Loading...
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 "md-cluster.h"
22
23#define MaxSector (~(sector_t)0)
24
25/*
26 * These flags should really be called "NO_RETRY" rather than
27 * "FAILFAST" because they don't make any promise about time lapse,
28 * only about the number of retries, which will be zero.
29 * REQ_FAILFAST_DRIVER is not included because
30 * Commit: 4a27446f3e39 ("[SCSI] modify scsi to handle new fail fast flags.")
31 * seems to suggest that the errors it avoids retrying should usually
32 * be retried.
33 */
34#define MD_FAILFAST (REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT)
35
36/*
37 * The struct embedded in rdev is used to serialize IO.
38 */
39struct serial_in_rdev {
40 struct rb_root_cached serial_rb;
41 spinlock_t serial_lock;
42 wait_queue_head_t serial_io_wait;
43};
44
45/*
46 * MD's 'extended' device
47 */
48struct md_rdev {
49 struct list_head same_set; /* RAID devices within the same set */
50
51 sector_t sectors; /* Device size (in 512bytes sectors) */
52 struct mddev *mddev; /* RAID array if running */
53 int last_events; /* IO event timestamp */
54
55 /*
56 * If meta_bdev is non-NULL, it means that a separate device is
57 * being used to store the metadata (superblock/bitmap) which
58 * would otherwise be contained on the same device as the data (bdev).
59 */
60 struct block_device *meta_bdev;
61 struct block_device *bdev; /* block device handle */
62
63 struct page *sb_page, *bb_page;
64 int sb_loaded;
65 __u64 sb_events;
66 sector_t data_offset; /* start of data in array */
67 sector_t new_data_offset;/* only relevant while reshaping */
68 sector_t sb_start; /* offset of the super block (in 512byte sectors) */
69 int sb_size; /* bytes in the superblock */
70 int preferred_minor; /* autorun support */
71
72 struct kobject kobj;
73
74 /* A device can be in one of three states based on two flags:
75 * Not working: faulty==1 in_sync==0
76 * Fully working: faulty==0 in_sync==1
77 * Working, but not
78 * in sync with array
79 * faulty==0 in_sync==0
80 *
81 * It can never have faulty==1, in_sync==1
82 * This reduces the burden of testing multiple flags in many cases
83 */
84
85 unsigned long flags; /* bit set of 'enum flag_bits' bits. */
86 wait_queue_head_t blocked_wait;
87
88 int desc_nr; /* descriptor index in the superblock */
89 int raid_disk; /* role of device in array */
90 int new_raid_disk; /* role that the device will have in
91 * the array after a level-change completes.
92 */
93 int saved_raid_disk; /* role that device used to have in the
94 * array and could again if we did a partial
95 * resync from the bitmap
96 */
97 union {
98 sector_t recovery_offset;/* If this device has been partially
99 * recovered, this is where we were
100 * up to.
101 */
102 sector_t journal_tail; /* If this device is a journal device,
103 * this is the journal tail (journal
104 * recovery start point)
105 */
106 };
107
108 atomic_t nr_pending; /* number of pending requests.
109 * only maintained for arrays that
110 * support hot removal
111 */
112 atomic_t read_errors; /* number of consecutive read errors that
113 * we have tried to ignore.
114 */
115 time64_t last_read_error; /* monotonic time since our
116 * last read error
117 */
118 atomic_t corrected_errors; /* number of corrected read errors,
119 * for reporting to userspace and storing
120 * in superblock.
121 */
122
123 struct serial_in_rdev *serial; /* used for raid1 io serialization */
124
125 struct work_struct del_work; /* used for delayed sysfs removal */
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 RemoveSynchronized, /* synchronize_rcu() was called after
195 * this device was known to be faulty,
196 * so it is safe to remove without
197 * another synchronize_rcu() call.
198 */
199 ExternalBbl, /* External metadata provides bad
200 * block management for a disk
201 */
202 FailFast, /* Minimal retries should be attempted on
203 * this device, so use REQ_FAILFAST_DEV.
204 * Also don't try to repair failed reads.
205 * It is expects that no bad block log
206 * is present.
207 */
208 LastDev, /* Seems to be the last working dev as
209 * it didn't fail, so don't use FailFast
210 * any more for metadata
211 */
212 CollisionCheck, /*
213 * check if there is collision between raid1
214 * serial bios.
215 */
216};
217
218static inline int is_badblock(struct md_rdev *rdev, sector_t s, int sectors,
219 sector_t *first_bad, int *bad_sectors)
220{
221 if (unlikely(rdev->badblocks.count)) {
222 int rv = badblocks_check(&rdev->badblocks, rdev->data_offset + s,
223 sectors,
224 first_bad, bad_sectors);
225 if (rv)
226 *first_bad -= rdev->data_offset;
227 return rv;
228 }
229 return 0;
230}
231extern int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
232 int is_new);
233extern int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
234 int is_new);
235struct md_cluster_info;
236
237/**
238 * enum mddev_flags - md device flags.
239 * @MD_ARRAY_FIRST_USE: First use of array, needs initialization.
240 * @MD_CLOSING: If set, we are closing the array, do not open it then.
241 * @MD_JOURNAL_CLEAN: A raid with journal is already clean.
242 * @MD_HAS_JOURNAL: The raid array has journal feature set.
243 * @MD_CLUSTER_RESYNC_LOCKED: cluster raid only, which means node, already took
244 * resync lock, need to release the lock.
245 * @MD_FAILFAST_SUPPORTED: Using MD_FAILFAST on metadata writes is supported as
246 * calls to md_error() will never cause the array to
247 * become failed.
248 * @MD_HAS_PPL: The raid array has PPL feature set.
249 * @MD_HAS_MULTIPLE_PPLS: The raid array has multiple PPLs feature set.
250 * @MD_ALLOW_SB_UPDATE: md_check_recovery is allowed to update the metadata
251 * without taking reconfig_mutex.
252 * @MD_UPDATING_SB: md_check_recovery is updating the metadata without
253 * explicitly holding reconfig_mutex.
254 * @MD_NOT_READY: do_md_run() is active, so 'array_state', ust not report that
255 * array is ready yet.
256 * @MD_BROKEN: This is used to stop writes and mark array as failed.
257 * @MD_DELETED: This device is being deleted
258 *
259 * change UNSUPPORTED_MDDEV_FLAGS for each array type if new flag is added
260 */
261enum mddev_flags {
262 MD_ARRAY_FIRST_USE,
263 MD_CLOSING,
264 MD_JOURNAL_CLEAN,
265 MD_HAS_JOURNAL,
266 MD_CLUSTER_RESYNC_LOCKED,
267 MD_FAILFAST_SUPPORTED,
268 MD_HAS_PPL,
269 MD_HAS_MULTIPLE_PPLS,
270 MD_ALLOW_SB_UPDATE,
271 MD_UPDATING_SB,
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 atomic_t active_io;
319 int ro;
320 int sysfs_active; /* set when sysfs deletes
321 * are happening, so run/
322 * takeover/stop are not safe
323 */
324 struct gendisk *gendisk;
325
326 struct kobject kobj;
327 int hold_active;
328#define UNTIL_IOCTL 1
329#define UNTIL_STOP 2
330
331 /* Superblock information */
332 int major_version,
333 minor_version,
334 patch_version;
335 int persistent;
336 int external; /* metadata is
337 * managed externally */
338 char metadata_type[17]; /* externally set*/
339 int chunk_sectors;
340 time64_t ctime, utime;
341 int level, layout;
342 char clevel[16];
343 int raid_disks;
344 int max_disks;
345 sector_t dev_sectors; /* used size of
346 * component devices */
347 sector_t array_sectors; /* exported array size */
348 int external_size; /* size managed
349 * externally */
350 __u64 events;
351 /* If the last 'event' was simply a clean->dirty transition, and
352 * we didn't write it to the spares, then it is safe and simple
353 * to just decrement the event count on a dirty->clean transition.
354 * So we record that possibility here.
355 */
356 int can_decrease_events;
357
358 char uuid[16];
359
360 /* If the array is being reshaped, we need to record the
361 * new shape and an indication of where we are up to.
362 * This is written to the superblock.
363 * If reshape_position is MaxSector, then no reshape is happening (yet).
364 */
365 sector_t reshape_position;
366 int delta_disks, new_level, new_layout;
367 int new_chunk_sectors;
368 int reshape_backwards;
369
370 struct md_thread *thread; /* management thread */
371 struct md_thread *sync_thread; /* doing resync or reconstruct */
372
373 /* 'last_sync_action' is initialized to "none". It is set when a
374 * sync operation (i.e "data-check", "requested-resync", "resync",
375 * "recovery", or "reshape") is started. It holds this value even
376 * when the sync thread is "frozen" (interrupted) or "idle" (stopped
377 * or finished). It is overwritten when a new sync operation is begun.
378 */
379 char *last_sync_action;
380 sector_t curr_resync; /* last block scheduled */
381 /* As resync requests can complete out of order, we cannot easily track
382 * how much resync has been completed. So we occasionally pause until
383 * everything completes, then set curr_resync_completed to curr_resync.
384 * As such it may be well behind the real resync mark, but it is a value
385 * we are certain of.
386 */
387 sector_t curr_resync_completed;
388 unsigned long resync_mark; /* a recent timestamp */
389 sector_t resync_mark_cnt;/* blocks written at resync_mark */
390 sector_t curr_mark_cnt; /* blocks scheduled now */
391
392 sector_t resync_max_sectors; /* may be set by personality */
393
394 atomic64_t resync_mismatches; /* count of sectors where
395 * parity/replica mismatch found
396 */
397
398 /* allow user-space to request suspension of IO to regions of the array */
399 sector_t suspend_lo;
400 sector_t suspend_hi;
401 /* if zero, use the system-wide default */
402 int sync_speed_min;
403 int sync_speed_max;
404
405 /* resync even though the same disks are shared among md-devices */
406 int parallel_resync;
407
408 int ok_start_degraded;
409
410 unsigned long recovery;
411 /* If a RAID personality determines that recovery (of a particular
412 * device) will fail due to a read error on the source device, it
413 * takes a copy of this number and does not attempt recovery again
414 * until this number changes.
415 */
416 int recovery_disabled;
417
418 int in_sync; /* know to not need resync */
419 /* 'open_mutex' avoids races between 'md_open' and 'do_md_stop', so
420 * that we are never stopping an array while it is open.
421 * 'reconfig_mutex' protects all other reconfiguration.
422 * These locks are separate due to conflicting interactions
423 * with disk->open_mutex.
424 * Lock ordering is:
425 * reconfig_mutex -> disk->open_mutex
426 * disk->open_mutex -> open_mutex: e.g. __blkdev_get -> md_open
427 */
428 struct mutex open_mutex;
429 struct mutex reconfig_mutex;
430 atomic_t active; /* general refcount */
431 atomic_t openers; /* number of active opens */
432
433 int changed; /* True if we might need to
434 * reread partition info */
435 int degraded; /* whether md should consider
436 * adding a spare
437 */
438
439 atomic_t recovery_active; /* blocks scheduled, but not written */
440 wait_queue_head_t recovery_wait;
441 sector_t recovery_cp;
442 sector_t resync_min; /* user requested sync
443 * starts here */
444 sector_t resync_max; /* resync should pause
445 * when it gets here */
446
447 struct kernfs_node *sysfs_state; /* handle for 'array_state'
448 * file in sysfs.
449 */
450 struct kernfs_node *sysfs_action; /* handle for 'sync_action' */
451 struct kernfs_node *sysfs_completed; /*handle for 'sync_completed' */
452 struct kernfs_node *sysfs_degraded; /*handle for 'degraded' */
453 struct kernfs_node *sysfs_level; /*handle for 'level' */
454
455 struct work_struct del_work; /* used for delayed sysfs removal */
456
457 /* "lock" protects:
458 * flush_bio transition from NULL to !NULL
459 * rdev superblocks, events
460 * clearing MD_CHANGE_*
461 * in_sync - and related safemode and MD_CHANGE changes
462 * pers (also protected by reconfig_mutex and pending IO).
463 * clearing ->bitmap
464 * clearing ->bitmap_info.file
465 * changing ->resync_{min,max}
466 * setting MD_RECOVERY_RUNNING (which interacts with resync_{min,max})
467 */
468 spinlock_t lock;
469 wait_queue_head_t sb_wait; /* for waiting on superblock updates */
470 atomic_t pending_writes; /* number of active superblock writes */
471
472 unsigned int safemode; /* if set, update "clean" superblock
473 * when no writes pending.
474 */
475 unsigned int safemode_delay;
476 struct timer_list safemode_timer;
477 struct percpu_ref writes_pending;
478 int sync_checkers; /* # of threads checking writes_pending */
479 struct request_queue *queue; /* for plugging ... */
480
481 struct bitmap *bitmap; /* the bitmap for the device */
482 struct {
483 struct file *file; /* the bitmap file */
484 loff_t offset; /* offset from superblock of
485 * start of bitmap. May be
486 * negative, but not '0'
487 * For external metadata, offset
488 * from start of device.
489 */
490 unsigned long space; /* space available at this offset */
491 loff_t default_offset; /* this is the offset to use when
492 * hot-adding a bitmap. It should
493 * eventually be settable by sysfs.
494 */
495 unsigned long default_space; /* space available at
496 * default offset */
497 struct mutex mutex;
498 unsigned long chunksize;
499 unsigned long daemon_sleep; /* how many jiffies between updates? */
500 unsigned long max_write_behind; /* write-behind mode */
501 int external;
502 int nodes; /* Maximum number of nodes in the cluster */
503 char cluster_name[64]; /* Name of the cluster */
504 } bitmap_info;
505
506 atomic_t max_corr_read_errors; /* max read retries */
507 struct list_head all_mddevs;
508
509 const struct attribute_group *to_remove;
510
511 struct bio_set bio_set;
512 struct bio_set sync_set; /* for sync operations like
513 * metadata and bitmap writes
514 */
515 struct bio_set io_acct_set; /* for raid0 and raid5 io accounting */
516
517 /* Generic flush handling.
518 * The last to finish preflush schedules a worker to submit
519 * the rest of the request (without the REQ_PREFLUSH flag).
520 */
521 struct bio *flush_bio;
522 atomic_t flush_pending;
523 ktime_t start_flush, prev_flush_start; /* prev_flush_start is when the previous completed
524 * flush was started.
525 */
526 struct work_struct flush_work;
527 struct work_struct event_work; /* used by dm to report failure event */
528 mempool_t *serial_info_pool;
529 void (*sync_super)(struct mddev *mddev, struct md_rdev *rdev);
530 struct md_cluster_info *cluster_info;
531 unsigned int good_device_nr; /* good device num within cluster raid */
532 unsigned int noio_flag; /* for memalloc scope API */
533
534 bool has_superblocks:1;
535 bool fail_last_dev:1;
536 bool serialize_policy:1;
537};
538
539enum recovery_flags {
540 /*
541 * If neither SYNC or RESHAPE are set, then it is a recovery.
542 */
543 MD_RECOVERY_RUNNING, /* a thread is running, or about to be started */
544 MD_RECOVERY_SYNC, /* actually doing a resync, not a recovery */
545 MD_RECOVERY_RECOVER, /* doing recovery, or need to try it. */
546 MD_RECOVERY_INTR, /* resync needs to be aborted for some reason */
547 MD_RECOVERY_DONE, /* thread is done and is waiting to be reaped */
548 MD_RECOVERY_NEEDED, /* we might need to start a resync/recover */
549 MD_RECOVERY_REQUESTED, /* user-space has requested a sync (used with SYNC) */
550 MD_RECOVERY_CHECK, /* user-space request for check-only, no repair */
551 MD_RECOVERY_RESHAPE, /* A reshape is happening */
552 MD_RECOVERY_FROZEN, /* User request to abort, and not restart, any action */
553 MD_RECOVERY_ERROR, /* sync-action interrupted because io-error */
554 MD_RECOVERY_WAIT, /* waiting for pers->start() to finish */
555 MD_RESYNCING_REMOTE, /* remote node is running resync thread */
556};
557
558static inline int __must_check mddev_lock(struct mddev *mddev)
559{
560 return mutex_lock_interruptible(&mddev->reconfig_mutex);
561}
562
563/* Sometimes we need to take the lock in a situation where
564 * failure due to interrupts is not acceptable.
565 */
566static inline void mddev_lock_nointr(struct mddev *mddev)
567{
568 mutex_lock(&mddev->reconfig_mutex);
569}
570
571static inline int mddev_trylock(struct mddev *mddev)
572{
573 return mutex_trylock(&mddev->reconfig_mutex);
574}
575extern void mddev_unlock(struct mddev *mddev);
576
577static inline void md_sync_acct(struct block_device *bdev, unsigned long nr_sectors)
578{
579 atomic_add(nr_sectors, &bdev->bd_disk->sync_io);
580}
581
582static inline void md_sync_acct_bio(struct bio *bio, unsigned long nr_sectors)
583{
584 md_sync_acct(bio->bi_bdev, nr_sectors);
585}
586
587struct md_personality
588{
589 char *name;
590 int level;
591 struct list_head list;
592 struct module *owner;
593 bool __must_check (*make_request)(struct mddev *mddev, struct bio *bio);
594 /*
595 * start up works that do NOT require md_thread. tasks that
596 * requires md_thread should go into start()
597 */
598 int (*run)(struct mddev *mddev);
599 /* start up works that require md threads */
600 int (*start)(struct mddev *mddev);
601 void (*free)(struct mddev *mddev, void *priv);
602 void (*status)(struct seq_file *seq, struct mddev *mddev);
603 /* error_handler must set ->faulty and clear ->in_sync
604 * if appropriate, and should abort recovery if needed
605 */
606 void (*error_handler)(struct mddev *mddev, struct md_rdev *rdev);
607 int (*hot_add_disk) (struct mddev *mddev, struct md_rdev *rdev);
608 int (*hot_remove_disk) (struct mddev *mddev, struct md_rdev *rdev);
609 int (*spare_active) (struct mddev *mddev);
610 sector_t (*sync_request)(struct mddev *mddev, sector_t sector_nr, int *skipped);
611 int (*resize) (struct mddev *mddev, sector_t sectors);
612 sector_t (*size) (struct mddev *mddev, sector_t sectors, int raid_disks);
613 int (*check_reshape) (struct mddev *mddev);
614 int (*start_reshape) (struct mddev *mddev);
615 void (*finish_reshape) (struct mddev *mddev);
616 void (*update_reshape_pos) (struct mddev *mddev);
617 /* quiesce suspends or resumes internal processing.
618 * 1 - stop new actions and wait for action io to complete
619 * 0 - return to normal behaviour
620 */
621 void (*quiesce) (struct mddev *mddev, int quiesce);
622 /* takeover is used to transition an array from one
623 * personality to another. The new personality must be able
624 * to handle the data in the current layout.
625 * e.g. 2drive raid1 -> 2drive raid5
626 * ndrive raid5 -> degraded n+1drive raid6 with special layout
627 * If the takeover succeeds, a new 'private' structure is returned.
628 * This needs to be installed and then ->run used to activate the
629 * array.
630 */
631 void *(*takeover) (struct mddev *mddev);
632 /* Changes the consistency policy of an active array. */
633 int (*change_consistency_policy)(struct mddev *mddev, const char *buf);
634};
635
636struct md_sysfs_entry {
637 struct attribute attr;
638 ssize_t (*show)(struct mddev *, char *);
639 ssize_t (*store)(struct mddev *, const char *, size_t);
640};
641extern const struct attribute_group md_bitmap_group;
642
643static inline struct kernfs_node *sysfs_get_dirent_safe(struct kernfs_node *sd, char *name)
644{
645 if (sd)
646 return sysfs_get_dirent(sd, name);
647 return sd;
648}
649static inline void sysfs_notify_dirent_safe(struct kernfs_node *sd)
650{
651 if (sd)
652 sysfs_notify_dirent(sd);
653}
654
655static inline char * mdname (struct mddev * mddev)
656{
657 return mddev->gendisk ? mddev->gendisk->disk_name : "mdX";
658}
659
660static inline int sysfs_link_rdev(struct mddev *mddev, struct md_rdev *rdev)
661{
662 char nm[20];
663 if (!test_bit(Replacement, &rdev->flags) &&
664 !test_bit(Journal, &rdev->flags) &&
665 mddev->kobj.sd) {
666 sprintf(nm, "rd%d", rdev->raid_disk);
667 return sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
668 } else
669 return 0;
670}
671
672static inline void sysfs_unlink_rdev(struct mddev *mddev, struct md_rdev *rdev)
673{
674 char nm[20];
675 if (!test_bit(Replacement, &rdev->flags) &&
676 !test_bit(Journal, &rdev->flags) &&
677 mddev->kobj.sd) {
678 sprintf(nm, "rd%d", rdev->raid_disk);
679 sysfs_remove_link(&mddev->kobj, nm);
680 }
681}
682
683/*
684 * iterates through some rdev ringlist. It's safe to remove the
685 * current 'rdev'. Dont touch 'tmp' though.
686 */
687#define rdev_for_each_list(rdev, tmp, head) \
688 list_for_each_entry_safe(rdev, tmp, head, same_set)
689
690/*
691 * iterates through the 'same array disks' ringlist
692 */
693#define rdev_for_each(rdev, mddev) \
694 list_for_each_entry(rdev, &((mddev)->disks), same_set)
695
696#define rdev_for_each_safe(rdev, tmp, mddev) \
697 list_for_each_entry_safe(rdev, tmp, &((mddev)->disks), same_set)
698
699#define rdev_for_each_rcu(rdev, mddev) \
700 list_for_each_entry_rcu(rdev, &((mddev)->disks), same_set)
701
702struct md_thread {
703 void (*run) (struct md_thread *thread);
704 struct mddev *mddev;
705 wait_queue_head_t wqueue;
706 unsigned long flags;
707 struct task_struct *tsk;
708 unsigned long timeout;
709 void *private;
710};
711
712struct md_io_acct {
713 struct bio *orig_bio;
714 unsigned long start_time;
715 struct bio bio_clone;
716};
717
718#define THREAD_WAKEUP 0
719
720static inline void safe_put_page(struct page *p)
721{
722 if (p) put_page(p);
723}
724
725extern int register_md_personality(struct md_personality *p);
726extern int unregister_md_personality(struct md_personality *p);
727extern int register_md_cluster_operations(struct md_cluster_operations *ops,
728 struct module *module);
729extern int unregister_md_cluster_operations(void);
730extern int md_setup_cluster(struct mddev *mddev, int nodes);
731extern void md_cluster_stop(struct mddev *mddev);
732extern struct md_thread *md_register_thread(
733 void (*run)(struct md_thread *thread),
734 struct mddev *mddev,
735 const char *name);
736extern void md_unregister_thread(struct md_thread **threadp);
737extern void md_wakeup_thread(struct md_thread *thread);
738extern void md_check_recovery(struct mddev *mddev);
739extern void md_reap_sync_thread(struct mddev *mddev);
740extern int mddev_init_writes_pending(struct mddev *mddev);
741extern bool md_write_start(struct mddev *mddev, struct bio *bi);
742extern void md_write_inc(struct mddev *mddev, struct bio *bi);
743extern void md_write_end(struct mddev *mddev);
744extern void md_done_sync(struct mddev *mddev, int blocks, int ok);
745extern void md_error(struct mddev *mddev, struct md_rdev *rdev);
746extern void md_finish_reshape(struct mddev *mddev);
747void md_submit_discard_bio(struct mddev *mddev, struct md_rdev *rdev,
748 struct bio *bio, sector_t start, sector_t size);
749int acct_bioset_init(struct mddev *mddev);
750void acct_bioset_exit(struct mddev *mddev);
751void md_account_bio(struct mddev *mddev, struct bio **bio);
752
753extern bool __must_check md_flush_request(struct mddev *mddev, struct bio *bio);
754extern void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
755 sector_t sector, int size, struct page *page);
756extern int md_super_wait(struct mddev *mddev);
757extern int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
758 struct page *page, blk_opf_t opf, bool metadata_op);
759extern void md_do_sync(struct md_thread *thread);
760extern void md_new_event(void);
761extern void md_allow_write(struct mddev *mddev);
762extern void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev);
763extern void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors);
764extern int md_check_no_bitmap(struct mddev *mddev);
765extern int md_integrity_register(struct mddev *mddev);
766extern int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev);
767extern int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale);
768
769extern void mddev_init(struct mddev *mddev);
770struct mddev *md_alloc(dev_t dev, char *name);
771void mddev_put(struct mddev *mddev);
772extern int md_run(struct mddev *mddev);
773extern int md_start(struct mddev *mddev);
774extern void md_stop(struct mddev *mddev);
775extern void md_stop_writes(struct mddev *mddev);
776extern int md_rdev_init(struct md_rdev *rdev);
777extern void md_rdev_clear(struct md_rdev *rdev);
778
779extern void md_handle_request(struct mddev *mddev, struct bio *bio);
780extern void mddev_suspend(struct mddev *mddev);
781extern void mddev_resume(struct mddev *mddev);
782
783extern void md_reload_sb(struct mddev *mddev, int raid_disk);
784extern void md_update_sb(struct mddev *mddev, int force);
785extern void mddev_create_serial_pool(struct mddev *mddev, struct md_rdev *rdev,
786 bool is_suspend);
787extern void mddev_destroy_serial_pool(struct mddev *mddev, struct md_rdev *rdev,
788 bool is_suspend);
789struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr);
790struct md_rdev *md_find_rdev_rcu(struct mddev *mddev, dev_t dev);
791
792static inline bool is_mddev_broken(struct md_rdev *rdev, const char *md_type)
793{
794 if (!disk_live(rdev->bdev->bd_disk)) {
795 if (!test_and_set_bit(MD_BROKEN, &rdev->mddev->flags))
796 pr_warn("md: %s: %s array has a missing/failed member\n",
797 mdname(rdev->mddev), md_type);
798 return true;
799 }
800 return false;
801}
802
803static inline void rdev_dec_pending(struct md_rdev *rdev, struct mddev *mddev)
804{
805 int faulty = test_bit(Faulty, &rdev->flags);
806 if (atomic_dec_and_test(&rdev->nr_pending) && faulty) {
807 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
808 md_wakeup_thread(mddev->thread);
809 }
810}
811
812extern struct md_cluster_operations *md_cluster_ops;
813static inline int mddev_is_clustered(struct mddev *mddev)
814{
815 return mddev->cluster_info && mddev->bitmap_info.nodes > 1;
816}
817
818/* clear unsupported mddev_flags */
819static inline void mddev_clear_unsupported_flags(struct mddev *mddev,
820 unsigned long unsupported_flags)
821{
822 mddev->flags &= ~unsupported_flags;
823}
824
825static inline void mddev_check_write_zeroes(struct mddev *mddev, struct bio *bio)
826{
827 if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
828 !bio->bi_bdev->bd_disk->queue->limits.max_write_zeroes_sectors)
829 mddev->queue->limits.max_write_zeroes_sectors = 0;
830}
831
832struct mdu_array_info_s;
833struct mdu_disk_info_s;
834
835extern int mdp_major;
836void md_autostart_arrays(int part);
837int md_set_array_info(struct mddev *mddev, struct mdu_array_info_s *info);
838int md_add_new_disk(struct mddev *mddev, struct mdu_disk_info_s *info);
839int do_md_run(struct mddev *mddev);
840
841extern const struct block_device_operations md_fops;
842
843#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 "md-cluster.h"
22
23#define MaxSector (~(sector_t)0)
24
25/*
26 * These flags should really be called "NO_RETRY" rather than
27 * "FAILFAST" because they don't make any promise about time lapse,
28 * only about the number of retries, which will be zero.
29 * REQ_FAILFAST_DRIVER is not included because
30 * Commit: 4a27446f3e39 ("[SCSI] modify scsi to handle new fail fast flags.")
31 * seems to suggest that the errors it avoids retrying should usually
32 * be retried.
33 */
34#define MD_FAILFAST (REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT)
35
36/*
37 * The struct embedded in rdev is used to serialize IO.
38 */
39struct serial_in_rdev {
40 struct rb_root_cached serial_rb;
41 spinlock_t serial_lock;
42 wait_queue_head_t serial_io_wait;
43};
44
45/*
46 * MD's 'extended' device
47 */
48struct md_rdev {
49 struct list_head same_set; /* RAID devices within the same set */
50
51 sector_t sectors; /* Device size (in 512bytes sectors) */
52 struct mddev *mddev; /* RAID array if running */
53 int last_events; /* IO event timestamp */
54
55 /*
56 * If meta_bdev is non-NULL, it means that a separate device is
57 * being used to store the metadata (superblock/bitmap) which
58 * would otherwise be contained on the same device as the data (bdev).
59 */
60 struct block_device *meta_bdev;
61 struct block_device *bdev; /* block device handle */
62 struct bdev_handle *bdev_handle; /* Handle from open for bdev */
63
64 struct page *sb_page, *bb_page;
65 int sb_loaded;
66 __u64 sb_events;
67 sector_t data_offset; /* start of data in array */
68 sector_t new_data_offset;/* only relevant while reshaping */
69 sector_t sb_start; /* offset of the super block (in 512byte sectors) */
70 int sb_size; /* bytes in the superblock */
71 int preferred_minor; /* autorun support */
72
73 struct kobject kobj;
74
75 /* A device can be in one of three states based on two flags:
76 * Not working: faulty==1 in_sync==0
77 * Fully working: faulty==0 in_sync==1
78 * Working, but not
79 * in sync with array
80 * faulty==0 in_sync==0
81 *
82 * It can never have faulty==1, in_sync==1
83 * This reduces the burden of testing multiple flags in many cases
84 */
85
86 unsigned long flags; /* bit set of 'enum flag_bits' bits. */
87 wait_queue_head_t blocked_wait;
88
89 int desc_nr; /* descriptor index in the superblock */
90 int raid_disk; /* role of device in array */
91 int new_raid_disk; /* role that the device will have in
92 * the array after a level-change completes.
93 */
94 int saved_raid_disk; /* role that device used to have in the
95 * array and could again if we did a partial
96 * resync from the bitmap
97 */
98 union {
99 sector_t recovery_offset;/* If this device has been partially
100 * recovered, this is where we were
101 * up to.
102 */
103 sector_t journal_tail; /* If this device is a journal device,
104 * this is the journal tail (journal
105 * recovery start point)
106 */
107 };
108
109 atomic_t nr_pending; /* number of pending requests.
110 * only maintained for arrays that
111 * support hot removal
112 */
113 atomic_t read_errors; /* number of consecutive read errors that
114 * we have tried to ignore.
115 */
116 time64_t last_read_error; /* monotonic time since our
117 * last read error
118 */
119 atomic_t corrected_errors; /* number of corrected read errors,
120 * for reporting to userspace and storing
121 * in superblock.
122 */
123
124 struct serial_in_rdev *serial; /* used for raid1 io serialization */
125
126 struct kernfs_node *sysfs_state; /* handle for 'state'
127 * sysfs entry */
128 /* handle for 'unacknowledged_bad_blocks' sysfs dentry */
129 struct kernfs_node *sysfs_unack_badblocks;
130 /* handle for 'bad_blocks' sysfs dentry */
131 struct kernfs_node *sysfs_badblocks;
132 struct badblocks badblocks;
133
134 struct {
135 short offset; /* Offset from superblock to start of PPL.
136 * Not used by external metadata. */
137 unsigned int size; /* Size in sectors of the PPL space */
138 sector_t sector; /* First sector of the PPL space */
139 } ppl;
140};
141enum flag_bits {
142 Faulty, /* device is known to have a fault */
143 In_sync, /* device is in_sync with rest of array */
144 Bitmap_sync, /* ..actually, not quite In_sync. Need a
145 * bitmap-based recovery to get fully in sync.
146 * The bit is only meaningful before device
147 * has been passed to pers->hot_add_disk.
148 */
149 WriteMostly, /* Avoid reading if at all possible */
150 AutoDetected, /* added by auto-detect */
151 Blocked, /* An error occurred but has not yet
152 * been acknowledged by the metadata
153 * handler, so don't allow writes
154 * until it is cleared */
155 WriteErrorSeen, /* A write error has been seen on this
156 * device
157 */
158 FaultRecorded, /* Intermediate state for clearing
159 * Blocked. The Fault is/will-be
160 * recorded in the metadata, but that
161 * metadata hasn't been stored safely
162 * on disk yet.
163 */
164 BlockedBadBlocks, /* A writer is blocked because they
165 * found an unacknowledged bad-block.
166 * This can safely be cleared at any
167 * time, and the writer will re-check.
168 * It may be set at any time, and at
169 * worst the writer will timeout and
170 * re-check. So setting it as
171 * accurately as possible is good, but
172 * not absolutely critical.
173 */
174 WantReplacement, /* This device is a candidate to be
175 * hot-replaced, either because it has
176 * reported some faults, or because
177 * of explicit request.
178 */
179 Replacement, /* This device is a replacement for
180 * a want_replacement device with same
181 * raid_disk number.
182 */
183 Candidate, /* For clustered environments only:
184 * This device is seen locally but not
185 * by the whole cluster
186 */
187 Journal, /* This device is used as journal for
188 * raid-5/6.
189 * Usually, this device should be faster
190 * than other devices in the array
191 */
192 ClusterRemove,
193 ExternalBbl, /* External metadata provides bad
194 * block management for a disk
195 */
196 FailFast, /* Minimal retries should be attempted on
197 * this device, so use REQ_FAILFAST_DEV.
198 * Also don't try to repair failed reads.
199 * It is expects that no bad block log
200 * is present.
201 */
202 LastDev, /* Seems to be the last working dev as
203 * it didn't fail, so don't use FailFast
204 * any more for metadata
205 */
206 CollisionCheck, /*
207 * check if there is collision between raid1
208 * serial bios.
209 */
210};
211
212static inline int is_badblock(struct md_rdev *rdev, sector_t s, int sectors,
213 sector_t *first_bad, int *bad_sectors)
214{
215 if (unlikely(rdev->badblocks.count)) {
216 int rv = badblocks_check(&rdev->badblocks, rdev->data_offset + s,
217 sectors,
218 first_bad, bad_sectors);
219 if (rv)
220 *first_bad -= rdev->data_offset;
221 return rv;
222 }
223 return 0;
224}
225extern int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
226 int is_new);
227extern int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
228 int is_new);
229struct md_cluster_info;
230
231/**
232 * enum mddev_flags - md device flags.
233 * @MD_ARRAY_FIRST_USE: First use of array, needs initialization.
234 * @MD_CLOSING: If set, we are closing the array, do not open it then.
235 * @MD_JOURNAL_CLEAN: A raid with journal is already clean.
236 * @MD_HAS_JOURNAL: The raid array has journal feature set.
237 * @MD_CLUSTER_RESYNC_LOCKED: cluster raid only, which means node, already took
238 * resync lock, need to release the lock.
239 * @MD_FAILFAST_SUPPORTED: Using MD_FAILFAST on metadata writes is supported as
240 * calls to md_error() will never cause the array to
241 * become failed.
242 * @MD_HAS_PPL: The raid array has PPL feature set.
243 * @MD_HAS_MULTIPLE_PPLS: The raid array has multiple PPLs feature set.
244 * @MD_NOT_READY: do_md_run() is active, so 'array_state', ust not report that
245 * array is ready yet.
246 * @MD_BROKEN: This is used to stop writes and mark array as failed.
247 * @MD_DELETED: This device is being deleted
248 *
249 * change UNSUPPORTED_MDDEV_FLAGS for each array type if new flag is added
250 */
251enum mddev_flags {
252 MD_ARRAY_FIRST_USE,
253 MD_CLOSING,
254 MD_JOURNAL_CLEAN,
255 MD_HAS_JOURNAL,
256 MD_CLUSTER_RESYNC_LOCKED,
257 MD_FAILFAST_SUPPORTED,
258 MD_HAS_PPL,
259 MD_HAS_MULTIPLE_PPLS,
260 MD_NOT_READY,
261 MD_BROKEN,
262 MD_DELETED,
263};
264
265enum mddev_sb_flags {
266 MD_SB_CHANGE_DEVS, /* Some device status has changed */
267 MD_SB_CHANGE_CLEAN, /* transition to or from 'clean' */
268 MD_SB_CHANGE_PENDING, /* switch from 'clean' to 'active' in progress */
269 MD_SB_NEED_REWRITE, /* metadata write needs to be repeated */
270};
271
272#define NR_SERIAL_INFOS 8
273/* record current range of serialize IOs */
274struct serial_info {
275 struct rb_node node;
276 sector_t start; /* start sector of rb node */
277 sector_t last; /* end sector of rb node */
278 sector_t _subtree_last; /* highest sector in subtree of rb node */
279};
280
281/*
282 * mddev->curr_resync stores the current sector of the resync but
283 * also has some overloaded values.
284 */
285enum {
286 /* No resync in progress */
287 MD_RESYNC_NONE = 0,
288 /* Yielded to allow another conflicting resync to commence */
289 MD_RESYNC_YIELDED = 1,
290 /* Delayed to check that there is no conflict with another sync */
291 MD_RESYNC_DELAYED = 2,
292 /* Any value greater than or equal to this is in an active resync */
293 MD_RESYNC_ACTIVE = 3,
294};
295
296struct mddev {
297 void *private;
298 struct md_personality *pers;
299 dev_t unit;
300 int md_minor;
301 struct list_head disks;
302 unsigned long flags;
303 unsigned long sb_flags;
304
305 int suspended;
306 struct mutex suspend_mutex;
307 struct percpu_ref active_io;
308 int ro;
309 int sysfs_active; /* set when sysfs deletes
310 * are happening, so run/
311 * takeover/stop are not safe
312 */
313 struct gendisk *gendisk;
314
315 struct kobject kobj;
316 int hold_active;
317#define UNTIL_IOCTL 1
318#define UNTIL_STOP 2
319
320 /* Superblock information */
321 int major_version,
322 minor_version,
323 patch_version;
324 int persistent;
325 int external; /* metadata is
326 * managed externally */
327 char metadata_type[17]; /* externally set*/
328 int chunk_sectors;
329 time64_t ctime, utime;
330 int level, layout;
331 char clevel[16];
332 int raid_disks;
333 int max_disks;
334 sector_t dev_sectors; /* used size of
335 * component devices */
336 sector_t array_sectors; /* exported array size */
337 int external_size; /* size managed
338 * externally */
339 __u64 events;
340 /* If the last 'event' was simply a clean->dirty transition, and
341 * we didn't write it to the spares, then it is safe and simple
342 * to just decrement the event count on a dirty->clean transition.
343 * So we record that possibility here.
344 */
345 int can_decrease_events;
346
347 char uuid[16];
348
349 /* If the array is being reshaped, we need to record the
350 * new shape and an indication of where we are up to.
351 * This is written to the superblock.
352 * If reshape_position is MaxSector, then no reshape is happening (yet).
353 */
354 sector_t reshape_position;
355 int delta_disks, new_level, new_layout;
356 int new_chunk_sectors;
357 int reshape_backwards;
358
359 struct md_thread __rcu *thread; /* management thread */
360 struct md_thread __rcu *sync_thread; /* doing resync or reconstruct */
361
362 /* 'last_sync_action' is initialized to "none". It is set when a
363 * sync operation (i.e "data-check", "requested-resync", "resync",
364 * "recovery", or "reshape") is started. It holds this value even
365 * when the sync thread is "frozen" (interrupted) or "idle" (stopped
366 * or finished). It is overwritten when a new sync operation is begun.
367 */
368 char *last_sync_action;
369 sector_t curr_resync; /* last block scheduled */
370 /* As resync requests can complete out of order, we cannot easily track
371 * how much resync has been completed. So we occasionally pause until
372 * everything completes, then set curr_resync_completed to curr_resync.
373 * As such it may be well behind the real resync mark, but it is a value
374 * we are certain of.
375 */
376 sector_t curr_resync_completed;
377 unsigned long resync_mark; /* a recent timestamp */
378 sector_t resync_mark_cnt;/* blocks written at resync_mark */
379 sector_t curr_mark_cnt; /* blocks scheduled now */
380
381 sector_t resync_max_sectors; /* may be set by personality */
382
383 atomic64_t resync_mismatches; /* count of sectors where
384 * parity/replica mismatch found
385 */
386
387 /* allow user-space to request suspension of IO to regions of the array */
388 sector_t suspend_lo;
389 sector_t suspend_hi;
390 /* if zero, use the system-wide default */
391 int sync_speed_min;
392 int sync_speed_max;
393
394 /* resync even though the same disks are shared among md-devices */
395 int parallel_resync;
396
397 int ok_start_degraded;
398
399 unsigned long recovery;
400 /* If a RAID personality determines that recovery (of a particular
401 * device) will fail due to a read error on the source device, it
402 * takes a copy of this number and does not attempt recovery again
403 * until this number changes.
404 */
405 int recovery_disabled;
406
407 int in_sync; /* know to not need resync */
408 /* 'open_mutex' avoids races between 'md_open' and 'do_md_stop', so
409 * that we are never stopping an array while it is open.
410 * 'reconfig_mutex' protects all other reconfiguration.
411 * These locks are separate due to conflicting interactions
412 * with disk->open_mutex.
413 * Lock ordering is:
414 * reconfig_mutex -> disk->open_mutex
415 * disk->open_mutex -> open_mutex: e.g. __blkdev_get -> md_open
416 */
417 struct mutex open_mutex;
418 struct mutex reconfig_mutex;
419 atomic_t active; /* general refcount */
420 atomic_t openers; /* number of active opens */
421
422 int changed; /* True if we might need to
423 * reread partition info */
424 int degraded; /* whether md should consider
425 * adding a spare
426 */
427
428 atomic_t recovery_active; /* blocks scheduled, but not written */
429 wait_queue_head_t recovery_wait;
430 sector_t recovery_cp;
431 sector_t resync_min; /* user requested sync
432 * starts here */
433 sector_t resync_max; /* resync should pause
434 * when it gets here */
435
436 struct kernfs_node *sysfs_state; /* handle for 'array_state'
437 * file in sysfs.
438 */
439 struct kernfs_node *sysfs_action; /* handle for 'sync_action' */
440 struct kernfs_node *sysfs_completed; /*handle for 'sync_completed' */
441 struct kernfs_node *sysfs_degraded; /*handle for 'degraded' */
442 struct kernfs_node *sysfs_level; /*handle for 'level' */
443
444 /* used for delayed sysfs removal */
445 struct work_struct del_work;
446 /* used for register new sync thread */
447 struct work_struct sync_work;
448
449 /* "lock" protects:
450 * flush_bio transition from NULL to !NULL
451 * rdev superblocks, events
452 * clearing MD_CHANGE_*
453 * in_sync - and related safemode and MD_CHANGE changes
454 * pers (also protected by reconfig_mutex and pending IO).
455 * clearing ->bitmap
456 * clearing ->bitmap_info.file
457 * changing ->resync_{min,max}
458 * setting MD_RECOVERY_RUNNING (which interacts with resync_{min,max})
459 */
460 spinlock_t lock;
461 wait_queue_head_t sb_wait; /* for waiting on superblock updates */
462 atomic_t pending_writes; /* number of active superblock writes */
463
464 unsigned int safemode; /* if set, update "clean" superblock
465 * when no writes pending.
466 */
467 unsigned int safemode_delay;
468 struct timer_list safemode_timer;
469 struct percpu_ref writes_pending;
470 int sync_checkers; /* # of threads checking writes_pending */
471 struct request_queue *queue; /* for plugging ... */
472
473 struct bitmap *bitmap; /* the bitmap for the device */
474 struct {
475 struct file *file; /* the bitmap file */
476 loff_t offset; /* offset from superblock of
477 * start of bitmap. May be
478 * negative, but not '0'
479 * For external metadata, offset
480 * from start of device.
481 */
482 unsigned long space; /* space available at this offset */
483 loff_t default_offset; /* this is the offset to use when
484 * hot-adding a bitmap. It should
485 * eventually be settable by sysfs.
486 */
487 unsigned long default_space; /* space available at
488 * default offset */
489 struct mutex mutex;
490 unsigned long chunksize;
491 unsigned long daemon_sleep; /* how many jiffies between updates? */
492 unsigned long max_write_behind; /* write-behind mode */
493 int external;
494 int nodes; /* Maximum number of nodes in the cluster */
495 char cluster_name[64]; /* Name of the cluster */
496 } bitmap_info;
497
498 atomic_t max_corr_read_errors; /* max read retries */
499 struct list_head all_mddevs;
500
501 const struct attribute_group *to_remove;
502
503 struct bio_set bio_set;
504 struct bio_set sync_set; /* for sync operations like
505 * metadata and bitmap writes
506 */
507 struct bio_set io_clone_set;
508
509 /* Generic flush handling.
510 * The last to finish preflush schedules a worker to submit
511 * the rest of the request (without the REQ_PREFLUSH flag).
512 */
513 struct bio *flush_bio;
514 atomic_t flush_pending;
515 ktime_t start_flush, prev_flush_start; /* prev_flush_start is when the previous completed
516 * flush was started.
517 */
518 struct work_struct flush_work;
519 struct work_struct event_work; /* used by dm to report failure event */
520 mempool_t *serial_info_pool;
521 void (*sync_super)(struct mddev *mddev, struct md_rdev *rdev);
522 struct md_cluster_info *cluster_info;
523 unsigned int good_device_nr; /* good device num within cluster raid */
524 unsigned int noio_flag; /* for memalloc scope API */
525
526 /*
527 * Temporarily store rdev that will be finally removed when
528 * reconfig_mutex is unlocked, protected by reconfig_mutex.
529 */
530 struct list_head deleting;
531
532 /* Used to synchronize idle and frozen for action_store() */
533 struct mutex sync_mutex;
534 /* The sequence number for sync thread */
535 atomic_t sync_seq;
536
537 bool has_superblocks:1;
538 bool fail_last_dev:1;
539 bool serialize_policy:1;
540};
541
542enum recovery_flags {
543 /*
544 * If neither SYNC or RESHAPE are set, then it is a recovery.
545 */
546 MD_RECOVERY_RUNNING, /* a thread is running, or about to be started */
547 MD_RECOVERY_SYNC, /* actually doing a resync, not a recovery */
548 MD_RECOVERY_RECOVER, /* doing recovery, or need to try it. */
549 MD_RECOVERY_INTR, /* resync needs to be aborted for some reason */
550 MD_RECOVERY_DONE, /* thread is done and is waiting to be reaped */
551 MD_RECOVERY_NEEDED, /* we might need to start a resync/recover */
552 MD_RECOVERY_REQUESTED, /* user-space has requested a sync (used with SYNC) */
553 MD_RECOVERY_CHECK, /* user-space request for check-only, no repair */
554 MD_RECOVERY_RESHAPE, /* A reshape is happening */
555 MD_RECOVERY_FROZEN, /* User request to abort, and not restart, any action */
556 MD_RECOVERY_ERROR, /* sync-action interrupted because io-error */
557 MD_RECOVERY_WAIT, /* waiting for pers->start() to finish */
558 MD_RESYNCING_REMOTE, /* remote node is running resync thread */
559};
560
561static inline int __must_check mddev_lock(struct mddev *mddev)
562{
563 return mutex_lock_interruptible(&mddev->reconfig_mutex);
564}
565
566/* Sometimes we need to take the lock in a situation where
567 * failure due to interrupts is not acceptable.
568 */
569static inline void mddev_lock_nointr(struct mddev *mddev)
570{
571 mutex_lock(&mddev->reconfig_mutex);
572}
573
574static inline int mddev_trylock(struct mddev *mddev)
575{
576 return mutex_trylock(&mddev->reconfig_mutex);
577}
578extern void mddev_unlock(struct mddev *mddev);
579
580static inline void md_sync_acct(struct block_device *bdev, unsigned long nr_sectors)
581{
582 atomic_add(nr_sectors, &bdev->bd_disk->sync_io);
583}
584
585static inline void md_sync_acct_bio(struct bio *bio, unsigned long nr_sectors)
586{
587 md_sync_acct(bio->bi_bdev, nr_sectors);
588}
589
590struct md_personality
591{
592 char *name;
593 int level;
594 struct list_head list;
595 struct module *owner;
596 bool __must_check (*make_request)(struct mddev *mddev, struct bio *bio);
597 /*
598 * start up works that do NOT require md_thread. tasks that
599 * requires md_thread should go into start()
600 */
601 int (*run)(struct mddev *mddev);
602 /* start up works that require md threads */
603 int (*start)(struct mddev *mddev);
604 void (*free)(struct mddev *mddev, void *priv);
605 void (*status)(struct seq_file *seq, struct mddev *mddev);
606 /* error_handler must set ->faulty and clear ->in_sync
607 * if appropriate, and should abort recovery if needed
608 */
609 void (*error_handler)(struct mddev *mddev, struct md_rdev *rdev);
610 int (*hot_add_disk) (struct mddev *mddev, struct md_rdev *rdev);
611 int (*hot_remove_disk) (struct mddev *mddev, struct md_rdev *rdev);
612 int (*spare_active) (struct mddev *mddev);
613 sector_t (*sync_request)(struct mddev *mddev, sector_t sector_nr, int *skipped);
614 int (*resize) (struct mddev *mddev, sector_t sectors);
615 sector_t (*size) (struct mddev *mddev, sector_t sectors, int raid_disks);
616 int (*check_reshape) (struct mddev *mddev);
617 int (*start_reshape) (struct mddev *mddev);
618 void (*finish_reshape) (struct mddev *mddev);
619 void (*update_reshape_pos) (struct mddev *mddev);
620 /* quiesce suspends or resumes internal processing.
621 * 1 - stop new actions and wait for action io to complete
622 * 0 - return to normal behaviour
623 */
624 void (*quiesce) (struct mddev *mddev, int quiesce);
625 /* takeover is used to transition an array from one
626 * personality to another. The new personality must be able
627 * to handle the data in the current layout.
628 * e.g. 2drive raid1 -> 2drive raid5
629 * ndrive raid5 -> degraded n+1drive raid6 with special layout
630 * If the takeover succeeds, a new 'private' structure is returned.
631 * This needs to be installed and then ->run used to activate the
632 * array.
633 */
634 void *(*takeover) (struct mddev *mddev);
635 /* Changes the consistency policy of an active array. */
636 int (*change_consistency_policy)(struct mddev *mddev, const char *buf);
637};
638
639struct md_sysfs_entry {
640 struct attribute attr;
641 ssize_t (*show)(struct mddev *, char *);
642 ssize_t (*store)(struct mddev *, const char *, size_t);
643};
644extern const struct attribute_group md_bitmap_group;
645
646static inline struct kernfs_node *sysfs_get_dirent_safe(struct kernfs_node *sd, char *name)
647{
648 if (sd)
649 return sysfs_get_dirent(sd, name);
650 return sd;
651}
652static inline void sysfs_notify_dirent_safe(struct kernfs_node *sd)
653{
654 if (sd)
655 sysfs_notify_dirent(sd);
656}
657
658static inline char * mdname (struct mddev * mddev)
659{
660 return mddev->gendisk ? mddev->gendisk->disk_name : "mdX";
661}
662
663static inline int sysfs_link_rdev(struct mddev *mddev, struct md_rdev *rdev)
664{
665 char nm[20];
666 if (!test_bit(Replacement, &rdev->flags) &&
667 !test_bit(Journal, &rdev->flags) &&
668 mddev->kobj.sd) {
669 sprintf(nm, "rd%d", rdev->raid_disk);
670 return sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
671 } else
672 return 0;
673}
674
675static inline void sysfs_unlink_rdev(struct mddev *mddev, struct md_rdev *rdev)
676{
677 char nm[20];
678 if (!test_bit(Replacement, &rdev->flags) &&
679 !test_bit(Journal, &rdev->flags) &&
680 mddev->kobj.sd) {
681 sprintf(nm, "rd%d", rdev->raid_disk);
682 sysfs_remove_link(&mddev->kobj, nm);
683 }
684}
685
686/*
687 * iterates through some rdev ringlist. It's safe to remove the
688 * current 'rdev'. Dont touch 'tmp' though.
689 */
690#define rdev_for_each_list(rdev, tmp, head) \
691 list_for_each_entry_safe(rdev, tmp, head, same_set)
692
693/*
694 * iterates through the 'same array disks' ringlist
695 */
696#define rdev_for_each(rdev, mddev) \
697 list_for_each_entry(rdev, &((mddev)->disks), same_set)
698
699#define rdev_for_each_safe(rdev, tmp, mddev) \
700 list_for_each_entry_safe(rdev, tmp, &((mddev)->disks), same_set)
701
702#define rdev_for_each_rcu(rdev, mddev) \
703 list_for_each_entry_rcu(rdev, &((mddev)->disks), same_set)
704
705struct md_thread {
706 void (*run) (struct md_thread *thread);
707 struct mddev *mddev;
708 wait_queue_head_t wqueue;
709 unsigned long flags;
710 struct task_struct *tsk;
711 unsigned long timeout;
712 void *private;
713};
714
715struct md_io_clone {
716 struct mddev *mddev;
717 struct bio *orig_bio;
718 unsigned long start_time;
719 struct bio bio_clone;
720};
721
722#define THREAD_WAKEUP 0
723
724static inline void safe_put_page(struct page *p)
725{
726 if (p) put_page(p);
727}
728
729extern int register_md_personality(struct md_personality *p);
730extern int unregister_md_personality(struct md_personality *p);
731extern int register_md_cluster_operations(struct md_cluster_operations *ops,
732 struct module *module);
733extern int unregister_md_cluster_operations(void);
734extern int md_setup_cluster(struct mddev *mddev, int nodes);
735extern void md_cluster_stop(struct mddev *mddev);
736extern struct md_thread *md_register_thread(
737 void (*run)(struct md_thread *thread),
738 struct mddev *mddev,
739 const char *name);
740extern void md_unregister_thread(struct mddev *mddev, struct md_thread __rcu **threadp);
741extern void md_wakeup_thread(struct md_thread __rcu *thread);
742extern void md_check_recovery(struct mddev *mddev);
743extern void md_reap_sync_thread(struct mddev *mddev);
744extern bool md_write_start(struct mddev *mddev, struct bio *bi);
745extern void md_write_inc(struct mddev *mddev, struct bio *bi);
746extern void md_write_end(struct mddev *mddev);
747extern void md_done_sync(struct mddev *mddev, int blocks, int ok);
748extern void md_error(struct mddev *mddev, struct md_rdev *rdev);
749extern void md_finish_reshape(struct mddev *mddev);
750void md_submit_discard_bio(struct mddev *mddev, struct md_rdev *rdev,
751 struct bio *bio, sector_t start, sector_t size);
752void md_account_bio(struct mddev *mddev, struct bio **bio);
753
754extern bool __must_check md_flush_request(struct mddev *mddev, struct bio *bio);
755extern void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
756 sector_t sector, int size, struct page *page);
757extern int md_super_wait(struct mddev *mddev);
758extern int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
759 struct page *page, blk_opf_t opf, bool metadata_op);
760extern void md_do_sync(struct md_thread *thread);
761extern void md_new_event(void);
762extern void md_allow_write(struct mddev *mddev);
763extern void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev);
764extern void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors);
765extern int md_check_no_bitmap(struct mddev *mddev);
766extern int md_integrity_register(struct mddev *mddev);
767extern int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev);
768extern int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale);
769
770extern int mddev_init(struct mddev *mddev);
771extern void mddev_destroy(struct mddev *mddev);
772struct mddev *md_alloc(dev_t dev, char *name);
773void mddev_put(struct mddev *mddev);
774extern int md_run(struct mddev *mddev);
775extern int md_start(struct mddev *mddev);
776extern void md_stop(struct mddev *mddev);
777extern void md_stop_writes(struct mddev *mddev);
778extern int md_rdev_init(struct md_rdev *rdev);
779extern void md_rdev_clear(struct md_rdev *rdev);
780
781extern void md_handle_request(struct mddev *mddev, struct bio *bio);
782extern int mddev_suspend(struct mddev *mddev, bool interruptible);
783extern void mddev_resume(struct mddev *mddev);
784
785extern void md_reload_sb(struct mddev *mddev, int raid_disk);
786extern void md_update_sb(struct mddev *mddev, int force);
787extern void mddev_create_serial_pool(struct mddev *mddev, struct md_rdev *rdev);
788extern void mddev_destroy_serial_pool(struct mddev *mddev,
789 struct md_rdev *rdev);
790struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr);
791struct md_rdev *md_find_rdev_rcu(struct mddev *mddev, dev_t dev);
792
793static inline bool is_rdev_broken(struct md_rdev *rdev)
794{
795 return !disk_live(rdev->bdev->bd_disk);
796}
797
798static inline void rdev_dec_pending(struct md_rdev *rdev, struct mddev *mddev)
799{
800 int faulty = test_bit(Faulty, &rdev->flags);
801 if (atomic_dec_and_test(&rdev->nr_pending) && faulty) {
802 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
803 md_wakeup_thread(mddev->thread);
804 }
805}
806
807extern struct md_cluster_operations *md_cluster_ops;
808static inline int mddev_is_clustered(struct mddev *mddev)
809{
810 return mddev->cluster_info && mddev->bitmap_info.nodes > 1;
811}
812
813/* clear unsupported mddev_flags */
814static inline void mddev_clear_unsupported_flags(struct mddev *mddev,
815 unsigned long unsupported_flags)
816{
817 mddev->flags &= ~unsupported_flags;
818}
819
820static inline void mddev_check_write_zeroes(struct mddev *mddev, struct bio *bio)
821{
822 if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
823 !bio->bi_bdev->bd_disk->queue->limits.max_write_zeroes_sectors)
824 mddev->queue->limits.max_write_zeroes_sectors = 0;
825}
826
827static inline int mddev_suspend_and_lock(struct mddev *mddev)
828{
829 int ret;
830
831 ret = mddev_suspend(mddev, true);
832 if (ret)
833 return ret;
834
835 ret = mddev_lock(mddev);
836 if (ret)
837 mddev_resume(mddev);
838
839 return ret;
840}
841
842static inline void mddev_suspend_and_lock_nointr(struct mddev *mddev)
843{
844 mddev_suspend(mddev, false);
845 mutex_lock(&mddev->reconfig_mutex);
846}
847
848static inline void mddev_unlock_and_resume(struct mddev *mddev)
849{
850 mddev_unlock(mddev);
851 mddev_resume(mddev);
852}
853
854struct mdu_array_info_s;
855struct mdu_disk_info_s;
856
857extern int mdp_major;
858extern struct workqueue_struct *md_bitmap_wq;
859void md_autostart_arrays(int part);
860int md_set_array_info(struct mddev *mddev, struct mdu_array_info_s *info);
861int md_add_new_disk(struct mddev *mddev, struct mdu_disk_info_s *info);
862int do_md_run(struct mddev *mddev);
863
864extern const struct block_device_operations md_fops;
865
866#endif /* _MD_MD_H */