Difference between revisions of "Ext4"

From ArchWiki
Jump to: navigation, search
(enable metadata checksums)
(kernel modules for csums)
Line 215: Line 215:
Then the file-system needs to be converted to 64bit: {{Ic |resize2fs -b /dev/sdX}}.   
Then the file-system needs to be converted to 64bit: {{Ic |resize2fs -b /dev/sdX}}.   
Finally checksums can be added: {{Ic |tune2fs -O metadata_csum /dev/sdX}}.
Finally checksums can be added: {{Ic |tune2fs -O metadata_csum /dev/sdX}}.
The file-system can then be mounted as usual.
If this is the root file-system these modules might need to be added to {{Ic|/etc/mkinitcpio.conf}}:
    MODULES="... crc32c_intel crc32c_generic"
And then regenerate the kernel image using {{Ic|mkinitcpio -p linux}}.
== See also ==
== See also ==

Revision as of 21:31, 24 July 2016


From Ext4 - Linux Kernel Newbies:

Ext4 is the evolution of the most used Linux filesystem, Ext3. In many ways, Ext4 is a deeper improvement over Ext3 than Ext3 was over Ext2. Ext3 was mostly about adding journaling to Ext2, but Ext4 modifies important data structures of the filesystem such as the ones destined to store the file data. The result is a filesystem with an improved design, better performance, reliability, and features.

Create a new ext4 filesystem

To format a partition do:

# mkfs.ext4 /dev/partition
Tip: See the mkfs.ext4 man page for more options; edit /etc/mke2fs.conf to view/configure default options.

Bytes-per-inode ratio

From man mkfs.ext4:

mke2fs creates an inode for every bytes-per-inode bytes of space on the disk. The larger the bytes-per-inode ratio, the fewer inodes will be created.

Creating a new file, directory, symlink etc. requires at least one free inode. If the inode count is too low, no file can be created on the filesystem even though there is still space left on it.

Because it is not possible to change either the bytes-per-inode ratio or the inode count after the filesystem is created, mkfs.ext4 uses by default a rather low ratio of one inode every 16384 bytes (16 Kb) to avoid this situation.

However, for partitions with size in the hundreds or thousands of GB and average file size in the megabyte range, this usually results in a much too large inode number because the number of files created never reaches the number of inodes.

This results in a waste of disk space, because all those unused inodes each take up 256 bytes on the filesystem (this is also set in /etc/mke2fs.conf but should not be changed). 256 * several millions = quite a few gigabytes wasted in unused inodes.

This situation can be evaluated by comparing the {I}Use% figures provided by df and df -i:

$ df -h /home
Filesystem              Size    Used   Avail  Use%   Mounted on
/dev/mapper/lvm-home    115G    56G    59G    49%    /home
$ df -hi /home
Filesystem              Inodes  IUsed  IFree  IUse%  Mounted on
/dev/mapper/lvm-home    1.8M    1.1K   1.8M   1%     /home

To specify a different bytes-per-inode ratio, you can use the -T usage-type option which hints at the expected usage of the filesystem using types defined in /etc/mke2fs.conf. Among those types are the bigger largefile and largefile4 which offer more relevant ratios of one inode every 1 MiB and 4 MiB respectively. It can be used as such:

# mkfs.ext4 -T largefile /dev/device

The bytes-per-inode ratio can also be set directly via the -i option: e.g. use -i 2097152 for a 2 MiB ratio and -i 6291456 for a 6 MiB ratio.

Tip: Conversely, if you are setting up a partition dedicated to host millions of small files like emails or newsgroup items, you can use smaller usage-type values such as news (one inode for every 4096 bytes) or small (same plus smaller inode and block sizes).
Warning: If you make a heavy use of symbolic links, make sure to keep the inode count high enough with a low bytes-per-inode ratio, because while not taking more space every new symbolic link consumes one new inode and therefore the filesystem may run out of them quickly.

Reserved blocks

By default, 5% of the filesystem blocks will be reserved for the super-user, to avoid fragmentation and "allow root-owned daemons to continue to function correctly after non-privileged processes are prevented from writing to the filesystem" (from man mkfs.ext4).

For modern high-capacity disks, this is higher than necessary if the partition is used as a long-term archive or not crucial to system operations (like /home). See this email for the opinion of ext4 developer Ted Ts'o on reserved blocks.

It is generally safe to reduce the percentage of reserved blocks to free up disk space when the partition is either:

  • Very large (for example > 50G)
  • Used as long-term archive, i.e., where files will not be deleted and created very often

The -m option of ext4-related utilities allows to specify the percentage of reserved blocks.

To totally prevent reserving blocks upon filesystem creation, use:

# mkfs.ext4 -m 0 /dev/device

To reduce it to 1% afterwards, use:

# tune2fs -m 1 /dev/device

You can use findmnt(8) to find the device name:

$ findmnt /the/mount/point

Migrating from ext2/ext3 to ext4

Mounting ext2/ext3 partitions as ext4 without converting


A compromise between fully converting to ext4 and simply remaining with ext2/ext3 is to mount the partitions as ext4.


  • Compatibility (the filesystem can continue to be mounted as ext3) – This allows users to still read the filesystem from other operating systems without ext4 support (e.g. Windows with ext2/ext3 drivers)
  • Improved performance (though not as much as a fully-converted ext4 partition).[1] [2]


  • Fewer features of ext4 are used (only those that do not change the disk format such as multiblock allocation and delayed allocation)
Note: Except for the relative novelty of ext4 (which can be seen as a risk), there is no major drawback to this technique.


  1. Edit /etc/fstab and change the 'type' from ext2/ext3 to ext4 for any partitions you would like to mount as ext4.
  2. Re-mount the affected partitions.

Converting ext2/ext3 partitions to ext4


To experience the benefits of ext4, an irreversible conversion process must be completed.


  • Improved performance and new features.[3] [4]


  • Partitions that contain mostly static files, such as a /boot partition, may not benefit from the new features. Also, adding a journal (which is implied by moving a ext2 partition to ext3/4) always incurs performance overhead.
  • Irreversible (ext4 partitions cannot be 'downgraded' to ext2/ext3. It is, however, backwards compatible until extent and other unique options are enabled)


These instructions were adapted from Kernel documentation and an BBS thread.

  • If you convert the system's root filesystem, ensure that the 'fallback' initramfs is available at reboot. Alternatively, add ext4 according to Mkinitcpio#MODULES and re-create the 'default' initial ramdisk with mkinitcpio -p linux before starting.
  • If you decide to convert a separate /boot partition, ensure the bootloader supports booting from ext4.

In the following steps /dev/sdxX denotes the path to the partition to be converted, such as /dev/sda1.

  1. BACK-UP! Back-up all data on any ext3 partitions that are to be converted to ext4. A useful package, especially for root partitions, is Clonezilla.
  2. Edit /etc/fstab and change the 'type' from ext3 to ext4 for any partitions that are to be converted to ext4.
  3. Boot the live medium (if necessary). The conversion process with e2fsprogs must be done when the drive is not mounted. If converting a root partition, the simplest way to achieve this is to boot from some other live medium.
  4. Ensure the partition is NOT mounted
  5. If you want to convert a ext2 partition, the first conversion step is to add a journal by running tune2fs -j /dev/sdxX as root; making it a ext3 partition.
  6. Run tune2fs -O extent,uninit_bg,dir_index /dev/sdxX as root. This command converts the ext3 filesystem to ext4 (irreversibly).
  7. Run fsck -f /dev/sdxX as root.
    • The user must fsck the filesystem, or it will be unreadable! This fsck run is needed to return the filesystem to a consistent state. It will find checksum errors in the group descriptors - this is expected. The -f option asks fsck to force checking even if the file system seems clean. The -p option may be used on top to 'automatically repair' (otherwise, the user will be asked for input for each error).
  8. Recommended: mount the partition and run e4defrag -c -v /dev/sdxX as root.
    • Even though the filesystem is now converted to ext4, all files that have been written before the conversion do not yet take advantage of the extent option of ext4, which will improve large file performance and reduce fragmentation and filesystem check time. In order to fully take advantage of ext4, all files would have to be rewritten on disk. Use e4defrag to take care of this problem.
  9. Reboot Arch Linux!

Using ext4 per directory encryption

Linux comes with an Ext4 feature to encrypt directories of a filesystem. See also Quarkslab's blog entry with a write-up of the features, an overview of the implementation state and practical test results with kernel 4.1.

Encryption keys are stored in the keyring. To get started, make sure you have enabled CONFIG_KEYS and CONFIG_EXT4_ENCRYPTION kernel options and you have kernel 4.1 or higher. Note the Arch default linux does not have CONFIG_EXT4_ENCRYPTION set yet.

First of all, you need to update e2fsprogs to at least version 1.43.

Let us make a directory that we will encrypt. Encryption policy can be set only on new empty directories. For example, if we are to encrypt /encrypted/dir, create the upper level directory:

# mkdir /encrypted

First generate a random salt value and store it in a safe place:

# head -c 16 /dev/random | xxd -p

Now generate and add a new key into your keyring: this step should be repeated every time you flush your keychain (reboot)

# e4crypt add_key -S 0x877282f53bd0adbbef92142fc4cac459
Enter passphrase (echo disabled): 
Added key with descriptor [f88747555a6115f5]

Now you know a descriptor for your key. Make sure you have added a key into your keychain:

# keyctl show
Session Keyring
1021618178 --alswrv   1000  1000  keyring: _ses
 176349519 --alsw-v   1000  1000   \_ logon: ext4:f88747555a6115f5

Almost done. Now set an encryption policy for a directory:

# e4crypt set_policy f88747555a6115f5 /encrypted/dir

That is all. If you try accessing the disk without adding a key into keychain, filenames and their contents will be seen as encrypted gibberish. Be careful running old versions of e2fsck on your filesystem - it will treat encrypted filenames as invalid.

Tips and tricks


E4rat is a preload application designed for the ext4 filesystem. It monitors files opened during boot, optimizes their placement on the partition to improve access time, and preloads them at the very beginning of the boot process. E4rat does not offer improvements with SSDs, whose access time is negligible compared to hard disks.

Barriers and performance

Since kernel 2.6.30, ext4 performance has decreased due to changes that serve to improve data integrity.[5]

Most file systems (XFS, ext3, ext4, reiserfs) send write barriers to disk after fsync or during transaction commits. Write barriers enforce proper ordering of writes, making volatile disk write caches safe to use (at some performance penalty). If your disks are battery-backed in one way or another, disabling barriers may safely improve performance.
Sending write barriers can be disabled using the barrier=0 mount option (for ext3, ext4, and reiserfs), or using the nobarrier mount option (for XFS).[6][dead link 2016-7-5]
Warning: Disabling barriers when disks cannot guarantee caches are properly written in case of power failure can lead to severe file system corruption and data loss.

To turn barriers off add the option barrier=0 to the desired filesystem. For example:

/dev/sda5    /    ext4    noatime,barrier=0    0    1

Enabling metadata checksums

Starting with e2fsprogs 1.43 we can safely enable support for ext4 checksums and 64bit. More about metadata checksums can be read here.

If your CPU supports SSE 4.2, make sure the crc32c_intel kernel module is loaded in order to enable the hardware accelerated CRC32C algorithm. If not you'll need to load the crc32c_generic module.

This needs to be done with the partition unmounted, so if you want to convert the root, you'll need to run off an USB live distro.

First the partition needs to be checked and optimized using e2fsck -D /dev/sdX.

Then the file-system needs to be converted to 64bit: resize2fs -b /dev/sdX.

Finally checksums can be added: tune2fs -O metadata_csum /dev/sdX.

The file-system can then be mounted as usual. If this is the root file-system these modules might need to be added to /etc/mkinitcpio.conf:

   MODULES="... crc32c_intel crc32c_generic"

And then regenerate the kernel image using mkinitcpio -p linux.

See also