2 * Copyright (C) 2003 Sistina Software Limited.
4 * This file is released under the GPL.
8 #include "dm-bio-list.h"
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>
23 static struct workqueue_struct *_kmirrord_wq;
24 static struct work_struct _kmirrord_work;
26 static inline void wake(void)
28 queue_work(_kmirrord_wq, &_kmirrord_work);
31 /*-----------------------------------------------------------------
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.
38 * In addition to being present in the hash table a region _may_
39 * be present on one of three lists.
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
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.
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
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
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 *---------------------------------------------------------------*/
69 struct mirror_set *ms;
71 unsigned region_shift;
73 /* holds persistent region state */
74 struct dirty_log *log;
78 mempool_t *region_pool;
80 unsigned int nr_buckets;
81 struct list_head *buckets;
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;
98 struct region_hash *rh; /* FIXME: can we get rid of this ? */
102 struct list_head hash_list;
103 struct list_head list;
106 struct bio_list delayed_bios;
112 static inline region_t bio_to_region(struct region_hash *rh, struct bio *bio)
114 return bio->bi_sector >> rh->region_shift;
117 static inline sector_t region_to_sector(struct region_hash *rh, region_t region)
119 return region << rh->region_shift;
122 /* FIXME move this */
123 static void queue_bio(struct mirror_set *ms, struct bio *bio, int rw);
125 #define MIN_REGIONS 64
126 #define MAX_RECOVERY 1
127 static int rh_init(struct region_hash *rh, struct mirror_set *ms,
128 struct dirty_log *log, uint32_t region_size,
131 unsigned int nr_buckets, max_buckets;
135 * Calculate a suitable number of buckets for our hash
138 max_buckets = nr_regions >> 6;
139 for (nr_buckets = 128u; nr_buckets < max_buckets; nr_buckets <<= 1)
145 rh->region_size = region_size;
146 rh->region_shift = ffs(region_size) - 1;
147 rwlock_init(&rh->hash_lock);
148 rh->mask = nr_buckets - 1;
149 rh->nr_buckets = nr_buckets;
151 rh->buckets = vmalloc(nr_buckets * sizeof(*rh->buckets));
153 DMERR("unable to allocate region hash memory");
157 for (i = 0; i < nr_buckets; i++)
158 INIT_LIST_HEAD(rh->buckets + i);
160 spin_lock_init(&rh->region_lock);
161 sema_init(&rh->recovery_count, 0);
162 INIT_LIST_HEAD(&rh->clean_regions);
163 INIT_LIST_HEAD(&rh->quiesced_regions);
164 INIT_LIST_HEAD(&rh->recovered_regions);
166 rh->region_pool = mempool_create_kmalloc_pool(MIN_REGIONS,
167 sizeof(struct region));
168 if (!rh->region_pool) {
177 static void rh_exit(struct region_hash *rh)
180 struct region *reg, *nreg;
182 BUG_ON(!list_empty(&rh->quiesced_regions));
183 for (h = 0; h < rh->nr_buckets; h++) {
184 list_for_each_entry_safe(reg, nreg, rh->buckets + h, hash_list) {
185 BUG_ON(atomic_read(®->pending));
186 mempool_free(reg, rh->region_pool);
191 dm_destroy_dirty_log(rh->log);
193 mempool_destroy(rh->region_pool);
197 #define RH_HASH_MULT 2654435387U
199 static inline unsigned int rh_hash(struct region_hash *rh, region_t region)
201 return (unsigned int) ((region * RH_HASH_MULT) >> 12) & rh->mask;
204 static struct region *__rh_lookup(struct region_hash *rh, region_t region)
208 list_for_each_entry (reg, rh->buckets + rh_hash(rh, region), hash_list)
209 if (reg->key == region)
215 static void __rh_insert(struct region_hash *rh, struct region *reg)
217 unsigned int h = rh_hash(rh, reg->key);
218 list_add(®->hash_list, rh->buckets + h);
221 static struct region *__rh_alloc(struct region_hash *rh, region_t region)
223 struct region *reg, *nreg;
225 read_unlock(&rh->hash_lock);
226 nreg = mempool_alloc(rh->region_pool, GFP_NOIO);
227 nreg->state = rh->log->type->in_sync(rh->log, region, 1) ?
228 RH_CLEAN : RH_NOSYNC;
232 INIT_LIST_HEAD(&nreg->list);
234 atomic_set(&nreg->pending, 0);
235 bio_list_init(&nreg->delayed_bios);
236 write_lock_irq(&rh->hash_lock);
238 reg = __rh_lookup(rh, region);
240 /* we lost the race */
241 mempool_free(nreg, rh->region_pool);
244 __rh_insert(rh, nreg);
245 if (nreg->state == RH_CLEAN) {
246 spin_lock(&rh->region_lock);
247 list_add(&nreg->list, &rh->clean_regions);
248 spin_unlock(&rh->region_lock);
252 write_unlock_irq(&rh->hash_lock);
253 read_lock(&rh->hash_lock);
258 static inline struct region *__rh_find(struct region_hash *rh, region_t region)
262 reg = __rh_lookup(rh, region);
264 reg = __rh_alloc(rh, region);
269 static int rh_state(struct region_hash *rh, region_t region, int may_block)
274 read_lock(&rh->hash_lock);
275 reg = __rh_lookup(rh, region);
276 read_unlock(&rh->hash_lock);
282 * The region wasn't in the hash, so we fall back to the
285 r = rh->log->type->in_sync(rh->log, region, may_block);
288 * Any error from the dirty log (eg. -EWOULDBLOCK) gets
289 * taken as a RH_NOSYNC
291 return r == 1 ? RH_CLEAN : RH_NOSYNC;
294 static inline int rh_in_sync(struct region_hash *rh,
295 region_t region, int may_block)
297 int state = rh_state(rh, region, may_block);
298 return state == RH_CLEAN || state == RH_DIRTY;
301 static void dispatch_bios(struct mirror_set *ms, struct bio_list *bio_list)
305 while ((bio = bio_list_pop(bio_list))) {
306 queue_bio(ms, bio, WRITE);
310 static void rh_update_states(struct region_hash *rh)
312 struct region *reg, *next;
315 LIST_HEAD(recovered);
318 * Quickly grab the lists.
320 write_lock_irq(&rh->hash_lock);
321 spin_lock(&rh->region_lock);
322 if (!list_empty(&rh->clean_regions)) {
323 list_splice(&rh->clean_regions, &clean);
324 INIT_LIST_HEAD(&rh->clean_regions);
326 list_for_each_entry (reg, &clean, list) {
327 rh->log->type->clear_region(rh->log, reg->key);
328 list_del(®->hash_list);
332 if (!list_empty(&rh->recovered_regions)) {
333 list_splice(&rh->recovered_regions, &recovered);
334 INIT_LIST_HEAD(&rh->recovered_regions);
336 list_for_each_entry (reg, &recovered, list)
337 list_del(®->hash_list);
339 spin_unlock(&rh->region_lock);
340 write_unlock_irq(&rh->hash_lock);
343 * All the regions on the recovered and clean lists have
344 * now been pulled out of the system, so no need to do
347 list_for_each_entry_safe (reg, next, &recovered, list) {
348 rh->log->type->clear_region(rh->log, reg->key);
349 rh->log->type->complete_resync_work(rh->log, reg->key, 1);
350 dispatch_bios(rh->ms, ®->delayed_bios);
351 up(&rh->recovery_count);
352 mempool_free(reg, rh->region_pool);
355 if (!list_empty(&recovered))
356 rh->log->type->flush(rh->log);
358 list_for_each_entry_safe (reg, next, &clean, list)
359 mempool_free(reg, rh->region_pool);
362 static void rh_inc(struct region_hash *rh, region_t region)
366 read_lock(&rh->hash_lock);
367 reg = __rh_find(rh, region);
369 spin_lock_irq(&rh->region_lock);
370 atomic_inc(®->pending);
372 if (reg->state == RH_CLEAN) {
373 reg->state = RH_DIRTY;
374 list_del_init(®->list); /* take off the clean list */
375 spin_unlock_irq(&rh->region_lock);
377 rh->log->type->mark_region(rh->log, reg->key);
379 spin_unlock_irq(&rh->region_lock);
382 read_unlock(&rh->hash_lock);
385 static void rh_inc_pending(struct region_hash *rh, struct bio_list *bios)
389 for (bio = bios->head; bio; bio = bio->bi_next)
390 rh_inc(rh, bio_to_region(rh, bio));
393 static void rh_dec(struct region_hash *rh, region_t region)
399 read_lock(&rh->hash_lock);
400 reg = __rh_lookup(rh, region);
401 read_unlock(&rh->hash_lock);
403 spin_lock_irqsave(&rh->region_lock, flags);
404 if (atomic_dec_and_test(®->pending)) {
405 if (reg->state == RH_RECOVERING) {
406 list_add_tail(®->list, &rh->quiesced_regions);
408 reg->state = RH_CLEAN;
409 list_add(®->list, &rh->clean_regions);
413 spin_unlock_irqrestore(&rh->region_lock, flags);
420 * Starts quiescing a region in preparation for recovery.
422 static int __rh_recovery_prepare(struct region_hash *rh)
429 * Ask the dirty log what's next.
431 r = rh->log->type->get_resync_work(rh->log, ®ion);
436 * Get this region, and start it quiescing by setting the
439 read_lock(&rh->hash_lock);
440 reg = __rh_find(rh, region);
441 read_unlock(&rh->hash_lock);
443 spin_lock_irq(&rh->region_lock);
444 reg->state = RH_RECOVERING;
446 /* Already quiesced ? */
447 if (atomic_read(®->pending))
448 list_del_init(®->list);
451 list_del_init(®->list);
452 list_add(®->list, &rh->quiesced_regions);
454 spin_unlock_irq(&rh->region_lock);
459 static void rh_recovery_prepare(struct region_hash *rh)
461 while (!down_trylock(&rh->recovery_count))
462 if (__rh_recovery_prepare(rh) <= 0) {
463 up(&rh->recovery_count);
469 * Returns any quiesced regions.
471 static struct region *rh_recovery_start(struct region_hash *rh)
473 struct region *reg = NULL;
475 spin_lock_irq(&rh->region_lock);
476 if (!list_empty(&rh->quiesced_regions)) {
477 reg = list_entry(rh->quiesced_regions.next,
478 struct region, list);
479 list_del_init(®->list); /* remove from the quiesced list */
481 spin_unlock_irq(&rh->region_lock);
486 /* FIXME: success ignored for now */
487 static void rh_recovery_end(struct region *reg, int success)
489 struct region_hash *rh = reg->rh;
491 spin_lock_irq(&rh->region_lock);
492 list_add(®->list, ®->rh->recovered_regions);
493 spin_unlock_irq(&rh->region_lock);
498 static void rh_flush(struct region_hash *rh)
500 rh->log->type->flush(rh->log);
503 static void rh_delay(struct region_hash *rh, struct bio *bio)
507 read_lock(&rh->hash_lock);
508 reg = __rh_find(rh, bio_to_region(rh, bio));
509 bio_list_add(®->delayed_bios, bio);
510 read_unlock(&rh->hash_lock);
513 static void rh_stop_recovery(struct region_hash *rh)
517 /* wait for any recovering regions */
518 for (i = 0; i < MAX_RECOVERY; i++)
519 down(&rh->recovery_count);
522 static void rh_start_recovery(struct region_hash *rh)
526 for (i = 0; i < MAX_RECOVERY; i++)
527 up(&rh->recovery_count);
532 /*-----------------------------------------------------------------
533 * Mirror set structures.
534 *---------------------------------------------------------------*/
536 atomic_t error_count;
542 struct dm_target *ti;
543 struct list_head list;
544 struct region_hash rh;
545 struct kcopyd_client *kcopyd_client;
547 spinlock_t lock; /* protects the next two lists */
548 struct bio_list reads;
549 struct bio_list writes;
555 struct mirror *default_mirror; /* Default mirror */
557 unsigned int nr_mirrors;
558 struct mirror mirror[0];
562 * Every mirror should look like this one.
564 #define DEFAULT_MIRROR 0
567 * This is yucky. We squirrel the mirror_set struct away inside
568 * bi_next for write buffers. This is safe since the bh
569 * doesn't get submitted to the lower levels of block layer.
571 static struct mirror_set *bio_get_ms(struct bio *bio)
573 return (struct mirror_set *) bio->bi_next;
576 static void bio_set_ms(struct bio *bio, struct mirror_set *ms)
578 bio->bi_next = (struct bio *) ms;
581 /*-----------------------------------------------------------------
584 * When a mirror is first activated we may find that some regions
585 * are in the no-sync state. We have to recover these by
586 * recopying from the default mirror to all the others.
587 *---------------------------------------------------------------*/
588 static void recovery_complete(int read_err, unsigned int write_err,
591 struct region *reg = (struct region *) context;
593 /* FIXME: better error handling */
594 rh_recovery_end(reg, read_err || write_err);
597 static int recover(struct mirror_set *ms, struct region *reg)
601 struct io_region from, to[KCOPYD_MAX_REGIONS], *dest;
603 unsigned long flags = 0;
605 /* fill in the source */
606 m = ms->default_mirror;
607 from.bdev = m->dev->bdev;
608 from.sector = m->offset + region_to_sector(reg->rh, reg->key);
609 if (reg->key == (ms->nr_regions - 1)) {
611 * The final region may be smaller than
614 from.count = ms->ti->len & (reg->rh->region_size - 1);
616 from.count = reg->rh->region_size;
618 from.count = reg->rh->region_size;
620 /* fill in the destinations */
621 for (i = 0, dest = to; i < ms->nr_mirrors; i++) {
622 if (&ms->mirror[i] == ms->default_mirror)
626 dest->bdev = m->dev->bdev;
627 dest->sector = m->offset + region_to_sector(reg->rh, reg->key);
628 dest->count = from.count;
633 set_bit(KCOPYD_IGNORE_ERROR, &flags);
634 r = kcopyd_copy(ms->kcopyd_client, &from, ms->nr_mirrors - 1, to, flags,
635 recovery_complete, reg);
640 static void do_recovery(struct mirror_set *ms)
644 struct dirty_log *log = ms->rh.log;
647 * Start quiescing some regions.
649 rh_recovery_prepare(&ms->rh);
652 * Copy any already quiesced regions.
654 while ((reg = rh_recovery_start(&ms->rh))) {
655 r = recover(ms, reg);
657 rh_recovery_end(reg, 0);
661 * Update the in sync flag.
664 (log->type->get_sync_count(log) == ms->nr_regions)) {
665 /* the sync is complete */
666 dm_table_event(ms->ti->table);
671 /*-----------------------------------------------------------------
673 *---------------------------------------------------------------*/
674 static struct mirror *choose_mirror(struct mirror_set *ms, sector_t sector)
676 /* FIXME: add read balancing */
677 return ms->default_mirror;
681 * remap a buffer to a particular mirror.
683 static void map_bio(struct mirror_set *ms, struct mirror *m, struct bio *bio)
685 bio->bi_bdev = m->dev->bdev;
686 bio->bi_sector = m->offset + (bio->bi_sector - ms->ti->begin);
689 static void do_reads(struct mirror_set *ms, struct bio_list *reads)
695 while ((bio = bio_list_pop(reads))) {
696 region = bio_to_region(&ms->rh, bio);
699 * We can only read balance if the region is in sync.
701 if (rh_in_sync(&ms->rh, region, 0))
702 m = choose_mirror(ms, bio->bi_sector);
704 m = ms->default_mirror;
707 generic_make_request(bio);
711 /*-----------------------------------------------------------------
714 * We do different things with the write io depending on the
715 * state of the region that it's in:
717 * SYNC: increment pending, use kcopyd to write to *all* mirrors
718 * RECOVERING: delay the io until recovery completes
719 * NOSYNC: increment pending, just write to the default mirror
720 *---------------------------------------------------------------*/
721 static void write_callback(unsigned long error, void *context)
725 struct bio *bio = (struct bio *) context;
726 struct mirror_set *ms;
728 ms = bio_get_ms(bio);
729 bio_set_ms(bio, NULL);
732 * NOTE: We don't decrement the pending count here,
733 * instead it is done by the targets endio function.
734 * This way we handle both writes to SYNC and NOSYNC
735 * regions with the same code.
740 * only error the io if all mirrors failed.
744 for (i = 0; i < ms->nr_mirrors; i++)
745 if (!test_bit(i, &error)) {
750 bio_endio(bio, bio->bi_size, 0);
753 static void do_write(struct mirror_set *ms, struct bio *bio)
756 struct io_region io[KCOPYD_MAX_REGIONS+1];
759 for (i = 0; i < ms->nr_mirrors; i++) {
762 io[i].bdev = m->dev->bdev;
763 io[i].sector = m->offset + (bio->bi_sector - ms->ti->begin);
764 io[i].count = bio->bi_size >> 9;
768 dm_io_async_bvec(ms->nr_mirrors, io, WRITE,
769 bio->bi_io_vec + bio->bi_idx,
770 write_callback, bio);
773 static void do_writes(struct mirror_set *ms, struct bio_list *writes)
777 struct bio_list sync, nosync, recover, *this_list = NULL;
783 * Classify each write.
785 bio_list_init(&sync);
786 bio_list_init(&nosync);
787 bio_list_init(&recover);
789 while ((bio = bio_list_pop(writes))) {
790 state = rh_state(&ms->rh, bio_to_region(&ms->rh, bio), 1);
802 this_list = &recover;
806 bio_list_add(this_list, bio);
810 * Increment the pending counts for any regions that will
811 * be written to (writes to recover regions are going to
814 rh_inc_pending(&ms->rh, &sync);
815 rh_inc_pending(&ms->rh, &nosync);
821 while ((bio = bio_list_pop(&sync)))
824 while ((bio = bio_list_pop(&recover)))
825 rh_delay(&ms->rh, bio);
827 while ((bio = bio_list_pop(&nosync))) {
828 map_bio(ms, ms->default_mirror, bio);
829 generic_make_request(bio);
833 /*-----------------------------------------------------------------
835 *---------------------------------------------------------------*/
836 static LIST_HEAD(_mirror_sets);
837 static DECLARE_RWSEM(_mirror_sets_lock);
839 static void do_mirror(struct mirror_set *ms)
841 struct bio_list reads, writes;
843 spin_lock(&ms->lock);
846 bio_list_init(&ms->reads);
847 bio_list_init(&ms->writes);
848 spin_unlock(&ms->lock);
850 rh_update_states(&ms->rh);
852 do_reads(ms, &reads);
853 do_writes(ms, &writes);
856 static void do_work(void *ignored)
858 struct mirror_set *ms;
860 down_read(&_mirror_sets_lock);
861 list_for_each_entry (ms, &_mirror_sets, list)
863 up_read(&_mirror_sets_lock);
866 /*-----------------------------------------------------------------
868 *---------------------------------------------------------------*/
869 static struct mirror_set *alloc_context(unsigned int nr_mirrors,
870 uint32_t region_size,
871 struct dm_target *ti,
872 struct dirty_log *dl)
875 struct mirror_set *ms = NULL;
877 if (array_too_big(sizeof(*ms), sizeof(ms->mirror[0]), nr_mirrors))
880 len = sizeof(*ms) + (sizeof(ms->mirror[0]) * nr_mirrors);
882 ms = kmalloc(len, GFP_KERNEL);
884 ti->error = "dm-mirror: Cannot allocate mirror context";
889 spin_lock_init(&ms->lock);
892 ms->nr_mirrors = nr_mirrors;
893 ms->nr_regions = dm_sector_div_up(ti->len, region_size);
895 ms->default_mirror = &ms->mirror[DEFAULT_MIRROR];
897 if (rh_init(&ms->rh, ms, dl, region_size, ms->nr_regions)) {
898 ti->error = "dm-mirror: Error creating dirty region hash";
906 static void free_context(struct mirror_set *ms, struct dm_target *ti,
910 dm_put_device(ti, ms->mirror[m].dev);
916 static inline int _check_region_size(struct dm_target *ti, uint32_t size)
918 return !(size % (PAGE_SIZE >> 9) || (size & (size - 1)) ||
922 static int get_mirror(struct mirror_set *ms, struct dm_target *ti,
923 unsigned int mirror, char **argv)
927 if (sscanf(argv[1], SECTOR_FORMAT, &offset) != 1) {
928 ti->error = "dm-mirror: Invalid offset";
932 if (dm_get_device(ti, argv[0], offset, ti->len,
933 dm_table_get_mode(ti->table),
934 &ms->mirror[mirror].dev)) {
935 ti->error = "dm-mirror: Device lookup failure";
939 ms->mirror[mirror].offset = offset;
944 static int add_mirror_set(struct mirror_set *ms)
946 down_write(&_mirror_sets_lock);
947 list_add_tail(&ms->list, &_mirror_sets);
948 up_write(&_mirror_sets_lock);
954 static void del_mirror_set(struct mirror_set *ms)
956 down_write(&_mirror_sets_lock);
958 up_write(&_mirror_sets_lock);
962 * Create dirty log: log_type #log_params <log_params>
964 static struct dirty_log *create_dirty_log(struct dm_target *ti,
965 unsigned int argc, char **argv,
966 unsigned int *args_used)
968 unsigned int param_count;
969 struct dirty_log *dl;
972 ti->error = "dm-mirror: Insufficient mirror log arguments";
976 if (sscanf(argv[1], "%u", ¶m_count) != 1) {
977 ti->error = "dm-mirror: Invalid mirror log argument count";
981 *args_used = 2 + param_count;
983 if (argc < *args_used) {
984 ti->error = "dm-mirror: Insufficient mirror log arguments";
988 dl = dm_create_dirty_log(argv[0], ti, param_count, argv + 2);
990 ti->error = "dm-mirror: Error creating mirror dirty log";
994 if (!_check_region_size(ti, dl->type->get_region_size(dl))) {
995 ti->error = "dm-mirror: Invalid region size";
996 dm_destroy_dirty_log(dl);
1004 * Construct a mirror mapping:
1006 * log_type #log_params <log_params>
1007 * #mirrors [mirror_path offset]{2,}
1009 * log_type is "core" or "disk"
1010 * #log_params is between 1 and 3
1012 #define DM_IO_PAGES 64
1013 static int mirror_ctr(struct dm_target *ti, unsigned int argc, char **argv)
1016 unsigned int nr_mirrors, m, args_used;
1017 struct mirror_set *ms;
1018 struct dirty_log *dl;
1020 dl = create_dirty_log(ti, argc, argv, &args_used);
1027 if (!argc || sscanf(argv[0], "%u", &nr_mirrors) != 1 ||
1028 nr_mirrors < 2 || nr_mirrors > KCOPYD_MAX_REGIONS + 1) {
1029 ti->error = "dm-mirror: Invalid number of mirrors";
1030 dm_destroy_dirty_log(dl);
1036 if (argc != nr_mirrors * 2) {
1037 ti->error = "dm-mirror: Wrong number of mirror arguments";
1038 dm_destroy_dirty_log(dl);
1042 ms = alloc_context(nr_mirrors, dl->type->get_region_size(dl), ti, dl);
1044 dm_destroy_dirty_log(dl);
1048 /* Get the mirror parameter sets */
1049 for (m = 0; m < nr_mirrors; m++) {
1050 r = get_mirror(ms, ti, m, argv);
1052 free_context(ms, ti, m);
1060 ti->split_io = ms->rh.region_size;
1062 r = kcopyd_client_create(DM_IO_PAGES, &ms->kcopyd_client);
1064 free_context(ms, ti, ms->nr_mirrors);
1072 static void mirror_dtr(struct dm_target *ti)
1074 struct mirror_set *ms = (struct mirror_set *) ti->private;
1077 kcopyd_client_destroy(ms->kcopyd_client);
1078 free_context(ms, ti, ms->nr_mirrors);
1081 static void queue_bio(struct mirror_set *ms, struct bio *bio, int rw)
1083 int should_wake = 0;
1084 struct bio_list *bl;
1086 bl = (rw == WRITE) ? &ms->writes : &ms->reads;
1087 spin_lock(&ms->lock);
1088 should_wake = !(bl->head);
1089 bio_list_add(bl, bio);
1090 spin_unlock(&ms->lock);
1097 * Mirror mapping function
1099 static int mirror_map(struct dm_target *ti, struct bio *bio,
1100 union map_info *map_context)
1102 int r, rw = bio_rw(bio);
1104 struct mirror_set *ms = ti->private;
1106 map_context->ll = bio->bi_sector >> ms->rh.region_shift;
1109 queue_bio(ms, bio, rw);
1113 r = ms->rh.log->type->in_sync(ms->rh.log,
1114 bio_to_region(&ms->rh, bio), 0);
1115 if (r < 0 && r != -EWOULDBLOCK)
1118 if (r == -EWOULDBLOCK) /* FIXME: ugly */
1122 * We don't want to fast track a recovery just for a read
1123 * ahead. So we just let it silently fail.
1124 * FIXME: get rid of this.
1126 if (!r && rw == READA)
1130 /* Pass this io over to the daemon */
1131 queue_bio(ms, bio, rw);
1135 m = choose_mirror(ms, bio->bi_sector);
1139 map_bio(ms, m, bio);
1143 static int mirror_end_io(struct dm_target *ti, struct bio *bio,
1144 int error, union map_info *map_context)
1146 int rw = bio_rw(bio);
1147 struct mirror_set *ms = (struct mirror_set *) ti->private;
1148 region_t region = map_context->ll;
1151 * We need to dec pending if this was a write.
1154 rh_dec(&ms->rh, region);
1159 static void mirror_postsuspend(struct dm_target *ti)
1161 struct mirror_set *ms = (struct mirror_set *) ti->private;
1162 struct dirty_log *log = ms->rh.log;
1164 rh_stop_recovery(&ms->rh);
1165 if (log->type->suspend && log->type->suspend(log))
1166 /* FIXME: need better error handling */
1167 DMWARN("log suspend failed");
1170 static void mirror_resume(struct dm_target *ti)
1172 struct mirror_set *ms = (struct mirror_set *) ti->private;
1173 struct dirty_log *log = ms->rh.log;
1174 if (log->type->resume && log->type->resume(log))
1175 /* FIXME: need better error handling */
1176 DMWARN("log resume failed");
1177 rh_start_recovery(&ms->rh);
1180 static int mirror_status(struct dm_target *ti, status_type_t type,
1181 char *result, unsigned int maxlen)
1184 struct mirror_set *ms = (struct mirror_set *) ti->private;
1186 sz = ms->rh.log->type->status(ms->rh.log, type, result, maxlen);
1189 case STATUSTYPE_INFO:
1190 DMEMIT("%d ", ms->nr_mirrors);
1191 for (m = 0; m < ms->nr_mirrors; m++)
1192 DMEMIT("%s ", ms->mirror[m].dev->name);
1194 DMEMIT(SECTOR_FORMAT "/" SECTOR_FORMAT,
1195 ms->rh.log->type->get_sync_count(ms->rh.log),
1199 case STATUSTYPE_TABLE:
1200 DMEMIT("%d ", ms->nr_mirrors);
1201 for (m = 0; m < ms->nr_mirrors; m++)
1202 DMEMIT("%s " SECTOR_FORMAT " ",
1203 ms->mirror[m].dev->name, ms->mirror[m].offset);
1209 static struct target_type mirror_target = {
1211 .version = {1, 0, 1},
1212 .module = THIS_MODULE,
1216 .end_io = mirror_end_io,
1217 .postsuspend = mirror_postsuspend,
1218 .resume = mirror_resume,
1219 .status = mirror_status,
1222 static int __init dm_mirror_init(void)
1226 r = dm_dirty_log_init();
1230 _kmirrord_wq = create_singlethread_workqueue("kmirrord");
1231 if (!_kmirrord_wq) {
1232 DMERR("couldn't start kmirrord");
1233 dm_dirty_log_exit();
1236 INIT_WORK(&_kmirrord_work, do_work, NULL);
1238 r = dm_register_target(&mirror_target);
1240 DMERR("%s: Failed to register mirror target",
1241 mirror_target.name);
1242 dm_dirty_log_exit();
1243 destroy_workqueue(_kmirrord_wq);
1249 static void __exit dm_mirror_exit(void)
1253 r = dm_unregister_target(&mirror_target);
1255 DMERR("%s: unregister failed %d", mirror_target.name, r);
1257 destroy_workqueue(_kmirrord_wq);
1258 dm_dirty_log_exit();
1262 module_init(dm_mirror_init);
1263 module_exit(dm_mirror_exit);
1265 MODULE_DESCRIPTION(DM_NAME " mirror target");
1266 MODULE_AUTHOR("Joe Thornber");
1267 MODULE_LICENSE("GPL");