Linux Audio

Check our new training course

Loading...
v3.1
 
  1/*
  2   md_k.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
 29typedef struct mddev_s mddev_t;
 30typedef struct mdk_rdev_s mdk_rdev_t;
 
 
 
 
 
 
 
 
 31
 32/* Bad block numbers are stored sorted in a single page.
 33 * 64bits is used for each block or extent.
 34 * 54 bits are sector number, 9 bits are extent size,
 35 * 1 bit is an 'acknowledged' flag.
 36 */
 37#define MD_MAX_BADBLOCKS	(PAGE_SIZE/8)
 
 
 
 
 38
 39/*
 40 * MD's 'extended' device
 41 */
 42struct mdk_rdev_s
 43{
 44	struct list_head same_set;	/* RAID devices within the same set */
 45
 46	sector_t sectors;		/* Device size (in 512bytes sectors) */
 47	mddev_t *mddev;			/* RAID array if running */
 48	int last_events;		/* IO event timestamp */
 49
 50	/*
 51	 * If meta_bdev is non-NULL, it means that a separate device is
 52	 * being used to store the metadata (superblock/bitmap) which
 53	 * would otherwise be contained on the same device as the data (bdev).
 54	 */
 55	struct block_device *meta_bdev;
 56	struct block_device *bdev;	/* block device handle */
 57
 58	struct page	*sb_page, *bb_page;
 59	int		sb_loaded;
 60	__u64		sb_events;
 61	sector_t	data_offset;	/* start of data in array */
 62	sector_t 	sb_start;	/* offset of the super block (in 512byte sectors) */
 
 63	int		sb_size;	/* bytes in the superblock */
 64	int		preferred_minor;	/* autorun support */
 65
 66	struct kobject	kobj;
 67
 68	/* A device can be in one of three states based on two flags:
 69	 * Not working:   faulty==1 in_sync==0
 70	 * Fully working: faulty==0 in_sync==1
 71	 * Working, but not
 72	 * in sync with array
 73	 *                faulty==0 in_sync==0
 74	 *
 75	 * It can never have faulty==1, in_sync==1
 76	 * This reduces the burden of testing multiple flags in many cases
 77	 */
 78
 79	unsigned long	flags;
 80#define	Faulty		1		/* device is known to have a fault */
 81#define	In_sync		2		/* device is in_sync with rest of array */
 82#define	WriteMostly	4		/* Avoid reading if at all possible */
 83#define	AutoDetected	7		/* added by auto-detect */
 84#define Blocked		8		/* An error occurred but has not yet
 85					 * been acknowledged by the metadata
 86					 * handler, so don't allow writes
 87					 * until it is cleared */
 88#define WriteErrorSeen	9		/* A write error has been seen on this
 89					 * device
 90					 */
 91#define FaultRecorded	10		/* Intermediate state for clearing
 92					 * Blocked.  The Fault is/will-be
 93					 * recorded in the metadata, but that
 94					 * metadata hasn't been stored safely
 95					 * on disk yet.
 96					 */
 97#define BlockedBadBlocks 11		/* A writer is blocked because they
 98					 * found an unacknowledged bad-block.
 99					 * This can safely be cleared at any
100					 * time, and the writer will re-check.
101					 * It may be set at any time, and at
102					 * worst the writer will timeout and
103					 * re-check.  So setting it as
104					 * accurately as possible is good, but
105					 * not absolutely critical.
106					 */
107	wait_queue_head_t blocked_wait;
108
109	int desc_nr;			/* descriptor index in the superblock */
110	int raid_disk;			/* role of device in array */
111	int new_raid_disk;		/* role that the device will have in
112					 * the array after a level-change completes.
113					 */
114	int saved_raid_disk;		/* role that device used to have in the
115					 * array and could again if we did a partial
116					 * resync from the bitmap
117					 */
118	sector_t	recovery_offset;/* If this device has been partially
 
119					 * recovered, this is where we were
120					 * up to.
121					 */
 
 
 
 
 
122
123	atomic_t	nr_pending;	/* number of pending requests.
124					 * only maintained for arrays that
125					 * support hot removal
126					 */
127	atomic_t	read_errors;	/* number of consecutive read errors that
128					 * we have tried to ignore.
129					 */
130	struct timespec last_read_error;	/* monotonic time since our
131						 * last read error
132						 */
133	atomic_t	corrected_errors; /* number of corrected read errors,
134					   * for reporting to userspace and storing
135					   * in superblock.
136					   */
 
 
 
137	struct work_struct del_work;	/* used for delayed sysfs removal */
138
139	struct sysfs_dirent *sysfs_state; /* handle for 'state'
140					   * sysfs entry */
 
 
 
 
 
141
142	struct badblocks {
143		int	count;		/* count of bad blocks */
144		int	unacked_exist;	/* there probably are unacknowledged
145					 * bad blocks.  This is only cleared
146					 * when a read discovers none
147					 */
148		int	shift;		/* shift from sectors to block size
149					 * a -ve shift means badblocks are
150					 * disabled.*/
151		u64	*page;		/* badblock list */
152		int	changed;
153		seqlock_t lock;
154
155		sector_t sector;
156		sector_t size;		/* in sectors */
157	} badblocks;
158};
159
160#define BB_LEN_MASK	(0x00000000000001FFULL)
161#define BB_OFFSET_MASK	(0x7FFFFFFFFFFFFE00ULL)
162#define BB_ACK_MASK	(0x8000000000000000ULL)
163#define BB_MAX_LEN	512
164#define BB_OFFSET(x)	(((x) & BB_OFFSET_MASK) >> 9)
165#define BB_LEN(x)	(((x) & BB_LEN_MASK) + 1)
166#define BB_ACK(x)	(!!((x) & BB_ACK_MASK))
167#define BB_MAKE(a, l, ack) (((a)<<9) | ((l)-1) | ((u64)(!!(ack)) << 63))
168
169extern int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
170			  sector_t *first_bad, int *bad_sectors);
171static inline int is_badblock(mdk_rdev_t *rdev, sector_t s, int sectors,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
172			      sector_t *first_bad, int *bad_sectors)
173{
174	if (unlikely(rdev->badblocks.count)) {
175		int rv = md_is_badblock(&rdev->badblocks, rdev->data_offset + s,
176					sectors,
177					first_bad, bad_sectors);
178		if (rv)
179			*first_bad -= rdev->data_offset;
180		return rv;
181	}
182	return 0;
183}
184extern int rdev_set_badblocks(mdk_rdev_t *rdev, sector_t s, int sectors,
185			      int acknowledged);
186extern int rdev_clear_badblocks(mdk_rdev_t *rdev, sector_t s, int sectors);
187extern void md_ack_all_badblocks(struct badblocks *bb);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
188
189struct mddev_s
190{
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
191	void				*private;
192	struct mdk_personality		*pers;
193	dev_t				unit;
194	int				md_minor;
195	struct list_head 		disks;
196	unsigned long			flags;
197#define MD_CHANGE_DEVS	0	/* Some device status has changed */
198#define MD_CHANGE_CLEAN 1	/* transition to or from 'clean' */
199#define MD_CHANGE_PENDING 2	/* switch from 'clean' to 'active' in progress */
200#define MD_ARRAY_FIRST_USE 3    /* First use of array, needs initialization */
201
202	int				suspended;
203	atomic_t			active_io;
204	int				ro;
205	int				sysfs_active; /* set when sysfs deletes
206						       * are happening, so run/
207						       * takeover/stop are not safe
208						       */
209	int				ready; /* See when safe to pass 
210						* IO requests down */
211	struct gendisk			*gendisk;
212
213	struct kobject			kobj;
214	int				hold_active;
215#define	UNTIL_IOCTL	1
216#define	UNTIL_STOP	2
217
218	/* Superblock information */
219	int				major_version,
220					minor_version,
221					patch_version;
222	int				persistent;
223	int 				external;	/* metadata is
224							 * managed externally */
225	char				metadata_type[17]; /* externally set*/
226	int				chunk_sectors;
227	time_t				ctime, utime;
228	int				level, layout;
229	char				clevel[16];
230	int				raid_disks;
231	int				max_disks;
232	sector_t			dev_sectors; 	/* used size of
233							 * component devices */
234	sector_t			array_sectors; /* exported array size */
235	int				external_size; /* size managed
236							* externally */
237	__u64				events;
238	/* If the last 'event' was simply a clean->dirty transition, and
239	 * we didn't write it to the spares, then it is safe and simple
240	 * to just decrement the event count on a dirty->clean transition.
241	 * So we record that possibility here.
242	 */
243	int				can_decrease_events;
244
245	char				uuid[16];
246
247	/* If the array is being reshaped, we need to record the
248	 * new shape and an indication of where we are up to.
249	 * This is written to the superblock.
250	 * If reshape_position is MaxSector, then no reshape is happening (yet).
251	 */
252	sector_t			reshape_position;
253	int				delta_disks, new_level, new_layout;
254	int				new_chunk_sectors;
 
255
256	atomic_t			plug_cnt;	/* If device is expecting
257							 * more bios soon.
258							 */
259	struct mdk_thread_s		*thread;	/* management thread */
260	struct mdk_thread_s		*sync_thread;	/* doing resync or reconstruct */
 
 
 
 
 
261	sector_t			curr_resync;	/* last block scheduled */
262	/* As resync requests can complete out of order, we cannot easily track
263	 * how much resync has been completed.  So we occasionally pause until
264	 * everything completes, then set curr_resync_completed to curr_resync.
265	 * As such it may be well behind the real resync mark, but it is a value
266	 * we are certain of.
267	 */
268	sector_t			curr_resync_completed;
269	unsigned long			resync_mark;	/* a recent timestamp */
270	sector_t			resync_mark_cnt;/* blocks written at resync_mark */
271	sector_t			curr_mark_cnt; /* blocks scheduled now */
272
273	sector_t			resync_max_sectors; /* may be set by personality */
274
275	sector_t			resync_mismatches; /* count of sectors where
276							    * parity/replica mismatch found
277							    */
278
279	/* allow user-space to request suspension of IO to regions of the array */
280	sector_t			suspend_lo;
281	sector_t			suspend_hi;
282	/* if zero, use the system-wide default */
283	int				sync_speed_min;
284	int				sync_speed_max;
285
286	/* resync even though the same disks are shared among md-devices */
287	int				parallel_resync;
288
289	int				ok_start_degraded;
290	/* recovery/resync flags 
291	 * NEEDED:   we might need to start a resync/recover
292	 * RUNNING:  a thread is running, or about to be started
293	 * SYNC:     actually doing a resync, not a recovery
294	 * RECOVER:  doing recovery, or need to try it.
295	 * INTR:     resync needs to be aborted for some reason
296	 * DONE:     thread is done and is waiting to be reaped
297	 * REQUEST:  user-space has requested a sync (used with SYNC)
298	 * CHECK:    user-space request for check-only, no repair
299	 * RESHAPE:  A reshape is happening
300	 *
301	 * If neither SYNC or RESHAPE are set, then it is a recovery.
302	 */
303#define	MD_RECOVERY_RUNNING	0
304#define	MD_RECOVERY_SYNC	1
305#define	MD_RECOVERY_RECOVER	2
306#define	MD_RECOVERY_INTR	3
307#define	MD_RECOVERY_DONE	4
308#define	MD_RECOVERY_NEEDED	5
309#define	MD_RECOVERY_REQUESTED	6
310#define	MD_RECOVERY_CHECK	7
311#define MD_RECOVERY_RESHAPE	8
312#define	MD_RECOVERY_FROZEN	9
313
314	unsigned long			recovery;
315	/* If a RAID personality determines that recovery (of a particular
316	 * device) will fail due to a read error on the source device, it
317	 * takes a copy of this number and does not attempt recovery again
318	 * until this number changes.
319	 */
320	int				recovery_disabled;
321
322	int				in_sync;	/* know to not need resync */
323	/* 'open_mutex' avoids races between 'md_open' and 'do_md_stop', so
324	 * that we are never stopping an array while it is open.
325	 * 'reconfig_mutex' protects all other reconfiguration.
326	 * These locks are separate due to conflicting interactions
327	 * with bdev->bd_mutex.
328	 * Lock ordering is:
329	 *  reconfig_mutex -> bd_mutex : e.g. do_md_run -> revalidate_disk
330	 *  bd_mutex -> open_mutex:  e.g. __blkdev_get -> md_open
331	 */
332	struct mutex			open_mutex;
333	struct mutex			reconfig_mutex;
334	atomic_t			active;		/* general refcount */
335	atomic_t			openers;	/* number of active opens */
336
337	int				changed;	/* True if we might need to
338							 * reread partition info */
339	int				degraded;	/* whether md should consider
340							 * adding a spare
341							 */
342
343	atomic_t			recovery_active; /* blocks scheduled, but not written */
344	wait_queue_head_t		recovery_wait;
345	sector_t			recovery_cp;
346	sector_t			resync_min;	/* user requested sync
347							 * starts here */
348	sector_t			resync_max;	/* resync should pause
349							 * when it gets here */
350
351	struct sysfs_dirent		*sysfs_state;	/* handle for 'array_state'
352							 * file in sysfs.
353							 */
354	struct sysfs_dirent		*sysfs_action;  /* handle for 'sync_action' */
 
 
 
355
356	struct work_struct del_work;	/* used for delayed sysfs removal */
357
358	spinlock_t			write_lock;
 
 
 
 
 
 
 
 
 
 
 
359	wait_queue_head_t		sb_wait;	/* for waiting on superblock updates */
360	atomic_t			pending_writes;	/* number of active superblock writes */
361
362	unsigned int			safemode;	/* if set, update "clean" superblock
363							 * when no writes pending.
364							 */ 
365	unsigned int			safemode_delay;
366	struct timer_list		safemode_timer;
367	atomic_t			writes_pending; 
 
368	struct request_queue		*queue;	/* for plugging ... */
369
370	struct bitmap                   *bitmap; /* the bitmap for the device */
371	struct {
372		struct file		*file; /* the bitmap file */
373		loff_t			offset; /* offset from superblock of
374						 * start of bitmap. May be
375						 * negative, but not '0'
376						 * For external metadata, offset
377						 * from start of device. 
378						 */
 
379		loff_t			default_offset; /* this is the offset to use when
380							 * hot-adding a bitmap.  It should
381							 * eventually be settable by sysfs.
382							 */
 
 
383		struct mutex		mutex;
384		unsigned long		chunksize;
385		unsigned long		daemon_sleep; /* how many jiffies between updates? */
386		unsigned long		max_write_behind; /* write-behind mode */
387		int			external;
 
 
388	} bitmap_info;
389
390	atomic_t 			max_corr_read_errors; /* max read retries */
391	struct list_head		all_mddevs;
392
393	struct attribute_group		*to_remove;
394
395	struct bio_set			*bio_set;
 
 
 
 
396
397	/* Generic flush handling.
398	 * The last to finish preflush schedules a worker to submit
399	 * the rest of the request (without the REQ_FLUSH flag).
400	 */
401	struct bio *flush_bio;
402	atomic_t flush_pending;
 
 
 
403	struct work_struct flush_work;
404	struct work_struct event_work;	/* used by dm to report failure event */
405	void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 
 
 
 
 
 
 
 
406};
407
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
408
409static inline void rdev_dec_pending(mdk_rdev_t *rdev, mddev_t *mddev)
410{
411	int faulty = test_bit(Faulty, &rdev->flags);
412	if (atomic_dec_and_test(&rdev->nr_pending) && faulty)
413		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
 
 
 
 
 
 
414}
415
 
 
 
 
 
 
416static inline void md_sync_acct(struct block_device *bdev, unsigned long nr_sectors)
417{
418        atomic_add(nr_sectors, &bdev->bd_contains->bd_disk->sync_io);
 
 
 
 
 
419}
420
421struct mdk_personality
422{
423	char *name;
424	int level;
425	struct list_head list;
426	struct module *owner;
427	int (*make_request)(mddev_t *mddev, struct bio *bio);
428	int (*run)(mddev_t *mddev);
429	int (*stop)(mddev_t *mddev);
430	void (*status)(struct seq_file *seq, mddev_t *mddev);
 
 
 
 
 
 
431	/* error_handler must set ->faulty and clear ->in_sync
432	 * if appropriate, and should abort recovery if needed 
433	 */
434	void (*error_handler)(mddev_t *mddev, mdk_rdev_t *rdev);
435	int (*hot_add_disk) (mddev_t *mddev, mdk_rdev_t *rdev);
436	int (*hot_remove_disk) (mddev_t *mddev, int number);
437	int (*spare_active) (mddev_t *mddev);
438	sector_t (*sync_request)(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster);
439	int (*resize) (mddev_t *mddev, sector_t sectors);
440	sector_t (*size) (mddev_t *mddev, sector_t sectors, int raid_disks);
441	int (*check_reshape) (mddev_t *mddev);
442	int (*start_reshape) (mddev_t *mddev);
443	void (*finish_reshape) (mddev_t *mddev);
444	/* quiesce moves between quiescence states
445	 * 0 - fully active
446	 * 1 - no new requests allowed
447	 * others - reserved
448	 */
449	void (*quiesce) (mddev_t *mddev, int state);
450	/* takeover is used to transition an array from one
451	 * personality to another.  The new personality must be able
452	 * to handle the data in the current layout.
453	 * e.g. 2drive raid1 -> 2drive raid5
454	 *      ndrive raid5 -> degraded n+1drive raid6 with special layout
455	 * If the takeover succeeds, a new 'private' structure is returned.
456	 * This needs to be installed and then ->run used to activate the
457	 * array.
458	 */
459	void *(*takeover) (mddev_t *mddev);
 
 
460};
461
462
463struct md_sysfs_entry {
464	struct attribute attr;
465	ssize_t (*show)(mddev_t *, char *);
466	ssize_t (*store)(mddev_t *, const char *, size_t);
467};
468extern struct attribute_group md_bitmap_group;
469
470static inline struct sysfs_dirent *sysfs_get_dirent_safe(struct sysfs_dirent *sd, char *name)
471{
472	if (sd)
473		return sysfs_get_dirent(sd, NULL, name);
474	return sd;
475}
476static inline void sysfs_notify_dirent_safe(struct sysfs_dirent *sd)
477{
478	if (sd)
479		sysfs_notify_dirent(sd);
480}
481
482static inline char * mdname (mddev_t * mddev)
483{
484	return mddev->gendisk ? mddev->gendisk->disk_name : "mdX";
485}
486
487static inline int sysfs_link_rdev(mddev_t *mddev, mdk_rdev_t *rdev)
488{
489	char nm[20];
490	sprintf(nm, "rd%d", rdev->raid_disk);
491	return sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
 
 
 
 
 
492}
493
494static inline void sysfs_unlink_rdev(mddev_t *mddev, mdk_rdev_t *rdev)
495{
496	char nm[20];
497	sprintf(nm, "rd%d", rdev->raid_disk);
498	sysfs_remove_link(&mddev->kobj, nm);
 
 
 
 
499}
500
501/*
502 * iterates through some rdev ringlist. It's safe to remove the
503 * current 'rdev'. Dont touch 'tmp' though.
504 */
505#define rdev_for_each_list(rdev, tmp, head)				\
506	list_for_each_entry_safe(rdev, tmp, head, same_set)
507
508/*
509 * iterates through the 'same array disks' ringlist
510 */
511#define rdev_for_each(rdev, tmp, mddev)				\
 
 
 
512	list_for_each_entry_safe(rdev, tmp, &((mddev)->disks), same_set)
513
514#define rdev_for_each_rcu(rdev, mddev)				\
515	list_for_each_entry_rcu(rdev, &((mddev)->disks), same_set)
516
517typedef struct mdk_thread_s {
518	void			(*run) (mddev_t *mddev);
519	mddev_t			*mddev;
520	wait_queue_head_t	wqueue;
521	unsigned long           flags;
522	struct task_struct	*tsk;
523	unsigned long		timeout;
524} mdk_thread_t;
 
525
526#define THREAD_WAKEUP  0
527
528#define __wait_event_lock_irq(wq, condition, lock, cmd) 		\
529do {									\
530	wait_queue_t __wait;						\
531	init_waitqueue_entry(&__wait, current);				\
532									\
533	add_wait_queue(&wq, &__wait);					\
534	for (;;) {							\
535		set_current_state(TASK_UNINTERRUPTIBLE);		\
536		if (condition)						\
537			break;						\
538		spin_unlock_irq(&lock);					\
539		cmd;							\
540		schedule();						\
541		spin_lock_irq(&lock);					\
542	}								\
543	current->state = TASK_RUNNING;					\
544	remove_wait_queue(&wq, &__wait);				\
545} while (0)
546
547#define wait_event_lock_irq(wq, condition, lock, cmd) 			\
548do {									\
549	if (condition)	 						\
550		break;							\
551	__wait_event_lock_irq(wq, condition, lock, cmd);		\
552} while (0)
553
554static inline void safe_put_page(struct page *p)
555{
556	if (p) put_page(p);
557}
558
559extern int register_md_personality(struct mdk_personality *p);
560extern int unregister_md_personality(struct mdk_personality *p);
561extern mdk_thread_t * md_register_thread(void (*run) (mddev_t *mddev),
562				mddev_t *mddev, const char *name);
563extern void md_unregister_thread(mdk_thread_t **threadp);
564extern void md_wakeup_thread(mdk_thread_t *thread);
565extern void md_check_recovery(mddev_t *mddev);
566extern void md_write_start(mddev_t *mddev, struct bio *bi);
567extern void md_write_end(mddev_t *mddev);
568extern void md_done_sync(mddev_t *mddev, int blocks, int ok);
569extern void md_error(mddev_t *mddev, mdk_rdev_t *rdev);
570
571extern int mddev_congested(mddev_t *mddev, int bits);
572extern void md_flush_request(mddev_t *mddev, struct bio *bio);
573extern void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
 
 
 
 
 
 
 
 
 
 
574			   sector_t sector, int size, struct page *page);
575extern void md_super_wait(mddev_t *mddev);
576extern int sync_page_io(mdk_rdev_t *rdev, sector_t sector, int size, 
577			struct page *page, int rw, bool metadata_op);
578extern void md_do_sync(mddev_t *mddev);
579extern void md_new_event(mddev_t *mddev);
580extern int md_allow_write(mddev_t *mddev);
581extern void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev);
582extern void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors);
583extern int md_check_no_bitmap(mddev_t *mddev);
584extern int md_integrity_register(mddev_t *mddev);
585extern void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev);
 
586extern int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale);
587extern void restore_bitmap_write_access(struct file *file);
588
589extern void mddev_init(mddev_t *mddev);
590extern int md_run(mddev_t *mddev);
591extern void md_stop(mddev_t *mddev);
592extern void md_stop_writes(mddev_t *mddev);
593extern int md_rdev_init(mdk_rdev_t *rdev);
594
595extern void mddev_suspend(mddev_t *mddev);
596extern void mddev_resume(mddev_t *mddev);
597extern struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
598				   mddev_t *mddev);
 
599extern struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
600				   mddev_t *mddev);
601extern int mddev_check_plugged(mddev_t *mddev);
602extern void md_trim_bio(struct bio *bio, int offset, int size);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
603#endif /* _MD_MD_H */
v5.9
  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/*
 37 * The struct embedded in rdev is used to serialize IO.
 
 
 38 */
 39struct serial_in_rdev {
 40	struct rb_root_cached serial_rb;
 41	spinlock_t serial_lock;
 42	wait_queue_head_t serial_io_wait;
 43};
 44
 45/*
 46 * MD's 'extended' device
 47 */
 48struct md_rdev {
 
 49	struct list_head same_set;	/* RAID devices within the same set */
 50
 51	sector_t sectors;		/* Device size (in 512bytes sectors) */
 52	struct mddev *mddev;		/* RAID array if running */
 53	int last_events;		/* IO event timestamp */
 54
 55	/*
 56	 * If meta_bdev is non-NULL, it means that a separate device is
 57	 * being used to store the metadata (superblock/bitmap) which
 58	 * would otherwise be contained on the same device as the data (bdev).
 59	 */
 60	struct block_device *meta_bdev;
 61	struct block_device *bdev;	/* block device handle */
 62
 63	struct page	*sb_page, *bb_page;
 64	int		sb_loaded;
 65	__u64		sb_events;
 66	sector_t	data_offset;	/* start of data in array */
 67	sector_t	new_data_offset;/* only relevant while reshaping */
 68	sector_t	sb_start;	/* offset of the super block (in 512byte sectors) */
 69	int		sb_size;	/* bytes in the superblock */
 70	int		preferred_minor;	/* autorun support */
 71
 72	struct kobject	kobj;
 73
 74	/* A device can be in one of three states based on two flags:
 75	 * Not working:   faulty==1 in_sync==0
 76	 * Fully working: faulty==0 in_sync==1
 77	 * Working, but not
 78	 * in sync with array
 79	 *                faulty==0 in_sync==0
 80	 *
 81	 * It can never have faulty==1, in_sync==1
 82	 * This reduces the burden of testing multiple flags in many cases
 83	 */
 84
 85	unsigned long	flags;	/* bit set of 'enum flag_bits' bits. */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 86	wait_queue_head_t blocked_wait;
 87
 88	int desc_nr;			/* descriptor index in the superblock */
 89	int raid_disk;			/* role of device in array */
 90	int new_raid_disk;		/* role that the device will have in
 91					 * the array after a level-change completes.
 92					 */
 93	int saved_raid_disk;		/* role that device used to have in the
 94					 * array and could again if we did a partial
 95					 * resync from the bitmap
 96					 */
 97	union {
 98		sector_t recovery_offset;/* If this device has been partially
 99					 * recovered, this is where we were
100					 * up to.
101					 */
102		sector_t journal_tail;	/* If this device is a journal device,
103					 * this is the journal tail (journal
104					 * recovery start point)
105					 */
106	};
107
108	atomic_t	nr_pending;	/* number of pending requests.
109					 * only maintained for arrays that
110					 * support hot removal
111					 */
112	atomic_t	read_errors;	/* number of consecutive read errors that
113					 * we have tried to ignore.
114					 */
115	time64_t	last_read_error;	/* monotonic time since our
116						 * last read error
117						 */
118	atomic_t	corrected_errors; /* number of corrected read errors,
119					   * for reporting to userspace and storing
120					   * in superblock.
121					   */
122
123	struct serial_in_rdev *serial;  /* used for raid1 io serialization */
124
125	struct work_struct del_work;	/* used for delayed sysfs removal */
126
127	struct kernfs_node *sysfs_state; /* handle for 'state'
128					   * sysfs entry */
129	/* handle for 'unacknowledged_bad_blocks' sysfs dentry */
130	struct kernfs_node *sysfs_unack_badblocks;
131	/* handle for 'bad_blocks' sysfs dentry */
132	struct kernfs_node *sysfs_badblocks;
133	struct badblocks badblocks;
134
135	struct {
136		short offset;	/* Offset from superblock to start of PPL.
137				 * Not used by external metadata. */
138		unsigned int size;	/* Size in sectors of the PPL space */
139		sector_t sector;	/* First sector of the PPL space */
140	} ppl;
141};
142enum flag_bits {
143	Faulty,			/* device is known to have a fault */
144	In_sync,		/* device is in_sync with rest of array */
145	Bitmap_sync,		/* ..actually, not quite In_sync.  Need a
146				 * bitmap-based recovery to get fully in sync.
147				 * The bit is only meaningful before device
148				 * has been passed to pers->hot_add_disk.
149				 */
150	WriteMostly,		/* Avoid reading if at all possible */
151	AutoDetected,		/* added by auto-detect */
152	Blocked,		/* An error occurred but has not yet
153				 * been acknowledged by the metadata
154				 * handler, so don't allow writes
155				 * until it is cleared */
156	WriteErrorSeen,		/* A write error has been seen on this
157				 * device
158				 */
159	FaultRecorded,		/* Intermediate state for clearing
160				 * Blocked.  The Fault is/will-be
161				 * recorded in the metadata, but that
162				 * metadata hasn't been stored safely
163				 * on disk yet.
164				 */
165	BlockedBadBlocks,	/* A writer is blocked because they
166				 * found an unacknowledged bad-block.
167				 * This can safely be cleared at any
168				 * time, and the writer will re-check.
169				 * It may be set at any time, and at
170				 * worst the writer will timeout and
171				 * re-check.  So setting it as
172				 * accurately as possible is good, but
173				 * not absolutely critical.
174				 */
175	WantReplacement,	/* This device is a candidate to be
176				 * hot-replaced, either because it has
177				 * reported some faults, or because
178				 * of explicit request.
179				 */
180	Replacement,		/* This device is a replacement for
181				 * a want_replacement device with same
182				 * raid_disk number.
183				 */
184	Candidate,		/* For clustered environments only:
185				 * This device is seen locally but not
186				 * by the whole cluster
187				 */
188	Journal,		/* This device is used as journal for
189				 * raid-5/6.
190				 * Usually, this device should be faster
191				 * than other devices in the array
192				 */
193	ClusterRemove,
194	RemoveSynchronized,	/* synchronize_rcu() was called after
195				 * this device was known to be faulty,
196				 * so it is safe to remove without
197				 * another synchronize_rcu() call.
198				 */
199	ExternalBbl,            /* External metadata provides bad
200				 * block management for a disk
201				 */
202	FailFast,		/* Minimal retries should be attempted on
203				 * this device, so use REQ_FAILFAST_DEV.
204				 * Also don't try to repair failed reads.
205				 * It is expects that no bad block log
206				 * is present.
207				 */
208	LastDev,		/* Seems to be the last working dev as
209				 * it didn't fail, so don't use FailFast
210				 * any more for metadata
211				 */
212	CollisionCheck,		/*
213				 * check if there is collision between raid1
214				 * serial bios.
215				 */
216};
217
218static inline int is_badblock(struct md_rdev *rdev, sector_t s, int sectors,
219			      sector_t *first_bad, int *bad_sectors)
220{
221	if (unlikely(rdev->badblocks.count)) {
222		int rv = badblocks_check(&rdev->badblocks, rdev->data_offset + s,
223					sectors,
224					first_bad, bad_sectors);
225		if (rv)
226			*first_bad -= rdev->data_offset;
227		return rv;
228	}
229	return 0;
230}
231extern int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
232			      int is_new);
233extern int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
234				int is_new);
235struct md_cluster_info;
236
237/* change UNSUPPORTED_MDDEV_FLAGS for each array type if new flag is added */
238enum mddev_flags {
239	MD_ARRAY_FIRST_USE,	/* First use of array, needs initialization */
240	MD_CLOSING,		/* If set, we are closing the array, do not open
241				 * it then */
242	MD_JOURNAL_CLEAN,	/* A raid with journal is already clean */
243	MD_HAS_JOURNAL,		/* The raid array has journal feature set */
244	MD_CLUSTER_RESYNC_LOCKED, /* cluster raid only, which means node
245				   * already took resync lock, need to
246				   * release the lock */
247	MD_FAILFAST_SUPPORTED,	/* Using MD_FAILFAST on metadata writes is
248				 * supported as calls to md_error() will
249				 * never cause the array to become failed.
250				 */
251	MD_HAS_PPL,		/* The raid array has PPL feature set */
252	MD_HAS_MULTIPLE_PPLS,	/* The raid array has multiple PPLs feature set */
253	MD_ALLOW_SB_UPDATE,	/* md_check_recovery is allowed to update
254				 * the metadata without taking reconfig_mutex.
255				 */
256	MD_UPDATING_SB,		/* md_check_recovery is updating the metadata
257				 * without explicitly holding reconfig_mutex.
258				 */
259	MD_NOT_READY,		/* do_md_run() is active, so 'array_state'
260				 * must not report that array is ready yet
261				 */
262	MD_BROKEN,              /* This is used in RAID-0/LINEAR only, to stop
263				 * I/O in case an array member is gone/failed.
264				 */
265};
266
267enum mddev_sb_flags {
268	MD_SB_CHANGE_DEVS,		/* Some device status has changed */
269	MD_SB_CHANGE_CLEAN,	/* transition to or from 'clean' */
270	MD_SB_CHANGE_PENDING,	/* switch from 'clean' to 'active' in progress */
271	MD_SB_NEED_REWRITE,	/* metadata write needs to be repeated */
272};
273
274#define NR_SERIAL_INFOS		8
275/* record current range of serialize IOs */
276struct serial_info {
277	struct rb_node node;
278	sector_t start;		/* start sector of rb node */
279	sector_t last;		/* end sector of rb node */
280	sector_t _subtree_last; /* highest sector in subtree of rb node */
281};
282
283struct mddev {
284	void				*private;
285	struct md_personality		*pers;
286	dev_t				unit;
287	int				md_minor;
288	struct list_head		disks;
289	unsigned long			flags;
290	unsigned long			sb_flags;
 
 
 
291
292	int				suspended;
293	atomic_t			active_io;
294	int				ro;
295	int				sysfs_active; /* set when sysfs deletes
296						       * are happening, so run/
297						       * takeover/stop are not safe
298						       */
 
 
299	struct gendisk			*gendisk;
300
301	struct kobject			kobj;
302	int				hold_active;
303#define	UNTIL_IOCTL	1
304#define	UNTIL_STOP	2
305
306	/* Superblock information */
307	int				major_version,
308					minor_version,
309					patch_version;
310	int				persistent;
311	int				external;	/* metadata is
312							 * managed externally */
313	char				metadata_type[17]; /* externally set*/
314	int				chunk_sectors;
315	time64_t			ctime, utime;
316	int				level, layout;
317	char				clevel[16];
318	int				raid_disks;
319	int				max_disks;
320	sector_t			dev_sectors;	/* used size of
321							 * component devices */
322	sector_t			array_sectors; /* exported array size */
323	int				external_size; /* size managed
324							* externally */
325	__u64				events;
326	/* If the last 'event' was simply a clean->dirty transition, and
327	 * we didn't write it to the spares, then it is safe and simple
328	 * to just decrement the event count on a dirty->clean transition.
329	 * So we record that possibility here.
330	 */
331	int				can_decrease_events;
332
333	char				uuid[16];
334
335	/* If the array is being reshaped, we need to record the
336	 * new shape and an indication of where we are up to.
337	 * This is written to the superblock.
338	 * If reshape_position is MaxSector, then no reshape is happening (yet).
339	 */
340	sector_t			reshape_position;
341	int				delta_disks, new_level, new_layout;
342	int				new_chunk_sectors;
343	int				reshape_backwards;
344
345	struct md_thread		*thread;	/* management thread */
346	struct md_thread		*sync_thread;	/* doing resync or reconstruct */
347
348	/* 'last_sync_action' is initialized to "none".  It is set when a
349	 * sync operation (i.e "data-check", "requested-resync", "resync",
350	 * "recovery", or "reshape") is started.  It holds this value even
351	 * when the sync thread is "frozen" (interrupted) or "idle" (stopped
352	 * or finished).  It is overwritten when a new sync operation is begun.
353	 */
354	char				*last_sync_action;
355	sector_t			curr_resync;	/* last block scheduled */
356	/* As resync requests can complete out of order, we cannot easily track
357	 * how much resync has been completed.  So we occasionally pause until
358	 * everything completes, then set curr_resync_completed to curr_resync.
359	 * As such it may be well behind the real resync mark, but it is a value
360	 * we are certain of.
361	 */
362	sector_t			curr_resync_completed;
363	unsigned long			resync_mark;	/* a recent timestamp */
364	sector_t			resync_mark_cnt;/* blocks written at resync_mark */
365	sector_t			curr_mark_cnt; /* blocks scheduled now */
366
367	sector_t			resync_max_sectors; /* may be set by personality */
368
369	atomic64_t			resync_mismatches; /* count of sectors where
370							    * parity/replica mismatch found
371							    */
372
373	/* allow user-space to request suspension of IO to regions of the array */
374	sector_t			suspend_lo;
375	sector_t			suspend_hi;
376	/* if zero, use the system-wide default */
377	int				sync_speed_min;
378	int				sync_speed_max;
379
380	/* resync even though the same disks are shared among md-devices */
381	int				parallel_resync;
382
383	int				ok_start_degraded;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
384
385	unsigned long			recovery;
386	/* If a RAID personality determines that recovery (of a particular
387	 * device) will fail due to a read error on the source device, it
388	 * takes a copy of this number and does not attempt recovery again
389	 * until this number changes.
390	 */
391	int				recovery_disabled;
392
393	int				in_sync;	/* know to not need resync */
394	/* 'open_mutex' avoids races between 'md_open' and 'do_md_stop', so
395	 * that we are never stopping an array while it is open.
396	 * 'reconfig_mutex' protects all other reconfiguration.
397	 * These locks are separate due to conflicting interactions
398	 * with bdev->bd_mutex.
399	 * Lock ordering is:
400	 *  reconfig_mutex -> bd_mutex : e.g. do_md_run -> revalidate_disk
401	 *  bd_mutex -> open_mutex:  e.g. __blkdev_get -> md_open
402	 */
403	struct mutex			open_mutex;
404	struct mutex			reconfig_mutex;
405	atomic_t			active;		/* general refcount */
406	atomic_t			openers;	/* number of active opens */
407
408	int				changed;	/* True if we might need to
409							 * reread partition info */
410	int				degraded;	/* whether md should consider
411							 * adding a spare
412							 */
413
414	atomic_t			recovery_active; /* blocks scheduled, but not written */
415	wait_queue_head_t		recovery_wait;
416	sector_t			recovery_cp;
417	sector_t			resync_min;	/* user requested sync
418							 * starts here */
419	sector_t			resync_max;	/* resync should pause
420							 * when it gets here */
421
422	struct kernfs_node		*sysfs_state;	/* handle for 'array_state'
423							 * file in sysfs.
424							 */
425	struct kernfs_node		*sysfs_action;  /* handle for 'sync_action' */
426	struct kernfs_node		*sysfs_completed;	/*handle for 'sync_completed' */
427	struct kernfs_node		*sysfs_degraded;	/*handle for 'degraded' */
428	struct kernfs_node		*sysfs_level;		/*handle for 'level' */
429
430	struct work_struct del_work;	/* used for delayed sysfs removal */
431
432	/* "lock" protects:
433	 *   flush_bio transition from NULL to !NULL
434	 *   rdev superblocks, events
435	 *   clearing MD_CHANGE_*
436	 *   in_sync - and related safemode and MD_CHANGE changes
437	 *   pers (also protected by reconfig_mutex and pending IO).
438	 *   clearing ->bitmap
439	 *   clearing ->bitmap_info.file
440	 *   changing ->resync_{min,max}
441	 *   setting MD_RECOVERY_RUNNING (which interacts with resync_{min,max})
442	 */
443	spinlock_t			lock;
444	wait_queue_head_t		sb_wait;	/* for waiting on superblock updates */
445	atomic_t			pending_writes;	/* number of active superblock writes */
446
447	unsigned int			safemode;	/* if set, update "clean" superblock
448							 * when no writes pending.
449							 */
450	unsigned int			safemode_delay;
451	struct timer_list		safemode_timer;
452	struct percpu_ref		writes_pending;
453	int				sync_checkers;	/* # of threads checking writes_pending */
454	struct request_queue		*queue;	/* for plugging ... */
455
456	struct bitmap			*bitmap; /* the bitmap for the device */
457	struct {
458		struct file		*file; /* the bitmap file */
459		loff_t			offset; /* offset from superblock of
460						 * start of bitmap. May be
461						 * negative, but not '0'
462						 * For external metadata, offset
463						 * from start of device.
464						 */
465		unsigned long		space; /* space available at this offset */
466		loff_t			default_offset; /* this is the offset to use when
467							 * hot-adding a bitmap.  It should
468							 * eventually be settable by sysfs.
469							 */
470		unsigned long		default_space; /* space available at
471							* default offset */
472		struct mutex		mutex;
473		unsigned long		chunksize;
474		unsigned long		daemon_sleep; /* how many jiffies between updates? */
475		unsigned long		max_write_behind; /* write-behind mode */
476		int			external;
477		int			nodes; /* Maximum number of nodes in the cluster */
478		char                    cluster_name[64]; /* Name of the cluster */
479	} bitmap_info;
480
481	atomic_t			max_corr_read_errors; /* max read retries */
482	struct list_head		all_mddevs;
483
484	struct attribute_group		*to_remove;
485
486	struct bio_set			bio_set;
487	struct bio_set			sync_set; /* for sync operations like
488						   * metadata and bitmap writes
489						   */
490	mempool_t			md_io_pool;
491
492	/* Generic flush handling.
493	 * The last to finish preflush schedules a worker to submit
494	 * the rest of the request (without the REQ_PREFLUSH flag).
495	 */
496	struct bio *flush_bio;
497	atomic_t flush_pending;
498	ktime_t start_flush, last_flush; /* last_flush is when the last completed
499					  * flush was started.
500					  */
501	struct work_struct flush_work;
502	struct work_struct event_work;	/* used by dm to report failure event */
503	mempool_t *serial_info_pool;
504	void (*sync_super)(struct mddev *mddev, struct md_rdev *rdev);
505	struct md_cluster_info		*cluster_info;
506	unsigned int			good_device_nr;	/* good device num within cluster raid */
507	unsigned int			noio_flag; /* for memalloc scope API */
508
509	bool	has_superblocks:1;
510	bool	fail_last_dev:1;
511	bool	serialize_policy:1;
512};
513
514enum recovery_flags {
515	/*
516	 * If neither SYNC or RESHAPE are set, then it is a recovery.
517	 */
518	MD_RECOVERY_RUNNING,	/* a thread is running, or about to be started */
519	MD_RECOVERY_SYNC,	/* actually doing a resync, not a recovery */
520	MD_RECOVERY_RECOVER,	/* doing recovery, or need to try it. */
521	MD_RECOVERY_INTR,	/* resync needs to be aborted for some reason */
522	MD_RECOVERY_DONE,	/* thread is done and is waiting to be reaped */
523	MD_RECOVERY_NEEDED,	/* we might need to start a resync/recover */
524	MD_RECOVERY_REQUESTED,	/* user-space has requested a sync (used with SYNC) */
525	MD_RECOVERY_CHECK,	/* user-space request for check-only, no repair */
526	MD_RECOVERY_RESHAPE,	/* A reshape is happening */
527	MD_RECOVERY_FROZEN,	/* User request to abort, and not restart, any action */
528	MD_RECOVERY_ERROR,	/* sync-action interrupted because io-error */
529	MD_RECOVERY_WAIT,	/* waiting for pers->start() to finish */
530	MD_RESYNCING_REMOTE,	/* remote node is running resync thread */
531};
532
533static inline int __must_check mddev_lock(struct mddev *mddev)
534{
535	return mutex_lock_interruptible(&mddev->reconfig_mutex);
536}
537
538/* Sometimes we need to take the lock in a situation where
539 * failure due to interrupts is not acceptable.
540 */
541static inline void mddev_lock_nointr(struct mddev *mddev)
542{
543	mutex_lock(&mddev->reconfig_mutex);
544}
545
546static inline int mddev_trylock(struct mddev *mddev)
547{
548	return mutex_trylock(&mddev->reconfig_mutex);
549}
550extern void mddev_unlock(struct mddev *mddev);
551
552static inline void md_sync_acct(struct block_device *bdev, unsigned long nr_sectors)
553{
554	atomic_add(nr_sectors, &bdev->bd_contains->bd_disk->sync_io);
555}
556
557static inline void md_sync_acct_bio(struct bio *bio, unsigned long nr_sectors)
558{
559	atomic_add(nr_sectors, &bio->bi_disk->sync_io);
560}
561
562struct md_personality
563{
564	char *name;
565	int level;
566	struct list_head list;
567	struct module *owner;
568	bool __must_check (*make_request)(struct mddev *mddev, struct bio *bio);
569	/*
570	 * start up works that do NOT require md_thread. tasks that
571	 * requires md_thread should go into start()
572	 */
573	int (*run)(struct mddev *mddev);
574	/* start up works that require md threads */
575	int (*start)(struct mddev *mddev);
576	void (*free)(struct mddev *mddev, void *priv);
577	void (*status)(struct seq_file *seq, struct mddev *mddev);
578	/* error_handler must set ->faulty and clear ->in_sync
579	 * if appropriate, and should abort recovery if needed
580	 */
581	void (*error_handler)(struct mddev *mddev, struct md_rdev *rdev);
582	int (*hot_add_disk) (struct mddev *mddev, struct md_rdev *rdev);
583	int (*hot_remove_disk) (struct mddev *mddev, struct md_rdev *rdev);
584	int (*spare_active) (struct mddev *mddev);
585	sector_t (*sync_request)(struct mddev *mddev, sector_t sector_nr, int *skipped);
586	int (*resize) (struct mddev *mddev, sector_t sectors);
587	sector_t (*size) (struct mddev *mddev, sector_t sectors, int raid_disks);
588	int (*check_reshape) (struct mddev *mddev);
589	int (*start_reshape) (struct mddev *mddev);
590	void (*finish_reshape) (struct mddev *mddev);
591	void (*update_reshape_pos) (struct mddev *mddev);
592	/* quiesce suspends or resumes internal processing.
593	 * 1 - stop new actions and wait for action io to complete
594	 * 0 - return to normal behaviour
595	 */
596	void (*quiesce) (struct mddev *mddev, int quiesce);
597	/* takeover is used to transition an array from one
598	 * personality to another.  The new personality must be able
599	 * to handle the data in the current layout.
600	 * e.g. 2drive raid1 -> 2drive raid5
601	 *      ndrive raid5 -> degraded n+1drive raid6 with special layout
602	 * If the takeover succeeds, a new 'private' structure is returned.
603	 * This needs to be installed and then ->run used to activate the
604	 * array.
605	 */
606	void *(*takeover) (struct mddev *mddev);
607	/* Changes the consistency policy of an active array. */
608	int (*change_consistency_policy)(struct mddev *mddev, const char *buf);
609};
610
 
611struct md_sysfs_entry {
612	struct attribute attr;
613	ssize_t (*show)(struct mddev *, char *);
614	ssize_t (*store)(struct mddev *, const char *, size_t);
615};
616extern struct attribute_group md_bitmap_group;
617
618static inline struct kernfs_node *sysfs_get_dirent_safe(struct kernfs_node *sd, char *name)
619{
620	if (sd)
621		return sysfs_get_dirent(sd, name);
622	return sd;
623}
624static inline void sysfs_notify_dirent_safe(struct kernfs_node *sd)
625{
626	if (sd)
627		sysfs_notify_dirent(sd);
628}
629
630static inline char * mdname (struct mddev * mddev)
631{
632	return mddev->gendisk ? mddev->gendisk->disk_name : "mdX";
633}
634
635static inline int sysfs_link_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		return sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
643	} else
644		return 0;
645}
646
647static inline void sysfs_unlink_rdev(struct mddev *mddev, struct md_rdev *rdev)
648{
649	char nm[20];
650	if (!test_bit(Replacement, &rdev->flags) &&
651	    !test_bit(Journal, &rdev->flags) &&
652	    mddev->kobj.sd) {
653		sprintf(nm, "rd%d", rdev->raid_disk);
654		sysfs_remove_link(&mddev->kobj, nm);
655	}
656}
657
658/*
659 * iterates through some rdev ringlist. It's safe to remove the
660 * current 'rdev'. Dont touch 'tmp' though.
661 */
662#define rdev_for_each_list(rdev, tmp, head)				\
663	list_for_each_entry_safe(rdev, tmp, head, same_set)
664
665/*
666 * iterates through the 'same array disks' ringlist
667 */
668#define rdev_for_each(rdev, mddev)				\
669	list_for_each_entry(rdev, &((mddev)->disks), same_set)
670
671#define rdev_for_each_safe(rdev, tmp, mddev)				\
672	list_for_each_entry_safe(rdev, tmp, &((mddev)->disks), same_set)
673
674#define rdev_for_each_rcu(rdev, mddev)				\
675	list_for_each_entry_rcu(rdev, &((mddev)->disks), same_set)
676
677struct md_thread {
678	void			(*run) (struct md_thread *thread);
679	struct mddev		*mddev;
680	wait_queue_head_t	wqueue;
681	unsigned long		flags;
682	struct task_struct	*tsk;
683	unsigned long		timeout;
684	void			*private;
685};
686
687#define THREAD_WAKEUP  0
688
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
689static inline void safe_put_page(struct page *p)
690{
691	if (p) put_page(p);
692}
693
694extern int register_md_personality(struct md_personality *p);
695extern int unregister_md_personality(struct md_personality *p);
696extern int register_md_cluster_operations(struct md_cluster_operations *ops,
697		struct module *module);
698extern int unregister_md_cluster_operations(void);
699extern int md_setup_cluster(struct mddev *mddev, int nodes);
700extern void md_cluster_stop(struct mddev *mddev);
701extern struct md_thread *md_register_thread(
702	void (*run)(struct md_thread *thread),
703	struct mddev *mddev,
704	const char *name);
705extern void md_unregister_thread(struct md_thread **threadp);
706extern void md_wakeup_thread(struct md_thread *thread);
707extern void md_check_recovery(struct mddev *mddev);
708extern void md_reap_sync_thread(struct mddev *mddev);
709extern int mddev_init_writes_pending(struct mddev *mddev);
710extern bool md_write_start(struct mddev *mddev, struct bio *bi);
711extern void md_write_inc(struct mddev *mddev, struct bio *bi);
712extern void md_write_end(struct mddev *mddev);
713extern void md_done_sync(struct mddev *mddev, int blocks, int ok);
714extern void md_error(struct mddev *mddev, struct md_rdev *rdev);
715extern void md_finish_reshape(struct mddev *mddev);
716
717extern bool __must_check md_flush_request(struct mddev *mddev, struct bio *bio);
718extern void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
719			   sector_t sector, int size, struct page *page);
720extern int md_super_wait(struct mddev *mddev);
721extern int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
722			struct page *page, int op, int op_flags,
723			bool metadata_op);
724extern void md_do_sync(struct md_thread *thread);
725extern void md_new_event(struct mddev *mddev);
726extern void md_allow_write(struct mddev *mddev);
727extern void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev);
728extern void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors);
729extern int md_check_no_bitmap(struct mddev *mddev);
730extern int md_integrity_register(struct mddev *mddev);
731extern int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev);
732extern int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale);
 
733
734extern void mddev_init(struct mddev *mddev);
735extern int md_run(struct mddev *mddev);
736extern int md_start(struct mddev *mddev);
737extern void md_stop(struct mddev *mddev);
738extern void md_stop_writes(struct mddev *mddev);
739extern int md_rdev_init(struct md_rdev *rdev);
740extern void md_rdev_clear(struct md_rdev *rdev);
741
742extern void md_handle_request(struct mddev *mddev, struct bio *bio);
743extern void mddev_suspend(struct mddev *mddev);
744extern void mddev_resume(struct mddev *mddev);
745extern struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
746				   struct mddev *mddev);
747
748extern void md_reload_sb(struct mddev *mddev, int raid_disk);
749extern void md_update_sb(struct mddev *mddev, int force);
750extern void md_kick_rdev_from_array(struct md_rdev * rdev);
751extern void mddev_create_serial_pool(struct mddev *mddev, struct md_rdev *rdev,
752				     bool is_suspend);
753extern void mddev_destroy_serial_pool(struct mddev *mddev, struct md_rdev *rdev,
754				      bool is_suspend);
755struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr);
756struct md_rdev *md_find_rdev_rcu(struct mddev *mddev, dev_t dev);
757
758static inline bool is_mddev_broken(struct md_rdev *rdev, const char *md_type)
759{
760	int flags = rdev->bdev->bd_disk->flags;
761
762	if (!(flags & GENHD_FL_UP)) {
763		if (!test_and_set_bit(MD_BROKEN, &rdev->mddev->flags))
764			pr_warn("md: %s: %s array has a missing/failed member\n",
765				mdname(rdev->mddev), md_type);
766		return true;
767	}
768	return false;
769}
770
771static inline void rdev_dec_pending(struct md_rdev *rdev, struct mddev *mddev)
772{
773	int faulty = test_bit(Faulty, &rdev->flags);
774	if (atomic_dec_and_test(&rdev->nr_pending) && faulty) {
775		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
776		md_wakeup_thread(mddev->thread);
777	}
778}
779
780extern struct md_cluster_operations *md_cluster_ops;
781static inline int mddev_is_clustered(struct mddev *mddev)
782{
783	return mddev->cluster_info && mddev->bitmap_info.nodes > 1;
784}
785
786/* clear unsupported mddev_flags */
787static inline void mddev_clear_unsupported_flags(struct mddev *mddev,
788	unsigned long unsupported_flags)
789{
790	mddev->flags &= ~unsupported_flags;
791}
792
793static inline void mddev_check_writesame(struct mddev *mddev, struct bio *bio)
794{
795	if (bio_op(bio) == REQ_OP_WRITE_SAME &&
796	    !bio->bi_disk->queue->limits.max_write_same_sectors)
797		mddev->queue->limits.max_write_same_sectors = 0;
798}
799
800static inline void mddev_check_write_zeroes(struct mddev *mddev, struct bio *bio)
801{
802	if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
803	    !bio->bi_disk->queue->limits.max_write_zeroes_sectors)
804		mddev->queue->limits.max_write_zeroes_sectors = 0;
805}
806
807struct mdu_array_info_s;
808struct mdu_disk_info_s;
809
810extern int mdp_major;
811void md_autostart_arrays(int part);
812int md_set_array_info(struct mddev *mddev, struct mdu_array_info_s *info);
813int md_add_new_disk(struct mddev *mddev, struct mdu_disk_info_s *info);
814int do_md_run(struct mddev *mddev);
815
816extern const struct block_device_operations md_fops;
817
818#endif /* _MD_MD_H */