Modified variant of the file_system found in xv6, which was designed after the original file system of UNIX version 6.
The disk is organized in blocks of 1024 bytes.
| Boot | Super Block | Log Area | Inode Area | BMap Area | Data Blocks | |
|---|---|---|---|---|---|---|
| Blocks (example) | 1 | 1 | 30 | 13 | 1 | 2002 |
Boot: The first block is reserved for the boot loader for historic reasons. This block is currently not used.
Super Block:
This block contains the struct xv6fs_superblock which describes the file_system, including the position and size of the following areas. Everything in here stays static after the file system got created.
Log Area:
The first block contains the log header struct logheader. The following LOGSIZE blocks are used to store the data for blocks to be committed from the log. For each of the LOGSIZE blocks the header stores the destination block address.
See xv6fs_log.
Inode Area:
The inodes are stored sequentially on disk in an array as struct xv6fs_dinode (disk inode). The inode number is the index into this array. The amount of blocks reserved for this inode array defines the maximum of supported inodes and thus files and directories.
Inode 1 is by definition the root directory of the file system.
Directories are stored as files containing an array of struct xv6fs_dirent.
Files are stored in blocks in the data blocks area. The disk inode contains an array of block addresses for the files content. An additional block can be allocated for block addresses of larger files.
BMap Area: Stores one bit per block as a use/free flag. The size of this area is defined by the size of the file system. Only blocks from the data blocks can be free, all blocks containing file system meta data (including this bitmap) are indicated as used.
Data Blocks: Unstructured data blocks with the contents of the files and directories. Size of this area defined the usable disk size of the file system.
Max files:
File and directory count are limited by the inode count. This is defined at file system creation in mkfs (see MAX_ACTIVE_INODESS). As directories references inodes with a 16-bit unsigned integer, the maximal number of inodes in xv6fs is 65,535 (0 is a reserved value).
Max files per directory;
A struct xv6fs_dirent is 16 bytes long, thus 64 directory entries can be stored in one block. A file can be up to 12+256 blocks large. This limits the number of directory entries per directory to 17152 entries. Each dir contains . and .., so 17150 files or sub directories can be stored in one xv6fs directory.
Max file size:
Currently a disk inode can reference 12 data blocks. One additional block can be used as a second array of 1024/sizeof(uint32_t) = 256 block addresses. The maximal file size supported is thus (12+256)*1024 = 268 kb.
Max file system size: Blocks are indexed by 32-bit unsigned integers. Thus file systems of up to 4 GB should be possible. xv6fs supports a maximum of 65,535 files with 268 kb each = 16 GB in total. So filling up such a huge disk is possible.
Max file name:
Just 14 chars, limited by XV6_NAME_MAX and indirectly by the size of one struct xv6fs_dirent.
- Refactored and renamed some defines, moved code to separate generic and xv6fs specific code.
- Moved out non-file system specific function calls (e.g. file system tree traversal calls) from xv6fs_log specific calls. This allows a virtual file system abstraction and mounting. It also changes the order of lock acquire/releases compared to xv6.
- Split device disk inodes into char and block devices to correctly store different types of devices.
Video about the original xv6 file system: https://www.youtube.com/watch?v=o9sYiWj1F28&ab_channel=hhp3
Overview: kernel | file_system
File System: init_filesystem | vfs | xv6fs | devfs | block_io | inode | file | directory