Objectives
- Explain the basic filesystem organization
- Understand the role of the VFS
- Know what filesystems are available in Linux and which ones can be used on your actual system
- Grasp why journalling filesystems represent significant advances
- Discuss the use of special filesystems in Linux
Filesystem Basics
- A file is an abstraction from the physical hard disk
- A file can be in a physical partition or a logical partition controlled by a Logical Volume Manager
- Filesystems can be also of a network nature (with their true physical part hidden over the network)
Filesystem Tree Organization
- / is the root of the filesystem
- the filesystem may be composed of other filesystems
- /dev - /proc - /sys - /run are virtual pseudo filesystems stored only in memory
Available File Systems
| Name | Description |
|---|---|
| ext2 | Native Linux filesystem |
| minix | Predecessor to ext2 |
| proc | Used for /proc/ |
| msdos | MSDOS |
| umsdos | Extensions to MSDOS |
| vfat | Windows VFAT (includes FAT32, FAT, etc.) |
| ntfs | Windows NT NTFS (read-only) |
| sysv | SystemV/Xenit/Coherent FS |
| hpfs | OS/2 HPFS |
| ufs | Sun |
| udf | CD R/W, DVD |
| hfs | Apple MacIntosh Filesystem (HFS) |
| hfs+ | Apple MacIntosh Extended HFS |
| befs | BeOS filesystem |
| jffs, jffs2 | Journalling Flash Filesystem |
| iso9660 | cdrom etc., including Joliet extensions |
| cramfs | Compressed read-only filesystem |
| romfs | Small read-only filesystem |
| tmpfs | Ram disk that is swappable |
| ramfs | Ram disk, for example of a filesystem |
| gfs2 | Clustering filesystem from Red Hat |
| nfs | Network Filesystem (through version 4) |
| smb | Samba networking |
| ncp | Novell Netware FS using NCP Protocol |
| coda | Experimental distributed filesystem |
| afs | Andrewdistributed filesystem, from Carnegie Mellon |
| ocfs2 | Extent-based, disk cluster filesystem from Oracle |
Journalling Filesystems
Journalling filesystems are those that implements a mechanism to recover from crashes because they maintain a journal in the modifications of the filesystem that allows to keep track of the changes, those at the same time are made through transactions.
These are some of the journalling filesystems freely available under Linux
- ext3
- ext4
- reiserfs
- JSF
- XFS
- btrfs
Current Filesystems Types
Execute the following command
$ cat /proc/filesystems
Special Filesystems
There are some special filesystems used to access certain kernel data structures or for tunning kernel behavior or to implement particular functions
| Filesystem | Mount Point | Purpose |
|---|---|---|
| rootfs | None | During kernel load, provides an empty root directory. |
| hugetlbfs | Anywhere | Provides extended page access (2 or 4 MB on X86). |
| bdev | None | Used for block devices. |
| proc | /proc | Pseudo filesystem access to many kernel structures and subsystems. |
| sockfs | None | Used by BSD Sockets. |
| tmpfs | Anywhere | RAM disk with swapping, re-sizing. |
| shm | None | Used by System V IPC Shared Memory. |
| pipefs | None | Used for pipes. |
| binfmt_misc | Anywhere | Used by various executable formats. |
| devpts | /dev/pts | Used by Unix98 pseudo-terminals. |
| usbfs | /proc/bus/usb | Used by USB sub-system for dynamical devices. |
| sysfs | /sys (or elsewhere) | Used as a device tree. |
| debugfs | /sys/kernel/debug (or elsewhere) | Used for simple debugging file access. |
LAB 1.1 The tmpfs Special Filesystem
tmpfs is one of many special filesystems used under Linux. Some of these are not really used as filesystems, but just take advantage of the filesystem abstraction. However, tmpfs is a real filesystem that applications can do I/O on.
Essentially, tmpfs functions as a ramdisk; it resides purely in memory. But it has some nice properties that old-fashioned conventional ramdisk implementations did not have:
1. The filesystem adjusts its size (and thus the memory that is used) dynamically; it starts at zero and expands as necessary up to the maximum size it was mounted with.
2. If your RAM gets exhausted, tmpfs can utilize swap space. (You still can’t try to put more in the filesystem than its maximum capacity allows, however.)
3. tmpfs does not require having a normal filesystem placed in it, such as ext3 or vfat; it has its own methods for dealing with files and I/O that are aware that it is really just space in memory (it is not actually a block device), and as such are optimized for speed.
Thus there is no need to pre-format the filesystem with a mkfs command; you merely just have to mount it and use it.
Mount a new instance of tmpfs anywhere on your directory structure with a command like:
$ sudo mkdir /mnt/tmpfs $ sudo mount -t tmpfs none /mnt/tmpfs
See how much space the filesystem has been given and how much it is using:
$ df -h /mnt/tmpfs
You should see it has been alotted a default value of half of your RAM; however, the usage is zero, and will only start to grow as you place files on /mnt/tmpfs.
You could change the allotted size as a mount option as in:
$ sudo mount -t tmpfs -o size=1G none /mnt/tmpfs
You might try filling it up until you reach full capacity and see what happens. Do not forget to unmount when you are done with:
$ sudo umount /mnt/tmpfs
Virutally all modern Linux distributions mount an instance of tmpfs at /dev/shm:
$ df -h /dev/shm Filesystem tmpfs Type Size Used Avail Use% Mounted on tmpfs 3.9G 24M 3.9G 1% /dev/shm
Many applications use this such as when they are using POSIX shared memory as an inter-process communication mechanism. Any user can create, read and write files in /dev/shm, so it is a good place to create temporary files in memory.
Create some files in /dev/shm and note how the filesystem is filling up with df.
In addition, many distributions mount multiple instances of tmpfs; for example, on a RHEL 7 system:
$ df -h | grep tmpfs
Some distributions (such as Fedora) may (by default) mount /tmp as a tmpfs system; in such cases one has to avoid putting large files in /tmp to avoid running out of memory. Or one can disable this behavior as we discussed earlier when describing /tmp.