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/kobject.h>
20 #include <linux/list.h>
22 #include <linux/mutex.h>
23 #include <linux/timer.h>
24 #include <linux/wait.h>
25 #include <linux/workqueue.h>
27 #define MaxSector (~(sector_t)0)
29 /* Bad block numbers are stored sorted in a single page.
30 * 64bits is used for each block or extent.
31 * 54 bits are sector number, 9 bits are extent size,
32 * 1 bit is an 'acknowledged' flag.
34 #define MD_MAX_BADBLOCKS (PAGE_SIZE/8)
37 * MD's 'extended' device
40 struct list_head same_set; /* RAID devices within the same set */
42 sector_t sectors; /* Device size (in 512bytes sectors) */
43 struct mddev *mddev; /* RAID array if running */
44 int last_events; /* IO event timestamp */
47 * If meta_bdev is non-NULL, it means that a separate device is
48 * being used to store the metadata (superblock/bitmap) which
49 * would otherwise be contained on the same device as the data (bdev).
51 struct block_device *meta_bdev;
52 struct block_device *bdev; /* block device handle */
54 struct page *sb_page, *bb_page;
57 sector_t data_offset; /* start of data in array */
58 sector_t new_data_offset;/* only relevant while reshaping */
59 sector_t sb_start; /* offset of the super block (in 512byte sectors) */
60 int sb_size; /* bytes in the superblock */
61 int preferred_minor; /* autorun support */
65 /* A device can be in one of three states based on two flags:
66 * Not working: faulty==1 in_sync==0
67 * Fully working: faulty==0 in_sync==1
70 * faulty==0 in_sync==0
72 * It can never have faulty==1, in_sync==1
73 * This reduces the burden of testing multiple flags in many cases
76 unsigned long flags; /* bit set of 'enum flag_bits' bits. */
77 wait_queue_head_t blocked_wait;
79 int desc_nr; /* descriptor index in the superblock */
80 int raid_disk; /* role of device in array */
81 int new_raid_disk; /* role that the device will have in
82 * the array after a level-change completes.
84 int saved_raid_disk; /* role that device used to have in the
85 * array and could again if we did a partial
86 * resync from the bitmap
88 sector_t recovery_offset;/* If this device has been partially
89 * recovered, this is where we were
93 atomic_t nr_pending; /* number of pending requests.
94 * only maintained for arrays that
97 atomic_t read_errors; /* number of consecutive read errors that
98 * we have tried to ignore.
100 struct timespec last_read_error; /* monotonic time since our
103 atomic_t corrected_errors; /* number of corrected read errors,
104 * for reporting to userspace and storing
107 struct work_struct del_work; /* used for delayed sysfs removal */
109 struct sysfs_dirent *sysfs_state; /* handle for 'state'
113 int count; /* count of bad blocks */
114 int unacked_exist; /* there probably are unacknowledged
115 * bad blocks. This is only cleared
116 * when a read discovers none
118 int shift; /* shift from sectors to block size
119 * a -ve shift means badblocks are
121 u64 *page; /* badblock list */
126 sector_t size; /* in sectors */
130 Faulty, /* device is known to have a fault */
131 In_sync, /* device is in_sync with rest of array */
132 Unmerged, /* device is being added to array and should
133 * be considerred for bvec_merge_fn but not
136 WriteMostly, /* Avoid reading if at all possible */
137 AutoDetected, /* added by auto-detect */
138 Blocked, /* An error occurred but has not yet
139 * been acknowledged by the metadata
140 * handler, so don't allow writes
141 * until it is cleared */
142 WriteErrorSeen, /* A write error has been seen on this
145 FaultRecorded, /* Intermediate state for clearing
146 * Blocked. The Fault is/will-be
147 * recorded in the metadata, but that
148 * metadata hasn't been stored safely
151 BlockedBadBlocks, /* A writer is blocked because they
152 * found an unacknowledged bad-block.
153 * This can safely be cleared at any
154 * time, and the writer will re-check.
155 * It may be set at any time, and at
156 * worst the writer will timeout and
157 * re-check. So setting it as
158 * accurately as possible is good, but
159 * not absolutely critical.
161 WantReplacement, /* This device is a candidate to be
162 * hot-replaced, either because it has
163 * reported some faults, or because
164 * of explicit request.
166 Replacement, /* This device is a replacement for
167 * a want_replacement device with same
172 #define BB_LEN_MASK (0x00000000000001FFULL)
173 #define BB_OFFSET_MASK (0x7FFFFFFFFFFFFE00ULL)
174 #define BB_ACK_MASK (0x8000000000000000ULL)
175 #define BB_MAX_LEN 512
176 #define BB_OFFSET(x) (((x) & BB_OFFSET_MASK) >> 9)
177 #define BB_LEN(x) (((x) & BB_LEN_MASK) + 1)
178 #define BB_ACK(x) (!!((x) & BB_ACK_MASK))
179 #define BB_MAKE(a, l, ack) (((a)<<9) | ((l)-1) | ((u64)(!!(ack)) << 63))
181 extern int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
182 sector_t *first_bad, int *bad_sectors);
183 static inline int is_badblock(struct md_rdev *rdev, sector_t s, int sectors,
184 sector_t *first_bad, int *bad_sectors)
186 if (unlikely(rdev->badblocks.count)) {
187 int rv = md_is_badblock(&rdev->badblocks, rdev->data_offset + s,
189 first_bad, bad_sectors);
191 *first_bad -= rdev->data_offset;
196 extern int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
198 extern int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
200 extern void md_ack_all_badblocks(struct badblocks *bb);
204 struct md_personality *pers;
207 struct list_head disks;
209 #define MD_CHANGE_DEVS 0 /* Some device status has changed */
210 #define MD_CHANGE_CLEAN 1 /* transition to or from 'clean' */
211 #define MD_CHANGE_PENDING 2 /* switch from 'clean' to 'active' in progress */
212 #define MD_UPDATE_SB_FLAGS (1 | 2 | 4) /* If these are set, md_update_sb needed */
213 #define MD_ARRAY_FIRST_USE 3 /* First use of array, needs initialization */
218 int sysfs_active; /* set when sysfs deletes
219 * are happening, so run/
220 * takeover/stop are not safe
222 int ready; /* See when safe to pass
223 * IO requests down */
224 struct gendisk *gendisk;
228 #define UNTIL_IOCTL 1
231 /* Superblock information */
236 int external; /* metadata is
237 * managed externally */
238 char metadata_type[17]; /* externally set*/
245 sector_t dev_sectors; /* used size of
246 * component devices */
247 sector_t array_sectors; /* exported array size */
248 int external_size; /* size managed
251 /* If the last 'event' was simply a clean->dirty transition, and
252 * we didn't write it to the spares, then it is safe and simple
253 * to just decrement the event count on a dirty->clean transition.
254 * So we record that possibility here.
256 int can_decrease_events;
260 /* If the array is being reshaped, we need to record the
261 * new shape and an indication of where we are up to.
262 * This is written to the superblock.
263 * If reshape_position is MaxSector, then no reshape is happening (yet).
265 sector_t reshape_position;
266 int delta_disks, new_level, new_layout;
267 int new_chunk_sectors;
268 int reshape_backwards;
270 struct md_thread *thread; /* management thread */
271 struct md_thread *sync_thread; /* doing resync or reconstruct */
273 /* 'last_sync_action' is initialized to "none". It is set when a
274 * sync operation (i.e "data-check", "requested-resync", "resync",
275 * "recovery", or "reshape") is started. It holds this value even
276 * when the sync thread is "frozen" (interrupted) or "idle" (stopped
277 * or finished). It is overwritten when a new sync operation is begun.
279 char *last_sync_action;
280 sector_t curr_resync; /* last block scheduled */
281 /* As resync requests can complete out of order, we cannot easily track
282 * how much resync has been completed. So we occasionally pause until
283 * everything completes, then set curr_resync_completed to curr_resync.
284 * As such it may be well behind the real resync mark, but it is a value
287 sector_t curr_resync_completed;
288 unsigned long resync_mark; /* a recent timestamp */
289 sector_t resync_mark_cnt;/* blocks written at resync_mark */
290 sector_t curr_mark_cnt; /* blocks scheduled now */
292 sector_t resync_max_sectors; /* may be set by personality */
294 atomic64_t resync_mismatches; /* count of sectors where
295 * parity/replica mismatch found
298 /* allow user-space to request suspension of IO to regions of the array */
301 /* if zero, use the system-wide default */
305 /* resync even though the same disks are shared among md-devices */
308 int ok_start_degraded;
309 /* recovery/resync flags
310 * NEEDED: we might need to start a resync/recover
311 * RUNNING: a thread is running, or about to be started
312 * SYNC: actually doing a resync, not a recovery
313 * RECOVER: doing recovery, or need to try it.
314 * INTR: resync needs to be aborted for some reason
315 * DONE: thread is done and is waiting to be reaped
316 * REQUEST: user-space has requested a sync (used with SYNC)
317 * CHECK: user-space request for check-only, no repair
318 * RESHAPE: A reshape is happening
319 * ERROR: sync-action interrupted because io-error
321 * If neither SYNC or RESHAPE are set, then it is a recovery.
323 #define MD_RECOVERY_RUNNING 0
324 #define MD_RECOVERY_SYNC 1
325 #define MD_RECOVERY_RECOVER 2
326 #define MD_RECOVERY_INTR 3
327 #define MD_RECOVERY_DONE 4
328 #define MD_RECOVERY_NEEDED 5
329 #define MD_RECOVERY_REQUESTED 6
330 #define MD_RECOVERY_CHECK 7
331 #define MD_RECOVERY_RESHAPE 8
332 #define MD_RECOVERY_FROZEN 9
333 #define MD_RECOVERY_ERROR 10
335 unsigned long recovery;
336 /* If a RAID personality determines that recovery (of a particular
337 * device) will fail due to a read error on the source device, it
338 * takes a copy of this number and does not attempt recovery again
339 * until this number changes.
341 int recovery_disabled;
343 int in_sync; /* know to not need resync */
344 /* 'open_mutex' avoids races between 'md_open' and 'do_md_stop', so
345 * that we are never stopping an array while it is open.
346 * 'reconfig_mutex' protects all other reconfiguration.
347 * These locks are separate due to conflicting interactions
348 * with bdev->bd_mutex.
350 * reconfig_mutex -> bd_mutex : e.g. do_md_run -> revalidate_disk
351 * bd_mutex -> open_mutex: e.g. __blkdev_get -> md_open
353 struct mutex open_mutex;
354 struct mutex reconfig_mutex;
355 atomic_t active; /* general refcount */
356 atomic_t openers; /* number of active opens */
358 int changed; /* True if we might need to
359 * reread partition info */
360 int degraded; /* whether md should consider
363 int merge_check_needed; /* at least one
368 atomic_t recovery_active; /* blocks scheduled, but not written */
369 wait_queue_head_t recovery_wait;
370 sector_t recovery_cp;
371 sector_t resync_min; /* user requested sync
373 sector_t resync_max; /* resync should pause
374 * when it gets here */
376 struct sysfs_dirent *sysfs_state; /* handle for 'array_state'
379 struct sysfs_dirent *sysfs_action; /* handle for 'sync_action' */
381 struct work_struct del_work; /* used for delayed sysfs removal */
383 spinlock_t write_lock;
384 wait_queue_head_t sb_wait; /* for waiting on superblock updates */
385 atomic_t pending_writes; /* number of active superblock writes */
387 unsigned int safemode; /* if set, update "clean" superblock
388 * when no writes pending.
390 unsigned int safemode_delay;
391 struct timer_list safemode_timer;
392 atomic_t writes_pending;
393 struct request_queue *queue; /* for plugging ... */
395 struct bitmap *bitmap; /* the bitmap for the device */
397 struct file *file; /* the bitmap file */
398 loff_t offset; /* offset from superblock of
399 * start of bitmap. May be
400 * negative, but not '0'
401 * For external metadata, offset
402 * from start of device.
404 unsigned long space; /* space available at this offset */
405 loff_t default_offset; /* this is the offset to use when
406 * hot-adding a bitmap. It should
407 * eventually be settable by sysfs.
409 unsigned long default_space; /* space available at
412 unsigned long chunksize;
413 unsigned long daemon_sleep; /* how many jiffies between updates? */
414 unsigned long max_write_behind; /* write-behind mode */
418 atomic_t max_corr_read_errors; /* max read retries */
419 struct list_head all_mddevs;
421 struct attribute_group *to_remove;
423 struct bio_set *bio_set;
425 /* Generic flush handling.
426 * The last to finish preflush schedules a worker to submit
427 * the rest of the request (without the REQ_FLUSH flag).
429 struct bio *flush_bio;
430 atomic_t flush_pending;
431 struct work_struct flush_work;
432 struct work_struct event_work; /* used by dm to report failure event */
433 void (*sync_super)(struct mddev *mddev, struct md_rdev *rdev);
437 static inline void rdev_dec_pending(struct md_rdev *rdev, struct mddev *mddev)
439 int faulty = test_bit(Faulty, &rdev->flags);
440 if (atomic_dec_and_test(&rdev->nr_pending) && faulty)
441 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
444 static inline void md_sync_acct(struct block_device *bdev, unsigned long nr_sectors)
446 atomic_add(nr_sectors, &bdev->bd_contains->bd_disk->sync_io);
449 struct md_personality
453 struct list_head list;
454 struct module *owner;
455 void (*make_request)(struct mddev *mddev, struct bio *bio);
456 int (*run)(struct mddev *mddev);
457 int (*stop)(struct mddev *mddev);
458 void (*status)(struct seq_file *seq, struct mddev *mddev);
459 /* error_handler must set ->faulty and clear ->in_sync
460 * if appropriate, and should abort recovery if needed
462 void (*error_handler)(struct mddev *mddev, struct md_rdev *rdev);
463 int (*hot_add_disk) (struct mddev *mddev, struct md_rdev *rdev);
464 int (*hot_remove_disk) (struct mddev *mddev, struct md_rdev *rdev);
465 int (*spare_active) (struct mddev *mddev);
466 sector_t (*sync_request)(struct mddev *mddev, sector_t sector_nr, int *skipped, int go_faster);
467 int (*resize) (struct mddev *mddev, sector_t sectors);
468 sector_t (*size) (struct mddev *mddev, sector_t sectors, int raid_disks);
469 int (*check_reshape) (struct mddev *mddev);
470 int (*start_reshape) (struct mddev *mddev);
471 void (*finish_reshape) (struct mddev *mddev);
472 /* quiesce moves between quiescence states
474 * 1 - no new requests allowed
477 void (*quiesce) (struct mddev *mddev, int state);
478 /* takeover is used to transition an array from one
479 * personality to another. The new personality must be able
480 * to handle the data in the current layout.
481 * e.g. 2drive raid1 -> 2drive raid5
482 * ndrive raid5 -> degraded n+1drive raid6 with special layout
483 * If the takeover succeeds, a new 'private' structure is returned.
484 * This needs to be installed and then ->run used to activate the
487 void *(*takeover) (struct mddev *mddev);
491 struct md_sysfs_entry {
492 struct attribute attr;
493 ssize_t (*show)(struct mddev *, char *);
494 ssize_t (*store)(struct mddev *, const char *, size_t);
496 extern struct attribute_group md_bitmap_group;
498 static inline struct sysfs_dirent *sysfs_get_dirent_safe(struct sysfs_dirent *sd, char *name)
501 return sysfs_get_dirent(sd, NULL, name);
504 static inline void sysfs_notify_dirent_safe(struct sysfs_dirent *sd)
507 sysfs_notify_dirent(sd);
510 static inline char * mdname (struct mddev * mddev)
512 return mddev->gendisk ? mddev->gendisk->disk_name : "mdX";
515 static inline int sysfs_link_rdev(struct mddev *mddev, struct md_rdev *rdev)
518 if (!test_bit(Replacement, &rdev->flags) && mddev->kobj.sd) {
519 sprintf(nm, "rd%d", rdev->raid_disk);
520 return sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
525 static inline void sysfs_unlink_rdev(struct mddev *mddev, struct md_rdev *rdev)
528 if (!test_bit(Replacement, &rdev->flags) && mddev->kobj.sd) {
529 sprintf(nm, "rd%d", rdev->raid_disk);
530 sysfs_remove_link(&mddev->kobj, nm);
535 * iterates through some rdev ringlist. It's safe to remove the
536 * current 'rdev'. Dont touch 'tmp' though.
538 #define rdev_for_each_list(rdev, tmp, head) \
539 list_for_each_entry_safe(rdev, tmp, head, same_set)
542 * iterates through the 'same array disks' ringlist
544 #define rdev_for_each(rdev, mddev) \
545 list_for_each_entry(rdev, &((mddev)->disks), same_set)
547 #define rdev_for_each_safe(rdev, tmp, mddev) \
548 list_for_each_entry_safe(rdev, tmp, &((mddev)->disks), same_set)
550 #define rdev_for_each_rcu(rdev, mddev) \
551 list_for_each_entry_rcu(rdev, &((mddev)->disks), same_set)
554 void (*run) (struct md_thread *thread);
556 wait_queue_head_t wqueue;
558 struct task_struct *tsk;
559 unsigned long timeout;
563 #define THREAD_WAKEUP 0
565 static inline void safe_put_page(struct page *p)
570 extern int register_md_personality(struct md_personality *p);
571 extern int unregister_md_personality(struct md_personality *p);
572 extern struct md_thread *md_register_thread(
573 void (*run)(struct md_thread *thread),
576 extern void md_unregister_thread(struct md_thread **threadp);
577 extern void md_wakeup_thread(struct md_thread *thread);
578 extern void md_check_recovery(struct mddev *mddev);
579 extern void md_reap_sync_thread(struct mddev *mddev);
580 extern void md_write_start(struct mddev *mddev, struct bio *bi);
581 extern void md_write_end(struct mddev *mddev);
582 extern void md_done_sync(struct mddev *mddev, int blocks, int ok);
583 extern void md_error(struct mddev *mddev, struct md_rdev *rdev);
584 extern void md_finish_reshape(struct mddev *mddev);
586 extern int mddev_congested(struct mddev *mddev, int bits);
587 extern void md_flush_request(struct mddev *mddev, struct bio *bio);
588 extern void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
589 sector_t sector, int size, struct page *page);
590 extern void md_super_wait(struct mddev *mddev);
591 extern int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
592 struct page *page, int rw, bool metadata_op);
593 extern void md_do_sync(struct md_thread *thread);
594 extern void md_new_event(struct mddev *mddev);
595 extern int md_allow_write(struct mddev *mddev);
596 extern void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev);
597 extern void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors);
598 extern int md_check_no_bitmap(struct mddev *mddev);
599 extern int md_integrity_register(struct mddev *mddev);
600 extern void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev);
601 extern int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale);
602 extern void restore_bitmap_write_access(struct file *file);
604 extern void mddev_init(struct mddev *mddev);
605 extern int md_run(struct mddev *mddev);
606 extern void md_stop(struct mddev *mddev);
607 extern void md_stop_writes(struct mddev *mddev);
608 extern int md_rdev_init(struct md_rdev *rdev);
609 extern void md_rdev_clear(struct md_rdev *rdev);
611 extern void mddev_suspend(struct mddev *mddev);
612 extern void mddev_resume(struct mddev *mddev);
613 extern struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
614 struct mddev *mddev);
615 extern struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
616 struct mddev *mddev);
617 extern void md_trim_bio(struct bio *bio, int offset, int size);
619 extern void md_unplug(struct blk_plug_cb *cb, bool from_schedule);
620 static inline int mddev_check_plugged(struct mddev *mddev)
622 return !!blk_check_plugged(md_unplug, mddev,
623 sizeof(struct blk_plug_cb));
625 #endif /* _MD_MD_H */