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 | */ |
39 | struct 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 | */ |
48 | struct 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 | }; |
141 | enum 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 | RemoveSynchronized, /* synchronize_rcu() was called after |
194 | * this device was known to be faulty, |
195 | * so it is safe to remove without |
196 | * another synchronize_rcu() call. |
197 | */ |
198 | ExternalBbl, /* External metadata provides bad |
199 | * block management for a disk |
200 | */ |
201 | FailFast, /* Minimal retries should be attempted on |
202 | * this device, so use REQ_FAILFAST_DEV. |
203 | * Also don't try to repair failed reads. |
204 | * It is expects that no bad block log |
205 | * is present. |
206 | */ |
207 | LastDev, /* Seems to be the last working dev as |
208 | * it didn't fail, so don't use FailFast |
209 | * any more for metadata |
210 | */ |
211 | CollisionCheck, /* |
212 | * check if there is collision between raid1 |
213 | * serial bios. |
214 | */ |
215 | }; |
216 | |
217 | static inline int is_badblock(struct md_rdev *rdev, sector_t s, int sectors, |
218 | sector_t *first_bad, int *bad_sectors) |
219 | { |
220 | if (unlikely(rdev->badblocks.count)) { |
221 | int rv = badblocks_check(bb: &rdev->badblocks, s: rdev->data_offset + s, |
222 | sectors, |
223 | first_bad, bad_sectors); |
224 | if (rv) |
225 | *first_bad -= rdev->data_offset; |
226 | return rv; |
227 | } |
228 | return 0; |
229 | } |
230 | extern int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors, |
231 | int is_new); |
232 | extern int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors, |
233 | int is_new); |
234 | struct md_cluster_info; |
235 | |
236 | /** |
237 | * enum mddev_flags - md device flags. |
238 | * @MD_ARRAY_FIRST_USE: First use of array, needs initialization. |
239 | * @MD_CLOSING: If set, we are closing the array, do not open it then. |
240 | * @MD_JOURNAL_CLEAN: A raid with journal is already clean. |
241 | * @MD_HAS_JOURNAL: The raid array has journal feature set. |
242 | * @MD_CLUSTER_RESYNC_LOCKED: cluster raid only, which means node, already took |
243 | * resync lock, need to release the lock. |
244 | * @MD_FAILFAST_SUPPORTED: Using MD_FAILFAST on metadata writes is supported as |
245 | * calls to md_error() will never cause the array to |
246 | * become failed. |
247 | * @MD_HAS_PPL: The raid array has PPL feature set. |
248 | * @MD_HAS_MULTIPLE_PPLS: The raid array has multiple PPLs feature set. |
249 | * @MD_NOT_READY: do_md_run() is active, so 'array_state', ust not report that |
250 | * array is ready yet. |
251 | * @MD_BROKEN: This is used to stop writes and mark array as failed. |
252 | * @MD_DELETED: This device is being deleted |
253 | * |
254 | * change UNSUPPORTED_MDDEV_FLAGS for each array type if new flag is added |
255 | */ |
256 | enum mddev_flags { |
257 | MD_ARRAY_FIRST_USE, |
258 | MD_CLOSING, |
259 | MD_JOURNAL_CLEAN, |
260 | MD_HAS_JOURNAL, |
261 | MD_CLUSTER_RESYNC_LOCKED, |
262 | MD_FAILFAST_SUPPORTED, |
263 | MD_HAS_PPL, |
264 | MD_HAS_MULTIPLE_PPLS, |
265 | MD_NOT_READY, |
266 | MD_BROKEN, |
267 | MD_DELETED, |
268 | }; |
269 | |
270 | enum mddev_sb_flags { |
271 | MD_SB_CHANGE_DEVS, /* Some device status has changed */ |
272 | MD_SB_CHANGE_CLEAN, /* transition to or from 'clean' */ |
273 | MD_SB_CHANGE_PENDING, /* switch from 'clean' to 'active' in progress */ |
274 | MD_SB_NEED_REWRITE, /* metadata write needs to be repeated */ |
275 | }; |
276 | |
277 | #define NR_SERIAL_INFOS 8 |
278 | /* record current range of serialize IOs */ |
279 | struct serial_info { |
280 | struct rb_node node; |
281 | sector_t start; /* start sector of rb node */ |
282 | sector_t last; /* end sector of rb node */ |
283 | sector_t _subtree_last; /* highest sector in subtree of rb node */ |
284 | }; |
285 | |
286 | /* |
287 | * mddev->curr_resync stores the current sector of the resync but |
288 | * also has some overloaded values. |
289 | */ |
290 | enum { |
291 | /* No resync in progress */ |
292 | MD_RESYNC_NONE = 0, |
293 | /* Yielded to allow another conflicting resync to commence */ |
294 | MD_RESYNC_YIELDED = 1, |
295 | /* Delayed to check that there is no conflict with another sync */ |
296 | MD_RESYNC_DELAYED = 2, |
297 | /* Any value greater than or equal to this is in an active resync */ |
298 | MD_RESYNC_ACTIVE = 3, |
299 | }; |
300 | |
301 | struct mddev { |
302 | void *private; |
303 | struct md_personality *pers; |
304 | dev_t unit; |
305 | int md_minor; |
306 | struct list_head disks; |
307 | unsigned long flags; |
308 | unsigned long sb_flags; |
309 | |
310 | int suspended; |
311 | struct mutex suspend_mutex; |
312 | struct percpu_ref active_io; |
313 | int ro; |
314 | int sysfs_active; /* set when sysfs deletes |
315 | * are happening, so run/ |
316 | * takeover/stop are not safe |
317 | */ |
318 | struct gendisk *gendisk; |
319 | |
320 | struct kobject kobj; |
321 | int hold_active; |
322 | #define UNTIL_IOCTL 1 |
323 | #define UNTIL_STOP 2 |
324 | |
325 | /* Superblock information */ |
326 | int major_version, |
327 | minor_version, |
328 | patch_version; |
329 | int persistent; |
330 | int external; /* metadata is |
331 | * managed externally */ |
332 | char metadata_type[17]; /* externally set*/ |
333 | int chunk_sectors; |
334 | time64_t ctime, utime; |
335 | int level, layout; |
336 | char clevel[16]; |
337 | int raid_disks; |
338 | int max_disks; |
339 | sector_t dev_sectors; /* used size of |
340 | * component devices */ |
341 | sector_t array_sectors; /* exported array size */ |
342 | int external_size; /* size managed |
343 | * externally */ |
344 | __u64 events; |
345 | /* If the last 'event' was simply a clean->dirty transition, and |
346 | * we didn't write it to the spares, then it is safe and simple |
347 | * to just decrement the event count on a dirty->clean transition. |
348 | * So we record that possibility here. |
349 | */ |
350 | int can_decrease_events; |
351 | |
352 | char uuid[16]; |
353 | |
354 | /* If the array is being reshaped, we need to record the |
355 | * new shape and an indication of where we are up to. |
356 | * This is written to the superblock. |
357 | * If reshape_position is MaxSector, then no reshape is happening (yet). |
358 | */ |
359 | sector_t reshape_position; |
360 | int delta_disks, new_level, new_layout; |
361 | int new_chunk_sectors; |
362 | int reshape_backwards; |
363 | |
364 | struct md_thread __rcu *thread; /* management thread */ |
365 | struct md_thread __rcu *sync_thread; /* doing resync or reconstruct */ |
366 | |
367 | /* 'last_sync_action' is initialized to "none". It is set when a |
368 | * sync operation (i.e "data-check", "requested-resync", "resync", |
369 | * "recovery", or "reshape") is started. It holds this value even |
370 | * when the sync thread is "frozen" (interrupted) or "idle" (stopped |
371 | * or finished). It is overwritten when a new sync operation is begun. |
372 | */ |
373 | char *last_sync_action; |
374 | sector_t curr_resync; /* last block scheduled */ |
375 | /* As resync requests can complete out of order, we cannot easily track |
376 | * how much resync has been completed. So we occasionally pause until |
377 | * everything completes, then set curr_resync_completed to curr_resync. |
378 | * As such it may be well behind the real resync mark, but it is a value |
379 | * we are certain of. |
380 | */ |
381 | sector_t curr_resync_completed; |
382 | unsigned long resync_mark; /* a recent timestamp */ |
383 | sector_t resync_mark_cnt;/* blocks written at resync_mark */ |
384 | sector_t curr_mark_cnt; /* blocks scheduled now */ |
385 | |
386 | sector_t resync_max_sectors; /* may be set by personality */ |
387 | |
388 | atomic64_t resync_mismatches; /* count of sectors where |
389 | * parity/replica mismatch found |
390 | */ |
391 | |
392 | /* allow user-space to request suspension of IO to regions of the array */ |
393 | sector_t suspend_lo; |
394 | sector_t suspend_hi; |
395 | /* if zero, use the system-wide default */ |
396 | int sync_speed_min; |
397 | int sync_speed_max; |
398 | |
399 | /* resync even though the same disks are shared among md-devices */ |
400 | int parallel_resync; |
401 | |
402 | int ok_start_degraded; |
403 | |
404 | unsigned long recovery; |
405 | /* If a RAID personality determines that recovery (of a particular |
406 | * device) will fail due to a read error on the source device, it |
407 | * takes a copy of this number and does not attempt recovery again |
408 | * until this number changes. |
409 | */ |
410 | int recovery_disabled; |
411 | |
412 | int in_sync; /* know to not need resync */ |
413 | /* 'open_mutex' avoids races between 'md_open' and 'do_md_stop', so |
414 | * that we are never stopping an array while it is open. |
415 | * 'reconfig_mutex' protects all other reconfiguration. |
416 | * These locks are separate due to conflicting interactions |
417 | * with disk->open_mutex. |
418 | * Lock ordering is: |
419 | * reconfig_mutex -> disk->open_mutex |
420 | * disk->open_mutex -> open_mutex: e.g. __blkdev_get -> md_open |
421 | */ |
422 | struct mutex open_mutex; |
423 | struct mutex reconfig_mutex; |
424 | atomic_t active; /* general refcount */ |
425 | atomic_t openers; /* number of active opens */ |
426 | |
427 | int changed; /* True if we might need to |
428 | * reread partition info */ |
429 | int degraded; /* whether md should consider |
430 | * adding a spare |
431 | */ |
432 | |
433 | atomic_t recovery_active; /* blocks scheduled, but not written */ |
434 | wait_queue_head_t recovery_wait; |
435 | sector_t recovery_cp; |
436 | sector_t resync_min; /* user requested sync |
437 | * starts here */ |
438 | sector_t resync_max; /* resync should pause |
439 | * when it gets here */ |
440 | |
441 | struct kernfs_node *sysfs_state; /* handle for 'array_state' |
442 | * file in sysfs. |
443 | */ |
444 | struct kernfs_node *sysfs_action; /* handle for 'sync_action' */ |
445 | struct kernfs_node *sysfs_completed; /*handle for 'sync_completed' */ |
446 | struct kernfs_node *sysfs_degraded; /*handle for 'degraded' */ |
447 | struct kernfs_node *sysfs_level; /*handle for 'level' */ |
448 | |
449 | /* used for delayed sysfs removal */ |
450 | struct work_struct del_work; |
451 | /* used for register new sync thread */ |
452 | struct work_struct sync_work; |
453 | |
454 | /* "lock" protects: |
455 | * flush_bio transition from NULL to !NULL |
456 | * rdev superblocks, events |
457 | * clearing MD_CHANGE_* |
458 | * in_sync - and related safemode and MD_CHANGE changes |
459 | * pers (also protected by reconfig_mutex and pending IO). |
460 | * clearing ->bitmap |
461 | * clearing ->bitmap_info.file |
462 | * changing ->resync_{min,max} |
463 | * setting MD_RECOVERY_RUNNING (which interacts with resync_{min,max}) |
464 | */ |
465 | spinlock_t lock; |
466 | wait_queue_head_t sb_wait; /* for waiting on superblock updates */ |
467 | atomic_t pending_writes; /* number of active superblock writes */ |
468 | |
469 | unsigned int safemode; /* if set, update "clean" superblock |
470 | * when no writes pending. |
471 | */ |
472 | unsigned int safemode_delay; |
473 | struct timer_list safemode_timer; |
474 | struct percpu_ref writes_pending; |
475 | int sync_checkers; /* # of threads checking writes_pending */ |
476 | struct request_queue *queue; /* for plugging ... */ |
477 | |
478 | struct bitmap *bitmap; /* the bitmap for the device */ |
479 | struct { |
480 | struct file *file; /* the bitmap file */ |
481 | loff_t offset; /* offset from superblock of |
482 | * start of bitmap. May be |
483 | * negative, but not '0' |
484 | * For external metadata, offset |
485 | * from start of device. |
486 | */ |
487 | unsigned long space; /* space available at this offset */ |
488 | loff_t default_offset; /* this is the offset to use when |
489 | * hot-adding a bitmap. It should |
490 | * eventually be settable by sysfs. |
491 | */ |
492 | unsigned long default_space; /* space available at |
493 | * default offset */ |
494 | struct mutex mutex; |
495 | unsigned long chunksize; |
496 | unsigned long daemon_sleep; /* how many jiffies between updates? */ |
497 | unsigned long max_write_behind; /* write-behind mode */ |
498 | int external; |
499 | int nodes; /* Maximum number of nodes in the cluster */ |
500 | char cluster_name[64]; /* Name of the cluster */ |
501 | } bitmap_info; |
502 | |
503 | atomic_t max_corr_read_errors; /* max read retries */ |
504 | struct list_head all_mddevs; |
505 | |
506 | const struct attribute_group *to_remove; |
507 | |
508 | struct bio_set bio_set; |
509 | struct bio_set sync_set; /* for sync operations like |
510 | * metadata and bitmap writes |
511 | */ |
512 | struct bio_set io_clone_set; |
513 | |
514 | /* Generic flush handling. |
515 | * The last to finish preflush schedules a worker to submit |
516 | * the rest of the request (without the REQ_PREFLUSH flag). |
517 | */ |
518 | struct bio *flush_bio; |
519 | atomic_t flush_pending; |
520 | ktime_t start_flush, prev_flush_start; /* prev_flush_start is when the previous completed |
521 | * flush was started. |
522 | */ |
523 | struct work_struct flush_work; |
524 | struct work_struct event_work; /* used by dm to report failure event */ |
525 | mempool_t *serial_info_pool; |
526 | void (*sync_super)(struct mddev *mddev, struct md_rdev *rdev); |
527 | struct md_cluster_info *cluster_info; |
528 | unsigned int good_device_nr; /* good device num within cluster raid */ |
529 | unsigned int noio_flag; /* for memalloc scope API */ |
530 | |
531 | /* |
532 | * Temporarily store rdev that will be finally removed when |
533 | * reconfig_mutex is unlocked, protected by reconfig_mutex. |
534 | */ |
535 | struct list_head deleting; |
536 | |
537 | /* Used to synchronize idle and frozen for action_store() */ |
538 | struct mutex sync_mutex; |
539 | /* The sequence number for sync thread */ |
540 | atomic_t sync_seq; |
541 | |
542 | bool has_superblocks:1; |
543 | bool fail_last_dev:1; |
544 | bool serialize_policy:1; |
545 | }; |
546 | |
547 | enum recovery_flags { |
548 | /* |
549 | * If neither SYNC or RESHAPE are set, then it is a recovery. |
550 | */ |
551 | MD_RECOVERY_RUNNING, /* a thread is running, or about to be started */ |
552 | MD_RECOVERY_SYNC, /* actually doing a resync, not a recovery */ |
553 | MD_RECOVERY_RECOVER, /* doing recovery, or need to try it. */ |
554 | MD_RECOVERY_INTR, /* resync needs to be aborted for some reason */ |
555 | MD_RECOVERY_DONE, /* thread is done and is waiting to be reaped */ |
556 | MD_RECOVERY_NEEDED, /* we might need to start a resync/recover */ |
557 | MD_RECOVERY_REQUESTED, /* user-space has requested a sync (used with SYNC) */ |
558 | MD_RECOVERY_CHECK, /* user-space request for check-only, no repair */ |
559 | MD_RECOVERY_RESHAPE, /* A reshape is happening */ |
560 | MD_RECOVERY_FROZEN, /* User request to abort, and not restart, any action */ |
561 | MD_RECOVERY_ERROR, /* sync-action interrupted because io-error */ |
562 | MD_RECOVERY_WAIT, /* waiting for pers->start() to finish */ |
563 | MD_RESYNCING_REMOTE, /* remote node is running resync thread */ |
564 | }; |
565 | |
566 | static inline int __must_check mddev_lock(struct mddev *mddev) |
567 | { |
568 | return mutex_lock_interruptible(&mddev->reconfig_mutex); |
569 | } |
570 | |
571 | /* Sometimes we need to take the lock in a situation where |
572 | * failure due to interrupts is not acceptable. |
573 | */ |
574 | static inline void mddev_lock_nointr(struct mddev *mddev) |
575 | { |
576 | mutex_lock(&mddev->reconfig_mutex); |
577 | } |
578 | |
579 | static inline int mddev_trylock(struct mddev *mddev) |
580 | { |
581 | return mutex_trylock(lock: &mddev->reconfig_mutex); |
582 | } |
583 | extern void mddev_unlock(struct mddev *mddev); |
584 | |
585 | static inline void md_sync_acct(struct block_device *bdev, unsigned long nr_sectors) |
586 | { |
587 | atomic_add(i: nr_sectors, v: &bdev->bd_disk->sync_io); |
588 | } |
589 | |
590 | static inline void md_sync_acct_bio(struct bio *bio, unsigned long nr_sectors) |
591 | { |
592 | md_sync_acct(bdev: bio->bi_bdev, nr_sectors); |
593 | } |
594 | |
595 | struct md_personality |
596 | { |
597 | char *name; |
598 | int level; |
599 | struct list_head list; |
600 | struct module *owner; |
601 | bool __must_check (*make_request)(struct mddev *mddev, struct bio *bio); |
602 | /* |
603 | * start up works that do NOT require md_thread. tasks that |
604 | * requires md_thread should go into start() |
605 | */ |
606 | int (*run)(struct mddev *mddev); |
607 | /* start up works that require md threads */ |
608 | int (*start)(struct mddev *mddev); |
609 | void (*free)(struct mddev *mddev, void *priv); |
610 | void (*status)(struct seq_file *seq, struct mddev *mddev); |
611 | /* error_handler must set ->faulty and clear ->in_sync |
612 | * if appropriate, and should abort recovery if needed |
613 | */ |
614 | void (*error_handler)(struct mddev *mddev, struct md_rdev *rdev); |
615 | int (*hot_add_disk) (struct mddev *mddev, struct md_rdev *rdev); |
616 | int (*hot_remove_disk) (struct mddev *mddev, struct md_rdev *rdev); |
617 | int (*spare_active) (struct mddev *mddev); |
618 | sector_t (*sync_request)(struct mddev *mddev, sector_t sector_nr, int *skipped); |
619 | int (*resize) (struct mddev *mddev, sector_t sectors); |
620 | sector_t (*size) (struct mddev *mddev, sector_t sectors, int raid_disks); |
621 | int (*check_reshape) (struct mddev *mddev); |
622 | int (*start_reshape) (struct mddev *mddev); |
623 | void (*finish_reshape) (struct mddev *mddev); |
624 | void (*update_reshape_pos) (struct mddev *mddev); |
625 | /* quiesce suspends or resumes internal processing. |
626 | * 1 - stop new actions and wait for action io to complete |
627 | * 0 - return to normal behaviour |
628 | */ |
629 | void (*quiesce) (struct mddev *mddev, int quiesce); |
630 | /* takeover is used to transition an array from one |
631 | * personality to another. The new personality must be able |
632 | * to handle the data in the current layout. |
633 | * e.g. 2drive raid1 -> 2drive raid5 |
634 | * ndrive raid5 -> degraded n+1drive raid6 with special layout |
635 | * If the takeover succeeds, a new 'private' structure is returned. |
636 | * This needs to be installed and then ->run used to activate the |
637 | * array. |
638 | */ |
639 | void *(*takeover) (struct mddev *mddev); |
640 | /* Changes the consistency policy of an active array. */ |
641 | int (*change_consistency_policy)(struct mddev *mddev, const char *buf); |
642 | }; |
643 | |
644 | struct md_sysfs_entry { |
645 | struct attribute attr; |
646 | ssize_t (*show)(struct mddev *, char *); |
647 | ssize_t (*store)(struct mddev *, const char *, size_t); |
648 | }; |
649 | extern const struct attribute_group md_bitmap_group; |
650 | |
651 | static inline struct kernfs_node *sysfs_get_dirent_safe(struct kernfs_node *sd, char *name) |
652 | { |
653 | if (sd) |
654 | return sysfs_get_dirent(parent: sd, name); |
655 | return sd; |
656 | } |
657 | static inline void sysfs_notify_dirent_safe(struct kernfs_node *sd) |
658 | { |
659 | if (sd) |
660 | sysfs_notify_dirent(kn: sd); |
661 | } |
662 | |
663 | static inline char * mdname (struct mddev * mddev) |
664 | { |
665 | return mddev->gendisk ? mddev->gendisk->disk_name : "mdX" ; |
666 | } |
667 | |
668 | static inline int sysfs_link_rdev(struct mddev *mddev, struct md_rdev *rdev) |
669 | { |
670 | char nm[20]; |
671 | if (!test_bit(Replacement, &rdev->flags) && |
672 | !test_bit(Journal, &rdev->flags) && |
673 | mddev->kobj.sd) { |
674 | sprintf(buf: nm, fmt: "rd%d" , rdev->raid_disk); |
675 | return sysfs_create_link(kobj: &mddev->kobj, target: &rdev->kobj, name: nm); |
676 | } else |
677 | return 0; |
678 | } |
679 | |
680 | static inline void sysfs_unlink_rdev(struct mddev *mddev, struct md_rdev *rdev) |
681 | { |
682 | char nm[20]; |
683 | if (!test_bit(Replacement, &rdev->flags) && |
684 | !test_bit(Journal, &rdev->flags) && |
685 | mddev->kobj.sd) { |
686 | sprintf(buf: nm, fmt: "rd%d" , rdev->raid_disk); |
687 | sysfs_remove_link(kobj: &mddev->kobj, name: nm); |
688 | } |
689 | } |
690 | |
691 | /* |
692 | * iterates through some rdev ringlist. It's safe to remove the |
693 | * current 'rdev'. Dont touch 'tmp' though. |
694 | */ |
695 | #define rdev_for_each_list(rdev, tmp, head) \ |
696 | list_for_each_entry_safe(rdev, tmp, head, same_set) |
697 | |
698 | /* |
699 | * iterates through the 'same array disks' ringlist |
700 | */ |
701 | #define rdev_for_each(rdev, mddev) \ |
702 | list_for_each_entry(rdev, &((mddev)->disks), same_set) |
703 | |
704 | #define rdev_for_each_safe(rdev, tmp, mddev) \ |
705 | list_for_each_entry_safe(rdev, tmp, &((mddev)->disks), same_set) |
706 | |
707 | #define rdev_for_each_rcu(rdev, mddev) \ |
708 | list_for_each_entry_rcu(rdev, &((mddev)->disks), same_set) |
709 | |
710 | struct md_thread { |
711 | void (*run) (struct md_thread *thread); |
712 | struct mddev *mddev; |
713 | wait_queue_head_t wqueue; |
714 | unsigned long flags; |
715 | struct task_struct *tsk; |
716 | unsigned long timeout; |
717 | void *private; |
718 | }; |
719 | |
720 | struct md_io_clone { |
721 | struct mddev *mddev; |
722 | struct bio *orig_bio; |
723 | unsigned long start_time; |
724 | struct bio bio_clone; |
725 | }; |
726 | |
727 | #define THREAD_WAKEUP 0 |
728 | |
729 | static inline void safe_put_page(struct page *p) |
730 | { |
731 | if (p) put_page(page: p); |
732 | } |
733 | |
734 | extern int register_md_personality(struct md_personality *p); |
735 | extern int unregister_md_personality(struct md_personality *p); |
736 | extern int register_md_cluster_operations(struct md_cluster_operations *ops, |
737 | struct module *module); |
738 | extern int unregister_md_cluster_operations(void); |
739 | extern int md_setup_cluster(struct mddev *mddev, int nodes); |
740 | extern void md_cluster_stop(struct mddev *mddev); |
741 | extern struct md_thread *md_register_thread( |
742 | void (*run)(struct md_thread *thread), |
743 | struct mddev *mddev, |
744 | const char *name); |
745 | extern void md_unregister_thread(struct mddev *mddev, struct md_thread __rcu **threadp); |
746 | extern void md_wakeup_thread(struct md_thread __rcu *thread); |
747 | extern void md_check_recovery(struct mddev *mddev); |
748 | extern void md_reap_sync_thread(struct mddev *mddev); |
749 | extern bool md_write_start(struct mddev *mddev, struct bio *bi); |
750 | extern void md_write_inc(struct mddev *mddev, struct bio *bi); |
751 | extern void md_write_end(struct mddev *mddev); |
752 | extern void md_done_sync(struct mddev *mddev, int blocks, int ok); |
753 | extern void md_error(struct mddev *mddev, struct md_rdev *rdev); |
754 | extern void md_finish_reshape(struct mddev *mddev); |
755 | void md_submit_discard_bio(struct mddev *mddev, struct md_rdev *rdev, |
756 | struct bio *bio, sector_t start, sector_t size); |
757 | void md_account_bio(struct mddev *mddev, struct bio **bio); |
758 | |
759 | extern bool __must_check md_flush_request(struct mddev *mddev, struct bio *bio); |
760 | extern void md_super_write(struct mddev *mddev, struct md_rdev *rdev, |
761 | sector_t sector, int size, struct page *page); |
762 | extern int md_super_wait(struct mddev *mddev); |
763 | extern int sync_page_io(struct md_rdev *rdev, sector_t sector, int size, |
764 | struct page *page, blk_opf_t opf, bool metadata_op); |
765 | extern void md_do_sync(struct md_thread *thread); |
766 | extern void md_new_event(void); |
767 | extern void md_allow_write(struct mddev *mddev); |
768 | extern void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev); |
769 | extern void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors); |
770 | extern int md_check_no_bitmap(struct mddev *mddev); |
771 | extern int md_integrity_register(struct mddev *mddev); |
772 | extern int md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev); |
773 | extern int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale); |
774 | |
775 | extern int mddev_init(struct mddev *mddev); |
776 | extern void mddev_destroy(struct mddev *mddev); |
777 | struct mddev *md_alloc(dev_t dev, char *name); |
778 | void mddev_put(struct mddev *mddev); |
779 | extern int md_run(struct mddev *mddev); |
780 | extern int md_start(struct mddev *mddev); |
781 | extern void md_stop(struct mddev *mddev); |
782 | extern void md_stop_writes(struct mddev *mddev); |
783 | extern int md_rdev_init(struct md_rdev *rdev); |
784 | extern void md_rdev_clear(struct md_rdev *rdev); |
785 | |
786 | extern void md_handle_request(struct mddev *mddev, struct bio *bio); |
787 | extern int mddev_suspend(struct mddev *mddev, bool interruptible); |
788 | extern void mddev_resume(struct mddev *mddev); |
789 | |
790 | extern void md_reload_sb(struct mddev *mddev, int raid_disk); |
791 | extern void md_update_sb(struct mddev *mddev, int force); |
792 | extern void mddev_create_serial_pool(struct mddev *mddev, struct md_rdev *rdev); |
793 | extern void mddev_destroy_serial_pool(struct mddev *mddev, |
794 | struct md_rdev *rdev); |
795 | struct md_rdev *md_find_rdev_nr_rcu(struct mddev *mddev, int nr); |
796 | struct md_rdev *md_find_rdev_rcu(struct mddev *mddev, dev_t dev); |
797 | |
798 | static inline bool is_rdev_broken(struct md_rdev *rdev) |
799 | { |
800 | return !disk_live(disk: rdev->bdev->bd_disk); |
801 | } |
802 | |
803 | static inline void rdev_dec_pending(struct md_rdev *rdev, struct mddev *mddev) |
804 | { |
805 | int faulty = test_bit(Faulty, &rdev->flags); |
806 | if (atomic_dec_and_test(v: &rdev->nr_pending) && faulty) { |
807 | set_bit(nr: MD_RECOVERY_NEEDED, addr: &mddev->recovery); |
808 | md_wakeup_thread(thread: mddev->thread); |
809 | } |
810 | } |
811 | |
812 | extern struct md_cluster_operations *md_cluster_ops; |
813 | static inline int mddev_is_clustered(struct mddev *mddev) |
814 | { |
815 | return mddev->cluster_info && mddev->bitmap_info.nodes > 1; |
816 | } |
817 | |
818 | /* clear unsupported mddev_flags */ |
819 | static inline void mddev_clear_unsupported_flags(struct mddev *mddev, |
820 | unsigned long unsupported_flags) |
821 | { |
822 | mddev->flags &= ~unsupported_flags; |
823 | } |
824 | |
825 | static inline void mddev_check_write_zeroes(struct mddev *mddev, struct bio *bio) |
826 | { |
827 | if (bio_op(bio) == REQ_OP_WRITE_ZEROES && |
828 | !bio->bi_bdev->bd_disk->queue->limits.max_write_zeroes_sectors) |
829 | mddev->queue->limits.max_write_zeroes_sectors = 0; |
830 | } |
831 | |
832 | static inline int mddev_suspend_and_lock(struct mddev *mddev) |
833 | { |
834 | int ret; |
835 | |
836 | ret = mddev_suspend(mddev, interruptible: true); |
837 | if (ret) |
838 | return ret; |
839 | |
840 | ret = mddev_lock(mddev); |
841 | if (ret) |
842 | mddev_resume(mddev); |
843 | |
844 | return ret; |
845 | } |
846 | |
847 | static inline void mddev_suspend_and_lock_nointr(struct mddev *mddev) |
848 | { |
849 | mddev_suspend(mddev, interruptible: false); |
850 | mutex_lock(&mddev->reconfig_mutex); |
851 | } |
852 | |
853 | static inline void mddev_unlock_and_resume(struct mddev *mddev) |
854 | { |
855 | mddev_unlock(mddev); |
856 | mddev_resume(mddev); |
857 | } |
858 | |
859 | struct mdu_array_info_s; |
860 | struct mdu_disk_info_s; |
861 | |
862 | extern int mdp_major; |
863 | extern struct workqueue_struct *md_bitmap_wq; |
864 | void md_autostart_arrays(int part); |
865 | int md_set_array_info(struct mddev *mddev, struct mdu_array_info_s *info); |
866 | int md_add_new_disk(struct mddev *mddev, struct mdu_disk_info_s *info); |
867 | int do_md_run(struct mddev *mddev); |
868 | |
869 | extern const struct block_device_operations md_fops; |
870 | |
871 | #endif /* _MD_MD_H */ |
872 | |