As should be clear by now, in reality, there is no such thing as the Linux filesystem. However, these filesystems share certain characteristics that make them viable as Linux filesystems.
A Linux filesystem adheres to the Filesystem Hierarchy Standard (FHS). This FHS is maintained by The Linux Foundation and is currently up to version 3.0. As with many things in the Linux ecosystem, it is based on a Unix predecessor: the Unix Filesystem Standard (UFS). It specifies the directory structure and its contents. We'll explore this structure together in the next part of this chapter.
Since Linux is most commonly used in servers, Linux filesystem implementations (often) have very advanced features on the topic of file integrity and disaster recovery. An example of such a disaster would be a system experiencing a power outage when it was in the middle of writing a business-critical file. If the write operation was stored in memory and aborted halfway through, the file would be in an inconsistent state. When the system is brought up again, the operating system does not have the write operation in memory anymore (since memory is cleared on each reboot), and only a part of the file will be written. Obviously, this is unwanted behavior and can lead to problems. Because of the properties of COW, Btrfs does not have this problem. However, ext4 and XFS are not COW filesystems. They both handle this issue in another way: with journaling:
As the preceding diagram shows, files are written to disk in three steps:
- Filesystem requests disk write from journal
- Journal writes on disk
- After file write, journal is updated
If the server crashes between steps 2 and 3, the write will be done again after power up, because the journal still contains the entry. The journal only contains some metadata about the operation, not the entire file. Since the journal contains a reference to the actual location on disk (the drive sectors), it will overwrite what was previously written, in this case, part of the file. If it finished successfully this time, the journal entry will be removed and the state of the file/disk is guaranteed. Should the server fail between steps 1 and 2, the actual instruction to write to disk has never been given and the software giving the instruction should account for that possibility.