dm-raid1.c

linux 内核源代码

 * Target functions
 *---------------------------------------------------------------*/
static struct mirror_set *alloc_context(unsigned int nr_mirrors,
					uint32_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 = kzalloc(len, GFP_KERNEL);
	if (!ms) {
		ti->error = "Cannot allocate mirror context";
		return NULL;
	}

	spin_lock_init(&ms->lock);

	ms->ti = ti;
	ms->nr_mirrors = nr_mirrors;
	ms->nr_regions = dm_sector_div_up(ti->len, region_size);
	ms->in_sync = 0;
	ms->default_mirror = &ms->mirror[DEFAULT_MIRROR];

	ms->io_client = dm_io_client_create(DM_IO_PAGES);
	if (IS_ERR(ms->io_client)) {
		ti->error = "Error creating dm_io client";
		kfree(ms);
 		return NULL;
	}

	if (rh_init(&ms->rh, ms, dl, region_size, ms->nr_regions)) {
		ti->error = "Error creating dirty region hash";
		dm_io_client_destroy(ms->io_client);
		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);

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

static inline int _check_region_size(struct dm_target *ti, uint32_t size)
{
	return !(size % (PAGE_SIZE >> 9) || !is_power_of_2(size) ||
		 size > ti->len);
}

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

	if (sscanf(argv[1], "%llu", &offset) != 1) {
		ti->error = "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 = "Device lookup failure";
		return -ENXIO;
	}

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

	return 0;
}

/*
 * 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 = "Insufficient mirror log arguments";
		return NULL;
	}

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

	*args_used = 2 + param_count;

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

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

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

	return dl;
}

static int parse_features(struct mirror_set *ms, unsigned argc, char **argv,
			  unsigned *args_used)
{
	unsigned num_features;
	struct dm_target *ti = ms->ti;

	*args_used = 0;

	if (!argc)
		return 0;

	if (sscanf(argv[0], "%u", &num_features) != 1) {
		ti->error = "Invalid number of features";
		return -EINVAL;
	}

	argc--;
	argv++;
	(*args_used)++;

	if (num_features > argc) {
		ti->error = "Not enough arguments to support feature count";
		return -EINVAL;
	}

	if (!strcmp("handle_errors", argv[0]))
		ms->features |= DM_RAID1_HANDLE_ERRORS;
	else {
		ti->error = "Unrecognised feature requested";
		return -EINVAL;
	}

	(*args_used)++;

	return 0;
}

/*
 * Construct a mirror mapping:
 *
 * log_type #log_params <log_params>
 * #mirrors [mirror_path offset]{2,}
 * [#features <features>]
 *
 * log_type is "core" or "disk"
 * #log_params is between 1 and 3
 *
 * If present, features must be "handle_errors".
 */
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 = "Invalid number of mirrors";
		dm_destroy_dirty_log(dl);
		return -EINVAL;
	}

	argv++, argc--;

	if (argc < nr_mirrors * 2) {
		ti->error = "Too few 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;
 	ti->split_io = ms->rh.region_size;

	ms->kmirrord_wq = create_singlethread_workqueue("kmirrord");
	if (!ms->kmirrord_wq) {
		DMERR("couldn't start kmirrord");
		r = -ENOMEM;
		goto err_free_context;
	}
	INIT_WORK(&ms->kmirrord_work, do_mirror);

	r = parse_features(ms, argc, argv, &args_used);
	if (r)
		goto err_destroy_wq;

	argv += args_used;
	argc -= args_used;

	/*
	 * Any read-balancing addition depends on the
	 * DM_RAID1_HANDLE_ERRORS flag being present.
	 * This is because the decision to balance depends
	 * on the sync state of a region.  If the above
	 * flag is not present, we ignore errors; and
	 * the sync state may be inaccurate.
	 */

	if (argc) {
		ti->error = "Too many mirror arguments";
		r = -EINVAL;
		goto err_destroy_wq;
	}

	r = kcopyd_client_create(DM_IO_PAGES, &ms->kcopyd_client);
	if (r)
		goto err_destroy_wq;

	wake(ms);
	return 0;

err_destroy_wq:
	destroy_workqueue(ms->kmirrord_wq);
err_free_context:
	free_context(ms, ti, ms->nr_mirrors);
	return r;
}

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

	flush_workqueue(ms->kmirrord_wq);
	kcopyd_client_destroy(ms->kcopyd_client);
	destroy_workqueue(ms->kmirrord_wq);
	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(ms);
}

/*
 * 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_to_region(&ms->rh, bio);

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

	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 = DM_MAPIO_SUBMITTED;

	/*
	 * 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 DM_MAPIO_SUBMITTED;
	}

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

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

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_postsuspend(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);

	/* Wait for all I/O we generated to complete */
	wait_event(_kmirrord_recovery_stopped,
		   !atomic_read(&ms->rh.recovery_in_flight));

	if (log->type->postsuspend && log->type->postsuspend(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)
{
	unsigned int m, sz = 0;
	struct mirror_set *ms = (struct mirror_set *) ti->private;

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

		DMEMIT("%llu/%llu 0 ",
			(unsigned long long)ms->rh.log->type->
				get_sync_count(ms->rh.log),
			(unsigned long long)ms->nr_regions);

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

		break;

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

		DMEMIT("%d", ms->nr_mirrors);
		for (m = 0; m < ms->nr_mirrors; m++)
			DMEMIT(" %s %llu", ms->mirror[m].dev->name,
				(unsigned long long)ms->mirror[m].offset);

		if (ms->features & DM_RAID1_HANDLE_ERRORS)
			DMEMIT(" 1 handle_errors");
	}

	return 0;
}

static struct target_type mirror_target = {
	.name	 = "mirror",
	.version = {1, 0, 3},
	.module	 = THIS_MODULE,
	.ctr	 = mirror_ctr,
	.dtr	 = mirror_dtr,
	.map	 = mirror_map,
	.end_io	 = mirror_end_io,
	.postsuspend = mirror_postsuspend,
	.resume	 = mirror_resume,
	.status	 = mirror_status,
};

static int __init dm_mirror_init(void)
{
	int r;

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

	r = dm_register_target(&mirror_target);
	if (r < 0) {
		DMERR("Failed to register mirror target");
		dm_dirty_log_exit();
	}

	return r;
}

static void __exit dm_mirror_exit(void)
{
	int r;

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

	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");