Btrfs: Add mount -o degraded to allow mounts to continue with missing devices

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

fs/btrfs/ctree.h

@@ -657,6 +657,7 @@ struct btrfs_root {
 #define BTRFS_MOUNT_NODATACOW		(1 << 1)
 #define BTRFS_MOUNT_NOBARRIER		(1 << 2)
 #define BTRFS_MOUNT_SSD			(1 << 3)
+#define BTRFS_MOUNT_DEGRADED		(1 << 4)
 
 #define btrfs_clear_opt(o, opt)		((o) &= ~BTRFS_MOUNT_##opt)
 #define btrfs_set_opt(o, opt)		((o) |= BTRFS_MOUNT_##opt)
@@ -1606,4 +1607,6 @@ int btrfs_delete_xattrs(struct btrfs_trans_handle *trans,
 			struct btrfs_root *root, struct inode *inode);
 /* super.c */
 u64 btrfs_parse_size(char *str);
+int btrfs_parse_options(char *options, struct btrfs_root *root,
+			char **subvol_name);
 #endif

fs/btrfs/disk-io.c

@@ -614,21 +614,6 @@ int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
 	return ret;
 }
 
-static int close_all_devices(struct btrfs_fs_info *fs_info)
-{
-	struct list_head *list;
-	struct list_head *next;
-	struct btrfs_device *device;
-
-	list = &fs_info->fs_devices->devices;
-	list_for_each(next, list) {
-		device = list_entry(next, struct btrfs_device, dev_list);
-		close_bdev_excl(device->bdev);
-		device->bdev = NULL;
-	}
-	return 0;
-}
-
 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
 					    u64 bytenr, u32 blocksize)
 {
@@ -927,6 +912,8 @@ static int btrfs_congested_fn(void *congested_data, int bdi_bits)
 
 	list_for_each(cur, &info->fs_devices->devices) {
 		device = list_entry(cur, struct btrfs_device, dev_list);
+		if (!device->bdev)
+			continue;
 		bdi = blk_get_backing_dev_info(device->bdev);
 		if (bdi && bdi_congested(bdi, bdi_bits)) {
 			ret = 1;
@@ -1140,7 +1127,8 @@ static void btrfs_async_submit_work(struct work_struct *work)
 }
 
 struct btrfs_root *open_ctree(struct super_block *sb,
-			      struct btrfs_fs_devices *fs_devices)
+			      struct btrfs_fs_devices *fs_devices,
+			      char *options)
 {
 	u32 sectorsize;
 	u32 nodesize;
@@ -1276,12 +1264,19 @@ struct btrfs_root *open_ctree(struct super_block *sb,
 	if (!btrfs_super_root(disk_super))
 		goto fail_sb_buffer;
 
-	if (btrfs_super_num_devices(disk_super) != fs_devices->num_devices) {
+	btrfs_parse_options(options, tree_root, NULL);
+
+	if (btrfs_super_num_devices(disk_super) > fs_devices->num_devices) {
 		printk("Btrfs: wanted %llu devices, but found %llu\n",
 		       (unsigned long long)btrfs_super_num_devices(disk_super),
 		       (unsigned long long)fs_devices->num_devices);
+		if (btrfs_test_opt(tree_root, DEGRADED))
+			printk("continuing in degraded mode\n");
+		else {
 			goto fail_sb_buffer;
 		}
+	}
+
 	fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
 
 	nodesize = btrfs_super_nodesize(disk_super);
@@ -1329,6 +1324,8 @@ struct btrfs_root *open_ctree(struct super_block *sb,
 	ret = btrfs_read_chunk_tree(chunk_root);
 	BUG_ON(ret);
 
+	btrfs_close_extra_devices(fs_devices);
+
 	blocksize = btrfs_level_size(tree_root,
 				     btrfs_super_root_level(disk_super));
 
@@ -1374,7 +1371,7 @@ struct btrfs_root *open_ctree(struct super_block *sb,
 fail_iput:
 	iput(fs_info->btree_inode);
 fail:
-	close_all_devices(fs_info);
+	btrfs_close_devices(fs_info->fs_devices);
 	btrfs_mapping_tree_free(&fs_info->mapping_tree);
 
 	kfree(extent_root);
@@ -1429,6 +1426,13 @@ int write_all_supers(struct btrfs_root *root)
 	dev_item = &sb->dev_item;
 	list_for_each(cur, head) {
 		dev = list_entry(cur, struct btrfs_device, dev_list);
+		if (!dev->bdev) {
+			total_errors++;
+			continue;
+		}
+		if (!dev->in_fs_metadata)
+			continue;
+
 		btrfs_set_stack_device_type(dev_item, dev->type);
 		btrfs_set_stack_device_id(dev_item, dev->devid);
 		btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
@@ -1482,6 +1486,11 @@ int write_all_supers(struct btrfs_root *root)
 
 	list_for_each(cur, head) {
 		dev = list_entry(cur, struct btrfs_device, dev_list);
+		if (!dev->bdev)
+			continue;
+		if (!dev->in_fs_metadata)
+			continue;
+
 		BUG_ON(!dev->pending_io);
 		bh = dev->pending_io;
 		wait_on_buffer(bh);
@@ -1631,7 +1640,7 @@ int close_ctree(struct btrfs_root *root)
 		kfree(hasher);
 	}
 #endif
-	close_all_devices(fs_info);
+	btrfs_close_devices(fs_info->fs_devices);
 	btrfs_mapping_tree_free(&fs_info->mapping_tree);
 
 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,23)

fs/btrfs/disk-io.h

@@ -33,7 +33,8 @@ struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
 int clean_tree_block(struct btrfs_trans_handle *trans,
 		     struct btrfs_root *root, struct extent_buffer *buf);
 struct btrfs_root *open_ctree(struct super_block *sb,
-			      struct btrfs_fs_devices *fs_devices);
+			      struct btrfs_fs_devices *fs_devices,
+			      char *options);
 int close_ctree(struct btrfs_root *root);
 int write_ctree_super(struct btrfs_trans_handle *trans,
 		      struct btrfs_root *root);

fs/btrfs/super.c

@@ -65,11 +65,13 @@ static void btrfs_put_super (struct super_block * sb)
 }
 
 enum {
-	Opt_subvol, Opt_nodatasum, Opt_nodatacow, Opt_max_extent,
-	Opt_max_inline, Opt_alloc_start, Opt_nobarrier, Opt_ssd, Opt_err,
+	Opt_degraded, Opt_subvol, Opt_nodatasum, Opt_nodatacow,
+	Opt_max_extent, Opt_max_inline, Opt_alloc_start, Opt_nobarrier,
+	Opt_ssd, Opt_err,
 };
 
 static match_table_t tokens = {
+	{Opt_degraded, "degraded"},
 	{Opt_subvol, "subvol=%s"},
 	{Opt_nodatasum, "nodatasum"},
 	{Opt_nodatacow, "nodatacow"},
@@ -106,8 +108,7 @@ u64 btrfs_parse_size(char *str)
 	return res;
 }
 
-static int parse_options (char * options,
-			  struct btrfs_root *root,
+int btrfs_parse_options(char *options, struct btrfs_root *root,
 			char **subvol_name)
 {
 	char * p;
@@ -135,6 +136,12 @@ static int parse_options (char * options,
 
 		token = match_token(p, tokens, args);
 		switch (token) {
+		case Opt_degraded:
+			if (info) {
+				printk("btrfs: allowing degraded mounts\n");
+				btrfs_set_opt(info->mount_opt, DEGRADED);
+			}
+			break;
 		case Opt_subvol:
 			if (subvol_name) {
 				*subvol_name = match_strdup(&args[0]);
@@ -234,7 +241,7 @@ static int btrfs_fill_super(struct super_block * sb,
 	sb->s_xattr = btrfs_xattr_handlers;
 	sb->s_time_gran = 1;
 
-	tree_root = open_ctree(sb, fs_devices);
+	tree_root = open_ctree(sb, fs_devices, (char *)data);
 
 	if (IS_ERR(tree_root)) {
 		printk("btrfs: open_ctree failed\n");
@@ -267,8 +274,6 @@ static int btrfs_fill_super(struct super_block * sb,
 		goto fail_close;
 	}
 
-	parse_options((char *)data, tree_root, NULL);
-
 	/* this does the super kobj at the same time */
 	err = btrfs_sysfs_add_super(tree_root->fs_info);
 	if (err)
@@ -341,7 +346,7 @@ int btrfs_get_sb_bdev(struct file_system_type *fs_type,
 	if (error)
 		return error;
 
-	bdev = fs_devices->lowest_bdev;
+	bdev = fs_devices->latest_bdev;
 	btrfs_lock_volumes();
 	s = sget(fs_type, btrfs_test_super, set_anon_super, fs_devices);
 	btrfs_unlock_volumes();
@@ -411,7 +416,7 @@ static int btrfs_get_sb(struct file_system_type *fs_type,
 	int ret;
 	char *subvol_name = NULL;
 
-	parse_options((char *)data, NULL, &subvol_name);
+	btrfs_parse_options((char *)data, NULL, &subvol_name);
 	ret = btrfs_get_sb_bdev(fs_type, flags, dev_name, data, mnt,
 			subvol_name ? subvol_name : "default");
 	if (subvol_name)

fs/btrfs/volumes.h

@@ -27,8 +27,10 @@ struct btrfs_device {
 	struct list_head dev_alloc_list;
 	struct btrfs_root *dev_root;
 	struct buffer_head *pending_io;
+	u64 generation;
 
 	int barriers;
+	int in_fs_metadata;
 
 	spinlock_t io_lock;
 
@@ -122,6 +124,7 @@ int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 int btrfs_scan_one_device(const char *path, int flags, void *holder,
 			  struct btrfs_fs_devices **fs_devices_ret);
 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices);
+int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices);
 int btrfs_add_device(struct btrfs_trans_handle *trans,
 		     struct btrfs_root *root,
 		     struct btrfs_device *device);