2 md.h : kernel internal structure of the Linux MD driver
3 Copyright (C) 1996-98 Ingo Molnar, Gadi Oxman
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)
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.
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>
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"
30 #define MaxSector (~(sector_t)0)
33 * MD's 'extended' device
36 struct list_head same_set; /* RAID devices within the same set */
38 sector_t sectors; /* Device size (in 512bytes sectors) */
39 struct mddev *mddev; /* RAID array if running */
40 int last_events; /* IO event timestamp */
43 * If meta_bdev is non-NULL, it means that a separate device is
44 * being used to store the metadata (superblock/bitmap) which
45 * would otherwise be contained on the same device as the data (bdev).
47 struct block_device *meta_bdev;
48 struct block_device *bdev; /* block device handle */
50 struct page *sb_page, *bb_page;
53 sector_t data_offset; /* start of data in array */
54 sector_t new_data_offset;/* only relevant while reshaping */
55 sector_t sb_start; /* offset of the super block (in 512byte sectors) */
56 int sb_size; /* bytes in the superblock */
57 int preferred_minor; /* autorun support */
61 /* A device can be in one of three states based on two flags:
62 * Not working: faulty==1 in_sync==0
63 * Fully working: faulty==0 in_sync==1
66 * faulty==0 in_sync==0
68 * It can never have faulty==1, in_sync==1
69 * This reduces the burden of testing multiple flags in many cases
72 unsigned long flags; /* bit set of 'enum flag_bits' bits. */
73 wait_queue_head_t blocked_wait;
75 int desc_nr; /* descriptor index in the superblock */
76 int raid_disk; /* role of device in array */
77 int new_raid_disk; /* role that the device will have in
78 * the array after a level-change completes.
80 int saved_raid_disk; /* role that device used to have in the
81 * array and could again if we did a partial
82 * resync from the bitmap
85 sector_t recovery_offset;/* If this device has been partially
86 * recovered, this is where we were
89 sector_t journal_tail; /* If this device is a journal device,
90 * this is the journal tail (journal
91 * recovery start point)
95 atomic_t nr_pending; /* number of pending requests.
96 * only maintained for arrays that
99 atomic_t read_errors; /* number of consecutive read errors that
100 * we have tried to ignore.
102 time64_t last_read_error; /* monotonic time since our
105 atomic_t corrected_errors; /* number of corrected read errors,
106 * for reporting to userspace and storing
109 struct work_struct del_work; /* used for delayed sysfs removal */
111 struct kernfs_node *sysfs_state; /* handle for 'state'
114 struct badblocks badblocks;
117 Faulty, /* device is known to have a fault */
118 In_sync, /* device is in_sync with rest of array */
119 Bitmap_sync, /* ..actually, not quite In_sync. Need a
120 * bitmap-based recovery to get fully in sync
122 WriteMostly, /* Avoid reading if at all possible */
123 AutoDetected, /* added by auto-detect */
124 Blocked, /* An error occurred but has not yet
125 * been acknowledged by the metadata
126 * handler, so don't allow writes
127 * until it is cleared */
128 WriteErrorSeen, /* A write error has been seen on this
131 FaultRecorded, /* Intermediate state for clearing
132 * Blocked. The Fault is/will-be
133 * recorded in the metadata, but that
134 * metadata hasn't been stored safely
137 BlockedBadBlocks, /* A writer is blocked because they
138 * found an unacknowledged bad-block.
139 * This can safely be cleared at any
140 * time, and the writer will re-check.
141 * It may be set at any time, and at
142 * worst the writer will timeout and
143 * re-check. So setting it as
144 * accurately as possible is good, but
145 * not absolutely critical.
147 WantReplacement, /* This device is a candidate to be
148 * hot-replaced, either because it has
149 * reported some faults, or because
150 * of explicit request.
152 Replacement, /* This device is a replacement for
153 * a want_replacement device with same
156 Candidate, /* For clustered environments only:
157 * This device is seen locally but not
158 * by the whole cluster
160 Journal, /* This device is used as journal for
162 * Usually, this device should be faster
163 * than other devices in the array
166 RemoveSynchronized, /* synchronize_rcu() was called after
167 * this device was known to be faulty,
168 * so it is safe to remove without
169 * another synchronize_rcu() call.
171 ExternalBbl, /* External metadata provides bad
172 * block management for a disk
174 FailFast, /* Minimal retries should be attempted on
175 * this device, so use REQ_FAILFAST_DEV.
176 * Also don't try to repair failed reads.
177 * It is expects that no bad block log
182 static inline int is_badblock(struct md_rdev *rdev, sector_t s, int sectors,
183 sector_t *first_bad, int *bad_sectors)
185 if (unlikely(rdev->badblocks.count)) {
186 int rv = badblocks_check(&rdev->badblocks, rdev->data_offset + s,
188 first_bad, bad_sectors);
190 *first_bad -= rdev->data_offset;
195 extern int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
197 extern int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
199 struct md_cluster_info;
202 MD_CHANGE_DEVS, /* Some device status has changed */
203 MD_CHANGE_CLEAN, /* transition to or from 'clean' */
204 MD_CHANGE_PENDING, /* switch from 'clean' to 'active' in progress */
205 MD_ARRAY_FIRST_USE, /* First use of array, needs initialization */
206 MD_CLOSING, /* If set, we are closing the array, do not open
208 MD_JOURNAL_CLEAN, /* A raid with journal is already clean */
209 MD_HAS_JOURNAL, /* The raid array has journal feature set */
210 MD_RELOAD_SB, /* Reload the superblock because another node
213 MD_CLUSTER_RESYNC_LOCKED, /* cluster raid only, which means node
214 * already took resync lock, need to
215 * release the lock */
217 #define MD_UPDATE_SB_FLAGS (BIT(MD_CHANGE_DEVS) | \
218 BIT(MD_CHANGE_CLEAN) | \
219 BIT(MD_CHANGE_PENDING)) /* If these are set, md_update_sb needed */
222 struct md_personality *pers;
225 struct list_head disks;
231 int sysfs_active; /* set when sysfs deletes
232 * are happening, so run/
233 * takeover/stop are not safe
235 struct gendisk *gendisk;
239 #define UNTIL_IOCTL 1
242 /* Superblock information */
247 int external; /* metadata is
248 * managed externally */
249 char metadata_type[17]; /* externally set*/
251 time64_t ctime, utime;
256 sector_t dev_sectors; /* used size of
257 * component devices */
258 sector_t array_sectors; /* exported array size */
259 int external_size; /* size managed
262 /* If the last 'event' was simply a clean->dirty transition, and
263 * we didn't write it to the spares, then it is safe and simple
264 * to just decrement the event count on a dirty->clean transition.
265 * So we record that possibility here.
267 int can_decrease_events;
271 /* If the array is being reshaped, we need to record the
272 * new shape and an indication of where we are up to.
273 * This is written to the superblock.
274 * If reshape_position is MaxSector, then no reshape is happening (yet).
276 sector_t reshape_position;
277 int delta_disks, new_level, new_layout;
278 int new_chunk_sectors;
279 int reshape_backwards;
281 struct md_thread *thread; /* management thread */
282 struct md_thread *sync_thread; /* doing resync or reconstruct */
284 /* 'last_sync_action' is initialized to "none". It is set when a
285 * sync operation (i.e "data-check", "requested-resync", "resync",
286 * "recovery", or "reshape") is started. It holds this value even
287 * when the sync thread is "frozen" (interrupted) or "idle" (stopped
288 * or finished). It is overwritten when a new sync operation is begun.
290 char *last_sync_action;
291 sector_t curr_resync; /* last block scheduled */
292 /* As resync requests can complete out of order, we cannot easily track
293 * how much resync has been completed. So we occasionally pause until
294 * everything completes, then set curr_resync_completed to curr_resync.
295 * As such it may be well behind the real resync mark, but it is a value
298 sector_t curr_resync_completed;
299 unsigned long resync_mark; /* a recent timestamp */
300 sector_t resync_mark_cnt;/* blocks written at resync_mark */
301 sector_t curr_mark_cnt; /* blocks scheduled now */
303 sector_t resync_max_sectors; /* may be set by personality */
305 atomic64_t resync_mismatches; /* count of sectors where
306 * parity/replica mismatch found
309 /* allow user-space to request suspension of IO to regions of the array */
312 /* if zero, use the system-wide default */
316 /* resync even though the same disks are shared among md-devices */
319 int ok_start_degraded;
321 unsigned long recovery;
322 /* If a RAID personality determines that recovery (of a particular
323 * device) will fail due to a read error on the source device, it
324 * takes a copy of this number and does not attempt recovery again
325 * until this number changes.
327 int recovery_disabled;
329 int in_sync; /* know to not need resync */
330 /* 'open_mutex' avoids races between 'md_open' and 'do_md_stop', so
331 * that we are never stopping an array while it is open.
332 * 'reconfig_mutex' protects all other reconfiguration.
333 * These locks are separate due to conflicting interactions
334 * with bdev->bd_mutex.
336 * reconfig_mutex -> bd_mutex : e.g. do_md_run -> revalidate_disk
337 * bd_mutex -> open_mutex: e.g. __blkdev_get -> md_open
339 struct mutex open_mutex;
340 struct mutex reconfig_mutex;
341 atomic_t active; /* general refcount */
342 atomic_t openers; /* number of active opens */
344 int changed; /* True if we might need to
345 * reread partition info */
346 int degraded; /* whether md should consider
350 atomic_t recovery_active; /* blocks scheduled, but not written */
351 wait_queue_head_t recovery_wait;
352 sector_t recovery_cp;
353 sector_t resync_min; /* user requested sync
355 sector_t resync_max; /* resync should pause
356 * when it gets here */
358 struct kernfs_node *sysfs_state; /* handle for 'array_state'
361 struct kernfs_node *sysfs_action; /* handle for 'sync_action' */
363 struct work_struct del_work; /* used for delayed sysfs removal */
366 * flush_bio transition from NULL to !NULL
367 * rdev superblocks, events
368 * clearing MD_CHANGE_*
369 * in_sync - and related safemode and MD_CHANGE changes
370 * pers (also protected by reconfig_mutex and pending IO).
372 * clearing ->bitmap_info.file
373 * changing ->resync_{min,max}
374 * setting MD_RECOVERY_RUNNING (which interacts with resync_{min,max})
377 wait_queue_head_t sb_wait; /* for waiting on superblock updates */
378 atomic_t pending_writes; /* number of active superblock writes */
380 unsigned int safemode; /* if set, update "clean" superblock
381 * when no writes pending.
383 unsigned int safemode_delay;
384 struct timer_list safemode_timer;
385 atomic_t writes_pending;
386 struct request_queue *queue; /* for plugging ... */
388 struct bitmap *bitmap; /* the bitmap for the device */
390 struct file *file; /* the bitmap file */
391 loff_t offset; /* offset from superblock of
392 * start of bitmap. May be
393 * negative, but not '0'
394 * For external metadata, offset
395 * from start of device.
397 unsigned long space; /* space available at this offset */
398 loff_t default_offset; /* this is the offset to use when
399 * hot-adding a bitmap. It should
400 * eventually be settable by sysfs.
402 unsigned long default_space; /* space available at
405 unsigned long chunksize;
406 unsigned long daemon_sleep; /* how many jiffies between updates? */
407 unsigned long max_write_behind; /* write-behind mode */
409 int nodes; /* Maximum number of nodes in the cluster */
410 char cluster_name[64]; /* Name of the cluster */
413 atomic_t max_corr_read_errors; /* max read retries */
414 struct list_head all_mddevs;
416 struct attribute_group *to_remove;
418 struct bio_set *bio_set;
420 /* Generic flush handling.
421 * The last to finish preflush schedules a worker to submit
422 * the rest of the request (without the REQ_PREFLUSH flag).
424 struct bio *flush_bio;
425 atomic_t flush_pending;
426 struct work_struct flush_work;
427 struct work_struct event_work; /* used by dm to report failure event */
428 void (*sync_super)(struct mddev *mddev, struct md_rdev *rdev);
429 struct md_cluster_info *cluster_info;
430 unsigned int good_device_nr; /* good device num within cluster raid */
433 enum recovery_flags {
435 * If neither SYNC or RESHAPE are set, then it is a recovery.
437 MD_RECOVERY_RUNNING, /* a thread is running, or about to be started */
438 MD_RECOVERY_SYNC, /* actually doing a resync, not a recovery */
439 MD_RECOVERY_RECOVER, /* doing recovery, or need to try it. */
440 MD_RECOVERY_INTR, /* resync needs to be aborted for some reason */
441 MD_RECOVERY_DONE, /* thread is done and is waiting to be reaped */
442 MD_RECOVERY_NEEDED, /* we might need to start a resync/recover */
443 MD_RECOVERY_REQUESTED, /* user-space has requested a sync (used with SYNC) */
444 MD_RECOVERY_CHECK, /* user-space request for check-only, no repair */
445 MD_RECOVERY_RESHAPE, /* A reshape is happening */
446 MD_RECOVERY_FROZEN, /* User request to abort, and not restart, any action */
447 MD_RECOVERY_ERROR, /* sync-action interrupted because io-error */
450 static inline int __must_check mddev_lock(struct mddev *mddev)
452 return mutex_lock_interruptible(&mddev->reconfig_mutex);
455 /* Sometimes we need to take the lock in a situation where
456 * failure due to interrupts is not acceptable.
458 static inline void mddev_lock_nointr(struct mddev *mddev)
460 mutex_lock(&mddev->reconfig_mutex);
463 static inline int mddev_is_locked(struct mddev *mddev)
465 return mutex_is_locked(&mddev->reconfig_mutex);
468 static inline int mddev_trylock(struct mddev *mddev)
470 return mutex_trylock(&mddev->reconfig_mutex);
472 extern void mddev_unlock(struct mddev *mddev);
474 static inline void md_sync_acct(struct block_device *bdev, unsigned long nr_sectors)
476 atomic_add(nr_sectors, &bdev->bd_contains->bd_disk->sync_io);
479 struct md_personality
483 struct list_head list;
484 struct module *owner;
485 void (*make_request)(struct mddev *mddev, struct bio *bio);
486 int (*run)(struct mddev *mddev);
487 void (*free)(struct mddev *mddev, void *priv);
488 void (*status)(struct seq_file *seq, struct mddev *mddev);
489 /* error_handler must set ->faulty and clear ->in_sync
490 * if appropriate, and should abort recovery if needed
492 void (*error_handler)(struct mddev *mddev, struct md_rdev *rdev);
493 int (*hot_add_disk) (struct mddev *mddev, struct md_rdev *rdev);
494 int (*hot_remove_disk) (struct mddev *mddev, struct md_rdev *rdev);
495 int (*spare_active) (struct mddev *mddev);
496 sector_t (*sync_request)(struct mddev *mddev, sector_t sector_nr, int *skipped);
497 int (*resize) (struct mddev *mddev, sector_t sectors);
498 sector_t (*size) (struct mddev *mddev, sector_t sectors, int raid_disks);
499 int (*check_reshape) (struct mddev *mddev);
500 int (*start_reshape) (struct mddev *mddev);
501 void (*finish_reshape) (struct mddev *mddev);
502 /* quiesce moves between quiescence states
504 * 1 - no new requests allowed
507 void (*quiesce) (struct mddev *mddev, int state);
508 /* takeover is used to transition an array from one
509 * personality to another. The new personality must be able
510 * to handle the data in the current layout.
511 * e.g. 2drive raid1 -> 2drive raid5
512 * ndrive raid5 -> degraded n+1drive raid6 with special layout
513 * If the takeover succeeds, a new 'private' structure is returned.
514 * This needs to be installed and then ->run used to activate the
517 void *(*takeover) (struct mddev *mddev);
518 /* congested implements bdi.congested_fn().
519 * Will not be called while array is 'suspended' */
520 int (*congested)(struct mddev *mddev, int bits);
523 struct md_sysfs_entry {
524 struct attribute attr;
525 ssize_t (*show)(struct mddev *, char *);
526 ssize_t (*store)(struct mddev *, const char *, size_t);
528 extern struct attribute_group md_bitmap_group;
530 static inline struct kernfs_node *sysfs_get_dirent_safe(struct kernfs_node *sd, char *name)
533 return sysfs_get_dirent(sd, name);
536 static inline void sysfs_notify_dirent_safe(struct kernfs_node *sd)
539 sysfs_notify_dirent(sd);
542 static inline char * mdname (struct mddev * mddev)
544 return mddev->gendisk ? mddev->gendisk->disk_name : "mdX";
547 static inline int sysfs_link_rdev(struct mddev *mddev, struct md_rdev *rdev)
550 if (!test_bit(Replacement, &rdev->flags) &&
551 !test_bit(Journal, &rdev->flags) &&
553 sprintf(nm, "rd%d", rdev->raid_disk);
554 return sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
559 static inline void sysfs_unlink_rdev(struct mddev *mddev, struct md_rdev *rdev)
562 if (!test_bit(Replacement, &rdev->flags) &&
563 !test_bit(Journal, &rdev->flags) &&
565 sprintf(nm, "rd%d", rdev->raid_disk);
566 sysfs_remove_link(&mddev->kobj, nm);
571 * iterates through some rdev ringlist. It's safe to remove the
572 * current 'rdev'. Dont touch 'tmp' though.
574 #define rdev_for_each_list(rdev, tmp, head) \
575 list_for_each_entry_safe(rdev, tmp, head, same_set)
578 * iterates through the 'same array disks' ringlist
580 #define rdev_for_each(rdev, mddev) \
581 list_for_each_entry(rdev, &((mddev)->disks), same_set)
583 #define rdev_for_each_safe(rdev, tmp, mddev) \
584 list_for_each_entry_safe(rdev, tmp, &((mddev)->disks), same_set)
586 #define rdev_for_each_rcu(rdev, mddev) \
587 list_for_each_entry_rcu(rdev, &((mddev)->disks), same_set)
590 void (*run) (struct md_thread *thread);
592 wait_queue_head_t wqueue;
594 struct task_struct *tsk;
595 unsigned long timeout;
599 #define THREAD_WAKEUP 0
601 static inline void safe_put_page(struct page *p)
606 extern int register_md_personality(struct md_personality *p);
607 extern int unregister_md_personality(struct md_personality *p);
608 extern int register_md_cluster_operations(struct md_cluster_operations *ops,
609 struct module *module);
610 extern int unregister_md_cluster_operations(void);
611 extern int md_setup_cluster(struct mddev *mddev, int nodes);
612 extern void md_cluster_stop(struct mddev *mddev);
613 extern struct md_thread *md_register_thread(
614 void (*run)(struct md_thread *thread),
617 extern void md_unregister_thread(struct md_thread **threadp);
618 extern void md_wakeup_thread(struct md_thread *thread);
619 extern void md_check_recovery(struct mddev *mddev);
620 extern void md_reap_sync_thread(struct mddev *mddev);
621 extern void md_write_start(struct mddev *mddev, struct bio *bi);
622 extern void md_write_end(struct mddev *mddev);
623 extern void md_done_sync(struct mddev *mddev, int blocks, int ok);
624 extern void md_error(struct mddev *mddev, struct md_rdev *rdev);
625 extern void md_finish_reshape(struct mddev *mddev);
627 extern int mddev_congested(struct mddev *mddev, int bits);
628 extern void md_flush_request(struct mddev *mddev, struct bio *bio);
629 extern void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
630 sector_t sector, int size, struct page *page);
631 extern void md_super_wait(struct mddev *mddev);
632 extern int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
633 struct page *page, int op, int op_flags,
635 extern void md_do_sync(struct md_thread *thread);
636 extern void md_new_event(struct mddev *mddev);
637 extern int md_allow_write(struct mddev *mddev);
638 extern void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev);
639 extern void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors);
640 extern int md_check_no_bitmap(struct mddev *mddev);
641 extern int md_integrity_register(struct mddev *mddev);
642 extern int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev);
643 extern int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale);
645 extern void mddev_init(struct mddev *mddev);
646 extern int md_run(struct mddev *mddev);
647 extern void md_stop(struct mddev *mddev);
648 extern void md_stop_writes(struct mddev *mddev);
649 extern int md_rdev_init(struct md_rdev *rdev);
650 extern void md_rdev_clear(struct md_rdev *rdev);
652 extern void mddev_suspend(struct mddev *mddev);
653 extern void mddev_resume(struct mddev *mddev);
654 extern struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
655 struct mddev *mddev);
656 extern struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
657 struct mddev *mddev);
659 extern void md_unplug(struct blk_plug_cb *cb, bool from_schedule);
660 extern void md_reload_sb(struct mddev *mddev, int raid_disk);
661 extern void md_update_sb(struct mddev *mddev, int force);
662 extern void md_kick_rdev_from_array(struct md_rdev * rdev);
663 struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr);
664 static inline int mddev_check_plugged(struct mddev *mddev)
666 return !!blk_check_plugged(md_unplug, mddev,
667 sizeof(struct blk_plug_cb));
670 static inline void rdev_dec_pending(struct md_rdev *rdev, struct mddev *mddev)
672 int faulty = test_bit(Faulty, &rdev->flags);
673 if (atomic_dec_and_test(&rdev->nr_pending) && faulty) {
674 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
675 md_wakeup_thread(mddev->thread);
679 extern struct md_cluster_operations *md_cluster_ops;
680 static inline int mddev_is_clustered(struct mddev *mddev)
682 return mddev->cluster_info && mddev->bitmap_info.nodes > 1;
684 #endif /* _MD_MD_H */