Linux Audio

Check our new training course

Linux BSP upgrade and security maintenance

Need help to get security updates for your Linux BSP?
Loading...
v3.15
 
  1/*
  2   md.h : kernel internal structure of the Linux MD driver
  3          Copyright (C) 1996-98 Ingo Molnar, Gadi Oxman
  4	  
  5   This program is free software; you can redistribute it and/or modify
  6   it under the terms of the GNU General Public License as published by
  7   the Free Software Foundation; either version 2, or (at your option)
  8   any later version.
  9   
 10   You should have received a copy of the GNU General Public License
 11   (for example /usr/src/linux/COPYING); if not, write to the Free
 12   Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.  
 13*/
 14
 15#ifndef _MD_MD_H
 16#define _MD_MD_H
 17
 18#include <linux/blkdev.h>
 
 
 19#include <linux/kobject.h>
 20#include <linux/list.h>
 21#include <linux/mm.h>
 22#include <linux/mutex.h>
 23#include <linux/timer.h>
 24#include <linux/wait.h>
 25#include <linux/workqueue.h>
 
 26
 27#define MaxSector (~(sector_t)0)
 28
 29/* Bad block numbers are stored sorted in a single page.
 30 * 64bits is used for each block or extent.
 31 * 54 bits are sector number, 9 bits are extent size,
 32 * 1 bit is an 'acknowledged' flag.
 
 
 
 
 33 */
 34#define MD_MAX_BADBLOCKS	(PAGE_SIZE/8)
 35
 36/*
 37 * MD's 'extended' device
 38 */
 39struct md_rdev {
 40	struct list_head same_set;	/* RAID devices within the same set */
 41
 42	sector_t sectors;		/* Device size (in 512bytes sectors) */
 43	struct mddev *mddev;		/* RAID array if running */
 44	int last_events;		/* IO event timestamp */
 45
 46	/*
 47	 * If meta_bdev is non-NULL, it means that a separate device is
 48	 * being used to store the metadata (superblock/bitmap) which
 49	 * would otherwise be contained on the same device as the data (bdev).
 50	 */
 51	struct block_device *meta_bdev;
 52	struct block_device *bdev;	/* block device handle */
 53
 54	struct page	*sb_page, *bb_page;
 55	int		sb_loaded;
 56	__u64		sb_events;
 57	sector_t	data_offset;	/* start of data in array */
 58	sector_t	new_data_offset;/* only relevant while reshaping */
 59	sector_t 	sb_start;	/* offset of the super block (in 512byte sectors) */
 60	int		sb_size;	/* bytes in the superblock */
 61	int		preferred_minor;	/* autorun support */
 62
 63	struct kobject	kobj;
 64
 65	/* A device can be in one of three states based on two flags:
 66	 * Not working:   faulty==1 in_sync==0
 67	 * Fully working: faulty==0 in_sync==1
 68	 * Working, but not
 69	 * in sync with array
 70	 *                faulty==0 in_sync==0
 71	 *
 72	 * It can never have faulty==1, in_sync==1
 73	 * This reduces the burden of testing multiple flags in many cases
 74	 */
 75
 76	unsigned long	flags;	/* bit set of 'enum flag_bits' bits. */
 77	wait_queue_head_t blocked_wait;
 78
 79	int desc_nr;			/* descriptor index in the superblock */
 80	int raid_disk;			/* role of device in array */
 81	int new_raid_disk;		/* role that the device will have in
 82					 * the array after a level-change completes.
 83					 */
 84	int saved_raid_disk;		/* role that device used to have in the
 85					 * array and could again if we did a partial
 86					 * resync from the bitmap
 87					 */
 88	sector_t	recovery_offset;/* If this device has been partially
 
 89					 * recovered, this is where we were
 90					 * up to.
 91					 */
 
 
 
 
 
 92
 93	atomic_t	nr_pending;	/* number of pending requests.
 94					 * only maintained for arrays that
 95					 * support hot removal
 96					 */
 97	atomic_t	read_errors;	/* number of consecutive read errors that
 98					 * we have tried to ignore.
 99					 */
100	struct timespec last_read_error;	/* monotonic time since our
101						 * last read error
102						 */
103	atomic_t	corrected_errors; /* number of corrected read errors,
104					   * for reporting to userspace and storing
105					   * in superblock.
106					   */
 
 
 
 
 
 
 
 
107	struct work_struct del_work;	/* used for delayed sysfs removal */
108
109	struct kernfs_node *sysfs_state; /* handle for 'state'
110					   * sysfs entry */
111
112	struct badblocks {
113		int	count;		/* count of bad blocks */
114		int	unacked_exist;	/* there probably are unacknowledged
115					 * bad blocks.  This is only cleared
116					 * when a read discovers none
117					 */
118		int	shift;		/* shift from sectors to block size
119					 * a -ve shift means badblocks are
120					 * disabled.*/
121		u64	*page;		/* badblock list */
122		int	changed;
123		seqlock_t lock;
124
125		sector_t sector;
126		sector_t size;		/* in sectors */
127	} badblocks;
128};
129enum flag_bits {
130	Faulty,			/* device is known to have a fault */
131	In_sync,		/* device is in_sync with rest of array */
132	Bitmap_sync,		/* ..actually, not quite In_sync.  Need a
133				 * bitmap-based recovery to get fully in sync
134				 */
135	Unmerged,		/* device is being added to array and should
136				 * be considerred for bvec_merge_fn but not
137				 * yet for actual IO
138				 */
139	WriteMostly,		/* Avoid reading if at all possible */
140	AutoDetected,		/* added by auto-detect */
141	Blocked,		/* An error occurred but has not yet
142				 * been acknowledged by the metadata
143				 * handler, so don't allow writes
144				 * until it is cleared */
145	WriteErrorSeen,		/* A write error has been seen on this
146				 * device
147				 */
148	FaultRecorded,		/* Intermediate state for clearing
149				 * Blocked.  The Fault is/will-be
150				 * recorded in the metadata, but that
151				 * metadata hasn't been stored safely
152				 * on disk yet.
153				 */
154	BlockedBadBlocks,	/* A writer is blocked because they
155				 * found an unacknowledged bad-block.
156				 * This can safely be cleared at any
157				 * time, and the writer will re-check.
158				 * It may be set at any time, and at
159				 * worst the writer will timeout and
160				 * re-check.  So setting it as
161				 * accurately as possible is good, but
162				 * not absolutely critical.
163				 */
164	WantReplacement,	/* This device is a candidate to be
165				 * hot-replaced, either because it has
166				 * reported some faults, or because
167				 * of explicit request.
168				 */
169	Replacement,		/* This device is a replacement for
170				 * a want_replacement device with same
171				 * raid_disk number.
172				 */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
173};
174
175#define BB_LEN_MASK	(0x00000000000001FFULL)
176#define BB_OFFSET_MASK	(0x7FFFFFFFFFFFFE00ULL)
177#define BB_ACK_MASK	(0x8000000000000000ULL)
178#define BB_MAX_LEN	512
179#define BB_OFFSET(x)	(((x) & BB_OFFSET_MASK) >> 9)
180#define BB_LEN(x)	(((x) & BB_LEN_MASK) + 1)
181#define BB_ACK(x)	(!!((x) & BB_ACK_MASK))
182#define BB_MAKE(a, l, ack) (((a)<<9) | ((l)-1) | ((u64)(!!(ack)) << 63))
183
184extern int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
185			  sector_t *first_bad, int *bad_sectors);
186static inline int is_badblock(struct md_rdev *rdev, sector_t s, int sectors,
187			      sector_t *first_bad, int *bad_sectors)
188{
189	if (unlikely(rdev->badblocks.count)) {
190		int rv = md_is_badblock(&rdev->badblocks, rdev->data_offset + s,
191					sectors,
192					first_bad, bad_sectors);
193		if (rv)
194			*first_bad -= rdev->data_offset;
195		return rv;
196	}
197	return 0;
198}
199extern int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
200			      int is_new);
201extern int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
202				int is_new);
203extern void md_ack_all_badblocks(struct badblocks *bb);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
204
205struct mddev {
206	void				*private;
207	struct md_personality		*pers;
208	dev_t				unit;
209	int				md_minor;
210	struct list_head		disks;
211	unsigned long			flags;
212#define MD_CHANGE_DEVS	0	/* Some device status has changed */
213#define MD_CHANGE_CLEAN 1	/* transition to or from 'clean' */
214#define MD_CHANGE_PENDING 2	/* switch from 'clean' to 'active' in progress */
215#define MD_UPDATE_SB_FLAGS (1 | 2 | 4)	/* If these are set, md_update_sb needed */
216#define MD_ARRAY_FIRST_USE 3    /* First use of array, needs initialization */
217#define MD_STILL_CLOSED	4	/* If set, then array has not been opened since
218				 * md_ioctl checked on it.
219				 */
220
221	int				suspended;
222	atomic_t			active_io;
223	int				ro;
224	int				sysfs_active; /* set when sysfs deletes
225						       * are happening, so run/
226						       * takeover/stop are not safe
227						       */
228	int				ready; /* See when safe to pass
229						* IO requests down */
230	struct gendisk			*gendisk;
231
232	struct kobject			kobj;
233	int				hold_active;
234#define	UNTIL_IOCTL	1
235#define	UNTIL_STOP	2
236
237	/* Superblock information */
238	int				major_version,
239					minor_version,
240					patch_version;
241	int				persistent;
242	int 				external;	/* metadata is
243							 * managed externally */
244	char				metadata_type[17]; /* externally set*/
245	int				chunk_sectors;
246	time_t				ctime, utime;
247	int				level, layout;
248	char				clevel[16];
249	int				raid_disks;
250	int				max_disks;
251	sector_t			dev_sectors; 	/* used size of
252							 * component devices */
253	sector_t			array_sectors; /* exported array size */
254	int				external_size; /* size managed
255							* externally */
256	__u64				events;
257	/* If the last 'event' was simply a clean->dirty transition, and
258	 * we didn't write it to the spares, then it is safe and simple
259	 * to just decrement the event count on a dirty->clean transition.
260	 * So we record that possibility here.
261	 */
262	int				can_decrease_events;
263
264	char				uuid[16];
265
266	/* If the array is being reshaped, we need to record the
267	 * new shape and an indication of where we are up to.
268	 * This is written to the superblock.
269	 * If reshape_position is MaxSector, then no reshape is happening (yet).
270	 */
271	sector_t			reshape_position;
272	int				delta_disks, new_level, new_layout;
273	int				new_chunk_sectors;
274	int				reshape_backwards;
275
276	struct md_thread		*thread;	/* management thread */
277	struct md_thread		*sync_thread;	/* doing resync or reconstruct */
278
279	/* 'last_sync_action' is initialized to "none".  It is set when a
280	 * sync operation (i.e "data-check", "requested-resync", "resync",
281	 * "recovery", or "reshape") is started.  It holds this value even
282	 * when the sync thread is "frozen" (interrupted) or "idle" (stopped
283	 * or finished).  It is overwritten when a new sync operation is begun.
284	 */
285	char				*last_sync_action;
286	sector_t			curr_resync;	/* last block scheduled */
287	/* As resync requests can complete out of order, we cannot easily track
288	 * how much resync has been completed.  So we occasionally pause until
289	 * everything completes, then set curr_resync_completed to curr_resync.
290	 * As such it may be well behind the real resync mark, but it is a value
291	 * we are certain of.
292	 */
293	sector_t			curr_resync_completed;
294	unsigned long			resync_mark;	/* a recent timestamp */
295	sector_t			resync_mark_cnt;/* blocks written at resync_mark */
296	sector_t			curr_mark_cnt; /* blocks scheduled now */
297
298	sector_t			resync_max_sectors; /* may be set by personality */
299
300	atomic64_t			resync_mismatches; /* count of sectors where
301							    * parity/replica mismatch found
302							    */
303
304	/* allow user-space to request suspension of IO to regions of the array */
305	sector_t			suspend_lo;
306	sector_t			suspend_hi;
307	/* if zero, use the system-wide default */
308	int				sync_speed_min;
309	int				sync_speed_max;
310
311	/* resync even though the same disks are shared among md-devices */
312	int				parallel_resync;
313
314	int				ok_start_degraded;
315	/* recovery/resync flags 
316	 * NEEDED:   we might need to start a resync/recover
317	 * RUNNING:  a thread is running, or about to be started
318	 * SYNC:     actually doing a resync, not a recovery
319	 * RECOVER:  doing recovery, or need to try it.
320	 * INTR:     resync needs to be aborted for some reason
321	 * DONE:     thread is done and is waiting to be reaped
322	 * REQUEST:  user-space has requested a sync (used with SYNC)
323	 * CHECK:    user-space request for check-only, no repair
324	 * RESHAPE:  A reshape is happening
325	 * ERROR:    sync-action interrupted because io-error
326	 *
327	 * If neither SYNC or RESHAPE are set, then it is a recovery.
328	 */
329#define	MD_RECOVERY_RUNNING	0
330#define	MD_RECOVERY_SYNC	1
331#define	MD_RECOVERY_RECOVER	2
332#define	MD_RECOVERY_INTR	3
333#define	MD_RECOVERY_DONE	4
334#define	MD_RECOVERY_NEEDED	5
335#define	MD_RECOVERY_REQUESTED	6
336#define	MD_RECOVERY_CHECK	7
337#define MD_RECOVERY_RESHAPE	8
338#define	MD_RECOVERY_FROZEN	9
339#define	MD_RECOVERY_ERROR	10
340
341	unsigned long			recovery;
342	/* If a RAID personality determines that recovery (of a particular
343	 * device) will fail due to a read error on the source device, it
344	 * takes a copy of this number and does not attempt recovery again
345	 * until this number changes.
346	 */
347	int				recovery_disabled;
348
349	int				in_sync;	/* know to not need resync */
350	/* 'open_mutex' avoids races between 'md_open' and 'do_md_stop', so
351	 * that we are never stopping an array while it is open.
352	 * 'reconfig_mutex' protects all other reconfiguration.
353	 * These locks are separate due to conflicting interactions
354	 * with bdev->bd_mutex.
355	 * Lock ordering is:
356	 *  reconfig_mutex -> bd_mutex : e.g. do_md_run -> revalidate_disk
357	 *  bd_mutex -> open_mutex:  e.g. __blkdev_get -> md_open
358	 */
359	struct mutex			open_mutex;
360	struct mutex			reconfig_mutex;
361	atomic_t			active;		/* general refcount */
362	atomic_t			openers;	/* number of active opens */
363
364	int				changed;	/* True if we might need to
365							 * reread partition info */
366	int				degraded;	/* whether md should consider
367							 * adding a spare
368							 */
369	int				merge_check_needed; /* at least one
370							     * member device
371							     * has a
372							     * merge_bvec_fn */
373
374	atomic_t			recovery_active; /* blocks scheduled, but not written */
375	wait_queue_head_t		recovery_wait;
376	sector_t			recovery_cp;
377	sector_t			resync_min;	/* user requested sync
378							 * starts here */
379	sector_t			resync_max;	/* resync should pause
380							 * when it gets here */
381
382	struct kernfs_node		*sysfs_state;	/* handle for 'array_state'
383							 * file in sysfs.
384							 */
385	struct kernfs_node		*sysfs_action;  /* handle for 'sync_action' */
386
387	struct work_struct del_work;	/* used for delayed sysfs removal */
388
389	spinlock_t			write_lock;
 
 
 
 
 
 
 
 
 
 
 
390	wait_queue_head_t		sb_wait;	/* for waiting on superblock updates */
391	atomic_t			pending_writes;	/* number of active superblock writes */
392
393	unsigned int			safemode;	/* if set, update "clean" superblock
394							 * when no writes pending.
395							 */ 
396	unsigned int			safemode_delay;
397	struct timer_list		safemode_timer;
398	atomic_t			writes_pending; 
 
399	struct request_queue		*queue;	/* for plugging ... */
400
401	struct bitmap                   *bitmap; /* the bitmap for the device */
402	struct {
403		struct file		*file; /* the bitmap file */
404		loff_t			offset; /* offset from superblock of
405						 * start of bitmap. May be
406						 * negative, but not '0'
407						 * For external metadata, offset
408						 * from start of device. 
409						 */
410		unsigned long		space; /* space available at this offset */
411		loff_t			default_offset; /* this is the offset to use when
412							 * hot-adding a bitmap.  It should
413							 * eventually be settable by sysfs.
414							 */
415		unsigned long		default_space; /* space available at
416							* default offset */
417		struct mutex		mutex;
418		unsigned long		chunksize;
419		unsigned long		daemon_sleep; /* how many jiffies between updates? */
420		unsigned long		max_write_behind; /* write-behind mode */
421		int			external;
 
 
422	} bitmap_info;
423
424	atomic_t 			max_corr_read_errors; /* max read retries */
425	struct list_head		all_mddevs;
426
427	struct attribute_group		*to_remove;
428
429	struct bio_set			*bio_set;
 
 
 
430
431	/* Generic flush handling.
432	 * The last to finish preflush schedules a worker to submit
433	 * the rest of the request (without the REQ_FLUSH flag).
434	 */
435	struct bio *flush_bio;
436	atomic_t flush_pending;
 
 
 
437	struct work_struct flush_work;
438	struct work_struct event_work;	/* used by dm to report failure event */
 
439	void (*sync_super)(struct mddev *mddev, struct md_rdev *rdev);
 
 
 
 
 
440};
441
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
442
443static inline void rdev_dec_pending(struct md_rdev *rdev, struct mddev *mddev)
444{
445	int faulty = test_bit(Faulty, &rdev->flags);
446	if (atomic_dec_and_test(&rdev->nr_pending) && faulty)
447		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
 
 
 
 
 
 
448}
449
 
 
 
 
 
 
450static inline void md_sync_acct(struct block_device *bdev, unsigned long nr_sectors)
451{
452        atomic_add(nr_sectors, &bdev->bd_contains->bd_disk->sync_io);
 
 
 
 
 
453}
454
455struct md_personality
456{
457	char *name;
458	int level;
459	struct list_head list;
460	struct module *owner;
461	void (*make_request)(struct mddev *mddev, struct bio *bio);
 
 
 
 
462	int (*run)(struct mddev *mddev);
463	int (*stop)(struct mddev *mddev);
 
 
464	void (*status)(struct seq_file *seq, struct mddev *mddev);
465	/* error_handler must set ->faulty and clear ->in_sync
466	 * if appropriate, and should abort recovery if needed 
467	 */
468	void (*error_handler)(struct mddev *mddev, struct md_rdev *rdev);
469	int (*hot_add_disk) (struct mddev *mddev, struct md_rdev *rdev);
470	int (*hot_remove_disk) (struct mddev *mddev, struct md_rdev *rdev);
471	int (*spare_active) (struct mddev *mddev);
472	sector_t (*sync_request)(struct mddev *mddev, sector_t sector_nr, int *skipped, int go_faster);
473	int (*resize) (struct mddev *mddev, sector_t sectors);
474	sector_t (*size) (struct mddev *mddev, sector_t sectors, int raid_disks);
475	int (*check_reshape) (struct mddev *mddev);
476	int (*start_reshape) (struct mddev *mddev);
477	void (*finish_reshape) (struct mddev *mddev);
478	/* quiesce moves between quiescence states
479	 * 0 - fully active
480	 * 1 - no new requests allowed
481	 * others - reserved
482	 */
483	void (*quiesce) (struct mddev *mddev, int state);
484	/* takeover is used to transition an array from one
485	 * personality to another.  The new personality must be able
486	 * to handle the data in the current layout.
487	 * e.g. 2drive raid1 -> 2drive raid5
488	 *      ndrive raid5 -> degraded n+1drive raid6 with special layout
489	 * If the takeover succeeds, a new 'private' structure is returned.
490	 * This needs to be installed and then ->run used to activate the
491	 * array.
492	 */
493	void *(*takeover) (struct mddev *mddev);
 
 
 
 
 
494};
495
496
497struct md_sysfs_entry {
498	struct attribute attr;
499	ssize_t (*show)(struct mddev *, char *);
500	ssize_t (*store)(struct mddev *, const char *, size_t);
501};
502extern struct attribute_group md_bitmap_group;
503
504static inline struct kernfs_node *sysfs_get_dirent_safe(struct kernfs_node *sd, char *name)
505{
506	if (sd)
507		return sysfs_get_dirent(sd, name);
508	return sd;
509}
510static inline void sysfs_notify_dirent_safe(struct kernfs_node *sd)
511{
512	if (sd)
513		sysfs_notify_dirent(sd);
514}
515
516static inline char * mdname (struct mddev * mddev)
517{
518	return mddev->gendisk ? mddev->gendisk->disk_name : "mdX";
519}
520
521static inline int sysfs_link_rdev(struct mddev *mddev, struct md_rdev *rdev)
522{
523	char nm[20];
524	if (!test_bit(Replacement, &rdev->flags) && mddev->kobj.sd) {
 
 
525		sprintf(nm, "rd%d", rdev->raid_disk);
526		return sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
527	} else
528		return 0;
529}
530
531static inline void sysfs_unlink_rdev(struct mddev *mddev, struct md_rdev *rdev)
532{
533	char nm[20];
534	if (!test_bit(Replacement, &rdev->flags) && mddev->kobj.sd) {
 
 
535		sprintf(nm, "rd%d", rdev->raid_disk);
536		sysfs_remove_link(&mddev->kobj, nm);
537	}
538}
539
540/*
541 * iterates through some rdev ringlist. It's safe to remove the
542 * current 'rdev'. Dont touch 'tmp' though.
543 */
544#define rdev_for_each_list(rdev, tmp, head)				\
545	list_for_each_entry_safe(rdev, tmp, head, same_set)
546
547/*
548 * iterates through the 'same array disks' ringlist
549 */
550#define rdev_for_each(rdev, mddev)				\
551	list_for_each_entry(rdev, &((mddev)->disks), same_set)
552
553#define rdev_for_each_safe(rdev, tmp, mddev)				\
554	list_for_each_entry_safe(rdev, tmp, &((mddev)->disks), same_set)
555
556#define rdev_for_each_rcu(rdev, mddev)				\
557	list_for_each_entry_rcu(rdev, &((mddev)->disks), same_set)
558
559struct md_thread {
560	void			(*run) (struct md_thread *thread);
561	struct mddev		*mddev;
562	wait_queue_head_t	wqueue;
563	unsigned long           flags;
564	struct task_struct	*tsk;
565	unsigned long		timeout;
566	void			*private;
567};
568
569#define THREAD_WAKEUP  0
570
571static inline void safe_put_page(struct page *p)
572{
573	if (p) put_page(p);
574}
575
576extern int register_md_personality(struct md_personality *p);
577extern int unregister_md_personality(struct md_personality *p);
 
 
 
 
 
578extern struct md_thread *md_register_thread(
579	void (*run)(struct md_thread *thread),
580	struct mddev *mddev,
581	const char *name);
582extern void md_unregister_thread(struct md_thread **threadp);
583extern void md_wakeup_thread(struct md_thread *thread);
584extern void md_check_recovery(struct mddev *mddev);
585extern void md_reap_sync_thread(struct mddev *mddev);
586extern void md_write_start(struct mddev *mddev, struct bio *bi);
 
 
587extern void md_write_end(struct mddev *mddev);
588extern void md_done_sync(struct mddev *mddev, int blocks, int ok);
589extern void md_error(struct mddev *mddev, struct md_rdev *rdev);
590extern void md_finish_reshape(struct mddev *mddev);
591
592extern int mddev_congested(struct mddev *mddev, int bits);
593extern void md_flush_request(struct mddev *mddev, struct bio *bio);
594extern void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
595			   sector_t sector, int size, struct page *page);
596extern void md_super_wait(struct mddev *mddev);
597extern int sync_page_io(struct md_rdev *rdev, sector_t sector, int size, 
598			struct page *page, int rw, bool metadata_op);
 
599extern void md_do_sync(struct md_thread *thread);
600extern void md_new_event(struct mddev *mddev);
601extern int md_allow_write(struct mddev *mddev);
602extern void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev);
603extern void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors);
604extern int md_check_no_bitmap(struct mddev *mddev);
605extern int md_integrity_register(struct mddev *mddev);
606extern void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev);
607extern int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale);
608
609extern void mddev_init(struct mddev *mddev);
610extern int md_run(struct mddev *mddev);
 
611extern void md_stop(struct mddev *mddev);
612extern void md_stop_writes(struct mddev *mddev);
613extern int md_rdev_init(struct md_rdev *rdev);
614extern void md_rdev_clear(struct md_rdev *rdev);
615
 
616extern void mddev_suspend(struct mddev *mddev);
617extern void mddev_resume(struct mddev *mddev);
618extern struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
619				   struct mddev *mddev);
620extern struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
621				   struct mddev *mddev);
622
623extern void md_unplug(struct blk_plug_cb *cb, bool from_schedule);
624static inline int mddev_check_plugged(struct mddev *mddev)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
625{
626	return !!blk_check_plugged(md_unplug, mddev,
627				   sizeof(struct blk_plug_cb));
 
628}
629#endif /* _MD_MD_H */
v5.4
  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 "md-cluster.h"
 22
 23#define MaxSector (~(sector_t)0)
 24
 25/*
 26 * These flags should really be called "NO_RETRY" rather than
 27 * "FAILFAST" because they don't make any promise about time lapse,
 28 * only about the number of retries, which will be zero.
 29 * REQ_FAILFAST_DRIVER is not included because
 30 * Commit: 4a27446f3e39 ("[SCSI] modify scsi to handle new fail fast flags.")
 31 * seems to suggest that the errors it avoids retrying should usually
 32 * be retried.
 33 */
 34#define	MD_FAILFAST	(REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT)
 
 35/*
 36 * MD's 'extended' device
 37 */
 38struct md_rdev {
 39	struct list_head same_set;	/* RAID devices within the same set */
 40
 41	sector_t sectors;		/* Device size (in 512bytes sectors) */
 42	struct mddev *mddev;		/* RAID array if running */
 43	int last_events;		/* IO event timestamp */
 44
 45	/*
 46	 * If meta_bdev is non-NULL, it means that a separate device is
 47	 * being used to store the metadata (superblock/bitmap) which
 48	 * would otherwise be contained on the same device as the data (bdev).
 49	 */
 50	struct block_device *meta_bdev;
 51	struct block_device *bdev;	/* block device handle */
 52
 53	struct page	*sb_page, *bb_page;
 54	int		sb_loaded;
 55	__u64		sb_events;
 56	sector_t	data_offset;	/* start of data in array */
 57	sector_t	new_data_offset;/* only relevant while reshaping */
 58	sector_t	sb_start;	/* offset of the super block (in 512byte sectors) */
 59	int		sb_size;	/* bytes in the superblock */
 60	int		preferred_minor;	/* autorun support */
 61
 62	struct kobject	kobj;
 63
 64	/* A device can be in one of three states based on two flags:
 65	 * Not working:   faulty==1 in_sync==0
 66	 * Fully working: faulty==0 in_sync==1
 67	 * Working, but not
 68	 * in sync with array
 69	 *                faulty==0 in_sync==0
 70	 *
 71	 * It can never have faulty==1, in_sync==1
 72	 * This reduces the burden of testing multiple flags in many cases
 73	 */
 74
 75	unsigned long	flags;	/* bit set of 'enum flag_bits' bits. */
 76	wait_queue_head_t blocked_wait;
 77
 78	int desc_nr;			/* descriptor index in the superblock */
 79	int raid_disk;			/* role of device in array */
 80	int new_raid_disk;		/* role that the device will have in
 81					 * the array after a level-change completes.
 82					 */
 83	int saved_raid_disk;		/* role that device used to have in the
 84					 * array and could again if we did a partial
 85					 * resync from the bitmap
 86					 */
 87	union {
 88		sector_t recovery_offset;/* If this device has been partially
 89					 * recovered, this is where we were
 90					 * up to.
 91					 */
 92		sector_t journal_tail;	/* If this device is a journal device,
 93					 * this is the journal tail (journal
 94					 * recovery start point)
 95					 */
 96	};
 97
 98	atomic_t	nr_pending;	/* number of pending requests.
 99					 * only maintained for arrays that
100					 * support hot removal
101					 */
102	atomic_t	read_errors;	/* number of consecutive read errors that
103					 * we have tried to ignore.
104					 */
105	time64_t	last_read_error;	/* monotonic time since our
106						 * last read error
107						 */
108	atomic_t	corrected_errors; /* number of corrected read errors,
109					   * for reporting to userspace and storing
110					   * in superblock.
111					   */
112
113	/*
114	 * The members for check collision of write behind IOs.
115	 */
116	struct list_head wb_list;
117	spinlock_t wb_list_lock;
118	wait_queue_head_t wb_io_wait;
119
120	struct work_struct del_work;	/* used for delayed sysfs removal */
121
122	struct kernfs_node *sysfs_state; /* handle for 'state'
123					   * sysfs entry */
124
125	struct badblocks badblocks;
126
127	struct {
128		short offset;	/* Offset from superblock to start of PPL.
129				 * Not used by external metadata. */
130		unsigned int size;	/* Size in sectors of the PPL space */
131		sector_t sector;	/* First sector of the PPL space */
132	} ppl;
 
 
 
 
 
 
 
 
133};
134enum flag_bits {
135	Faulty,			/* device is known to have a fault */
136	In_sync,		/* device is in_sync with rest of array */
137	Bitmap_sync,		/* ..actually, not quite In_sync.  Need a
138				 * bitmap-based recovery to get fully in sync.
139				 * The bit is only meaningful before device
140				 * has been passed to pers->hot_add_disk.
 
 
141				 */
142	WriteMostly,		/* Avoid reading if at all possible */
143	AutoDetected,		/* added by auto-detect */
144	Blocked,		/* An error occurred but has not yet
145				 * been acknowledged by the metadata
146				 * handler, so don't allow writes
147				 * until it is cleared */
148	WriteErrorSeen,		/* A write error has been seen on this
149				 * device
150				 */
151	FaultRecorded,		/* Intermediate state for clearing
152				 * Blocked.  The Fault is/will-be
153				 * recorded in the metadata, but that
154				 * metadata hasn't been stored safely
155				 * on disk yet.
156				 */
157	BlockedBadBlocks,	/* A writer is blocked because they
158				 * found an unacknowledged bad-block.
159				 * This can safely be cleared at any
160				 * time, and the writer will re-check.
161				 * It may be set at any time, and at
162				 * worst the writer will timeout and
163				 * re-check.  So setting it as
164				 * accurately as possible is good, but
165				 * not absolutely critical.
166				 */
167	WantReplacement,	/* This device is a candidate to be
168				 * hot-replaced, either because it has
169				 * reported some faults, or because
170				 * of explicit request.
171				 */
172	Replacement,		/* This device is a replacement for
173				 * a want_replacement device with same
174				 * raid_disk number.
175				 */
176	Candidate,		/* For clustered environments only:
177				 * This device is seen locally but not
178				 * by the whole cluster
179				 */
180	Journal,		/* This device is used as journal for
181				 * raid-5/6.
182				 * Usually, this device should be faster
183				 * than other devices in the array
184				 */
185	ClusterRemove,
186	RemoveSynchronized,	/* synchronize_rcu() was called after
187				 * this device was known to be faulty,
188				 * so it is safe to remove without
189				 * another synchronize_rcu() call.
190				 */
191	ExternalBbl,            /* External metadata provides bad
192				 * block management for a disk
193				 */
194	FailFast,		/* Minimal retries should be attempted on
195				 * this device, so use REQ_FAILFAST_DEV.
196				 * Also don't try to repair failed reads.
197				 * It is expects that no bad block log
198				 * is present.
199				 */
200	LastDev,		/* Seems to be the last working dev as
201				 * it didn't fail, so don't use FailFast
202				 * any more for metadata
203				 */
204	WBCollisionCheck,	/*
205				 * multiqueue device should check if there
206				 * is collision between write behind bios.
207				 */
208};
209
 
 
 
 
 
 
 
 
 
 
 
210static inline int is_badblock(struct md_rdev *rdev, sector_t s, int sectors,
211			      sector_t *first_bad, int *bad_sectors)
212{
213	if (unlikely(rdev->badblocks.count)) {
214		int rv = badblocks_check(&rdev->badblocks, rdev->data_offset + s,
215					sectors,
216					first_bad, bad_sectors);
217		if (rv)
218			*first_bad -= rdev->data_offset;
219		return rv;
220	}
221	return 0;
222}
223extern int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
224			      int is_new);
225extern int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
226				int is_new);
227struct md_cluster_info;
228
229/* change UNSUPPORTED_MDDEV_FLAGS for each array type if new flag is added */
230enum mddev_flags {
231	MD_ARRAY_FIRST_USE,	/* First use of array, needs initialization */
232	MD_CLOSING,		/* If set, we are closing the array, do not open
233				 * it then */
234	MD_JOURNAL_CLEAN,	/* A raid with journal is already clean */
235	MD_HAS_JOURNAL,		/* The raid array has journal feature set */
236	MD_CLUSTER_RESYNC_LOCKED, /* cluster raid only, which means node
237				   * already took resync lock, need to
238				   * release the lock */
239	MD_FAILFAST_SUPPORTED,	/* Using MD_FAILFAST on metadata writes is
240				 * supported as calls to md_error() will
241				 * never cause the array to become failed.
242				 */
243	MD_HAS_PPL,		/* The raid array has PPL feature set */
244	MD_HAS_MULTIPLE_PPLS,	/* The raid array has multiple PPLs feature set */
245	MD_ALLOW_SB_UPDATE,	/* md_check_recovery is allowed to update
246				 * the metadata without taking reconfig_mutex.
247				 */
248	MD_UPDATING_SB,		/* md_check_recovery is updating the metadata
249				 * without explicitly holding reconfig_mutex.
250				 */
251	MD_NOT_READY,		/* do_md_run() is active, so 'array_state'
252				 * must not report that array is ready yet
253				 */
254	MD_BROKEN,              /* This is used in RAID-0/LINEAR only, to stop
255				 * I/O in case an array member is gone/failed.
256				 */
257};
258
259enum mddev_sb_flags {
260	MD_SB_CHANGE_DEVS,		/* Some device status has changed */
261	MD_SB_CHANGE_CLEAN,	/* transition to or from 'clean' */
262	MD_SB_CHANGE_PENDING,	/* switch from 'clean' to 'active' in progress */
263	MD_SB_NEED_REWRITE,	/* metadata write needs to be repeated */
264};
265
266#define NR_WB_INFOS	8
267/* record current range of write behind IOs */
268struct wb_info {
269	sector_t lo;
270	sector_t hi;
271	struct list_head list;
272};
273
274struct mddev {
275	void				*private;
276	struct md_personality		*pers;
277	dev_t				unit;
278	int				md_minor;
279	struct list_head		disks;
280	unsigned long			flags;
281	unsigned long			sb_flags;
 
 
 
 
 
 
 
282
283	int				suspended;
284	atomic_t			active_io;
285	int				ro;
286	int				sysfs_active; /* set when sysfs deletes
287						       * are happening, so run/
288						       * takeover/stop are not safe
289						       */
 
 
290	struct gendisk			*gendisk;
291
292	struct kobject			kobj;
293	int				hold_active;
294#define	UNTIL_IOCTL	1
295#define	UNTIL_STOP	2
296
297	/* Superblock information */
298	int				major_version,
299					minor_version,
300					patch_version;
301	int				persistent;
302	int				external;	/* metadata is
303							 * managed externally */
304	char				metadata_type[17]; /* externally set*/
305	int				chunk_sectors;
306	time64_t			ctime, utime;
307	int				level, layout;
308	char				clevel[16];
309	int				raid_disks;
310	int				max_disks;
311	sector_t			dev_sectors;	/* used size of
312							 * component devices */
313	sector_t			array_sectors; /* exported array size */
314	int				external_size; /* size managed
315							* externally */
316	__u64				events;
317	/* If the last 'event' was simply a clean->dirty transition, and
318	 * we didn't write it to the spares, then it is safe and simple
319	 * to just decrement the event count on a dirty->clean transition.
320	 * So we record that possibility here.
321	 */
322	int				can_decrease_events;
323
324	char				uuid[16];
325
326	/* If the array is being reshaped, we need to record the
327	 * new shape and an indication of where we are up to.
328	 * This is written to the superblock.
329	 * If reshape_position is MaxSector, then no reshape is happening (yet).
330	 */
331	sector_t			reshape_position;
332	int				delta_disks, new_level, new_layout;
333	int				new_chunk_sectors;
334	int				reshape_backwards;
335
336	struct md_thread		*thread;	/* management thread */
337	struct md_thread		*sync_thread;	/* doing resync or reconstruct */
338
339	/* 'last_sync_action' is initialized to "none".  It is set when a
340	 * sync operation (i.e "data-check", "requested-resync", "resync",
341	 * "recovery", or "reshape") is started.  It holds this value even
342	 * when the sync thread is "frozen" (interrupted) or "idle" (stopped
343	 * or finished).  It is overwritten when a new sync operation is begun.
344	 */
345	char				*last_sync_action;
346	sector_t			curr_resync;	/* last block scheduled */
347	/* As resync requests can complete out of order, we cannot easily track
348	 * how much resync has been completed.  So we occasionally pause until
349	 * everything completes, then set curr_resync_completed to curr_resync.
350	 * As such it may be well behind the real resync mark, but it is a value
351	 * we are certain of.
352	 */
353	sector_t			curr_resync_completed;
354	unsigned long			resync_mark;	/* a recent timestamp */
355	sector_t			resync_mark_cnt;/* blocks written at resync_mark */
356	sector_t			curr_mark_cnt; /* blocks scheduled now */
357
358	sector_t			resync_max_sectors; /* may be set by personality */
359
360	atomic64_t			resync_mismatches; /* count of sectors where
361							    * parity/replica mismatch found
362							    */
363
364	/* allow user-space to request suspension of IO to regions of the array */
365	sector_t			suspend_lo;
366	sector_t			suspend_hi;
367	/* if zero, use the system-wide default */
368	int				sync_speed_min;
369	int				sync_speed_max;
370
371	/* resync even though the same disks are shared among md-devices */
372	int				parallel_resync;
373
374	int				ok_start_degraded;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
375
376	unsigned long			recovery;
377	/* If a RAID personality determines that recovery (of a particular
378	 * device) will fail due to a read error on the source device, it
379	 * takes a copy of this number and does not attempt recovery again
380	 * until this number changes.
381	 */
382	int				recovery_disabled;
383
384	int				in_sync;	/* know to not need resync */
385	/* 'open_mutex' avoids races between 'md_open' and 'do_md_stop', so
386	 * that we are never stopping an array while it is open.
387	 * 'reconfig_mutex' protects all other reconfiguration.
388	 * These locks are separate due to conflicting interactions
389	 * with bdev->bd_mutex.
390	 * Lock ordering is:
391	 *  reconfig_mutex -> bd_mutex : e.g. do_md_run -> revalidate_disk
392	 *  bd_mutex -> open_mutex:  e.g. __blkdev_get -> md_open
393	 */
394	struct mutex			open_mutex;
395	struct mutex			reconfig_mutex;
396	atomic_t			active;		/* general refcount */
397	atomic_t			openers;	/* number of active opens */
398
399	int				changed;	/* True if we might need to
400							 * reread partition info */
401	int				degraded;	/* whether md should consider
402							 * adding a spare
403							 */
 
 
 
 
404
405	atomic_t			recovery_active; /* blocks scheduled, but not written */
406	wait_queue_head_t		recovery_wait;
407	sector_t			recovery_cp;
408	sector_t			resync_min;	/* user requested sync
409							 * starts here */
410	sector_t			resync_max;	/* resync should pause
411							 * when it gets here */
412
413	struct kernfs_node		*sysfs_state;	/* handle for 'array_state'
414							 * file in sysfs.
415							 */
416	struct kernfs_node		*sysfs_action;  /* handle for 'sync_action' */
417
418	struct work_struct del_work;	/* used for delayed sysfs removal */
419
420	/* "lock" protects:
421	 *   flush_bio transition from NULL to !NULL
422	 *   rdev superblocks, events
423	 *   clearing MD_CHANGE_*
424	 *   in_sync - and related safemode and MD_CHANGE changes
425	 *   pers (also protected by reconfig_mutex and pending IO).
426	 *   clearing ->bitmap
427	 *   clearing ->bitmap_info.file
428	 *   changing ->resync_{min,max}
429	 *   setting MD_RECOVERY_RUNNING (which interacts with resync_{min,max})
430	 */
431	spinlock_t			lock;
432	wait_queue_head_t		sb_wait;	/* for waiting on superblock updates */
433	atomic_t			pending_writes;	/* number of active superblock writes */
434
435	unsigned int			safemode;	/* if set, update "clean" superblock
436							 * when no writes pending.
437							 */
438	unsigned int			safemode_delay;
439	struct timer_list		safemode_timer;
440	struct percpu_ref		writes_pending;
441	int				sync_checkers;	/* # of threads checking writes_pending */
442	struct request_queue		*queue;	/* for plugging ... */
443
444	struct bitmap			*bitmap; /* the bitmap for the device */
445	struct {
446		struct file		*file; /* the bitmap file */
447		loff_t			offset; /* offset from superblock of
448						 * start of bitmap. May be
449						 * negative, but not '0'
450						 * For external metadata, offset
451						 * from start of device.
452						 */
453		unsigned long		space; /* space available at this offset */
454		loff_t			default_offset; /* this is the offset to use when
455							 * hot-adding a bitmap.  It should
456							 * eventually be settable by sysfs.
457							 */
458		unsigned long		default_space; /* space available at
459							* default offset */
460		struct mutex		mutex;
461		unsigned long		chunksize;
462		unsigned long		daemon_sleep; /* how many jiffies between updates? */
463		unsigned long		max_write_behind; /* write-behind mode */
464		int			external;
465		int			nodes; /* Maximum number of nodes in the cluster */
466		char                    cluster_name[64]; /* Name of the cluster */
467	} bitmap_info;
468
469	atomic_t			max_corr_read_errors; /* max read retries */
470	struct list_head		all_mddevs;
471
472	struct attribute_group		*to_remove;
473
474	struct bio_set			bio_set;
475	struct bio_set			sync_set; /* for sync operations like
476						   * metadata and bitmap writes
477						   */
478
479	/* Generic flush handling.
480	 * The last to finish preflush schedules a worker to submit
481	 * the rest of the request (without the REQ_PREFLUSH flag).
482	 */
483	struct bio *flush_bio;
484	atomic_t flush_pending;
485	ktime_t start_flush, last_flush; /* last_flush is when the last completed
486					  * flush was started.
487					  */
488	struct work_struct flush_work;
489	struct work_struct event_work;	/* used by dm to report failure event */
490	mempool_t *wb_info_pool;
491	void (*sync_super)(struct mddev *mddev, struct md_rdev *rdev);
492	struct md_cluster_info		*cluster_info;
493	unsigned int			good_device_nr;	/* good device num within cluster raid */
494
495	bool	has_superblocks:1;
496	bool	fail_last_dev:1;
497};
498
499enum recovery_flags {
500	/*
501	 * If neither SYNC or RESHAPE are set, then it is a recovery.
502	 */
503	MD_RECOVERY_RUNNING,	/* a thread is running, or about to be started */
504	MD_RECOVERY_SYNC,	/* actually doing a resync, not a recovery */
505	MD_RECOVERY_RECOVER,	/* doing recovery, or need to try it. */
506	MD_RECOVERY_INTR,	/* resync needs to be aborted for some reason */
507	MD_RECOVERY_DONE,	/* thread is done and is waiting to be reaped */
508	MD_RECOVERY_NEEDED,	/* we might need to start a resync/recover */
509	MD_RECOVERY_REQUESTED,	/* user-space has requested a sync (used with SYNC) */
510	MD_RECOVERY_CHECK,	/* user-space request for check-only, no repair */
511	MD_RECOVERY_RESHAPE,	/* A reshape is happening */
512	MD_RECOVERY_FROZEN,	/* User request to abort, and not restart, any action */
513	MD_RECOVERY_ERROR,	/* sync-action interrupted because io-error */
514	MD_RECOVERY_WAIT,	/* waiting for pers->start() to finish */
515	MD_RESYNCING_REMOTE,	/* remote node is running resync thread */
516};
517
518static inline int __must_check mddev_lock(struct mddev *mddev)
519{
520	return mutex_lock_interruptible(&mddev->reconfig_mutex);
521}
522
523/* Sometimes we need to take the lock in a situation where
524 * failure due to interrupts is not acceptable.
525 */
526static inline void mddev_lock_nointr(struct mddev *mddev)
527{
528	mutex_lock(&mddev->reconfig_mutex);
529}
530
531static inline int mddev_trylock(struct mddev *mddev)
532{
533	return mutex_trylock(&mddev->reconfig_mutex);
534}
535extern void mddev_unlock(struct mddev *mddev);
536
537static inline void md_sync_acct(struct block_device *bdev, unsigned long nr_sectors)
538{
539	atomic_add(nr_sectors, &bdev->bd_contains->bd_disk->sync_io);
540}
541
542static inline void md_sync_acct_bio(struct bio *bio, unsigned long nr_sectors)
543{
544	atomic_add(nr_sectors, &bio->bi_disk->sync_io);
545}
546
547struct md_personality
548{
549	char *name;
550	int level;
551	struct list_head list;
552	struct module *owner;
553	bool (*make_request)(struct mddev *mddev, struct bio *bio);
554	/*
555	 * start up works that do NOT require md_thread. tasks that
556	 * requires md_thread should go into start()
557	 */
558	int (*run)(struct mddev *mddev);
559	/* start up works that require md threads */
560	int (*start)(struct mddev *mddev);
561	void (*free)(struct mddev *mddev, void *priv);
562	void (*status)(struct seq_file *seq, struct mddev *mddev);
563	/* error_handler must set ->faulty and clear ->in_sync
564	 * if appropriate, and should abort recovery if needed
565	 */
566	void (*error_handler)(struct mddev *mddev, struct md_rdev *rdev);
567	int (*hot_add_disk) (struct mddev *mddev, struct md_rdev *rdev);
568	int (*hot_remove_disk) (struct mddev *mddev, struct md_rdev *rdev);
569	int (*spare_active) (struct mddev *mddev);
570	sector_t (*sync_request)(struct mddev *mddev, sector_t sector_nr, int *skipped);
571	int (*resize) (struct mddev *mddev, sector_t sectors);
572	sector_t (*size) (struct mddev *mddev, sector_t sectors, int raid_disks);
573	int (*check_reshape) (struct mddev *mddev);
574	int (*start_reshape) (struct mddev *mddev);
575	void (*finish_reshape) (struct mddev *mddev);
576	void (*update_reshape_pos) (struct mddev *mddev);
577	/* quiesce suspends or resumes internal processing.
578	 * 1 - stop new actions and wait for action io to complete
579	 * 0 - return to normal behaviour
580	 */
581	void (*quiesce) (struct mddev *mddev, int quiesce);
582	/* takeover is used to transition an array from one
583	 * personality to another.  The new personality must be able
584	 * to handle the data in the current layout.
585	 * e.g. 2drive raid1 -> 2drive raid5
586	 *      ndrive raid5 -> degraded n+1drive raid6 with special layout
587	 * If the takeover succeeds, a new 'private' structure is returned.
588	 * This needs to be installed and then ->run used to activate the
589	 * array.
590	 */
591	void *(*takeover) (struct mddev *mddev);
592	/* congested implements bdi.congested_fn().
593	 * Will not be called while array is 'suspended' */
594	int (*congested)(struct mddev *mddev, int bits);
595	/* Changes the consistency policy of an active array. */
596	int (*change_consistency_policy)(struct mddev *mddev, const char *buf);
597};
598
 
599struct md_sysfs_entry {
600	struct attribute attr;
601	ssize_t (*show)(struct mddev *, char *);
602	ssize_t (*store)(struct mddev *, const char *, size_t);
603};
604extern struct attribute_group md_bitmap_group;
605
606static inline struct kernfs_node *sysfs_get_dirent_safe(struct kernfs_node *sd, char *name)
607{
608	if (sd)
609		return sysfs_get_dirent(sd, name);
610	return sd;
611}
612static inline void sysfs_notify_dirent_safe(struct kernfs_node *sd)
613{
614	if (sd)
615		sysfs_notify_dirent(sd);
616}
617
618static inline char * mdname (struct mddev * mddev)
619{
620	return mddev->gendisk ? mddev->gendisk->disk_name : "mdX";
621}
622
623static inline int sysfs_link_rdev(struct mddev *mddev, struct md_rdev *rdev)
624{
625	char nm[20];
626	if (!test_bit(Replacement, &rdev->flags) &&
627	    !test_bit(Journal, &rdev->flags) &&
628	    mddev->kobj.sd) {
629		sprintf(nm, "rd%d", rdev->raid_disk);
630		return sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
631	} else
632		return 0;
633}
634
635static inline void sysfs_unlink_rdev(struct mddev *mddev, struct md_rdev *rdev)
636{
637	char nm[20];
638	if (!test_bit(Replacement, &rdev->flags) &&
639	    !test_bit(Journal, &rdev->flags) &&
640	    mddev->kobj.sd) {
641		sprintf(nm, "rd%d", rdev->raid_disk);
642		sysfs_remove_link(&mddev->kobj, nm);
643	}
644}
645
646/*
647 * iterates through some rdev ringlist. It's safe to remove the
648 * current 'rdev'. Dont touch 'tmp' though.
649 */
650#define rdev_for_each_list(rdev, tmp, head)				\
651	list_for_each_entry_safe(rdev, tmp, head, same_set)
652
653/*
654 * iterates through the 'same array disks' ringlist
655 */
656#define rdev_for_each(rdev, mddev)				\
657	list_for_each_entry(rdev, &((mddev)->disks), same_set)
658
659#define rdev_for_each_safe(rdev, tmp, mddev)				\
660	list_for_each_entry_safe(rdev, tmp, &((mddev)->disks), same_set)
661
662#define rdev_for_each_rcu(rdev, mddev)				\
663	list_for_each_entry_rcu(rdev, &((mddev)->disks), same_set)
664
665struct md_thread {
666	void			(*run) (struct md_thread *thread);
667	struct mddev		*mddev;
668	wait_queue_head_t	wqueue;
669	unsigned long		flags;
670	struct task_struct	*tsk;
671	unsigned long		timeout;
672	void			*private;
673};
674
675#define THREAD_WAKEUP  0
676
677static inline void safe_put_page(struct page *p)
678{
679	if (p) put_page(p);
680}
681
682extern int register_md_personality(struct md_personality *p);
683extern int unregister_md_personality(struct md_personality *p);
684extern int register_md_cluster_operations(struct md_cluster_operations *ops,
685		struct module *module);
686extern int unregister_md_cluster_operations(void);
687extern int md_setup_cluster(struct mddev *mddev, int nodes);
688extern void md_cluster_stop(struct mddev *mddev);
689extern struct md_thread *md_register_thread(
690	void (*run)(struct md_thread *thread),
691	struct mddev *mddev,
692	const char *name);
693extern void md_unregister_thread(struct md_thread **threadp);
694extern void md_wakeup_thread(struct md_thread *thread);
695extern void md_check_recovery(struct mddev *mddev);
696extern void md_reap_sync_thread(struct mddev *mddev);
697extern int mddev_init_writes_pending(struct mddev *mddev);
698extern bool md_write_start(struct mddev *mddev, struct bio *bi);
699extern void md_write_inc(struct mddev *mddev, struct bio *bi);
700extern void md_write_end(struct mddev *mddev);
701extern void md_done_sync(struct mddev *mddev, int blocks, int ok);
702extern void md_error(struct mddev *mddev, struct md_rdev *rdev);
703extern void md_finish_reshape(struct mddev *mddev);
704
705extern int mddev_congested(struct mddev *mddev, int bits);
706extern void md_flush_request(struct mddev *mddev, struct bio *bio);
707extern void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
708			   sector_t sector, int size, struct page *page);
709extern int md_super_wait(struct mddev *mddev);
710extern int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
711			struct page *page, int op, int op_flags,
712			bool metadata_op);
713extern void md_do_sync(struct md_thread *thread);
714extern void md_new_event(struct mddev *mddev);
715extern void md_allow_write(struct mddev *mddev);
716extern void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev);
717extern void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors);
718extern int md_check_no_bitmap(struct mddev *mddev);
719extern int md_integrity_register(struct mddev *mddev);
720extern int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev);
721extern int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale);
722
723extern void mddev_init(struct mddev *mddev);
724extern int md_run(struct mddev *mddev);
725extern int md_start(struct mddev *mddev);
726extern void md_stop(struct mddev *mddev);
727extern void md_stop_writes(struct mddev *mddev);
728extern int md_rdev_init(struct md_rdev *rdev);
729extern void md_rdev_clear(struct md_rdev *rdev);
730
731extern void md_handle_request(struct mddev *mddev, struct bio *bio);
732extern void mddev_suspend(struct mddev *mddev);
733extern void mddev_resume(struct mddev *mddev);
 
 
734extern struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
735				   struct mddev *mddev);
736
737extern void md_reload_sb(struct mddev *mddev, int raid_disk);
738extern void md_update_sb(struct mddev *mddev, int force);
739extern void md_kick_rdev_from_array(struct md_rdev * rdev);
740extern void mddev_create_wb_pool(struct mddev *mddev, struct md_rdev *rdev,
741				 bool is_suspend);
742struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr);
743struct md_rdev *md_find_rdev_rcu(struct mddev *mddev, dev_t dev);
744
745static inline bool is_mddev_broken(struct md_rdev *rdev, const char *md_type)
746{
747	int flags = rdev->bdev->bd_disk->flags;
748
749	if (!(flags & GENHD_FL_UP)) {
750		if (!test_and_set_bit(MD_BROKEN, &rdev->mddev->flags))
751			pr_warn("md: %s: %s array has a missing/failed member\n",
752				mdname(rdev->mddev), md_type);
753		return true;
754	}
755	return false;
756}
757
758static inline void rdev_dec_pending(struct md_rdev *rdev, struct mddev *mddev)
759{
760	int faulty = test_bit(Faulty, &rdev->flags);
761	if (atomic_dec_and_test(&rdev->nr_pending) && faulty) {
762		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
763		md_wakeup_thread(mddev->thread);
764	}
765}
766
767extern struct md_cluster_operations *md_cluster_ops;
768static inline int mddev_is_clustered(struct mddev *mddev)
769{
770	return mddev->cluster_info && mddev->bitmap_info.nodes > 1;
771}
772
773/* clear unsupported mddev_flags */
774static inline void mddev_clear_unsupported_flags(struct mddev *mddev,
775	unsigned long unsupported_flags)
776{
777	mddev->flags &= ~unsupported_flags;
778}
779
780static inline void mddev_check_writesame(struct mddev *mddev, struct bio *bio)
781{
782	if (bio_op(bio) == REQ_OP_WRITE_SAME &&
783	    !bio->bi_disk->queue->limits.max_write_same_sectors)
784		mddev->queue->limits.max_write_same_sectors = 0;
785}
786
787static inline void mddev_check_write_zeroes(struct mddev *mddev, struct bio *bio)
788{
789	if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
790	    !bio->bi_disk->queue->limits.max_write_zeroes_sectors)
791		mddev->queue->limits.max_write_zeroes_sectors = 0;
792}
793#endif /* _MD_MD_H */