]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Copyright (C) 2003 Sistina Software Limited. | |
3 | * | |
4 | * This file is released under the GPL. | |
5 | */ | |
6 | ||
586e80e6 MP |
7 | #include <linux/device-mapper.h> |
8 | ||
1da177e4 | 9 | #include "dm-bio-list.h" |
06386bbf | 10 | #include "dm-bio-record.h" |
1da177e4 LT |
11 | |
12 | #include <linux/ctype.h> | |
13 | #include <linux/init.h> | |
14 | #include <linux/mempool.h> | |
15 | #include <linux/module.h> | |
16 | #include <linux/pagemap.h> | |
17 | #include <linux/slab.h> | |
18 | #include <linux/time.h> | |
19 | #include <linux/vmalloc.h> | |
20 | #include <linux/workqueue.h> | |
6f3c3f0a | 21 | #include <linux/log2.h> |
72f4b314 | 22 | #include <linux/hardirq.h> |
a765e20e AK |
23 | #include <linux/dm-io.h> |
24 | #include <linux/dm-dirty-log.h> | |
25 | #include <linux/dm-kcopyd.h> | |
1da177e4 | 26 | |
72d94861 | 27 | #define DM_MSG_PREFIX "raid1" |
88be163a | 28 | #define DM_IO_PAGES 64 |
72d94861 | 29 | |
a8e6afa2 | 30 | #define DM_RAID1_HANDLE_ERRORS 0x01 |
f44db678 | 31 | #define errors_handled(p) ((p)->features & DM_RAID1_HANDLE_ERRORS) |
a8e6afa2 | 32 | |
33184048 | 33 | static DECLARE_WAIT_QUEUE_HEAD(_kmirrord_recovery_stopped); |
1da177e4 | 34 | |
1da177e4 LT |
35 | /*----------------------------------------------------------------- |
36 | * Region hash | |
37 | * | |
38 | * The mirror splits itself up into discrete regions. Each | |
39 | * region can be in one of three states: clean, dirty, | |
40 | * nosync. There is no need to put clean regions in the hash. | |
41 | * | |
42 | * In addition to being present in the hash table a region _may_ | |
43 | * be present on one of three lists. | |
44 | * | |
45 | * clean_regions: Regions on this list have no io pending to | |
46 | * them, they are in sync, we are no longer interested in them, | |
47 | * they are dull. rh_update_states() will remove them from the | |
48 | * hash table. | |
49 | * | |
50 | * quiesced_regions: These regions have been spun down, ready | |
51 | * for recovery. rh_recovery_start() will remove regions from | |
52 | * this list and hand them to kmirrord, which will schedule the | |
53 | * recovery io with kcopyd. | |
54 | * | |
55 | * recovered_regions: Regions that kcopyd has successfully | |
56 | * recovered. rh_update_states() will now schedule any delayed | |
57 | * io, up the recovery_count, and remove the region from the | |
58 | * hash. | |
59 | * | |
60 | * There are 2 locks: | |
61 | * A rw spin lock 'hash_lock' protects just the hash table, | |
62 | * this is never held in write mode from interrupt context, | |
63 | * which I believe means that we only have to disable irqs when | |
64 | * doing a write lock. | |
65 | * | |
66 | * An ordinary spin lock 'region_lock' that protects the three | |
67 | * lists in the region_hash, with the 'state', 'list' and | |
68 | * 'bhs_delayed' fields of the regions. This is used from irq | |
69 | * context, so all other uses will have to suspend local irqs. | |
70 | *---------------------------------------------------------------*/ | |
71 | struct mirror_set; | |
72 | struct region_hash { | |
73 | struct mirror_set *ms; | |
74 | uint32_t region_size; | |
75 | unsigned region_shift; | |
76 | ||
77 | /* holds persistent region state */ | |
416cd17b | 78 | struct dm_dirty_log *log; |
1da177e4 LT |
79 | |
80 | /* hash table */ | |
81 | rwlock_t hash_lock; | |
82 | mempool_t *region_pool; | |
83 | unsigned int mask; | |
84 | unsigned int nr_buckets; | |
85 | struct list_head *buckets; | |
86 | ||
87 | spinlock_t region_lock; | |
33184048 | 88 | atomic_t recovery_in_flight; |
1da177e4 LT |
89 | struct semaphore recovery_count; |
90 | struct list_head clean_regions; | |
91 | struct list_head quiesced_regions; | |
92 | struct list_head recovered_regions; | |
f44db678 | 93 | struct list_head failed_recovered_regions; |
1da177e4 LT |
94 | }; |
95 | ||
96 | enum { | |
97 | RH_CLEAN, | |
98 | RH_DIRTY, | |
99 | RH_NOSYNC, | |
100 | RH_RECOVERING | |
101 | }; | |
102 | ||
103 | struct region { | |
104 | struct region_hash *rh; /* FIXME: can we get rid of this ? */ | |
105 | region_t key; | |
106 | int state; | |
107 | ||
108 | struct list_head hash_list; | |
109 | struct list_head list; | |
110 | ||
111 | atomic_t pending; | |
112 | struct bio_list delayed_bios; | |
113 | }; | |
114 | ||
e4c8b3ba NB |
115 | |
116 | /*----------------------------------------------------------------- | |
117 | * Mirror set structures. | |
118 | *---------------------------------------------------------------*/ | |
72f4b314 JB |
119 | enum dm_raid1_error { |
120 | DM_RAID1_WRITE_ERROR, | |
121 | DM_RAID1_SYNC_ERROR, | |
122 | DM_RAID1_READ_ERROR | |
123 | }; | |
124 | ||
e4c8b3ba | 125 | struct mirror { |
aa5617c5 | 126 | struct mirror_set *ms; |
e4c8b3ba | 127 | atomic_t error_count; |
39ed7adb | 128 | unsigned long error_type; |
e4c8b3ba NB |
129 | struct dm_dev *dev; |
130 | sector_t offset; | |
131 | }; | |
132 | ||
133 | struct mirror_set { | |
134 | struct dm_target *ti; | |
135 | struct list_head list; | |
136 | struct region_hash rh; | |
eb69aca5 | 137 | struct dm_kcopyd_client *kcopyd_client; |
a8e6afa2 | 138 | uint64_t features; |
e4c8b3ba | 139 | |
72f4b314 | 140 | spinlock_t lock; /* protects the lists */ |
e4c8b3ba NB |
141 | struct bio_list reads; |
142 | struct bio_list writes; | |
72f4b314 | 143 | struct bio_list failures; |
e4c8b3ba | 144 | |
88be163a | 145 | struct dm_io_client *io_client; |
06386bbf | 146 | mempool_t *read_record_pool; |
88be163a | 147 | |
e4c8b3ba NB |
148 | /* recovery */ |
149 | region_t nr_regions; | |
150 | int in_sync; | |
fc1ff958 | 151 | int log_failure; |
b80aa7a0 | 152 | atomic_t suspend; |
e4c8b3ba | 153 | |
72f4b314 | 154 | atomic_t default_mirror; /* Default mirror */ |
e4c8b3ba | 155 | |
6ad36fe2 HS |
156 | struct workqueue_struct *kmirrord_wq; |
157 | struct work_struct kmirrord_work; | |
a2aebe03 MP |
158 | struct timer_list timer; |
159 | unsigned long timer_pending; | |
160 | ||
72f4b314 | 161 | struct work_struct trigger_event; |
6ad36fe2 | 162 | |
e4c8b3ba NB |
163 | unsigned int nr_mirrors; |
164 | struct mirror mirror[0]; | |
165 | }; | |
166 | ||
1da177e4 LT |
167 | /* |
168 | * Conversion fns | |
169 | */ | |
170 | static inline region_t bio_to_region(struct region_hash *rh, struct bio *bio) | |
171 | { | |
e4c8b3ba | 172 | return (bio->bi_sector - rh->ms->ti->begin) >> rh->region_shift; |
1da177e4 LT |
173 | } |
174 | ||
175 | static inline sector_t region_to_sector(struct region_hash *rh, region_t region) | |
176 | { | |
177 | return region << rh->region_shift; | |
178 | } | |
179 | ||
6ad36fe2 HS |
180 | static void wake(struct mirror_set *ms) |
181 | { | |
182 | queue_work(ms->kmirrord_wq, &ms->kmirrord_work); | |
183 | } | |
184 | ||
a2aebe03 MP |
185 | static void delayed_wake_fn(unsigned long data) |
186 | { | |
187 | struct mirror_set *ms = (struct mirror_set *) data; | |
188 | ||
189 | clear_bit(0, &ms->timer_pending); | |
190 | wake(ms); | |
191 | } | |
192 | ||
193 | static void delayed_wake(struct mirror_set *ms) | |
194 | { | |
195 | if (test_and_set_bit(0, &ms->timer_pending)) | |
196 | return; | |
197 | ||
198 | ms->timer.expires = jiffies + HZ / 5; | |
199 | ms->timer.data = (unsigned long) ms; | |
200 | ms->timer.function = delayed_wake_fn; | |
201 | add_timer(&ms->timer); | |
202 | } | |
203 | ||
1da177e4 LT |
204 | /* FIXME move this */ |
205 | static void queue_bio(struct mirror_set *ms, struct bio *bio, int rw); | |
206 | ||
1da177e4 LT |
207 | #define MIN_REGIONS 64 |
208 | #define MAX_RECOVERY 1 | |
209 | static int rh_init(struct region_hash *rh, struct mirror_set *ms, | |
416cd17b | 210 | struct dm_dirty_log *log, uint32_t region_size, |
1da177e4 LT |
211 | region_t nr_regions) |
212 | { | |
213 | unsigned int nr_buckets, max_buckets; | |
214 | size_t i; | |
215 | ||
216 | /* | |
217 | * Calculate a suitable number of buckets for our hash | |
218 | * table. | |
219 | */ | |
220 | max_buckets = nr_regions >> 6; | |
221 | for (nr_buckets = 128u; nr_buckets < max_buckets; nr_buckets <<= 1) | |
222 | ; | |
223 | nr_buckets >>= 1; | |
224 | ||
225 | rh->ms = ms; | |
226 | rh->log = log; | |
227 | rh->region_size = region_size; | |
228 | rh->region_shift = ffs(region_size) - 1; | |
229 | rwlock_init(&rh->hash_lock); | |
230 | rh->mask = nr_buckets - 1; | |
231 | rh->nr_buckets = nr_buckets; | |
232 | ||
233 | rh->buckets = vmalloc(nr_buckets * sizeof(*rh->buckets)); | |
234 | if (!rh->buckets) { | |
235 | DMERR("unable to allocate region hash memory"); | |
236 | return -ENOMEM; | |
237 | } | |
238 | ||
239 | for (i = 0; i < nr_buckets; i++) | |
240 | INIT_LIST_HEAD(rh->buckets + i); | |
241 | ||
242 | spin_lock_init(&rh->region_lock); | |
243 | sema_init(&rh->recovery_count, 0); | |
33184048 | 244 | atomic_set(&rh->recovery_in_flight, 0); |
1da177e4 LT |
245 | INIT_LIST_HEAD(&rh->clean_regions); |
246 | INIT_LIST_HEAD(&rh->quiesced_regions); | |
247 | INIT_LIST_HEAD(&rh->recovered_regions); | |
f44db678 | 248 | INIT_LIST_HEAD(&rh->failed_recovered_regions); |
1da177e4 | 249 | |
0eaae62a MD |
250 | rh->region_pool = mempool_create_kmalloc_pool(MIN_REGIONS, |
251 | sizeof(struct region)); | |
1da177e4 LT |
252 | if (!rh->region_pool) { |
253 | vfree(rh->buckets); | |
254 | rh->buckets = NULL; | |
255 | return -ENOMEM; | |
256 | } | |
257 | ||
258 | return 0; | |
259 | } | |
260 | ||
261 | static void rh_exit(struct region_hash *rh) | |
262 | { | |
263 | unsigned int h; | |
264 | struct region *reg, *nreg; | |
265 | ||
266 | BUG_ON(!list_empty(&rh->quiesced_regions)); | |
267 | for (h = 0; h < rh->nr_buckets; h++) { | |
268 | list_for_each_entry_safe(reg, nreg, rh->buckets + h, hash_list) { | |
269 | BUG_ON(atomic_read(®->pending)); | |
270 | mempool_free(reg, rh->region_pool); | |
271 | } | |
272 | } | |
273 | ||
274 | if (rh->log) | |
416cd17b | 275 | dm_dirty_log_destroy(rh->log); |
1da177e4 LT |
276 | if (rh->region_pool) |
277 | mempool_destroy(rh->region_pool); | |
278 | vfree(rh->buckets); | |
279 | } | |
280 | ||
281 | #define RH_HASH_MULT 2654435387U | |
282 | ||
283 | static inline unsigned int rh_hash(struct region_hash *rh, region_t region) | |
284 | { | |
285 | return (unsigned int) ((region * RH_HASH_MULT) >> 12) & rh->mask; | |
286 | } | |
287 | ||
288 | static struct region *__rh_lookup(struct region_hash *rh, region_t region) | |
289 | { | |
290 | struct region *reg; | |
291 | ||
292 | list_for_each_entry (reg, rh->buckets + rh_hash(rh, region), hash_list) | |
293 | if (reg->key == region) | |
294 | return reg; | |
295 | ||
296 | return NULL; | |
297 | } | |
298 | ||
299 | static void __rh_insert(struct region_hash *rh, struct region *reg) | |
300 | { | |
301 | unsigned int h = rh_hash(rh, reg->key); | |
302 | list_add(®->hash_list, rh->buckets + h); | |
303 | } | |
304 | ||
305 | static struct region *__rh_alloc(struct region_hash *rh, region_t region) | |
306 | { | |
307 | struct region *reg, *nreg; | |
308 | ||
309 | read_unlock(&rh->hash_lock); | |
c06aad85 DK |
310 | nreg = mempool_alloc(rh->region_pool, GFP_ATOMIC); |
311 | if (unlikely(!nreg)) | |
312 | nreg = kmalloc(sizeof(struct region), GFP_NOIO); | |
1da177e4 LT |
313 | nreg->state = rh->log->type->in_sync(rh->log, region, 1) ? |
314 | RH_CLEAN : RH_NOSYNC; | |
315 | nreg->rh = rh; | |
316 | nreg->key = region; | |
317 | ||
318 | INIT_LIST_HEAD(&nreg->list); | |
319 | ||
320 | atomic_set(&nreg->pending, 0); | |
321 | bio_list_init(&nreg->delayed_bios); | |
322 | write_lock_irq(&rh->hash_lock); | |
323 | ||
324 | reg = __rh_lookup(rh, region); | |
325 | if (reg) | |
326 | /* we lost the race */ | |
327 | mempool_free(nreg, rh->region_pool); | |
328 | ||
329 | else { | |
330 | __rh_insert(rh, nreg); | |
331 | if (nreg->state == RH_CLEAN) { | |
332 | spin_lock(&rh->region_lock); | |
333 | list_add(&nreg->list, &rh->clean_regions); | |
334 | spin_unlock(&rh->region_lock); | |
335 | } | |
336 | reg = nreg; | |
337 | } | |
338 | write_unlock_irq(&rh->hash_lock); | |
339 | read_lock(&rh->hash_lock); | |
340 | ||
341 | return reg; | |
342 | } | |
343 | ||
344 | static inline struct region *__rh_find(struct region_hash *rh, region_t region) | |
345 | { | |
346 | struct region *reg; | |
347 | ||
348 | reg = __rh_lookup(rh, region); | |
349 | if (!reg) | |
350 | reg = __rh_alloc(rh, region); | |
351 | ||
352 | return reg; | |
353 | } | |
354 | ||
355 | static int rh_state(struct region_hash *rh, region_t region, int may_block) | |
356 | { | |
357 | int r; | |
358 | struct region *reg; | |
359 | ||
360 | read_lock(&rh->hash_lock); | |
361 | reg = __rh_lookup(rh, region); | |
362 | read_unlock(&rh->hash_lock); | |
363 | ||
364 | if (reg) | |
365 | return reg->state; | |
366 | ||
367 | /* | |
368 | * The region wasn't in the hash, so we fall back to the | |
369 | * dirty log. | |
370 | */ | |
371 | r = rh->log->type->in_sync(rh->log, region, may_block); | |
372 | ||
373 | /* | |
374 | * Any error from the dirty log (eg. -EWOULDBLOCK) gets | |
375 | * taken as a RH_NOSYNC | |
376 | */ | |
377 | return r == 1 ? RH_CLEAN : RH_NOSYNC; | |
378 | } | |
379 | ||
380 | static inline int rh_in_sync(struct region_hash *rh, | |
381 | region_t region, int may_block) | |
382 | { | |
383 | int state = rh_state(rh, region, may_block); | |
384 | return state == RH_CLEAN || state == RH_DIRTY; | |
385 | } | |
386 | ||
387 | static void dispatch_bios(struct mirror_set *ms, struct bio_list *bio_list) | |
388 | { | |
389 | struct bio *bio; | |
390 | ||
391 | while ((bio = bio_list_pop(bio_list))) { | |
392 | queue_bio(ms, bio, WRITE); | |
393 | } | |
394 | } | |
395 | ||
f3ee6b2f JB |
396 | static void complete_resync_work(struct region *reg, int success) |
397 | { | |
398 | struct region_hash *rh = reg->rh; | |
399 | ||
400 | rh->log->type->set_region_sync(rh->log, reg->key, success); | |
b80aa7a0 JB |
401 | |
402 | /* | |
403 | * Dispatch the bios before we call 'wake_up_all'. | |
404 | * This is important because if we are suspending, | |
405 | * we want to know that recovery is complete and | |
406 | * the work queue is flushed. If we wake_up_all | |
407 | * before we dispatch_bios (queue bios and call wake()), | |
408 | * then we risk suspending before the work queue | |
409 | * has been properly flushed. | |
410 | */ | |
f3ee6b2f JB |
411 | dispatch_bios(rh->ms, ®->delayed_bios); |
412 | if (atomic_dec_and_test(&rh->recovery_in_flight)) | |
413 | wake_up_all(&_kmirrord_recovery_stopped); | |
414 | up(&rh->recovery_count); | |
415 | } | |
416 | ||
1da177e4 LT |
417 | static void rh_update_states(struct region_hash *rh) |
418 | { | |
419 | struct region *reg, *next; | |
420 | ||
421 | LIST_HEAD(clean); | |
422 | LIST_HEAD(recovered); | |
f44db678 | 423 | LIST_HEAD(failed_recovered); |
1da177e4 LT |
424 | |
425 | /* | |
426 | * Quickly grab the lists. | |
427 | */ | |
428 | write_lock_irq(&rh->hash_lock); | |
429 | spin_lock(&rh->region_lock); | |
430 | if (!list_empty(&rh->clean_regions)) { | |
c12bfc92 | 431 | list_splice_init(&rh->clean_regions, &clean); |
1da177e4 | 432 | |
943317ef | 433 | list_for_each_entry(reg, &clean, list) |
1da177e4 | 434 | list_del(®->hash_list); |
1da177e4 LT |
435 | } |
436 | ||
437 | if (!list_empty(&rh->recovered_regions)) { | |
c12bfc92 | 438 | list_splice_init(&rh->recovered_regions, &recovered); |
1da177e4 LT |
439 | |
440 | list_for_each_entry (reg, &recovered, list) | |
441 | list_del(®->hash_list); | |
442 | } | |
f44db678 JB |
443 | |
444 | if (!list_empty(&rh->failed_recovered_regions)) { | |
c12bfc92 RD |
445 | list_splice_init(&rh->failed_recovered_regions, |
446 | &failed_recovered); | |
f44db678 JB |
447 | |
448 | list_for_each_entry(reg, &failed_recovered, list) | |
449 | list_del(®->hash_list); | |
450 | } | |
451 | ||
1da177e4 LT |
452 | spin_unlock(&rh->region_lock); |
453 | write_unlock_irq(&rh->hash_lock); | |
454 | ||
455 | /* | |
456 | * All the regions on the recovered and clean lists have | |
457 | * now been pulled out of the system, so no need to do | |
458 | * any more locking. | |
459 | */ | |
460 | list_for_each_entry_safe (reg, next, &recovered, list) { | |
461 | rh->log->type->clear_region(rh->log, reg->key); | |
f3ee6b2f | 462 | complete_resync_work(reg, 1); |
1da177e4 LT |
463 | mempool_free(reg, rh->region_pool); |
464 | } | |
465 | ||
f44db678 JB |
466 | list_for_each_entry_safe(reg, next, &failed_recovered, list) { |
467 | complete_resync_work(reg, errors_handled(rh->ms) ? 0 : 1); | |
468 | mempool_free(reg, rh->region_pool); | |
469 | } | |
470 | ||
943317ef JB |
471 | list_for_each_entry_safe(reg, next, &clean, list) { |
472 | rh->log->type->clear_region(rh->log, reg->key); | |
1da177e4 | 473 | mempool_free(reg, rh->region_pool); |
943317ef JB |
474 | } |
475 | ||
476 | rh->log->type->flush(rh->log); | |
1da177e4 LT |
477 | } |
478 | ||
479 | static void rh_inc(struct region_hash *rh, region_t region) | |
480 | { | |
481 | struct region *reg; | |
482 | ||
483 | read_lock(&rh->hash_lock); | |
484 | reg = __rh_find(rh, region); | |
844e8d90 | 485 | |
7692c5dd | 486 | spin_lock_irq(&rh->region_lock); |
844e8d90 JN |
487 | atomic_inc(®->pending); |
488 | ||
1da177e4 | 489 | if (reg->state == RH_CLEAN) { |
1da177e4 LT |
490 | reg->state = RH_DIRTY; |
491 | list_del_init(®->list); /* take off the clean list */ | |
7692c5dd JB |
492 | spin_unlock_irq(&rh->region_lock); |
493 | ||
494 | rh->log->type->mark_region(rh->log, reg->key); | |
495 | } else | |
496 | spin_unlock_irq(&rh->region_lock); | |
497 | ||
1da177e4 | 498 | |
1da177e4 LT |
499 | read_unlock(&rh->hash_lock); |
500 | } | |
501 | ||
502 | static void rh_inc_pending(struct region_hash *rh, struct bio_list *bios) | |
503 | { | |
504 | struct bio *bio; | |
505 | ||
506 | for (bio = bios->head; bio; bio = bio->bi_next) | |
507 | rh_inc(rh, bio_to_region(rh, bio)); | |
508 | } | |
509 | ||
510 | static void rh_dec(struct region_hash *rh, region_t region) | |
511 | { | |
512 | unsigned long flags; | |
513 | struct region *reg; | |
514 | int should_wake = 0; | |
515 | ||
516 | read_lock(&rh->hash_lock); | |
517 | reg = __rh_lookup(rh, region); | |
518 | read_unlock(&rh->hash_lock); | |
519 | ||
7692c5dd | 520 | spin_lock_irqsave(&rh->region_lock, flags); |
1da177e4 | 521 | if (atomic_dec_and_test(®->pending)) { |
930d332a JN |
522 | /* |
523 | * There is no pending I/O for this region. | |
524 | * We can move the region to corresponding list for next action. | |
525 | * At this point, the region is not yet connected to any list. | |
526 | * | |
527 | * If the state is RH_NOSYNC, the region should be kept off | |
528 | * from clean list. | |
529 | * The hash entry for RH_NOSYNC will remain in memory | |
530 | * until the region is recovered or the map is reloaded. | |
531 | */ | |
532 | ||
533 | /* do nothing for RH_NOSYNC */ | |
1da177e4 LT |
534 | if (reg->state == RH_RECOVERING) { |
535 | list_add_tail(®->list, &rh->quiesced_regions); | |
930d332a | 536 | } else if (reg->state == RH_DIRTY) { |
1da177e4 LT |
537 | reg->state = RH_CLEAN; |
538 | list_add(®->list, &rh->clean_regions); | |
539 | } | |
1da177e4 LT |
540 | should_wake = 1; |
541 | } | |
7692c5dd | 542 | spin_unlock_irqrestore(&rh->region_lock, flags); |
1da177e4 LT |
543 | |
544 | if (should_wake) | |
6ad36fe2 | 545 | wake(rh->ms); |
1da177e4 LT |
546 | } |
547 | ||
548 | /* | |
549 | * Starts quiescing a region in preparation for recovery. | |
550 | */ | |
551 | static int __rh_recovery_prepare(struct region_hash *rh) | |
552 | { | |
553 | int r; | |
554 | struct region *reg; | |
555 | region_t region; | |
556 | ||
557 | /* | |
558 | * Ask the dirty log what's next. | |
559 | */ | |
560 | r = rh->log->type->get_resync_work(rh->log, ®ion); | |
561 | if (r <= 0) | |
562 | return r; | |
563 | ||
564 | /* | |
565 | * Get this region, and start it quiescing by setting the | |
566 | * recovering flag. | |
567 | */ | |
568 | read_lock(&rh->hash_lock); | |
569 | reg = __rh_find(rh, region); | |
570 | read_unlock(&rh->hash_lock); | |
571 | ||
572 | spin_lock_irq(&rh->region_lock); | |
573 | reg->state = RH_RECOVERING; | |
574 | ||
575 | /* Already quiesced ? */ | |
576 | if (atomic_read(®->pending)) | |
577 | list_del_init(®->list); | |
179e0917 AM |
578 | else |
579 | list_move(®->list, &rh->quiesced_regions); | |
1da177e4 | 580 | |
1da177e4 LT |
581 | spin_unlock_irq(&rh->region_lock); |
582 | ||
583 | return 1; | |
584 | } | |
585 | ||
586 | static void rh_recovery_prepare(struct region_hash *rh) | |
587 | { | |
33184048 JB |
588 | /* Extra reference to avoid race with rh_stop_recovery */ |
589 | atomic_inc(&rh->recovery_in_flight); | |
590 | ||
591 | while (!down_trylock(&rh->recovery_count)) { | |
592 | atomic_inc(&rh->recovery_in_flight); | |
1da177e4 | 593 | if (__rh_recovery_prepare(rh) <= 0) { |
33184048 | 594 | atomic_dec(&rh->recovery_in_flight); |
1da177e4 LT |
595 | up(&rh->recovery_count); |
596 | break; | |
597 | } | |
33184048 JB |
598 | } |
599 | ||
600 | /* Drop the extra reference */ | |
601 | if (atomic_dec_and_test(&rh->recovery_in_flight)) | |
602 | wake_up_all(&_kmirrord_recovery_stopped); | |
1da177e4 LT |
603 | } |
604 | ||
605 | /* | |
606 | * Returns any quiesced regions. | |
607 | */ | |
608 | static struct region *rh_recovery_start(struct region_hash *rh) | |
609 | { | |
610 | struct region *reg = NULL; | |
611 | ||
612 | spin_lock_irq(&rh->region_lock); | |
613 | if (!list_empty(&rh->quiesced_regions)) { | |
614 | reg = list_entry(rh->quiesced_regions.next, | |
615 | struct region, list); | |
616 | list_del_init(®->list); /* remove from the quiesced list */ | |
617 | } | |
618 | spin_unlock_irq(&rh->region_lock); | |
619 | ||
620 | return reg; | |
621 | } | |
622 | ||
1da177e4 LT |
623 | static void rh_recovery_end(struct region *reg, int success) |
624 | { | |
625 | struct region_hash *rh = reg->rh; | |
626 | ||
627 | spin_lock_irq(&rh->region_lock); | |
f44db678 JB |
628 | if (success) |
629 | list_add(®->list, ®->rh->recovered_regions); | |
630 | else { | |
631 | reg->state = RH_NOSYNC; | |
632 | list_add(®->list, ®->rh->failed_recovered_regions); | |
633 | } | |
1da177e4 LT |
634 | spin_unlock_irq(&rh->region_lock); |
635 | ||
6ad36fe2 | 636 | wake(rh->ms); |
1da177e4 LT |
637 | } |
638 | ||
fc1ff958 | 639 | static int rh_flush(struct region_hash *rh) |
1da177e4 | 640 | { |
fc1ff958 | 641 | return rh->log->type->flush(rh->log); |
1da177e4 LT |
642 | } |
643 | ||
644 | static void rh_delay(struct region_hash *rh, struct bio *bio) | |
645 | { | |
646 | struct region *reg; | |
647 | ||
648 | read_lock(&rh->hash_lock); | |
649 | reg = __rh_find(rh, bio_to_region(rh, bio)); | |
650 | bio_list_add(®->delayed_bios, bio); | |
651 | read_unlock(&rh->hash_lock); | |
652 | } | |
653 | ||
654 | static void rh_stop_recovery(struct region_hash *rh) | |
655 | { | |
656 | int i; | |
657 | ||
658 | /* wait for any recovering regions */ | |
659 | for (i = 0; i < MAX_RECOVERY; i++) | |
660 | down(&rh->recovery_count); | |
661 | } | |
662 | ||
663 | static void rh_start_recovery(struct region_hash *rh) | |
664 | { | |
665 | int i; | |
666 | ||
667 | for (i = 0; i < MAX_RECOVERY; i++) | |
668 | up(&rh->recovery_count); | |
669 | ||
6ad36fe2 | 670 | wake(rh->ms); |
1da177e4 LT |
671 | } |
672 | ||
06386bbf JB |
673 | #define MIN_READ_RECORDS 20 |
674 | struct dm_raid1_read_record { | |
675 | struct mirror *m; | |
676 | struct dm_bio_details details; | |
677 | }; | |
678 | ||
1da177e4 LT |
679 | /* |
680 | * Every mirror should look like this one. | |
681 | */ | |
682 | #define DEFAULT_MIRROR 0 | |
683 | ||
684 | /* | |
06386bbf JB |
685 | * This is yucky. We squirrel the mirror struct away inside |
686 | * bi_next for read/write buffers. This is safe since the bh | |
1da177e4 LT |
687 | * doesn't get submitted to the lower levels of block layer. |
688 | */ | |
06386bbf | 689 | static struct mirror *bio_get_m(struct bio *bio) |
1da177e4 | 690 | { |
06386bbf | 691 | return (struct mirror *) bio->bi_next; |
1da177e4 LT |
692 | } |
693 | ||
06386bbf | 694 | static void bio_set_m(struct bio *bio, struct mirror *m) |
1da177e4 | 695 | { |
06386bbf | 696 | bio->bi_next = (struct bio *) m; |
1da177e4 LT |
697 | } |
698 | ||
72f4b314 JB |
699 | static struct mirror *get_default_mirror(struct mirror_set *ms) |
700 | { | |
701 | return &ms->mirror[atomic_read(&ms->default_mirror)]; | |
702 | } | |
703 | ||
704 | static void set_default_mirror(struct mirror *m) | |
705 | { | |
706 | struct mirror_set *ms = m->ms; | |
707 | struct mirror *m0 = &(ms->mirror[0]); | |
708 | ||
709 | atomic_set(&ms->default_mirror, m - m0); | |
710 | } | |
711 | ||
712 | /* fail_mirror | |
713 | * @m: mirror device to fail | |
714 | * @error_type: one of the enum's, DM_RAID1_*_ERROR | |
715 | * | |
716 | * If errors are being handled, record the type of | |
717 | * error encountered for this device. If this type | |
718 | * of error has already been recorded, we can return; | |
719 | * otherwise, we must signal userspace by triggering | |
720 | * an event. Additionally, if the device is the | |
721 | * primary device, we must choose a new primary, but | |
722 | * only if the mirror is in-sync. | |
723 | * | |
724 | * This function must not block. | |
725 | */ | |
726 | static void fail_mirror(struct mirror *m, enum dm_raid1_error error_type) | |
727 | { | |
728 | struct mirror_set *ms = m->ms; | |
729 | struct mirror *new; | |
730 | ||
731 | if (!errors_handled(ms)) | |
732 | return; | |
733 | ||
734 | /* | |
735 | * error_count is used for nothing more than a | |
736 | * simple way to tell if a device has encountered | |
737 | * errors. | |
738 | */ | |
739 | atomic_inc(&m->error_count); | |
740 | ||
741 | if (test_and_set_bit(error_type, &m->error_type)) | |
742 | return; | |
743 | ||
744 | if (m != get_default_mirror(ms)) | |
745 | goto out; | |
746 | ||
747 | if (!ms->in_sync) { | |
748 | /* | |
749 | * Better to issue requests to same failing device | |
750 | * than to risk returning corrupt data. | |
751 | */ | |
752 | DMERR("Primary mirror (%s) failed while out-of-sync: " | |
753 | "Reads may fail.", m->dev->name); | |
754 | goto out; | |
755 | } | |
756 | ||
757 | for (new = ms->mirror; new < ms->mirror + ms->nr_mirrors; new++) | |
758 | if (!atomic_read(&new->error_count)) { | |
759 | set_default_mirror(new); | |
760 | break; | |
761 | } | |
762 | ||
763 | if (unlikely(new == ms->mirror + ms->nr_mirrors)) | |
764 | DMWARN("All sides of mirror have failed."); | |
765 | ||
766 | out: | |
767 | schedule_work(&ms->trigger_event); | |
768 | } | |
769 | ||
1da177e4 LT |
770 | /*----------------------------------------------------------------- |
771 | * Recovery. | |
772 | * | |
773 | * When a mirror is first activated we may find that some regions | |
774 | * are in the no-sync state. We have to recover these by | |
775 | * recopying from the default mirror to all the others. | |
776 | *---------------------------------------------------------------*/ | |
4cdc1d1f | 777 | static void recovery_complete(int read_err, unsigned long write_err, |
1da177e4 LT |
778 | void *context) |
779 | { | |
8f0205b7 JB |
780 | struct region *reg = (struct region *)context; |
781 | struct mirror_set *ms = reg->rh->ms; | |
782 | int m, bit = 0; | |
1da177e4 | 783 | |
8f0205b7 | 784 | if (read_err) { |
f44db678 JB |
785 | /* Read error means the failure of default mirror. */ |
786 | DMERR_LIMIT("Unable to read primary mirror during recovery"); | |
8f0205b7 JB |
787 | fail_mirror(get_default_mirror(ms), DM_RAID1_SYNC_ERROR); |
788 | } | |
f44db678 | 789 | |
8f0205b7 | 790 | if (write_err) { |
4cdc1d1f | 791 | DMERR_LIMIT("Write error during recovery (error = 0x%lx)", |
f44db678 | 792 | write_err); |
8f0205b7 JB |
793 | /* |
794 | * Bits correspond to devices (excluding default mirror). | |
795 | * The default mirror cannot change during recovery. | |
796 | */ | |
797 | for (m = 0; m < ms->nr_mirrors; m++) { | |
798 | if (&ms->mirror[m] == get_default_mirror(ms)) | |
799 | continue; | |
800 | if (test_bit(bit, &write_err)) | |
801 | fail_mirror(ms->mirror + m, | |
802 | DM_RAID1_SYNC_ERROR); | |
803 | bit++; | |
804 | } | |
805 | } | |
f44db678 | 806 | |
ce503f59 | 807 | rh_recovery_end(reg, !(read_err || write_err)); |
1da177e4 LT |
808 | } |
809 | ||
810 | static int recover(struct mirror_set *ms, struct region *reg) | |
811 | { | |
812 | int r; | |
813 | unsigned int i; | |
eb69aca5 | 814 | struct dm_io_region from, to[DM_KCOPYD_MAX_REGIONS], *dest; |
1da177e4 LT |
815 | struct mirror *m; |
816 | unsigned long flags = 0; | |
817 | ||
818 | /* fill in the source */ | |
72f4b314 | 819 | m = get_default_mirror(ms); |
1da177e4 LT |
820 | from.bdev = m->dev->bdev; |
821 | from.sector = m->offset + region_to_sector(reg->rh, reg->key); | |
822 | if (reg->key == (ms->nr_regions - 1)) { | |
823 | /* | |
824 | * The final region may be smaller than | |
825 | * region_size. | |
826 | */ | |
827 | from.count = ms->ti->len & (reg->rh->region_size - 1); | |
828 | if (!from.count) | |
829 | from.count = reg->rh->region_size; | |
830 | } else | |
831 | from.count = reg->rh->region_size; | |
832 | ||
833 | /* fill in the destinations */ | |
834 | for (i = 0, dest = to; i < ms->nr_mirrors; i++) { | |
72f4b314 | 835 | if (&ms->mirror[i] == get_default_mirror(ms)) |
1da177e4 LT |
836 | continue; |
837 | ||
838 | m = ms->mirror + i; | |
839 | dest->bdev = m->dev->bdev; | |
840 | dest->sector = m->offset + region_to_sector(reg->rh, reg->key); | |
841 | dest->count = from.count; | |
842 | dest++; | |
843 | } | |
844 | ||
845 | /* hand to kcopyd */ | |
f7c83e2e JB |
846 | if (!errors_handled(ms)) |
847 | set_bit(DM_KCOPYD_IGNORE_ERROR, &flags); | |
848 | ||
eb69aca5 HM |
849 | r = dm_kcopyd_copy(ms->kcopyd_client, &from, ms->nr_mirrors - 1, to, |
850 | flags, recovery_complete, reg); | |
1da177e4 LT |
851 | |
852 | return r; | |
853 | } | |
854 | ||
855 | static void do_recovery(struct mirror_set *ms) | |
856 | { | |
857 | int r; | |
858 | struct region *reg; | |
416cd17b | 859 | struct dm_dirty_log *log = ms->rh.log; |
1da177e4 LT |
860 | |
861 | /* | |
862 | * Start quiescing some regions. | |
863 | */ | |
864 | rh_recovery_prepare(&ms->rh); | |
865 | ||
866 | /* | |
867 | * Copy any already quiesced regions. | |
868 | */ | |
869 | while ((reg = rh_recovery_start(&ms->rh))) { | |
870 | r = recover(ms, reg); | |
871 | if (r) | |
872 | rh_recovery_end(reg, 0); | |
873 | } | |
874 | ||
875 | /* | |
876 | * Update the in sync flag. | |
877 | */ | |
878 | if (!ms->in_sync && | |
879 | (log->type->get_sync_count(log) == ms->nr_regions)) { | |
880 | /* the sync is complete */ | |
881 | dm_table_event(ms->ti->table); | |
882 | ms->in_sync = 1; | |
883 | } | |
884 | } | |
885 | ||
886 | /*----------------------------------------------------------------- | |
887 | * Reads | |
888 | *---------------------------------------------------------------*/ | |
889 | static struct mirror *choose_mirror(struct mirror_set *ms, sector_t sector) | |
890 | { | |
06386bbf JB |
891 | struct mirror *m = get_default_mirror(ms); |
892 | ||
893 | do { | |
894 | if (likely(!atomic_read(&m->error_count))) | |
895 | return m; | |
896 | ||
897 | if (m-- == ms->mirror) | |
898 | m += ms->nr_mirrors; | |
899 | } while (m != get_default_mirror(ms)); | |
900 | ||
901 | return NULL; | |
902 | } | |
903 | ||
904 | static int default_ok(struct mirror *m) | |
905 | { | |
906 | struct mirror *default_mirror = get_default_mirror(m->ms); | |
907 | ||
908 | return !atomic_read(&default_mirror->error_count); | |
909 | } | |
910 | ||
911 | static int mirror_available(struct mirror_set *ms, struct bio *bio) | |
912 | { | |
913 | region_t region = bio_to_region(&ms->rh, bio); | |
914 | ||
915 | if (ms->rh.log->type->in_sync(ms->rh.log, region, 0)) | |
916 | return choose_mirror(ms, bio->bi_sector) ? 1 : 0; | |
917 | ||
918 | return 0; | |
1da177e4 LT |
919 | } |
920 | ||
921 | /* | |
922 | * remap a buffer to a particular mirror. | |
923 | */ | |
06386bbf JB |
924 | static sector_t map_sector(struct mirror *m, struct bio *bio) |
925 | { | |
926 | return m->offset + (bio->bi_sector - m->ms->ti->begin); | |
927 | } | |
928 | ||
929 | static void map_bio(struct mirror *m, struct bio *bio) | |
1da177e4 LT |
930 | { |
931 | bio->bi_bdev = m->dev->bdev; | |
06386bbf JB |
932 | bio->bi_sector = map_sector(m, bio); |
933 | } | |
934 | ||
22a1ceb1 | 935 | static void map_region(struct dm_io_region *io, struct mirror *m, |
06386bbf JB |
936 | struct bio *bio) |
937 | { | |
938 | io->bdev = m->dev->bdev; | |
939 | io->sector = map_sector(m, bio); | |
940 | io->count = bio->bi_size >> 9; | |
941 | } | |
942 | ||
943 | /*----------------------------------------------------------------- | |
944 | * Reads | |
945 | *---------------------------------------------------------------*/ | |
946 | static void read_callback(unsigned long error, void *context) | |
947 | { | |
948 | struct bio *bio = context; | |
949 | struct mirror *m; | |
950 | ||
951 | m = bio_get_m(bio); | |
952 | bio_set_m(bio, NULL); | |
953 | ||
954 | if (likely(!error)) { | |
955 | bio_endio(bio, 0); | |
956 | return; | |
957 | } | |
958 | ||
959 | fail_mirror(m, DM_RAID1_READ_ERROR); | |
960 | ||
961 | if (likely(default_ok(m)) || mirror_available(m->ms, bio)) { | |
962 | DMWARN_LIMIT("Read failure on mirror device %s. " | |
963 | "Trying alternative device.", | |
964 | m->dev->name); | |
965 | queue_bio(m->ms, bio, bio_rw(bio)); | |
966 | return; | |
967 | } | |
968 | ||
969 | DMERR_LIMIT("Read failure on mirror device %s. Failing I/O.", | |
970 | m->dev->name); | |
971 | bio_endio(bio, -EIO); | |
972 | } | |
973 | ||
974 | /* Asynchronous read. */ | |
975 | static void read_async_bio(struct mirror *m, struct bio *bio) | |
976 | { | |
22a1ceb1 | 977 | struct dm_io_region io; |
06386bbf JB |
978 | struct dm_io_request io_req = { |
979 | .bi_rw = READ, | |
980 | .mem.type = DM_IO_BVEC, | |
981 | .mem.ptr.bvec = bio->bi_io_vec + bio->bi_idx, | |
982 | .notify.fn = read_callback, | |
983 | .notify.context = bio, | |
984 | .client = m->ms->io_client, | |
985 | }; | |
986 | ||
987 | map_region(&io, m, bio); | |
988 | bio_set_m(bio, m); | |
989 | (void) dm_io(&io_req, 1, &io, NULL); | |
1da177e4 LT |
990 | } |
991 | ||
992 | static void do_reads(struct mirror_set *ms, struct bio_list *reads) | |
993 | { | |
994 | region_t region; | |
995 | struct bio *bio; | |
996 | struct mirror *m; | |
997 | ||
998 | while ((bio = bio_list_pop(reads))) { | |
999 | region = bio_to_region(&ms->rh, bio); | |
06386bbf | 1000 | m = get_default_mirror(ms); |
1da177e4 LT |
1001 | |
1002 | /* | |
1003 | * We can only read balance if the region is in sync. | |
1004 | */ | |
06386bbf | 1005 | if (likely(rh_in_sync(&ms->rh, region, 1))) |
1da177e4 | 1006 | m = choose_mirror(ms, bio->bi_sector); |
06386bbf JB |
1007 | else if (m && atomic_read(&m->error_count)) |
1008 | m = NULL; | |
1da177e4 | 1009 | |
06386bbf JB |
1010 | if (likely(m)) |
1011 | read_async_bio(m, bio); | |
1012 | else | |
1013 | bio_endio(bio, -EIO); | |
1da177e4 LT |
1014 | } |
1015 | } | |
1016 | ||
1017 | /*----------------------------------------------------------------- | |
1018 | * Writes. | |
1019 | * | |
1020 | * We do different things with the write io depending on the | |
1021 | * state of the region that it's in: | |
1022 | * | |
1023 | * SYNC: increment pending, use kcopyd to write to *all* mirrors | |
1024 | * RECOVERING: delay the io until recovery completes | |
1025 | * NOSYNC: increment pending, just write to the default mirror | |
1026 | *---------------------------------------------------------------*/ | |
72f4b314 JB |
1027 | |
1028 | /* __bio_mark_nosync | |
1029 | * @ms | |
1030 | * @bio | |
1031 | * @done | |
1032 | * @error | |
1033 | * | |
1034 | * The bio was written on some mirror(s) but failed on other mirror(s). | |
1035 | * We can successfully endio the bio but should avoid the region being | |
1036 | * marked clean by setting the state RH_NOSYNC. | |
1037 | * | |
1038 | * This function is _not_ safe in interrupt context! | |
1039 | */ | |
1040 | static void __bio_mark_nosync(struct mirror_set *ms, | |
1041 | struct bio *bio, unsigned done, int error) | |
1042 | { | |
1043 | unsigned long flags; | |
1044 | struct region_hash *rh = &ms->rh; | |
416cd17b | 1045 | struct dm_dirty_log *log = ms->rh.log; |
72f4b314 JB |
1046 | struct region *reg; |
1047 | region_t region = bio_to_region(rh, bio); | |
1048 | int recovering = 0; | |
1049 | ||
1050 | /* We must inform the log that the sync count has changed. */ | |
1051 | log->type->set_region_sync(log, region, 0); | |
1052 | ms->in_sync = 0; | |
1053 | ||
1054 | read_lock(&rh->hash_lock); | |
1055 | reg = __rh_find(rh, region); | |
1056 | read_unlock(&rh->hash_lock); | |
1057 | ||
1058 | /* region hash entry should exist because write was in-flight */ | |
1059 | BUG_ON(!reg); | |
1060 | BUG_ON(!list_empty(®->list)); | |
1061 | ||
1062 | spin_lock_irqsave(&rh->region_lock, flags); | |
1063 | /* | |
1064 | * Possible cases: | |
1065 | * 1) RH_DIRTY | |
1066 | * 2) RH_NOSYNC: was dirty, other preceeding writes failed | |
1067 | * 3) RH_RECOVERING: flushing pending writes | |
1068 | * Either case, the region should have not been connected to list. | |
1069 | */ | |
1070 | recovering = (reg->state == RH_RECOVERING); | |
1071 | reg->state = RH_NOSYNC; | |
1072 | BUG_ON(!list_empty(®->list)); | |
1073 | spin_unlock_irqrestore(&rh->region_lock, flags); | |
1074 | ||
1075 | bio_endio(bio, error); | |
1076 | if (recovering) | |
1077 | complete_resync_work(reg, 0); | |
1078 | } | |
1079 | ||
1da177e4 LT |
1080 | static void write_callback(unsigned long error, void *context) |
1081 | { | |
72f4b314 | 1082 | unsigned i, ret = 0; |
1da177e4 LT |
1083 | struct bio *bio = (struct bio *) context; |
1084 | struct mirror_set *ms; | |
72f4b314 JB |
1085 | int uptodate = 0; |
1086 | int should_wake = 0; | |
1087 | unsigned long flags; | |
1da177e4 | 1088 | |
06386bbf JB |
1089 | ms = bio_get_m(bio)->ms; |
1090 | bio_set_m(bio, NULL); | |
1da177e4 LT |
1091 | |
1092 | /* | |
1093 | * NOTE: We don't decrement the pending count here, | |
1094 | * instead it is done by the targets endio function. | |
1095 | * This way we handle both writes to SYNC and NOSYNC | |
1096 | * regions with the same code. | |
1097 | */ | |
72f4b314 JB |
1098 | if (likely(!error)) |
1099 | goto out; | |
1da177e4 | 1100 | |
72f4b314 JB |
1101 | for (i = 0; i < ms->nr_mirrors; i++) |
1102 | if (test_bit(i, &error)) | |
1103 | fail_mirror(ms->mirror + i, DM_RAID1_WRITE_ERROR); | |
1104 | else | |
1105 | uptodate = 1; | |
1106 | ||
1107 | if (unlikely(!uptodate)) { | |
1108 | DMERR("All replicated volumes dead, failing I/O"); | |
1109 | /* None of the writes succeeded, fail the I/O. */ | |
1110 | ret = -EIO; | |
1111 | } else if (errors_handled(ms)) { | |
1da177e4 | 1112 | /* |
72f4b314 JB |
1113 | * Need to raise event. Since raising |
1114 | * events can block, we need to do it in | |
1115 | * the main thread. | |
1da177e4 | 1116 | */ |
72f4b314 JB |
1117 | spin_lock_irqsave(&ms->lock, flags); |
1118 | if (!ms->failures.head) | |
1119 | should_wake = 1; | |
1120 | bio_list_add(&ms->failures, bio); | |
1121 | spin_unlock_irqrestore(&ms->lock, flags); | |
1122 | if (should_wake) | |
1123 | wake(ms); | |
1124 | return; | |
1da177e4 | 1125 | } |
72f4b314 JB |
1126 | out: |
1127 | bio_endio(bio, ret); | |
1da177e4 LT |
1128 | } |
1129 | ||
1130 | static void do_write(struct mirror_set *ms, struct bio *bio) | |
1131 | { | |
1132 | unsigned int i; | |
22a1ceb1 | 1133 | struct dm_io_region io[ms->nr_mirrors], *dest = io; |
1da177e4 | 1134 | struct mirror *m; |
88be163a MB |
1135 | struct dm_io_request io_req = { |
1136 | .bi_rw = WRITE, | |
1137 | .mem.type = DM_IO_BVEC, | |
1138 | .mem.ptr.bvec = bio->bi_io_vec + bio->bi_idx, | |
1139 | .notify.fn = write_callback, | |
1140 | .notify.context = bio, | |
1141 | .client = ms->io_client, | |
1142 | }; | |
1da177e4 | 1143 | |
06386bbf JB |
1144 | for (i = 0, m = ms->mirror; i < ms->nr_mirrors; i++, m++) |
1145 | map_region(dest++, m, bio); | |
1da177e4 | 1146 | |
06386bbf JB |
1147 | /* |
1148 | * Use default mirror because we only need it to retrieve the reference | |
1149 | * to the mirror set in write_callback(). | |
1150 | */ | |
1151 | bio_set_m(bio, get_default_mirror(ms)); | |
88be163a MB |
1152 | |
1153 | (void) dm_io(&io_req, ms->nr_mirrors, io, NULL); | |
1da177e4 LT |
1154 | } |
1155 | ||
1156 | static void do_writes(struct mirror_set *ms, struct bio_list *writes) | |
1157 | { | |
1158 | int state; | |
1159 | struct bio *bio; | |
1160 | struct bio_list sync, nosync, recover, *this_list = NULL; | |
1161 | ||
1162 | if (!writes->head) | |
1163 | return; | |
1164 | ||
1165 | /* | |
1166 | * Classify each write. | |
1167 | */ | |
1168 | bio_list_init(&sync); | |
1169 | bio_list_init(&nosync); | |
1170 | bio_list_init(&recover); | |
1171 | ||
1172 | while ((bio = bio_list_pop(writes))) { | |
1173 | state = rh_state(&ms->rh, bio_to_region(&ms->rh, bio), 1); | |
1174 | switch (state) { | |
1175 | case RH_CLEAN: | |
1176 | case RH_DIRTY: | |
1177 | this_list = &sync; | |
1178 | break; | |
1179 | ||
1180 | case RH_NOSYNC: | |
1181 | this_list = &nosync; | |
1182 | break; | |
1183 | ||
1184 | case RH_RECOVERING: | |
1185 | this_list = &recover; | |
1186 | break; | |
1187 | } | |
1188 | ||
1189 | bio_list_add(this_list, bio); | |
1190 | } | |
1191 | ||
1192 | /* | |
1193 | * Increment the pending counts for any regions that will | |
1194 | * be written to (writes to recover regions are going to | |
1195 | * be delayed). | |
1196 | */ | |
1197 | rh_inc_pending(&ms->rh, &sync); | |
1198 | rh_inc_pending(&ms->rh, &nosync); | |
fc1ff958 | 1199 | ms->log_failure = rh_flush(&ms->rh) ? 1 : 0; |
1da177e4 LT |
1200 | |
1201 | /* | |
1202 | * Dispatch io. | |
1203 | */ | |
b80aa7a0 JB |
1204 | if (unlikely(ms->log_failure)) { |
1205 | spin_lock_irq(&ms->lock); | |
1206 | bio_list_merge(&ms->failures, &sync); | |
1207 | spin_unlock_irq(&ms->lock); | |
a2aebe03 | 1208 | wake(ms); |
b80aa7a0 | 1209 | } else |
fc1ff958 | 1210 | while ((bio = bio_list_pop(&sync))) |
b80aa7a0 | 1211 | do_write(ms, bio); |
1da177e4 LT |
1212 | |
1213 | while ((bio = bio_list_pop(&recover))) | |
1214 | rh_delay(&ms->rh, bio); | |
1215 | ||
1216 | while ((bio = bio_list_pop(&nosync))) { | |
06386bbf | 1217 | map_bio(get_default_mirror(ms), bio); |
1da177e4 LT |
1218 | generic_make_request(bio); |
1219 | } | |
1220 | } | |
1221 | ||
72f4b314 JB |
1222 | static void do_failures(struct mirror_set *ms, struct bio_list *failures) |
1223 | { | |
1224 | struct bio *bio; | |
1225 | ||
1226 | if (!failures->head) | |
1227 | return; | |
1228 | ||
b80aa7a0 JB |
1229 | if (!ms->log_failure) { |
1230 | while ((bio = bio_list_pop(failures))) | |
1231 | __bio_mark_nosync(ms, bio, bio->bi_size, 0); | |
1232 | return; | |
1233 | } | |
1234 | ||
1235 | /* | |
1236 | * If the log has failed, unattempted writes are being | |
1237 | * put on the failures list. We can't issue those writes | |
1238 | * until a log has been marked, so we must store them. | |
1239 | * | |
1240 | * If a 'noflush' suspend is in progress, we can requeue | |
1241 | * the I/O's to the core. This give userspace a chance | |
1242 | * to reconfigure the mirror, at which point the core | |
1243 | * will reissue the writes. If the 'noflush' flag is | |
1244 | * not set, we have no choice but to return errors. | |
1245 | * | |
1246 | * Some writes on the failures list may have been | |
1247 | * submitted before the log failure and represent a | |
1248 | * failure to write to one of the devices. It is ok | |
1249 | * for us to treat them the same and requeue them | |
1250 | * as well. | |
1251 | */ | |
1252 | if (dm_noflush_suspending(ms->ti)) { | |
1253 | while ((bio = bio_list_pop(failures))) | |
1254 | bio_endio(bio, DM_ENDIO_REQUEUE); | |
1255 | return; | |
1256 | } | |
1257 | ||
1258 | if (atomic_read(&ms->suspend)) { | |
1259 | while ((bio = bio_list_pop(failures))) | |
1260 | bio_endio(bio, -EIO); | |
1261 | return; | |
1262 | } | |
1263 | ||
1264 | spin_lock_irq(&ms->lock); | |
1265 | bio_list_merge(&ms->failures, failures); | |
1266 | spin_unlock_irq(&ms->lock); | |
1267 | ||
a2aebe03 | 1268 | delayed_wake(ms); |
72f4b314 JB |
1269 | } |
1270 | ||
1271 | static void trigger_event(struct work_struct *work) | |
1272 | { | |
1273 | struct mirror_set *ms = | |
1274 | container_of(work, struct mirror_set, trigger_event); | |
1275 | ||
1276 | dm_table_event(ms->ti->table); | |
1277 | } | |
1278 | ||
1da177e4 LT |
1279 | /*----------------------------------------------------------------- |
1280 | * kmirrord | |
1281 | *---------------------------------------------------------------*/ | |
a2aebe03 | 1282 | static void do_mirror(struct work_struct *work) |
1da177e4 | 1283 | { |
6ad36fe2 HS |
1284 | struct mirror_set *ms =container_of(work, struct mirror_set, |
1285 | kmirrord_work); | |
72f4b314 JB |
1286 | struct bio_list reads, writes, failures; |
1287 | unsigned long flags; | |
1da177e4 | 1288 | |
72f4b314 | 1289 | spin_lock_irqsave(&ms->lock, flags); |
1da177e4 LT |
1290 | reads = ms->reads; |
1291 | writes = ms->writes; | |
72f4b314 | 1292 | failures = ms->failures; |
1da177e4 LT |
1293 | bio_list_init(&ms->reads); |
1294 | bio_list_init(&ms->writes); | |
72f4b314 JB |
1295 | bio_list_init(&ms->failures); |
1296 | spin_unlock_irqrestore(&ms->lock, flags); | |
1da177e4 LT |
1297 | |
1298 | rh_update_states(&ms->rh); | |
1299 | do_recovery(ms); | |
1300 | do_reads(ms, &reads); | |
1301 | do_writes(ms, &writes); | |
72f4b314 | 1302 | do_failures(ms, &failures); |
7ff14a36 MP |
1303 | |
1304 | dm_table_unplug_all(ms->ti->table); | |
1da177e4 LT |
1305 | } |
1306 | ||
72f4b314 | 1307 | |
1da177e4 LT |
1308 | /*----------------------------------------------------------------- |
1309 | * Target functions | |
1310 | *---------------------------------------------------------------*/ | |
1311 | static struct mirror_set *alloc_context(unsigned int nr_mirrors, | |
1312 | uint32_t region_size, | |
1313 | struct dm_target *ti, | |
416cd17b | 1314 | struct dm_dirty_log *dl) |
1da177e4 LT |
1315 | { |
1316 | size_t len; | |
1317 | struct mirror_set *ms = NULL; | |
1318 | ||
1da177e4 LT |
1319 | len = sizeof(*ms) + (sizeof(ms->mirror[0]) * nr_mirrors); |
1320 | ||
dd00cc48 | 1321 | ms = kzalloc(len, GFP_KERNEL); |
1da177e4 | 1322 | if (!ms) { |
72d94861 | 1323 | ti->error = "Cannot allocate mirror context"; |
1da177e4 LT |
1324 | return NULL; |
1325 | } | |
1326 | ||
1da177e4 LT |
1327 | spin_lock_init(&ms->lock); |
1328 | ||
1329 | ms->ti = ti; | |
1330 | ms->nr_mirrors = nr_mirrors; | |
1331 | ms->nr_regions = dm_sector_div_up(ti->len, region_size); | |
1332 | ms->in_sync = 0; | |
b80aa7a0 JB |
1333 | ms->log_failure = 0; |
1334 | atomic_set(&ms->suspend, 0); | |
72f4b314 | 1335 | atomic_set(&ms->default_mirror, DEFAULT_MIRROR); |
1da177e4 | 1336 | |
06386bbf JB |
1337 | len = sizeof(struct dm_raid1_read_record); |
1338 | ms->read_record_pool = mempool_create_kmalloc_pool(MIN_READ_RECORDS, | |
1339 | len); | |
1340 | if (!ms->read_record_pool) { | |
1341 | ti->error = "Error creating mirror read_record_pool"; | |
1342 | kfree(ms); | |
1343 | return NULL; | |
1344 | } | |
1345 | ||
88be163a MB |
1346 | ms->io_client = dm_io_client_create(DM_IO_PAGES); |
1347 | if (IS_ERR(ms->io_client)) { | |
1348 | ti->error = "Error creating dm_io client"; | |
06386bbf | 1349 | mempool_destroy(ms->read_record_pool); |
88be163a MB |
1350 | kfree(ms); |
1351 | return NULL; | |
1352 | } | |
1353 | ||
1da177e4 | 1354 | if (rh_init(&ms->rh, ms, dl, region_size, ms->nr_regions)) { |
72d94861 | 1355 | ti->error = "Error creating dirty region hash"; |
a72cf737 | 1356 | dm_io_client_destroy(ms->io_client); |
06386bbf | 1357 | mempool_destroy(ms->read_record_pool); |
1da177e4 LT |
1358 | kfree(ms); |
1359 | return NULL; | |
1360 | } | |
1361 | ||
1362 | return ms; | |
1363 | } | |
1364 | ||
1365 | static void free_context(struct mirror_set *ms, struct dm_target *ti, | |
1366 | unsigned int m) | |
1367 | { | |
1368 | while (m--) | |
1369 | dm_put_device(ti, ms->mirror[m].dev); | |
1370 | ||
88be163a | 1371 | dm_io_client_destroy(ms->io_client); |
1da177e4 | 1372 | rh_exit(&ms->rh); |
06386bbf | 1373 | mempool_destroy(ms->read_record_pool); |
1da177e4 LT |
1374 | kfree(ms); |
1375 | } | |
1376 | ||
1377 | static inline int _check_region_size(struct dm_target *ti, uint32_t size) | |
1378 | { | |
6f3c3f0a | 1379 | return !(size % (PAGE_SIZE >> 9) || !is_power_of_2(size) || |
1da177e4 LT |
1380 | size > ti->len); |
1381 | } | |
1382 | ||
1383 | static int get_mirror(struct mirror_set *ms, struct dm_target *ti, | |
1384 | unsigned int mirror, char **argv) | |
1385 | { | |
4ee218cd | 1386 | unsigned long long offset; |
1da177e4 | 1387 | |
4ee218cd | 1388 | if (sscanf(argv[1], "%llu", &offset) != 1) { |
72d94861 | 1389 | ti->error = "Invalid offset"; |
1da177e4 LT |
1390 | return -EINVAL; |
1391 | } | |
1392 | ||
1393 | if (dm_get_device(ti, argv[0], offset, ti->len, | |
1394 | dm_table_get_mode(ti->table), | |
1395 | &ms->mirror[mirror].dev)) { | |
72d94861 | 1396 | ti->error = "Device lookup failure"; |
1da177e4 LT |
1397 | return -ENXIO; |
1398 | } | |
1399 | ||
aa5617c5 | 1400 | ms->mirror[mirror].ms = ms; |
72f4b314 JB |
1401 | atomic_set(&(ms->mirror[mirror].error_count), 0); |
1402 | ms->mirror[mirror].error_type = 0; | |
1da177e4 LT |
1403 | ms->mirror[mirror].offset = offset; |
1404 | ||
1405 | return 0; | |
1406 | } | |
1407 | ||
1da177e4 LT |
1408 | /* |
1409 | * Create dirty log: log_type #log_params <log_params> | |
1410 | */ | |
416cd17b | 1411 | static struct dm_dirty_log *create_dirty_log(struct dm_target *ti, |
1da177e4 LT |
1412 | unsigned int argc, char **argv, |
1413 | unsigned int *args_used) | |
1414 | { | |
1415 | unsigned int param_count; | |
416cd17b | 1416 | struct dm_dirty_log *dl; |
1da177e4 LT |
1417 | |
1418 | if (argc < 2) { | |
72d94861 | 1419 | ti->error = "Insufficient mirror log arguments"; |
1da177e4 LT |
1420 | return NULL; |
1421 | } | |
1422 | ||
1423 | if (sscanf(argv[1], "%u", ¶m_count) != 1) { | |
72d94861 | 1424 | ti->error = "Invalid mirror log argument count"; |
1da177e4 LT |
1425 | return NULL; |
1426 | } | |
1427 | ||
1428 | *args_used = 2 + param_count; | |
1429 | ||
1430 | if (argc < *args_used) { | |
72d94861 | 1431 | ti->error = "Insufficient mirror log arguments"; |
1da177e4 LT |
1432 | return NULL; |
1433 | } | |
1434 | ||
416cd17b | 1435 | dl = dm_dirty_log_create(argv[0], ti, param_count, argv + 2); |
1da177e4 | 1436 | if (!dl) { |
72d94861 | 1437 | ti->error = "Error creating mirror dirty log"; |
1da177e4 LT |
1438 | return NULL; |
1439 | } | |
1440 | ||
1441 | if (!_check_region_size(ti, dl->type->get_region_size(dl))) { | |
72d94861 | 1442 | ti->error = "Invalid region size"; |
416cd17b | 1443 | dm_dirty_log_destroy(dl); |
1da177e4 LT |
1444 | return NULL; |
1445 | } | |
1446 | ||
1447 | return dl; | |
1448 | } | |
1449 | ||
a8e6afa2 JB |
1450 | static int parse_features(struct mirror_set *ms, unsigned argc, char **argv, |
1451 | unsigned *args_used) | |
1452 | { | |
1453 | unsigned num_features; | |
1454 | struct dm_target *ti = ms->ti; | |
1455 | ||
1456 | *args_used = 0; | |
1457 | ||
1458 | if (!argc) | |
1459 | return 0; | |
1460 | ||
1461 | if (sscanf(argv[0], "%u", &num_features) != 1) { | |
1462 | ti->error = "Invalid number of features"; | |
1463 | return -EINVAL; | |
1464 | } | |
1465 | ||
1466 | argc--; | |
1467 | argv++; | |
1468 | (*args_used)++; | |
1469 | ||
1470 | if (num_features > argc) { | |
1471 | ti->error = "Not enough arguments to support feature count"; | |
1472 | return -EINVAL; | |
1473 | } | |
1474 | ||
1475 | if (!strcmp("handle_errors", argv[0])) | |
1476 | ms->features |= DM_RAID1_HANDLE_ERRORS; | |
1477 | else { | |
1478 | ti->error = "Unrecognised feature requested"; | |
1479 | return -EINVAL; | |
1480 | } | |
1481 | ||
1482 | (*args_used)++; | |
1483 | ||
1484 | return 0; | |
1485 | } | |
1486 | ||
1da177e4 LT |
1487 | /* |
1488 | * Construct a mirror mapping: | |
1489 | * | |
1490 | * log_type #log_params <log_params> | |
1491 | * #mirrors [mirror_path offset]{2,} | |
a8e6afa2 | 1492 | * [#features <features>] |
1da177e4 LT |
1493 | * |
1494 | * log_type is "core" or "disk" | |
1495 | * #log_params is between 1 and 3 | |
a8e6afa2 JB |
1496 | * |
1497 | * If present, features must be "handle_errors". | |
1da177e4 | 1498 | */ |
1da177e4 LT |
1499 | static int mirror_ctr(struct dm_target *ti, unsigned int argc, char **argv) |
1500 | { | |
1501 | int r; | |
1502 | unsigned int nr_mirrors, m, args_used; | |
1503 | struct mirror_set *ms; | |
416cd17b | 1504 | struct dm_dirty_log *dl; |
1da177e4 LT |
1505 | |
1506 | dl = create_dirty_log(ti, argc, argv, &args_used); | |
1507 | if (!dl) | |
1508 | return -EINVAL; | |
1509 | ||
1510 | argv += args_used; | |
1511 | argc -= args_used; | |
1512 | ||
1513 | if (!argc || sscanf(argv[0], "%u", &nr_mirrors) != 1 || | |
eb69aca5 | 1514 | nr_mirrors < 2 || nr_mirrors > DM_KCOPYD_MAX_REGIONS + 1) { |
72d94861 | 1515 | ti->error = "Invalid number of mirrors"; |
416cd17b | 1516 | dm_dirty_log_destroy(dl); |
1da177e4 LT |
1517 | return -EINVAL; |
1518 | } | |
1519 | ||
1520 | argv++, argc--; | |
1521 | ||
a8e6afa2 JB |
1522 | if (argc < nr_mirrors * 2) { |
1523 | ti->error = "Too few mirror arguments"; | |
416cd17b | 1524 | dm_dirty_log_destroy(dl); |
1da177e4 LT |
1525 | return -EINVAL; |
1526 | } | |
1527 | ||
1528 | ms = alloc_context(nr_mirrors, dl->type->get_region_size(dl), ti, dl); | |
1529 | if (!ms) { | |
416cd17b | 1530 | dm_dirty_log_destroy(dl); |
1da177e4 LT |
1531 | return -ENOMEM; |
1532 | } | |
1533 | ||
1534 | /* Get the mirror parameter sets */ | |
1535 | for (m = 0; m < nr_mirrors; m++) { | |
1536 | r = get_mirror(ms, ti, m, argv); | |
1537 | if (r) { | |
1538 | free_context(ms, ti, m); | |
1539 | return r; | |
1540 | } | |
1541 | argv += 2; | |
1542 | argc -= 2; | |
1543 | } | |
1544 | ||
1545 | ti->private = ms; | |
d88854f0 | 1546 | ti->split_io = ms->rh.region_size; |
1da177e4 | 1547 | |
6ad36fe2 HS |
1548 | ms->kmirrord_wq = create_singlethread_workqueue("kmirrord"); |
1549 | if (!ms->kmirrord_wq) { | |
1550 | DMERR("couldn't start kmirrord"); | |
a72cf737 DM |
1551 | r = -ENOMEM; |
1552 | goto err_free_context; | |
6ad36fe2 HS |
1553 | } |
1554 | INIT_WORK(&ms->kmirrord_work, do_mirror); | |
a2aebe03 MP |
1555 | init_timer(&ms->timer); |
1556 | ms->timer_pending = 0; | |
72f4b314 | 1557 | INIT_WORK(&ms->trigger_event, trigger_event); |
6ad36fe2 | 1558 | |
a8e6afa2 | 1559 | r = parse_features(ms, argc, argv, &args_used); |
a72cf737 DM |
1560 | if (r) |
1561 | goto err_destroy_wq; | |
a8e6afa2 JB |
1562 | |
1563 | argv += args_used; | |
1564 | argc -= args_used; | |
1565 | ||
f44db678 JB |
1566 | /* |
1567 | * Any read-balancing addition depends on the | |
1568 | * DM_RAID1_HANDLE_ERRORS flag being present. | |
1569 | * This is because the decision to balance depends | |
1570 | * on the sync state of a region. If the above | |
1571 | * flag is not present, we ignore errors; and | |
1572 | * the sync state may be inaccurate. | |
1573 | */ | |
1574 | ||
a8e6afa2 JB |
1575 | if (argc) { |
1576 | ti->error = "Too many mirror arguments"; | |
a72cf737 DM |
1577 | r = -EINVAL; |
1578 | goto err_destroy_wq; | |
a8e6afa2 JB |
1579 | } |
1580 | ||
eb69aca5 | 1581 | r = dm_kcopyd_client_create(DM_IO_PAGES, &ms->kcopyd_client); |
a72cf737 DM |
1582 | if (r) |
1583 | goto err_destroy_wq; | |
1da177e4 | 1584 | |
6ad36fe2 | 1585 | wake(ms); |
1da177e4 | 1586 | return 0; |
a72cf737 DM |
1587 | |
1588 | err_destroy_wq: | |
1589 | destroy_workqueue(ms->kmirrord_wq); | |
1590 | err_free_context: | |
1591 | free_context(ms, ti, ms->nr_mirrors); | |
1592 | return r; | |
1da177e4 LT |
1593 | } |
1594 | ||
1595 | static void mirror_dtr(struct dm_target *ti) | |
1596 | { | |
1597 | struct mirror_set *ms = (struct mirror_set *) ti->private; | |
1598 | ||
a2aebe03 | 1599 | del_timer_sync(&ms->timer); |
6ad36fe2 | 1600 | flush_workqueue(ms->kmirrord_wq); |
eb69aca5 | 1601 | dm_kcopyd_client_destroy(ms->kcopyd_client); |
6ad36fe2 | 1602 | destroy_workqueue(ms->kmirrord_wq); |
1da177e4 LT |
1603 | free_context(ms, ti, ms->nr_mirrors); |
1604 | } | |
1605 | ||
1606 | static void queue_bio(struct mirror_set *ms, struct bio *bio, int rw) | |
1607 | { | |
72f4b314 | 1608 | unsigned long flags; |
1da177e4 LT |
1609 | int should_wake = 0; |
1610 | struct bio_list *bl; | |
1611 | ||
1612 | bl = (rw == WRITE) ? &ms->writes : &ms->reads; | |
72f4b314 | 1613 | spin_lock_irqsave(&ms->lock, flags); |
1da177e4 LT |
1614 | should_wake = !(bl->head); |
1615 | bio_list_add(bl, bio); | |
72f4b314 | 1616 | spin_unlock_irqrestore(&ms->lock, flags); |
1da177e4 LT |
1617 | |
1618 | if (should_wake) | |
6ad36fe2 | 1619 | wake(ms); |
1da177e4 LT |
1620 | } |
1621 | ||
1622 | /* | |
1623 | * Mirror mapping function | |
1624 | */ | |
1625 | static int mirror_map(struct dm_target *ti, struct bio *bio, | |
1626 | union map_info *map_context) | |
1627 | { | |
1628 | int r, rw = bio_rw(bio); | |
1629 | struct mirror *m; | |
1630 | struct mirror_set *ms = ti->private; | |
06386bbf | 1631 | struct dm_raid1_read_record *read_record = NULL; |
1da177e4 LT |
1632 | |
1633 | if (rw == WRITE) { | |
06386bbf JB |
1634 | /* Save region for mirror_end_io() handler */ |
1635 | map_context->ll = bio_to_region(&ms->rh, bio); | |
1da177e4 | 1636 | queue_bio(ms, bio, rw); |
d2a7ad29 | 1637 | return DM_MAPIO_SUBMITTED; |
1da177e4 LT |
1638 | } |
1639 | ||
1640 | r = ms->rh.log->type->in_sync(ms->rh.log, | |
1641 | bio_to_region(&ms->rh, bio), 0); | |
1642 | if (r < 0 && r != -EWOULDBLOCK) | |
1643 | return r; | |
1644 | ||
1da177e4 | 1645 | /* |
06386bbf | 1646 | * If region is not in-sync queue the bio. |
1da177e4 | 1647 | */ |
06386bbf JB |
1648 | if (!r || (r == -EWOULDBLOCK)) { |
1649 | if (rw == READA) | |
1650 | return -EWOULDBLOCK; | |
1da177e4 | 1651 | |
1da177e4 | 1652 | queue_bio(ms, bio, rw); |
d2a7ad29 | 1653 | return DM_MAPIO_SUBMITTED; |
1da177e4 LT |
1654 | } |
1655 | ||
06386bbf JB |
1656 | /* |
1657 | * The region is in-sync and we can perform reads directly. | |
1658 | * Store enough information so we can retry if it fails. | |
1659 | */ | |
1da177e4 | 1660 | m = choose_mirror(ms, bio->bi_sector); |
06386bbf | 1661 | if (unlikely(!m)) |
1da177e4 LT |
1662 | return -EIO; |
1663 | ||
06386bbf JB |
1664 | read_record = mempool_alloc(ms->read_record_pool, GFP_NOIO); |
1665 | if (likely(read_record)) { | |
1666 | dm_bio_record(&read_record->details, bio); | |
1667 | map_context->ptr = read_record; | |
1668 | read_record->m = m; | |
1669 | } | |
1670 | ||
1671 | map_bio(m, bio); | |
1672 | ||
d2a7ad29 | 1673 | return DM_MAPIO_REMAPPED; |
1da177e4 LT |
1674 | } |
1675 | ||
1676 | static int mirror_end_io(struct dm_target *ti, struct bio *bio, | |
1677 | int error, union map_info *map_context) | |
1678 | { | |
1679 | int rw = bio_rw(bio); | |
1680 | struct mirror_set *ms = (struct mirror_set *) ti->private; | |
06386bbf JB |
1681 | struct mirror *m = NULL; |
1682 | struct dm_bio_details *bd = NULL; | |
1683 | struct dm_raid1_read_record *read_record = map_context->ptr; | |
1da177e4 LT |
1684 | |
1685 | /* | |
1686 | * We need to dec pending if this was a write. | |
1687 | */ | |
06386bbf JB |
1688 | if (rw == WRITE) { |
1689 | rh_dec(&ms->rh, map_context->ll); | |
1690 | return error; | |
1691 | } | |
1da177e4 | 1692 | |
06386bbf JB |
1693 | if (error == -EOPNOTSUPP) |
1694 | goto out; | |
1695 | ||
1696 | if ((error == -EWOULDBLOCK) && bio_rw_ahead(bio)) | |
1697 | goto out; | |
1698 | ||
1699 | if (unlikely(error)) { | |
1700 | if (!read_record) { | |
1701 | /* | |
1702 | * There wasn't enough memory to record necessary | |
1703 | * information for a retry or there was no other | |
1704 | * mirror in-sync. | |
1705 | */ | |
e03f1a84 | 1706 | DMERR_LIMIT("Mirror read failed."); |
06386bbf JB |
1707 | return -EIO; |
1708 | } | |
e03f1a84 AB |
1709 | |
1710 | m = read_record->m; | |
1711 | ||
06386bbf JB |
1712 | DMERR("Mirror read failed from %s. Trying alternative device.", |
1713 | m->dev->name); | |
1714 | ||
06386bbf JB |
1715 | fail_mirror(m, DM_RAID1_READ_ERROR); |
1716 | ||
1717 | /* | |
1718 | * A failed read is requeued for another attempt using an intact | |
1719 | * mirror. | |
1720 | */ | |
1721 | if (default_ok(m) || mirror_available(ms, bio)) { | |
1722 | bd = &read_record->details; | |
1723 | ||
1724 | dm_bio_restore(bd, bio); | |
1725 | mempool_free(read_record, ms->read_record_pool); | |
1726 | map_context->ptr = NULL; | |
1727 | queue_bio(ms, bio, rw); | |
1728 | return 1; | |
1729 | } | |
1730 | DMERR("All replicated volumes dead, failing I/O"); | |
1731 | } | |
1732 | ||
1733 | out: | |
1734 | if (read_record) { | |
1735 | mempool_free(read_record, ms->read_record_pool); | |
1736 | map_context->ptr = NULL; | |
1737 | } | |
1738 | ||
1739 | return error; | |
1da177e4 LT |
1740 | } |
1741 | ||
b80aa7a0 | 1742 | static void mirror_presuspend(struct dm_target *ti) |
1da177e4 LT |
1743 | { |
1744 | struct mirror_set *ms = (struct mirror_set *) ti->private; | |
416cd17b | 1745 | struct dm_dirty_log *log = ms->rh.log; |
1da177e4 | 1746 | |
b80aa7a0 JB |
1747 | atomic_set(&ms->suspend, 1); |
1748 | ||
1749 | /* | |
1750 | * We must finish up all the work that we've | |
1751 | * generated (i.e. recovery work). | |
1752 | */ | |
1da177e4 | 1753 | rh_stop_recovery(&ms->rh); |
33184048 | 1754 | |
33184048 JB |
1755 | wait_event(_kmirrord_recovery_stopped, |
1756 | !atomic_read(&ms->rh.recovery_in_flight)); | |
1757 | ||
b80aa7a0 JB |
1758 | if (log->type->presuspend && log->type->presuspend(log)) |
1759 | /* FIXME: need better error handling */ | |
1760 | DMWARN("log presuspend failed"); | |
1761 | ||
1762 | /* | |
1763 | * Now that recovery is complete/stopped and the | |
1764 | * delayed bios are queued, we need to wait for | |
1765 | * the worker thread to complete. This way, | |
1766 | * we know that all of our I/O has been pushed. | |
1767 | */ | |
1768 | flush_workqueue(ms->kmirrord_wq); | |
1769 | } | |
1770 | ||
1771 | static void mirror_postsuspend(struct dm_target *ti) | |
1772 | { | |
1773 | struct mirror_set *ms = ti->private; | |
416cd17b | 1774 | struct dm_dirty_log *log = ms->rh.log; |
b80aa7a0 | 1775 | |
6b3df0d7 | 1776 | if (log->type->postsuspend && log->type->postsuspend(log)) |
1da177e4 | 1777 | /* FIXME: need better error handling */ |
b80aa7a0 | 1778 | DMWARN("log postsuspend failed"); |
1da177e4 LT |
1779 | } |
1780 | ||
1781 | static void mirror_resume(struct dm_target *ti) | |
1782 | { | |
b80aa7a0 | 1783 | struct mirror_set *ms = ti->private; |
416cd17b | 1784 | struct dm_dirty_log *log = ms->rh.log; |
b80aa7a0 JB |
1785 | |
1786 | atomic_set(&ms->suspend, 0); | |
1da177e4 LT |
1787 | if (log->type->resume && log->type->resume(log)) |
1788 | /* FIXME: need better error handling */ | |
1789 | DMWARN("log resume failed"); | |
1790 | rh_start_recovery(&ms->rh); | |
1791 | } | |
1792 | ||
af195ac8 JB |
1793 | /* |
1794 | * device_status_char | |
1795 | * @m: mirror device/leg we want the status of | |
1796 | * | |
1797 | * We return one character representing the most severe error | |
1798 | * we have encountered. | |
1799 | * A => Alive - No failures | |
1800 | * D => Dead - A write failure occurred leaving mirror out-of-sync | |
1801 | * S => Sync - A sychronization failure occurred, mirror out-of-sync | |
1802 | * R => Read - A read failure occurred, mirror data unaffected | |
1803 | * | |
1804 | * Returns: <char> | |
1805 | */ | |
1806 | static char device_status_char(struct mirror *m) | |
1807 | { | |
1808 | if (!atomic_read(&(m->error_count))) | |
1809 | return 'A'; | |
1810 | ||
1811 | return (test_bit(DM_RAID1_WRITE_ERROR, &(m->error_type))) ? 'D' : | |
1812 | (test_bit(DM_RAID1_SYNC_ERROR, &(m->error_type))) ? 'S' : | |
1813 | (test_bit(DM_RAID1_READ_ERROR, &(m->error_type))) ? 'R' : 'U'; | |
1814 | } | |
1815 | ||
1816 | ||
1da177e4 LT |
1817 | static int mirror_status(struct dm_target *ti, status_type_t type, |
1818 | char *result, unsigned int maxlen) | |
1819 | { | |
315dcc22 | 1820 | unsigned int m, sz = 0; |
1da177e4 | 1821 | struct mirror_set *ms = (struct mirror_set *) ti->private; |
416cd17b | 1822 | struct dm_dirty_log *log = ms->rh.log; |
af195ac8 | 1823 | char buffer[ms->nr_mirrors + 1]; |
1da177e4 | 1824 | |
1da177e4 LT |
1825 | switch (type) { |
1826 | case STATUSTYPE_INFO: | |
1827 | DMEMIT("%d ", ms->nr_mirrors); | |
af195ac8 | 1828 | for (m = 0; m < ms->nr_mirrors; m++) { |
1da177e4 | 1829 | DMEMIT("%s ", ms->mirror[m].dev->name); |
af195ac8 JB |
1830 | buffer[m] = device_status_char(&(ms->mirror[m])); |
1831 | } | |
1832 | buffer[m] = '\0'; | |
1da177e4 | 1833 | |
af195ac8 JB |
1834 | DMEMIT("%llu/%llu 1 %s ", |
1835 | (unsigned long long)log->type->get_sync_count(ms->rh.log), | |
1836 | (unsigned long long)ms->nr_regions, buffer); | |
315dcc22 | 1837 | |
af195ac8 | 1838 | sz += log->type->status(ms->rh.log, type, result+sz, maxlen-sz); |
315dcc22 | 1839 | |
1da177e4 LT |
1840 | break; |
1841 | ||
1842 | case STATUSTYPE_TABLE: | |
af195ac8 | 1843 | sz = log->type->status(ms->rh.log, type, result, maxlen); |
315dcc22 | 1844 | |
e52b8f6d | 1845 | DMEMIT("%d", ms->nr_mirrors); |
1da177e4 | 1846 | for (m = 0; m < ms->nr_mirrors; m++) |
e52b8f6d | 1847 | DMEMIT(" %s %llu", ms->mirror[m].dev->name, |
b80aa7a0 | 1848 | (unsigned long long)ms->mirror[m].offset); |
a8e6afa2 JB |
1849 | |
1850 | if (ms->features & DM_RAID1_HANDLE_ERRORS) | |
1851 | DMEMIT(" 1 handle_errors"); | |
1da177e4 LT |
1852 | } |
1853 | ||
1854 | return 0; | |
1855 | } | |
1856 | ||
1857 | static struct target_type mirror_target = { | |
1858 | .name = "mirror", | |
af195ac8 | 1859 | .version = {1, 0, 20}, |
1da177e4 LT |
1860 | .module = THIS_MODULE, |
1861 | .ctr = mirror_ctr, | |
1862 | .dtr = mirror_dtr, | |
1863 | .map = mirror_map, | |
1864 | .end_io = mirror_end_io, | |
b80aa7a0 | 1865 | .presuspend = mirror_presuspend, |
1da177e4 LT |
1866 | .postsuspend = mirror_postsuspend, |
1867 | .resume = mirror_resume, | |
1868 | .status = mirror_status, | |
1869 | }; | |
1870 | ||
1871 | static int __init dm_mirror_init(void) | |
1872 | { | |
1873 | int r; | |
1874 | ||
1da177e4 | 1875 | r = dm_register_target(&mirror_target); |
769aef30 | 1876 | if (r < 0) |
0cd33124 | 1877 | DMERR("Failed to register mirror target"); |
1da177e4 LT |
1878 | |
1879 | return r; | |
1880 | } | |
1881 | ||
1882 | static void __exit dm_mirror_exit(void) | |
1883 | { | |
1884 | int r; | |
1885 | ||
1886 | r = dm_unregister_target(&mirror_target); | |
1887 | if (r < 0) | |
0cd33124 | 1888 | DMERR("unregister failed %d", r); |
1da177e4 LT |
1889 | } |
1890 | ||
1891 | /* Module hooks */ | |
1892 | module_init(dm_mirror_init); | |
1893 | module_exit(dm_mirror_exit); | |
1894 | ||
1895 | MODULE_DESCRIPTION(DM_NAME " mirror target"); | |
1896 | MODULE_AUTHOR("Joe Thornber"); | |
1897 | MODULE_LICENSE("GPL"); |