mddev *mddev);
660 	int (*start_reshape) (struct mddev *mddev);
661 	void (*finish_reshape) (struct mddev *mddev);
662 	void (*update_reshape_pos) (struct mddev *mddev);
663 	void (*prepare_suspend) (struct mddev *mddev);
664 	/* quiesce suspends or resumes internal processing.
665 	 * 1 - stop new actions and wait for action io to complete
666 	 * 0 - return to normal behaviour
667 	 */
668 	void (*quiesce) (struct mddev *mddev, int quiesce);
669 	/* takeover is used to transition an array from one
670 	 * personality to another.  The new personality must be able
671 	 * to handle the data in the current layout.
672 	 * e.g. 2drive raid1 -> 2drive raid5
673 	 *      ndrive raid5 -> degraded n+1drive raid6 with special layout
674 	 * If the takeover succeeds, a new 'private' structure is returned.
675 	 * This needs to be installed and then ->run used to activate the
676 	 * array.
677 	 */
678 	void *(*takeover) (struct mddev *mddev);
679 	/* Changes the consistency policy of an active array. */
680 	int (*change_consistency_policy)(struct mddev *mddev, const char *buf);
681 };
682 
683 struct md_sysfs_entry {
684 	struct attribute attr;
685 	ssize_t (*show)(struct mddev *, char *);
686 	ssize_t (*store)(struct mddev *, const char *, size_t);
687 };
688 extern const struct attribute_group md_bitmap_group;
689 
690 static inline struct kernfs_node *sysfs_get_dirent_safe(struct kernfs_node *sd, char *name)
691 {
692 	if (sd)
693 		return sysfs_get_dirent(sd, name);
694 	return sd;
695 }
696 static inline void sysfs_notify_dirent_safe(struct kernfs_node *sd)
697 {
698 	if (sd)
699 		sysfs_notify_dirent(sd);
700 }
701 
702 static inline char * mdname (struct mddev * mddev)
703 {
704 	return mddev->gendisk ? mddev->gendisk->disk_name : "mdX";
705 }
706 
707 static inline int sysfs_link_rdev(struct mddev *mddev, struct md_rdev *rdev)
708 {
709 	char nm[20];
710 	if (!test_bit(Replacement, &rdev->flags) &&
711 	    !test_bit(Journal, &rdev->flags) &&
712 	    mddev->kobj.sd) {
713 		sprintf(nm, "rd%d", rdev->raid_disk);
714 		return sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
715 	} else
716 		return 0;
717 }
718 
719 static inline void sysfs_unlink_rdev(struct mddev *mddev, struct md_rdev *rdev)
720 {
721 	char nm[20];
722 	if (!test_bit(Replacement, &rdev->flags) &&
723 	    !test_bit(Journal, &rdev->flags) &&
724 	    mddev->kobj.sd) {
725 		sprintf(nm, "rd%d", rdev->raid_disk);
726 		sysfs_remove_link(&mddev->kobj, nm);
727 	}
728 }
729 
730 /*
731  * iterates through some rdev ringlist. It's safe to remove the
732  * current 'rdev'. Dont touch 'tmp' though.
733  */
734 #define rdev_for_each_list(rdev, tmp, head)				\
735 	list_for_each_entry_safe(rdev, tmp, head, same_set)
736 
737 /*
738  * iterates through the 'same array disks' ringlist
739  */
740 #define rdev_for_each(rdev, mddev)				\
741 	list_for_each_entry(rdev, &((mddev)->disks), same_set)
742 
743 #define rdev_for_each_safe(rdev, tmp, mddev)				\
744 	list_for_each_entry_safe(rdev, tmp, &((mddev)->disks), same_set)
745 
746 #define rdev_for_each_rcu(rdev, mddev)				\
747 	list_for_each_entry_rcu(rdev, &((mddev)->disks), same_set)
748 
749 struct md_thread {
750 	void			(*run) (struct md_thread *thread);
751 	struct mddev		*mddev;
752 	wait_queue_head_t	wqueue;
753 	unsigned long		flags;
754 	struct task_struct	*tsk;
755 	unsigned long		timeout;
756 	void			*private;
757 };
758 
759 struct md_io_clone {
760 	struct mddev	*mddev;
761 	struct bio	*orig_bio;
762 	unsigned long	start_time;
763 	struct bio	bio_clone;
764 };
765 
766 #define THREAD_WAKEUP  0
767 
768 static inline void safe_put_page(struct page *p)
769 {
770 	if (p) put_page(p);
771 }
772 
773 extern int register_md_personality(struct md_personality *p);
774 extern int unregister_md_personality(struct md_personality *p);
775 extern int register_md_cluster_operations(struct md_cluster_operations *ops,
776 		struct module *module);
777 extern int unregister_md_cluster_operations(void);
778 extern int md_setup_cluster(struct mddev *mddev, int nodes);
779 extern void md_cluster_stop(struct mddev *mddev);
780 extern struct md_thread *md_register_thread(
781 	void (*run)(struct md_thread *thread),
782 	struct mddev *mddev,
783 	const char *name);
784 extern void md_unregister_thread(struct mddev *mddev, struct md_thread __rcu **threadp);
785 extern void md_wakeup_thread(struct md_thread __rcu *thread);
786 extern void md_check_recovery(struct mddev *mddev);
787 extern void md_reap_sync_thread(struct mddev *mddev);
788 extern bool md_write_start(struct mddev *mddev, struct bio *bi);
789 extern void md_write_inc(struct mddev *mddev, struct bio *bi);
790 extern void md_write_end(struct mddev *mddev);
791 extern void md_done_sync(struct mddev *mddev, int blocks, int ok);
792 extern void md_error(struct mddev *mddev, struct md_rdev *rdev);
793 extern void md_finish_reshape(struct mddev *mddev);
794 void md_submit_discard_bio(struct mddev *mddev, struct md_rdev *rdev,
795 			struct bio *bio, sector_t start, sector_t size);
796 void md_account_bio(struct mddev *mddev, struct bio **bio);
797 void md_free_cloned_bio(struct bio *bio);
798 
799 extern bool __must_check md_flush_request(struct mddev *mddev, struct bio *bio);
800 extern void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
801 			   sector_t sector, int size, struct page *page);
802 extern int md_super_wait(struct mddev *mddev);
803 extern int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
804 		struct page *page, blk_opf_t opf, bool metadata_op);
805 extern void md_do_sync(struct md_thread *thread);
806 extern void md_new_event(void);
807 extern void md_allow_write(struct mddev *mddev);
808 extern void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev);
809 extern void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors);
810 extern int md_check_no_bitmap(struct mddev *mddev);
811 extern int md_integrity_register(struct mddev *mddev);
812 extern int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev);
813 extern int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale);
814 
815 extern int mddev_init(struct mddev *mddev);
816 extern void mddev_destroy(struct mddev *mddev);
817 struct mddev *md_alloc(dev_t dev, char *name);
818 void mddev_put(struct mddev *mddev);
819 extern int md_run(struct mddev *mddev);
820 extern int md_start(struct mddev *mddev);
821 extern void md_stop(struct mddev *mddev);
822 extern void md_stop_writes(struct mddev *mddev);
823 extern int md_rdev_init(struct md_rdev *rdev);
824 extern void md_rdev_clear(struct md_rdev *rdev);
825 
826 extern bool md_handle_request(struct mddev *mddev, struct bio *bio);
827 extern int mddev_suspend(struct mddev *mddev, bool interruptible);
828 extern void mddev_resume(struct mddev *mddev);
829 extern void md_idle_sync_thread(struct mddev *mddev);
830 extern void md_frozen_sync_thread(struct mddev *mddev);
831 extern void md_unfrozen_sync_thread(struct mddev *mddev);
832 
833 extern void md_reload_sb(struct mddev *mddev, int raid_disk);
834 extern void md_update_sb(struct mddev *mddev, int force);
835 extern void mddev_create_serial_pool(struct mddev *mddev, struct md_rdev *rdev);
836 extern void mddev_destroy_serial_pool(struct mddev *mddev,
837 				      struct md_rdev *rdev);
838 struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr);
839 struct md_rdev *md_find_rdev_rcu(struct mddev *mddev, dev_t dev);
840 
841 static inline bool is_rdev_broken(struct md_rdev *rdev)
842 {
843 	return !disk_live(rdev->bdev->bd_disk);
844 }
845 
846 static inline void rdev_dec_pending(struct md_rdev *rdev, struct mddev *mddev)
847 {
848 	int faulty = test_bit(Faulty, &rdev->flags);
849 	if (atomic_dec_and_test(&rdev->nr_pending) && faulty) {
850 		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
851 		md_wakeup_thread(mddev->thread);
852 	}
853 }
854 
855 extern struct md_cluster_operations *md_cluster_ops;
856 static inline int mddev_is_clustered(struct mddev *mddev)
857 {
858 	return mddev->cluster_info && mddev->bitmap_info.nodes > 1;
859 }
860 
861 /* clear unsupported mddev_flags */
862 static inline void mddev_clear_unsupported_flags(struct mddev *mddev,
863 	unsigned long unsupported_flags)
864 {
865 	mddev->flags &= ~unsupported_flags;
866 }
867 
868 static inline void mddev_check_write_zeroes(struct mddev *mddev, struct bio *bio)
869 {
870 	if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
871 	    !bio->bi_bdev->bd_disk->queue->limits.max_write_zeroes_sectors)
872 		mddev->gendisk->queue->limits.max_write_zeroes_sectors = 0;
873 }
874 
875 static inline int mddev_suspend_and_lock(struct mddev *mddev)
876 {
877 	int ret;
878 
879 	ret = mddev_suspend(mddev, true);
880 	if (ret)
881 		return ret;
882 
883 	ret = mddev_lock(mddev);
884 	if (ret)
885 		mddev_resume(mddev);
886 
887 	return ret;
888 }
889 
890 static inline void mddev_suspend_and_lock_nointr(struct mddev *mddev)
891 {
892 	mddev_suspend(mddev, false);
893 	mutex_lock(&mddev->reconfig_mutex);
894 }
895 
896 static inline void mddev_unlock_and_resume(struct mddev *mddev)
897 {
898 	mddev_unlock(mddev);
899 	mddev_resume(mddev);
900 }
901 
902 struct mdu_array_info_s;
903 struct mdu_disk_info_s;
904 
905 extern int mdp_major;
906 extern struct workqueue_struct *md_bitmap_wq;
907 void md_autostart_arrays(int part);
908 int md_set_array_info(struct mddev *mddev, struct mdu_array_info_s *info);
909 int md_add_new_disk(struct mddev *mddev, struct mdu_disk_info_s *info);
910 int do_md_run(struct mddev *mddev);
911 void mddev_stack_rdev_limits(struct mddev *mddev, struct queue_limits *lim);
912 int mddev_stack_new_rdev(struct mddev *mddev, struct md_rdev *rdev);
913 void mddev_update_io_opt(struct mddev *mddev, unsigned int nr_stripes);
914 
915 extern const struct block_device_operations md_fops;
916 
917 /*
918  * MD devices can be used undeneath by DM, in which case ->gendisk is NULL.
919  */
920 static inline bool mddev_is_dm(struct mddev *mddev)
921 {
922 	return !mddev->gendisk;
923 }
924 
925 static inline void mddev_trace_remap(struct mddev *mddev, struct bio *bio,
926 		sector_t sector)
927 {
928 	if (!mddev_is_dm(mddev))
929 		trace_block_bio_remap(bio, disk_devt(mddev->gendisk), sector);
930 }
931 
932 #define mddev_add_trace_msg(mddev, fmt, args...)			\
933 do {									\
934 	if (!mddev_is_dm(mddev))					\
935 		blk_add_trace_msg((mddev)->gendisk->queue, fmt, ##args); \
936 } while (0)
937 
938 #endif /* _MD_MD_H */
939