Linux Audio

Check our new training course

Loading...
v6.8
  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	struct bdev_handle *bdev_handle;	/* Handle from open for bdev */
 63
 64	struct page	*sb_page, *bb_page;
 65	int		sb_loaded;
 66	__u64		sb_events;
 67	sector_t	data_offset;	/* start of data in array */
 68	sector_t	new_data_offset;/* only relevant while reshaping */
 69	sector_t	sb_start;	/* offset of the super block (in 512byte sectors) */
 70	int		sb_size;	/* bytes in the superblock */
 71	int		preferred_minor;	/* autorun support */
 72
 73	struct kobject	kobj;
 74
 75	/* A device can be in one of three states based on two flags:
 76	 * Not working:   faulty==1 in_sync==0
 77	 * Fully working: faulty==0 in_sync==1
 78	 * Working, but not
 79	 * in sync with array
 80	 *                faulty==0 in_sync==0
 81	 *
 82	 * It can never have faulty==1, in_sync==1
 83	 * This reduces the burden of testing multiple flags in many cases
 84	 */
 85
 86	unsigned long	flags;	/* bit set of 'enum flag_bits' bits. */
 87	wait_queue_head_t blocked_wait;
 88
 89	int desc_nr;			/* descriptor index in the superblock */
 90	int raid_disk;			/* role of device in array */
 91	int new_raid_disk;		/* role that the device will have in
 92					 * the array after a level-change completes.
 93					 */
 94	int saved_raid_disk;		/* role that device used to have in the
 95					 * array and could again if we did a partial
 96					 * resync from the bitmap
 97					 */
 98	union {
 99		sector_t recovery_offset;/* If this device has been partially
100					 * recovered, this is where we were
101					 * up to.
102					 */
103		sector_t journal_tail;	/* If this device is a journal device,
104					 * this is the journal tail (journal
105					 * recovery start point)
106					 */
107	};
108
109	atomic_t	nr_pending;	/* number of pending requests.
110					 * only maintained for arrays that
111					 * support hot removal
112					 */
113	atomic_t	read_errors;	/* number of consecutive read errors that
114					 * we have tried to ignore.
115					 */
116	time64_t	last_read_error;	/* monotonic time since our
117						 * last read error
118						 */
119	atomic_t	corrected_errors; /* number of corrected read errors,
120					   * for reporting to userspace and storing
121					   * in superblock.
122					   */
123
124	struct serial_in_rdev *serial;  /* used for raid1 io serialization */
125
126	struct kernfs_node *sysfs_state; /* handle for 'state'
127					   * sysfs entry */
128	/* handle for 'unacknowledged_bad_blocks' sysfs dentry */
129	struct kernfs_node *sysfs_unack_badblocks;
130	/* handle for 'bad_blocks' sysfs dentry */
131	struct kernfs_node *sysfs_badblocks;
132	struct badblocks badblocks;
133
134	struct {
135		short offset;	/* Offset from superblock to start of PPL.
136				 * Not used by external metadata. */
137		unsigned int size;	/* Size in sectors of the PPL space */
138		sector_t sector;	/* First sector of the PPL space */
139	} ppl;
140};
141enum flag_bits {
142	Faulty,			/* device is known to have a fault */
143	In_sync,		/* device is in_sync with rest of array */
144	Bitmap_sync,		/* ..actually, not quite In_sync.  Need a
145				 * bitmap-based recovery to get fully in sync.
146				 * The bit is only meaningful before device
147				 * has been passed to pers->hot_add_disk.
148				 */
149	WriteMostly,		/* Avoid reading if at all possible */
150	AutoDetected,		/* added by auto-detect */
151	Blocked,		/* An error occurred but has not yet
152				 * been acknowledged by the metadata
153				 * handler, so don't allow writes
154				 * until it is cleared */
155	WriteErrorSeen,		/* A write error has been seen on this
156				 * device
157				 */
158	FaultRecorded,		/* Intermediate state for clearing
159				 * Blocked.  The Fault is/will-be
160				 * recorded in the metadata, but that
161				 * metadata hasn't been stored safely
162				 * on disk yet.
163				 */
164	BlockedBadBlocks,	/* A writer is blocked because they
165				 * found an unacknowledged bad-block.
166				 * This can safely be cleared at any
167				 * time, and the writer will re-check.
168				 * It may be set at any time, and at
169				 * worst the writer will timeout and
170				 * re-check.  So setting it as
171				 * accurately as possible is good, but
172				 * not absolutely critical.
173				 */
174	WantReplacement,	/* This device is a candidate to be
175				 * hot-replaced, either because it has
176				 * reported some faults, or because
177				 * of explicit request.
178				 */
179	Replacement,		/* This device is a replacement for
180				 * a want_replacement device with same
181				 * raid_disk number.
182				 */
183	Candidate,		/* For clustered environments only:
184				 * This device is seen locally but not
185				 * by the whole cluster
186				 */
187	Journal,		/* This device is used as journal for
188				 * raid-5/6.
189				 * Usually, this device should be faster
190				 * than other devices in the array
191				 */
192	ClusterRemove,
 
 
 
 
 
193	ExternalBbl,            /* External metadata provides bad
194				 * block management for a disk
195				 */
196	FailFast,		/* Minimal retries should be attempted on
197				 * this device, so use REQ_FAILFAST_DEV.
198				 * Also don't try to repair failed reads.
199				 * It is expects that no bad block log
200				 * is present.
201				 */
202	LastDev,		/* Seems to be the last working dev as
203				 * it didn't fail, so don't use FailFast
204				 * any more for metadata
205				 */
206	CollisionCheck,		/*
207				 * check if there is collision between raid1
208				 * serial bios.
209				 */
210};
211
212static inline int is_badblock(struct md_rdev *rdev, sector_t s, int sectors,
213			      sector_t *first_bad, int *bad_sectors)
214{
215	if (unlikely(rdev->badblocks.count)) {
216		int rv = badblocks_check(&rdev->badblocks, rdev->data_offset + s,
217					sectors,
218					first_bad, bad_sectors);
219		if (rv)
220			*first_bad -= rdev->data_offset;
221		return rv;
222	}
223	return 0;
224}
225extern int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
226			      int is_new);
227extern int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
228				int is_new);
229struct md_cluster_info;
230
231/**
232 * enum mddev_flags - md device flags.
233 * @MD_ARRAY_FIRST_USE: First use of array, needs initialization.
234 * @MD_CLOSING: If set, we are closing the array, do not open it then.
235 * @MD_JOURNAL_CLEAN: A raid with journal is already clean.
236 * @MD_HAS_JOURNAL: The raid array has journal feature set.
237 * @MD_CLUSTER_RESYNC_LOCKED: cluster raid only, which means node, already took
238 *			       resync lock, need to release the lock.
239 * @MD_FAILFAST_SUPPORTED: Using MD_FAILFAST on metadata writes is supported as
240 *			    calls to md_error() will never cause the array to
241 *			    become failed.
242 * @MD_HAS_PPL:  The raid array has PPL feature set.
243 * @MD_HAS_MULTIPLE_PPLS: The raid array has multiple PPLs feature set.
244 * @MD_NOT_READY: do_md_run() is active, so 'array_state', ust not report that
245 *		   array is ready yet.
246 * @MD_BROKEN: This is used to stop writes and mark array as failed.
247 * @MD_DELETED: This device is being deleted
248 *
249 * change UNSUPPORTED_MDDEV_FLAGS for each array type if new flag is added
250 */
251enum mddev_flags {
252	MD_ARRAY_FIRST_USE,
253	MD_CLOSING,
254	MD_JOURNAL_CLEAN,
255	MD_HAS_JOURNAL,
256	MD_CLUSTER_RESYNC_LOCKED,
257	MD_FAILFAST_SUPPORTED,
258	MD_HAS_PPL,
259	MD_HAS_MULTIPLE_PPLS,
260	MD_NOT_READY,
261	MD_BROKEN,
262	MD_DELETED,
 
 
 
 
263};
264
265enum mddev_sb_flags {
266	MD_SB_CHANGE_DEVS,		/* Some device status has changed */
267	MD_SB_CHANGE_CLEAN,	/* transition to or from 'clean' */
268	MD_SB_CHANGE_PENDING,	/* switch from 'clean' to 'active' in progress */
269	MD_SB_NEED_REWRITE,	/* metadata write needs to be repeated */
270};
271
272#define NR_SERIAL_INFOS		8
273/* record current range of serialize IOs */
274struct serial_info {
275	struct rb_node node;
276	sector_t start;		/* start sector of rb node */
277	sector_t last;		/* end sector of rb node */
278	sector_t _subtree_last; /* highest sector in subtree of rb node */
279};
280
281/*
282 * mddev->curr_resync stores the current sector of the resync but
283 * also has some overloaded values.
284 */
285enum {
286	/* No resync in progress */
287	MD_RESYNC_NONE = 0,
288	/* Yielded to allow another conflicting resync to commence */
289	MD_RESYNC_YIELDED = 1,
290	/* Delayed to check that there is no conflict with another sync */
291	MD_RESYNC_DELAYED = 2,
292	/* Any value greater than or equal to this is in an active resync */
293	MD_RESYNC_ACTIVE = 3,
294};
295
296struct mddev {
297	void				*private;
298	struct md_personality		*pers;
299	dev_t				unit;
300	int				md_minor;
301	struct list_head		disks;
302	unsigned long			flags;
303	unsigned long			sb_flags;
304
305	int				suspended;
306	struct mutex			suspend_mutex;
307	struct percpu_ref		active_io;
308	int				ro;
309	int				sysfs_active; /* set when sysfs deletes
310						       * are happening, so run/
311						       * takeover/stop are not safe
312						       */
313	struct gendisk			*gendisk;
314
315	struct kobject			kobj;
316	int				hold_active;
317#define	UNTIL_IOCTL	1
318#define	UNTIL_STOP	2
319
320	/* Superblock information */
321	int				major_version,
322					minor_version,
323					patch_version;
324	int				persistent;
325	int				external;	/* metadata is
326							 * managed externally */
327	char				metadata_type[17]; /* externally set*/
328	int				chunk_sectors;
329	time64_t			ctime, utime;
330	int				level, layout;
331	char				clevel[16];
332	int				raid_disks;
333	int				max_disks;
334	sector_t			dev_sectors;	/* used size of
335							 * component devices */
336	sector_t			array_sectors; /* exported array size */
337	int				external_size; /* size managed
338							* externally */
339	__u64				events;
340	/* If the last 'event' was simply a clean->dirty transition, and
341	 * we didn't write it to the spares, then it is safe and simple
342	 * to just decrement the event count on a dirty->clean transition.
343	 * So we record that possibility here.
344	 */
345	int				can_decrease_events;
346
347	char				uuid[16];
348
349	/* If the array is being reshaped, we need to record the
350	 * new shape and an indication of where we are up to.
351	 * This is written to the superblock.
352	 * If reshape_position is MaxSector, then no reshape is happening (yet).
353	 */
354	sector_t			reshape_position;
355	int				delta_disks, new_level, new_layout;
356	int				new_chunk_sectors;
357	int				reshape_backwards;
358
359	struct md_thread __rcu		*thread;	/* management thread */
360	struct md_thread __rcu		*sync_thread;	/* doing resync or reconstruct */
361
362	/* 'last_sync_action' is initialized to "none".  It is set when a
363	 * sync operation (i.e "data-check", "requested-resync", "resync",
364	 * "recovery", or "reshape") is started.  It holds this value even
365	 * when the sync thread is "frozen" (interrupted) or "idle" (stopped
366	 * or finished).  It is overwritten when a new sync operation is begun.
367	 */
368	char				*last_sync_action;
369	sector_t			curr_resync;	/* last block scheduled */
370	/* As resync requests can complete out of order, we cannot easily track
371	 * how much resync has been completed.  So we occasionally pause until
372	 * everything completes, then set curr_resync_completed to curr_resync.
373	 * As such it may be well behind the real resync mark, but it is a value
374	 * we are certain of.
375	 */
376	sector_t			curr_resync_completed;
377	unsigned long			resync_mark;	/* a recent timestamp */
378	sector_t			resync_mark_cnt;/* blocks written at resync_mark */
379	sector_t			curr_mark_cnt; /* blocks scheduled now */
380
381	sector_t			resync_max_sectors; /* may be set by personality */
382
383	atomic64_t			resync_mismatches; /* count of sectors where
384							    * parity/replica mismatch found
385							    */
386
387	/* allow user-space to request suspension of IO to regions of the array */
388	sector_t			suspend_lo;
389	sector_t			suspend_hi;
390	/* if zero, use the system-wide default */
391	int				sync_speed_min;
392	int				sync_speed_max;
393
394	/* resync even though the same disks are shared among md-devices */
395	int				parallel_resync;
396
397	int				ok_start_degraded;
398
399	unsigned long			recovery;
400	/* If a RAID personality determines that recovery (of a particular
401	 * device) will fail due to a read error on the source device, it
402	 * takes a copy of this number and does not attempt recovery again
403	 * until this number changes.
404	 */
405	int				recovery_disabled;
406
407	int				in_sync;	/* know to not need resync */
408	/* 'open_mutex' avoids races between 'md_open' and 'do_md_stop', so
409	 * that we are never stopping an array while it is open.
410	 * 'reconfig_mutex' protects all other reconfiguration.
411	 * These locks are separate due to conflicting interactions
412	 * with disk->open_mutex.
413	 * Lock ordering is:
414	 *  reconfig_mutex -> disk->open_mutex
415	 *  disk->open_mutex -> open_mutex:  e.g. __blkdev_get -> md_open
416	 */
417	struct mutex			open_mutex;
418	struct mutex			reconfig_mutex;
419	atomic_t			active;		/* general refcount */
420	atomic_t			openers;	/* number of active opens */
421
422	int				changed;	/* True if we might need to
423							 * reread partition info */
424	int				degraded;	/* whether md should consider
425							 * adding a spare
426							 */
427
428	atomic_t			recovery_active; /* blocks scheduled, but not written */
429	wait_queue_head_t		recovery_wait;
430	sector_t			recovery_cp;
431	sector_t			resync_min;	/* user requested sync
432							 * starts here */
433	sector_t			resync_max;	/* resync should pause
434							 * when it gets here */
435
436	struct kernfs_node		*sysfs_state;	/* handle for 'array_state'
437							 * file in sysfs.
438							 */
439	struct kernfs_node		*sysfs_action;  /* handle for 'sync_action' */
440	struct kernfs_node		*sysfs_completed;	/*handle for 'sync_completed' */
441	struct kernfs_node		*sysfs_degraded;	/*handle for 'degraded' */
442	struct kernfs_node		*sysfs_level;		/*handle for 'level' */
443
444	/* used for delayed sysfs removal */
445	struct work_struct del_work;
446	/* used for register new sync thread */
447	struct work_struct sync_work;
448
449	/* "lock" protects:
450	 *   flush_bio transition from NULL to !NULL
451	 *   rdev superblocks, events
452	 *   clearing MD_CHANGE_*
453	 *   in_sync - and related safemode and MD_CHANGE changes
454	 *   pers (also protected by reconfig_mutex and pending IO).
455	 *   clearing ->bitmap
456	 *   clearing ->bitmap_info.file
457	 *   changing ->resync_{min,max}
458	 *   setting MD_RECOVERY_RUNNING (which interacts with resync_{min,max})
459	 */
460	spinlock_t			lock;
461	wait_queue_head_t		sb_wait;	/* for waiting on superblock updates */
462	atomic_t			pending_writes;	/* number of active superblock writes */
463
464	unsigned int			safemode;	/* if set, update "clean" superblock
465							 * when no writes pending.
466							 */
467	unsigned int			safemode_delay;
468	struct timer_list		safemode_timer;
469	struct percpu_ref		writes_pending;
470	int				sync_checkers;	/* # of threads checking writes_pending */
471	struct request_queue		*queue;	/* for plugging ... */
472
473	struct bitmap			*bitmap; /* the bitmap for the device */
474	struct {
475		struct file		*file; /* the bitmap file */
476		loff_t			offset; /* offset from superblock of
477						 * start of bitmap. May be
478						 * negative, but not '0'
479						 * For external metadata, offset
480						 * from start of device.
481						 */
482		unsigned long		space; /* space available at this offset */
483		loff_t			default_offset; /* this is the offset to use when
484							 * hot-adding a bitmap.  It should
485							 * eventually be settable by sysfs.
486							 */
487		unsigned long		default_space; /* space available at
488							* default offset */
489		struct mutex		mutex;
490		unsigned long		chunksize;
491		unsigned long		daemon_sleep; /* how many jiffies between updates? */
492		unsigned long		max_write_behind; /* write-behind mode */
493		int			external;
494		int			nodes; /* Maximum number of nodes in the cluster */
495		char                    cluster_name[64]; /* Name of the cluster */
496	} bitmap_info;
497
498	atomic_t			max_corr_read_errors; /* max read retries */
499	struct list_head		all_mddevs;
500
501	const struct attribute_group	*to_remove;
502
503	struct bio_set			bio_set;
504	struct bio_set			sync_set; /* for sync operations like
505						   * metadata and bitmap writes
506						   */
507	struct bio_set			io_clone_set;
508
509	/* Generic flush handling.
510	 * The last to finish preflush schedules a worker to submit
511	 * the rest of the request (without the REQ_PREFLUSH flag).
512	 */
513	struct bio *flush_bio;
514	atomic_t flush_pending;
515	ktime_t start_flush, prev_flush_start; /* prev_flush_start is when the previous completed
516						* flush was started.
517						*/
518	struct work_struct flush_work;
519	struct work_struct event_work;	/* used by dm to report failure event */
520	mempool_t *serial_info_pool;
521	void (*sync_super)(struct mddev *mddev, struct md_rdev *rdev);
522	struct md_cluster_info		*cluster_info;
523	unsigned int			good_device_nr;	/* good device num within cluster raid */
524	unsigned int			noio_flag; /* for memalloc scope API */
525
526	/*
527	 * Temporarily store rdev that will be finally removed when
528	 * reconfig_mutex is unlocked, protected by reconfig_mutex.
529	 */
530	struct list_head		deleting;
531
532	/* Used to synchronize idle and frozen for action_store() */
533	struct mutex			sync_mutex;
534	/* The sequence number for sync thread */
535	atomic_t sync_seq;
536
537	bool	has_superblocks:1;
538	bool	fail_last_dev:1;
539	bool	serialize_policy:1;
540};
541
542enum recovery_flags {
543	/*
544	 * If neither SYNC or RESHAPE are set, then it is a recovery.
545	 */
546	MD_RECOVERY_RUNNING,	/* a thread is running, or about to be started */
547	MD_RECOVERY_SYNC,	/* actually doing a resync, not a recovery */
548	MD_RECOVERY_RECOVER,	/* doing recovery, or need to try it. */
549	MD_RECOVERY_INTR,	/* resync needs to be aborted for some reason */
550	MD_RECOVERY_DONE,	/* thread is done and is waiting to be reaped */
551	MD_RECOVERY_NEEDED,	/* we might need to start a resync/recover */
552	MD_RECOVERY_REQUESTED,	/* user-space has requested a sync (used with SYNC) */
553	MD_RECOVERY_CHECK,	/* user-space request for check-only, no repair */
554	MD_RECOVERY_RESHAPE,	/* A reshape is happening */
555	MD_RECOVERY_FROZEN,	/* User request to abort, and not restart, any action */
556	MD_RECOVERY_ERROR,	/* sync-action interrupted because io-error */
557	MD_RECOVERY_WAIT,	/* waiting for pers->start() to finish */
558	MD_RESYNCING_REMOTE,	/* remote node is running resync thread */
559};
560
561static inline int __must_check mddev_lock(struct mddev *mddev)
562{
563	return mutex_lock_interruptible(&mddev->reconfig_mutex);
564}
565
566/* Sometimes we need to take the lock in a situation where
567 * failure due to interrupts is not acceptable.
568 */
569static inline void mddev_lock_nointr(struct mddev *mddev)
570{
571	mutex_lock(&mddev->reconfig_mutex);
572}
573
 
 
 
 
 
574static inline int mddev_trylock(struct mddev *mddev)
575{
576	return mutex_trylock(&mddev->reconfig_mutex);
577}
578extern void mddev_unlock(struct mddev *mddev);
579
580static inline void md_sync_acct(struct block_device *bdev, unsigned long nr_sectors)
581{
582	atomic_add(nr_sectors, &bdev->bd_disk->sync_io);
583}
584
585static inline void md_sync_acct_bio(struct bio *bio, unsigned long nr_sectors)
586{
587	md_sync_acct(bio->bi_bdev, nr_sectors);
588}
589
590struct md_personality
591{
592	char *name;
593	int level;
594	struct list_head list;
595	struct module *owner;
596	bool __must_check (*make_request)(struct mddev *mddev, struct bio *bio);
597	/*
598	 * start up works that do NOT require md_thread. tasks that
599	 * requires md_thread should go into start()
600	 */
601	int (*run)(struct mddev *mddev);
602	/* start up works that require md threads */
603	int (*start)(struct mddev *mddev);
604	void (*free)(struct mddev *mddev, void *priv);
605	void (*status)(struct seq_file *seq, struct mddev *mddev);
606	/* error_handler must set ->faulty and clear ->in_sync
607	 * if appropriate, and should abort recovery if needed
608	 */
609	void (*error_handler)(struct mddev *mddev, struct md_rdev *rdev);
610	int (*hot_add_disk) (struct mddev *mddev, struct md_rdev *rdev);
611	int (*hot_remove_disk) (struct mddev *mddev, struct md_rdev *rdev);
612	int (*spare_active) (struct mddev *mddev);
613	sector_t (*sync_request)(struct mddev *mddev, sector_t sector_nr, int *skipped);
614	int (*resize) (struct mddev *mddev, sector_t sectors);
615	sector_t (*size) (struct mddev *mddev, sector_t sectors, int raid_disks);
616	int (*check_reshape) (struct mddev *mddev);
617	int (*start_reshape) (struct mddev *mddev);
618	void (*finish_reshape) (struct mddev *mddev);
619	void (*update_reshape_pos) (struct mddev *mddev);
620	/* quiesce suspends or resumes internal processing.
621	 * 1 - stop new actions and wait for action io to complete
622	 * 0 - return to normal behaviour
623	 */
624	void (*quiesce) (struct mddev *mddev, int quiesce);
625	/* takeover is used to transition an array from one
626	 * personality to another.  The new personality must be able
627	 * to handle the data in the current layout.
628	 * e.g. 2drive raid1 -> 2drive raid5
629	 *      ndrive raid5 -> degraded n+1drive raid6 with special layout
630	 * If the takeover succeeds, a new 'private' structure is returned.
631	 * This needs to be installed and then ->run used to activate the
632	 * array.
633	 */
634	void *(*takeover) (struct mddev *mddev);
635	/* Changes the consistency policy of an active array. */
636	int (*change_consistency_policy)(struct mddev *mddev, const char *buf);
 
637};
638
639struct md_sysfs_entry {
640	struct attribute attr;
641	ssize_t (*show)(struct mddev *, char *);
642	ssize_t (*store)(struct mddev *, const char *, size_t);
643};
644extern const struct attribute_group md_bitmap_group;
645
646static inline struct kernfs_node *sysfs_get_dirent_safe(struct kernfs_node *sd, char *name)
647{
648	if (sd)
649		return sysfs_get_dirent(sd, name);
650	return sd;
651}
652static inline void sysfs_notify_dirent_safe(struct kernfs_node *sd)
653{
654	if (sd)
655		sysfs_notify_dirent(sd);
656}
657
658static inline char * mdname (struct mddev * mddev)
659{
660	return mddev->gendisk ? mddev->gendisk->disk_name : "mdX";
661}
662
663static inline int sysfs_link_rdev(struct mddev *mddev, struct md_rdev *rdev)
664{
665	char nm[20];
666	if (!test_bit(Replacement, &rdev->flags) &&
667	    !test_bit(Journal, &rdev->flags) &&
668	    mddev->kobj.sd) {
669		sprintf(nm, "rd%d", rdev->raid_disk);
670		return sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
671	} else
672		return 0;
673}
674
675static inline void sysfs_unlink_rdev(struct mddev *mddev, struct md_rdev *rdev)
676{
677	char nm[20];
678	if (!test_bit(Replacement, &rdev->flags) &&
679	    !test_bit(Journal, &rdev->flags) &&
680	    mddev->kobj.sd) {
681		sprintf(nm, "rd%d", rdev->raid_disk);
682		sysfs_remove_link(&mddev->kobj, nm);
683	}
684}
685
686/*
687 * iterates through some rdev ringlist. It's safe to remove the
688 * current 'rdev'. Dont touch 'tmp' though.
689 */
690#define rdev_for_each_list(rdev, tmp, head)				\
691	list_for_each_entry_safe(rdev, tmp, head, same_set)
692
693/*
694 * iterates through the 'same array disks' ringlist
695 */
696#define rdev_for_each(rdev, mddev)				\
697	list_for_each_entry(rdev, &((mddev)->disks), same_set)
698
699#define rdev_for_each_safe(rdev, tmp, mddev)				\
700	list_for_each_entry_safe(rdev, tmp, &((mddev)->disks), same_set)
701
702#define rdev_for_each_rcu(rdev, mddev)				\
703	list_for_each_entry_rcu(rdev, &((mddev)->disks), same_set)
704
705struct md_thread {
706	void			(*run) (struct md_thread *thread);
707	struct mddev		*mddev;
708	wait_queue_head_t	wqueue;
709	unsigned long		flags;
710	struct task_struct	*tsk;
711	unsigned long		timeout;
712	void			*private;
713};
714
715struct md_io_clone {
716	struct mddev	*mddev;
717	struct bio	*orig_bio;
718	unsigned long	start_time;
719	struct bio	bio_clone;
720};
721
722#define THREAD_WAKEUP  0
723
724static inline void safe_put_page(struct page *p)
725{
726	if (p) put_page(p);
727}
728
729extern int register_md_personality(struct md_personality *p);
730extern int unregister_md_personality(struct md_personality *p);
731extern int register_md_cluster_operations(struct md_cluster_operations *ops,
732		struct module *module);
733extern int unregister_md_cluster_operations(void);
734extern int md_setup_cluster(struct mddev *mddev, int nodes);
735extern void md_cluster_stop(struct mddev *mddev);
736extern struct md_thread *md_register_thread(
737	void (*run)(struct md_thread *thread),
738	struct mddev *mddev,
739	const char *name);
740extern void md_unregister_thread(struct mddev *mddev, struct md_thread __rcu **threadp);
741extern void md_wakeup_thread(struct md_thread __rcu *thread);
742extern void md_check_recovery(struct mddev *mddev);
743extern void md_reap_sync_thread(struct mddev *mddev);
744extern bool md_write_start(struct mddev *mddev, struct bio *bi);
745extern void md_write_inc(struct mddev *mddev, struct bio *bi);
746extern void md_write_end(struct mddev *mddev);
747extern void md_done_sync(struct mddev *mddev, int blocks, int ok);
748extern void md_error(struct mddev *mddev, struct md_rdev *rdev);
749extern void md_finish_reshape(struct mddev *mddev);
750void md_submit_discard_bio(struct mddev *mddev, struct md_rdev *rdev,
751			struct bio *bio, sector_t start, sector_t size);
752void md_account_bio(struct mddev *mddev, struct bio **bio);
753
754extern bool __must_check md_flush_request(struct mddev *mddev, struct bio *bio);
 
755extern void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
756			   sector_t sector, int size, struct page *page);
757extern int md_super_wait(struct mddev *mddev);
758extern int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
759		struct page *page, blk_opf_t opf, bool metadata_op);
 
760extern void md_do_sync(struct md_thread *thread);
761extern void md_new_event(void);
762extern void md_allow_write(struct mddev *mddev);
763extern void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev);
764extern void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors);
765extern int md_check_no_bitmap(struct mddev *mddev);
766extern int md_integrity_register(struct mddev *mddev);
767extern int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev);
768extern int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale);
769
770extern int mddev_init(struct mddev *mddev);
771extern void mddev_destroy(struct mddev *mddev);
772struct mddev *md_alloc(dev_t dev, char *name);
773void mddev_put(struct mddev *mddev);
774extern int md_run(struct mddev *mddev);
775extern int md_start(struct mddev *mddev);
776extern void md_stop(struct mddev *mddev);
777extern void md_stop_writes(struct mddev *mddev);
778extern int md_rdev_init(struct md_rdev *rdev);
779extern void md_rdev_clear(struct md_rdev *rdev);
780
781extern void md_handle_request(struct mddev *mddev, struct bio *bio);
782extern int mddev_suspend(struct mddev *mddev, bool interruptible);
783extern void mddev_resume(struct mddev *mddev);
 
 
 
 
784
 
785extern void md_reload_sb(struct mddev *mddev, int raid_disk);
786extern void md_update_sb(struct mddev *mddev, int force);
787extern void mddev_create_serial_pool(struct mddev *mddev, struct md_rdev *rdev);
788extern void mddev_destroy_serial_pool(struct mddev *mddev,
789				      struct md_rdev *rdev);
790struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr);
791struct md_rdev *md_find_rdev_rcu(struct mddev *mddev, dev_t dev);
792
793static inline bool is_rdev_broken(struct md_rdev *rdev)
794{
795	return !disk_live(rdev->bdev->bd_disk);
 
796}
797
798static inline void rdev_dec_pending(struct md_rdev *rdev, struct mddev *mddev)
799{
800	int faulty = test_bit(Faulty, &rdev->flags);
801	if (atomic_dec_and_test(&rdev->nr_pending) && faulty) {
802		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
803		md_wakeup_thread(mddev->thread);
804	}
805}
806
807extern struct md_cluster_operations *md_cluster_ops;
808static inline int mddev_is_clustered(struct mddev *mddev)
809{
810	return mddev->cluster_info && mddev->bitmap_info.nodes > 1;
811}
812
813/* clear unsupported mddev_flags */
814static inline void mddev_clear_unsupported_flags(struct mddev *mddev,
815	unsigned long unsupported_flags)
816{
817	mddev->flags &= ~unsupported_flags;
818}
819
820static inline void mddev_check_write_zeroes(struct mddev *mddev, struct bio *bio)
821{
822	if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
823	    !bio->bi_bdev->bd_disk->queue->limits.max_write_zeroes_sectors)
824		mddev->queue->limits.max_write_zeroes_sectors = 0;
825}
826
827static inline int mddev_suspend_and_lock(struct mddev *mddev)
828{
829	int ret;
830
831	ret = mddev_suspend(mddev, true);
832	if (ret)
833		return ret;
834
835	ret = mddev_lock(mddev);
836	if (ret)
837		mddev_resume(mddev);
838
839	return ret;
840}
841
842static inline void mddev_suspend_and_lock_nointr(struct mddev *mddev)
843{
844	mddev_suspend(mddev, false);
845	mutex_lock(&mddev->reconfig_mutex);
846}
847
848static inline void mddev_unlock_and_resume(struct mddev *mddev)
849{
850	mddev_unlock(mddev);
851	mddev_resume(mddev);
852}
853
854struct mdu_array_info_s;
855struct mdu_disk_info_s;
856
857extern int mdp_major;
858extern struct workqueue_struct *md_bitmap_wq;
859void md_autostart_arrays(int part);
860int md_set_array_info(struct mddev *mddev, struct mdu_array_info_s *info);
861int md_add_new_disk(struct mddev *mddev, struct mdu_disk_info_s *info);
862int do_md_run(struct mddev *mddev);
863
864extern const struct block_device_operations md_fops;
865
866#endif /* _MD_MD_H */
v4.10.11
 
  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};
126enum flag_bits {
127	Faulty,			/* device is known to have a fault */
128	In_sync,		/* device is in_sync with rest of array */
129	Bitmap_sync,		/* ..actually, not quite In_sync.  Need a
130				 * bitmap-based recovery to get fully in sync
 
 
131				 */
132	WriteMostly,		/* Avoid reading if at all possible */
133	AutoDetected,		/* added by auto-detect */
134	Blocked,		/* An error occurred but has not yet
135				 * been acknowledged by the metadata
136				 * handler, so don't allow writes
137				 * until it is cleared */
138	WriteErrorSeen,		/* A write error has been seen on this
139				 * device
140				 */
141	FaultRecorded,		/* Intermediate state for clearing
142				 * Blocked.  The Fault is/will-be
143				 * recorded in the metadata, but that
144				 * metadata hasn't been stored safely
145				 * on disk yet.
146				 */
147	BlockedBadBlocks,	/* A writer is blocked because they
148				 * found an unacknowledged bad-block.
149				 * This can safely be cleared at any
150				 * time, and the writer will re-check.
151				 * It may be set at any time, and at
152				 * worst the writer will timeout and
153				 * re-check.  So setting it as
154				 * accurately as possible is good, but
155				 * not absolutely critical.
156				 */
157	WantReplacement,	/* This device is a candidate to be
158				 * hot-replaced, either because it has
159				 * reported some faults, or because
160				 * of explicit request.
161				 */
162	Replacement,		/* This device is a replacement for
163				 * a want_replacement device with same
164				 * raid_disk number.
165				 */
166	Candidate,		/* For clustered environments only:
167				 * This device is seen locally but not
168				 * by the whole cluster
169				 */
170	Journal,		/* This device is used as journal for
171				 * raid-5/6.
172				 * Usually, this device should be faster
173				 * than other devices in the array
174				 */
175	ClusterRemove,
176	RemoveSynchronized,	/* synchronize_rcu() was called after
177				 * this device was known to be faulty,
178				 * so it is safe to remove without
179				 * another synchronize_rcu() call.
180				 */
181	ExternalBbl,            /* External metadata provides bad
182				 * block management for a disk
183				 */
184	FailFast,		/* Minimal retries should be attempted on
185				 * this device, so use REQ_FAILFAST_DEV.
186				 * Also don't try to repair failed reads.
187				 * It is expects that no bad block log
188				 * is present.
189				 */
190	LastDev,		/* Seems to be the last working dev as
191				 * it didn't fail, so don't use FailFast
192				 * any more for metadata
193				 */
 
 
 
 
194};
195
196static inline int is_badblock(struct md_rdev *rdev, sector_t s, int sectors,
197			      sector_t *first_bad, int *bad_sectors)
198{
199	if (unlikely(rdev->badblocks.count)) {
200		int rv = badblocks_check(&rdev->badblocks, rdev->data_offset + s,
201					sectors,
202					first_bad, bad_sectors);
203		if (rv)
204			*first_bad -= rdev->data_offset;
205		return rv;
206	}
207	return 0;
208}
209extern int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
210			      int is_new);
211extern int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
212				int is_new);
213struct md_cluster_info;
214
215/* change UNSUPPORTED_MDDEV_FLAGS for each array type if new flag is added */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
216enum mddev_flags {
217	MD_ARRAY_FIRST_USE,	/* First use of array, needs initialization */
218	MD_CLOSING,		/* If set, we are closing the array, do not open
219				 * it then */
220	MD_JOURNAL_CLEAN,	/* A raid with journal is already clean */
221	MD_HAS_JOURNAL,		/* The raid array has journal feature set */
222	MD_RELOAD_SB,		/* Reload the superblock because another node
223				 * updated it.
224				 */
225	MD_CLUSTER_RESYNC_LOCKED, /* cluster raid only, which means node
226				   * already took resync lock, need to
227				   * release the lock */
228	MD_FAILFAST_SUPPORTED,	/* Using MD_FAILFAST on metadata writes is
229				 * supported as calls to md_error() will
230				 * never cause the array to become failed.
231				 */
232};
233
234enum mddev_sb_flags {
235	MD_SB_CHANGE_DEVS,		/* Some device status has changed */
236	MD_SB_CHANGE_CLEAN,	/* transition to or from 'clean' */
237	MD_SB_CHANGE_PENDING,	/* switch from 'clean' to 'active' in progress */
238	MD_SB_NEED_REWRITE,	/* metadata write needs to be repeated */
239};
240
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
241struct mddev {
242	void				*private;
243	struct md_personality		*pers;
244	dev_t				unit;
245	int				md_minor;
246	struct list_head		disks;
247	unsigned long			flags;
248	unsigned long			sb_flags;
249
250	int				suspended;
251	atomic_t			active_io;
 
252	int				ro;
253	int				sysfs_active; /* set when sysfs deletes
254						       * are happening, so run/
255						       * takeover/stop are not safe
256						       */
257	struct gendisk			*gendisk;
258
259	struct kobject			kobj;
260	int				hold_active;
261#define	UNTIL_IOCTL	1
262#define	UNTIL_STOP	2
263
264	/* Superblock information */
265	int				major_version,
266					minor_version,
267					patch_version;
268	int				persistent;
269	int				external;	/* metadata is
270							 * managed externally */
271	char				metadata_type[17]; /* externally set*/
272	int				chunk_sectors;
273	time64_t			ctime, utime;
274	int				level, layout;
275	char				clevel[16];
276	int				raid_disks;
277	int				max_disks;
278	sector_t			dev_sectors;	/* used size of
279							 * component devices */
280	sector_t			array_sectors; /* exported array size */
281	int				external_size; /* size managed
282							* externally */
283	__u64				events;
284	/* If the last 'event' was simply a clean->dirty transition, and
285	 * we didn't write it to the spares, then it is safe and simple
286	 * to just decrement the event count on a dirty->clean transition.
287	 * So we record that possibility here.
288	 */
289	int				can_decrease_events;
290
291	char				uuid[16];
292
293	/* If the array is being reshaped, we need to record the
294	 * new shape and an indication of where we are up to.
295	 * This is written to the superblock.
296	 * If reshape_position is MaxSector, then no reshape is happening (yet).
297	 */
298	sector_t			reshape_position;
299	int				delta_disks, new_level, new_layout;
300	int				new_chunk_sectors;
301	int				reshape_backwards;
302
303	struct md_thread		*thread;	/* management thread */
304	struct md_thread		*sync_thread;	/* doing resync or reconstruct */
305
306	/* 'last_sync_action' is initialized to "none".  It is set when a
307	 * sync operation (i.e "data-check", "requested-resync", "resync",
308	 * "recovery", or "reshape") is started.  It holds this value even
309	 * when the sync thread is "frozen" (interrupted) or "idle" (stopped
310	 * or finished).  It is overwritten when a new sync operation is begun.
311	 */
312	char				*last_sync_action;
313	sector_t			curr_resync;	/* last block scheduled */
314	/* As resync requests can complete out of order, we cannot easily track
315	 * how much resync has been completed.  So we occasionally pause until
316	 * everything completes, then set curr_resync_completed to curr_resync.
317	 * As such it may be well behind the real resync mark, but it is a value
318	 * we are certain of.
319	 */
320	sector_t			curr_resync_completed;
321	unsigned long			resync_mark;	/* a recent timestamp */
322	sector_t			resync_mark_cnt;/* blocks written at resync_mark */
323	sector_t			curr_mark_cnt; /* blocks scheduled now */
324
325	sector_t			resync_max_sectors; /* may be set by personality */
326
327	atomic64_t			resync_mismatches; /* count of sectors where
328							    * parity/replica mismatch found
329							    */
330
331	/* allow user-space to request suspension of IO to regions of the array */
332	sector_t			suspend_lo;
333	sector_t			suspend_hi;
334	/* if zero, use the system-wide default */
335	int				sync_speed_min;
336	int				sync_speed_max;
337
338	/* resync even though the same disks are shared among md-devices */
339	int				parallel_resync;
340
341	int				ok_start_degraded;
342
343	unsigned long			recovery;
344	/* If a RAID personality determines that recovery (of a particular
345	 * device) will fail due to a read error on the source device, it
346	 * takes a copy of this number and does not attempt recovery again
347	 * until this number changes.
348	 */
349	int				recovery_disabled;
350
351	int				in_sync;	/* know to not need resync */
352	/* 'open_mutex' avoids races between 'md_open' and 'do_md_stop', so
353	 * that we are never stopping an array while it is open.
354	 * 'reconfig_mutex' protects all other reconfiguration.
355	 * These locks are separate due to conflicting interactions
356	 * with bdev->bd_mutex.
357	 * Lock ordering is:
358	 *  reconfig_mutex -> bd_mutex : e.g. do_md_run -> revalidate_disk
359	 *  bd_mutex -> open_mutex:  e.g. __blkdev_get -> md_open
360	 */
361	struct mutex			open_mutex;
362	struct mutex			reconfig_mutex;
363	atomic_t			active;		/* general refcount */
364	atomic_t			openers;	/* number of active opens */
365
366	int				changed;	/* True if we might need to
367							 * reread partition info */
368	int				degraded;	/* whether md should consider
369							 * adding a spare
370							 */
371
372	atomic_t			recovery_active; /* blocks scheduled, but not written */
373	wait_queue_head_t		recovery_wait;
374	sector_t			recovery_cp;
375	sector_t			resync_min;	/* user requested sync
376							 * starts here */
377	sector_t			resync_max;	/* resync should pause
378							 * when it gets here */
379
380	struct kernfs_node		*sysfs_state;	/* handle for 'array_state'
381							 * file in sysfs.
382							 */
383	struct kernfs_node		*sysfs_action;  /* handle for 'sync_action' */
384
385	struct work_struct del_work;	/* used for delayed sysfs removal */
 
 
 
 
 
 
386
387	/* "lock" protects:
388	 *   flush_bio transition from NULL to !NULL
389	 *   rdev superblocks, events
390	 *   clearing MD_CHANGE_*
391	 *   in_sync - and related safemode and MD_CHANGE changes
392	 *   pers (also protected by reconfig_mutex and pending IO).
393	 *   clearing ->bitmap
394	 *   clearing ->bitmap_info.file
395	 *   changing ->resync_{min,max}
396	 *   setting MD_RECOVERY_RUNNING (which interacts with resync_{min,max})
397	 */
398	spinlock_t			lock;
399	wait_queue_head_t		sb_wait;	/* for waiting on superblock updates */
400	atomic_t			pending_writes;	/* number of active superblock writes */
401
402	unsigned int			safemode;	/* if set, update "clean" superblock
403							 * when no writes pending.
404							 */
405	unsigned int			safemode_delay;
406	struct timer_list		safemode_timer;
407	atomic_t			writes_pending;
 
408	struct request_queue		*queue;	/* for plugging ... */
409
410	struct bitmap			*bitmap; /* the bitmap for the device */
411	struct {
412		struct file		*file; /* the bitmap file */
413		loff_t			offset; /* offset from superblock of
414						 * start of bitmap. May be
415						 * negative, but not '0'
416						 * For external metadata, offset
417						 * from start of device.
418						 */
419		unsigned long		space; /* space available at this offset */
420		loff_t			default_offset; /* this is the offset to use when
421							 * hot-adding a bitmap.  It should
422							 * eventually be settable by sysfs.
423							 */
424		unsigned long		default_space; /* space available at
425							* default offset */
426		struct mutex		mutex;
427		unsigned long		chunksize;
428		unsigned long		daemon_sleep; /* how many jiffies between updates? */
429		unsigned long		max_write_behind; /* write-behind mode */
430		int			external;
431		int			nodes; /* Maximum number of nodes in the cluster */
432		char                    cluster_name[64]; /* Name of the cluster */
433	} bitmap_info;
434
435	atomic_t			max_corr_read_errors; /* max read retries */
436	struct list_head		all_mddevs;
437
438	struct attribute_group		*to_remove;
439
440	struct bio_set			*bio_set;
 
 
 
 
441
442	/* Generic flush handling.
443	 * The last to finish preflush schedules a worker to submit
444	 * the rest of the request (without the REQ_PREFLUSH flag).
445	 */
446	struct bio *flush_bio;
447	atomic_t flush_pending;
 
 
 
448	struct work_struct flush_work;
449	struct work_struct event_work;	/* used by dm to report failure event */
 
450	void (*sync_super)(struct mddev *mddev, struct md_rdev *rdev);
451	struct md_cluster_info		*cluster_info;
452	unsigned int			good_device_nr;	/* good device num within cluster raid */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
453};
454
455enum recovery_flags {
456	/*
457	 * If neither SYNC or RESHAPE are set, then it is a recovery.
458	 */
459	MD_RECOVERY_RUNNING,	/* a thread is running, or about to be started */
460	MD_RECOVERY_SYNC,	/* actually doing a resync, not a recovery */
461	MD_RECOVERY_RECOVER,	/* doing recovery, or need to try it. */
462	MD_RECOVERY_INTR,	/* resync needs to be aborted for some reason */
463	MD_RECOVERY_DONE,	/* thread is done and is waiting to be reaped */
464	MD_RECOVERY_NEEDED,	/* we might need to start a resync/recover */
465	MD_RECOVERY_REQUESTED,	/* user-space has requested a sync (used with SYNC) */
466	MD_RECOVERY_CHECK,	/* user-space request for check-only, no repair */
467	MD_RECOVERY_RESHAPE,	/* A reshape is happening */
468	MD_RECOVERY_FROZEN,	/* User request to abort, and not restart, any action */
469	MD_RECOVERY_ERROR,	/* sync-action interrupted because io-error */
 
 
470};
471
472static inline int __must_check mddev_lock(struct mddev *mddev)
473{
474	return mutex_lock_interruptible(&mddev->reconfig_mutex);
475}
476
477/* Sometimes we need to take the lock in a situation where
478 * failure due to interrupts is not acceptable.
479 */
480static inline void mddev_lock_nointr(struct mddev *mddev)
481{
482	mutex_lock(&mddev->reconfig_mutex);
483}
484
485static inline int mddev_is_locked(struct mddev *mddev)
486{
487	return mutex_is_locked(&mddev->reconfig_mutex);
488}
489
490static inline int mddev_trylock(struct mddev *mddev)
491{
492	return mutex_trylock(&mddev->reconfig_mutex);
493}
494extern void mddev_unlock(struct mddev *mddev);
495
496static inline void md_sync_acct(struct block_device *bdev, unsigned long nr_sectors)
497{
498	atomic_add(nr_sectors, &bdev->bd_contains->bd_disk->sync_io);
 
 
 
 
 
499}
500
501struct md_personality
502{
503	char *name;
504	int level;
505	struct list_head list;
506	struct module *owner;
507	void (*make_request)(struct mddev *mddev, struct bio *bio);
 
 
 
 
508	int (*run)(struct mddev *mddev);
 
 
509	void (*free)(struct mddev *mddev, void *priv);
510	void (*status)(struct seq_file *seq, struct mddev *mddev);
511	/* error_handler must set ->faulty and clear ->in_sync
512	 * if appropriate, and should abort recovery if needed
513	 */
514	void (*error_handler)(struct mddev *mddev, struct md_rdev *rdev);
515	int (*hot_add_disk) (struct mddev *mddev, struct md_rdev *rdev);
516	int (*hot_remove_disk) (struct mddev *mddev, struct md_rdev *rdev);
517	int (*spare_active) (struct mddev *mddev);
518	sector_t (*sync_request)(struct mddev *mddev, sector_t sector_nr, int *skipped);
519	int (*resize) (struct mddev *mddev, sector_t sectors);
520	sector_t (*size) (struct mddev *mddev, sector_t sectors, int raid_disks);
521	int (*check_reshape) (struct mddev *mddev);
522	int (*start_reshape) (struct mddev *mddev);
523	void (*finish_reshape) (struct mddev *mddev);
524	/* quiesce moves between quiescence states
525	 * 0 - fully active
526	 * 1 - no new requests allowed
527	 * others - reserved
528	 */
529	void (*quiesce) (struct mddev *mddev, int state);
530	/* takeover is used to transition an array from one
531	 * personality to another.  The new personality must be able
532	 * to handle the data in the current layout.
533	 * e.g. 2drive raid1 -> 2drive raid5
534	 *      ndrive raid5 -> degraded n+1drive raid6 with special layout
535	 * If the takeover succeeds, a new 'private' structure is returned.
536	 * This needs to be installed and then ->run used to activate the
537	 * array.
538	 */
539	void *(*takeover) (struct mddev *mddev);
540	/* congested implements bdi.congested_fn().
541	 * Will not be called while array is 'suspended' */
542	int (*congested)(struct mddev *mddev, int bits);
543};
544
545struct md_sysfs_entry {
546	struct attribute attr;
547	ssize_t (*show)(struct mddev *, char *);
548	ssize_t (*store)(struct mddev *, const char *, size_t);
549};
550extern struct attribute_group md_bitmap_group;
551
552static inline struct kernfs_node *sysfs_get_dirent_safe(struct kernfs_node *sd, char *name)
553{
554	if (sd)
555		return sysfs_get_dirent(sd, name);
556	return sd;
557}
558static inline void sysfs_notify_dirent_safe(struct kernfs_node *sd)
559{
560	if (sd)
561		sysfs_notify_dirent(sd);
562}
563
564static inline char * mdname (struct mddev * mddev)
565{
566	return mddev->gendisk ? mddev->gendisk->disk_name : "mdX";
567}
568
569static inline int sysfs_link_rdev(struct mddev *mddev, struct md_rdev *rdev)
570{
571	char nm[20];
572	if (!test_bit(Replacement, &rdev->flags) &&
573	    !test_bit(Journal, &rdev->flags) &&
574	    mddev->kobj.sd) {
575		sprintf(nm, "rd%d", rdev->raid_disk);
576		return sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
577	} else
578		return 0;
579}
580
581static inline void sysfs_unlink_rdev(struct mddev *mddev, struct md_rdev *rdev)
582{
583	char nm[20];
584	if (!test_bit(Replacement, &rdev->flags) &&
585	    !test_bit(Journal, &rdev->flags) &&
586	    mddev->kobj.sd) {
587		sprintf(nm, "rd%d", rdev->raid_disk);
588		sysfs_remove_link(&mddev->kobj, nm);
589	}
590}
591
592/*
593 * iterates through some rdev ringlist. It's safe to remove the
594 * current 'rdev'. Dont touch 'tmp' though.
595 */
596#define rdev_for_each_list(rdev, tmp, head)				\
597	list_for_each_entry_safe(rdev, tmp, head, same_set)
598
599/*
600 * iterates through the 'same array disks' ringlist
601 */
602#define rdev_for_each(rdev, mddev)				\
603	list_for_each_entry(rdev, &((mddev)->disks), same_set)
604
605#define rdev_for_each_safe(rdev, tmp, mddev)				\
606	list_for_each_entry_safe(rdev, tmp, &((mddev)->disks), same_set)
607
608#define rdev_for_each_rcu(rdev, mddev)				\
609	list_for_each_entry_rcu(rdev, &((mddev)->disks), same_set)
610
611struct md_thread {
612	void			(*run) (struct md_thread *thread);
613	struct mddev		*mddev;
614	wait_queue_head_t	wqueue;
615	unsigned long		flags;
616	struct task_struct	*tsk;
617	unsigned long		timeout;
618	void			*private;
619};
620
 
 
 
 
 
 
 
621#define THREAD_WAKEUP  0
622
623static inline void safe_put_page(struct page *p)
624{
625	if (p) put_page(p);
626}
627
628extern int register_md_personality(struct md_personality *p);
629extern int unregister_md_personality(struct md_personality *p);
630extern int register_md_cluster_operations(struct md_cluster_operations *ops,
631		struct module *module);
632extern int unregister_md_cluster_operations(void);
633extern int md_setup_cluster(struct mddev *mddev, int nodes);
634extern void md_cluster_stop(struct mddev *mddev);
635extern struct md_thread *md_register_thread(
636	void (*run)(struct md_thread *thread),
637	struct mddev *mddev,
638	const char *name);
639extern void md_unregister_thread(struct md_thread **threadp);
640extern void md_wakeup_thread(struct md_thread *thread);
641extern void md_check_recovery(struct mddev *mddev);
642extern void md_reap_sync_thread(struct mddev *mddev);
643extern void md_write_start(struct mddev *mddev, struct bio *bi);
 
644extern void md_write_end(struct mddev *mddev);
645extern void md_done_sync(struct mddev *mddev, int blocks, int ok);
646extern void md_error(struct mddev *mddev, struct md_rdev *rdev);
647extern void md_finish_reshape(struct mddev *mddev);
 
 
 
648
649extern int mddev_congested(struct mddev *mddev, int bits);
650extern void md_flush_request(struct mddev *mddev, struct bio *bio);
651extern void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
652			   sector_t sector, int size, struct page *page);
653extern int md_super_wait(struct mddev *mddev);
654extern int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
655			struct page *page, int op, int op_flags,
656			bool metadata_op);
657extern void md_do_sync(struct md_thread *thread);
658extern void md_new_event(struct mddev *mddev);
659extern int md_allow_write(struct mddev *mddev);
660extern void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev);
661extern void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors);
662extern int md_check_no_bitmap(struct mddev *mddev);
663extern int md_integrity_register(struct mddev *mddev);
664extern int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev);
665extern int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale);
666
667extern void mddev_init(struct mddev *mddev);
 
 
 
668extern int md_run(struct mddev *mddev);
 
669extern void md_stop(struct mddev *mddev);
670extern void md_stop_writes(struct mddev *mddev);
671extern int md_rdev_init(struct md_rdev *rdev);
672extern void md_rdev_clear(struct md_rdev *rdev);
673
674extern void mddev_suspend(struct mddev *mddev);
 
675extern void mddev_resume(struct mddev *mddev);
676extern struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
677				   struct mddev *mddev);
678extern struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
679				   struct mddev *mddev);
680
681extern void md_unplug(struct blk_plug_cb *cb, bool from_schedule);
682extern void md_reload_sb(struct mddev *mddev, int raid_disk);
683extern void md_update_sb(struct mddev *mddev, int force);
684extern void md_kick_rdev_from_array(struct md_rdev * rdev);
 
 
685struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr);
686static inline int mddev_check_plugged(struct mddev *mddev)
 
 
687{
688	return !!blk_check_plugged(md_unplug, mddev,
689				   sizeof(struct blk_plug_cb));
690}
691
692static inline void rdev_dec_pending(struct md_rdev *rdev, struct mddev *mddev)
693{
694	int faulty = test_bit(Faulty, &rdev->flags);
695	if (atomic_dec_and_test(&rdev->nr_pending) && faulty) {
696		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
697		md_wakeup_thread(mddev->thread);
698	}
699}
700
701extern struct md_cluster_operations *md_cluster_ops;
702static inline int mddev_is_clustered(struct mddev *mddev)
703{
704	return mddev->cluster_info && mddev->bitmap_info.nodes > 1;
705}
706
707/* clear unsupported mddev_flags */
708static inline void mddev_clear_unsupported_flags(struct mddev *mddev,
709	unsigned long unsupported_flags)
710{
711	mddev->flags &= ~unsupported_flags;
712}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
713#endif /* _MD_MD_H */