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