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 #define DM_MSG_PREFIX "raid1"
25 static struct workqueue_struct *_kmirrord_wq;
26 static struct work_struct _kmirrord_work;
27 static DECLARE_WAIT_QUEUE_HEAD(_kmirrord_recovery_stopped);
29 static inline void wake(void)
31 queue_work(_kmirrord_wq, &_kmirrord_work);
34 /*-----------------------------------------------------------------
37 * The mirror splits itself up into discrete regions. Each
38 * region can be in one of three states: clean, dirty,
39 * nosync. There is no need to put clean regions in the hash.
41 * In addition to being present in the hash table a region _may_
42 * be present on one of three lists.
44 * clean_regions: Regions on this list have no io pending to
45 * them, they are in sync, we are no longer interested in them,
46 * they are dull. rh_update_states() will remove them from the
49 * quiesced_regions: These regions have been spun down, ready
50 * for recovery. rh_recovery_start() will remove regions from
51 * this list and hand them to kmirrord, which will schedule the
52 * recovery io with kcopyd.
54 * recovered_regions: Regions that kcopyd has successfully
55 * recovered. rh_update_states() will now schedule any delayed
56 * io, up the recovery_count, and remove the region from the
60 * A rw spin lock 'hash_lock' protects just the hash table,
61 * this is never held in write mode from interrupt context,
62 * which I believe means that we only have to disable irqs when
65 * An ordinary spin lock 'region_lock' that protects the three
66 * lists in the region_hash, with the 'state', 'list' and
67 * 'bhs_delayed' fields of the regions. This is used from irq
68 * context, so all other uses will have to suspend local irqs.
69 *---------------------------------------------------------------*/
72 struct mirror_set *ms;
74 unsigned region_shift;
76 /* holds persistent region state */
77 struct dirty_log *log;
81 mempool_t *region_pool;
83 unsigned int nr_buckets;
84 struct list_head *buckets;
86 spinlock_t region_lock;
87 atomic_t recovery_in_flight;
88 struct semaphore recovery_count;
89 struct list_head clean_regions;
90 struct list_head quiesced_regions;
91 struct list_head recovered_regions;
102 struct region_hash *rh; /* FIXME: can we get rid of this ? */
106 struct list_head hash_list;
107 struct list_head list;
110 struct bio_list delayed_bios;
114 /*-----------------------------------------------------------------
115 * Mirror set structures.
116 *---------------------------------------------------------------*/
118 atomic_t error_count;
124 struct dm_target *ti;
125 struct list_head list;
126 struct region_hash rh;
127 struct kcopyd_client *kcopyd_client;
129 spinlock_t lock; /* protects the next two lists */
130 struct bio_list reads;
131 struct bio_list writes;
137 struct mirror *default_mirror; /* Default mirror */
139 unsigned int nr_mirrors;
140 struct mirror mirror[0];
146 static inline region_t bio_to_region(struct region_hash *rh, struct bio *bio)
148 return (bio->bi_sector - rh->ms->ti->begin) >> rh->region_shift;
151 static inline sector_t region_to_sector(struct region_hash *rh, region_t region)
153 return region << rh->region_shift;
156 /* FIXME move this */
157 static void queue_bio(struct mirror_set *ms, struct bio *bio, int rw);
159 #define MIN_REGIONS 64
160 #define MAX_RECOVERY 1
161 static int rh_init(struct region_hash *rh, struct mirror_set *ms,
162 struct dirty_log *log, uint32_t region_size,
165 unsigned int nr_buckets, max_buckets;
169 * Calculate a suitable number of buckets for our hash
172 max_buckets = nr_regions >> 6;
173 for (nr_buckets = 128u; nr_buckets < max_buckets; nr_buckets <<= 1)
179 rh->region_size = region_size;
180 rh->region_shift = ffs(region_size) - 1;
181 rwlock_init(&rh->hash_lock);
182 rh->mask = nr_buckets - 1;
183 rh->nr_buckets = nr_buckets;
185 rh->buckets = vmalloc(nr_buckets * sizeof(*rh->buckets));
187 DMERR("unable to allocate region hash memory");
191 for (i = 0; i < nr_buckets; i++)
192 INIT_LIST_HEAD(rh->buckets + i);
194 spin_lock_init(&rh->region_lock);
195 sema_init(&rh->recovery_count, 0);
196 atomic_set(&rh->recovery_in_flight, 0);
197 INIT_LIST_HEAD(&rh->clean_regions);
198 INIT_LIST_HEAD(&rh->quiesced_regions);
199 INIT_LIST_HEAD(&rh->recovered_regions);
201 rh->region_pool = mempool_create_kmalloc_pool(MIN_REGIONS,
202 sizeof(struct region));
203 if (!rh->region_pool) {
212 static void rh_exit(struct region_hash *rh)
215 struct region *reg, *nreg;
217 BUG_ON(!list_empty(&rh->quiesced_regions));
218 for (h = 0; h < rh->nr_buckets; h++) {
219 list_for_each_entry_safe(reg, nreg, rh->buckets + h, hash_list) {
220 BUG_ON(atomic_read(®->pending));
221 mempool_free(reg, rh->region_pool);
226 dm_destroy_dirty_log(rh->log);
228 mempool_destroy(rh->region_pool);
232 #define RH_HASH_MULT 2654435387U
234 static inline unsigned int rh_hash(struct region_hash *rh, region_t region)
236 return (unsigned int) ((region * RH_HASH_MULT) >> 12) & rh->mask;
239 static struct region *__rh_lookup(struct region_hash *rh, region_t region)
243 list_for_each_entry (reg, rh->buckets + rh_hash(rh, region), hash_list)
244 if (reg->key == region)
250 static void __rh_insert(struct region_hash *rh, struct region *reg)
252 unsigned int h = rh_hash(rh, reg->key);
253 list_add(®->hash_list, rh->buckets + h);
256 static struct region *__rh_alloc(struct region_hash *rh, region_t region)
258 struct region *reg, *nreg;
260 read_unlock(&rh->hash_lock);
261 nreg = mempool_alloc(rh->region_pool, GFP_ATOMIC);
263 nreg = kmalloc(sizeof(struct region), GFP_NOIO);
264 nreg->state = rh->log->type->in_sync(rh->log, region, 1) ?
265 RH_CLEAN : RH_NOSYNC;
269 INIT_LIST_HEAD(&nreg->list);
271 atomic_set(&nreg->pending, 0);
272 bio_list_init(&nreg->delayed_bios);
273 write_lock_irq(&rh->hash_lock);
275 reg = __rh_lookup(rh, region);
277 /* we lost the race */
278 mempool_free(nreg, rh->region_pool);
281 __rh_insert(rh, nreg);
282 if (nreg->state == RH_CLEAN) {
283 spin_lock(&rh->region_lock);
284 list_add(&nreg->list, &rh->clean_regions);
285 spin_unlock(&rh->region_lock);
289 write_unlock_irq(&rh->hash_lock);
290 read_lock(&rh->hash_lock);
295 static inline struct region *__rh_find(struct region_hash *rh, region_t region)
299 reg = __rh_lookup(rh, region);
301 reg = __rh_alloc(rh, region);
306 static int rh_state(struct region_hash *rh, region_t region, int may_block)
311 read_lock(&rh->hash_lock);
312 reg = __rh_lookup(rh, region);
313 read_unlock(&rh->hash_lock);
319 * The region wasn't in the hash, so we fall back to the
322 r = rh->log->type->in_sync(rh->log, region, may_block);
325 * Any error from the dirty log (eg. -EWOULDBLOCK) gets
326 * taken as a RH_NOSYNC
328 return r == 1 ? RH_CLEAN : RH_NOSYNC;
331 static inline int rh_in_sync(struct region_hash *rh,
332 region_t region, int may_block)
334 int state = rh_state(rh, region, may_block);
335 return state == RH_CLEAN || state == RH_DIRTY;
338 static void dispatch_bios(struct mirror_set *ms, struct bio_list *bio_list)
342 while ((bio = bio_list_pop(bio_list))) {
343 queue_bio(ms, bio, WRITE);
347 static void complete_resync_work(struct region *reg, int success)
349 struct region_hash *rh = reg->rh;
351 rh->log->type->set_region_sync(rh->log, reg->key, success);
352 dispatch_bios(rh->ms, ®->delayed_bios);
353 if (atomic_dec_and_test(&rh->recovery_in_flight))
354 wake_up_all(&_kmirrord_recovery_stopped);
355 up(&rh->recovery_count);
358 static void rh_update_states(struct region_hash *rh)
360 struct region *reg, *next;
363 LIST_HEAD(recovered);
366 * Quickly grab the lists.
368 write_lock_irq(&rh->hash_lock);
369 spin_lock(&rh->region_lock);
370 if (!list_empty(&rh->clean_regions)) {
371 list_splice(&rh->clean_regions, &clean);
372 INIT_LIST_HEAD(&rh->clean_regions);
374 list_for_each_entry (reg, &clean, list) {
375 rh->log->type->clear_region(rh->log, reg->key);
376 list_del(®->hash_list);
380 if (!list_empty(&rh->recovered_regions)) {
381 list_splice(&rh->recovered_regions, &recovered);
382 INIT_LIST_HEAD(&rh->recovered_regions);
384 list_for_each_entry (reg, &recovered, list)
385 list_del(®->hash_list);
387 spin_unlock(&rh->region_lock);
388 write_unlock_irq(&rh->hash_lock);
391 * All the regions on the recovered and clean lists have
392 * now been pulled out of the system, so no need to do
395 list_for_each_entry_safe (reg, next, &recovered, list) {
396 rh->log->type->clear_region(rh->log, reg->key);
397 complete_resync_work(reg, 1);
398 mempool_free(reg, rh->region_pool);
401 if (!list_empty(&recovered))
402 rh->log->type->flush(rh->log);
404 list_for_each_entry_safe (reg, next, &clean, list)
405 mempool_free(reg, rh->region_pool);
408 static void rh_inc(struct region_hash *rh, region_t region)
412 read_lock(&rh->hash_lock);
413 reg = __rh_find(rh, region);
415 spin_lock_irq(&rh->region_lock);
416 atomic_inc(®->pending);
418 if (reg->state == RH_CLEAN) {
419 reg->state = RH_DIRTY;
420 list_del_init(®->list); /* take off the clean list */
421 spin_unlock_irq(&rh->region_lock);
423 rh->log->type->mark_region(rh->log, reg->key);
425 spin_unlock_irq(&rh->region_lock);
428 read_unlock(&rh->hash_lock);
431 static void rh_inc_pending(struct region_hash *rh, struct bio_list *bios)
435 for (bio = bios->head; bio; bio = bio->bi_next)
436 rh_inc(rh, bio_to_region(rh, bio));
439 static void rh_dec(struct region_hash *rh, region_t region)
445 read_lock(&rh->hash_lock);
446 reg = __rh_lookup(rh, region);
447 read_unlock(&rh->hash_lock);
449 spin_lock_irqsave(&rh->region_lock, flags);
450 if (atomic_dec_and_test(®->pending)) {
452 * There is no pending I/O for this region.
453 * We can move the region to corresponding list for next action.
454 * At this point, the region is not yet connected to any list.
456 * If the state is RH_NOSYNC, the region should be kept off
458 * The hash entry for RH_NOSYNC will remain in memory
459 * until the region is recovered or the map is reloaded.
462 /* do nothing for RH_NOSYNC */
463 if (reg->state == RH_RECOVERING) {
464 list_add_tail(®->list, &rh->quiesced_regions);
465 } else if (reg->state == RH_DIRTY) {
466 reg->state = RH_CLEAN;
467 list_add(®->list, &rh->clean_regions);
471 spin_unlock_irqrestore(&rh->region_lock, flags);
478 * Starts quiescing a region in preparation for recovery.
480 static int __rh_recovery_prepare(struct region_hash *rh)
487 * Ask the dirty log what's next.
489 r = rh->log->type->get_resync_work(rh->log, ®ion);
494 * Get this region, and start it quiescing by setting the
497 read_lock(&rh->hash_lock);
498 reg = __rh_find(rh, region);
499 read_unlock(&rh->hash_lock);
501 spin_lock_irq(&rh->region_lock);
502 reg->state = RH_RECOVERING;
504 /* Already quiesced ? */
505 if (atomic_read(®->pending))
506 list_del_init(®->list);
508 list_move(®->list, &rh->quiesced_regions);
510 spin_unlock_irq(&rh->region_lock);
515 static void rh_recovery_prepare(struct region_hash *rh)
517 /* Extra reference to avoid race with rh_stop_recovery */
518 atomic_inc(&rh->recovery_in_flight);
520 while (!down_trylock(&rh->recovery_count)) {
521 atomic_inc(&rh->recovery_in_flight);
522 if (__rh_recovery_prepare(rh) <= 0) {
523 atomic_dec(&rh->recovery_in_flight);
524 up(&rh->recovery_count);
529 /* Drop the extra reference */
530 if (atomic_dec_and_test(&rh->recovery_in_flight))
531 wake_up_all(&_kmirrord_recovery_stopped);
535 * Returns any quiesced regions.
537 static struct region *rh_recovery_start(struct region_hash *rh)
539 struct region *reg = NULL;
541 spin_lock_irq(&rh->region_lock);
542 if (!list_empty(&rh->quiesced_regions)) {
543 reg = list_entry(rh->quiesced_regions.next,
544 struct region, list);
545 list_del_init(®->list); /* remove from the quiesced list */
547 spin_unlock_irq(&rh->region_lock);
552 /* FIXME: success ignored for now */
553 static void rh_recovery_end(struct region *reg, int success)
555 struct region_hash *rh = reg->rh;
557 spin_lock_irq(&rh->region_lock);
558 list_add(®->list, ®->rh->recovered_regions);
559 spin_unlock_irq(&rh->region_lock);
564 static void rh_flush(struct region_hash *rh)
566 rh->log->type->flush(rh->log);
569 static void rh_delay(struct region_hash *rh, struct bio *bio)
573 read_lock(&rh->hash_lock);
574 reg = __rh_find(rh, bio_to_region(rh, bio));
575 bio_list_add(®->delayed_bios, bio);
576 read_unlock(&rh->hash_lock);
579 static void rh_stop_recovery(struct region_hash *rh)
583 /* wait for any recovering regions */
584 for (i = 0; i < MAX_RECOVERY; i++)
585 down(&rh->recovery_count);
588 static void rh_start_recovery(struct region_hash *rh)
592 for (i = 0; i < MAX_RECOVERY; i++)
593 up(&rh->recovery_count);
599 * Every mirror should look like this one.
601 #define DEFAULT_MIRROR 0
604 * This is yucky. We squirrel the mirror_set struct away inside
605 * bi_next for write buffers. This is safe since the bh
606 * doesn't get submitted to the lower levels of block layer.
608 static struct mirror_set *bio_get_ms(struct bio *bio)
610 return (struct mirror_set *) bio->bi_next;
613 static void bio_set_ms(struct bio *bio, struct mirror_set *ms)
615 bio->bi_next = (struct bio *) ms;
618 /*-----------------------------------------------------------------
621 * When a mirror is first activated we may find that some regions
622 * are in the no-sync state. We have to recover these by
623 * recopying from the default mirror to all the others.
624 *---------------------------------------------------------------*/
625 static void recovery_complete(int read_err, unsigned int write_err,
628 struct region *reg = (struct region *) context;
630 /* FIXME: better error handling */
631 rh_recovery_end(reg, !(read_err || write_err));
634 static int recover(struct mirror_set *ms, struct region *reg)
638 struct io_region from, to[KCOPYD_MAX_REGIONS], *dest;
640 unsigned long flags = 0;
642 /* fill in the source */
643 m = ms->default_mirror;
644 from.bdev = m->dev->bdev;
645 from.sector = m->offset + region_to_sector(reg->rh, reg->key);
646 if (reg->key == (ms->nr_regions - 1)) {
648 * The final region may be smaller than
651 from.count = ms->ti->len & (reg->rh->region_size - 1);
653 from.count = reg->rh->region_size;
655 from.count = reg->rh->region_size;
657 /* fill in the destinations */
658 for (i = 0, dest = to; i < ms->nr_mirrors; i++) {
659 if (&ms->mirror[i] == ms->default_mirror)
663 dest->bdev = m->dev->bdev;
664 dest->sector = m->offset + region_to_sector(reg->rh, reg->key);
665 dest->count = from.count;
670 set_bit(KCOPYD_IGNORE_ERROR, &flags);
671 r = kcopyd_copy(ms->kcopyd_client, &from, ms->nr_mirrors - 1, to, flags,
672 recovery_complete, reg);
677 static void do_recovery(struct mirror_set *ms)
681 struct dirty_log *log = ms->rh.log;
684 * Start quiescing some regions.
686 rh_recovery_prepare(&ms->rh);
689 * Copy any already quiesced regions.
691 while ((reg = rh_recovery_start(&ms->rh))) {
692 r = recover(ms, reg);
694 rh_recovery_end(reg, 0);
698 * Update the in sync flag.
701 (log->type->get_sync_count(log) == ms->nr_regions)) {
702 /* the sync is complete */
703 dm_table_event(ms->ti->table);
708 /*-----------------------------------------------------------------
710 *---------------------------------------------------------------*/
711 static struct mirror *choose_mirror(struct mirror_set *ms, sector_t sector)
713 /* FIXME: add read balancing */
714 return ms->default_mirror;
718 * remap a buffer to a particular mirror.
720 static void map_bio(struct mirror_set *ms, struct mirror *m, struct bio *bio)
722 bio->bi_bdev = m->dev->bdev;
723 bio->bi_sector = m->offset + (bio->bi_sector - ms->ti->begin);
726 static void do_reads(struct mirror_set *ms, struct bio_list *reads)
732 while ((bio = bio_list_pop(reads))) {
733 region = bio_to_region(&ms->rh, bio);
736 * We can only read balance if the region is in sync.
738 if (rh_in_sync(&ms->rh, region, 0))
739 m = choose_mirror(ms, bio->bi_sector);
741 m = ms->default_mirror;
744 generic_make_request(bio);
748 /*-----------------------------------------------------------------
751 * We do different things with the write io depending on the
752 * state of the region that it's in:
754 * SYNC: increment pending, use kcopyd to write to *all* mirrors
755 * RECOVERING: delay the io until recovery completes
756 * NOSYNC: increment pending, just write to the default mirror
757 *---------------------------------------------------------------*/
758 static void write_callback(unsigned long error, void *context)
762 struct bio *bio = (struct bio *) context;
763 struct mirror_set *ms;
765 ms = bio_get_ms(bio);
766 bio_set_ms(bio, NULL);
769 * NOTE: We don't decrement the pending count here,
770 * instead it is done by the targets endio function.
771 * This way we handle both writes to SYNC and NOSYNC
772 * regions with the same code.
777 * only error the io if all mirrors failed.
781 for (i = 0; i < ms->nr_mirrors; i++)
782 if (!test_bit(i, &error)) {
787 bio_endio(bio, bio->bi_size, 0);
790 static void do_write(struct mirror_set *ms, struct bio *bio)
793 struct io_region io[KCOPYD_MAX_REGIONS+1];
796 for (i = 0; i < ms->nr_mirrors; i++) {
799 io[i].bdev = m->dev->bdev;
800 io[i].sector = m->offset + (bio->bi_sector - ms->ti->begin);
801 io[i].count = bio->bi_size >> 9;
805 dm_io_async_bvec(ms->nr_mirrors, io, WRITE,
806 bio->bi_io_vec + bio->bi_idx,
807 write_callback, bio);
810 static void do_writes(struct mirror_set *ms, struct bio_list *writes)
814 struct bio_list sync, nosync, recover, *this_list = NULL;
820 * Classify each write.
822 bio_list_init(&sync);
823 bio_list_init(&nosync);
824 bio_list_init(&recover);
826 while ((bio = bio_list_pop(writes))) {
827 state = rh_state(&ms->rh, bio_to_region(&ms->rh, bio), 1);
839 this_list = &recover;
843 bio_list_add(this_list, bio);
847 * Increment the pending counts for any regions that will
848 * be written to (writes to recover regions are going to
851 rh_inc_pending(&ms->rh, &sync);
852 rh_inc_pending(&ms->rh, &nosync);
858 while ((bio = bio_list_pop(&sync)))
861 while ((bio = bio_list_pop(&recover)))
862 rh_delay(&ms->rh, bio);
864 while ((bio = bio_list_pop(&nosync))) {
865 map_bio(ms, ms->default_mirror, bio);
866 generic_make_request(bio);
870 /*-----------------------------------------------------------------
872 *---------------------------------------------------------------*/
873 static LIST_HEAD(_mirror_sets);
874 static DECLARE_RWSEM(_mirror_sets_lock);
876 static void do_mirror(struct mirror_set *ms)
878 struct bio_list reads, writes;
880 spin_lock(&ms->lock);
883 bio_list_init(&ms->reads);
884 bio_list_init(&ms->writes);
885 spin_unlock(&ms->lock);
887 rh_update_states(&ms->rh);
889 do_reads(ms, &reads);
890 do_writes(ms, &writes);
893 static void do_work(struct work_struct *ignored)
895 struct mirror_set *ms;
897 down_read(&_mirror_sets_lock);
898 list_for_each_entry (ms, &_mirror_sets, list)
900 up_read(&_mirror_sets_lock);
903 /*-----------------------------------------------------------------
905 *---------------------------------------------------------------*/
906 static struct mirror_set *alloc_context(unsigned int nr_mirrors,
907 uint32_t region_size,
908 struct dm_target *ti,
909 struct dirty_log *dl)
912 struct mirror_set *ms = NULL;
914 if (array_too_big(sizeof(*ms), sizeof(ms->mirror[0]), nr_mirrors))
917 len = sizeof(*ms) + (sizeof(ms->mirror[0]) * nr_mirrors);
919 ms = kmalloc(len, GFP_KERNEL);
921 ti->error = "Cannot allocate mirror context";
926 spin_lock_init(&ms->lock);
929 ms->nr_mirrors = nr_mirrors;
930 ms->nr_regions = dm_sector_div_up(ti->len, region_size);
932 ms->default_mirror = &ms->mirror[DEFAULT_MIRROR];
934 if (rh_init(&ms->rh, ms, dl, region_size, ms->nr_regions)) {
935 ti->error = "Error creating dirty region hash";
943 static void free_context(struct mirror_set *ms, struct dm_target *ti,
947 dm_put_device(ti, ms->mirror[m].dev);
953 static inline int _check_region_size(struct dm_target *ti, uint32_t size)
955 return !(size % (PAGE_SIZE >> 9) || (size & (size - 1)) ||
959 static int get_mirror(struct mirror_set *ms, struct dm_target *ti,
960 unsigned int mirror, char **argv)
962 unsigned long long offset;
964 if (sscanf(argv[1], "%llu", &offset) != 1) {
965 ti->error = "Invalid offset";
969 if (dm_get_device(ti, argv[0], offset, ti->len,
970 dm_table_get_mode(ti->table),
971 &ms->mirror[mirror].dev)) {
972 ti->error = "Device lookup failure";
976 ms->mirror[mirror].offset = offset;
981 static int add_mirror_set(struct mirror_set *ms)
983 down_write(&_mirror_sets_lock);
984 list_add_tail(&ms->list, &_mirror_sets);
985 up_write(&_mirror_sets_lock);
991 static void del_mirror_set(struct mirror_set *ms)
993 down_write(&_mirror_sets_lock);
995 up_write(&_mirror_sets_lock);
999 * Create dirty log: log_type #log_params <log_params>
1001 static struct dirty_log *create_dirty_log(struct dm_target *ti,
1002 unsigned int argc, char **argv,
1003 unsigned int *args_used)
1005 unsigned int param_count;
1006 struct dirty_log *dl;
1009 ti->error = "Insufficient mirror log arguments";
1013 if (sscanf(argv[1], "%u", ¶m_count) != 1) {
1014 ti->error = "Invalid mirror log argument count";
1018 *args_used = 2 + param_count;
1020 if (argc < *args_used) {
1021 ti->error = "Insufficient mirror log arguments";
1025 dl = dm_create_dirty_log(argv[0], ti, param_count, argv + 2);
1027 ti->error = "Error creating mirror dirty log";
1031 if (!_check_region_size(ti, dl->type->get_region_size(dl))) {
1032 ti->error = "Invalid region size";
1033 dm_destroy_dirty_log(dl);
1041 * Construct a mirror mapping:
1043 * log_type #log_params <log_params>
1044 * #mirrors [mirror_path offset]{2,}
1046 * log_type is "core" or "disk"
1047 * #log_params is between 1 and 3
1049 #define DM_IO_PAGES 64
1050 static int mirror_ctr(struct dm_target *ti, unsigned int argc, char **argv)
1053 unsigned int nr_mirrors, m, args_used;
1054 struct mirror_set *ms;
1055 struct dirty_log *dl;
1057 dl = create_dirty_log(ti, argc, argv, &args_used);
1064 if (!argc || sscanf(argv[0], "%u", &nr_mirrors) != 1 ||
1065 nr_mirrors < 2 || nr_mirrors > KCOPYD_MAX_REGIONS + 1) {
1066 ti->error = "Invalid number of mirrors";
1067 dm_destroy_dirty_log(dl);
1073 if (argc != nr_mirrors * 2) {
1074 ti->error = "Wrong number of mirror arguments";
1075 dm_destroy_dirty_log(dl);
1079 ms = alloc_context(nr_mirrors, dl->type->get_region_size(dl), ti, dl);
1081 dm_destroy_dirty_log(dl);
1085 /* Get the mirror parameter sets */
1086 for (m = 0; m < nr_mirrors; m++) {
1087 r = get_mirror(ms, ti, m, argv);
1089 free_context(ms, ti, m);
1097 ti->split_io = ms->rh.region_size;
1099 r = kcopyd_client_create(DM_IO_PAGES, &ms->kcopyd_client);
1101 free_context(ms, ti, ms->nr_mirrors);
1109 static void mirror_dtr(struct dm_target *ti)
1111 struct mirror_set *ms = (struct mirror_set *) ti->private;
1114 kcopyd_client_destroy(ms->kcopyd_client);
1115 free_context(ms, ti, ms->nr_mirrors);
1118 static void queue_bio(struct mirror_set *ms, struct bio *bio, int rw)
1120 int should_wake = 0;
1121 struct bio_list *bl;
1123 bl = (rw == WRITE) ? &ms->writes : &ms->reads;
1124 spin_lock(&ms->lock);
1125 should_wake = !(bl->head);
1126 bio_list_add(bl, bio);
1127 spin_unlock(&ms->lock);
1134 * Mirror mapping function
1136 static int mirror_map(struct dm_target *ti, struct bio *bio,
1137 union map_info *map_context)
1139 int r, rw = bio_rw(bio);
1141 struct mirror_set *ms = ti->private;
1143 map_context->ll = bio_to_region(&ms->rh, bio);
1146 queue_bio(ms, bio, rw);
1147 return DM_MAPIO_SUBMITTED;
1150 r = ms->rh.log->type->in_sync(ms->rh.log,
1151 bio_to_region(&ms->rh, bio), 0);
1152 if (r < 0 && r != -EWOULDBLOCK)
1155 if (r == -EWOULDBLOCK) /* FIXME: ugly */
1156 r = DM_MAPIO_SUBMITTED;
1159 * We don't want to fast track a recovery just for a read
1160 * ahead. So we just let it silently fail.
1161 * FIXME: get rid of this.
1163 if (!r && rw == READA)
1167 /* Pass this io over to the daemon */
1168 queue_bio(ms, bio, rw);
1169 return DM_MAPIO_SUBMITTED;
1172 m = choose_mirror(ms, bio->bi_sector);
1176 map_bio(ms, m, bio);
1177 return DM_MAPIO_REMAPPED;
1180 static int mirror_end_io(struct dm_target *ti, struct bio *bio,
1181 int error, union map_info *map_context)
1183 int rw = bio_rw(bio);
1184 struct mirror_set *ms = (struct mirror_set *) ti->private;
1185 region_t region = map_context->ll;
1188 * We need to dec pending if this was a write.
1191 rh_dec(&ms->rh, region);
1196 static void mirror_postsuspend(struct dm_target *ti)
1198 struct mirror_set *ms = (struct mirror_set *) ti->private;
1199 struct dirty_log *log = ms->rh.log;
1201 rh_stop_recovery(&ms->rh);
1203 /* Wait for all I/O we generated to complete */
1204 wait_event(_kmirrord_recovery_stopped,
1205 !atomic_read(&ms->rh.recovery_in_flight));
1207 if (log->type->suspend && log->type->suspend(log))
1208 /* FIXME: need better error handling */
1209 DMWARN("log suspend failed");
1212 static void mirror_resume(struct dm_target *ti)
1214 struct mirror_set *ms = (struct mirror_set *) ti->private;
1215 struct dirty_log *log = ms->rh.log;
1216 if (log->type->resume && log->type->resume(log))
1217 /* FIXME: need better error handling */
1218 DMWARN("log resume failed");
1219 rh_start_recovery(&ms->rh);
1222 static int mirror_status(struct dm_target *ti, status_type_t type,
1223 char *result, unsigned int maxlen)
1226 struct mirror_set *ms = (struct mirror_set *) ti->private;
1228 sz = ms->rh.log->type->status(ms->rh.log, type, result, maxlen);
1231 case STATUSTYPE_INFO:
1232 DMEMIT("%d ", ms->nr_mirrors);
1233 for (m = 0; m < ms->nr_mirrors; m++)
1234 DMEMIT("%s ", ms->mirror[m].dev->name);
1237 (unsigned long long)ms->rh.log->type->
1238 get_sync_count(ms->rh.log),
1239 (unsigned long long)ms->nr_regions);
1242 case STATUSTYPE_TABLE:
1243 DMEMIT("%d", ms->nr_mirrors);
1244 for (m = 0; m < ms->nr_mirrors; m++)
1245 DMEMIT(" %s %llu", ms->mirror[m].dev->name,
1246 (unsigned long long)ms->mirror[m].offset);
1252 static struct target_type mirror_target = {
1254 .version = {1, 0, 2},
1255 .module = THIS_MODULE,
1259 .end_io = mirror_end_io,
1260 .postsuspend = mirror_postsuspend,
1261 .resume = mirror_resume,
1262 .status = mirror_status,
1265 static int __init dm_mirror_init(void)
1269 r = dm_dirty_log_init();
1273 _kmirrord_wq = create_singlethread_workqueue("kmirrord");
1274 if (!_kmirrord_wq) {
1275 DMERR("couldn't start kmirrord");
1276 dm_dirty_log_exit();
1279 INIT_WORK(&_kmirrord_work, do_work);
1281 r = dm_register_target(&mirror_target);
1283 DMERR("%s: Failed to register mirror target",
1284 mirror_target.name);
1285 dm_dirty_log_exit();
1286 destroy_workqueue(_kmirrord_wq);
1292 static void __exit dm_mirror_exit(void)
1296 r = dm_unregister_target(&mirror_target);
1298 DMERR("%s: unregister failed %d", mirror_target.name, r);
1300 destroy_workqueue(_kmirrord_wq);
1301 dm_dirty_log_exit();
1305 module_init(dm_mirror_init);
1306 module_exit(dm_mirror_exit);
1308 MODULE_DESCRIPTION(DM_NAME " mirror target");
1309 MODULE_AUTHOR("Joe Thornber");
1310 MODULE_LICENSE("GPL");