blob: 85c32b22a420a67daee1aaf64ccfdd354092d33a [file] [log] [blame]
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +01001/*
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>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090014#include <linux/slab.h>
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +010015#include <linux/vmalloc.h>
16
17#include "dm.h"
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +010018
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
Mikulas Patocka41841532009-12-10 23:51:59 +000083 /*
Tejun Heod87f4c12010-09-03 11:56:19 +020084 * If there was a flush failure no regions can be marked clean.
Mikulas Patocka41841532009-12-10 23:51:59 +000085 */
Tejun Heod87f4c12010-09-03 11:56:19 +020086 int flush_failure;
Mikulas Patocka41841532009-12-10 23:51:59 +000087
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +010088 void *context;
89 sector_t target_begin;
90
91 /* Callback function to schedule bios writes */
92 void (*dispatch_bios)(void *context, struct bio_list *bios);
93
94 /* Callback function to wakeup callers worker thread. */
95 void (*wakeup_workers)(void *context);
96
97 /* Callback function to wakeup callers recovery waiters. */
98 void (*wakeup_all_recovery_waiters)(void *context);
99};
100
101struct dm_region {
102 struct dm_region_hash *rh; /* FIXME: can we get rid of this ? */
103 region_t key;
104 int state;
105
106 struct list_head hash_list;
107 struct list_head list;
108
109 atomic_t pending;
110 struct bio_list delayed_bios;
111};
112
113/*
114 * Conversion fns
115 */
116static region_t dm_rh_sector_to_region(struct dm_region_hash *rh, sector_t sector)
117{
118 return sector >> rh->region_shift;
119}
120
121sector_t dm_rh_region_to_sector(struct dm_region_hash *rh, region_t region)
122{
123 return region << rh->region_shift;
124}
125EXPORT_SYMBOL_GPL(dm_rh_region_to_sector);
126
127region_t dm_rh_bio_to_region(struct dm_region_hash *rh, struct bio *bio)
128{
Kent Overstreet4f024f32013-10-11 15:44:27 -0700129 return dm_rh_sector_to_region(rh, bio->bi_iter.bi_sector -
130 rh->target_begin);
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100131}
132EXPORT_SYMBOL_GPL(dm_rh_bio_to_region);
133
134void *dm_rh_region_context(struct dm_region *reg)
135{
136 return reg->rh->context;
137}
138EXPORT_SYMBOL_GPL(dm_rh_region_context);
139
140region_t dm_rh_get_region_key(struct dm_region *reg)
141{
142 return reg->key;
143}
144EXPORT_SYMBOL_GPL(dm_rh_get_region_key);
145
146sector_t dm_rh_get_region_size(struct dm_region_hash *rh)
147{
148 return rh->region_size;
149}
150EXPORT_SYMBOL_GPL(dm_rh_get_region_size);
151
152/*
153 * FIXME: shall we pass in a structure instead of all these args to
154 * dm_region_hash_create()????
155 */
156#define RH_HASH_MULT 2654435387U
157#define RH_HASH_SHIFT 12
158
159#define MIN_REGIONS 64
160struct dm_region_hash *dm_region_hash_create(
161 void *context, void (*dispatch_bios)(void *context,
162 struct bio_list *bios),
163 void (*wakeup_workers)(void *context),
164 void (*wakeup_all_recovery_waiters)(void *context),
165 sector_t target_begin, unsigned max_recovery,
166 struct dm_dirty_log *log, uint32_t region_size,
167 region_t nr_regions)
168{
169 struct dm_region_hash *rh;
170 unsigned nr_buckets, max_buckets;
171 size_t i;
172
173 /*
174 * Calculate a suitable number of buckets for our hash
175 * table.
176 */
177 max_buckets = nr_regions >> 6;
178 for (nr_buckets = 128u; nr_buckets < max_buckets; nr_buckets <<= 1)
179 ;
180 nr_buckets >>= 1;
181
182 rh = kmalloc(sizeof(*rh), GFP_KERNEL);
183 if (!rh) {
184 DMERR("unable to allocate region hash memory");
185 return ERR_PTR(-ENOMEM);
186 }
187
188 rh->context = context;
189 rh->dispatch_bios = dispatch_bios;
190 rh->wakeup_workers = wakeup_workers;
191 rh->wakeup_all_recovery_waiters = wakeup_all_recovery_waiters;
192 rh->target_begin = target_begin;
193 rh->max_recovery = max_recovery;
194 rh->log = log;
195 rh->region_size = region_size;
Mikulas Patockaa3d939a2015-10-02 11:21:24 -0400196 rh->region_shift = __ffs(region_size);
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100197 rwlock_init(&rh->hash_lock);
198 rh->mask = nr_buckets - 1;
199 rh->nr_buckets = nr_buckets;
200
201 rh->shift = RH_HASH_SHIFT;
202 rh->prime = RH_HASH_MULT;
203
204 rh->buckets = vmalloc(nr_buckets * sizeof(*rh->buckets));
205 if (!rh->buckets) {
206 DMERR("unable to allocate region hash bucket memory");
207 kfree(rh);
208 return ERR_PTR(-ENOMEM);
209 }
210
211 for (i = 0; i < nr_buckets; i++)
212 INIT_LIST_HEAD(rh->buckets + i);
213
214 spin_lock_init(&rh->region_lock);
215 sema_init(&rh->recovery_count, 0);
216 atomic_set(&rh->recovery_in_flight, 0);
217 INIT_LIST_HEAD(&rh->clean_regions);
218 INIT_LIST_HEAD(&rh->quiesced_regions);
219 INIT_LIST_HEAD(&rh->recovered_regions);
220 INIT_LIST_HEAD(&rh->failed_recovered_regions);
Tejun Heod87f4c12010-09-03 11:56:19 +0200221 rh->flush_failure = 0;
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100222
223 rh->region_pool = mempool_create_kmalloc_pool(MIN_REGIONS,
224 sizeof(struct dm_region));
225 if (!rh->region_pool) {
226 vfree(rh->buckets);
227 kfree(rh);
228 rh = ERR_PTR(-ENOMEM);
229 }
230
231 return rh;
232}
233EXPORT_SYMBOL_GPL(dm_region_hash_create);
234
235void dm_region_hash_destroy(struct dm_region_hash *rh)
236{
237 unsigned h;
238 struct dm_region *reg, *nreg;
239
240 BUG_ON(!list_empty(&rh->quiesced_regions));
241 for (h = 0; h < rh->nr_buckets; h++) {
242 list_for_each_entry_safe(reg, nreg, rh->buckets + h,
243 hash_list) {
244 BUG_ON(atomic_read(&reg->pending));
245 mempool_free(reg, rh->region_pool);
246 }
247 }
248
249 if (rh->log)
250 dm_dirty_log_destroy(rh->log);
251
Julia Lawall6f659852015-09-13 14:15:05 +0200252 mempool_destroy(rh->region_pool);
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100253 vfree(rh->buckets);
254 kfree(rh);
255}
256EXPORT_SYMBOL_GPL(dm_region_hash_destroy);
257
258struct dm_dirty_log *dm_rh_dirty_log(struct dm_region_hash *rh)
259{
260 return rh->log;
261}
262EXPORT_SYMBOL_GPL(dm_rh_dirty_log);
263
264static unsigned rh_hash(struct dm_region_hash *rh, region_t region)
265{
266 return (unsigned) ((region * rh->prime) >> rh->shift) & rh->mask;
267}
268
269static struct dm_region *__rh_lookup(struct dm_region_hash *rh, region_t region)
270{
271 struct dm_region *reg;
272 struct list_head *bucket = rh->buckets + rh_hash(rh, region);
273
274 list_for_each_entry(reg, bucket, hash_list)
275 if (reg->key == region)
276 return reg;
277
278 return NULL;
279}
280
281static void __rh_insert(struct dm_region_hash *rh, struct dm_region *reg)
282{
283 list_add(&reg->hash_list, rh->buckets + rh_hash(rh, reg->key));
284}
285
286static struct dm_region *__rh_alloc(struct dm_region_hash *rh, region_t region)
287{
288 struct dm_region *reg, *nreg;
289
290 nreg = mempool_alloc(rh->region_pool, GFP_ATOMIC);
291 if (unlikely(!nreg))
Mikulas Patockaa72986c2009-06-22 10:12:13 +0100292 nreg = kmalloc(sizeof(*nreg), GFP_NOIO | __GFP_NOFAIL);
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100293
294 nreg->state = rh->log->type->in_sync(rh->log, region, 1) ?
295 DM_RH_CLEAN : DM_RH_NOSYNC;
296 nreg->rh = rh;
297 nreg->key = region;
298 INIT_LIST_HEAD(&nreg->list);
299 atomic_set(&nreg->pending, 0);
300 bio_list_init(&nreg->delayed_bios);
301
302 write_lock_irq(&rh->hash_lock);
303 reg = __rh_lookup(rh, region);
304 if (reg)
305 /* We lost the race. */
306 mempool_free(nreg, rh->region_pool);
307 else {
308 __rh_insert(rh, nreg);
309 if (nreg->state == DM_RH_CLEAN) {
310 spin_lock(&rh->region_lock);
311 list_add(&nreg->list, &rh->clean_regions);
312 spin_unlock(&rh->region_lock);
313 }
314
315 reg = nreg;
316 }
317 write_unlock_irq(&rh->hash_lock);
318
319 return reg;
320}
321
322static struct dm_region *__rh_find(struct dm_region_hash *rh, region_t region)
323{
324 struct dm_region *reg;
325
326 reg = __rh_lookup(rh, region);
327 if (!reg) {
328 read_unlock(&rh->hash_lock);
329 reg = __rh_alloc(rh, region);
330 read_lock(&rh->hash_lock);
331 }
332
333 return reg;
334}
335
336int dm_rh_get_state(struct dm_region_hash *rh, region_t region, int may_block)
337{
338 int r;
339 struct dm_region *reg;
340
341 read_lock(&rh->hash_lock);
342 reg = __rh_lookup(rh, region);
343 read_unlock(&rh->hash_lock);
344
345 if (reg)
346 return reg->state;
347
348 /*
349 * The region wasn't in the hash, so we fall back to the
350 * dirty log.
351 */
352 r = rh->log->type->in_sync(rh->log, region, may_block);
353
354 /*
355 * Any error from the dirty log (eg. -EWOULDBLOCK) gets
356 * taken as a DM_RH_NOSYNC
357 */
358 return r == 1 ? DM_RH_CLEAN : DM_RH_NOSYNC;
359}
360EXPORT_SYMBOL_GPL(dm_rh_get_state);
361
362static void complete_resync_work(struct dm_region *reg, int success)
363{
364 struct dm_region_hash *rh = reg->rh;
365
366 rh->log->type->set_region_sync(rh->log, reg->key, success);
367
368 /*
369 * Dispatch the bios before we call 'wake_up_all'.
370 * This is important because if we are suspending,
371 * we want to know that recovery is complete and
372 * the work queue is flushed. If we wake_up_all
373 * before we dispatch_bios (queue bios and call wake()),
374 * then we risk suspending before the work queue
375 * has been properly flushed.
376 */
377 rh->dispatch_bios(rh->context, &reg->delayed_bios);
378 if (atomic_dec_and_test(&rh->recovery_in_flight))
379 rh->wakeup_all_recovery_waiters(rh->context);
380 up(&rh->recovery_count);
381}
382
383/* dm_rh_mark_nosync
384 * @ms
385 * @bio
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100386 *
387 * The bio was written on some mirror(s) but failed on other mirror(s).
388 * We can successfully endio the bio but should avoid the region being
389 * marked clean by setting the state DM_RH_NOSYNC.
390 *
391 * This function is _not_ safe in interrupt context!
392 */
Mikulas Patockac58098b2009-12-10 23:52:05 +0000393void dm_rh_mark_nosync(struct dm_region_hash *rh, struct bio *bio)
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100394{
395 unsigned long flags;
396 struct dm_dirty_log *log = rh->log;
397 struct dm_region *reg;
398 region_t region = dm_rh_bio_to_region(rh, bio);
399 int recovering = 0;
400
Jens Axboe1eff9d32016-08-05 15:35:16 -0600401 if (bio->bi_opf & REQ_PREFLUSH) {
Tejun Heod87f4c12010-09-03 11:56:19 +0200402 rh->flush_failure = 1;
Mikulas Patocka41841532009-12-10 23:51:59 +0000403 return;
404 }
405
Mike Christiee6047142016-06-05 14:32:04 -0500406 if (bio_op(bio) == REQ_OP_DISCARD)
Mikulas Patocka751f1882012-07-20 14:25:03 +0100407 return;
408
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100409 /* We must inform the log that the sync count has changed. */
410 log->type->set_region_sync(log, region, 0);
411
412 read_lock(&rh->hash_lock);
413 reg = __rh_find(rh, region);
414 read_unlock(&rh->hash_lock);
415
416 /* region hash entry should exist because write was in-flight */
417 BUG_ON(!reg);
418 BUG_ON(!list_empty(&reg->list));
419
420 spin_lock_irqsave(&rh->region_lock, flags);
421 /*
422 * Possible cases:
423 * 1) DM_RH_DIRTY
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300424 * 2) DM_RH_NOSYNC: was dirty, other preceding writes failed
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100425 * 3) DM_RH_RECOVERING: flushing pending writes
426 * Either case, the region should have not been connected to list.
427 */
428 recovering = (reg->state == DM_RH_RECOVERING);
429 reg->state = DM_RH_NOSYNC;
430 BUG_ON(!list_empty(&reg->list));
431 spin_unlock_irqrestore(&rh->region_lock, flags);
432
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100433 if (recovering)
434 complete_resync_work(reg, 0);
435}
436EXPORT_SYMBOL_GPL(dm_rh_mark_nosync);
437
438void dm_rh_update_states(struct dm_region_hash *rh, int errors_handled)
439{
440 struct dm_region *reg, *next;
441
442 LIST_HEAD(clean);
443 LIST_HEAD(recovered);
444 LIST_HEAD(failed_recovered);
445
446 /*
447 * Quickly grab the lists.
448 */
449 write_lock_irq(&rh->hash_lock);
450 spin_lock(&rh->region_lock);
451 if (!list_empty(&rh->clean_regions)) {
452 list_splice_init(&rh->clean_regions, &clean);
453
454 list_for_each_entry(reg, &clean, list)
455 list_del(&reg->hash_list);
456 }
457
458 if (!list_empty(&rh->recovered_regions)) {
459 list_splice_init(&rh->recovered_regions, &recovered);
460
461 list_for_each_entry(reg, &recovered, list)
462 list_del(&reg->hash_list);
463 }
464
465 if (!list_empty(&rh->failed_recovered_regions)) {
466 list_splice_init(&rh->failed_recovered_regions,
467 &failed_recovered);
468
469 list_for_each_entry(reg, &failed_recovered, list)
470 list_del(&reg->hash_list);
471 }
472
473 spin_unlock(&rh->region_lock);
474 write_unlock_irq(&rh->hash_lock);
475
476 /*
477 * All the regions on the recovered and clean lists have
478 * now been pulled out of the system, so no need to do
479 * any more locking.
480 */
481 list_for_each_entry_safe(reg, next, &recovered, list) {
482 rh->log->type->clear_region(rh->log, reg->key);
483 complete_resync_work(reg, 1);
484 mempool_free(reg, rh->region_pool);
485 }
486
487 list_for_each_entry_safe(reg, next, &failed_recovered, list) {
488 complete_resync_work(reg, errors_handled ? 0 : 1);
489 mempool_free(reg, rh->region_pool);
490 }
491
492 list_for_each_entry_safe(reg, next, &clean, list) {
493 rh->log->type->clear_region(rh->log, reg->key);
494 mempool_free(reg, rh->region_pool);
495 }
496
497 rh->log->type->flush(rh->log);
498}
499EXPORT_SYMBOL_GPL(dm_rh_update_states);
500
501static void rh_inc(struct dm_region_hash *rh, region_t region)
502{
503 struct dm_region *reg;
504
505 read_lock(&rh->hash_lock);
506 reg = __rh_find(rh, region);
507
508 spin_lock_irq(&rh->region_lock);
509 atomic_inc(&reg->pending);
510
511 if (reg->state == DM_RH_CLEAN) {
512 reg->state = DM_RH_DIRTY;
513 list_del_init(&reg->list); /* take off the clean list */
514 spin_unlock_irq(&rh->region_lock);
515
516 rh->log->type->mark_region(rh->log, reg->key);
517 } else
518 spin_unlock_irq(&rh->region_lock);
519
520
521 read_unlock(&rh->hash_lock);
522}
523
524void dm_rh_inc_pending(struct dm_region_hash *rh, struct bio_list *bios)
525{
526 struct bio *bio;
527
Mikulas Patocka41841532009-12-10 23:51:59 +0000528 for (bio = bios->head; bio; bio = bio->bi_next) {
Jens Axboe1eff9d32016-08-05 15:35:16 -0600529 if (bio->bi_opf & REQ_PREFLUSH || bio_op(bio) == REQ_OP_DISCARD)
Mikulas Patocka41841532009-12-10 23:51:59 +0000530 continue;
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100531 rh_inc(rh, dm_rh_bio_to_region(rh, bio));
Mikulas Patocka41841532009-12-10 23:51:59 +0000532 }
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100533}
534EXPORT_SYMBOL_GPL(dm_rh_inc_pending);
535
536void dm_rh_dec(struct dm_region_hash *rh, region_t region)
537{
538 unsigned long flags;
539 struct dm_region *reg;
540 int should_wake = 0;
541
542 read_lock(&rh->hash_lock);
543 reg = __rh_lookup(rh, region);
544 read_unlock(&rh->hash_lock);
545
546 spin_lock_irqsave(&rh->region_lock, flags);
547 if (atomic_dec_and_test(&reg->pending)) {
548 /*
549 * There is no pending I/O for this region.
550 * We can move the region to corresponding list for next action.
551 * At this point, the region is not yet connected to any list.
552 *
553 * If the state is DM_RH_NOSYNC, the region should be kept off
554 * from clean list.
555 * The hash entry for DM_RH_NOSYNC will remain in memory
556 * until the region is recovered or the map is reloaded.
557 */
558
559 /* do nothing for DM_RH_NOSYNC */
Tejun Heod87f4c12010-09-03 11:56:19 +0200560 if (unlikely(rh->flush_failure)) {
Mikulas Patocka41841532009-12-10 23:51:59 +0000561 /*
Tejun Heod87f4c12010-09-03 11:56:19 +0200562 * If a write flush failed some time ago, we
Mikulas Patocka41841532009-12-10 23:51:59 +0000563 * don't know whether or not this write made it
564 * to the disk, so we must resync the device.
565 */
566 reg->state = DM_RH_NOSYNC;
567 } else if (reg->state == DM_RH_RECOVERING) {
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100568 list_add_tail(&reg->list, &rh->quiesced_regions);
569 } else if (reg->state == DM_RH_DIRTY) {
570 reg->state = DM_RH_CLEAN;
571 list_add(&reg->list, &rh->clean_regions);
572 }
573 should_wake = 1;
574 }
575 spin_unlock_irqrestore(&rh->region_lock, flags);
576
577 if (should_wake)
578 rh->wakeup_workers(rh->context);
579}
580EXPORT_SYMBOL_GPL(dm_rh_dec);
581
582/*
583 * Starts quiescing a region in preparation for recovery.
584 */
585static int __rh_recovery_prepare(struct dm_region_hash *rh)
586{
587 int r;
588 region_t region;
589 struct dm_region *reg;
590
591 /*
592 * Ask the dirty log what's next.
593 */
594 r = rh->log->type->get_resync_work(rh->log, &region);
595 if (r <= 0)
596 return r;
597
598 /*
599 * Get this region, and start it quiescing by setting the
600 * recovering flag.
601 */
602 read_lock(&rh->hash_lock);
603 reg = __rh_find(rh, region);
604 read_unlock(&rh->hash_lock);
605
606 spin_lock_irq(&rh->region_lock);
607 reg->state = DM_RH_RECOVERING;
608
609 /* Already quiesced ? */
610 if (atomic_read(&reg->pending))
611 list_del_init(&reg->list);
612 else
613 list_move(&reg->list, &rh->quiesced_regions);
614
615 spin_unlock_irq(&rh->region_lock);
616
617 return 1;
618}
619
620void dm_rh_recovery_prepare(struct dm_region_hash *rh)
621{
622 /* Extra reference to avoid race with dm_rh_stop_recovery */
623 atomic_inc(&rh->recovery_in_flight);
624
625 while (!down_trylock(&rh->recovery_count)) {
626 atomic_inc(&rh->recovery_in_flight);
627 if (__rh_recovery_prepare(rh) <= 0) {
628 atomic_dec(&rh->recovery_in_flight);
629 up(&rh->recovery_count);
630 break;
631 }
632 }
633
634 /* Drop the extra reference */
635 if (atomic_dec_and_test(&rh->recovery_in_flight))
636 rh->wakeup_all_recovery_waiters(rh->context);
637}
638EXPORT_SYMBOL_GPL(dm_rh_recovery_prepare);
639
640/*
641 * Returns any quiesced regions.
642 */
643struct dm_region *dm_rh_recovery_start(struct dm_region_hash *rh)
644{
645 struct dm_region *reg = NULL;
646
647 spin_lock_irq(&rh->region_lock);
648 if (!list_empty(&rh->quiesced_regions)) {
649 reg = list_entry(rh->quiesced_regions.next,
650 struct dm_region, list);
651 list_del_init(&reg->list); /* remove from the quiesced list */
652 }
653 spin_unlock_irq(&rh->region_lock);
654
655 return reg;
656}
657EXPORT_SYMBOL_GPL(dm_rh_recovery_start);
658
659void dm_rh_recovery_end(struct dm_region *reg, int success)
660{
661 struct dm_region_hash *rh = reg->rh;
662
663 spin_lock_irq(&rh->region_lock);
664 if (success)
665 list_add(&reg->list, &reg->rh->recovered_regions);
Takahiro Yasui558569a2010-02-16 18:42:58 +0000666 else
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100667 list_add(&reg->list, &reg->rh->failed_recovered_regions);
Takahiro Yasui558569a2010-02-16 18:42:58 +0000668
Heinz Mauelshagen1f965b12008-10-21 17:45:06 +0100669 spin_unlock_irq(&rh->region_lock);
670
671 rh->wakeup_workers(rh->context);
672}
673EXPORT_SYMBOL_GPL(dm_rh_recovery_end);
674
675/* Return recovery in flight count. */
676int dm_rh_recovery_in_flight(struct dm_region_hash *rh)
677{
678 return atomic_read(&rh->recovery_in_flight);
679}
680EXPORT_SYMBOL_GPL(dm_rh_recovery_in_flight);
681
682int dm_rh_flush(struct dm_region_hash *rh)
683{
684 return rh->log->type->flush(rh->log);
685}
686EXPORT_SYMBOL_GPL(dm_rh_flush);
687
688void dm_rh_delay(struct dm_region_hash *rh, struct bio *bio)
689{
690 struct dm_region *reg;
691
692 read_lock(&rh->hash_lock);
693 reg = __rh_find(rh, dm_rh_bio_to_region(rh, bio));
694 bio_list_add(&reg->delayed_bios, bio);
695 read_unlock(&rh->hash_lock);
696}
697EXPORT_SYMBOL_GPL(dm_rh_delay);
698
699void dm_rh_stop_recovery(struct dm_region_hash *rh)
700{
701 int i;
702
703 /* wait for any recovering regions */
704 for (i = 0; i < rh->max_recovery; i++)
705 down(&rh->recovery_count);
706}
707EXPORT_SYMBOL_GPL(dm_rh_stop_recovery);
708
709void dm_rh_start_recovery(struct dm_region_hash *rh)
710{
711 int i;
712
713 for (i = 0; i < rh->max_recovery; i++)
714 up(&rh->recovery_count);
715
716 rh->wakeup_workers(rh->context);
717}
718EXPORT_SYMBOL_GPL(dm_rh_start_recovery);
719
720MODULE_DESCRIPTION(DM_NAME " region hash");
721MODULE_AUTHOR("Joe Thornber/Heinz Mauelshagen <dm-devel@redhat.com>");
722MODULE_LICENSE("GPL");