dm-raid1.c

Linux Kernel 2.6.9 for OMAP1710

	r = kcopyd_copy(ms->kcopyd_client, &from, ms->nr_mirrors - 1, to, flags,
			recovery_complete, reg);

	return r;
}

static void do_recovery(struct mirror_set *ms)
{
	int r;
	struct region *reg;
	struct dirty_log *log = ms->rh.log;

	/*
	 * Start quiescing some regions.
	 */
	rh_recovery_prepare(&ms->rh);

	/*
	 * Copy any already quiesced regions.
	 */
	while ((reg = rh_recovery_start(&ms->rh))) {
		r = recover(ms, reg);
		if (r)
			rh_recovery_end(reg, 0);
	}

	/*
	 * Update the in sync flag.
	 */
	if (!ms->in_sync &&
	    (log->type->get_sync_count(log) == ms->nr_regions)) {
		/* the sync is complete */
		dm_table_event(ms->ti->table);
		ms->in_sync = 1;
	}
}

/*-----------------------------------------------------------------
 * Reads
 *---------------------------------------------------------------*/
static struct mirror *choose_mirror(struct mirror_set *ms, sector_t sector)
{
	/* FIXME: add read balancing */
	return ms->mirror + DEFAULT_MIRROR;
}

/*
 * remap a buffer to a particular mirror.
 */
static void map_bio(struct mirror_set *ms, struct mirror *m, struct bio *bio)
{
	bio->bi_bdev = m->dev->bdev;
	bio->bi_sector = m->offset + (bio->bi_sector - ms->ti->begin);
}

static void do_reads(struct mirror_set *ms, struct bio_list *reads)
{
	region_t region;
	struct bio *bio;
	struct mirror *m;

	while ((bio = bio_list_pop(reads))) {
		region = bio_to_region(&ms->rh, bio);

		/*
		 * We can only read balance if the region is in sync.
		 */
		if (rh_in_sync(&ms->rh, region, 0))
			m = choose_mirror(ms, bio->bi_sector);
		else
			m = ms->mirror + DEFAULT_MIRROR;

		map_bio(ms, m, bio);
		generic_make_request(bio);
	}
}

/*-----------------------------------------------------------------
 * Writes.
 *
 * We do different things with the write io depending on the
 * state of the region that it's in:
 *
 * SYNC: 	increment pending, use kcopyd to write to *all* mirrors
 * RECOVERING:	delay the io until recovery completes
 * NOSYNC:	increment pending, just write to the default mirror
 *---------------------------------------------------------------*/
static void write_callback(unsigned long error, void *context)
{
	unsigned int i;
	int uptodate = 1;
	struct bio *bio = (struct bio *) context;
	struct mirror_set *ms;

	ms = bio_get_ms(bio);
	bio_set_ms(bio, NULL);

	/*
	 * NOTE: We don't decrement the pending count here,
	 * instead it is done by the targets endio function.
	 * This way we handle both writes to SYNC and NOSYNC
	 * regions with the same code.
	 */

	if (error) {
		/*
		 * only error the io if all mirrors failed.
		 * FIXME: bogus
		 */
		uptodate = 0;
		for (i = 0; i < ms->nr_mirrors; i++)
			if (!test_bit(i, &error)) {
				uptodate = 1;
				break;
			}
	}
	bio_endio(bio, bio->bi_size, 0);
}

static void do_write(struct mirror_set *ms, struct bio *bio)
{
	unsigned int i;
	struct io_region io[KCOPYD_MAX_REGIONS+1];
	struct mirror *m;

	for (i = 0; i < ms->nr_mirrors; i++) {
		m = ms->mirror + i;

		io[i].bdev = m->dev->bdev;
		io[i].sector = m->offset + (bio->bi_sector - ms->ti->begin);
		io[i].count = bio->bi_size >> 9;
	}

	bio_set_ms(bio, ms);
	dm_io_async_bvec(ms->nr_mirrors, io, WRITE,
			 bio->bi_io_vec + bio->bi_idx,
			 write_callback, bio);
}

static void do_writes(struct mirror_set *ms, struct bio_list *writes)
{
	int state;
	struct bio *bio;
	struct bio_list sync, nosync, recover, *this_list = NULL;

	if (!writes->head)
		return;

	/*
	 * Classify each write.
	 */
	bio_list_init(&sync);
	bio_list_init(&nosync);
	bio_list_init(&recover);

	while ((bio = bio_list_pop(writes))) {
		state = rh_state(&ms->rh, bio_to_region(&ms->rh, bio), 1);
		switch (state) {
		case RH_CLEAN:
		case RH_DIRTY:
			this_list = &sync;
			break;

		case RH_NOSYNC:
			this_list = &nosync;
			break;

		case RH_RECOVERING:
			this_list = &recover;
			break;
		}

		bio_list_add(this_list, bio);
	}

	/*
	 * Increment the pending counts for any regions that will
	 * be written to (writes to recover regions are going to
	 * be delayed).
	 */
	rh_inc_pending(&ms->rh, &sync);
	rh_inc_pending(&ms->rh, &nosync);
	rh_flush(&ms->rh);

	/*
	 * Dispatch io.
	 */
	while ((bio = bio_list_pop(&sync)))
		do_write(ms, bio);

	while ((bio = bio_list_pop(&recover)))
		rh_delay(&ms->rh, bio);

	while ((bio = bio_list_pop(&nosync))) {
		map_bio(ms, ms->mirror + DEFAULT_MIRROR, bio);
		generic_make_request(bio);
	}
}

/*-----------------------------------------------------------------
 * kmirrord
 *---------------------------------------------------------------*/
static LIST_HEAD(_mirror_sets);
static DECLARE_RWSEM(_mirror_sets_lock);

static void do_mirror(struct mirror_set *ms)
{
	struct bio_list reads, writes;

	spin_lock(&ms->lock);
	reads = ms->reads;
	writes = ms->writes;
	bio_list_init(&ms->reads);
	bio_list_init(&ms->writes);
	spin_unlock(&ms->lock);

	rh_update_states(&ms->rh);
	do_recovery(ms);
	do_reads(ms, &reads);
	do_writes(ms, &writes);
}

static void do_work(void *ignored)
{
	struct mirror_set *ms;

	down_read(&_mirror_sets_lock);
	list_for_each_entry (ms, &_mirror_sets, list)
		do_mirror(ms);
	up_read(&_mirror_sets_lock);
}

/*-----------------------------------------------------------------
 * Target functions
 *---------------------------------------------------------------*/
static struct mirror_set *alloc_context(unsigned int nr_mirrors,
					sector_t region_size,
					struct dm_target *ti,
					struct dirty_log *dl)
{
	size_t len;
	struct mirror_set *ms = NULL;

	if (array_too_big(sizeof(*ms), sizeof(ms->mirror[0]), nr_mirrors))
		return NULL;

	len = sizeof(*ms) + (sizeof(ms->mirror[0]) * nr_mirrors);

	ms = kmalloc(len, GFP_KERNEL);
	if (!ms) {
		ti->error = "dm-mirror: Cannot allocate mirror context";
		return NULL;
	}

	memset(ms, 0, len);
	spin_lock_init(&ms->lock);

	ms->ti = ti;
	ms->nr_mirrors = nr_mirrors;
	ms->nr_regions = dm_div_up(ti->len, region_size);
	ms->in_sync = 0;

	if (rh_init(&ms->rh, ms, dl, region_size, ms->nr_regions)) {
		ti->error = "dm-mirror: Error creating dirty region hash";
		kfree(ms);
		return NULL;
	}

	return ms;
}

static void free_context(struct mirror_set *ms, struct dm_target *ti,
			 unsigned int m)
{
	while (m--)
		dm_put_device(ti, ms->mirror[m].dev);

	rh_exit(&ms->rh);
	kfree(ms);
}

static inline int _check_region_size(struct dm_target *ti, sector_t size)
{
	return !(size % (PAGE_SIZE >> 9) || (size & (size - 1)) ||
		 size > ti->len);
}

static int get_mirror(struct mirror_set *ms, struct dm_target *ti,
		      unsigned int mirror, char **argv)
{
	sector_t offset;

	if (sscanf(argv[1], SECTOR_FORMAT, &offset) != 1) {
		ti->error = "dm-mirror: Invalid offset";
		return -EINVAL;
	}

	if (dm_get_device(ti, argv[0], offset, ti->len,
			  dm_table_get_mode(ti->table),
			  &ms->mirror[mirror].dev)) {
		ti->error = "dm-mirror: Device lookup failure";
		return -ENXIO;
	}

	ms->mirror[mirror].offset = offset;

	return 0;
}

static int add_mirror_set(struct mirror_set *ms)
{
	down_write(&_mirror_sets_lock);
	list_add_tail(&ms->list, &_mirror_sets);
	up_write(&_mirror_sets_lock);
	wake();

	return 0;
}

static void del_mirror_set(struct mirror_set *ms)
{
	down_write(&_mirror_sets_lock);
	list_del(&ms->list);
	up_write(&_mirror_sets_lock);
}

/*
 * Create dirty log: log_type #log_params <log_params>
 */
static struct dirty_log *create_dirty_log(struct dm_target *ti,
					  unsigned int argc, char **argv,
					  unsigned int *args_used)
{
	unsigned int param_count;
	struct dirty_log *dl;

	if (argc < 2) {
		ti->error = "dm-mirror: Insufficient mirror log arguments";
		return NULL;
	}

	if (sscanf(argv[1], "%u", &param_count) != 1) {
		ti->error = "dm-mirror: Invalid mirror log argument count";
		return NULL;
	}

	*args_used = 2 + param_count;

	if (argc < *args_used) {
		ti->error = "dm-mirror: Insufficient mirror log arguments";
		return NULL;
	}

	dl = dm_create_dirty_log(argv[0], ti, param_count, argv + 2);
	if (!dl) {
		ti->error = "dm-mirror: Error creating mirror dirty log";
		return NULL;
	}

	if (!_check_region_size(ti, dl->type->get_region_size(dl))) {
		ti->error = "dm-mirror: Invalid region size";
		dm_destroy_dirty_log(dl);
		return NULL;
	}

	return dl;
}

/*
 * Construct a mirror mapping:
 *
 * log_type #log_params <log_params>
 * #mirrors [mirror_path offset]{2,}
 *
 * log_type is "core" or "disk"
 * #log_params is between 1 and 3
 */
#define DM_IO_PAGES 64
static int mirror_ctr(struct dm_target *ti, unsigned int argc, char **argv)
{
	int r;
	unsigned int nr_mirrors, m, args_used;
	struct mirror_set *ms;
	struct dirty_log *dl;

	dl = create_dirty_log(ti, argc, argv, &args_used);
	if (!dl)
		return -EINVAL;

	argv += args_used;
	argc -= args_used;

	if (!argc || sscanf(argv[0], "%u", &nr_mirrors) != 1 ||
	    nr_mirrors < 2 || nr_mirrors > KCOPYD_MAX_REGIONS + 1) {
		ti->error = "dm-mirror: Invalid number of mirrors";
		dm_destroy_dirty_log(dl);
		return -EINVAL;
	}

	argv++, argc--;

	if (argc != nr_mirrors * 2) {
		ti->error = "dm-mirror: Wrong number of mirror arguments";
		dm_destroy_dirty_log(dl);
		return -EINVAL;
	}

	ms = alloc_context(nr_mirrors, dl->type->get_region_size(dl), ti, dl);
	if (!ms) {
		dm_destroy_dirty_log(dl);
		return -ENOMEM;
	}

	/* Get the mirror parameter sets */
	for (m = 0; m < nr_mirrors; m++) {
		r = get_mirror(ms, ti, m, argv);
		if (r) {
			free_context(ms, ti, m);
			return r;
		}
		argv += 2;
		argc -= 2;
	}

	ti->private = ms;

	r = kcopyd_client_create(DM_IO_PAGES, &ms->kcopyd_client);
	if (r) {
		free_context(ms, ti, ms->nr_mirrors);
		return r;
	}

	add_mirror_set(ms);
	return 0;
}

static void mirror_dtr(struct dm_target *ti)
{
	struct mirror_set *ms = (struct mirror_set *) ti->private;

	del_mirror_set(ms);
	kcopyd_client_destroy(ms->kcopyd_client);
	free_context(ms, ti, ms->nr_mirrors);
}

static void queue_bio(struct mirror_set *ms, struct bio *bio, int rw)
{
	int should_wake = 0;
	struct bio_list *bl;

	bl = (rw == WRITE) ? &ms->writes : &ms->reads;
	spin_lock(&ms->lock);
	should_wake = !(bl->head);
	bio_list_add(bl, bio);
	spin_unlock(&ms->lock);

	if (should_wake)
		wake();
}

/*
 * Mirror mapping function
 */
static int mirror_map(struct dm_target *ti, struct bio *bio,
		      union map_info *map_context)
{
	int r, rw = bio_rw(bio);
	struct mirror *m;
	struct mirror_set *ms = ti->private;

	map_context->ll = bio->bi_sector >> ms->rh.region_shift;

	if (rw == WRITE) {
		queue_bio(ms, bio, rw);
		return 0;
	}

	r = ms->rh.log->type->in_sync(ms->rh.log,
				      bio_to_region(&ms->rh, bio), 0);
	if (r < 0 && r != -EWOULDBLOCK)
		return r;

	if (r == -EWOULDBLOCK)	/* FIXME: ugly */
		r = 0;

	/*
	 * We don't want to fast track a recovery just for a read
	 * ahead.  So we just let it silently fail.
	 * FIXME: get rid of this.
	 */
	if (!r && rw == READA)
		return -EIO;

	if (!r) {
		/* Pass this io over to the daemon */
		queue_bio(ms, bio, rw);
		return 0;
	}

	m = choose_mirror(ms, bio->bi_sector);
	if (!m)
		return -EIO;

	map_bio(ms, m, bio);
	return 1;
}

static int mirror_end_io(struct dm_target *ti, struct bio *bio,
			 int error, union map_info *map_context)
{
	int rw = bio_rw(bio);
	struct mirror_set *ms = (struct mirror_set *) ti->private;
	region_t region = map_context->ll;

	/*
	 * We need to dec pending if this was a write.
	 */
	if (rw == WRITE)
		rh_dec(&ms->rh, region);

	return 0;
}

static void mirror_suspend(struct dm_target *ti)
{
	struct mirror_set *ms = (struct mirror_set *) ti->private;
	struct dirty_log *log = ms->rh.log;
	rh_stop_recovery(&ms->rh);
	if (log->type->suspend && log->type->suspend(log))
		/* FIXME: need better error handling */
		DMWARN("log suspend failed");
}

static void mirror_resume(struct dm_target *ti)
{
	struct mirror_set *ms = (struct mirror_set *) ti->private;
	struct dirty_log *log = ms->rh.log;
	if (log->type->resume && log->type->resume(log))
		/* FIXME: need better error handling */
		DMWARN("log resume failed");
	rh_start_recovery(&ms->rh);
}

static int mirror_status(struct dm_target *ti, status_type_t type,
			 char *result, unsigned int maxlen)
{
	char buffer[32];
	unsigned int m, sz;
	struct mirror_set *ms = (struct mirror_set *) ti->private;

	sz = ms->rh.log->type->status(ms->rh.log, type, result, maxlen);

	switch (type) {
	case STATUSTYPE_INFO:
		DMEMIT("%d ", ms->nr_mirrors);
		for (m = 0; m < ms->nr_mirrors; m++) {
			format_dev_t(buffer, ms->mirror[m].dev->bdev->bd_dev);
			DMEMIT("%s ", buffer);
		}

		DMEMIT(SECTOR_FORMAT "/" SECTOR_FORMAT,
		       ms->rh.log->type->get_sync_count(ms->rh.log),
		       ms->nr_regions);
		break;

	case STATUSTYPE_TABLE:
		DMEMIT("%d ", ms->nr_mirrors);
		for (m = 0; m < ms->nr_mirrors; m++) {
			format_dev_t(buffer, ms->mirror[m].dev->bdev->bd_dev);
			DMEMIT("%s " SECTOR_FORMAT " ",
			       buffer, ms->mirror[m].offset);
		}
	}

	return 0;
}

static struct target_type mirror_target = {
	.name	 = "mirror",
	.version = {1, 0, 1},
	.module	 = THIS_MODULE,
	.ctr	 = mirror_ctr,
	.dtr	 = mirror_dtr,
	.map	 = mirror_map,
	.end_io	 = mirror_end_io,
	.suspend = mirror_suspend,
	.resume	 = mirror_resume,
	.status	 = mirror_status,
};

static int __init dm_mirror_init(void)
{
	int r;

	r = dm_dirty_log_init();
	if (r)
		return r;

	_kmirrord_wq = create_workqueue("kmirrord");
	if (!_kmirrord_wq) {
		DMERR("couldn't start kmirrord");
		dm_dirty_log_exit();
		return r;
	}
	INIT_WORK(&_kmirrord_work, do_work, NULL);

	r = dm_register_target(&mirror_target);
	if (r < 0) {
		DMERR("%s: Failed to register mirror target",
		      mirror_target.name);
		dm_dirty_log_exit();
		destroy_workqueue(_kmirrord_wq);
	}

	return r;
}

static void __exit dm_mirror_exit(void)
{
	int r;

	r = dm_unregister_target(&mirror_target);
	if (r < 0)
		DMERR("%s: unregister failed %d", mirror_target.name, r);

	destroy_workqueue(_kmirrord_wq);
	dm_dirty_log_exit();
}

/* Module hooks */
module_init(dm_mirror_init);
module_exit(dm_mirror_exit);

MODULE_DESCRIPTION(DM_NAME " mirror target");
MODULE_AUTHOR("Joe Thornber");
MODULE_LICENSE("GPL");