Loading...
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 */
39struct 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 new_data_offset;/* only relevant while reshaping */
59 sector_t sb_start; /* offset of the super block (in 512byte sectors) */
60 int sb_size; /* bytes in the superblock */
61 int preferred_minor; /* autorun support */
62
63 struct kobject kobj;
64
65 /* A device can be in one of three states based on two flags:
66 * Not working: faulty==1 in_sync==0
67 * Fully working: faulty==0 in_sync==1
68 * Working, but not
69 * in sync with array
70 * faulty==0 in_sync==0
71 *
72 * It can never have faulty==1, in_sync==1
73 * This reduces the burden of testing multiple flags in many cases
74 */
75
76 unsigned long flags; /* bit set of 'enum flag_bits' bits. */
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 struct badblocks {
113 int count; /* count of bad blocks */
114 int unacked_exist; /* there probably are unacknowledged
115 * bad blocks. This is only cleared
116 * when a read discovers none
117 */
118 int shift; /* shift from sectors to block size
119 * a -ve shift means badblocks are
120 * disabled.*/
121 u64 *page; /* badblock list */
122 int changed;
123 seqlock_t lock;
124
125 sector_t sector;
126 sector_t size; /* in sectors */
127 } badblocks;
128};
129enum flag_bits {
130 Faulty, /* device is known to have a fault */
131 In_sync, /* device is in_sync with rest of array */
132 Unmerged, /* device is being added to array and should
133 * be considerred for bvec_merge_fn but not
134 * yet for actual IO
135 */
136 WriteMostly, /* Avoid reading if at all possible */
137 AutoDetected, /* added by auto-detect */
138 Blocked, /* An error occurred but has not yet
139 * been acknowledged by the metadata
140 * handler, so don't allow writes
141 * until it is cleared */
142 WriteErrorSeen, /* A write error has been seen on this
143 * device
144 */
145 FaultRecorded, /* Intermediate state for clearing
146 * Blocked. The Fault is/will-be
147 * recorded in the metadata, but that
148 * metadata hasn't been stored safely
149 * on disk yet.
150 */
151 BlockedBadBlocks, /* A writer is blocked because they
152 * found an unacknowledged bad-block.
153 * This can safely be cleared at any
154 * time, and the writer will re-check.
155 * It may be set at any time, and at
156 * worst the writer will timeout and
157 * re-check. So setting it as
158 * accurately as possible is good, but
159 * not absolutely critical.
160 */
161 WantReplacement, /* This device is a candidate to be
162 * hot-replaced, either because it has
163 * reported some faults, or because
164 * of explicit request.
165 */
166 Replacement, /* This device is a replacement for
167 * a want_replacement device with same
168 * raid_disk number.
169 */
170};
171
172#define BB_LEN_MASK (0x00000000000001FFULL)
173#define BB_OFFSET_MASK (0x7FFFFFFFFFFFFE00ULL)
174#define BB_ACK_MASK (0x8000000000000000ULL)
175#define BB_MAX_LEN 512
176#define BB_OFFSET(x) (((x) & BB_OFFSET_MASK) >> 9)
177#define BB_LEN(x) (((x) & BB_LEN_MASK) + 1)
178#define BB_ACK(x) (!!((x) & BB_ACK_MASK))
179#define BB_MAKE(a, l, ack) (((a)<<9) | ((l)-1) | ((u64)(!!(ack)) << 63))
180
181extern int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
182 sector_t *first_bad, int *bad_sectors);
183static inline int is_badblock(struct md_rdev *rdev, sector_t s, int sectors,
184 sector_t *first_bad, int *bad_sectors)
185{
186 if (unlikely(rdev->badblocks.count)) {
187 int rv = md_is_badblock(&rdev->badblocks, rdev->data_offset + s,
188 sectors,
189 first_bad, bad_sectors);
190 if (rv)
191 *first_bad -= rdev->data_offset;
192 return rv;
193 }
194 return 0;
195}
196extern int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
197 int is_new);
198extern int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
199 int is_new);
200extern void md_ack_all_badblocks(struct badblocks *bb);
201
202struct mddev {
203 void *private;
204 struct md_personality *pers;
205 dev_t unit;
206 int md_minor;
207 struct list_head disks;
208 unsigned long flags;
209#define MD_CHANGE_DEVS 0 /* Some device status has changed */
210#define MD_CHANGE_CLEAN 1 /* transition to or from 'clean' */
211#define MD_CHANGE_PENDING 2 /* switch from 'clean' to 'active' in progress */
212#define MD_ARRAY_FIRST_USE 3 /* First use of array, needs initialization */
213
214 int suspended;
215 atomic_t active_io;
216 int ro;
217 int sysfs_active; /* set when sysfs deletes
218 * are happening, so run/
219 * takeover/stop are not safe
220 */
221 int ready; /* See when safe to pass
222 * IO requests down */
223 struct gendisk *gendisk;
224
225 struct kobject kobj;
226 int hold_active;
227#define UNTIL_IOCTL 1
228#define UNTIL_STOP 2
229
230 /* Superblock information */
231 int major_version,
232 minor_version,
233 patch_version;
234 int persistent;
235 int external; /* metadata is
236 * managed externally */
237 char metadata_type[17]; /* externally set*/
238 int chunk_sectors;
239 time_t ctime, utime;
240 int level, layout;
241 char clevel[16];
242 int raid_disks;
243 int max_disks;
244 sector_t dev_sectors; /* used size of
245 * component devices */
246 sector_t array_sectors; /* exported array size */
247 int external_size; /* size managed
248 * externally */
249 __u64 events;
250 /* If the last 'event' was simply a clean->dirty transition, and
251 * we didn't write it to the spares, then it is safe and simple
252 * to just decrement the event count on a dirty->clean transition.
253 * So we record that possibility here.
254 */
255 int can_decrease_events;
256
257 char uuid[16];
258
259 /* If the array is being reshaped, we need to record the
260 * new shape and an indication of where we are up to.
261 * This is written to the superblock.
262 * If reshape_position is MaxSector, then no reshape is happening (yet).
263 */
264 sector_t reshape_position;
265 int delta_disks, new_level, new_layout;
266 int new_chunk_sectors;
267 int reshape_backwards;
268
269 atomic_t plug_cnt; /* If device is expecting
270 * more bios soon.
271 */
272 struct md_thread *thread; /* management thread */
273 struct md_thread *sync_thread; /* doing resync or reconstruct */
274 sector_t curr_resync; /* last block scheduled */
275 /* As resync requests can complete out of order, we cannot easily track
276 * how much resync has been completed. So we occasionally pause until
277 * everything completes, then set curr_resync_completed to curr_resync.
278 * As such it may be well behind the real resync mark, but it is a value
279 * we are certain of.
280 */
281 sector_t curr_resync_completed;
282 unsigned long resync_mark; /* a recent timestamp */
283 sector_t resync_mark_cnt;/* blocks written at resync_mark */
284 sector_t curr_mark_cnt; /* blocks scheduled now */
285
286 sector_t resync_max_sectors; /* may be set by personality */
287
288 sector_t resync_mismatches; /* count of sectors where
289 * parity/replica mismatch found
290 */
291
292 /* allow user-space to request suspension of IO to regions of the array */
293 sector_t suspend_lo;
294 sector_t suspend_hi;
295 /* if zero, use the system-wide default */
296 int sync_speed_min;
297 int sync_speed_max;
298
299 /* resync even though the same disks are shared among md-devices */
300 int parallel_resync;
301
302 int ok_start_degraded;
303 /* recovery/resync flags
304 * NEEDED: we might need to start a resync/recover
305 * RUNNING: a thread is running, or about to be started
306 * SYNC: actually doing a resync, not a recovery
307 * RECOVER: doing recovery, or need to try it.
308 * INTR: resync needs to be aborted for some reason
309 * DONE: thread is done and is waiting to be reaped
310 * REQUEST: user-space has requested a sync (used with SYNC)
311 * CHECK: user-space request for check-only, no repair
312 * RESHAPE: A reshape is happening
313 *
314 * If neither SYNC or RESHAPE are set, then it is a recovery.
315 */
316#define MD_RECOVERY_RUNNING 0
317#define MD_RECOVERY_SYNC 1
318#define MD_RECOVERY_RECOVER 2
319#define MD_RECOVERY_INTR 3
320#define MD_RECOVERY_DONE 4
321#define MD_RECOVERY_NEEDED 5
322#define MD_RECOVERY_REQUESTED 6
323#define MD_RECOVERY_CHECK 7
324#define MD_RECOVERY_RESHAPE 8
325#define MD_RECOVERY_FROZEN 9
326
327 unsigned long recovery;
328 /* If a RAID personality determines that recovery (of a particular
329 * device) will fail due to a read error on the source device, it
330 * takes a copy of this number and does not attempt recovery again
331 * until this number changes.
332 */
333 int recovery_disabled;
334
335 int in_sync; /* know to not need resync */
336 /* 'open_mutex' avoids races between 'md_open' and 'do_md_stop', so
337 * that we are never stopping an array while it is open.
338 * 'reconfig_mutex' protects all other reconfiguration.
339 * These locks are separate due to conflicting interactions
340 * with bdev->bd_mutex.
341 * Lock ordering is:
342 * reconfig_mutex -> bd_mutex : e.g. do_md_run -> revalidate_disk
343 * bd_mutex -> open_mutex: e.g. __blkdev_get -> md_open
344 */
345 struct mutex open_mutex;
346 struct mutex reconfig_mutex;
347 atomic_t active; /* general refcount */
348 atomic_t openers; /* number of active opens */
349
350 int changed; /* True if we might need to
351 * reread partition info */
352 int degraded; /* whether md should consider
353 * adding a spare
354 */
355 int merge_check_needed; /* at least one
356 * member device
357 * has a
358 * merge_bvec_fn */
359
360 atomic_t recovery_active; /* blocks scheduled, but not written */
361 wait_queue_head_t recovery_wait;
362 sector_t recovery_cp;
363 sector_t resync_min; /* user requested sync
364 * starts here */
365 sector_t resync_max; /* resync should pause
366 * when it gets here */
367
368 struct sysfs_dirent *sysfs_state; /* handle for 'array_state'
369 * file in sysfs.
370 */
371 struct sysfs_dirent *sysfs_action; /* handle for 'sync_action' */
372
373 struct work_struct del_work; /* used for delayed sysfs removal */
374
375 spinlock_t write_lock;
376 wait_queue_head_t sb_wait; /* for waiting on superblock updates */
377 atomic_t pending_writes; /* number of active superblock writes */
378
379 unsigned int safemode; /* if set, update "clean" superblock
380 * when no writes pending.
381 */
382 unsigned int safemode_delay;
383 struct timer_list safemode_timer;
384 atomic_t writes_pending;
385 struct request_queue *queue; /* for plugging ... */
386
387 struct bitmap *bitmap; /* the bitmap for the device */
388 struct {
389 struct file *file; /* the bitmap file */
390 loff_t offset; /* offset from superblock of
391 * start of bitmap. May be
392 * negative, but not '0'
393 * For external metadata, offset
394 * from start of device.
395 */
396 unsigned long space; /* space available at this offset */
397 loff_t default_offset; /* this is the offset to use when
398 * hot-adding a bitmap. It should
399 * eventually be settable by sysfs.
400 */
401 unsigned long default_space; /* space available at
402 * default offset */
403 struct mutex mutex;
404 unsigned long chunksize;
405 unsigned long daemon_sleep; /* how many jiffies between updates? */
406 unsigned long max_write_behind; /* write-behind mode */
407 int external;
408 } bitmap_info;
409
410 atomic_t max_corr_read_errors; /* max read retries */
411 struct list_head all_mddevs;
412
413 struct attribute_group *to_remove;
414
415 struct bio_set *bio_set;
416
417 /* Generic flush handling.
418 * The last to finish preflush schedules a worker to submit
419 * the rest of the request (without the REQ_FLUSH flag).
420 */
421 struct bio *flush_bio;
422 atomic_t flush_pending;
423 struct work_struct flush_work;
424 struct work_struct event_work; /* used by dm to report failure event */
425 void (*sync_super)(struct mddev *mddev, struct md_rdev *rdev);
426};
427
428
429static inline void rdev_dec_pending(struct md_rdev *rdev, struct mddev *mddev)
430{
431 int faulty = test_bit(Faulty, &rdev->flags);
432 if (atomic_dec_and_test(&rdev->nr_pending) && faulty)
433 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
434}
435
436static inline void md_sync_acct(struct block_device *bdev, unsigned long nr_sectors)
437{
438 atomic_add(nr_sectors, &bdev->bd_contains->bd_disk->sync_io);
439}
440
441struct md_personality
442{
443 char *name;
444 int level;
445 struct list_head list;
446 struct module *owner;
447 void (*make_request)(struct mddev *mddev, struct bio *bio);
448 int (*run)(struct mddev *mddev);
449 int (*stop)(struct mddev *mddev);
450 void (*status)(struct seq_file *seq, struct mddev *mddev);
451 /* error_handler must set ->faulty and clear ->in_sync
452 * if appropriate, and should abort recovery if needed
453 */
454 void (*error_handler)(struct mddev *mddev, struct md_rdev *rdev);
455 int (*hot_add_disk) (struct mddev *mddev, struct md_rdev *rdev);
456 int (*hot_remove_disk) (struct mddev *mddev, struct md_rdev *rdev);
457 int (*spare_active) (struct mddev *mddev);
458 sector_t (*sync_request)(struct mddev *mddev, sector_t sector_nr, int *skipped, int go_faster);
459 int (*resize) (struct mddev *mddev, sector_t sectors);
460 sector_t (*size) (struct mddev *mddev, sector_t sectors, int raid_disks);
461 int (*check_reshape) (struct mddev *mddev);
462 int (*start_reshape) (struct mddev *mddev);
463 void (*finish_reshape) (struct mddev *mddev);
464 /* quiesce moves between quiescence states
465 * 0 - fully active
466 * 1 - no new requests allowed
467 * others - reserved
468 */
469 void (*quiesce) (struct mddev *mddev, int state);
470 /* takeover is used to transition an array from one
471 * personality to another. The new personality must be able
472 * to handle the data in the current layout.
473 * e.g. 2drive raid1 -> 2drive raid5
474 * ndrive raid5 -> degraded n+1drive raid6 with special layout
475 * If the takeover succeeds, a new 'private' structure is returned.
476 * This needs to be installed and then ->run used to activate the
477 * array.
478 */
479 void *(*takeover) (struct mddev *mddev);
480};
481
482
483struct md_sysfs_entry {
484 struct attribute attr;
485 ssize_t (*show)(struct mddev *, char *);
486 ssize_t (*store)(struct mddev *, const char *, size_t);
487};
488extern struct attribute_group md_bitmap_group;
489
490static inline struct sysfs_dirent *sysfs_get_dirent_safe(struct sysfs_dirent *sd, char *name)
491{
492 if (sd)
493 return sysfs_get_dirent(sd, NULL, name);
494 return sd;
495}
496static inline void sysfs_notify_dirent_safe(struct sysfs_dirent *sd)
497{
498 if (sd)
499 sysfs_notify_dirent(sd);
500}
501
502static inline char * mdname (struct mddev * mddev)
503{
504 return mddev->gendisk ? mddev->gendisk->disk_name : "mdX";
505}
506
507static inline int sysfs_link_rdev(struct mddev *mddev, struct md_rdev *rdev)
508{
509 char nm[20];
510 if (!test_bit(Replacement, &rdev->flags)) {
511 sprintf(nm, "rd%d", rdev->raid_disk);
512 return sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
513 } else
514 return 0;
515}
516
517static inline void sysfs_unlink_rdev(struct mddev *mddev, struct md_rdev *rdev)
518{
519 char nm[20];
520 if (!test_bit(Replacement, &rdev->flags)) {
521 sprintf(nm, "rd%d", rdev->raid_disk);
522 sysfs_remove_link(&mddev->kobj, nm);
523 }
524}
525
526/*
527 * iterates through some rdev ringlist. It's safe to remove the
528 * current 'rdev'. Dont touch 'tmp' though.
529 */
530#define rdev_for_each_list(rdev, tmp, head) \
531 list_for_each_entry_safe(rdev, tmp, head, same_set)
532
533/*
534 * iterates through the 'same array disks' ringlist
535 */
536#define rdev_for_each(rdev, mddev) \
537 list_for_each_entry(rdev, &((mddev)->disks), same_set)
538
539#define rdev_for_each_safe(rdev, tmp, mddev) \
540 list_for_each_entry_safe(rdev, tmp, &((mddev)->disks), same_set)
541
542#define rdev_for_each_rcu(rdev, mddev) \
543 list_for_each_entry_rcu(rdev, &((mddev)->disks), same_set)
544
545struct md_thread {
546 void (*run) (struct mddev *mddev);
547 struct mddev *mddev;
548 wait_queue_head_t wqueue;
549 unsigned long flags;
550 struct task_struct *tsk;
551 unsigned long timeout;
552};
553
554#define THREAD_WAKEUP 0
555
556#define __wait_event_lock_irq(wq, condition, lock, cmd) \
557do { \
558 wait_queue_t __wait; \
559 init_waitqueue_entry(&__wait, current); \
560 \
561 add_wait_queue(&wq, &__wait); \
562 for (;;) { \
563 set_current_state(TASK_UNINTERRUPTIBLE); \
564 if (condition) \
565 break; \
566 spin_unlock_irq(&lock); \
567 cmd; \
568 schedule(); \
569 spin_lock_irq(&lock); \
570 } \
571 current->state = TASK_RUNNING; \
572 remove_wait_queue(&wq, &__wait); \
573} while (0)
574
575#define wait_event_lock_irq(wq, condition, lock, cmd) \
576do { \
577 if (condition) \
578 break; \
579 __wait_event_lock_irq(wq, condition, lock, cmd); \
580} while (0)
581
582static inline void safe_put_page(struct page *p)
583{
584 if (p) put_page(p);
585}
586
587extern int register_md_personality(struct md_personality *p);
588extern int unregister_md_personality(struct md_personality *p);
589extern struct md_thread *md_register_thread(
590 void (*run)(struct mddev *mddev),
591 struct mddev *mddev,
592 const char *name);
593extern void md_unregister_thread(struct md_thread **threadp);
594extern void md_wakeup_thread(struct md_thread *thread);
595extern void md_check_recovery(struct mddev *mddev);
596extern void md_write_start(struct mddev *mddev, struct bio *bi);
597extern void md_write_end(struct mddev *mddev);
598extern void md_done_sync(struct mddev *mddev, int blocks, int ok);
599extern void md_error(struct mddev *mddev, struct md_rdev *rdev);
600extern void md_finish_reshape(struct mddev *mddev);
601
602extern int mddev_congested(struct mddev *mddev, int bits);
603extern void md_flush_request(struct mddev *mddev, struct bio *bio);
604extern void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
605 sector_t sector, int size, struct page *page);
606extern void md_super_wait(struct mddev *mddev);
607extern int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
608 struct page *page, int rw, bool metadata_op);
609extern void md_do_sync(struct mddev *mddev);
610extern void md_new_event(struct mddev *mddev);
611extern int md_allow_write(struct mddev *mddev);
612extern void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev);
613extern void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors);
614extern int md_check_no_bitmap(struct mddev *mddev);
615extern int md_integrity_register(struct mddev *mddev);
616extern void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev);
617extern int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale);
618extern void restore_bitmap_write_access(struct file *file);
619
620extern void mddev_init(struct mddev *mddev);
621extern int md_run(struct mddev *mddev);
622extern void md_stop(struct mddev *mddev);
623extern void md_stop_writes(struct mddev *mddev);
624extern int md_rdev_init(struct md_rdev *rdev);
625extern void md_rdev_clear(struct md_rdev *rdev);
626
627extern void mddev_suspend(struct mddev *mddev);
628extern void mddev_resume(struct mddev *mddev);
629extern struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
630 struct mddev *mddev);
631extern struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
632 struct mddev *mddev);
633extern int mddev_check_plugged(struct mddev *mddev);
634extern void md_trim_bio(struct bio *bio, int offset, int size);
635#endif /* _MD_MD_H */
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/backing-dev.h>
20#include <linux/badblocks.h>
21#include <linux/kobject.h>
22#include <linux/list.h>
23#include <linux/mm.h>
24#include <linux/mutex.h>
25#include <linux/timer.h>
26#include <linux/wait.h>
27#include <linux/workqueue.h>
28#include "md-cluster.h"
29
30#define MaxSector (~(sector_t)0)
31
32/*
33 * MD's 'extended' device
34 */
35struct md_rdev {
36 struct list_head same_set; /* RAID devices within the same set */
37
38 sector_t sectors; /* Device size (in 512bytes sectors) */
39 struct mddev *mddev; /* RAID array if running */
40 int last_events; /* IO event timestamp */
41
42 /*
43 * If meta_bdev is non-NULL, it means that a separate device is
44 * being used to store the metadata (superblock/bitmap) which
45 * would otherwise be contained on the same device as the data (bdev).
46 */
47 struct block_device *meta_bdev;
48 struct block_device *bdev; /* block device handle */
49
50 struct page *sb_page, *bb_page;
51 int sb_loaded;
52 __u64 sb_events;
53 sector_t data_offset; /* start of data in array */
54 sector_t new_data_offset;/* only relevant while reshaping */
55 sector_t sb_start; /* offset of the super block (in 512byte sectors) */
56 int sb_size; /* bytes in the superblock */
57 int preferred_minor; /* autorun support */
58
59 struct kobject kobj;
60
61 /* A device can be in one of three states based on two flags:
62 * Not working: faulty==1 in_sync==0
63 * Fully working: faulty==0 in_sync==1
64 * Working, but not
65 * in sync with array
66 * faulty==0 in_sync==0
67 *
68 * It can never have faulty==1, in_sync==1
69 * This reduces the burden of testing multiple flags in many cases
70 */
71
72 unsigned long flags; /* bit set of 'enum flag_bits' bits. */
73 wait_queue_head_t blocked_wait;
74
75 int desc_nr; /* descriptor index in the superblock */
76 int raid_disk; /* role of device in array */
77 int new_raid_disk; /* role that the device will have in
78 * the array after a level-change completes.
79 */
80 int saved_raid_disk; /* role that device used to have in the
81 * array and could again if we did a partial
82 * resync from the bitmap
83 */
84 union {
85 sector_t recovery_offset;/* If this device has been partially
86 * recovered, this is where we were
87 * up to.
88 */
89 sector_t journal_tail; /* If this device is a journal device,
90 * this is the journal tail (journal
91 * recovery start point)
92 */
93 };
94
95 atomic_t nr_pending; /* number of pending requests.
96 * only maintained for arrays that
97 * support hot removal
98 */
99 atomic_t read_errors; /* number of consecutive read errors that
100 * we have tried to ignore.
101 */
102 struct timespec last_read_error; /* monotonic time since our
103 * last read error
104 */
105 atomic_t corrected_errors; /* number of corrected read errors,
106 * for reporting to userspace and storing
107 * in superblock.
108 */
109 struct work_struct del_work; /* used for delayed sysfs removal */
110
111 struct kernfs_node *sysfs_state; /* handle for 'state'
112 * sysfs entry */
113
114 struct badblocks badblocks;
115};
116enum flag_bits {
117 Faulty, /* device is known to have a fault */
118 In_sync, /* device is in_sync with rest of array */
119 Bitmap_sync, /* ..actually, not quite In_sync. Need a
120 * bitmap-based recovery to get fully in sync
121 */
122 WriteMostly, /* Avoid reading if at all possible */
123 AutoDetected, /* added by auto-detect */
124 Blocked, /* An error occurred but has not yet
125 * been acknowledged by the metadata
126 * handler, so don't allow writes
127 * until it is cleared */
128 WriteErrorSeen, /* A write error has been seen on this
129 * device
130 */
131 FaultRecorded, /* Intermediate state for clearing
132 * Blocked. The Fault is/will-be
133 * recorded in the metadata, but that
134 * metadata hasn't been stored safely
135 * on disk yet.
136 */
137 BlockedBadBlocks, /* A writer is blocked because they
138 * found an unacknowledged bad-block.
139 * This can safely be cleared at any
140 * time, and the writer will re-check.
141 * It may be set at any time, and at
142 * worst the writer will timeout and
143 * re-check. So setting it as
144 * accurately as possible is good, but
145 * not absolutely critical.
146 */
147 WantReplacement, /* This device is a candidate to be
148 * hot-replaced, either because it has
149 * reported some faults, or because
150 * of explicit request.
151 */
152 Replacement, /* This device is a replacement for
153 * a want_replacement device with same
154 * raid_disk number.
155 */
156 Candidate, /* For clustered environments only:
157 * This device is seen locally but not
158 * by the whole cluster
159 */
160 Journal, /* This device is used as journal for
161 * raid-5/6.
162 * Usually, this device should be faster
163 * than other devices in the array
164 */
165 ClusterRemove,
166};
167
168static inline int is_badblock(struct md_rdev *rdev, sector_t s, int sectors,
169 sector_t *first_bad, int *bad_sectors)
170{
171 if (unlikely(rdev->badblocks.count)) {
172 int rv = badblocks_check(&rdev->badblocks, rdev->data_offset + s,
173 sectors,
174 first_bad, bad_sectors);
175 if (rv)
176 *first_bad -= rdev->data_offset;
177 return rv;
178 }
179 return 0;
180}
181extern int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
182 int is_new);
183extern int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
184 int is_new);
185struct md_cluster_info;
186
187struct mddev {
188 void *private;
189 struct md_personality *pers;
190 dev_t unit;
191 int md_minor;
192 struct list_head disks;
193 unsigned long flags;
194#define MD_CHANGE_DEVS 0 /* Some device status has changed */
195#define MD_CHANGE_CLEAN 1 /* transition to or from 'clean' */
196#define MD_CHANGE_PENDING 2 /* switch from 'clean' to 'active' in progress */
197#define MD_UPDATE_SB_FLAGS (1 | 2 | 4) /* If these are set, md_update_sb needed */
198#define MD_ARRAY_FIRST_USE 3 /* First use of array, needs initialization */
199#define MD_STILL_CLOSED 4 /* If set, then array has not been opened since
200 * md_ioctl checked on it.
201 */
202#define MD_JOURNAL_CLEAN 5 /* A raid with journal is already clean */
203#define MD_HAS_JOURNAL 6 /* The raid array has journal feature set */
204#define MD_RELOAD_SB 7 /* Reload the superblock because another node
205 * updated it.
206 */
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 struct gendisk *gendisk;
216
217 struct kobject kobj;
218 int hold_active;
219#define UNTIL_IOCTL 1
220#define UNTIL_STOP 2
221
222 /* Superblock information */
223 int major_version,
224 minor_version,
225 patch_version;
226 int persistent;
227 int external; /* metadata is
228 * managed externally */
229 char metadata_type[17]; /* externally set*/
230 int chunk_sectors;
231 time64_t ctime, utime;
232 int level, layout;
233 char clevel[16];
234 int raid_disks;
235 int max_disks;
236 sector_t dev_sectors; /* used size of
237 * component devices */
238 sector_t array_sectors; /* exported array size */
239 int external_size; /* size managed
240 * externally */
241 __u64 events;
242 /* If the last 'event' was simply a clean->dirty transition, and
243 * we didn't write it to the spares, then it is safe and simple
244 * to just decrement the event count on a dirty->clean transition.
245 * So we record that possibility here.
246 */
247 int can_decrease_events;
248
249 char uuid[16];
250
251 /* If the array is being reshaped, we need to record the
252 * new shape and an indication of where we are up to.
253 * This is written to the superblock.
254 * If reshape_position is MaxSector, then no reshape is happening (yet).
255 */
256 sector_t reshape_position;
257 int delta_disks, new_level, new_layout;
258 int new_chunk_sectors;
259 int reshape_backwards;
260
261 struct md_thread *thread; /* management thread */
262 struct md_thread *sync_thread; /* doing resync or reconstruct */
263
264 /* 'last_sync_action' is initialized to "none". It is set when a
265 * sync operation (i.e "data-check", "requested-resync", "resync",
266 * "recovery", or "reshape") is started. It holds this value even
267 * when the sync thread is "frozen" (interrupted) or "idle" (stopped
268 * or finished). It is overwritten when a new sync operation is begun.
269 */
270 char *last_sync_action;
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 atomic64_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 * ERROR: sync-action interrupted because io-error
311 *
312 * If neither SYNC or RESHAPE are set, then it is a recovery.
313 */
314#define MD_RECOVERY_RUNNING 0
315#define MD_RECOVERY_SYNC 1
316#define MD_RECOVERY_RECOVER 2
317#define MD_RECOVERY_INTR 3
318#define MD_RECOVERY_DONE 4
319#define MD_RECOVERY_NEEDED 5
320#define MD_RECOVERY_REQUESTED 6
321#define MD_RECOVERY_CHECK 7
322#define MD_RECOVERY_RESHAPE 8
323#define MD_RECOVERY_FROZEN 9
324#define MD_RECOVERY_ERROR 10
325
326 unsigned long recovery;
327 /* If a RAID personality determines that recovery (of a particular
328 * device) will fail due to a read error on the source device, it
329 * takes a copy of this number and does not attempt recovery again
330 * until this number changes.
331 */
332 int recovery_disabled;
333
334 int in_sync; /* know to not need resync */
335 /* 'open_mutex' avoids races between 'md_open' and 'do_md_stop', so
336 * that we are never stopping an array while it is open.
337 * 'reconfig_mutex' protects all other reconfiguration.
338 * These locks are separate due to conflicting interactions
339 * with bdev->bd_mutex.
340 * Lock ordering is:
341 * reconfig_mutex -> bd_mutex : e.g. do_md_run -> revalidate_disk
342 * bd_mutex -> open_mutex: e.g. __blkdev_get -> md_open
343 */
344 struct mutex open_mutex;
345 struct mutex reconfig_mutex;
346 atomic_t active; /* general refcount */
347 atomic_t openers; /* number of active opens */
348
349 int changed; /* True if we might need to
350 * reread partition info */
351 int degraded; /* whether md should consider
352 * adding a spare
353 */
354
355 atomic_t recovery_active; /* blocks scheduled, but not written */
356 wait_queue_head_t recovery_wait;
357 sector_t recovery_cp;
358 sector_t resync_min; /* user requested sync
359 * starts here */
360 sector_t resync_max; /* resync should pause
361 * when it gets here */
362
363 struct kernfs_node *sysfs_state; /* handle for 'array_state'
364 * file in sysfs.
365 */
366 struct kernfs_node *sysfs_action; /* handle for 'sync_action' */
367
368 struct work_struct del_work; /* used for delayed sysfs removal */
369
370 /* "lock" protects:
371 * flush_bio transition from NULL to !NULL
372 * rdev superblocks, events
373 * clearing MD_CHANGE_*
374 * in_sync - and related safemode and MD_CHANGE changes
375 * pers (also protected by reconfig_mutex and pending IO).
376 * clearing ->bitmap
377 * clearing ->bitmap_info.file
378 * changing ->resync_{min,max}
379 * setting MD_RECOVERY_RUNNING (which interacts with resync_{min,max})
380 */
381 spinlock_t lock;
382 wait_queue_head_t sb_wait; /* for waiting on superblock updates */
383 atomic_t pending_writes; /* number of active superblock writes */
384
385 unsigned int safemode; /* if set, update "clean" superblock
386 * when no writes pending.
387 */
388 unsigned int safemode_delay;
389 struct timer_list safemode_timer;
390 atomic_t writes_pending;
391 struct request_queue *queue; /* for plugging ... */
392
393 struct bitmap *bitmap; /* the bitmap for the device */
394 struct {
395 struct file *file; /* the bitmap file */
396 loff_t offset; /* offset from superblock of
397 * start of bitmap. May be
398 * negative, but not '0'
399 * For external metadata, offset
400 * from start of device.
401 */
402 unsigned long space; /* space available at this offset */
403 loff_t default_offset; /* this is the offset to use when
404 * hot-adding a bitmap. It should
405 * eventually be settable by sysfs.
406 */
407 unsigned long default_space; /* space available at
408 * default offset */
409 struct mutex mutex;
410 unsigned long chunksize;
411 unsigned long daemon_sleep; /* how many jiffies between updates? */
412 unsigned long max_write_behind; /* write-behind mode */
413 int external;
414 int nodes; /* Maximum number of nodes in the cluster */
415 char cluster_name[64]; /* Name of the cluster */
416 } bitmap_info;
417
418 atomic_t max_corr_read_errors; /* max read retries */
419 struct list_head all_mddevs;
420
421 struct attribute_group *to_remove;
422
423 struct bio_set *bio_set;
424
425 /* Generic flush handling.
426 * The last to finish preflush schedules a worker to submit
427 * the rest of the request (without the REQ_FLUSH flag).
428 */
429 struct bio *flush_bio;
430 atomic_t flush_pending;
431 struct work_struct flush_work;
432 struct work_struct event_work; /* used by dm to report failure event */
433 void (*sync_super)(struct mddev *mddev, struct md_rdev *rdev);
434 struct md_cluster_info *cluster_info;
435 unsigned int good_device_nr; /* good device num within cluster raid */
436};
437
438static inline int __must_check mddev_lock(struct mddev *mddev)
439{
440 return mutex_lock_interruptible(&mddev->reconfig_mutex);
441}
442
443/* Sometimes we need to take the lock in a situation where
444 * failure due to interrupts is not acceptable.
445 */
446static inline void mddev_lock_nointr(struct mddev *mddev)
447{
448 mutex_lock(&mddev->reconfig_mutex);
449}
450
451static inline int mddev_is_locked(struct mddev *mddev)
452{
453 return mutex_is_locked(&mddev->reconfig_mutex);
454}
455
456static inline int mddev_trylock(struct mddev *mddev)
457{
458 return mutex_trylock(&mddev->reconfig_mutex);
459}
460extern void mddev_unlock(struct mddev *mddev);
461
462static inline void md_sync_acct(struct block_device *bdev, unsigned long nr_sectors)
463{
464 atomic_add(nr_sectors, &bdev->bd_contains->bd_disk->sync_io);
465}
466
467struct md_personality
468{
469 char *name;
470 int level;
471 struct list_head list;
472 struct module *owner;
473 void (*make_request)(struct mddev *mddev, struct bio *bio);
474 int (*run)(struct mddev *mddev);
475 void (*free)(struct mddev *mddev, void *priv);
476 void (*status)(struct seq_file *seq, struct mddev *mddev);
477 /* error_handler must set ->faulty and clear ->in_sync
478 * if appropriate, and should abort recovery if needed
479 */
480 void (*error_handler)(struct mddev *mddev, struct md_rdev *rdev);
481 int (*hot_add_disk) (struct mddev *mddev, struct md_rdev *rdev);
482 int (*hot_remove_disk) (struct mddev *mddev, struct md_rdev *rdev);
483 int (*spare_active) (struct mddev *mddev);
484 sector_t (*sync_request)(struct mddev *mddev, sector_t sector_nr, int *skipped);
485 int (*resize) (struct mddev *mddev, sector_t sectors);
486 sector_t (*size) (struct mddev *mddev, sector_t sectors, int raid_disks);
487 int (*check_reshape) (struct mddev *mddev);
488 int (*start_reshape) (struct mddev *mddev);
489 void (*finish_reshape) (struct mddev *mddev);
490 /* quiesce moves between quiescence states
491 * 0 - fully active
492 * 1 - no new requests allowed
493 * others - reserved
494 */
495 void (*quiesce) (struct mddev *mddev, int state);
496 /* takeover is used to transition an array from one
497 * personality to another. The new personality must be able
498 * to handle the data in the current layout.
499 * e.g. 2drive raid1 -> 2drive raid5
500 * ndrive raid5 -> degraded n+1drive raid6 with special layout
501 * If the takeover succeeds, a new 'private' structure is returned.
502 * This needs to be installed and then ->run used to activate the
503 * array.
504 */
505 void *(*takeover) (struct mddev *mddev);
506 /* congested implements bdi.congested_fn().
507 * Will not be called while array is 'suspended' */
508 int (*congested)(struct mddev *mddev, int bits);
509};
510
511struct md_sysfs_entry {
512 struct attribute attr;
513 ssize_t (*show)(struct mddev *, char *);
514 ssize_t (*store)(struct mddev *, const char *, size_t);
515};
516extern struct attribute_group md_bitmap_group;
517
518static inline struct kernfs_node *sysfs_get_dirent_safe(struct kernfs_node *sd, char *name)
519{
520 if (sd)
521 return sysfs_get_dirent(sd, name);
522 return sd;
523}
524static inline void sysfs_notify_dirent_safe(struct kernfs_node *sd)
525{
526 if (sd)
527 sysfs_notify_dirent(sd);
528}
529
530static inline char * mdname (struct mddev * mddev)
531{
532 return mddev->gendisk ? mddev->gendisk->disk_name : "mdX";
533}
534
535static inline int sysfs_link_rdev(struct mddev *mddev, struct md_rdev *rdev)
536{
537 char nm[20];
538 if (!test_bit(Replacement, &rdev->flags) &&
539 !test_bit(Journal, &rdev->flags) &&
540 mddev->kobj.sd) {
541 sprintf(nm, "rd%d", rdev->raid_disk);
542 return sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
543 } else
544 return 0;
545}
546
547static inline void sysfs_unlink_rdev(struct mddev *mddev, struct md_rdev *rdev)
548{
549 char nm[20];
550 if (!test_bit(Replacement, &rdev->flags) &&
551 !test_bit(Journal, &rdev->flags) &&
552 mddev->kobj.sd) {
553 sprintf(nm, "rd%d", rdev->raid_disk);
554 sysfs_remove_link(&mddev->kobj, nm);
555 }
556}
557
558/*
559 * iterates through some rdev ringlist. It's safe to remove the
560 * current 'rdev'. Dont touch 'tmp' though.
561 */
562#define rdev_for_each_list(rdev, tmp, head) \
563 list_for_each_entry_safe(rdev, tmp, head, same_set)
564
565/*
566 * iterates through the 'same array disks' ringlist
567 */
568#define rdev_for_each(rdev, mddev) \
569 list_for_each_entry(rdev, &((mddev)->disks), same_set)
570
571#define rdev_for_each_safe(rdev, tmp, mddev) \
572 list_for_each_entry_safe(rdev, tmp, &((mddev)->disks), same_set)
573
574#define rdev_for_each_rcu(rdev, mddev) \
575 list_for_each_entry_rcu(rdev, &((mddev)->disks), same_set)
576
577struct md_thread {
578 void (*run) (struct md_thread *thread);
579 struct mddev *mddev;
580 wait_queue_head_t wqueue;
581 unsigned long flags;
582 struct task_struct *tsk;
583 unsigned long timeout;
584 void *private;
585};
586
587#define THREAD_WAKEUP 0
588
589static inline void safe_put_page(struct page *p)
590{
591 if (p) put_page(p);
592}
593
594extern int register_md_personality(struct md_personality *p);
595extern int unregister_md_personality(struct md_personality *p);
596extern int register_md_cluster_operations(struct md_cluster_operations *ops,
597 struct module *module);
598extern int unregister_md_cluster_operations(void);
599extern int md_setup_cluster(struct mddev *mddev, int nodes);
600extern void md_cluster_stop(struct mddev *mddev);
601extern struct md_thread *md_register_thread(
602 void (*run)(struct md_thread *thread),
603 struct mddev *mddev,
604 const char *name);
605extern void md_unregister_thread(struct md_thread **threadp);
606extern void md_wakeup_thread(struct md_thread *thread);
607extern void md_check_recovery(struct mddev *mddev);
608extern void md_reap_sync_thread(struct mddev *mddev);
609extern void md_write_start(struct mddev *mddev, struct bio *bi);
610extern void md_write_end(struct mddev *mddev);
611extern void md_done_sync(struct mddev *mddev, int blocks, int ok);
612extern void md_error(struct mddev *mddev, struct md_rdev *rdev);
613extern void md_finish_reshape(struct mddev *mddev);
614
615extern int mddev_congested(struct mddev *mddev, int bits);
616extern void md_flush_request(struct mddev *mddev, struct bio *bio);
617extern void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
618 sector_t sector, int size, struct page *page);
619extern void md_super_wait(struct mddev *mddev);
620extern int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
621 struct page *page, int rw, bool metadata_op);
622extern void md_do_sync(struct md_thread *thread);
623extern void md_new_event(struct mddev *mddev);
624extern int md_allow_write(struct mddev *mddev);
625extern void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev);
626extern void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors);
627extern int md_check_no_bitmap(struct mddev *mddev);
628extern int md_integrity_register(struct mddev *mddev);
629extern int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev);
630extern int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale);
631
632extern void mddev_init(struct mddev *mddev);
633extern int md_run(struct mddev *mddev);
634extern void md_stop(struct mddev *mddev);
635extern void md_stop_writes(struct mddev *mddev);
636extern int md_rdev_init(struct md_rdev *rdev);
637extern void md_rdev_clear(struct md_rdev *rdev);
638
639extern void mddev_suspend(struct mddev *mddev);
640extern void mddev_resume(struct mddev *mddev);
641extern struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
642 struct mddev *mddev);
643extern struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
644 struct mddev *mddev);
645
646extern void md_unplug(struct blk_plug_cb *cb, bool from_schedule);
647extern void md_reload_sb(struct mddev *mddev, int raid_disk);
648extern void md_update_sb(struct mddev *mddev, int force);
649extern void md_kick_rdev_from_array(struct md_rdev * rdev);
650struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr);
651static inline int mddev_check_plugged(struct mddev *mddev)
652{
653 return !!blk_check_plugged(md_unplug, mddev,
654 sizeof(struct blk_plug_cb));
655}
656
657static inline void rdev_dec_pending(struct md_rdev *rdev, struct mddev *mddev)
658{
659 int faulty = test_bit(Faulty, &rdev->flags);
660 if (atomic_dec_and_test(&rdev->nr_pending) && faulty) {
661 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
662 md_wakeup_thread(mddev->thread);
663 }
664}
665
666extern struct md_cluster_operations *md_cluster_ops;
667static inline int mddev_is_clustered(struct mddev *mddev)
668{
669 return mddev->cluster_info && mddev->bitmap_info.nodes > 1;
670}
671#endif /* _MD_MD_H */