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 Unmerged, /* device is being added to array and should
132 * be considerred for bvec_merge_fn but not
133 * yet for actual IO
134 */
135 WriteMostly, /* Avoid reading if at all possible */
136 AutoDetected, /* added by auto-detect */
137 Blocked, /* An error occurred but has not yet
138 * been acknowledged by the metadata
139 * handler, so don't allow writes
140 * until it is cleared */
141 WriteErrorSeen, /* A write error has been seen on this
142 * device
143 */
144 FaultRecorded, /* Intermediate state for clearing
145 * Blocked. The Fault is/will-be
146 * recorded in the metadata, but that
147 * metadata hasn't been stored safely
148 * on disk yet.
149 */
150 BlockedBadBlocks, /* A writer is blocked because they
151 * found an unacknowledged bad-block.
152 * This can safely be cleared at any
153 * time, and the writer will re-check.
154 * It may be set at any time, and at
155 * worst the writer will timeout and
156 * re-check. So setting it as
157 * accurately as possible is good, but
158 * not absolutely critical.
159 */
160 WantReplacement, /* This device is a candidate to be
161 * hot-replaced, either because it has
162 * reported some faults, or because
163 * of explicit request.
164 */
165 Replacement, /* This device is a replacement for
166 * a want_replacement device with same
167 * raid_disk number.
168 */
169 };
170
171 #define BB_LEN_MASK (0x00000000000001FFULL)
172 #define BB_OFFSET_MASK (0x7FFFFFFFFFFFFE00ULL)
173 #define BB_ACK_MASK (0x8000000000000000ULL)
174 #define BB_MAX_LEN 512
175 #define BB_OFFSET(x) (((x) & BB_OFFSET_MASK) >> 9)
176 #define BB_LEN(x) (((x) & BB_LEN_MASK) + 1)
177 #define BB_ACK(x) (!!((x) & BB_ACK_MASK))
178 #define BB_MAKE(a, l, ack) (((a)<<9) | ((l)-1) | ((u64)(!!(ack)) << 63))
179
180 extern int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
181 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)182 static inline int is_badblock(struct md_rdev *rdev, sector_t s, int sectors,
183 sector_t *first_bad, int *bad_sectors)
184 {
185 if (unlikely(rdev->badblocks.count)) {
186 int rv = md_is_badblock(&rdev->badblocks, rdev->data_offset + s,
187 sectors,
188 first_bad, bad_sectors);
189 if (rv)
190 *first_bad -= rdev->data_offset;
191 return rv;
192 }
193 return 0;
194 }
195 extern int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
196 int acknowledged);
197 extern int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors);
198 extern void md_ack_all_badblocks(struct badblocks *bb);
199
200 struct mddev {
201 void *private;
202 struct md_personality *pers;
203 dev_t unit;
204 int md_minor;
205 struct list_head disks;
206 unsigned long flags;
207 #define MD_CHANGE_DEVS 0 /* Some device status has changed */
208 #define MD_CHANGE_CLEAN 1 /* transition to or from 'clean' */
209 #define MD_CHANGE_PENDING 2 /* switch from 'clean' to 'active' in progress */
210 #define MD_ARRAY_FIRST_USE 3 /* First use of array, needs initialization */
211
212 int suspended;
213 atomic_t active_io;
214 int ro;
215 int sysfs_active; /* set when sysfs deletes
216 * are happening, so run/
217 * takeover/stop are not safe
218 */
219 int ready; /* See when safe to pass
220 * IO requests down */
221 struct gendisk *gendisk;
222
223 struct kobject kobj;
224 int hold_active;
225 #define UNTIL_IOCTL 1
226 #define UNTIL_STOP 2
227
228 /* Superblock information */
229 int major_version,
230 minor_version,
231 patch_version;
232 int persistent;
233 int external; /* metadata is
234 * managed externally */
235 char metadata_type[17]; /* externally set*/
236 int chunk_sectors;
237 time_t ctime, utime;
238 int level, layout;
239 char clevel[16];
240 int raid_disks;
241 int max_disks;
242 sector_t dev_sectors; /* used size of
243 * component devices */
244 sector_t array_sectors; /* exported array size */
245 int external_size; /* size managed
246 * externally */
247 __u64 events;
248 /* If the last 'event' was simply a clean->dirty transition, and
249 * we didn't write it to the spares, then it is safe and simple
250 * to just decrement the event count on a dirty->clean transition.
251 * So we record that possibility here.
252 */
253 int can_decrease_events;
254
255 char uuid[16];
256
257 /* If the array is being reshaped, we need to record the
258 * new shape and an indication of where we are up to.
259 * This is written to the superblock.
260 * If reshape_position is MaxSector, then no reshape is happening (yet).
261 */
262 sector_t reshape_position;
263 int delta_disks, new_level, new_layout;
264 int new_chunk_sectors;
265
266 atomic_t plug_cnt; /* If device is expecting
267 * more bios soon.
268 */
269 struct md_thread *thread; /* management thread */
270 struct md_thread *sync_thread; /* doing resync or reconstruct */
271 sector_t curr_resync; /* last block scheduled */
272 /* As resync requests can complete out of order, we cannot easily track
273 * how much resync has been completed. So we occasionally pause until
274 * everything completes, then set curr_resync_completed to curr_resync.
275 * As such it may be well behind the real resync mark, but it is a value
276 * we are certain of.
277 */
278 sector_t curr_resync_completed;
279 unsigned long resync_mark; /* a recent timestamp */
280 sector_t resync_mark_cnt;/* blocks written at resync_mark */
281 sector_t curr_mark_cnt; /* blocks scheduled now */
282
283 sector_t resync_max_sectors; /* may be set by personality */
284
285 sector_t resync_mismatches; /* count of sectors where
286 * parity/replica mismatch found
287 */
288
289 /* allow user-space to request suspension of IO to regions of the array */
290 sector_t suspend_lo;
291 sector_t suspend_hi;
292 /* if zero, use the system-wide default */
293 int sync_speed_min;
294 int sync_speed_max;
295
296 /* resync even though the same disks are shared among md-devices */
297 int parallel_resync;
298
299 int ok_start_degraded;
300 /* recovery/resync flags
301 * NEEDED: we might need to start a resync/recover
302 * RUNNING: a thread is running, or about to be started
303 * SYNC: actually doing a resync, not a recovery
304 * RECOVER: doing recovery, or need to try it.
305 * INTR: resync needs to be aborted for some reason
306 * DONE: thread is done and is waiting to be reaped
307 * REQUEST: user-space has requested a sync (used with SYNC)
308 * CHECK: user-space request for check-only, no repair
309 * RESHAPE: A reshape is happening
310 *
311 * If neither SYNC or RESHAPE are set, then it is a recovery.
312 */
313 #define MD_RECOVERY_RUNNING 0
314 #define MD_RECOVERY_SYNC 1
315 #define MD_RECOVERY_RECOVER 2
316 #define MD_RECOVERY_INTR 3
317 #define MD_RECOVERY_DONE 4
318 #define MD_RECOVERY_NEEDED 5
319 #define MD_RECOVERY_REQUESTED 6
320 #define MD_RECOVERY_CHECK 7
321 #define MD_RECOVERY_RESHAPE 8
322 #define MD_RECOVERY_FROZEN 9
323
324 unsigned long recovery;
325 /* If a RAID personality determines that recovery (of a particular
326 * device) will fail due to a read error on the source device, it
327 * takes a copy of this number and does not attempt recovery again
328 * until this number changes.
329 */
330 int recovery_disabled;
331
332 int in_sync; /* know to not need resync */
333 /* 'open_mutex' avoids races between 'md_open' and 'do_md_stop', so
334 * that we are never stopping an array while it is open.
335 * 'reconfig_mutex' protects all other reconfiguration.
336 * These locks are separate due to conflicting interactions
337 * with bdev->bd_mutex.
338 * Lock ordering is:
339 * reconfig_mutex -> bd_mutex : e.g. do_md_run -> revalidate_disk
340 * bd_mutex -> open_mutex: e.g. __blkdev_get -> md_open
341 */
342 struct mutex open_mutex;
343 struct mutex reconfig_mutex;
344 atomic_t active; /* general refcount */
345 atomic_t openers; /* number of active opens */
346
347 int changed; /* True if we might need to
348 * reread partition info */
349 int degraded; /* whether md should consider
350 * adding a spare
351 */
352 int merge_check_needed; /* at least one
353 * member device
354 * has a
355 * merge_bvec_fn */
356
357 atomic_t recovery_active; /* blocks scheduled, but not written */
358 wait_queue_head_t recovery_wait;
359 sector_t recovery_cp;
360 sector_t resync_min; /* user requested sync
361 * starts here */
362 sector_t resync_max; /* resync should pause
363 * when it gets here */
364
365 struct sysfs_dirent *sysfs_state; /* handle for 'array_state'
366 * file in sysfs.
367 */
368 struct sysfs_dirent *sysfs_action; /* handle for 'sync_action' */
369
370 struct work_struct del_work; /* used for delayed sysfs removal */
371
372 spinlock_t write_lock;
373 wait_queue_head_t sb_wait; /* for waiting on superblock updates */
374 atomic_t pending_writes; /* number of active superblock writes */
375
376 unsigned int safemode; /* if set, update "clean" superblock
377 * when no writes pending.
378 */
379 unsigned int safemode_delay;
380 struct timer_list safemode_timer;
381 atomic_t writes_pending;
382 struct request_queue *queue; /* for plugging ... */
383
384 struct bitmap *bitmap; /* the bitmap for the device */
385 struct {
386 struct file *file; /* the bitmap file */
387 loff_t offset; /* offset from superblock of
388 * start of bitmap. May be
389 * negative, but not '0'
390 * For external metadata, offset
391 * from start of device.
392 */
393 loff_t default_offset; /* this is the offset to use when
394 * hot-adding a bitmap. It should
395 * eventually be settable by sysfs.
396 */
397 struct mutex mutex;
398 unsigned long chunksize;
399 unsigned long daemon_sleep; /* how many jiffies between updates? */
400 unsigned long max_write_behind; /* write-behind mode */
401 int external;
402 } bitmap_info;
403
404 atomic_t max_corr_read_errors; /* max read retries */
405 struct list_head all_mddevs;
406
407 struct attribute_group *to_remove;
408
409 struct bio_set *bio_set;
410
411 /* Generic flush handling.
412 * The last to finish preflush schedules a worker to submit
413 * the rest of the request (without the REQ_FLUSH flag).
414 */
415 struct bio *flush_bio;
416 atomic_t flush_pending;
417 struct work_struct flush_work;
418 struct work_struct event_work; /* used by dm to report failure event */
419 void (*sync_super)(struct mddev *mddev, struct md_rdev *rdev);
420 };
421
422
rdev_dec_pending(struct md_rdev * rdev,struct mddev * mddev)423 static inline void rdev_dec_pending(struct md_rdev *rdev, struct mddev *mddev)
424 {
425 int faulty = test_bit(Faulty, &rdev->flags);
426 if (atomic_dec_and_test(&rdev->nr_pending) && faulty)
427 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
428 }
429
md_sync_acct(struct block_device * bdev,unsigned long nr_sectors)430 static inline void md_sync_acct(struct block_device *bdev, unsigned long nr_sectors)
431 {
432 atomic_add(nr_sectors, &bdev->bd_contains->bd_disk->sync_io);
433 }
434
435 struct md_personality
436 {
437 char *name;
438 int level;
439 struct list_head list;
440 struct module *owner;
441 void (*make_request)(struct mddev *mddev, struct bio *bio);
442 int (*run)(struct mddev *mddev);
443 int (*stop)(struct mddev *mddev);
444 void (*status)(struct seq_file *seq, struct mddev *mddev);
445 /* error_handler must set ->faulty and clear ->in_sync
446 * if appropriate, and should abort recovery if needed
447 */
448 void (*error_handler)(struct mddev *mddev, struct md_rdev *rdev);
449 int (*hot_add_disk) (struct mddev *mddev, struct md_rdev *rdev);
450 int (*hot_remove_disk) (struct mddev *mddev, struct md_rdev *rdev);
451 int (*spare_active) (struct mddev *mddev);
452 sector_t (*sync_request)(struct mddev *mddev, sector_t sector_nr, int *skipped, int go_faster);
453 int (*resize) (struct mddev *mddev, sector_t sectors);
454 sector_t (*size) (struct mddev *mddev, sector_t sectors, int raid_disks);
455 int (*check_reshape) (struct mddev *mddev);
456 int (*start_reshape) (struct mddev *mddev);
457 void (*finish_reshape) (struct mddev *mddev);
458 /* quiesce moves between quiescence states
459 * 0 - fully active
460 * 1 - no new requests allowed
461 * others - reserved
462 */
463 void (*quiesce) (struct mddev *mddev, int state);
464 /* takeover is used to transition an array from one
465 * personality to another. The new personality must be able
466 * to handle the data in the current layout.
467 * e.g. 2drive raid1 -> 2drive raid5
468 * ndrive raid5 -> degraded n+1drive raid6 with special layout
469 * If the takeover succeeds, a new 'private' structure is returned.
470 * This needs to be installed and then ->run used to activate the
471 * array.
472 */
473 void *(*takeover) (struct mddev *mddev);
474 };
475
476
477 struct md_sysfs_entry {
478 struct attribute attr;
479 ssize_t (*show)(struct mddev *, char *);
480 ssize_t (*store)(struct mddev *, const char *, size_t);
481 };
482 extern struct attribute_group md_bitmap_group;
483
sysfs_get_dirent_safe(struct sysfs_dirent * sd,char * name)484 static inline struct sysfs_dirent *sysfs_get_dirent_safe(struct sysfs_dirent *sd, char *name)
485 {
486 if (sd)
487 return sysfs_get_dirent(sd, NULL, name);
488 return sd;
489 }
sysfs_notify_dirent_safe(struct sysfs_dirent * sd)490 static inline void sysfs_notify_dirent_safe(struct sysfs_dirent *sd)
491 {
492 if (sd)
493 sysfs_notify_dirent(sd);
494 }
495
mdname(struct mddev * mddev)496 static inline char * mdname (struct mddev * mddev)
497 {
498 return mddev->gendisk ? mddev->gendisk->disk_name : "mdX";
499 }
500
sysfs_link_rdev(struct mddev * mddev,struct md_rdev * rdev)501 static inline int sysfs_link_rdev(struct mddev *mddev, struct md_rdev *rdev)
502 {
503 char nm[20];
504 if (!test_bit(Replacement, &rdev->flags)) {
505 sprintf(nm, "rd%d", rdev->raid_disk);
506 return sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
507 } else
508 return 0;
509 }
510
sysfs_unlink_rdev(struct mddev * mddev,struct md_rdev * rdev)511 static inline void sysfs_unlink_rdev(struct mddev *mddev, struct md_rdev *rdev)
512 {
513 char nm[20];
514 if (!test_bit(Replacement, &rdev->flags)) {
515 sprintf(nm, "rd%d", rdev->raid_disk);
516 sysfs_remove_link(&mddev->kobj, nm);
517 }
518 }
519
520 /*
521 * iterates through some rdev ringlist. It's safe to remove the
522 * current 'rdev'. Dont touch 'tmp' though.
523 */
524 #define rdev_for_each_list(rdev, tmp, head) \
525 list_for_each_entry_safe(rdev, tmp, head, same_set)
526
527 /*
528 * iterates through the 'same array disks' ringlist
529 */
530 #define rdev_for_each(rdev, mddev) \
531 list_for_each_entry(rdev, &((mddev)->disks), same_set)
532
533 #define rdev_for_each_safe(rdev, tmp, mddev) \
534 list_for_each_entry_safe(rdev, tmp, &((mddev)->disks), same_set)
535
536 #define rdev_for_each_rcu(rdev, mddev) \
537 list_for_each_entry_rcu(rdev, &((mddev)->disks), same_set)
538
539 struct md_thread {
540 void (*run) (struct mddev *mddev);
541 struct mddev *mddev;
542 wait_queue_head_t wqueue;
543 unsigned long flags;
544 struct task_struct *tsk;
545 unsigned long timeout;
546 };
547
548 #define THREAD_WAKEUP 0
549
550 #define __wait_event_lock_irq(wq, condition, lock, cmd) \
551 do { \
552 wait_queue_t __wait; \
553 init_waitqueue_entry(&__wait, current); \
554 \
555 add_wait_queue(&wq, &__wait); \
556 for (;;) { \
557 set_current_state(TASK_UNINTERRUPTIBLE); \
558 if (condition) \
559 break; \
560 spin_unlock_irq(&lock); \
561 cmd; \
562 schedule(); \
563 spin_lock_irq(&lock); \
564 } \
565 current->state = TASK_RUNNING; \
566 remove_wait_queue(&wq, &__wait); \
567 } while (0)
568
569 #define wait_event_lock_irq(wq, condition, lock, cmd) \
570 do { \
571 if (condition) \
572 break; \
573 __wait_event_lock_irq(wq, condition, lock, cmd); \
574 } while (0)
575
safe_put_page(struct page * p)576 static inline void safe_put_page(struct page *p)
577 {
578 if (p) put_page(p);
579 }
580
581 extern int register_md_personality(struct md_personality *p);
582 extern int unregister_md_personality(struct md_personality *p);
583 extern struct md_thread *md_register_thread(
584 void (*run)(struct mddev *mddev),
585 struct mddev *mddev,
586 const char *name);
587 extern void md_unregister_thread(struct md_thread **threadp);
588 extern void md_wakeup_thread(struct md_thread *thread);
589 extern void md_check_recovery(struct mddev *mddev);
590 extern void md_write_start(struct mddev *mddev, struct bio *bi);
591 extern void md_write_end(struct mddev *mddev);
592 extern void md_done_sync(struct mddev *mddev, int blocks, int ok);
593 extern void md_error(struct mddev *mddev, struct md_rdev *rdev);
594
595 extern int mddev_congested(struct mddev *mddev, int bits);
596 extern void md_flush_request(struct mddev *mddev, struct bio *bio);
597 extern void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
598 sector_t sector, int size, struct page *page);
599 extern void md_super_wait(struct mddev *mddev);
600 extern int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
601 struct page *page, int rw, bool metadata_op);
602 extern void md_do_sync(struct mddev *mddev);
603 extern void md_new_event(struct mddev *mddev);
604 extern int md_allow_write(struct mddev *mddev);
605 extern void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev);
606 extern void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors);
607 extern int md_check_no_bitmap(struct mddev *mddev);
608 extern int md_integrity_register(struct mddev *mddev);
609 extern void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev);
610 extern int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale);
611 extern void restore_bitmap_write_access(struct file *file);
612
613 extern void mddev_init(struct mddev *mddev);
614 extern int md_run(struct mddev *mddev);
615 extern void md_stop(struct mddev *mddev);
616 extern void md_stop_writes(struct mddev *mddev);
617 extern int md_rdev_init(struct md_rdev *rdev);
618
619 extern void mddev_suspend(struct mddev *mddev);
620 extern void mddev_resume(struct mddev *mddev);
621 extern struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
622 struct mddev *mddev);
623 extern struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
624 struct mddev *mddev);
625 extern int mddev_check_plugged(struct mddev *mddev);
626 extern void md_trim_bio(struct bio *bio, int offset, int size);
627 #endif /* _MD_MD_H */
628