Revision History | ||
---|---|---|
Revision 1.9 | 2008-07-24 | Revised by: |
Text corrections. | ||
Revision 1.8 | 2008-01-06 | Revised by: |
Changes according to SquashFS release 3.3. Some parts added. | ||
Revision 1.7 | 2005-03-25 | Revised by: |
Changes according to SquashFS release 2.1. | ||
Revision 1.6 | 2004-11-10 | Revised by: |
Changes according to SquashFS release 2.0. Text corrections. | ||
Revision 1.5 | 2004-06-07 | Revised by: |
Changes according to SquashFS release 2.0 alpha. Lots of description improvements and clarifications. Split instructions for Linux kernels of 2.6.x (new) and 2.4.x series. | ||
Revision 1.1 | 2004-05-22 | Revised by: |
Changes according to SquashFS release 1.3r3. | ||
Revision 1.0 | 2004-02-19 | Revised by: |
Initial Release, reviewed by LDP. | ||
Revision 0.2 | 2003-12-08 | Revised by: |
Text corrections, license added. | ||
Revision 0.1 | 2003-11-24 | Revised by: |
Initial version. Instructions for SquashFS release 1.3r2. |
Abstract
This HOWTO describes the usage of SquashFS - a highly-compressed read-only file system for Linux, which is intended for use in tiny-sized and embedded systems, and anywhere else you'd want to use a compressed file system. With this document, you'll learn how to prepare a SquashFS-ready Linux kernel, create a squashed file system and happily use it.
Home of this HOWTO
The SquashFS HOWTO lives at http://artemio.net/projects/linuxdoc/squashfs. There you will always find the latest version of the document, and will be able to send your feedback.
When creating tiny-sized and embedded Linux systems, every byte of the storage device (floppy, flash disk, etc.) is very important, so compression is used everywhere possible. Also, compressed file systems are frequently needed for archiving purposes. For huge public archives, as well as for personal media archives, this is essential.
SquashFS brings all this to a new level. It is a read-only file system that lets you compress whole file systems or single directories, write them to other devices/partitions or to ordinary files, and then mount them directly (if a device) or using a loopback device (if it is a file). The modular, compact system design of SquashFS is bliss. For archiving purposes, SquashFS gives you a lot more flexibility and performance speed than a tarball archive.
SquashFS is distributed as a Linux kernel source patch (which enables SquashFS read support in your kernel), the mksquashfs tool, which creates squashed file systems (in a file or on a block device) and the unsquashfs tool, which extracts multiple files from an existing squashed file system.
The latest SquashFS release tree is 3.x, the former one was 2.x. This document describes both these releases with proper notes given. For example, if some feature or parameter is different in these release trees, it will be written as follows: new value (3.x) or old value (2.x)
Data, inodes and directories are compressed
SquashFS stores full uid/gids (32 bits), and file creation time
Files up to 2^64 bytes are supported; file systems can be up to 2^64 bytes
Inode and directory data are highly compacted, and packed on byte boundaries; each compressed inode is on average 8 bytes in length (the exact length varies on file type, i.e. regular file, directory, symbolic link, and block/character device inodes have different sizes)
SquashFS can use block sizes up to up to 64 Kb (2.x) and 1Mb (3.x). The default size is 128Kb (3.x), which achieves greater compression ratios than the normal 4K block size
By the 2.x release it was introduced the concept of fragment blocks: an ability to join multiple files smaller than block size into a single block, achieving greater compression ratios
File duplicates are detected and removed
Both big and little endian architectures are supported; SquashFS can mount file systems created on different byte-order machines
Now let's make sure any further discussions will be clearer for you to understand. The procedure of getting SquashFS working basically consists of the following steps:
Patching and recompiling the target Linux kernel to enable SquashFS support
Compiling the mksquashfs and the unsquashfs tools
Creating a compressed file system with mksquashfs
Testing: mounting a squashed file system to a temporary location
Modifying the /etc/fstab or startup scripts of your target Linux system to mount the new squashed file system when needed
The SquashFS home site is located at http://squashfs.sourceforge.net/ - it contains news for the latest release and it's changelog, as well as general information about SquashFS. You can grab the latest version at the SquashFS project page at SourceForge.
Squashfs is also avaliable with LZMA compression at http://www.squashfs-lzma.org/
In order to read SquashFS, you need it supported in your kernel - just as if it was a reiserfs or ext3 file system. You have to make sure there is an appropriate patch for your kernel version. It should be located in kernel-patches/linux-2.x.y subdirectory of the SquashFS source tree. Also, remember that in most cases you will need a clean (original) Linux kernel source from kernel.org. If your kernel source is from a distro vendor, it may be already pre-patched with custom vendor patches, and patching with a SquashFS patch will almost surely not work, as SquashFS patches are made against original Linux kernels. However some distributions make avaliable SquashFS-kernel-modules and SquashFS-tools packages from the repositories. If you consider to use these packages from your distro, you don't need to patch your kernel or to build SquashFS tools from source. In such case, make sure to get the proper module for your kernel from the distro repositories. Please note that doing so you will have an easy installation step, but no control on kernel configuration parameters if you mind to use SquashFS for particular purposes (eg. for embedded systems).
With a kernel source and a proper SquashFS patch present, all you have to do is (we'll assume that you have your Linux kernel source in /usr/src/linux and that you have the SquashFS source in /usr/src/squashfs):
Change to the SquashFS source directory and copy the kernel patch (we'll assume it's named squashfs-patch) to /usr/src/linux.
bash# cd /usr/src/squashfs bash# cp linux-2.x.y/squashfs-patch /usr/src/linux |
Go to the linux kernel source directory /usr/src/linux:
bash# cd /usr/src/linux |
Note: please remember that we will not be leaving this directory during all further kernel-related procedures, and all paths will be given relative to /usr/src/linux.
Now patch the source with the SquashFS patch:
bash# patch -p1 < squashfs-patch |
Cleanup and prepare the kernel source:
bash# make distclean bash# make mrproper |
Configure the kernel using your favourite method (config/menuconfig/xconfig/gconfig):
bash# make menuconfig |
In the "File systems" section, "Miscellaneous file systems" subsection, enable the "Squashed filesystem" option, whether as module or bundled with the kernel. It is only obligatory to compile SquashFS inside the kernel if you plan using squashed initial RAM disks (initrd).
In the same subsection, "do NOT" enable the "Additional option for memory-constrained system", unless you are configuring your kernel for an embedded system.
If you would like to use a squashed initial RAM disk, enable the "Initial RAM disk support" in the "Device drivers" section, "Block devices" subsection.
If you want to be able to mount the squashed file system via a loopback device in future, you should enable "Loopback device support" in the "Device drivers" section, "Block devices" subsection.
Now you may compile the kernel and modules:
bash# make |
Configure the kernel:
bash# make menuconfig |
In the "File systems" section, enable the "Squashed filesystem" option, whether as module or bundled with the kernel. It is only obligatory to compile SquashFS inside the kernel if you plan using squashed initial RAM disks (initrd).
If you would like to use a squashed initial RAM disk, enable the "Initial RAM disk support" in the "Block devices" section.
If you want to be able to mount the squashed file system via a loopback device in future, you should enable " Loopback device support " in the "Block devices " section.
Now you may compile the kernel and modules:
bash# make dep bash# make bzImage bash# make modules |
It's time to install your new SquashFS-enabled kernel. The instructions below are for installing and booting the kernel on the host machine. You may want to install and test it on the target system.
We assume that the kernel was compiled for a x86 architecture, and the compressed kernel image is located in the arch/i386/boot/ subdirectory of the kernel tree. Now copy the kernel to the /boot directory (and name it bzImage-sqsh for convenience, if you like):
bash# cp arch/i386/boot/bzImage /boot/bzImage-sqsh |
Don't forget to install the kernel modules if you have any:
bash# make modules_install |
Modify your boot loader's configuration file to include your new kernel and install (update) the boot loader. Now you may reboot with your new kernel. When it boots, check that everything went fine:
bash# cat /proc/filesystems |
Or, if you built SquashFS support as a kernel module:
bash# insmod squashfs bash# cat /proc/filesystems |
If you see the squashfs line among other file systems, this means you have successfully enabled SquashFS in your kernel.
Now you need to compile mksquashfs - the tool for creating squashed file systems and the unsquashfs which extracts files from an existing squashed file system.
bash# cd /usr/src/squashfs/squashfs-tools |
Compile and install the tools:
bash# make bash# cp mksquashfs /usr/sbin bash# cp unsquashfs /usr/sbin |
If everything went fine, typing mksquashfs or unsquashfs at the shell prompt should print the "usage" message.
If you use Debian (or onother Linux distribution) you may want to get SquashFS module and tools from the distro repositories. With Debian you have to install the proper kernel module and the tools with the following commands:
(Assuming your architecture is x86)
bash# apt-get install squashfs-modules-2.6-486 squashfs-tools |
Now load the squashfs module for the Linux kernel and if it was correctly loaded you should find it in the relative list
bash# modprobe squashfs bash# lsmod|grep squash squashfs 39620 0 |
Then you have to add the squashfs module to /etc/modules if you need it loaded at boot time.
bash# echo squashfs >> /etc/modules |
Note that at the time of writing, the Debian packages (Etch. 4.0 r2) relates to squashfs 3.1 release. Some latest options and features of 3.2 release may be not supported. Look at the next section for details.
mksquashfs is the tool for creating new squashed file systems, and for appending new data to existing squashed file systems. The general command-line format for mksquashfs is:
bash# mksquashfs source1 source2 ... destination [options] |
source1, source2, etc.: files and directories to be added to the resulting file system, given with relative and/or absolute paths
destination: a regular file (filesystem image file), or a block device (such as /dev/fd0 or /dev/hda3) where you want to have your squashed file system
Notes for default mksquashfs behavior:
When the new files are added to the new file system or appended to an existing one, mksquashfs will automatically rename files with duplicate names: if two or more files named text will appear in the same resulting directory, the second file will be renamed to text_1, third one to text_2 and so on.
Duplicate files will be removed, so there will be only one physical instance (By the SquashFS 2.x, you can disable the detection/removal of the duplicates with the -no-duplicates option).
If destination has a pre-existing SquashFS file system on it, by default, the new source items will be appended to the existing root directory. Examine the options table below to force mksquashfs to overwrite the whole destination and/or change the way new source items are added.
If a single source file or directory is given, it becomes the root in a newly created file system. If two or more source files and/or directories are given, they will all become sub-items in the root of the new file system.
The resulting filesystem will be padded to a multiple of 4 Kb: this is required for filesystems to be used on block devices. If you are very sure you don't ned this, use the -nopad option to disable this operation.
See the next section for more details about all possible options.
All possible options for mksquashfs are shown in the table below.
Table 1. Command-line options of the mksquashfs tool
Option | Description |
---|---|
-2.0 | force mksquashfs version 2.1 to create a version 2.0 filesystem |
-all-root or -root-owned | make all files in the target file system owned by root (UID=0, GID=0) |
-always-use-fragments | divide all files greater than block size into fragments (by the version 2.x). It will result in greater compression ratios |
-b [block size] | use [block size] filesystem block size (32 Kbytes default for 2.x, 128 kbytes for 3.x) - this can be either 4096, 8192, 16384, 32768, 65536 or 131072 |
-be or -le | force a big or little endian file system, respectively |
-check-data | enable additional file system checks |
-e [file1] ( [file2] ... ) | specify which files and/or directories to omit from the new file system that is to be created |
-ef [file] | specify a file which contains the list of files/directories to exclude |
-force-gid [GID] | set all group IDs in target file system to [GID] (can be specified as a name or a number) |
-force-uid [UID] | set all user IDs in target file system to [UID] (can be specified as a name or a number) |
-info | print files, their original size and compression ratio, as they are added to the file system |
-keep-as-directory | if the source is a single directory, force this directory to be a subdirectory of the root in the created file system |
-noappend | if the destination file/device already contains a squashed file system, overwrite it, rather than append the new data to an existing file system |
-no-duplicates | do not detect/remove duplicate file names |
-noD or -noDataCompression | do not compress the data |
-noF or -noFragmentCompression | do not compress the fragments (avaliable by 2.x) |
-no-fragments | do not generate fragment blocks (avaliable by 2.x, this will produce almost the same filesystem as 1.x did) |
-noI or -noInodeCompression | do not compress the inode table |
-nopad | do not pad the resulting file system to a multiple of 4 KBytes |
-root-becomes [name] | can be used while appending to a pre-existing squashed file system: it will make a new root, and [name] directory will contain all pre-existing files/directories |
-version | print the version, copyright and license message |
-recover [name] | recover filesystem data using recovery file [name] (3.3) |
-no-recovery | don't create a recovery file (3.3). |
-no-exports | don't make avaliable file system to export via NFS (3.x) |
-no-sparse | don't check for sparse files (3.x) |
-processors [number] | set the number of CPU to create file system. By default it will be used all avaliable processors (3.x) |
--read-queue [size] | set input queue to [size] Mb. (Default is 64 Mb)(3.x) |
-write-queue [size] | set output queue to [size] Mb (3.x) |
-sort [sort_file] | sort files relating to priorities in [sort_file] (3.x) |
-wildcards | enable the extended shell wildcards to exclude directories/files (to be used with -e) |
-regex | enable to use POSIX regular expressions (3.3) |
In most cases, you should leave all compression/block options by default, as they allow mksquashfs to achieve the best possible compression ratios.
unsquashfs is the tool for extracting data from squashed file systems. The general command-line format for unsquashfs is:
unsquashfs [options] target [files/directories to extract] |
target is the squashed file system to extract.
Notes for unsquashfs behavior:
By not specifying any destination path, unsquashfs extracts the compressed file system in the ./squashfs-root directory.
The tool does not extract a squashed file system on already exsisting directory unless the -f option is specified.
You can specify on the command line, a multiple number of files/directories to extract and the items to be extracted can be also be given in a file with -e [file] option.
All possible options for unsquashfs are shown in the table below.
Table 2. Command-line options of the unsquashfs tool
Option | Description |
---|---|
-v[ersion] | print the version, licence and copyright message |
-i[nfo] | print the files as they are extracted from the file system |
-l[ist] | list the squashed file system without extracting files |
-li | list files with theyr attributes as they are unsquashed (3.3) |
-ll | list the squashed file system files with attributes without any extraction (3.3) |
-d[estination] path | specify a destination path for unsquashed items |
-f[orce] | if files exist overwrite them |
-s[tat] | display file system's superblock informations (it can discover the file system version and the options used to compress it) (3.3) |
-e[f] [extract file] | list of directories or files to extract (entries given one per line) (3.3) |
-r[egex] | treat extract names as POSIX regular expressions (3.3) |
Note that by 3.x release you can extract 1.x and 2.x squashed file system too.
In order to create a squashed file system out of a single directory (say, /some/dir), and output it to a regular file (thus, producing a file system image), you need to say only one magic phrase:
bash# mksquashfs /some/dir dir.sqsh |
mksquashfs will perform the squashing and print the resulting number of inodes and size of data written, as well as the average compression ratio. Now you have your /some/dir directory image in the dir.sqsh file. You can now use the mount command to mount it using a loopback device:
bash# mkdir /mnt/dir bash# mount dir.sqsh /mnt/dir -t squashfs -o loop |
To check if you have what's expected:
bash# ls /mnt/dir |
If you want to output the file system directly into a device (say, your floppy at /dev/fd0):
bash# mksquashfs /some/dir /dev/fd0 |
Then just mount the device:
bash# mount /dev/fd0 /mnt/floppy -t squashfs |
And check if it's okay:
bash# ls /mnt/floppy |
Operations described here correspond to most cases where a read-only compressed file system can be used, whether you want it to be on a block device or in a file. This could be anything from large FTP/HTTP-served archives that don't change often, to having a squashed /usr partition and anything alike with these.
Let's suppose you have a /var/arch directory with lots of files and that you want to turn it into a squashed file system and keep it on your root partition as a file (it will be a file system image that you will mount via a loopback device). The operations needed to perform are as follows.
Squash the directory, then mount it via loopback to test it:
bash# mksquashfs /var/arch /var/arch.sqsh bash# mkdir /mnt/tmp bash# mount /var/arch.sqsh /mnt/tmp -t squashfs -o loop bash# ls /mnt/tmp |
If everything is as expected, make this file system mount automatically at boot time by adding this line to your /etc/fstab:
/var/arch.sqsh /var/arch squashfs ro,defaults 0 0 |
Unmount the file system from the temporary mount point, and mount using it's fstab entry:
bash# umount /mnt/tmp bash# mount /var/arch |
Now just ensure that everything works fine:
bash# ls /var/arch |
Say you have two hard disk partitions, /dev/hda6 (which is empty) and /dev/hda7 (which is bigger than /dev/hda6, mounted at /var/arch, contains some data and is full). Now, say you want to squash the /dev/hda7 file system and move it to /dev/hda6, then use /dev/hda7 for some other purposes. We will suppose you have the following line in /etc/fstab (reiserfs is just an example file system used on /dev/hda7):
/dev/hda7 /var/arch reiserfs defaults 0 0 |
In the same fashion as with the previous example:
bash# mksquashfs /var/arch /var/arch.sqsh bash# mkdir /mnt/tmp bash# mount /var/arch.sqsh /mnt/tmp -t squashfs -o loop bash# ls /mnt/tmp |
If everything went fine, unmount /dev/hda7 (if needed) and use dd to copy /var/arch.sqsh to /dev/hda6:
bash# umount /dev/hda7 bash# dd if=/var/arch.sqsh of=/dev/hda6 |
Now change the line in /etc/fstab for /dev/hda7 to:
/dev/hda6 /var/arch squashfs ro,defaults 0 0 |
Mount the new file system and check to see if all went fine:
bash# mount /var/arch bash# ls /var/arch |
Don't forget to erase the unneeded file system image:
bash# rm /var/arch.sqsh |
By saying "tiny/embedded", I mean Linux systems that are being built for booting from floppy disks, IDE/USB flash disks, iso9660 CD-ROMs, small-sized hard drives and the like. Whether you want to have your whole root file system on a single media (a single partition, a single floppy), or have a modular system (several floppies or disk partitions), the procedure is almost identical. Creating such Linux systems themselves is out of scope of this HOWTO - there are dedicated HOWTOs and guides for this (like the Bootdisk HOWTO and Linux From Scratch - visit www.tldp.org to retrieve these documents).
In order to use SquashFS for creating Linux systems on small disks, you just have to follow the usual steps for creating a minimal system, performing the following operations at respective points:
When developing a kernel for your system, make sure you enable SquashFS support so it can mount squashed file systems
Use mksquashfs for creating read-only initial ram disks and/or root and/or other file systems
Don't forget to set file system types to squashfs in /etc/fstab and/or the startup scripts of your system for mounting squashed file systems
Floppy example. Let's say you have your floppy system tree at /home/user/floppylinux and you want to place the root file system on one floppy and /usr on another. What you should do is:
bash# cd /home/user bash# mksquashfs floppylinux root.sqsh -e usr bash# mksquashfs floppylinux/usr usr.sqsh |
Note 1: you can see here how we use the -e option to exclude the /usr directory for root file system's image.
Note 2: don't forget to specify squashfs in your root disk's /etc/fstab or startup scripts when mounting the /usr file system.
Insert a root disk in your 3.5" floppy drive (I assume you have a lilo or grub on it, and, thus, a file system exists on this floppy, and the root file system will reside under the /boot directory of this file system):
bash# mount /mnt/floppy bash# cp root.sqsh /mnt/floppy/boot |
When done, unmount the root floppy, change the floppy to a /usr disk and use dd to transfer the usr file system:
bash# dd if=usr.sqsh of=/dev/fd0 |
With SquashFS, you can compress large file systems that will be used in live CDs (just as an example). For this purpose SquashFS is also used with UnionFS.
Enable SquashFS in the linux kernel of the target system
Create a squashed root file system
Modify the /etc/fstab or startup scripts of the target system to mount the squashd file system when you need it
If you create a root file system out of a running Linux system, use the -e option for mksquashfs to exclude all pseudo-filesystems such as /proc, /sys (on linux kernels after 2.5.x) and /dev (when using DevFS). Also, don't forget to add the file system image itself that is being created with mksquashfs (I think you know the reasons for these exclusions).
As mentioned, another interesting use for SquashFS is with Unionfs filesystem, which provides copy-on-write semantics for the read-only file systems, enahancing the possibilities. (For unionfs you can look at http://www.filesystems.org/project-unionfs.html)
Just to make an example, you may want to make your /home/user squashed, to compress and backup your files without losing the possibility to apply changes or writing new files.
Create the ro.fs squashed file system and the rw.fs dir.
bash# mksquashfs /home/user1 ro.fs bash# mkdir /home/rw.fs |
Mount the squashed ro.fs file system using the loopback device
bash# mount -t squashfs ro.fs /mnt -o loop |
mount the unionfs file system, that makes /mnt and /home/rw.fs apparently merged under /home/user1 location.
bash# cd /home bash# mount -t unionfs -o dirs=rw.fs=rw:/mnt=ro unionfs user1 |
As you can see, now you can create new files in /home/user1.
bash# cd /home/user1 bash# touch file1 bash# ls |
umount the unionfs and the squashfs file systems and list the content of /home/user1 and /home/rw.fs dir
bash# cd .. bash# umount /home/user1 bash# umount /mnt bash# ls /home/user1 bash# ls /home/rw.fs |
You can see that the new file1 was created in /home/rw.fs
When you want to add the new created files to the stable and compressed squashed file system, you have to add them to the exsisting one.
bash# mksquashfs /home/rw.fs /home/ro.fs |
Now, to mount your squashed user home directory at system startup, you can do as follow:
Make squashfs and unionfs modules loaded at boot time.
bash# echo squashfs >> /etc/modules bash# echo unionfs >> /etc/modules |
Change the owner of the writeble branch to match user1.
chown user1 /home/rw.fs |
Add these lines to /etc/fstab file to mount squashfs and unionfs at boot time.
... /home/ro.fs /mnt squashfs loop 0 0 unionfs /home/user1 unionfs dirs=/home/rw.fs=rw:/mnt=ro 0 0 |
I would like to express my sincere thanks and immeasurable respect to:
Phillip Lougher - for his brilliant work under squashfs, for creating an exculsive patch for linux-2.4.18, for his help with polishing this howto and answers to my mails
Tabatha Marshall at TLDP for helping me with bringing this HOWTO to the final 1.0 release
Everybody at The Linux Documentation Project for their great work under all the HOWTOs and guides that helped me a lot with exploring and hacking Linux
All those at the TLDP mailing lists who helped me with getting started
Endless thanks and respect to everybody who develops open-source software
Artemiy I. Pavlov
I want to thank Artemiy for his patience answering my emails, allowing me to work to this howto. I would express also my thanks and respect to everybody who is involved in free software
Marco Cecchetti
This document may be used and distributed under the terms and conditions set forth in the Open Content licence. In short, this means that you can freely modify and re-distribute the HOWTO under the main condition that you keep the author and copyright the article along. The full text of the licence is available at http://www.opencontent.org/opl.shtml