bcache in userspace; userspace fsck

1 files changed, 2503 insertions, 0 deletions

diff --git a/libbcache/super.c b/libbcache/super.c
new file mode 100644
index 0000000..5f6a85e
--- /dev/null
+++ b/libbcache/super.c
@@ -0,0 +1,2503 @@
+/*
+ * bcache setup/teardown code, and some metadata io - read a superblock and
+ * figure out what to do with it.
+ *
+ * Copyright 2010, 2011 Kent Overstreet <kent.overstreet@gmail.com>
+ * Copyright 2012 Google, Inc.
+ */
+
+#include "bcache.h"
+#include "blockdev.h"
+#include "alloc.h"
+#include "btree_cache.h"
+#include "btree_gc.h"
+#include "btree_update.h"
+#include "btree_io.h"
+#include "chardev.h"
+#include "checksum.h"
+#include "clock.h"
+#include "compress.h"
+#include "debug.h"
+#include "error.h"
+#include "fs.h"
+#include "fs-gc.h"
+#include "inode.h"
+#include "io.h"
+#include "journal.h"
+#include "keylist.h"
+#include "move.h"
+#include "migrate.h"
+#include "movinggc.h"
+#include "notify.h"
+#include "stats.h"
+#include "super.h"
+#include "tier.h"
+#include "writeback.h"
+
+#include <linux/backing-dev.h>
+#include <linux/blkdev.h>
+#include <linux/debugfs.h>
+#include <linux/genhd.h>
+#include <linux/idr.h>
+#include <linux/kthread.h>
+#include <linux/module.h>
+#include <linux/percpu.h>
+#include <linux/random.h>
+#include <linux/reboot.h>
+#include <linux/sysfs.h>
+#include <crypto/hash.h>
+
+#include <trace/events/bcache.h>
+
+MODULE_LICENSE("GPL");
+MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
+
+static const uuid_le invalid_uuid = {
+	.b = {
+		0xa0, 0x3e, 0xf8, 0xed, 0x3e, 0xe1, 0xb8, 0x78,
+		0xc8, 0x50, 0xfc, 0x5e, 0xcb, 0x16, 0xcd, 0x99
+	}
+};
+
+static struct kset *bcache_kset;
+struct mutex bch_register_lock;
+LIST_HEAD(bch_cache_sets);
+
+static int bch_chardev_major;
+static struct class *bch_chardev_class;
+static struct device *bch_chardev;
+static DEFINE_IDR(bch_chardev_minor);
+static DECLARE_WAIT_QUEUE_HEAD(bch_read_only_wait);
+struct workqueue_struct *bcache_io_wq;
+struct crypto_shash *bch_sha1;
+
+static void bch_cache_stop(struct cache *);
+static int bch_cache_online(struct cache *);
+
+static bool bch_is_open_cache(struct block_device *bdev)
+{
+	struct cache_set *c;
+	struct cache *ca;
+	unsigned i;
+
+	rcu_read_lock();
+	list_for_each_entry(c, &bch_cache_sets, list)
+		for_each_cache_rcu(ca, c, i)
+			if (ca->disk_sb.bdev == bdev) {
+				rcu_read_unlock();
+				return true;
+			}
+	rcu_read_unlock();
+	return false;
+}
+
+static bool bch_is_open(struct block_device *bdev)
+{
+	lockdep_assert_held(&bch_register_lock);
+
+	return bch_is_open_cache(bdev) || bch_is_open_backing_dev(bdev);
+}
+
+static const char *bch_blkdev_open(const char *path, void *holder,
+				   struct block_device **ret)
+{
+	struct block_device *bdev;
+	const char *err;
+
+	*ret = NULL;
+	bdev = blkdev_get_by_path(path, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
+				  holder);
+
+	if (bdev == ERR_PTR(-EBUSY)) {
+		bdev = lookup_bdev(path);
+		if (IS_ERR(bdev))
+			return "device busy";
+
+		err = bch_is_open(bdev)
+			? "device already registered"
+			: "device busy";
+
+		bdput(bdev);
+		return err;
+	}
+
+	if (IS_ERR(bdev))
+		return "failed to open device";
+
+	bdev_get_queue(bdev)->backing_dev_info.capabilities |= BDI_CAP_STABLE_WRITES;
+
+	*ret = bdev;
+	return NULL;
+}
+
+static int bch_congested_fn(void *data, int bdi_bits)
+{
+	struct backing_dev_info *bdi;
+	struct cache_set *c = data;
+	struct cache *ca;
+	unsigned i;
+	int ret = 0;
+
+	rcu_read_lock();
+	if (bdi_bits & (1 << WB_sync_congested)) {
+		/* Reads - check all devices: */
+		for_each_cache_rcu(ca, c, i) {
+			bdi = blk_get_backing_dev_info(ca->disk_sb.bdev);
+
+			if (bdi_congested(bdi, bdi_bits)) {
+				ret = 1;
+				break;
+			}
+		}
+	} else {
+		/* Writes only go to tier 0: */
+		group_for_each_cache_rcu(ca, &c->cache_tiers[0], i) {
+			bdi = blk_get_backing_dev_info(ca->disk_sb.bdev);
+
+			if (bdi_congested(bdi, bdi_bits)) {
+				ret = 1;
+				break;
+			}
+		}
+	}
+	rcu_read_unlock();
+
+	return ret;
+}
+
+/* Superblock */
+
+static struct cache_member_cpu cache_mi_to_cpu_mi(struct cache_member *mi)
+{
+	return (struct cache_member_cpu) {
+		.nbuckets	= le64_to_cpu(mi->nbuckets),
+		.first_bucket	= le16_to_cpu(mi->first_bucket),
+		.bucket_size	= le16_to_cpu(mi->bucket_size),
+		.state		= CACHE_STATE(mi),
+		.tier		= CACHE_TIER(mi),
+		.replication_set= CACHE_REPLICATION_SET(mi),
+		.has_metadata	= CACHE_HAS_METADATA(mi),
+		.has_data	= CACHE_HAS_DATA(mi),
+		.replacement	= CACHE_REPLACEMENT(mi),
+		.discard	= CACHE_DISCARD(mi),
+		.valid		= !bch_is_zero(mi->uuid.b, sizeof(uuid_le)),
+	};
+}
+
+static const char *validate_cache_super(struct bcache_superblock *disk_sb)
+{
+	struct cache_sb *sb = disk_sb->sb;
+	struct cache_member_cpu	mi;
+	u16 block_size;
+	unsigned i;
+
+	switch (le64_to_cpu(sb->version)) {
+	case BCACHE_SB_VERSION_CDEV_V0:
+	case BCACHE_SB_VERSION_CDEV_WITH_UUID:
+	case BCACHE_SB_VERSION_CDEV_V2:
+	case BCACHE_SB_VERSION_CDEV_V3:
+		break;
+	default:
+		return"Unsupported superblock version";
+	}
+
+	if (CACHE_SET_SYNC(sb) &&
+	    le64_to_cpu(sb->version) != BCACHE_SB_VERSION_CDEV_V3)
+		return "Unsupported superblock version";
+
+	block_size = le16_to_cpu(sb->block_size);
+
+	if (!is_power_of_2(block_size) ||
+	    block_size > PAGE_SECTORS)
+		return "Bad block size";
+
+	if (bch_is_zero(sb->disk_uuid.b, sizeof(uuid_le)))
+		return "Bad disk UUID";
+
+	if (bch_is_zero(sb->user_uuid.b, sizeof(uuid_le)))
+		return "Bad user UUID";
+
+	if (bch_is_zero(sb->set_uuid.b, sizeof(uuid_le)))
+		return "Bad set UUID";
+
+	if (!sb->nr_in_set ||
+	    sb->nr_in_set <= sb->nr_this_dev ||
+	    sb->nr_in_set > MAX_CACHES_PER_SET)
+		return "Bad cache device number in set";
+
+	if (!CACHE_SET_META_REPLICAS_WANT(sb) ||
+	    CACHE_SET_META_REPLICAS_WANT(sb) >= BCH_REPLICAS_MAX)
+		return "Invalid number of metadata replicas";
+
+	if (!CACHE_SET_META_REPLICAS_HAVE(sb) ||
+	    CACHE_SET_META_REPLICAS_HAVE(sb) >
+	    CACHE_SET_META_REPLICAS_WANT(sb))
+		return "Invalid number of metadata replicas";
+
+	if (!CACHE_SET_DATA_REPLICAS_WANT(sb) ||
+	    CACHE_SET_DATA_REPLICAS_WANT(sb) >= BCH_REPLICAS_MAX)
+		return "Invalid number of data replicas";
+
+	if (!CACHE_SET_DATA_REPLICAS_HAVE(sb) ||
+	    CACHE_SET_DATA_REPLICAS_HAVE(sb) >
+	    CACHE_SET_DATA_REPLICAS_WANT(sb))
+		return "Invalid number of data replicas";
+
+	if (CACHE_SB_CSUM_TYPE(sb) >= BCH_CSUM_NR)
+		return "Invalid checksum type";
+
+	if (!CACHE_SET_BTREE_NODE_SIZE(sb))
+		return "Btree node size not set";
+
+	if (!is_power_of_2(CACHE_SET_BTREE_NODE_SIZE(sb)))
+		return "Btree node size not a power of two";
+
+	if (CACHE_SET_BTREE_NODE_SIZE(sb) > BTREE_NODE_SIZE_MAX)
+		return "Btree node size too large";
+
+	/* Default value, for old filesystems: */
+	if (!CACHE_SET_GC_RESERVE(sb))
+		SET_CACHE_SET_GC_RESERVE(sb, 10);
+
+	if (CACHE_SET_GC_RESERVE(sb) < 5)
+		return "gc reserve percentage too small";
+
+	if (!CACHE_SET_JOURNAL_ENTRY_SIZE(sb))
+		SET_CACHE_SET_JOURNAL_ENTRY_SIZE(sb, 9);
+
+	/* 4 mb max: */
+	if (512U << CACHE_SET_JOURNAL_ENTRY_SIZE(sb) > JOURNAL_ENTRY_SIZE_MAX)
+		return "max journal entry size too big";
+
+	if (le16_to_cpu(sb->u64s) < bch_journal_buckets_offset(sb))
+		return "Invalid superblock: member info area missing";
+
+	mi = cache_mi_to_cpu_mi(sb->members + sb->nr_this_dev);
+
+	if (mi.nbuckets > LONG_MAX)
+		return "Too many buckets";
+
+	if (mi.nbuckets < 1 << 8)
+		return "Not enough buckets";
+
+	if (!is_power_of_2(mi.bucket_size) ||
+	    mi.bucket_size < PAGE_SECTORS ||
+	    mi.bucket_size < block_size)
+		return "Bad bucket size";
+
+	if (get_capacity(disk_sb->bdev->bd_disk) <
+	    mi.bucket_size * mi.nbuckets)
+		return "Invalid superblock: device too small";
+
+	if (le64_to_cpu(sb->offset) +
+	    (__set_blocks(sb, le16_to_cpu(sb->u64s),
+			  block_size << 9) * block_size) >
+	    mi.first_bucket * mi.bucket_size)
+		return "Invalid superblock: first bucket comes before end of super";
+
+	for (i = 0; i < bch_nr_journal_buckets(sb); i++)
+		if (journal_bucket(sb, i) <  mi.first_bucket ||
+		    journal_bucket(sb, i) >= mi.nbuckets)
+			return "bad journal bucket";
+
+	return NULL;
+}
+
+void free_super(struct bcache_superblock *sb)
+{
+	if (sb->bio)
+		bio_put(sb->bio);
+	if (!IS_ERR_OR_NULL(sb->bdev))
+		blkdev_put(sb->bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
+
+	free_pages((unsigned long) sb->sb, sb->page_order);
+	memset(sb, 0, sizeof(*sb));
+}
+
+static int __bch_super_realloc(struct bcache_superblock *sb, unsigned order)
+{
+	struct cache_sb *new_sb;
+	struct bio *bio;
+
+	if (sb->page_order >= order && sb->sb)
+		return 0;
+
+	new_sb = (void *) __get_free_pages(GFP_KERNEL, order);
+	if (!new_sb)
+		return -ENOMEM;
+
+	bio = (dynamic_fault("bcache:add:super_realloc")
+	       ? NULL
+	       : bio_kmalloc(GFP_KERNEL, 1 << order));
+	if (!bio) {
+		free_pages((unsigned long) new_sb, order);
+		return -ENOMEM;
+	}
+
+	if (sb->sb)
+		memcpy(new_sb, sb->sb, PAGE_SIZE << sb->page_order);
+
+	free_pages((unsigned long) sb->sb, sb->page_order);
+	sb->sb = new_sb;
+
+	if (sb->bio)
+		bio_put(sb->bio);
+	sb->bio = bio;
+
+	sb->page_order = order;
+
+	return 0;
+}
+
+int bch_super_realloc(struct bcache_superblock *sb, unsigned u64s)
+{
+	struct cache_member *mi = sb->sb->members + sb->sb->nr_this_dev;
+	char buf[BDEVNAME_SIZE];
+	size_t bytes = __set_bytes((struct cache_sb *) NULL, u64s);
+	u64 want = bytes + (SB_SECTOR << 9);
+
+	u64 first_bucket_offset = (u64) le16_to_cpu(mi->first_bucket) *
+		((u64) le16_to_cpu(mi->bucket_size) << 9);
+
+	if (want > first_bucket_offset) {
+		pr_err("%s: superblock too big: want %llu but have %llu",
+		       bdevname(sb->bdev, buf), want, first_bucket_offset);
+		return -ENOSPC;
+	}
+
+	return __bch_super_realloc(sb, get_order(bytes));
+}
+
+static const char *read_super(struct bcache_superblock *sb,
+			      const char *path)
+{
+	const char *err;
+	unsigned order = 0;
+
+	lockdep_assert_held(&bch_register_lock);
+
+	memset(sb, 0, sizeof(*sb));
+
+	err = bch_blkdev_open(path, &sb, &sb->bdev);
+	if (err)
+		return err;
+retry:
+	err = "cannot allocate memory";
+	if (__bch_super_realloc(sb, order))
+		goto err;
+
+	err = "dynamic fault";
+	if (cache_set_init_fault("read_super"))
+		goto err;
+
+	bio_reset(sb->bio);
+	sb->bio->bi_bdev = sb->bdev;
+	sb->bio->bi_iter.bi_sector = SB_SECTOR;
+	sb->bio->bi_iter.bi_size = PAGE_SIZE << sb->page_order;
+	bio_set_op_attrs(sb->bio, REQ_OP_READ, REQ_SYNC|REQ_META);
+	bch_bio_map(sb->bio, sb->sb);
+
+	err = "IO error";
+	if (submit_bio_wait(sb->bio))
+		goto err;
+
+	err = "Not a bcache superblock";
+	if (uuid_le_cmp(sb->sb->magic, BCACHE_MAGIC))
+		goto err;
+
+	err = "Superblock has incorrect offset";
+	if (le64_to_cpu(sb->sb->offset) != SB_SECTOR)
+		goto err;
+
+	pr_debug("read sb version %llu, flags %llu, seq %llu, journal size %u",
+		 le64_to_cpu(sb->sb->version),
+		 le64_to_cpu(sb->sb->flags),
+		 le64_to_cpu(sb->sb->seq),
+		 le16_to_cpu(sb->sb->u64s));
+
+	err = "Superblock block size smaller than device block size";
+	if (le16_to_cpu(sb->sb->block_size) << 9 <
+	    bdev_logical_block_size(sb->bdev))
+		goto err;
+
+	order = get_order(__set_bytes(sb->sb, le16_to_cpu(sb->sb->u64s)));
+	if (order > sb->page_order)
+		goto retry;
+
+	err = "bad checksum reading superblock";
+	if (le64_to_cpu(sb->sb->csum) !=
+	    __csum_set(sb->sb, le16_to_cpu(sb->sb->u64s),
+		       le64_to_cpu(sb->sb->version) <
+		       BCACHE_SB_VERSION_CDEV_V3
+		       ? BCH_CSUM_CRC64
+		       : CACHE_SB_CSUM_TYPE(sb->sb)))
+		goto err;
+
+	return NULL;
+err:
+	free_super(sb);
+	return err;
+}
+
+void __write_super(struct cache_set *c, struct bcache_superblock *disk_sb)
+{
+	struct cache_sb *sb = disk_sb->sb;
+	struct bio *bio = disk_sb->bio;
+
+	bio->bi_bdev		= disk_sb->bdev;
+	bio->bi_iter.bi_sector	= SB_SECTOR;
+	bio->bi_iter.bi_size	=
+		roundup(__set_bytes(sb, le16_to_cpu(sb->u64s)),
+			bdev_logical_block_size(disk_sb->bdev));
+	bio_set_op_attrs(bio, REQ_OP_WRITE, REQ_SYNC|REQ_META);
+	bch_bio_map(bio, sb);
+
+	pr_debug("ver %llu, flags %llu, seq %llu",
+		 le64_to_cpu(sb->version),
+		 le64_to_cpu(sb->flags),
+		 le64_to_cpu(sb->seq));
+
+	bch_generic_make_request(bio, c);
+}
+
+static void write_super_endio(struct bio *bio)
+{
+	struct cache *ca = bio->bi_private;
+
+	/* XXX: return errors directly */
+
+	cache_fatal_io_err_on(bio->bi_error, ca, "superblock write");
+
+	bch_account_io_completion(ca);
+
+	closure_put(&ca->set->sb_write);
+	percpu_ref_put(&ca->ref);
+}
+
+static void bcache_write_super_unlock(struct closure *cl)
+{
+	struct cache_set *c = container_of(cl, struct cache_set, sb_write);
+
+	up(&c->sb_write_mutex);
+}
+
+/* Update cached mi: */
+static int cache_set_mi_update(struct cache_set *c,
+			       struct cache_member *mi,
+			       unsigned nr_in_set)
+{
+	struct cache_member_rcu *new, *old;
+	struct cache *ca;
+	unsigned i;
+
+	mutex_lock(&c->mi_lock);
+
+	new = kzalloc(sizeof(struct cache_member_rcu) +
+		      sizeof(struct cache_member_cpu) * nr_in_set,
+		      GFP_KERNEL);
+	if (!new) {
+		mutex_unlock(&c->mi_lock);
+		return -ENOMEM;
+	}
+
+	new->nr_in_set = nr_in_set;
+
+	for (i = 0; i < nr_in_set; i++)
+		new->m[i] = cache_mi_to_cpu_mi(&mi[i]);
+
+	rcu_read_lock();
+	for_each_cache(ca, c, i)
+		ca->mi = new->m[i];
+	rcu_read_unlock();
+
+	old = rcu_dereference_protected(c->members,
+				lockdep_is_held(&c->mi_lock));
+
+	rcu_assign_pointer(c->members, new);
+	if (old)
+		kfree_rcu(old, rcu);
+
+	mutex_unlock(&c->mi_lock);
+	return 0;
+}
+
+/* doesn't copy member info */
+static void __copy_super(struct cache_sb *dst, struct cache_sb *src)
+{
+	dst->version		= src->version;
+	dst->seq		= src->seq;
+	dst->user_uuid		= src->user_uuid;
+	dst->set_uuid		= src->set_uuid;
+	memcpy(dst->label, src->label, SB_LABEL_SIZE);
+	dst->flags		= src->flags;
+	dst->flags2		= src->flags2;
+	dst->nr_in_set		= src->nr_in_set;
+	dst->block_size		= src->block_size;
+}
+
+static int cache_sb_to_cache_set(struct cache_set *c, struct cache_sb *src)
+{
+	struct cache_member *new;
+
+	lockdep_assert_held(&bch_register_lock);
+
+	new = kzalloc(sizeof(struct cache_member) * src->nr_in_set,
+		      GFP_KERNEL);
+	if (!new)
+		return -ENOMEM;
+
+	memcpy(new, src->members,
+	       src->nr_in_set * sizeof(struct cache_member));
+
+	if (cache_set_mi_update(c, new, src->nr_in_set)) {
+		kfree(new);
+		return -ENOMEM;
+	}
+
+	kfree(c->disk_mi);
+	c->disk_mi = new;
+
+	__copy_super(&c->disk_sb, src);
+
+	c->sb.block_size	= le16_to_cpu(src->block_size);
+	c->sb.btree_node_size	= CACHE_SET_BTREE_NODE_SIZE(src);
+	c->sb.nr_in_set		= src->nr_in_set;
+	c->sb.clean		= CACHE_SET_CLEAN(src);
+	c->sb.meta_replicas_have= CACHE_SET_META_REPLICAS_HAVE(src);
+	c->sb.data_replicas_have= CACHE_SET_DATA_REPLICAS_HAVE(src);
+	c->sb.str_hash_type	= CACHE_SET_STR_HASH_TYPE(src);
+
+	return 0;
+}
+
+static int cache_sb_from_cache_set(struct cache_set *c, struct cache *ca)
+{
+	struct cache_sb *src = &c->disk_sb, *dst = ca->disk_sb.sb;
+
+	if (src->nr_in_set != dst->nr_in_set) {
+		/*
+		 * We have to preserve the list of journal buckets on the
+		 * cache's superblock:
+		 */
+		unsigned old_offset = bch_journal_buckets_offset(dst);
+		unsigned u64s = bch_journal_buckets_offset(src)
+			+ bch_nr_journal_buckets(dst);
+		int ret = bch_super_realloc(&ca->disk_sb, u64s);
+
+		if (ret)
+			return ret;
+
+		dst->nr_in_set	= src->nr_in_set;
+		dst->u64s	= cpu_to_le16(u64s);
+
+		memmove(dst->_data + bch_journal_buckets_offset(dst),
+			dst->_data + old_offset,
+			bch_nr_journal_buckets(dst) * sizeof(u64));
+	}
+
+	memcpy(dst->_data,
+	       c->disk_mi,
+	       src->nr_in_set * sizeof(struct cache_member));
+
+	__copy_super(dst, src);
+
+	return 0;
+}
+
+static void __bcache_write_super(struct cache_set *c)
+{
+	struct closure *cl = &c->sb_write;
+	struct cache *ca;
+	unsigned i;
+
+	cache_set_mi_update(c, c->disk_mi, c->sb.nr_in_set);
+
+	closure_init(cl, &c->cl);
+
+	le64_add_cpu(&c->disk_sb.seq, 1);
+
+	for_each_cache(ca, c, i) {
+		struct cache_sb *sb = ca->disk_sb.sb;
+		struct bio *bio = ca->disk_sb.bio;
+
+		cache_sb_from_cache_set(c, ca);
+
+		SET_CACHE_SB_CSUM_TYPE(sb, c->opts.metadata_checksum);
+		sb->csum = cpu_to_le64(__csum_set(sb,
+						  le16_to_cpu(sb->u64s),
+						  CACHE_SB_CSUM_TYPE(sb)));
+
+		bio_reset(bio);
+		bio->bi_bdev	= ca->disk_sb.bdev;
+		bio->bi_end_io	= write_super_endio;
+		bio->bi_private = ca;
+
+		closure_get(cl);
+		percpu_ref_get(&ca->ref);
+		__write_super(c, &ca->disk_sb);
+	}
+
+	closure_return_with_destructor(cl, bcache_write_super_unlock);
+}
+
+void bcache_write_super(struct cache_set *c)
+{
+	down(&c->sb_write_mutex);
+	__bcache_write_super(c);
+}
+
+void bch_check_mark_super_slowpath(struct cache_set *c, const struct bkey_i *k,
+				   bool meta)
+{
+	struct cache_member *mi;
+	struct bkey_s_c_extent e = bkey_i_to_s_c_extent(k);
+	const struct bch_extent_ptr *ptr;
+
+	if (!CACHE_SET_SYNC(&c->disk_sb))
+		return;
+
+	down(&c->sb_write_mutex);
+
+	/* recheck, might have raced */
+	if (bch_check_super_marked(c, k, meta)) {
+		up(&c->sb_write_mutex);
+		return;
+	}
+
+	mi = c->disk_mi;
+
+	extent_for_each_ptr(e, ptr)
+		if (bch_extent_ptr_is_dirty(c, e, ptr))
+			(meta
+			 ? SET_CACHE_HAS_METADATA
+			 : SET_CACHE_HAS_DATA)(mi + ptr->dev, true);
+
+	__bcache_write_super(c);
+}
+
+/* Cache set RO/RW: */
+
+/*
+ * For startup/shutdown of RW stuff, the dependencies are:
+ *
+ * - foreground writes depend on copygc and tiering (to free up space)
+ *
+ * - copygc and tiering depend on mark and sweep gc (they actually probably
+ *   don't because they either reserve ahead of time or don't block if
+ *   allocations fail, but allocations can require mark and sweep gc to run
+ *   because of generation number wraparound)
+ *
+ * - all of the above depends on the allocator threads
+ *
+ * - allocator depends on the journal (when it rewrites prios and gens)
+ */
+
+static void __bch_cache_set_read_only(struct cache_set *c)
+{
+	struct cache *ca;
+	unsigned i;
+
+	c->tiering_pd.rate.rate = UINT_MAX;
+	bch_ratelimit_reset(&c->tiering_pd.rate);
+	bch_tiering_read_stop(c);
+
+	for_each_cache(ca, c, i)
+		bch_moving_gc_stop(ca);
+
+	bch_gc_thread_stop(c);
+
+	bch_btree_flush(c);
+
+	for_each_cache(ca, c, i)
+		bch_cache_allocator_stop(ca);
+
+	/*
+	 * Write a journal entry after flushing the btree, so we don't end up
+	 * replaying everything we just flushed:
+	 */
+	if (test_bit(JOURNAL_STARTED, &c->journal.flags)) {
+		int ret;
+
+		bch_journal_flush_async(&c->journal, NULL);
+		ret = bch_journal_meta(&c->journal);
+		BUG_ON(ret && !bch_journal_error(&c->journal));
+	}
+
+	cancel_delayed_work_sync(&c->journal.write_work);
+	cancel_delayed_work_sync(&c->journal.reclaim_work);
+}
+
+static void bch_writes_disabled(struct percpu_ref *writes)
+{
+	struct cache_set *c = container_of(writes, struct cache_set, writes);
+
+	set_bit(CACHE_SET_WRITE_DISABLE_COMPLETE, &c->flags);
+	wake_up(&bch_read_only_wait);
+}
+
+static void bch_cache_set_read_only_work(struct work_struct *work)
+{
+	struct cache_set *c =
+		container_of(work, struct cache_set, read_only_work);
+
+	percpu_ref_put(&c->writes);
+
+	del_timer(&c->foreground_write_wakeup);
+	cancel_delayed_work(&c->pd_controllers_update);
+
+	c->foreground_write_pd.rate.rate = UINT_MAX;
+	bch_wake_delayed_writes((unsigned long) c);
+
+	if (!test_bit(CACHE_SET_EMERGENCY_RO, &c->flags)) {
+		/*
+		 * If we're not doing an emergency shutdown, we want to wait on
+		 * outstanding writes to complete so they don't see spurious
+		 * errors due to shutting down the allocator:
+		 */
+		wait_event(bch_read_only_wait,
+			   test_bit(CACHE_SET_WRITE_DISABLE_COMPLETE, &c->flags));
+
+		__bch_cache_set_read_only(c);
+
+		if (!bch_journal_error(&c->journal) &&
+		    !test_bit(CACHE_SET_ERROR, &c->flags)) {
+			SET_CACHE_SET_CLEAN(&c->disk_sb, true);
+			bcache_write_super(c);
+		}
+	} else {
+		/*
+		 * If we are doing an emergency shutdown outstanding writes may
+		 * hang until we shutdown the allocator so we don't want to wait
+		 * on outstanding writes before shutting everything down - but
+		 * we do need to wait on them before returning and signalling
+		 * that going RO is complete:
+		 */
+		__bch_cache_set_read_only(c);
+
+		wait_event(bch_read_only_wait,
+			   test_bit(CACHE_SET_WRITE_DISABLE_COMPLETE, &c->flags));
+	}
+
+	bch_notify_cache_set_read_only(c);
+	trace_bcache_cache_set_read_only_done(c);
+
+	set_bit(CACHE_SET_RO_COMPLETE, &c->flags);
+	wake_up(&bch_read_only_wait);
+}
+
+bool bch_cache_set_read_only(struct cache_set *c)
+{
+	if (test_and_set_bit(CACHE_SET_RO, &c->flags))
+		return false;
+
+	trace_bcache_cache_set_read_only(c);
+
+	percpu_ref_get(&c->writes);
+
+	/*
+	 * Block new foreground-end write operations from starting - any new
+	 * writes will return -EROFS:
+	 *
+	 * (This is really blocking new _allocations_, writes to previously
+	 * allocated space can still happen until stopping the allocator in
+	 * bch_cache_allocator_stop()).
+	 */
+	percpu_ref_kill(&c->writes);
+
+	queue_work(system_freezable_wq, &c->read_only_work);
+	return true;
+}
+
+bool bch_cache_set_emergency_read_only(struct cache_set *c)
+{
+	bool ret = !test_and_set_bit(CACHE_SET_EMERGENCY_RO, &c->flags);
+
+	bch_cache_set_read_only(c);
+	bch_journal_halt(&c->journal);
+
+	wake_up(&bch_read_only_wait);
+	return ret;
+}
+
+void bch_cache_set_read_only_sync(struct cache_set *c)
+{
+	/* so we don't race with bch_cache_set_read_write() */
+	lockdep_assert_held(&bch_register_lock);
+
+	bch_cache_set_read_only(c);
+
+	wait_event(bch_read_only_wait,
+		   test_bit(CACHE_SET_RO_COMPLETE, &c->flags) &&
+		   test_bit(CACHE_SET_WRITE_DISABLE_COMPLETE, &c->flags));
+}
+
+static const char *__bch_cache_set_read_write(struct cache_set *c)
+{
+	struct cache *ca;
+	const char *err;
+	unsigned i;
+
+	lockdep_assert_held(&bch_register_lock);
+
+	err = "error starting allocator thread";
+	for_each_cache(ca, c, i)
+		if (ca->mi.state == CACHE_ACTIVE &&
+		    bch_cache_allocator_start(ca)) {
+			percpu_ref_put(&ca->ref);
+			goto err;
+		}
+
+	err = "error starting btree GC thread";
+	if (bch_gc_thread_start(c))
+		goto err;
+
+	for_each_cache(ca, c, i) {
+		if (ca->mi.state != CACHE_ACTIVE)
+			continue;
+
+		err = "error starting moving GC thread";
+		if (bch_moving_gc_thread_start(ca)) {
+			percpu_ref_put(&ca->ref);
+			goto err;
+		}
+	}
+
+	err = "error starting tiering thread";
+	if (bch_tiering_read_start(c))
+		goto err;
+
+	schedule_delayed_work(&c->pd_controllers_update, 5 * HZ);
+
+	return NULL;
+err:
+	__bch_cache_set_read_only(c);
+	return err;
+}
+
+const char *bch_cache_set_read_write(struct cache_set *c)
+{
+	const char *err;
+
+	lockdep_assert_held(&bch_register_lock);
+
+	if (!test_bit(CACHE_SET_RO_COMPLETE, &c->flags))
+		return NULL;
+
+	err = __bch_cache_set_read_write(c);
+	if (err)
+		return err;
+
+	percpu_ref_reinit(&c->writes);
+
+	clear_bit(CACHE_SET_WRITE_DISABLE_COMPLETE, &c->flags);
+	clear_bit(CACHE_SET_EMERGENCY_RO, &c->flags);
+	clear_bit(CACHE_SET_RO_COMPLETE, &c->flags);
+	clear_bit(CACHE_SET_RO, &c->flags);
+	return NULL;
+}
+
+/* Cache set startup/shutdown: */
+
+static void cache_set_free(struct cache_set *c)
+{
+	del_timer_sync(&c->foreground_write_wakeup);
+	cancel_delayed_work_sync(&c->pd_controllers_update);
+	cancel_work_sync(&c->read_only_work);
+	cancel_work_sync(&c->bio_submit_work);
+	cancel_work_sync(&c->read_retry_work);
+
+	bch_btree_cache_free(c);
+	bch_journal_free(&c->journal);
+	bch_io_clock_exit(&c->io_clock[WRITE]);
+	bch_io_clock_exit(&c->io_clock[READ]);
+	bch_compress_free(c);
+	bdi_destroy(&c->bdi);
+	lg_lock_free(&c->bucket_stats_lock);
+	free_percpu(c->bucket_stats_percpu);
+	mempool_exit(&c->btree_bounce_pool);
+	mempool_exit(&c->bio_bounce_pages);
+	bioset_exit(&c->bio_write);
+	bioset_exit(&c->bio_read_split);
+	bioset_exit(&c->bio_read);
+	bioset_exit(&c->btree_read_bio);
+	mempool_exit(&c->btree_interior_update_pool);
+	mempool_exit(&c->btree_reserve_pool);
+	mempool_exit(&c->fill_iter);
+	mempool_exit(&c->search);
+	percpu_ref_exit(&c->writes);
+
+	if (c->copygc_wq)
+		destroy_workqueue(c->copygc_wq);
+	if (c->wq)
+		destroy_workqueue(c->wq);
+
+	kfree_rcu(rcu_dereference_protected(c->members, 1), rcu); /* shutting down */
+	kfree(c->disk_mi);
+	kfree(c);
+	module_put(THIS_MODULE);
+}
+
+/*
+ * should be __cache_set_stop4 - block devices are closed, now we can finally
+ * free it
+ */
+void bch_cache_set_release(struct kobject *kobj)
+{
+	struct cache_set *c = container_of(kobj, struct cache_set, kobj);
+	struct completion *stop_completion = c->stop_completion;
+
+	bch_notify_cache_set_stopped(c);
+	bch_info(c, "stopped");
+
+	cache_set_free(c);
+
+	if (stop_completion)
+		complete(stop_completion);
+}
+
+/*
+ * All activity on the cache_set should have stopped now - close devices:
+ */
+static void __cache_set_stop3(struct closure *cl)
+{
+	struct cache_set *c = container_of(cl, struct cache_set, cl);
+	struct cache *ca;
+	unsigned i;
+
+	mutex_lock(&bch_register_lock);
+	for_each_cache(ca, c, i)
+		bch_cache_stop(ca);
+	mutex_unlock(&bch_register_lock);
+
+	mutex_lock(&bch_register_lock);
+	list_del(&c->list);
+	if (c->minor >= 0)
+		idr_remove(&bch_chardev_minor, c->minor);
+	mutex_unlock(&bch_register_lock);
+
+	closure_debug_destroy(&c->cl);
+	kobject_put(&c->kobj);
+}
+
+/*
+ * Openers (i.e. block devices) should have exited, shutdown all userspace
+ * interfaces and wait for &c->cl to hit 0
+ */
+static void __cache_set_stop2(struct closure *cl)
+{
+	struct cache_set *c = container_of(cl, struct cache_set, caching);
+
+	bch_debug_exit_cache_set(c);
+
+	if (!IS_ERR_OR_NULL(c->chardev))
+		device_unregister(c->chardev);
+
+	if (c->kobj.state_in_sysfs)
+		kobject_del(&c->kobj);
+
+	bch_cache_accounting_destroy(&c->accounting);
+
+	kobject_put(&c->time_stats);
+	kobject_put(&c->opts_dir);
+	kobject_put(&c->internal);
+
+	mutex_lock(&bch_register_lock);
+	bch_cache_set_read_only_sync(c);
+	mutex_unlock(&bch_register_lock);
+
+	closure_return(cl);
+}
+
+/*
+ * First phase of the shutdown process that's kicked off by cache_set_stop(); we
+ * haven't waited for anything to stop yet, we're just punting to process
+ * context to shut down block devices:
+ */
+static void __cache_set_stop1(struct closure *cl)
+{
+	struct cache_set *c = container_of(cl, struct cache_set, caching);
+
+	bch_blockdevs_stop(c);
+
+	continue_at(cl, __cache_set_stop2, system_wq);
+}
+
+void bch_cache_set_stop(struct cache_set *c)
+{
+	if (!test_and_set_bit(CACHE_SET_STOPPING, &c->flags))
+		closure_queue(&c->caching);
+}
+
+void bch_cache_set_unregister(struct cache_set *c)
+{
+	if (!test_and_set_bit(CACHE_SET_UNREGISTERING, &c->flags))
+		bch_cache_set_stop(c);
+}
+
+static unsigned cache_set_nr_devices(struct cache_set *c)
+{
+	unsigned i, nr = 0;
+	struct cache_member *mi = c->disk_mi;
+
+	lockdep_assert_held(&bch_register_lock);
+
+	for (i = 0; i < c->disk_sb.nr_in_set; i++)
+		if (!bch_is_zero(mi[i].uuid.b, sizeof(uuid_le)))
+			nr++;
+
+	return nr;
+}
+
+static unsigned cache_set_nr_online_devices(struct cache_set *c)
+{
+	unsigned i, nr = 0;
+
+	for (i = 0; i < c->sb.nr_in_set; i++)
+		if (c->cache[i])
+			nr++;
+
+	return nr;
+}
+
+#define alloc_bucket_pages(gfp, ca)			\
+	((void *) __get_free_pages(__GFP_ZERO|gfp, ilog2(bucket_pages(ca))))
+
+static struct cache_set *bch_cache_set_alloc(struct cache_sb *sb,
+					     struct cache_set_opts opts)
+{
+	struct cache_set *c;
+	unsigned iter_size, journal_entry_bytes;
+
+	c = kzalloc(sizeof(struct cache_set), GFP_KERNEL);
+	if (!c)
+		return NULL;
+
+	__module_get(THIS_MODULE);
+
+	c->minor		= -1;
+
+	sema_init(&c->sb_write_mutex, 1);
+	INIT_RADIX_TREE(&c->devices, GFP_KERNEL);
+	mutex_init(&c->btree_cache_lock);
+	mutex_init(&c->bucket_lock);
+	mutex_init(&c->btree_root_lock);
+	INIT_WORK(&c->read_only_work, bch_cache_set_read_only_work);
+	mutex_init(&c->mi_lock);
+
+	init_rwsem(&c->gc_lock);
+
+#define BCH_TIME_STAT(name, frequency_units, duration_units)		\
+	spin_lock_init(&c->name##_time.lock);
+	BCH_TIME_STATS()
+#undef BCH_TIME_STAT
+
+	bch_open_buckets_init(c);
+	bch_tiering_init_cache_set(c);
+
+	INIT_LIST_HEAD(&c->list);
+	INIT_LIST_HEAD(&c->cached_devs);
+	INIT_LIST_HEAD(&c->btree_cache);
+	INIT_LIST_HEAD(&c->btree_cache_freeable);
+	INIT_LIST_HEAD(&c->btree_cache_freed);
+
+	INIT_LIST_HEAD(&c->btree_interior_update_list);
+	mutex_init(&c->btree_reserve_cache_lock);
+	mutex_init(&c->btree_interior_update_lock);
+
+	mutex_init(&c->bio_bounce_pages_lock);
+	INIT_WORK(&c->bio_submit_work, bch_bio_submit_work);
+	spin_lock_init(&c->bio_submit_lock);
+	bio_list_init(&c->read_retry_list);
+	spin_lock_init(&c->read_retry_lock);
+	INIT_WORK(&c->read_retry_work, bch_read_retry_work);
+	mutex_init(&c->zlib_workspace_lock);
+
+	seqcount_init(&c->gc_pos_lock);
+
+	c->prio_clock[READ].hand = 1;
+	c->prio_clock[READ].min_prio = 0;
+	c->prio_clock[WRITE].hand = 1;
+	c->prio_clock[WRITE].min_prio = 0;
+
+	c->congested_read_threshold_us	= 2000;
+	c->congested_write_threshold_us	= 20000;
+	c->error_limit	= 16 << IO_ERROR_SHIFT;
+	init_waitqueue_head(&c->writeback_wait);
+
+	c->writeback_pages_max = (256 << 10) / PAGE_SIZE;
+
+	c->copy_gc_enabled = 1;
+	c->tiering_enabled = 1;
+	c->tiering_percent = 10;
+
+	c->foreground_target_percent = 20;
+
+	c->journal.write_time	= &c->journal_write_time;
+	c->journal.delay_time	= &c->journal_delay_time;
+	c->journal.blocked_time	= &c->journal_blocked_time;
+	c->journal.flush_seq_time = &c->journal_flush_seq_time;
+
+	mutex_init(&c->uevent_lock);
+
+	if (cache_sb_to_cache_set(c, sb))
+		goto err;
+
+	scnprintf(c->name, sizeof(c->name), "%pU", &c->disk_sb.user_uuid);
+
+	c->opts = cache_superblock_opts(sb);
+	cache_set_opts_apply(&c->opts, opts);
+
+	c->block_bits		= ilog2(c->sb.block_size);
+
+	if (cache_set_init_fault("cache_set_alloc"))
+		goto err;
+
+	iter_size = (btree_blocks(c) + 1) * 2 *
+		sizeof(struct btree_node_iter_set);
+
+	journal_entry_bytes = 512U << CACHE_SET_JOURNAL_ENTRY_SIZE(sb);
+
+	if (!(c->wq = alloc_workqueue("bcache",
+				WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_HIGHPRI, 1)) ||
+	    !(c->copygc_wq = alloc_workqueue("bcache_copygc",
+				WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_HIGHPRI, 1)) ||
+	    percpu_ref_init(&c->writes, bch_writes_disabled, 0, GFP_KERNEL) ||
+	    mempool_init_slab_pool(&c->search, 1, bch_search_cache) ||
+	    mempool_init_kmalloc_pool(&c->btree_reserve_pool, 1,
+				      sizeof(struct btree_reserve)) ||
+	    mempool_init_kmalloc_pool(&c->btree_interior_update_pool, 1,
+				      sizeof(struct btree_interior_update)) ||
+	    mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
+	    bioset_init(&c->btree_read_bio, 1, 0) ||
+	    bioset_init(&c->bio_read, 1, offsetof(struct bch_read_bio, bio)) ||
+	    bioset_init(&c->bio_read_split, 1, offsetof(struct bch_read_bio, bio)) ||
+	    bioset_init(&c->bio_write, 1, offsetof(struct bch_write_bio, bio)) ||
+	    mempool_init_page_pool(&c->bio_bounce_pages,
+				   max_t(unsigned,
+					 c->sb.btree_node_size,
+					 CRC32_EXTENT_SIZE_MAX) /
+				   PAGE_SECTORS, 0) ||
+	    !(c->bucket_stats_percpu = alloc_percpu(struct bucket_stats_cache_set)) ||
+	    lg_lock_init(&c->bucket_stats_lock) ||
+	    mempool_init_page_pool(&c->btree_bounce_pool, 1,
+				   ilog2(btree_pages(c))) ||
+	    bdi_setup_and_register(&c->bdi, "bcache") ||
+	    bch_io_clock_init(&c->io_clock[READ]) ||
+	    bch_io_clock_init(&c->io_clock[WRITE]) ||
+	    bch_journal_alloc(&c->journal, journal_entry_bytes) ||
+	    bch_btree_cache_alloc(c) ||
+	    bch_compress_init(c))
+		goto err;
+
+	c->bdi.ra_pages		= VM_MAX_READAHEAD * 1024 / PAGE_SIZE;
+	c->bdi.congested_fn	= bch_congested_fn;
+	c->bdi.congested_data	= c;
+
+	/*
+	 * Now that all allocations have succeeded, init various refcounty
+	 * things that let us shutdown:
+	 */
+	closure_init(&c->cl, NULL);
+
+	c->kobj.kset = bcache_kset;
+	kobject_init(&c->kobj, &bch_cache_set_ktype);
+	kobject_init(&c->internal, &bch_cache_set_internal_ktype);
+	kobject_init(&c->opts_dir, &bch_cache_set_opts_dir_ktype);
+	kobject_init(&c->time_stats, &bch_cache_set_time_stats_ktype);
+
+	bch_cache_accounting_init(&c->accounting, &c->cl);
+
+	closure_init(&c->caching, &c->cl);
+	set_closure_fn(&c->caching, __cache_set_stop1, system_wq);
+
+	continue_at_noreturn(&c->cl, __cache_set_stop3, system_wq);
+	return c;
+err:
+	cache_set_free(c);
+	return NULL;
+}
+
+static int bch_cache_set_online(struct cache_set *c)
+{
+	struct cache *ca;
+	unsigned i;
+
+	lockdep_assert_held(&bch_register_lock);
+
+	if (c->kobj.state_in_sysfs)
+		return 0;
+
+	c->minor = idr_alloc(&bch_chardev_minor, c, 0, 0, GFP_KERNEL);
+	if (c->minor < 0)
+		return c->minor;
+
+	c->chardev = device_create(bch_chardev_class, NULL,
+				   MKDEV(bch_chardev_major, c->minor), NULL,
+				   "bcache%u-ctl", c->minor);
+	if (IS_ERR(c->chardev))
+		return PTR_ERR(c->chardev);
+
+	if (kobject_add(&c->kobj, NULL, "%pU", c->disk_sb.user_uuid.b) ||
+	    kobject_add(&c->internal, &c->kobj, "internal") ||
+	    kobject_add(&c->opts_dir, &c->kobj, "options") ||
+	    kobject_add(&c->time_stats, &c->kobj, "time_stats") ||
+	    bch_cache_accounting_add_kobjs(&c->accounting, &c->kobj))
+		return -1;
+
+	for_each_cache(ca, c, i)
+		if (bch_cache_online(ca)) {
+			percpu_ref_put(&ca->ref);
+			return -1;
+		}
+
+	list_add(&c->list, &bch_cache_sets);
+	return 0;
+}
+
+static const char *run_cache_set(struct cache_set *c)
+{
+	const char *err = "cannot allocate memory";
+	struct cache *ca;
+	unsigned i, id;
+	time64_t now;
+	LIST_HEAD(journal);
+	struct jset *j;
+	int ret = -EINVAL;
+
+	lockdep_assert_held(&bch_register_lock);
+	BUG_ON(test_bit(CACHE_SET_RUNNING, &c->flags));
+
+	/* We don't want bch_fatal_error() to free underneath us */
+	closure_get(&c->caching);
+
+	/*
+	 * Make sure that each cache object's mi is up to date before
+	 * we start testing it.
+	 */
+	for_each_cache(ca, c, i)
+		cache_sb_from_cache_set(c, ca);
+
+	/*
+	 * CACHE_SET_SYNC is true if the cache set has already been run
+	 * and potentially has data.
+	 * It is false if it is the first time it is run.
+	 */
+
+	if (CACHE_SET_SYNC(&c->disk_sb)) {
+		ret = bch_journal_read(c, &journal);
+		if (ret)
+			goto err;
+
+		pr_debug("btree_journal_read() done");
+
+		j = &list_entry(journal.prev, struct journal_replay, list)->j;
+
+		err = "error reading priorities";
+		for_each_cache(ca, c, i) {
+			ret = bch_prio_read(ca);
+			if (ret) {
+				percpu_ref_put(&ca->ref);
+				goto err;
+			}
+		}
+
+		c->prio_clock[READ].hand = le16_to_cpu(j->read_clock);
+		c->prio_clock[WRITE].hand = le16_to_cpu(j->write_clock);
+
+		for_each_cache(ca, c, i) {
+			bch_recalc_min_prio(ca, READ);
+			bch_recalc_min_prio(ca, WRITE);
+		}
+
+		/*
+		 * If bch_prio_read() fails it'll call cache_set_error and we'll
+		 * tear everything down right away, but if we perhaps checked
+		 * sooner we could avoid journal replay.
+		 */
+
+		for (id = 0; id < BTREE_ID_NR; id++) {
+			unsigned level;
+			struct bkey_i *k;
+
+			err = "bad btree root";
+			k = bch_journal_find_btree_root(c, j, id, &level);
+			if (!k && id == BTREE_ID_EXTENTS)
+				goto err;
+			if (!k) {
+				pr_debug("missing btree root: %d", id);
+				continue;
+			}
+
+			err = "error reading btree root";
+			if (bch_btree_root_read(c, id, k, level))
+				goto err;
+		}
+
+		bch_verbose(c, "starting mark and sweep:");
+
+		err = "error in recovery";
+		if (bch_initial_gc(c, &journal))
+			goto err;
+
+		bch_verbose(c, "mark and sweep done");
+
+		/*
+		 * bch_journal_start() can't happen sooner, or btree_gc_finish()
+		 * will give spurious errors about oldest_gen > bucket_gen -
+		 * this is a hack but oh well.
+		 */
+		bch_journal_start(c);
+
+		err = "error starting allocator thread";
+		for_each_cache(ca, c, i)
+			if (ca->mi.state == CACHE_ACTIVE &&
+			    bch_cache_allocator_start(ca)) {
+				percpu_ref_put(&ca->ref);
+				goto err;
+			}
+
+		bch_verbose(c, "starting journal replay:");
+
+		err = "journal replay failed";
+		ret = bch_journal_replay(c, &journal);
+		if (ret)
+			goto err;
+
+		bch_verbose(c, "journal replay done");
+
+		/*
+		 * Write a new journal entry _before_ we start journalling new
+		 * data - otherwise, we could end up with btree node bsets with
+		 * journal seqs arbitrarily far in the future vs. the most
+		 * recently written journal entry on disk, if we crash before
+		 * writing the next journal entry:
+		 */
+		err = "error writing journal entry";
+		if (bch_journal_meta(&c->journal))
+			goto err;
+
+		bch_verbose(c, "starting fs gc:");
+		err = "error in fs gc";
+		ret = bch_gc_inode_nlinks(c);
+		if (ret)
+			goto err;
+		bch_verbose(c, "fs gc done");
+
+		if (!c->opts.nofsck) {
+			bch_verbose(c, "starting fsck:");
+			err = "error in fsck";
+			ret = bch_fsck(c);
+			if (ret)
+				goto err;
+			bch_verbose(c, "fsck done");
+		}
+	} else {
+		struct bkey_i_inode inode;
+		struct closure cl;
+
+		closure_init_stack(&cl);
+
+		bch_notice(c, "initializing new filesystem");
+
+		err = "unable to allocate journal buckets";
+		for_each_cache(ca, c, i)
+			if (bch_cache_journal_alloc(ca)) {
+				percpu_ref_put(&ca->ref);
+				goto err;
+			}
+
+		bch_initial_gc(c, NULL);
+
+		/*
+		 * journal_res_get() will crash if called before this has
+		 * set up the journal.pin FIFO and journal.cur pointer:
+		 */
+		bch_journal_start(c);
+		bch_journal_set_replay_done(&c->journal);
+
+		err = "error starting allocator thread";
+		for_each_cache(ca, c, i)
+			if (ca->mi.state == CACHE_ACTIVE &&
+			    bch_cache_allocator_start(ca)) {
+				percpu_ref_put(&ca->ref);
+				goto err;
+			}
+
+		err = "cannot allocate new btree root";
+		for (id = 0; id < BTREE_ID_NR; id++)
+			if (bch_btree_root_alloc(c, id, &cl)) {
+				closure_sync(&cl);
+				goto err;
+			}
+
+		/* Wait for new btree roots to be written: */
+		closure_sync(&cl);
+
+		bkey_inode_init(&inode.k_i);
+		inode.k.p.inode = BCACHE_ROOT_INO;
+		inode.v.i_mode = cpu_to_le16(S_IFDIR|S_IRWXU|S_IRUGO|S_IXUGO);
+		inode.v.i_nlink = cpu_to_le32(2);
+		get_random_bytes(&inode.v.i_hash_seed, sizeof(inode.v.i_hash_seed));
+		SET_INODE_STR_HASH_TYPE(&inode.v, c->sb.str_hash_type);
+
+		err = "error creating root directory";
+		if (bch_btree_insert(c, BTREE_ID_INODES, &inode.k_i,
+				     NULL, NULL, NULL, 0))
+			goto err;
+
+		err = "error writing first journal entry";
+		if (bch_journal_meta(&c->journal))
+			goto err;
+	}
+
+	if (c->opts.read_only) {
+		bch_cache_set_read_only_sync(c);
+	} else {
+		err = __bch_cache_set_read_write(c);
+		if (err)
+			goto err;
+	}
+
+	now = ktime_get_seconds();
+	rcu_read_lock();
+	for_each_cache_rcu(ca, c, i)
+		c->disk_mi[ca->sb.nr_this_dev].last_mount = cpu_to_le64(now);
+	rcu_read_unlock();
+
+	/* Mark cache set as initialized: */
+	SET_CACHE_SET_SYNC(&c->disk_sb, true);
+	SET_CACHE_SET_CLEAN(&c->disk_sb, false);
+	bcache_write_super(c);
+
+	err = "dynamic fault";
+	if (cache_set_init_fault("run_cache_set"))
+		goto err;
+
+	err = "error creating kobject";
+	if (bch_cache_set_online(c))
+		goto err;
+
+	err = "can't bring up blockdev volumes";
+	if (bch_blockdev_volumes_start(c))
+		goto err;
+
+	bch_debug_init_cache_set(c);
+	set_bit(CACHE_SET_RUNNING, &c->flags);
+	bch_attach_backing_devs(c);
+
+	closure_put(&c->caching);
+
+	bch_notify_cache_set_read_write(c);
+
+	BUG_ON(!list_empty(&journal));
+	return NULL;
+err:
+	switch (ret) {
+	case BCH_FSCK_ERRORS_NOT_FIXED:
+		bch_err(c, "filesystem contains errors: please report this to the developers");
+		pr_cont("mount with -o fix_errors to repair");
+		err = "fsck error";
+		break;
+	case BCH_FSCK_REPAIR_UNIMPLEMENTED:
+		bch_err(c, "filesystem contains errors: please report this to the developers");
+		pr_cont("repair unimplemented: inform the developers so that it can be added");
+		err = "fsck error";
+		break;
+	case BCH_FSCK_REPAIR_IMPOSSIBLE:
+		bch_err(c, "filesystem contains errors, but repair impossible");
+		err = "fsck error";
+		break;
+	case BCH_FSCK_UNKNOWN_VERSION:
+		err = "unknown metadata version";;
+		break;
+	case -ENOMEM:
+		err = "cannot allocate memory";
+		break;
+	case -EIO:
+		err = "IO error";
+		break;
+	}
+
+	BUG_ON(!err);
+
+	bch_journal_entries_free(&journal);
+	set_bit(CACHE_SET_ERROR, &c->flags);
+	bch_cache_set_unregister(c);
+	closure_put(&c->caching);
+	return err;
+}
+
+static const char *can_add_cache(struct cache_sb *sb,
+				 struct cache_set *c)
+{
+	if (le16_to_cpu(sb->block_size) != c->sb.block_size)
+		return "mismatched block size";
+
+	if (le16_to_cpu(sb->members[sb->nr_this_dev].bucket_size) <
+	    CACHE_SET_BTREE_NODE_SIZE(&c->disk_sb))
+		return "new cache bucket_size is too small";
+
+	return NULL;
+}
+
+static const char *can_attach_cache(struct cache_sb *sb, struct cache_set *c)
+{
+	const char *err;
+	bool match;
+
+	err = can_add_cache(sb, c);
+	if (err)
+		return err;
+
+	/*
+	 * When attaching an existing device, the cache set superblock must
+	 * already contain member_info with a matching UUID
+	 */
+	match = le64_to_cpu(sb->seq) <= le64_to_cpu(c->disk_sb.seq)
+		? (sb->nr_this_dev < c->disk_sb.nr_in_set &&
+		   !memcmp(&c->disk_mi[sb->nr_this_dev].uuid,
+			   &sb->disk_uuid, sizeof(uuid_le)))
+		: (sb->nr_this_dev < sb->nr_in_set &&
+		   !memcmp(&sb->members[sb->nr_this_dev].uuid,
+			   &sb->disk_uuid, sizeof(uuid_le)));
+
+	if (!match)
+		return "cache sb does not match set";
+
+	return NULL;
+}
+
+/* Cache device */
+
+bool bch_cache_read_only(struct cache *ca)
+{
+	struct cache_set *c = ca->set;
+	char buf[BDEVNAME_SIZE];
+
+	bdevname(ca->disk_sb.bdev, buf);
+
+	lockdep_assert_held(&bch_register_lock);
+
+	if (ca->mi.state != CACHE_ACTIVE)
+		return false;
+
+	if (!bch_cache_may_remove(ca)) {
+		bch_err(c, "required member %s going RO, forcing fs RO", buf);
+		bch_cache_set_read_only_sync(c);
+	}
+
+	trace_bcache_cache_read_only(ca);
+
+	bch_moving_gc_stop(ca);
+
+	/*
+	 * This stops new data writes (e.g. to existing open data
+	 * buckets) and then waits for all existing writes to
+	 * complete.
+	 */
+	bch_cache_allocator_stop(ca);
+
+	bch_cache_group_remove_cache(&c->journal.devs, ca);
+
+	/*
+	 * Device data write barrier -- no non-meta-data writes should
+	 * occur after this point.  However, writes to btree buckets,
+	 * journal buckets, and the superblock can still occur.
+	 */
+	trace_bcache_cache_read_only_done(ca);
+
+	bch_notice(c, "%s read only", bdevname(ca->disk_sb.bdev, buf));
+	bch_notify_cache_read_only(ca);
+
+	SET_CACHE_STATE(&c->disk_mi[ca->sb.nr_this_dev], CACHE_RO);
+	bcache_write_super(c);
+	return true;
+}
+
+static const char *__bch_cache_read_write(struct cache_set *c, struct cache *ca)
+{
+	lockdep_assert_held(&bch_register_lock);
+
+	if (ca->mi.state == CACHE_ACTIVE)
+		return NULL;
+
+	if (test_bit(CACHE_DEV_REMOVING, &ca->flags))
+		return "removing";
+
+	trace_bcache_cache_read_write(ca);
+
+	if (bch_cache_allocator_start(ca))
+		return "error starting allocator thread";
+
+	if (bch_moving_gc_thread_start(ca))
+		return "error starting moving GC thread";
+
+	bch_cache_group_add_cache(&c->journal.devs, ca);
+
+	wake_up_process(c->tiering_read);
+
+	bch_notify_cache_read_write(ca);
+	trace_bcache_cache_read_write_done(ca);
+
+	return NULL;
+}
+
+const char *bch_cache_read_write(struct cache *ca)
+{
+	struct cache_set *c = ca->set;
+	const char *err;
+
+	err = __bch_cache_read_write(c, ca);
+	if (err)
+		return err;
+
+	SET_CACHE_STATE(&c->disk_mi[ca->sb.nr_this_dev], CACHE_ACTIVE);
+	bcache_write_super(c);
+
+	return NULL;
+}
+
+/*
+ * bch_cache_stop has already returned, so we no longer hold the register
+ * lock at the point this is called.
+ */
+
+void bch_cache_release(struct kobject *kobj)
+{
+	struct cache *ca = container_of(kobj, struct cache, kobj);
+
+	percpu_ref_exit(&ca->ref);
+	kfree(ca);
+}
+
+static void bch_cache_free_work(struct work_struct *work)
+{
+	struct cache *ca = container_of(work, struct cache, free_work);
+	struct cache_set *c = ca->set;
+	unsigned i;
+
+	cancel_work_sync(&ca->io_error_work);
+
+	if (c && c->kobj.state_in_sysfs) {
+		char buf[12];
+
+		sprintf(buf, "cache%u", ca->sb.nr_this_dev);
+		sysfs_remove_link(&c->kobj, buf);
+	}
+
+	if (ca->kobj.state_in_sysfs)
+		kobject_del(&ca->kobj);
+
+	free_super(&ca->disk_sb);
+
+	/*
+	 * bch_cache_stop can be called in the middle of initialization
+	 * of the struct cache object.
+	 * As such, not all the sub-structures may be initialized.
+	 * However, they were zeroed when the object was allocated.
+	 */
+
+	free_percpu(ca->sectors_written);
+	bioset_exit(&ca->replica_set);
+	free_percpu(ca->bucket_stats_percpu);
+	kfree(ca->journal.bucket_seq);
+	free_pages((unsigned long) ca->disk_buckets, ilog2(bucket_pages(ca)));
+	kfree(ca->prio_buckets);
+	kfree(ca->bio_prio);
+	kfree(ca->journal.bio);
+	vfree(ca->buckets);
+	vfree(ca->oldest_gens);
+	free_heap(&ca->heap);
+	free_fifo(&ca->free_inc);
+
+	for (i = 0; i < RESERVE_NR; i++)
+		free_fifo(&ca->free[i]);
+
+	kobject_put(&ca->kobj);
+
+	if (c)
+		kobject_put(&c->kobj);
+}
+
+static void bch_cache_percpu_ref_release(struct percpu_ref *ref)
+{
+	struct cache *ca = container_of(ref, struct cache, ref);
+
+	schedule_work(&ca->free_work);
+}
+
+static void bch_cache_free_rcu(struct rcu_head *rcu)
+{
+	struct cache *ca = container_of(rcu, struct cache, free_rcu);
+
+	/*
+	 * This decrements the ref count to ca, and once the ref count
+	 * is 0 (outstanding bios to the ca also incremented it and
+	 * decrement it on completion/error), bch_cache_percpu_ref_release
+	 * is called, and that eventually results in bch_cache_free_work
+	 * being called, which in turn results in bch_cache_release being
+	 * called.
+	 *
+	 * In particular, these functions won't be called until there are no
+	 * bios outstanding (the per-cpu ref counts are all 0), so it
+	 * is safe to remove the actual sysfs device at that point,
+	 * and that can indicate success to the user.
+	 */
+
+	percpu_ref_kill(&ca->ref);
+}
+
+static void bch_cache_stop(struct cache *ca)
+{
+	struct cache_set *c = ca->set;
+
+	lockdep_assert_held(&bch_register_lock);
+
+	if (c) {
+		BUG_ON(rcu_access_pointer(c->cache[ca->sb.nr_this_dev]) != ca);
+		rcu_assign_pointer(c->cache[ca->sb.nr_this_dev], NULL);
+	}
+
+	call_rcu(&ca->free_rcu, bch_cache_free_rcu);
+}
+
+static void bch_cache_remove_work(struct work_struct *work)
+{
+	struct cache *ca = container_of(work, struct cache, remove_work);
+	struct cache_set *c = ca->set;
+	char name[BDEVNAME_SIZE];
+	bool force = test_bit(CACHE_DEV_FORCE_REMOVE, &ca->flags);
+	unsigned dev = ca->sb.nr_this_dev;
+
+	bdevname(ca->disk_sb.bdev, name);
+
+	/*
+	 * Device should already be RO, now migrate data off:
+	 *
+	 * XXX: locking is sketchy, bch_cache_read_write() has to check
+	 * CACHE_DEV_REMOVING bit
+	 */
+	if (!ca->mi.has_data) {
+		/* Nothing to do: */
+	} else if (!bch_move_data_off_device(ca)) {
+		lockdep_assert_held(&bch_register_lock);
+		SET_CACHE_HAS_DATA(&c->disk_mi[ca->sb.nr_this_dev], false);
+
+		bcache_write_super(c);
+	} else if (force) {
+		bch_flag_data_bad(ca);
+
+		lockdep_assert_held(&bch_register_lock);
+		SET_CACHE_HAS_DATA(&c->disk_mi[ca->sb.nr_this_dev], false);
+
+		bcache_write_super(c);
+	} else {
+		bch_err(c, "Remove of %s failed, unable to migrate data off",
+			name);
+		clear_bit(CACHE_DEV_REMOVING, &ca->flags);
+		return;
+	}
+
+	/* Now metadata: */
+
+	if (!ca->mi.has_metadata) {
+		/* Nothing to do: */
+	} else if (!bch_move_meta_data_off_device(ca)) {
+		lockdep_assert_held(&bch_register_lock);
+		SET_CACHE_HAS_METADATA(&c->disk_mi[ca->sb.nr_this_dev], false);
+
+		bcache_write_super(c);
+	} else {
+		bch_err(c, "Remove of %s failed, unable to migrate metadata off",
+			name);
+		clear_bit(CACHE_DEV_REMOVING, &ca->flags);
+		return;
+	}
+
+	/*
+	 * Ok, really doing the remove:
+	 * Drop device's prio pointer before removing it from superblock:
+	 */
+	bch_notify_cache_removed(ca);
+
+	spin_lock(&c->journal.lock);
+	c->journal.prio_buckets[dev] = 0;
+	spin_unlock(&c->journal.lock);
+
+	bch_journal_meta(&c->journal);
+
+	/*
+	 * Stop device before removing it from the cache set's list of devices -
+	 * and get our own ref on cache set since ca is going away:
+	 */
+	closure_get(&c->cl);
+
+	mutex_lock(&bch_register_lock);
+	bch_cache_stop(ca);
+
+	/*
+	 * RCU barrier between dropping between c->cache and dropping from
+	 * member info:
+	 */
+	synchronize_rcu();
+
+	lockdep_assert_held(&bch_register_lock);
+
+	/*
+	 * Free this device's slot in the cache_member array - all pointers to
+	 * this device must be gone:
+	 */
+	memset(&c->disk_mi[dev].uuid, 0, sizeof(c->disk_mi[dev].uuid));
+
+	bcache_write_super(c);
+	mutex_unlock(&bch_register_lock);
+
+	closure_put(&c->cl);
+}
+
+bool bch_cache_remove(struct cache *ca, bool force)
+{
+	mutex_lock(&bch_register_lock);
+
+	if (test_bit(CACHE_DEV_REMOVING, &ca->flags))
+		return false;
+
+	if (!bch_cache_may_remove(ca)) {
+		bch_err(ca->set, "Can't remove last device in tier %u",
+			ca->mi.tier);
+		bch_notify_cache_remove_failed(ca);
+		return false;
+	}
+
+	/* First, go RO before we try to migrate data off: */
+	bch_cache_read_only(ca);
+
+	if (force)
+		set_bit(CACHE_DEV_FORCE_REMOVE, &ca->flags);
+	set_bit(CACHE_DEV_REMOVING, &ca->flags);
+	bch_notify_cache_removing(ca);
+
+	mutex_unlock(&bch_register_lock);
+
+	/* Migrate the data and finish removal asynchronously: */
+
+	queue_work(system_long_wq, &ca->remove_work);
+	return true;
+}
+
+static int bch_cache_online(struct cache *ca)
+{
+	char buf[12];
+
+	lockdep_assert_held(&bch_register_lock);
+
+	sprintf(buf, "cache%u", ca->sb.nr_this_dev);
+
+	if (kobject_add(&ca->kobj,
+			&part_to_dev(ca->disk_sb.bdev->bd_part)->kobj,
+			"bcache") ||
+	    sysfs_create_link(&ca->kobj, &ca->set->kobj, "set") ||
+	    sysfs_create_link(&ca->set->kobj, &ca->kobj, buf))
+		return -1;
+
+	return 0;
+}
+
+static const char *cache_alloc(struct bcache_superblock *sb,
+			       struct cache_set *c,
+			       struct cache **ret)
+{
+	size_t reserve_none, movinggc_reserve, free_inc_reserve, total_reserve;
+	size_t heap_size;
+	unsigned i, journal_entry_pages;
+	const char *err = "cannot allocate memory";
+	struct cache *ca;
+
+	if (c->sb.nr_in_set == 1)
+		bdevname(sb->bdev, c->name);
+
+	if (cache_set_init_fault("cache_alloc"))
+		return err;
+
+	ca = kzalloc(sizeof(*ca), GFP_KERNEL);
+	if (!ca)
+		return err;
+
+	if (percpu_ref_init(&ca->ref, bch_cache_percpu_ref_release,
+			    0, GFP_KERNEL)) {
+		kfree(ca);
+		return err;
+	}
+
+	kobject_init(&ca->kobj, &bch_cache_ktype);
+
+	spin_lock_init(&ca->self.lock);
+	ca->self.nr_devices = 1;
+	rcu_assign_pointer(ca->self.d[0].dev, ca);
+	ca->sb.nr_this_dev = sb->sb->nr_this_dev;
+
+	INIT_WORK(&ca->free_work, bch_cache_free_work);
+	INIT_WORK(&ca->remove_work, bch_cache_remove_work);
+	spin_lock_init(&ca->freelist_lock);
+	spin_lock_init(&ca->prio_buckets_lock);
+	mutex_init(&ca->heap_lock);
+	bch_moving_init_cache(ca);
+
+	ca->disk_sb = *sb;
+	ca->disk_sb.bdev->bd_holder = ca;
+	memset(sb, 0, sizeof(*sb));
+
+	INIT_WORK(&ca->io_error_work, bch_nonfatal_io_error_work);
+
+	err = "dynamic fault";
+	if (cache_set_init_fault("cache_alloc"))
+		goto err;
+
+	ca->mi = cache_mi_to_cpu_mi(ca->disk_sb.sb->members +
+				    ca->disk_sb.sb->nr_this_dev);
+	ca->bucket_bits = ilog2(ca->mi.bucket_size);
+
+	/* XXX: tune these */
+	movinggc_reserve = max_t(size_t, 16, ca->mi.nbuckets >> 7);
+	reserve_none = max_t(size_t, 4, ca->mi.nbuckets >> 9);
+	/*
+	 * free_inc must be smaller than the copygc reserve: if it was bigger,
+	 * one copygc iteration might not make enough buckets available to fill
+	 * up free_inc and allow the allocator to make forward progress
+	 */
+	free_inc_reserve = movinggc_reserve / 2;
+	heap_size = movinggc_reserve * 8;
+
+	journal_entry_pages =
+		DIV_ROUND_UP(1U << CACHE_SET_JOURNAL_ENTRY_SIZE(ca->disk_sb.sb),
+			     PAGE_SECTORS);
+
+	if (!init_fifo(&ca->free[RESERVE_PRIO], prio_buckets(ca), GFP_KERNEL) ||
+	    !init_fifo(&ca->free[RESERVE_BTREE], BTREE_NODE_RESERVE, GFP_KERNEL) ||
+	    !init_fifo(&ca->free[RESERVE_MOVINGGC],
+		       movinggc_reserve, GFP_KERNEL) ||
+	    !init_fifo(&ca->free[RESERVE_NONE], reserve_none, GFP_KERNEL) ||
+	    !init_fifo(&ca->free_inc,	free_inc_reserve, GFP_KERNEL) ||
+	    !init_heap(&ca->heap,	heap_size, GFP_KERNEL) ||
+	    !(ca->oldest_gens	= vzalloc(sizeof(u8) *
+					  ca->mi.nbuckets)) ||
+	    !(ca->buckets	= vzalloc(sizeof(struct bucket) *
+					  ca->mi.nbuckets)) ||
+	    !(ca->prio_buckets	= kzalloc(sizeof(uint64_t) * prio_buckets(ca) *
+					  2, GFP_KERNEL)) ||
+	    !(ca->disk_buckets	= alloc_bucket_pages(GFP_KERNEL, ca)) ||
+	    !(ca->bucket_stats_percpu = alloc_percpu(struct bucket_stats_cache)) ||
+	    !(ca->journal.bucket_seq = kcalloc(bch_nr_journal_buckets(ca->disk_sb.sb),
+					       sizeof(u64), GFP_KERNEL)) ||
+	    !(ca->journal.bio = bio_kmalloc(GFP_KERNEL, journal_entry_pages)) ||
+	    !(ca->bio_prio = bio_kmalloc(GFP_KERNEL, bucket_pages(ca))) ||
+	    bioset_init(&ca->replica_set, 4,
+			offsetof(struct bch_write_bio, bio)) ||
+	    !(ca->sectors_written = alloc_percpu(*ca->sectors_written)))
+		goto err;
+
+	ca->prio_last_buckets = ca->prio_buckets + prio_buckets(ca);
+
+	total_reserve = ca->free_inc.size;
+	for (i = 0; i < RESERVE_NR; i++)
+		total_reserve += ca->free[i].size;
+	pr_debug("%zu buckets reserved", total_reserve);
+
+	ca->copygc_write_point.group = &ca->self;
+	ca->tiering_write_point.group = &ca->self;
+
+	kobject_get(&c->kobj);
+	ca->set = c;
+
+	kobject_get(&ca->kobj);
+	rcu_assign_pointer(c->cache[ca->sb.nr_this_dev], ca);
+
+	if (le64_to_cpu(ca->disk_sb.sb->seq) > le64_to_cpu(c->disk_sb.seq))
+		cache_sb_to_cache_set(c, ca->disk_sb.sb);
+
+	/*
+	 * Increase journal write timeout if flushes to this device are
+	 * expensive:
+	 */
+	if (!blk_queue_nonrot(bdev_get_queue(ca->disk_sb.bdev)) &&
+	    journal_flushes_device(ca))
+		c->journal.write_delay_ms =
+			max(c->journal.write_delay_ms, 1000U);
+
+	err = "error creating kobject";
+	if (c->kobj.state_in_sysfs &&
+	    bch_cache_online(ca))
+		goto err;
+
+	if (ret)
+		*ret = ca;
+	else
+		kobject_put(&ca->kobj);
+	return NULL;
+err:
+	bch_cache_stop(ca);
+	return err;
+}
+
+static struct cache_set *cache_set_lookup(uuid_le uuid)
+{
+	struct cache_set *c;
+
+	lockdep_assert_held(&bch_register_lock);
+
+	list_for_each_entry(c, &bch_cache_sets, list)
+		if (!memcmp(&c->disk_sb.set_uuid, &uuid, sizeof(uuid_le)))
+			return c;
+
+	return NULL;
+}
+
+static const char *register_cache(struct bcache_superblock *sb,
+				  struct cache_set_opts opts)
+{
+	char name[BDEVNAME_SIZE];
+	const char *err = "cannot allocate memory";
+	struct cache_set *c;
+
+	err = validate_cache_super(sb);
+	if (err)
+		return err;
+
+	bdevname(sb->bdev, name);
+
+	c = cache_set_lookup(sb->sb->set_uuid);
+	if (c) {
+		if ((err = (can_attach_cache(sb->sb, c) ?:
+			    cache_alloc(sb, c, NULL))))
+			return err;
+
+		if (cache_set_nr_online_devices(c) == cache_set_nr_devices(c)) {
+			err = run_cache_set(c);
+			if (err)
+				return err;
+		}
+		goto out;
+	}
+
+	c = bch_cache_set_alloc(sb->sb, opts);
+	if (!c)
+		return err;
+
+	err = cache_alloc(sb, c, NULL);
+	if (err)
+		goto err_stop;
+
+	if (cache_set_nr_online_devices(c) == cache_set_nr_devices(c)) {
+		err = run_cache_set(c);
+		if (err)
+			goto err_stop;
+	}
+
+	err = "error creating kobject";
+	if (bch_cache_set_online(c))
+		goto err_stop;
+out:
+
+	bch_info(c, "started");
+	return NULL;
+err_stop:
+	bch_cache_set_stop(c);
+	return err;
+}
+
+int bch_cache_set_add_cache(struct cache_set *c, const char *path)
+{
+	struct bcache_superblock sb;
+	const char *err;
+	struct cache *ca;
+	struct cache_member *new_mi = NULL;
+	struct cache_member mi;
+	unsigned nr_this_dev, nr_in_set, u64s;
+	int ret = -EINVAL;
+
+	mutex_lock(&bch_register_lock);
+
+	err = read_super(&sb, path);
+	if (err)
+		goto err_unlock;
+
+	err = validate_cache_super(&sb);
+	if (err)
+		goto err_unlock;
+
+	err = can_add_cache(sb.sb, c);
+	if (err)
+		goto err_unlock;
+
+	/*
+	 * Preserve the old cache member information (esp. tier)
+	 * before we start bashing the disk stuff.
+	 */
+	mi = sb.sb->members[sb.sb->nr_this_dev];
+	mi.last_mount = cpu_to_le64(ktime_get_seconds());
+
+	down_read(&c->gc_lock);
+
+	if (dynamic_fault("bcache:add:no_slot"))
+		goto no_slot;
+
+	if (test_bit(CACHE_SET_GC_FAILURE, &c->flags))
+		goto no_slot;
+
+	for (nr_this_dev = 0; nr_this_dev < MAX_CACHES_PER_SET; nr_this_dev++)
+		if (nr_this_dev >= c->sb.nr_in_set ||
+		    bch_is_zero(c->disk_mi[nr_this_dev].uuid.b,
+				 sizeof(uuid_le)))
+			goto have_slot;
+no_slot:
+	up_read(&c->gc_lock);
+
+	err = "no slots available in superblock";
+	ret = -ENOSPC;
+	goto err_unlock;
+
+have_slot:
+	nr_in_set = max_t(unsigned, nr_this_dev + 1, c->sb.nr_in_set);
+	up_read(&c->gc_lock);
+
+	u64s = nr_in_set * (sizeof(struct cache_member) / sizeof(u64));
+	err = "no space in superblock for member info";
+	if (bch_super_realloc(&sb, u64s))
+		goto err_unlock;
+
+	new_mi = dynamic_fault("bcache:add:member_info_realloc")
+		? NULL
+		: kmalloc(sizeof(struct cache_member) * nr_in_set,
+			  GFP_KERNEL);
+	if (!new_mi) {
+		err = "cannot allocate memory";
+		ret = -ENOMEM;
+		goto err_unlock;
+	}
+
+	memcpy(new_mi, c->disk_mi,
+	       sizeof(struct cache_member) * nr_in_set);
+	new_mi[nr_this_dev] = mi;
+
+	sb.sb->nr_this_dev	= nr_this_dev;
+	sb.sb->nr_in_set	= nr_in_set;
+	sb.sb->u64s		= cpu_to_le16(u64s);
+	memcpy(sb.sb->members, new_mi,
+	       sizeof(struct cache_member) * nr_in_set);
+
+	if (cache_set_mi_update(c, new_mi, nr_in_set)) {
+		err = "cannot allocate memory";
+		ret = -ENOMEM;
+		goto err_unlock;
+	}
+
+	/* commit new member info */
+	swap(c->disk_mi, new_mi);
+	kfree(new_mi);
+	new_mi = NULL;
+	c->disk_sb.nr_in_set = nr_in_set;
+	c->sb.nr_in_set = nr_in_set;
+
+	err = cache_alloc(&sb, c, &ca);
+	if (err)
+		goto err_unlock;
+
+	bcache_write_super(c);
+
+	err = "journal alloc failed";
+	if (bch_cache_journal_alloc(ca))
+		goto err_put;
+
+	bch_notify_cache_added(ca);
+
+	if (ca->mi.state == CACHE_ACTIVE) {
+		err = __bch_cache_read_write(c, ca);
+		if (err)
+			goto err_put;
+	}
+
+	kobject_put(&ca->kobj);
+	mutex_unlock(&bch_register_lock);
+	return 0;
+err_put:
+	bch_cache_stop(ca);
+err_unlock:
+	kfree(new_mi);
+	free_super(&sb);
+	mutex_unlock(&bch_register_lock);
+
+	bch_err(c, "Unable to add device: %s", err);
+	return ret ?: -EINVAL;
+}
+
+const char *bch_register_cache_set(char * const *devices, unsigned nr_devices,
+				   struct cache_set_opts opts,
+				   struct cache_set **ret)
+{
+	const char *err;
+	struct cache_set *c = NULL;
+	struct bcache_superblock *sb;
+	uuid_le uuid;
+	unsigned i;
+
+	memset(&uuid, 0, sizeof(uuid_le));
+
+	if (!nr_devices)
+		return "need at least one device";
+
+	if (!try_module_get(THIS_MODULE))
+		return "module unloading";
+
+	err = "cannot allocate memory";
+	sb = kcalloc(nr_devices, sizeof(*sb), GFP_KERNEL);
+	if (!sb)
+		goto err;
+
+	/*
+	 * read_super() needs to happen under register_lock, so that the
+	 * exclusive open is atomic with adding the new cache set to the list of
+	 * cache sets:
+	 */
+	mutex_lock(&bch_register_lock);
+
+	for (i = 0; i < nr_devices; i++) {
+		err = read_super(&sb[i], devices[i]);
+		if (err)
+			goto err_unlock;
+
+		err = "attempting to register backing device";
+		if (__SB_IS_BDEV(le64_to_cpu(sb[i].sb->version)))
+			goto err_unlock;
+
+		err = validate_cache_super(&sb[i]);
+		if (err)
+			goto err_unlock;
+	}
+
+	err = "cache set already registered";
+	if (cache_set_lookup(sb->sb->set_uuid))
+		goto err_unlock;
+
+	err = "cannot allocate memory";
+	c = bch_cache_set_alloc(sb[0].sb, opts);
+	if (!c)
+		goto err_unlock;
+
+	for (i = 0; i < nr_devices; i++) {
+		err = cache_alloc(&sb[i], c, NULL);
+		if (err)
+			goto err_unlock;
+	}
+
+	err = "insufficient devices";
+	if (cache_set_nr_online_devices(c) != cache_set_nr_devices(c))
+		goto err_unlock;
+
+	err = run_cache_set(c);
+	if (err)
+		goto err_unlock;
+
+	err = "error creating kobject";
+	if (bch_cache_set_online(c))
+		goto err_unlock;
+
+	if (ret) {
+		closure_get(&c->cl);
+		*ret = c;
+	}
+
+	mutex_unlock(&bch_register_lock);
+
+	err = NULL;
+out:
+	kfree(sb);
+	module_put(THIS_MODULE);
+	return err;
+err_unlock:
+	if (c)
+		bch_cache_set_stop(c);
+	mutex_unlock(&bch_register_lock);
+err:
+	for (i = 0; i < nr_devices; i++)
+		free_super(&sb[i]);
+	goto out;
+}
+
+const char *bch_register_one(const char *path)
+{
+	struct bcache_superblock sb;
+	const char *err;
+
+	mutex_lock(&bch_register_lock);
+
+	err = read_super(&sb, path);
+	if (err)
+		goto err;
+
+	if (__SB_IS_BDEV(le64_to_cpu(sb.sb->version)))
+		err = bch_backing_dev_register(&sb);
+	else
+		err = register_cache(&sb, cache_set_opts_empty());
+
+	free_super(&sb);
+err:
+	mutex_unlock(&bch_register_lock);
+	return err;
+}
+
+/* Global interfaces/init */
+
+#define kobj_attribute_write(n, fn)					\
+	static struct kobj_attribute ksysfs_##n = __ATTR(n, S_IWUSR, NULL, fn)
+
+#define kobj_attribute_rw(n, show, store)				\
+	static struct kobj_attribute ksysfs_##n =			\
+		__ATTR(n, S_IWUSR|S_IRUSR, show, store)
+
+static ssize_t register_bcache(struct kobject *, struct kobj_attribute *,
+			       const char *, size_t);
+
+kobj_attribute_write(register,		register_bcache);
+kobj_attribute_write(register_quiet,	register_bcache);
+
+static ssize_t register_bcache(struct kobject *k, struct kobj_attribute *attr,
+			       const char *buffer, size_t size)
+{
+	ssize_t ret = -EINVAL;
+	const char *err = "cannot allocate memory";
+	char *path = NULL;
+
+	if (!try_module_get(THIS_MODULE))
+		return -EBUSY;
+
+	if (!(path = kstrndup(skip_spaces(buffer), size, GFP_KERNEL)))
+		goto err;
+
+	err = bch_register_one(strim(path));
+	if (err)
+		goto err;
+
+	ret = size;
+out:
+	kfree(path);
+	module_put(THIS_MODULE);
+	return ret;
+err:
+	pr_err("error opening %s: %s", path, err);
+	goto out;
+}
+
+static int bcache_reboot(struct notifier_block *n, unsigned long code, void *x)
+{
+	if (code == SYS_DOWN ||
+	    code == SYS_HALT ||
+	    code == SYS_POWER_OFF) {
+		struct cache_set *c;
+
+		mutex_lock(&bch_register_lock);
+
+		if (!list_empty(&bch_cache_sets))
+			pr_info("Setting all devices read only:");
+
+		list_for_each_entry(c, &bch_cache_sets, list)
+			bch_cache_set_read_only(c);
+
+		list_for_each_entry(c, &bch_cache_sets, list)
+			bch_cache_set_read_only_sync(c);
+
+		mutex_unlock(&bch_register_lock);
+	}
+
+	return NOTIFY_DONE;
+}
+
+static struct notifier_block reboot = {
+	.notifier_call	= bcache_reboot,
+	.priority	= INT_MAX, /* before any real devices */
+};
+
+static ssize_t reboot_test(struct kobject *k, struct kobj_attribute *attr,
+			   const char *buffer, size_t size)
+{
+	bcache_reboot(NULL, SYS_DOWN, NULL);
+	return size;
+}
+
+kobj_attribute_write(reboot,		reboot_test);
+
+static void bcache_exit(void)
+{
+	bch_debug_exit();
+	bch_fs_exit();
+	bch_blockdev_exit();
+	if (bcache_kset)
+		kset_unregister(bcache_kset);
+	if (bcache_io_wq)
+		destroy_workqueue(bcache_io_wq);
+	if (!IS_ERR_OR_NULL(bch_chardev_class))
+		device_destroy(bch_chardev_class,
+			       MKDEV(bch_chardev_major, 0));
+	if (!IS_ERR_OR_NULL(bch_chardev_class))
+		class_destroy(bch_chardev_class);
+	if (bch_chardev_major > 0)
+		unregister_chrdev(bch_chardev_major, "bcache");
+	if (!IS_ERR_OR_NULL(bch_sha1))
+		crypto_free_shash(bch_sha1);
+	unregister_reboot_notifier(&reboot);
+}
+
+static int __init bcache_init(void)
+{
+	static const struct attribute *files[] = {
+		&ksysfs_register.attr,
+		&ksysfs_register_quiet.attr,
+		&ksysfs_reboot.attr,
+		NULL
+	};
+
+	mutex_init(&bch_register_lock);
+	register_reboot_notifier(&reboot);
+	closure_debug_init();
+	bkey_pack_test();
+
+	bch_sha1 = crypto_alloc_shash("sha1", 0, 0);
+	if (IS_ERR(bch_sha1))
+		goto err;
+
+	bch_chardev_major = register_chrdev(0, "bcache-ctl", &bch_chardev_fops);
+	if (bch_chardev_major < 0)
+		goto err;
+
+	bch_chardev_class = class_create(THIS_MODULE, "bcache");
+	if (IS_ERR(bch_chardev_class))
+		goto err;
+
+	bch_chardev = device_create(bch_chardev_class, NULL,
+				    MKDEV(bch_chardev_major, 255),
+				    NULL, "bcache-ctl");
+	if (IS_ERR(bch_chardev))
+		goto err;
+
+	if (!(bcache_io_wq = create_freezable_workqueue("bcache_io")) ||
+	    !(bcache_kset = kset_create_and_add("bcache", NULL, fs_kobj)) ||
+	    sysfs_create_files(&bcache_kset->kobj, files) ||
+	    bch_blockdev_init() ||
+	    bch_fs_init() ||
+	    bch_debug_init())
+		goto err;
+
+	return 0;
+err:
+	bcache_exit();
+	return -ENOMEM;
+}
+
+#define BCH_DEBUG_PARAM(name, description)			\
+	bool bch_##name;					\
+	module_param_named(name, bch_##name, bool, 0644);	\
+	MODULE_PARM_DESC(name, description);
+BCH_DEBUG_PARAMS()
+#undef BCH_DEBUG_PARAM
+
+module_exit(bcache_exit);
+module_init(bcache_init);