Linux Kernel  3.7.1
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
md.h
Go to the documentation of this file.
1 /*
2  md.h : kernel internal structure of the Linux MD driver
3  Copyright (C) 1996-98 Ingo Molnar, Gadi Oxman
4 
5  This program is free software; you can redistribute it and/or modify
6  it under the terms of the GNU General Public License as published by
7  the Free Software Foundation; either version 2, or (at your option)
8  any later version.
9 
10  You should have received a copy of the GNU General Public License
11  (for example /usr/src/linux/COPYING); if not, write to the Free
12  Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
13 */
14 
15 #ifndef _MD_MD_H
16 #define _MD_MD_H
17 
18 #include <linux/blkdev.h>
19 #include <linux/kobject.h>
20 #include <linux/list.h>
21 #include <linux/mm.h>
22 #include <linux/mutex.h>
23 #include <linux/timer.h>
24 #include <linux/wait.h>
25 #include <linux/workqueue.h>
26 
27 #define MaxSector (~(sector_t)0)
28 
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.
33  */
34 #define MD_MAX_BADBLOCKS (PAGE_SIZE/8)
35 
36 /*
37  * MD's 'extended' device
38  */
39 struct md_rdev {
40  struct list_head same_set; /* RAID devices within the same set */
41 
42  sector_t sectors; /* Device size (in 512bytes sectors) */
43  struct mddev *mddev; /* RAID array if running */
44  int last_events; /* IO event timestamp */
45 
46  /*
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).
50  */
52  struct block_device *bdev; /* block device handle */
53 
54  struct page *sb_page, *bb_page;
55  int sb_loaded;
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 */
62 
63  struct kobject kobj;
64 
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
68  * Working, but not
69  * in sync with array
70  * faulty==0 in_sync==0
71  *
72  * It can never have faulty==1, in_sync==1
73  * This reduces the burden of testing multiple flags in many cases
74  */
75 
76  unsigned long flags; /* bit set of 'enum flag_bits' bits. */
78 
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.
83  */
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
87  */
88  sector_t recovery_offset;/* If this device has been partially
89  * recovered, this is where we were
90  * up to.
91  */
92 
93  atomic_t nr_pending; /* number of pending requests.
94  * only maintained for arrays that
95  * support hot removal
96  */
97  atomic_t read_errors; /* number of consecutive read errors that
98  * we have tried to ignore.
99  */
100  struct timespec last_read_error; /* monotonic time since our
101  * last read error
102  */
103  atomic_t corrected_errors; /* number of corrected read errors,
104  * for reporting to userspace and storing
105  * in superblock.
106  */
107  struct work_struct del_work; /* used for delayed sysfs removal */
108 
109  struct sysfs_dirent *sysfs_state; /* handle for 'state'
110  * sysfs entry */
111 
112  struct badblocks {
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
117  */
118  int shift; /* shift from sectors to block size
119  * a -ve shift means badblocks are
120  * disabled.*/
121  u64 *page; /* badblock list */
122  int changed;
124 
126  sector_t size; /* in sectors */
127  } badblocks;
128 };
129 enum flag_bits {
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
134  * yet for actual IO
135  */
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
143  * device
144  */
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
149  * on disk yet.
150  */
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.
160  */
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.
165  */
166  Replacement, /* This device is a replacement for
167  * a want_replacement device with same
168  * raid_disk number.
169  */
170 };
171 
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))
180 
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)
185 {
186  if (unlikely(rdev->badblocks.count)) {
187  int rv = md_is_badblock(&rdev->badblocks, rdev->data_offset + s,
188  sectors,
189  first_bad, bad_sectors);
190  if (rv)
191  *first_bad -= rdev->data_offset;
192  return rv;
193  }
194  return 0;
195 }
196 extern int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
197  int is_new);
198 extern int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
199  int is_new);
200 extern void md_ack_all_badblocks(struct badblocks *bb);
201 
202 struct mddev {
203  void *private;
206  int md_minor;
207  struct list_head disks;
208  unsigned long flags;
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_ARRAY_FIRST_USE 3 /* First use of array, needs initialization */
213 
216  int ro;
217  int sysfs_active; /* set when sysfs deletes
218  * are happening, so run/
219  * takeover/stop are not safe
220  */
221  int ready; /* See when safe to pass
222  * IO requests down */
223  struct gendisk *gendisk;
224 
225  struct kobject kobj;
227 #define UNTIL_IOCTL 1
228 #define UNTIL_STOP 2
229 
230  /* Superblock information */
235  int external; /* metadata is
236  * managed externally */
237  char metadata_type[17]; /* externally set*/
240  int level, layout;
241  char clevel[16];
244  sector_t dev_sectors; /* used size of
245  * component devices */
246  sector_t array_sectors; /* exported array size */
247  int external_size; /* size managed
248  * externally */
250  /* If the last 'event' was simply a clean->dirty transition, and
251  * we didn't write it to the spares, then it is safe and simple
252  * to just decrement the event count on a dirty->clean transition.
253  * So we record that possibility here.
254  */
256 
257  char uuid[16];
258 
259  /* If the array is being reshaped, we need to record the
260  * new shape and an indication of where we are up to.
261  * This is written to the superblock.
262  * If reshape_position is MaxSector, then no reshape is happening (yet).
263  */
268 
269  struct md_thread *thread; /* management thread */
270  struct md_thread *sync_thread; /* doing resync or reconstruct */
271  sector_t curr_resync; /* last block scheduled */
272  /* As resync requests can complete out of order, we cannot easily track
273  * how much resync has been completed. So we occasionally pause until
274  * everything completes, then set curr_resync_completed to curr_resync.
275  * As such it may be well behind the real resync mark, but it is a value
276  * we are certain of.
277  */
279  unsigned long resync_mark; /* a recent timestamp */
280  sector_t resync_mark_cnt;/* blocks written at resync_mark */
281  sector_t curr_mark_cnt; /* blocks scheduled now */
282 
283  sector_t resync_max_sectors; /* may be set by personality */
284 
285  atomic64_t resync_mismatches; /* count of sectors where
286  * parity/replica mismatch found
287  */
288 
289  /* allow user-space to request suspension of IO to regions of the array */
292  /* if zero, use the system-wide default */
295 
296  /* resync even though the same disks are shared among md-devices */
298 
300  /* recovery/resync flags
301  * NEEDED: we might need to start a resync/recover
302  * RUNNING: a thread is running, or about to be started
303  * SYNC: actually doing a resync, not a recovery
304  * RECOVER: doing recovery, or need to try it.
305  * INTR: resync needs to be aborted for some reason
306  * DONE: thread is done and is waiting to be reaped
307  * REQUEST: user-space has requested a sync (used with SYNC)
308  * CHECK: user-space request for check-only, no repair
309  * RESHAPE: A reshape is happening
310  *
311  * If neither SYNC or RESHAPE are set, then it is a recovery.
312  */
313 #define MD_RECOVERY_RUNNING 0
314 #define MD_RECOVERY_SYNC 1
315 #define MD_RECOVERY_RECOVER 2
316 #define MD_RECOVERY_INTR 3
317 #define MD_RECOVERY_DONE 4
318 #define MD_RECOVERY_NEEDED 5
319 #define MD_RECOVERY_REQUESTED 6
320 #define MD_RECOVERY_CHECK 7
321 #define MD_RECOVERY_RESHAPE 8
322 #define MD_RECOVERY_FROZEN 9
323 
324  unsigned long recovery;
325  /* If a RAID personality determines that recovery (of a particular
326  * device) will fail due to a read error on the source device, it
327  * takes a copy of this number and does not attempt recovery again
328  * until this number changes.
329  */
331 
332  int in_sync; /* know to not need resync */
333  /* 'open_mutex' avoids races between 'md_open' and 'do_md_stop', so
334  * that we are never stopping an array while it is open.
335  * 'reconfig_mutex' protects all other reconfiguration.
336  * These locks are separate due to conflicting interactions
337  * with bdev->bd_mutex.
338  * Lock ordering is:
339  * reconfig_mutex -> bd_mutex : e.g. do_md_run -> revalidate_disk
340  * bd_mutex -> open_mutex: e.g. __blkdev_get -> md_open
341  */
344  atomic_t active; /* general refcount */
345  atomic_t openers; /* number of active opens */
346 
347  int changed; /* True if we might need to
348  * reread partition info */
349  int degraded; /* whether md should consider
350  * adding a spare
351  */
352  int merge_check_needed; /* at least one
353  * member device
354  * has a
355  * merge_bvec_fn */
356 
357  atomic_t recovery_active; /* blocks scheduled, but not written */
360  sector_t resync_min; /* user requested sync
361  * starts here */
362  sector_t resync_max; /* resync should pause
363  * when it gets here */
364 
365  struct sysfs_dirent *sysfs_state; /* handle for 'array_state'
366  * file in sysfs.
367  */
368  struct sysfs_dirent *sysfs_action; /* handle for 'sync_action' */
369 
370  struct work_struct del_work; /* used for delayed sysfs removal */
371 
373  wait_queue_head_t sb_wait; /* for waiting on superblock updates */
374  atomic_t pending_writes; /* number of active superblock writes */
375 
376  unsigned int safemode; /* if set, update "clean" superblock
377  * when no writes pending.
378  */
379  unsigned int safemode_delay;
382  struct request_queue *queue; /* for plugging ... */
383 
384  struct bitmap *bitmap; /* the bitmap for the device */
385  struct {
386  struct file *file; /* the bitmap file */
387  loff_t offset; /* offset from superblock of
388  * start of bitmap. May be
389  * negative, but not '0'
390  * For external metadata, offset
391  * from start of device.
392  */
393  unsigned long space; /* space available at this offset */
394  loff_t default_offset; /* this is the offset to use when
395  * hot-adding a bitmap. It should
396  * eventually be settable by sysfs.
397  */
398  unsigned long default_space; /* space available at
399  * default offset */
400  struct mutex mutex;
401  unsigned long chunksize;
402  unsigned long daemon_sleep; /* how many jiffies between updates? */
403  unsigned long max_write_behind; /* write-behind mode */
404  int external;
405  } bitmap_info;
406 
407  atomic_t max_corr_read_errors; /* max read retries */
409 
411 
412  struct bio_set *bio_set;
413 
414  /* Generic flush handling.
415  * The last to finish preflush schedules a worker to submit
416  * the rest of the request (without the REQ_FLUSH flag).
417  */
418  struct bio *flush_bio;
421  struct work_struct event_work; /* used by dm to report failure event */
422  void (*sync_super)(struct mddev *mddev, struct md_rdev *rdev);
423 };
424 
425 
426 static inline void rdev_dec_pending(struct md_rdev *rdev, struct mddev *mddev)
427 {
428  int faulty = test_bit(Faulty, &rdev->flags);
429  if (atomic_dec_and_test(&rdev->nr_pending) && faulty)
431 }
432 
433 static inline void md_sync_acct(struct block_device *bdev, unsigned long nr_sectors)
434 {
435  atomic_add(nr_sectors, &bdev->bd_contains->bd_disk->sync_io);
436 }
437 
439 {
440  char *name;
441  int level;
442  struct list_head list;
443  struct module *owner;
444  void (*make_request)(struct mddev *mddev, struct bio *bio);
445  int (*run)(struct mddev *mddev);
446  int (*stop)(struct mddev *mddev);
447  void (*status)(struct seq_file *seq, struct mddev *mddev);
448  /* error_handler must set ->faulty and clear ->in_sync
449  * if appropriate, and should abort recovery if needed
450  */
451  void (*error_handler)(struct mddev *mddev, struct md_rdev *rdev);
452  int (*hot_add_disk) (struct mddev *mddev, struct md_rdev *rdev);
453  int (*hot_remove_disk) (struct mddev *mddev, struct md_rdev *rdev);
454  int (*spare_active) (struct mddev *mddev);
455  sector_t (*sync_request)(struct mddev *mddev, sector_t sector_nr, int *skipped, int go_faster);
458  int (*check_reshape) (struct mddev *mddev);
459  int (*start_reshape) (struct mddev *mddev);
461  /* quiesce moves between quiescence states
462  * 0 - fully active
463  * 1 - no new requests allowed
464  * others - reserved
465  */
466  void (*quiesce) (struct mddev *mddev, int state);
467  /* takeover is used to transition an array from one
468  * personality to another. The new personality must be able
469  * to handle the data in the current layout.
470  * e.g. 2drive raid1 -> 2drive raid5
471  * ndrive raid5 -> degraded n+1drive raid6 with special layout
472  * If the takeover succeeds, a new 'private' structure is returned.
473  * This needs to be installed and then ->run used to activate the
474  * array.
475  */
476  void *(*takeover) (struct mddev *mddev);
477 };
478 
479 
481  struct attribute attr;
482  ssize_t (*show)(struct mddev *, char *);
483  ssize_t (*store)(struct mddev *, const char *, size_t);
484 };
485 extern struct attribute_group md_bitmap_group;
486 
487 static inline struct sysfs_dirent *sysfs_get_dirent_safe(struct sysfs_dirent *sd, char *name)
488 {
489  if (sd)
490  return sysfs_get_dirent(sd, NULL, name);
491  return sd;
492 }
493 static inline void sysfs_notify_dirent_safe(struct sysfs_dirent *sd)
494 {
495  if (sd)
497 }
498 
499 static inline char * mdname (struct mddev * mddev)
500 {
501  return mddev->gendisk ? mddev->gendisk->disk_name : "mdX";
502 }
503 
504 static inline int sysfs_link_rdev(struct mddev *mddev, struct md_rdev *rdev)
505 {
506  char nm[20];
507  if (!test_bit(Replacement, &rdev->flags)) {
508  sprintf(nm, "rd%d", rdev->raid_disk);
509  return sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
510  } else
511  return 0;
512 }
513 
514 static inline void sysfs_unlink_rdev(struct mddev *mddev, struct md_rdev *rdev)
515 {
516  char nm[20];
517  if (!test_bit(Replacement, &rdev->flags)) {
518  sprintf(nm, "rd%d", rdev->raid_disk);
519  sysfs_remove_link(&mddev->kobj, nm);
520  }
521 }
522 
523 /*
524  * iterates through some rdev ringlist. It's safe to remove the
525  * current 'rdev'. Dont touch 'tmp' though.
526  */
527 #define rdev_for_each_list(rdev, tmp, head) \
528  list_for_each_entry_safe(rdev, tmp, head, same_set)
529 
530 /*
531  * iterates through the 'same array disks' ringlist
532  */
533 #define rdev_for_each(rdev, mddev) \
534  list_for_each_entry(rdev, &((mddev)->disks), same_set)
535 
536 #define rdev_for_each_safe(rdev, tmp, mddev) \
537  list_for_each_entry_safe(rdev, tmp, &((mddev)->disks), same_set)
538 
539 #define rdev_for_each_rcu(rdev, mddev) \
540  list_for_each_entry_rcu(rdev, &((mddev)->disks), same_set)
541 
542 struct md_thread {
543  void (*run) (struct md_thread *thread);
544  struct mddev *mddev;
546  unsigned long flags;
547  struct task_struct *tsk;
548  unsigned long timeout;
549  void *private;
550 };
551 
552 #define THREAD_WAKEUP 0
553 
554 #define __wait_event_lock_irq(wq, condition, lock, cmd) \
555 do { \
556  wait_queue_t __wait; \
557  init_waitqueue_entry(&__wait, current); \
558  \
559  add_wait_queue(&wq, &__wait); \
560  for (;;) { \
561  set_current_state(TASK_UNINTERRUPTIBLE); \
562  if (condition) \
563  break; \
564  spin_unlock_irq(&lock); \
565  cmd; \
566  schedule(); \
567  spin_lock_irq(&lock); \
568  } \
569  current->state = TASK_RUNNING; \
570  remove_wait_queue(&wq, &__wait); \
571 } while (0)
572 
573 #define wait_event_lock_irq(wq, condition, lock, cmd) \
574 do { \
575  if (condition) \
576  break; \
577  __wait_event_lock_irq(wq, condition, lock, cmd); \
578 } while (0)
579 
580 static inline void safe_put_page(struct page *p)
581 {
582  if (p) put_page(p);
583 }
584 
585 extern int register_md_personality(struct md_personality *p);
586 extern int unregister_md_personality(struct md_personality *p);
587 extern struct md_thread *md_register_thread(
588  void (*run)(struct md_thread *thread),
589  struct mddev *mddev,
590  const char *name);
591 extern void md_unregister_thread(struct md_thread **threadp);
592 extern void md_wakeup_thread(struct md_thread *thread);
593 extern void md_check_recovery(struct mddev *mddev);
594 extern void md_write_start(struct mddev *mddev, struct bio *bi);
595 extern void md_write_end(struct mddev *mddev);
596 extern void md_done_sync(struct mddev *mddev, int blocks, int ok);
597 extern void md_error(struct mddev *mddev, struct md_rdev *rdev);
598 extern void md_finish_reshape(struct mddev *mddev);
599 
600 extern int mddev_congested(struct mddev *mddev, int bits);
601 extern void md_flush_request(struct mddev *mddev, struct bio *bio);
602 extern void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
603  sector_t sector, int size, struct page *page);
604 extern void md_super_wait(struct mddev *mddev);
605 extern int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
606  struct page *page, int rw, bool metadata_op);
607 extern void md_do_sync(struct md_thread *thread);
608 extern void md_new_event(struct mddev *mddev);
609 extern int md_allow_write(struct mddev *mddev);
610 extern void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev);
611 extern void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors);
612 extern int md_check_no_bitmap(struct mddev *mddev);
613 extern int md_integrity_register(struct mddev *mddev);
614 extern void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev);
615 extern int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale);
616 extern void restore_bitmap_write_access(struct file *file);
617 
618 extern void mddev_init(struct mddev *mddev);
619 extern int md_run(struct mddev *mddev);
620 extern void md_stop(struct mddev *mddev);
621 extern void md_stop_writes(struct mddev *mddev);
622 extern int md_rdev_init(struct md_rdev *rdev);
623 extern void md_rdev_clear(struct md_rdev *rdev);
624 
625 extern void mddev_suspend(struct mddev *mddev);
626 extern void mddev_resume(struct mddev *mddev);
627 extern struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
628  struct mddev *mddev);
629 extern struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
630  struct mddev *mddev);
631 extern void md_trim_bio(struct bio *bio, int offset, int size);
632 
633 extern void md_unplug(struct blk_plug_cb *cb, bool from_schedule);
634 static inline int mddev_check_plugged(struct mddev *mddev)
635 {
636  return !!blk_check_plugged(md_unplug, mddev,
637  sizeof(struct blk_plug_cb));
638 }
639 #endif /* _MD_MD_H */