blob: 84f2ed6d9c5b077b909d77e289097646c5554185 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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>
22
23static struct workqueue_struct *_kmirrord_wq;
24static struct work_struct _kmirrord_work;
25
26static inline void wake(void)
27{
28 queue_work(_kmirrord_wq, &_kmirrord_work);
29}
30
31/*-----------------------------------------------------------------
32 * Region hash
33 *
34 * The mirror splits itself up into discrete regions. Each
35 * region can be in one of three states: clean, dirty,
36 * nosync. There is no need to put clean regions in the hash.
37 *
38 * In addition to being present in the hash table a region _may_
39 * be present on one of three lists.
40 *
41 * clean_regions: Regions on this list have no io pending to
42 * them, they are in sync, we are no longer interested in them,
43 * they are dull. rh_update_states() will remove them from the
44 * hash table.
45 *
46 * quiesced_regions: These regions have been spun down, ready
47 * for recovery. rh_recovery_start() will remove regions from
48 * this list and hand them to kmirrord, which will schedule the
49 * recovery io with kcopyd.
50 *
51 * recovered_regions: Regions that kcopyd has successfully
52 * recovered. rh_update_states() will now schedule any delayed
53 * io, up the recovery_count, and remove the region from the
54 * hash.
55 *
56 * There are 2 locks:
57 * A rw spin lock 'hash_lock' protects just the hash table,
58 * this is never held in write mode from interrupt context,
59 * which I believe means that we only have to disable irqs when
60 * doing a write lock.
61 *
62 * An ordinary spin lock 'region_lock' that protects the three
63 * lists in the region_hash, with the 'state', 'list' and
64 * 'bhs_delayed' fields of the regions. This is used from irq
65 * context, so all other uses will have to suspend local irqs.
66 *---------------------------------------------------------------*/
67struct mirror_set;
68struct region_hash {
69 struct mirror_set *ms;
70 uint32_t region_size;
71 unsigned region_shift;
72
73 /* holds persistent region state */
74 struct dirty_log *log;
75
76 /* hash table */
77 rwlock_t hash_lock;
78 mempool_t *region_pool;
79 unsigned int mask;
80 unsigned int nr_buckets;
81 struct list_head *buckets;
82
83 spinlock_t region_lock;
84 struct semaphore recovery_count;
85 struct list_head clean_regions;
86 struct list_head quiesced_regions;
87 struct list_head recovered_regions;
88};
89
90enum {
91 RH_CLEAN,
92 RH_DIRTY,
93 RH_NOSYNC,
94 RH_RECOVERING
95};
96
97struct region {
98 struct region_hash *rh; /* FIXME: can we get rid of this ? */
99 region_t key;
100 int state;
101
102 struct list_head hash_list;
103 struct list_head list;
104
105 atomic_t pending;
106 struct bio_list delayed_bios;
107};
108
Neil Browne4c8b3b2006-06-26 00:27:26 -0700109
110/*-----------------------------------------------------------------
111 * Mirror set structures.
112 *---------------------------------------------------------------*/
113struct mirror {
114 atomic_t error_count;
115 struct dm_dev *dev;
116 sector_t offset;
117};
118
119struct mirror_set {
120 struct dm_target *ti;
121 struct list_head list;
122 struct region_hash rh;
123 struct kcopyd_client *kcopyd_client;
124
125 spinlock_t lock; /* protects the next two lists */
126 struct bio_list reads;
127 struct bio_list writes;
128
129 /* recovery */
130 region_t nr_regions;
131 int in_sync;
132
133 struct mirror *default_mirror; /* Default mirror */
134
135 unsigned int nr_mirrors;
136 struct mirror mirror[0];
137};
138
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139/*
140 * Conversion fns
141 */
142static inline region_t bio_to_region(struct region_hash *rh, struct bio *bio)
143{
Neil Browne4c8b3b2006-06-26 00:27:26 -0700144 return (bio->bi_sector - rh->ms->ti->begin) >> rh->region_shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145}
146
147static inline sector_t region_to_sector(struct region_hash *rh, region_t region)
148{
149 return region << rh->region_shift;
150}
151
152/* FIXME move this */
153static void queue_bio(struct mirror_set *ms, struct bio *bio, int rw);
154
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155#define MIN_REGIONS 64
156#define MAX_RECOVERY 1
157static int rh_init(struct region_hash *rh, struct mirror_set *ms,
158 struct dirty_log *log, uint32_t region_size,
159 region_t nr_regions)
160{
161 unsigned int nr_buckets, max_buckets;
162 size_t i;
163
164 /*
165 * Calculate a suitable number of buckets for our hash
166 * table.
167 */
168 max_buckets = nr_regions >> 6;
169 for (nr_buckets = 128u; nr_buckets < max_buckets; nr_buckets <<= 1)
170 ;
171 nr_buckets >>= 1;
172
173 rh->ms = ms;
174 rh->log = log;
175 rh->region_size = region_size;
176 rh->region_shift = ffs(region_size) - 1;
177 rwlock_init(&rh->hash_lock);
178 rh->mask = nr_buckets - 1;
179 rh->nr_buckets = nr_buckets;
180
181 rh->buckets = vmalloc(nr_buckets * sizeof(*rh->buckets));
182 if (!rh->buckets) {
183 DMERR("unable to allocate region hash memory");
184 return -ENOMEM;
185 }
186
187 for (i = 0; i < nr_buckets; i++)
188 INIT_LIST_HEAD(rh->buckets + i);
189
190 spin_lock_init(&rh->region_lock);
191 sema_init(&rh->recovery_count, 0);
192 INIT_LIST_HEAD(&rh->clean_regions);
193 INIT_LIST_HEAD(&rh->quiesced_regions);
194 INIT_LIST_HEAD(&rh->recovered_regions);
195
Matthew Dobson0eaae62a2006-03-26 01:37:47 -0800196 rh->region_pool = mempool_create_kmalloc_pool(MIN_REGIONS,
197 sizeof(struct region));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 if (!rh->region_pool) {
199 vfree(rh->buckets);
200 rh->buckets = NULL;
201 return -ENOMEM;
202 }
203
204 return 0;
205}
206
207static void rh_exit(struct region_hash *rh)
208{
209 unsigned int h;
210 struct region *reg, *nreg;
211
212 BUG_ON(!list_empty(&rh->quiesced_regions));
213 for (h = 0; h < rh->nr_buckets; h++) {
214 list_for_each_entry_safe(reg, nreg, rh->buckets + h, hash_list) {
215 BUG_ON(atomic_read(&reg->pending));
216 mempool_free(reg, rh->region_pool);
217 }
218 }
219
220 if (rh->log)
221 dm_destroy_dirty_log(rh->log);
222 if (rh->region_pool)
223 mempool_destroy(rh->region_pool);
224 vfree(rh->buckets);
225}
226
227#define RH_HASH_MULT 2654435387U
228
229static inline unsigned int rh_hash(struct region_hash *rh, region_t region)
230{
231 return (unsigned int) ((region * RH_HASH_MULT) >> 12) & rh->mask;
232}
233
234static struct region *__rh_lookup(struct region_hash *rh, region_t region)
235{
236 struct region *reg;
237
238 list_for_each_entry (reg, rh->buckets + rh_hash(rh, region), hash_list)
239 if (reg->key == region)
240 return reg;
241
242 return NULL;
243}
244
245static void __rh_insert(struct region_hash *rh, struct region *reg)
246{
247 unsigned int h = rh_hash(rh, reg->key);
248 list_add(&reg->hash_list, rh->buckets + h);
249}
250
251static struct region *__rh_alloc(struct region_hash *rh, region_t region)
252{
253 struct region *reg, *nreg;
254
255 read_unlock(&rh->hash_lock);
256 nreg = mempool_alloc(rh->region_pool, GFP_NOIO);
257 nreg->state = rh->log->type->in_sync(rh->log, region, 1) ?
258 RH_CLEAN : RH_NOSYNC;
259 nreg->rh = rh;
260 nreg->key = region;
261
262 INIT_LIST_HEAD(&nreg->list);
263
264 atomic_set(&nreg->pending, 0);
265 bio_list_init(&nreg->delayed_bios);
266 write_lock_irq(&rh->hash_lock);
267
268 reg = __rh_lookup(rh, region);
269 if (reg)
270 /* we lost the race */
271 mempool_free(nreg, rh->region_pool);
272
273 else {
274 __rh_insert(rh, nreg);
275 if (nreg->state == RH_CLEAN) {
276 spin_lock(&rh->region_lock);
277 list_add(&nreg->list, &rh->clean_regions);
278 spin_unlock(&rh->region_lock);
279 }
280 reg = nreg;
281 }
282 write_unlock_irq(&rh->hash_lock);
283 read_lock(&rh->hash_lock);
284
285 return reg;
286}
287
288static inline struct region *__rh_find(struct region_hash *rh, region_t region)
289{
290 struct region *reg;
291
292 reg = __rh_lookup(rh, region);
293 if (!reg)
294 reg = __rh_alloc(rh, region);
295
296 return reg;
297}
298
299static int rh_state(struct region_hash *rh, region_t region, int may_block)
300{
301 int r;
302 struct region *reg;
303
304 read_lock(&rh->hash_lock);
305 reg = __rh_lookup(rh, region);
306 read_unlock(&rh->hash_lock);
307
308 if (reg)
309 return reg->state;
310
311 /*
312 * The region wasn't in the hash, so we fall back to the
313 * dirty log.
314 */
315 r = rh->log->type->in_sync(rh->log, region, may_block);
316
317 /*
318 * Any error from the dirty log (eg. -EWOULDBLOCK) gets
319 * taken as a RH_NOSYNC
320 */
321 return r == 1 ? RH_CLEAN : RH_NOSYNC;
322}
323
324static inline int rh_in_sync(struct region_hash *rh,
325 region_t region, int may_block)
326{
327 int state = rh_state(rh, region, may_block);
328 return state == RH_CLEAN || state == RH_DIRTY;
329}
330
331static void dispatch_bios(struct mirror_set *ms, struct bio_list *bio_list)
332{
333 struct bio *bio;
334
335 while ((bio = bio_list_pop(bio_list))) {
336 queue_bio(ms, bio, WRITE);
337 }
338}
339
340static void rh_update_states(struct region_hash *rh)
341{
342 struct region *reg, *next;
343
344 LIST_HEAD(clean);
345 LIST_HEAD(recovered);
346
347 /*
348 * Quickly grab the lists.
349 */
350 write_lock_irq(&rh->hash_lock);
351 spin_lock(&rh->region_lock);
352 if (!list_empty(&rh->clean_regions)) {
353 list_splice(&rh->clean_regions, &clean);
354 INIT_LIST_HEAD(&rh->clean_regions);
355
356 list_for_each_entry (reg, &clean, list) {
357 rh->log->type->clear_region(rh->log, reg->key);
358 list_del(&reg->hash_list);
359 }
360 }
361
362 if (!list_empty(&rh->recovered_regions)) {
363 list_splice(&rh->recovered_regions, &recovered);
364 INIT_LIST_HEAD(&rh->recovered_regions);
365
366 list_for_each_entry (reg, &recovered, list)
367 list_del(&reg->hash_list);
368 }
369 spin_unlock(&rh->region_lock);
370 write_unlock_irq(&rh->hash_lock);
371
372 /*
373 * All the regions on the recovered and clean lists have
374 * now been pulled out of the system, so no need to do
375 * any more locking.
376 */
377 list_for_each_entry_safe (reg, next, &recovered, list) {
378 rh->log->type->clear_region(rh->log, reg->key);
379 rh->log->type->complete_resync_work(rh->log, reg->key, 1);
380 dispatch_bios(rh->ms, &reg->delayed_bios);
381 up(&rh->recovery_count);
382 mempool_free(reg, rh->region_pool);
383 }
384
385 if (!list_empty(&recovered))
386 rh->log->type->flush(rh->log);
387
388 list_for_each_entry_safe (reg, next, &clean, list)
389 mempool_free(reg, rh->region_pool);
390}
391
392static void rh_inc(struct region_hash *rh, region_t region)
393{
394 struct region *reg;
395
396 read_lock(&rh->hash_lock);
397 reg = __rh_find(rh, region);
Jun'ichi Nomura844e8d92005-09-09 16:23:42 -0700398
Jonathan E Brassow7692c5d2005-11-21 21:32:37 -0800399 spin_lock_irq(&rh->region_lock);
Jun'ichi Nomura844e8d92005-09-09 16:23:42 -0700400 atomic_inc(&reg->pending);
401
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 if (reg->state == RH_CLEAN) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403 reg->state = RH_DIRTY;
404 list_del_init(&reg->list); /* take off the clean list */
Jonathan E Brassow7692c5d2005-11-21 21:32:37 -0800405 spin_unlock_irq(&rh->region_lock);
406
407 rh->log->type->mark_region(rh->log, reg->key);
408 } else
409 spin_unlock_irq(&rh->region_lock);
410
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412 read_unlock(&rh->hash_lock);
413}
414
415static void rh_inc_pending(struct region_hash *rh, struct bio_list *bios)
416{
417 struct bio *bio;
418
419 for (bio = bios->head; bio; bio = bio->bi_next)
420 rh_inc(rh, bio_to_region(rh, bio));
421}
422
423static void rh_dec(struct region_hash *rh, region_t region)
424{
425 unsigned long flags;
426 struct region *reg;
427 int should_wake = 0;
428
429 read_lock(&rh->hash_lock);
430 reg = __rh_lookup(rh, region);
431 read_unlock(&rh->hash_lock);
432
Jonathan E Brassow7692c5d2005-11-21 21:32:37 -0800433 spin_lock_irqsave(&rh->region_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434 if (atomic_dec_and_test(&reg->pending)) {
Jun'ichi Nomura930d3322006-03-27 01:17:47 -0800435 /*
436 * There is no pending I/O for this region.
437 * We can move the region to corresponding list for next action.
438 * At this point, the region is not yet connected to any list.
439 *
440 * If the state is RH_NOSYNC, the region should be kept off
441 * from clean list.
442 * The hash entry for RH_NOSYNC will remain in memory
443 * until the region is recovered or the map is reloaded.
444 */
445
446 /* do nothing for RH_NOSYNC */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 if (reg->state == RH_RECOVERING) {
448 list_add_tail(&reg->list, &rh->quiesced_regions);
Jun'ichi Nomura930d3322006-03-27 01:17:47 -0800449 } else if (reg->state == RH_DIRTY) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 reg->state = RH_CLEAN;
451 list_add(&reg->list, &rh->clean_regions);
452 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 should_wake = 1;
454 }
Jonathan E Brassow7692c5d2005-11-21 21:32:37 -0800455 spin_unlock_irqrestore(&rh->region_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456
457 if (should_wake)
458 wake();
459}
460
461/*
462 * Starts quiescing a region in preparation for recovery.
463 */
464static int __rh_recovery_prepare(struct region_hash *rh)
465{
466 int r;
467 struct region *reg;
468 region_t region;
469
470 /*
471 * Ask the dirty log what's next.
472 */
473 r = rh->log->type->get_resync_work(rh->log, &region);
474 if (r <= 0)
475 return r;
476
477 /*
478 * Get this region, and start it quiescing by setting the
479 * recovering flag.
480 */
481 read_lock(&rh->hash_lock);
482 reg = __rh_find(rh, region);
483 read_unlock(&rh->hash_lock);
484
485 spin_lock_irq(&rh->region_lock);
486 reg->state = RH_RECOVERING;
487
488 /* Already quiesced ? */
489 if (atomic_read(&reg->pending))
490 list_del_init(&reg->list);
Akinobu Mita179e0912006-06-26 00:24:41 -0700491 else
492 list_move(&reg->list, &rh->quiesced_regions);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494 spin_unlock_irq(&rh->region_lock);
495
496 return 1;
497}
498
499static void rh_recovery_prepare(struct region_hash *rh)
500{
501 while (!down_trylock(&rh->recovery_count))
502 if (__rh_recovery_prepare(rh) <= 0) {
503 up(&rh->recovery_count);
504 break;
505 }
506}
507
508/*
509 * Returns any quiesced regions.
510 */
511static struct region *rh_recovery_start(struct region_hash *rh)
512{
513 struct region *reg = NULL;
514
515 spin_lock_irq(&rh->region_lock);
516 if (!list_empty(&rh->quiesced_regions)) {
517 reg = list_entry(rh->quiesced_regions.next,
518 struct region, list);
519 list_del_init(&reg->list); /* remove from the quiesced list */
520 }
521 spin_unlock_irq(&rh->region_lock);
522
523 return reg;
524}
525
526/* FIXME: success ignored for now */
527static void rh_recovery_end(struct region *reg, int success)
528{
529 struct region_hash *rh = reg->rh;
530
531 spin_lock_irq(&rh->region_lock);
532 list_add(&reg->list, &reg->rh->recovered_regions);
533 spin_unlock_irq(&rh->region_lock);
534
535 wake();
536}
537
538static void rh_flush(struct region_hash *rh)
539{
540 rh->log->type->flush(rh->log);
541}
542
543static void rh_delay(struct region_hash *rh, struct bio *bio)
544{
545 struct region *reg;
546
547 read_lock(&rh->hash_lock);
548 reg = __rh_find(rh, bio_to_region(rh, bio));
549 bio_list_add(&reg->delayed_bios, bio);
550 read_unlock(&rh->hash_lock);
551}
552
553static void rh_stop_recovery(struct region_hash *rh)
554{
555 int i;
556
557 /* wait for any recovering regions */
558 for (i = 0; i < MAX_RECOVERY; i++)
559 down(&rh->recovery_count);
560}
561
562static void rh_start_recovery(struct region_hash *rh)
563{
564 int i;
565
566 for (i = 0; i < MAX_RECOVERY; i++)
567 up(&rh->recovery_count);
568
569 wake();
570}
571
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572/*
573 * Every mirror should look like this one.
574 */
575#define DEFAULT_MIRROR 0
576
577/*
578 * This is yucky. We squirrel the mirror_set struct away inside
579 * bi_next for write buffers. This is safe since the bh
580 * doesn't get submitted to the lower levels of block layer.
581 */
582static struct mirror_set *bio_get_ms(struct bio *bio)
583{
584 return (struct mirror_set *) bio->bi_next;
585}
586
587static void bio_set_ms(struct bio *bio, struct mirror_set *ms)
588{
589 bio->bi_next = (struct bio *) ms;
590}
591
592/*-----------------------------------------------------------------
593 * Recovery.
594 *
595 * When a mirror is first activated we may find that some regions
596 * are in the no-sync state. We have to recover these by
597 * recopying from the default mirror to all the others.
598 *---------------------------------------------------------------*/
599static void recovery_complete(int read_err, unsigned int write_err,
600 void *context)
601{
602 struct region *reg = (struct region *) context;
603
604 /* FIXME: better error handling */
605 rh_recovery_end(reg, read_err || write_err);
606}
607
608static int recover(struct mirror_set *ms, struct region *reg)
609{
610 int r;
611 unsigned int i;
612 struct io_region from, to[KCOPYD_MAX_REGIONS], *dest;
613 struct mirror *m;
614 unsigned long flags = 0;
615
616 /* fill in the source */
Jonathan E Brassowa1a19082006-01-06 00:20:05 -0800617 m = ms->default_mirror;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 from.bdev = m->dev->bdev;
619 from.sector = m->offset + region_to_sector(reg->rh, reg->key);
620 if (reg->key == (ms->nr_regions - 1)) {
621 /*
622 * The final region may be smaller than
623 * region_size.
624 */
625 from.count = ms->ti->len & (reg->rh->region_size - 1);
626 if (!from.count)
627 from.count = reg->rh->region_size;
628 } else
629 from.count = reg->rh->region_size;
630
631 /* fill in the destinations */
632 for (i = 0, dest = to; i < ms->nr_mirrors; i++) {
Jonathan E Brassowa1a19082006-01-06 00:20:05 -0800633 if (&ms->mirror[i] == ms->default_mirror)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 continue;
635
636 m = ms->mirror + i;
637 dest->bdev = m->dev->bdev;
638 dest->sector = m->offset + region_to_sector(reg->rh, reg->key);
639 dest->count = from.count;
640 dest++;
641 }
642
643 /* hand to kcopyd */
644 set_bit(KCOPYD_IGNORE_ERROR, &flags);
645 r = kcopyd_copy(ms->kcopyd_client, &from, ms->nr_mirrors - 1, to, flags,
646 recovery_complete, reg);
647
648 return r;
649}
650
651static void do_recovery(struct mirror_set *ms)
652{
653 int r;
654 struct region *reg;
655 struct dirty_log *log = ms->rh.log;
656
657 /*
658 * Start quiescing some regions.
659 */
660 rh_recovery_prepare(&ms->rh);
661
662 /*
663 * Copy any already quiesced regions.
664 */
665 while ((reg = rh_recovery_start(&ms->rh))) {
666 r = recover(ms, reg);
667 if (r)
668 rh_recovery_end(reg, 0);
669 }
670
671 /*
672 * Update the in sync flag.
673 */
674 if (!ms->in_sync &&
675 (log->type->get_sync_count(log) == ms->nr_regions)) {
676 /* the sync is complete */
677 dm_table_event(ms->ti->table);
678 ms->in_sync = 1;
679 }
680}
681
682/*-----------------------------------------------------------------
683 * Reads
684 *---------------------------------------------------------------*/
685static struct mirror *choose_mirror(struct mirror_set *ms, sector_t sector)
686{
687 /* FIXME: add read balancing */
Jonathan E Brassowa1a19082006-01-06 00:20:05 -0800688 return ms->default_mirror;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689}
690
691/*
692 * remap a buffer to a particular mirror.
693 */
694static void map_bio(struct mirror_set *ms, struct mirror *m, struct bio *bio)
695{
696 bio->bi_bdev = m->dev->bdev;
697 bio->bi_sector = m->offset + (bio->bi_sector - ms->ti->begin);
698}
699
700static void do_reads(struct mirror_set *ms, struct bio_list *reads)
701{
702 region_t region;
703 struct bio *bio;
704 struct mirror *m;
705
706 while ((bio = bio_list_pop(reads))) {
707 region = bio_to_region(&ms->rh, bio);
708
709 /*
710 * We can only read balance if the region is in sync.
711 */
712 if (rh_in_sync(&ms->rh, region, 0))
713 m = choose_mirror(ms, bio->bi_sector);
714 else
Jonathan E Brassowa1a19082006-01-06 00:20:05 -0800715 m = ms->default_mirror;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716
717 map_bio(ms, m, bio);
718 generic_make_request(bio);
719 }
720}
721
722/*-----------------------------------------------------------------
723 * Writes.
724 *
725 * We do different things with the write io depending on the
726 * state of the region that it's in:
727 *
728 * SYNC: increment pending, use kcopyd to write to *all* mirrors
729 * RECOVERING: delay the io until recovery completes
730 * NOSYNC: increment pending, just write to the default mirror
731 *---------------------------------------------------------------*/
732static void write_callback(unsigned long error, void *context)
733{
734 unsigned int i;
735 int uptodate = 1;
736 struct bio *bio = (struct bio *) context;
737 struct mirror_set *ms;
738
739 ms = bio_get_ms(bio);
740 bio_set_ms(bio, NULL);
741
742 /*
743 * NOTE: We don't decrement the pending count here,
744 * instead it is done by the targets endio function.
745 * This way we handle both writes to SYNC and NOSYNC
746 * regions with the same code.
747 */
748
749 if (error) {
750 /*
751 * only error the io if all mirrors failed.
752 * FIXME: bogus
753 */
754 uptodate = 0;
755 for (i = 0; i < ms->nr_mirrors; i++)
756 if (!test_bit(i, &error)) {
757 uptodate = 1;
758 break;
759 }
760 }
761 bio_endio(bio, bio->bi_size, 0);
762}
763
764static void do_write(struct mirror_set *ms, struct bio *bio)
765{
766 unsigned int i;
767 struct io_region io[KCOPYD_MAX_REGIONS+1];
768 struct mirror *m;
769
770 for (i = 0; i < ms->nr_mirrors; i++) {
771 m = ms->mirror + i;
772
773 io[i].bdev = m->dev->bdev;
774 io[i].sector = m->offset + (bio->bi_sector - ms->ti->begin);
775 io[i].count = bio->bi_size >> 9;
776 }
777
778 bio_set_ms(bio, ms);
779 dm_io_async_bvec(ms->nr_mirrors, io, WRITE,
780 bio->bi_io_vec + bio->bi_idx,
781 write_callback, bio);
782}
783
784static void do_writes(struct mirror_set *ms, struct bio_list *writes)
785{
786 int state;
787 struct bio *bio;
788 struct bio_list sync, nosync, recover, *this_list = NULL;
789
790 if (!writes->head)
791 return;
792
793 /*
794 * Classify each write.
795 */
796 bio_list_init(&sync);
797 bio_list_init(&nosync);
798 bio_list_init(&recover);
799
800 while ((bio = bio_list_pop(writes))) {
801 state = rh_state(&ms->rh, bio_to_region(&ms->rh, bio), 1);
802 switch (state) {
803 case RH_CLEAN:
804 case RH_DIRTY:
805 this_list = &sync;
806 break;
807
808 case RH_NOSYNC:
809 this_list = &nosync;
810 break;
811
812 case RH_RECOVERING:
813 this_list = &recover;
814 break;
815 }
816
817 bio_list_add(this_list, bio);
818 }
819
820 /*
821 * Increment the pending counts for any regions that will
822 * be written to (writes to recover regions are going to
823 * be delayed).
824 */
825 rh_inc_pending(&ms->rh, &sync);
826 rh_inc_pending(&ms->rh, &nosync);
827 rh_flush(&ms->rh);
828
829 /*
830 * Dispatch io.
831 */
832 while ((bio = bio_list_pop(&sync)))
833 do_write(ms, bio);
834
835 while ((bio = bio_list_pop(&recover)))
836 rh_delay(&ms->rh, bio);
837
838 while ((bio = bio_list_pop(&nosync))) {
Jonathan E Brassowa1a19082006-01-06 00:20:05 -0800839 map_bio(ms, ms->default_mirror, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 generic_make_request(bio);
841 }
842}
843
844/*-----------------------------------------------------------------
845 * kmirrord
846 *---------------------------------------------------------------*/
847static LIST_HEAD(_mirror_sets);
848static DECLARE_RWSEM(_mirror_sets_lock);
849
850static void do_mirror(struct mirror_set *ms)
851{
852 struct bio_list reads, writes;
853
854 spin_lock(&ms->lock);
855 reads = ms->reads;
856 writes = ms->writes;
857 bio_list_init(&ms->reads);
858 bio_list_init(&ms->writes);
859 spin_unlock(&ms->lock);
860
861 rh_update_states(&ms->rh);
862 do_recovery(ms);
863 do_reads(ms, &reads);
864 do_writes(ms, &writes);
865}
866
867static void do_work(void *ignored)
868{
869 struct mirror_set *ms;
870
871 down_read(&_mirror_sets_lock);
872 list_for_each_entry (ms, &_mirror_sets, list)
873 do_mirror(ms);
874 up_read(&_mirror_sets_lock);
875}
876
877/*-----------------------------------------------------------------
878 * Target functions
879 *---------------------------------------------------------------*/
880static struct mirror_set *alloc_context(unsigned int nr_mirrors,
881 uint32_t region_size,
882 struct dm_target *ti,
883 struct dirty_log *dl)
884{
885 size_t len;
886 struct mirror_set *ms = NULL;
887
888 if (array_too_big(sizeof(*ms), sizeof(ms->mirror[0]), nr_mirrors))
889 return NULL;
890
891 len = sizeof(*ms) + (sizeof(ms->mirror[0]) * nr_mirrors);
892
893 ms = kmalloc(len, GFP_KERNEL);
894 if (!ms) {
895 ti->error = "dm-mirror: Cannot allocate mirror context";
896 return NULL;
897 }
898
899 memset(ms, 0, len);
900 spin_lock_init(&ms->lock);
901
902 ms->ti = ti;
903 ms->nr_mirrors = nr_mirrors;
904 ms->nr_regions = dm_sector_div_up(ti->len, region_size);
905 ms->in_sync = 0;
Jonathan E Brassowa1a19082006-01-06 00:20:05 -0800906 ms->default_mirror = &ms->mirror[DEFAULT_MIRROR];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907
908 if (rh_init(&ms->rh, ms, dl, region_size, ms->nr_regions)) {
909 ti->error = "dm-mirror: Error creating dirty region hash";
910 kfree(ms);
911 return NULL;
912 }
913
914 return ms;
915}
916
917static void free_context(struct mirror_set *ms, struct dm_target *ti,
918 unsigned int m)
919{
920 while (m--)
921 dm_put_device(ti, ms->mirror[m].dev);
922
923 rh_exit(&ms->rh);
924 kfree(ms);
925}
926
927static inline int _check_region_size(struct dm_target *ti, uint32_t size)
928{
929 return !(size % (PAGE_SIZE >> 9) || (size & (size - 1)) ||
930 size > ti->len);
931}
932
933static int get_mirror(struct mirror_set *ms, struct dm_target *ti,
934 unsigned int mirror, char **argv)
935{
Andrew Morton4ee218c2006-03-27 01:17:48 -0800936 unsigned long long offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937
Andrew Morton4ee218c2006-03-27 01:17:48 -0800938 if (sscanf(argv[1], "%llu", &offset) != 1) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 ti->error = "dm-mirror: Invalid offset";
940 return -EINVAL;
941 }
942
943 if (dm_get_device(ti, argv[0], offset, ti->len,
944 dm_table_get_mode(ti->table),
945 &ms->mirror[mirror].dev)) {
946 ti->error = "dm-mirror: Device lookup failure";
947 return -ENXIO;
948 }
949
950 ms->mirror[mirror].offset = offset;
951
952 return 0;
953}
954
955static int add_mirror_set(struct mirror_set *ms)
956{
957 down_write(&_mirror_sets_lock);
958 list_add_tail(&ms->list, &_mirror_sets);
959 up_write(&_mirror_sets_lock);
960 wake();
961
962 return 0;
963}
964
965static void del_mirror_set(struct mirror_set *ms)
966{
967 down_write(&_mirror_sets_lock);
968 list_del(&ms->list);
969 up_write(&_mirror_sets_lock);
970}
971
972/*
973 * Create dirty log: log_type #log_params <log_params>
974 */
975static struct dirty_log *create_dirty_log(struct dm_target *ti,
976 unsigned int argc, char **argv,
977 unsigned int *args_used)
978{
979 unsigned int param_count;
980 struct dirty_log *dl;
981
982 if (argc < 2) {
983 ti->error = "dm-mirror: Insufficient mirror log arguments";
984 return NULL;
985 }
986
987 if (sscanf(argv[1], "%u", &param_count) != 1) {
988 ti->error = "dm-mirror: Invalid mirror log argument count";
989 return NULL;
990 }
991
992 *args_used = 2 + param_count;
993
994 if (argc < *args_used) {
995 ti->error = "dm-mirror: Insufficient mirror log arguments";
996 return NULL;
997 }
998
999 dl = dm_create_dirty_log(argv[0], ti, param_count, argv + 2);
1000 if (!dl) {
1001 ti->error = "dm-mirror: Error creating mirror dirty log";
1002 return NULL;
1003 }
1004
1005 if (!_check_region_size(ti, dl->type->get_region_size(dl))) {
1006 ti->error = "dm-mirror: Invalid region size";
1007 dm_destroy_dirty_log(dl);
1008 return NULL;
1009 }
1010
1011 return dl;
1012}
1013
1014/*
1015 * Construct a mirror mapping:
1016 *
1017 * log_type #log_params <log_params>
1018 * #mirrors [mirror_path offset]{2,}
1019 *
1020 * log_type is "core" or "disk"
1021 * #log_params is between 1 and 3
1022 */
1023#define DM_IO_PAGES 64
1024static int mirror_ctr(struct dm_target *ti, unsigned int argc, char **argv)
1025{
1026 int r;
1027 unsigned int nr_mirrors, m, args_used;
1028 struct mirror_set *ms;
1029 struct dirty_log *dl;
1030
1031 dl = create_dirty_log(ti, argc, argv, &args_used);
1032 if (!dl)
1033 return -EINVAL;
1034
1035 argv += args_used;
1036 argc -= args_used;
1037
1038 if (!argc || sscanf(argv[0], "%u", &nr_mirrors) != 1 ||
1039 nr_mirrors < 2 || nr_mirrors > KCOPYD_MAX_REGIONS + 1) {
1040 ti->error = "dm-mirror: Invalid number of mirrors";
1041 dm_destroy_dirty_log(dl);
1042 return -EINVAL;
1043 }
1044
1045 argv++, argc--;
1046
1047 if (argc != nr_mirrors * 2) {
1048 ti->error = "dm-mirror: Wrong number of mirror arguments";
1049 dm_destroy_dirty_log(dl);
1050 return -EINVAL;
1051 }
1052
1053 ms = alloc_context(nr_mirrors, dl->type->get_region_size(dl), ti, dl);
1054 if (!ms) {
1055 dm_destroy_dirty_log(dl);
1056 return -ENOMEM;
1057 }
1058
1059 /* Get the mirror parameter sets */
1060 for (m = 0; m < nr_mirrors; m++) {
1061 r = get_mirror(ms, ti, m, argv);
1062 if (r) {
1063 free_context(ms, ti, m);
1064 return r;
1065 }
1066 argv += 2;
1067 argc -= 2;
1068 }
1069
1070 ti->private = ms;
Alasdair G Kergond88854f2005-07-07 17:59:34 -07001071 ti->split_io = ms->rh.region_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072
1073 r = kcopyd_client_create(DM_IO_PAGES, &ms->kcopyd_client);
1074 if (r) {
1075 free_context(ms, ti, ms->nr_mirrors);
1076 return r;
1077 }
1078
1079 add_mirror_set(ms);
1080 return 0;
1081}
1082
1083static void mirror_dtr(struct dm_target *ti)
1084{
1085 struct mirror_set *ms = (struct mirror_set *) ti->private;
1086
1087 del_mirror_set(ms);
1088 kcopyd_client_destroy(ms->kcopyd_client);
1089 free_context(ms, ti, ms->nr_mirrors);
1090}
1091
1092static void queue_bio(struct mirror_set *ms, struct bio *bio, int rw)
1093{
1094 int should_wake = 0;
1095 struct bio_list *bl;
1096
1097 bl = (rw == WRITE) ? &ms->writes : &ms->reads;
1098 spin_lock(&ms->lock);
1099 should_wake = !(bl->head);
1100 bio_list_add(bl, bio);
1101 spin_unlock(&ms->lock);
1102
1103 if (should_wake)
1104 wake();
1105}
1106
1107/*
1108 * Mirror mapping function
1109 */
1110static int mirror_map(struct dm_target *ti, struct bio *bio,
1111 union map_info *map_context)
1112{
1113 int r, rw = bio_rw(bio);
1114 struct mirror *m;
1115 struct mirror_set *ms = ti->private;
1116
Neil Browne4c8b3b2006-06-26 00:27:26 -07001117 map_context->ll = bio_to_region(&ms->rh, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118
1119 if (rw == WRITE) {
1120 queue_bio(ms, bio, rw);
1121 return 0;
1122 }
1123
1124 r = ms->rh.log->type->in_sync(ms->rh.log,
1125 bio_to_region(&ms->rh, bio), 0);
1126 if (r < 0 && r != -EWOULDBLOCK)
1127 return r;
1128
1129 if (r == -EWOULDBLOCK) /* FIXME: ugly */
1130 r = 0;
1131
1132 /*
1133 * We don't want to fast track a recovery just for a read
1134 * ahead. So we just let it silently fail.
1135 * FIXME: get rid of this.
1136 */
1137 if (!r && rw == READA)
1138 return -EIO;
1139
1140 if (!r) {
1141 /* Pass this io over to the daemon */
1142 queue_bio(ms, bio, rw);
1143 return 0;
1144 }
1145
1146 m = choose_mirror(ms, bio->bi_sector);
1147 if (!m)
1148 return -EIO;
1149
1150 map_bio(ms, m, bio);
1151 return 1;
1152}
1153
1154static int mirror_end_io(struct dm_target *ti, struct bio *bio,
1155 int error, union map_info *map_context)
1156{
1157 int rw = bio_rw(bio);
1158 struct mirror_set *ms = (struct mirror_set *) ti->private;
1159 region_t region = map_context->ll;
1160
1161 /*
1162 * We need to dec pending if this was a write.
1163 */
1164 if (rw == WRITE)
1165 rh_dec(&ms->rh, region);
1166
1167 return 0;
1168}
1169
1170static void mirror_postsuspend(struct dm_target *ti)
1171{
1172 struct mirror_set *ms = (struct mirror_set *) ti->private;
1173 struct dirty_log *log = ms->rh.log;
1174
1175 rh_stop_recovery(&ms->rh);
1176 if (log->type->suspend && log->type->suspend(log))
1177 /* FIXME: need better error handling */
1178 DMWARN("log suspend failed");
1179}
1180
1181static void mirror_resume(struct dm_target *ti)
1182{
1183 struct mirror_set *ms = (struct mirror_set *) ti->private;
1184 struct dirty_log *log = ms->rh.log;
1185 if (log->type->resume && log->type->resume(log))
1186 /* FIXME: need better error handling */
1187 DMWARN("log resume failed");
1188 rh_start_recovery(&ms->rh);
1189}
1190
1191static int mirror_status(struct dm_target *ti, status_type_t type,
1192 char *result, unsigned int maxlen)
1193{
1194 unsigned int m, sz;
1195 struct mirror_set *ms = (struct mirror_set *) ti->private;
1196
1197 sz = ms->rh.log->type->status(ms->rh.log, type, result, maxlen);
1198
1199 switch (type) {
1200 case STATUSTYPE_INFO:
1201 DMEMIT("%d ", ms->nr_mirrors);
1202 for (m = 0; m < ms->nr_mirrors; m++)
1203 DMEMIT("%s ", ms->mirror[m].dev->name);
1204
Andrew Morton4ee218c2006-03-27 01:17:48 -08001205 DMEMIT("%llu/%llu",
1206 (unsigned long long)ms->rh.log->type->
1207 get_sync_count(ms->rh.log),
1208 (unsigned long long)ms->nr_regions);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209 break;
1210
1211 case STATUSTYPE_TABLE:
1212 DMEMIT("%d ", ms->nr_mirrors);
1213 for (m = 0; m < ms->nr_mirrors; m++)
Andrew Morton4ee218c2006-03-27 01:17:48 -08001214 DMEMIT("%s %llu ", ms->mirror[m].dev->name,
1215 (unsigned long long)ms->mirror[m].offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 }
1217
1218 return 0;
1219}
1220
1221static struct target_type mirror_target = {
1222 .name = "mirror",
Kevin Corry702ca6f2006-06-26 00:27:28 -07001223 .version = {1, 0, 2},
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 .module = THIS_MODULE,
1225 .ctr = mirror_ctr,
1226 .dtr = mirror_dtr,
1227 .map = mirror_map,
1228 .end_io = mirror_end_io,
1229 .postsuspend = mirror_postsuspend,
1230 .resume = mirror_resume,
1231 .status = mirror_status,
1232};
1233
1234static int __init dm_mirror_init(void)
1235{
1236 int r;
1237
1238 r = dm_dirty_log_init();
1239 if (r)
1240 return r;
1241
Alasdair G Kergon48f1f532005-08-04 12:53:37 -07001242 _kmirrord_wq = create_singlethread_workqueue("kmirrord");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243 if (!_kmirrord_wq) {
1244 DMERR("couldn't start kmirrord");
1245 dm_dirty_log_exit();
1246 return r;
1247 }
1248 INIT_WORK(&_kmirrord_work, do_work, NULL);
1249
1250 r = dm_register_target(&mirror_target);
1251 if (r < 0) {
1252 DMERR("%s: Failed to register mirror target",
1253 mirror_target.name);
1254 dm_dirty_log_exit();
1255 destroy_workqueue(_kmirrord_wq);
1256 }
1257
1258 return r;
1259}
1260
1261static void __exit dm_mirror_exit(void)
1262{
1263 int r;
1264
1265 r = dm_unregister_target(&mirror_target);
1266 if (r < 0)
1267 DMERR("%s: unregister failed %d", mirror_target.name, r);
1268
1269 destroy_workqueue(_kmirrord_wq);
1270 dm_dirty_log_exit();
1271}
1272
1273/* Module hooks */
1274module_init(dm_mirror_init);
1275module_exit(dm_mirror_exit);
1276
1277MODULE_DESCRIPTION(DM_NAME " mirror target");
1278MODULE_AUTHOR("Joe Thornber");
1279MODULE_LICENSE("GPL");