Linux Audio

Check our new training course

Loading...
v3.5.6
  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 sysfs_dirent *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	Unmerged,		/* device is being added to array and should
133				 * be considerred for bvec_merge_fn but not
134				 * yet for actual IO
 
135				 */
136	WriteMostly,		/* Avoid reading if at all possible */
137	AutoDetected,		/* added by auto-detect */
138	Blocked,		/* An error occurred but has not yet
139				 * been acknowledged by the metadata
140				 * handler, so don't allow writes
141				 * until it is cleared */
142	WriteErrorSeen,		/* A write error has been seen on this
143				 * device
144				 */
145	FaultRecorded,		/* Intermediate state for clearing
146				 * Blocked.  The Fault is/will-be
147				 * recorded in the metadata, but that
148				 * metadata hasn't been stored safely
149				 * on disk yet.
150				 */
151	BlockedBadBlocks,	/* A writer is blocked because they
152				 * found an unacknowledged bad-block.
153				 * This can safely be cleared at any
154				 * time, and the writer will re-check.
155				 * It may be set at any time, and at
156				 * worst the writer will timeout and
157				 * re-check.  So setting it as
158				 * accurately as possible is good, but
159				 * not absolutely critical.
160				 */
161	WantReplacement,	/* This device is a candidate to be
162				 * hot-replaced, either because it has
163				 * reported some faults, or because
164				 * of explicit request.
165				 */
166	Replacement,		/* This device is a replacement for
167				 * a want_replacement device with same
168				 * raid_disk number.
169				 */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
170};
171
172#define BB_LEN_MASK	(0x00000000000001FFULL)
173#define BB_OFFSET_MASK	(0x7FFFFFFFFFFFFE00ULL)
174#define BB_ACK_MASK	(0x8000000000000000ULL)
175#define BB_MAX_LEN	512
176#define BB_OFFSET(x)	(((x) & BB_OFFSET_MASK) >> 9)
177#define BB_LEN(x)	(((x) & BB_LEN_MASK) + 1)
178#define BB_ACK(x)	(!!((x) & BB_ACK_MASK))
179#define BB_MAKE(a, l, ack) (((a)<<9) | ((l)-1) | ((u64)(!!(ack)) << 63))
180
181extern int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
182			  sector_t *first_bad, int *bad_sectors);
183static inline int is_badblock(struct md_rdev *rdev, sector_t s, int sectors,
184			      sector_t *first_bad, int *bad_sectors)
185{
186	if (unlikely(rdev->badblocks.count)) {
187		int rv = md_is_badblock(&rdev->badblocks, rdev->data_offset + s,
188					sectors,
189					first_bad, bad_sectors);
190		if (rv)
191			*first_bad -= rdev->data_offset;
192		return rv;
193	}
194	return 0;
195}
196extern int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
197			      int is_new);
198extern int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
199				int is_new);
200extern void md_ack_all_badblocks(struct badblocks *bb);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
201
202struct mddev {
203	void				*private;
204	struct md_personality		*pers;
205	dev_t				unit;
206	int				md_minor;
207	struct list_head 		disks;
208	unsigned long			flags;
209#define MD_CHANGE_DEVS	0	/* Some device status has changed */
210#define MD_CHANGE_CLEAN 1	/* transition to or from 'clean' */
211#define MD_CHANGE_PENDING 2	/* switch from 'clean' to 'active' in progress */
212#define MD_ARRAY_FIRST_USE 3    /* First use of array, needs initialization */
213
214	int				suspended;
215	atomic_t			active_io;
216	int				ro;
217	int				sysfs_active; /* set when sysfs deletes
218						       * are happening, so run/
219						       * takeover/stop are not safe
220						       */
221	int				ready; /* See when safe to pass 
222						* IO requests down */
223	struct gendisk			*gendisk;
224
225	struct kobject			kobj;
226	int				hold_active;
227#define	UNTIL_IOCTL	1
228#define	UNTIL_STOP	2
229
230	/* Superblock information */
231	int				major_version,
232					minor_version,
233					patch_version;
234	int				persistent;
235	int 				external;	/* metadata is
236							 * managed externally */
237	char				metadata_type[17]; /* externally set*/
238	int				chunk_sectors;
239	time_t				ctime, utime;
240	int				level, layout;
241	char				clevel[16];
242	int				raid_disks;
243	int				max_disks;
244	sector_t			dev_sectors; 	/* used size of
245							 * component devices */
246	sector_t			array_sectors; /* exported array size */
247	int				external_size; /* size managed
248							* externally */
249	__u64				events;
250	/* If the last 'event' was simply a clean->dirty transition, and
251	 * we didn't write it to the spares, then it is safe and simple
252	 * to just decrement the event count on a dirty->clean transition.
253	 * So we record that possibility here.
254	 */
255	int				can_decrease_events;
256
257	char				uuid[16];
258
259	/* If the array is being reshaped, we need to record the
260	 * new shape and an indication of where we are up to.
261	 * This is written to the superblock.
262	 * If reshape_position is MaxSector, then no reshape is happening (yet).
263	 */
264	sector_t			reshape_position;
265	int				delta_disks, new_level, new_layout;
266	int				new_chunk_sectors;
267	int				reshape_backwards;
268
269	atomic_t			plug_cnt;	/* If device is expecting
270							 * more bios soon.
271							 */
272	struct md_thread		*thread;	/* management thread */
273	struct md_thread		*sync_thread;	/* doing resync or reconstruct */
 
 
 
 
 
 
 
 
274	sector_t			curr_resync;	/* last block scheduled */
275	/* As resync requests can complete out of order, we cannot easily track
276	 * how much resync has been completed.  So we occasionally pause until
277	 * everything completes, then set curr_resync_completed to curr_resync.
278	 * As such it may be well behind the real resync mark, but it is a value
279	 * we are certain of.
280	 */
281	sector_t			curr_resync_completed;
282	unsigned long			resync_mark;	/* a recent timestamp */
283	sector_t			resync_mark_cnt;/* blocks written at resync_mark */
284	sector_t			curr_mark_cnt; /* blocks scheduled now */
285
286	sector_t			resync_max_sectors; /* may be set by personality */
287
288	sector_t			resync_mismatches; /* count of sectors where
289							    * parity/replica mismatch found
290							    */
291
292	/* allow user-space to request suspension of IO to regions of the array */
293	sector_t			suspend_lo;
294	sector_t			suspend_hi;
295	/* if zero, use the system-wide default */
296	int				sync_speed_min;
297	int				sync_speed_max;
298
299	/* resync even though the same disks are shared among md-devices */
300	int				parallel_resync;
301
302	int				ok_start_degraded;
303	/* recovery/resync flags 
304	 * NEEDED:   we might need to start a resync/recover
305	 * RUNNING:  a thread is running, or about to be started
306	 * SYNC:     actually doing a resync, not a recovery
307	 * RECOVER:  doing recovery, or need to try it.
308	 * INTR:     resync needs to be aborted for some reason
309	 * DONE:     thread is done and is waiting to be reaped
310	 * REQUEST:  user-space has requested a sync (used with SYNC)
311	 * CHECK:    user-space request for check-only, no repair
312	 * RESHAPE:  A reshape is happening
313	 *
314	 * If neither SYNC or RESHAPE are set, then it is a recovery.
315	 */
316#define	MD_RECOVERY_RUNNING	0
317#define	MD_RECOVERY_SYNC	1
318#define	MD_RECOVERY_RECOVER	2
319#define	MD_RECOVERY_INTR	3
320#define	MD_RECOVERY_DONE	4
321#define	MD_RECOVERY_NEEDED	5
322#define	MD_RECOVERY_REQUESTED	6
323#define	MD_RECOVERY_CHECK	7
324#define MD_RECOVERY_RESHAPE	8
325#define	MD_RECOVERY_FROZEN	9
326
327	unsigned long			recovery;
328	/* If a RAID personality determines that recovery (of a particular
329	 * device) will fail due to a read error on the source device, it
330	 * takes a copy of this number and does not attempt recovery again
331	 * until this number changes.
332	 */
333	int				recovery_disabled;
334
335	int				in_sync;	/* know to not need resync */
336	/* 'open_mutex' avoids races between 'md_open' and 'do_md_stop', so
337	 * that we are never stopping an array while it is open.
338	 * 'reconfig_mutex' protects all other reconfiguration.
339	 * These locks are separate due to conflicting interactions
340	 * with bdev->bd_mutex.
341	 * Lock ordering is:
342	 *  reconfig_mutex -> bd_mutex : e.g. do_md_run -> revalidate_disk
343	 *  bd_mutex -> open_mutex:  e.g. __blkdev_get -> md_open
344	 */
345	struct mutex			open_mutex;
346	struct mutex			reconfig_mutex;
347	atomic_t			active;		/* general refcount */
348	atomic_t			openers;	/* number of active opens */
349
350	int				changed;	/* True if we might need to
351							 * reread partition info */
352	int				degraded;	/* whether md should consider
353							 * adding a spare
354							 */
355	int				merge_check_needed; /* at least one
356							     * member device
357							     * has a
358							     * merge_bvec_fn */
359
360	atomic_t			recovery_active; /* blocks scheduled, but not written */
361	wait_queue_head_t		recovery_wait;
362	sector_t			recovery_cp;
363	sector_t			resync_min;	/* user requested sync
364							 * starts here */
365	sector_t			resync_max;	/* resync should pause
366							 * when it gets here */
367
368	struct sysfs_dirent		*sysfs_state;	/* handle for 'array_state'
369							 * file in sysfs.
370							 */
371	struct sysfs_dirent		*sysfs_action;  /* handle for 'sync_action' */
372
373	struct work_struct del_work;	/* used for delayed sysfs removal */
374
375	spinlock_t			write_lock;
 
 
 
 
 
 
 
 
 
 
 
376	wait_queue_head_t		sb_wait;	/* for waiting on superblock updates */
377	atomic_t			pending_writes;	/* number of active superblock writes */
378
379	unsigned int			safemode;	/* if set, update "clean" superblock
380							 * when no writes pending.
381							 */ 
382	unsigned int			safemode_delay;
383	struct timer_list		safemode_timer;
384	atomic_t			writes_pending; 
 
385	struct request_queue		*queue;	/* for plugging ... */
386
387	struct bitmap                   *bitmap; /* the bitmap for the device */
388	struct {
389		struct file		*file; /* the bitmap file */
390		loff_t			offset; /* offset from superblock of
391						 * start of bitmap. May be
392						 * negative, but not '0'
393						 * For external metadata, offset
394						 * from start of device. 
395						 */
396		unsigned long		space; /* space available at this offset */
397		loff_t			default_offset; /* this is the offset to use when
398							 * hot-adding a bitmap.  It should
399							 * eventually be settable by sysfs.
400							 */
401		unsigned long		default_space; /* space available at
402							* default offset */
403		struct mutex		mutex;
404		unsigned long		chunksize;
405		unsigned long		daemon_sleep; /* how many jiffies between updates? */
406		unsigned long		max_write_behind; /* write-behind mode */
407		int			external;
 
 
408	} bitmap_info;
409
410	atomic_t 			max_corr_read_errors; /* max read retries */
411	struct list_head		all_mddevs;
412
413	struct attribute_group		*to_remove;
414
415	struct bio_set			*bio_set;
 
 
 
416
417	/* Generic flush handling.
418	 * The last to finish preflush schedules a worker to submit
419	 * the rest of the request (without the REQ_FLUSH flag).
420	 */
421	struct bio *flush_bio;
422	atomic_t flush_pending;
423	struct work_struct flush_work;
424	struct work_struct event_work;	/* used by dm to report failure event */
425	void (*sync_super)(struct mddev *mddev, struct md_rdev *rdev);
 
 
 
 
426};
427
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
428
429static inline void rdev_dec_pending(struct md_rdev *rdev, struct mddev *mddev)
430{
431	int faulty = test_bit(Faulty, &rdev->flags);
432	if (atomic_dec_and_test(&rdev->nr_pending) && faulty)
433		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
 
 
 
 
 
 
434}
435
 
 
 
 
 
 
436static inline void md_sync_acct(struct block_device *bdev, unsigned long nr_sectors)
437{
438        atomic_add(nr_sectors, &bdev->bd_contains->bd_disk->sync_io);
 
 
 
 
 
439}
440
441struct md_personality
442{
443	char *name;
444	int level;
445	struct list_head list;
446	struct module *owner;
447	void (*make_request)(struct mddev *mddev, struct bio *bio);
 
 
 
 
448	int (*run)(struct mddev *mddev);
449	int (*stop)(struct mddev *mddev);
 
 
450	void (*status)(struct seq_file *seq, struct mddev *mddev);
451	/* error_handler must set ->faulty and clear ->in_sync
452	 * if appropriate, and should abort recovery if needed 
453	 */
454	void (*error_handler)(struct mddev *mddev, struct md_rdev *rdev);
455	int (*hot_add_disk) (struct mddev *mddev, struct md_rdev *rdev);
456	int (*hot_remove_disk) (struct mddev *mddev, struct md_rdev *rdev);
457	int (*spare_active) (struct mddev *mddev);
458	sector_t (*sync_request)(struct mddev *mddev, sector_t sector_nr, int *skipped, int go_faster);
459	int (*resize) (struct mddev *mddev, sector_t sectors);
460	sector_t (*size) (struct mddev *mddev, sector_t sectors, int raid_disks);
461	int (*check_reshape) (struct mddev *mddev);
462	int (*start_reshape) (struct mddev *mddev);
463	void (*finish_reshape) (struct mddev *mddev);
464	/* quiesce moves between quiescence states
465	 * 0 - fully active
466	 * 1 - no new requests allowed
467	 * others - reserved
468	 */
469	void (*quiesce) (struct mddev *mddev, int state);
470	/* takeover is used to transition an array from one
471	 * personality to another.  The new personality must be able
472	 * to handle the data in the current layout.
473	 * e.g. 2drive raid1 -> 2drive raid5
474	 *      ndrive raid5 -> degraded n+1drive raid6 with special layout
475	 * If the takeover succeeds, a new 'private' structure is returned.
476	 * This needs to be installed and then ->run used to activate the
477	 * array.
478	 */
479	void *(*takeover) (struct mddev *mddev);
 
 
 
 
 
480};
481
482
483struct md_sysfs_entry {
484	struct attribute attr;
485	ssize_t (*show)(struct mddev *, char *);
486	ssize_t (*store)(struct mddev *, const char *, size_t);
487};
488extern struct attribute_group md_bitmap_group;
489
490static inline struct sysfs_dirent *sysfs_get_dirent_safe(struct sysfs_dirent *sd, char *name)
491{
492	if (sd)
493		return sysfs_get_dirent(sd, NULL, name);
494	return sd;
495}
496static inline void sysfs_notify_dirent_safe(struct sysfs_dirent *sd)
497{
498	if (sd)
499		sysfs_notify_dirent(sd);
500}
501
502static inline char * mdname (struct mddev * mddev)
503{
504	return mddev->gendisk ? mddev->gendisk->disk_name : "mdX";
505}
506
507static inline int sysfs_link_rdev(struct mddev *mddev, struct md_rdev *rdev)
508{
509	char nm[20];
510	if (!test_bit(Replacement, &rdev->flags)) {
 
 
511		sprintf(nm, "rd%d", rdev->raid_disk);
512		return sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
513	} else
514		return 0;
515}
516
517static inline void sysfs_unlink_rdev(struct mddev *mddev, struct md_rdev *rdev)
518{
519	char nm[20];
520	if (!test_bit(Replacement, &rdev->flags)) {
 
 
521		sprintf(nm, "rd%d", rdev->raid_disk);
522		sysfs_remove_link(&mddev->kobj, nm);
523	}
524}
525
526/*
527 * iterates through some rdev ringlist. It's safe to remove the
528 * current 'rdev'. Dont touch 'tmp' though.
529 */
530#define rdev_for_each_list(rdev, tmp, head)				\
531	list_for_each_entry_safe(rdev, tmp, head, same_set)
532
533/*
534 * iterates through the 'same array disks' ringlist
535 */
536#define rdev_for_each(rdev, mddev)				\
537	list_for_each_entry(rdev, &((mddev)->disks), same_set)
538
539#define rdev_for_each_safe(rdev, tmp, mddev)				\
540	list_for_each_entry_safe(rdev, tmp, &((mddev)->disks), same_set)
541
542#define rdev_for_each_rcu(rdev, mddev)				\
543	list_for_each_entry_rcu(rdev, &((mddev)->disks), same_set)
544
545struct md_thread {
546	void			(*run) (struct mddev *mddev);
547	struct mddev		*mddev;
548	wait_queue_head_t	wqueue;
549	unsigned long           flags;
550	struct task_struct	*tsk;
551	unsigned long		timeout;
 
552};
553
554#define THREAD_WAKEUP  0
555
556#define __wait_event_lock_irq(wq, condition, lock, cmd) 		\
557do {									\
558	wait_queue_t __wait;						\
559	init_waitqueue_entry(&__wait, current);				\
560									\
561	add_wait_queue(&wq, &__wait);					\
562	for (;;) {							\
563		set_current_state(TASK_UNINTERRUPTIBLE);		\
564		if (condition)						\
565			break;						\
566		spin_unlock_irq(&lock);					\
567		cmd;							\
568		schedule();						\
569		spin_lock_irq(&lock);					\
570	}								\
571	current->state = TASK_RUNNING;					\
572	remove_wait_queue(&wq, &__wait);				\
573} while (0)
574
575#define wait_event_lock_irq(wq, condition, lock, cmd) 			\
576do {									\
577	if (condition)	 						\
578		break;							\
579	__wait_event_lock_irq(wq, condition, lock, cmd);		\
580} while (0)
581
582static inline void safe_put_page(struct page *p)
583{
584	if (p) put_page(p);
585}
586
587extern int register_md_personality(struct md_personality *p);
588extern int unregister_md_personality(struct md_personality *p);
 
 
 
 
 
589extern struct md_thread *md_register_thread(
590	void (*run)(struct mddev *mddev),
591	struct mddev *mddev,
592	const char *name);
593extern void md_unregister_thread(struct md_thread **threadp);
594extern void md_wakeup_thread(struct md_thread *thread);
595extern void md_check_recovery(struct mddev *mddev);
596extern void md_write_start(struct mddev *mddev, struct bio *bi);
 
 
 
597extern void md_write_end(struct mddev *mddev);
598extern void md_done_sync(struct mddev *mddev, int blocks, int ok);
599extern void md_error(struct mddev *mddev, struct md_rdev *rdev);
600extern void md_finish_reshape(struct mddev *mddev);
601
602extern int mddev_congested(struct mddev *mddev, int bits);
603extern void md_flush_request(struct mddev *mddev, struct bio *bio);
604extern void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
605			   sector_t sector, int size, struct page *page);
606extern void md_super_wait(struct mddev *mddev);
607extern int sync_page_io(struct md_rdev *rdev, sector_t sector, int size, 
608			struct page *page, int rw, bool metadata_op);
609extern void md_do_sync(struct mddev *mddev);
 
610extern void md_new_event(struct mddev *mddev);
611extern int md_allow_write(struct mddev *mddev);
612extern void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev);
613extern void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors);
614extern int md_check_no_bitmap(struct mddev *mddev);
615extern int md_integrity_register(struct mddev *mddev);
616extern void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev);
617extern int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale);
618extern void restore_bitmap_write_access(struct file *file);
619
620extern void mddev_init(struct mddev *mddev);
621extern int md_run(struct mddev *mddev);
 
622extern void md_stop(struct mddev *mddev);
623extern void md_stop_writes(struct mddev *mddev);
624extern int md_rdev_init(struct md_rdev *rdev);
625extern void md_rdev_clear(struct md_rdev *rdev);
626
 
627extern void mddev_suspend(struct mddev *mddev);
628extern void mddev_resume(struct mddev *mddev);
629extern struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
630				   struct mddev *mddev);
631extern struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
632				   struct mddev *mddev);
633extern int mddev_check_plugged(struct mddev *mddev);
634extern void md_trim_bio(struct bio *bio, int offset, int size);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
635#endif /* _MD_MD_H */
v4.17
  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/backing-dev.h>
 20#include <linux/badblocks.h>
 21#include <linux/kobject.h>
 22#include <linux/list.h>
 23#include <linux/mm.h>
 24#include <linux/mutex.h>
 25#include <linux/timer.h>
 26#include <linux/wait.h>
 27#include <linux/workqueue.h>
 28#include "md-cluster.h"
 29
 30#define MaxSector (~(sector_t)0)
 31
 32/*
 33 * These flags should really be called "NO_RETRY" rather than
 34 * "FAILFAST" because they don't make any promise about time lapse,
 35 * only about the number of retries, which will be zero.
 36 * REQ_FAILFAST_DRIVER is not included because
 37 * Commit: 4a27446f3e39 ("[SCSI] modify scsi to handle new fail fast flags.")
 38 * seems to suggest that the errors it avoids retrying should usually
 39 * be retried.
 40 */
 41#define	MD_FAILFAST	(REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT)
 
 42/*
 43 * MD's 'extended' device
 44 */
 45struct md_rdev {
 46	struct list_head same_set;	/* RAID devices within the same set */
 47
 48	sector_t sectors;		/* Device size (in 512bytes sectors) */
 49	struct mddev *mddev;		/* RAID array if running */
 50	int last_events;		/* IO event timestamp */
 51
 52	/*
 53	 * If meta_bdev is non-NULL, it means that a separate device is
 54	 * being used to store the metadata (superblock/bitmap) which
 55	 * would otherwise be contained on the same device as the data (bdev).
 56	 */
 57	struct block_device *meta_bdev;
 58	struct block_device *bdev;	/* block device handle */
 59
 60	struct page	*sb_page, *bb_page;
 61	int		sb_loaded;
 62	__u64		sb_events;
 63	sector_t	data_offset;	/* start of data in array */
 64	sector_t	new_data_offset;/* only relevant while reshaping */
 65	sector_t	sb_start;	/* offset of the super block (in 512byte sectors) */
 66	int		sb_size;	/* bytes in the superblock */
 67	int		preferred_minor;	/* autorun support */
 68
 69	struct kobject	kobj;
 70
 71	/* A device can be in one of three states based on two flags:
 72	 * Not working:   faulty==1 in_sync==0
 73	 * Fully working: faulty==0 in_sync==1
 74	 * Working, but not
 75	 * in sync with array
 76	 *                faulty==0 in_sync==0
 77	 *
 78	 * It can never have faulty==1, in_sync==1
 79	 * This reduces the burden of testing multiple flags in many cases
 80	 */
 81
 82	unsigned long	flags;	/* bit set of 'enum flag_bits' bits. */
 83	wait_queue_head_t blocked_wait;
 84
 85	int desc_nr;			/* descriptor index in the superblock */
 86	int raid_disk;			/* role of device in array */
 87	int new_raid_disk;		/* role that the device will have in
 88					 * the array after a level-change completes.
 89					 */
 90	int saved_raid_disk;		/* role that device used to have in the
 91					 * array and could again if we did a partial
 92					 * resync from the bitmap
 93					 */
 94	union {
 95		sector_t recovery_offset;/* If this device has been partially
 96					 * recovered, this is where we were
 97					 * up to.
 98					 */
 99		sector_t journal_tail;	/* If this device is a journal device,
100					 * this is the journal tail (journal
101					 * recovery start point)
102					 */
103	};
104
105	atomic_t	nr_pending;	/* number of pending requests.
106					 * only maintained for arrays that
107					 * support hot removal
108					 */
109	atomic_t	read_errors;	/* number of consecutive read errors that
110					 * we have tried to ignore.
111					 */
112	time64_t	last_read_error;	/* monotonic time since our
113						 * last read error
114						 */
115	atomic_t	corrected_errors; /* number of corrected read errors,
116					   * for reporting to userspace and storing
117					   * in superblock.
118					   */
119	struct work_struct del_work;	/* used for delayed sysfs removal */
120
121	struct kernfs_node *sysfs_state; /* handle for 'state'
122					   * sysfs entry */
123
124	struct badblocks badblocks;
125
126	struct {
127		short offset;	/* Offset from superblock to start of PPL.
128				 * Not used by external metadata. */
129		unsigned int size;	/* Size in sectors of the PPL space */
130		sector_t sector;	/* First sector of the PPL space */
131	} ppl;
 
 
 
 
 
 
 
 
132};
133enum flag_bits {
134	Faulty,			/* device is known to have a fault */
135	In_sync,		/* device is in_sync with rest of array */
136	Bitmap_sync,		/* ..actually, not quite In_sync.  Need a
137				 * bitmap-based recovery to get fully in sync.
138				 * The bit is only meaningful before device
139				 * has been passed to pers->hot_add_disk.
140				 */
141	WriteMostly,		/* Avoid reading if at all possible */
142	AutoDetected,		/* added by auto-detect */
143	Blocked,		/* An error occurred but has not yet
144				 * been acknowledged by the metadata
145				 * handler, so don't allow writes
146				 * until it is cleared */
147	WriteErrorSeen,		/* A write error has been seen on this
148				 * device
149				 */
150	FaultRecorded,		/* Intermediate state for clearing
151				 * Blocked.  The Fault is/will-be
152				 * recorded in the metadata, but that
153				 * metadata hasn't been stored safely
154				 * on disk yet.
155				 */
156	BlockedBadBlocks,	/* A writer is blocked because they
157				 * found an unacknowledged bad-block.
158				 * This can safely be cleared at any
159				 * time, and the writer will re-check.
160				 * It may be set at any time, and at
161				 * worst the writer will timeout and
162				 * re-check.  So setting it as
163				 * accurately as possible is good, but
164				 * not absolutely critical.
165				 */
166	WantReplacement,	/* This device is a candidate to be
167				 * hot-replaced, either because it has
168				 * reported some faults, or because
169				 * of explicit request.
170				 */
171	Replacement,		/* This device is a replacement for
172				 * a want_replacement device with same
173				 * raid_disk number.
174				 */
175	Candidate,		/* For clustered environments only:
176				 * This device is seen locally but not
177				 * by the whole cluster
178				 */
179	Journal,		/* This device is used as journal for
180				 * raid-5/6.
181				 * Usually, this device should be faster
182				 * than other devices in the array
183				 */
184	ClusterRemove,
185	RemoveSynchronized,	/* synchronize_rcu() was called after
186				 * this device was known to be faulty,
187				 * so it is safe to remove without
188				 * another synchronize_rcu() call.
189				 */
190	ExternalBbl,            /* External metadata provides bad
191				 * block management for a disk
192				 */
193	FailFast,		/* Minimal retries should be attempted on
194				 * this device, so use REQ_FAILFAST_DEV.
195				 * Also don't try to repair failed reads.
196				 * It is expects that no bad block log
197				 * is present.
198				 */
199	LastDev,		/* Seems to be the last working dev as
200				 * it didn't fail, so don't use FailFast
201				 * any more for metadata
202				 */
203};
204
 
 
 
 
 
 
 
 
 
 
 
205static inline int is_badblock(struct md_rdev *rdev, sector_t s, int sectors,
206			      sector_t *first_bad, int *bad_sectors)
207{
208	if (unlikely(rdev->badblocks.count)) {
209		int rv = badblocks_check(&rdev->badblocks, rdev->data_offset + s,
210					sectors,
211					first_bad, bad_sectors);
212		if (rv)
213			*first_bad -= rdev->data_offset;
214		return rv;
215	}
216	return 0;
217}
218extern int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
219			      int is_new);
220extern int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
221				int is_new);
222struct md_cluster_info;
223
224/* change UNSUPPORTED_MDDEV_FLAGS for each array type if new flag is added */
225enum mddev_flags {
226	MD_ARRAY_FIRST_USE,	/* First use of array, needs initialization */
227	MD_CLOSING,		/* If set, we are closing the array, do not open
228				 * it then */
229	MD_JOURNAL_CLEAN,	/* A raid with journal is already clean */
230	MD_HAS_JOURNAL,		/* The raid array has journal feature set */
231	MD_CLUSTER_RESYNC_LOCKED, /* cluster raid only, which means node
232				   * already took resync lock, need to
233				   * release the lock */
234	MD_FAILFAST_SUPPORTED,	/* Using MD_FAILFAST on metadata writes is
235				 * supported as calls to md_error() will
236				 * never cause the array to become failed.
237				 */
238	MD_HAS_PPL,		/* The raid array has PPL feature set */
239	MD_HAS_MULTIPLE_PPLS,	/* The raid array has multiple PPLs feature set */
240	MD_ALLOW_SB_UPDATE,	/* md_check_recovery is allowed to update
241				 * the metadata without taking reconfig_mutex.
242				 */
243	MD_UPDATING_SB,		/* md_check_recovery is updating the metadata
244				 * without explicitly holding reconfig_mutex.
245				 */
246};
247
248enum mddev_sb_flags {
249	MD_SB_CHANGE_DEVS,		/* Some device status has changed */
250	MD_SB_CHANGE_CLEAN,	/* transition to or from 'clean' */
251	MD_SB_CHANGE_PENDING,	/* switch from 'clean' to 'active' in progress */
252	MD_SB_NEED_REWRITE,	/* metadata write needs to be repeated */
253};
254
255struct mddev {
256	void				*private;
257	struct md_personality		*pers;
258	dev_t				unit;
259	int				md_minor;
260	struct list_head		disks;
261	unsigned long			flags;
262	unsigned long			sb_flags;
 
 
 
263
264	int				suspended;
265	atomic_t			active_io;
266	int				ro;
267	int				sysfs_active; /* set when sysfs deletes
268						       * are happening, so run/
269						       * takeover/stop are not safe
270						       */
 
 
271	struct gendisk			*gendisk;
272
273	struct kobject			kobj;
274	int				hold_active;
275#define	UNTIL_IOCTL	1
276#define	UNTIL_STOP	2
277
278	/* Superblock information */
279	int				major_version,
280					minor_version,
281					patch_version;
282	int				persistent;
283	int				external;	/* metadata is
284							 * managed externally */
285	char				metadata_type[17]; /* externally set*/
286	int				chunk_sectors;
287	time64_t			ctime, utime;
288	int				level, layout;
289	char				clevel[16];
290	int				raid_disks;
291	int				max_disks;
292	sector_t			dev_sectors;	/* used size of
293							 * component devices */
294	sector_t			array_sectors; /* exported array size */
295	int				external_size; /* size managed
296							* externally */
297	__u64				events;
298	/* If the last 'event' was simply a clean->dirty transition, and
299	 * we didn't write it to the spares, then it is safe and simple
300	 * to just decrement the event count on a dirty->clean transition.
301	 * So we record that possibility here.
302	 */
303	int				can_decrease_events;
304
305	char				uuid[16];
306
307	/* If the array is being reshaped, we need to record the
308	 * new shape and an indication of where we are up to.
309	 * This is written to the superblock.
310	 * If reshape_position is MaxSector, then no reshape is happening (yet).
311	 */
312	sector_t			reshape_position;
313	int				delta_disks, new_level, new_layout;
314	int				new_chunk_sectors;
315	int				reshape_backwards;
316
 
 
 
317	struct md_thread		*thread;	/* management thread */
318	struct md_thread		*sync_thread;	/* doing resync or reconstruct */
319
320	/* 'last_sync_action' is initialized to "none".  It is set when a
321	 * sync operation (i.e "data-check", "requested-resync", "resync",
322	 * "recovery", or "reshape") is started.  It holds this value even
323	 * when the sync thread is "frozen" (interrupted) or "idle" (stopped
324	 * or finished).  It is overwritten when a new sync operation is begun.
325	 */
326	char				*last_sync_action;
327	sector_t			curr_resync;	/* last block scheduled */
328	/* As resync requests can complete out of order, we cannot easily track
329	 * how much resync has been completed.  So we occasionally pause until
330	 * everything completes, then set curr_resync_completed to curr_resync.
331	 * As such it may be well behind the real resync mark, but it is a value
332	 * we are certain of.
333	 */
334	sector_t			curr_resync_completed;
335	unsigned long			resync_mark;	/* a recent timestamp */
336	sector_t			resync_mark_cnt;/* blocks written at resync_mark */
337	sector_t			curr_mark_cnt; /* blocks scheduled now */
338
339	sector_t			resync_max_sectors; /* may be set by personality */
340
341	atomic64_t			resync_mismatches; /* count of sectors where
342							    * parity/replica mismatch found
343							    */
344
345	/* allow user-space to request suspension of IO to regions of the array */
346	sector_t			suspend_lo;
347	sector_t			suspend_hi;
348	/* if zero, use the system-wide default */
349	int				sync_speed_min;
350	int				sync_speed_max;
351
352	/* resync even though the same disks are shared among md-devices */
353	int				parallel_resync;
354
355	int				ok_start_degraded;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
356
357	unsigned long			recovery;
358	/* If a RAID personality determines that recovery (of a particular
359	 * device) will fail due to a read error on the source device, it
360	 * takes a copy of this number and does not attempt recovery again
361	 * until this number changes.
362	 */
363	int				recovery_disabled;
364
365	int				in_sync;	/* know to not need resync */
366	/* 'open_mutex' avoids races between 'md_open' and 'do_md_stop', so
367	 * that we are never stopping an array while it is open.
368	 * 'reconfig_mutex' protects all other reconfiguration.
369	 * These locks are separate due to conflicting interactions
370	 * with bdev->bd_mutex.
371	 * Lock ordering is:
372	 *  reconfig_mutex -> bd_mutex : e.g. do_md_run -> revalidate_disk
373	 *  bd_mutex -> open_mutex:  e.g. __blkdev_get -> md_open
374	 */
375	struct mutex			open_mutex;
376	struct mutex			reconfig_mutex;
377	atomic_t			active;		/* general refcount */
378	atomic_t			openers;	/* number of active opens */
379
380	int				changed;	/* True if we might need to
381							 * reread partition info */
382	int				degraded;	/* whether md should consider
383							 * adding a spare
384							 */
 
 
 
 
385
386	atomic_t			recovery_active; /* blocks scheduled, but not written */
387	wait_queue_head_t		recovery_wait;
388	sector_t			recovery_cp;
389	sector_t			resync_min;	/* user requested sync
390							 * starts here */
391	sector_t			resync_max;	/* resync should pause
392							 * when it gets here */
393
394	struct kernfs_node		*sysfs_state;	/* handle for 'array_state'
395							 * file in sysfs.
396							 */
397	struct kernfs_node		*sysfs_action;  /* handle for 'sync_action' */
398
399	struct work_struct del_work;	/* used for delayed sysfs removal */
400
401	/* "lock" protects:
402	 *   flush_bio transition from NULL to !NULL
403	 *   rdev superblocks, events
404	 *   clearing MD_CHANGE_*
405	 *   in_sync - and related safemode and MD_CHANGE changes
406	 *   pers (also protected by reconfig_mutex and pending IO).
407	 *   clearing ->bitmap
408	 *   clearing ->bitmap_info.file
409	 *   changing ->resync_{min,max}
410	 *   setting MD_RECOVERY_RUNNING (which interacts with resync_{min,max})
411	 */
412	spinlock_t			lock;
413	wait_queue_head_t		sb_wait;	/* for waiting on superblock updates */
414	atomic_t			pending_writes;	/* number of active superblock writes */
415
416	unsigned int			safemode;	/* if set, update "clean" superblock
417							 * when no writes pending.
418							 */
419	unsigned int			safemode_delay;
420	struct timer_list		safemode_timer;
421	struct percpu_ref		writes_pending;
422	int				sync_checkers;	/* # of threads checking writes_pending */
423	struct request_queue		*queue;	/* for plugging ... */
424
425	struct bitmap			*bitmap; /* the bitmap for the device */
426	struct {
427		struct file		*file; /* the bitmap file */
428		loff_t			offset; /* offset from superblock of
429						 * start of bitmap. May be
430						 * negative, but not '0'
431						 * For external metadata, offset
432						 * from start of device.
433						 */
434		unsigned long		space; /* space available at this offset */
435		loff_t			default_offset; /* this is the offset to use when
436							 * hot-adding a bitmap.  It should
437							 * eventually be settable by sysfs.
438							 */
439		unsigned long		default_space; /* space available at
440							* default offset */
441		struct mutex		mutex;
442		unsigned long		chunksize;
443		unsigned long		daemon_sleep; /* how many jiffies between updates? */
444		unsigned long		max_write_behind; /* write-behind mode */
445		int			external;
446		int			nodes; /* Maximum number of nodes in the cluster */
447		char                    cluster_name[64]; /* Name of the cluster */
448	} bitmap_info;
449
450	atomic_t			max_corr_read_errors; /* max read retries */
451	struct list_head		all_mddevs;
452
453	struct attribute_group		*to_remove;
454
455	struct bio_set			*bio_set;
456	struct bio_set			*sync_set; /* for sync operations like
457						   * metadata and bitmap writes
458						   */
459
460	/* Generic flush handling.
461	 * The last to finish preflush schedules a worker to submit
462	 * the rest of the request (without the REQ_PREFLUSH flag).
463	 */
464	struct bio *flush_bio;
465	atomic_t flush_pending;
466	struct work_struct flush_work;
467	struct work_struct event_work;	/* used by dm to report failure event */
468	void (*sync_super)(struct mddev *mddev, struct md_rdev *rdev);
469	struct md_cluster_info		*cluster_info;
470	unsigned int			good_device_nr;	/* good device num within cluster raid */
471
472	bool	has_superblocks:1;
473};
474
475enum recovery_flags {
476	/*
477	 * If neither SYNC or RESHAPE are set, then it is a recovery.
478	 */
479	MD_RECOVERY_RUNNING,	/* a thread is running, or about to be started */
480	MD_RECOVERY_SYNC,	/* actually doing a resync, not a recovery */
481	MD_RECOVERY_RECOVER,	/* doing recovery, or need to try it. */
482	MD_RECOVERY_INTR,	/* resync needs to be aborted for some reason */
483	MD_RECOVERY_DONE,	/* thread is done and is waiting to be reaped */
484	MD_RECOVERY_NEEDED,	/* we might need to start a resync/recover */
485	MD_RECOVERY_REQUESTED,	/* user-space has requested a sync (used with SYNC) */
486	MD_RECOVERY_CHECK,	/* user-space request for check-only, no repair */
487	MD_RECOVERY_RESHAPE,	/* A reshape is happening */
488	MD_RECOVERY_FROZEN,	/* User request to abort, and not restart, any action */
489	MD_RECOVERY_ERROR,	/* sync-action interrupted because io-error */
490	MD_RECOVERY_WAIT,	/* waiting for pers->start() to finish */
491};
492
493static inline int __must_check mddev_lock(struct mddev *mddev)
494{
495	return mutex_lock_interruptible(&mddev->reconfig_mutex);
496}
497
498/* Sometimes we need to take the lock in a situation where
499 * failure due to interrupts is not acceptable.
500 */
501static inline void mddev_lock_nointr(struct mddev *mddev)
502{
503	mutex_lock(&mddev->reconfig_mutex);
504}
505
506static inline int mddev_trylock(struct mddev *mddev)
507{
508	return mutex_trylock(&mddev->reconfig_mutex);
509}
510extern void mddev_unlock(struct mddev *mddev);
511
512static inline void md_sync_acct(struct block_device *bdev, unsigned long nr_sectors)
513{
514	atomic_add(nr_sectors, &bdev->bd_contains->bd_disk->sync_io);
515}
516
517static inline void md_sync_acct_bio(struct bio *bio, unsigned long nr_sectors)
518{
519	atomic_add(nr_sectors, &bio->bi_disk->sync_io);
520}
521
522struct md_personality
523{
524	char *name;
525	int level;
526	struct list_head list;
527	struct module *owner;
528	bool (*make_request)(struct mddev *mddev, struct bio *bio);
529	/*
530	 * start up works that do NOT require md_thread. tasks that
531	 * requires md_thread should go into start()
532	 */
533	int (*run)(struct mddev *mddev);
534	/* start up works that require md threads */
535	int (*start)(struct mddev *mddev);
536	void (*free)(struct mddev *mddev, void *priv);
537	void (*status)(struct seq_file *seq, struct mddev *mddev);
538	/* error_handler must set ->faulty and clear ->in_sync
539	 * if appropriate, and should abort recovery if needed
540	 */
541	void (*error_handler)(struct mddev *mddev, struct md_rdev *rdev);
542	int (*hot_add_disk) (struct mddev *mddev, struct md_rdev *rdev);
543	int (*hot_remove_disk) (struct mddev *mddev, struct md_rdev *rdev);
544	int (*spare_active) (struct mddev *mddev);
545	sector_t (*sync_request)(struct mddev *mddev, sector_t sector_nr, int *skipped);
546	int (*resize) (struct mddev *mddev, sector_t sectors);
547	sector_t (*size) (struct mddev *mddev, sector_t sectors, int raid_disks);
548	int (*check_reshape) (struct mddev *mddev);
549	int (*start_reshape) (struct mddev *mddev);
550	void (*finish_reshape) (struct mddev *mddev);
551	/* quiesce suspends or resumes internal processing.
552	 * 1 - stop new actions and wait for action io to complete
553	 * 0 - return to normal behaviour
 
554	 */
555	void (*quiesce) (struct mddev *mddev, int quiesce);
556	/* takeover is used to transition an array from one
557	 * personality to another.  The new personality must be able
558	 * to handle the data in the current layout.
559	 * e.g. 2drive raid1 -> 2drive raid5
560	 *      ndrive raid5 -> degraded n+1drive raid6 with special layout
561	 * If the takeover succeeds, a new 'private' structure is returned.
562	 * This needs to be installed and then ->run used to activate the
563	 * array.
564	 */
565	void *(*takeover) (struct mddev *mddev);
566	/* congested implements bdi.congested_fn().
567	 * Will not be called while array is 'suspended' */
568	int (*congested)(struct mddev *mddev, int bits);
569	/* Changes the consistency policy of an active array. */
570	int (*change_consistency_policy)(struct mddev *mddev, const char *buf);
571};
572
 
573struct md_sysfs_entry {
574	struct attribute attr;
575	ssize_t (*show)(struct mddev *, char *);
576	ssize_t (*store)(struct mddev *, const char *, size_t);
577};
578extern struct attribute_group md_bitmap_group;
579
580static inline struct kernfs_node *sysfs_get_dirent_safe(struct kernfs_node *sd, char *name)
581{
582	if (sd)
583		return sysfs_get_dirent(sd, name);
584	return sd;
585}
586static inline void sysfs_notify_dirent_safe(struct kernfs_node *sd)
587{
588	if (sd)
589		sysfs_notify_dirent(sd);
590}
591
592static inline char * mdname (struct mddev * mddev)
593{
594	return mddev->gendisk ? mddev->gendisk->disk_name : "mdX";
595}
596
597static inline int sysfs_link_rdev(struct mddev *mddev, struct md_rdev *rdev)
598{
599	char nm[20];
600	if (!test_bit(Replacement, &rdev->flags) &&
601	    !test_bit(Journal, &rdev->flags) &&
602	    mddev->kobj.sd) {
603		sprintf(nm, "rd%d", rdev->raid_disk);
604		return sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
605	} else
606		return 0;
607}
608
609static inline void sysfs_unlink_rdev(struct mddev *mddev, struct md_rdev *rdev)
610{
611	char nm[20];
612	if (!test_bit(Replacement, &rdev->flags) &&
613	    !test_bit(Journal, &rdev->flags) &&
614	    mddev->kobj.sd) {
615		sprintf(nm, "rd%d", rdev->raid_disk);
616		sysfs_remove_link(&mddev->kobj, nm);
617	}
618}
619
620/*
621 * iterates through some rdev ringlist. It's safe to remove the
622 * current 'rdev'. Dont touch 'tmp' though.
623 */
624#define rdev_for_each_list(rdev, tmp, head)				\
625	list_for_each_entry_safe(rdev, tmp, head, same_set)
626
627/*
628 * iterates through the 'same array disks' ringlist
629 */
630#define rdev_for_each(rdev, mddev)				\
631	list_for_each_entry(rdev, &((mddev)->disks), same_set)
632
633#define rdev_for_each_safe(rdev, tmp, mddev)				\
634	list_for_each_entry_safe(rdev, tmp, &((mddev)->disks), same_set)
635
636#define rdev_for_each_rcu(rdev, mddev)				\
637	list_for_each_entry_rcu(rdev, &((mddev)->disks), same_set)
638
639struct md_thread {
640	void			(*run) (struct md_thread *thread);
641	struct mddev		*mddev;
642	wait_queue_head_t	wqueue;
643	unsigned long		flags;
644	struct task_struct	*tsk;
645	unsigned long		timeout;
646	void			*private;
647};
648
649#define THREAD_WAKEUP  0
650
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
651static inline void safe_put_page(struct page *p)
652{
653	if (p) put_page(p);
654}
655
656extern int register_md_personality(struct md_personality *p);
657extern int unregister_md_personality(struct md_personality *p);
658extern int register_md_cluster_operations(struct md_cluster_operations *ops,
659		struct module *module);
660extern int unregister_md_cluster_operations(void);
661extern int md_setup_cluster(struct mddev *mddev, int nodes);
662extern void md_cluster_stop(struct mddev *mddev);
663extern struct md_thread *md_register_thread(
664	void (*run)(struct md_thread *thread),
665	struct mddev *mddev,
666	const char *name);
667extern void md_unregister_thread(struct md_thread **threadp);
668extern void md_wakeup_thread(struct md_thread *thread);
669extern void md_check_recovery(struct mddev *mddev);
670extern void md_reap_sync_thread(struct mddev *mddev);
671extern int mddev_init_writes_pending(struct mddev *mddev);
672extern bool md_write_start(struct mddev *mddev, struct bio *bi);
673extern void md_write_inc(struct mddev *mddev, struct bio *bi);
674extern void md_write_end(struct mddev *mddev);
675extern void md_done_sync(struct mddev *mddev, int blocks, int ok);
676extern void md_error(struct mddev *mddev, struct md_rdev *rdev);
677extern void md_finish_reshape(struct mddev *mddev);
678
679extern int mddev_congested(struct mddev *mddev, int bits);
680extern void md_flush_request(struct mddev *mddev, struct bio *bio);
681extern void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
682			   sector_t sector, int size, struct page *page);
683extern int md_super_wait(struct mddev *mddev);
684extern int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
685			struct page *page, int op, int op_flags,
686			bool metadata_op);
687extern void md_do_sync(struct md_thread *thread);
688extern void md_new_event(struct mddev *mddev);
689extern void md_allow_write(struct mddev *mddev);
690extern void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev);
691extern void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors);
692extern int md_check_no_bitmap(struct mddev *mddev);
693extern int md_integrity_register(struct mddev *mddev);
694extern int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev);
695extern int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale);
 
696
697extern void mddev_init(struct mddev *mddev);
698extern int md_run(struct mddev *mddev);
699extern int md_start(struct mddev *mddev);
700extern void md_stop(struct mddev *mddev);
701extern void md_stop_writes(struct mddev *mddev);
702extern int md_rdev_init(struct md_rdev *rdev);
703extern void md_rdev_clear(struct md_rdev *rdev);
704
705extern void md_handle_request(struct mddev *mddev, struct bio *bio);
706extern void mddev_suspend(struct mddev *mddev);
707extern void mddev_resume(struct mddev *mddev);
 
 
708extern struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
709				   struct mddev *mddev);
710
711extern void md_reload_sb(struct mddev *mddev, int raid_disk);
712extern void md_update_sb(struct mddev *mddev, int force);
713extern void md_kick_rdev_from_array(struct md_rdev * rdev);
714struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr);
715struct md_rdev *md_find_rdev_rcu(struct mddev *mddev, dev_t dev);
716
717static inline void rdev_dec_pending(struct md_rdev *rdev, struct mddev *mddev)
718{
719	int faulty = test_bit(Faulty, &rdev->flags);
720	if (atomic_dec_and_test(&rdev->nr_pending) && faulty) {
721		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
722		md_wakeup_thread(mddev->thread);
723	}
724}
725
726extern struct md_cluster_operations *md_cluster_ops;
727static inline int mddev_is_clustered(struct mddev *mddev)
728{
729	return mddev->cluster_info && mddev->bitmap_info.nodes > 1;
730}
731
732/* clear unsupported mddev_flags */
733static inline void mddev_clear_unsupported_flags(struct mddev *mddev,
734	unsigned long unsupported_flags)
735{
736	mddev->flags &= ~unsupported_flags;
737}
738
739static inline void mddev_check_writesame(struct mddev *mddev, struct bio *bio)
740{
741	if (bio_op(bio) == REQ_OP_WRITE_SAME &&
742	    !bio->bi_disk->queue->limits.max_write_same_sectors)
743		mddev->queue->limits.max_write_same_sectors = 0;
744}
745
746static inline void mddev_check_write_zeroes(struct mddev *mddev, struct bio *bio)
747{
748	if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
749	    !bio->bi_disk->queue->limits.max_write_zeroes_sectors)
750		mddev->queue->limits.max_write_zeroes_sectors = 0;
751}
752#endif /* _MD_MD_H */