1 /*
2 md.h : kernel internal structure of the Linux MD driver
3 Copyright (C) 1996-98 Ingo Molnar, Gadi Oxman
4
5 This program is free software; you can redistribute it and/or modify
6 it under the terms of the GNU General Public License as published by
7 the Free Software Foundation; either version 2, or (at your option)
8 any later version.
9
10 You should have received a copy of the GNU General Public License
11 (for example /usr/src/linux/COPYING); if not, write to the Free
12 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
13 */
14
15 #ifndef _MD_MD_H
16 #define _MD_MD_H
17
18 #include <linux/blkdev.h>
19 #include <linux/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 /* Bad block numbers are stored sorted in a single page.
30 * 64bits is used for each block or extent.
31 * 54 bits are sector number, 9 bits are extent size,
32 * 1 bit is an 'acknowledged' flag.
33 */
34 #define MD_MAX_BADBLOCKS (PAGE_SIZE/8)
35
36 /*
37 * MD's 'extended' device
38 */
39 struct md_rdev {
40 struct list_head same_set; /* RAID devices within the same set */
41
42 sector_t sectors; /* Device size (in 512bytes sectors) */
43 struct mddev *mddev; /* RAID array if running */
44 int last_events; /* IO event timestamp */
45
46 /*
47 * If meta_bdev is non-NULL, it means that a separate device is
48 * being used to store the metadata (superblock/bitmap) which
49 * would otherwise be contained on the same device as the data (bdev).
50 */
51 struct block_device *meta_bdev;
52 struct block_device *bdev; /* block device handle */
53
54 struct page *sb_page, *bb_page;
55 int sb_loaded;
56 __u64 sb_events;
57 sector_t data_offset; /* start of data in array */
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 sector_t recovery_offset;/* If this device has been partially
88 * recovered, this is where we were
89 * up to.
90 */
91
92 atomic_t nr_pending; /* number of pending requests.
93 * only maintained for arrays that
94 * support hot removal
95 */
96 atomic_t read_errors; /* number of consecutive read errors that
97 * we have tried to ignore.
98 */
99 struct timespec last_read_error; /* monotonic time since our
100 * last read error
101 */
102 atomic_t corrected_errors; /* number of corrected read errors,
103 * for reporting to userspace and storing
104 * in superblock.
105 */
106 struct work_struct del_work; /* used for delayed sysfs removal */
107
108 struct sysfs_dirent *sysfs_state; /* handle for 'state'
109 * sysfs entry */
110
111 struct badblocks {
112 int count; /* count of bad blocks */
113 int unacked_exist; /* there probably are unacknowledged
114 * bad blocks. This is only cleared
115 * when a read discovers none
116 */
117 int shift; /* shift from sectors to block size
118 * a -ve shift means badblocks are
119 * disabled.*/
120 u64 *page; /* badblock list */
121 int changed;
122 seqlock_t lock;
123
124 sector_t sector;
125 sector_t size; /* in sectors */
126 } badblocks;
127 };
128 enum flag_bits {
129 Faulty, /* device is known to have a fault */
130 In_sync, /* device is in_sync with rest of array */
131 WriteMostly, /* Avoid reading if at all possible */
132 AutoDetected, /* added by auto-detect */
133 Blocked, /* An error occurred but has not yet
134 * been acknowledged by the metadata
135 * handler, so don't allow writes
136 * until it is cleared */
137 WriteErrorSeen, /* A write error has been seen on this
138 * device
139 */
140 FaultRecorded, /* Intermediate state for clearing
141 * Blocked. The Fault is/will-be
142 * recorded in the metadata, but that
143 * metadata hasn't been stored safely
144 * on disk yet.
145 */
146 BlockedBadBlocks, /* A writer is blocked because they
147 * found an unacknowledged bad-block.
148 * This can safely be cleared at any
149 * time, and the writer will re-check.
150 * It may be set at any time, and at
151 * worst the writer will timeout and
152 * re-check. So setting it as
153 * accurately as possible is good, but
154 * not absolutely critical.
155 */
156 WantReplacement, /* This device is a candidate to be
157 * hot-replaced, either because it has
158 * reported some faults, or because
159 * of explicit request.
160 */
161 Replacement, /* This device is a replacement for
162 * a want_replacement device with same
163 * raid_disk number.
164 */
165 };
166
167 #define BB_LEN_MASK (0x00000000000001FFULL)
168 #define BB_OFFSET_MASK (0x7FFFFFFFFFFFFE00ULL)
169 #define BB_ACK_MASK (0x8000000000000000ULL)
170 #define BB_MAX_LEN 512
171 #define BB_OFFSET(x) (((x) & BB_OFFSET_MASK) >> 9)
172 #define BB_LEN(x) (((x) & BB_LEN_MASK) + 1)
173 #define BB_ACK(x) (!!((x) & BB_ACK_MASK))
174 #define BB_MAKE(a, l, ack) (((a)<<9) | ((l)-1) | ((u64)(!!(ack)) << 63))
175
176 extern int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
177 sector_t *first_bad, int *bad_sectors);
is_badblock(struct md_rdev * rdev,sector_t s,int sectors,sector_t * first_bad,int * bad_sectors)178 static inline int is_badblock(struct md_rdev *rdev, sector_t s, int sectors,
179 sector_t *first_bad, int *bad_sectors)
180 {
181 if (unlikely(rdev->badblocks.count)) {
182 int rv = md_is_badblock(&rdev->badblocks, rdev->data_offset + s,
183 sectors,
184 first_bad, bad_sectors);
185 if (rv)
186 *first_bad -= rdev->data_offset;
187 return rv;
188 }
189 return 0;
190 }
191 extern int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
192 int acknowledged);
193 extern int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors);
194 extern void md_ack_all_badblocks(struct badblocks *bb);
195
196 struct mddev {
197 void *private;
198 struct md_personality *pers;
199 dev_t unit;
200 int md_minor;
201 struct list_head disks;
202 unsigned long flags;
203 #define MD_CHANGE_DEVS 0 /* Some device status has changed */
204 #define MD_CHANGE_CLEAN 1 /* transition to or from 'clean' */
205 #define MD_CHANGE_PENDING 2 /* switch from 'clean' to 'active' in progress */
206 #define MD_ARRAY_FIRST_USE 3 /* First use of array, needs initialization */
207
208 int suspended;
209 atomic_t active_io;
210 int ro;
211 int sysfs_active; /* set when sysfs deletes
212 * are happening, so run/
213 * takeover/stop are not safe
214 */
215 int ready; /* See when safe to pass
216 * IO requests down */
217 struct gendisk *gendisk;
218
219 struct kobject kobj;
220 int hold_active;
221 #define UNTIL_IOCTL 1
222 #define UNTIL_STOP 2
223
224 /* Superblock information */
225 int major_version,
226 minor_version,
227 patch_version;
228 int persistent;
229 int external; /* metadata is
230 * managed externally */
231 char metadata_type[17]; /* externally set*/
232 int chunk_sectors;
233 time_t ctime, utime;
234 int level, layout;
235 char clevel[16];
236 int raid_disks;
237 int max_disks;
238 sector_t dev_sectors; /* used size of
239 * component devices */
240 sector_t array_sectors; /* exported array size */
241 int external_size; /* size managed
242 * externally */
243 __u64 events;
244 /* If the last 'event' was simply a clean->dirty transition, and
245 * we didn't write it to the spares, then it is safe and simple
246 * to just decrement the event count on a dirty->clean transition.
247 * So we record that possibility here.
248 */
249 int can_decrease_events;
250
251 char uuid[16];
252
253 /* If the array is being reshaped, we need to record the
254 * new shape and an indication of where we are up to.
255 * This is written to the superblock.
256 * If reshape_position is MaxSector, then no reshape is happening (yet).
257 */
258 sector_t reshape_position;
259 int delta_disks, new_level, new_layout;
260 int new_chunk_sectors;
261
262 atomic_t plug_cnt; /* If device is expecting
263 * more bios soon.
264 */
265 struct md_thread *thread; /* management thread */
266 struct md_thread *sync_thread; /* doing resync or reconstruct */
267 sector_t curr_resync; /* last block scheduled */
268 /* As resync requests can complete out of order, we cannot easily track
269 * how much resync has been completed. So we occasionally pause until
270 * everything completes, then set curr_resync_completed to curr_resync.
271 * As such it may be well behind the real resync mark, but it is a value
272 * we are certain of.
273 */
274 sector_t curr_resync_completed;
275 unsigned long resync_mark; /* a recent timestamp */
276 sector_t resync_mark_cnt;/* blocks written at resync_mark */
277 sector_t curr_mark_cnt; /* blocks scheduled now */
278
279 sector_t resync_max_sectors; /* may be set by personality */
280
281 sector_t resync_mismatches; /* count of sectors where
282 * parity/replica mismatch found
283 */
284
285 /* allow user-space to request suspension of IO to regions of the array */
286 sector_t suspend_lo;
287 sector_t suspend_hi;
288 /* if zero, use the system-wide default */
289 int sync_speed_min;
290 int sync_speed_max;
291
292 /* resync even though the same disks are shared among md-devices */
293 int parallel_resync;
294
295 int ok_start_degraded;
296 /* recovery/resync flags
297 * NEEDED: we might need to start a resync/recover
298 * RUNNING: a thread is running, or about to be started
299 * SYNC: actually doing a resync, not a recovery
300 * RECOVER: doing recovery, or need to try it.
301 * INTR: resync needs to be aborted for some reason
302 * DONE: thread is done and is waiting to be reaped
303 * REQUEST: user-space has requested a sync (used with SYNC)
304 * CHECK: user-space request for check-only, no repair
305 * RESHAPE: A reshape is happening
306 *
307 * If neither SYNC or RESHAPE are set, then it is a recovery.
308 */
309 #define MD_RECOVERY_RUNNING 0
310 #define MD_RECOVERY_SYNC 1
311 #define MD_RECOVERY_RECOVER 2
312 #define MD_RECOVERY_INTR 3
313 #define MD_RECOVERY_DONE 4
314 #define MD_RECOVERY_NEEDED 5
315 #define MD_RECOVERY_REQUESTED 6
316 #define MD_RECOVERY_CHECK 7
317 #define MD_RECOVERY_RESHAPE 8
318 #define MD_RECOVERY_FROZEN 9
319
320 unsigned long recovery;
321 /* If a RAID personality determines that recovery (of a particular
322 * device) will fail due to a read error on the source device, it
323 * takes a copy of this number and does not attempt recovery again
324 * until this number changes.
325 */
326 int recovery_disabled;
327
328 int in_sync; /* know to not need resync */
329 /* 'open_mutex' avoids races between 'md_open' and 'do_md_stop', so
330 * that we are never stopping an array while it is open.
331 * 'reconfig_mutex' protects all other reconfiguration.
332 * These locks are separate due to conflicting interactions
333 * with bdev->bd_mutex.
334 * Lock ordering is:
335 * reconfig_mutex -> bd_mutex : e.g. do_md_run -> revalidate_disk
336 * bd_mutex -> open_mutex: e.g. __blkdev_get -> md_open
337 */
338 struct mutex open_mutex;
339 struct mutex reconfig_mutex;
340 atomic_t active; /* general refcount */
341 atomic_t openers; /* number of active opens */
342
343 int changed; /* True if we might need to
344 * reread partition info */
345 int degraded; /* whether md should consider
346 * adding a spare
347 */
348
349 atomic_t recovery_active; /* blocks scheduled, but not written */
350 wait_queue_head_t recovery_wait;
351 sector_t recovery_cp;
352 sector_t resync_min; /* user requested sync
353 * starts here */
354 sector_t resync_max; /* resync should pause
355 * when it gets here */
356
357 struct sysfs_dirent *sysfs_state; /* handle for 'array_state'
358 * file in sysfs.
359 */
360 struct sysfs_dirent *sysfs_action; /* handle for 'sync_action' */
361
362 struct work_struct del_work; /* used for delayed sysfs removal */
363
364 spinlock_t write_lock;
365 wait_queue_head_t sb_wait; /* for waiting on superblock updates */
366 atomic_t pending_writes; /* number of active superblock writes */
367
368 unsigned int safemode; /* if set, update "clean" superblock
369 * when no writes pending.
370 */
371 unsigned int safemode_delay;
372 struct timer_list safemode_timer;
373 atomic_t writes_pending;
374 struct request_queue *queue; /* for plugging ... */
375
376 struct bitmap *bitmap; /* the bitmap for the device */
377 struct {
378 struct file *file; /* the bitmap file */
379 loff_t offset; /* offset from superblock of
380 * start of bitmap. May be
381 * negative, but not '0'
382 * For external metadata, offset
383 * from start of device.
384 */
385 loff_t default_offset; /* this is the offset to use when
386 * hot-adding a bitmap. It should
387 * eventually be settable by sysfs.
388 */
389 struct mutex mutex;
390 unsigned long chunksize;
391 unsigned long daemon_sleep; /* how many jiffies between updates? */
392 unsigned long max_write_behind; /* write-behind mode */
393 int external;
394 } bitmap_info;
395
396 atomic_t max_corr_read_errors; /* max read retries */
397 struct list_head all_mddevs;
398
399 struct attribute_group *to_remove;
400
401 struct bio_set *bio_set;
402
403 /* Generic flush handling.
404 * The last to finish preflush schedules a worker to submit
405 * the rest of the request (without the REQ_FLUSH flag).
406 */
407 struct bio *flush_bio;
408 atomic_t flush_pending;
409 struct work_struct flush_work;
410 struct work_struct event_work; /* used by dm to report failure event */
411 void (*sync_super)(struct mddev *mddev, struct md_rdev *rdev);
412 };
413
414
rdev_dec_pending(struct md_rdev * rdev,struct mddev * mddev)415 static inline void rdev_dec_pending(struct md_rdev *rdev, struct mddev *mddev)
416 {
417 int faulty = test_bit(Faulty, &rdev->flags);
418 if (atomic_dec_and_test(&rdev->nr_pending) && faulty)
419 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
420 }
421
md_sync_acct(struct block_device * bdev,unsigned long nr_sectors)422 static inline void md_sync_acct(struct block_device *bdev, unsigned long nr_sectors)
423 {
424 atomic_add(nr_sectors, &bdev->bd_contains->bd_disk->sync_io);
425 }
426
427 struct md_personality
428 {
429 char *name;
430 int level;
431 struct list_head list;
432 struct module *owner;
433 void (*make_request)(struct mddev *mddev, struct bio *bio);
434 int (*run)(struct mddev *mddev);
435 int (*stop)(struct mddev *mddev);
436 void (*status)(struct seq_file *seq, struct mddev *mddev);
437 /* error_handler must set ->faulty and clear ->in_sync
438 * if appropriate, and should abort recovery if needed
439 */
440 void (*error_handler)(struct mddev *mddev, struct md_rdev *rdev);
441 int (*hot_add_disk) (struct mddev *mddev, struct md_rdev *rdev);
442 int (*hot_remove_disk) (struct mddev *mddev, struct md_rdev *rdev);
443 int (*spare_active) (struct mddev *mddev);
444 sector_t (*sync_request)(struct mddev *mddev, sector_t sector_nr, int *skipped, int go_faster);
445 int (*resize) (struct mddev *mddev, sector_t sectors);
446 sector_t (*size) (struct mddev *mddev, sector_t sectors, int raid_disks);
447 int (*check_reshape) (struct mddev *mddev);
448 int (*start_reshape) (struct mddev *mddev);
449 void (*finish_reshape) (struct mddev *mddev);
450 /* quiesce moves between quiescence states
451 * 0 - fully active
452 * 1 - no new requests allowed
453 * others - reserved
454 */
455 void (*quiesce) (struct mddev *mddev, int state);
456 /* takeover is used to transition an array from one
457 * personality to another. The new personality must be able
458 * to handle the data in the current layout.
459 * e.g. 2drive raid1 -> 2drive raid5
460 * ndrive raid5 -> degraded n+1drive raid6 with special layout
461 * If the takeover succeeds, a new 'private' structure is returned.
462 * This needs to be installed and then ->run used to activate the
463 * array.
464 */
465 void *(*takeover) (struct mddev *mddev);
466 };
467
468
469 struct md_sysfs_entry {
470 struct attribute attr;
471 ssize_t (*show)(struct mddev *, char *);
472 ssize_t (*store)(struct mddev *, const char *, size_t);
473 };
474 extern struct attribute_group md_bitmap_group;
475
sysfs_get_dirent_safe(struct sysfs_dirent * sd,char * name)476 static inline struct sysfs_dirent *sysfs_get_dirent_safe(struct sysfs_dirent *sd, char *name)
477 {
478 if (sd)
479 return sysfs_get_dirent(sd, NULL, name);
480 return sd;
481 }
sysfs_notify_dirent_safe(struct sysfs_dirent * sd)482 static inline void sysfs_notify_dirent_safe(struct sysfs_dirent *sd)
483 {
484 if (sd)
485 sysfs_notify_dirent(sd);
486 }
487
mdname(struct mddev * mddev)488 static inline char * mdname (struct mddev * mddev)
489 {
490 return mddev->gendisk ? mddev->gendisk->disk_name : "mdX";
491 }
492
sysfs_link_rdev(struct mddev * mddev,struct md_rdev * rdev)493 static inline int sysfs_link_rdev(struct mddev *mddev, struct md_rdev *rdev)
494 {
495 char nm[20];
496 if (!test_bit(Replacement, &rdev->flags)) {
497 sprintf(nm, "rd%d", rdev->raid_disk);
498 return sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
499 } else
500 return 0;
501 }
502
sysfs_unlink_rdev(struct mddev * mddev,struct md_rdev * rdev)503 static inline void sysfs_unlink_rdev(struct mddev *mddev, struct md_rdev *rdev)
504 {
505 char nm[20];
506 if (!test_bit(Replacement, &rdev->flags)) {
507 sprintf(nm, "rd%d", rdev->raid_disk);
508 sysfs_remove_link(&mddev->kobj, nm);
509 }
510 }
511
512 /*
513 * iterates through some rdev ringlist. It's safe to remove the
514 * current 'rdev'. Dont touch 'tmp' though.
515 */
516 #define rdev_for_each_list(rdev, tmp, head) \
517 list_for_each_entry_safe(rdev, tmp, head, same_set)
518
519 /*
520 * iterates through the 'same array disks' ringlist
521 */
522 #define rdev_for_each(rdev, tmp, mddev) \
523 list_for_each_entry_safe(rdev, tmp, &((mddev)->disks), same_set)
524
525 #define rdev_for_each_rcu(rdev, mddev) \
526 list_for_each_entry_rcu(rdev, &((mddev)->disks), same_set)
527
528 struct md_thread {
529 void (*run) (struct mddev *mddev);
530 struct mddev *mddev;
531 wait_queue_head_t wqueue;
532 unsigned long flags;
533 struct task_struct *tsk;
534 unsigned long timeout;
535 };
536
537 #define THREAD_WAKEUP 0
538
539 #define __wait_event_lock_irq(wq, condition, lock, cmd) \
540 do { \
541 wait_queue_t __wait; \
542 init_waitqueue_entry(&__wait, current); \
543 \
544 add_wait_queue(&wq, &__wait); \
545 for (;;) { \
546 set_current_state(TASK_UNINTERRUPTIBLE); \
547 if (condition) \
548 break; \
549 spin_unlock_irq(&lock); \
550 cmd; \
551 schedule(); \
552 spin_lock_irq(&lock); \
553 } \
554 current->state = TASK_RUNNING; \
555 remove_wait_queue(&wq, &__wait); \
556 } while (0)
557
558 #define wait_event_lock_irq(wq, condition, lock, cmd) \
559 do { \
560 if (condition) \
561 break; \
562 __wait_event_lock_irq(wq, condition, lock, cmd); \
563 } while (0)
564
safe_put_page(struct page * p)565 static inline void safe_put_page(struct page *p)
566 {
567 if (p) put_page(p);
568 }
569
570 extern int register_md_personality(struct md_personality *p);
571 extern int unregister_md_personality(struct md_personality *p);
572 extern struct md_thread *md_register_thread(
573 void (*run)(struct mddev *mddev),
574 struct mddev *mddev,
575 const char *name);
576 extern void md_unregister_thread(struct md_thread **threadp);
577 extern void md_wakeup_thread(struct md_thread *thread);
578 extern void md_check_recovery(struct mddev *mddev);
579 extern void md_write_start(struct mddev *mddev, struct bio *bi);
580 extern void md_write_end(struct mddev *mddev);
581 extern void md_done_sync(struct mddev *mddev, int blocks, int ok);
582 extern void md_error(struct mddev *mddev, struct md_rdev *rdev);
583
584 extern int mddev_congested(struct mddev *mddev, int bits);
585 extern void md_flush_request(struct mddev *mddev, struct bio *bio);
586 extern void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
587 sector_t sector, int size, struct page *page);
588 extern void md_super_wait(struct mddev *mddev);
589 extern int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
590 struct page *page, int rw, bool metadata_op);
591 extern void md_do_sync(struct mddev *mddev);
592 extern void md_new_event(struct mddev *mddev);
593 extern int md_allow_write(struct mddev *mddev);
594 extern void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev);
595 extern void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors);
596 extern int md_check_no_bitmap(struct mddev *mddev);
597 extern int md_integrity_register(struct mddev *mddev);
598 extern void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev);
599 extern int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale);
600 extern void restore_bitmap_write_access(struct file *file);
601
602 extern void mddev_init(struct mddev *mddev);
603 extern int md_run(struct mddev *mddev);
604 extern void md_stop(struct mddev *mddev);
605 extern void md_stop_writes(struct mddev *mddev);
606 extern int md_rdev_init(struct md_rdev *rdev);
607
608 extern void mddev_suspend(struct mddev *mddev);
609 extern void mddev_resume(struct mddev *mddev);
610 extern struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
611 struct mddev *mddev);
612 extern struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
613 struct mddev *mddev);
614 extern int mddev_check_plugged(struct mddev *mddev);
615 extern void md_trim_bio(struct bio *bio, int offset, int size);
616 #endif /* _MD_MD_H */
617