aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
Diffstat
-rw-r--r--drivers/md/Makefile2
-rw-r--r--drivers/md/dm-raid1.c789
-rw-r--r--drivers/md/dm-region-hash.c704
-rw-r--r--include/linux/dm-region-hash.h104
4 files changed, 912 insertions, 687 deletions
diff --git a/drivers/md/Makefile b/drivers/md/Makefile
index f1ef33dfd8cf..1c615804ea76 100644
--- a/drivers/md/Makefile
+++ b/drivers/md/Makefile
@@ -34,7 +34,7 @@ obj-$(CONFIG_DM_CRYPT) += dm-crypt.o
34obj-$(CONFIG_DM_DELAY) += dm-delay.o 34obj-$(CONFIG_DM_DELAY) += dm-delay.o
35obj-$(CONFIG_DM_MULTIPATH) += dm-multipath.o dm-round-robin.o 35obj-$(CONFIG_DM_MULTIPATH) += dm-multipath.o dm-round-robin.o
36obj-$(CONFIG_DM_SNAPSHOT) += dm-snapshot.o 36obj-$(CONFIG_DM_SNAPSHOT) += dm-snapshot.o
37obj-$(CONFIG_DM_MIRROR) += dm-mirror.o dm-log.o 37obj-$(CONFIG_DM_MIRROR) += dm-mirror.o dm-log.o dm-region-hash.o
38obj-$(CONFIG_DM_ZERO) += dm-zero.o 38obj-$(CONFIG_DM_ZERO) += dm-zero.o
39 39
40quiet_cmd_unroll = UNROLL $@ 40quiet_cmd_unroll = UNROLL $@
diff --git a/drivers/md/dm-raid1.c b/drivers/md/dm-raid1.c
index f358853af5cf..92dcc06832a4 100644
--- a/drivers/md/dm-raid1.c
+++ b/drivers/md/dm-raid1.c
@@ -1,31 +1,30 @@
1/* 1/*
2 * Copyright (C) 2003 Sistina Software Limited. 2 * Copyright (C) 2003 Sistina Software Limited.
3 * Copyright (C) 2005-2008 Red Hat, Inc. All rights reserved.
3 * 4 *
4 * This file is released under the GPL. 5 * This file is released under the GPL.
5 */ 6 */
6 7
7#include <linux/device-mapper.h>
8
9#include "dm-bio-list.h" 8#include "dm-bio-list.h"
10#include "dm-bio-record.h" 9#include "dm-bio-record.h"
11 10
12#include <linux/ctype.h>
13#include <linux/init.h> 11#include <linux/init.h>
14#include <linux/mempool.h> 12#include <linux/mempool.h>
15#include <linux/module.h> 13#include <linux/module.h>
16#include <linux/pagemap.h> 14#include <linux/pagemap.h>
17#include <linux/slab.h> 15#include <linux/slab.h>
18#include <linux/time.h>
19#include <linux/vmalloc.h>
20#include <linux/workqueue.h> 16#include <linux/workqueue.h>
21#include <linux/log2.h> 17#include <linux/device-mapper.h>
22#include <linux/hardirq.h>
23#include <linux/dm-io.h> 18#include <linux/dm-io.h>
24#include <linux/dm-dirty-log.h> 19#include <linux/dm-dirty-log.h>
25#include <linux/dm-kcopyd.h> 20#include <linux/dm-kcopyd.h>
21#include <linux/dm-region-hash.h>
26 22
27#define DM_MSG_PREFIX "raid1" 23#define DM_MSG_PREFIX "raid1"
24
25#define MAX_RECOVERY 1 /* Maximum number of regions recovered in parallel. */
28#define DM_IO_PAGES 64 26#define DM_IO_PAGES 64
27#define DM_KCOPYD_PAGES 64
29 28
30#define DM_RAID1_HANDLE_ERRORS 0x01 29#define DM_RAID1_HANDLE_ERRORS 0x01
31#define errors_handled(p) ((p)->features & DM_RAID1_HANDLE_ERRORS) 30#define errors_handled(p) ((p)->features & DM_RAID1_HANDLE_ERRORS)
@@ -33,87 +32,6 @@
33static DECLARE_WAIT_QUEUE_HEAD(_kmirrord_recovery_stopped); 32static DECLARE_WAIT_QUEUE_HEAD(_kmirrord_recovery_stopped);
34 33
35/*----------------------------------------------------------------- 34/*-----------------------------------------------------------------
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 *---------------------------------------------------------------*/
71struct mirror_set;
72struct region_hash {
73 struct mirror_set *ms;
74 uint32_t region_size;
75 unsigned region_shift;
76
77 /* holds persistent region state */
78 struct dm_dirty_log *log;
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;
88 atomic_t recovery_in_flight;
89 struct semaphore recovery_count;
90 struct list_head clean_regions;
91 struct list_head quiesced_regions;
92 struct list_head recovered_regions;
93 struct list_head failed_recovered_regions;
94};
95
96enum {
97 RH_CLEAN,
98 RH_DIRTY,
99 RH_NOSYNC,
100 RH_RECOVERING
101};
102
103struct 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
115
116/*-----------------------------------------------------------------
117 * Mirror set structures. 35 * Mirror set structures.
118 *---------------------------------------------------------------*/ 36 *---------------------------------------------------------------*/
119enum dm_raid1_error { 37enum dm_raid1_error {
@@ -133,8 +51,7 @@ struct mirror {
133struct mirror_set { 51struct mirror_set {
134 struct dm_target *ti; 52 struct dm_target *ti;
135 struct list_head list; 53 struct list_head list;
136 struct region_hash rh; 54
137 struct dm_kcopyd_client *kcopyd_client;
138 uint64_t features; 55 uint64_t features;
139 56
140 spinlock_t lock; /* protects the lists */ 57 spinlock_t lock; /* protects the lists */
@@ -142,6 +59,8 @@ struct mirror_set {
142 struct bio_list writes; 59 struct bio_list writes;
143 struct bio_list failures; 60 struct bio_list failures;
144 61
62 struct dm_region_hash *rh;
63 struct dm_kcopyd_client *kcopyd_client;
145 struct dm_io_client *io_client; 64 struct dm_io_client *io_client;
146 mempool_t *read_record_pool; 65 mempool_t *read_record_pool;
147 66
@@ -160,25 +79,14 @@ struct mirror_set {
160 79
161 struct work_struct trigger_event; 80 struct work_struct trigger_event;
162 81
163 unsigned int nr_mirrors; 82 unsigned nr_mirrors;
164 struct mirror mirror[0]; 83 struct mirror mirror[0];
165}; 84};
166 85
167/* 86static void wakeup_mirrord(void *context)
168 * Conversion fns
169 */
170static inline region_t bio_to_region(struct region_hash *rh, struct bio *bio)
171{ 87{
172 return (bio->bi_sector - rh->ms->ti->begin) >> rh->region_shift; 88 struct mirror_set *ms = context;
173}
174 89
175static inline sector_t region_to_sector(struct region_hash *rh, region_t region)
176{
177 return region << rh->region_shift;
178}
179
180static void wake(struct mirror_set *ms)
181{
182 queue_work(ms->kmirrord_wq, &ms->kmirrord_work); 90 queue_work(ms->kmirrord_wq, &ms->kmirrord_work);
183} 91}
184 92
@@ -187,7 +95,7 @@ static void delayed_wake_fn(unsigned long data)
187 struct mirror_set *ms = (struct mirror_set *) data; 95 struct mirror_set *ms = (struct mirror_set *) data;
188 96
189 clear_bit(0, &ms->timer_pending); 97 clear_bit(0, &ms->timer_pending);
190 wake(ms); 98 wakeup_mirrord(ms);
191} 99}
192 100
193static void delayed_wake(struct mirror_set *ms) 101static void delayed_wake(struct mirror_set *ms)
@@ -201,473 +109,34 @@ static void delayed_wake(struct mirror_set *ms)
201 add_timer(&ms->timer); 109 add_timer(&ms->timer);
202} 110}
203 111
204/* FIXME move this */ 112static void wakeup_all_recovery_waiters(void *context)
205static void queue_bio(struct mirror_set *ms, struct bio *bio, int rw);
206
207#define MIN_REGIONS 64
208#define MAX_RECOVERY 1
209static int rh_init(struct region_hash *rh, struct mirror_set *ms,
210 struct dm_dirty_log *log, uint32_t region_size,
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);
244 atomic_set(&rh->recovery_in_flight, 0);
245 INIT_LIST_HEAD(&rh->clean_regions);
246 INIT_LIST_HEAD(&rh->quiesced_regions);
247 INIT_LIST_HEAD(&rh->recovered_regions);
248 INIT_LIST_HEAD(&rh->failed_recovered_regions);
249
250 rh->region_pool = mempool_create_kmalloc_pool(MIN_REGIONS,
251 sizeof(struct region));
252 if (!rh->region_pool) {
253 vfree(rh->buckets);
254 rh->buckets = NULL;
255 return -ENOMEM;
256 }
257
258 return 0;
259}
260
261static 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(&reg->pending));
270 mempool_free(reg, rh->region_pool);
271 }
272 }
273
274 if (rh->log)
275 dm_dirty_log_destroy(rh->log);
276 if (rh->region_pool)
277 mempool_destroy(rh->region_pool);
278 vfree(rh->buckets);
279}
280
281#define RH_HASH_MULT 2654435387U
282
283static 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
288static 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
299static void __rh_insert(struct region_hash *rh, struct region *reg)
300{
301 unsigned int h = rh_hash(rh, reg->key);
302 list_add(&reg->hash_list, rh->buckets + h);
303}
304
305static struct region *__rh_alloc(struct region_hash *rh, region_t region)
306{
307 struct region *reg, *nreg;
308
309 read_unlock(&rh->hash_lock);
310 nreg = mempool_alloc(rh->region_pool, GFP_ATOMIC);
311 if (unlikely(!nreg))
312 nreg = kmalloc(sizeof(struct region), GFP_NOIO);
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
344static 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
355static 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
380static 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
387static 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
396static 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);
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 */
411 dispatch_bios(rh->ms, &reg->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
417static void rh_update_states(struct region_hash *rh)
418{
419 struct region *reg, *next;
420
421 LIST_HEAD(clean);
422 LIST_HEAD(recovered);
423 LIST_HEAD(failed_recovered);
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)) {
431 list_splice_init(&rh->clean_regions, &clean);
432
433 list_for_each_entry(reg, &clean, list)
434 list_del(&reg->hash_list);
435 }
436
437 if (!list_empty(&rh->recovered_regions)) {
438 list_splice_init(&rh->recovered_regions, &recovered);
439
440 list_for_each_entry (reg, &recovered, list)
441 list_del(&reg->hash_list);
442 }
443
444 if (!list_empty(&rh->failed_recovered_regions)) {
445 list_splice_init(&rh->failed_recovered_regions,
446 &failed_recovered);
447
448 list_for_each_entry(reg, &failed_recovered, list)
449 list_del(&reg->hash_list);
450 }
451
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);
462 complete_resync_work(reg, 1);
463 mempool_free(reg, rh->region_pool);
464 }
465
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
471 list_for_each_entry_safe(reg, next, &clean, list) {
472 rh->log->type->clear_region(rh->log, reg->key);
473 mempool_free(reg, rh->region_pool);
474 }
475
476 rh->log->type->flush(rh->log);
477}
478
479static 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);
485
486 spin_lock_irq(&rh->region_lock);
487 atomic_inc(&reg->pending);
488
489 if (reg->state == RH_CLEAN) {
490 reg->state = RH_DIRTY;
491 list_del_init(&reg->list); /* take off the clean list */
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
498
499 read_unlock(&rh->hash_lock);
500}
501
502static void rh_inc_pending(struct region_hash *rh, struct bio_list *bios)
503{ 113{
504 struct bio *bio; 114 wake_up_all(&_kmirrord_recovery_stopped);
505
506 for (bio = bios->head; bio; bio = bio->bi_next)
507 rh_inc(rh, bio_to_region(rh, bio));
508} 115}
509 116
510static void rh_dec(struct region_hash *rh, region_t region) 117static void queue_bio(struct mirror_set *ms, struct bio *bio, int rw)
511{ 118{
512 unsigned long flags; 119 unsigned long flags;
513 struct region *reg;
514 int should_wake = 0; 120 int should_wake = 0;
121 struct bio_list *bl;
515 122
516 read_lock(&rh->hash_lock); 123 bl = (rw == WRITE) ? &ms->writes : &ms->reads;
517 reg = __rh_lookup(rh, region); 124 spin_lock_irqsave(&ms->lock, flags);
518 read_unlock(&rh->hash_lock); 125 should_wake = !(bl->head);
519 126 bio_list_add(bl, bio);
520 spin_lock_irqsave(&rh->region_lock, flags); 127 spin_unlock_irqrestore(&ms->lock, flags);
521 if (atomic_dec_and_test(&reg->pending)) {
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 */
534 if (reg->state == RH_RECOVERING) {
535 list_add_tail(&reg->list, &rh->quiesced_regions);
536 } else if (reg->state == RH_DIRTY) {
537 reg->state = RH_CLEAN;
538 list_add(&reg->list, &rh->clean_regions);
539 }
540 should_wake = 1;
541 }
542 spin_unlock_irqrestore(&rh->region_lock, flags);
543 128
544 if (should_wake) 129 if (should_wake)
545 wake(rh->ms); 130 wakeup_mirrord(ms);
546} 131}
547 132
548/* 133static void dispatch_bios(void *context, struct bio_list *bio_list)
549 * Starts quiescing a region in preparation for recovery.
550 */
551static int __rh_recovery_prepare(struct region_hash *rh)
552{ 134{
553 int r; 135 struct mirror_set *ms = context;
554 struct region *reg; 136 struct bio *bio;
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, &region);
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(&reg->pending))
577 list_del_init(&reg->list);
578 else
579 list_move(&reg->list, &rh->quiesced_regions);
580
581 spin_unlock_irq(&rh->region_lock);
582
583 return 1;
584}
585
586static void rh_recovery_prepare(struct region_hash *rh)
587{
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);
593 if (__rh_recovery_prepare(rh) <= 0) {
594 atomic_dec(&rh->recovery_in_flight);
595 up(&rh->recovery_count);
596 break;
597 }
598 }
599
600 /* Drop the extra reference */
601 if (atomic_dec_and_test(&rh->recovery_in_flight))
602 wake_up_all(&_kmirrord_recovery_stopped);
603}
604
605/*
606 * Returns any quiesced regions.
607 */
608static 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(&reg->list); /* remove from the quiesced list */
617 }
618 spin_unlock_irq(&rh->region_lock);
619
620 return reg;
621}
622
623static 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);
628 if (success)
629 list_add(&reg->list, &reg->rh->recovered_regions);
630 else {
631 reg->state = RH_NOSYNC;
632 list_add(&reg->list, &reg->rh->failed_recovered_regions);
633 }
634 spin_unlock_irq(&rh->region_lock);
635
636 wake(rh->ms);
637}
638
639static int rh_flush(struct region_hash *rh)
640{
641 return rh->log->type->flush(rh->log);
642}
643
644static 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(&reg->delayed_bios, bio);
651 read_unlock(&rh->hash_lock);
652}
653
654static 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
663static 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);
669 137
670 wake(rh->ms); 138 while ((bio = bio_list_pop(bio_list)))
139 queue_bio(ms, bio, WRITE);
671} 140}
672 141
673#define MIN_READ_RECORDS 20 142#define MIN_READ_RECORDS 20
@@ -777,8 +246,8 @@ out:
777static void recovery_complete(int read_err, unsigned long write_err, 246static void recovery_complete(int read_err, unsigned long write_err,
778 void *context) 247 void *context)
779{ 248{
780 struct region *reg = (struct region *)context; 249 struct dm_region *reg = context;
781 struct mirror_set *ms = reg->rh->ms; 250 struct mirror_set *ms = dm_rh_region_context(reg);
782 int m, bit = 0; 251 int m, bit = 0;
783 252
784 if (read_err) { 253 if (read_err) {
@@ -804,31 +273,33 @@ static void recovery_complete(int read_err, unsigned long write_err,
804 } 273 }
805 } 274 }
806 275
807 rh_recovery_end(reg, !(read_err || write_err)); 276 dm_rh_recovery_end(reg, !(read_err || write_err));
808} 277}
809 278
810static int recover(struct mirror_set *ms, struct region *reg) 279static int recover(struct mirror_set *ms, struct dm_region *reg)
811{ 280{
812 int r; 281 int r;
813 unsigned int i; 282 unsigned i;
814 struct dm_io_region from, to[DM_KCOPYD_MAX_REGIONS], *dest; 283 struct dm_io_region from, to[DM_KCOPYD_MAX_REGIONS], *dest;
815 struct mirror *m; 284 struct mirror *m;
816 unsigned long flags = 0; 285 unsigned long flags = 0;
286 region_t key = dm_rh_get_region_key(reg);
287 sector_t region_size = dm_rh_get_region_size(ms->rh);
817 288
818 /* fill in the source */ 289 /* fill in the source */
819 m = get_default_mirror(ms); 290 m = get_default_mirror(ms);
820 from.bdev = m->dev->bdev; 291 from.bdev = m->dev->bdev;
821 from.sector = m->offset + region_to_sector(reg->rh, reg->key); 292 from.sector = m->offset + dm_rh_region_to_sector(ms->rh, key);
822 if (reg->key == (ms->nr_regions - 1)) { 293 if (key == (ms->nr_regions - 1)) {
823 /* 294 /*
824 * The final region may be smaller than 295 * The final region may be smaller than
825 * region_size. 296 * region_size.
826 */ 297 */
827 from.count = ms->ti->len & (reg->rh->region_size - 1); 298 from.count = ms->ti->len & (region_size - 1);
828 if (!from.count) 299 if (!from.count)
829 from.count = reg->rh->region_size; 300 from.count = region_size;
830 } else 301 } else
831 from.count = reg->rh->region_size; 302 from.count = region_size;
832 303
833 /* fill in the destinations */ 304 /* fill in the destinations */
834 for (i = 0, dest = to; i < ms->nr_mirrors; i++) { 305 for (i = 0, dest = to; i < ms->nr_mirrors; i++) {
@@ -837,7 +308,7 @@ static int recover(struct mirror_set *ms, struct region *reg)
837 308
838 m = ms->mirror + i; 309 m = ms->mirror + i;
839 dest->bdev = m->dev->bdev; 310 dest->bdev = m->dev->bdev;
840 dest->sector = m->offset + region_to_sector(reg->rh, reg->key); 311 dest->sector = m->offset + dm_rh_region_to_sector(ms->rh, key);
841 dest->count = from.count; 312 dest->count = from.count;
842 dest++; 313 dest++;
843 } 314 }
@@ -854,22 +325,22 @@ static int recover(struct mirror_set *ms, struct region *reg)
854 325
855static void do_recovery(struct mirror_set *ms) 326static void do_recovery(struct mirror_set *ms)
856{ 327{
328 struct dm_region *reg;
329 struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
857 int r; 330 int r;
858 struct region *reg;
859 struct dm_dirty_log *log = ms->rh.log;
860 331
861 /* 332 /*
862 * Start quiescing some regions. 333 * Start quiescing some regions.
863 */ 334 */
864 rh_recovery_prepare(&ms->rh); 335 dm_rh_recovery_prepare(ms->rh);
865 336
866 /* 337 /*
867 * Copy any already quiesced regions. 338 * Copy any already quiesced regions.
868 */ 339 */
869 while ((reg = rh_recovery_start(&ms->rh))) { 340 while ((reg = dm_rh_recovery_start(ms->rh))) {
870 r = recover(ms, reg); 341 r = recover(ms, reg);
871 if (r) 342 if (r)
872 rh_recovery_end(reg, 0); 343 dm_rh_recovery_end(reg, 0);
873 } 344 }
874 345
875 /* 346 /*
@@ -910,9 +381,10 @@ static int default_ok(struct mirror *m)
910 381
911static int mirror_available(struct mirror_set *ms, struct bio *bio) 382static int mirror_available(struct mirror_set *ms, struct bio *bio)
912{ 383{
913 region_t region = bio_to_region(&ms->rh, bio); 384 struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
385 region_t region = dm_rh_bio_to_region(ms->rh, bio);
914 386
915 if (ms->rh.log->type->in_sync(ms->rh.log, region, 0)) 387 if (log->type->in_sync(log, region, 0))
916 return choose_mirror(ms, bio->bi_sector) ? 1 : 0; 388 return choose_mirror(ms, bio->bi_sector) ? 1 : 0;
917 389
918 return 0; 390 return 0;
@@ -986,7 +458,14 @@ static void read_async_bio(struct mirror *m, struct bio *bio)
986 458
987 map_region(&io, m, bio); 459 map_region(&io, m, bio);
988 bio_set_m(bio, m); 460 bio_set_m(bio, m);
989 (void) dm_io(&io_req, 1, &io, NULL); 461 BUG_ON(dm_io(&io_req, 1, &io, NULL));
462}
463
464static inline int region_in_sync(struct mirror_set *ms, region_t region,
465 int may_block)
466{
467 int state = dm_rh_get_state(ms->rh, region, may_block);
468 return state == DM_RH_CLEAN || state == DM_RH_DIRTY;
990} 469}
991 470
992static void do_reads(struct mirror_set *ms, struct bio_list *reads) 471static void do_reads(struct mirror_set *ms, struct bio_list *reads)
@@ -996,13 +475,13 @@ static void do_reads(struct mirror_set *ms, struct bio_list *reads)
996 struct mirror *m; 475 struct mirror *m;
997 476
998 while ((bio = bio_list_pop(reads))) { 477 while ((bio = bio_list_pop(reads))) {
999 region = bio_to_region(&ms->rh, bio); 478 region = dm_rh_bio_to_region(ms->rh, bio);
1000 m = get_default_mirror(ms); 479 m = get_default_mirror(ms);
1001 480
1002 /* 481 /*
1003 * We can only read balance if the region is in sync. 482 * We can only read balance if the region is in sync.
1004 */ 483 */
1005 if (likely(rh_in_sync(&ms->rh, region, 1))) 484 if (likely(region_in_sync(ms, region, 1)))
1006 m = choose_mirror(ms, bio->bi_sector); 485 m = choose_mirror(ms, bio->bi_sector);
1007 else if (m && atomic_read(&m->error_count)) 486 else if (m && atomic_read(&m->error_count))
1008 m = NULL; 487 m = NULL;
@@ -1025,57 +504,6 @@ static void do_reads(struct mirror_set *ms, struct bio_list *reads)
1025 * NOSYNC: increment pending, just write to the default mirror 504 * NOSYNC: increment pending, just write to the default mirror
1026 *---------------------------------------------------------------*/ 505 *---------------------------------------------------------------*/
1027 506
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 */
1040static 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;
1045 struct dm_dirty_log *log = ms->rh.log;
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(&reg->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(&reg->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 507
1080static void write_callback(unsigned long error, void *context) 508static void write_callback(unsigned long error, void *context)
1081{ 509{
@@ -1120,7 +548,7 @@ static void write_callback(unsigned long error, void *context)
1120 bio_list_add(&ms->failures, bio); 548 bio_list_add(&ms->failures, bio);
1121 spin_unlock_irqrestore(&ms->lock, flags); 549 spin_unlock_irqrestore(&ms->lock, flags);
1122 if (should_wake) 550 if (should_wake)
1123 wake(ms); 551 wakeup_mirrord(ms);
1124 return; 552 return;
1125 } 553 }
1126out: 554out:
@@ -1150,7 +578,7 @@ static void do_write(struct mirror_set *ms, struct bio *bio)
1150 */ 578 */
1151 bio_set_m(bio, get_default_mirror(ms)); 579 bio_set_m(bio, get_default_mirror(ms));
1152 580
1153 (void) dm_io(&io_req, ms->nr_mirrors, io, NULL); 581 BUG_ON(dm_io(&io_req, ms->nr_mirrors, io, NULL));
1154} 582}
1155 583
1156static void do_writes(struct mirror_set *ms, struct bio_list *writes) 584static void do_writes(struct mirror_set *ms, struct bio_list *writes)
@@ -1170,18 +598,19 @@ static void do_writes(struct mirror_set *ms, struct bio_list *writes)
1170 bio_list_init(&recover); 598 bio_list_init(&recover);
1171 599
1172 while ((bio = bio_list_pop(writes))) { 600 while ((bio = bio_list_pop(writes))) {
1173 state = rh_state(&ms->rh, bio_to_region(&ms->rh, bio), 1); 601 state = dm_rh_get_state(ms->rh,
602 dm_rh_bio_to_region(ms->rh, bio), 1);
1174 switch (state) { 603 switch (state) {
1175 case RH_CLEAN: 604 case DM_RH_CLEAN:
1176 case RH_DIRTY: 605 case DM_RH_DIRTY:
1177 this_list = &sync; 606 this_list = &sync;
1178 break; 607 break;
1179 608
1180 case RH_NOSYNC: 609 case DM_RH_NOSYNC:
1181 this_list = &nosync; 610 this_list = &nosync;
1182 break; 611 break;
1183 612
1184 case RH_RECOVERING: 613 case DM_RH_RECOVERING:
1185 this_list = &recover; 614 this_list = &recover;
1186 break; 615 break;
1187 } 616 }
@@ -1194,9 +623,9 @@ static void do_writes(struct mirror_set *ms, struct bio_list *writes)
1194 * be written to (writes to recover regions are going to 623 * be written to (writes to recover regions are going to
1195 * be delayed). 624 * be delayed).
1196 */ 625 */
1197 rh_inc_pending(&ms->rh, &sync); 626 dm_rh_inc_pending(ms->rh, &sync);
1198 rh_inc_pending(&ms->rh, &nosync); 627 dm_rh_inc_pending(ms->rh, &nosync);
1199 ms->log_failure = rh_flush(&ms->rh) ? 1 : 0; 628 ms->log_failure = dm_rh_flush(ms->rh) ? 1 : 0;
1200 629
1201 /* 630 /*
1202 * Dispatch io. 631 * Dispatch io.
@@ -1205,13 +634,13 @@ static void do_writes(struct mirror_set *ms, struct bio_list *writes)
1205 spin_lock_irq(&ms->lock); 634 spin_lock_irq(&ms->lock);
1206 bio_list_merge(&ms->failures, &sync); 635 bio_list_merge(&ms->failures, &sync);
1207 spin_unlock_irq(&ms->lock); 636 spin_unlock_irq(&ms->lock);
1208 wake(ms); 637 wakeup_mirrord(ms);
1209 } else 638 } else
1210 while ((bio = bio_list_pop(&sync))) 639 while ((bio = bio_list_pop(&sync)))
1211 do_write(ms, bio); 640 do_write(ms, bio);
1212 641
1213 while ((bio = bio_list_pop(&recover))) 642 while ((bio = bio_list_pop(&recover)))
1214 rh_delay(&ms->rh, bio); 643 dm_rh_delay(ms->rh, bio);
1215 644
1216 while ((bio = bio_list_pop(&nosync))) { 645 while ((bio = bio_list_pop(&nosync))) {
1217 map_bio(get_default_mirror(ms), bio); 646 map_bio(get_default_mirror(ms), bio);
@@ -1228,7 +657,8 @@ static void do_failures(struct mirror_set *ms, struct bio_list *failures)
1228 657
1229 if (!ms->log_failure) { 658 if (!ms->log_failure) {
1230 while ((bio = bio_list_pop(failures))) 659 while ((bio = bio_list_pop(failures)))
1231 __bio_mark_nosync(ms, bio, bio->bi_size, 0); 660 ms->in_sync = 0;
661 dm_rh_mark_nosync(ms->rh, bio, bio->bi_size, 0);
1232 return; 662 return;
1233 } 663 }
1234 664
@@ -1281,8 +711,8 @@ static void trigger_event(struct work_struct *work)
1281 *---------------------------------------------------------------*/ 711 *---------------------------------------------------------------*/
1282static void do_mirror(struct work_struct *work) 712static void do_mirror(struct work_struct *work)
1283{ 713{
1284 struct mirror_set *ms =container_of(work, struct mirror_set, 714 struct mirror_set *ms = container_of(work, struct mirror_set,
1285 kmirrord_work); 715 kmirrord_work);
1286 struct bio_list reads, writes, failures; 716 struct bio_list reads, writes, failures;
1287 unsigned long flags; 717 unsigned long flags;
1288 718
@@ -1295,7 +725,7 @@ static void do_mirror(struct work_struct *work)
1295 bio_list_init(&ms->failures); 725 bio_list_init(&ms->failures);
1296 spin_unlock_irqrestore(&ms->lock, flags); 726 spin_unlock_irqrestore(&ms->lock, flags);
1297 727
1298 rh_update_states(&ms->rh); 728 dm_rh_update_states(ms->rh, errors_handled(ms));
1299 do_recovery(ms); 729 do_recovery(ms);
1300 do_reads(ms, &reads); 730 do_reads(ms, &reads);
1301 do_writes(ms, &writes); 731 do_writes(ms, &writes);
@@ -1304,7 +734,6 @@ static void do_mirror(struct work_struct *work)
1304 dm_table_unplug_all(ms->ti->table); 734 dm_table_unplug_all(ms->ti->table);
1305} 735}
1306 736
1307
1308/*----------------------------------------------------------------- 737/*-----------------------------------------------------------------
1309 * Target functions 738 * Target functions
1310 *---------------------------------------------------------------*/ 739 *---------------------------------------------------------------*/
@@ -1351,7 +780,11 @@ static struct mirror_set *alloc_context(unsigned int nr_mirrors,
1351 return NULL; 780 return NULL;
1352 } 781 }
1353 782
1354 if (rh_init(&ms->rh, ms, dl, region_size, ms->nr_regions)) { 783 ms->rh = dm_region_hash_create(ms, dispatch_bios, wakeup_mirrord,
784 wakeup_all_recovery_waiters,
785 ms->ti->begin, MAX_RECOVERY,
786 dl, region_size, ms->nr_regions);
787 if (IS_ERR(ms->rh)) {
1355 ti->error = "Error creating dirty region hash"; 788 ti->error = "Error creating dirty region hash";
1356 dm_io_client_destroy(ms->io_client); 789 dm_io_client_destroy(ms->io_client);
1357 mempool_destroy(ms->read_record_pool); 790 mempool_destroy(ms->read_record_pool);
@@ -1369,7 +802,7 @@ static void free_context(struct mirror_set *ms, struct dm_target *ti,
1369 dm_put_device(ti, ms->mirror[m].dev); 802 dm_put_device(ti, ms->mirror[m].dev);
1370 803
1371 dm_io_client_destroy(ms->io_client); 804 dm_io_client_destroy(ms->io_client);
1372 rh_exit(&ms->rh); 805 dm_region_hash_destroy(ms->rh);
1373 mempool_destroy(ms->read_record_pool); 806 mempool_destroy(ms->read_record_pool);
1374 kfree(ms); 807 kfree(ms);
1375} 808}
@@ -1409,10 +842,10 @@ static int get_mirror(struct mirror_set *ms, struct dm_target *ti,
1409 * Create dirty log: log_type #log_params <log_params> 842 * Create dirty log: log_type #log_params <log_params>
1410 */ 843 */
1411static struct dm_dirty_log *create_dirty_log(struct dm_target *ti, 844static struct dm_dirty_log *create_dirty_log(struct dm_target *ti,
1412 unsigned int argc, char **argv, 845 unsigned argc, char **argv,
1413 unsigned int *args_used) 846 unsigned *args_used)
1414{ 847{
1415 unsigned int param_count; 848 unsigned param_count;
1416 struct dm_dirty_log *dl; 849 struct dm_dirty_log *dl;
1417 850
1418 if (argc < 2) { 851 if (argc < 2) {
@@ -1543,7 +976,7 @@ static int mirror_ctr(struct dm_target *ti, unsigned int argc, char **argv)
1543 } 976 }
1544 977
1545 ti->private = ms; 978 ti->private = ms;
1546 ti->split_io = ms->rh.region_size; 979 ti->split_io = dm_rh_get_region_size(ms->rh);
1547 980
1548 ms->kmirrord_wq = create_singlethread_workqueue("kmirrord"); 981 ms->kmirrord_wq = create_singlethread_workqueue("kmirrord");
1549 if (!ms->kmirrord_wq) { 982 if (!ms->kmirrord_wq) {
@@ -1578,11 +1011,11 @@ static int mirror_ctr(struct dm_target *ti, unsigned int argc, char **argv)
1578 goto err_destroy_wq; 1011 goto err_destroy_wq;
1579 } 1012 }
1580 1013
1581 r = dm_kcopyd_client_create(DM_IO_PAGES, &ms->kcopyd_client); 1014 r = dm_kcopyd_client_create(DM_KCOPYD_PAGES, &ms->kcopyd_client);
1582 if (r) 1015 if (r)
1583 goto err_destroy_wq; 1016 goto err_destroy_wq;
1584 1017
1585 wake(ms); 1018 wakeup_mirrord(ms);
1586 return 0; 1019 return 0;
1587 1020
1588err_destroy_wq: 1021err_destroy_wq:
@@ -1603,22 +1036,6 @@ static void mirror_dtr(struct dm_target *ti)
1603 free_context(ms, ti, ms->nr_mirrors); 1036 free_context(ms, ti, ms->nr_mirrors);
1604} 1037}
1605 1038
1606static void queue_bio(struct mirror_set *ms, struct bio *bio, int rw)
1607{
1608 unsigned long flags;
1609 int should_wake = 0;
1610 struct bio_list *bl;
1611
1612 bl = (rw == WRITE) ? &ms->writes : &ms->reads;
1613 spin_lock_irqsave(&ms->lock, flags);
1614 should_wake = !(bl->head);
1615 bio_list_add(bl, bio);
1616 spin_unlock_irqrestore(&ms->lock, flags);
1617
1618 if (should_wake)
1619 wake(ms);
1620}
1621
1622/* 1039/*
1623 * Mirror mapping function 1040 * Mirror mapping function
1624 */ 1041 */
@@ -1629,16 +1046,16 @@ static int mirror_map(struct dm_target *ti, struct bio *bio,
1629 struct mirror *m; 1046 struct mirror *m;
1630 struct mirror_set *ms = ti->private; 1047 struct mirror_set *ms = ti->private;
1631 struct dm_raid1_read_record *read_record = NULL; 1048 struct dm_raid1_read_record *read_record = NULL;
1049 struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
1632 1050
1633 if (rw == WRITE) { 1051 if (rw == WRITE) {
1634 /* Save region for mirror_end_io() handler */ 1052 /* Save region for mirror_end_io() handler */
1635 map_context->ll = bio_to_region(&ms->rh, bio); 1053 map_context->ll = dm_rh_bio_to_region(ms->rh, bio);
1636 queue_bio(ms, bio, rw); 1054 queue_bio(ms, bio, rw);
1637 return DM_MAPIO_SUBMITTED; 1055 return DM_MAPIO_SUBMITTED;
1638 } 1056 }
1639 1057
1640 r = ms->rh.log->type->in_sync(ms->rh.log, 1058 r = log->type->in_sync(log, dm_rh_bio_to_region(ms->rh, bio), 0);
1641 bio_to_region(&ms->rh, bio), 0);
1642 if (r < 0 && r != -EWOULDBLOCK) 1059 if (r < 0 && r != -EWOULDBLOCK)
1643 return r; 1060 return r;
1644 1061
@@ -1686,7 +1103,7 @@ static int mirror_end_io(struct dm_target *ti, struct bio *bio,
1686 * We need to dec pending if this was a write. 1103 * We need to dec pending if this was a write.
1687 */ 1104 */
1688 if (rw == WRITE) { 1105 if (rw == WRITE) {
1689 rh_dec(&ms->rh, map_context->ll); 1106 dm_rh_dec(ms->rh, map_context->ll);
1690 return error; 1107 return error;
1691 } 1108 }
1692 1109
@@ -1742,7 +1159,7 @@ out:
1742static void mirror_presuspend(struct dm_target *ti) 1159static void mirror_presuspend(struct dm_target *ti)
1743{ 1160{
1744 struct mirror_set *ms = (struct mirror_set *) ti->private; 1161 struct mirror_set *ms = (struct mirror_set *) ti->private;
1745 struct dm_dirty_log *log = ms->rh.log; 1162 struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
1746 1163
1747 atomic_set(&ms->suspend, 1); 1164 atomic_set(&ms->suspend, 1);
1748 1165
@@ -1750,10 +1167,10 @@ static void mirror_presuspend(struct dm_target *ti)
1750 * We must finish up all the work that we've 1167 * We must finish up all the work that we've
1751 * generated (i.e. recovery work). 1168 * generated (i.e. recovery work).
1752 */ 1169 */
1753 rh_stop_recovery(&ms->rh); 1170 dm_rh_stop_recovery(ms->rh);
1754 1171
1755 wait_event(_kmirrord_recovery_stopped, 1172 wait_event(_kmirrord_recovery_stopped,
1756 !atomic_read(&ms->rh.recovery_in_flight)); 1173 !dm_rh_recovery_in_flight(ms->rh));
1757 1174
1758 if (log->type->presuspend && log->type->presuspend(log)) 1175 if (log->type->presuspend && log->type->presuspend(log))
1759 /* FIXME: need better error handling */ 1176 /* FIXME: need better error handling */
@@ -1771,7 +1188,7 @@ static void mirror_presuspend(struct dm_target *ti)
1771static void mirror_postsuspend(struct dm_target *ti) 1188static void mirror_postsuspend(struct dm_target *ti)
1772{ 1189{
1773 struct mirror_set *ms = ti->private; 1190 struct mirror_set *ms = ti->private;
1774 struct dm_dirty_log *log = ms->rh.log; 1191 struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
1775 1192
1776 if (log->type->postsuspend && log->type->postsuspend(log)) 1193 if (log->type->postsuspend && log->type->postsuspend(log))
1777 /* FIXME: need better error handling */ 1194 /* FIXME: need better error handling */
@@ -1781,13 +1198,13 @@ static void mirror_postsuspend(struct dm_target *ti)
1781static void mirror_resume(struct dm_target *ti) 1198static void mirror_resume(struct dm_target *ti)
1782{ 1199{
1783 struct mirror_set *ms = ti->private; 1200 struct mirror_set *ms = ti->private;
1784 struct dm_dirty_log *log = ms->rh.log; 1201 struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
1785 1202
1786 atomic_set(&ms->suspend, 0); 1203 atomic_set(&ms->suspend, 0);
1787 if (log->type->resume && log->type->resume(log)) 1204 if (log->type->resume && log->type->resume(log))
1788 /* FIXME: need better error handling */ 1205 /* FIXME: need better error handling */
1789 DMWARN("log resume failed"); 1206 DMWARN("log resume failed");
1790 rh_start_recovery(&ms->rh); 1207 dm_rh_start_recovery(ms->rh);
1791} 1208}
1792 1209
1793/* 1210/*
@@ -1819,7 +1236,7 @@ static int mirror_status(struct dm_target *ti, status_type_t type,
1819{ 1236{
1820 unsigned int m, sz = 0; 1237 unsigned int m, sz = 0;
1821 struct mirror_set *ms = (struct mirror_set *) ti->private; 1238 struct mirror_set *ms = (struct mirror_set *) ti->private;
1822 struct dm_dirty_log *log = ms->rh.log; 1239 struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
1823 char buffer[ms->nr_mirrors + 1]; 1240 char buffer[ms->nr_mirrors + 1];
1824 1241
1825 switch (type) { 1242 switch (type) {
@@ -1832,15 +1249,15 @@ static int mirror_status(struct dm_target *ti, status_type_t type,
1832 buffer[m] = '\0'; 1249 buffer[m] = '\0';
1833 1250
1834 DMEMIT("%llu/%llu 1 %s ", 1251 DMEMIT("%llu/%llu 1 %s ",
1835 (unsigned long long)log->type->get_sync_count(ms->rh.log), 1252 (unsigned long long)log->type->get_sync_count(log),
1836 (unsigned long long)ms->nr_regions, buffer); 1253 (unsigned long long)ms->nr_regions, buffer);
1837 1254
1838 sz += log->type->status(ms->rh.log, type, result+sz, maxlen-sz); 1255 sz += log->type->status(log, type, result+sz, maxlen-sz);
1839 1256
1840 break; 1257 break;
1841 1258
1842 case STATUSTYPE_TABLE: 1259 case STATUSTYPE_TABLE:
1843 sz = log->type->status(ms->rh.log, type, result, maxlen); 1260 sz = log->type->status(log, type, result, maxlen);
1844 1261
1845 DMEMIT("%d", ms->nr_mirrors); 1262 DMEMIT("%d", ms->nr_mirrors);
1846 for (m = 0; m < ms->nr_mirrors; m++) 1263 for (m = 0; m < ms->nr_mirrors; m++)
diff --git a/drivers/md/dm-region-hash.c b/drivers/md/dm-region-hash.c
new file mode 100644
index 000000000000..59f8d9df9e1a
--- /dev/null
+++ b/drivers/md/dm-region-hash.c
@@ -0,0 +1,704 @@
1/*
2 * Copyright (C) 2003 Sistina Software Limited.
3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
4 *
5 * This file is released under the GPL.
6 */
7
8#include <linux/dm-dirty-log.h>
9#include <linux/dm-region-hash.h>
10
11#include <linux/ctype.h>
12#include <linux/init.h>
13#include <linux/module.h>
14#include <linux/vmalloc.h>
15
16#include "dm.h"
17#include "dm-bio-list.h"
18
19#define DM_MSG_PREFIX "region hash"
20
21/*-----------------------------------------------------------------
22 * Region hash
23 *
24 * The mirror splits itself up into discrete regions. Each
25 * region can be in one of three states: clean, dirty,
26 * nosync. There is no need to put clean regions in the hash.
27 *
28 * In addition to being present in the hash table a region _may_
29 * be present on one of three lists.
30 *
31 * clean_regions: Regions on this list have no io pending to
32 * them, they are in sync, we are no longer interested in them,
33 * they are dull. dm_rh_update_states() will remove them from the
34 * hash table.
35 *
36 * quiesced_regions: These regions have been spun down, ready
37 * for recovery. rh_recovery_start() will remove regions from
38 * this list and hand them to kmirrord, which will schedule the
39 * recovery io with kcopyd.
40 *
41 * recovered_regions: Regions that kcopyd has successfully
42 * recovered. dm_rh_update_states() will now schedule any delayed
43 * io, up the recovery_count, and remove the region from the
44 * hash.
45 *
46 * There are 2 locks:
47 * A rw spin lock 'hash_lock' protects just the hash table,
48 * this is never held in write mode from interrupt context,
49 * which I believe means that we only have to disable irqs when
50 * doing a write lock.
51 *
52 * An ordinary spin lock 'region_lock' that protects the three
53 * lists in the region_hash, with the 'state', 'list' and
54 * 'delayed_bios' fields of the regions. This is used from irq
55 * context, so all other uses will have to suspend local irqs.
56 *---------------------------------------------------------------*/
57struct dm_region_hash {
58 uint32_t region_size;
59 unsigned region_shift;
60
61 /* holds persistent region state */
62 struct dm_dirty_log *log;
63
64 /* hash table */
65 rwlock_t hash_lock;
66 mempool_t *region_pool;
67 unsigned mask;
68 unsigned nr_buckets;
69 unsigned prime;
70 unsigned shift;
71 struct list_head *buckets;
72
73 unsigned max_recovery; /* Max # of regions to recover in parallel */
74
75 spinlock_t region_lock;
76 atomic_t recovery_in_flight;
77 struct semaphore recovery_count;
78 struct list_head clean_regions;
79 struct list_head quiesced_regions;
80 struct list_head recovered_regions;
81 struct list_head failed_recovered_regions;
82
83 void *context;
84 sector_t target_begin;
85
86 /* Callback function to schedule bios writes */
87 void (*dispatch_bios)(void *context, struct bio_list *bios);
88
89 /* Callback function to wakeup callers worker thread. */
90 void (*wakeup_workers)(void *context);
91
92 /* Callback function to wakeup callers recovery waiters. */
93 void (*wakeup_all_recovery_waiters)(void *context);
94};
95
96struct dm_region {
97 struct dm_region_hash *rh; /* FIXME: can we get rid of this ? */
98 region_t key;
99 int state;
100
101 struct list_head hash_list;
102 struct list_head list;
103
104 atomic_t pending;
105 struct bio_list delayed_bios;
106};
107
108/*
109 * Conversion fns
110 */
111static region_t dm_rh_sector_to_region(struct dm_region_hash *rh, sector_t sector)
112{
113 return sector >> rh->region_shift;
114}
115
116sector_t dm_rh_region_to_sector(struct dm_region_hash *rh, region_t region)
117{
118 return region << rh->region_shift;
119}
120EXPORT_SYMBOL_GPL(dm_rh_region_to_sector);
121
122region_t dm_rh_bio_to_region(struct dm_region_hash *rh, struct bio *bio)
123{
124 return dm_rh_sector_to_region(rh, bio->bi_sector - rh->target_begin);
125}
126EXPORT_SYMBOL_GPL(dm_rh_bio_to_region);
127
128void *dm_rh_region_context(struct dm_region *reg)
129{
130 return reg->rh->context;
131}
132EXPORT_SYMBOL_GPL(dm_rh_region_context);
133
134region_t dm_rh_get_region_key(struct dm_region *reg)
135{
136 return reg->key;
137}
138EXPORT_SYMBOL_GPL(dm_rh_get_region_key);
139
140sector_t dm_rh_get_region_size(struct dm_region_hash *rh)
141{
142 return rh->region_size;
143}
144EXPORT_SYMBOL_GPL(dm_rh_get_region_size);
145
146/*
147 * FIXME: shall we pass in a structure instead of all these args to
148 * dm_region_hash_create()????
149 */
150#define RH_HASH_MULT 2654435387U
151#define RH_HASH_SHIFT 12
152
153#define MIN_REGIONS 64
154struct dm_region_hash *dm_region_hash_create(
155 void *context, void (*dispatch_bios)(void *context,
156 struct bio_list *bios),
157 void (*wakeup_workers)(void *context),
158 void (*wakeup_all_recovery_waiters)(void *context),
159 sector_t target_begin, unsigned max_recovery,
160 struct dm_dirty_log *log, uint32_t region_size,
161 region_t nr_regions)
162{
163 struct dm_region_hash *rh;
164 unsigned nr_buckets, max_buckets;
165 size_t i;
166
167 /*
168 * Calculate a suitable number of buckets for our hash
169 * table.
170 */
171 max_buckets = nr_regions >> 6;
172 for (nr_buckets = 128u; nr_buckets < max_buckets; nr_buckets <<= 1)
173 ;
174 nr_buckets >>= 1;
175
176 rh = kmalloc(sizeof(*rh), GFP_KERNEL);
177 if (!rh) {
178 DMERR("unable to allocate region hash memory");
179 return ERR_PTR(-ENOMEM);
180 }
181
182 rh->context = context;
183 rh->dispatch_bios = dispatch_bios;
184 rh->wakeup_workers = wakeup_workers;
185 rh->wakeup_all_recovery_waiters = wakeup_all_recovery_waiters;
186 rh->target_begin = target_begin;
187 rh->max_recovery = max_recovery;
188 rh->log = log;
189 rh->region_size = region_size;
190 rh->region_shift = ffs(region_size) - 1;
191 rwlock_init(&rh->hash_lock);
192 rh->mask = nr_buckets - 1;
193 rh->nr_buckets = nr_buckets;
194
195 rh->shift = RH_HASH_SHIFT;
196 rh->prime = RH_HASH_MULT;
197
198 rh->buckets = vmalloc(nr_buckets * sizeof(*rh->buckets));
199 if (!rh->buckets) {
200 DMERR("unable to allocate region hash bucket memory");
201 kfree(rh);
202 return ERR_PTR(-ENOMEM);
203 }
204
205 for (i = 0; i < nr_buckets; i++)
206 INIT_LIST_HEAD(rh->buckets + i);
207
208 spin_lock_init(&rh->region_lock);
209 sema_init(&rh->recovery_count, 0);
210 atomic_set(&rh->recovery_in_flight, 0);
211 INIT_LIST_HEAD(&rh->clean_regions);
212 INIT_LIST_HEAD(&rh->quiesced_regions);
213 INIT_LIST_HEAD(&rh->recovered_regions);
214 INIT_LIST_HEAD(&rh->failed_recovered_regions);
215
216 rh->region_pool = mempool_create_kmalloc_pool(MIN_REGIONS,
217 sizeof(struct dm_region));
218 if (!rh->region_pool) {
219 vfree(rh->buckets);
220 kfree(rh);
221 rh = ERR_PTR(-ENOMEM);
222 }
223
224 return rh;
225}
226EXPORT_SYMBOL_GPL(dm_region_hash_create);
227
228void dm_region_hash_destroy(struct dm_region_hash *rh)
229{
230 unsigned h;
231 struct dm_region *reg, *nreg;
232
233 BUG_ON(!list_empty(&rh->quiesced_regions));
234 for (h = 0; h < rh->nr_buckets; h++) {
235 list_for_each_entry_safe(reg, nreg, rh->buckets + h,
236 hash_list) {
237 BUG_ON(atomic_read(&reg->pending));
238 mempool_free(reg, rh->region_pool);
239 }
240 }
241
242 if (rh->log)
243 dm_dirty_log_destroy(rh->log);
244
245 if (rh->region_pool)
246 mempool_destroy(rh->region_pool);
247
248 vfree(rh->buckets);
249 kfree(rh);
250}
251EXPORT_SYMBOL_GPL(dm_region_hash_destroy);
252
253struct dm_dirty_log *dm_rh_dirty_log(struct dm_region_hash *rh)
254{
255 return rh->log;
256}
257EXPORT_SYMBOL_GPL(dm_rh_dirty_log);
258
259static unsigned rh_hash(struct dm_region_hash *rh, region_t region)
260{
261 return (unsigned) ((region * rh->prime) >> rh->shift) & rh->mask;
262}
263
264static struct dm_region *__rh_lookup(struct dm_region_hash *rh, region_t region)
265{
266 struct dm_region *reg;
267 struct list_head *bucket = rh->buckets + rh_hash(rh, region);
268
269 list_for_each_entry(reg, bucket, hash_list)
270 if (reg->key == region)
271 return reg;
272
273 return NULL;
274}
275
276static void __rh_insert(struct dm_region_hash *rh, struct dm_region *reg)
277{
278 list_add(&reg->hash_list, rh->buckets + rh_hash(rh, reg->key));
279}
280
281static struct dm_region *__rh_alloc(struct dm_region_hash *rh, region_t region)
282{
283 struct dm_region *reg, *nreg;
284
285 nreg = mempool_alloc(rh->region_pool, GFP_ATOMIC);
286 if (unlikely(!nreg))
287 nreg = kmalloc(sizeof(*nreg), GFP_NOIO);
288
289 nreg->state = rh->log->type->in_sync(rh->log, region, 1) ?
290 DM_RH_CLEAN : DM_RH_NOSYNC;
291 nreg->rh = rh;
292 nreg->key = region;
293 INIT_LIST_HEAD(&nreg->list);
294 atomic_set(&nreg->pending, 0);
295 bio_list_init(&nreg->delayed_bios);
296
297 write_lock_irq(&rh->hash_lock);
298 reg = __rh_lookup(rh, region);
299 if (reg)
300 /* We lost the race. */
301 mempool_free(nreg, rh->region_pool);
302 else {
303 __rh_insert(rh, nreg);
304 if (nreg->state == DM_RH_CLEAN) {
305 spin_lock(&rh->region_lock);
306 list_add(&nreg->list, &rh->clean_regions);
307 spin_unlock(&rh->region_lock);
308 }
309
310 reg = nreg;
311 }
312 write_unlock_irq(&rh->hash_lock);
313
314 return reg;
315}
316
317static struct dm_region *__rh_find(struct dm_region_hash *rh, region_t region)
318{
319 struct dm_region *reg;
320
321 reg = __rh_lookup(rh, region);
322 if (!reg) {
323 read_unlock(&rh->hash_lock);
324 reg = __rh_alloc(rh, region);
325 read_lock(&rh->hash_lock);
326 }
327
328 return reg;
329}
330
331int dm_rh_get_state(struct dm_region_hash *rh, region_t region, int may_block)
332{
333 int r;
334 struct dm_region *reg;
335
336 read_lock(&rh->hash_lock);
337 reg = __rh_lookup(rh, region);
338 read_unlock(&rh->hash_lock);
339
340 if (reg)
341 return reg->state;
342
343 /*
344 * The region wasn't in the hash, so we fall back to the
345 * dirty log.
346 */
347 r = rh->log->type->in_sync(rh->log, region, may_block);
348
349 /*
350 * Any error from the dirty log (eg. -EWOULDBLOCK) gets
351 * taken as a DM_RH_NOSYNC
352 */
353 return r == 1 ? DM_RH_CLEAN : DM_RH_NOSYNC;
354}
355EXPORT_SYMBOL_GPL(dm_rh_get_state);
356
357static void complete_resync_work(struct dm_region *reg, int success)
358{
359 struct dm_region_hash *rh = reg->rh;
360
361 rh->log->type->set_region_sync(rh->log, reg->key, success);
362
363 /*
364 * Dispatch the bios before we call 'wake_up_all'.
365 * This is important because if we are suspending,
366 * we want to know that recovery is complete and
367 * the work queue is flushed. If we wake_up_all
368 * before we dispatch_bios (queue bios and call wake()),
369 * then we risk suspending before the work queue
370 * has been properly flushed.
371 */
372 rh->dispatch_bios(rh->context, &reg->delayed_bios);
373 if (atomic_dec_and_test(&rh->recovery_in_flight))
374 rh->wakeup_all_recovery_waiters(rh->context);
375 up(&rh->recovery_count);
376}
377
378/* dm_rh_mark_nosync
379 * @ms
380 * @bio
381 * @done
382 * @error
383 *
384 * The bio was written on some mirror(s) but failed on other mirror(s).
385 * We can successfully endio the bio but should avoid the region being
386 * marked clean by setting the state DM_RH_NOSYNC.
387 *
388 * This function is _not_ safe in interrupt context!
389 */
390void dm_rh_mark_nosync(struct dm_region_hash *rh,
391 struct bio *bio, unsigned done, int error)
392{
393 unsigned long flags;
394 struct dm_dirty_log *log = rh->log;
395 struct dm_region *reg;
396 region_t region = dm_rh_bio_to_region(rh, bio);
397 int recovering = 0;
398
399 /* We must inform the log that the sync count has changed. */
400 log->type->set_region_sync(log, region, 0);
401
402 read_lock(&rh->hash_lock);
403 reg = __rh_find(rh, region);
404 read_unlock(&rh->hash_lock);
405
406 /* region hash entry should exist because write was in-flight */
407 BUG_ON(!reg);
408 BUG_ON(!list_empty(&reg->list));
409
410 spin_lock_irqsave(&rh->region_lock, flags);
411 /*
412 * Possible cases:
413 * 1) DM_RH_DIRTY
414 * 2) DM_RH_NOSYNC: was dirty, other preceeding writes failed
415 * 3) DM_RH_RECOVERING: flushing pending writes
416 * Either case, the region should have not been connected to list.
417 */
418 recovering = (reg->state == DM_RH_RECOVERING);
419 reg->state = DM_RH_NOSYNC;
420 BUG_ON(!list_empty(&reg->list));
421 spin_unlock_irqrestore(&rh->region_lock, flags);
422
423 bio_endio(bio, error);
424 if (recovering)
425 complete_resync_work(reg, 0);
426}
427EXPORT_SYMBOL_GPL(dm_rh_mark_nosync);
428
429void dm_rh_update_states(struct dm_region_hash *rh, int errors_handled)
430{
431 struct dm_region *reg, *next;
432
433 LIST_HEAD(clean);
434 LIST_HEAD(recovered);
435 LIST_HEAD(failed_recovered);
436
437 /*
438 * Quickly grab the lists.
439 */
440 write_lock_irq(&rh->hash_lock);
441 spin_lock(&rh->region_lock);
442 if (!list_empty(&rh->clean_regions)) {
443 list_splice_init(&rh->clean_regions, &clean);
444
445 list_for_each_entry(reg, &clean, list)
446 list_del(&reg->hash_list);
447 }
448
449 if (!list_empty(&rh->recovered_regions)) {
450 list_splice_init(&rh->recovered_regions, &recovered);
451
452 list_for_each_entry(reg, &recovered, list)
453 list_del(&reg->hash_list);
454 }
455
456 if (!list_empty(&rh->failed_recovered_regions)) {
457 list_splice_init(&rh->failed_recovered_regions,
458 &failed_recovered);
459
460 list_for_each_entry(reg, &failed_recovered, list)
461 list_del(&reg->hash_list);
462 }
463
464 spin_unlock(&rh->region_lock);
465 write_unlock_irq(&rh->hash_lock);
466
467 /*
468 * All the regions on the recovered and clean lists have
469 * now been pulled out of the system, so no need to do
470 * any more locking.
471 */
472 list_for_each_entry_safe(reg, next, &recovered, list) {
473 rh->log->type->clear_region(rh->log, reg->key);
474 complete_resync_work(reg, 1);
475 mempool_free(reg, rh->region_pool);
476 }
477
478 list_for_each_entry_safe(reg, next, &failed_recovered, list) {
479 complete_resync_work(reg, errors_handled ? 0 : 1);
480 mempool_free(reg, rh->region_pool);
481 }
482
483 list_for_each_entry_safe(reg, next, &clean, list) {
484 rh->log->type->clear_region(rh->log, reg->key);
485 mempool_free(reg, rh->region_pool);
486 }
487
488 rh->log->type->flush(rh->log);
489}
490EXPORT_SYMBOL_GPL(dm_rh_update_states);
491
492static void rh_inc(struct dm_region_hash *rh, region_t region)
493{
494 struct dm_region *reg;
495
496 read_lock(&rh->hash_lock);
497 reg = __rh_find(rh, region);
498
499 spin_lock_irq(&rh->region_lock);
500 atomic_inc(&reg->pending);
501
502 if (reg->state == DM_RH_CLEAN) {
503 reg->state = DM_RH_DIRTY;
504 list_del_init(&reg->list); /* take off the clean list */
505 spin_unlock_irq(&rh->region_lock);
506
507 rh->log->type->mark_region(rh->log, reg->key);
508 } else
509 spin_unlock_irq(&rh->region_lock);
510
511
512 read_unlock(&rh->hash_lock);
513}
514
515void dm_rh_inc_pending(struct dm_region_hash *rh, struct bio_list *bios)
516{
517 struct bio *bio;
518
519 for (bio = bios->head; bio; bio = bio->bi_next)
520 rh_inc(rh, dm_rh_bio_to_region(rh, bio));
521}
522EXPORT_SYMBOL_GPL(dm_rh_inc_pending);
523
524void dm_rh_dec(struct dm_region_hash *rh, region_t region)
525{
526 unsigned long flags;
527 struct dm_region *reg;
528 int should_wake = 0;
529
530 read_lock(&rh->hash_lock);
531 reg = __rh_lookup(rh, region);
532 read_unlock(&rh->hash_lock);
533
534 spin_lock_irqsave(&rh->region_lock, flags);
535 if (atomic_dec_and_test(&reg->pending)) {
536 /*
537 * There is no pending I/O for this region.
538 * We can move the region to corresponding list for next action.
539 * At this point, the region is not yet connected to any list.
540 *
541 * If the state is DM_RH_NOSYNC, the region should be kept off
542 * from clean list.
543 * The hash entry for DM_RH_NOSYNC will remain in memory
544 * until the region is recovered or the map is reloaded.
545 */
546
547 /* do nothing for DM_RH_NOSYNC */
548 if (reg->state == DM_RH_RECOVERING) {
549 list_add_tail(&reg->list, &rh->quiesced_regions);
550 } else if (reg->state == DM_RH_DIRTY) {
551 reg->state = DM_RH_CLEAN;
552 list_add(&reg->list, &rh->clean_regions);
553 }
554 should_wake = 1;
555 }
556 spin_unlock_irqrestore(&rh->region_lock, flags);
557
558 if (should_wake)
559 rh->wakeup_workers(rh->context);
560}
561EXPORT_SYMBOL_GPL(dm_rh_dec);
562
563/*
564 * Starts quiescing a region in preparation for recovery.
565 */
566static int __rh_recovery_prepare(struct dm_region_hash *rh)
567{
568 int r;
569 region_t region;
570 struct dm_region *reg;
571
572 /*
573 * Ask the dirty log what's next.
574 */
575 r = rh->log->type->get_resync_work(rh->log, &region);
576 if (r <= 0)
577 return r;
578
579 /*
580 * Get this region, and start it quiescing by setting the
581 * recovering flag.
582 */
583 read_lock(&rh->hash_lock);
584 reg = __rh_find(rh, region);
585 read_unlock(&rh->hash_lock);
586
587 spin_lock_irq(&rh->region_lock);
588 reg->state = DM_RH_RECOVERING;
589
590 /* Already quiesced ? */
591 if (atomic_read(&reg->pending))
592 list_del_init(&reg->list);
593 else
594 list_move(&reg->list, &rh->quiesced_regions);
595
596 spin_unlock_irq(&rh->region_lock);
597
598 return 1;
599}
600
601void dm_rh_recovery_prepare(struct dm_region_hash *rh)
602{
603 /* Extra reference to avoid race with dm_rh_stop_recovery */
604 atomic_inc(&rh->recovery_in_flight);
605
606 while (!down_trylock(&rh->recovery_count)) {
607 atomic_inc(&rh->recovery_in_flight);
608 if (__rh_recovery_prepare(rh) <= 0) {
609 atomic_dec(&rh->recovery_in_flight);
610 up(&rh->recovery_count);
611 break;
612 }
613 }
614
615 /* Drop the extra reference */
616 if (atomic_dec_and_test(&rh->recovery_in_flight))
617 rh->wakeup_all_recovery_waiters(rh->context);
618}
619EXPORT_SYMBOL_GPL(dm_rh_recovery_prepare);
620
621/*
622 * Returns any quiesced regions.
623 */
624struct dm_region *dm_rh_recovery_start(struct dm_region_hash *rh)
625{
626 struct dm_region *reg = NULL;
627
628 spin_lock_irq(&rh->region_lock);
629 if (!list_empty(&rh->quiesced_regions)) {
630 reg = list_entry(rh->quiesced_regions.next,
631 struct dm_region, list);
632 list_del_init(&reg->list); /* remove from the quiesced list */
633 }
634 spin_unlock_irq(&rh->region_lock);
635
636 return reg;
637}
638EXPORT_SYMBOL_GPL(dm_rh_recovery_start);
639
640void dm_rh_recovery_end(struct dm_region *reg, int success)
641{
642 struct dm_region_hash *rh = reg->rh;
643
644 spin_lock_irq(&rh->region_lock);
645 if (success)
646 list_add(&reg->list, &reg->rh->recovered_regions);
647 else {
648 reg->state = DM_RH_NOSYNC;
649 list_add(&reg->list, &reg->rh->failed_recovered_regions);
650 }
651 spin_unlock_irq(&rh->region_lock);
652
653 rh->wakeup_workers(rh->context);
654}
655EXPORT_SYMBOL_GPL(dm_rh_recovery_end);
656
657/* Return recovery in flight count. */
658int dm_rh_recovery_in_flight(struct dm_region_hash *rh)
659{
660 return atomic_read(&rh->recovery_in_flight);
661}
662EXPORT_SYMBOL_GPL(dm_rh_recovery_in_flight);
663
664int dm_rh_flush(struct dm_region_hash *rh)
665{
666 return rh->log->type->flush(rh->log);
667}
668EXPORT_SYMBOL_GPL(dm_rh_flush);
669
670void dm_rh_delay(struct dm_region_hash *rh, struct bio *bio)
671{
672 struct dm_region *reg;
673
674 read_lock(&rh->hash_lock);
675 reg = __rh_find(rh, dm_rh_bio_to_region(rh, bio));
676 bio_list_add(&reg->delayed_bios, bio);
677 read_unlock(&rh->hash_lock);
678}
679EXPORT_SYMBOL_GPL(dm_rh_delay);
680
681void dm_rh_stop_recovery(struct dm_region_hash *rh)
682{
683 int i;
684
685 /* wait for any recovering regions */
686 for (i = 0; i < rh->max_recovery; i++)
687 down(&rh->recovery_count);
688}
689EXPORT_SYMBOL_GPL(dm_rh_stop_recovery);
690
691void dm_rh_start_recovery(struct dm_region_hash *rh)
692{
693 int i;
694
695 for (i = 0; i < rh->max_recovery; i++)
696 up(&rh->recovery_count);
697
698 rh->wakeup_workers(rh->context);
699}
700EXPORT_SYMBOL_GPL(dm_rh_start_recovery);
701
702MODULE_DESCRIPTION(DM_NAME " region hash");
703MODULE_AUTHOR("Joe Thornber/Heinz Mauelshagen <dm-devel@redhat.com>");
704MODULE_LICENSE("GPL");
diff --git a/include/linux/dm-region-hash.h b/include/linux/dm-region-hash.h
new file mode 100644
index 000000000000..a9e652a41373
--- /dev/null
+++ b/include/linux/dm-region-hash.h
@@ -0,0 +1,104 @@
1/*
2 * Copyright (C) 2003 Sistina Software Limited.
3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
4 *
5 * Device-Mapper dirty region hash interface.
6 *
7 * This file is released under the GPL.
8 */
9
10#ifndef DM_REGION_HASH_H
11#define DM_REGION_HASH_H
12
13#include <linux/dm-dirty-log.h>
14
15/*-----------------------------------------------------------------
16 * Region hash
17 *----------------------------------------------------------------*/
18struct dm_region_hash;
19struct dm_region;
20
21/*
22 * States a region can have.
23 */
24enum dm_rh_region_states {
25 DM_RH_CLEAN = 0x01, /* No writes in flight. */
26 DM_RH_DIRTY = 0x02, /* Writes in flight. */
27 DM_RH_NOSYNC = 0x04, /* Out of sync. */
28 DM_RH_RECOVERING = 0x08, /* Under resynchronization. */
29};
30
31/*
32 * Region hash create/destroy.
33 */
34struct bio_list;
35struct dm_region_hash *dm_region_hash_create(
36 void *context, void (*dispatch_bios)(void *context,
37 struct bio_list *bios),
38 void (*wakeup_workers)(void *context),
39 void (*wakeup_all_recovery_waiters)(void *context),
40 sector_t target_begin, unsigned max_recovery,
41 struct dm_dirty_log *log, uint32_t region_size,
42 region_t nr_regions);
43void dm_region_hash_destroy(struct dm_region_hash *rh);
44
45struct dm_dirty_log *dm_rh_dirty_log(struct dm_region_hash *rh);
46
47/*
48 * Conversion functions.
49 */
50region_t dm_rh_bio_to_region(struct dm_region_hash *rh, struct bio *bio);
51sector_t dm_rh_region_to_sector(struct dm_region_hash *rh, region_t region);
52void *dm_rh_region_context(struct dm_region *reg);
53
54/*
55 * Get region size and key (ie. number of the region).
56 */
57sector_t dm_rh_get_region_size(struct dm_region_hash *rh);
58region_t dm_rh_get_region_key(struct dm_region *reg);
59
60/*
61 * Get/set/update region state (and dirty log).
62 *
63 */
64int dm_rh_get_state(struct dm_region_hash *rh, region_t region, int may_block);
65void dm_rh_set_state(struct dm_region_hash *rh, region_t region,
66 enum dm_rh_region_states state, int may_block);
67
68/* Non-zero errors_handled leaves the state of the region NOSYNC */
69void dm_rh_update_states(struct dm_region_hash *rh, int errors_handled);
70
71/* Flush the region hash and dirty log. */
72int dm_rh_flush(struct dm_region_hash *rh);
73
74/* Inc/dec pending count on regions. */
75void dm_rh_inc_pending(struct dm_region_hash *rh, struct bio_list *bios);
76void dm_rh_dec(struct dm_region_hash *rh, region_t region);
77
78/* Delay bios on regions. */
79void dm_rh_delay(struct dm_region_hash *rh, struct bio *bio);
80
81void dm_rh_mark_nosync(struct dm_region_hash *rh,
82 struct bio *bio, unsigned done, int error);
83
84/*
85 * Region recovery control.
86 */
87
88/* Prepare some regions for recovery by starting to quiesce them. */
89void dm_rh_recovery_prepare(struct dm_region_hash *rh);
90
91/* Try fetching a quiesced region for recovery. */
92struct dm_region *dm_rh_recovery_start(struct dm_region_hash *rh);
93
94/* Report recovery end on a region. */
95void dm_rh_recovery_end(struct dm_region *reg, int error);
96
97/* Returns number of regions with recovery work outstanding. */
98int dm_rh_recovery_in_flight(struct dm_region_hash *rh);
99
100/* Start/stop recovery. */
101void dm_rh_start_recovery(struct dm_region_hash *rh);
102void dm_rh_stop_recovery(struct dm_region_hash *rh);
103
104#endif /* DM_REGION_HASH_H */
generated by cgit v1.2.2 (git 2.25.0) at 2024-11-26 17:34:00 -0500