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"
24 #define DM_IO_PAGES 64
26 #define DM_RAID1_HANDLE_ERRORS 0x01
28 static DECLARE_WAIT_QUEUE_HEAD(_kmirrord_recovery_stopped);
30 /*-----------------------------------------------------------------
33 * The mirror splits itself up into discrete regions. Each
34 * region can be in one of three states: clean, dirty,
35 * nosync. There is no need to put clean regions in the hash.
37 * In addition to being present in the hash table a region _may_
38 * be present on one of three lists.
40 * clean_regions: Regions on this list have no io pending to
41 * them, they are in sync, we are no longer interested in them,
42 * they are dull. rh_update_states() will remove them from the
45 * quiesced_regions: These regions have been spun down, ready
46 * for recovery. rh_recovery_start() will remove regions from
47 * this list and hand them to kmirrord, which will schedule the
48 * recovery io with kcopyd.
50 * recovered_regions: Regions that kcopyd has successfully
51 * recovered. rh_update_states() will now schedule any delayed
52 * io, up the recovery_count, and remove the region from the
56 * A rw spin lock 'hash_lock' protects just the hash table,
57 * this is never held in write mode from interrupt context,
58 * which I believe means that we only have to disable irqs when
61 * An ordinary spin lock 'region_lock' that protects the three
62 * lists in the region_hash, with the 'state', 'list' and
63 * 'bhs_delayed' fields of the regions. This is used from irq
64 * context, so all other uses will have to suspend local irqs.
65 *---------------------------------------------------------------*/
68 struct mirror_set *ms;
70 unsigned region_shift;
72 /* holds persistent region state */
73 struct dirty_log *log;
77 mempool_t *region_pool;
79 unsigned int nr_buckets;
80 struct list_head *buckets;
82 spinlock_t region_lock;
83 atomic_t recovery_in_flight;
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;
110 /*-----------------------------------------------------------------
111 * Mirror set structures.
112 *---------------------------------------------------------------*/
114 atomic_t error_count;
120 struct dm_target *ti;
121 struct list_head list;
122 struct region_hash rh;
123 struct kcopyd_client *kcopyd_client;
126 spinlock_t lock; /* protects the next two lists */
127 struct bio_list reads;
128 struct bio_list writes;
130 struct dm_io_client *io_client;
136 struct mirror *default_mirror; /* Default mirror */
138 struct workqueue_struct *kmirrord_wq;
139 struct work_struct kmirrord_work;
141 unsigned int nr_mirrors;
142 struct mirror mirror[0];
148 static inline region_t bio_to_region(struct region_hash *rh, struct bio *bio)
150 return (bio->bi_sector - rh->ms->ti->begin) >> rh->region_shift;
153 static inline sector_t region_to_sector(struct region_hash *rh, region_t region)
155 return region << rh->region_shift;
158 static void wake(struct mirror_set *ms)
160 queue_work(ms->kmirrord_wq, &ms->kmirrord_work);
163 /* FIXME move this */
164 static void queue_bio(struct mirror_set *ms, struct bio *bio, int rw);
166 #define MIN_REGIONS 64
167 #define MAX_RECOVERY 1
168 static int rh_init(struct region_hash *rh, struct mirror_set *ms,
169 struct dirty_log *log, uint32_t region_size,
172 unsigned int nr_buckets, max_buckets;
176 * Calculate a suitable number of buckets for our hash
179 max_buckets = nr_regions >> 6;
180 for (nr_buckets = 128u; nr_buckets < max_buckets; nr_buckets <<= 1)
186 rh->region_size = region_size;
187 rh->region_shift = ffs(region_size) - 1;
188 rwlock_init(&rh->hash_lock);
189 rh->mask = nr_buckets - 1;
190 rh->nr_buckets = nr_buckets;
192 rh->buckets = vmalloc(nr_buckets * sizeof(*rh->buckets));
194 DMERR("unable to allocate region hash memory");
198 for (i = 0; i < nr_buckets; i++)
199 INIT_LIST_HEAD(rh->buckets + i);
201 spin_lock_init(&rh->region_lock);
202 sema_init(&rh->recovery_count, 0);
203 atomic_set(&rh->recovery_in_flight, 0);
204 INIT_LIST_HEAD(&rh->clean_regions);
205 INIT_LIST_HEAD(&rh->quiesced_regions);
206 INIT_LIST_HEAD(&rh->recovered_regions);
208 rh->region_pool = mempool_create_kmalloc_pool(MIN_REGIONS,
209 sizeof(struct region));
210 if (!rh->region_pool) {
219 static void rh_exit(struct region_hash *rh)
222 struct region *reg, *nreg;
224 BUG_ON(!list_empty(&rh->quiesced_regions));
225 for (h = 0; h < rh->nr_buckets; h++) {
226 list_for_each_entry_safe(reg, nreg, rh->buckets + h, hash_list) {
227 BUG_ON(atomic_read(®->pending));
228 mempool_free(reg, rh->region_pool);
233 dm_destroy_dirty_log(rh->log);
235 mempool_destroy(rh->region_pool);
239 #define RH_HASH_MULT 2654435387U
241 static inline unsigned int rh_hash(struct region_hash *rh, region_t region)
243 return (unsigned int) ((region * RH_HASH_MULT) >> 12) & rh->mask;
246 static struct region *__rh_lookup(struct region_hash *rh, region_t region)
250 list_for_each_entry (reg, rh->buckets + rh_hash(rh, region), hash_list)
251 if (reg->key == region)
257 static void __rh_insert(struct region_hash *rh, struct region *reg)
259 unsigned int h = rh_hash(rh, reg->key);
260 list_add(®->hash_list, rh->buckets + h);
263 static struct region *__rh_alloc(struct region_hash *rh, region_t region)
265 struct region *reg, *nreg;
267 read_unlock(&rh->hash_lock);
268 nreg = mempool_alloc(rh->region_pool, GFP_ATOMIC);
270 nreg = kmalloc(sizeof(struct region), GFP_NOIO);
271 nreg->state = rh->log->type->in_sync(rh->log, region, 1) ?
272 RH_CLEAN : RH_NOSYNC;
276 INIT_LIST_HEAD(&nreg->list);
278 atomic_set(&nreg->pending, 0);
279 bio_list_init(&nreg->delayed_bios);
280 write_lock_irq(&rh->hash_lock);
282 reg = __rh_lookup(rh, region);
284 /* we lost the race */
285 mempool_free(nreg, rh->region_pool);
288 __rh_insert(rh, nreg);
289 if (nreg->state == RH_CLEAN) {
290 spin_lock(&rh->region_lock);
291 list_add(&nreg->list, &rh->clean_regions);
292 spin_unlock(&rh->region_lock);
296 write_unlock_irq(&rh->hash_lock);
297 read_lock(&rh->hash_lock);
302 static inline struct region *__rh_find(struct region_hash *rh, region_t region)
306 reg = __rh_lookup(rh, region);
308 reg = __rh_alloc(rh, region);
313 static int rh_state(struct region_hash *rh, region_t region, int may_block)
318 read_lock(&rh->hash_lock);
319 reg = __rh_lookup(rh, region);
320 read_unlock(&rh->hash_lock);
326 * The region wasn't in the hash, so we fall back to the
329 r = rh->log->type->in_sync(rh->log, region, may_block);
332 * Any error from the dirty log (eg. -EWOULDBLOCK) gets
333 * taken as a RH_NOSYNC
335 return r == 1 ? RH_CLEAN : RH_NOSYNC;
338 static inline int rh_in_sync(struct region_hash *rh,
339 region_t region, int may_block)
341 int state = rh_state(rh, region, may_block);
342 return state == RH_CLEAN || state == RH_DIRTY;
345 static void dispatch_bios(struct mirror_set *ms, struct bio_list *bio_list)
349 while ((bio = bio_list_pop(bio_list))) {
350 queue_bio(ms, bio, WRITE);
354 static void complete_resync_work(struct region *reg, int success)
356 struct region_hash *rh = reg->rh;
358 rh->log->type->set_region_sync(rh->log, reg->key, success);
359 dispatch_bios(rh->ms, ®->delayed_bios);
360 if (atomic_dec_and_test(&rh->recovery_in_flight))
361 wake_up_all(&_kmirrord_recovery_stopped);
362 up(&rh->recovery_count);
365 static void rh_update_states(struct region_hash *rh)
367 struct region *reg, *next;
370 LIST_HEAD(recovered);
373 * Quickly grab the lists.
375 write_lock_irq(&rh->hash_lock);
376 spin_lock(&rh->region_lock);
377 if (!list_empty(&rh->clean_regions)) {
378 list_splice(&rh->clean_regions, &clean);
379 INIT_LIST_HEAD(&rh->clean_regions);
381 list_for_each_entry (reg, &clean, list) {
382 rh->log->type->clear_region(rh->log, reg->key);
383 list_del(®->hash_list);
387 if (!list_empty(&rh->recovered_regions)) {
388 list_splice(&rh->recovered_regions, &recovered);
389 INIT_LIST_HEAD(&rh->recovered_regions);
391 list_for_each_entry (reg, &recovered, list)
392 list_del(®->hash_list);
394 spin_unlock(&rh->region_lock);
395 write_unlock_irq(&rh->hash_lock);
398 * All the regions on the recovered and clean lists have
399 * now been pulled out of the system, so no need to do
402 list_for_each_entry_safe (reg, next, &recovered, list) {
403 rh->log->type->clear_region(rh->log, reg->key);
404 complete_resync_work(reg, 1);
405 mempool_free(reg, rh->region_pool);
408 if (!list_empty(&recovered))
409 rh->log->type->flush(rh->log);
411 list_for_each_entry_safe (reg, next, &clean, list)
412 mempool_free(reg, rh->region_pool);
415 static void rh_inc(struct region_hash *rh, region_t region)
419 read_lock(&rh->hash_lock);
420 reg = __rh_find(rh, region);
422 spin_lock_irq(&rh->region_lock);
423 atomic_inc(®->pending);
425 if (reg->state == RH_CLEAN) {
426 reg->state = RH_DIRTY;
427 list_del_init(®->list); /* take off the clean list */
428 spin_unlock_irq(&rh->region_lock);
430 rh->log->type->mark_region(rh->log, reg->key);
432 spin_unlock_irq(&rh->region_lock);
435 read_unlock(&rh->hash_lock);
438 static void rh_inc_pending(struct region_hash *rh, struct bio_list *bios)
442 for (bio = bios->head; bio; bio = bio->bi_next)
443 rh_inc(rh, bio_to_region(rh, bio));
446 static void rh_dec(struct region_hash *rh, region_t region)
452 read_lock(&rh->hash_lock);
453 reg = __rh_lookup(rh, region);
454 read_unlock(&rh->hash_lock);
456 spin_lock_irqsave(&rh->region_lock, flags);
457 if (atomic_dec_and_test(®->pending)) {
459 * There is no pending I/O for this region.
460 * We can move the region to corresponding list for next action.
461 * At this point, the region is not yet connected to any list.
463 * If the state is RH_NOSYNC, the region should be kept off
465 * The hash entry for RH_NOSYNC will remain in memory
466 * until the region is recovered or the map is reloaded.
469 /* do nothing for RH_NOSYNC */
470 if (reg->state == RH_RECOVERING) {
471 list_add_tail(®->list, &rh->quiesced_regions);
472 } else if (reg->state == RH_DIRTY) {
473 reg->state = RH_CLEAN;
474 list_add(®->list, &rh->clean_regions);
478 spin_unlock_irqrestore(&rh->region_lock, flags);
485 * Starts quiescing a region in preparation for recovery.
487 static int __rh_recovery_prepare(struct region_hash *rh)
494 * Ask the dirty log what's next.
496 r = rh->log->type->get_resync_work(rh->log, ®ion);
501 * Get this region, and start it quiescing by setting the
504 read_lock(&rh->hash_lock);
505 reg = __rh_find(rh, region);
506 read_unlock(&rh->hash_lock);
508 spin_lock_irq(&rh->region_lock);
509 reg->state = RH_RECOVERING;
511 /* Already quiesced ? */
512 if (atomic_read(®->pending))
513 list_del_init(®->list);
515 list_move(®->list, &rh->quiesced_regions);
517 spin_unlock_irq(&rh->region_lock);
522 static void rh_recovery_prepare(struct region_hash *rh)
524 /* Extra reference to avoid race with rh_stop_recovery */
525 atomic_inc(&rh->recovery_in_flight);
527 while (!down_trylock(&rh->recovery_count)) {
528 atomic_inc(&rh->recovery_in_flight);
529 if (__rh_recovery_prepare(rh) <= 0) {
530 atomic_dec(&rh->recovery_in_flight);
531 up(&rh->recovery_count);
536 /* Drop the extra reference */
537 if (atomic_dec_and_test(&rh->recovery_in_flight))
538 wake_up_all(&_kmirrord_recovery_stopped);
542 * Returns any quiesced regions.
544 static struct region *rh_recovery_start(struct region_hash *rh)
546 struct region *reg = NULL;
548 spin_lock_irq(&rh->region_lock);
549 if (!list_empty(&rh->quiesced_regions)) {
550 reg = list_entry(rh->quiesced_regions.next,
551 struct region, list);
552 list_del_init(®->list); /* remove from the quiesced list */
554 spin_unlock_irq(&rh->region_lock);
559 /* FIXME: success ignored for now */
560 static void rh_recovery_end(struct region *reg, int success)
562 struct region_hash *rh = reg->rh;
564 spin_lock_irq(&rh->region_lock);
565 list_add(®->list, ®->rh->recovered_regions);
566 spin_unlock_irq(&rh->region_lock);
571 static void rh_flush(struct region_hash *rh)
573 rh->log->type->flush(rh->log);
576 static void rh_delay(struct region_hash *rh, struct bio *bio)
580 read_lock(&rh->hash_lock);
581 reg = __rh_find(rh, bio_to_region(rh, bio));
582 bio_list_add(®->delayed_bios, bio);
583 read_unlock(&rh->hash_lock);
586 static void rh_stop_recovery(struct region_hash *rh)
590 /* wait for any recovering regions */
591 for (i = 0; i < MAX_RECOVERY; i++)
592 down(&rh->recovery_count);
595 static void rh_start_recovery(struct region_hash *rh)
599 for (i = 0; i < MAX_RECOVERY; i++)
600 up(&rh->recovery_count);
606 * Every mirror should look like this one.
608 #define DEFAULT_MIRROR 0
611 * This is yucky. We squirrel the mirror_set struct away inside
612 * bi_next for write buffers. This is safe since the bh
613 * doesn't get submitted to the lower levels of block layer.
615 static struct mirror_set *bio_get_ms(struct bio *bio)
617 return (struct mirror_set *) bio->bi_next;
620 static void bio_set_ms(struct bio *bio, struct mirror_set *ms)
622 bio->bi_next = (struct bio *) ms;
625 /*-----------------------------------------------------------------
628 * When a mirror is first activated we may find that some regions
629 * are in the no-sync state. We have to recover these by
630 * recopying from the default mirror to all the others.
631 *---------------------------------------------------------------*/
632 static void recovery_complete(int read_err, unsigned int write_err,
635 struct region *reg = (struct region *) context;
637 /* FIXME: better error handling */
638 rh_recovery_end(reg, !(read_err || write_err));
641 static int recover(struct mirror_set *ms, struct region *reg)
645 struct io_region from, to[KCOPYD_MAX_REGIONS], *dest;
647 unsigned long flags = 0;
649 /* fill in the source */
650 m = ms->default_mirror;
651 from.bdev = m->dev->bdev;
652 from.sector = m->offset + region_to_sector(reg->rh, reg->key);
653 if (reg->key == (ms->nr_regions - 1)) {
655 * The final region may be smaller than
658 from.count = ms->ti->len & (reg->rh->region_size - 1);
660 from.count = reg->rh->region_size;
662 from.count = reg->rh->region_size;
664 /* fill in the destinations */
665 for (i = 0, dest = to; i < ms->nr_mirrors; i++) {
666 if (&ms->mirror[i] == ms->default_mirror)
670 dest->bdev = m->dev->bdev;
671 dest->sector = m->offset + region_to_sector(reg->rh, reg->key);
672 dest->count = from.count;
677 set_bit(KCOPYD_IGNORE_ERROR, &flags);
678 r = kcopyd_copy(ms->kcopyd_client, &from, ms->nr_mirrors - 1, to, flags,
679 recovery_complete, reg);
684 static void do_recovery(struct mirror_set *ms)
688 struct dirty_log *log = ms->rh.log;
691 * Start quiescing some regions.
693 rh_recovery_prepare(&ms->rh);
696 * Copy any already quiesced regions.
698 while ((reg = rh_recovery_start(&ms->rh))) {
699 r = recover(ms, reg);
701 rh_recovery_end(reg, 0);
705 * Update the in sync flag.
708 (log->type->get_sync_count(log) == ms->nr_regions)) {
709 /* the sync is complete */
710 dm_table_event(ms->ti->table);
715 /*-----------------------------------------------------------------
717 *---------------------------------------------------------------*/
718 static struct mirror *choose_mirror(struct mirror_set *ms, sector_t sector)
720 /* FIXME: add read balancing */
721 return ms->default_mirror;
725 * remap a buffer to a particular mirror.
727 static void map_bio(struct mirror_set *ms, struct mirror *m, struct bio *bio)
729 bio->bi_bdev = m->dev->bdev;
730 bio->bi_sector = m->offset + (bio->bi_sector - ms->ti->begin);
733 static void do_reads(struct mirror_set *ms, struct bio_list *reads)
739 while ((bio = bio_list_pop(reads))) {
740 region = bio_to_region(&ms->rh, bio);
743 * We can only read balance if the region is in sync.
745 if (rh_in_sync(&ms->rh, region, 0))
746 m = choose_mirror(ms, bio->bi_sector);
748 m = ms->default_mirror;
751 generic_make_request(bio);
755 /*-----------------------------------------------------------------
758 * We do different things with the write io depending on the
759 * state of the region that it's in:
761 * SYNC: increment pending, use kcopyd to write to *all* mirrors
762 * RECOVERING: delay the io until recovery completes
763 * NOSYNC: increment pending, just write to the default mirror
764 *---------------------------------------------------------------*/
765 static void write_callback(unsigned long error, void *context)
769 struct bio *bio = (struct bio *) context;
770 struct mirror_set *ms;
772 ms = bio_get_ms(bio);
773 bio_set_ms(bio, NULL);
776 * NOTE: We don't decrement the pending count here,
777 * instead it is done by the targets endio function.
778 * This way we handle both writes to SYNC and NOSYNC
779 * regions with the same code.
784 * only error the io if all mirrors failed.
788 for (i = 0; i < ms->nr_mirrors; i++)
789 if (!test_bit(i, &error)) {
794 bio_endio(bio, bio->bi_size, 0);
797 static void do_write(struct mirror_set *ms, struct bio *bio)
800 struct io_region io[KCOPYD_MAX_REGIONS+1];
802 struct dm_io_request io_req = {
804 .mem.type = DM_IO_BVEC,
805 .mem.ptr.bvec = bio->bi_io_vec + bio->bi_idx,
806 .notify.fn = write_callback,
807 .notify.context = bio,
808 .client = ms->io_client,
811 for (i = 0; i < ms->nr_mirrors; i++) {
814 io[i].bdev = m->dev->bdev;
815 io[i].sector = m->offset + (bio->bi_sector - ms->ti->begin);
816 io[i].count = bio->bi_size >> 9;
821 (void) dm_io(&io_req, ms->nr_mirrors, io, NULL);
824 static void do_writes(struct mirror_set *ms, struct bio_list *writes)
828 struct bio_list sync, nosync, recover, *this_list = NULL;
834 * Classify each write.
836 bio_list_init(&sync);
837 bio_list_init(&nosync);
838 bio_list_init(&recover);
840 while ((bio = bio_list_pop(writes))) {
841 state = rh_state(&ms->rh, bio_to_region(&ms->rh, bio), 1);
853 this_list = &recover;
857 bio_list_add(this_list, bio);
861 * Increment the pending counts for any regions that will
862 * be written to (writes to recover regions are going to
865 rh_inc_pending(&ms->rh, &sync);
866 rh_inc_pending(&ms->rh, &nosync);
872 while ((bio = bio_list_pop(&sync)))
875 while ((bio = bio_list_pop(&recover)))
876 rh_delay(&ms->rh, bio);
878 while ((bio = bio_list_pop(&nosync))) {
879 map_bio(ms, ms->default_mirror, bio);
880 generic_make_request(bio);
884 /*-----------------------------------------------------------------
886 *---------------------------------------------------------------*/
887 static void do_mirror(struct work_struct *work)
889 struct mirror_set *ms =container_of(work, struct mirror_set,
891 struct bio_list reads, writes;
893 spin_lock(&ms->lock);
896 bio_list_init(&ms->reads);
897 bio_list_init(&ms->writes);
898 spin_unlock(&ms->lock);
900 rh_update_states(&ms->rh);
902 do_reads(ms, &reads);
903 do_writes(ms, &writes);
906 /*-----------------------------------------------------------------
908 *---------------------------------------------------------------*/
909 static struct mirror_set *alloc_context(unsigned int nr_mirrors,
910 uint32_t region_size,
911 struct dm_target *ti,
912 struct dirty_log *dl)
915 struct mirror_set *ms = NULL;
917 if (array_too_big(sizeof(*ms), sizeof(ms->mirror[0]), nr_mirrors))
920 len = sizeof(*ms) + (sizeof(ms->mirror[0]) * nr_mirrors);
922 ms = kmalloc(len, GFP_KERNEL);
924 ti->error = "Cannot allocate mirror context";
929 spin_lock_init(&ms->lock);
932 ms->nr_mirrors = nr_mirrors;
933 ms->nr_regions = dm_sector_div_up(ti->len, region_size);
935 ms->default_mirror = &ms->mirror[DEFAULT_MIRROR];
937 ms->io_client = dm_io_client_create(DM_IO_PAGES);
938 if (IS_ERR(ms->io_client)) {
939 ti->error = "Error creating dm_io client";
944 if (rh_init(&ms->rh, ms, dl, region_size, ms->nr_regions)) {
945 ti->error = "Error creating dirty region hash";
953 static void free_context(struct mirror_set *ms, struct dm_target *ti,
957 dm_put_device(ti, ms->mirror[m].dev);
959 dm_io_client_destroy(ms->io_client);
964 static inline int _check_region_size(struct dm_target *ti, uint32_t size)
966 return !(size % (PAGE_SIZE >> 9) || (size & (size - 1)) ||
970 static int get_mirror(struct mirror_set *ms, struct dm_target *ti,
971 unsigned int mirror, char **argv)
973 unsigned long long offset;
975 if (sscanf(argv[1], "%llu", &offset) != 1) {
976 ti->error = "Invalid offset";
980 if (dm_get_device(ti, argv[0], offset, ti->len,
981 dm_table_get_mode(ti->table),
982 &ms->mirror[mirror].dev)) {
983 ti->error = "Device lookup failure";
987 ms->mirror[mirror].offset = offset;
993 * Create dirty log: log_type #log_params <log_params>
995 static struct dirty_log *create_dirty_log(struct dm_target *ti,
996 unsigned int argc, char **argv,
997 unsigned int *args_used)
999 unsigned int param_count;
1000 struct dirty_log *dl;
1003 ti->error = "Insufficient mirror log arguments";
1007 if (sscanf(argv[1], "%u", ¶m_count) != 1) {
1008 ti->error = "Invalid mirror log argument count";
1012 *args_used = 2 + param_count;
1014 if (argc < *args_used) {
1015 ti->error = "Insufficient mirror log arguments";
1019 dl = dm_create_dirty_log(argv[0], ti, param_count, argv + 2);
1021 ti->error = "Error creating mirror dirty log";
1025 if (!_check_region_size(ti, dl->type->get_region_size(dl))) {
1026 ti->error = "Invalid region size";
1027 dm_destroy_dirty_log(dl);
1034 static int parse_features(struct mirror_set *ms, unsigned argc, char **argv,
1035 unsigned *args_used)
1037 unsigned num_features;
1038 struct dm_target *ti = ms->ti;
1045 if (sscanf(argv[0], "%u", &num_features) != 1) {
1046 ti->error = "Invalid number of features";
1054 if (num_features > argc) {
1055 ti->error = "Not enough arguments to support feature count";
1059 if (!strcmp("handle_errors", argv[0]))
1060 ms->features |= DM_RAID1_HANDLE_ERRORS;
1062 ti->error = "Unrecognised feature requested";
1072 * Construct a mirror mapping:
1074 * log_type #log_params <log_params>
1075 * #mirrors [mirror_path offset]{2,}
1076 * [#features <features>]
1078 * log_type is "core" or "disk"
1079 * #log_params is between 1 and 3
1081 * If present, features must be "handle_errors".
1083 static int mirror_ctr(struct dm_target *ti, unsigned int argc, char **argv)
1086 unsigned int nr_mirrors, m, args_used;
1087 struct mirror_set *ms;
1088 struct dirty_log *dl;
1090 dl = create_dirty_log(ti, argc, argv, &args_used);
1097 if (!argc || sscanf(argv[0], "%u", &nr_mirrors) != 1 ||
1098 nr_mirrors < 2 || nr_mirrors > KCOPYD_MAX_REGIONS + 1) {
1099 ti->error = "Invalid number of mirrors";
1100 dm_destroy_dirty_log(dl);
1106 if (argc < nr_mirrors * 2) {
1107 ti->error = "Too few mirror arguments";
1108 dm_destroy_dirty_log(dl);
1112 ms = alloc_context(nr_mirrors, dl->type->get_region_size(dl), ti, dl);
1114 dm_destroy_dirty_log(dl);
1118 /* Get the mirror parameter sets */
1119 for (m = 0; m < nr_mirrors; m++) {
1120 r = get_mirror(ms, ti, m, argv);
1122 free_context(ms, ti, m);
1130 ti->split_io = ms->rh.region_size;
1132 ms->kmirrord_wq = create_singlethread_workqueue("kmirrord");
1133 if (!ms->kmirrord_wq) {
1134 DMERR("couldn't start kmirrord");
1135 free_context(ms, ti, m);
1138 INIT_WORK(&ms->kmirrord_work, do_mirror);
1140 r = parse_features(ms, argc, argv, &args_used);
1142 free_context(ms, ti, ms->nr_mirrors);
1150 ti->error = "Too many mirror arguments";
1151 free_context(ms, ti, ms->nr_mirrors);
1155 r = kcopyd_client_create(DM_IO_PAGES, &ms->kcopyd_client);
1157 destroy_workqueue(ms->kmirrord_wq);
1158 free_context(ms, ti, ms->nr_mirrors);
1166 static void mirror_dtr(struct dm_target *ti)
1168 struct mirror_set *ms = (struct mirror_set *) ti->private;
1170 flush_workqueue(ms->kmirrord_wq);
1171 kcopyd_client_destroy(ms->kcopyd_client);
1172 destroy_workqueue(ms->kmirrord_wq);
1173 free_context(ms, ti, ms->nr_mirrors);
1176 static void queue_bio(struct mirror_set *ms, struct bio *bio, int rw)
1178 int should_wake = 0;
1179 struct bio_list *bl;
1181 bl = (rw == WRITE) ? &ms->writes : &ms->reads;
1182 spin_lock(&ms->lock);
1183 should_wake = !(bl->head);
1184 bio_list_add(bl, bio);
1185 spin_unlock(&ms->lock);
1192 * Mirror mapping function
1194 static int mirror_map(struct dm_target *ti, struct bio *bio,
1195 union map_info *map_context)
1197 int r, rw = bio_rw(bio);
1199 struct mirror_set *ms = ti->private;
1201 map_context->ll = bio_to_region(&ms->rh, bio);
1204 queue_bio(ms, bio, rw);
1205 return DM_MAPIO_SUBMITTED;
1208 r = ms->rh.log->type->in_sync(ms->rh.log,
1209 bio_to_region(&ms->rh, bio), 0);
1210 if (r < 0 && r != -EWOULDBLOCK)
1213 if (r == -EWOULDBLOCK) /* FIXME: ugly */
1214 r = DM_MAPIO_SUBMITTED;
1217 * We don't want to fast track a recovery just for a read
1218 * ahead. So we just let it silently fail.
1219 * FIXME: get rid of this.
1221 if (!r && rw == READA)
1225 /* Pass this io over to the daemon */
1226 queue_bio(ms, bio, rw);
1227 return DM_MAPIO_SUBMITTED;
1230 m = choose_mirror(ms, bio->bi_sector);
1234 map_bio(ms, m, bio);
1235 return DM_MAPIO_REMAPPED;
1238 static int mirror_end_io(struct dm_target *ti, struct bio *bio,
1239 int error, union map_info *map_context)
1241 int rw = bio_rw(bio);
1242 struct mirror_set *ms = (struct mirror_set *) ti->private;
1243 region_t region = map_context->ll;
1246 * We need to dec pending if this was a write.
1249 rh_dec(&ms->rh, region);
1254 static void mirror_postsuspend(struct dm_target *ti)
1256 struct mirror_set *ms = (struct mirror_set *) ti->private;
1257 struct dirty_log *log = ms->rh.log;
1259 rh_stop_recovery(&ms->rh);
1261 /* Wait for all I/O we generated to complete */
1262 wait_event(_kmirrord_recovery_stopped,
1263 !atomic_read(&ms->rh.recovery_in_flight));
1265 if (log->type->suspend && log->type->suspend(log))
1266 /* FIXME: need better error handling */
1267 DMWARN("log suspend failed");
1270 static void mirror_resume(struct dm_target *ti)
1272 struct mirror_set *ms = (struct mirror_set *) ti->private;
1273 struct dirty_log *log = ms->rh.log;
1274 if (log->type->resume && log->type->resume(log))
1275 /* FIXME: need better error handling */
1276 DMWARN("log resume failed");
1277 rh_start_recovery(&ms->rh);
1280 static int mirror_status(struct dm_target *ti, status_type_t type,
1281 char *result, unsigned int maxlen)
1283 unsigned int m, sz = 0;
1284 struct mirror_set *ms = (struct mirror_set *) ti->private;
1287 case STATUSTYPE_INFO:
1288 DMEMIT("%d ", ms->nr_mirrors);
1289 for (m = 0; m < ms->nr_mirrors; m++)
1290 DMEMIT("%s ", ms->mirror[m].dev->name);
1293 (unsigned long long)ms->rh.log->type->
1294 get_sync_count(ms->rh.log),
1295 (unsigned long long)ms->nr_regions);
1297 sz = ms->rh.log->type->status(ms->rh.log, type, result, maxlen);
1301 case STATUSTYPE_TABLE:
1302 sz = ms->rh.log->type->status(ms->rh.log, type, result, maxlen);
1304 DMEMIT("%d", ms->nr_mirrors);
1305 for (m = 0; m < ms->nr_mirrors; m++)
1306 DMEMIT(" %s %llu", ms->mirror[m].dev->name,
1307 (unsigned long long)ms->mirror[m].offset);
1309 if (ms->features & DM_RAID1_HANDLE_ERRORS)
1310 DMEMIT(" 1 handle_errors");
1316 static struct target_type mirror_target = {
1318 .version = {1, 0, 3},
1319 .module = THIS_MODULE,
1323 .end_io = mirror_end_io,
1324 .postsuspend = mirror_postsuspend,
1325 .resume = mirror_resume,
1326 .status = mirror_status,
1329 static int __init dm_mirror_init(void)
1333 r = dm_dirty_log_init();
1337 r = dm_register_target(&mirror_target);
1339 DMERR("%s: Failed to register mirror target",
1340 mirror_target.name);
1341 dm_dirty_log_exit();
1347 static void __exit dm_mirror_exit(void)
1351 r = dm_unregister_target(&mirror_target);
1353 DMERR("%s: unregister failed %d", mirror_target.name, r);
1355 dm_dirty_log_exit();
1359 module_init(dm_mirror_init);
1360 module_exit(dm_mirror_exit);
1362 MODULE_DESCRIPTION(DM_NAME " mirror target");
1363 MODULE_AUTHOR("Joe Thornber");
1364 MODULE_LICENSE("GPL");