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 */
v5.4
  1/* SPDX-License-Identifier: GPL-2.0-or-later */
  2/*
  3   md.h : kernel internal structure of the Linux MD driver
  4          Copyright (C) 1996-98 Ingo Molnar, Gadi Oxman
  5
  6*/
  7
  8#ifndef _MD_MD_H
  9#define _MD_MD_H
 10
 11#include <linux/blkdev.h>
 12#include <linux/backing-dev.h>
 13#include <linux/badblocks.h>
 14#include <linux/kobject.h>
 15#include <linux/list.h>
 16#include <linux/mm.h>
 17#include <linux/mutex.h>
 18#include <linux/timer.h>
 19#include <linux/wait.h>
 20#include <linux/workqueue.h>
 21#include "md-cluster.h"
 22
 23#define MaxSector (~(sector_t)0)
 24
 25/*
 26 * These flags should really be called "NO_RETRY" rather than
 27 * "FAILFAST" because they don't make any promise about time lapse,
 28 * only about the number of retries, which will be zero.
 29 * REQ_FAILFAST_DRIVER is not included because
 30 * Commit: 4a27446f3e39 ("[SCSI] modify scsi to handle new fail fast flags.")
 31 * seems to suggest that the errors it avoids retrying should usually
 32 * be retried.
 33 */
 34#define	MD_FAILFAST	(REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT)
 
 
 
 
 
 
 
 
 
 
 35/*
 36 * MD's 'extended' device
 37 */
 38struct md_rdev {
 39	struct list_head same_set;	/* RAID devices within the same set */
 40
 41	sector_t sectors;		/* Device size (in 512bytes sectors) */
 42	struct mddev *mddev;		/* RAID array if running */
 43	int last_events;		/* IO event timestamp */
 44
 45	/*
 46	 * If meta_bdev is non-NULL, it means that a separate device is
 47	 * being used to store the metadata (superblock/bitmap) which
 48	 * would otherwise be contained on the same device as the data (bdev).
 49	 */
 50	struct block_device *meta_bdev;
 51	struct block_device *bdev;	/* block device handle */
 
 52
 53	struct page	*sb_page, *bb_page;
 54	int		sb_loaded;
 55	__u64		sb_events;
 56	sector_t	data_offset;	/* start of data in array */
 57	sector_t	new_data_offset;/* only relevant while reshaping */
 58	sector_t	sb_start;	/* offset of the super block (in 512byte sectors) */
 59	int		sb_size;	/* bytes in the superblock */
 60	int		preferred_minor;	/* autorun support */
 61
 62	struct kobject	kobj;
 63
 64	/* A device can be in one of three states based on two flags:
 65	 * Not working:   faulty==1 in_sync==0
 66	 * Fully working: faulty==0 in_sync==1
 67	 * Working, but not
 68	 * in sync with array
 69	 *                faulty==0 in_sync==0
 70	 *
 71	 * It can never have faulty==1, in_sync==1
 72	 * This reduces the burden of testing multiple flags in many cases
 73	 */
 74
 75	unsigned long	flags;	/* bit set of 'enum flag_bits' bits. */
 76	wait_queue_head_t blocked_wait;
 77
 78	int desc_nr;			/* descriptor index in the superblock */
 79	int raid_disk;			/* role of device in array */
 80	int new_raid_disk;		/* role that the device will have in
 81					 * the array after a level-change completes.
 82					 */
 83	int saved_raid_disk;		/* role that device used to have in the
 84					 * array and could again if we did a partial
 85					 * resync from the bitmap
 86					 */
 87	union {
 88		sector_t recovery_offset;/* If this device has been partially
 89					 * recovered, this is where we were
 90					 * up to.
 91					 */
 92		sector_t journal_tail;	/* If this device is a journal device,
 93					 * this is the journal tail (journal
 94					 * recovery start point)
 95					 */
 96	};
 97
 98	atomic_t	nr_pending;	/* number of pending requests.
 99					 * only maintained for arrays that
100					 * support hot removal
101					 */
102	atomic_t	read_errors;	/* number of consecutive read errors that
103					 * we have tried to ignore.
104					 */
105	time64_t	last_read_error;	/* monotonic time since our
106						 * last read error
107						 */
108	atomic_t	corrected_errors; /* number of corrected read errors,
109					   * for reporting to userspace and storing
110					   * in superblock.
111					   */
112
113	/*
114	 * The members for check collision of write behind IOs.
115	 */
116	struct list_head wb_list;
117	spinlock_t wb_list_lock;
118	wait_queue_head_t wb_io_wait;
119
120	struct work_struct del_work;	/* used for delayed sysfs removal */
121
122	struct kernfs_node *sysfs_state; /* handle for 'state'
123					   * sysfs entry */
124
 
 
 
125	struct badblocks badblocks;
126
127	struct {
128		short offset;	/* Offset from superblock to start of PPL.
129				 * Not used by external metadata. */
130		unsigned int size;	/* Size in sectors of the PPL space */
131		sector_t sector;	/* First sector of the PPL space */
132	} ppl;
133};
134enum flag_bits {
135	Faulty,			/* device is known to have a fault */
136	In_sync,		/* device is in_sync with rest of array */
137	Bitmap_sync,		/* ..actually, not quite In_sync.  Need a
138				 * bitmap-based recovery to get fully in sync.
139				 * The bit is only meaningful before device
140				 * has been passed to pers->hot_add_disk.
141				 */
142	WriteMostly,		/* Avoid reading if at all possible */
143	AutoDetected,		/* added by auto-detect */
144	Blocked,		/* An error occurred but has not yet
145				 * been acknowledged by the metadata
146				 * handler, so don't allow writes
147				 * until it is cleared */
148	WriteErrorSeen,		/* A write error has been seen on this
149				 * device
150				 */
151	FaultRecorded,		/* Intermediate state for clearing
152				 * Blocked.  The Fault is/will-be
153				 * recorded in the metadata, but that
154				 * metadata hasn't been stored safely
155				 * on disk yet.
156				 */
157	BlockedBadBlocks,	/* A writer is blocked because they
158				 * found an unacknowledged bad-block.
159				 * This can safely be cleared at any
160				 * time, and the writer will re-check.
161				 * It may be set at any time, and at
162				 * worst the writer will timeout and
163				 * re-check.  So setting it as
164				 * accurately as possible is good, but
165				 * not absolutely critical.
166				 */
167	WantReplacement,	/* This device is a candidate to be
168				 * hot-replaced, either because it has
169				 * reported some faults, or because
170				 * of explicit request.
171				 */
172	Replacement,		/* This device is a replacement for
173				 * a want_replacement device with same
174				 * raid_disk number.
175				 */
176	Candidate,		/* For clustered environments only:
177				 * This device is seen locally but not
178				 * by the whole cluster
179				 */
180	Journal,		/* This device is used as journal for
181				 * raid-5/6.
182				 * Usually, this device should be faster
183				 * than other devices in the array
184				 */
185	ClusterRemove,
186	RemoveSynchronized,	/* synchronize_rcu() was called after
187				 * this device was known to be faulty,
188				 * so it is safe to remove without
189				 * another synchronize_rcu() call.
190				 */
191	ExternalBbl,            /* External metadata provides bad
192				 * block management for a disk
193				 */
194	FailFast,		/* Minimal retries should be attempted on
195				 * this device, so use REQ_FAILFAST_DEV.
196				 * Also don't try to repair failed reads.
197				 * It is expects that no bad block log
198				 * is present.
199				 */
200	LastDev,		/* Seems to be the last working dev as
201				 * it didn't fail, so don't use FailFast
202				 * any more for metadata
203				 */
204	WBCollisionCheck,	/*
205				 * multiqueue device should check if there
206				 * is collision between write behind bios.
207				 */
208};
209
210static inline int is_badblock(struct md_rdev *rdev, sector_t s, int sectors,
211			      sector_t *first_bad, int *bad_sectors)
212{
213	if (unlikely(rdev->badblocks.count)) {
214		int rv = badblocks_check(&rdev->badblocks, rdev->data_offset + s,
215					sectors,
216					first_bad, bad_sectors);
217		if (rv)
218			*first_bad -= rdev->data_offset;
219		return rv;
220	}
221	return 0;
222}
223extern int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
224			      int is_new);
225extern int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
226				int is_new);
227struct md_cluster_info;
228
229/* change UNSUPPORTED_MDDEV_FLAGS for each array type if new flag is added */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
230enum mddev_flags {
231	MD_ARRAY_FIRST_USE,	/* First use of array, needs initialization */
232	MD_CLOSING,		/* If set, we are closing the array, do not open
233				 * it then */
234	MD_JOURNAL_CLEAN,	/* A raid with journal is already clean */
235	MD_HAS_JOURNAL,		/* The raid array has journal feature set */
236	MD_CLUSTER_RESYNC_LOCKED, /* cluster raid only, which means node
237				   * already took resync lock, need to
238				   * release the lock */
239	MD_FAILFAST_SUPPORTED,	/* Using MD_FAILFAST on metadata writes is
240				 * supported as calls to md_error() will
241				 * never cause the array to become failed.
242				 */
243	MD_HAS_PPL,		/* The raid array has PPL feature set */
244	MD_HAS_MULTIPLE_PPLS,	/* The raid array has multiple PPLs feature set */
245	MD_ALLOW_SB_UPDATE,	/* md_check_recovery is allowed to update
246				 * the metadata without taking reconfig_mutex.
247				 */
248	MD_UPDATING_SB,		/* md_check_recovery is updating the metadata
249				 * without explicitly holding reconfig_mutex.
250				 */
251	MD_NOT_READY,		/* do_md_run() is active, so 'array_state'
252				 * must not report that array is ready yet
253				 */
254	MD_BROKEN,              /* This is used in RAID-0/LINEAR only, to stop
255				 * I/O in case an array member is gone/failed.
256				 */
257};
258
259enum mddev_sb_flags {
260	MD_SB_CHANGE_DEVS,		/* Some device status has changed */
261	MD_SB_CHANGE_CLEAN,	/* transition to or from 'clean' */
262	MD_SB_CHANGE_PENDING,	/* switch from 'clean' to 'active' in progress */
263	MD_SB_NEED_REWRITE,	/* metadata write needs to be repeated */
264};
265
266#define NR_WB_INFOS	8
267/* record current range of write behind IOs */
268struct wb_info {
269	sector_t lo;
270	sector_t hi;
271	struct list_head list;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
272};
273
274struct mddev {
275	void				*private;
276	struct md_personality		*pers;
277	dev_t				unit;
278	int				md_minor;
279	struct list_head		disks;
280	unsigned long			flags;
281	unsigned long			sb_flags;
282
283	int				suspended;
284	atomic_t			active_io;
 
285	int				ro;
286	int				sysfs_active; /* set when sysfs deletes
287						       * are happening, so run/
288						       * takeover/stop are not safe
289						       */
290	struct gendisk			*gendisk;
291
292	struct kobject			kobj;
293	int				hold_active;
294#define	UNTIL_IOCTL	1
295#define	UNTIL_STOP	2
296
297	/* Superblock information */
298	int				major_version,
299					minor_version,
300					patch_version;
301	int				persistent;
302	int				external;	/* metadata is
303							 * managed externally */
304	char				metadata_type[17]; /* externally set*/
305	int				chunk_sectors;
306	time64_t			ctime, utime;
307	int				level, layout;
308	char				clevel[16];
309	int				raid_disks;
310	int				max_disks;
311	sector_t			dev_sectors;	/* used size of
312							 * component devices */
313	sector_t			array_sectors; /* exported array size */
314	int				external_size; /* size managed
315							* externally */
316	__u64				events;
317	/* If the last 'event' was simply a clean->dirty transition, and
318	 * we didn't write it to the spares, then it is safe and simple
319	 * to just decrement the event count on a dirty->clean transition.
320	 * So we record that possibility here.
321	 */
322	int				can_decrease_events;
323
324	char				uuid[16];
325
326	/* If the array is being reshaped, we need to record the
327	 * new shape and an indication of where we are up to.
328	 * This is written to the superblock.
329	 * If reshape_position is MaxSector, then no reshape is happening (yet).
330	 */
331	sector_t			reshape_position;
332	int				delta_disks, new_level, new_layout;
333	int				new_chunk_sectors;
334	int				reshape_backwards;
335
336	struct md_thread		*thread;	/* management thread */
337	struct md_thread		*sync_thread;	/* doing resync or reconstruct */
338
339	/* 'last_sync_action' is initialized to "none".  It is set when a
340	 * sync operation (i.e "data-check", "requested-resync", "resync",
341	 * "recovery", or "reshape") is started.  It holds this value even
342	 * when the sync thread is "frozen" (interrupted) or "idle" (stopped
343	 * or finished).  It is overwritten when a new sync operation is begun.
344	 */
345	char				*last_sync_action;
346	sector_t			curr_resync;	/* last block scheduled */
347	/* As resync requests can complete out of order, we cannot easily track
348	 * how much resync has been completed.  So we occasionally pause until
349	 * everything completes, then set curr_resync_completed to curr_resync.
350	 * As such it may be well behind the real resync mark, but it is a value
351	 * we are certain of.
352	 */
353	sector_t			curr_resync_completed;
354	unsigned long			resync_mark;	/* a recent timestamp */
355	sector_t			resync_mark_cnt;/* blocks written at resync_mark */
356	sector_t			curr_mark_cnt; /* blocks scheduled now */
357
358	sector_t			resync_max_sectors; /* may be set by personality */
359
360	atomic64_t			resync_mismatches; /* count of sectors where
361							    * parity/replica mismatch found
362							    */
363
364	/* allow user-space to request suspension of IO to regions of the array */
365	sector_t			suspend_lo;
366	sector_t			suspend_hi;
367	/* if zero, use the system-wide default */
368	int				sync_speed_min;
369	int				sync_speed_max;
370
371	/* resync even though the same disks are shared among md-devices */
372	int				parallel_resync;
373
374	int				ok_start_degraded;
375
376	unsigned long			recovery;
377	/* If a RAID personality determines that recovery (of a particular
378	 * device) will fail due to a read error on the source device, it
379	 * takes a copy of this number and does not attempt recovery again
380	 * until this number changes.
381	 */
382	int				recovery_disabled;
383
384	int				in_sync;	/* know to not need resync */
385	/* 'open_mutex' avoids races between 'md_open' and 'do_md_stop', so
386	 * that we are never stopping an array while it is open.
387	 * 'reconfig_mutex' protects all other reconfiguration.
388	 * These locks are separate due to conflicting interactions
389	 * with bdev->bd_mutex.
390	 * Lock ordering is:
391	 *  reconfig_mutex -> bd_mutex : e.g. do_md_run -> revalidate_disk
392	 *  bd_mutex -> open_mutex:  e.g. __blkdev_get -> md_open
393	 */
394	struct mutex			open_mutex;
395	struct mutex			reconfig_mutex;
396	atomic_t			active;		/* general refcount */
397	atomic_t			openers;	/* number of active opens */
398
399	int				changed;	/* True if we might need to
400							 * reread partition info */
401	int				degraded;	/* whether md should consider
402							 * adding a spare
403							 */
404
405	atomic_t			recovery_active; /* blocks scheduled, but not written */
406	wait_queue_head_t		recovery_wait;
407	sector_t			recovery_cp;
408	sector_t			resync_min;	/* user requested sync
409							 * starts here */
410	sector_t			resync_max;	/* resync should pause
411							 * when it gets here */
412
413	struct kernfs_node		*sysfs_state;	/* handle for 'array_state'
414							 * file in sysfs.
415							 */
416	struct kernfs_node		*sysfs_action;  /* handle for 'sync_action' */
417
418	struct work_struct del_work;	/* used for delayed sysfs removal */
 
 
 
 
 
 
419
420	/* "lock" protects:
421	 *   flush_bio transition from NULL to !NULL
422	 *   rdev superblocks, events
423	 *   clearing MD_CHANGE_*
424	 *   in_sync - and related safemode and MD_CHANGE changes
425	 *   pers (also protected by reconfig_mutex and pending IO).
426	 *   clearing ->bitmap
427	 *   clearing ->bitmap_info.file
428	 *   changing ->resync_{min,max}
429	 *   setting MD_RECOVERY_RUNNING (which interacts with resync_{min,max})
430	 */
431	spinlock_t			lock;
432	wait_queue_head_t		sb_wait;	/* for waiting on superblock updates */
433	atomic_t			pending_writes;	/* number of active superblock writes */
434
435	unsigned int			safemode;	/* if set, update "clean" superblock
436							 * when no writes pending.
437							 */
438	unsigned int			safemode_delay;
439	struct timer_list		safemode_timer;
440	struct percpu_ref		writes_pending;
441	int				sync_checkers;	/* # of threads checking writes_pending */
442	struct request_queue		*queue;	/* for plugging ... */
443
444	struct bitmap			*bitmap; /* the bitmap for the device */
445	struct {
446		struct file		*file; /* the bitmap file */
447		loff_t			offset; /* offset from superblock of
448						 * start of bitmap. May be
449						 * negative, but not '0'
450						 * For external metadata, offset
451						 * from start of device.
452						 */
453		unsigned long		space; /* space available at this offset */
454		loff_t			default_offset; /* this is the offset to use when
455							 * hot-adding a bitmap.  It should
456							 * eventually be settable by sysfs.
457							 */
458		unsigned long		default_space; /* space available at
459							* default offset */
460		struct mutex		mutex;
461		unsigned long		chunksize;
462		unsigned long		daemon_sleep; /* how many jiffies between updates? */
463		unsigned long		max_write_behind; /* write-behind mode */
464		int			external;
465		int			nodes; /* Maximum number of nodes in the cluster */
466		char                    cluster_name[64]; /* Name of the cluster */
467	} bitmap_info;
468
469	atomic_t			max_corr_read_errors; /* max read retries */
470	struct list_head		all_mddevs;
471
472	struct attribute_group		*to_remove;
473
474	struct bio_set			bio_set;
475	struct bio_set			sync_set; /* for sync operations like
476						   * metadata and bitmap writes
477						   */
 
478
479	/* Generic flush handling.
480	 * The last to finish preflush schedules a worker to submit
481	 * the rest of the request (without the REQ_PREFLUSH flag).
482	 */
483	struct bio *flush_bio;
484	atomic_t flush_pending;
485	ktime_t start_flush, last_flush; /* last_flush is when the last completed
486					  * flush was started.
487					  */
488	struct work_struct flush_work;
489	struct work_struct event_work;	/* used by dm to report failure event */
490	mempool_t *wb_info_pool;
491	void (*sync_super)(struct mddev *mddev, struct md_rdev *rdev);
492	struct md_cluster_info		*cluster_info;
493	unsigned int			good_device_nr;	/* good device num within cluster raid */
 
 
 
 
 
 
 
 
 
 
 
 
494
495	bool	has_superblocks:1;
496	bool	fail_last_dev:1;
 
497};
498
499enum recovery_flags {
500	/*
501	 * If neither SYNC or RESHAPE are set, then it is a recovery.
502	 */
503	MD_RECOVERY_RUNNING,	/* a thread is running, or about to be started */
504	MD_RECOVERY_SYNC,	/* actually doing a resync, not a recovery */
505	MD_RECOVERY_RECOVER,	/* doing recovery, or need to try it. */
506	MD_RECOVERY_INTR,	/* resync needs to be aborted for some reason */
507	MD_RECOVERY_DONE,	/* thread is done and is waiting to be reaped */
508	MD_RECOVERY_NEEDED,	/* we might need to start a resync/recover */
509	MD_RECOVERY_REQUESTED,	/* user-space has requested a sync (used with SYNC) */
510	MD_RECOVERY_CHECK,	/* user-space request for check-only, no repair */
511	MD_RECOVERY_RESHAPE,	/* A reshape is happening */
512	MD_RECOVERY_FROZEN,	/* User request to abort, and not restart, any action */
513	MD_RECOVERY_ERROR,	/* sync-action interrupted because io-error */
514	MD_RECOVERY_WAIT,	/* waiting for pers->start() to finish */
515	MD_RESYNCING_REMOTE,	/* remote node is running resync thread */
516};
517
518static inline int __must_check mddev_lock(struct mddev *mddev)
519{
520	return mutex_lock_interruptible(&mddev->reconfig_mutex);
521}
522
523/* Sometimes we need to take the lock in a situation where
524 * failure due to interrupts is not acceptable.
525 */
526static inline void mddev_lock_nointr(struct mddev *mddev)
527{
528	mutex_lock(&mddev->reconfig_mutex);
529}
530
531static inline int mddev_trylock(struct mddev *mddev)
532{
533	return mutex_trylock(&mddev->reconfig_mutex);
534}
535extern void mddev_unlock(struct mddev *mddev);
536
537static inline void md_sync_acct(struct block_device *bdev, unsigned long nr_sectors)
538{
539	atomic_add(nr_sectors, &bdev->bd_contains->bd_disk->sync_io);
540}
541
542static inline void md_sync_acct_bio(struct bio *bio, unsigned long nr_sectors)
543{
544	atomic_add(nr_sectors, &bio->bi_disk->sync_io);
545}
546
547struct md_personality
548{
549	char *name;
550	int level;
551	struct list_head list;
552	struct module *owner;
553	bool (*make_request)(struct mddev *mddev, struct bio *bio);
554	/*
555	 * start up works that do NOT require md_thread. tasks that
556	 * requires md_thread should go into start()
557	 */
558	int (*run)(struct mddev *mddev);
559	/* start up works that require md threads */
560	int (*start)(struct mddev *mddev);
561	void (*free)(struct mddev *mddev, void *priv);
562	void (*status)(struct seq_file *seq, struct mddev *mddev);
563	/* error_handler must set ->faulty and clear ->in_sync
564	 * if appropriate, and should abort recovery if needed
565	 */
566	void (*error_handler)(struct mddev *mddev, struct md_rdev *rdev);
567	int (*hot_add_disk) (struct mddev *mddev, struct md_rdev *rdev);
568	int (*hot_remove_disk) (struct mddev *mddev, struct md_rdev *rdev);
569	int (*spare_active) (struct mddev *mddev);
570	sector_t (*sync_request)(struct mddev *mddev, sector_t sector_nr, int *skipped);
571	int (*resize) (struct mddev *mddev, sector_t sectors);
572	sector_t (*size) (struct mddev *mddev, sector_t sectors, int raid_disks);
573	int (*check_reshape) (struct mddev *mddev);
574	int (*start_reshape) (struct mddev *mddev);
575	void (*finish_reshape) (struct mddev *mddev);
576	void (*update_reshape_pos) (struct mddev *mddev);
577	/* quiesce suspends or resumes internal processing.
578	 * 1 - stop new actions and wait for action io to complete
579	 * 0 - return to normal behaviour
580	 */
581	void (*quiesce) (struct mddev *mddev, int quiesce);
582	/* takeover is used to transition an array from one
583	 * personality to another.  The new personality must be able
584	 * to handle the data in the current layout.
585	 * e.g. 2drive raid1 -> 2drive raid5
586	 *      ndrive raid5 -> degraded n+1drive raid6 with special layout
587	 * If the takeover succeeds, a new 'private' structure is returned.
588	 * This needs to be installed and then ->run used to activate the
589	 * array.
590	 */
591	void *(*takeover) (struct mddev *mddev);
592	/* congested implements bdi.congested_fn().
593	 * Will not be called while array is 'suspended' */
594	int (*congested)(struct mddev *mddev, int bits);
595	/* Changes the consistency policy of an active array. */
596	int (*change_consistency_policy)(struct mddev *mddev, const char *buf);
597};
598
599struct md_sysfs_entry {
600	struct attribute attr;
601	ssize_t (*show)(struct mddev *, char *);
602	ssize_t (*store)(struct mddev *, const char *, size_t);
603};
604extern struct attribute_group md_bitmap_group;
605
606static inline struct kernfs_node *sysfs_get_dirent_safe(struct kernfs_node *sd, char *name)
607{
608	if (sd)
609		return sysfs_get_dirent(sd, name);
610	return sd;
611}
612static inline void sysfs_notify_dirent_safe(struct kernfs_node *sd)
613{
614	if (sd)
615		sysfs_notify_dirent(sd);
616}
617
618static inline char * mdname (struct mddev * mddev)
619{
620	return mddev->gendisk ? mddev->gendisk->disk_name : "mdX";
621}
622
623static inline int sysfs_link_rdev(struct mddev *mddev, struct md_rdev *rdev)
624{
625	char nm[20];
626	if (!test_bit(Replacement, &rdev->flags) &&
627	    !test_bit(Journal, &rdev->flags) &&
628	    mddev->kobj.sd) {
629		sprintf(nm, "rd%d", rdev->raid_disk);
630		return sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
631	} else
632		return 0;
633}
634
635static inline void sysfs_unlink_rdev(struct mddev *mddev, struct md_rdev *rdev)
636{
637	char nm[20];
638	if (!test_bit(Replacement, &rdev->flags) &&
639	    !test_bit(Journal, &rdev->flags) &&
640	    mddev->kobj.sd) {
641		sprintf(nm, "rd%d", rdev->raid_disk);
642		sysfs_remove_link(&mddev->kobj, nm);
643	}
644}
645
646/*
647 * iterates through some rdev ringlist. It's safe to remove the
648 * current 'rdev'. Dont touch 'tmp' though.
649 */
650#define rdev_for_each_list(rdev, tmp, head)				\
651	list_for_each_entry_safe(rdev, tmp, head, same_set)
652
653/*
654 * iterates through the 'same array disks' ringlist
655 */
656#define rdev_for_each(rdev, mddev)				\
657	list_for_each_entry(rdev, &((mddev)->disks), same_set)
658
659#define rdev_for_each_safe(rdev, tmp, mddev)				\
660	list_for_each_entry_safe(rdev, tmp, &((mddev)->disks), same_set)
661
662#define rdev_for_each_rcu(rdev, mddev)				\
663	list_for_each_entry_rcu(rdev, &((mddev)->disks), same_set)
664
665struct md_thread {
666	void			(*run) (struct md_thread *thread);
667	struct mddev		*mddev;
668	wait_queue_head_t	wqueue;
669	unsigned long		flags;
670	struct task_struct	*tsk;
671	unsigned long		timeout;
672	void			*private;
673};
674
 
 
 
 
 
 
 
675#define THREAD_WAKEUP  0
676
677static inline void safe_put_page(struct page *p)
678{
679	if (p) put_page(p);
680}
681
682extern int register_md_personality(struct md_personality *p);
683extern int unregister_md_personality(struct md_personality *p);
684extern int register_md_cluster_operations(struct md_cluster_operations *ops,
685		struct module *module);
686extern int unregister_md_cluster_operations(void);
687extern int md_setup_cluster(struct mddev *mddev, int nodes);
688extern void md_cluster_stop(struct mddev *mddev);
689extern struct md_thread *md_register_thread(
690	void (*run)(struct md_thread *thread),
691	struct mddev *mddev,
692	const char *name);
693extern void md_unregister_thread(struct md_thread **threadp);
694extern void md_wakeup_thread(struct md_thread *thread);
695extern void md_check_recovery(struct mddev *mddev);
696extern void md_reap_sync_thread(struct mddev *mddev);
697extern int mddev_init_writes_pending(struct mddev *mddev);
698extern bool md_write_start(struct mddev *mddev, struct bio *bi);
699extern void md_write_inc(struct mddev *mddev, struct bio *bi);
700extern void md_write_end(struct mddev *mddev);
701extern void md_done_sync(struct mddev *mddev, int blocks, int ok);
702extern void md_error(struct mddev *mddev, struct md_rdev *rdev);
703extern void md_finish_reshape(struct mddev *mddev);
 
 
 
704
705extern int mddev_congested(struct mddev *mddev, int bits);
706extern void md_flush_request(struct mddev *mddev, struct bio *bio);
707extern void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
708			   sector_t sector, int size, struct page *page);
709extern int md_super_wait(struct mddev *mddev);
710extern int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
711			struct page *page, int op, int op_flags,
712			bool metadata_op);
713extern void md_do_sync(struct md_thread *thread);
714extern void md_new_event(struct mddev *mddev);
715extern void md_allow_write(struct mddev *mddev);
716extern void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev);
717extern void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors);
718extern int md_check_no_bitmap(struct mddev *mddev);
719extern int md_integrity_register(struct mddev *mddev);
720extern int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev);
721extern int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale);
722
723extern void mddev_init(struct mddev *mddev);
 
 
 
724extern int md_run(struct mddev *mddev);
725extern int md_start(struct mddev *mddev);
726extern void md_stop(struct mddev *mddev);
727extern void md_stop_writes(struct mddev *mddev);
728extern int md_rdev_init(struct md_rdev *rdev);
729extern void md_rdev_clear(struct md_rdev *rdev);
730
731extern void md_handle_request(struct mddev *mddev, struct bio *bio);
732extern void mddev_suspend(struct mddev *mddev);
733extern void mddev_resume(struct mddev *mddev);
734extern struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
735				   struct mddev *mddev);
736
737extern void md_reload_sb(struct mddev *mddev, int raid_disk);
738extern void md_update_sb(struct mddev *mddev, int force);
739extern void md_kick_rdev_from_array(struct md_rdev * rdev);
740extern void mddev_create_wb_pool(struct mddev *mddev, struct md_rdev *rdev,
741				 bool is_suspend);
742struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr);
743struct md_rdev *md_find_rdev_rcu(struct mddev *mddev, dev_t dev);
744
745static inline bool is_mddev_broken(struct md_rdev *rdev, const char *md_type)
746{
747	int flags = rdev->bdev->bd_disk->flags;
748
749	if (!(flags & GENHD_FL_UP)) {
750		if (!test_and_set_bit(MD_BROKEN, &rdev->mddev->flags))
751			pr_warn("md: %s: %s array has a missing/failed member\n",
752				mdname(rdev->mddev), md_type);
753		return true;
754	}
755	return false;
756}
757
758static inline void rdev_dec_pending(struct md_rdev *rdev, struct mddev *mddev)
759{
760	int faulty = test_bit(Faulty, &rdev->flags);
761	if (atomic_dec_and_test(&rdev->nr_pending) && faulty) {
762		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
763		md_wakeup_thread(mddev->thread);
764	}
765}
766
767extern struct md_cluster_operations *md_cluster_ops;
768static inline int mddev_is_clustered(struct mddev *mddev)
769{
770	return mddev->cluster_info && mddev->bitmap_info.nodes > 1;
771}
772
773/* clear unsupported mddev_flags */
774static inline void mddev_clear_unsupported_flags(struct mddev *mddev,
775	unsigned long unsupported_flags)
776{
777	mddev->flags &= ~unsupported_flags;
778}
779
780static inline void mddev_check_writesame(struct mddev *mddev, struct bio *bio)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
781{
782	if (bio_op(bio) == REQ_OP_WRITE_SAME &&
783	    !bio->bi_disk->queue->limits.max_write_same_sectors)
784		mddev->queue->limits.max_write_same_sectors = 0;
785}
786
787static inline void mddev_check_write_zeroes(struct mddev *mddev, struct bio *bio)
788{
789	if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
790	    !bio->bi_disk->queue->limits.max_write_zeroes_sectors)
791		mddev->queue->limits.max_write_zeroes_sectors = 0;
792}
 
 
 
 
 
 
 
 
 
 
 
 
 
793#endif /* _MD_MD_H */