Btrfs: finish off inode indexing in dirs, add overflows

Signed-off-by: Chris Mason <chris.mason@oracle.com>

fs/btrfs/ctree.h

@@ -270,6 +270,10 @@ struct btrfs_root {
 #define BTRFS_KEY_TYPE_MAX	256
 #define BTRFS_KEY_TYPE_MASK	(BTRFS_KEY_TYPE_MAX - 1)
 
+#define BTRFS_KEY_OVERFLOW_MAX 128
+#define BTRFS_KEY_OVERFLOW_SHIFT 8
+#define BTRFS_KEY_OVERFLOW_MASK (0x7FULL << BTRFS_KEY_OVERFLOW_SHIFT)
+
 /*
  * inode items have the data typically returned from stat and store other
  * info about object characteristics.  There is one for every file and dir in
@@ -588,6 +592,19 @@ static inline void btrfs_set_disk_key_flags(struct btrfs_disk_key *disk,
 	disk->flags = cpu_to_le32(val);
 }
 
+static inline u32 btrfs_key_overflow(struct btrfs_key *key)
+{
+	u32 over = key->flags & BTRFS_KEY_OVERFLOW_MASK;
+	return over >> BTRFS_KEY_OVERFLOW_SHIFT;
+}
+
+static inline void btrfs_set_key_overflow(struct btrfs_key *key, u32 over)
+{
+	BUG_ON(over > BTRFS_KEY_OVERFLOW_MAX);
+	over = over << BTRFS_KEY_OVERFLOW_SHIFT;
+	key->flags = (key->flags & ~((u64)BTRFS_KEY_OVERFLOW_MASK)) | over;
+}
+
 static inline u32 btrfs_key_type(struct btrfs_key *key)
 {
 	return key->flags & BTRFS_KEY_TYPE_MASK;
@@ -612,6 +629,22 @@ static inline void btrfs_set_disk_key_type(struct btrfs_disk_key *key, u32 type)
 	btrfs_set_disk_key_flags(key, flags);
 }
 
+static inline u32 btrfs_disk_key_overflow(struct btrfs_disk_key *key)
+{
+	u32 over = le32_to_cpu(key->flags) & BTRFS_KEY_OVERFLOW_MASK;
+	return over >> BTRFS_KEY_OVERFLOW_SHIFT;
+}
+
+static inline void btrfs_set_disK_key_overflow(struct btrfs_disk_key *key,
+					       u32 over)
+{
+	u32 flags = btrfs_disk_key_flags(key);
+	BUG_ON(over > BTRFS_KEY_OVERFLOW_MAX);
+	over = over << BTRFS_KEY_OVERFLOW_SHIFT;
+	flags = (flags & ~((u64)BTRFS_KEY_OVERFLOW_MASK)) | over;
+	btrfs_set_disk_key_flags(key, flags);
+}
+
 static inline u64 btrfs_header_blocknr(struct btrfs_header *h)
 {
 	return le64_to_cpu(h->blocknr);

fs/btrfs/dir-item.c

@@ -4,6 +4,26 @@
 #include "hash.h"
 #include "transaction.h"
 
+int insert_with_overflow(struct btrfs_trans_handle *trans, struct btrfs_root
+			    *root, struct btrfs_path *path, struct btrfs_key
+			    *cpu_key, u32 data_size)
+{
+	int overflow;
+	int ret;
+
+	ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
+	overflow = btrfs_key_overflow(cpu_key);
+
+	while(ret == -EEXIST && overflow < BTRFS_KEY_OVERFLOW_MAX) {
+		overflow++;
+		btrfs_set_key_overflow(cpu_key, overflow);
+		btrfs_release_path(root, path);
+		ret = btrfs_insert_empty_item(trans, root, path, cpu_key,
+					      data_size);
+	}
+	return ret;
+}
+
 int btrfs_insert_dir_item(struct btrfs_trans_handle *trans, struct btrfs_root
 			  *root, const char *name, int name_len, u64 dir, u64
 			  objectid, u8 type)
@@ -23,7 +43,7 @@ int btrfs_insert_dir_item(struct btrfs_trans_handle *trans, struct btrfs_root
 	path = btrfs_alloc_path();
 	btrfs_init_path(path);
 	data_size = sizeof(*dir_item) + name_len;
-	ret = btrfs_insert_empty_item(trans, root, path, &key, data_size);
+	ret = insert_with_overflow(trans, root, path, &key, data_size);
 	if (ret)
 		goto out;
 
@@ -41,7 +61,7 @@ int btrfs_insert_dir_item(struct btrfs_trans_handle *trans, struct btrfs_root
 
 	btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
 	key.offset = objectid;
-	ret = btrfs_insert_empty_item(trans, root, path, &key, data_size);
+	ret = insert_with_overflow(trans, root, path, &key, data_size);
 	// FIXME clear the dirindex bit
 	if (ret)
 		goto out;
@@ -70,14 +90,40 @@ int btrfs_lookup_dir_item(struct btrfs_trans_handle *trans, struct btrfs_root
 	struct btrfs_key key;
 	int ins_len = mod < 0 ? -1 : 0;
 	int cow = mod != 0;
+	struct btrfs_disk_key *found_key;
+	struct btrfs_leaf *leaf;
 
 	key.objectid = dir;
 	key.flags = 0;
 	btrfs_set_key_type(&key, BTRFS_DIR_ITEM_KEY);
+	btrfs_set_key_overflow(&key, BTRFS_KEY_OVERFLOW_MAX - 1);
 	ret = btrfs_name_hash(name, name_len, &key.offset);
 	BUG_ON(ret);
+	while(1) {
 		ret = btrfs_search_slot(trans, root, &key, path, ins_len, cow);
+		if (ret < 0)
 			return ret;
+		if (ret > 0) {
+			if (path->slots[0] == 0)
+				return 1;
+			path->slots[0]--;
+		}
+		leaf = btrfs_buffer_leaf(path->nodes[0]);
+		found_key = &leaf->items[path->slots[0]].key;
+
+		if (btrfs_disk_key_objectid(found_key) != dir ||
+		    btrfs_disk_key_type(found_key) != BTRFS_DIR_ITEM_KEY ||
+		    btrfs_disk_key_offset(found_key) != key.offset)
+			return 1;
+
+		if (btrfs_match_dir_item_name(root, path, name, name_len))
+			return 0;
+
+		if (btrfs_disk_key_overflow(found_key) == 0)
+			return 1;
+		btrfs_release_path(root, path);
+	}
+	return 1;
 }
 
 int btrfs_lookup_dir_index_item(struct btrfs_trans_handle *trans,
@@ -89,13 +135,31 @@ int btrfs_lookup_dir_index_item(struct btrfs_trans_handle *trans,
 	struct btrfs_key key;
 	int ins_len = mod < 0 ? -1 : 0;
 	int cow = mod != 0;
+	struct btrfs_disk_key *found_key;
+	struct btrfs_leaf *leaf;
 
 	key.objectid = dir;
 	key.flags = 0;
 	btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
+	btrfs_set_key_overflow(&key, BTRFS_KEY_OVERFLOW_MAX - 1);
 	key.offset = objectid;
 	ret = btrfs_search_slot(trans, root, &key, path, ins_len, cow);
+	if (ret < 0)
 		return ret;
+	if (ret > 0) {
+		if (path->slots[0] == 0)
+			return 1;
+		path->slots[0]--;
+	}
+	leaf = btrfs_buffer_leaf(path->nodes[0]);
+	found_key = &leaf->items[path->slots[0]].key;
+
+	if (btrfs_disk_key_objectid(found_key) != dir ||
+	    btrfs_disk_key_type(found_key) != BTRFS_DIR_INDEX_KEY)
+		return 1;
+	if (btrfs_disk_key_offset(found_key) == objectid)
+		return 0;
+	return 1;
 }
 
 int btrfs_match_dir_item_name(struct btrfs_root *root,

fs/btrfs/super.c

@@ -486,19 +486,18 @@ static int btrfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
 			continue;
 		if (btrfs_disk_key_offset(&item->key) < filp->f_pos)
 			continue;
-
+		filp->f_pos = btrfs_disk_key_offset(&item->key);
 		advance = 1;
 		di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
 		over = filldir(dirent, (const char *)(di + 1),
 			       btrfs_dir_name_len(di),
 			       btrfs_disk_key_offset(&item->key),
 			       btrfs_dir_objectid(di), d_type);
-		if (over) {
-			filp->f_pos = btrfs_disk_key_offset(&item->key);
-			break;
-		}
-		filp->f_pos = btrfs_disk_key_offset(&item->key) + 1;
+		if (over)
+			goto nopos;
 	}
+	filp->f_pos++;
+nopos:
 	ret = 0;
 err:
 	btrfs_release_path(root, path);