xref: /openbmc/linux/drivers/md/dm-raid1.c (revision bc5aa3a0)
1 /*
2  * Copyright (C) 2003 Sistina Software Limited.
3  * Copyright (C) 2005-2008 Red Hat, Inc. All rights reserved.
4  *
5  * This file is released under the GPL.
6  */
7 
8 #include "dm-bio-record.h"
9 
10 #include <linux/init.h>
11 #include <linux/mempool.h>
12 #include <linux/module.h>
13 #include <linux/pagemap.h>
14 #include <linux/slab.h>
15 #include <linux/workqueue.h>
16 #include <linux/device-mapper.h>
17 #include <linux/dm-io.h>
18 #include <linux/dm-dirty-log.h>
19 #include <linux/dm-kcopyd.h>
20 #include <linux/dm-region-hash.h>
21 
22 #define DM_MSG_PREFIX "raid1"
23 
24 #define MAX_RECOVERY 1	/* Maximum number of regions recovered in parallel. */
25 
26 #define DM_RAID1_HANDLE_ERRORS	0x01
27 #define DM_RAID1_KEEP_LOG	0x02
28 #define errors_handled(p)	((p)->features & DM_RAID1_HANDLE_ERRORS)
29 #define keep_log(p)		((p)->features & DM_RAID1_KEEP_LOG)
30 
31 static DECLARE_WAIT_QUEUE_HEAD(_kmirrord_recovery_stopped);
32 
33 /*-----------------------------------------------------------------
34  * Mirror set structures.
35  *---------------------------------------------------------------*/
36 enum dm_raid1_error {
37 	DM_RAID1_WRITE_ERROR,
38 	DM_RAID1_FLUSH_ERROR,
39 	DM_RAID1_SYNC_ERROR,
40 	DM_RAID1_READ_ERROR
41 };
42 
43 struct mirror {
44 	struct mirror_set *ms;
45 	atomic_t error_count;
46 	unsigned long error_type;
47 	struct dm_dev *dev;
48 	sector_t offset;
49 };
50 
51 struct mirror_set {
52 	struct dm_target *ti;
53 	struct list_head list;
54 
55 	uint64_t features;
56 
57 	spinlock_t lock;	/* protects the lists */
58 	struct bio_list reads;
59 	struct bio_list writes;
60 	struct bio_list failures;
61 	struct bio_list holds;	/* bios are waiting until suspend */
62 
63 	struct dm_region_hash *rh;
64 	struct dm_kcopyd_client *kcopyd_client;
65 	struct dm_io_client *io_client;
66 
67 	/* recovery */
68 	region_t nr_regions;
69 	int in_sync;
70 	int log_failure;
71 	int leg_failure;
72 	atomic_t suspend;
73 
74 	atomic_t default_mirror;	/* Default mirror */
75 
76 	struct workqueue_struct *kmirrord_wq;
77 	struct work_struct kmirrord_work;
78 	struct timer_list timer;
79 	unsigned long timer_pending;
80 
81 	struct work_struct trigger_event;
82 
83 	unsigned nr_mirrors;
84 	struct mirror mirror[0];
85 };
86 
87 DECLARE_DM_KCOPYD_THROTTLE_WITH_MODULE_PARM(raid1_resync_throttle,
88 		"A percentage of time allocated for raid resynchronization");
89 
90 static void wakeup_mirrord(void *context)
91 {
92 	struct mirror_set *ms = context;
93 
94 	queue_work(ms->kmirrord_wq, &ms->kmirrord_work);
95 }
96 
97 static void delayed_wake_fn(unsigned long data)
98 {
99 	struct mirror_set *ms = (struct mirror_set *) data;
100 
101 	clear_bit(0, &ms->timer_pending);
102 	wakeup_mirrord(ms);
103 }
104 
105 static void delayed_wake(struct mirror_set *ms)
106 {
107 	if (test_and_set_bit(0, &ms->timer_pending))
108 		return;
109 
110 	ms->timer.expires = jiffies + HZ / 5;
111 	ms->timer.data = (unsigned long) ms;
112 	ms->timer.function = delayed_wake_fn;
113 	add_timer(&ms->timer);
114 }
115 
116 static void wakeup_all_recovery_waiters(void *context)
117 {
118 	wake_up_all(&_kmirrord_recovery_stopped);
119 }
120 
121 static void queue_bio(struct mirror_set *ms, struct bio *bio, int rw)
122 {
123 	unsigned long flags;
124 	int should_wake = 0;
125 	struct bio_list *bl;
126 
127 	bl = (rw == WRITE) ? &ms->writes : &ms->reads;
128 	spin_lock_irqsave(&ms->lock, flags);
129 	should_wake = !(bl->head);
130 	bio_list_add(bl, bio);
131 	spin_unlock_irqrestore(&ms->lock, flags);
132 
133 	if (should_wake)
134 		wakeup_mirrord(ms);
135 }
136 
137 static void dispatch_bios(void *context, struct bio_list *bio_list)
138 {
139 	struct mirror_set *ms = context;
140 	struct bio *bio;
141 
142 	while ((bio = bio_list_pop(bio_list)))
143 		queue_bio(ms, bio, WRITE);
144 }
145 
146 struct dm_raid1_bio_record {
147 	struct mirror *m;
148 	/* if details->bi_bdev == NULL, details were not saved */
149 	struct dm_bio_details details;
150 	region_t write_region;
151 };
152 
153 /*
154  * Every mirror should look like this one.
155  */
156 #define DEFAULT_MIRROR 0
157 
158 /*
159  * This is yucky.  We squirrel the mirror struct away inside
160  * bi_next for read/write buffers.  This is safe since the bh
161  * doesn't get submitted to the lower levels of block layer.
162  */
163 static struct mirror *bio_get_m(struct bio *bio)
164 {
165 	return (struct mirror *) bio->bi_next;
166 }
167 
168 static void bio_set_m(struct bio *bio, struct mirror *m)
169 {
170 	bio->bi_next = (struct bio *) m;
171 }
172 
173 static struct mirror *get_default_mirror(struct mirror_set *ms)
174 {
175 	return &ms->mirror[atomic_read(&ms->default_mirror)];
176 }
177 
178 static void set_default_mirror(struct mirror *m)
179 {
180 	struct mirror_set *ms = m->ms;
181 	struct mirror *m0 = &(ms->mirror[0]);
182 
183 	atomic_set(&ms->default_mirror, m - m0);
184 }
185 
186 static struct mirror *get_valid_mirror(struct mirror_set *ms)
187 {
188 	struct mirror *m;
189 
190 	for (m = ms->mirror; m < ms->mirror + ms->nr_mirrors; m++)
191 		if (!atomic_read(&m->error_count))
192 			return m;
193 
194 	return NULL;
195 }
196 
197 /* fail_mirror
198  * @m: mirror device to fail
199  * @error_type: one of the enum's, DM_RAID1_*_ERROR
200  *
201  * If errors are being handled, record the type of
202  * error encountered for this device.  If this type
203  * of error has already been recorded, we can return;
204  * otherwise, we must signal userspace by triggering
205  * an event.  Additionally, if the device is the
206  * primary device, we must choose a new primary, but
207  * only if the mirror is in-sync.
208  *
209  * This function must not block.
210  */
211 static void fail_mirror(struct mirror *m, enum dm_raid1_error error_type)
212 {
213 	struct mirror_set *ms = m->ms;
214 	struct mirror *new;
215 
216 	ms->leg_failure = 1;
217 
218 	/*
219 	 * error_count is used for nothing more than a
220 	 * simple way to tell if a device has encountered
221 	 * errors.
222 	 */
223 	atomic_inc(&m->error_count);
224 
225 	if (test_and_set_bit(error_type, &m->error_type))
226 		return;
227 
228 	if (!errors_handled(ms))
229 		return;
230 
231 	if (m != get_default_mirror(ms))
232 		goto out;
233 
234 	if (!ms->in_sync && !keep_log(ms)) {
235 		/*
236 		 * Better to issue requests to same failing device
237 		 * than to risk returning corrupt data.
238 		 */
239 		DMERR("Primary mirror (%s) failed while out-of-sync: "
240 		      "Reads may fail.", m->dev->name);
241 		goto out;
242 	}
243 
244 	new = get_valid_mirror(ms);
245 	if (new)
246 		set_default_mirror(new);
247 	else
248 		DMWARN("All sides of mirror have failed.");
249 
250 out:
251 	schedule_work(&ms->trigger_event);
252 }
253 
254 static int mirror_flush(struct dm_target *ti)
255 {
256 	struct mirror_set *ms = ti->private;
257 	unsigned long error_bits;
258 
259 	unsigned int i;
260 	struct dm_io_region io[ms->nr_mirrors];
261 	struct mirror *m;
262 	struct dm_io_request io_req = {
263 		.bi_op = REQ_OP_WRITE,
264 		.bi_op_flags = WRITE_FLUSH,
265 		.mem.type = DM_IO_KMEM,
266 		.mem.ptr.addr = NULL,
267 		.client = ms->io_client,
268 	};
269 
270 	for (i = 0, m = ms->mirror; i < ms->nr_mirrors; i++, m++) {
271 		io[i].bdev = m->dev->bdev;
272 		io[i].sector = 0;
273 		io[i].count = 0;
274 	}
275 
276 	error_bits = -1;
277 	dm_io(&io_req, ms->nr_mirrors, io, &error_bits);
278 	if (unlikely(error_bits != 0)) {
279 		for (i = 0; i < ms->nr_mirrors; i++)
280 			if (test_bit(i, &error_bits))
281 				fail_mirror(ms->mirror + i,
282 					    DM_RAID1_FLUSH_ERROR);
283 		return -EIO;
284 	}
285 
286 	return 0;
287 }
288 
289 /*-----------------------------------------------------------------
290  * Recovery.
291  *
292  * When a mirror is first activated we may find that some regions
293  * are in the no-sync state.  We have to recover these by
294  * recopying from the default mirror to all the others.
295  *---------------------------------------------------------------*/
296 static void recovery_complete(int read_err, unsigned long write_err,
297 			      void *context)
298 {
299 	struct dm_region *reg = context;
300 	struct mirror_set *ms = dm_rh_region_context(reg);
301 	int m, bit = 0;
302 
303 	if (read_err) {
304 		/* Read error means the failure of default mirror. */
305 		DMERR_LIMIT("Unable to read primary mirror during recovery");
306 		fail_mirror(get_default_mirror(ms), DM_RAID1_SYNC_ERROR);
307 	}
308 
309 	if (write_err) {
310 		DMERR_LIMIT("Write error during recovery (error = 0x%lx)",
311 			    write_err);
312 		/*
313 		 * Bits correspond to devices (excluding default mirror).
314 		 * The default mirror cannot change during recovery.
315 		 */
316 		for (m = 0; m < ms->nr_mirrors; m++) {
317 			if (&ms->mirror[m] == get_default_mirror(ms))
318 				continue;
319 			if (test_bit(bit, &write_err))
320 				fail_mirror(ms->mirror + m,
321 					    DM_RAID1_SYNC_ERROR);
322 			bit++;
323 		}
324 	}
325 
326 	dm_rh_recovery_end(reg, !(read_err || write_err));
327 }
328 
329 static int recover(struct mirror_set *ms, struct dm_region *reg)
330 {
331 	int r;
332 	unsigned i;
333 	struct dm_io_region from, to[DM_KCOPYD_MAX_REGIONS], *dest;
334 	struct mirror *m;
335 	unsigned long flags = 0;
336 	region_t key = dm_rh_get_region_key(reg);
337 	sector_t region_size = dm_rh_get_region_size(ms->rh);
338 
339 	/* fill in the source */
340 	m = get_default_mirror(ms);
341 	from.bdev = m->dev->bdev;
342 	from.sector = m->offset + dm_rh_region_to_sector(ms->rh, key);
343 	if (key == (ms->nr_regions - 1)) {
344 		/*
345 		 * The final region may be smaller than
346 		 * region_size.
347 		 */
348 		from.count = ms->ti->len & (region_size - 1);
349 		if (!from.count)
350 			from.count = region_size;
351 	} else
352 		from.count = region_size;
353 
354 	/* fill in the destinations */
355 	for (i = 0, dest = to; i < ms->nr_mirrors; i++) {
356 		if (&ms->mirror[i] == get_default_mirror(ms))
357 			continue;
358 
359 		m = ms->mirror + i;
360 		dest->bdev = m->dev->bdev;
361 		dest->sector = m->offset + dm_rh_region_to_sector(ms->rh, key);
362 		dest->count = from.count;
363 		dest++;
364 	}
365 
366 	/* hand to kcopyd */
367 	if (!errors_handled(ms))
368 		set_bit(DM_KCOPYD_IGNORE_ERROR, &flags);
369 
370 	r = dm_kcopyd_copy(ms->kcopyd_client, &from, ms->nr_mirrors - 1, to,
371 			   flags, recovery_complete, reg);
372 
373 	return r;
374 }
375 
376 static void reset_ms_flags(struct mirror_set *ms)
377 {
378 	unsigned int m;
379 
380 	ms->leg_failure = 0;
381 	for (m = 0; m < ms->nr_mirrors; m++) {
382 		atomic_set(&(ms->mirror[m].error_count), 0);
383 		ms->mirror[m].error_type = 0;
384 	}
385 }
386 
387 static void do_recovery(struct mirror_set *ms)
388 {
389 	struct dm_region *reg;
390 	struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
391 	int r;
392 
393 	/*
394 	 * Start quiescing some regions.
395 	 */
396 	dm_rh_recovery_prepare(ms->rh);
397 
398 	/*
399 	 * Copy any already quiesced regions.
400 	 */
401 	while ((reg = dm_rh_recovery_start(ms->rh))) {
402 		r = recover(ms, reg);
403 		if (r)
404 			dm_rh_recovery_end(reg, 0);
405 	}
406 
407 	/*
408 	 * Update the in sync flag.
409 	 */
410 	if (!ms->in_sync &&
411 	    (log->type->get_sync_count(log) == ms->nr_regions)) {
412 		/* the sync is complete */
413 		dm_table_event(ms->ti->table);
414 		ms->in_sync = 1;
415 		reset_ms_flags(ms);
416 	}
417 }
418 
419 /*-----------------------------------------------------------------
420  * Reads
421  *---------------------------------------------------------------*/
422 static struct mirror *choose_mirror(struct mirror_set *ms, sector_t sector)
423 {
424 	struct mirror *m = get_default_mirror(ms);
425 
426 	do {
427 		if (likely(!atomic_read(&m->error_count)))
428 			return m;
429 
430 		if (m-- == ms->mirror)
431 			m += ms->nr_mirrors;
432 	} while (m != get_default_mirror(ms));
433 
434 	return NULL;
435 }
436 
437 static int default_ok(struct mirror *m)
438 {
439 	struct mirror *default_mirror = get_default_mirror(m->ms);
440 
441 	return !atomic_read(&default_mirror->error_count);
442 }
443 
444 static int mirror_available(struct mirror_set *ms, struct bio *bio)
445 {
446 	struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
447 	region_t region = dm_rh_bio_to_region(ms->rh, bio);
448 
449 	if (log->type->in_sync(log, region, 0))
450 		return choose_mirror(ms,  bio->bi_iter.bi_sector) ? 1 : 0;
451 
452 	return 0;
453 }
454 
455 /*
456  * remap a buffer to a particular mirror.
457  */
458 static sector_t map_sector(struct mirror *m, struct bio *bio)
459 {
460 	if (unlikely(!bio->bi_iter.bi_size))
461 		return 0;
462 	return m->offset + dm_target_offset(m->ms->ti, bio->bi_iter.bi_sector);
463 }
464 
465 static void map_bio(struct mirror *m, struct bio *bio)
466 {
467 	bio->bi_bdev = m->dev->bdev;
468 	bio->bi_iter.bi_sector = map_sector(m, bio);
469 }
470 
471 static void map_region(struct dm_io_region *io, struct mirror *m,
472 		       struct bio *bio)
473 {
474 	io->bdev = m->dev->bdev;
475 	io->sector = map_sector(m, bio);
476 	io->count = bio_sectors(bio);
477 }
478 
479 static void hold_bio(struct mirror_set *ms, struct bio *bio)
480 {
481 	/*
482 	 * Lock is required to avoid race condition during suspend
483 	 * process.
484 	 */
485 	spin_lock_irq(&ms->lock);
486 
487 	if (atomic_read(&ms->suspend)) {
488 		spin_unlock_irq(&ms->lock);
489 
490 		/*
491 		 * If device is suspended, complete the bio.
492 		 */
493 		if (dm_noflush_suspending(ms->ti))
494 			bio->bi_error = DM_ENDIO_REQUEUE;
495 		else
496 			bio->bi_error = -EIO;
497 
498 		bio_endio(bio);
499 		return;
500 	}
501 
502 	/*
503 	 * Hold bio until the suspend is complete.
504 	 */
505 	bio_list_add(&ms->holds, bio);
506 	spin_unlock_irq(&ms->lock);
507 }
508 
509 /*-----------------------------------------------------------------
510  * Reads
511  *---------------------------------------------------------------*/
512 static void read_callback(unsigned long error, void *context)
513 {
514 	struct bio *bio = context;
515 	struct mirror *m;
516 
517 	m = bio_get_m(bio);
518 	bio_set_m(bio, NULL);
519 
520 	if (likely(!error)) {
521 		bio_endio(bio);
522 		return;
523 	}
524 
525 	fail_mirror(m, DM_RAID1_READ_ERROR);
526 
527 	if (likely(default_ok(m)) || mirror_available(m->ms, bio)) {
528 		DMWARN_LIMIT("Read failure on mirror device %s.  "
529 			     "Trying alternative device.",
530 			     m->dev->name);
531 		queue_bio(m->ms, bio, bio_data_dir(bio));
532 		return;
533 	}
534 
535 	DMERR_LIMIT("Read failure on mirror device %s.  Failing I/O.",
536 		    m->dev->name);
537 	bio_io_error(bio);
538 }
539 
540 /* Asynchronous read. */
541 static void read_async_bio(struct mirror *m, struct bio *bio)
542 {
543 	struct dm_io_region io;
544 	struct dm_io_request io_req = {
545 		.bi_op = REQ_OP_READ,
546 		.bi_op_flags = 0,
547 		.mem.type = DM_IO_BIO,
548 		.mem.ptr.bio = bio,
549 		.notify.fn = read_callback,
550 		.notify.context = bio,
551 		.client = m->ms->io_client,
552 	};
553 
554 	map_region(&io, m, bio);
555 	bio_set_m(bio, m);
556 	BUG_ON(dm_io(&io_req, 1, &io, NULL));
557 }
558 
559 static inline int region_in_sync(struct mirror_set *ms, region_t region,
560 				 int may_block)
561 {
562 	int state = dm_rh_get_state(ms->rh, region, may_block);
563 	return state == DM_RH_CLEAN || state == DM_RH_DIRTY;
564 }
565 
566 static void do_reads(struct mirror_set *ms, struct bio_list *reads)
567 {
568 	region_t region;
569 	struct bio *bio;
570 	struct mirror *m;
571 
572 	while ((bio = bio_list_pop(reads))) {
573 		region = dm_rh_bio_to_region(ms->rh, bio);
574 		m = get_default_mirror(ms);
575 
576 		/*
577 		 * We can only read balance if the region is in sync.
578 		 */
579 		if (likely(region_in_sync(ms, region, 1)))
580 			m = choose_mirror(ms, bio->bi_iter.bi_sector);
581 		else if (m && atomic_read(&m->error_count))
582 			m = NULL;
583 
584 		if (likely(m))
585 			read_async_bio(m, bio);
586 		else
587 			bio_io_error(bio);
588 	}
589 }
590 
591 /*-----------------------------------------------------------------
592  * Writes.
593  *
594  * We do different things with the write io depending on the
595  * state of the region that it's in:
596  *
597  * SYNC: 	increment pending, use kcopyd to write to *all* mirrors
598  * RECOVERING:	delay the io until recovery completes
599  * NOSYNC:	increment pending, just write to the default mirror
600  *---------------------------------------------------------------*/
601 
602 
603 static void write_callback(unsigned long error, void *context)
604 {
605 	unsigned i;
606 	struct bio *bio = (struct bio *) context;
607 	struct mirror_set *ms;
608 	int should_wake = 0;
609 	unsigned long flags;
610 
611 	ms = bio_get_m(bio)->ms;
612 	bio_set_m(bio, NULL);
613 
614 	/*
615 	 * NOTE: We don't decrement the pending count here,
616 	 * instead it is done by the targets endio function.
617 	 * This way we handle both writes to SYNC and NOSYNC
618 	 * regions with the same code.
619 	 */
620 	if (likely(!error)) {
621 		bio_endio(bio);
622 		return;
623 	}
624 
625 	/*
626 	 * If the bio is discard, return an error, but do not
627 	 * degrade the array.
628 	 */
629 	if (bio_op(bio) == REQ_OP_DISCARD) {
630 		bio->bi_error = -EOPNOTSUPP;
631 		bio_endio(bio);
632 		return;
633 	}
634 
635 	for (i = 0; i < ms->nr_mirrors; i++)
636 		if (test_bit(i, &error))
637 			fail_mirror(ms->mirror + i, DM_RAID1_WRITE_ERROR);
638 
639 	/*
640 	 * Need to raise event.  Since raising
641 	 * events can block, we need to do it in
642 	 * the main thread.
643 	 */
644 	spin_lock_irqsave(&ms->lock, flags);
645 	if (!ms->failures.head)
646 		should_wake = 1;
647 	bio_list_add(&ms->failures, bio);
648 	spin_unlock_irqrestore(&ms->lock, flags);
649 	if (should_wake)
650 		wakeup_mirrord(ms);
651 }
652 
653 static void do_write(struct mirror_set *ms, struct bio *bio)
654 {
655 	unsigned int i;
656 	struct dm_io_region io[ms->nr_mirrors], *dest = io;
657 	struct mirror *m;
658 	struct dm_io_request io_req = {
659 		.bi_op = REQ_OP_WRITE,
660 		.bi_op_flags = bio->bi_opf & WRITE_FLUSH_FUA,
661 		.mem.type = DM_IO_BIO,
662 		.mem.ptr.bio = bio,
663 		.notify.fn = write_callback,
664 		.notify.context = bio,
665 		.client = ms->io_client,
666 	};
667 
668 	if (bio_op(bio) == REQ_OP_DISCARD) {
669 		io_req.bi_op = REQ_OP_DISCARD;
670 		io_req.mem.type = DM_IO_KMEM;
671 		io_req.mem.ptr.addr = NULL;
672 	}
673 
674 	for (i = 0, m = ms->mirror; i < ms->nr_mirrors; i++, m++)
675 		map_region(dest++, m, bio);
676 
677 	/*
678 	 * Use default mirror because we only need it to retrieve the reference
679 	 * to the mirror set in write_callback().
680 	 */
681 	bio_set_m(bio, get_default_mirror(ms));
682 
683 	BUG_ON(dm_io(&io_req, ms->nr_mirrors, io, NULL));
684 }
685 
686 static void do_writes(struct mirror_set *ms, struct bio_list *writes)
687 {
688 	int state;
689 	struct bio *bio;
690 	struct bio_list sync, nosync, recover, *this_list = NULL;
691 	struct bio_list requeue;
692 	struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
693 	region_t region;
694 
695 	if (!writes->head)
696 		return;
697 
698 	/*
699 	 * Classify each write.
700 	 */
701 	bio_list_init(&sync);
702 	bio_list_init(&nosync);
703 	bio_list_init(&recover);
704 	bio_list_init(&requeue);
705 
706 	while ((bio = bio_list_pop(writes))) {
707 		if ((bio->bi_opf & REQ_PREFLUSH) ||
708 		    (bio_op(bio) == REQ_OP_DISCARD)) {
709 			bio_list_add(&sync, bio);
710 			continue;
711 		}
712 
713 		region = dm_rh_bio_to_region(ms->rh, bio);
714 
715 		if (log->type->is_remote_recovering &&
716 		    log->type->is_remote_recovering(log, region)) {
717 			bio_list_add(&requeue, bio);
718 			continue;
719 		}
720 
721 		state = dm_rh_get_state(ms->rh, region, 1);
722 		switch (state) {
723 		case DM_RH_CLEAN:
724 		case DM_RH_DIRTY:
725 			this_list = &sync;
726 			break;
727 
728 		case DM_RH_NOSYNC:
729 			this_list = &nosync;
730 			break;
731 
732 		case DM_RH_RECOVERING:
733 			this_list = &recover;
734 			break;
735 		}
736 
737 		bio_list_add(this_list, bio);
738 	}
739 
740 	/*
741 	 * Add bios that are delayed due to remote recovery
742 	 * back on to the write queue
743 	 */
744 	if (unlikely(requeue.head)) {
745 		spin_lock_irq(&ms->lock);
746 		bio_list_merge(&ms->writes, &requeue);
747 		spin_unlock_irq(&ms->lock);
748 		delayed_wake(ms);
749 	}
750 
751 	/*
752 	 * Increment the pending counts for any regions that will
753 	 * be written to (writes to recover regions are going to
754 	 * be delayed).
755 	 */
756 	dm_rh_inc_pending(ms->rh, &sync);
757 	dm_rh_inc_pending(ms->rh, &nosync);
758 
759 	/*
760 	 * If the flush fails on a previous call and succeeds here,
761 	 * we must not reset the log_failure variable.  We need
762 	 * userspace interaction to do that.
763 	 */
764 	ms->log_failure = dm_rh_flush(ms->rh) ? 1 : ms->log_failure;
765 
766 	/*
767 	 * Dispatch io.
768 	 */
769 	if (unlikely(ms->log_failure) && errors_handled(ms)) {
770 		spin_lock_irq(&ms->lock);
771 		bio_list_merge(&ms->failures, &sync);
772 		spin_unlock_irq(&ms->lock);
773 		wakeup_mirrord(ms);
774 	} else
775 		while ((bio = bio_list_pop(&sync)))
776 			do_write(ms, bio);
777 
778 	while ((bio = bio_list_pop(&recover)))
779 		dm_rh_delay(ms->rh, bio);
780 
781 	while ((bio = bio_list_pop(&nosync))) {
782 		if (unlikely(ms->leg_failure) && errors_handled(ms) && !keep_log(ms)) {
783 			spin_lock_irq(&ms->lock);
784 			bio_list_add(&ms->failures, bio);
785 			spin_unlock_irq(&ms->lock);
786 			wakeup_mirrord(ms);
787 		} else {
788 			map_bio(get_default_mirror(ms), bio);
789 			generic_make_request(bio);
790 		}
791 	}
792 }
793 
794 static void do_failures(struct mirror_set *ms, struct bio_list *failures)
795 {
796 	struct bio *bio;
797 
798 	if (likely(!failures->head))
799 		return;
800 
801 	/*
802 	 * If the log has failed, unattempted writes are being
803 	 * put on the holds list.  We can't issue those writes
804 	 * until a log has been marked, so we must store them.
805 	 *
806 	 * If a 'noflush' suspend is in progress, we can requeue
807 	 * the I/O's to the core.  This give userspace a chance
808 	 * to reconfigure the mirror, at which point the core
809 	 * will reissue the writes.  If the 'noflush' flag is
810 	 * not set, we have no choice but to return errors.
811 	 *
812 	 * Some writes on the failures list may have been
813 	 * submitted before the log failure and represent a
814 	 * failure to write to one of the devices.  It is ok
815 	 * for us to treat them the same and requeue them
816 	 * as well.
817 	 */
818 	while ((bio = bio_list_pop(failures))) {
819 		if (!ms->log_failure) {
820 			ms->in_sync = 0;
821 			dm_rh_mark_nosync(ms->rh, bio);
822 		}
823 
824 		/*
825 		 * If all the legs are dead, fail the I/O.
826 		 * If the device has failed and keep_log is enabled,
827 		 * fail the I/O.
828 		 *
829 		 * If we have been told to handle errors, and keep_log
830 		 * isn't enabled, hold the bio and wait for userspace to
831 		 * deal with the problem.
832 		 *
833 		 * Otherwise pretend that the I/O succeeded. (This would
834 		 * be wrong if the failed leg returned after reboot and
835 		 * got replicated back to the good legs.)
836 		 */
837 		if (unlikely(!get_valid_mirror(ms) || (keep_log(ms) && ms->log_failure)))
838 			bio_io_error(bio);
839 		else if (errors_handled(ms) && !keep_log(ms))
840 			hold_bio(ms, bio);
841 		else
842 			bio_endio(bio);
843 	}
844 }
845 
846 static void trigger_event(struct work_struct *work)
847 {
848 	struct mirror_set *ms =
849 		container_of(work, struct mirror_set, trigger_event);
850 
851 	dm_table_event(ms->ti->table);
852 }
853 
854 /*-----------------------------------------------------------------
855  * kmirrord
856  *---------------------------------------------------------------*/
857 static void do_mirror(struct work_struct *work)
858 {
859 	struct mirror_set *ms = container_of(work, struct mirror_set,
860 					     kmirrord_work);
861 	struct bio_list reads, writes, failures;
862 	unsigned long flags;
863 
864 	spin_lock_irqsave(&ms->lock, flags);
865 	reads = ms->reads;
866 	writes = ms->writes;
867 	failures = ms->failures;
868 	bio_list_init(&ms->reads);
869 	bio_list_init(&ms->writes);
870 	bio_list_init(&ms->failures);
871 	spin_unlock_irqrestore(&ms->lock, flags);
872 
873 	dm_rh_update_states(ms->rh, errors_handled(ms));
874 	do_recovery(ms);
875 	do_reads(ms, &reads);
876 	do_writes(ms, &writes);
877 	do_failures(ms, &failures);
878 }
879 
880 /*-----------------------------------------------------------------
881  * Target functions
882  *---------------------------------------------------------------*/
883 static struct mirror_set *alloc_context(unsigned int nr_mirrors,
884 					uint32_t region_size,
885 					struct dm_target *ti,
886 					struct dm_dirty_log *dl)
887 {
888 	size_t len;
889 	struct mirror_set *ms = NULL;
890 
891 	len = sizeof(*ms) + (sizeof(ms->mirror[0]) * nr_mirrors);
892 
893 	ms = kzalloc(len, GFP_KERNEL);
894 	if (!ms) {
895 		ti->error = "Cannot allocate mirror context";
896 		return NULL;
897 	}
898 
899 	spin_lock_init(&ms->lock);
900 	bio_list_init(&ms->reads);
901 	bio_list_init(&ms->writes);
902 	bio_list_init(&ms->failures);
903 	bio_list_init(&ms->holds);
904 
905 	ms->ti = ti;
906 	ms->nr_mirrors = nr_mirrors;
907 	ms->nr_regions = dm_sector_div_up(ti->len, region_size);
908 	ms->in_sync = 0;
909 	ms->log_failure = 0;
910 	ms->leg_failure = 0;
911 	atomic_set(&ms->suspend, 0);
912 	atomic_set(&ms->default_mirror, DEFAULT_MIRROR);
913 
914 	ms->io_client = dm_io_client_create();
915 	if (IS_ERR(ms->io_client)) {
916 		ti->error = "Error creating dm_io client";
917 		kfree(ms);
918  		return NULL;
919 	}
920 
921 	ms->rh = dm_region_hash_create(ms, dispatch_bios, wakeup_mirrord,
922 				       wakeup_all_recovery_waiters,
923 				       ms->ti->begin, MAX_RECOVERY,
924 				       dl, region_size, ms->nr_regions);
925 	if (IS_ERR(ms->rh)) {
926 		ti->error = "Error creating dirty region hash";
927 		dm_io_client_destroy(ms->io_client);
928 		kfree(ms);
929 		return NULL;
930 	}
931 
932 	return ms;
933 }
934 
935 static void free_context(struct mirror_set *ms, struct dm_target *ti,
936 			 unsigned int m)
937 {
938 	while (m--)
939 		dm_put_device(ti, ms->mirror[m].dev);
940 
941 	dm_io_client_destroy(ms->io_client);
942 	dm_region_hash_destroy(ms->rh);
943 	kfree(ms);
944 }
945 
946 static int get_mirror(struct mirror_set *ms, struct dm_target *ti,
947 		      unsigned int mirror, char **argv)
948 {
949 	unsigned long long offset;
950 	char dummy;
951 	int ret;
952 
953 	if (sscanf(argv[1], "%llu%c", &offset, &dummy) != 1) {
954 		ti->error = "Invalid offset";
955 		return -EINVAL;
956 	}
957 
958 	ret = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table),
959 			    &ms->mirror[mirror].dev);
960 	if (ret) {
961 		ti->error = "Device lookup failure";
962 		return ret;
963 	}
964 
965 	ms->mirror[mirror].ms = ms;
966 	atomic_set(&(ms->mirror[mirror].error_count), 0);
967 	ms->mirror[mirror].error_type = 0;
968 	ms->mirror[mirror].offset = offset;
969 
970 	return 0;
971 }
972 
973 /*
974  * Create dirty log: log_type #log_params <log_params>
975  */
976 static struct dm_dirty_log *create_dirty_log(struct dm_target *ti,
977 					     unsigned argc, char **argv,
978 					     unsigned *args_used)
979 {
980 	unsigned param_count;
981 	struct dm_dirty_log *dl;
982 	char dummy;
983 
984 	if (argc < 2) {
985 		ti->error = "Insufficient mirror log arguments";
986 		return NULL;
987 	}
988 
989 	if (sscanf(argv[1], "%u%c", &param_count, &dummy) != 1) {
990 		ti->error = "Invalid mirror log argument count";
991 		return NULL;
992 	}
993 
994 	*args_used = 2 + param_count;
995 
996 	if (argc < *args_used) {
997 		ti->error = "Insufficient mirror log arguments";
998 		return NULL;
999 	}
1000 
1001 	dl = dm_dirty_log_create(argv[0], ti, mirror_flush, param_count,
1002 				 argv + 2);
1003 	if (!dl) {
1004 		ti->error = "Error creating mirror dirty log";
1005 		return NULL;
1006 	}
1007 
1008 	return dl;
1009 }
1010 
1011 static int parse_features(struct mirror_set *ms, unsigned argc, char **argv,
1012 			  unsigned *args_used)
1013 {
1014 	unsigned num_features;
1015 	struct dm_target *ti = ms->ti;
1016 	char dummy;
1017 	int i;
1018 
1019 	*args_used = 0;
1020 
1021 	if (!argc)
1022 		return 0;
1023 
1024 	if (sscanf(argv[0], "%u%c", &num_features, &dummy) != 1) {
1025 		ti->error = "Invalid number of features";
1026 		return -EINVAL;
1027 	}
1028 
1029 	argc--;
1030 	argv++;
1031 	(*args_used)++;
1032 
1033 	if (num_features > argc) {
1034 		ti->error = "Not enough arguments to support feature count";
1035 		return -EINVAL;
1036 	}
1037 
1038 	for (i = 0; i < num_features; i++) {
1039 		if (!strcmp("handle_errors", argv[0]))
1040 			ms->features |= DM_RAID1_HANDLE_ERRORS;
1041 		else if (!strcmp("keep_log", argv[0]))
1042 			ms->features |= DM_RAID1_KEEP_LOG;
1043 		else {
1044 			ti->error = "Unrecognised feature requested";
1045 			return -EINVAL;
1046 		}
1047 
1048 		argc--;
1049 		argv++;
1050 		(*args_used)++;
1051 	}
1052 	if (!errors_handled(ms) && keep_log(ms)) {
1053 		ti->error = "keep_log feature requires the handle_errors feature";
1054 		return -EINVAL;
1055 	}
1056 
1057 	return 0;
1058 }
1059 
1060 /*
1061  * Construct a mirror mapping:
1062  *
1063  * log_type #log_params <log_params>
1064  * #mirrors [mirror_path offset]{2,}
1065  * [#features <features>]
1066  *
1067  * log_type is "core" or "disk"
1068  * #log_params is between 1 and 3
1069  *
1070  * If present, supported features are "handle_errors" and "keep_log".
1071  */
1072 static int mirror_ctr(struct dm_target *ti, unsigned int argc, char **argv)
1073 {
1074 	int r;
1075 	unsigned int nr_mirrors, m, args_used;
1076 	struct mirror_set *ms;
1077 	struct dm_dirty_log *dl;
1078 	char dummy;
1079 
1080 	dl = create_dirty_log(ti, argc, argv, &args_used);
1081 	if (!dl)
1082 		return -EINVAL;
1083 
1084 	argv += args_used;
1085 	argc -= args_used;
1086 
1087 	if (!argc || sscanf(argv[0], "%u%c", &nr_mirrors, &dummy) != 1 ||
1088 	    nr_mirrors < 2 || nr_mirrors > DM_KCOPYD_MAX_REGIONS + 1) {
1089 		ti->error = "Invalid number of mirrors";
1090 		dm_dirty_log_destroy(dl);
1091 		return -EINVAL;
1092 	}
1093 
1094 	argv++, argc--;
1095 
1096 	if (argc < nr_mirrors * 2) {
1097 		ti->error = "Too few mirror arguments";
1098 		dm_dirty_log_destroy(dl);
1099 		return -EINVAL;
1100 	}
1101 
1102 	ms = alloc_context(nr_mirrors, dl->type->get_region_size(dl), ti, dl);
1103 	if (!ms) {
1104 		dm_dirty_log_destroy(dl);
1105 		return -ENOMEM;
1106 	}
1107 
1108 	/* Get the mirror parameter sets */
1109 	for (m = 0; m < nr_mirrors; m++) {
1110 		r = get_mirror(ms, ti, m, argv);
1111 		if (r) {
1112 			free_context(ms, ti, m);
1113 			return r;
1114 		}
1115 		argv += 2;
1116 		argc -= 2;
1117 	}
1118 
1119 	ti->private = ms;
1120 
1121 	r = dm_set_target_max_io_len(ti, dm_rh_get_region_size(ms->rh));
1122 	if (r)
1123 		goto err_free_context;
1124 
1125 	ti->num_flush_bios = 1;
1126 	ti->num_discard_bios = 1;
1127 	ti->per_io_data_size = sizeof(struct dm_raid1_bio_record);
1128 	ti->discard_zeroes_data_unsupported = true;
1129 
1130 	ms->kmirrord_wq = alloc_workqueue("kmirrord", WQ_MEM_RECLAIM, 0);
1131 	if (!ms->kmirrord_wq) {
1132 		DMERR("couldn't start kmirrord");
1133 		r = -ENOMEM;
1134 		goto err_free_context;
1135 	}
1136 	INIT_WORK(&ms->kmirrord_work, do_mirror);
1137 	init_timer(&ms->timer);
1138 	ms->timer_pending = 0;
1139 	INIT_WORK(&ms->trigger_event, trigger_event);
1140 
1141 	r = parse_features(ms, argc, argv, &args_used);
1142 	if (r)
1143 		goto err_destroy_wq;
1144 
1145 	argv += args_used;
1146 	argc -= args_used;
1147 
1148 	/*
1149 	 * Any read-balancing addition depends on the
1150 	 * DM_RAID1_HANDLE_ERRORS flag being present.
1151 	 * This is because the decision to balance depends
1152 	 * on the sync state of a region.  If the above
1153 	 * flag is not present, we ignore errors; and
1154 	 * the sync state may be inaccurate.
1155 	 */
1156 
1157 	if (argc) {
1158 		ti->error = "Too many mirror arguments";
1159 		r = -EINVAL;
1160 		goto err_destroy_wq;
1161 	}
1162 
1163 	ms->kcopyd_client = dm_kcopyd_client_create(&dm_kcopyd_throttle);
1164 	if (IS_ERR(ms->kcopyd_client)) {
1165 		r = PTR_ERR(ms->kcopyd_client);
1166 		goto err_destroy_wq;
1167 	}
1168 
1169 	wakeup_mirrord(ms);
1170 	return 0;
1171 
1172 err_destroy_wq:
1173 	destroy_workqueue(ms->kmirrord_wq);
1174 err_free_context:
1175 	free_context(ms, ti, ms->nr_mirrors);
1176 	return r;
1177 }
1178 
1179 static void mirror_dtr(struct dm_target *ti)
1180 {
1181 	struct mirror_set *ms = (struct mirror_set *) ti->private;
1182 
1183 	del_timer_sync(&ms->timer);
1184 	flush_workqueue(ms->kmirrord_wq);
1185 	flush_work(&ms->trigger_event);
1186 	dm_kcopyd_client_destroy(ms->kcopyd_client);
1187 	destroy_workqueue(ms->kmirrord_wq);
1188 	free_context(ms, ti, ms->nr_mirrors);
1189 }
1190 
1191 /*
1192  * Mirror mapping function
1193  */
1194 static int mirror_map(struct dm_target *ti, struct bio *bio)
1195 {
1196 	int r, rw = bio_data_dir(bio);
1197 	struct mirror *m;
1198 	struct mirror_set *ms = ti->private;
1199 	struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
1200 	struct dm_raid1_bio_record *bio_record =
1201 	  dm_per_bio_data(bio, sizeof(struct dm_raid1_bio_record));
1202 
1203 	bio_record->details.bi_bdev = NULL;
1204 
1205 	if (rw == WRITE) {
1206 		/* Save region for mirror_end_io() handler */
1207 		bio_record->write_region = dm_rh_bio_to_region(ms->rh, bio);
1208 		queue_bio(ms, bio, rw);
1209 		return DM_MAPIO_SUBMITTED;
1210 	}
1211 
1212 	r = log->type->in_sync(log, dm_rh_bio_to_region(ms->rh, bio), 0);
1213 	if (r < 0 && r != -EWOULDBLOCK)
1214 		return r;
1215 
1216 	/*
1217 	 * If region is not in-sync queue the bio.
1218 	 */
1219 	if (!r || (r == -EWOULDBLOCK)) {
1220 		if (bio->bi_opf & REQ_RAHEAD)
1221 			return -EWOULDBLOCK;
1222 
1223 		queue_bio(ms, bio, rw);
1224 		return DM_MAPIO_SUBMITTED;
1225 	}
1226 
1227 	/*
1228 	 * The region is in-sync and we can perform reads directly.
1229 	 * Store enough information so we can retry if it fails.
1230 	 */
1231 	m = choose_mirror(ms, bio->bi_iter.bi_sector);
1232 	if (unlikely(!m))
1233 		return -EIO;
1234 
1235 	dm_bio_record(&bio_record->details, bio);
1236 	bio_record->m = m;
1237 
1238 	map_bio(m, bio);
1239 
1240 	return DM_MAPIO_REMAPPED;
1241 }
1242 
1243 static int mirror_end_io(struct dm_target *ti, struct bio *bio, int error)
1244 {
1245 	int rw = bio_data_dir(bio);
1246 	struct mirror_set *ms = (struct mirror_set *) ti->private;
1247 	struct mirror *m = NULL;
1248 	struct dm_bio_details *bd = NULL;
1249 	struct dm_raid1_bio_record *bio_record =
1250 	  dm_per_bio_data(bio, sizeof(struct dm_raid1_bio_record));
1251 
1252 	/*
1253 	 * We need to dec pending if this was a write.
1254 	 */
1255 	if (rw == WRITE) {
1256 		if (!(bio->bi_opf & REQ_PREFLUSH) &&
1257 		    bio_op(bio) != REQ_OP_DISCARD)
1258 			dm_rh_dec(ms->rh, bio_record->write_region);
1259 		return error;
1260 	}
1261 
1262 	if (error == -EOPNOTSUPP)
1263 		goto out;
1264 
1265 	if ((error == -EWOULDBLOCK) && (bio->bi_opf & REQ_RAHEAD))
1266 		goto out;
1267 
1268 	if (unlikely(error)) {
1269 		if (!bio_record->details.bi_bdev) {
1270 			/*
1271 			 * There wasn't enough memory to record necessary
1272 			 * information for a retry or there was no other
1273 			 * mirror in-sync.
1274 			 */
1275 			DMERR_LIMIT("Mirror read failed.");
1276 			return -EIO;
1277 		}
1278 
1279 		m = bio_record->m;
1280 
1281 		DMERR("Mirror read failed from %s. Trying alternative device.",
1282 		      m->dev->name);
1283 
1284 		fail_mirror(m, DM_RAID1_READ_ERROR);
1285 
1286 		/*
1287 		 * A failed read is requeued for another attempt using an intact
1288 		 * mirror.
1289 		 */
1290 		if (default_ok(m) || mirror_available(ms, bio)) {
1291 			bd = &bio_record->details;
1292 
1293 			dm_bio_restore(bd, bio);
1294 			bio_record->details.bi_bdev = NULL;
1295 
1296 			queue_bio(ms, bio, rw);
1297 			return DM_ENDIO_INCOMPLETE;
1298 		}
1299 		DMERR("All replicated volumes dead, failing I/O");
1300 	}
1301 
1302 out:
1303 	bio_record->details.bi_bdev = NULL;
1304 
1305 	return error;
1306 }
1307 
1308 static void mirror_presuspend(struct dm_target *ti)
1309 {
1310 	struct mirror_set *ms = (struct mirror_set *) ti->private;
1311 	struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
1312 
1313 	struct bio_list holds;
1314 	struct bio *bio;
1315 
1316 	atomic_set(&ms->suspend, 1);
1317 
1318 	/*
1319 	 * Process bios in the hold list to start recovery waiting
1320 	 * for bios in the hold list. After the process, no bio has
1321 	 * a chance to be added in the hold list because ms->suspend
1322 	 * is set.
1323 	 */
1324 	spin_lock_irq(&ms->lock);
1325 	holds = ms->holds;
1326 	bio_list_init(&ms->holds);
1327 	spin_unlock_irq(&ms->lock);
1328 
1329 	while ((bio = bio_list_pop(&holds)))
1330 		hold_bio(ms, bio);
1331 
1332 	/*
1333 	 * We must finish up all the work that we've
1334 	 * generated (i.e. recovery work).
1335 	 */
1336 	dm_rh_stop_recovery(ms->rh);
1337 
1338 	wait_event(_kmirrord_recovery_stopped,
1339 		   !dm_rh_recovery_in_flight(ms->rh));
1340 
1341 	if (log->type->presuspend && log->type->presuspend(log))
1342 		/* FIXME: need better error handling */
1343 		DMWARN("log presuspend failed");
1344 
1345 	/*
1346 	 * Now that recovery is complete/stopped and the
1347 	 * delayed bios are queued, we need to wait for
1348 	 * the worker thread to complete.  This way,
1349 	 * we know that all of our I/O has been pushed.
1350 	 */
1351 	flush_workqueue(ms->kmirrord_wq);
1352 }
1353 
1354 static void mirror_postsuspend(struct dm_target *ti)
1355 {
1356 	struct mirror_set *ms = ti->private;
1357 	struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
1358 
1359 	if (log->type->postsuspend && log->type->postsuspend(log))
1360 		/* FIXME: need better error handling */
1361 		DMWARN("log postsuspend failed");
1362 }
1363 
1364 static void mirror_resume(struct dm_target *ti)
1365 {
1366 	struct mirror_set *ms = ti->private;
1367 	struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
1368 
1369 	atomic_set(&ms->suspend, 0);
1370 	if (log->type->resume && log->type->resume(log))
1371 		/* FIXME: need better error handling */
1372 		DMWARN("log resume failed");
1373 	dm_rh_start_recovery(ms->rh);
1374 }
1375 
1376 /*
1377  * device_status_char
1378  * @m: mirror device/leg we want the status of
1379  *
1380  * We return one character representing the most severe error
1381  * we have encountered.
1382  *    A => Alive - No failures
1383  *    D => Dead - A write failure occurred leaving mirror out-of-sync
1384  *    S => Sync - A sychronization failure occurred, mirror out-of-sync
1385  *    R => Read - A read failure occurred, mirror data unaffected
1386  *
1387  * Returns: <char>
1388  */
1389 static char device_status_char(struct mirror *m)
1390 {
1391 	if (!atomic_read(&(m->error_count)))
1392 		return 'A';
1393 
1394 	return (test_bit(DM_RAID1_FLUSH_ERROR, &(m->error_type))) ? 'F' :
1395 		(test_bit(DM_RAID1_WRITE_ERROR, &(m->error_type))) ? 'D' :
1396 		(test_bit(DM_RAID1_SYNC_ERROR, &(m->error_type))) ? 'S' :
1397 		(test_bit(DM_RAID1_READ_ERROR, &(m->error_type))) ? 'R' : 'U';
1398 }
1399 
1400 
1401 static void mirror_status(struct dm_target *ti, status_type_t type,
1402 			  unsigned status_flags, char *result, unsigned maxlen)
1403 {
1404 	unsigned int m, sz = 0;
1405 	int num_feature_args = 0;
1406 	struct mirror_set *ms = (struct mirror_set *) ti->private;
1407 	struct dm_dirty_log *log = dm_rh_dirty_log(ms->rh);
1408 	char buffer[ms->nr_mirrors + 1];
1409 
1410 	switch (type) {
1411 	case STATUSTYPE_INFO:
1412 		DMEMIT("%d ", ms->nr_mirrors);
1413 		for (m = 0; m < ms->nr_mirrors; m++) {
1414 			DMEMIT("%s ", ms->mirror[m].dev->name);
1415 			buffer[m] = device_status_char(&(ms->mirror[m]));
1416 		}
1417 		buffer[m] = '\0';
1418 
1419 		DMEMIT("%llu/%llu 1 %s ",
1420 		      (unsigned long long)log->type->get_sync_count(log),
1421 		      (unsigned long long)ms->nr_regions, buffer);
1422 
1423 		sz += log->type->status(log, type, result+sz, maxlen-sz);
1424 
1425 		break;
1426 
1427 	case STATUSTYPE_TABLE:
1428 		sz = log->type->status(log, type, result, maxlen);
1429 
1430 		DMEMIT("%d", ms->nr_mirrors);
1431 		for (m = 0; m < ms->nr_mirrors; m++)
1432 			DMEMIT(" %s %llu", ms->mirror[m].dev->name,
1433 			       (unsigned long long)ms->mirror[m].offset);
1434 
1435 		num_feature_args += !!errors_handled(ms);
1436 		num_feature_args += !!keep_log(ms);
1437 		if (num_feature_args) {
1438 			DMEMIT(" %d", num_feature_args);
1439 			if (errors_handled(ms))
1440 				DMEMIT(" handle_errors");
1441 			if (keep_log(ms))
1442 				DMEMIT(" keep_log");
1443 		}
1444 
1445 		break;
1446 	}
1447 }
1448 
1449 static int mirror_iterate_devices(struct dm_target *ti,
1450 				  iterate_devices_callout_fn fn, void *data)
1451 {
1452 	struct mirror_set *ms = ti->private;
1453 	int ret = 0;
1454 	unsigned i;
1455 
1456 	for (i = 0; !ret && i < ms->nr_mirrors; i++)
1457 		ret = fn(ti, ms->mirror[i].dev,
1458 			 ms->mirror[i].offset, ti->len, data);
1459 
1460 	return ret;
1461 }
1462 
1463 static struct target_type mirror_target = {
1464 	.name	 = "mirror",
1465 	.version = {1, 14, 0},
1466 	.module	 = THIS_MODULE,
1467 	.ctr	 = mirror_ctr,
1468 	.dtr	 = mirror_dtr,
1469 	.map	 = mirror_map,
1470 	.end_io	 = mirror_end_io,
1471 	.presuspend = mirror_presuspend,
1472 	.postsuspend = mirror_postsuspend,
1473 	.resume	 = mirror_resume,
1474 	.status	 = mirror_status,
1475 	.iterate_devices = mirror_iterate_devices,
1476 };
1477 
1478 static int __init dm_mirror_init(void)
1479 {
1480 	int r;
1481 
1482 	r = dm_register_target(&mirror_target);
1483 	if (r < 0) {
1484 		DMERR("Failed to register mirror target");
1485 		goto bad_target;
1486 	}
1487 
1488 	return 0;
1489 
1490 bad_target:
1491 	return r;
1492 }
1493 
1494 static void __exit dm_mirror_exit(void)
1495 {
1496 	dm_unregister_target(&mirror_target);
1497 }
1498 
1499 /* Module hooks */
1500 module_init(dm_mirror_init);
1501 module_exit(dm_mirror_exit);
1502 
1503 MODULE_DESCRIPTION(DM_NAME " mirror target");
1504 MODULE_AUTHOR("Joe Thornber");
1505 MODULE_LICENSE("GPL");
1506