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
27 #define errors_handled(p) ((p)->features & DM_RAID1_HANDLE_ERRORS)
29 static DECLARE_WAIT_QUEUE_HEAD(_kmirrord_recovery_stopped);
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 atomic_t recovery_in_flight;
85 struct semaphore recovery_count;
86 struct list_head clean_regions;
87 struct list_head quiesced_regions;
88 struct list_head recovered_regions;
89 struct list_head failed_recovered_regions;
100 struct region_hash *rh; /* FIXME: can we get rid of this ? */
104 struct list_head hash_list;
105 struct list_head list;
108 struct bio_list delayed_bios;
112 /*-----------------------------------------------------------------
113 * Mirror set structures.
114 *---------------------------------------------------------------*/
116 atomic_t error_count;
122 struct dm_target *ti;
123 struct list_head list;
124 struct region_hash rh;
125 struct kcopyd_client *kcopyd_client;
128 spinlock_t lock; /* protects the next two lists */
129 struct bio_list reads;
130 struct bio_list writes;
132 struct dm_io_client *io_client;
139 struct mirror *default_mirror; /* Default mirror */
141 struct workqueue_struct *kmirrord_wq;
142 struct work_struct kmirrord_work;
144 unsigned int nr_mirrors;
145 struct mirror mirror[0];
151 static inline region_t bio_to_region(struct region_hash *rh, struct bio *bio)
153 return (bio->bi_sector - rh->ms->ti->begin) >> rh->region_shift;
156 static inline sector_t region_to_sector(struct region_hash *rh, region_t region)
158 return region << rh->region_shift;
161 static void wake(struct mirror_set *ms)
163 queue_work(ms->kmirrord_wq, &ms->kmirrord_work);
166 /* FIXME move this */
167 static void queue_bio(struct mirror_set *ms, struct bio *bio, int rw);
169 #define MIN_REGIONS 64
170 #define MAX_RECOVERY 1
171 static int rh_init(struct region_hash *rh, struct mirror_set *ms,
172 struct dirty_log *log, uint32_t region_size,
175 unsigned int nr_buckets, max_buckets;
179 * Calculate a suitable number of buckets for our hash
182 max_buckets = nr_regions >> 6;
183 for (nr_buckets = 128u; nr_buckets < max_buckets; nr_buckets <<= 1)
189 rh->region_size = region_size;
190 rh->region_shift = ffs(region_size) - 1;
191 rwlock_init(&rh->hash_lock);
192 rh->mask = nr_buckets - 1;
193 rh->nr_buckets = nr_buckets;
195 rh->buckets = vmalloc(nr_buckets * sizeof(*rh->buckets));
197 DMERR("unable to allocate region hash memory");
201 for (i = 0; i < nr_buckets; i++)
202 INIT_LIST_HEAD(rh->buckets + i);
204 spin_lock_init(&rh->region_lock);
205 sema_init(&rh->recovery_count, 0);
206 atomic_set(&rh->recovery_in_flight, 0);
207 INIT_LIST_HEAD(&rh->clean_regions);
208 INIT_LIST_HEAD(&rh->quiesced_regions);
209 INIT_LIST_HEAD(&rh->recovered_regions);
210 INIT_LIST_HEAD(&rh->failed_recovered_regions);
212 rh->region_pool = mempool_create_kmalloc_pool(MIN_REGIONS,
213 sizeof(struct region));
214 if (!rh->region_pool) {
223 static void rh_exit(struct region_hash *rh)
226 struct region *reg, *nreg;
228 BUG_ON(!list_empty(&rh->quiesced_regions));
229 for (h = 0; h < rh->nr_buckets; h++) {
230 list_for_each_entry_safe(reg, nreg, rh->buckets + h, hash_list) {
231 BUG_ON(atomic_read(®->pending));
232 mempool_free(reg, rh->region_pool);
237 dm_destroy_dirty_log(rh->log);
239 mempool_destroy(rh->region_pool);
243 #define RH_HASH_MULT 2654435387U
245 static inline unsigned int rh_hash(struct region_hash *rh, region_t region)
247 return (unsigned int) ((region * RH_HASH_MULT) >> 12) & rh->mask;
250 static struct region *__rh_lookup(struct region_hash *rh, region_t region)
254 list_for_each_entry (reg, rh->buckets + rh_hash(rh, region), hash_list)
255 if (reg->key == region)
261 static void __rh_insert(struct region_hash *rh, struct region *reg)
263 unsigned int h = rh_hash(rh, reg->key);
264 list_add(®->hash_list, rh->buckets + h);
267 static struct region *__rh_alloc(struct region_hash *rh, region_t region)
269 struct region *reg, *nreg;
271 read_unlock(&rh->hash_lock);
272 nreg = mempool_alloc(rh->region_pool, GFP_ATOMIC);
274 nreg = kmalloc(sizeof(struct region), GFP_NOIO);
275 nreg->state = rh->log->type->in_sync(rh->log, region, 1) ?
276 RH_CLEAN : RH_NOSYNC;
280 INIT_LIST_HEAD(&nreg->list);
282 atomic_set(&nreg->pending, 0);
283 bio_list_init(&nreg->delayed_bios);
284 write_lock_irq(&rh->hash_lock);
286 reg = __rh_lookup(rh, region);
288 /* we lost the race */
289 mempool_free(nreg, rh->region_pool);
292 __rh_insert(rh, nreg);
293 if (nreg->state == RH_CLEAN) {
294 spin_lock(&rh->region_lock);
295 list_add(&nreg->list, &rh->clean_regions);
296 spin_unlock(&rh->region_lock);
300 write_unlock_irq(&rh->hash_lock);
301 read_lock(&rh->hash_lock);
306 static inline struct region *__rh_find(struct region_hash *rh, region_t region)
310 reg = __rh_lookup(rh, region);
312 reg = __rh_alloc(rh, region);
317 static int rh_state(struct region_hash *rh, region_t region, int may_block)
322 read_lock(&rh->hash_lock);
323 reg = __rh_lookup(rh, region);
324 read_unlock(&rh->hash_lock);
330 * The region wasn't in the hash, so we fall back to the
333 r = rh->log->type->in_sync(rh->log, region, may_block);
336 * Any error from the dirty log (eg. -EWOULDBLOCK) gets
337 * taken as a RH_NOSYNC
339 return r == 1 ? RH_CLEAN : RH_NOSYNC;
342 static inline int rh_in_sync(struct region_hash *rh,
343 region_t region, int may_block)
345 int state = rh_state(rh, region, may_block);
346 return state == RH_CLEAN || state == RH_DIRTY;
349 static void dispatch_bios(struct mirror_set *ms, struct bio_list *bio_list)
353 while ((bio = bio_list_pop(bio_list))) {
354 queue_bio(ms, bio, WRITE);
358 static void complete_resync_work(struct region *reg, int success)
360 struct region_hash *rh = reg->rh;
362 rh->log->type->set_region_sync(rh->log, reg->key, success);
363 dispatch_bios(rh->ms, ®->delayed_bios);
364 if (atomic_dec_and_test(&rh->recovery_in_flight))
365 wake_up_all(&_kmirrord_recovery_stopped);
366 up(&rh->recovery_count);
369 static void rh_update_states(struct region_hash *rh)
371 struct region *reg, *next;
374 LIST_HEAD(recovered);
375 LIST_HEAD(failed_recovered);
378 * Quickly grab the lists.
380 write_lock_irq(&rh->hash_lock);
381 spin_lock(&rh->region_lock);
382 if (!list_empty(&rh->clean_regions)) {
383 list_splice(&rh->clean_regions, &clean);
384 INIT_LIST_HEAD(&rh->clean_regions);
386 list_for_each_entry(reg, &clean, list)
387 list_del(®->hash_list);
390 if (!list_empty(&rh->recovered_regions)) {
391 list_splice(&rh->recovered_regions, &recovered);
392 INIT_LIST_HEAD(&rh->recovered_regions);
394 list_for_each_entry (reg, &recovered, list)
395 list_del(®->hash_list);
398 if (!list_empty(&rh->failed_recovered_regions)) {
399 list_splice(&rh->failed_recovered_regions, &failed_recovered);
400 INIT_LIST_HEAD(&rh->failed_recovered_regions);
402 list_for_each_entry(reg, &failed_recovered, list)
403 list_del(®->hash_list);
406 spin_unlock(&rh->region_lock);
407 write_unlock_irq(&rh->hash_lock);
410 * All the regions on the recovered and clean lists have
411 * now been pulled out of the system, so no need to do
414 list_for_each_entry_safe (reg, next, &recovered, list) {
415 rh->log->type->clear_region(rh->log, reg->key);
416 complete_resync_work(reg, 1);
417 mempool_free(reg, rh->region_pool);
420 list_for_each_entry_safe(reg, next, &failed_recovered, list) {
421 complete_resync_work(reg, errors_handled(rh->ms) ? 0 : 1);
422 mempool_free(reg, rh->region_pool);
425 list_for_each_entry_safe(reg, next, &clean, list) {
426 rh->log->type->clear_region(rh->log, reg->key);
427 mempool_free(reg, rh->region_pool);
430 rh->log->type->flush(rh->log);
433 static void rh_inc(struct region_hash *rh, region_t region)
437 read_lock(&rh->hash_lock);
438 reg = __rh_find(rh, region);
440 spin_lock_irq(&rh->region_lock);
441 atomic_inc(®->pending);
443 if (reg->state == RH_CLEAN) {
444 reg->state = RH_DIRTY;
445 list_del_init(®->list); /* take off the clean list */
446 spin_unlock_irq(&rh->region_lock);
448 rh->log->type->mark_region(rh->log, reg->key);
450 spin_unlock_irq(&rh->region_lock);
453 read_unlock(&rh->hash_lock);
456 static void rh_inc_pending(struct region_hash *rh, struct bio_list *bios)
460 for (bio = bios->head; bio; bio = bio->bi_next)
461 rh_inc(rh, bio_to_region(rh, bio));
464 static void rh_dec(struct region_hash *rh, region_t region)
470 read_lock(&rh->hash_lock);
471 reg = __rh_lookup(rh, region);
472 read_unlock(&rh->hash_lock);
474 spin_lock_irqsave(&rh->region_lock, flags);
475 if (atomic_dec_and_test(®->pending)) {
477 * There is no pending I/O for this region.
478 * We can move the region to corresponding list for next action.
479 * At this point, the region is not yet connected to any list.
481 * If the state is RH_NOSYNC, the region should be kept off
483 * The hash entry for RH_NOSYNC will remain in memory
484 * until the region is recovered or the map is reloaded.
487 /* do nothing for RH_NOSYNC */
488 if (reg->state == RH_RECOVERING) {
489 list_add_tail(®->list, &rh->quiesced_regions);
490 } else if (reg->state == RH_DIRTY) {
491 reg->state = RH_CLEAN;
492 list_add(®->list, &rh->clean_regions);
496 spin_unlock_irqrestore(&rh->region_lock, flags);
503 * Starts quiescing a region in preparation for recovery.
505 static int __rh_recovery_prepare(struct region_hash *rh)
512 * Ask the dirty log what's next.
514 r = rh->log->type->get_resync_work(rh->log, ®ion);
519 * Get this region, and start it quiescing by setting the
522 read_lock(&rh->hash_lock);
523 reg = __rh_find(rh, region);
524 read_unlock(&rh->hash_lock);
526 spin_lock_irq(&rh->region_lock);
527 reg->state = RH_RECOVERING;
529 /* Already quiesced ? */
530 if (atomic_read(®->pending))
531 list_del_init(®->list);
533 list_move(®->list, &rh->quiesced_regions);
535 spin_unlock_irq(&rh->region_lock);
540 static void rh_recovery_prepare(struct region_hash *rh)
542 /* Extra reference to avoid race with rh_stop_recovery */
543 atomic_inc(&rh->recovery_in_flight);
545 while (!down_trylock(&rh->recovery_count)) {
546 atomic_inc(&rh->recovery_in_flight);
547 if (__rh_recovery_prepare(rh) <= 0) {
548 atomic_dec(&rh->recovery_in_flight);
549 up(&rh->recovery_count);
554 /* Drop the extra reference */
555 if (atomic_dec_and_test(&rh->recovery_in_flight))
556 wake_up_all(&_kmirrord_recovery_stopped);
560 * Returns any quiesced regions.
562 static struct region *rh_recovery_start(struct region_hash *rh)
564 struct region *reg = NULL;
566 spin_lock_irq(&rh->region_lock);
567 if (!list_empty(&rh->quiesced_regions)) {
568 reg = list_entry(rh->quiesced_regions.next,
569 struct region, list);
570 list_del_init(®->list); /* remove from the quiesced list */
572 spin_unlock_irq(&rh->region_lock);
577 static void rh_recovery_end(struct region *reg, int success)
579 struct region_hash *rh = reg->rh;
581 spin_lock_irq(&rh->region_lock);
583 list_add(®->list, ®->rh->recovered_regions);
585 reg->state = RH_NOSYNC;
586 list_add(®->list, ®->rh->failed_recovered_regions);
588 spin_unlock_irq(&rh->region_lock);
593 static int rh_flush(struct region_hash *rh)
595 return rh->log->type->flush(rh->log);
598 static void rh_delay(struct region_hash *rh, struct bio *bio)
602 read_lock(&rh->hash_lock);
603 reg = __rh_find(rh, bio_to_region(rh, bio));
604 bio_list_add(®->delayed_bios, bio);
605 read_unlock(&rh->hash_lock);
608 static void rh_stop_recovery(struct region_hash *rh)
612 /* wait for any recovering regions */
613 for (i = 0; i < MAX_RECOVERY; i++)
614 down(&rh->recovery_count);
617 static void rh_start_recovery(struct region_hash *rh)
621 for (i = 0; i < MAX_RECOVERY; i++)
622 up(&rh->recovery_count);
628 * Every mirror should look like this one.
630 #define DEFAULT_MIRROR 0
633 * This is yucky. We squirrel the mirror_set struct away inside
634 * bi_next for write buffers. This is safe since the bh
635 * doesn't get submitted to the lower levels of block layer.
637 static struct mirror_set *bio_get_ms(struct bio *bio)
639 return (struct mirror_set *) bio->bi_next;
642 static void bio_set_ms(struct bio *bio, struct mirror_set *ms)
644 bio->bi_next = (struct bio *) ms;
647 /*-----------------------------------------------------------------
650 * When a mirror is first activated we may find that some regions
651 * are in the no-sync state. We have to recover these by
652 * recopying from the default mirror to all the others.
653 *---------------------------------------------------------------*/
654 static void recovery_complete(int read_err, unsigned int write_err,
657 struct region *reg = (struct region *) context;
660 /* Read error means the failure of default mirror. */
661 DMERR_LIMIT("Unable to read primary mirror during recovery");
664 DMERR_LIMIT("Write error during recovery (error = 0x%x)",
667 rh_recovery_end(reg, !(read_err || write_err));
670 static int recover(struct mirror_set *ms, struct region *reg)
674 struct io_region from, to[KCOPYD_MAX_REGIONS], *dest;
676 unsigned long flags = 0;
678 /* fill in the source */
679 m = ms->default_mirror;
680 from.bdev = m->dev->bdev;
681 from.sector = m->offset + region_to_sector(reg->rh, reg->key);
682 if (reg->key == (ms->nr_regions - 1)) {
684 * The final region may be smaller than
687 from.count = ms->ti->len & (reg->rh->region_size - 1);
689 from.count = reg->rh->region_size;
691 from.count = reg->rh->region_size;
693 /* fill in the destinations */
694 for (i = 0, dest = to; i < ms->nr_mirrors; i++) {
695 if (&ms->mirror[i] == ms->default_mirror)
699 dest->bdev = m->dev->bdev;
700 dest->sector = m->offset + region_to_sector(reg->rh, reg->key);
701 dest->count = from.count;
706 set_bit(KCOPYD_IGNORE_ERROR, &flags);
707 r = kcopyd_copy(ms->kcopyd_client, &from, ms->nr_mirrors - 1, to, flags,
708 recovery_complete, reg);
713 static void do_recovery(struct mirror_set *ms)
717 struct dirty_log *log = ms->rh.log;
720 * Start quiescing some regions.
722 rh_recovery_prepare(&ms->rh);
725 * Copy any already quiesced regions.
727 while ((reg = rh_recovery_start(&ms->rh))) {
728 r = recover(ms, reg);
730 rh_recovery_end(reg, 0);
734 * Update the in sync flag.
737 (log->type->get_sync_count(log) == ms->nr_regions)) {
738 /* the sync is complete */
739 dm_table_event(ms->ti->table);
744 /*-----------------------------------------------------------------
746 *---------------------------------------------------------------*/
747 static struct mirror *choose_mirror(struct mirror_set *ms, sector_t sector)
749 /* FIXME: add read balancing */
750 return ms->default_mirror;
754 * remap a buffer to a particular mirror.
756 static void map_bio(struct mirror_set *ms, struct mirror *m, struct bio *bio)
758 bio->bi_bdev = m->dev->bdev;
759 bio->bi_sector = m->offset + (bio->bi_sector - ms->ti->begin);
762 static void do_reads(struct mirror_set *ms, struct bio_list *reads)
768 while ((bio = bio_list_pop(reads))) {
769 region = bio_to_region(&ms->rh, bio);
772 * We can only read balance if the region is in sync.
774 if (rh_in_sync(&ms->rh, region, 1))
775 m = choose_mirror(ms, bio->bi_sector);
777 m = ms->default_mirror;
780 generic_make_request(bio);
784 /*-----------------------------------------------------------------
787 * We do different things with the write io depending on the
788 * state of the region that it's in:
790 * SYNC: increment pending, use kcopyd to write to *all* mirrors
791 * RECOVERING: delay the io until recovery completes
792 * NOSYNC: increment pending, just write to the default mirror
793 *---------------------------------------------------------------*/
794 static void write_callback(unsigned long error, void *context)
798 struct bio *bio = (struct bio *) context;
799 struct mirror_set *ms;
801 ms = bio_get_ms(bio);
802 bio_set_ms(bio, NULL);
805 * NOTE: We don't decrement the pending count here,
806 * instead it is done by the targets endio function.
807 * This way we handle both writes to SYNC and NOSYNC
808 * regions with the same code.
813 * only error the io if all mirrors failed.
817 for (i = 0; i < ms->nr_mirrors; i++)
818 if (!test_bit(i, &error)) {
826 static void do_write(struct mirror_set *ms, struct bio *bio)
829 struct io_region io[KCOPYD_MAX_REGIONS+1];
831 struct dm_io_request io_req = {
833 .mem.type = DM_IO_BVEC,
834 .mem.ptr.bvec = bio->bi_io_vec + bio->bi_idx,
835 .notify.fn = write_callback,
836 .notify.context = bio,
837 .client = ms->io_client,
840 for (i = 0; i < ms->nr_mirrors; i++) {
843 io[i].bdev = m->dev->bdev;
844 io[i].sector = m->offset + (bio->bi_sector - ms->ti->begin);
845 io[i].count = bio->bi_size >> 9;
850 (void) dm_io(&io_req, ms->nr_mirrors, io, NULL);
853 static void do_writes(struct mirror_set *ms, struct bio_list *writes)
857 struct bio_list sync, nosync, recover, *this_list = NULL;
863 * Classify each write.
865 bio_list_init(&sync);
866 bio_list_init(&nosync);
867 bio_list_init(&recover);
869 while ((bio = bio_list_pop(writes))) {
870 state = rh_state(&ms->rh, bio_to_region(&ms->rh, bio), 1);
882 this_list = &recover;
886 bio_list_add(this_list, bio);
890 * Increment the pending counts for any regions that will
891 * be written to (writes to recover regions are going to
894 rh_inc_pending(&ms->rh, &sync);
895 rh_inc_pending(&ms->rh, &nosync);
896 ms->log_failure = rh_flush(&ms->rh) ? 1 : 0;
901 if (unlikely(ms->log_failure))
902 while ((bio = bio_list_pop(&sync)))
903 bio_endio(bio, -EIO);
904 else while ((bio = bio_list_pop(&sync)))
907 while ((bio = bio_list_pop(&recover)))
908 rh_delay(&ms->rh, bio);
910 while ((bio = bio_list_pop(&nosync))) {
911 map_bio(ms, ms->default_mirror, bio);
912 generic_make_request(bio);
916 /*-----------------------------------------------------------------
918 *---------------------------------------------------------------*/
919 static void do_mirror(struct work_struct *work)
921 struct mirror_set *ms =container_of(work, struct mirror_set,
923 struct bio_list reads, writes;
925 spin_lock(&ms->lock);
928 bio_list_init(&ms->reads);
929 bio_list_init(&ms->writes);
930 spin_unlock(&ms->lock);
932 rh_update_states(&ms->rh);
934 do_reads(ms, &reads);
935 do_writes(ms, &writes);
938 /*-----------------------------------------------------------------
940 *---------------------------------------------------------------*/
941 static struct mirror_set *alloc_context(unsigned int nr_mirrors,
942 uint32_t region_size,
943 struct dm_target *ti,
944 struct dirty_log *dl)
947 struct mirror_set *ms = NULL;
949 if (array_too_big(sizeof(*ms), sizeof(ms->mirror[0]), nr_mirrors))
952 len = sizeof(*ms) + (sizeof(ms->mirror[0]) * nr_mirrors);
954 ms = kzalloc(len, GFP_KERNEL);
956 ti->error = "Cannot allocate mirror context";
960 spin_lock_init(&ms->lock);
963 ms->nr_mirrors = nr_mirrors;
964 ms->nr_regions = dm_sector_div_up(ti->len, region_size);
966 ms->default_mirror = &ms->mirror[DEFAULT_MIRROR];
968 ms->io_client = dm_io_client_create(DM_IO_PAGES);
969 if (IS_ERR(ms->io_client)) {
970 ti->error = "Error creating dm_io client";
975 if (rh_init(&ms->rh, ms, dl, region_size, ms->nr_regions)) {
976 ti->error = "Error creating dirty region hash";
984 static void free_context(struct mirror_set *ms, struct dm_target *ti,
988 dm_put_device(ti, ms->mirror[m].dev);
990 dm_io_client_destroy(ms->io_client);
995 static inline int _check_region_size(struct dm_target *ti, uint32_t size)
997 return !(size % (PAGE_SIZE >> 9) || (size & (size - 1)) ||
1001 static int get_mirror(struct mirror_set *ms, struct dm_target *ti,
1002 unsigned int mirror, char **argv)
1004 unsigned long long offset;
1006 if (sscanf(argv[1], "%llu", &offset) != 1) {
1007 ti->error = "Invalid offset";
1011 if (dm_get_device(ti, argv[0], offset, ti->len,
1012 dm_table_get_mode(ti->table),
1013 &ms->mirror[mirror].dev)) {
1014 ti->error = "Device lookup failure";
1018 ms->mirror[mirror].offset = offset;
1024 * Create dirty log: log_type #log_params <log_params>
1026 static struct dirty_log *create_dirty_log(struct dm_target *ti,
1027 unsigned int argc, char **argv,
1028 unsigned int *args_used)
1030 unsigned int param_count;
1031 struct dirty_log *dl;
1034 ti->error = "Insufficient mirror log arguments";
1038 if (sscanf(argv[1], "%u", ¶m_count) != 1) {
1039 ti->error = "Invalid mirror log argument count";
1043 *args_used = 2 + param_count;
1045 if (argc < *args_used) {
1046 ti->error = "Insufficient mirror log arguments";
1050 dl = dm_create_dirty_log(argv[0], ti, param_count, argv + 2);
1052 ti->error = "Error creating mirror dirty log";
1056 if (!_check_region_size(ti, dl->type->get_region_size(dl))) {
1057 ti->error = "Invalid region size";
1058 dm_destroy_dirty_log(dl);
1065 static int parse_features(struct mirror_set *ms, unsigned argc, char **argv,
1066 unsigned *args_used)
1068 unsigned num_features;
1069 struct dm_target *ti = ms->ti;
1076 if (sscanf(argv[0], "%u", &num_features) != 1) {
1077 ti->error = "Invalid number of features";
1085 if (num_features > argc) {
1086 ti->error = "Not enough arguments to support feature count";
1090 if (!strcmp("handle_errors", argv[0]))
1091 ms->features |= DM_RAID1_HANDLE_ERRORS;
1093 ti->error = "Unrecognised feature requested";
1103 * Construct a mirror mapping:
1105 * log_type #log_params <log_params>
1106 * #mirrors [mirror_path offset]{2,}
1107 * [#features <features>]
1109 * log_type is "core" or "disk"
1110 * #log_params is between 1 and 3
1112 * If present, features must be "handle_errors".
1114 static int mirror_ctr(struct dm_target *ti, unsigned int argc, char **argv)
1117 unsigned int nr_mirrors, m, args_used;
1118 struct mirror_set *ms;
1119 struct dirty_log *dl;
1121 dl = create_dirty_log(ti, argc, argv, &args_used);
1128 if (!argc || sscanf(argv[0], "%u", &nr_mirrors) != 1 ||
1129 nr_mirrors < 2 || nr_mirrors > KCOPYD_MAX_REGIONS + 1) {
1130 ti->error = "Invalid number of mirrors";
1131 dm_destroy_dirty_log(dl);
1137 if (argc < nr_mirrors * 2) {
1138 ti->error = "Too few mirror arguments";
1139 dm_destroy_dirty_log(dl);
1143 ms = alloc_context(nr_mirrors, dl->type->get_region_size(dl), ti, dl);
1145 dm_destroy_dirty_log(dl);
1149 /* Get the mirror parameter sets */
1150 for (m = 0; m < nr_mirrors; m++) {
1151 r = get_mirror(ms, ti, m, argv);
1153 free_context(ms, ti, m);
1161 ti->split_io = ms->rh.region_size;
1163 ms->kmirrord_wq = create_singlethread_workqueue("kmirrord");
1164 if (!ms->kmirrord_wq) {
1165 DMERR("couldn't start kmirrord");
1166 free_context(ms, ti, m);
1169 INIT_WORK(&ms->kmirrord_work, do_mirror);
1171 r = parse_features(ms, argc, argv, &args_used);
1173 free_context(ms, ti, ms->nr_mirrors);
1181 * Any read-balancing addition depends on the
1182 * DM_RAID1_HANDLE_ERRORS flag being present.
1183 * This is because the decision to balance depends
1184 * on the sync state of a region. If the above
1185 * flag is not present, we ignore errors; and
1186 * the sync state may be inaccurate.
1190 ti->error = "Too many mirror arguments";
1191 free_context(ms, ti, ms->nr_mirrors);
1195 r = kcopyd_client_create(DM_IO_PAGES, &ms->kcopyd_client);
1197 destroy_workqueue(ms->kmirrord_wq);
1198 free_context(ms, ti, ms->nr_mirrors);
1206 static void mirror_dtr(struct dm_target *ti)
1208 struct mirror_set *ms = (struct mirror_set *) ti->private;
1210 flush_workqueue(ms->kmirrord_wq);
1211 kcopyd_client_destroy(ms->kcopyd_client);
1212 destroy_workqueue(ms->kmirrord_wq);
1213 free_context(ms, ti, ms->nr_mirrors);
1216 static void queue_bio(struct mirror_set *ms, struct bio *bio, int rw)
1218 int should_wake = 0;
1219 struct bio_list *bl;
1221 bl = (rw == WRITE) ? &ms->writes : &ms->reads;
1222 spin_lock(&ms->lock);
1223 should_wake = !(bl->head);
1224 bio_list_add(bl, bio);
1225 spin_unlock(&ms->lock);
1232 * Mirror mapping function
1234 static int mirror_map(struct dm_target *ti, struct bio *bio,
1235 union map_info *map_context)
1237 int r, rw = bio_rw(bio);
1239 struct mirror_set *ms = ti->private;
1241 map_context->ll = bio_to_region(&ms->rh, bio);
1244 queue_bio(ms, bio, rw);
1245 return DM_MAPIO_SUBMITTED;
1248 r = ms->rh.log->type->in_sync(ms->rh.log,
1249 bio_to_region(&ms->rh, bio), 0);
1250 if (r < 0 && r != -EWOULDBLOCK)
1253 if (r == -EWOULDBLOCK) /* FIXME: ugly */
1254 r = DM_MAPIO_SUBMITTED;
1257 * We don't want to fast track a recovery just for a read
1258 * ahead. So we just let it silently fail.
1259 * FIXME: get rid of this.
1261 if (!r && rw == READA)
1265 /* Pass this io over to the daemon */
1266 queue_bio(ms, bio, rw);
1267 return DM_MAPIO_SUBMITTED;
1270 m = choose_mirror(ms, bio->bi_sector);
1274 map_bio(ms, m, bio);
1275 return DM_MAPIO_REMAPPED;
1278 static int mirror_end_io(struct dm_target *ti, struct bio *bio,
1279 int error, union map_info *map_context)
1281 int rw = bio_rw(bio);
1282 struct mirror_set *ms = (struct mirror_set *) ti->private;
1283 region_t region = map_context->ll;
1286 * We need to dec pending if this was a write.
1289 rh_dec(&ms->rh, region);
1294 static void mirror_postsuspend(struct dm_target *ti)
1296 struct mirror_set *ms = (struct mirror_set *) ti->private;
1297 struct dirty_log *log = ms->rh.log;
1299 rh_stop_recovery(&ms->rh);
1301 /* Wait for all I/O we generated to complete */
1302 wait_event(_kmirrord_recovery_stopped,
1303 !atomic_read(&ms->rh.recovery_in_flight));
1305 if (log->type->suspend && log->type->suspend(log))
1306 /* FIXME: need better error handling */
1307 DMWARN("log suspend failed");
1310 static void mirror_resume(struct dm_target *ti)
1312 struct mirror_set *ms = (struct mirror_set *) ti->private;
1313 struct dirty_log *log = ms->rh.log;
1314 if (log->type->resume && log->type->resume(log))
1315 /* FIXME: need better error handling */
1316 DMWARN("log resume failed");
1317 rh_start_recovery(&ms->rh);
1320 static int mirror_status(struct dm_target *ti, status_type_t type,
1321 char *result, unsigned int maxlen)
1323 unsigned int m, sz = 0;
1324 struct mirror_set *ms = (struct mirror_set *) ti->private;
1327 case STATUSTYPE_INFO:
1328 DMEMIT("%d ", ms->nr_mirrors);
1329 for (m = 0; m < ms->nr_mirrors; m++)
1330 DMEMIT("%s ", ms->mirror[m].dev->name);
1332 DMEMIT("%llu/%llu 0 ",
1333 (unsigned long long)ms->rh.log->type->
1334 get_sync_count(ms->rh.log),
1335 (unsigned long long)ms->nr_regions);
1337 sz += ms->rh.log->type->status(ms->rh.log, type, result+sz, maxlen-sz);
1341 case STATUSTYPE_TABLE:
1342 sz = ms->rh.log->type->status(ms->rh.log, type, result, maxlen);
1344 DMEMIT("%d", ms->nr_mirrors);
1345 for (m = 0; m < ms->nr_mirrors; m++)
1346 DMEMIT(" %s %llu", ms->mirror[m].dev->name,
1347 (unsigned long long)ms->mirror[m].offset);
1349 if (ms->features & DM_RAID1_HANDLE_ERRORS)
1350 DMEMIT(" 1 handle_errors");
1356 static struct target_type mirror_target = {
1358 .version = {1, 0, 3},
1359 .module = THIS_MODULE,
1363 .end_io = mirror_end_io,
1364 .postsuspend = mirror_postsuspend,
1365 .resume = mirror_resume,
1366 .status = mirror_status,
1369 static int __init dm_mirror_init(void)
1373 r = dm_dirty_log_init();
1377 r = dm_register_target(&mirror_target);
1379 DMERR("Failed to register mirror target");
1380 dm_dirty_log_exit();
1386 static void __exit dm_mirror_exit(void)
1390 r = dm_unregister_target(&mirror_target);
1392 DMERR("unregister failed %d", r);
1394 dm_dirty_log_exit();
1398 module_init(dm_mirror_init);
1399 module_exit(dm_mirror_exit);
1401 MODULE_DESCRIPTION(DM_NAME " mirror target");
1402 MODULE_AUTHOR("Joe Thornber");
1403 MODULE_LICENSE("GPL");