xref: /linux/drivers/md/md.h (revision 36239c6704b71da7fb8e2a9429e159a84d0c5a3e)
1 /*
2    md_k.h : kernel internal structure of the Linux MD driver
3           Copyright (C) 1996-98 Ingo Molnar, Gadi Oxman
4 
5    This program is free software; you can redistribute it and/or modify
6    it under the terms of the GNU General Public License as published by
7    the Free Software Foundation; either version 2, or (at your option)
8    any later version.
9 
10    You should have received a copy of the GNU General Public License
11    (for example /usr/src/linux/COPYING); if not, write to the Free
12    Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
13 */
14 
15 #ifndef _MD_MD_H
16 #define _MD_MD_H
17 
18 #include <linux/blkdev.h>
19 #include <linux/kobject.h>
20 #include <linux/list.h>
21 #include <linux/mm.h>
22 #include <linux/mutex.h>
23 #include <linux/timer.h>
24 #include <linux/wait.h>
25 #include <linux/workqueue.h>
26 
27 #define MaxSector (~(sector_t)0)
28 
29 typedef struct mddev_s mddev_t;
30 typedef struct mdk_rdev_s mdk_rdev_t;
31 
32 /*
33  * MD's 'extended' device
34  */
35 struct mdk_rdev_s
36 {
37 	struct list_head same_set;	/* RAID devices within the same set */
38 
39 	sector_t sectors;		/* Device size (in 512bytes sectors) */
40 	mddev_t *mddev;			/* RAID array if running */
41 	int last_events;		/* IO event timestamp */
42 
43 	struct block_device *bdev;	/* block device handle */
44 
45 	struct page	*sb_page;
46 	int		sb_loaded;
47 	__u64		sb_events;
48 	sector_t	data_offset;	/* start of data in array */
49 	sector_t 	sb_start;	/* offset of the super block (in 512byte sectors) */
50 	int		sb_size;	/* bytes in the superblock */
51 	int		preferred_minor;	/* autorun support */
52 
53 	struct kobject	kobj;
54 
55 	/* A device can be in one of three states based on two flags:
56 	 * Not working:   faulty==1 in_sync==0
57 	 * Fully working: faulty==0 in_sync==1
58 	 * Working, but not
59 	 * in sync with array
60 	 *                faulty==0 in_sync==0
61 	 *
62 	 * It can never have faulty==1, in_sync==1
63 	 * This reduces the burden of testing multiple flags in many cases
64 	 */
65 
66 	unsigned long	flags;
67 #define	Faulty		1		/* device is known to have a fault */
68 #define	In_sync		2		/* device is in_sync with rest of array */
69 #define	WriteMostly	4		/* Avoid reading if at all possible */
70 #define	BarriersNotsupp	5		/* BIO_RW_BARRIER is not supported */
71 #define	AllReserved	6		/* If whole device is reserved for
72 					 * one array */
73 #define	AutoDetected	7		/* added by auto-detect */
74 #define Blocked		8		/* An error occured on an externally
75 					 * managed array, don't allow writes
76 					 * until it is cleared */
77 	wait_queue_head_t blocked_wait;
78 
79 	int desc_nr;			/* descriptor index in the superblock */
80 	int raid_disk;			/* role of device in array */
81 	int new_raid_disk;		/* role that the device will have in
82 					 * the array after a level-change completes.
83 					 */
84 	int saved_raid_disk;		/* role that device used to have in the
85 					 * array and could again if we did a partial
86 					 * resync from the bitmap
87 					 */
88 	sector_t	recovery_offset;/* If this device has been partially
89 					 * recovered, this is where we were
90 					 * up to.
91 					 */
92 
93 	atomic_t	nr_pending;	/* number of pending requests.
94 					 * only maintained for arrays that
95 					 * support hot removal
96 					 */
97 	atomic_t	read_errors;	/* number of consecutive read errors that
98 					 * we have tried to ignore.
99 					 */
100 	struct timespec last_read_error;	/* monotonic time since our
101 						 * last read error
102 						 */
103 	atomic_t	corrected_errors; /* number of corrected read errors,
104 					   * for reporting to userspace and storing
105 					   * in superblock.
106 					   */
107 	struct work_struct del_work;	/* used for delayed sysfs removal */
108 
109 	struct sysfs_dirent *sysfs_state; /* handle for 'state'
110 					   * sysfs entry */
111 };
112 
113 struct mddev_s
114 {
115 	void				*private;
116 	struct mdk_personality		*pers;
117 	dev_t				unit;
118 	int				md_minor;
119 	struct list_head 		disks;
120 	unsigned long			flags;
121 #define MD_CHANGE_DEVS	0	/* Some device status has changed */
122 #define MD_CHANGE_CLEAN 1	/* transition to or from 'clean' */
123 #define MD_CHANGE_PENDING 2	/* superblock update in progress */
124 
125 	int				suspended;
126 	atomic_t			active_io;
127 	int				ro;
128 
129 	struct gendisk			*gendisk;
130 
131 	struct kobject			kobj;
132 	int				hold_active;
133 #define	UNTIL_IOCTL	1
134 #define	UNTIL_STOP	2
135 
136 	/* Superblock information */
137 	int				major_version,
138 					minor_version,
139 					patch_version;
140 	int				persistent;
141 	int 				external;	/* metadata is
142 							 * managed externally */
143 	char				metadata_type[17]; /* externally set*/
144 	int				chunk_sectors;
145 	time_t				ctime, utime;
146 	int				level, layout;
147 	char				clevel[16];
148 	int				raid_disks;
149 	int				max_disks;
150 	sector_t			dev_sectors; 	/* used size of
151 							 * component devices */
152 	sector_t			array_sectors; /* exported array size */
153 	int				external_size; /* size managed
154 							* externally */
155 	__u64				events;
156 	/* If the last 'event' was simply a clean->dirty transition, and
157 	 * we didn't write it to the spares, then it is safe and simple
158 	 * to just decrement the event count on a dirty->clean transition.
159 	 * So we record that possibility here.
160 	 */
161 	int				can_decrease_events;
162 
163 	char				uuid[16];
164 
165 	/* If the array is being reshaped, we need to record the
166 	 * new shape and an indication of where we are up to.
167 	 * This is written to the superblock.
168 	 * If reshape_position is MaxSector, then no reshape is happening (yet).
169 	 */
170 	sector_t			reshape_position;
171 	int				delta_disks, new_level, new_layout;
172 	int				new_chunk_sectors;
173 
174 	struct mdk_thread_s		*thread;	/* management thread */
175 	struct mdk_thread_s		*sync_thread;	/* doing resync or reconstruct */
176 	sector_t			curr_resync;	/* last block scheduled */
177 	/* As resync requests can complete out of order, we cannot easily track
178 	 * how much resync has been completed.  So we occasionally pause until
179 	 * everything completes, then set curr_resync_completed to curr_resync.
180 	 * As such it may be well behind the real resync mark, but it is a value
181 	 * we are certain of.
182 	 */
183 	sector_t			curr_resync_completed;
184 	unsigned long			resync_mark;	/* a recent timestamp */
185 	sector_t			resync_mark_cnt;/* blocks written at resync_mark */
186 	sector_t			curr_mark_cnt; /* blocks scheduled now */
187 
188 	sector_t			resync_max_sectors; /* may be set by personality */
189 
190 	sector_t			resync_mismatches; /* count of sectors where
191 							    * parity/replica mismatch found
192 							    */
193 
194 	/* allow user-space to request suspension of IO to regions of the array */
195 	sector_t			suspend_lo;
196 	sector_t			suspend_hi;
197 	/* if zero, use the system-wide default */
198 	int				sync_speed_min;
199 	int				sync_speed_max;
200 
201 	/* resync even though the same disks are shared among md-devices */
202 	int				parallel_resync;
203 
204 	int				ok_start_degraded;
205 	/* recovery/resync flags
206 	 * NEEDED:   we might need to start a resync/recover
207 	 * RUNNING:  a thread is running, or about to be started
208 	 * SYNC:     actually doing a resync, not a recovery
209 	 * RECOVER:  doing recovery, or need to try it.
210 	 * INTR:     resync needs to be aborted for some reason
211 	 * DONE:     thread is done and is waiting to be reaped
212 	 * REQUEST:  user-space has requested a sync (used with SYNC)
213 	 * CHECK:    user-space request for check-only, no repair
214 	 * RESHAPE:  A reshape is happening
215 	 *
216 	 * If neither SYNC or RESHAPE are set, then it is a recovery.
217 	 */
218 #define	MD_RECOVERY_RUNNING	0
219 #define	MD_RECOVERY_SYNC	1
220 #define	MD_RECOVERY_RECOVER	2
221 #define	MD_RECOVERY_INTR	3
222 #define	MD_RECOVERY_DONE	4
223 #define	MD_RECOVERY_NEEDED	5
224 #define	MD_RECOVERY_REQUESTED	6
225 #define	MD_RECOVERY_CHECK	7
226 #define MD_RECOVERY_RESHAPE	8
227 #define	MD_RECOVERY_FROZEN	9
228 
229 	unsigned long			recovery;
230 	int				recovery_disabled; /* if we detect that recovery
231 							    * will always fail, set this
232 							    * so we don't loop trying */
233 
234 	int				in_sync;	/* know to not need resync */
235 	/* 'open_mutex' avoids races between 'md_open' and 'do_md_stop', so
236 	 * that we are never stopping an array while it is open.
237 	 * 'reconfig_mutex' protects all other reconfiguration.
238 	 * These locks are separate due to conflicting interactions
239 	 * with bdev->bd_mutex.
240 	 * Lock ordering is:
241 	 *  reconfig_mutex -> bd_mutex : e.g. do_md_run -> revalidate_disk
242 	 *  bd_mutex -> open_mutex:  e.g. __blkdev_get -> md_open
243 	 */
244 	struct mutex			open_mutex;
245 	struct mutex			reconfig_mutex;
246 	atomic_t			active;		/* general refcount */
247 	atomic_t			openers;	/* number of active opens */
248 
249 	int				degraded;	/* whether md should consider
250 							 * adding a spare
251 							 */
252 	int				barriers_work;	/* initialised to true, cleared as soon
253 							 * as a barrier request to slave
254 							 * fails.  Only supported
255 							 */
256 	struct bio			*biolist; 	/* bios that need to be retried
257 							 * because BIO_RW_BARRIER is not supported
258 							 */
259 
260 	atomic_t			recovery_active; /* blocks scheduled, but not written */
261 	wait_queue_head_t		recovery_wait;
262 	sector_t			recovery_cp;
263 	sector_t			resync_min;	/* user requested sync
264 							 * starts here */
265 	sector_t			resync_max;	/* resync should pause
266 							 * when it gets here */
267 
268 	struct sysfs_dirent		*sysfs_state;	/* handle for 'array_state'
269 							 * file in sysfs.
270 							 */
271 	struct sysfs_dirent		*sysfs_action;  /* handle for 'sync_action' */
272 
273 	struct work_struct del_work;	/* used for delayed sysfs removal */
274 
275 	spinlock_t			write_lock;
276 	wait_queue_head_t		sb_wait;	/* for waiting on superblock updates */
277 	atomic_t			pending_writes;	/* number of active superblock writes */
278 
279 	unsigned int			safemode;	/* if set, update "clean" superblock
280 							 * when no writes pending.
281 							 */
282 	unsigned int			safemode_delay;
283 	struct timer_list		safemode_timer;
284 	atomic_t			writes_pending;
285 	struct request_queue		*queue;	/* for plugging ... */
286 
287 	struct bitmap                   *bitmap; /* the bitmap for the device */
288 	struct {
289 		struct file		*file; /* the bitmap file */
290 		loff_t			offset; /* offset from superblock of
291 						 * start of bitmap. May be
292 						 * negative, but not '0'
293 						 * For external metadata, offset
294 						 * from start of device.
295 						 */
296 		loff_t			default_offset; /* this is the offset to use when
297 							 * hot-adding a bitmap.  It should
298 							 * eventually be settable by sysfs.
299 							 */
300 		struct mutex		mutex;
301 		unsigned long		chunksize;
302 		unsigned long		daemon_sleep; /* how many seconds between updates? */
303 		unsigned long		max_write_behind; /* write-behind mode */
304 		int			external;
305 	} bitmap_info;
306 
307 	atomic_t 			max_corr_read_errors; /* max read retries */
308 	struct list_head		all_mddevs;
309 
310 	struct attribute_group		*to_remove;
311 	/* Generic barrier handling.
312 	 * If there is a pending barrier request, all other
313 	 * writes are blocked while the devices are flushed.
314 	 * The last to finish a flush schedules a worker to
315 	 * submit the barrier request (without the barrier flag),
316 	 * then submit more flush requests.
317 	 */
318 	struct bio *barrier;
319 	atomic_t flush_pending;
320 	struct work_struct barrier_work;
321 };
322 
323 
324 static inline void rdev_dec_pending(mdk_rdev_t *rdev, mddev_t *mddev)
325 {
326 	int faulty = test_bit(Faulty, &rdev->flags);
327 	if (atomic_dec_and_test(&rdev->nr_pending) && faulty)
328 		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
329 }
330 
331 static inline void md_sync_acct(struct block_device *bdev, unsigned long nr_sectors)
332 {
333         atomic_add(nr_sectors, &bdev->bd_contains->bd_disk->sync_io);
334 }
335 
336 struct mdk_personality
337 {
338 	char *name;
339 	int level;
340 	struct list_head list;
341 	struct module *owner;
342 	int (*make_request)(mddev_t *mddev, struct bio *bio);
343 	int (*run)(mddev_t *mddev);
344 	int (*stop)(mddev_t *mddev);
345 	void (*status)(struct seq_file *seq, mddev_t *mddev);
346 	/* error_handler must set ->faulty and clear ->in_sync
347 	 * if appropriate, and should abort recovery if needed
348 	 */
349 	void (*error_handler)(mddev_t *mddev, mdk_rdev_t *rdev);
350 	int (*hot_add_disk) (mddev_t *mddev, mdk_rdev_t *rdev);
351 	int (*hot_remove_disk) (mddev_t *mddev, int number);
352 	int (*spare_active) (mddev_t *mddev);
353 	sector_t (*sync_request)(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster);
354 	int (*resize) (mddev_t *mddev, sector_t sectors);
355 	sector_t (*size) (mddev_t *mddev, sector_t sectors, int raid_disks);
356 	int (*check_reshape) (mddev_t *mddev);
357 	int (*start_reshape) (mddev_t *mddev);
358 	void (*finish_reshape) (mddev_t *mddev);
359 	/* quiesce moves between quiescence states
360 	 * 0 - fully active
361 	 * 1 - no new requests allowed
362 	 * others - reserved
363 	 */
364 	void (*quiesce) (mddev_t *mddev, int state);
365 	/* takeover is used to transition an array from one
366 	 * personality to another.  The new personality must be able
367 	 * to handle the data in the current layout.
368 	 * e.g. 2drive raid1 -> 2drive raid5
369 	 *      ndrive raid5 -> degraded n+1drive raid6 with special layout
370 	 * If the takeover succeeds, a new 'private' structure is returned.
371 	 * This needs to be installed and then ->run used to activate the
372 	 * array.
373 	 */
374 	void *(*takeover) (mddev_t *mddev);
375 };
376 
377 
378 struct md_sysfs_entry {
379 	struct attribute attr;
380 	ssize_t (*show)(mddev_t *, char *);
381 	ssize_t (*store)(mddev_t *, const char *, size_t);
382 };
383 extern struct attribute_group md_bitmap_group;
384 
385 static inline char * mdname (mddev_t * mddev)
386 {
387 	return mddev->gendisk ? mddev->gendisk->disk_name : "mdX";
388 }
389 
390 /*
391  * iterates through some rdev ringlist. It's safe to remove the
392  * current 'rdev'. Dont touch 'tmp' though.
393  */
394 #define rdev_for_each_list(rdev, tmp, head)				\
395 	list_for_each_entry_safe(rdev, tmp, head, same_set)
396 
397 /*
398  * iterates through the 'same array disks' ringlist
399  */
400 #define rdev_for_each(rdev, tmp, mddev)				\
401 	list_for_each_entry_safe(rdev, tmp, &((mddev)->disks), same_set)
402 
403 #define rdev_for_each_rcu(rdev, mddev)				\
404 	list_for_each_entry_rcu(rdev, &((mddev)->disks), same_set)
405 
406 typedef struct mdk_thread_s {
407 	void			(*run) (mddev_t *mddev);
408 	mddev_t			*mddev;
409 	wait_queue_head_t	wqueue;
410 	unsigned long           flags;
411 	struct task_struct	*tsk;
412 	unsigned long		timeout;
413 } mdk_thread_t;
414 
415 #define THREAD_WAKEUP  0
416 
417 #define __wait_event_lock_irq(wq, condition, lock, cmd) 		\
418 do {									\
419 	wait_queue_t __wait;						\
420 	init_waitqueue_entry(&__wait, current);				\
421 									\
422 	add_wait_queue(&wq, &__wait);					\
423 	for (;;) {							\
424 		set_current_state(TASK_UNINTERRUPTIBLE);		\
425 		if (condition)						\
426 			break;						\
427 		spin_unlock_irq(&lock);					\
428 		cmd;							\
429 		schedule();						\
430 		spin_lock_irq(&lock);					\
431 	}								\
432 	current->state = TASK_RUNNING;					\
433 	remove_wait_queue(&wq, &__wait);				\
434 } while (0)
435 
436 #define wait_event_lock_irq(wq, condition, lock, cmd) 			\
437 do {									\
438 	if (condition)	 						\
439 		break;							\
440 	__wait_event_lock_irq(wq, condition, lock, cmd);		\
441 } while (0)
442 
443 static inline void safe_put_page(struct page *p)
444 {
445 	if (p) put_page(p);
446 }
447 
448 extern int register_md_personality(struct mdk_personality *p);
449 extern int unregister_md_personality(struct mdk_personality *p);
450 extern mdk_thread_t * md_register_thread(void (*run) (mddev_t *mddev),
451 				mddev_t *mddev, const char *name);
452 extern void md_unregister_thread(mdk_thread_t *thread);
453 extern void md_wakeup_thread(mdk_thread_t *thread);
454 extern void md_check_recovery(mddev_t *mddev);
455 extern void md_write_start(mddev_t *mddev, struct bio *bi);
456 extern void md_write_end(mddev_t *mddev);
457 extern void md_done_sync(mddev_t *mddev, int blocks, int ok);
458 extern void md_error(mddev_t *mddev, mdk_rdev_t *rdev);
459 
460 extern int mddev_congested(mddev_t *mddev, int bits);
461 extern void md_barrier_request(mddev_t *mddev, struct bio *bio);
462 extern void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
463 			   sector_t sector, int size, struct page *page);
464 extern void md_super_wait(mddev_t *mddev);
465 extern int sync_page_io(struct block_device *bdev, sector_t sector, int size,
466 			struct page *page, int rw);
467 extern void md_do_sync(mddev_t *mddev);
468 extern void md_new_event(mddev_t *mddev);
469 extern int md_allow_write(mddev_t *mddev);
470 extern void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev);
471 extern void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors);
472 extern int md_check_no_bitmap(mddev_t *mddev);
473 extern int md_integrity_register(mddev_t *mddev);
474 extern void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev);
475 extern int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale);
476 extern void restore_bitmap_write_access(struct file *file);
477 
478 #endif /* _MD_MD_H */
479