bcache: Add options for gc, root user reserves

5 files changed, 56 insertions, 39 deletions

diff --git a/drivers/md/bcache/alloc.c b/drivers/md/bcache/alloc.c
index 5bea5472a6be..0fedca0e2a0a 100644
--- a/drivers/md/bcache/alloc.c
+++ b/drivers/md/bcache/alloc.c
@@ -1555,7 +1555,7 @@ static void bch_recalc_capacity(struct cache_set *c)
 
 	total_capacity = capacity;
 
-	capacity *= (100 - c->sector_reserve_percent);
+	capacity *= (100 - c->opts.gc_reserve_percent);
 	capacity = div64_u64(capacity, 100);
 
 	BUG_ON(capacity + reserved_sectors > total_capacity);
diff --git a/drivers/md/bcache/bcache.h b/drivers/md/bcache/bcache.h
index 52f6ab20aee2..6903ab40ac6f 100644
--- a/drivers/md/bcache/bcache.h
+++ b/drivers/md/bcache/bcache.h
@@ -809,11 +809,6 @@ struct cache_set {
 	 * buckets is below this percentage
 	 */
 	unsigned		foreground_target_percent;
-	/*
-	 * foreground writes will fail when the number of free sectors is
-	 * below this percentage
-	 */
-	unsigned		sector_reserve_percent;
 
 #define BCH_DEBUG_PARAM(name, description) bool name;
 	BCH_DEBUG_PARAMS_ALL()
diff --git a/drivers/md/bcache/super.c b/drivers/md/bcache/super.c
index 99aff0b5879e..c9f2978a23bb 100644
--- a/drivers/md/bcache/super.c
+++ b/drivers/md/bcache/super.c
@@ -220,7 +220,7 @@ static const char *validate_cache_super(struct bcache_superblock *disk_sb)
 		return"Unsupported superblock version";
 	}
 
-	if (CACHE_SYNC(sb) &&
+	if (CACHE_SET_SYNC(sb) &&
 	    le64_to_cpu(sb->version) != BCACHE_SB_VERSION_CDEV_V3)
 		return "Unsupported superblock version";
 
@@ -265,15 +265,22 @@ static const char *validate_cache_super(struct bcache_superblock *disk_sb)
 	if (CACHE_SB_CSUM_TYPE(sb) >= BCH_CSUM_NR)
 		return "Invalid checksum type";
 
-	if (!CACHE_BTREE_NODE_SIZE(sb))
+	if (!CACHE_SET_BTREE_NODE_SIZE(sb))
 		return "Btree node size not set";
 
-	if (!is_power_of_2(CACHE_BTREE_NODE_SIZE(sb)))
+	if (!is_power_of_2(CACHE_SET_BTREE_NODE_SIZE(sb)))
 		return "Btree node size not a power of two";
 
-	if (CACHE_BTREE_NODE_SIZE(sb) > BTREE_NODE_SIZE_MAX)
+	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 (le16_to_cpu(sb->u64s) < bch_journal_buckets_offset(sb))
 		return "Invalid superblock: member info area missing";
 
@@ -560,7 +567,7 @@ static int cache_sb_to_cache_set(struct cache_set *c, struct cache_sb *src)
 	dst->block_size		= src->block_size;
 
 	c->sb.block_size	= le16_to_cpu(src->block_size);
-	c->sb.btree_node_size	= CACHE_BTREE_NODE_SIZE(src);
+	c->sb.btree_node_size	= CACHE_SET_BTREE_NODE_SIZE(src);
 
 	c->sb.nr_in_set		= src->nr_in_set;
 
@@ -1129,7 +1136,6 @@ static struct cache_set *bch_cache_set_alloc(struct cache_sb *sb,
 	c->tiering_percent = 10;
 
 	c->foreground_target_percent = 20;
-	c->sector_reserve_percent = 20;
 
 	c->journal.write_time	= &c->journal_write_time;
 	c->journal.delay_time	= &c->journal_delay_time;
@@ -1282,12 +1288,12 @@ static const char *run_cache_set(struct cache_set *c)
 		cache_sb_from_cache_set(c, ca);
 
 	/*
-	 * CACHE_SYNC is true if the cache set has already been run
+	 * 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_SYNC(&c->disk_sb)) {
+	if (CACHE_SET_SYNC(&c->disk_sb)) {
 		LIST_HEAD(journal);
 		struct jset *j;
 
@@ -1424,7 +1430,7 @@ static const char *run_cache_set(struct cache_set *c)
 			goto err;
 
 		/* Mark cache set as initialized: */
-		SET_CACHE_SYNC(&c->disk_sb, true);
+		SET_CACHE_SET_SYNC(&c->disk_sb, true);
 	}
 
 	if (c->opts.read_only) {
@@ -1477,7 +1483,7 @@ static const char *can_add_cache(struct cache_sb *sb,
 		return "mismatched block size";
 
 	if (le16_to_cpu(sb->members[sb->nr_this_dev].bucket_size) <
-	    CACHE_BTREE_NODE_SIZE(&c->disk_sb))
+	    CACHE_SET_BTREE_NODE_SIZE(&c->disk_sb))
 		return "new cache bucket_size is too small";
 
 	return NULL;
@@ -1879,10 +1885,6 @@ static const char *cache_alloc(struct bcache_superblock *sb,
 	if (c->sb.nr_in_set == 1)
 		bdevname(sb->bdev, c->name);
 
-	err = validate_cache_super(sb);
-	if (err)
-		return err;
-
 	if (cache_set_init_fault("cache_alloc"))
 		return err;
 
@@ -2023,6 +2025,10 @@ static const char *register_cache(struct bcache_superblock *sb,
 	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);
@@ -2234,6 +2240,10 @@ const char *bch_register_cache_set(char * const *devices, unsigned nr_devices,
 		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);
+		if (err)
+			return err;
 	}
 
 	err = "cache set already registered";
diff --git a/drivers/md/bcache/sysfs.c b/drivers/md/bcache/sysfs.c
index 61d01bfd66a6..596d43571641 100644
--- a/drivers/md/bcache/sysfs.c
+++ b/drivers/md/bcache/sysfs.c
@@ -153,7 +153,6 @@ rw_attribute(btree_flush_delay);
 rw_attribute(pd_controllers_update_seconds);
 
 rw_attribute(foreground_target_percent);
-rw_attribute(sector_reserve_percent);
 
 rw_attribute(size);
 read_attribute(meta_replicas_have);
@@ -647,10 +646,8 @@ SHOW(bch_cache_set)
 
 	sysfs_hprint(block_size,		block_bytes(c));
 	sysfs_print(block_size_bytes,		block_bytes(c));
-	sysfs_hprint(btree_node_size,
-		     CACHE_BTREE_NODE_SIZE(&c->disk_sb) << 9);
-	sysfs_print(btree_node_size_bytes,
-		    CACHE_BTREE_NODE_SIZE(&c->disk_sb) << 9);
+	sysfs_hprint(btree_node_size,		c->sb.btree_node_size << 9);
+	sysfs_print(btree_node_size_bytes,	c->sb.btree_node_size << 9);
 
 	sysfs_hprint(btree_cache_size,		bch_cache_size(c));
 	sysfs_print(cache_available_percent,	bch_cache_available_percent(c));
@@ -691,7 +688,6 @@ SHOW(bch_cache_set)
 	sysfs_print(pd_controllers_update_seconds,
 		    c->pd_controllers_update_seconds);
 	sysfs_print(foreground_target_percent, c->foreground_target_percent);
-	sysfs_print(sector_reserve_percent, c->sector_reserve_percent);
 
 	sysfs_printf(tiering_enabled,		"%i", c->tiering_enabled);
 	sysfs_print(tiering_percent,		c->tiering_percent);
@@ -808,7 +804,6 @@ STORE(__bch_cache_set)
 	sysfs_strtoul(pd_controllers_update_seconds,
 		      c->pd_controllers_update_seconds);
 	sysfs_strtoul(foreground_target_percent, c->foreground_target_percent);
-	sysfs_strtoul(sector_reserve_percent, c->sector_reserve_percent);
 
 	sysfs_strtoul(tiering_percent,		c->tiering_percent);
 	sysfs_pd_controller_store(tiering,	&c->tiering_pd);
@@ -903,7 +898,6 @@ static struct attribute *bch_cache_set_files[] = {
 
 	&sysfs_btree_flush_delay,
 	&sysfs_foreground_target_percent,
-	&sysfs_sector_reserve_percent,
 	&sysfs_tiering_percent,
 
 	&sysfs_journal_flush,
diff --git a/include/uapi/linux/bcache.h b/include/uapi/linux/bcache.h
index 0b04508c0a7b..efc4271f8428 100644
--- a/include/uapi/linux/bcache.h
+++ b/include/uapi/linux/bcache.h
@@ -705,7 +705,9 @@ struct cache_sb {
 	 * to change:
 	 */
 	uuid_le			user_uuid;
-	__le64			pad1[6];
+
+	__le64			flags2;
+	__le64			pad1[5];
 
 	/* Number of cache_member entries: */
 	__u8			nr_in_set;
@@ -732,9 +734,11 @@ struct cache_sb {
 	};
 };
 
-LE64_BITMASK(CACHE_SYNC,		struct cache_sb, flags, 0, 1);
+/* XXX: rename CACHE_SET -> BCH_FS or something? */
 
-LE64_BITMASK(CACHE_ERROR_ACTION,	struct cache_sb, flags, 1, 4);
+LE64_BITMASK(CACHE_SET_SYNC,		struct cache_sb, flags, 0, 1);
+
+LE64_BITMASK(CACHE_SET_ERROR_ACTION,	struct cache_sb, flags, 1, 4);
 #define BCH_ON_ERROR_CONTINUE		0U
 #define BCH_ON_ERROR_RO			1U
 #define BCH_ON_ERROR_PANIC		2U
@@ -747,13 +751,13 @@ LE64_BITMASK(CACHE_SET_DATA_REPLICAS_WANT,struct cache_sb, flags, 8, 12);
 
 LE64_BITMASK(CACHE_SB_CSUM_TYPE,	struct cache_sb, flags, 12, 16);
 
-LE64_BITMASK(CACHE_META_PREFERRED_CSUM_TYPE,struct cache_sb, flags, 16, 20);
+LE64_BITMASK(CACHE_SET_META_PREFERRED_CSUM_TYPE,struct cache_sb, flags, 16, 20);
 #define BCH_CSUM_NONE			0U
 #define BCH_CSUM_CRC32C			1U
 #define BCH_CSUM_CRC64			2U
 #define BCH_CSUM_NR			3U
 
-LE64_BITMASK(CACHE_BTREE_NODE_SIZE,	struct cache_sb, flags, 20, 36);
+LE64_BITMASK(CACHE_SET_BTREE_NODE_SIZE,	struct cache_sb, flags, 20, 36);
 
 LE64_BITMASK(CACHE_SET_META_REPLICAS_HAVE,struct cache_sb, flags, 36, 40);
 LE64_BITMASK(CACHE_SET_DATA_REPLICAS_HAVE,struct cache_sb, flags, 40, 44);
@@ -768,9 +772,9 @@ enum bch_str_hash_type {
 
 #define BCH_STR_HASH_NR			4
 
-LE64_BITMASK(CACHE_DATA_PREFERRED_CSUM_TYPE, struct cache_sb, flags, 48, 52);
+LE64_BITMASK(CACHE_SET_DATA_PREFERRED_CSUM_TYPE, struct cache_sb, flags, 48, 52);
 
-LE64_BITMASK(CACHE_COMPRESSION_TYPE,	struct cache_sb, flags, 52, 56);
+LE64_BITMASK(CACHE_SET_COMPRESSION_TYPE, struct cache_sb, flags, 52, 56);
 enum {
 	BCH_COMPRESSION_NONE		= 0,
 	BCH_COMPRESSION_LZ4		= 1,
@@ -782,6 +786,10 @@ enum {
 /* Limit inode numbers to 32 bits: */
 LE64_BITMASK(CACHE_INODE_32BIT,		struct cache_sb, flags, 56, 57);
 
+LE64_BITMASK(CACHE_SET_GC_RESERVE,	struct cache_sb, flags, 57, 63);
+
+LE64_BITMASK(CACHE_SET_ROOT_RESERVE,	struct cache_sb, flags2, 0,  6);
+
 /* options: */
 
 /**
@@ -807,7 +815,7 @@ LE64_BITMASK(CACHE_INODE_32BIT,		struct cache_sb, flags, 56, 57);
 	CACHE_SET_OPT(errors,					\
 		      bch_error_actions,			\
 		      0, BCH_NR_ERROR_ACTIONS,			\
-		      CACHE_ERROR_ACTION,			\
+		      CACHE_SET_ERROR_ACTION,			\
 		      true)					\
 	CACHE_SET_OPT(metadata_replicas,			\
 		      bch_uint_opt,				\
@@ -822,17 +830,17 @@ LE64_BITMASK(CACHE_INODE_32BIT,		struct cache_sb, flags, 56, 57);
 	CACHE_SET_OPT(metadata_checksum,			\
 		      bch_csum_types,				\
 		      0, BCH_CSUM_NR,				\
-		      CACHE_META_PREFERRED_CSUM_TYPE,		\
+		      CACHE_SET_META_PREFERRED_CSUM_TYPE,	\
 		      true)					\
 	CACHE_SET_OPT(data_checksum,				\
 		      bch_csum_types,				\
 		      0, BCH_CSUM_NR,				\
-		      CACHE_DATA_PREFERRED_CSUM_TYPE,		\
+		      CACHE_SET_DATA_PREFERRED_CSUM_TYPE,	\
 		      true)					\
 	CACHE_SET_OPT(compression,				\
 		      bch_compression_types,			\
 		      0, BCH_COMPRESSION_NR,			\
-		      CACHE_COMPRESSION_TYPE,			\
+		      CACHE_SET_COMPRESSION_TYPE,		\
 		      true)					\
 	CACHE_SET_OPT(str_hash,					\
 		      bch_str_hash_types,			\
@@ -843,6 +851,16 @@ LE64_BITMASK(CACHE_INODE_32BIT,		struct cache_sb, flags, 56, 57);
 		      bch_bool_opt, 0, 2,			\
 		      CACHE_INODE_32BIT,			\
 		      true)					\
+	CACHE_SET_OPT(gc_reserve_percent,			\
+		      bch_uint_opt,				\
+		      5, 21,					\
+		      CACHE_SET_GC_RESERVE,			\
+		      false)					\
+	CACHE_SET_OPT(root_reserve_percent,			\
+		      bch_uint_opt,				\
+		      0, 21,					\
+		      CACHE_SET_ROOT_RESERVE,			\
+		      false)
 
 /* backing device specific stuff: */