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 <trace/events/block.h>
22#include "md-cluster.h"
23
24#define MaxSector (~(sector_t)0)
25
26/*
27 * These flags should really be called "NO_RETRY" rather than
28 * "FAILFAST" because they don't make any promise about time lapse,
29 * only about the number of retries, which will be zero.
30 * REQ_FAILFAST_DRIVER is not included because
31 * Commit: 4a27446f3e39 ("[SCSI] modify scsi to handle new fail fast flags.")
32 * seems to suggest that the errors it avoids retrying should usually
33 * be retried.
34 */
35#define MD_FAILFAST (REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT)
36
37/* Status of sync thread. */
38enum sync_action {
39 /*
40 * Represent by MD_RECOVERY_SYNC, start when:
41 * 1) after assemble, sync data from first rdev to other copies, this
42 * must be done first before other sync actions and will only execute
43 * once;
44 * 2) resize the array(notice that this is not reshape), sync data for
45 * the new range;
46 */
47 ACTION_RESYNC,
48 /*
49 * Represent by MD_RECOVERY_RECOVER, start when:
50 * 1) for new replacement, sync data based on the replace rdev or
51 * available copies from other rdev;
52 * 2) for new member disk while the array is degraded, sync data from
53 * other rdev;
54 * 3) reassemble after power failure or re-add a hot removed rdev, sync
55 * data from first rdev to other copies based on bitmap;
56 */
57 ACTION_RECOVER,
58 /*
59 * Represent by MD_RECOVERY_SYNC | MD_RECOVERY_REQUESTED |
60 * MD_RECOVERY_CHECK, start when user echo "check" to sysfs api
61 * sync_action, used to check if data copies from differenct rdev are
62 * the same. The number of mismatch sectors will be exported to user
63 * by sysfs api mismatch_cnt;
64 */
65 ACTION_CHECK,
66 /*
67 * Represent by MD_RECOVERY_SYNC | MD_RECOVERY_REQUESTED, start when
68 * user echo "repair" to sysfs api sync_action, usually paired with
69 * ACTION_CHECK, used to force syncing data once user found that there
70 * are inconsistent data,
71 */
72 ACTION_REPAIR,
73 /*
74 * Represent by MD_RECOVERY_RESHAPE, start when new member disk is added
75 * to the conf, notice that this is different from spares or
76 * replacement;
77 */
78 ACTION_RESHAPE,
79 /*
80 * Represent by MD_RECOVERY_FROZEN, can be set by sysfs api sync_action
81 * or internal usage like setting the array read-only, will forbid above
82 * actions.
83 */
84 ACTION_FROZEN,
85 /*
86 * All above actions don't match.
87 */
88 ACTION_IDLE,
89 NR_SYNC_ACTIONS,
90};
91
92/*
93 * The struct embedded in rdev is used to serialize IO.
94 */
95struct serial_in_rdev {
96 struct rb_root_cached serial_rb;
97 spinlock_t serial_lock;
98 wait_queue_head_t serial_io_wait;
99};
100
101/*
102 * MD's 'extended' device
103 */
104struct md_rdev {
105 struct list_head same_set; /* RAID devices within the same set */
106
107 sector_t sectors; /* Device size (in 512bytes sectors) */
108 struct mddev *mddev; /* RAID array if running */
109 int last_events; /* IO event timestamp */
110
111 /*
112 * If meta_bdev is non-NULL, it means that a separate device is
113 * being used to store the metadata (superblock/bitmap) which
114 * would otherwise be contained on the same device as the data (bdev).
115 */
116 struct block_device *meta_bdev;
117 struct block_device *bdev; /* block device handle */
118 struct file *bdev_file; /* Handle from open for bdev */
119
120 struct page *sb_page, *bb_page;
121 int sb_loaded;
122 __u64 sb_events;
123 sector_t data_offset; /* start of data in array */
124 sector_t new_data_offset;/* only relevant while reshaping */
125 sector_t sb_start; /* offset of the super block (in 512byte sectors) */
126 int sb_size; /* bytes in the superblock */
127 int preferred_minor; /* autorun support */
128
129 struct kobject kobj;
130
131 /* A device can be in one of three states based on two flags:
132 * Not working: faulty==1 in_sync==0
133 * Fully working: faulty==0 in_sync==1
134 * Working, but not
135 * in sync with array
136 * faulty==0 in_sync==0
137 *
138 * It can never have faulty==1, in_sync==1
139 * This reduces the burden of testing multiple flags in many cases
140 */
141
142 unsigned long flags; /* bit set of 'enum flag_bits' bits. */
143 wait_queue_head_t blocked_wait;
144
145 int desc_nr; /* descriptor index in the superblock */
146 int raid_disk; /* role of device in array */
147 int new_raid_disk; /* role that the device will have in
148 * the array after a level-change completes.
149 */
150 int saved_raid_disk; /* role that device used to have in the
151 * array and could again if we did a partial
152 * resync from the bitmap
153 */
154 union {
155 sector_t recovery_offset;/* If this device has been partially
156 * recovered, this is where we were
157 * up to.
158 */
159 sector_t journal_tail; /* If this device is a journal device,
160 * this is the journal tail (journal
161 * recovery start point)
162 */
163 };
164
165 atomic_t nr_pending; /* number of pending requests.
166 * only maintained for arrays that
167 * support hot removal
168 */
169 atomic_t read_errors; /* number of consecutive read errors that
170 * we have tried to ignore.
171 */
172 time64_t last_read_error; /* monotonic time since our
173 * last read error
174 */
175 atomic_t corrected_errors; /* number of corrected read errors,
176 * for reporting to userspace and storing
177 * in superblock.
178 */
179
180 struct serial_in_rdev *serial; /* used for raid1 io serialization */
181
182 struct kernfs_node *sysfs_state; /* handle for 'state'
183 * sysfs entry */
184 /* handle for 'unacknowledged_bad_blocks' sysfs dentry */
185 struct kernfs_node *sysfs_unack_badblocks;
186 /* handle for 'bad_blocks' sysfs dentry */
187 struct kernfs_node *sysfs_badblocks;
188 struct badblocks badblocks;
189
190 struct {
191 short offset; /* Offset from superblock to start of PPL.
192 * Not used by external metadata. */
193 unsigned int size; /* Size in sectors of the PPL space */
194 sector_t sector; /* First sector of the PPL space */
195 } ppl;
196};
197enum flag_bits {
198 Faulty, /* device is known to have a fault */
199 In_sync, /* device is in_sync with rest of array */
200 Bitmap_sync, /* ..actually, not quite In_sync. Need a
201 * bitmap-based recovery to get fully in sync.
202 * The bit is only meaningful before device
203 * has been passed to pers->hot_add_disk.
204 */
205 WriteMostly, /* Avoid reading if at all possible */
206 AutoDetected, /* added by auto-detect */
207 Blocked, /* An error occurred but has not yet
208 * been acknowledged by the metadata
209 * handler, so don't allow writes
210 * until it is cleared */
211 WriteErrorSeen, /* A write error has been seen on this
212 * device
213 */
214 FaultRecorded, /* Intermediate state for clearing
215 * Blocked. The Fault is/will-be
216 * recorded in the metadata, but that
217 * metadata hasn't been stored safely
218 * on disk yet.
219 */
220 BlockedBadBlocks, /* A writer is blocked because they
221 * found an unacknowledged bad-block.
222 * This can safely be cleared at any
223 * time, and the writer will re-check.
224 * It may be set at any time, and at
225 * worst the writer will timeout and
226 * re-check. So setting it as
227 * accurately as possible is good, but
228 * not absolutely critical.
229 */
230 WantReplacement, /* This device is a candidate to be
231 * hot-replaced, either because it has
232 * reported some faults, or because
233 * of explicit request.
234 */
235 Replacement, /* This device is a replacement for
236 * a want_replacement device with same
237 * raid_disk number.
238 */
239 Candidate, /* For clustered environments only:
240 * This device is seen locally but not
241 * by the whole cluster
242 */
243 Journal, /* This device is used as journal for
244 * raid-5/6.
245 * Usually, this device should be faster
246 * than other devices in the array
247 */
248 ClusterRemove,
249 ExternalBbl, /* External metadata provides bad
250 * block management for a disk
251 */
252 FailFast, /* Minimal retries should be attempted on
253 * this device, so use REQ_FAILFAST_DEV.
254 * Also don't try to repair failed reads.
255 * It is expects that no bad block log
256 * is present.
257 */
258 LastDev, /* Seems to be the last working dev as
259 * it didn't fail, so don't use FailFast
260 * any more for metadata
261 */
262 CollisionCheck, /*
263 * check if there is collision between raid1
264 * serial bios.
265 */
266 Nonrot, /* non-rotational device (SSD) */
267};
268
269static inline int is_badblock(struct md_rdev *rdev, sector_t s, int sectors,
270 sector_t *first_bad, int *bad_sectors)
271{
272 if (unlikely(rdev->badblocks.count)) {
273 int rv = badblocks_check(&rdev->badblocks, rdev->data_offset + s,
274 sectors,
275 first_bad, bad_sectors);
276 if (rv)
277 *first_bad -= rdev->data_offset;
278 return rv;
279 }
280 return 0;
281}
282
283static inline int rdev_has_badblock(struct md_rdev *rdev, sector_t s,
284 int sectors)
285{
286 sector_t first_bad;
287 int bad_sectors;
288
289 return is_badblock(rdev, s, sectors, &first_bad, &bad_sectors);
290}
291
292extern int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
293 int is_new);
294extern int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
295 int is_new);
296struct md_cluster_info;
297
298/**
299 * enum mddev_flags - md device flags.
300 * @MD_ARRAY_FIRST_USE: First use of array, needs initialization.
301 * @MD_CLOSING: If set, we are closing the array, do not open it then.
302 * @MD_JOURNAL_CLEAN: A raid with journal is already clean.
303 * @MD_HAS_JOURNAL: The raid array has journal feature set.
304 * @MD_CLUSTER_RESYNC_LOCKED: cluster raid only, which means node, already took
305 * resync lock, need to release the lock.
306 * @MD_FAILFAST_SUPPORTED: Using MD_FAILFAST on metadata writes is supported as
307 * calls to md_error() will never cause the array to
308 * become failed.
309 * @MD_HAS_PPL: The raid array has PPL feature set.
310 * @MD_HAS_MULTIPLE_PPLS: The raid array has multiple PPLs feature set.
311 * @MD_NOT_READY: do_md_run() is active, so 'array_state', ust not report that
312 * array is ready yet.
313 * @MD_BROKEN: This is used to stop writes and mark array as failed.
314 * @MD_DELETED: This device is being deleted
315 *
316 * change UNSUPPORTED_MDDEV_FLAGS for each array type if new flag is added
317 */
318enum mddev_flags {
319 MD_ARRAY_FIRST_USE,
320 MD_CLOSING,
321 MD_JOURNAL_CLEAN,
322 MD_HAS_JOURNAL,
323 MD_CLUSTER_RESYNC_LOCKED,
324 MD_FAILFAST_SUPPORTED,
325 MD_HAS_PPL,
326 MD_HAS_MULTIPLE_PPLS,
327 MD_NOT_READY,
328 MD_BROKEN,
329 MD_DELETED,
330};
331
332enum mddev_sb_flags {
333 MD_SB_CHANGE_DEVS, /* Some device status has changed */
334 MD_SB_CHANGE_CLEAN, /* transition to or from 'clean' */
335 MD_SB_CHANGE_PENDING, /* switch from 'clean' to 'active' in progress */
336 MD_SB_NEED_REWRITE, /* metadata write needs to be repeated */
337};
338
339#define NR_SERIAL_INFOS 8
340/* record current range of serialize IOs */
341struct serial_info {
342 struct rb_node node;
343 sector_t start; /* start sector of rb node */
344 sector_t last; /* end sector of rb node */
345 sector_t _subtree_last; /* highest sector in subtree of rb node */
346};
347
348/*
349 * mddev->curr_resync stores the current sector of the resync but
350 * also has some overloaded values.
351 */
352enum {
353 /* No resync in progress */
354 MD_RESYNC_NONE = 0,
355 /* Yielded to allow another conflicting resync to commence */
356 MD_RESYNC_YIELDED = 1,
357 /* Delayed to check that there is no conflict with another sync */
358 MD_RESYNC_DELAYED = 2,
359 /* Any value greater than or equal to this is in an active resync */
360 MD_RESYNC_ACTIVE = 3,
361};
362
363struct mddev {
364 void *private;
365 struct md_personality *pers;
366 dev_t unit;
367 int md_minor;
368 struct list_head disks;
369 unsigned long flags;
370 unsigned long sb_flags;
371
372 int suspended;
373 struct mutex suspend_mutex;
374 struct percpu_ref active_io;
375 int ro;
376 int sysfs_active; /* set when sysfs deletes
377 * are happening, so run/
378 * takeover/stop are not safe
379 */
380 struct gendisk *gendisk;
381
382 struct kobject kobj;
383 int hold_active;
384#define UNTIL_IOCTL 1
385#define UNTIL_STOP 2
386
387 /* Superblock information */
388 int major_version,
389 minor_version,
390 patch_version;
391 int persistent;
392 int external; /* metadata is
393 * managed externally */
394 char metadata_type[17]; /* externally set*/
395 int chunk_sectors;
396 time64_t ctime, utime;
397 int level, layout;
398 char clevel[16];
399 int raid_disks;
400 int max_disks;
401 sector_t dev_sectors; /* used size of
402 * component devices */
403 sector_t array_sectors; /* exported array size */
404 int external_size; /* size managed
405 * externally */
406 __u64 events;
407 /* If the last 'event' was simply a clean->dirty transition, and
408 * we didn't write it to the spares, then it is safe and simple
409 * to just decrement the event count on a dirty->clean transition.
410 * So we record that possibility here.
411 */
412 int can_decrease_events;
413
414 char uuid[16];
415
416 /* If the array is being reshaped, we need to record the
417 * new shape and an indication of where we are up to.
418 * This is written to the superblock.
419 * If reshape_position is MaxSector, then no reshape is happening (yet).
420 */
421 sector_t reshape_position;
422 int delta_disks, new_level, new_layout;
423 int new_chunk_sectors;
424 int reshape_backwards;
425
426 struct md_thread __rcu *thread; /* management thread */
427 struct md_thread __rcu *sync_thread; /* doing resync or reconstruct */
428
429 /*
430 * Set when a sync operation is started. It holds this value even
431 * when the sync thread is "frozen" (interrupted) or "idle" (stopped
432 * or finished). It is overwritten when a new sync operation is begun.
433 */
434 enum sync_action last_sync_action;
435 sector_t curr_resync; /* last block scheduled */
436 /* As resync requests can complete out of order, we cannot easily track
437 * how much resync has been completed. So we occasionally pause until
438 * everything completes, then set curr_resync_completed to curr_resync.
439 * As such it may be well behind the real resync mark, but it is a value
440 * we are certain of.
441 */
442 sector_t curr_resync_completed;
443 unsigned long resync_mark; /* a recent timestamp */
444 sector_t resync_mark_cnt;/* blocks written at resync_mark */
445 sector_t curr_mark_cnt; /* blocks scheduled now */
446
447 sector_t resync_max_sectors; /* may be set by personality */
448
449 atomic64_t resync_mismatches; /* count of sectors where
450 * parity/replica mismatch found
451 */
452
453 /* allow user-space to request suspension of IO to regions of the array */
454 sector_t suspend_lo;
455 sector_t suspend_hi;
456 /* if zero, use the system-wide default */
457 int sync_speed_min;
458 int sync_speed_max;
459
460 /* resync even though the same disks are shared among md-devices */
461 int parallel_resync;
462
463 int ok_start_degraded;
464
465 unsigned long recovery;
466 /* If a RAID personality determines that recovery (of a particular
467 * device) will fail due to a read error on the source device, it
468 * takes a copy of this number and does not attempt recovery again
469 * until this number changes.
470 */
471 int recovery_disabled;
472
473 int in_sync; /* know to not need resync */
474 /* 'open_mutex' avoids races between 'md_open' and 'do_md_stop', so
475 * that we are never stopping an array while it is open.
476 * 'reconfig_mutex' protects all other reconfiguration.
477 * These locks are separate due to conflicting interactions
478 * with disk->open_mutex.
479 * Lock ordering is:
480 * reconfig_mutex -> disk->open_mutex
481 * disk->open_mutex -> open_mutex: e.g. __blkdev_get -> md_open
482 */
483 struct mutex open_mutex;
484 struct mutex reconfig_mutex;
485 atomic_t active; /* general refcount */
486 atomic_t openers; /* number of active opens */
487
488 int changed; /* True if we might need to
489 * reread partition info */
490 int degraded; /* whether md should consider
491 * adding a spare
492 */
493
494 atomic_t recovery_active; /* blocks scheduled, but not written */
495 wait_queue_head_t recovery_wait;
496 sector_t recovery_cp;
497 sector_t resync_min; /* user requested sync
498 * starts here */
499 sector_t resync_max; /* resync should pause
500 * when it gets here */
501
502 struct kernfs_node *sysfs_state; /* handle for 'array_state'
503 * file in sysfs.
504 */
505 struct kernfs_node *sysfs_action; /* handle for 'sync_action' */
506 struct kernfs_node *sysfs_completed; /*handle for 'sync_completed' */
507 struct kernfs_node *sysfs_degraded; /*handle for 'degraded' */
508 struct kernfs_node *sysfs_level; /*handle for 'level' */
509
510 /* used for delayed sysfs removal */
511 struct work_struct del_work;
512 /* used for register new sync thread */
513 struct work_struct sync_work;
514
515 /* "lock" protects:
516 * flush_bio transition from NULL to !NULL
517 * rdev superblocks, events
518 * clearing MD_CHANGE_*
519 * in_sync - and related safemode and MD_CHANGE changes
520 * pers (also protected by reconfig_mutex and pending IO).
521 * clearing ->bitmap
522 * clearing ->bitmap_info.file
523 * changing ->resync_{min,max}
524 * setting MD_RECOVERY_RUNNING (which interacts with resync_{min,max})
525 */
526 spinlock_t lock;
527 wait_queue_head_t sb_wait; /* for waiting on superblock updates */
528 atomic_t pending_writes; /* number of active superblock writes */
529
530 unsigned int safemode; /* if set, update "clean" superblock
531 * when no writes pending.
532 */
533 unsigned int safemode_delay;
534 struct timer_list safemode_timer;
535 struct percpu_ref writes_pending;
536 int sync_checkers; /* # of threads checking writes_pending */
537
538 void *bitmap; /* the bitmap for the device */
539 struct bitmap_operations *bitmap_ops;
540 struct {
541 struct file *file; /* the bitmap file */
542 loff_t offset; /* offset from superblock of
543 * start of bitmap. May be
544 * negative, but not '0'
545 * For external metadata, offset
546 * from start of device.
547 */
548 unsigned long space; /* space available at this offset */
549 loff_t default_offset; /* this is the offset to use when
550 * hot-adding a bitmap. It should
551 * eventually be settable by sysfs.
552 */
553 unsigned long default_space; /* space available at
554 * default offset */
555 struct mutex mutex;
556 unsigned long chunksize;
557 unsigned long daemon_sleep; /* how many jiffies between updates? */
558 unsigned long max_write_behind; /* write-behind mode */
559 int external;
560 int nodes; /* Maximum number of nodes in the cluster */
561 char cluster_name[64]; /* Name of the cluster */
562 } bitmap_info;
563
564 atomic_t max_corr_read_errors; /* max read retries */
565 struct list_head all_mddevs;
566
567 const struct attribute_group *to_remove;
568
569 struct bio_set bio_set;
570 struct bio_set sync_set; /* for sync operations like
571 * metadata and bitmap writes
572 */
573 struct bio_set io_clone_set;
574
575 struct work_struct event_work; /* used by dm to report failure event */
576 mempool_t *serial_info_pool;
577 void (*sync_super)(struct mddev *mddev, struct md_rdev *rdev);
578 struct md_cluster_info *cluster_info;
579 unsigned int good_device_nr; /* good device num within cluster raid */
580 unsigned int noio_flag; /* for memalloc scope API */
581
582 /*
583 * Temporarily store rdev that will be finally removed when
584 * reconfig_mutex is unlocked, protected by reconfig_mutex.
585 */
586 struct list_head deleting;
587
588 /* The sequence number for sync thread */
589 atomic_t sync_seq;
590
591 bool has_superblocks:1;
592 bool fail_last_dev:1;
593 bool serialize_policy:1;
594};
595
596enum recovery_flags {
597 /* flags for sync thread running status */
598
599 /*
600 * set when one of sync action is set and new sync thread need to be
601 * registered, or just add/remove spares from conf.
602 */
603 MD_RECOVERY_NEEDED,
604 /* sync thread is running, or about to be started */
605 MD_RECOVERY_RUNNING,
606 /* sync thread needs to be aborted for some reason */
607 MD_RECOVERY_INTR,
608 /* sync thread is done and is waiting to be unregistered */
609 MD_RECOVERY_DONE,
610 /* running sync thread must abort immediately, and not restart */
611 MD_RECOVERY_FROZEN,
612 /* waiting for pers->start() to finish */
613 MD_RECOVERY_WAIT,
614 /* interrupted because io-error */
615 MD_RECOVERY_ERROR,
616
617 /* flags determines sync action, see details in enum sync_action */
618
619 /* if just this flag is set, action is resync. */
620 MD_RECOVERY_SYNC,
621 /*
622 * paired with MD_RECOVERY_SYNC, if MD_RECOVERY_CHECK is not set,
623 * action is repair, means user requested resync.
624 */
625 MD_RECOVERY_REQUESTED,
626 /*
627 * paired with MD_RECOVERY_SYNC and MD_RECOVERY_REQUESTED, action is
628 * check.
629 */
630 MD_RECOVERY_CHECK,
631 /* recovery, or need to try it */
632 MD_RECOVERY_RECOVER,
633 /* reshape */
634 MD_RECOVERY_RESHAPE,
635 /* remote node is running resync thread */
636 MD_RESYNCING_REMOTE,
637};
638
639enum md_ro_state {
640 MD_RDWR,
641 MD_RDONLY,
642 MD_AUTO_READ,
643 MD_MAX_STATE
644};
645
646static inline bool md_is_rdwr(struct mddev *mddev)
647{
648 return (mddev->ro == MD_RDWR);
649}
650
651static inline bool reshape_interrupted(struct mddev *mddev)
652{
653 /* reshape never start */
654 if (mddev->reshape_position == MaxSector)
655 return false;
656
657 /* interrupted */
658 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
659 return true;
660
661 /* running reshape will be interrupted soon. */
662 if (test_bit(MD_RECOVERY_WAIT, &mddev->recovery) ||
663 test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
664 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
665 return true;
666
667 return false;
668}
669
670static inline int __must_check mddev_lock(struct mddev *mddev)
671{
672 return mutex_lock_interruptible(&mddev->reconfig_mutex);
673}
674
675/* Sometimes we need to take the lock in a situation where
676 * failure due to interrupts is not acceptable.
677 */
678static inline void mddev_lock_nointr(struct mddev *mddev)
679{
680 mutex_lock(&mddev->reconfig_mutex);
681}
682
683static inline int mddev_trylock(struct mddev *mddev)
684{
685 return mutex_trylock(&mddev->reconfig_mutex);
686}
687extern void mddev_unlock(struct mddev *mddev);
688
689static inline void md_sync_acct(struct block_device *bdev, unsigned long nr_sectors)
690{
691 if (blk_queue_io_stat(bdev->bd_disk->queue))
692 atomic_add(nr_sectors, &bdev->bd_disk->sync_io);
693}
694
695static inline void md_sync_acct_bio(struct bio *bio, unsigned long nr_sectors)
696{
697 md_sync_acct(bio->bi_bdev, nr_sectors);
698}
699
700struct md_personality
701{
702 char *name;
703 int level;
704 struct list_head list;
705 struct module *owner;
706 bool __must_check (*make_request)(struct mddev *mddev, struct bio *bio);
707 /*
708 * start up works that do NOT require md_thread. tasks that
709 * requires md_thread should go into start()
710 */
711 int (*run)(struct mddev *mddev);
712 /* start up works that require md threads */
713 int (*start)(struct mddev *mddev);
714 void (*free)(struct mddev *mddev, void *priv);
715 void (*status)(struct seq_file *seq, struct mddev *mddev);
716 /* error_handler must set ->faulty and clear ->in_sync
717 * if appropriate, and should abort recovery if needed
718 */
719 void (*error_handler)(struct mddev *mddev, struct md_rdev *rdev);
720 int (*hot_add_disk) (struct mddev *mddev, struct md_rdev *rdev);
721 int (*hot_remove_disk) (struct mddev *mddev, struct md_rdev *rdev);
722 int (*spare_active) (struct mddev *mddev);
723 sector_t (*sync_request)(struct mddev *mddev, sector_t sector_nr,
724 sector_t max_sector, int *skipped);
725 int (*resize) (struct mddev *mddev, sector_t sectors);
726 sector_t (*size) (struct mddev *mddev, sector_t sectors, int raid_disks);
727 int (*check_reshape) (struct mddev *mddev);
728 int (*start_reshape) (struct mddev *mddev);
729 void (*finish_reshape) (struct mddev *mddev);
730 void (*update_reshape_pos) (struct mddev *mddev);
731 void (*prepare_suspend) (struct mddev *mddev);
732 /* quiesce suspends or resumes internal processing.
733 * 1 - stop new actions and wait for action io to complete
734 * 0 - return to normal behaviour
735 */
736 void (*quiesce) (struct mddev *mddev, int quiesce);
737 /* takeover is used to transition an array from one
738 * personality to another. The new personality must be able
739 * to handle the data in the current layout.
740 * e.g. 2drive raid1 -> 2drive raid5
741 * ndrive raid5 -> degraded n+1drive raid6 with special layout
742 * If the takeover succeeds, a new 'private' structure is returned.
743 * This needs to be installed and then ->run used to activate the
744 * array.
745 */
746 void *(*takeover) (struct mddev *mddev);
747 /* Changes the consistency policy of an active array. */
748 int (*change_consistency_policy)(struct mddev *mddev, const char *buf);
749 /* convert io ranges from array to bitmap */
750 void (*bitmap_sector)(struct mddev *mddev, sector_t *offset,
751 unsigned long *sectors);
752};
753
754struct md_sysfs_entry {
755 struct attribute attr;
756 ssize_t (*show)(struct mddev *, char *);
757 ssize_t (*store)(struct mddev *, const char *, size_t);
758};
759extern const struct attribute_group md_bitmap_group;
760
761static inline struct kernfs_node *sysfs_get_dirent_safe(struct kernfs_node *sd, char *name)
762{
763 if (sd)
764 return sysfs_get_dirent(sd, name);
765 return sd;
766}
767static inline void sysfs_notify_dirent_safe(struct kernfs_node *sd)
768{
769 if (sd)
770 sysfs_notify_dirent(sd);
771}
772
773static inline char * mdname (struct mddev * mddev)
774{
775 return mddev->gendisk ? mddev->gendisk->disk_name : "mdX";
776}
777
778static inline int sysfs_link_rdev(struct mddev *mddev, struct md_rdev *rdev)
779{
780 char nm[20];
781 if (!test_bit(Replacement, &rdev->flags) &&
782 !test_bit(Journal, &rdev->flags) &&
783 mddev->kobj.sd) {
784 sprintf(nm, "rd%d", rdev->raid_disk);
785 return sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
786 } else
787 return 0;
788}
789
790static inline void sysfs_unlink_rdev(struct mddev *mddev, struct md_rdev *rdev)
791{
792 char nm[20];
793 if (!test_bit(Replacement, &rdev->flags) &&
794 !test_bit(Journal, &rdev->flags) &&
795 mddev->kobj.sd) {
796 sprintf(nm, "rd%d", rdev->raid_disk);
797 sysfs_remove_link(&mddev->kobj, nm);
798 }
799}
800
801/*
802 * iterates through some rdev ringlist. It's safe to remove the
803 * current 'rdev'. Dont touch 'tmp' though.
804 */
805#define rdev_for_each_list(rdev, tmp, head) \
806 list_for_each_entry_safe(rdev, tmp, head, same_set)
807
808/*
809 * iterates through the 'same array disks' ringlist
810 */
811#define rdev_for_each(rdev, mddev) \
812 list_for_each_entry(rdev, &((mddev)->disks), same_set)
813
814#define rdev_for_each_safe(rdev, tmp, mddev) \
815 list_for_each_entry_safe(rdev, tmp, &((mddev)->disks), same_set)
816
817#define rdev_for_each_rcu(rdev, mddev) \
818 list_for_each_entry_rcu(rdev, &((mddev)->disks), same_set)
819
820struct md_thread {
821 void (*run) (struct md_thread *thread);
822 struct mddev *mddev;
823 wait_queue_head_t wqueue;
824 unsigned long flags;
825 struct task_struct *tsk;
826 unsigned long timeout;
827 void *private;
828};
829
830struct md_io_clone {
831 struct mddev *mddev;
832 struct bio *orig_bio;
833 unsigned long start_time;
834 sector_t offset;
835 unsigned long sectors;
836 struct bio bio_clone;
837};
838
839#define THREAD_WAKEUP 0
840
841static inline void safe_put_page(struct page *p)
842{
843 if (p) put_page(p);
844}
845
846extern int register_md_personality(struct md_personality *p);
847extern int unregister_md_personality(struct md_personality *p);
848extern int register_md_cluster_operations(const struct md_cluster_operations *ops,
849 struct module *module);
850extern int unregister_md_cluster_operations(void);
851extern int md_setup_cluster(struct mddev *mddev, int nodes);
852extern void md_cluster_stop(struct mddev *mddev);
853extern struct md_thread *md_register_thread(
854 void (*run)(struct md_thread *thread),
855 struct mddev *mddev,
856 const char *name);
857extern void md_unregister_thread(struct mddev *mddev, struct md_thread __rcu **threadp);
858extern void md_wakeup_thread(struct md_thread __rcu *thread);
859extern void md_check_recovery(struct mddev *mddev);
860extern void md_reap_sync_thread(struct mddev *mddev);
861extern enum sync_action md_sync_action(struct mddev *mddev);
862extern enum sync_action md_sync_action_by_name(const char *page);
863extern const char *md_sync_action_name(enum sync_action action);
864extern void md_write_start(struct mddev *mddev, struct bio *bi);
865extern void md_write_inc(struct mddev *mddev, struct bio *bi);
866extern void md_write_end(struct mddev *mddev);
867extern void md_done_sync(struct mddev *mddev, int blocks, int ok);
868extern void md_error(struct mddev *mddev, struct md_rdev *rdev);
869extern void md_finish_reshape(struct mddev *mddev);
870void md_submit_discard_bio(struct mddev *mddev, struct md_rdev *rdev,
871 struct bio *bio, sector_t start, sector_t size);
872void md_account_bio(struct mddev *mddev, struct bio **bio);
873void md_free_cloned_bio(struct bio *bio);
874
875extern bool __must_check md_flush_request(struct mddev *mddev, struct bio *bio);
876extern void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
877 sector_t sector, int size, struct page *page);
878extern int md_super_wait(struct mddev *mddev);
879extern int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
880 struct page *page, blk_opf_t opf, bool metadata_op);
881extern void md_do_sync(struct md_thread *thread);
882extern void md_new_event(void);
883extern void md_allow_write(struct mddev *mddev);
884extern void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev);
885extern void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors);
886extern int md_check_no_bitmap(struct mddev *mddev);
887extern int md_integrity_register(struct mddev *mddev);
888extern int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale);
889
890extern int mddev_init(struct mddev *mddev);
891extern void mddev_destroy(struct mddev *mddev);
892void md_init_stacking_limits(struct queue_limits *lim);
893struct mddev *md_alloc(dev_t dev, char *name);
894void mddev_put(struct mddev *mddev);
895extern int md_run(struct mddev *mddev);
896extern int md_start(struct mddev *mddev);
897extern void md_stop(struct mddev *mddev);
898extern void md_stop_writes(struct mddev *mddev);
899extern int md_rdev_init(struct md_rdev *rdev);
900extern void md_rdev_clear(struct md_rdev *rdev);
901
902extern bool md_handle_request(struct mddev *mddev, struct bio *bio);
903extern int mddev_suspend(struct mddev *mddev, bool interruptible);
904extern void mddev_resume(struct mddev *mddev);
905extern void md_idle_sync_thread(struct mddev *mddev);
906extern void md_frozen_sync_thread(struct mddev *mddev);
907extern void md_unfrozen_sync_thread(struct mddev *mddev);
908
909extern void md_reload_sb(struct mddev *mddev, int raid_disk);
910extern void md_update_sb(struct mddev *mddev, int force);
911extern void mddev_create_serial_pool(struct mddev *mddev, struct md_rdev *rdev);
912extern void mddev_destroy_serial_pool(struct mddev *mddev,
913 struct md_rdev *rdev);
914struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr);
915struct md_rdev *md_find_rdev_rcu(struct mddev *mddev, dev_t dev);
916
917static inline bool is_rdev_broken(struct md_rdev *rdev)
918{
919 return !disk_live(rdev->bdev->bd_disk);
920}
921
922static inline void rdev_dec_pending(struct md_rdev *rdev, struct mddev *mddev)
923{
924 int faulty = test_bit(Faulty, &rdev->flags);
925 if (atomic_dec_and_test(&rdev->nr_pending) && faulty) {
926 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
927 md_wakeup_thread(mddev->thread);
928 }
929}
930
931extern const struct md_cluster_operations *md_cluster_ops;
932static inline int mddev_is_clustered(struct mddev *mddev)
933{
934 return mddev->cluster_info && mddev->bitmap_info.nodes > 1;
935}
936
937/* clear unsupported mddev_flags */
938static inline void mddev_clear_unsupported_flags(struct mddev *mddev,
939 unsigned long unsupported_flags)
940{
941 mddev->flags &= ~unsupported_flags;
942}
943
944static inline void mddev_check_write_zeroes(struct mddev *mddev, struct bio *bio)
945{
946 if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
947 !bio->bi_bdev->bd_disk->queue->limits.max_write_zeroes_sectors)
948 mddev->gendisk->queue->limits.max_write_zeroes_sectors = 0;
949}
950
951static inline int mddev_suspend_and_lock(struct mddev *mddev)
952{
953 int ret;
954
955 ret = mddev_suspend(mddev, true);
956 if (ret)
957 return ret;
958
959 ret = mddev_lock(mddev);
960 if (ret)
961 mddev_resume(mddev);
962
963 return ret;
964}
965
966static inline void mddev_suspend_and_lock_nointr(struct mddev *mddev)
967{
968 mddev_suspend(mddev, false);
969 mutex_lock(&mddev->reconfig_mutex);
970}
971
972static inline void mddev_unlock_and_resume(struct mddev *mddev)
973{
974 mddev_unlock(mddev);
975 mddev_resume(mddev);
976}
977
978struct mdu_array_info_s;
979struct mdu_disk_info_s;
980
981extern int mdp_major;
982extern struct workqueue_struct *md_bitmap_wq;
983void md_autostart_arrays(int part);
984int md_set_array_info(struct mddev *mddev, struct mdu_array_info_s *info);
985int md_add_new_disk(struct mddev *mddev, struct mdu_disk_info_s *info);
986int do_md_run(struct mddev *mddev);
987#define MDDEV_STACK_INTEGRITY (1u << 0)
988int mddev_stack_rdev_limits(struct mddev *mddev, struct queue_limits *lim,
989 unsigned int flags);
990int mddev_stack_new_rdev(struct mddev *mddev, struct md_rdev *rdev);
991void mddev_update_io_opt(struct mddev *mddev, unsigned int nr_stripes);
992
993extern const struct block_device_operations md_fops;
994
995/*
996 * MD devices can be used undeneath by DM, in which case ->gendisk is NULL.
997 */
998static inline bool mddev_is_dm(struct mddev *mddev)
999{
1000 return !mddev->gendisk;
1001}
1002
1003static inline void mddev_trace_remap(struct mddev *mddev, struct bio *bio,
1004 sector_t sector)
1005{
1006 if (!mddev_is_dm(mddev))
1007 trace_block_bio_remap(bio, disk_devt(mddev->gendisk), sector);
1008}
1009
1010static inline bool rdev_blocked(struct md_rdev *rdev)
1011{
1012 /*
1013 * Blocked will be set by error handler and cleared by daemon after
1014 * updating superblock, meanwhile write IO should be blocked to prevent
1015 * reading old data after power failure.
1016 */
1017 if (test_bit(Blocked, &rdev->flags))
1018 return true;
1019
1020 /*
1021 * Faulty device should not be accessed anymore, there is no need to
1022 * wait for bad block to be acknowledged.
1023 */
1024 if (test_bit(Faulty, &rdev->flags))
1025 return false;
1026
1027 /* rdev is blocked by badblocks. */
1028 if (test_bit(BlockedBadBlocks, &rdev->flags))
1029 return true;
1030
1031 return false;
1032}
1033
1034#define mddev_add_trace_msg(mddev, fmt, args...) \
1035do { \
1036 if (!mddev_is_dm(mddev)) \
1037 blk_add_trace_msg((mddev)->gendisk->queue, fmt, ##args); \
1038} while (0)
1039
1040#endif /* _MD_MD_H */