2 * Copyright (C) 2003 Sistina Software Limited.
3 * Copyright (C) 2005-2008 Red Hat, Inc. All rights reserved.
5 * This file is released under the GPL.
8 #include "dm-bio-list.h"
9 #include "dm-bio-record.h"
11 #include <linux/init.h>
12 #include <linux/mempool.h>
13 #include <linux/module.h>
14 #include <linux/pagemap.h>
15 #include <linux/slab.h>
16 #include <linux/workqueue.h>
17 #include <linux/device-mapper.h>
18 #include <linux/dm-io.h>
19 #include <linux/dm-dirty-log.h>
20 #include <linux/dm-kcopyd.h>
21 #include <linux/dm-region-hash.h>
23 #define DM_MSG_PREFIX "raid1"
25 #define MAX_RECOVERY 1 /* Maximum number of regions recovered in parallel. */
26 #define DM_IO_PAGES 64
27 #define DM_KCOPYD_PAGES 64
29 #define DM_RAID1_HANDLE_ERRORS 0x01
30 #define errors_handled(p) ((p)->features & DM_RAID1_HANDLE_ERRORS)
32 static DECLARE_WAIT_QUEUE_HEAD(_kmirrord_recovery_stopped);
34 /*-----------------------------------------------------------------
35 * Mirror set structures.
36 *---------------------------------------------------------------*/
44 struct mirror_set *ms;
46 unsigned long error_type;
53 struct list_head list;
57 spinlock_t lock; /* protects the lists */
58 struct bio_list reads;
59 struct bio_list writes;
60 struct bio_list failures;
62 struct dm_region_hash *rh;
63 struct dm_kcopyd_client *kcopyd_client;
64 struct dm_io_client *io_client;
65 mempool_t *read_record_pool;
73 atomic_t default_mirror; /* Default mirror */
75 struct workqueue_struct *kmirrord_wq;
76 struct work_struct kmirrord_work;
77 struct timer_list timer;
78 unsigned long timer_pending;
80 struct work_struct trigger_event;
83 struct mirror mirror[0];
86 static void wakeup_mirrord(void *context)
88 struct mirror_set *ms = context;
90 queue_work(ms->kmirrord_wq, &ms->kmirrord_work);
93 static void delayed_wake_fn(unsigned long data)
95 struct mirror_set *ms = (struct mirror_set *) data;
97 clear_bit(0, &ms->timer_pending);
101 static void delayed_wake(struct mirror_set *ms)
103 if (test_and_set_bit(0, &ms->timer_pending))
106 ms->timer.expires = jiffies + HZ / 5;
107 ms->timer.data = (unsigned long) ms;
108 ms->timer.function = delayed_wake_fn;
109 add_timer(&ms->timer);
112 static void wakeup_all_recovery_waiters(void *context)
114 wake_up_all(&_kmirrord_recovery_stopped);
117 static void queue_bio(struct mirror_set *ms, struct bio *bio, int rw)
123 bl = (rw == WRITE) ? &ms->writes : &ms->reads;
124 spin_lock_irqsave(&ms->lock, flags);
125 should_wake = !(bl->head);
126 bio_list_add(bl, bio);
127 spin_unlock_irqrestore(&ms->lock, flags);
133 static void dispatch_bios(void *context, struct bio_list *bio_list)
135 struct mirror_set *ms = context;
138 while ((bio = bio_list_pop(bio_list)))
139 queue_bio(ms, bio, WRITE);
142 #define MIN_READ_RECORDS 20
143 struct dm_raid1_read_record {
145 struct dm_bio_details details;
148 static struct kmem_cache *_dm_raid1_read_record_cache;
151 * Every mirror should look like this one.
153 #define DEFAULT_MIRROR 0
156 * This is yucky. We squirrel the mirror struct away inside
157 * bi_next for read/write buffers. This is safe since the bh
158 * doesn't get submitted to the lower levels of block layer.
160 static struct mirror *bio_get_m(struct bio *bio)
162 return (struct mirror *) bio->bi_next;
165 static void bio_set_m(struct bio *bio, struct mirror *m)
167 bio->bi_next = (struct bio *) m;
170 static struct mirror *get_default_mirror(struct mirror_set *ms)
172 return &ms->mirror[atomic_read(&ms->default_mirror)];
175 static void set_default_mirror(struct mirror *m)
177 struct mirror_set *ms = m->ms;
178 struct mirror *m0 = &(ms->mirror[0]);
180 atomic_set(&ms->default_mirror, m - m0);
184 * @m: mirror device to fail
185 * @error_type: one of the enum's, DM_RAID1_*_ERROR
187 * If errors are being handled, record the type of
188 * error encountered for this device. If this type
189 * of error has already been recorded, we can return;
190 * otherwise, we must signal userspace by triggering
191 * an event. Additionally, if the device is the
192 * primary device, we must choose a new primary, but
193 * only if the mirror is in-sync.
195 * This function must not block.
197 static void fail_mirror(struct mirror *m, enum dm_raid1_error error_type)
199 struct mirror_set *ms = m->ms;
203 * error_count is used for nothing more than a
204 * simple way to tell if a device has encountered
207 atomic_inc(&m->error_count);
209 if (test_and_set_bit(error_type, &m->error_type))
212 if (!errors_handled(ms))
215 if (m != get_default_mirror(ms))
220 * Better to issue requests to same failing device
221 * than to risk returning corrupt data.
223 DMERR("Primary mirror (%s) failed while out-of-sync: "
224 "Reads may fail.", m->dev->name);
228 for (new = ms->mirror; new < ms->mirror + ms->nr_mirrors; new++)
229 if (!atomic_read(&new->error_count)) {
230 set_default_mirror(new);
234 if (unlikely(new == ms->mirror + ms->nr_mirrors))
235 DMWARN("All sides of mirror have failed.");
238 schedule_work(&ms->trigger_event);
241 /*-----------------------------------------------------------------
244 * When a mirror is first activated we may find that some regions
245 * are in the no-sync state. We have to recover these by
246 * recopying from the default mirror to all the others.
247 *---------------------------------------------------------------*/
248 static void recovery_complete(int read_err, unsigned long write_err,
251 struct dm_region *reg = context;
252 struct mirror_set *ms = dm_rh_region_context(reg);
256 /* Read error means the failure of default mirror. */
257 DMERR_LIMIT("Unable to read primary mirror during recovery");
258 fail_mirror(get_default_mirror(ms), DM_RAID1_SYNC_ERROR);
262 DMERR_LIMIT("Write error during recovery (error = 0x%lx)",
265 * Bits correspond to devices (excluding default mirror).
266 * The default mirror cannot change during recovery.
268 for (m = 0; m < ms->nr_mirrors; m++) {
269 if (&ms->mirror[m] == get_default_mirror(ms))
271 if (test_bit(bit, &write_err))
272 fail_mirror(ms->mirror + m,
273 DM_RAID1_SYNC_ERROR);
278 dm_rh_recovery_end(reg, !(read_err || write_err));
281 static int recover(struct mirror_set *ms, struct dm_region *reg)
285 struct dm_io_region from, to[DM_KCOPYD_MAX_REGIONS], *dest;
287 unsigned long flags = 0;
288 region_t key = dm_rh_get_region_key(reg);
289 sector_t region_size = dm_rh_get_region_size(ms->rh);
291 /* fill in the source */
292 m = get_default_mirror(ms);
293 from.bdev = m->dev->bdev;
294 from.sector = m->offset + dm_rh_region_to_sector(ms->rh, key);
295 if (key == (ms->nr_regions - 1)) {
297 * The final region may be smaller than
300 from.count = ms->ti->len & (region_size - 1);
302 from.count = region_size;
304 from.count = region_size;
306 /* fill in the destinations */
307 for (i = 0, dest = to; i < ms->nr_mirrors; i++) {
308 if (&ms->mirror[i] == get_default_mirror(ms))
312 dest->bdev = m->dev->bdev;
313 dest->sector = m->offset + dm_rh_region_to_sector(ms->rh, key);
314 dest->count = from.count;
319 if (!errors_handled(ms))
320 set_bit(DM_KCOPYD_IGNORE_ERROR, &flags);
322 r = dm_kcopyd_copy(ms->kcopyd_client, &from, ms->nr_mirrors - 1, to,
323 flags, recovery_complete, reg);
328 static void do_recovery(struct mirror_set *ms)
330 struct dm_region *reg;
331 struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
335 * Start quiescing some regions.
337 dm_rh_recovery_prepare(ms->rh);
340 * Copy any already quiesced regions.
342 while ((reg = dm_rh_recovery_start(ms->rh))) {
343 r = recover(ms, reg);
345 dm_rh_recovery_end(reg, 0);
349 * Update the in sync flag.
352 (log->type->get_sync_count(log) == ms->nr_regions)) {
353 /* the sync is complete */
354 dm_table_event(ms->ti->table);
359 /*-----------------------------------------------------------------
361 *---------------------------------------------------------------*/
362 static struct mirror *choose_mirror(struct mirror_set *ms, sector_t sector)
364 struct mirror *m = get_default_mirror(ms);
367 if (likely(!atomic_read(&m->error_count)))
370 if (m-- == ms->mirror)
372 } while (m != get_default_mirror(ms));
377 static int default_ok(struct mirror *m)
379 struct mirror *default_mirror = get_default_mirror(m->ms);
381 return !atomic_read(&default_mirror->error_count);
384 static int mirror_available(struct mirror_set *ms, struct bio *bio)
386 struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
387 region_t region = dm_rh_bio_to_region(ms->rh, bio);
389 if (log->type->in_sync(log, region, 0))
390 return choose_mirror(ms, bio->bi_sector) ? 1 : 0;
396 * remap a buffer to a particular mirror.
398 static sector_t map_sector(struct mirror *m, struct bio *bio)
400 return m->offset + (bio->bi_sector - m->ms->ti->begin);
403 static void map_bio(struct mirror *m, struct bio *bio)
405 bio->bi_bdev = m->dev->bdev;
406 bio->bi_sector = map_sector(m, bio);
409 static void map_region(struct dm_io_region *io, struct mirror *m,
412 io->bdev = m->dev->bdev;
413 io->sector = map_sector(m, bio);
414 io->count = bio->bi_size >> 9;
417 /*-----------------------------------------------------------------
419 *---------------------------------------------------------------*/
420 static void read_callback(unsigned long error, void *context)
422 struct bio *bio = context;
426 bio_set_m(bio, NULL);
428 if (likely(!error)) {
433 fail_mirror(m, DM_RAID1_READ_ERROR);
435 if (likely(default_ok(m)) || mirror_available(m->ms, bio)) {
436 DMWARN_LIMIT("Read failure on mirror device %s. "
437 "Trying alternative device.",
439 queue_bio(m->ms, bio, bio_rw(bio));
443 DMERR_LIMIT("Read failure on mirror device %s. Failing I/O.",
445 bio_endio(bio, -EIO);
448 /* Asynchronous read. */
449 static void read_async_bio(struct mirror *m, struct bio *bio)
451 struct dm_io_region io;
452 struct dm_io_request io_req = {
454 .mem.type = DM_IO_BVEC,
455 .mem.ptr.bvec = bio->bi_io_vec + bio->bi_idx,
456 .notify.fn = read_callback,
457 .notify.context = bio,
458 .client = m->ms->io_client,
461 map_region(&io, m, bio);
463 BUG_ON(dm_io(&io_req, 1, &io, NULL));
466 static inline int region_in_sync(struct mirror_set *ms, region_t region,
469 int state = dm_rh_get_state(ms->rh, region, may_block);
470 return state == DM_RH_CLEAN || state == DM_RH_DIRTY;
473 static void do_reads(struct mirror_set *ms, struct bio_list *reads)
479 while ((bio = bio_list_pop(reads))) {
480 region = dm_rh_bio_to_region(ms->rh, bio);
481 m = get_default_mirror(ms);
484 * We can only read balance if the region is in sync.
486 if (likely(region_in_sync(ms, region, 1)))
487 m = choose_mirror(ms, bio->bi_sector);
488 else if (m && atomic_read(&m->error_count))
492 read_async_bio(m, bio);
494 bio_endio(bio, -EIO);
498 /*-----------------------------------------------------------------
501 * We do different things with the write io depending on the
502 * state of the region that it's in:
504 * SYNC: increment pending, use kcopyd to write to *all* mirrors
505 * RECOVERING: delay the io until recovery completes
506 * NOSYNC: increment pending, just write to the default mirror
507 *---------------------------------------------------------------*/
510 static void write_callback(unsigned long error, void *context)
513 struct bio *bio = (struct bio *) context;
514 struct mirror_set *ms;
519 ms = bio_get_m(bio)->ms;
520 bio_set_m(bio, NULL);
523 * NOTE: We don't decrement the pending count here,
524 * instead it is done by the targets endio function.
525 * This way we handle both writes to SYNC and NOSYNC
526 * regions with the same code.
531 for (i = 0; i < ms->nr_mirrors; i++)
532 if (test_bit(i, &error))
533 fail_mirror(ms->mirror + i, DM_RAID1_WRITE_ERROR);
537 if (unlikely(!uptodate)) {
538 DMERR("All replicated volumes dead, failing I/O");
539 /* None of the writes succeeded, fail the I/O. */
541 } else if (errors_handled(ms)) {
543 * Need to raise event. Since raising
544 * events can block, we need to do it in
547 spin_lock_irqsave(&ms->lock, flags);
548 if (!ms->failures.head)
550 bio_list_add(&ms->failures, bio);
551 spin_unlock_irqrestore(&ms->lock, flags);
560 static void do_write(struct mirror_set *ms, struct bio *bio)
563 struct dm_io_region io[ms->nr_mirrors], *dest = io;
565 struct dm_io_request io_req = {
567 .mem.type = DM_IO_BVEC,
568 .mem.ptr.bvec = bio->bi_io_vec + bio->bi_idx,
569 .notify.fn = write_callback,
570 .notify.context = bio,
571 .client = ms->io_client,
574 for (i = 0, m = ms->mirror; i < ms->nr_mirrors; i++, m++)
575 map_region(dest++, m, bio);
578 * Use default mirror because we only need it to retrieve the reference
579 * to the mirror set in write_callback().
581 bio_set_m(bio, get_default_mirror(ms));
583 BUG_ON(dm_io(&io_req, ms->nr_mirrors, io, NULL));
586 static void do_writes(struct mirror_set *ms, struct bio_list *writes)
590 struct bio_list sync, nosync, recover, *this_list = NULL;
591 struct bio_list requeue;
592 struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
599 * Classify each write.
601 bio_list_init(&sync);
602 bio_list_init(&nosync);
603 bio_list_init(&recover);
604 bio_list_init(&requeue);
606 while ((bio = bio_list_pop(writes))) {
607 region = dm_rh_bio_to_region(ms->rh, bio);
609 if (log->type->is_remote_recovering &&
610 log->type->is_remote_recovering(log, region)) {
611 bio_list_add(&requeue, bio);
615 state = dm_rh_get_state(ms->rh, region, 1);
626 case DM_RH_RECOVERING:
627 this_list = &recover;
631 bio_list_add(this_list, bio);
635 * Add bios that are delayed due to remote recovery
636 * back on to the write queue
638 if (unlikely(requeue.head)) {
639 spin_lock_irq(&ms->lock);
640 bio_list_merge(&ms->writes, &requeue);
641 spin_unlock_irq(&ms->lock);
645 * Increment the pending counts for any regions that will
646 * be written to (writes to recover regions are going to
649 dm_rh_inc_pending(ms->rh, &sync);
650 dm_rh_inc_pending(ms->rh, &nosync);
651 ms->log_failure = dm_rh_flush(ms->rh) ? 1 : 0;
656 if (unlikely(ms->log_failure)) {
657 spin_lock_irq(&ms->lock);
658 bio_list_merge(&ms->failures, &sync);
659 spin_unlock_irq(&ms->lock);
662 while ((bio = bio_list_pop(&sync)))
665 while ((bio = bio_list_pop(&recover)))
666 dm_rh_delay(ms->rh, bio);
668 while ((bio = bio_list_pop(&nosync))) {
669 map_bio(get_default_mirror(ms), bio);
670 generic_make_request(bio);
674 static void do_failures(struct mirror_set *ms, struct bio_list *failures)
681 if (!ms->log_failure) {
682 while ((bio = bio_list_pop(failures))) {
684 dm_rh_mark_nosync(ms->rh, bio, bio->bi_size, 0);
690 * If the log has failed, unattempted writes are being
691 * put on the failures list. We can't issue those writes
692 * until a log has been marked, so we must store them.
694 * If a 'noflush' suspend is in progress, we can requeue
695 * the I/O's to the core. This give userspace a chance
696 * to reconfigure the mirror, at which point the core
697 * will reissue the writes. If the 'noflush' flag is
698 * not set, we have no choice but to return errors.
700 * Some writes on the failures list may have been
701 * submitted before the log failure and represent a
702 * failure to write to one of the devices. It is ok
703 * for us to treat them the same and requeue them
706 if (dm_noflush_suspending(ms->ti)) {
707 while ((bio = bio_list_pop(failures)))
708 bio_endio(bio, DM_ENDIO_REQUEUE);
712 if (atomic_read(&ms->suspend)) {
713 while ((bio = bio_list_pop(failures)))
714 bio_endio(bio, -EIO);
718 spin_lock_irq(&ms->lock);
719 bio_list_merge(&ms->failures, failures);
720 spin_unlock_irq(&ms->lock);
725 static void trigger_event(struct work_struct *work)
727 struct mirror_set *ms =
728 container_of(work, struct mirror_set, trigger_event);
730 dm_table_event(ms->ti->table);
733 /*-----------------------------------------------------------------
735 *---------------------------------------------------------------*/
736 static void do_mirror(struct work_struct *work)
738 struct mirror_set *ms = container_of(work, struct mirror_set,
740 struct bio_list reads, writes, failures;
743 spin_lock_irqsave(&ms->lock, flags);
746 failures = ms->failures;
747 bio_list_init(&ms->reads);
748 bio_list_init(&ms->writes);
749 bio_list_init(&ms->failures);
750 spin_unlock_irqrestore(&ms->lock, flags);
752 dm_rh_update_states(ms->rh, errors_handled(ms));
754 do_reads(ms, &reads);
755 do_writes(ms, &writes);
756 do_failures(ms, &failures);
758 dm_table_unplug_all(ms->ti->table);
761 /*-----------------------------------------------------------------
763 *---------------------------------------------------------------*/
764 static struct mirror_set *alloc_context(unsigned int nr_mirrors,
765 uint32_t region_size,
766 struct dm_target *ti,
767 struct dm_dirty_log *dl)
770 struct mirror_set *ms = NULL;
772 len = sizeof(*ms) + (sizeof(ms->mirror[0]) * nr_mirrors);
774 ms = kzalloc(len, GFP_KERNEL);
776 ti->error = "Cannot allocate mirror context";
780 spin_lock_init(&ms->lock);
783 ms->nr_mirrors = nr_mirrors;
784 ms->nr_regions = dm_sector_div_up(ti->len, region_size);
787 atomic_set(&ms->suspend, 0);
788 atomic_set(&ms->default_mirror, DEFAULT_MIRROR);
790 ms->read_record_pool = mempool_create_slab_pool(MIN_READ_RECORDS,
791 _dm_raid1_read_record_cache);
793 if (!ms->read_record_pool) {
794 ti->error = "Error creating mirror read_record_pool";
799 ms->io_client = dm_io_client_create(DM_IO_PAGES);
800 if (IS_ERR(ms->io_client)) {
801 ti->error = "Error creating dm_io client";
802 mempool_destroy(ms->read_record_pool);
807 ms->rh = dm_region_hash_create(ms, dispatch_bios, wakeup_mirrord,
808 wakeup_all_recovery_waiters,
809 ms->ti->begin, MAX_RECOVERY,
810 dl, region_size, ms->nr_regions);
811 if (IS_ERR(ms->rh)) {
812 ti->error = "Error creating dirty region hash";
813 dm_io_client_destroy(ms->io_client);
814 mempool_destroy(ms->read_record_pool);
822 static void free_context(struct mirror_set *ms, struct dm_target *ti,
826 dm_put_device(ti, ms->mirror[m].dev);
828 dm_io_client_destroy(ms->io_client);
829 dm_region_hash_destroy(ms->rh);
830 mempool_destroy(ms->read_record_pool);
834 static int get_mirror(struct mirror_set *ms, struct dm_target *ti,
835 unsigned int mirror, char **argv)
837 unsigned long long offset;
839 if (sscanf(argv[1], "%llu", &offset) != 1) {
840 ti->error = "Invalid offset";
844 if (dm_get_device(ti, argv[0], offset, ti->len,
845 dm_table_get_mode(ti->table),
846 &ms->mirror[mirror].dev)) {
847 ti->error = "Device lookup failure";
851 ms->mirror[mirror].ms = ms;
852 atomic_set(&(ms->mirror[mirror].error_count), 0);
853 ms->mirror[mirror].error_type = 0;
854 ms->mirror[mirror].offset = offset;
860 * Create dirty log: log_type #log_params <log_params>
862 static struct dm_dirty_log *create_dirty_log(struct dm_target *ti,
863 unsigned argc, char **argv,
866 unsigned param_count;
867 struct dm_dirty_log *dl;
870 ti->error = "Insufficient mirror log arguments";
874 if (sscanf(argv[1], "%u", ¶m_count) != 1) {
875 ti->error = "Invalid mirror log argument count";
879 *args_used = 2 + param_count;
881 if (argc < *args_used) {
882 ti->error = "Insufficient mirror log arguments";
886 dl = dm_dirty_log_create(argv[0], ti, param_count, argv + 2);
888 ti->error = "Error creating mirror dirty log";
895 static int parse_features(struct mirror_set *ms, unsigned argc, char **argv,
898 unsigned num_features;
899 struct dm_target *ti = ms->ti;
906 if (sscanf(argv[0], "%u", &num_features) != 1) {
907 ti->error = "Invalid number of features";
915 if (num_features > argc) {
916 ti->error = "Not enough arguments to support feature count";
920 if (!strcmp("handle_errors", argv[0]))
921 ms->features |= DM_RAID1_HANDLE_ERRORS;
923 ti->error = "Unrecognised feature requested";
933 * Construct a mirror mapping:
935 * log_type #log_params <log_params>
936 * #mirrors [mirror_path offset]{2,}
937 * [#features <features>]
939 * log_type is "core" or "disk"
940 * #log_params is between 1 and 3
942 * If present, features must be "handle_errors".
944 static int mirror_ctr(struct dm_target *ti, unsigned int argc, char **argv)
947 unsigned int nr_mirrors, m, args_used;
948 struct mirror_set *ms;
949 struct dm_dirty_log *dl;
951 dl = create_dirty_log(ti, argc, argv, &args_used);
958 if (!argc || sscanf(argv[0], "%u", &nr_mirrors) != 1 ||
959 nr_mirrors < 2 || nr_mirrors > DM_KCOPYD_MAX_REGIONS + 1) {
960 ti->error = "Invalid number of mirrors";
961 dm_dirty_log_destroy(dl);
967 if (argc < nr_mirrors * 2) {
968 ti->error = "Too few mirror arguments";
969 dm_dirty_log_destroy(dl);
973 ms = alloc_context(nr_mirrors, dl->type->get_region_size(dl), ti, dl);
975 dm_dirty_log_destroy(dl);
979 /* Get the mirror parameter sets */
980 for (m = 0; m < nr_mirrors; m++) {
981 r = get_mirror(ms, ti, m, argv);
983 free_context(ms, ti, m);
991 ti->split_io = dm_rh_get_region_size(ms->rh);
993 ms->kmirrord_wq = create_singlethread_workqueue("kmirrord");
994 if (!ms->kmirrord_wq) {
995 DMERR("couldn't start kmirrord");
997 goto err_free_context;
999 INIT_WORK(&ms->kmirrord_work, do_mirror);
1000 init_timer(&ms->timer);
1001 ms->timer_pending = 0;
1002 INIT_WORK(&ms->trigger_event, trigger_event);
1004 r = parse_features(ms, argc, argv, &args_used);
1006 goto err_destroy_wq;
1012 * Any read-balancing addition depends on the
1013 * DM_RAID1_HANDLE_ERRORS flag being present.
1014 * This is because the decision to balance depends
1015 * on the sync state of a region. If the above
1016 * flag is not present, we ignore errors; and
1017 * the sync state may be inaccurate.
1021 ti->error = "Too many mirror arguments";
1023 goto err_destroy_wq;
1026 r = dm_kcopyd_client_create(DM_KCOPYD_PAGES, &ms->kcopyd_client);
1028 goto err_destroy_wq;
1034 destroy_workqueue(ms->kmirrord_wq);
1036 free_context(ms, ti, ms->nr_mirrors);
1040 static void mirror_dtr(struct dm_target *ti)
1042 struct mirror_set *ms = (struct mirror_set *) ti->private;
1044 del_timer_sync(&ms->timer);
1045 flush_workqueue(ms->kmirrord_wq);
1046 flush_scheduled_work();
1047 dm_kcopyd_client_destroy(ms->kcopyd_client);
1048 destroy_workqueue(ms->kmirrord_wq);
1049 free_context(ms, ti, ms->nr_mirrors);
1053 * Mirror mapping function
1055 static int mirror_map(struct dm_target *ti, struct bio *bio,
1056 union map_info *map_context)
1058 int r, rw = bio_rw(bio);
1060 struct mirror_set *ms = ti->private;
1061 struct dm_raid1_read_record *read_record = NULL;
1062 struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
1065 /* Save region for mirror_end_io() handler */
1066 map_context->ll = dm_rh_bio_to_region(ms->rh, bio);
1067 queue_bio(ms, bio, rw);
1068 return DM_MAPIO_SUBMITTED;
1071 r = log->type->in_sync(log, dm_rh_bio_to_region(ms->rh, bio), 0);
1072 if (r < 0 && r != -EWOULDBLOCK)
1076 * If region is not in-sync queue the bio.
1078 if (!r || (r == -EWOULDBLOCK)) {
1080 return -EWOULDBLOCK;
1082 queue_bio(ms, bio, rw);
1083 return DM_MAPIO_SUBMITTED;
1087 * The region is in-sync and we can perform reads directly.
1088 * Store enough information so we can retry if it fails.
1090 m = choose_mirror(ms, bio->bi_sector);
1094 read_record = mempool_alloc(ms->read_record_pool, GFP_NOIO);
1095 if (likely(read_record)) {
1096 dm_bio_record(&read_record->details, bio);
1097 map_context->ptr = read_record;
1103 return DM_MAPIO_REMAPPED;
1106 static int mirror_end_io(struct dm_target *ti, struct bio *bio,
1107 int error, union map_info *map_context)
1109 int rw = bio_rw(bio);
1110 struct mirror_set *ms = (struct mirror_set *) ti->private;
1111 struct mirror *m = NULL;
1112 struct dm_bio_details *bd = NULL;
1113 struct dm_raid1_read_record *read_record = map_context->ptr;
1116 * We need to dec pending if this was a write.
1119 dm_rh_dec(ms->rh, map_context->ll);
1123 if (error == -EOPNOTSUPP)
1126 if ((error == -EWOULDBLOCK) && bio_rw_ahead(bio))
1129 if (unlikely(error)) {
1132 * There wasn't enough memory to record necessary
1133 * information for a retry or there was no other
1136 DMERR_LIMIT("Mirror read failed.");
1142 DMERR("Mirror read failed from %s. Trying alternative device.",
1145 fail_mirror(m, DM_RAID1_READ_ERROR);
1148 * A failed read is requeued for another attempt using an intact
1151 if (default_ok(m) || mirror_available(ms, bio)) {
1152 bd = &read_record->details;
1154 dm_bio_restore(bd, bio);
1155 mempool_free(read_record, ms->read_record_pool);
1156 map_context->ptr = NULL;
1157 queue_bio(ms, bio, rw);
1160 DMERR("All replicated volumes dead, failing I/O");
1165 mempool_free(read_record, ms->read_record_pool);
1166 map_context->ptr = NULL;
1172 static void mirror_presuspend(struct dm_target *ti)
1174 struct mirror_set *ms = (struct mirror_set *) ti->private;
1175 struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
1177 atomic_set(&ms->suspend, 1);
1180 * We must finish up all the work that we've
1181 * generated (i.e. recovery work).
1183 dm_rh_stop_recovery(ms->rh);
1185 wait_event(_kmirrord_recovery_stopped,
1186 !dm_rh_recovery_in_flight(ms->rh));
1188 if (log->type->presuspend && log->type->presuspend(log))
1189 /* FIXME: need better error handling */
1190 DMWARN("log presuspend failed");
1193 * Now that recovery is complete/stopped and the
1194 * delayed bios are queued, we need to wait for
1195 * the worker thread to complete. This way,
1196 * we know that all of our I/O has been pushed.
1198 flush_workqueue(ms->kmirrord_wq);
1201 static void mirror_postsuspend(struct dm_target *ti)
1203 struct mirror_set *ms = ti->private;
1204 struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
1206 if (log->type->postsuspend && log->type->postsuspend(log))
1207 /* FIXME: need better error handling */
1208 DMWARN("log postsuspend failed");
1211 static void mirror_resume(struct dm_target *ti)
1213 struct mirror_set *ms = ti->private;
1214 struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
1216 atomic_set(&ms->suspend, 0);
1217 if (log->type->resume && log->type->resume(log))
1218 /* FIXME: need better error handling */
1219 DMWARN("log resume failed");
1220 dm_rh_start_recovery(ms->rh);
1224 * device_status_char
1225 * @m: mirror device/leg we want the status of
1227 * We return one character representing the most severe error
1228 * we have encountered.
1229 * A => Alive - No failures
1230 * D => Dead - A write failure occurred leaving mirror out-of-sync
1231 * S => Sync - A sychronization failure occurred, mirror out-of-sync
1232 * R => Read - A read failure occurred, mirror data unaffected
1236 static char device_status_char(struct mirror *m)
1238 if (!atomic_read(&(m->error_count)))
1241 return (test_bit(DM_RAID1_WRITE_ERROR, &(m->error_type))) ? 'D' :
1242 (test_bit(DM_RAID1_SYNC_ERROR, &(m->error_type))) ? 'S' :
1243 (test_bit(DM_RAID1_READ_ERROR, &(m->error_type))) ? 'R' : 'U';
1247 static int mirror_status(struct dm_target *ti, status_type_t type,
1248 char *result, unsigned int maxlen)
1250 unsigned int m, sz = 0;
1251 struct mirror_set *ms = (struct mirror_set *) ti->private;
1252 struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
1253 char buffer[ms->nr_mirrors + 1];
1256 case STATUSTYPE_INFO:
1257 DMEMIT("%d ", ms->nr_mirrors);
1258 for (m = 0; m < ms->nr_mirrors; m++) {
1259 DMEMIT("%s ", ms->mirror[m].dev->name);
1260 buffer[m] = device_status_char(&(ms->mirror[m]));
1264 DMEMIT("%llu/%llu 1 %s ",
1265 (unsigned long long)log->type->get_sync_count(log),
1266 (unsigned long long)ms->nr_regions, buffer);
1268 sz += log->type->status(log, type, result+sz, maxlen-sz);
1272 case STATUSTYPE_TABLE:
1273 sz = log->type->status(log, type, result, maxlen);
1275 DMEMIT("%d", ms->nr_mirrors);
1276 for (m = 0; m < ms->nr_mirrors; m++)
1277 DMEMIT(" %s %llu", ms->mirror[m].dev->name,
1278 (unsigned long long)ms->mirror[m].offset);
1280 if (ms->features & DM_RAID1_HANDLE_ERRORS)
1281 DMEMIT(" 1 handle_errors");
1287 static struct target_type mirror_target = {
1289 .version = {1, 0, 20},
1290 .module = THIS_MODULE,
1294 .end_io = mirror_end_io,
1295 .presuspend = mirror_presuspend,
1296 .postsuspend = mirror_postsuspend,
1297 .resume = mirror_resume,
1298 .status = mirror_status,
1301 static int __init dm_mirror_init(void)
1305 _dm_raid1_read_record_cache = KMEM_CACHE(dm_raid1_read_record, 0);
1306 if (!_dm_raid1_read_record_cache) {
1307 DMERR("Can't allocate dm_raid1_read_record cache");
1312 r = dm_register_target(&mirror_target);
1314 DMERR("Failed to register mirror target");
1321 kmem_cache_destroy(_dm_raid1_read_record_cache);
1326 static void __exit dm_mirror_exit(void)
1328 dm_unregister_target(&mirror_target);
1329 kmem_cache_destroy(_dm_raid1_read_record_cache);
1333 module_init(dm_mirror_init);
1334 module_exit(dm_mirror_exit);
1336 MODULE_DESCRIPTION(DM_NAME " mirror target");
1337 MODULE_AUTHOR("Joe Thornber");
1338 MODULE_LICENSE("GPL");