]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Copyright (C) 2003 Sistina Software Limited. | |
3 | * | |
4 | * This file is released under the GPL. | |
5 | */ | |
6 | ||
7 | #include "dm.h" | |
8 | #include "dm-bio-list.h" | |
9 | #include "dm-io.h" | |
10 | #include "dm-log.h" | |
11 | #include "kcopyd.h" | |
12 | ||
13 | #include <linux/ctype.h> | |
14 | #include <linux/init.h> | |
15 | #include <linux/mempool.h> | |
16 | #include <linux/module.h> | |
17 | #include <linux/pagemap.h> | |
18 | #include <linux/slab.h> | |
19 | #include <linux/time.h> | |
20 | #include <linux/vmalloc.h> | |
21 | #include <linux/workqueue.h> | |
6f3c3f0a | 22 | #include <linux/log2.h> |
1da177e4 | 23 | |
72d94861 | 24 | #define DM_MSG_PREFIX "raid1" |
88be163a | 25 | #define DM_IO_PAGES 64 |
72d94861 | 26 | |
a8e6afa2 | 27 | #define DM_RAID1_HANDLE_ERRORS 0x01 |
f44db678 | 28 | #define errors_handled(p) ((p)->features & DM_RAID1_HANDLE_ERRORS) |
a8e6afa2 | 29 | |
33184048 | 30 | static DECLARE_WAIT_QUEUE_HEAD(_kmirrord_recovery_stopped); |
1da177e4 | 31 | |
1da177e4 LT |
32 | /*----------------------------------------------------------------- |
33 | * Region hash | |
34 | * | |
35 | * The mirror splits itself up into discrete regions. Each | |
36 | * region can be in one of three states: clean, dirty, | |
37 | * nosync. There is no need to put clean regions in the hash. | |
38 | * | |
39 | * In addition to being present in the hash table a region _may_ | |
40 | * be present on one of three lists. | |
41 | * | |
42 | * clean_regions: Regions on this list have no io pending to | |
43 | * them, they are in sync, we are no longer interested in them, | |
44 | * they are dull. rh_update_states() will remove them from the | |
45 | * hash table. | |
46 | * | |
47 | * quiesced_regions: These regions have been spun down, ready | |
48 | * for recovery. rh_recovery_start() will remove regions from | |
49 | * this list and hand them to kmirrord, which will schedule the | |
50 | * recovery io with kcopyd. | |
51 | * | |
52 | * recovered_regions: Regions that kcopyd has successfully | |
53 | * recovered. rh_update_states() will now schedule any delayed | |
54 | * io, up the recovery_count, and remove the region from the | |
55 | * hash. | |
56 | * | |
57 | * There are 2 locks: | |
58 | * A rw spin lock 'hash_lock' protects just the hash table, | |
59 | * this is never held in write mode from interrupt context, | |
60 | * which I believe means that we only have to disable irqs when | |
61 | * doing a write lock. | |
62 | * | |
63 | * An ordinary spin lock 'region_lock' that protects the three | |
64 | * lists in the region_hash, with the 'state', 'list' and | |
65 | * 'bhs_delayed' fields of the regions. This is used from irq | |
66 | * context, so all other uses will have to suspend local irqs. | |
67 | *---------------------------------------------------------------*/ | |
68 | struct mirror_set; | |
69 | struct region_hash { | |
70 | struct mirror_set *ms; | |
71 | uint32_t region_size; | |
72 | unsigned region_shift; | |
73 | ||
74 | /* holds persistent region state */ | |
75 | struct dirty_log *log; | |
76 | ||
77 | /* hash table */ | |
78 | rwlock_t hash_lock; | |
79 | mempool_t *region_pool; | |
80 | unsigned int mask; | |
81 | unsigned int nr_buckets; | |
82 | struct list_head *buckets; | |
83 | ||
84 | spinlock_t region_lock; | |
33184048 | 85 | atomic_t recovery_in_flight; |
1da177e4 LT |
86 | struct semaphore recovery_count; |
87 | struct list_head clean_regions; | |
88 | struct list_head quiesced_regions; | |
89 | struct list_head recovered_regions; | |
f44db678 | 90 | struct list_head failed_recovered_regions; |
1da177e4 LT |
91 | }; |
92 | ||
93 | enum { | |
94 | RH_CLEAN, | |
95 | RH_DIRTY, | |
96 | RH_NOSYNC, | |
97 | RH_RECOVERING | |
98 | }; | |
99 | ||
100 | struct region { | |
101 | struct region_hash *rh; /* FIXME: can we get rid of this ? */ | |
102 | region_t key; | |
103 | int state; | |
104 | ||
105 | struct list_head hash_list; | |
106 | struct list_head list; | |
107 | ||
108 | atomic_t pending; | |
109 | struct bio_list delayed_bios; | |
110 | }; | |
111 | ||
e4c8b3ba NB |
112 | |
113 | /*----------------------------------------------------------------- | |
114 | * Mirror set structures. | |
115 | *---------------------------------------------------------------*/ | |
116 | struct mirror { | |
aa5617c5 | 117 | struct mirror_set *ms; |
e4c8b3ba NB |
118 | atomic_t error_count; |
119 | struct dm_dev *dev; | |
120 | sector_t offset; | |
121 | }; | |
122 | ||
123 | struct mirror_set { | |
124 | struct dm_target *ti; | |
125 | struct list_head list; | |
126 | struct region_hash rh; | |
127 | struct kcopyd_client *kcopyd_client; | |
a8e6afa2 | 128 | uint64_t features; |
e4c8b3ba NB |
129 | |
130 | spinlock_t lock; /* protects the next two lists */ | |
131 | struct bio_list reads; | |
132 | struct bio_list writes; | |
133 | ||
88be163a MB |
134 | struct dm_io_client *io_client; |
135 | ||
e4c8b3ba NB |
136 | /* recovery */ |
137 | region_t nr_regions; | |
138 | int in_sync; | |
fc1ff958 | 139 | int log_failure; |
e4c8b3ba NB |
140 | |
141 | struct mirror *default_mirror; /* Default mirror */ | |
142 | ||
6ad36fe2 HS |
143 | struct workqueue_struct *kmirrord_wq; |
144 | struct work_struct kmirrord_work; | |
145 | ||
e4c8b3ba NB |
146 | unsigned int nr_mirrors; |
147 | struct mirror mirror[0]; | |
148 | }; | |
149 | ||
1da177e4 LT |
150 | /* |
151 | * Conversion fns | |
152 | */ | |
153 | static inline region_t bio_to_region(struct region_hash *rh, struct bio *bio) | |
154 | { | |
e4c8b3ba | 155 | return (bio->bi_sector - rh->ms->ti->begin) >> rh->region_shift; |
1da177e4 LT |
156 | } |
157 | ||
158 | static inline sector_t region_to_sector(struct region_hash *rh, region_t region) | |
159 | { | |
160 | return region << rh->region_shift; | |
161 | } | |
162 | ||
6ad36fe2 HS |
163 | static void wake(struct mirror_set *ms) |
164 | { | |
165 | queue_work(ms->kmirrord_wq, &ms->kmirrord_work); | |
166 | } | |
167 | ||
1da177e4 LT |
168 | /* FIXME move this */ |
169 | static void queue_bio(struct mirror_set *ms, struct bio *bio, int rw); | |
170 | ||
1da177e4 LT |
171 | #define MIN_REGIONS 64 |
172 | #define MAX_RECOVERY 1 | |
173 | static int rh_init(struct region_hash *rh, struct mirror_set *ms, | |
174 | struct dirty_log *log, uint32_t region_size, | |
175 | region_t nr_regions) | |
176 | { | |
177 | unsigned int nr_buckets, max_buckets; | |
178 | size_t i; | |
179 | ||
180 | /* | |
181 | * Calculate a suitable number of buckets for our hash | |
182 | * table. | |
183 | */ | |
184 | max_buckets = nr_regions >> 6; | |
185 | for (nr_buckets = 128u; nr_buckets < max_buckets; nr_buckets <<= 1) | |
186 | ; | |
187 | nr_buckets >>= 1; | |
188 | ||
189 | rh->ms = ms; | |
190 | rh->log = log; | |
191 | rh->region_size = region_size; | |
192 | rh->region_shift = ffs(region_size) - 1; | |
193 | rwlock_init(&rh->hash_lock); | |
194 | rh->mask = nr_buckets - 1; | |
195 | rh->nr_buckets = nr_buckets; | |
196 | ||
197 | rh->buckets = vmalloc(nr_buckets * sizeof(*rh->buckets)); | |
198 | if (!rh->buckets) { | |
199 | DMERR("unable to allocate region hash memory"); | |
200 | return -ENOMEM; | |
201 | } | |
202 | ||
203 | for (i = 0; i < nr_buckets; i++) | |
204 | INIT_LIST_HEAD(rh->buckets + i); | |
205 | ||
206 | spin_lock_init(&rh->region_lock); | |
207 | sema_init(&rh->recovery_count, 0); | |
33184048 | 208 | atomic_set(&rh->recovery_in_flight, 0); |
1da177e4 LT |
209 | INIT_LIST_HEAD(&rh->clean_regions); |
210 | INIT_LIST_HEAD(&rh->quiesced_regions); | |
211 | INIT_LIST_HEAD(&rh->recovered_regions); | |
f44db678 | 212 | INIT_LIST_HEAD(&rh->failed_recovered_regions); |
1da177e4 | 213 | |
0eaae62a MD |
214 | rh->region_pool = mempool_create_kmalloc_pool(MIN_REGIONS, |
215 | sizeof(struct region)); | |
1da177e4 LT |
216 | if (!rh->region_pool) { |
217 | vfree(rh->buckets); | |
218 | rh->buckets = NULL; | |
219 | return -ENOMEM; | |
220 | } | |
221 | ||
222 | return 0; | |
223 | } | |
224 | ||
225 | static void rh_exit(struct region_hash *rh) | |
226 | { | |
227 | unsigned int h; | |
228 | struct region *reg, *nreg; | |
229 | ||
230 | BUG_ON(!list_empty(&rh->quiesced_regions)); | |
231 | for (h = 0; h < rh->nr_buckets; h++) { | |
232 | list_for_each_entry_safe(reg, nreg, rh->buckets + h, hash_list) { | |
233 | BUG_ON(atomic_read(®->pending)); | |
234 | mempool_free(reg, rh->region_pool); | |
235 | } | |
236 | } | |
237 | ||
238 | if (rh->log) | |
239 | dm_destroy_dirty_log(rh->log); | |
240 | if (rh->region_pool) | |
241 | mempool_destroy(rh->region_pool); | |
242 | vfree(rh->buckets); | |
243 | } | |
244 | ||
245 | #define RH_HASH_MULT 2654435387U | |
246 | ||
247 | static inline unsigned int rh_hash(struct region_hash *rh, region_t region) | |
248 | { | |
249 | return (unsigned int) ((region * RH_HASH_MULT) >> 12) & rh->mask; | |
250 | } | |
251 | ||
252 | static struct region *__rh_lookup(struct region_hash *rh, region_t region) | |
253 | { | |
254 | struct region *reg; | |
255 | ||
256 | list_for_each_entry (reg, rh->buckets + rh_hash(rh, region), hash_list) | |
257 | if (reg->key == region) | |
258 | return reg; | |
259 | ||
260 | return NULL; | |
261 | } | |
262 | ||
263 | static void __rh_insert(struct region_hash *rh, struct region *reg) | |
264 | { | |
265 | unsigned int h = rh_hash(rh, reg->key); | |
266 | list_add(®->hash_list, rh->buckets + h); | |
267 | } | |
268 | ||
269 | static struct region *__rh_alloc(struct region_hash *rh, region_t region) | |
270 | { | |
271 | struct region *reg, *nreg; | |
272 | ||
273 | read_unlock(&rh->hash_lock); | |
c06aad85 DK |
274 | nreg = mempool_alloc(rh->region_pool, GFP_ATOMIC); |
275 | if (unlikely(!nreg)) | |
276 | nreg = kmalloc(sizeof(struct region), GFP_NOIO); | |
1da177e4 LT |
277 | nreg->state = rh->log->type->in_sync(rh->log, region, 1) ? |
278 | RH_CLEAN : RH_NOSYNC; | |
279 | nreg->rh = rh; | |
280 | nreg->key = region; | |
281 | ||
282 | INIT_LIST_HEAD(&nreg->list); | |
283 | ||
284 | atomic_set(&nreg->pending, 0); | |
285 | bio_list_init(&nreg->delayed_bios); | |
286 | write_lock_irq(&rh->hash_lock); | |
287 | ||
288 | reg = __rh_lookup(rh, region); | |
289 | if (reg) | |
290 | /* we lost the race */ | |
291 | mempool_free(nreg, rh->region_pool); | |
292 | ||
293 | else { | |
294 | __rh_insert(rh, nreg); | |
295 | if (nreg->state == RH_CLEAN) { | |
296 | spin_lock(&rh->region_lock); | |
297 | list_add(&nreg->list, &rh->clean_regions); | |
298 | spin_unlock(&rh->region_lock); | |
299 | } | |
300 | reg = nreg; | |
301 | } | |
302 | write_unlock_irq(&rh->hash_lock); | |
303 | read_lock(&rh->hash_lock); | |
304 | ||
305 | return reg; | |
306 | } | |
307 | ||
308 | static inline struct region *__rh_find(struct region_hash *rh, region_t region) | |
309 | { | |
310 | struct region *reg; | |
311 | ||
312 | reg = __rh_lookup(rh, region); | |
313 | if (!reg) | |
314 | reg = __rh_alloc(rh, region); | |
315 | ||
316 | return reg; | |
317 | } | |
318 | ||
319 | static int rh_state(struct region_hash *rh, region_t region, int may_block) | |
320 | { | |
321 | int r; | |
322 | struct region *reg; | |
323 | ||
324 | read_lock(&rh->hash_lock); | |
325 | reg = __rh_lookup(rh, region); | |
326 | read_unlock(&rh->hash_lock); | |
327 | ||
328 | if (reg) | |
329 | return reg->state; | |
330 | ||
331 | /* | |
332 | * The region wasn't in the hash, so we fall back to the | |
333 | * dirty log. | |
334 | */ | |
335 | r = rh->log->type->in_sync(rh->log, region, may_block); | |
336 | ||
337 | /* | |
338 | * Any error from the dirty log (eg. -EWOULDBLOCK) gets | |
339 | * taken as a RH_NOSYNC | |
340 | */ | |
341 | return r == 1 ? RH_CLEAN : RH_NOSYNC; | |
342 | } | |
343 | ||
344 | static inline int rh_in_sync(struct region_hash *rh, | |
345 | region_t region, int may_block) | |
346 | { | |
347 | int state = rh_state(rh, region, may_block); | |
348 | return state == RH_CLEAN || state == RH_DIRTY; | |
349 | } | |
350 | ||
351 | static void dispatch_bios(struct mirror_set *ms, struct bio_list *bio_list) | |
352 | { | |
353 | struct bio *bio; | |
354 | ||
355 | while ((bio = bio_list_pop(bio_list))) { | |
356 | queue_bio(ms, bio, WRITE); | |
357 | } | |
358 | } | |
359 | ||
f3ee6b2f JB |
360 | static void complete_resync_work(struct region *reg, int success) |
361 | { | |
362 | struct region_hash *rh = reg->rh; | |
363 | ||
364 | rh->log->type->set_region_sync(rh->log, reg->key, success); | |
365 | dispatch_bios(rh->ms, ®->delayed_bios); | |
366 | if (atomic_dec_and_test(&rh->recovery_in_flight)) | |
367 | wake_up_all(&_kmirrord_recovery_stopped); | |
368 | up(&rh->recovery_count); | |
369 | } | |
370 | ||
1da177e4 LT |
371 | static void rh_update_states(struct region_hash *rh) |
372 | { | |
373 | struct region *reg, *next; | |
374 | ||
375 | LIST_HEAD(clean); | |
376 | LIST_HEAD(recovered); | |
f44db678 | 377 | LIST_HEAD(failed_recovered); |
1da177e4 LT |
378 | |
379 | /* | |
380 | * Quickly grab the lists. | |
381 | */ | |
382 | write_lock_irq(&rh->hash_lock); | |
383 | spin_lock(&rh->region_lock); | |
384 | if (!list_empty(&rh->clean_regions)) { | |
385 | list_splice(&rh->clean_regions, &clean); | |
386 | INIT_LIST_HEAD(&rh->clean_regions); | |
387 | ||
943317ef | 388 | list_for_each_entry(reg, &clean, list) |
1da177e4 | 389 | list_del(®->hash_list); |
1da177e4 LT |
390 | } |
391 | ||
392 | if (!list_empty(&rh->recovered_regions)) { | |
393 | list_splice(&rh->recovered_regions, &recovered); | |
394 | INIT_LIST_HEAD(&rh->recovered_regions); | |
395 | ||
396 | list_for_each_entry (reg, &recovered, list) | |
397 | list_del(®->hash_list); | |
398 | } | |
f44db678 JB |
399 | |
400 | if (!list_empty(&rh->failed_recovered_regions)) { | |
401 | list_splice(&rh->failed_recovered_regions, &failed_recovered); | |
402 | INIT_LIST_HEAD(&rh->failed_recovered_regions); | |
403 | ||
404 | list_for_each_entry(reg, &failed_recovered, list) | |
405 | list_del(®->hash_list); | |
406 | } | |
407 | ||
1da177e4 LT |
408 | spin_unlock(&rh->region_lock); |
409 | write_unlock_irq(&rh->hash_lock); | |
410 | ||
411 | /* | |
412 | * All the regions on the recovered and clean lists have | |
413 | * now been pulled out of the system, so no need to do | |
414 | * any more locking. | |
415 | */ | |
416 | list_for_each_entry_safe (reg, next, &recovered, list) { | |
417 | rh->log->type->clear_region(rh->log, reg->key); | |
f3ee6b2f | 418 | complete_resync_work(reg, 1); |
1da177e4 LT |
419 | mempool_free(reg, rh->region_pool); |
420 | } | |
421 | ||
f44db678 JB |
422 | list_for_each_entry_safe(reg, next, &failed_recovered, list) { |
423 | complete_resync_work(reg, errors_handled(rh->ms) ? 0 : 1); | |
424 | mempool_free(reg, rh->region_pool); | |
425 | } | |
426 | ||
943317ef JB |
427 | list_for_each_entry_safe(reg, next, &clean, list) { |
428 | rh->log->type->clear_region(rh->log, reg->key); | |
1da177e4 | 429 | mempool_free(reg, rh->region_pool); |
943317ef JB |
430 | } |
431 | ||
432 | rh->log->type->flush(rh->log); | |
1da177e4 LT |
433 | } |
434 | ||
435 | static void rh_inc(struct region_hash *rh, region_t region) | |
436 | { | |
437 | struct region *reg; | |
438 | ||
439 | read_lock(&rh->hash_lock); | |
440 | reg = __rh_find(rh, region); | |
844e8d90 | 441 | |
7692c5dd | 442 | spin_lock_irq(&rh->region_lock); |
844e8d90 JN |
443 | atomic_inc(®->pending); |
444 | ||
1da177e4 | 445 | if (reg->state == RH_CLEAN) { |
1da177e4 LT |
446 | reg->state = RH_DIRTY; |
447 | list_del_init(®->list); /* take off the clean list */ | |
7692c5dd JB |
448 | spin_unlock_irq(&rh->region_lock); |
449 | ||
450 | rh->log->type->mark_region(rh->log, reg->key); | |
451 | } else | |
452 | spin_unlock_irq(&rh->region_lock); | |
453 | ||
1da177e4 | 454 | |
1da177e4 LT |
455 | read_unlock(&rh->hash_lock); |
456 | } | |
457 | ||
458 | static void rh_inc_pending(struct region_hash *rh, struct bio_list *bios) | |
459 | { | |
460 | struct bio *bio; | |
461 | ||
462 | for (bio = bios->head; bio; bio = bio->bi_next) | |
463 | rh_inc(rh, bio_to_region(rh, bio)); | |
464 | } | |
465 | ||
466 | static void rh_dec(struct region_hash *rh, region_t region) | |
467 | { | |
468 | unsigned long flags; | |
469 | struct region *reg; | |
470 | int should_wake = 0; | |
471 | ||
472 | read_lock(&rh->hash_lock); | |
473 | reg = __rh_lookup(rh, region); | |
474 | read_unlock(&rh->hash_lock); | |
475 | ||
7692c5dd | 476 | spin_lock_irqsave(&rh->region_lock, flags); |
1da177e4 | 477 | if (atomic_dec_and_test(®->pending)) { |
930d332a JN |
478 | /* |
479 | * There is no pending I/O for this region. | |
480 | * We can move the region to corresponding list for next action. | |
481 | * At this point, the region is not yet connected to any list. | |
482 | * | |
483 | * If the state is RH_NOSYNC, the region should be kept off | |
484 | * from clean list. | |
485 | * The hash entry for RH_NOSYNC will remain in memory | |
486 | * until the region is recovered or the map is reloaded. | |
487 | */ | |
488 | ||
489 | /* do nothing for RH_NOSYNC */ | |
1da177e4 LT |
490 | if (reg->state == RH_RECOVERING) { |
491 | list_add_tail(®->list, &rh->quiesced_regions); | |
930d332a | 492 | } else if (reg->state == RH_DIRTY) { |
1da177e4 LT |
493 | reg->state = RH_CLEAN; |
494 | list_add(®->list, &rh->clean_regions); | |
495 | } | |
1da177e4 LT |
496 | should_wake = 1; |
497 | } | |
7692c5dd | 498 | spin_unlock_irqrestore(&rh->region_lock, flags); |
1da177e4 LT |
499 | |
500 | if (should_wake) | |
6ad36fe2 | 501 | wake(rh->ms); |
1da177e4 LT |
502 | } |
503 | ||
504 | /* | |
505 | * Starts quiescing a region in preparation for recovery. | |
506 | */ | |
507 | static int __rh_recovery_prepare(struct region_hash *rh) | |
508 | { | |
509 | int r; | |
510 | struct region *reg; | |
511 | region_t region; | |
512 | ||
513 | /* | |
514 | * Ask the dirty log what's next. | |
515 | */ | |
516 | r = rh->log->type->get_resync_work(rh->log, ®ion); | |
517 | if (r <= 0) | |
518 | return r; | |
519 | ||
520 | /* | |
521 | * Get this region, and start it quiescing by setting the | |
522 | * recovering flag. | |
523 | */ | |
524 | read_lock(&rh->hash_lock); | |
525 | reg = __rh_find(rh, region); | |
526 | read_unlock(&rh->hash_lock); | |
527 | ||
528 | spin_lock_irq(&rh->region_lock); | |
529 | reg->state = RH_RECOVERING; | |
530 | ||
531 | /* Already quiesced ? */ | |
532 | if (atomic_read(®->pending)) | |
533 | list_del_init(®->list); | |
179e0917 AM |
534 | else |
535 | list_move(®->list, &rh->quiesced_regions); | |
1da177e4 | 536 | |
1da177e4 LT |
537 | spin_unlock_irq(&rh->region_lock); |
538 | ||
539 | return 1; | |
540 | } | |
541 | ||
542 | static void rh_recovery_prepare(struct region_hash *rh) | |
543 | { | |
33184048 JB |
544 | /* Extra reference to avoid race with rh_stop_recovery */ |
545 | atomic_inc(&rh->recovery_in_flight); | |
546 | ||
547 | while (!down_trylock(&rh->recovery_count)) { | |
548 | atomic_inc(&rh->recovery_in_flight); | |
1da177e4 | 549 | if (__rh_recovery_prepare(rh) <= 0) { |
33184048 | 550 | atomic_dec(&rh->recovery_in_flight); |
1da177e4 LT |
551 | up(&rh->recovery_count); |
552 | break; | |
553 | } | |
33184048 JB |
554 | } |
555 | ||
556 | /* Drop the extra reference */ | |
557 | if (atomic_dec_and_test(&rh->recovery_in_flight)) | |
558 | wake_up_all(&_kmirrord_recovery_stopped); | |
1da177e4 LT |
559 | } |
560 | ||
561 | /* | |
562 | * Returns any quiesced regions. | |
563 | */ | |
564 | static struct region *rh_recovery_start(struct region_hash *rh) | |
565 | { | |
566 | struct region *reg = NULL; | |
567 | ||
568 | spin_lock_irq(&rh->region_lock); | |
569 | if (!list_empty(&rh->quiesced_regions)) { | |
570 | reg = list_entry(rh->quiesced_regions.next, | |
571 | struct region, list); | |
572 | list_del_init(®->list); /* remove from the quiesced list */ | |
573 | } | |
574 | spin_unlock_irq(&rh->region_lock); | |
575 | ||
576 | return reg; | |
577 | } | |
578 | ||
1da177e4 LT |
579 | static void rh_recovery_end(struct region *reg, int success) |
580 | { | |
581 | struct region_hash *rh = reg->rh; | |
582 | ||
583 | spin_lock_irq(&rh->region_lock); | |
f44db678 JB |
584 | if (success) |
585 | list_add(®->list, ®->rh->recovered_regions); | |
586 | else { | |
587 | reg->state = RH_NOSYNC; | |
588 | list_add(®->list, ®->rh->failed_recovered_regions); | |
589 | } | |
1da177e4 LT |
590 | spin_unlock_irq(&rh->region_lock); |
591 | ||
6ad36fe2 | 592 | wake(rh->ms); |
1da177e4 LT |
593 | } |
594 | ||
fc1ff958 | 595 | static int rh_flush(struct region_hash *rh) |
1da177e4 | 596 | { |
fc1ff958 | 597 | return rh->log->type->flush(rh->log); |
1da177e4 LT |
598 | } |
599 | ||
600 | static void rh_delay(struct region_hash *rh, struct bio *bio) | |
601 | { | |
602 | struct region *reg; | |
603 | ||
604 | read_lock(&rh->hash_lock); | |
605 | reg = __rh_find(rh, bio_to_region(rh, bio)); | |
606 | bio_list_add(®->delayed_bios, bio); | |
607 | read_unlock(&rh->hash_lock); | |
608 | } | |
609 | ||
610 | static void rh_stop_recovery(struct region_hash *rh) | |
611 | { | |
612 | int i; | |
613 | ||
614 | /* wait for any recovering regions */ | |
615 | for (i = 0; i < MAX_RECOVERY; i++) | |
616 | down(&rh->recovery_count); | |
617 | } | |
618 | ||
619 | static void rh_start_recovery(struct region_hash *rh) | |
620 | { | |
621 | int i; | |
622 | ||
623 | for (i = 0; i < MAX_RECOVERY; i++) | |
624 | up(&rh->recovery_count); | |
625 | ||
6ad36fe2 | 626 | wake(rh->ms); |
1da177e4 LT |
627 | } |
628 | ||
1da177e4 LT |
629 | /* |
630 | * Every mirror should look like this one. | |
631 | */ | |
632 | #define DEFAULT_MIRROR 0 | |
633 | ||
634 | /* | |
635 | * This is yucky. We squirrel the mirror_set struct away inside | |
636 | * bi_next for write buffers. This is safe since the bh | |
637 | * doesn't get submitted to the lower levels of block layer. | |
638 | */ | |
639 | static struct mirror_set *bio_get_ms(struct bio *bio) | |
640 | { | |
641 | return (struct mirror_set *) bio->bi_next; | |
642 | } | |
643 | ||
644 | static void bio_set_ms(struct bio *bio, struct mirror_set *ms) | |
645 | { | |
646 | bio->bi_next = (struct bio *) ms; | |
647 | } | |
648 | ||
649 | /*----------------------------------------------------------------- | |
650 | * Recovery. | |
651 | * | |
652 | * When a mirror is first activated we may find that some regions | |
653 | * are in the no-sync state. We have to recover these by | |
654 | * recopying from the default mirror to all the others. | |
655 | *---------------------------------------------------------------*/ | |
656 | static void recovery_complete(int read_err, unsigned int write_err, | |
657 | void *context) | |
658 | { | |
659 | struct region *reg = (struct region *) context; | |
660 | ||
f44db678 JB |
661 | if (read_err) |
662 | /* Read error means the failure of default mirror. */ | |
663 | DMERR_LIMIT("Unable to read primary mirror during recovery"); | |
664 | ||
665 | if (write_err) | |
666 | DMERR_LIMIT("Write error during recovery (error = 0x%x)", | |
667 | write_err); | |
668 | ||
ce503f59 | 669 | rh_recovery_end(reg, !(read_err || write_err)); |
1da177e4 LT |
670 | } |
671 | ||
672 | static int recover(struct mirror_set *ms, struct region *reg) | |
673 | { | |
674 | int r; | |
675 | unsigned int i; | |
676 | struct io_region from, to[KCOPYD_MAX_REGIONS], *dest; | |
677 | struct mirror *m; | |
678 | unsigned long flags = 0; | |
679 | ||
680 | /* fill in the source */ | |
a1a19080 | 681 | m = ms->default_mirror; |
1da177e4 LT |
682 | from.bdev = m->dev->bdev; |
683 | from.sector = m->offset + region_to_sector(reg->rh, reg->key); | |
684 | if (reg->key == (ms->nr_regions - 1)) { | |
685 | /* | |
686 | * The final region may be smaller than | |
687 | * region_size. | |
688 | */ | |
689 | from.count = ms->ti->len & (reg->rh->region_size - 1); | |
690 | if (!from.count) | |
691 | from.count = reg->rh->region_size; | |
692 | } else | |
693 | from.count = reg->rh->region_size; | |
694 | ||
695 | /* fill in the destinations */ | |
696 | for (i = 0, dest = to; i < ms->nr_mirrors; i++) { | |
a1a19080 | 697 | if (&ms->mirror[i] == ms->default_mirror) |
1da177e4 LT |
698 | continue; |
699 | ||
700 | m = ms->mirror + i; | |
701 | dest->bdev = m->dev->bdev; | |
702 | dest->sector = m->offset + region_to_sector(reg->rh, reg->key); | |
703 | dest->count = from.count; | |
704 | dest++; | |
705 | } | |
706 | ||
707 | /* hand to kcopyd */ | |
708 | set_bit(KCOPYD_IGNORE_ERROR, &flags); | |
709 | r = kcopyd_copy(ms->kcopyd_client, &from, ms->nr_mirrors - 1, to, flags, | |
710 | recovery_complete, reg); | |
711 | ||
712 | return r; | |
713 | } | |
714 | ||
715 | static void do_recovery(struct mirror_set *ms) | |
716 | { | |
717 | int r; | |
718 | struct region *reg; | |
719 | struct dirty_log *log = ms->rh.log; | |
720 | ||
721 | /* | |
722 | * Start quiescing some regions. | |
723 | */ | |
724 | rh_recovery_prepare(&ms->rh); | |
725 | ||
726 | /* | |
727 | * Copy any already quiesced regions. | |
728 | */ | |
729 | while ((reg = rh_recovery_start(&ms->rh))) { | |
730 | r = recover(ms, reg); | |
731 | if (r) | |
732 | rh_recovery_end(reg, 0); | |
733 | } | |
734 | ||
735 | /* | |
736 | * Update the in sync flag. | |
737 | */ | |
738 | if (!ms->in_sync && | |
739 | (log->type->get_sync_count(log) == ms->nr_regions)) { | |
740 | /* the sync is complete */ | |
741 | dm_table_event(ms->ti->table); | |
742 | ms->in_sync = 1; | |
743 | } | |
744 | } | |
745 | ||
746 | /*----------------------------------------------------------------- | |
747 | * Reads | |
748 | *---------------------------------------------------------------*/ | |
749 | static struct mirror *choose_mirror(struct mirror_set *ms, sector_t sector) | |
750 | { | |
751 | /* FIXME: add read balancing */ | |
a1a19080 | 752 | return ms->default_mirror; |
1da177e4 LT |
753 | } |
754 | ||
755 | /* | |
756 | * remap a buffer to a particular mirror. | |
757 | */ | |
758 | static void map_bio(struct mirror_set *ms, struct mirror *m, struct bio *bio) | |
759 | { | |
760 | bio->bi_bdev = m->dev->bdev; | |
761 | bio->bi_sector = m->offset + (bio->bi_sector - ms->ti->begin); | |
762 | } | |
763 | ||
764 | static void do_reads(struct mirror_set *ms, struct bio_list *reads) | |
765 | { | |
766 | region_t region; | |
767 | struct bio *bio; | |
768 | struct mirror *m; | |
769 | ||
770 | while ((bio = bio_list_pop(reads))) { | |
771 | region = bio_to_region(&ms->rh, bio); | |
772 | ||
773 | /* | |
774 | * We can only read balance if the region is in sync. | |
775 | */ | |
b997b82d | 776 | if (rh_in_sync(&ms->rh, region, 1)) |
1da177e4 LT |
777 | m = choose_mirror(ms, bio->bi_sector); |
778 | else | |
a1a19080 | 779 | m = ms->default_mirror; |
1da177e4 LT |
780 | |
781 | map_bio(ms, m, bio); | |
782 | generic_make_request(bio); | |
783 | } | |
784 | } | |
785 | ||
786 | /*----------------------------------------------------------------- | |
787 | * Writes. | |
788 | * | |
789 | * We do different things with the write io depending on the | |
790 | * state of the region that it's in: | |
791 | * | |
792 | * SYNC: increment pending, use kcopyd to write to *all* mirrors | |
793 | * RECOVERING: delay the io until recovery completes | |
794 | * NOSYNC: increment pending, just write to the default mirror | |
795 | *---------------------------------------------------------------*/ | |
796 | static void write_callback(unsigned long error, void *context) | |
797 | { | |
798 | unsigned int i; | |
799 | int uptodate = 1; | |
800 | struct bio *bio = (struct bio *) context; | |
801 | struct mirror_set *ms; | |
802 | ||
803 | ms = bio_get_ms(bio); | |
804 | bio_set_ms(bio, NULL); | |
805 | ||
806 | /* | |
807 | * NOTE: We don't decrement the pending count here, | |
808 | * instead it is done by the targets endio function. | |
809 | * This way we handle both writes to SYNC and NOSYNC | |
810 | * regions with the same code. | |
811 | */ | |
812 | ||
813 | if (error) { | |
814 | /* | |
815 | * only error the io if all mirrors failed. | |
816 | * FIXME: bogus | |
817 | */ | |
818 | uptodate = 0; | |
819 | for (i = 0; i < ms->nr_mirrors; i++) | |
820 | if (!test_bit(i, &error)) { | |
821 | uptodate = 1; | |
822 | break; | |
823 | } | |
824 | } | |
6712ecf8 | 825 | bio_endio(bio, 0); |
1da177e4 LT |
826 | } |
827 | ||
828 | static void do_write(struct mirror_set *ms, struct bio *bio) | |
829 | { | |
830 | unsigned int i; | |
831 | struct io_region io[KCOPYD_MAX_REGIONS+1]; | |
832 | struct mirror *m; | |
88be163a MB |
833 | struct dm_io_request io_req = { |
834 | .bi_rw = WRITE, | |
835 | .mem.type = DM_IO_BVEC, | |
836 | .mem.ptr.bvec = bio->bi_io_vec + bio->bi_idx, | |
837 | .notify.fn = write_callback, | |
838 | .notify.context = bio, | |
839 | .client = ms->io_client, | |
840 | }; | |
1da177e4 LT |
841 | |
842 | for (i = 0; i < ms->nr_mirrors; i++) { | |
843 | m = ms->mirror + i; | |
844 | ||
845 | io[i].bdev = m->dev->bdev; | |
846 | io[i].sector = m->offset + (bio->bi_sector - ms->ti->begin); | |
847 | io[i].count = bio->bi_size >> 9; | |
848 | } | |
849 | ||
850 | bio_set_ms(bio, ms); | |
88be163a MB |
851 | |
852 | (void) dm_io(&io_req, ms->nr_mirrors, io, NULL); | |
1da177e4 LT |
853 | } |
854 | ||
855 | static void do_writes(struct mirror_set *ms, struct bio_list *writes) | |
856 | { | |
857 | int state; | |
858 | struct bio *bio; | |
859 | struct bio_list sync, nosync, recover, *this_list = NULL; | |
860 | ||
861 | if (!writes->head) | |
862 | return; | |
863 | ||
864 | /* | |
865 | * Classify each write. | |
866 | */ | |
867 | bio_list_init(&sync); | |
868 | bio_list_init(&nosync); | |
869 | bio_list_init(&recover); | |
870 | ||
871 | while ((bio = bio_list_pop(writes))) { | |
872 | state = rh_state(&ms->rh, bio_to_region(&ms->rh, bio), 1); | |
873 | switch (state) { | |
874 | case RH_CLEAN: | |
875 | case RH_DIRTY: | |
876 | this_list = &sync; | |
877 | break; | |
878 | ||
879 | case RH_NOSYNC: | |
880 | this_list = &nosync; | |
881 | break; | |
882 | ||
883 | case RH_RECOVERING: | |
884 | this_list = &recover; | |
885 | break; | |
886 | } | |
887 | ||
888 | bio_list_add(this_list, bio); | |
889 | } | |
890 | ||
891 | /* | |
892 | * Increment the pending counts for any regions that will | |
893 | * be written to (writes to recover regions are going to | |
894 | * be delayed). | |
895 | */ | |
896 | rh_inc_pending(&ms->rh, &sync); | |
897 | rh_inc_pending(&ms->rh, &nosync); | |
fc1ff958 | 898 | ms->log_failure = rh_flush(&ms->rh) ? 1 : 0; |
1da177e4 LT |
899 | |
900 | /* | |
901 | * Dispatch io. | |
902 | */ | |
fc1ff958 JB |
903 | if (unlikely(ms->log_failure)) |
904 | while ((bio = bio_list_pop(&sync))) | |
6712ecf8 | 905 | bio_endio(bio, -EIO); |
fc1ff958 | 906 | else while ((bio = bio_list_pop(&sync))) |
1da177e4 LT |
907 | do_write(ms, bio); |
908 | ||
909 | while ((bio = bio_list_pop(&recover))) | |
910 | rh_delay(&ms->rh, bio); | |
911 | ||
912 | while ((bio = bio_list_pop(&nosync))) { | |
a1a19080 | 913 | map_bio(ms, ms->default_mirror, bio); |
1da177e4 LT |
914 | generic_make_request(bio); |
915 | } | |
916 | } | |
917 | ||
918 | /*----------------------------------------------------------------- | |
919 | * kmirrord | |
920 | *---------------------------------------------------------------*/ | |
6ad36fe2 | 921 | static void do_mirror(struct work_struct *work) |
1da177e4 | 922 | { |
6ad36fe2 HS |
923 | struct mirror_set *ms =container_of(work, struct mirror_set, |
924 | kmirrord_work); | |
1da177e4 LT |
925 | struct bio_list reads, writes; |
926 | ||
927 | spin_lock(&ms->lock); | |
928 | reads = ms->reads; | |
929 | writes = ms->writes; | |
930 | bio_list_init(&ms->reads); | |
931 | bio_list_init(&ms->writes); | |
932 | spin_unlock(&ms->lock); | |
933 | ||
934 | rh_update_states(&ms->rh); | |
935 | do_recovery(ms); | |
936 | do_reads(ms, &reads); | |
937 | do_writes(ms, &writes); | |
938 | } | |
939 | ||
1da177e4 LT |
940 | /*----------------------------------------------------------------- |
941 | * Target functions | |
942 | *---------------------------------------------------------------*/ | |
943 | static struct mirror_set *alloc_context(unsigned int nr_mirrors, | |
944 | uint32_t region_size, | |
945 | struct dm_target *ti, | |
946 | struct dirty_log *dl) | |
947 | { | |
948 | size_t len; | |
949 | struct mirror_set *ms = NULL; | |
950 | ||
951 | if (array_too_big(sizeof(*ms), sizeof(ms->mirror[0]), nr_mirrors)) | |
952 | return NULL; | |
953 | ||
954 | len = sizeof(*ms) + (sizeof(ms->mirror[0]) * nr_mirrors); | |
955 | ||
dd00cc48 | 956 | ms = kzalloc(len, GFP_KERNEL); |
1da177e4 | 957 | if (!ms) { |
72d94861 | 958 | ti->error = "Cannot allocate mirror context"; |
1da177e4 LT |
959 | return NULL; |
960 | } | |
961 | ||
1da177e4 LT |
962 | spin_lock_init(&ms->lock); |
963 | ||
964 | ms->ti = ti; | |
965 | ms->nr_mirrors = nr_mirrors; | |
966 | ms->nr_regions = dm_sector_div_up(ti->len, region_size); | |
967 | ms->in_sync = 0; | |
a1a19080 | 968 | ms->default_mirror = &ms->mirror[DEFAULT_MIRROR]; |
1da177e4 | 969 | |
88be163a MB |
970 | ms->io_client = dm_io_client_create(DM_IO_PAGES); |
971 | if (IS_ERR(ms->io_client)) { | |
972 | ti->error = "Error creating dm_io client"; | |
973 | kfree(ms); | |
974 | return NULL; | |
975 | } | |
976 | ||
1da177e4 | 977 | if (rh_init(&ms->rh, ms, dl, region_size, ms->nr_regions)) { |
72d94861 | 978 | ti->error = "Error creating dirty region hash"; |
a72cf737 | 979 | dm_io_client_destroy(ms->io_client); |
1da177e4 LT |
980 | kfree(ms); |
981 | return NULL; | |
982 | } | |
983 | ||
984 | return ms; | |
985 | } | |
986 | ||
987 | static void free_context(struct mirror_set *ms, struct dm_target *ti, | |
988 | unsigned int m) | |
989 | { | |
990 | while (m--) | |
991 | dm_put_device(ti, ms->mirror[m].dev); | |
992 | ||
88be163a | 993 | dm_io_client_destroy(ms->io_client); |
1da177e4 LT |
994 | rh_exit(&ms->rh); |
995 | kfree(ms); | |
996 | } | |
997 | ||
998 | static inline int _check_region_size(struct dm_target *ti, uint32_t size) | |
999 | { | |
6f3c3f0a | 1000 | return !(size % (PAGE_SIZE >> 9) || !is_power_of_2(size) || |
1da177e4 LT |
1001 | size > ti->len); |
1002 | } | |
1003 | ||
1004 | static int get_mirror(struct mirror_set *ms, struct dm_target *ti, | |
1005 | unsigned int mirror, char **argv) | |
1006 | { | |
4ee218cd | 1007 | unsigned long long offset; |
1da177e4 | 1008 | |
4ee218cd | 1009 | if (sscanf(argv[1], "%llu", &offset) != 1) { |
72d94861 | 1010 | ti->error = "Invalid offset"; |
1da177e4 LT |
1011 | return -EINVAL; |
1012 | } | |
1013 | ||
1014 | if (dm_get_device(ti, argv[0], offset, ti->len, | |
1015 | dm_table_get_mode(ti->table), | |
1016 | &ms->mirror[mirror].dev)) { | |
72d94861 | 1017 | ti->error = "Device lookup failure"; |
1da177e4 LT |
1018 | return -ENXIO; |
1019 | } | |
1020 | ||
aa5617c5 | 1021 | ms->mirror[mirror].ms = ms; |
1da177e4 LT |
1022 | ms->mirror[mirror].offset = offset; |
1023 | ||
1024 | return 0; | |
1025 | } | |
1026 | ||
1da177e4 LT |
1027 | /* |
1028 | * Create dirty log: log_type #log_params <log_params> | |
1029 | */ | |
1030 | static struct dirty_log *create_dirty_log(struct dm_target *ti, | |
1031 | unsigned int argc, char **argv, | |
1032 | unsigned int *args_used) | |
1033 | { | |
1034 | unsigned int param_count; | |
1035 | struct dirty_log *dl; | |
1036 | ||
1037 | if (argc < 2) { | |
72d94861 | 1038 | ti->error = "Insufficient mirror log arguments"; |
1da177e4 LT |
1039 | return NULL; |
1040 | } | |
1041 | ||
1042 | if (sscanf(argv[1], "%u", ¶m_count) != 1) { | |
72d94861 | 1043 | ti->error = "Invalid mirror log argument count"; |
1da177e4 LT |
1044 | return NULL; |
1045 | } | |
1046 | ||
1047 | *args_used = 2 + param_count; | |
1048 | ||
1049 | if (argc < *args_used) { | |
72d94861 | 1050 | ti->error = "Insufficient mirror log arguments"; |
1da177e4 LT |
1051 | return NULL; |
1052 | } | |
1053 | ||
1054 | dl = dm_create_dirty_log(argv[0], ti, param_count, argv + 2); | |
1055 | if (!dl) { | |
72d94861 | 1056 | ti->error = "Error creating mirror dirty log"; |
1da177e4 LT |
1057 | return NULL; |
1058 | } | |
1059 | ||
1060 | if (!_check_region_size(ti, dl->type->get_region_size(dl))) { | |
72d94861 | 1061 | ti->error = "Invalid region size"; |
1da177e4 LT |
1062 | dm_destroy_dirty_log(dl); |
1063 | return NULL; | |
1064 | } | |
1065 | ||
1066 | return dl; | |
1067 | } | |
1068 | ||
a8e6afa2 JB |
1069 | static int parse_features(struct mirror_set *ms, unsigned argc, char **argv, |
1070 | unsigned *args_used) | |
1071 | { | |
1072 | unsigned num_features; | |
1073 | struct dm_target *ti = ms->ti; | |
1074 | ||
1075 | *args_used = 0; | |
1076 | ||
1077 | if (!argc) | |
1078 | return 0; | |
1079 | ||
1080 | if (sscanf(argv[0], "%u", &num_features) != 1) { | |
1081 | ti->error = "Invalid number of features"; | |
1082 | return -EINVAL; | |
1083 | } | |
1084 | ||
1085 | argc--; | |
1086 | argv++; | |
1087 | (*args_used)++; | |
1088 | ||
1089 | if (num_features > argc) { | |
1090 | ti->error = "Not enough arguments to support feature count"; | |
1091 | return -EINVAL; | |
1092 | } | |
1093 | ||
1094 | if (!strcmp("handle_errors", argv[0])) | |
1095 | ms->features |= DM_RAID1_HANDLE_ERRORS; | |
1096 | else { | |
1097 | ti->error = "Unrecognised feature requested"; | |
1098 | return -EINVAL; | |
1099 | } | |
1100 | ||
1101 | (*args_used)++; | |
1102 | ||
1103 | return 0; | |
1104 | } | |
1105 | ||
1da177e4 LT |
1106 | /* |
1107 | * Construct a mirror mapping: | |
1108 | * | |
1109 | * log_type #log_params <log_params> | |
1110 | * #mirrors [mirror_path offset]{2,} | |
a8e6afa2 | 1111 | * [#features <features>] |
1da177e4 LT |
1112 | * |
1113 | * log_type is "core" or "disk" | |
1114 | * #log_params is between 1 and 3 | |
a8e6afa2 JB |
1115 | * |
1116 | * If present, features must be "handle_errors". | |
1da177e4 | 1117 | */ |
1da177e4 LT |
1118 | static int mirror_ctr(struct dm_target *ti, unsigned int argc, char **argv) |
1119 | { | |
1120 | int r; | |
1121 | unsigned int nr_mirrors, m, args_used; | |
1122 | struct mirror_set *ms; | |
1123 | struct dirty_log *dl; | |
1124 | ||
1125 | dl = create_dirty_log(ti, argc, argv, &args_used); | |
1126 | if (!dl) | |
1127 | return -EINVAL; | |
1128 | ||
1129 | argv += args_used; | |
1130 | argc -= args_used; | |
1131 | ||
1132 | if (!argc || sscanf(argv[0], "%u", &nr_mirrors) != 1 || | |
1133 | nr_mirrors < 2 || nr_mirrors > KCOPYD_MAX_REGIONS + 1) { | |
72d94861 | 1134 | ti->error = "Invalid number of mirrors"; |
1da177e4 LT |
1135 | dm_destroy_dirty_log(dl); |
1136 | return -EINVAL; | |
1137 | } | |
1138 | ||
1139 | argv++, argc--; | |
1140 | ||
a8e6afa2 JB |
1141 | if (argc < nr_mirrors * 2) { |
1142 | ti->error = "Too few mirror arguments"; | |
1da177e4 LT |
1143 | dm_destroy_dirty_log(dl); |
1144 | return -EINVAL; | |
1145 | } | |
1146 | ||
1147 | ms = alloc_context(nr_mirrors, dl->type->get_region_size(dl), ti, dl); | |
1148 | if (!ms) { | |
1149 | dm_destroy_dirty_log(dl); | |
1150 | return -ENOMEM; | |
1151 | } | |
1152 | ||
1153 | /* Get the mirror parameter sets */ | |
1154 | for (m = 0; m < nr_mirrors; m++) { | |
1155 | r = get_mirror(ms, ti, m, argv); | |
1156 | if (r) { | |
1157 | free_context(ms, ti, m); | |
1158 | return r; | |
1159 | } | |
1160 | argv += 2; | |
1161 | argc -= 2; | |
1162 | } | |
1163 | ||
1164 | ti->private = ms; | |
d88854f0 | 1165 | ti->split_io = ms->rh.region_size; |
1da177e4 | 1166 | |
6ad36fe2 HS |
1167 | ms->kmirrord_wq = create_singlethread_workqueue("kmirrord"); |
1168 | if (!ms->kmirrord_wq) { | |
1169 | DMERR("couldn't start kmirrord"); | |
a72cf737 DM |
1170 | r = -ENOMEM; |
1171 | goto err_free_context; | |
6ad36fe2 HS |
1172 | } |
1173 | INIT_WORK(&ms->kmirrord_work, do_mirror); | |
1174 | ||
a8e6afa2 | 1175 | r = parse_features(ms, argc, argv, &args_used); |
a72cf737 DM |
1176 | if (r) |
1177 | goto err_destroy_wq; | |
a8e6afa2 JB |
1178 | |
1179 | argv += args_used; | |
1180 | argc -= args_used; | |
1181 | ||
f44db678 JB |
1182 | /* |
1183 | * Any read-balancing addition depends on the | |
1184 | * DM_RAID1_HANDLE_ERRORS flag being present. | |
1185 | * This is because the decision to balance depends | |
1186 | * on the sync state of a region. If the above | |
1187 | * flag is not present, we ignore errors; and | |
1188 | * the sync state may be inaccurate. | |
1189 | */ | |
1190 | ||
a8e6afa2 JB |
1191 | if (argc) { |
1192 | ti->error = "Too many mirror arguments"; | |
a72cf737 DM |
1193 | r = -EINVAL; |
1194 | goto err_destroy_wq; | |
a8e6afa2 JB |
1195 | } |
1196 | ||
1da177e4 | 1197 | r = kcopyd_client_create(DM_IO_PAGES, &ms->kcopyd_client); |
a72cf737 DM |
1198 | if (r) |
1199 | goto err_destroy_wq; | |
1da177e4 | 1200 | |
6ad36fe2 | 1201 | wake(ms); |
1da177e4 | 1202 | return 0; |
a72cf737 DM |
1203 | |
1204 | err_destroy_wq: | |
1205 | destroy_workqueue(ms->kmirrord_wq); | |
1206 | err_free_context: | |
1207 | free_context(ms, ti, ms->nr_mirrors); | |
1208 | return r; | |
1da177e4 LT |
1209 | } |
1210 | ||
1211 | static void mirror_dtr(struct dm_target *ti) | |
1212 | { | |
1213 | struct mirror_set *ms = (struct mirror_set *) ti->private; | |
1214 | ||
6ad36fe2 | 1215 | flush_workqueue(ms->kmirrord_wq); |
1da177e4 | 1216 | kcopyd_client_destroy(ms->kcopyd_client); |
6ad36fe2 | 1217 | destroy_workqueue(ms->kmirrord_wq); |
1da177e4 LT |
1218 | free_context(ms, ti, ms->nr_mirrors); |
1219 | } | |
1220 | ||
1221 | static void queue_bio(struct mirror_set *ms, struct bio *bio, int rw) | |
1222 | { | |
1223 | int should_wake = 0; | |
1224 | struct bio_list *bl; | |
1225 | ||
1226 | bl = (rw == WRITE) ? &ms->writes : &ms->reads; | |
1227 | spin_lock(&ms->lock); | |
1228 | should_wake = !(bl->head); | |
1229 | bio_list_add(bl, bio); | |
1230 | spin_unlock(&ms->lock); | |
1231 | ||
1232 | if (should_wake) | |
6ad36fe2 | 1233 | wake(ms); |
1da177e4 LT |
1234 | } |
1235 | ||
1236 | /* | |
1237 | * Mirror mapping function | |
1238 | */ | |
1239 | static int mirror_map(struct dm_target *ti, struct bio *bio, | |
1240 | union map_info *map_context) | |
1241 | { | |
1242 | int r, rw = bio_rw(bio); | |
1243 | struct mirror *m; | |
1244 | struct mirror_set *ms = ti->private; | |
1245 | ||
e4c8b3ba | 1246 | map_context->ll = bio_to_region(&ms->rh, bio); |
1da177e4 LT |
1247 | |
1248 | if (rw == WRITE) { | |
1249 | queue_bio(ms, bio, rw); | |
d2a7ad29 | 1250 | return DM_MAPIO_SUBMITTED; |
1da177e4 LT |
1251 | } |
1252 | ||
1253 | r = ms->rh.log->type->in_sync(ms->rh.log, | |
1254 | bio_to_region(&ms->rh, bio), 0); | |
1255 | if (r < 0 && r != -EWOULDBLOCK) | |
1256 | return r; | |
1257 | ||
1258 | if (r == -EWOULDBLOCK) /* FIXME: ugly */ | |
d2a7ad29 | 1259 | r = DM_MAPIO_SUBMITTED; |
1da177e4 LT |
1260 | |
1261 | /* | |
1262 | * We don't want to fast track a recovery just for a read | |
1263 | * ahead. So we just let it silently fail. | |
1264 | * FIXME: get rid of this. | |
1265 | */ | |
1266 | if (!r && rw == READA) | |
1267 | return -EIO; | |
1268 | ||
1269 | if (!r) { | |
1270 | /* Pass this io over to the daemon */ | |
1271 | queue_bio(ms, bio, rw); | |
d2a7ad29 | 1272 | return DM_MAPIO_SUBMITTED; |
1da177e4 LT |
1273 | } |
1274 | ||
1275 | m = choose_mirror(ms, bio->bi_sector); | |
1276 | if (!m) | |
1277 | return -EIO; | |
1278 | ||
1279 | map_bio(ms, m, bio); | |
d2a7ad29 | 1280 | return DM_MAPIO_REMAPPED; |
1da177e4 LT |
1281 | } |
1282 | ||
1283 | static int mirror_end_io(struct dm_target *ti, struct bio *bio, | |
1284 | int error, union map_info *map_context) | |
1285 | { | |
1286 | int rw = bio_rw(bio); | |
1287 | struct mirror_set *ms = (struct mirror_set *) ti->private; | |
1288 | region_t region = map_context->ll; | |
1289 | ||
1290 | /* | |
1291 | * We need to dec pending if this was a write. | |
1292 | */ | |
1293 | if (rw == WRITE) | |
1294 | rh_dec(&ms->rh, region); | |
1295 | ||
1296 | return 0; | |
1297 | } | |
1298 | ||
1299 | static void mirror_postsuspend(struct dm_target *ti) | |
1300 | { | |
1301 | struct mirror_set *ms = (struct mirror_set *) ti->private; | |
1302 | struct dirty_log *log = ms->rh.log; | |
1303 | ||
1304 | rh_stop_recovery(&ms->rh); | |
33184048 JB |
1305 | |
1306 | /* Wait for all I/O we generated to complete */ | |
1307 | wait_event(_kmirrord_recovery_stopped, | |
1308 | !atomic_read(&ms->rh.recovery_in_flight)); | |
1309 | ||
6b3df0d7 | 1310 | if (log->type->postsuspend && log->type->postsuspend(log)) |
1da177e4 LT |
1311 | /* FIXME: need better error handling */ |
1312 | DMWARN("log suspend failed"); | |
1313 | } | |
1314 | ||
1315 | static void mirror_resume(struct dm_target *ti) | |
1316 | { | |
1317 | struct mirror_set *ms = (struct mirror_set *) ti->private; | |
1318 | struct dirty_log *log = ms->rh.log; | |
1319 | if (log->type->resume && log->type->resume(log)) | |
1320 | /* FIXME: need better error handling */ | |
1321 | DMWARN("log resume failed"); | |
1322 | rh_start_recovery(&ms->rh); | |
1323 | } | |
1324 | ||
1325 | static int mirror_status(struct dm_target *ti, status_type_t type, | |
1326 | char *result, unsigned int maxlen) | |
1327 | { | |
315dcc22 | 1328 | unsigned int m, sz = 0; |
1da177e4 LT |
1329 | struct mirror_set *ms = (struct mirror_set *) ti->private; |
1330 | ||
1da177e4 LT |
1331 | switch (type) { |
1332 | case STATUSTYPE_INFO: | |
1333 | DMEMIT("%d ", ms->nr_mirrors); | |
1334 | for (m = 0; m < ms->nr_mirrors; m++) | |
1335 | DMEMIT("%s ", ms->mirror[m].dev->name); | |
1336 | ||
c95bc206 | 1337 | DMEMIT("%llu/%llu 0 ", |
4ee218cd AM |
1338 | (unsigned long long)ms->rh.log->type-> |
1339 | get_sync_count(ms->rh.log), | |
1340 | (unsigned long long)ms->nr_regions); | |
315dcc22 | 1341 | |
c95bc206 | 1342 | sz += ms->rh.log->type->status(ms->rh.log, type, result+sz, maxlen-sz); |
315dcc22 | 1343 | |
1da177e4 LT |
1344 | break; |
1345 | ||
1346 | case STATUSTYPE_TABLE: | |
315dcc22 JB |
1347 | sz = ms->rh.log->type->status(ms->rh.log, type, result, maxlen); |
1348 | ||
e52b8f6d | 1349 | DMEMIT("%d", ms->nr_mirrors); |
1da177e4 | 1350 | for (m = 0; m < ms->nr_mirrors; m++) |
e52b8f6d | 1351 | DMEMIT(" %s %llu", ms->mirror[m].dev->name, |
4ee218cd | 1352 | (unsigned long long)ms->mirror[m].offset); |
a8e6afa2 JB |
1353 | |
1354 | if (ms->features & DM_RAID1_HANDLE_ERRORS) | |
1355 | DMEMIT(" 1 handle_errors"); | |
1da177e4 LT |
1356 | } |
1357 | ||
1358 | return 0; | |
1359 | } | |
1360 | ||
1361 | static struct target_type mirror_target = { | |
1362 | .name = "mirror", | |
315dcc22 | 1363 | .version = {1, 0, 3}, |
1da177e4 LT |
1364 | .module = THIS_MODULE, |
1365 | .ctr = mirror_ctr, | |
1366 | .dtr = mirror_dtr, | |
1367 | .map = mirror_map, | |
1368 | .end_io = mirror_end_io, | |
1369 | .postsuspend = mirror_postsuspend, | |
1370 | .resume = mirror_resume, | |
1371 | .status = mirror_status, | |
1372 | }; | |
1373 | ||
1374 | static int __init dm_mirror_init(void) | |
1375 | { | |
1376 | int r; | |
1377 | ||
1378 | r = dm_dirty_log_init(); | |
1379 | if (r) | |
1380 | return r; | |
1381 | ||
1da177e4 LT |
1382 | r = dm_register_target(&mirror_target); |
1383 | if (r < 0) { | |
0cd33124 | 1384 | DMERR("Failed to register mirror target"); |
1da177e4 | 1385 | dm_dirty_log_exit(); |
1da177e4 LT |
1386 | } |
1387 | ||
1388 | return r; | |
1389 | } | |
1390 | ||
1391 | static void __exit dm_mirror_exit(void) | |
1392 | { | |
1393 | int r; | |
1394 | ||
1395 | r = dm_unregister_target(&mirror_target); | |
1396 | if (r < 0) | |
0cd33124 | 1397 | DMERR("unregister failed %d", r); |
1da177e4 | 1398 | |
1da177e4 LT |
1399 | dm_dirty_log_exit(); |
1400 | } | |
1401 | ||
1402 | /* Module hooks */ | |
1403 | module_init(dm_mirror_init); | |
1404 | module_exit(dm_mirror_exit); | |
1405 | ||
1406 | MODULE_DESCRIPTION(DM_NAME " mirror target"); | |
1407 | MODULE_AUTHOR("Joe Thornber"); | |
1408 | MODULE_LICENSE("GPL"); |