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;
149 * Every mirror should look like this one.
151 #define DEFAULT_MIRROR 0
154 * This is yucky. We squirrel the mirror struct away inside
155 * bi_next for read/write buffers. This is safe since the bh
156 * doesn't get submitted to the lower levels of block layer.
158 static struct mirror *bio_get_m(struct bio *bio)
160 return (struct mirror *) bio->bi_next;
163 static void bio_set_m(struct bio *bio, struct mirror *m)
165 bio->bi_next = (struct bio *) m;
168 static struct mirror *get_default_mirror(struct mirror_set *ms)
170 return &ms->mirror[atomic_read(&ms->default_mirror)];
173 static void set_default_mirror(struct mirror *m)
175 struct mirror_set *ms = m->ms;
176 struct mirror *m0 = &(ms->mirror[0]);
178 atomic_set(&ms->default_mirror, m - m0);
182 * @m: mirror device to fail
183 * @error_type: one of the enum's, DM_RAID1_*_ERROR
185 * If errors are being handled, record the type of
186 * error encountered for this device. If this type
187 * of error has already been recorded, we can return;
188 * otherwise, we must signal userspace by triggering
189 * an event. Additionally, if the device is the
190 * primary device, we must choose a new primary, but
191 * only if the mirror is in-sync.
193 * This function must not block.
195 static void fail_mirror(struct mirror *m, enum dm_raid1_error error_type)
197 struct mirror_set *ms = m->ms;
200 if (!errors_handled(ms))
204 * error_count is used for nothing more than a
205 * simple way to tell if a device has encountered
208 atomic_inc(&m->error_count);
210 if (test_and_set_bit(error_type, &m->error_type))
213 if (m != get_default_mirror(ms))
218 * Better to issue requests to same failing device
219 * than to risk returning corrupt data.
221 DMERR("Primary mirror (%s) failed while out-of-sync: "
222 "Reads may fail.", m->dev->name);
226 for (new = ms->mirror; new < ms->mirror + ms->nr_mirrors; new++)
227 if (!atomic_read(&new->error_count)) {
228 set_default_mirror(new);
232 if (unlikely(new == ms->mirror + ms->nr_mirrors))
233 DMWARN("All sides of mirror have failed.");
236 schedule_work(&ms->trigger_event);
239 /*-----------------------------------------------------------------
242 * When a mirror is first activated we may find that some regions
243 * are in the no-sync state. We have to recover these by
244 * recopying from the default mirror to all the others.
245 *---------------------------------------------------------------*/
246 static void recovery_complete(int read_err, unsigned long write_err,
249 struct dm_region *reg = context;
250 struct mirror_set *ms = dm_rh_region_context(reg);
254 /* Read error means the failure of default mirror. */
255 DMERR_LIMIT("Unable to read primary mirror during recovery");
256 fail_mirror(get_default_mirror(ms), DM_RAID1_SYNC_ERROR);
260 DMERR_LIMIT("Write error during recovery (error = 0x%lx)",
263 * Bits correspond to devices (excluding default mirror).
264 * The default mirror cannot change during recovery.
266 for (m = 0; m < ms->nr_mirrors; m++) {
267 if (&ms->mirror[m] == get_default_mirror(ms))
269 if (test_bit(bit, &write_err))
270 fail_mirror(ms->mirror + m,
271 DM_RAID1_SYNC_ERROR);
276 dm_rh_recovery_end(reg, !(read_err || write_err));
279 static int recover(struct mirror_set *ms, struct dm_region *reg)
283 struct dm_io_region from, to[DM_KCOPYD_MAX_REGIONS], *dest;
285 unsigned long flags = 0;
286 region_t key = dm_rh_get_region_key(reg);
287 sector_t region_size = dm_rh_get_region_size(ms->rh);
289 /* fill in the source */
290 m = get_default_mirror(ms);
291 from.bdev = m->dev->bdev;
292 from.sector = m->offset + dm_rh_region_to_sector(ms->rh, key);
293 if (key == (ms->nr_regions - 1)) {
295 * The final region may be smaller than
298 from.count = ms->ti->len & (region_size - 1);
300 from.count = region_size;
302 from.count = region_size;
304 /* fill in the destinations */
305 for (i = 0, dest = to; i < ms->nr_mirrors; i++) {
306 if (&ms->mirror[i] == get_default_mirror(ms))
310 dest->bdev = m->dev->bdev;
311 dest->sector = m->offset + dm_rh_region_to_sector(ms->rh, key);
312 dest->count = from.count;
317 if (!errors_handled(ms))
318 set_bit(DM_KCOPYD_IGNORE_ERROR, &flags);
320 r = dm_kcopyd_copy(ms->kcopyd_client, &from, ms->nr_mirrors - 1, to,
321 flags, recovery_complete, reg);
326 static void do_recovery(struct mirror_set *ms)
328 struct dm_region *reg;
329 struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
333 * Start quiescing some regions.
335 dm_rh_recovery_prepare(ms->rh);
338 * Copy any already quiesced regions.
340 while ((reg = dm_rh_recovery_start(ms->rh))) {
341 r = recover(ms, reg);
343 dm_rh_recovery_end(reg, 0);
347 * Update the in sync flag.
350 (log->type->get_sync_count(log) == ms->nr_regions)) {
351 /* the sync is complete */
352 dm_table_event(ms->ti->table);
357 /*-----------------------------------------------------------------
359 *---------------------------------------------------------------*/
360 static struct mirror *choose_mirror(struct mirror_set *ms, sector_t sector)
362 struct mirror *m = get_default_mirror(ms);
365 if (likely(!atomic_read(&m->error_count)))
368 if (m-- == ms->mirror)
370 } while (m != get_default_mirror(ms));
375 static int default_ok(struct mirror *m)
377 struct mirror *default_mirror = get_default_mirror(m->ms);
379 return !atomic_read(&default_mirror->error_count);
382 static int mirror_available(struct mirror_set *ms, struct bio *bio)
384 struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
385 region_t region = dm_rh_bio_to_region(ms->rh, bio);
387 if (log->type->in_sync(log, region, 0))
388 return choose_mirror(ms, bio->bi_sector) ? 1 : 0;
394 * remap a buffer to a particular mirror.
396 static sector_t map_sector(struct mirror *m, struct bio *bio)
398 return m->offset + (bio->bi_sector - m->ms->ti->begin);
401 static void map_bio(struct mirror *m, struct bio *bio)
403 bio->bi_bdev = m->dev->bdev;
404 bio->bi_sector = map_sector(m, bio);
407 static void map_region(struct dm_io_region *io, struct mirror *m,
410 io->bdev = m->dev->bdev;
411 io->sector = map_sector(m, bio);
412 io->count = bio->bi_size >> 9;
415 /*-----------------------------------------------------------------
417 *---------------------------------------------------------------*/
418 static void read_callback(unsigned long error, void *context)
420 struct bio *bio = context;
424 bio_set_m(bio, NULL);
426 if (likely(!error)) {
431 fail_mirror(m, DM_RAID1_READ_ERROR);
433 if (likely(default_ok(m)) || mirror_available(m->ms, bio)) {
434 DMWARN_LIMIT("Read failure on mirror device %s. "
435 "Trying alternative device.",
437 queue_bio(m->ms, bio, bio_rw(bio));
441 DMERR_LIMIT("Read failure on mirror device %s. Failing I/O.",
443 bio_endio(bio, -EIO);
446 /* Asynchronous read. */
447 static void read_async_bio(struct mirror *m, struct bio *bio)
449 struct dm_io_region io;
450 struct dm_io_request io_req = {
452 .mem.type = DM_IO_BVEC,
453 .mem.ptr.bvec = bio->bi_io_vec + bio->bi_idx,
454 .notify.fn = read_callback,
455 .notify.context = bio,
456 .client = m->ms->io_client,
459 map_region(&io, m, bio);
461 BUG_ON(dm_io(&io_req, 1, &io, NULL));
464 static inline int region_in_sync(struct mirror_set *ms, region_t region,
467 int state = dm_rh_get_state(ms->rh, region, may_block);
468 return state == DM_RH_CLEAN || state == DM_RH_DIRTY;
471 static void do_reads(struct mirror_set *ms, struct bio_list *reads)
477 while ((bio = bio_list_pop(reads))) {
478 region = dm_rh_bio_to_region(ms->rh, bio);
479 m = get_default_mirror(ms);
482 * We can only read balance if the region is in sync.
484 if (likely(region_in_sync(ms, region, 1)))
485 m = choose_mirror(ms, bio->bi_sector);
486 else if (m && atomic_read(&m->error_count))
490 read_async_bio(m, bio);
492 bio_endio(bio, -EIO);
496 /*-----------------------------------------------------------------
499 * We do different things with the write io depending on the
500 * state of the region that it's in:
502 * SYNC: increment pending, use kcopyd to write to *all* mirrors
503 * RECOVERING: delay the io until recovery completes
504 * NOSYNC: increment pending, just write to the default mirror
505 *---------------------------------------------------------------*/
508 static void write_callback(unsigned long error, void *context)
511 struct bio *bio = (struct bio *) context;
512 struct mirror_set *ms;
517 ms = bio_get_m(bio)->ms;
518 bio_set_m(bio, NULL);
521 * NOTE: We don't decrement the pending count here,
522 * instead it is done by the targets endio function.
523 * This way we handle both writes to SYNC and NOSYNC
524 * regions with the same code.
529 for (i = 0; i < ms->nr_mirrors; i++)
530 if (test_bit(i, &error))
531 fail_mirror(ms->mirror + i, DM_RAID1_WRITE_ERROR);
535 if (unlikely(!uptodate)) {
536 DMERR("All replicated volumes dead, failing I/O");
537 /* None of the writes succeeded, fail the I/O. */
539 } else if (errors_handled(ms)) {
541 * Need to raise event. Since raising
542 * events can block, we need to do it in
545 spin_lock_irqsave(&ms->lock, flags);
546 if (!ms->failures.head)
548 bio_list_add(&ms->failures, bio);
549 spin_unlock_irqrestore(&ms->lock, flags);
558 static void do_write(struct mirror_set *ms, struct bio *bio)
561 struct dm_io_region io[ms->nr_mirrors], *dest = io;
563 struct dm_io_request io_req = {
565 .mem.type = DM_IO_BVEC,
566 .mem.ptr.bvec = bio->bi_io_vec + bio->bi_idx,
567 .notify.fn = write_callback,
568 .notify.context = bio,
569 .client = ms->io_client,
572 for (i = 0, m = ms->mirror; i < ms->nr_mirrors; i++, m++)
573 map_region(dest++, m, bio);
576 * Use default mirror because we only need it to retrieve the reference
577 * to the mirror set in write_callback().
579 bio_set_m(bio, get_default_mirror(ms));
581 BUG_ON(dm_io(&io_req, ms->nr_mirrors, io, NULL));
584 static void do_writes(struct mirror_set *ms, struct bio_list *writes)
588 struct bio_list sync, nosync, recover, *this_list = NULL;
594 * Classify each write.
596 bio_list_init(&sync);
597 bio_list_init(&nosync);
598 bio_list_init(&recover);
600 while ((bio = bio_list_pop(writes))) {
601 state = dm_rh_get_state(ms->rh,
602 dm_rh_bio_to_region(ms->rh, bio), 1);
613 case DM_RH_RECOVERING:
614 this_list = &recover;
618 bio_list_add(this_list, bio);
622 * Increment the pending counts for any regions that will
623 * be written to (writes to recover regions are going to
626 dm_rh_inc_pending(ms->rh, &sync);
627 dm_rh_inc_pending(ms->rh, &nosync);
628 ms->log_failure = dm_rh_flush(ms->rh) ? 1 : 0;
633 if (unlikely(ms->log_failure)) {
634 spin_lock_irq(&ms->lock);
635 bio_list_merge(&ms->failures, &sync);
636 spin_unlock_irq(&ms->lock);
639 while ((bio = bio_list_pop(&sync)))
642 while ((bio = bio_list_pop(&recover)))
643 dm_rh_delay(ms->rh, bio);
645 while ((bio = bio_list_pop(&nosync))) {
646 map_bio(get_default_mirror(ms), bio);
647 generic_make_request(bio);
651 static void do_failures(struct mirror_set *ms, struct bio_list *failures)
658 if (!ms->log_failure) {
659 while ((bio = bio_list_pop(failures))) {
661 dm_rh_mark_nosync(ms->rh, bio, bio->bi_size, 0);
667 * If the log has failed, unattempted writes are being
668 * put on the failures list. We can't issue those writes
669 * until a log has been marked, so we must store them.
671 * If a 'noflush' suspend is in progress, we can requeue
672 * the I/O's to the core. This give userspace a chance
673 * to reconfigure the mirror, at which point the core
674 * will reissue the writes. If the 'noflush' flag is
675 * not set, we have no choice but to return errors.
677 * Some writes on the failures list may have been
678 * submitted before the log failure and represent a
679 * failure to write to one of the devices. It is ok
680 * for us to treat them the same and requeue them
683 if (dm_noflush_suspending(ms->ti)) {
684 while ((bio = bio_list_pop(failures)))
685 bio_endio(bio, DM_ENDIO_REQUEUE);
689 if (atomic_read(&ms->suspend)) {
690 while ((bio = bio_list_pop(failures)))
691 bio_endio(bio, -EIO);
695 spin_lock_irq(&ms->lock);
696 bio_list_merge(&ms->failures, failures);
697 spin_unlock_irq(&ms->lock);
702 static void trigger_event(struct work_struct *work)
704 struct mirror_set *ms =
705 container_of(work, struct mirror_set, trigger_event);
707 dm_table_event(ms->ti->table);
710 /*-----------------------------------------------------------------
712 *---------------------------------------------------------------*/
713 static void do_mirror(struct work_struct *work)
715 struct mirror_set *ms = container_of(work, struct mirror_set,
717 struct bio_list reads, writes, failures;
720 spin_lock_irqsave(&ms->lock, flags);
723 failures = ms->failures;
724 bio_list_init(&ms->reads);
725 bio_list_init(&ms->writes);
726 bio_list_init(&ms->failures);
727 spin_unlock_irqrestore(&ms->lock, flags);
729 dm_rh_update_states(ms->rh, errors_handled(ms));
731 do_reads(ms, &reads);
732 do_writes(ms, &writes);
733 do_failures(ms, &failures);
735 dm_table_unplug_all(ms->ti->table);
738 /*-----------------------------------------------------------------
740 *---------------------------------------------------------------*/
741 static struct mirror_set *alloc_context(unsigned int nr_mirrors,
742 uint32_t region_size,
743 struct dm_target *ti,
744 struct dm_dirty_log *dl)
747 struct mirror_set *ms = NULL;
749 len = sizeof(*ms) + (sizeof(ms->mirror[0]) * nr_mirrors);
751 ms = kzalloc(len, GFP_KERNEL);
753 ti->error = "Cannot allocate mirror context";
757 spin_lock_init(&ms->lock);
760 ms->nr_mirrors = nr_mirrors;
761 ms->nr_regions = dm_sector_div_up(ti->len, region_size);
764 atomic_set(&ms->suspend, 0);
765 atomic_set(&ms->default_mirror, DEFAULT_MIRROR);
767 len = sizeof(struct dm_raid1_read_record);
768 ms->read_record_pool = mempool_create_kmalloc_pool(MIN_READ_RECORDS,
770 if (!ms->read_record_pool) {
771 ti->error = "Error creating mirror read_record_pool";
776 ms->io_client = dm_io_client_create(DM_IO_PAGES);
777 if (IS_ERR(ms->io_client)) {
778 ti->error = "Error creating dm_io client";
779 mempool_destroy(ms->read_record_pool);
784 ms->rh = dm_region_hash_create(ms, dispatch_bios, wakeup_mirrord,
785 wakeup_all_recovery_waiters,
786 ms->ti->begin, MAX_RECOVERY,
787 dl, region_size, ms->nr_regions);
788 if (IS_ERR(ms->rh)) {
789 ti->error = "Error creating dirty region hash";
790 dm_io_client_destroy(ms->io_client);
791 mempool_destroy(ms->read_record_pool);
799 static void free_context(struct mirror_set *ms, struct dm_target *ti,
803 dm_put_device(ti, ms->mirror[m].dev);
805 dm_io_client_destroy(ms->io_client);
806 dm_region_hash_destroy(ms->rh);
807 mempool_destroy(ms->read_record_pool);
811 static inline int _check_region_size(struct dm_target *ti, uint32_t size)
813 return !(size % (PAGE_SIZE >> 9) || !is_power_of_2(size) ||
817 static int get_mirror(struct mirror_set *ms, struct dm_target *ti,
818 unsigned int mirror, char **argv)
820 unsigned long long offset;
822 if (sscanf(argv[1], "%llu", &offset) != 1) {
823 ti->error = "Invalid offset";
827 if (dm_get_device(ti, argv[0], offset, ti->len,
828 dm_table_get_mode(ti->table),
829 &ms->mirror[mirror].dev)) {
830 ti->error = "Device lookup failure";
834 ms->mirror[mirror].ms = ms;
835 atomic_set(&(ms->mirror[mirror].error_count), 0);
836 ms->mirror[mirror].error_type = 0;
837 ms->mirror[mirror].offset = offset;
843 * Create dirty log: log_type #log_params <log_params>
845 static struct dm_dirty_log *create_dirty_log(struct dm_target *ti,
846 unsigned argc, char **argv,
849 unsigned param_count;
850 struct dm_dirty_log *dl;
853 ti->error = "Insufficient mirror log arguments";
857 if (sscanf(argv[1], "%u", ¶m_count) != 1) {
858 ti->error = "Invalid mirror log argument count";
862 *args_used = 2 + param_count;
864 if (argc < *args_used) {
865 ti->error = "Insufficient mirror log arguments";
869 dl = dm_dirty_log_create(argv[0], ti, param_count, argv + 2);
871 ti->error = "Error creating mirror dirty log";
875 if (!_check_region_size(ti, dl->type->get_region_size(dl))) {
876 ti->error = "Invalid region size";
877 dm_dirty_log_destroy(dl);
884 static int parse_features(struct mirror_set *ms, unsigned argc, char **argv,
887 unsigned num_features;
888 struct dm_target *ti = ms->ti;
895 if (sscanf(argv[0], "%u", &num_features) != 1) {
896 ti->error = "Invalid number of features";
904 if (num_features > argc) {
905 ti->error = "Not enough arguments to support feature count";
909 if (!strcmp("handle_errors", argv[0]))
910 ms->features |= DM_RAID1_HANDLE_ERRORS;
912 ti->error = "Unrecognised feature requested";
922 * Construct a mirror mapping:
924 * log_type #log_params <log_params>
925 * #mirrors [mirror_path offset]{2,}
926 * [#features <features>]
928 * log_type is "core" or "disk"
929 * #log_params is between 1 and 3
931 * If present, features must be "handle_errors".
933 static int mirror_ctr(struct dm_target *ti, unsigned int argc, char **argv)
936 unsigned int nr_mirrors, m, args_used;
937 struct mirror_set *ms;
938 struct dm_dirty_log *dl;
940 dl = create_dirty_log(ti, argc, argv, &args_used);
947 if (!argc || sscanf(argv[0], "%u", &nr_mirrors) != 1 ||
948 nr_mirrors < 2 || nr_mirrors > DM_KCOPYD_MAX_REGIONS + 1) {
949 ti->error = "Invalid number of mirrors";
950 dm_dirty_log_destroy(dl);
956 if (argc < nr_mirrors * 2) {
957 ti->error = "Too few mirror arguments";
958 dm_dirty_log_destroy(dl);
962 ms = alloc_context(nr_mirrors, dl->type->get_region_size(dl), ti, dl);
964 dm_dirty_log_destroy(dl);
968 /* Get the mirror parameter sets */
969 for (m = 0; m < nr_mirrors; m++) {
970 r = get_mirror(ms, ti, m, argv);
972 free_context(ms, ti, m);
980 ti->split_io = dm_rh_get_region_size(ms->rh);
982 ms->kmirrord_wq = create_singlethread_workqueue("kmirrord");
983 if (!ms->kmirrord_wq) {
984 DMERR("couldn't start kmirrord");
986 goto err_free_context;
988 INIT_WORK(&ms->kmirrord_work, do_mirror);
989 init_timer(&ms->timer);
990 ms->timer_pending = 0;
991 INIT_WORK(&ms->trigger_event, trigger_event);
993 r = parse_features(ms, argc, argv, &args_used);
1001 * Any read-balancing addition depends on the
1002 * DM_RAID1_HANDLE_ERRORS flag being present.
1003 * This is because the decision to balance depends
1004 * on the sync state of a region. If the above
1005 * flag is not present, we ignore errors; and
1006 * the sync state may be inaccurate.
1010 ti->error = "Too many mirror arguments";
1012 goto err_destroy_wq;
1015 r = dm_kcopyd_client_create(DM_KCOPYD_PAGES, &ms->kcopyd_client);
1017 goto err_destroy_wq;
1023 destroy_workqueue(ms->kmirrord_wq);
1025 free_context(ms, ti, ms->nr_mirrors);
1029 static void mirror_dtr(struct dm_target *ti)
1031 struct mirror_set *ms = (struct mirror_set *) ti->private;
1033 del_timer_sync(&ms->timer);
1034 flush_workqueue(ms->kmirrord_wq);
1035 dm_kcopyd_client_destroy(ms->kcopyd_client);
1036 destroy_workqueue(ms->kmirrord_wq);
1037 free_context(ms, ti, ms->nr_mirrors);
1041 * Mirror mapping function
1043 static int mirror_map(struct dm_target *ti, struct bio *bio,
1044 union map_info *map_context)
1046 int r, rw = bio_rw(bio);
1048 struct mirror_set *ms = ti->private;
1049 struct dm_raid1_read_record *read_record = NULL;
1050 struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
1053 /* Save region for mirror_end_io() handler */
1054 map_context->ll = dm_rh_bio_to_region(ms->rh, bio);
1055 queue_bio(ms, bio, rw);
1056 return DM_MAPIO_SUBMITTED;
1059 r = log->type->in_sync(log, dm_rh_bio_to_region(ms->rh, bio), 0);
1060 if (r < 0 && r != -EWOULDBLOCK)
1064 * If region is not in-sync queue the bio.
1066 if (!r || (r == -EWOULDBLOCK)) {
1068 return -EWOULDBLOCK;
1070 queue_bio(ms, bio, rw);
1071 return DM_MAPIO_SUBMITTED;
1075 * The region is in-sync and we can perform reads directly.
1076 * Store enough information so we can retry if it fails.
1078 m = choose_mirror(ms, bio->bi_sector);
1082 read_record = mempool_alloc(ms->read_record_pool, GFP_NOIO);
1083 if (likely(read_record)) {
1084 dm_bio_record(&read_record->details, bio);
1085 map_context->ptr = read_record;
1091 return DM_MAPIO_REMAPPED;
1094 static int mirror_end_io(struct dm_target *ti, struct bio *bio,
1095 int error, union map_info *map_context)
1097 int rw = bio_rw(bio);
1098 struct mirror_set *ms = (struct mirror_set *) ti->private;
1099 struct mirror *m = NULL;
1100 struct dm_bio_details *bd = NULL;
1101 struct dm_raid1_read_record *read_record = map_context->ptr;
1104 * We need to dec pending if this was a write.
1107 dm_rh_dec(ms->rh, map_context->ll);
1111 if (error == -EOPNOTSUPP)
1114 if ((error == -EWOULDBLOCK) && bio_rw_ahead(bio))
1117 if (unlikely(error)) {
1120 * There wasn't enough memory to record necessary
1121 * information for a retry or there was no other
1124 DMERR_LIMIT("Mirror read failed.");
1130 DMERR("Mirror read failed from %s. Trying alternative device.",
1133 fail_mirror(m, DM_RAID1_READ_ERROR);
1136 * A failed read is requeued for another attempt using an intact
1139 if (default_ok(m) || mirror_available(ms, bio)) {
1140 bd = &read_record->details;
1142 dm_bio_restore(bd, bio);
1143 mempool_free(read_record, ms->read_record_pool);
1144 map_context->ptr = NULL;
1145 queue_bio(ms, bio, rw);
1148 DMERR("All replicated volumes dead, failing I/O");
1153 mempool_free(read_record, ms->read_record_pool);
1154 map_context->ptr = NULL;
1160 static void mirror_presuspend(struct dm_target *ti)
1162 struct mirror_set *ms = (struct mirror_set *) ti->private;
1163 struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
1165 atomic_set(&ms->suspend, 1);
1168 * We must finish up all the work that we've
1169 * generated (i.e. recovery work).
1171 dm_rh_stop_recovery(ms->rh);
1173 wait_event(_kmirrord_recovery_stopped,
1174 !dm_rh_recovery_in_flight(ms->rh));
1176 if (log->type->presuspend && log->type->presuspend(log))
1177 /* FIXME: need better error handling */
1178 DMWARN("log presuspend failed");
1181 * Now that recovery is complete/stopped and the
1182 * delayed bios are queued, we need to wait for
1183 * the worker thread to complete. This way,
1184 * we know that all of our I/O has been pushed.
1186 flush_workqueue(ms->kmirrord_wq);
1189 static void mirror_postsuspend(struct dm_target *ti)
1191 struct mirror_set *ms = ti->private;
1192 struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
1194 if (log->type->postsuspend && log->type->postsuspend(log))
1195 /* FIXME: need better error handling */
1196 DMWARN("log postsuspend failed");
1199 static void mirror_resume(struct dm_target *ti)
1201 struct mirror_set *ms = ti->private;
1202 struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
1204 atomic_set(&ms->suspend, 0);
1205 if (log->type->resume && log->type->resume(log))
1206 /* FIXME: need better error handling */
1207 DMWARN("log resume failed");
1208 dm_rh_start_recovery(ms->rh);
1212 * device_status_char
1213 * @m: mirror device/leg we want the status of
1215 * We return one character representing the most severe error
1216 * we have encountered.
1217 * A => Alive - No failures
1218 * D => Dead - A write failure occurred leaving mirror out-of-sync
1219 * S => Sync - A sychronization failure occurred, mirror out-of-sync
1220 * R => Read - A read failure occurred, mirror data unaffected
1224 static char device_status_char(struct mirror *m)
1226 if (!atomic_read(&(m->error_count)))
1229 return (test_bit(DM_RAID1_WRITE_ERROR, &(m->error_type))) ? 'D' :
1230 (test_bit(DM_RAID1_SYNC_ERROR, &(m->error_type))) ? 'S' :
1231 (test_bit(DM_RAID1_READ_ERROR, &(m->error_type))) ? 'R' : 'U';
1235 static int mirror_status(struct dm_target *ti, status_type_t type,
1236 char *result, unsigned int maxlen)
1238 unsigned int m, sz = 0;
1239 struct mirror_set *ms = (struct mirror_set *) ti->private;
1240 struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
1241 char buffer[ms->nr_mirrors + 1];
1244 case STATUSTYPE_INFO:
1245 DMEMIT("%d ", ms->nr_mirrors);
1246 for (m = 0; m < ms->nr_mirrors; m++) {
1247 DMEMIT("%s ", ms->mirror[m].dev->name);
1248 buffer[m] = device_status_char(&(ms->mirror[m]));
1252 DMEMIT("%llu/%llu 1 %s ",
1253 (unsigned long long)log->type->get_sync_count(log),
1254 (unsigned long long)ms->nr_regions, buffer);
1256 sz += log->type->status(log, type, result+sz, maxlen-sz);
1260 case STATUSTYPE_TABLE:
1261 sz = log->type->status(log, type, result, maxlen);
1263 DMEMIT("%d", ms->nr_mirrors);
1264 for (m = 0; m < ms->nr_mirrors; m++)
1265 DMEMIT(" %s %llu", ms->mirror[m].dev->name,
1266 (unsigned long long)ms->mirror[m].offset);
1268 if (ms->features & DM_RAID1_HANDLE_ERRORS)
1269 DMEMIT(" 1 handle_errors");
1275 static struct target_type mirror_target = {
1277 .version = {1, 0, 20},
1278 .module = THIS_MODULE,
1282 .end_io = mirror_end_io,
1283 .presuspend = mirror_presuspend,
1284 .postsuspend = mirror_postsuspend,
1285 .resume = mirror_resume,
1286 .status = mirror_status,
1289 static int __init dm_mirror_init(void)
1293 r = dm_register_target(&mirror_target);
1295 DMERR("Failed to register mirror target");
1300 static void __exit dm_mirror_exit(void)
1304 r = dm_unregister_target(&mirror_target);
1306 DMERR("unregister failed %d", r);
1310 module_init(dm_mirror_init);
1311 module_exit(dm_mirror_exit);
1313 MODULE_DESCRIPTION(DM_NAME " mirror target");
1314 MODULE_AUTHOR("Joe Thornber");
1315 MODULE_LICENSE("GPL");