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