The Second Extended Filesystem (ext2fs) is a rewrite of the original Extended Filesystem and as such, is also based around the concept of "inodes." Ext2 served as the de facto filesystem of Linux for nearly a decade from the early 1990s to the early 2000s when it was superseded by the journaling file systems ext3 and ReiserFS. It has native support for UNIX ownership / access rights, symbolic- and hard-links, and other properties that are common among UNIX-like operating systems. Organizationally, it divides disk space up into groups called "block groups." Having these groups results in distribution of data across the disk which helps to minimize head movement as well as the impact of fragmentation. Further, some (if not all) groups are required to contain backups of important data that can be used to rebuild the file system in the event of disaster.
Note: Most of the information here is based off of work done by Dave Poirier on the ext2-doc project (see the links section) which is graciously released under the GNU Free Documentation License. Be sure to buy him a beer the next time you see him.
Important Note: All values are little-endian unless otherwise specified
What is a Block?
The Ext2 file system divides up disk space into logical blocks of contiguous space. The size of blocks need not be the same size as the sector size of the disk the file system resides on. The size of blocks can be determined by reading the field starting at byte 24 in the Superblock.
What is a Block Group?
Blocks, along with inodes, are divided up into "block groups." These are nothing more than contiguous groups of blocks.
Each block group reserves a few of its blocks for special purposes such as:
- A bitmap of free/allocated blocks within the group
- A bitmap of allocated inodes within the group
- A table of inode structures that belong to the group
- Depending upon the revision of Ext2 used, some or all block groups may also contain a backup copy of the Superblock and the Block Group Descriptor Table.
What is an Inode?
An inode is a structure on the disk that represents a file, directory, symbolic link, etc. Inodes do not contain the data of the file / directory / etc. that they represent. Instead, they link to the blocks that actually contain the data. This lets the inodes themselves have a well-defined size which lets them be placed in easily indexed arrays. Each block group has an array of inodes it is responsible for, and conversely every inode within a file system belongs to one of such tables (and one of such block groups).
The first step in implementing an Ext2 driver is to find, extract, and parse the superblock. The Superblock contains all information about the layout of the file system and possibly contains other important information like what optional features were used to create the file system. Once you have finished with the Superblock, the next step is to look at the Block Group Descriptor Table
Locating the Superblock
The Superblock is always located at byte 1024 from the beginning of the volume and is exactly 1024 bytes in length. For example, if the disk uses 512 byte sectors, the Superblock will begin at LBA 2 and will occupy all of sector 2 and 3.
Determining the Number of Block Groups
From the Superblock, extract the size of each block, the total number of inodes, the total number of blocks, the number of blocks per block group, and the number of inodes in each block group. From this information we can infer the number of block groups there are by:
- Rounding up the total number of blocks divided by the number of blocks per block group
- Rounding up the total number of inodes divided by the number of inodes per block group
- Both (and check them against each other)
Base Superblock Fields
These fields are present in all versions of Ext2
|0||3||4||Total number of inodes in file system|
|4||7||4||Total number of blocks in file system|
|8||11||4||Number of blocks reserved for superuser (see offset 80)|
|12||15||4||Total number of unallocated blocks|
|16||19||4||Total number of unallocated inodes|
|20||23||4||Block number of the block containing the superblock (also the starting block number, NOT always zero.)|
|24||27||4||log2 (block size) - 10. (In other words, the number to shift 1,024 to the left by to obtain the block size)|
|28||31||4||log2 (fragment size) - 10. (In other words, the number to shift 1,024 to the left by to obtain the fragment size)|
|32||35||4||Number of blocks in each block group|
|36||39||4||Number of fragments in each block group|
|40||43||4||Number of inodes in each block group|
|44||47||4||Last mount time (in POSIX time)|
|48||51||4||Last written time (in POSIX time)|
|52||53||2||Number of times the volume has been mounted since its last consistency check (fsck)|
|54||55||2||Number of mounts allowed before a consistency check (fsck) must be done|
|56||57||2||Ext2 signature (0xef53), used to help confirm the presence of Ext2 on a volume|
|58||59||2||File system state (see below)|
|60||61||2||What to do when an error is detected (see below)|
|62||63||2||Minor portion of version (combine with Major portion below to construct full version field)|
|64||67||4||POSIX time of last consistency check (fsck)|
|68||71||4||Interval (in POSIX time) between forced consistency checks (fsck)|
|72||75||4||Operating system ID from which the filesystem on this volume was created (see below)|
|76||79||4||Major portion of version (combine with Minor portion above to construct full version field)|
|80||81||2||User ID that can use reserved blocks|
|82||83||2||Group ID that can use reserved blocks|
File System States
|1||File system is clean|
|2||File system has errors|
Error Handling Methods
|Value||Action to Take|
|1||Ignore the error (continue on)|
|2||Remount file system as read-only|
Creator Operating System IDs
|2||MASIX (an operating system developed by Rémy Card, one of the developers of ext2)|
|4||Other "Lites" (BSD4.4-Lite derivatives such as NetBSD, OpenBSD, XNU/Darwin, etc.)|
Extended Superblock Fields
These fields are only present if Major version (specified in the base superblock fields), is greater than or equal to 1.
|84||87||4||First non-reserved inode in file system. (In versions < 1.0, this is fixed as 11)|
|88||89||2||Size of each inode structure in bytes. (In versions < 1.0, this is fixed as 128)|
|90||91||2||Block group that this superblock is part of (if backup copy)|
|92||95||4||Optional features present (features that are not required to read or write, but usually result in a performance increase. see below)|
|96||99||4||Required features present (features that are required to be supported to read or write. see below)|
|100||103||4||Features that if not supported, the volume must be mounted read-only see below)|
|104||119||16||File system ID (what is output by blkid)|
|120||135||16||Volume name (C-style string: characters terminated by a 0 byte)|
|136||199||64||Path volume was last mounted to (C-style string: characters terminated by a 0 byte)|
|200||203||4||Compression algorithms used (see Required features above)|
|204||204||1||Number of blocks to preallocate for files|
|205||205||1||Number of blocks to preallocate for directories|
|208||223||16||Journal ID (same style as the File system ID above)|
|232||235||4||Head of orphan inode list|
Optional Feature Flags
These are optional features for an implementation to support, but offer performance or reliability gains to implementations that do support them.
|0x0001||Preallocate some number of (contiguous?) blocks (see byte 205 in the superblock) to a directory when creating a new one (to reduce fragmentation?)|
|0x0002||AFS server inodes exist|
|0x0004||File system has a journal (Ext3)|
|0x0008||Inodes have extended attributes|
|0x0010||File system can resize itself for larger partitions|
|0x0020||Directories use hash index|
Required Feature Flags
These features if present on a file system are required to be supported by an implementation in order to correctly read from or write to the file system.
|0x0001||Compression is used|
|0x0002||Directory entries contain a type field|
|0x0004||File system needs to replay its journal|
|0x0008||File system uses a journal device|
Read-Only Feature Flags
These features, if present on a file system, are required in order for an implementation to write to the file system, but are not required to read from the file system.
|0x0001||Sparse superblocks and group descriptor tables|
|0x0002||File system uses a 64-bit file size|
|0x0004||Directory contents are stored in the form of a Binary Tree|
Block Group Descriptor Table
The Block Group Descriptor Table contains a descriptor for each block group within the file system. The number of block groups within the file system, and correspondingly, the number of entries in the Block Group Descriptor Table, is described above. Each descriptor contains information regarding where important data structures for that group are located.
Locating the Block Group Descriptor Table
The table is located in the block immediately following the Superblock. So if the block size (determined from a field in the superblock) is 1024 bytes per block, the Block Group Descriptor Table will begin at block 2. For any other block size, it will begin at block 1. Remember that blocks are numbered starting at 0, and that block numbers don't usually correspond to physical block addresses.
Block Group Descriptor
A Block Group Descriptor contains information regarding where important data structures for that block group are located.
|0||3||4||Block address of block usage bitmap|
|4||7||4||Block address of inode usage bitmap|
|8||11||4||Starting block address of inode table|
|12||13||2||Number of unallocated blocks in group|
|14||15||2||Number of unallocated inodes in group|
|16||17||2||Number of directories in group|
Like blocks, each inode has a numerical address. It is extremely important to note that unlike block addresses, inode addresses start at 1.
With Ext2 versions prior to Major version 1, inodes 1 to 10 are reserved and should be in an allocated state. Starting with version 1, the first non-reserved inode is indicated via a field in the Superblock. Of the reserved inodes, number 2 subjectively has the most significance as it is used for the root directory.
Inodes have a fixed size of either 128 for version 0 Ext2 file systems, or as dictated by the field in the Superblock for version 1 file systems. All inodes reside in inode tables that belong to block groups. Therefore, looking up an inode is simply a matter of determining which block group it belongs to and indexing that block group's inode table.
Determining which Block Group contains an Inode
From an inode address (remember that they start at 1), we can determine which group the inode is in, by using the formula:
block group = (inode – 1) / INODES_PER_GROUP
where INODES_PER_GROUP is a field in the Superblock
Finding an inode inside of a Block Group
Once we know which group an inode resides in, we can look up the actual inode by first retrieving that block group's inode table's starting address (see Block Group Descriptor above). The index of our inode in this block group's inode table can be determined by using the formula:
index = (inode – 1) % INODES_PER_GROUP
where % denotes the Modulo operation and INODES_PER_GROUP is a field in the Superblock (the same field which was used to determine which block group the inode belongs to).
Next, we have to determine which block contains our inode. This is achieved from:
containing block = (index * INODE_SIZE) / BLOCK_SIZE
where INODE_SIZE is either fixed at 128 if VERSION < 1 or defined by a field in the Superblock if VERSION >= 1.0, and BLOCK_SIZE is defined by a field in the Superblock.
Finally, mask and shift as necessary to extract only the inode data from the containing block.
Reading the contents of an inode
Each inode contains 12 direct pointers, one singly indirect pointer, one doubly indirect block pointer, and one triply indirect pointer. The direct space "overflows" into the singly indirect space, which overflows into the doubly indirect space, which overflows into the triply indirect space.
Direct Block Pointers: There are 12 direct block pointers. If valid, the value is non-zero. Each pointer is the block address of a block containing data for this inode.
Singly Indirect Block Pointer: If a file needs more than 12 blocks, a separate block is allocated to store the block addresses of the remaining data blocks needed to store its contents. This separate block is called an indirect block because it adds an extra step (a level of indirection) between an inode and its data. The block addresses stored in the block are all 32-bit, and the capacity of stored addresses in this block is a function of the block size. The address of this indirect block is stored in the inode in the "Singly Indirect Block Pointer" field.
Doubly Indirect Block Pointer: If a file has more blocks than can fit in the 12 direct pointers and the indirect block, a double indirect block is used. A double indirect block is an extension of the indirect block described above only now we have two intermediate blocks between the inode and data blocks. The inode structure has a "Doubly Indirect Block Pointer" field that points to this block if necessary.
Triply Indirect Block Pointer: Lastly, if a file needs still more space, it can use a triple indirect block. Again, this is an extension of the double indirect block. So, a triple indirect block contains addresses of double indirect blocks, which contain addresses of single indirect blocks, which contain address of data blocks. The inode structure has a "Triply Indirect Block Pointer" field that points to this block if present.
This image from Wikipedia illustrates what is described above pretty well.
Inode Data Structure
|0||1||2||Type and Permissions (see below)|
|4||7||4||Lower 32 bits of size in bytes|
|8||11||4||Last Access Time (in POSIX time)|
|12||15||4||Creation Time (in POSIX time)|
|16||19||4||Last Modification time (in POSIX time)|
|20||23||4||Deletion time (in POSIX time)|
|26||27||2||Count of hard links (directory entries) to this inode. When this reaches 0, the data blocks are marked as unallocated.|
|28||31||4||Count of disk sectors (not Ext2 blocks) in use by this inode, not counting the actual inode structure nor directory entries linking to the inode.|
|32||35||4||Flags (see below)|
|36||39||4||Operating System Specific value #1|
|40||43||4||Direct Block Pointer 0|
|44||47||4||Direct Block Pointer 1|
|48||51||4||Direct Block Pointer 2|
|52||55||4||Direct Block Pointer 3|
|56||59||4||Direct Block Pointer 4|
|60||63||4||Direct Block Pointer 5|
|64||67||4||Direct Block Pointer 6|
|68||71||4||Direct Block Pointer 7|
|72||75||4||Direct Block Pointer 8|
|76||79||4||Direct Block Pointer 9|
|80||83||4||Direct Block Pointer 10|
|84||87||4||Direct Block Pointer 11|
|88||91||4||Singly Indirect Block Pointer (Points to a block that is a list of block pointers to data)|
|92||95||4||Doubly Indirect Block Pointer (Points to a block that is a list of block pointers to Singly Indirect Blocks)|
|96||99||4||Triply Indirect Block Pointer (Points to a block that is a list of block pointers to Doubly Indirect Blocks)|
|100||103||4||Generation number (Primarily used for NFS)|
|104||107||4||In Ext2 version 0, this field is reserved. In version >= 1, Extended attribute block (File ACL).|
|108||111||4||In Ext2 version 0, this field is reserved. In version >= 1, Upper 32 bits of file size (if feature bit set) if it's a file, Directory ACL if it's a directory|
|112||115||4||Block address of fragment|
|116||127||12||Operating System Specific Value #2|
Inode Type and Permissions
| Type value
value in hex
value in octal
|0x400||02000||Set group ID|
|0x800||04000||Set user ID|
|0x00000001||Secure deletion (not used)|
|0x00000002||Keep a copy of data when deleted (not used)|
|0x00000004||File compression (not used)|
|0x00000008||Synchronous updates—new data is written immediately to disk|
|0x00000010||Immutable file (content cannot be changed)|
|0x00000040||File is not included in 'dump' command|
|0x00000080||Last accessed time should not updated|
|0x00010000||Hash indexed directory|
|0x00040000||Journal file data|
OS Specific Value 1
|How they use this field|
OS Specific Value 2
|How they use this field|
Directories are inodes which contain some number of "entries" as their contents. These entries are nothing more than a name/inode pair. For instance the inode corresponding to the root directory might have an entry with the name of "etc" and an inode value of 50. A directory inode stores these entries in a linked-list fashion in its contents blocks.
The root directory is Inode 2.
The total size of a directory entry may be longer then the length of the name would imply (The name may not span to the end of the record), and records have to be aligned to 4-byte boundaries. Directory entries are also not allowed to span multiple blocks on the file-system, so there may be empty space in-between directory entries. Empty space is however not allowed in-between directory entries, so any possible empty space will be used as part of the preceding record by increasing its record length to include the empty space. Empty space may also be equivalently marked by a separate directory entry with an inode number of zero, indicating that directory entry should be skipped.
|4||5||2||Total size of this entry (Including all subfields)|
|6||6||1||Name Length least-significant 8 bits|
|7||7||1||Type indicator (only if the feature bit for "directory entries have file type byte" is set, else this is the most-significant 8 bits of the Name Length)|
Directory Entry Type Indicators
|7||Symbolic link (soft link)|
How To Read An Inode
- Read the Superblock to find the size of each block, the number of blocks per group, number Inodes per group, and the starting block of the first group (Block Group Descriptor Table).
- Determine which block group the inode belongs to.
- Read the Block Group Descriptor corresponding to the Block Group which contains the inode to be looked up.
- From the Block Group Descriptor, extract the location of the block group's inode table.
- Determine the index of the inode in the inode table.
- Index the inode table (taking into account non-standard inode size).
Directory entry information and file contents are located within the data blocks that the Inode points to.
How To Read the Root Directory
The root directory's inode is defined to always be 2. Read/parse the contents of inode 2.
- ext2-doc project: Second Extended File System - implementation-oriented documentation, describes internal structure in human language.
- Design and Implementation of the Second Extended Filesystem (overview)
- State of the Art: Where we are with the Ext3 filesystem - Paper by Mingming Cao, Theodore Y. Ts'o, Badari Pulavarty, and Suparna Bhattacharya describing extended features for ext2