xref: /openbmc/linux/drivers/md/md.h (revision 82ced6fd)
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  * options passed in raidrun:
34  */
35 
36 /* Currently this must fit in an 'int' */
37 #define MAX_CHUNK_SIZE (1<<30)
38 
39 /*
40  * MD's 'extended' device
41  */
42 struct mdk_rdev_s
43 {
44 	struct list_head same_set;	/* RAID devices within the same set */
45 
46 	sector_t sectors;		/* Device size (in 512bytes sectors) */
47 	mddev_t *mddev;			/* RAID array if running */
48 	int last_events;		/* IO event timestamp */
49 
50 	struct block_device *bdev;	/* block device handle */
51 
52 	struct page	*sb_page;
53 	int		sb_loaded;
54 	__u64		sb_events;
55 	sector_t	data_offset;	/* start of data in array */
56 	sector_t 	sb_start;	/* offset of the super block (in 512byte sectors) */
57 	int		sb_size;	/* bytes in the superblock */
58 	int		preferred_minor;	/* autorun support */
59 
60 	struct kobject	kobj;
61 
62 	/* A device can be in one of three states based on two flags:
63 	 * Not working:   faulty==1 in_sync==0
64 	 * Fully working: faulty==0 in_sync==1
65 	 * Working, but not
66 	 * in sync with array
67 	 *                faulty==0 in_sync==0
68 	 *
69 	 * It can never have faulty==1, in_sync==1
70 	 * This reduces the burden of testing multiple flags in many cases
71 	 */
72 
73 	unsigned long	flags;
74 #define	Faulty		1		/* device is known to have a fault */
75 #define	In_sync		2		/* device is in_sync with rest of array */
76 #define	WriteMostly	4		/* Avoid reading if at all possible */
77 #define	BarriersNotsupp	5		/* BIO_RW_BARRIER is not supported */
78 #define	AllReserved	6		/* If whole device is reserved for
79 					 * one array */
80 #define	AutoDetected	7		/* added by auto-detect */
81 #define Blocked		8		/* An error occured on an externally
82 					 * managed array, don't allow writes
83 					 * until it is cleared */
84 #define StateChanged	9		/* Faulty or Blocked has changed during
85 					 * interrupt, so it needs to be
86 					 * notified by the thread */
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 saved_raid_disk;		/* role that device used to have in the
92 					 * array and could again if we did a partial
93 					 * resync from the bitmap
94 					 */
95 	sector_t	recovery_offset;/* If this device has been partially
96 					 * recovered, this is where we were
97 					 * up to.
98 					 */
99 
100 	atomic_t	nr_pending;	/* number of pending requests.
101 					 * only maintained for arrays that
102 					 * support hot removal
103 					 */
104 	atomic_t	read_errors;	/* number of consecutive read errors that
105 					 * we have tried to ignore.
106 					 */
107 	atomic_t	corrected_errors; /* number of corrected read errors,
108 					   * for reporting to userspace and storing
109 					   * in superblock.
110 					   */
111 	struct work_struct del_work;	/* used for delayed sysfs removal */
112 
113 	struct sysfs_dirent *sysfs_state; /* handle for 'state'
114 					   * sysfs entry */
115 };
116 
117 struct mddev_s
118 {
119 	void				*private;
120 	struct mdk_personality		*pers;
121 	dev_t				unit;
122 	int				md_minor;
123 	struct list_head 		disks;
124 	unsigned long			flags;
125 #define MD_CHANGE_DEVS	0	/* Some device status has changed */
126 #define MD_CHANGE_CLEAN 1	/* transition to or from 'clean' */
127 #define MD_CHANGE_PENDING 2	/* superblock update in progress */
128 
129 	int				suspended;
130 	atomic_t			active_io;
131 	int				ro;
132 
133 	struct gendisk			*gendisk;
134 
135 	struct kobject			kobj;
136 	int				hold_active;
137 #define	UNTIL_IOCTL	1
138 #define	UNTIL_STOP	2
139 
140 	/* Superblock information */
141 	int				major_version,
142 					minor_version,
143 					patch_version;
144 	int				persistent;
145 	int 				external;	/* metadata is
146 							 * managed externally */
147 	char				metadata_type[17]; /* externally set*/
148 	int				chunk_size;
149 	time_t				ctime, utime;
150 	int				level, layout;
151 	char				clevel[16];
152 	int				raid_disks;
153 	int				max_disks;
154 	sector_t			dev_sectors; 	/* used size of
155 							 * component devices */
156 	sector_t			array_sectors; /* exported array size */
157 	int				external_size; /* size managed
158 							* externally */
159 	__u64				events;
160 
161 	char				uuid[16];
162 
163 	/* If the array is being reshaped, we need to record the
164 	 * new shape and an indication of where we are up to.
165 	 * This is written to the superblock.
166 	 * If reshape_position is MaxSector, then no reshape is happening (yet).
167 	 */
168 	sector_t			reshape_position;
169 	int				delta_disks, new_level, new_layout, new_chunk;
170 
171 	struct mdk_thread_s		*thread;	/* management thread */
172 	struct mdk_thread_s		*sync_thread;	/* doing resync or reconstruct */
173 	sector_t			curr_resync;	/* last block scheduled */
174 	/* As resync requests can complete out of order, we cannot easily track
175 	 * how much resync has been completed.  So we occasionally pause until
176 	 * everything completes, then set curr_resync_completed to curr_resync.
177 	 * As such it may be well behind the real resync mark, but it is a value
178 	 * we are certain of.
179 	 */
180 	sector_t			curr_resync_completed;
181 	unsigned long			resync_mark;	/* a recent timestamp */
182 	sector_t			resync_mark_cnt;/* blocks written at resync_mark */
183 	sector_t			curr_mark_cnt; /* blocks scheduled now */
184 
185 	sector_t			resync_max_sectors; /* may be set by personality */
186 
187 	sector_t			resync_mismatches; /* count of sectors where
188 							    * parity/replica mismatch found
189 							    */
190 
191 	/* allow user-space to request suspension of IO to regions of the array */
192 	sector_t			suspend_lo;
193 	sector_t			suspend_hi;
194 	/* if zero, use the system-wide default */
195 	int				sync_speed_min;
196 	int				sync_speed_max;
197 
198 	/* resync even though the same disks are shared among md-devices */
199 	int				parallel_resync;
200 
201 	int				ok_start_degraded;
202 	/* recovery/resync flags
203 	 * NEEDED:   we might need to start a resync/recover
204 	 * RUNNING:  a thread is running, or about to be started
205 	 * SYNC:     actually doing a resync, not a recovery
206 	 * RECOVER:  doing recovery, or need to try it.
207 	 * INTR:     resync needs to be aborted for some reason
208 	 * DONE:     thread is done and is waiting to be reaped
209 	 * REQUEST:  user-space has requested a sync (used with SYNC)
210 	 * CHECK:    user-space request for for check-only, no repair
211 	 * RESHAPE:  A reshape is happening
212 	 *
213 	 * If neither SYNC or RESHAPE are set, then it is a recovery.
214 	 */
215 #define	MD_RECOVERY_RUNNING	0
216 #define	MD_RECOVERY_SYNC	1
217 #define	MD_RECOVERY_RECOVER	2
218 #define	MD_RECOVERY_INTR	3
219 #define	MD_RECOVERY_DONE	4
220 #define	MD_RECOVERY_NEEDED	5
221 #define	MD_RECOVERY_REQUESTED	6
222 #define	MD_RECOVERY_CHECK	7
223 #define MD_RECOVERY_RESHAPE	8
224 #define	MD_RECOVERY_FROZEN	9
225 
226 	unsigned long			recovery;
227 	int				recovery_disabled; /* if we detect that recovery
228 							    * will always fail, set this
229 							    * so we don't loop trying */
230 
231 	int				in_sync;	/* know to not need resync */
232 	struct mutex			reconfig_mutex;
233 	atomic_t			active;		/* general refcount */
234 	atomic_t			openers;	/* number of active opens */
235 
236 	int				changed;	/* true if we might need to reread partition info */
237 	int				degraded;	/* whether md should consider
238 							 * adding a spare
239 							 */
240 	int				barriers_work;	/* initialised to true, cleared as soon
241 							 * as a barrier request to slave
242 							 * fails.  Only supported
243 							 */
244 	struct bio			*biolist; 	/* bios that need to be retried
245 							 * because BIO_RW_BARRIER is not supported
246 							 */
247 
248 	atomic_t			recovery_active; /* blocks scheduled, but not written */
249 	wait_queue_head_t		recovery_wait;
250 	sector_t			recovery_cp;
251 	sector_t			resync_min;	/* user requested sync
252 							 * starts here */
253 	sector_t			resync_max;	/* resync should pause
254 							 * when it gets here */
255 
256 	struct sysfs_dirent		*sysfs_state;	/* handle for 'array_state'
257 							 * file in sysfs.
258 							 */
259 	struct sysfs_dirent		*sysfs_action;  /* handle for 'sync_action' */
260 
261 	struct work_struct del_work;	/* used for delayed sysfs removal */
262 
263 	spinlock_t			write_lock;
264 	wait_queue_head_t		sb_wait;	/* for waiting on superblock updates */
265 	atomic_t			pending_writes;	/* number of active superblock writes */
266 
267 	unsigned int			safemode;	/* if set, update "clean" superblock
268 							 * when no writes pending.
269 							 */
270 	unsigned int			safemode_delay;
271 	struct timer_list		safemode_timer;
272 	atomic_t			writes_pending;
273 	struct request_queue		*queue;	/* for plugging ... */
274 
275 	atomic_t                        write_behind; /* outstanding async IO */
276 	unsigned int                    max_write_behind; /* 0 = sync */
277 
278 	struct bitmap                   *bitmap; /* the bitmap for the device */
279 	struct file			*bitmap_file; /* the bitmap file */
280 	long				bitmap_offset; /* offset from superblock of
281 							* start of bitmap. May be
282 							* negative, but not '0'
283 							*/
284 	long				default_bitmap_offset; /* this is the offset to use when
285 								* hot-adding a bitmap.  It should
286 								* eventually be settable by sysfs.
287 								*/
288 
289 	struct list_head		all_mddevs;
290 };
291 
292 
293 static inline void rdev_dec_pending(mdk_rdev_t *rdev, mddev_t *mddev)
294 {
295 	int faulty = test_bit(Faulty, &rdev->flags);
296 	if (atomic_dec_and_test(&rdev->nr_pending) && faulty)
297 		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
298 }
299 
300 static inline void md_sync_acct(struct block_device *bdev, unsigned long nr_sectors)
301 {
302         atomic_add(nr_sectors, &bdev->bd_contains->bd_disk->sync_io);
303 }
304 
305 struct mdk_personality
306 {
307 	char *name;
308 	int level;
309 	struct list_head list;
310 	struct module *owner;
311 	int (*make_request)(struct request_queue *q, struct bio *bio);
312 	int (*run)(mddev_t *mddev);
313 	int (*stop)(mddev_t *mddev);
314 	void (*status)(struct seq_file *seq, mddev_t *mddev);
315 	/* error_handler must set ->faulty and clear ->in_sync
316 	 * if appropriate, and should abort recovery if needed
317 	 */
318 	void (*error_handler)(mddev_t *mddev, mdk_rdev_t *rdev);
319 	int (*hot_add_disk) (mddev_t *mddev, mdk_rdev_t *rdev);
320 	int (*hot_remove_disk) (mddev_t *mddev, int number);
321 	int (*spare_active) (mddev_t *mddev);
322 	sector_t (*sync_request)(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster);
323 	int (*resize) (mddev_t *mddev, sector_t sectors);
324 	sector_t (*size) (mddev_t *mddev, sector_t sectors, int raid_disks);
325 	int (*check_reshape) (mddev_t *mddev);
326 	int (*start_reshape) (mddev_t *mddev);
327 	void (*finish_reshape) (mddev_t *mddev);
328 	int (*reconfig) (mddev_t *mddev, int layout, int chunk_size);
329 	/* quiesce moves between quiescence states
330 	 * 0 - fully active
331 	 * 1 - no new requests allowed
332 	 * others - reserved
333 	 */
334 	void (*quiesce) (mddev_t *mddev, int state);
335 	/* takeover is used to transition an array from one
336 	 * personality to another.  The new personality must be able
337 	 * to handle the data in the current layout.
338 	 * e.g. 2drive raid1 -> 2drive raid5
339 	 *      ndrive raid5 -> degraded n+1drive raid6 with special layout
340 	 * If the takeover succeeds, a new 'private' structure is returned.
341 	 * This needs to be installed and then ->run used to activate the
342 	 * array.
343 	 */
344 	void *(*takeover) (mddev_t *mddev);
345 };
346 
347 
348 struct md_sysfs_entry {
349 	struct attribute attr;
350 	ssize_t (*show)(mddev_t *, char *);
351 	ssize_t (*store)(mddev_t *, const char *, size_t);
352 };
353 
354 
355 static inline char * mdname (mddev_t * mddev)
356 {
357 	return mddev->gendisk ? mddev->gendisk->disk_name : "mdX";
358 }
359 
360 /*
361  * iterates through some rdev ringlist. It's safe to remove the
362  * current 'rdev'. Dont touch 'tmp' though.
363  */
364 #define rdev_for_each_list(rdev, tmp, head)				\
365 	list_for_each_entry_safe(rdev, tmp, head, same_set)
366 
367 /*
368  * iterates through the 'same array disks' ringlist
369  */
370 #define rdev_for_each(rdev, tmp, mddev)				\
371 	list_for_each_entry_safe(rdev, tmp, &((mddev)->disks), same_set)
372 
373 #define rdev_for_each_rcu(rdev, mddev)				\
374 	list_for_each_entry_rcu(rdev, &((mddev)->disks), same_set)
375 
376 typedef struct mdk_thread_s {
377 	void			(*run) (mddev_t *mddev);
378 	mddev_t			*mddev;
379 	wait_queue_head_t	wqueue;
380 	unsigned long           flags;
381 	struct task_struct	*tsk;
382 	unsigned long		timeout;
383 } mdk_thread_t;
384 
385 #define THREAD_WAKEUP  0
386 
387 #define __wait_event_lock_irq(wq, condition, lock, cmd) 		\
388 do {									\
389 	wait_queue_t __wait;						\
390 	init_waitqueue_entry(&__wait, current);				\
391 									\
392 	add_wait_queue(&wq, &__wait);					\
393 	for (;;) {							\
394 		set_current_state(TASK_UNINTERRUPTIBLE);		\
395 		if (condition)						\
396 			break;						\
397 		spin_unlock_irq(&lock);					\
398 		cmd;							\
399 		schedule();						\
400 		spin_lock_irq(&lock);					\
401 	}								\
402 	current->state = TASK_RUNNING;					\
403 	remove_wait_queue(&wq, &__wait);				\
404 } while (0)
405 
406 #define wait_event_lock_irq(wq, condition, lock, cmd) 			\
407 do {									\
408 	if (condition)	 						\
409 		break;							\
410 	__wait_event_lock_irq(wq, condition, lock, cmd);		\
411 } while (0)
412 
413 static inline void safe_put_page(struct page *p)
414 {
415 	if (p) put_page(p);
416 }
417 
418 extern int register_md_personality(struct mdk_personality *p);
419 extern int unregister_md_personality(struct mdk_personality *p);
420 extern mdk_thread_t * md_register_thread(void (*run) (mddev_t *mddev),
421 				mddev_t *mddev, const char *name);
422 extern void md_unregister_thread(mdk_thread_t *thread);
423 extern void md_wakeup_thread(mdk_thread_t *thread);
424 extern void md_check_recovery(mddev_t *mddev);
425 extern void md_write_start(mddev_t *mddev, struct bio *bi);
426 extern void md_write_end(mddev_t *mddev);
427 extern void md_done_sync(mddev_t *mddev, int blocks, int ok);
428 extern void md_error(mddev_t *mddev, mdk_rdev_t *rdev);
429 
430 extern void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
431 			   sector_t sector, int size, struct page *page);
432 extern void md_super_wait(mddev_t *mddev);
433 extern int sync_page_io(struct block_device *bdev, sector_t sector, int size,
434 			struct page *page, int rw);
435 extern void md_do_sync(mddev_t *mddev);
436 extern void md_new_event(mddev_t *mddev);
437 extern int md_allow_write(mddev_t *mddev);
438 extern void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev);
439 extern void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors);
440 
441 #endif /* _MD_MD_H */
442