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/*
2 md.h : kernel internal structure of the Linux MD driver
3 Copyright (C) 1996-98 Ingo Molnar, Gadi Oxman
4
5 This program is free software; you can redistribute it and/or modify
6 it under the terms of the GNU General Public License as published by
7 the Free Software Foundation; either version 2, or (at your option)
8 any later version.
9
10 You should have received a copy of the GNU General Public License
11 (for example /usr/src/linux/COPYING); if not, write to the Free
12 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
13*/
14
15#ifndef _MD_MD_H
16#define _MD_MD_H
17
18#include <linux/blkdev.h>
19#include <linux/backing-dev.h>
20#include <linux/badblocks.h>
21#include <linux/kobject.h>
22#include <linux/list.h>
23#include <linux/mm.h>
24#include <linux/mutex.h>
25#include <linux/timer.h>
26#include <linux/wait.h>
27#include <linux/workqueue.h>
28#include "md-cluster.h"
29
30#define MaxSector (~(sector_t)0)
31
32/*
33 * These flags should really be called "NO_RETRY" rather than
34 * "FAILFAST" because they don't make any promise about time lapse,
35 * only about the number of retries, which will be zero.
36 * REQ_FAILFAST_DRIVER is not included because
37 * Commit: 4a27446f3e39 ("[SCSI] modify scsi to handle new fail fast flags.")
38 * seems to suggest that the errors it avoids retrying should usually
39 * be retried.
40 */
41#define MD_FAILFAST (REQ_FAILFAST_DEV | REQ_FAILFAST_TRANSPORT)
42/*
43 * MD's 'extended' device
44 */
45struct md_rdev {
46 struct list_head same_set; /* RAID devices within the same set */
47
48 sector_t sectors; /* Device size (in 512bytes sectors) */
49 struct mddev *mddev; /* RAID array if running */
50 int last_events; /* IO event timestamp */
51
52 /*
53 * If meta_bdev is non-NULL, it means that a separate device is
54 * being used to store the metadata (superblock/bitmap) which
55 * would otherwise be contained on the same device as the data (bdev).
56 */
57 struct block_device *meta_bdev;
58 struct block_device *bdev; /* block device handle */
59
60 struct page *sb_page, *bb_page;
61 int sb_loaded;
62 __u64 sb_events;
63 sector_t data_offset; /* start of data in array */
64 sector_t new_data_offset;/* only relevant while reshaping */
65 sector_t sb_start; /* offset of the super block (in 512byte sectors) */
66 int sb_size; /* bytes in the superblock */
67 int preferred_minor; /* autorun support */
68
69 struct kobject kobj;
70
71 /* A device can be in one of three states based on two flags:
72 * Not working: faulty==1 in_sync==0
73 * Fully working: faulty==0 in_sync==1
74 * Working, but not
75 * in sync with array
76 * faulty==0 in_sync==0
77 *
78 * It can never have faulty==1, in_sync==1
79 * This reduces the burden of testing multiple flags in many cases
80 */
81
82 unsigned long flags; /* bit set of 'enum flag_bits' bits. */
83 wait_queue_head_t blocked_wait;
84
85 int desc_nr; /* descriptor index in the superblock */
86 int raid_disk; /* role of device in array */
87 int new_raid_disk; /* role that the device will have in
88 * the array after a level-change completes.
89 */
90 int saved_raid_disk; /* role that device used to have in the
91 * array and could again if we did a partial
92 * resync from the bitmap
93 */
94 union {
95 sector_t recovery_offset;/* If this device has been partially
96 * recovered, this is where we were
97 * up to.
98 */
99 sector_t journal_tail; /* If this device is a journal device,
100 * this is the journal tail (journal
101 * recovery start point)
102 */
103 };
104
105 atomic_t nr_pending; /* number of pending requests.
106 * only maintained for arrays that
107 * support hot removal
108 */
109 atomic_t read_errors; /* number of consecutive read errors that
110 * we have tried to ignore.
111 */
112 time64_t last_read_error; /* monotonic time since our
113 * last read error
114 */
115 atomic_t corrected_errors; /* number of corrected read errors,
116 * for reporting to userspace and storing
117 * in superblock.
118 */
119 struct work_struct del_work; /* used for delayed sysfs removal */
120
121 struct kernfs_node *sysfs_state; /* handle for 'state'
122 * sysfs entry */
123
124 struct badblocks badblocks;
125
126 struct {
127 short offset; /* Offset from superblock to start of PPL.
128 * Not used by external metadata. */
129 unsigned int size; /* Size in sectors of the PPL space */
130 sector_t sector; /* First sector of the PPL space */
131 } ppl;
132};
133enum flag_bits {
134 Faulty, /* device is known to have a fault */
135 In_sync, /* device is in_sync with rest of array */
136 Bitmap_sync, /* ..actually, not quite In_sync. Need a
137 * bitmap-based recovery to get fully in sync.
138 * The bit is only meaningful before device
139 * has been passed to pers->hot_add_disk.
140 */
141 WriteMostly, /* Avoid reading if at all possible */
142 AutoDetected, /* added by auto-detect */
143 Blocked, /* An error occurred but has not yet
144 * been acknowledged by the metadata
145 * handler, so don't allow writes
146 * until it is cleared */
147 WriteErrorSeen, /* A write error has been seen on this
148 * device
149 */
150 FaultRecorded, /* Intermediate state for clearing
151 * Blocked. The Fault is/will-be
152 * recorded in the metadata, but that
153 * metadata hasn't been stored safely
154 * on disk yet.
155 */
156 BlockedBadBlocks, /* A writer is blocked because they
157 * found an unacknowledged bad-block.
158 * This can safely be cleared at any
159 * time, and the writer will re-check.
160 * It may be set at any time, and at
161 * worst the writer will timeout and
162 * re-check. So setting it as
163 * accurately as possible is good, but
164 * not absolutely critical.
165 */
166 WantReplacement, /* This device is a candidate to be
167 * hot-replaced, either because it has
168 * reported some faults, or because
169 * of explicit request.
170 */
171 Replacement, /* This device is a replacement for
172 * a want_replacement device with same
173 * raid_disk number.
174 */
175 Candidate, /* For clustered environments only:
176 * This device is seen locally but not
177 * by the whole cluster
178 */
179 Journal, /* This device is used as journal for
180 * raid-5/6.
181 * Usually, this device should be faster
182 * than other devices in the array
183 */
184 ClusterRemove,
185 RemoveSynchronized, /* synchronize_rcu() was called after
186 * this device was known to be faulty,
187 * so it is safe to remove without
188 * another synchronize_rcu() call.
189 */
190 ExternalBbl, /* External metadata provides bad
191 * block management for a disk
192 */
193 FailFast, /* Minimal retries should be attempted on
194 * this device, so use REQ_FAILFAST_DEV.
195 * Also don't try to repair failed reads.
196 * It is expects that no bad block log
197 * is present.
198 */
199 LastDev, /* Seems to be the last working dev as
200 * it didn't fail, so don't use FailFast
201 * any more for metadata
202 */
203};
204
205static inline int is_badblock(struct md_rdev *rdev, sector_t s, int sectors,
206 sector_t *first_bad, int *bad_sectors)
207{
208 if (unlikely(rdev->badblocks.count)) {
209 int rv = badblocks_check(&rdev->badblocks, rdev->data_offset + s,
210 sectors,
211 first_bad, bad_sectors);
212 if (rv)
213 *first_bad -= rdev->data_offset;
214 return rv;
215 }
216 return 0;
217}
218extern int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
219 int is_new);
220extern int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
221 int is_new);
222struct md_cluster_info;
223
224/* change UNSUPPORTED_MDDEV_FLAGS for each array type if new flag is added */
225enum mddev_flags {
226 MD_ARRAY_FIRST_USE, /* First use of array, needs initialization */
227 MD_CLOSING, /* If set, we are closing the array, do not open
228 * it then */
229 MD_JOURNAL_CLEAN, /* A raid with journal is already clean */
230 MD_HAS_JOURNAL, /* The raid array has journal feature set */
231 MD_CLUSTER_RESYNC_LOCKED, /* cluster raid only, which means node
232 * already took resync lock, need to
233 * release the lock */
234 MD_FAILFAST_SUPPORTED, /* Using MD_FAILFAST on metadata writes is
235 * supported as calls to md_error() will
236 * never cause the array to become failed.
237 */
238 MD_HAS_PPL, /* The raid array has PPL feature set */
239 MD_HAS_MULTIPLE_PPLS, /* The raid array has multiple PPLs feature set */
240 MD_ALLOW_SB_UPDATE, /* md_check_recovery is allowed to update
241 * the metadata without taking reconfig_mutex.
242 */
243 MD_UPDATING_SB, /* md_check_recovery is updating the metadata
244 * without explicitly holding reconfig_mutex.
245 */
246};
247
248enum mddev_sb_flags {
249 MD_SB_CHANGE_DEVS, /* Some device status has changed */
250 MD_SB_CHANGE_CLEAN, /* transition to or from 'clean' */
251 MD_SB_CHANGE_PENDING, /* switch from 'clean' to 'active' in progress */
252 MD_SB_NEED_REWRITE, /* metadata write needs to be repeated */
253};
254
255struct mddev {
256 void *private;
257 struct md_personality *pers;
258 dev_t unit;
259 int md_minor;
260 struct list_head disks;
261 unsigned long flags;
262 unsigned long sb_flags;
263
264 int suspended;
265 atomic_t active_io;
266 int ro;
267 int sysfs_active; /* set when sysfs deletes
268 * are happening, so run/
269 * takeover/stop are not safe
270 */
271 struct gendisk *gendisk;
272
273 struct kobject kobj;
274 int hold_active;
275#define UNTIL_IOCTL 1
276#define UNTIL_STOP 2
277
278 /* Superblock information */
279 int major_version,
280 minor_version,
281 patch_version;
282 int persistent;
283 int external; /* metadata is
284 * managed externally */
285 char metadata_type[17]; /* externally set*/
286 int chunk_sectors;
287 time64_t ctime, utime;
288 int level, layout;
289 char clevel[16];
290 int raid_disks;
291 int max_disks;
292 sector_t dev_sectors; /* used size of
293 * component devices */
294 sector_t array_sectors; /* exported array size */
295 int external_size; /* size managed
296 * externally */
297 __u64 events;
298 /* If the last 'event' was simply a clean->dirty transition, and
299 * we didn't write it to the spares, then it is safe and simple
300 * to just decrement the event count on a dirty->clean transition.
301 * So we record that possibility here.
302 */
303 int can_decrease_events;
304
305 char uuid[16];
306
307 /* If the array is being reshaped, we need to record the
308 * new shape and an indication of where we are up to.
309 * This is written to the superblock.
310 * If reshape_position is MaxSector, then no reshape is happening (yet).
311 */
312 sector_t reshape_position;
313 int delta_disks, new_level, new_layout;
314 int new_chunk_sectors;
315 int reshape_backwards;
316
317 struct md_thread *thread; /* management thread */
318 struct md_thread *sync_thread; /* doing resync or reconstruct */
319
320 /* 'last_sync_action' is initialized to "none". It is set when a
321 * sync operation (i.e "data-check", "requested-resync", "resync",
322 * "recovery", or "reshape") is started. It holds this value even
323 * when the sync thread is "frozen" (interrupted) or "idle" (stopped
324 * or finished). It is overwritten when a new sync operation is begun.
325 */
326 char *last_sync_action;
327 sector_t curr_resync; /* last block scheduled */
328 /* As resync requests can complete out of order, we cannot easily track
329 * how much resync has been completed. So we occasionally pause until
330 * everything completes, then set curr_resync_completed to curr_resync.
331 * As such it may be well behind the real resync mark, but it is a value
332 * we are certain of.
333 */
334 sector_t curr_resync_completed;
335 unsigned long resync_mark; /* a recent timestamp */
336 sector_t resync_mark_cnt;/* blocks written at resync_mark */
337 sector_t curr_mark_cnt; /* blocks scheduled now */
338
339 sector_t resync_max_sectors; /* may be set by personality */
340
341 atomic64_t resync_mismatches; /* count of sectors where
342 * parity/replica mismatch found
343 */
344
345 /* allow user-space to request suspension of IO to regions of the array */
346 sector_t suspend_lo;
347 sector_t suspend_hi;
348 /* if zero, use the system-wide default */
349 int sync_speed_min;
350 int sync_speed_max;
351
352 /* resync even though the same disks are shared among md-devices */
353 int parallel_resync;
354
355 int ok_start_degraded;
356
357 unsigned long recovery;
358 /* If a RAID personality determines that recovery (of a particular
359 * device) will fail due to a read error on the source device, it
360 * takes a copy of this number and does not attempt recovery again
361 * until this number changes.
362 */
363 int recovery_disabled;
364
365 int in_sync; /* know to not need resync */
366 /* 'open_mutex' avoids races between 'md_open' and 'do_md_stop', so
367 * that we are never stopping an array while it is open.
368 * 'reconfig_mutex' protects all other reconfiguration.
369 * These locks are separate due to conflicting interactions
370 * with bdev->bd_mutex.
371 * Lock ordering is:
372 * reconfig_mutex -> bd_mutex : e.g. do_md_run -> revalidate_disk
373 * bd_mutex -> open_mutex: e.g. __blkdev_get -> md_open
374 */
375 struct mutex open_mutex;
376 struct mutex reconfig_mutex;
377 atomic_t active; /* general refcount */
378 atomic_t openers; /* number of active opens */
379
380 int changed; /* True if we might need to
381 * reread partition info */
382 int degraded; /* whether md should consider
383 * adding a spare
384 */
385
386 atomic_t recovery_active; /* blocks scheduled, but not written */
387 wait_queue_head_t recovery_wait;
388 sector_t recovery_cp;
389 sector_t resync_min; /* user requested sync
390 * starts here */
391 sector_t resync_max; /* resync should pause
392 * when it gets here */
393
394 struct kernfs_node *sysfs_state; /* handle for 'array_state'
395 * file in sysfs.
396 */
397 struct kernfs_node *sysfs_action; /* handle for 'sync_action' */
398
399 struct work_struct del_work; /* used for delayed sysfs removal */
400
401 /* "lock" protects:
402 * flush_bio transition from NULL to !NULL
403 * rdev superblocks, events
404 * clearing MD_CHANGE_*
405 * in_sync - and related safemode and MD_CHANGE changes
406 * pers (also protected by reconfig_mutex and pending IO).
407 * clearing ->bitmap
408 * clearing ->bitmap_info.file
409 * changing ->resync_{min,max}
410 * setting MD_RECOVERY_RUNNING (which interacts with resync_{min,max})
411 */
412 spinlock_t lock;
413 wait_queue_head_t sb_wait; /* for waiting on superblock updates */
414 atomic_t pending_writes; /* number of active superblock writes */
415
416 unsigned int safemode; /* if set, update "clean" superblock
417 * when no writes pending.
418 */
419 unsigned int safemode_delay;
420 struct timer_list safemode_timer;
421 struct percpu_ref writes_pending;
422 int sync_checkers; /* # of threads checking writes_pending */
423 struct request_queue *queue; /* for plugging ... */
424
425 struct bitmap *bitmap; /* the bitmap for the device */
426 struct {
427 struct file *file; /* the bitmap file */
428 loff_t offset; /* offset from superblock of
429 * start of bitmap. May be
430 * negative, but not '0'
431 * For external metadata, offset
432 * from start of device.
433 */
434 unsigned long space; /* space available at this offset */
435 loff_t default_offset; /* this is the offset to use when
436 * hot-adding a bitmap. It should
437 * eventually be settable by sysfs.
438 */
439 unsigned long default_space; /* space available at
440 * default offset */
441 struct mutex mutex;
442 unsigned long chunksize;
443 unsigned long daemon_sleep; /* how many jiffies between updates? */
444 unsigned long max_write_behind; /* write-behind mode */
445 int external;
446 int nodes; /* Maximum number of nodes in the cluster */
447 char cluster_name[64]; /* Name of the cluster */
448 } bitmap_info;
449
450 atomic_t max_corr_read_errors; /* max read retries */
451 struct list_head all_mddevs;
452
453 struct attribute_group *to_remove;
454
455 struct bio_set *bio_set;
456 struct bio_set *sync_set; /* for sync operations like
457 * metadata and bitmap writes
458 */
459
460 /* Generic flush handling.
461 * The last to finish preflush schedules a worker to submit
462 * the rest of the request (without the REQ_PREFLUSH flag).
463 */
464 struct bio *flush_bio;
465 atomic_t flush_pending;
466 struct work_struct flush_work;
467 struct work_struct event_work; /* used by dm to report failure event */
468 void (*sync_super)(struct mddev *mddev, struct md_rdev *rdev);
469 struct md_cluster_info *cluster_info;
470 unsigned int good_device_nr; /* good device num within cluster raid */
471
472 bool has_superblocks:1;
473};
474
475enum recovery_flags {
476 /*
477 * If neither SYNC or RESHAPE are set, then it is a recovery.
478 */
479 MD_RECOVERY_RUNNING, /* a thread is running, or about to be started */
480 MD_RECOVERY_SYNC, /* actually doing a resync, not a recovery */
481 MD_RECOVERY_RECOVER, /* doing recovery, or need to try it. */
482 MD_RECOVERY_INTR, /* resync needs to be aborted for some reason */
483 MD_RECOVERY_DONE, /* thread is done and is waiting to be reaped */
484 MD_RECOVERY_NEEDED, /* we might need to start a resync/recover */
485 MD_RECOVERY_REQUESTED, /* user-space has requested a sync (used with SYNC) */
486 MD_RECOVERY_CHECK, /* user-space request for check-only, no repair */
487 MD_RECOVERY_RESHAPE, /* A reshape is happening */
488 MD_RECOVERY_FROZEN, /* User request to abort, and not restart, any action */
489 MD_RECOVERY_ERROR, /* sync-action interrupted because io-error */
490 MD_RECOVERY_WAIT, /* waiting for pers->start() to finish */
491};
492
493static inline int __must_check mddev_lock(struct mddev *mddev)
494{
495 return mutex_lock_interruptible(&mddev->reconfig_mutex);
496}
497
498/* Sometimes we need to take the lock in a situation where
499 * failure due to interrupts is not acceptable.
500 */
501static inline void mddev_lock_nointr(struct mddev *mddev)
502{
503 mutex_lock(&mddev->reconfig_mutex);
504}
505
506static inline int mddev_trylock(struct mddev *mddev)
507{
508 return mutex_trylock(&mddev->reconfig_mutex);
509}
510extern void mddev_unlock(struct mddev *mddev);
511
512static inline void md_sync_acct(struct block_device *bdev, unsigned long nr_sectors)
513{
514 atomic_add(nr_sectors, &bdev->bd_contains->bd_disk->sync_io);
515}
516
517static inline void md_sync_acct_bio(struct bio *bio, unsigned long nr_sectors)
518{
519 atomic_add(nr_sectors, &bio->bi_disk->sync_io);
520}
521
522struct md_personality
523{
524 char *name;
525 int level;
526 struct list_head list;
527 struct module *owner;
528 bool (*make_request)(struct mddev *mddev, struct bio *bio);
529 /*
530 * start up works that do NOT require md_thread. tasks that
531 * requires md_thread should go into start()
532 */
533 int (*run)(struct mddev *mddev);
534 /* start up works that require md threads */
535 int (*start)(struct mddev *mddev);
536 void (*free)(struct mddev *mddev, void *priv);
537 void (*status)(struct seq_file *seq, struct mddev *mddev);
538 /* error_handler must set ->faulty and clear ->in_sync
539 * if appropriate, and should abort recovery if needed
540 */
541 void (*error_handler)(struct mddev *mddev, struct md_rdev *rdev);
542 int (*hot_add_disk) (struct mddev *mddev, struct md_rdev *rdev);
543 int (*hot_remove_disk) (struct mddev *mddev, struct md_rdev *rdev);
544 int (*spare_active) (struct mddev *mddev);
545 sector_t (*sync_request)(struct mddev *mddev, sector_t sector_nr, int *skipped);
546 int (*resize) (struct mddev *mddev, sector_t sectors);
547 sector_t (*size) (struct mddev *mddev, sector_t sectors, int raid_disks);
548 int (*check_reshape) (struct mddev *mddev);
549 int (*start_reshape) (struct mddev *mddev);
550 void (*finish_reshape) (struct mddev *mddev);
551 /* quiesce suspends or resumes internal processing.
552 * 1 - stop new actions and wait for action io to complete
553 * 0 - return to normal behaviour
554 */
555 void (*quiesce) (struct mddev *mddev, int quiesce);
556 /* takeover is used to transition an array from one
557 * personality to another. The new personality must be able
558 * to handle the data in the current layout.
559 * e.g. 2drive raid1 -> 2drive raid5
560 * ndrive raid5 -> degraded n+1drive raid6 with special layout
561 * If the takeover succeeds, a new 'private' structure is returned.
562 * This needs to be installed and then ->run used to activate the
563 * array.
564 */
565 void *(*takeover) (struct mddev *mddev);
566 /* congested implements bdi.congested_fn().
567 * Will not be called while array is 'suspended' */
568 int (*congested)(struct mddev *mddev, int bits);
569 /* Changes the consistency policy of an active array. */
570 int (*change_consistency_policy)(struct mddev *mddev, const char *buf);
571};
572
573struct md_sysfs_entry {
574 struct attribute attr;
575 ssize_t (*show)(struct mddev *, char *);
576 ssize_t (*store)(struct mddev *, const char *, size_t);
577};
578extern struct attribute_group md_bitmap_group;
579
580static inline struct kernfs_node *sysfs_get_dirent_safe(struct kernfs_node *sd, char *name)
581{
582 if (sd)
583 return sysfs_get_dirent(sd, name);
584 return sd;
585}
586static inline void sysfs_notify_dirent_safe(struct kernfs_node *sd)
587{
588 if (sd)
589 sysfs_notify_dirent(sd);
590}
591
592static inline char * mdname (struct mddev * mddev)
593{
594 return mddev->gendisk ? mddev->gendisk->disk_name : "mdX";
595}
596
597static inline int sysfs_link_rdev(struct mddev *mddev, struct md_rdev *rdev)
598{
599 char nm[20];
600 if (!test_bit(Replacement, &rdev->flags) &&
601 !test_bit(Journal, &rdev->flags) &&
602 mddev->kobj.sd) {
603 sprintf(nm, "rd%d", rdev->raid_disk);
604 return sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
605 } else
606 return 0;
607}
608
609static inline void sysfs_unlink_rdev(struct mddev *mddev, struct md_rdev *rdev)
610{
611 char nm[20];
612 if (!test_bit(Replacement, &rdev->flags) &&
613 !test_bit(Journal, &rdev->flags) &&
614 mddev->kobj.sd) {
615 sprintf(nm, "rd%d", rdev->raid_disk);
616 sysfs_remove_link(&mddev->kobj, nm);
617 }
618}
619
620/*
621 * iterates through some rdev ringlist. It's safe to remove the
622 * current 'rdev'. Dont touch 'tmp' though.
623 */
624#define rdev_for_each_list(rdev, tmp, head) \
625 list_for_each_entry_safe(rdev, tmp, head, same_set)
626
627/*
628 * iterates through the 'same array disks' ringlist
629 */
630#define rdev_for_each(rdev, mddev) \
631 list_for_each_entry(rdev, &((mddev)->disks), same_set)
632
633#define rdev_for_each_safe(rdev, tmp, mddev) \
634 list_for_each_entry_safe(rdev, tmp, &((mddev)->disks), same_set)
635
636#define rdev_for_each_rcu(rdev, mddev) \
637 list_for_each_entry_rcu(rdev, &((mddev)->disks), same_set)
638
639struct md_thread {
640 void (*run) (struct md_thread *thread);
641 struct mddev *mddev;
642 wait_queue_head_t wqueue;
643 unsigned long flags;
644 struct task_struct *tsk;
645 unsigned long timeout;
646 void *private;
647};
648
649#define THREAD_WAKEUP 0
650
651static inline void safe_put_page(struct page *p)
652{
653 if (p) put_page(p);
654}
655
656extern int register_md_personality(struct md_personality *p);
657extern int unregister_md_personality(struct md_personality *p);
658extern int register_md_cluster_operations(struct md_cluster_operations *ops,
659 struct module *module);
660extern int unregister_md_cluster_operations(void);
661extern int md_setup_cluster(struct mddev *mddev, int nodes);
662extern void md_cluster_stop(struct mddev *mddev);
663extern struct md_thread *md_register_thread(
664 void (*run)(struct md_thread *thread),
665 struct mddev *mddev,
666 const char *name);
667extern void md_unregister_thread(struct md_thread **threadp);
668extern void md_wakeup_thread(struct md_thread *thread);
669extern void md_check_recovery(struct mddev *mddev);
670extern void md_reap_sync_thread(struct mddev *mddev);
671extern int mddev_init_writes_pending(struct mddev *mddev);
672extern bool md_write_start(struct mddev *mddev, struct bio *bi);
673extern void md_write_inc(struct mddev *mddev, struct bio *bi);
674extern void md_write_end(struct mddev *mddev);
675extern void md_done_sync(struct mddev *mddev, int blocks, int ok);
676extern void md_error(struct mddev *mddev, struct md_rdev *rdev);
677extern void md_finish_reshape(struct mddev *mddev);
678
679extern int mddev_congested(struct mddev *mddev, int bits);
680extern void md_flush_request(struct mddev *mddev, struct bio *bio);
681extern void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
682 sector_t sector, int size, struct page *page);
683extern int md_super_wait(struct mddev *mddev);
684extern int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
685 struct page *page, int op, int op_flags,
686 bool metadata_op);
687extern void md_do_sync(struct md_thread *thread);
688extern void md_new_event(struct mddev *mddev);
689extern void md_allow_write(struct mddev *mddev);
690extern void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev);
691extern void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors);
692extern int md_check_no_bitmap(struct mddev *mddev);
693extern int md_integrity_register(struct mddev *mddev);
694extern int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev);
695extern int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale);
696
697extern void mddev_init(struct mddev *mddev);
698extern int md_run(struct mddev *mddev);
699extern int md_start(struct mddev *mddev);
700extern void md_stop(struct mddev *mddev);
701extern void md_stop_writes(struct mddev *mddev);
702extern int md_rdev_init(struct md_rdev *rdev);
703extern void md_rdev_clear(struct md_rdev *rdev);
704
705extern void md_handle_request(struct mddev *mddev, struct bio *bio);
706extern void mddev_suspend(struct mddev *mddev);
707extern void mddev_resume(struct mddev *mddev);
708extern struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
709 struct mddev *mddev);
710
711extern void md_reload_sb(struct mddev *mddev, int raid_disk);
712extern void md_update_sb(struct mddev *mddev, int force);
713extern void md_kick_rdev_from_array(struct md_rdev * rdev);
714struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr);
715struct md_rdev *md_find_rdev_rcu(struct mddev *mddev, dev_t dev);
716
717static inline void rdev_dec_pending(struct md_rdev *rdev, struct mddev *mddev)
718{
719 int faulty = test_bit(Faulty, &rdev->flags);
720 if (atomic_dec_and_test(&rdev->nr_pending) && faulty) {
721 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
722 md_wakeup_thread(mddev->thread);
723 }
724}
725
726extern struct md_cluster_operations *md_cluster_ops;
727static inline int mddev_is_clustered(struct mddev *mddev)
728{
729 return mddev->cluster_info && mddev->bitmap_info.nodes > 1;
730}
731
732/* clear unsupported mddev_flags */
733static inline void mddev_clear_unsupported_flags(struct mddev *mddev,
734 unsigned long unsupported_flags)
735{
736 mddev->flags &= ~unsupported_flags;
737}
738
739static inline void mddev_check_writesame(struct mddev *mddev, struct bio *bio)
740{
741 if (bio_op(bio) == REQ_OP_WRITE_SAME &&
742 !bio->bi_disk->queue->limits.max_write_same_sectors)
743 mddev->queue->limits.max_write_same_sectors = 0;
744}
745
746static inline void mddev_check_write_zeroes(struct mddev *mddev, struct bio *bio)
747{
748 if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
749 !bio->bi_disk->queue->limits.max_write_zeroes_sectors)
750 mddev->queue->limits.max_write_zeroes_sectors = 0;
751}
752#endif /* _MD_MD_H */