Loading...
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
29typedef struct mddev_s mddev_t;
30typedef struct mdk_rdev_s mdk_rdev_t;
31
32/* Bad block numbers are stored sorted in a single page.
33 * 64bits is used for each block or extent.
34 * 54 bits are sector number, 9 bits are extent size,
35 * 1 bit is an 'acknowledged' flag.
36 */
37#define MD_MAX_BADBLOCKS (PAGE_SIZE/8)
38
39/*
40 * MD's 'extended' device
41 */
42struct 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 /*
51 * If meta_bdev is non-NULL, it means that a separate device is
52 * being used to store the metadata (superblock/bitmap) which
53 * would otherwise be contained on the same device as the data (bdev).
54 */
55 struct block_device *meta_bdev;
56 struct block_device *bdev; /* block device handle */
57
58 struct page *sb_page, *bb_page;
59 int sb_loaded;
60 __u64 sb_events;
61 sector_t data_offset; /* start of data in array */
62 sector_t sb_start; /* offset of the super block (in 512byte sectors) */
63 int sb_size; /* bytes in the superblock */
64 int preferred_minor; /* autorun support */
65
66 struct kobject kobj;
67
68 /* A device can be in one of three states based on two flags:
69 * Not working: faulty==1 in_sync==0
70 * Fully working: faulty==0 in_sync==1
71 * Working, but not
72 * in sync with array
73 * faulty==0 in_sync==0
74 *
75 * It can never have faulty==1, in_sync==1
76 * This reduces the burden of testing multiple flags in many cases
77 */
78
79 unsigned long flags;
80#define Faulty 1 /* device is known to have a fault */
81#define In_sync 2 /* device is in_sync with rest of array */
82#define WriteMostly 4 /* Avoid reading if at all possible */
83#define AutoDetected 7 /* added by auto-detect */
84#define Blocked 8 /* An error occurred but has not yet
85 * been acknowledged by the metadata
86 * handler, so don't allow writes
87 * until it is cleared */
88#define WriteErrorSeen 9 /* A write error has been seen on this
89 * device
90 */
91#define FaultRecorded 10 /* Intermediate state for clearing
92 * Blocked. The Fault is/will-be
93 * recorded in the metadata, but that
94 * metadata hasn't been stored safely
95 * on disk yet.
96 */
97#define BlockedBadBlocks 11 /* A writer is blocked because they
98 * found an unacknowledged bad-block.
99 * This can safely be cleared at any
100 * time, and the writer will re-check.
101 * It may be set at any time, and at
102 * worst the writer will timeout and
103 * re-check. So setting it as
104 * accurately as possible is good, but
105 * not absolutely critical.
106 */
107 wait_queue_head_t blocked_wait;
108
109 int desc_nr; /* descriptor index in the superblock */
110 int raid_disk; /* role of device in array */
111 int new_raid_disk; /* role that the device will have in
112 * the array after a level-change completes.
113 */
114 int saved_raid_disk; /* role that device used to have in the
115 * array and could again if we did a partial
116 * resync from the bitmap
117 */
118 sector_t recovery_offset;/* If this device has been partially
119 * recovered, this is where we were
120 * up to.
121 */
122
123 atomic_t nr_pending; /* number of pending requests.
124 * only maintained for arrays that
125 * support hot removal
126 */
127 atomic_t read_errors; /* number of consecutive read errors that
128 * we have tried to ignore.
129 */
130 struct timespec last_read_error; /* monotonic time since our
131 * last read error
132 */
133 atomic_t corrected_errors; /* number of corrected read errors,
134 * for reporting to userspace and storing
135 * in superblock.
136 */
137 struct work_struct del_work; /* used for delayed sysfs removal */
138
139 struct sysfs_dirent *sysfs_state; /* handle for 'state'
140 * sysfs entry */
141
142 struct badblocks {
143 int count; /* count of bad blocks */
144 int unacked_exist; /* there probably are unacknowledged
145 * bad blocks. This is only cleared
146 * when a read discovers none
147 */
148 int shift; /* shift from sectors to block size
149 * a -ve shift means badblocks are
150 * disabled.*/
151 u64 *page; /* badblock list */
152 int changed;
153 seqlock_t lock;
154
155 sector_t sector;
156 sector_t size; /* in sectors */
157 } badblocks;
158};
159
160#define BB_LEN_MASK (0x00000000000001FFULL)
161#define BB_OFFSET_MASK (0x7FFFFFFFFFFFFE00ULL)
162#define BB_ACK_MASK (0x8000000000000000ULL)
163#define BB_MAX_LEN 512
164#define BB_OFFSET(x) (((x) & BB_OFFSET_MASK) >> 9)
165#define BB_LEN(x) (((x) & BB_LEN_MASK) + 1)
166#define BB_ACK(x) (!!((x) & BB_ACK_MASK))
167#define BB_MAKE(a, l, ack) (((a)<<9) | ((l)-1) | ((u64)(!!(ack)) << 63))
168
169extern int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
170 sector_t *first_bad, int *bad_sectors);
171static inline int is_badblock(mdk_rdev_t *rdev, sector_t s, int sectors,
172 sector_t *first_bad, int *bad_sectors)
173{
174 if (unlikely(rdev->badblocks.count)) {
175 int rv = md_is_badblock(&rdev->badblocks, rdev->data_offset + s,
176 sectors,
177 first_bad, bad_sectors);
178 if (rv)
179 *first_bad -= rdev->data_offset;
180 return rv;
181 }
182 return 0;
183}
184extern int rdev_set_badblocks(mdk_rdev_t *rdev, sector_t s, int sectors,
185 int acknowledged);
186extern int rdev_clear_badblocks(mdk_rdev_t *rdev, sector_t s, int sectors);
187extern void md_ack_all_badblocks(struct badblocks *bb);
188
189struct mddev_s
190{
191 void *private;
192 struct mdk_personality *pers;
193 dev_t unit;
194 int md_minor;
195 struct list_head disks;
196 unsigned long flags;
197#define MD_CHANGE_DEVS 0 /* Some device status has changed */
198#define MD_CHANGE_CLEAN 1 /* transition to or from 'clean' */
199#define MD_CHANGE_PENDING 2 /* switch from 'clean' to 'active' in progress */
200#define MD_ARRAY_FIRST_USE 3 /* First use of array, needs initialization */
201
202 int suspended;
203 atomic_t active_io;
204 int ro;
205 int sysfs_active; /* set when sysfs deletes
206 * are happening, so run/
207 * takeover/stop are not safe
208 */
209 int ready; /* See when safe to pass
210 * IO requests down */
211 struct gendisk *gendisk;
212
213 struct kobject kobj;
214 int hold_active;
215#define UNTIL_IOCTL 1
216#define UNTIL_STOP 2
217
218 /* Superblock information */
219 int major_version,
220 minor_version,
221 patch_version;
222 int persistent;
223 int external; /* metadata is
224 * managed externally */
225 char metadata_type[17]; /* externally set*/
226 int chunk_sectors;
227 time_t ctime, utime;
228 int level, layout;
229 char clevel[16];
230 int raid_disks;
231 int max_disks;
232 sector_t dev_sectors; /* used size of
233 * component devices */
234 sector_t array_sectors; /* exported array size */
235 int external_size; /* size managed
236 * externally */
237 __u64 events;
238 /* If the last 'event' was simply a clean->dirty transition, and
239 * we didn't write it to the spares, then it is safe and simple
240 * to just decrement the event count on a dirty->clean transition.
241 * So we record that possibility here.
242 */
243 int can_decrease_events;
244
245 char uuid[16];
246
247 /* If the array is being reshaped, we need to record the
248 * new shape and an indication of where we are up to.
249 * This is written to the superblock.
250 * If reshape_position is MaxSector, then no reshape is happening (yet).
251 */
252 sector_t reshape_position;
253 int delta_disks, new_level, new_layout;
254 int new_chunk_sectors;
255
256 atomic_t plug_cnt; /* If device is expecting
257 * more bios soon.
258 */
259 struct mdk_thread_s *thread; /* management thread */
260 struct mdk_thread_s *sync_thread; /* doing resync or reconstruct */
261 sector_t curr_resync; /* last block scheduled */
262 /* As resync requests can complete out of order, we cannot easily track
263 * how much resync has been completed. So we occasionally pause until
264 * everything completes, then set curr_resync_completed to curr_resync.
265 * As such it may be well behind the real resync mark, but it is a value
266 * we are certain of.
267 */
268 sector_t curr_resync_completed;
269 unsigned long resync_mark; /* a recent timestamp */
270 sector_t resync_mark_cnt;/* blocks written at resync_mark */
271 sector_t curr_mark_cnt; /* blocks scheduled now */
272
273 sector_t resync_max_sectors; /* may be set by personality */
274
275 sector_t resync_mismatches; /* count of sectors where
276 * parity/replica mismatch found
277 */
278
279 /* allow user-space to request suspension of IO to regions of the array */
280 sector_t suspend_lo;
281 sector_t suspend_hi;
282 /* if zero, use the system-wide default */
283 int sync_speed_min;
284 int sync_speed_max;
285
286 /* resync even though the same disks are shared among md-devices */
287 int parallel_resync;
288
289 int ok_start_degraded;
290 /* recovery/resync flags
291 * NEEDED: we might need to start a resync/recover
292 * RUNNING: a thread is running, or about to be started
293 * SYNC: actually doing a resync, not a recovery
294 * RECOVER: doing recovery, or need to try it.
295 * INTR: resync needs to be aborted for some reason
296 * DONE: thread is done and is waiting to be reaped
297 * REQUEST: user-space has requested a sync (used with SYNC)
298 * CHECK: user-space request for check-only, no repair
299 * RESHAPE: A reshape is happening
300 *
301 * If neither SYNC or RESHAPE are set, then it is a recovery.
302 */
303#define MD_RECOVERY_RUNNING 0
304#define MD_RECOVERY_SYNC 1
305#define MD_RECOVERY_RECOVER 2
306#define MD_RECOVERY_INTR 3
307#define MD_RECOVERY_DONE 4
308#define MD_RECOVERY_NEEDED 5
309#define MD_RECOVERY_REQUESTED 6
310#define MD_RECOVERY_CHECK 7
311#define MD_RECOVERY_RESHAPE 8
312#define MD_RECOVERY_FROZEN 9
313
314 unsigned long recovery;
315 /* If a RAID personality determines that recovery (of a particular
316 * device) will fail due to a read error on the source device, it
317 * takes a copy of this number and does not attempt recovery again
318 * until this number changes.
319 */
320 int recovery_disabled;
321
322 int in_sync; /* know to not need resync */
323 /* 'open_mutex' avoids races between 'md_open' and 'do_md_stop', so
324 * that we are never stopping an array while it is open.
325 * 'reconfig_mutex' protects all other reconfiguration.
326 * These locks are separate due to conflicting interactions
327 * with bdev->bd_mutex.
328 * Lock ordering is:
329 * reconfig_mutex -> bd_mutex : e.g. do_md_run -> revalidate_disk
330 * bd_mutex -> open_mutex: e.g. __blkdev_get -> md_open
331 */
332 struct mutex open_mutex;
333 struct mutex reconfig_mutex;
334 atomic_t active; /* general refcount */
335 atomic_t openers; /* number of active opens */
336
337 int changed; /* True if we might need to
338 * reread partition info */
339 int degraded; /* whether md should consider
340 * adding a spare
341 */
342
343 atomic_t recovery_active; /* blocks scheduled, but not written */
344 wait_queue_head_t recovery_wait;
345 sector_t recovery_cp;
346 sector_t resync_min; /* user requested sync
347 * starts here */
348 sector_t resync_max; /* resync should pause
349 * when it gets here */
350
351 struct sysfs_dirent *sysfs_state; /* handle for 'array_state'
352 * file in sysfs.
353 */
354 struct sysfs_dirent *sysfs_action; /* handle for 'sync_action' */
355
356 struct work_struct del_work; /* used for delayed sysfs removal */
357
358 spinlock_t write_lock;
359 wait_queue_head_t sb_wait; /* for waiting on superblock updates */
360 atomic_t pending_writes; /* number of active superblock writes */
361
362 unsigned int safemode; /* if set, update "clean" superblock
363 * when no writes pending.
364 */
365 unsigned int safemode_delay;
366 struct timer_list safemode_timer;
367 atomic_t writes_pending;
368 struct request_queue *queue; /* for plugging ... */
369
370 struct bitmap *bitmap; /* the bitmap for the device */
371 struct {
372 struct file *file; /* the bitmap file */
373 loff_t offset; /* offset from superblock of
374 * start of bitmap. May be
375 * negative, but not '0'
376 * For external metadata, offset
377 * from start of device.
378 */
379 loff_t default_offset; /* this is the offset to use when
380 * hot-adding a bitmap. It should
381 * eventually be settable by sysfs.
382 */
383 struct mutex mutex;
384 unsigned long chunksize;
385 unsigned long daemon_sleep; /* how many jiffies between updates? */
386 unsigned long max_write_behind; /* write-behind mode */
387 int external;
388 } bitmap_info;
389
390 atomic_t max_corr_read_errors; /* max read retries */
391 struct list_head all_mddevs;
392
393 struct attribute_group *to_remove;
394
395 struct bio_set *bio_set;
396
397 /* Generic flush handling.
398 * The last to finish preflush schedules a worker to submit
399 * the rest of the request (without the REQ_FLUSH flag).
400 */
401 struct bio *flush_bio;
402 atomic_t flush_pending;
403 struct work_struct flush_work;
404 struct work_struct event_work; /* used by dm to report failure event */
405 void (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
406};
407
408
409static inline void rdev_dec_pending(mdk_rdev_t *rdev, mddev_t *mddev)
410{
411 int faulty = test_bit(Faulty, &rdev->flags);
412 if (atomic_dec_and_test(&rdev->nr_pending) && faulty)
413 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
414}
415
416static inline void md_sync_acct(struct block_device *bdev, unsigned long nr_sectors)
417{
418 atomic_add(nr_sectors, &bdev->bd_contains->bd_disk->sync_io);
419}
420
421struct mdk_personality
422{
423 char *name;
424 int level;
425 struct list_head list;
426 struct module *owner;
427 int (*make_request)(mddev_t *mddev, struct bio *bio);
428 int (*run)(mddev_t *mddev);
429 int (*stop)(mddev_t *mddev);
430 void (*status)(struct seq_file *seq, mddev_t *mddev);
431 /* error_handler must set ->faulty and clear ->in_sync
432 * if appropriate, and should abort recovery if needed
433 */
434 void (*error_handler)(mddev_t *mddev, mdk_rdev_t *rdev);
435 int (*hot_add_disk) (mddev_t *mddev, mdk_rdev_t *rdev);
436 int (*hot_remove_disk) (mddev_t *mddev, int number);
437 int (*spare_active) (mddev_t *mddev);
438 sector_t (*sync_request)(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster);
439 int (*resize) (mddev_t *mddev, sector_t sectors);
440 sector_t (*size) (mddev_t *mddev, sector_t sectors, int raid_disks);
441 int (*check_reshape) (mddev_t *mddev);
442 int (*start_reshape) (mddev_t *mddev);
443 void (*finish_reshape) (mddev_t *mddev);
444 /* quiesce moves between quiescence states
445 * 0 - fully active
446 * 1 - no new requests allowed
447 * others - reserved
448 */
449 void (*quiesce) (mddev_t *mddev, int state);
450 /* takeover is used to transition an array from one
451 * personality to another. The new personality must be able
452 * to handle the data in the current layout.
453 * e.g. 2drive raid1 -> 2drive raid5
454 * ndrive raid5 -> degraded n+1drive raid6 with special layout
455 * If the takeover succeeds, a new 'private' structure is returned.
456 * This needs to be installed and then ->run used to activate the
457 * array.
458 */
459 void *(*takeover) (mddev_t *mddev);
460};
461
462
463struct md_sysfs_entry {
464 struct attribute attr;
465 ssize_t (*show)(mddev_t *, char *);
466 ssize_t (*store)(mddev_t *, const char *, size_t);
467};
468extern struct attribute_group md_bitmap_group;
469
470static inline struct sysfs_dirent *sysfs_get_dirent_safe(struct sysfs_dirent *sd, char *name)
471{
472 if (sd)
473 return sysfs_get_dirent(sd, NULL, name);
474 return sd;
475}
476static inline void sysfs_notify_dirent_safe(struct sysfs_dirent *sd)
477{
478 if (sd)
479 sysfs_notify_dirent(sd);
480}
481
482static inline char * mdname (mddev_t * mddev)
483{
484 return mddev->gendisk ? mddev->gendisk->disk_name : "mdX";
485}
486
487static inline int sysfs_link_rdev(mddev_t *mddev, mdk_rdev_t *rdev)
488{
489 char nm[20];
490 sprintf(nm, "rd%d", rdev->raid_disk);
491 return sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
492}
493
494static inline void sysfs_unlink_rdev(mddev_t *mddev, mdk_rdev_t *rdev)
495{
496 char nm[20];
497 sprintf(nm, "rd%d", rdev->raid_disk);
498 sysfs_remove_link(&mddev->kobj, nm);
499}
500
501/*
502 * iterates through some rdev ringlist. It's safe to remove the
503 * current 'rdev'. Dont touch 'tmp' though.
504 */
505#define rdev_for_each_list(rdev, tmp, head) \
506 list_for_each_entry_safe(rdev, tmp, head, same_set)
507
508/*
509 * iterates through the 'same array disks' ringlist
510 */
511#define rdev_for_each(rdev, tmp, mddev) \
512 list_for_each_entry_safe(rdev, tmp, &((mddev)->disks), same_set)
513
514#define rdev_for_each_rcu(rdev, mddev) \
515 list_for_each_entry_rcu(rdev, &((mddev)->disks), same_set)
516
517typedef struct mdk_thread_s {
518 void (*run) (mddev_t *mddev);
519 mddev_t *mddev;
520 wait_queue_head_t wqueue;
521 unsigned long flags;
522 struct task_struct *tsk;
523 unsigned long timeout;
524} mdk_thread_t;
525
526#define THREAD_WAKEUP 0
527
528#define __wait_event_lock_irq(wq, condition, lock, cmd) \
529do { \
530 wait_queue_t __wait; \
531 init_waitqueue_entry(&__wait, current); \
532 \
533 add_wait_queue(&wq, &__wait); \
534 for (;;) { \
535 set_current_state(TASK_UNINTERRUPTIBLE); \
536 if (condition) \
537 break; \
538 spin_unlock_irq(&lock); \
539 cmd; \
540 schedule(); \
541 spin_lock_irq(&lock); \
542 } \
543 current->state = TASK_RUNNING; \
544 remove_wait_queue(&wq, &__wait); \
545} while (0)
546
547#define wait_event_lock_irq(wq, condition, lock, cmd) \
548do { \
549 if (condition) \
550 break; \
551 __wait_event_lock_irq(wq, condition, lock, cmd); \
552} while (0)
553
554static inline void safe_put_page(struct page *p)
555{
556 if (p) put_page(p);
557}
558
559extern int register_md_personality(struct mdk_personality *p);
560extern int unregister_md_personality(struct mdk_personality *p);
561extern mdk_thread_t * md_register_thread(void (*run) (mddev_t *mddev),
562 mddev_t *mddev, const char *name);
563extern void md_unregister_thread(mdk_thread_t **threadp);
564extern void md_wakeup_thread(mdk_thread_t *thread);
565extern void md_check_recovery(mddev_t *mddev);
566extern void md_write_start(mddev_t *mddev, struct bio *bi);
567extern void md_write_end(mddev_t *mddev);
568extern void md_done_sync(mddev_t *mddev, int blocks, int ok);
569extern void md_error(mddev_t *mddev, mdk_rdev_t *rdev);
570
571extern int mddev_congested(mddev_t *mddev, int bits);
572extern void md_flush_request(mddev_t *mddev, struct bio *bio);
573extern void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
574 sector_t sector, int size, struct page *page);
575extern void md_super_wait(mddev_t *mddev);
576extern int sync_page_io(mdk_rdev_t *rdev, sector_t sector, int size,
577 struct page *page, int rw, bool metadata_op);
578extern void md_do_sync(mddev_t *mddev);
579extern void md_new_event(mddev_t *mddev);
580extern int md_allow_write(mddev_t *mddev);
581extern void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev);
582extern void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors);
583extern int md_check_no_bitmap(mddev_t *mddev);
584extern int md_integrity_register(mddev_t *mddev);
585extern void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev);
586extern int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale);
587extern void restore_bitmap_write_access(struct file *file);
588
589extern void mddev_init(mddev_t *mddev);
590extern int md_run(mddev_t *mddev);
591extern void md_stop(mddev_t *mddev);
592extern void md_stop_writes(mddev_t *mddev);
593extern int md_rdev_init(mdk_rdev_t *rdev);
594
595extern void mddev_suspend(mddev_t *mddev);
596extern void mddev_resume(mddev_t *mddev);
597extern struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
598 mddev_t *mddev);
599extern struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
600 mddev_t *mddev);
601extern int mddev_check_plugged(mddev_t *mddev);
602extern void md_trim_bio(struct bio *bio, int offset, int size);
603#endif /* _MD_MD_H */
1/* SPDX-License-Identifier: GPL-2.0-or-later */
2/*
3 md.h : kernel internal structure of the Linux MD driver
4 Copyright (C) 1996-98 Ingo Molnar, Gadi Oxman
5
6*/
7
8#ifndef _MD_MD_H
9#define _MD_MD_H
10
11#include <linux/blkdev.h>
12#include <linux/backing-dev.h>
13#include <linux/badblocks.h>
14#include <linux/kobject.h>
15#include <linux/list.h>
16#include <linux/mm.h>
17#include <linux/mutex.h>
18#include <linux/timer.h>
19#include <linux/wait.h>
20#include <linux/workqueue.h>
21#include <trace/events/block.h>
22#include "md-cluster.h"
23
24#define MaxSector (~(sector_t)0)
25
26/*
27 * These flags should really be called "NO_RETRY" rather than
28 * "FAILFAST" because they don't make any promise about time lapse,
29 * only about the number of retries, which will be zero.
30 * REQ_FAILFAST_DRIVER is not included because
31 * Commit: 4a27446f3e39 ("[SCSI] modify scsi to handle new fail fast flags.")
32 * seems to suggest that the errors it avoids retrying should usually
33 * be retried.
34 */
35#define MD_FAILFAST (REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT)
36
37/*
38 * The struct embedded in rdev is used to serialize IO.
39 */
40struct serial_in_rdev {
41 struct rb_root_cached serial_rb;
42 spinlock_t serial_lock;
43 wait_queue_head_t serial_io_wait;
44};
45
46/*
47 * MD's 'extended' device
48 */
49struct md_rdev {
50 struct list_head same_set; /* RAID devices within the same set */
51
52 sector_t sectors; /* Device size (in 512bytes sectors) */
53 struct mddev *mddev; /* RAID array if running */
54 int last_events; /* IO event timestamp */
55
56 /*
57 * If meta_bdev is non-NULL, it means that a separate device is
58 * being used to store the metadata (superblock/bitmap) which
59 * would otherwise be contained on the same device as the data (bdev).
60 */
61 struct block_device *meta_bdev;
62 struct block_device *bdev; /* block device handle */
63 struct file *bdev_file; /* Handle from open for bdev */
64
65 struct page *sb_page, *bb_page;
66 int sb_loaded;
67 __u64 sb_events;
68 sector_t data_offset; /* start of data in array */
69 sector_t new_data_offset;/* only relevant while reshaping */
70 sector_t sb_start; /* offset of the super block (in 512byte sectors) */
71 int sb_size; /* bytes in the superblock */
72 int preferred_minor; /* autorun support */
73
74 struct kobject kobj;
75
76 /* A device can be in one of three states based on two flags:
77 * Not working: faulty==1 in_sync==0
78 * Fully working: faulty==0 in_sync==1
79 * Working, but not
80 * in sync with array
81 * faulty==0 in_sync==0
82 *
83 * It can never have faulty==1, in_sync==1
84 * This reduces the burden of testing multiple flags in many cases
85 */
86
87 unsigned long flags; /* bit set of 'enum flag_bits' bits. */
88 wait_queue_head_t blocked_wait;
89
90 int desc_nr; /* descriptor index in the superblock */
91 int raid_disk; /* role of device in array */
92 int new_raid_disk; /* role that the device will have in
93 * the array after a level-change completes.
94 */
95 int saved_raid_disk; /* role that device used to have in the
96 * array and could again if we did a partial
97 * resync from the bitmap
98 */
99 union {
100 sector_t recovery_offset;/* If this device has been partially
101 * recovered, this is where we were
102 * up to.
103 */
104 sector_t journal_tail; /* If this device is a journal device,
105 * this is the journal tail (journal
106 * recovery start point)
107 */
108 };
109
110 atomic_t nr_pending; /* number of pending requests.
111 * only maintained for arrays that
112 * support hot removal
113 */
114 atomic_t read_errors; /* number of consecutive read errors that
115 * we have tried to ignore.
116 */
117 time64_t last_read_error; /* monotonic time since our
118 * last read error
119 */
120 atomic_t corrected_errors; /* number of corrected read errors,
121 * for reporting to userspace and storing
122 * in superblock.
123 */
124
125 struct serial_in_rdev *serial; /* used for raid1 io serialization */
126
127 struct kernfs_node *sysfs_state; /* handle for 'state'
128 * sysfs entry */
129 /* handle for 'unacknowledged_bad_blocks' sysfs dentry */
130 struct kernfs_node *sysfs_unack_badblocks;
131 /* handle for 'bad_blocks' sysfs dentry */
132 struct kernfs_node *sysfs_badblocks;
133 struct badblocks badblocks;
134
135 struct {
136 short offset; /* Offset from superblock to start of PPL.
137 * Not used by external metadata. */
138 unsigned int size; /* Size in sectors of the PPL space */
139 sector_t sector; /* First sector of the PPL space */
140 } ppl;
141};
142enum flag_bits {
143 Faulty, /* device is known to have a fault */
144 In_sync, /* device is in_sync with rest of array */
145 Bitmap_sync, /* ..actually, not quite In_sync. Need a
146 * bitmap-based recovery to get fully in sync.
147 * The bit is only meaningful before device
148 * has been passed to pers->hot_add_disk.
149 */
150 WriteMostly, /* Avoid reading if at all possible */
151 AutoDetected, /* added by auto-detect */
152 Blocked, /* An error occurred but has not yet
153 * been acknowledged by the metadata
154 * handler, so don't allow writes
155 * until it is cleared */
156 WriteErrorSeen, /* A write error has been seen on this
157 * device
158 */
159 FaultRecorded, /* Intermediate state for clearing
160 * Blocked. The Fault is/will-be
161 * recorded in the metadata, but that
162 * metadata hasn't been stored safely
163 * on disk yet.
164 */
165 BlockedBadBlocks, /* A writer is blocked because they
166 * found an unacknowledged bad-block.
167 * This can safely be cleared at any
168 * time, and the writer will re-check.
169 * It may be set at any time, and at
170 * worst the writer will timeout and
171 * re-check. So setting it as
172 * accurately as possible is good, but
173 * not absolutely critical.
174 */
175 WantReplacement, /* This device is a candidate to be
176 * hot-replaced, either because it has
177 * reported some faults, or because
178 * of explicit request.
179 */
180 Replacement, /* This device is a replacement for
181 * a want_replacement device with same
182 * raid_disk number.
183 */
184 Candidate, /* For clustered environments only:
185 * This device is seen locally but not
186 * by the whole cluster
187 */
188 Journal, /* This device is used as journal for
189 * raid-5/6.
190 * Usually, this device should be faster
191 * than other devices in the array
192 */
193 ClusterRemove,
194 ExternalBbl, /* External metadata provides bad
195 * block management for a disk
196 */
197 FailFast, /* Minimal retries should be attempted on
198 * this device, so use REQ_FAILFAST_DEV.
199 * Also don't try to repair failed reads.
200 * It is expects that no bad block log
201 * is present.
202 */
203 LastDev, /* Seems to be the last working dev as
204 * it didn't fail, so don't use FailFast
205 * any more for metadata
206 */
207 CollisionCheck, /*
208 * check if there is collision between raid1
209 * serial bios.
210 */
211 Nonrot, /* non-rotational device (SSD) */
212};
213
214static inline int is_badblock(struct md_rdev *rdev, sector_t s, int sectors,
215 sector_t *first_bad, int *bad_sectors)
216{
217 if (unlikely(rdev->badblocks.count)) {
218 int rv = badblocks_check(&rdev->badblocks, rdev->data_offset + s,
219 sectors,
220 first_bad, bad_sectors);
221 if (rv)
222 *first_bad -= rdev->data_offset;
223 return rv;
224 }
225 return 0;
226}
227
228static inline int rdev_has_badblock(struct md_rdev *rdev, sector_t s,
229 int sectors)
230{
231 sector_t first_bad;
232 int bad_sectors;
233
234 return is_badblock(rdev, s, sectors, &first_bad, &bad_sectors);
235}
236
237extern int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
238 int is_new);
239extern int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
240 int is_new);
241struct md_cluster_info;
242
243/**
244 * enum mddev_flags - md device flags.
245 * @MD_ARRAY_FIRST_USE: First use of array, needs initialization.
246 * @MD_CLOSING: If set, we are closing the array, do not open it then.
247 * @MD_JOURNAL_CLEAN: A raid with journal is already clean.
248 * @MD_HAS_JOURNAL: The raid array has journal feature set.
249 * @MD_CLUSTER_RESYNC_LOCKED: cluster raid only, which means node, already took
250 * resync lock, need to release the lock.
251 * @MD_FAILFAST_SUPPORTED: Using MD_FAILFAST on metadata writes is supported as
252 * calls to md_error() will never cause the array to
253 * become failed.
254 * @MD_HAS_PPL: The raid array has PPL feature set.
255 * @MD_HAS_MULTIPLE_PPLS: The raid array has multiple PPLs feature set.
256 * @MD_NOT_READY: do_md_run() is active, so 'array_state', ust not report that
257 * array is ready yet.
258 * @MD_BROKEN: This is used to stop writes and mark array as failed.
259 * @MD_DELETED: This device is being deleted
260 *
261 * change UNSUPPORTED_MDDEV_FLAGS for each array type if new flag is added
262 */
263enum mddev_flags {
264 MD_ARRAY_FIRST_USE,
265 MD_CLOSING,
266 MD_JOURNAL_CLEAN,
267 MD_HAS_JOURNAL,
268 MD_CLUSTER_RESYNC_LOCKED,
269 MD_FAILFAST_SUPPORTED,
270 MD_HAS_PPL,
271 MD_HAS_MULTIPLE_PPLS,
272 MD_NOT_READY,
273 MD_BROKEN,
274 MD_DELETED,
275};
276
277enum mddev_sb_flags {
278 MD_SB_CHANGE_DEVS, /* Some device status has changed */
279 MD_SB_CHANGE_CLEAN, /* transition to or from 'clean' */
280 MD_SB_CHANGE_PENDING, /* switch from 'clean' to 'active' in progress */
281 MD_SB_NEED_REWRITE, /* metadata write needs to be repeated */
282};
283
284#define NR_SERIAL_INFOS 8
285/* record current range of serialize IOs */
286struct serial_info {
287 struct rb_node node;
288 sector_t start; /* start sector of rb node */
289 sector_t last; /* end sector of rb node */
290 sector_t _subtree_last; /* highest sector in subtree of rb node */
291};
292
293/*
294 * mddev->curr_resync stores the current sector of the resync but
295 * also has some overloaded values.
296 */
297enum {
298 /* No resync in progress */
299 MD_RESYNC_NONE = 0,
300 /* Yielded to allow another conflicting resync to commence */
301 MD_RESYNC_YIELDED = 1,
302 /* Delayed to check that there is no conflict with another sync */
303 MD_RESYNC_DELAYED = 2,
304 /* Any value greater than or equal to this is in an active resync */
305 MD_RESYNC_ACTIVE = 3,
306};
307
308struct mddev {
309 void *private;
310 struct md_personality *pers;
311 dev_t unit;
312 int md_minor;
313 struct list_head disks;
314 unsigned long flags;
315 unsigned long sb_flags;
316
317 int suspended;
318 struct mutex suspend_mutex;
319 struct percpu_ref active_io;
320 int ro;
321 int sysfs_active; /* set when sysfs deletes
322 * are happening, so run/
323 * takeover/stop are not safe
324 */
325 struct gendisk *gendisk;
326
327 struct kobject kobj;
328 int hold_active;
329#define UNTIL_IOCTL 1
330#define UNTIL_STOP 2
331
332 /* Superblock information */
333 int major_version,
334 minor_version,
335 patch_version;
336 int persistent;
337 int external; /* metadata is
338 * managed externally */
339 char metadata_type[17]; /* externally set*/
340 int chunk_sectors;
341 time64_t ctime, utime;
342 int level, layout;
343 char clevel[16];
344 int raid_disks;
345 int max_disks;
346 sector_t dev_sectors; /* used size of
347 * component devices */
348 sector_t array_sectors; /* exported array size */
349 int external_size; /* size managed
350 * externally */
351 __u64 events;
352 /* If the last 'event' was simply a clean->dirty transition, and
353 * we didn't write it to the spares, then it is safe and simple
354 * to just decrement the event count on a dirty->clean transition.
355 * So we record that possibility here.
356 */
357 int can_decrease_events;
358
359 char uuid[16];
360
361 /* If the array is being reshaped, we need to record the
362 * new shape and an indication of where we are up to.
363 * This is written to the superblock.
364 * If reshape_position is MaxSector, then no reshape is happening (yet).
365 */
366 sector_t reshape_position;
367 int delta_disks, new_level, new_layout;
368 int new_chunk_sectors;
369 int reshape_backwards;
370
371 struct md_thread __rcu *thread; /* management thread */
372 struct md_thread __rcu *sync_thread; /* doing resync or reconstruct */
373
374 /* 'last_sync_action' is initialized to "none". It is set when a
375 * sync operation (i.e "data-check", "requested-resync", "resync",
376 * "recovery", or "reshape") is started. It holds this value even
377 * when the sync thread is "frozen" (interrupted) or "idle" (stopped
378 * or finished). It is overwritten when a new sync operation is begun.
379 */
380 char *last_sync_action;
381 sector_t curr_resync; /* last block scheduled */
382 /* As resync requests can complete out of order, we cannot easily track
383 * how much resync has been completed. So we occasionally pause until
384 * everything completes, then set curr_resync_completed to curr_resync.
385 * As such it may be well behind the real resync mark, but it is a value
386 * we are certain of.
387 */
388 sector_t curr_resync_completed;
389 unsigned long resync_mark; /* a recent timestamp */
390 sector_t resync_mark_cnt;/* blocks written at resync_mark */
391 sector_t curr_mark_cnt; /* blocks scheduled now */
392
393 sector_t resync_max_sectors; /* may be set by personality */
394
395 atomic64_t resync_mismatches; /* count of sectors where
396 * parity/replica mismatch found
397 */
398
399 /* allow user-space to request suspension of IO to regions of the array */
400 sector_t suspend_lo;
401 sector_t suspend_hi;
402 /* if zero, use the system-wide default */
403 int sync_speed_min;
404 int sync_speed_max;
405
406 /* resync even though the same disks are shared among md-devices */
407 int parallel_resync;
408
409 int ok_start_degraded;
410
411 unsigned long recovery;
412 /* If a RAID personality determines that recovery (of a particular
413 * device) will fail due to a read error on the source device, it
414 * takes a copy of this number and does not attempt recovery again
415 * until this number changes.
416 */
417 int recovery_disabled;
418
419 int in_sync; /* know to not need resync */
420 /* 'open_mutex' avoids races between 'md_open' and 'do_md_stop', so
421 * that we are never stopping an array while it is open.
422 * 'reconfig_mutex' protects all other reconfiguration.
423 * These locks are separate due to conflicting interactions
424 * with disk->open_mutex.
425 * Lock ordering is:
426 * reconfig_mutex -> disk->open_mutex
427 * disk->open_mutex -> open_mutex: e.g. __blkdev_get -> md_open
428 */
429 struct mutex open_mutex;
430 struct mutex reconfig_mutex;
431 atomic_t active; /* general refcount */
432 atomic_t openers; /* number of active opens */
433
434 int changed; /* True if we might need to
435 * reread partition info */
436 int degraded; /* whether md should consider
437 * adding a spare
438 */
439
440 atomic_t recovery_active; /* blocks scheduled, but not written */
441 wait_queue_head_t recovery_wait;
442 sector_t recovery_cp;
443 sector_t resync_min; /* user requested sync
444 * starts here */
445 sector_t resync_max; /* resync should pause
446 * when it gets here */
447
448 struct kernfs_node *sysfs_state; /* handle for 'array_state'
449 * file in sysfs.
450 */
451 struct kernfs_node *sysfs_action; /* handle for 'sync_action' */
452 struct kernfs_node *sysfs_completed; /*handle for 'sync_completed' */
453 struct kernfs_node *sysfs_degraded; /*handle for 'degraded' */
454 struct kernfs_node *sysfs_level; /*handle for 'level' */
455
456 /* used for delayed sysfs removal */
457 struct work_struct del_work;
458 /* used for register new sync thread */
459 struct work_struct sync_work;
460
461 /* "lock" protects:
462 * flush_bio transition from NULL to !NULL
463 * rdev superblocks, events
464 * clearing MD_CHANGE_*
465 * in_sync - and related safemode and MD_CHANGE changes
466 * pers (also protected by reconfig_mutex and pending IO).
467 * clearing ->bitmap
468 * clearing ->bitmap_info.file
469 * changing ->resync_{min,max}
470 * setting MD_RECOVERY_RUNNING (which interacts with resync_{min,max})
471 */
472 spinlock_t lock;
473 wait_queue_head_t sb_wait; /* for waiting on superblock updates */
474 atomic_t pending_writes; /* number of active superblock writes */
475
476 unsigned int safemode; /* if set, update "clean" superblock
477 * when no writes pending.
478 */
479 unsigned int safemode_delay;
480 struct timer_list safemode_timer;
481 struct percpu_ref writes_pending;
482 int sync_checkers; /* # of threads checking writes_pending */
483
484 struct bitmap *bitmap; /* the bitmap for the device */
485 struct {
486 struct file *file; /* the bitmap file */
487 loff_t offset; /* offset from superblock of
488 * start of bitmap. May be
489 * negative, but not '0'
490 * For external metadata, offset
491 * from start of device.
492 */
493 unsigned long space; /* space available at this offset */
494 loff_t default_offset; /* this is the offset to use when
495 * hot-adding a bitmap. It should
496 * eventually be settable by sysfs.
497 */
498 unsigned long default_space; /* space available at
499 * default offset */
500 struct mutex mutex;
501 unsigned long chunksize;
502 unsigned long daemon_sleep; /* how many jiffies between updates? */
503 unsigned long max_write_behind; /* write-behind mode */
504 int external;
505 int nodes; /* Maximum number of nodes in the cluster */
506 char cluster_name[64]; /* Name of the cluster */
507 } bitmap_info;
508
509 atomic_t max_corr_read_errors; /* max read retries */
510 struct list_head all_mddevs;
511
512 const struct attribute_group *to_remove;
513
514 struct bio_set bio_set;
515 struct bio_set sync_set; /* for sync operations like
516 * metadata and bitmap writes
517 */
518 struct bio_set io_clone_set;
519
520 /* Generic flush handling.
521 * The last to finish preflush schedules a worker to submit
522 * the rest of the request (without the REQ_PREFLUSH flag).
523 */
524 struct bio *flush_bio;
525 atomic_t flush_pending;
526 ktime_t start_flush, prev_flush_start; /* prev_flush_start is when the previous completed
527 * flush was started.
528 */
529 struct work_struct flush_work;
530 struct work_struct event_work; /* used by dm to report failure event */
531 mempool_t *serial_info_pool;
532 void (*sync_super)(struct mddev *mddev, struct md_rdev *rdev);
533 struct md_cluster_info *cluster_info;
534 unsigned int good_device_nr; /* good device num within cluster raid */
535 unsigned int noio_flag; /* for memalloc scope API */
536
537 /*
538 * Temporarily store rdev that will be finally removed when
539 * reconfig_mutex is unlocked, protected by reconfig_mutex.
540 */
541 struct list_head deleting;
542
543 /* Used to synchronize idle and frozen for action_store() */
544 struct mutex sync_mutex;
545 /* The sequence number for sync thread */
546 atomic_t sync_seq;
547
548 bool has_superblocks:1;
549 bool fail_last_dev:1;
550 bool serialize_policy:1;
551};
552
553enum recovery_flags {
554 /*
555 * If neither SYNC or RESHAPE are set, then it is a recovery.
556 */
557 MD_RECOVERY_RUNNING, /* a thread is running, or about to be started */
558 MD_RECOVERY_SYNC, /* actually doing a resync, not a recovery */
559 MD_RECOVERY_RECOVER, /* doing recovery, or need to try it. */
560 MD_RECOVERY_INTR, /* resync needs to be aborted for some reason */
561 MD_RECOVERY_DONE, /* thread is done and is waiting to be reaped */
562 MD_RECOVERY_NEEDED, /* we might need to start a resync/recover */
563 MD_RECOVERY_REQUESTED, /* user-space has requested a sync (used with SYNC) */
564 MD_RECOVERY_CHECK, /* user-space request for check-only, no repair */
565 MD_RECOVERY_RESHAPE, /* A reshape is happening */
566 MD_RECOVERY_FROZEN, /* User request to abort, and not restart, any action */
567 MD_RECOVERY_ERROR, /* sync-action interrupted because io-error */
568 MD_RECOVERY_WAIT, /* waiting for pers->start() to finish */
569 MD_RESYNCING_REMOTE, /* remote node is running resync thread */
570};
571
572enum md_ro_state {
573 MD_RDWR,
574 MD_RDONLY,
575 MD_AUTO_READ,
576 MD_MAX_STATE
577};
578
579static inline bool md_is_rdwr(struct mddev *mddev)
580{
581 return (mddev->ro == MD_RDWR);
582}
583
584static inline bool reshape_interrupted(struct mddev *mddev)
585{
586 /* reshape never start */
587 if (mddev->reshape_position == MaxSector)
588 return false;
589
590 /* interrupted */
591 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
592 return true;
593
594 /* running reshape will be interrupted soon. */
595 if (test_bit(MD_RECOVERY_WAIT, &mddev->recovery) ||
596 test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
597 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
598 return true;
599
600 return false;
601}
602
603static inline int __must_check mddev_lock(struct mddev *mddev)
604{
605 return mutex_lock_interruptible(&mddev->reconfig_mutex);
606}
607
608/* Sometimes we need to take the lock in a situation where
609 * failure due to interrupts is not acceptable.
610 */
611static inline void mddev_lock_nointr(struct mddev *mddev)
612{
613 mutex_lock(&mddev->reconfig_mutex);
614}
615
616static inline int mddev_trylock(struct mddev *mddev)
617{
618 return mutex_trylock(&mddev->reconfig_mutex);
619}
620extern void mddev_unlock(struct mddev *mddev);
621
622static inline void md_sync_acct(struct block_device *bdev, unsigned long nr_sectors)
623{
624 atomic_add(nr_sectors, &bdev->bd_disk->sync_io);
625}
626
627static inline void md_sync_acct_bio(struct bio *bio, unsigned long nr_sectors)
628{
629 md_sync_acct(bio->bi_bdev, nr_sectors);
630}
631
632struct md_personality
633{
634 char *name;
635 int level;
636 struct list_head list;
637 struct module *owner;
638 bool __must_check (*make_request)(struct mddev *mddev, struct bio *bio);
639 /*
640 * start up works that do NOT require md_thread. tasks that
641 * requires md_thread should go into start()
642 */
643 int (*run)(struct mddev *mddev);
644 /* start up works that require md threads */
645 int (*start)(struct mddev *mddev);
646 void (*free)(struct mddev *mddev, void *priv);
647 void (*status)(struct seq_file *seq, struct mddev *mddev);
648 /* error_handler must set ->faulty and clear ->in_sync
649 * if appropriate, and should abort recovery if needed
650 */
651 void (*error_handler)(struct mddev *mddev, struct md_rdev *rdev);
652 int (*hot_add_disk) (struct mddev *mddev, struct md_rdev *rdev);
653 int (*hot_remove_disk) (struct mddev *mddev, struct md_rdev *rdev);
654 int (*spare_active) (struct mddev *mddev);
655 sector_t (*sync_request)(struct mddev *mddev, sector_t sector_nr, int *skipped);
656 int (*resize) (struct mddev *mddev, sector_t sectors);
657 sector_t (*size) (struct mddev *mddev, sector_t sectors, int raid_disks);
658 int (*check_reshape) (struct mddev *mddev);
659 int (*start_reshape) (struct mddev *mddev);
660 void (*finish_reshape) (struct mddev *mddev);
661 void (*update_reshape_pos) (struct mddev *mddev);
662 void (*prepare_suspend) (struct mddev *mddev);
663 /* quiesce suspends or resumes internal processing.
664 * 1 - stop new actions and wait for action io to complete
665 * 0 - return to normal behaviour
666 */
667 void (*quiesce) (struct mddev *mddev, int quiesce);
668 /* takeover is used to transition an array from one
669 * personality to another. The new personality must be able
670 * to handle the data in the current layout.
671 * e.g. 2drive raid1 -> 2drive raid5
672 * ndrive raid5 -> degraded n+1drive raid6 with special layout
673 * If the takeover succeeds, a new 'private' structure is returned.
674 * This needs to be installed and then ->run used to activate the
675 * array.
676 */
677 void *(*takeover) (struct mddev *mddev);
678 /* Changes the consistency policy of an active array. */
679 int (*change_consistency_policy)(struct mddev *mddev, const char *buf);
680};
681
682struct md_sysfs_entry {
683 struct attribute attr;
684 ssize_t (*show)(struct mddev *, char *);
685 ssize_t (*store)(struct mddev *, const char *, size_t);
686};
687extern const struct attribute_group md_bitmap_group;
688
689static inline struct kernfs_node *sysfs_get_dirent_safe(struct kernfs_node *sd, char *name)
690{
691 if (sd)
692 return sysfs_get_dirent(sd, name);
693 return sd;
694}
695static inline void sysfs_notify_dirent_safe(struct kernfs_node *sd)
696{
697 if (sd)
698 sysfs_notify_dirent(sd);
699}
700
701static inline char * mdname (struct mddev * mddev)
702{
703 return mddev->gendisk ? mddev->gendisk->disk_name : "mdX";
704}
705
706static inline int sysfs_link_rdev(struct mddev *mddev, struct md_rdev *rdev)
707{
708 char nm[20];
709 if (!test_bit(Replacement, &rdev->flags) &&
710 !test_bit(Journal, &rdev->flags) &&
711 mddev->kobj.sd) {
712 sprintf(nm, "rd%d", rdev->raid_disk);
713 return sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
714 } else
715 return 0;
716}
717
718static inline void sysfs_unlink_rdev(struct mddev *mddev, struct md_rdev *rdev)
719{
720 char nm[20];
721 if (!test_bit(Replacement, &rdev->flags) &&
722 !test_bit(Journal, &rdev->flags) &&
723 mddev->kobj.sd) {
724 sprintf(nm, "rd%d", rdev->raid_disk);
725 sysfs_remove_link(&mddev->kobj, nm);
726 }
727}
728
729/*
730 * iterates through some rdev ringlist. It's safe to remove the
731 * current 'rdev'. Dont touch 'tmp' though.
732 */
733#define rdev_for_each_list(rdev, tmp, head) \
734 list_for_each_entry_safe(rdev, tmp, head, same_set)
735
736/*
737 * iterates through the 'same array disks' ringlist
738 */
739#define rdev_for_each(rdev, mddev) \
740 list_for_each_entry(rdev, &((mddev)->disks), same_set)
741
742#define rdev_for_each_safe(rdev, tmp, mddev) \
743 list_for_each_entry_safe(rdev, tmp, &((mddev)->disks), same_set)
744
745#define rdev_for_each_rcu(rdev, mddev) \
746 list_for_each_entry_rcu(rdev, &((mddev)->disks), same_set)
747
748struct md_thread {
749 void (*run) (struct md_thread *thread);
750 struct mddev *mddev;
751 wait_queue_head_t wqueue;
752 unsigned long flags;
753 struct task_struct *tsk;
754 unsigned long timeout;
755 void *private;
756};
757
758struct md_io_clone {
759 struct mddev *mddev;
760 struct bio *orig_bio;
761 unsigned long start_time;
762 struct bio bio_clone;
763};
764
765#define THREAD_WAKEUP 0
766
767static inline void safe_put_page(struct page *p)
768{
769 if (p) put_page(p);
770}
771
772extern int register_md_personality(struct md_personality *p);
773extern int unregister_md_personality(struct md_personality *p);
774extern int register_md_cluster_operations(struct md_cluster_operations *ops,
775 struct module *module);
776extern int unregister_md_cluster_operations(void);
777extern int md_setup_cluster(struct mddev *mddev, int nodes);
778extern void md_cluster_stop(struct mddev *mddev);
779extern struct md_thread *md_register_thread(
780 void (*run)(struct md_thread *thread),
781 struct mddev *mddev,
782 const char *name);
783extern void md_unregister_thread(struct mddev *mddev, struct md_thread __rcu **threadp);
784extern void md_wakeup_thread(struct md_thread __rcu *thread);
785extern void md_check_recovery(struct mddev *mddev);
786extern void md_reap_sync_thread(struct mddev *mddev);
787extern bool md_write_start(struct mddev *mddev, struct bio *bi);
788extern void md_write_inc(struct mddev *mddev, struct bio *bi);
789extern void md_write_end(struct mddev *mddev);
790extern void md_done_sync(struct mddev *mddev, int blocks, int ok);
791extern void md_error(struct mddev *mddev, struct md_rdev *rdev);
792extern void md_finish_reshape(struct mddev *mddev);
793void md_submit_discard_bio(struct mddev *mddev, struct md_rdev *rdev,
794 struct bio *bio, sector_t start, sector_t size);
795void md_account_bio(struct mddev *mddev, struct bio **bio);
796void md_free_cloned_bio(struct bio *bio);
797
798extern bool __must_check md_flush_request(struct mddev *mddev, struct bio *bio);
799extern void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
800 sector_t sector, int size, struct page *page);
801extern int md_super_wait(struct mddev *mddev);
802extern int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
803 struct page *page, blk_opf_t opf, bool metadata_op);
804extern void md_do_sync(struct md_thread *thread);
805extern void md_new_event(void);
806extern void md_allow_write(struct mddev *mddev);
807extern void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev);
808extern void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors);
809extern int md_check_no_bitmap(struct mddev *mddev);
810extern int md_integrity_register(struct mddev *mddev);
811extern int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev);
812extern int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale);
813
814extern int mddev_init(struct mddev *mddev);
815extern void mddev_destroy(struct mddev *mddev);
816struct mddev *md_alloc(dev_t dev, char *name);
817void mddev_put(struct mddev *mddev);
818extern int md_run(struct mddev *mddev);
819extern int md_start(struct mddev *mddev);
820extern void md_stop(struct mddev *mddev);
821extern void md_stop_writes(struct mddev *mddev);
822extern int md_rdev_init(struct md_rdev *rdev);
823extern void md_rdev_clear(struct md_rdev *rdev);
824
825extern bool md_handle_request(struct mddev *mddev, struct bio *bio);
826extern int mddev_suspend(struct mddev *mddev, bool interruptible);
827extern void mddev_resume(struct mddev *mddev);
828extern void md_idle_sync_thread(struct mddev *mddev);
829extern void md_frozen_sync_thread(struct mddev *mddev);
830extern void md_unfrozen_sync_thread(struct mddev *mddev);
831
832extern void md_reload_sb(struct mddev *mddev, int raid_disk);
833extern void md_update_sb(struct mddev *mddev, int force);
834extern void mddev_create_serial_pool(struct mddev *mddev, struct md_rdev *rdev);
835extern void mddev_destroy_serial_pool(struct mddev *mddev,
836 struct md_rdev *rdev);
837struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr);
838struct md_rdev *md_find_rdev_rcu(struct mddev *mddev, dev_t dev);
839
840static inline bool is_rdev_broken(struct md_rdev *rdev)
841{
842 return !disk_live(rdev->bdev->bd_disk);
843}
844
845static inline void rdev_dec_pending(struct md_rdev *rdev, struct mddev *mddev)
846{
847 int faulty = test_bit(Faulty, &rdev->flags);
848 if (atomic_dec_and_test(&rdev->nr_pending) && faulty) {
849 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
850 md_wakeup_thread(mddev->thread);
851 }
852}
853
854extern struct md_cluster_operations *md_cluster_ops;
855static inline int mddev_is_clustered(struct mddev *mddev)
856{
857 return mddev->cluster_info && mddev->bitmap_info.nodes > 1;
858}
859
860/* clear unsupported mddev_flags */
861static inline void mddev_clear_unsupported_flags(struct mddev *mddev,
862 unsigned long unsupported_flags)
863{
864 mddev->flags &= ~unsupported_flags;
865}
866
867static inline void mddev_check_write_zeroes(struct mddev *mddev, struct bio *bio)
868{
869 if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
870 !bio->bi_bdev->bd_disk->queue->limits.max_write_zeroes_sectors)
871 mddev->gendisk->queue->limits.max_write_zeroes_sectors = 0;
872}
873
874static inline int mddev_suspend_and_lock(struct mddev *mddev)
875{
876 int ret;
877
878 ret = mddev_suspend(mddev, true);
879 if (ret)
880 return ret;
881
882 ret = mddev_lock(mddev);
883 if (ret)
884 mddev_resume(mddev);
885
886 return ret;
887}
888
889static inline void mddev_suspend_and_lock_nointr(struct mddev *mddev)
890{
891 mddev_suspend(mddev, false);
892 mutex_lock(&mddev->reconfig_mutex);
893}
894
895static inline void mddev_unlock_and_resume(struct mddev *mddev)
896{
897 mddev_unlock(mddev);
898 mddev_resume(mddev);
899}
900
901struct mdu_array_info_s;
902struct mdu_disk_info_s;
903
904extern int mdp_major;
905extern struct workqueue_struct *md_bitmap_wq;
906void md_autostart_arrays(int part);
907int md_set_array_info(struct mddev *mddev, struct mdu_array_info_s *info);
908int md_add_new_disk(struct mddev *mddev, struct mdu_disk_info_s *info);
909int do_md_run(struct mddev *mddev);
910void mddev_stack_rdev_limits(struct mddev *mddev, struct queue_limits *lim);
911int mddev_stack_new_rdev(struct mddev *mddev, struct md_rdev *rdev);
912void mddev_update_io_opt(struct mddev *mddev, unsigned int nr_stripes);
913
914extern const struct block_device_operations md_fops;
915
916/*
917 * MD devices can be used undeneath by DM, in which case ->gendisk is NULL.
918 */
919static inline bool mddev_is_dm(struct mddev *mddev)
920{
921 return !mddev->gendisk;
922}
923
924static inline void mddev_trace_remap(struct mddev *mddev, struct bio *bio,
925 sector_t sector)
926{
927 if (!mddev_is_dm(mddev))
928 trace_block_bio_remap(bio, disk_devt(mddev->gendisk), sector);
929}
930
931#define mddev_add_trace_msg(mddev, fmt, args...) \
932do { \
933 if (!mddev_is_dm(mddev)) \
934 blk_add_trace_msg((mddev)->gendisk->queue, fmt, ##args); \
935} while (0)
936
937#endif /* _MD_MD_H */