ext4

ext4 is the default file system for many Linux distributions including Debian and Ubuntu.[12]

Large file system

The ext4 filesystem can support volumes with sizes up to 1 exbibyte (EiB) and single files with sizes up to 16 tebibytes (TiB) with the standard 4 KiB block size.[13] The maximum file, directory, and filesystem size limits grow at least proportionately with the filesystem block size up to the maximum 64 KiB block size available on ARM and PowerPC/Power ISA CPUs.

Extents

Extents replace the traditional block mapping scheme used by ext2 and ext3. An extent is a range of contiguous physical blocks, improving large-file performance and reducing fragmentation. A single extent in ext4 can map up to 128 MiB of contiguous space with a 4 KiB block size.[4] There can be four extents stored directly in the inode. When there are more than four extents to a file, the rest of the extents are indexed in a tree.[14]

Backward compatibility

ext4 is backward-compatible with ext3 and ext2, making it possible to mount ext3 and ext2 as ext4. This will slightly improve performance, because certain new features of the ext4 implementation can also be used with ext3 and ext2, such as the new block allocation algorithm, without affecting the on-disk format.

ext3 is partially forward-compatible with ext4. Practically, ext4 will not mount as an ext3 filesystem out of the box, unless certain new features are disabled when creating it, such as ^extent, ^flex_bg, ^huge_file, ^uninit_bg, ^dir_nlink, and ^extra_isize.[15]

Persistent pre-allocation

ext4 can pre-allocate on-disk space for a file. To do this on most file systems, zeroes would be written to the file when created. In ext4 (and some other file systems such as XFS) fallocate(), a new system call in the Linux kernel, can be used. The allocated space would be guaranteed and likely contiguous. This situation has applications for media streaming and databases.

Delayed allocation

ext4 uses a performance technique called allocate-on-flush, also known as delayed allocation. That is, ext4 delays block allocation until data is flushed to disk; in contrast, some file systems allocate blocks immediately, even when the data goes into a write cache. Delayed allocation improves performance and reduces fragmentation by effectively allocating larger amounts of data at a time.

Unlimited number of subdirectories

ext4 does not limit the number of subdirectories in a single directory, except by the inherent size limit of the directory itself. (In ext3 a directory can have at most 32,000 subdirectories.)[16] To allow for larger directories and continued performance, ext4 in Linux 2.6.23 and later turns on HTree indices (a specialized version of a B-tree) by default, which allows directories up to approximately 10–12 million entries to be stored in the 2-level HTree index and 2 GB directory size limit for 4 KiB block size, depending on the filename length. In Linux 4.12 and later the largedir feature enabled a 3-level HTree and directory sizes over 2 GB, allowing approximately 6 billion entries in a single directory.

Journal checksums

ext4 uses checksums[17] in the journal to improve reliability, since the journal is one of the most used files of the disk. This feature has a side benefit: it can safely avoid a disk I/O wait during journaling, improving performance slightly. Journal checksumming was inspired by a research article from the University of Wisconsin, titled IRON File Systems[18] (specifically, section 6, called "transaction checksums"), with modifications within the implementation of compound transactions performed by the IRON file system (originally proposed by Sam Naghshineh in the RedHat summit).

Metadata checksumming

Since Linux kernel 3.5 released in 2012[19][20]

Faster file-system checking

In ext4 unallocated block groups and sections of the inode table are marked as such. This enables e2fsck to skip them entirely and greatly reduces the time it takes to check the file system. Linux 2.6.24 implements this feature.

fsck time dependence on inode count (ext3 vs. ext4)

Multiblock allocator

When ext3 appends to a file, it calls the block allocator, once for each block. Consequently, if there are multiple concurrent writers, files can easily become fragmented on disk. However, ext4 uses delayed allocation, which allows it to buffer data and allocate groups of blocks. Consequently, the multiblock allocator can make better choices about allocating files contiguously on disk. The multiblock allocator can also be used when files are opened in O_DIRECT mode. This feature does not affect the disk format.

Improved timestamps

As computers become faster in general, and as Linux becomes used more for mission-critical applications, the granularity of second-based timestamps becomes insufficient. To solve this, ext4 provides timestamps measured in nanoseconds. In addition, 2 bits of the expanded timestamp field are added to the most significant bits of the seconds field of the timestamps to defer the year 2038 problem for an additional 408 years.[3]

ext4 also adds support for time-of-creation timestamps. But, as Theodore Ts'o points out, while it is easy to add an extra creation-date field in the inode (thus technically enabling support for these timestamps in ext4), it is more difficult to modify or add the necessary system calls, like stat() (which would probably require a new version) and the various libraries that depend on them (like glibc). These changes will require coordination of many projects.[21] Therefore, the creation date stored by ext4 is currently only available to user programs on Linux via the statx() API.[22]

Project quotas

Support for project quotas was added in Linux kernel 4.4 on 8 Jan 2016. This feature allows assigning disk quota limits to a particular project ID. The project ID of a file is a 32-bit number stored on each file and is inherited by all files and subdirectories created beneath a parent directory with an assigned project ID. This allows assigning quota limits to a particular subdirectory tree independent of file access permissions on the file, such as user and project quotas that are dependent on the UID and GID. While this is similar to a directory quota, the main difference is that the same project ID can be assigned to multiple top-level directories and is not strictly hierarchical.[23]

Transparent encryption

Support for transparent encryption was added in Linux kernel 4.1 on June 2015.[24]

Lazy initialization

The lazyinit feature allows cleaning of inode tables in background, speeding initialization when creating a new ext4 file system.[25] It is available since 2010 in Linux kernel version 2.6.37.[26]

Write barriers

ext4 enables write barriers by default. It ensures that file system metadata is correctly written and ordered on disk, even when write caches lose power. This goes with a performance cost especially for applications that use fsync heavily or create and delete many small files. For disks with a battery-backed write cache, disabling barriers (option 'barrier=0') may safely improve performance.[27]

In 2008, the principal developer of the ext3 and ext4 file systems, Theodore Ts'o, stated that although ext4 has improved features, it is not a major advance, it uses old technology, and is a stop-gap. Ts'o believes that Btrfs is the better direction because "it offers improvements in scalability, reliability, and ease of management".[28] Btrfs also has "a number of the same design ideas that reiser3/4 had".[29] However, ext4 has continued to gain new features such as file encryption and metadata checksums.

The ext4 file system does not honor the "secure deletion" file attribute, which is supposed to cause overwriting of files upon deletion. A patch to implement secure deletion was proposed in 2011, but did not solve the problem of sensitive data ending up in the file-system journal.[30]

Because delayed allocation changes the behavior that programmers have been relying on with ext3, the feature poses some additional risk of data loss in cases where the system crashes or loses power before all of the data has been written to disk. Due to this, ext4 in kernel versions 2.6.30 and later automatically handles these cases as ext3 does.

The typical scenario in which this might occur is a program replacing the contents of a file without forcing a write to the disk with fsync. There are two common ways of replacing the contents of a file on Unix systems:[31]

• fd=open("file", O_TRUNC); write(fd, data); close(fd);

In this case, an existing file is truncated at the time of open (due to O_TRUNC flag), then new data is written out. Since the write can take some time, there is an opportunity of losing contents even with ext3, but usually very small. However, because ext4 can delay writing file data for a long time, this opportunity is much greater.

There are several problems that can arise:

1. If the write does not succeed (which may be due to error conditions in the writing program, or due to external conditions such as a full disk), then both the original version and the new version of the file will be lost, and the file may be corrupted because only a part of it has been written.
2. If other processes access the file while it is being written, they see a corrupted version.
3. If other processes have the file open and do not expect its contents to change, those processes may crash. One notable example is a shared library file which is mapped into running programs.

Because of these issues, often the following idiom is preferred over the one above:

• fd=open("file.new"); write(fd, data); close(fd); rename("file.new", "file");

A new temporary file ("file.new") is created, which initially contains the new contents. Then the new file is renamed over the old one. Replacing files by the rename() call is guaranteed to be atomic by POSIX standards – i.e. either the old file remains, or it's overwritten with the new one. Because the ext3 default "ordered" journaling mode guarantees file data is written out on disk before metadata, this technique guarantees that either the old or the new file contents will persist on disk. ext4's delayed allocation breaks this expectation, because the file write can be delayed for a long time, and the rename is usually carried out before new file contents reach the disk.

Using fsync() more often to reduce the risk for ext4 could lead to performance penalties on ext3 filesystems mounted with the data=ordered flag (the default on most Linux distributions). Given that both file systems will be in use for some time, this complicates matters for end-user application developers. In response, ext4 in Linux kernels 2.6.30 and newer detect the occurrence of these common cases and force the files to be allocated immediately. For a small cost in performance, this provides semantics similar to ext3 ordered mode and increases the chance that either version of the file will survive the crash. This new behavior is enabled by default, but can be disabled with the "noauto_da_alloc" mount option.[31]

The new patches have become part of the mainline kernel 2.6.30, but various distributions chose to backport them to 2.6.28 or 2.6.29.[32]

These patches don't completely prevent potential data loss or help at all with new files. The only way to be safe is to write and use software that does fsync() when it needs to. Performance problems can be minimized by limiting crucial disk writes that need fsync() to occur less frequently.[33]

Simplified structure of the Linux kernel: ext4 is implemented between the Linux kernel Virtual File System and the generic block layer.

Linux kernel Virtual File System is a subsystem or layer inside of the Linux kernel. It is the result of the very serious attempt to integrate multiple file systems into an orderly single structure. The key idea, which dates back to the pioneering work done by Sun Microsystems employees in 1986,[34] is to abstract out that part of the file system that is common to all file systems and put that code in a separate layer that calls the underlying concrete file systems to actually manage the data.

All system calls related to files (or pseudo files) are directed to the Linux kernel Virtual File System for initial processing. These calls, coming from user processes, are the standard POSIX calls, such as open, read, write, lseek, etc.

• Btrfs
• Comparison of file systems
• Extended file attributes
• e2fsprogs
• Ext2Fsd
• JFS
• List of file systems
• Reiser4
• XFS
• ZFS

• ext4 documentation in Linux kernel source
• Theodore Ts'o's discussion on ext4, 29 June 2006
• "ext4 online defragmentation" (materials from Ottawa Linux Symposium 2007)
• "The new ext4 filesystem: current status and future plans" (materials from Ottawa Linux Symposium 2007)
• Kernel Log: Ext4 completes development phase as interim step to btrfs, 17 October 2008
• "Ext4 block and inode allocator improvements" (materials from Ottawa Linux Symposium 2008)
• "Ext4: The Next Generation of Ext2/3 Filesystem"
• Ext4 (and Ext2/Ext3) Wiki
• Ext4 wiki at kernelnewbies.org
• Native Windows port of Ext4 and other FS in CROSSMETA
• Ext2read A windows application to read/copy ext2/ext3/ext4 files with extent and LVM2 support.
• Ext2Fsd Open source ext2/ext3/ext4 read/write file system driver for Windows. ext4 is supported from version 0.50 onwards
• Ext4fuse Open source read-only ext4 driver for FUSE. (Supports Mac OS X 10.5 and later, using MacFuse)