9.5. Root File System: Design and Building
http://www.denx.de/wiki/DULG/RootFileSystemDesignAndBuilding
It is not an easy task to design the root file systemfor an embedded system.There are three major problems to be solved:
- what to put in it
- which file system type to use
- where to store and how to boot it
For now we will assume that the contents of the root file systemis aready known;for example, it is given to us as a directory treeor a tarball which contains all the required files.
We will also assume that our system is a typical resource-limitedembedded system so we will especially look for solutions where theroot file system can be stored on on-board flash memory or otherflash memory based devices like CompactFlash or SD cards, MMC orUSB memory sticks.
A widespread approach to build a root file system is to use someLinux distribution (like the ELDK) and to remove things not needed.This approach may be pretty common, but it is almost always terriblywrong. You also don't build a family home by taking a skyscraper andremoving parts.Like a house, a root file system should be built bottom up, startingfrom scratch and adding things you know you need. Never add anythingwhere you don't exactly know what it's needed for.
But our focus here is on the second item:the options we have forchosing a file system type and the consequences this has.
In all cases we will base our experiments on the same content of theroot filesystem; we use the images of the SELF (Simple Embedded Linux Framework) that come with the ELDK. In a first step we will transform theSELF images into a tarball to meet the requirements mentioned above:
In a ELDK installation, the SELF images can be found in the/opt/eldk/<architecture>/images/
directory.There is already a compressed ramdisk image in this directory,which we will use (ramdisk_image.gz
):
- Uncompress ramdisk image:
bash$ gzip -d -c -v /opt/eldk/ppc_8xx/images/ramdisk_image.gz >/tmp/ramdisk_image /opt/eldk/ppc_8xx/images/ramdisk_image.gz: 61.4%
Note: The following steps require root permissions! - Mount ramdisk image:
bash# mount -o loop /tmp/ramdisk_image /mnt/tmp
- Create tarball; to avoid the need for
root
permissions in the following steps we don't include the device files in our tarball:bash# cd /mnt/tmp bash# tar -zc --exclude='dev/*' -f /tmp/rootfs.tar.gz *
- Instead, we create a separate tarball which contains only the device entries so we can use them when necessary (with
cramfs
):bash# tar -zcf /tmp/devices.tar.gz dev/ bash# cd /tmp
- Unmount ramdisk image:
bash# umount /mnt/tmp
We will use the /tmp/rootfs.tar.gz
tarball as master file in all following experiments.
9.5.1. Root File System on a Ramdisk
Ram disks are used very often to hold the root file system of embedded systems. They have several advantages:
- well-known
- well-supported by the Linux kernel
- simple to build
- simple to use - you can even combine the ramdisk with the Linux kernel into a single image file
- RAM based, thus pretty fast
- writable file system
- original state of file system after each reboot = easy recovery from accidental or malicious data corruption etc.
- big memory footprint: you always have to load the complete filesystem into RAM, even if only small parts of are actually used
- slow boot time: you have to load (and uncompress) the whole image before the first application process can start
- only the whole image can be replaced (not individual files)
- additional storage needed for writable persistent data
In almost all cases you will use an ext2
file system in your ramdiskimage. The following steps are needed to create it:
- Create a directory tree with the content of the target root filesystem. We do this by unpacking our master tarball:
$ mkdir rootfs $ cd rootfs $ tar zxf /tmp/rootfs.tar.gz
- We use the
genext2fs
tool to create the ramdisk image as this allows to use a simple text file to describe which devices shall be created in the generated file system image. That means that noroot
permissions are required at all. We use the following device tablerootfs_devices.tab
:#<name> <type> <mode> <uid> <gid> <major> <minor> <start> <inc> <count> /dev d 755 0 0 - - - - - /dev/console c 640 0 0 5 1 - - - /dev/fb0 c 640 0 0 29 0 - - - /dev/full c 640 0 0 1 7 - - - /dev/hda b 640 0 0 3 0 - - - /dev/hda b 640 0 0 3 1 1 1 16 /dev/kmem c 640 0 0 1 2 - - - /dev/mem c 640 0 0 1 1 - - - /dev/mtd c 640 0 0 90 0 0 2 16 /dev/mtdblock b 640 0 0 31 0 0 1 16 /dev/mtdr c 640 0 0 90 1 0 2 16 /dev/nftla b 640 0 0 93 0 - - - /dev/nftla b 640 0 0 93 1 1 1 8 /dev/nftlb b 640 0 0 93 16 - - - /dev/nftlb b 640 0 0 93 17 1 1 8 /dev/null c 640 0 0 1 3 - - - /dev/ptyp c 640 0 0 2 0 0 1 10 /dev/ptypa c 640 0 0 2 10 - - - /dev/ptypb c 640 0 0 2 11 - - - /dev/ptypc c 640 0 0 2 12 - - - /dev/ptypd c 640 0 0 2 13 - - - /dev/ptype c 640 0 0 2 14 - - - /dev/ptypf c 640 0 0 2 15 - - - /dev/ram b 640 0 0 1 0 0 1 2 /dev/ram b 640 0 0 1 1 - - - /dev/rtc c 640 0 0 10 135 - - - /dev/tty c 640 0 0 4 0 0 1 4 /dev/tty c 640 0 0 5 0 - - - /dev/ttyS c 640 0 0 4 64 0 1 8 /dev/ttyp c 640 0 0 3 0 0 1 10 /dev/ttypa c 640 0 0 3 10 - - - /dev/ttypb c 640 0 0 3 11 - - - /dev/ttypc c 640 0 0 3 12 - - - /dev/ttypd c 640 0 0 3 13 - - - /dev/ttype c 640 0 0 3 14 - - - /dev/ttypf c 640 0 0 3 15 - - - /dev/zero c 640 0 0 1 5 - - -
A description of the format of this table is part of the manual page for thegenext2fs
tool, genext2fs(8). - We can now create an
ext2
file system image using thegenext2fs
tool:$ ROOTFS_DIR=rootfs # directory with root file system content $ ROOTFS_SIZE=3700 # size of file system image $ ROOTFS_FREE=100 # free space wanted $ ROOTFS_INODES=380 # number of inodes $ ROOTFS_DEVICES=rootfs_devices.tab # device description file $ ROOTFS_IMAGE=ramdisk.img # generated file system image $ genext2fs -U \ -d ${ROOTFS_DIR} \ -D ${ROOTFS_DEVICES} \ -b ${ROOTFS_SIZE} \ -r ${ROOTFS_FREE} \ -i ${ROOTFS_INODES} \ ${ROOTFS_IMAGE}
- Compress the file system image:
$ gzip -v9 ramdisk.img rootfs.img: 55.6% -- replaced with ramdisk.img.gz
- Create an U-Boot image file from it:
$ mkimage -T ramdisk -C gzip -n 'Test Ramdisk Image' \ > -d ramdisk.img.gz uRamdisk Image Name: Test Ramdisk Image Created: Sun Jun 12 16:58:06 2005 Image Type: PowerPC Linux RAMDisk Image (gzip compressed) Data Size: 1618547 Bytes = 1580.61 kB = 1.54 MB Load Address: 0x00000000 Entry Point: 0x00000000
We now have a root file system image uRamdisk
that can beused with U-Boot.
9.5.2. Root File System on a JFFS2 File System
JFFS2 (Journalling Flash File System version 2)was specifically designed for use on flash memory devicesin embedded systems.It is a log-structured file system which means that it is robustagainst loss of power, crashes or other unorderly shutdowns of the system("robust" means that data that is just being written when thesystem goes down may be lost,but the file system itself does not get corruptedand the system can be rebootet without need for any kindof file system check).
Some of the advantages of using JFFS2 as root file system in embedded systemsare:
- file system uses compression, thus making efficient use of flash memory
- log-structured file system, thus robust against unorderly shutdown
- flash sector wear-leveling
- writable flash file system
Disadvantages are:
- long mount times (especially older versions)
- slow when reading: files to be read get uncompressed on the fly which eats CPU cycles and takes time
- slow when writing: files to be written get compressed, which eats CPU cycles and takes time, but it may even take much longer until data gets actually stored in flash if the file system becomes full and blocks must be erased first or - even worse - if garbage collection becomes necessary
- The garbage collector thread may run at any time, consuming CPU cycles and blocking accesses to the file system.
Despite the aforementioned disadvantages,systems using a JFFS2 based root file system are easy to build,make efficient use of the available resourcesand can run pretty reliably.
To create a JFFS2 based root file system please proceed as follows:
- Create a directory tree with the content of the target root filesystem. We do this by unpacking our master tarball:
$ mkdir rootfs $ cd rootfs $ tar zxf /tmp/rootfs.tar.gz
- We can now create a
JFFS2
file system image using themkfs.jffs2
tool:$ ROOTFS_DIR=rootfs # directory with root file system content $ ROOTFS_EBSIZE=0x20000 # erase block size of flash memory $ ROOTFS_ENDIAN=b # target system is big endian $ ROOTFS_DEVICES=rootfs_devices.tab # device description file $ ROOTFS_IMAGE=jffs2.img # generated file system image $ mkfs.jffs2 -U \ -d ${ROOTFS_DIR} \ -D ${ROOTFS_DEVICES} \ -${ROOTFS_ENDIAN} \ -e ${ROOTFS_EBSIZE} \ -o ${ROOTFS_IMAGE} mkfs.jffs2: skipping device_table entry '/dev': no parent directory!
JFFS2
file system image to a NAND flashdevice, you should also add the
"-n"
(or
"--no-cleanmarkers"
) option,as cleanmarkers are not needed then.
When booting the Linux kernel prints the followingmessages showing the default partition mapwhich is used for the flash memoryon the TQM8xxL boards:
TQM flash bank 0: Using static image partition definition Creating 7 MTD partitions on "TQM8xxL0": 0x00000000-0x00040000 : "u-boot" 0x00040000-0x00100000 : "kernel" 0x00100000-0x00200000 : "user" 0x00200000-0x00400000 : "initrd" 0x00400000-0x00600000 : "cramfs" 0x00600000-0x00800000 : "jffs" 0x00400000-0x00800000 : "big_fs"We use U-Boot to load and store the JFFS2 image into the lastpartition and set up the Linux boot arguments to use this asroot device:
- Erase flash:
=> era 40400000 407FFFFF ................. done Erased 35 sectors
- Download JFFS2 image:
=> tftp 100000 /tftpboot/TQM860L/jffs2.img Using FEC ETHERNET device TFTP from server 192.168.3.1; our IP address is 192.168.3.80 Filename '/tftpboot/TQM860L/jffs2.img'. Load address: 0x100000 Loading: ################################################################# ################################################################# ################################################################# ################################################################# ################################################################# ################################################################# ######## done Bytes transferred = 2033888 (1f08e0 hex)
- Copy image to flash:
=> cp.b 100000 40400000 ${filesize} Copy to Flash... done
- set up boot arguments to use flash partition 6 as root device:
=> setenv mtd_args setenv bootargs root=/dev/mtdblock6 rw rootfstype=jffs2 => printenv addip addip=setenv bootargs ${bootargs} ip=${ipaddr}:${serverip}:${gatewayip}:${netmask}:${hostname}:${netdev}:off panic=1 => setenv flash_mtd 'run mtd_args addip;bootm ${kernel_addr}' => run flash_mtd Using FEC ETHERNET device TFTP from server 192.168.3.1; our IP address is 192.168.3.80 Filename '/tftpboot/TQM860L/uImage'. Load address: 0x200000 Loading: ################################################################# ################################################################# ########### done Bytes transferred = 719233 (af981 hex) ## Booting image at 40040000 ... Image Name: Linux-2.4.25 Created: 2005-06-12 16:32:24 UTC Image Type: PowerPC Linux Kernel Image (gzip compressed) Data Size: 782219 Bytes = 763.9 kB Load Address: 00000000 Entry Point: 00000000 Verifying Checksum ... OK Uncompressing Kernel Image ... OK Linux version 2.4.25 (wd@xpert) (gcc version 3.3.3 (DENX ELDK 3.1.1 3.3.3-9)) #1 Sun Jun 12 18:32:18 MEST 2005 On node 0 totalpages: 4096 zone(0): 4096 pages. zone(1): 0 pages. zone(2): 0 pages. Kernel command line: root=/dev/mtdblock6 rw rootfstype=jffs2 ip=192.168.3.80:192.168.3.1::255.255.255.0:tqm860l:eth1:off panic=1 Decrementer Frequency = 187500000/60 Calibrating delay loop... 49.86 BogoMIPS ... NET4: Unix domain sockets 1.0/SMP for Linux NET4.0. VFS: Mounted root (jffs2 filesystem). Freeing unused kernel memory: 56k init BusyBox v0.60.5 (2005.03.07-06:54+0000) Built-in shell (msh) Enter 'help' for a list of built-in commands. # ### Application running ... # mount rootfs on / type rootfs (rw) /dev/mtdblock6 on / type jffs2 (rw) /proc on /proc type proc (rw) # df / Filesystem 1k-blocks Used Available Use% Mounted on rootfs 4096 2372 1724 58% /
cramfs is a compressed, read-only file system.
Advantages are:
- file system uses compression, thus making efficient use of flash memory
- Allows for quick boot times as only used files get loaded and uncompressed
Disadvantages are:
- only the whole image can be replaced (not individual files)
- additional storage needed for writable persistent data
mkcramfs
tool does not support device table, so we need root permissions to create the required device files
To create a cramfs
based root file system please proceed as follows:
- Create a directory tree with the content of the target root filesystem. We do this by unpacking our master tarball:
$ mkdir rootfs $ cd rootfs $ tar -zxf /tmp/rootfs.tar.gz
- Create the required device files. We do this here by unpacking a special tarball which holds only the device file entries. Note: this requires root permissions!
# cd rootfs # tar -zxf /tmp/devices.tar.gz
- Many tools require some storage place in a filesystem, so we must provide at least one (small) writable filesystem. For all data which may be lost when the system goes down, a
"tmpfs"
filesystem is the optimal choice. To create such a writable tmpfs filesystem we add the following lines to the/etc/rc.sh
script:# mount TMPFS because root-fs is readonly /bin/mount -t tmpfs -o size=2M tmpfs /tmpfs
Some tools require write permissions on some device nodes (for example, to change ownership and permissions), or dynamically (re-) create such files (for example,/dev/log
which is usually a Unix Domain socket). The files are placed in a writable filesystem; in the root filesystem symbolic links are used to point to their new locations:dev/ptyp0 → /tmpfs/dev/ptyp0 dev/ttyp0 → /tmpfs/dev/ttyp0 dev/ptyp1 → /tmpfs/dev/ptyp1 dev/ttyp1 → /tmpfs/dev/ttyp1 dev/ptyp2 → /tmpfs/dev/ptyp2 dev/ttyp2 → /tmpfs/dev/ttyp2 dev/ptyp3 → /tmpfs/dev/ptyp3 dev/ttyp3 → /tmpfs/dev/ttyp3 dev/ptyp4 → /tmpfs/dev/ptyp4 dev/ttyp4 → /tmpfs/dev/ttyp4 dev/ptyp5 → /tmpfs/dev/ptyp5 dev/ttyp5 → /tmpfs/dev/ttyp5 dev/ptyp6 → /tmpfs/dev/ptyp6 dev/ttyp6 → /tmpfs/dev/ttyp6 dev/ptyp7 → /tmpfs/dev/ptyp7 dev/ttyp7 → /tmpfs/dev/ttyp7 dev/ptyp8 → /tmpfs/dev/ptyp8 dev/ttyp8 → /tmpfs/dev/ttyp8 dev/ptyp9 → /tmpfs/dev/ptyp9 dev/ttyp9 → /tmpfs/dev/ttyp9 dev/ptypa → /tmpfs/dev/ptypa dev/ttypa → /tmpfs/dev/ttypa dev/ptypb → /tmpfs/dev/ptypb dev/ttypb → /tmpfs/dev/ttypb dev/ptypc → /tmpfs/dev/ptypc dev/ttypc → /tmpfs/dev/ttypc dev/ptypd → /tmpfs/dev/ptypd dev/ttypd → /tmpfs/dev/ttypd dev/ptype → /tmpfs/dev/ptype dev/ttype → /tmpfs/dev/ttype dev/ptypf → /tmpfs/dev/ptypf dev/ttypf → /tmpfs/dev/ttypf tmp → /tmpfs/tmp var → /tmpfs/var dev/log → /var/log/log In case you use dhclient also: etc/dhclient.conf → /tmpfs/var/lib/dhclient.conf etc/resolv.conf → /tmpfs/var/lib/resolv.conf
To place the corresponding directories and device files in thetmpfs
file system, the following code is added to the/etc/rc.sh
script:mkdir -p /tmpfs/tmp /tmpfs/dev \ /tmpfs/var/lib/dhcp /tmpfs/var/lock /tmpfs/var/run while read name minor do mknod /tmpfs/dev/ptyp$name c 2 $minor mknod /tmpfs/dev/ttyp$name c 3 $minor done <<__EOD__ 0 0 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 a 10 b 11 c 12 d 13 e 14 f 15 __EOD__ chmod 0666 /tmpfs/dev/*
- We can now create a
cramfs
file system image using themkcramfs
tool:$ ROOTFS_DIR=rootfs # directory with root file system content $ ROOTFS_ENDIAN="-r" # target system has reversed (big) endianess $ ROOTFS_IMAGE=cramfs.img # generated file system image PATH=/opt/eldk/usr/bin:$PATH mkcramfs ${ROOTFS_ENDIAN} ${DEVICES} ${ROOTFS_DIR} ${ROOTFS_IMAGE} Swapping filesystem endian-ness bin dev etc ... -48.78% (-86348 bytes) in.ftpd -46.02% (-16280 bytes) in.telnetd -45.31% (-74444 bytes) xinetd Everything: 1864 kilobytes Super block: 76 bytes CRC: c166be6d warning: gids truncated to 8 bits. (This may be a security concern.)
- We can use the same setup as before for the JFFS2 filesystem, just changing the bootargument to
"rootfstype=cramfs"
9.5.4. Root File System on a Read-Only ext2 File System
When storing the root file system in on-board flash memoryit seems only natural to look for special flash filesystemslike JFFS2, or for other file system types that are designedfor such environments like cramfs.It seems to be a bad idea to use a standard ext2
file systembecause it contains neither any type of wear leveling whichis needed for writable file systems in flash memory,nor is it robust against unorderly shutdowns.
The situation changes if we use an ext2
file system which we mountread-only. Such a configuration can be very useful in some situations.
Advantages:
- very fast
- low RAM memory footprint
Disadvantages:
- high flash memory footprint because no compression
To create an ext2
imagethat can be used as a read-only root file systemthe following steps are necessary:
- Create a directory tree with the content of the target root filesystem. We do this by unpacking our master tarball:
$ mkdir rootfs $ cd rootfs $ tar -zxf /tmp/rootfs.tar.gz
- Like with the
cramfs
root file system, we use"tmpfs"
for cases where a writable file system is needed and add the following lines to the/etc/rc.sh
script:# mount TMPFS because root-fs is readonly /bin/mount -t tmpfs -o size=2M tmpfs /tmpfs
We also create the same symbolic links for device files that must be placed in a writable filesystem:dev/ptyp0 → /tmpfs/dev/ptyp0 dev/ttyp0 → /tmpfs/dev/ttyp0 dev/ptyp1 → /tmpfs/dev/ptyp1 dev/ttyp1 → /tmpfs/dev/ttyp1 dev/ptyp2 → /tmpfs/dev/ptyp2 dev/ttyp2 → /tmpfs/dev/ttyp2 dev/ptyp3 → /tmpfs/dev/ptyp3 dev/ttyp3 → /tmpfs/dev/ttyp3 dev/ptyp4 → /tmpfs/dev/ptyp4 dev/ttyp4 → /tmpfs/dev/ttyp4 dev/ptyp5 → /tmpfs/dev/ptyp5 dev/ttyp5 → /tmpfs/dev/ttyp5 dev/ptyp6 → /tmpfs/dev/ptyp6 dev/ttyp6 → /tmpfs/dev/ttyp6 dev/ptyp7 → /tmpfs/dev/ptyp7 dev/ttyp7 → /tmpfs/dev/ttyp7 dev/ptyp8 → /tmpfs/dev/ptyp8 dev/ttyp8 → /tmpfs/dev/ttyp8 dev/ptyp9 → /tmpfs/dev/ptyp9 dev/ttyp9 → /tmpfs/dev/ttyp9 dev/ptypa → /tmpfs/dev/ptypa dev/ttypa → /tmpfs/dev/ttypa dev/ptypb → /tmpfs/dev/ptypb dev/ttypb → /tmpfs/dev/ttypb dev/ptypc → /tmpfs/dev/ptypc dev/ttypc → /tmpfs/dev/ttypc dev/ptypd → /tmpfs/dev/ptypd dev/ttypd → /tmpfs/dev/ttypd dev/ptype → /tmpfs/dev/ptype dev/ttype → /tmpfs/dev/ttype dev/ptypf → /tmpfs/dev/ptypf dev/ttypf → /tmpfs/dev/ttypf tmp → /tmpfs/tmp var → /tmpfs/var dev/log → /var/log/log In case you use dhclient also: etc/dhclient.conf → /tmpfs/var/lib/dhclient.conf etc/resolv.conf → /tmpfs/var/lib/resolv.conf
To place the corresponding directories and device files in thetmpfs
file system, the following code is added to the/etc/rc.sh
script:mkdir -p /tmpfs/tmp /tmpfs/dev \ /tmpfs/var/lib/dhcp /tmpfs/var/lock /tmpfs/var/run while read name minor do mknod /tmpfs/dev/ptyp$name c 2 $minor mknod /tmpfs/dev/ttyp$name c 3 $minor done <<__EOD__ 0 0 1 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 a 10 b 11 c 12 d 13 e 14 f 15 __EOD__ chmod 0666 /tmpfs/dev/*
- Like we did for the ramdisk, we now create an
ext2
file system image using thegenext2fs
tool:$ ROOTFS_DIR=rootfs # directory with root file system content $ ROOTFS_SIZE=3700 # size of file system image $ ROOTFS_FREE=100 # free space wanted $ ROOTFS_INODES=380 # number of inodes $ ROOTFS_DEVICES=rootfs_devices.tab # device description file $ ROOTFS_IMAGE=ext2.img # generated file system image $ genext2fs -U \ -d ${ROOTFS_DIR} \ -D ${ROOTFS_DEVICES} \ -b ${ROOTFS_SIZE} \ -r ${ROOTFS_FREE} \ -i ${ROOTFS_INODES} \ ${ROOTFS_IMAGE}
- We can again use the same setup as before for the JFFS2 filesystem, just changing the boot argument to
"rootfstype=ext2"
(or simply omit it completely as this is the default anyway), and we must change the"rw"
argument into"ro"
to mount our root file system really read-only:... Linux version 2.4.25 (wd@xpert) (gcc version 3.3.3 (DENX ELDK 3.1.1 3.3.3-9)) #1 Sun Jun 12 18:32:18 MEST 2005 On node 0 totalpages: 4096 zone(0): 4096 pages. zone(1): 0 pages. zone(2): 0 pages. Kernel command line: root=/dev/mtdblock6 ro rootfstype=ext2 ip=192.168.3.80:192.168.3.1::255.255.255.0:tqm860l:eth1:off panic=1 Decrementer Frequency = 187500000/60 Calibrating delay loop... 49.86 BogoMIPS ...
9.5.5. Root File System on a Flash Card
Using an ext2
file system on a flash memory card(like CompactFlash, SD, MMC or a USB memory stick)is standard technology.To avoid unnecessary flash wear it is a good idea to mountthe root file system read-only,or at least using the "noatime"
mount option.
For our test we can use the "ext2.img"
file from the previous step without changes:
- In this test we use a standard CompactFlash card which comes with a single partition on it. We use U-Boot to copy the
ext2
file system image into this partition:=> tftp 100000 /tftpboot/TQM860L/ext2.img Using FEC ETHERNET device TFTP from server 192.168.3.1; our IP address is 192.168.3.80 Filename '/tftpboot/TQM860L/ext2.img'. Load address: 0x100000 Loading: ################################################################# ################################################################# ################################################################# ################################################################# ################################################################# ################################################################# ################################################################# ################################################################# ################################################################# ################################################################# ################################################################# ########################## done Bytes transferred = 3788800 (39d000 hex) => ide part Partition Map for IDE device 0 -- Partition Type: DOS Partition Start Sector Num Sectors Type 1 32 500704 6 => ide write 100000 20 1ce8 IDE write: device 0 block # 32, count 7400 ... 7400 blocks written: OK
Note that the"ide write"
command takes parameters as hex numbers, and the write count is in terms of disk blocks of 512 bytes each. So we have to use 0x20 for the starts sector of the first partition, and 3788800 / 512 = 7400 = 0x1CE8 for the block count. - We now prepare the Linux boot arguments to take this partition as read-only root device:
=> setenv cf_args setenv bootargs root=/dev/hda1 ro => setenv flash_cf 'run cf_args addip;bootm ${kernel_addr} - ${fdt_addr}' => setenv bootcmd run flash_cf
- ...and boot the system:
... Linux version 2.4.25 (wd@xpert) (gcc version 3.3.3 (DENX ELDK 3.1.1 3.3.3-9)) #1 Sun Jun 12 18:32:18 MEST 2005 On node 0 totalpages: 4096 zone(0): 4096 pages. zone(1): 0 pages. zone(2): 0 pages. Kernel command line: root=/dev/hda1 ro ip=192.168.3.80:192.168.3.1::255.255.255.0:tqm860l:eth1:off panic=1 Decrementer Frequency = 187500000/60 Calibrating delay loop... 49.86 BogoMIPS ...
9.5.6. Root File System in a Read-Only File in a FAT File System
This is a more complicated example that shows that - dependingon project requirements - many other alternatives for chosing a rootfile system for your embedded system exist.
The scenario is as follows: on your embedded device you use acheap and popular storage medium like CompactFlash, MMC or SD cardsor USB memory sticksto store both the Linux kernel and your root file system.You want to distribute software updates over the internet:your customers can download the file from your web site,or you sent the images by email.Your customers may use any flash card or memory stick they happen to find,so you have no information about brand or size of the storage device.
Unfortunately most of your customers use Windows systems.And they don't want to be bothered with long instructionshow to create special partitions on the storage deviceor how to write binary images or things like that.A simple "copy file" operation is nearly exhausting their capabilities.
What to do?Well, if copying a file is all your customers can do we shouldnot ask for more. Storage devices like CompactFlash cards etc.typically come with a single partition on it, which holds a FAT
orVFAT
file system. This cannot be used as a Linux root file system directly,so we have to use some trickery.
Here is one possible solution:Your software distribution consistes of two files:The first file is the Linux kernel with a minimal ramdisk image attached(using the multi-file image format for U-Boot); U-Boot can load and boot such files from a FAT or VFAT file system.The second file is your root file system.For convenience and speed we use again an image of an ext2
file system.When Linux boots, it will initially use the attached ramdiskas root file system.The programs in this ramdisk will mount the FAT or VFAT file system -read-only.Then we can use a loop device (see losetup(8))to associate the root file system image with a block devicewhich can be used as a mount point.And finally we use pivot_root(8) to change the root file systemto our image on the CF card.
This sounds not so complicated,and actually it is quite simple onceyou understand what needs to be done.Here is a more detailed description:
- The root file system image is easy: as mantioned before, we will use an
ext2
file system image, and to avoid wearing the flash storage device we will use it in read-only mode - we did a read-onlyext2
root file system image before, and here we can just re-use the existing image file. - The initial ramdisk image that performs the pivot_root step must be created from scratch, but we already know how to create ramdisk images, so we just have to figure out what to put in it.
The most important tool here isnash
, a script interpreter that was specifically designed for such purposes (see nash(8)). We don't need any additional tools, and if we use static linking, then thenash
binary plus a small script to control it is all we need for our initial ramdisk.
To be precise, we need a couple of (empty) directories (bin
,dev
,etc
,lib
,loopfs
,mnt
,proc
, andsysroot
), thebin/nash
binary, thelinuxrc
script and a symbolic linksbin
pointing tobin
:drwxr-xr-x 2 wd users 4096 Apr 13 01:11 bin -rwxr-xr-x 1 wd users 469512 Apr 11 22:47 bin/nash drwxr-xr-x 2 wd users 4096 Apr 12 00:04 dev drwxr-xr-x 2 wd users 4096 Apr 12 00:04 etc drwxr-xr-x 2 wd users 4096 Apr 12 00:04 lib -rwxr-xr-x 1 wd users 511 Apr 13 01:28 linuxrc drwxr-xr-x 2 wd users 4096 Apr 12 00:04 loopfs drwxr-xr-x 2 wd users 4096 Apr 12 00:09 mnt drwxr-xr-x 2 wd users 4096 Apr 12 00:04 proc lrwxrwxrwx 1 wd users 3 Jun 12 18:54 sbin -> bin drwxr-xr-x 2 wd users 4096 Apr 12 00:04 sysroot
- We also need only a minimal device table for creating the initial ramdisk:
#<name> <type> <mode> <uid> <gid> <major> <minor> <start> <inc> <count> /dev d 755 0 0 - - - - - /dev/console c 640 0 0 5 1 - - - /dev/hda b 640 0 0 3 0 - - - /dev/hda b 640 0 0 3 1 1 1 8 /dev/loop b 640 0 0 7 0 0 1 4 /dev/null c 640 0 0 1 3 - - - /dev/ram b 640 0 0 1 0 0 1 2 /dev/ram b 640 0 0 1 1 - - - /dev/tty c 640 0 0 4 0 0 1 4 /dev/tty c 640 0 0 5 0 - - - /dev/ttyS c 640 0 0 4 64 0 1 4 /dev/zero c 640 0 0 1 5 - - -
- To create the initial ramdisk we perform the usual steps:
$ INITRD_DIR=initrd $ INITRD_SIZE=490 $ INITRD_FREE=0 $ INITRD_INODES=54 $ INITRD_DEVICES=initrd_devices.tab $ INITRD_IMAGE=initrd.img $ genext2fs -U \ -d ${INITRD_DIR} \ -D ${INITRD_DEVICES} \ -b ${INITRD_SIZE} \ -r ${INITRD_FREE} \ -i ${INITRD_INODES} \ ${INITRD_IMAGE} $ gzip -v9 ${INITRD_IMAGE}
The result is a really small (233 kB) compressed ramdisk image. - Assuming you already have your Linux kernel image, you can now use
mkimage
to build an U-Boot multi-file image that combines the Linux kernel and the initial ramdisk:$ LINUX_KERNEL=linuxppc_2_4_devel/arch/ppc/boot/images/vmlinux.gz $ mkimage -A ppc -O Linux -T multi -C gzip \ > -n 'Linux with Pivot Root Helper' \ > -d ${LINUX_KERNEL}:${INITRD_IMAGE}.gz linux.img Image Name: Linux with Pivot Root Helper Created: Mon Jun 13 01:48:11 2005 Image Type: PowerPC Linux Multi-File Image (gzip compressed) Data Size: 1020665 Bytes = 996.74 kB = 0.97 MB Load Address: 0x00000000 Entry Point: 0x00000000 Contents: Image 0: 782219 Bytes = 763 kB = 0 MB Image 1: 238433 Bytes = 232 kB = 0 MB
The newly created filelinux.img
is the second image we have to copy to the CF card.
We are done.
But wait - one essential part was not mentioned yet:the linuxrc
script in our initial ramdisk imagewhich contains all the magic.This script is quite simple:
#!/bin/nash echo Mounting /proc filesystem mount -t proc /proc /proc echo Creating block devices mkdevices /dev echo Creating root device mkrootdev /dev/root echo 0x0100 > /proc/sys/kernel/real-root-dev echo Mounting flash card mount -o noatime -t vfat /dev/hda1 /mnt echo losetup for filesystem image losetup /dev/loop0 /mnt/rootfs.img echo Mounting root filesystem image mount -o defaults --ro -t ext2 /dev/loop0 /sysroot echo Running pivot_root pivot_root /sysroot /sysroot/initrd umount /initrd/procLet's go though it step by step:
- The first line says that it's a script file for the
/bin/nash
interpreter.
Note: even if this file looks like a shell script it is NOT interpreted by a shell, but by thenash
interpreter. For a complete list of availablenash
commands and their syntax please refer to the manual page, nash(8). - The first action is to mount the
/proc
pseudo file system which is needed to find out some required information. - Then we create block device entries for all partitions listed in
/proc/partitions
(mkdevices
command). - In the next step a block device for our new root file system is created (
mkrootdev
command). - Then we mount the CF card. We assume that there is only a single partition on it (
/dev/hda1
) which is of typeVFAT
(which also will work withFAT
file systems). These assumptions work fine with basicly all memory devices used under Windows. - We further assume that the file name of the root file system image on the CF card is
"rootfs.img"
- this file now gets mounted using a loop device (losetup
andmount
commands). - Our file system image, is now mounted on the
/sysroot
directory. In the last step we usepivot_root
to make this the new root file system. - As a final cleanup we unmount the
/proc
file system which is not needed any more.
And how does this work on our target?
- First we copy the two images to the CF card; we do this on the target under Linux:
bash-2.05b# fdisk -l /dev/hda Disk /dev/hda: 256 MB, 256376832 bytes 16 heads, 32 sectors/track, 978 cylinders Units = cylinders of 512 * 512 = 262144 bytes Device Boot Start End Blocks Id System /dev/hda1 * 1 978 250352 6 FAT16 bash-2.05b# mkfs.vfat /dev/hda1 mkfs.vfat 2.8 (28 Feb 2001) bash-2.05b# mount -t vfat /dev/hda1 /mnt bash-2.05b# cp -v linux.img rootfs.img /mnt/ `linux.img' -> `/mnt/linux.img' `rootfs.img' -> `/mnt/rootfs.img' bash-2.05b# ls -l /mnt total 4700 -rwxr--r-- 1 root root 1020729 Jun 14 05:36 linux.img -rwxr--r-- 1 root root 3788800 Jun 14 05:36 rootfs.img bash-2.05b# umount /mnt
- We now prepare U-Boot to load the
"uMulti"
file (combined Linux kernel and initial ramdisk) from the CF card and boot it:=> setenv fat_args setenv bootargs rw => setenv fat_boot 'run fat_args addip;fatload ide 0:1 200000 linux.img;bootm' => setenv bootcmd run fat_boot
- And finally we try it out:
U-Boot 1.1.3 (Jun 13 2005 - 02:24:00) CPU: XPC86xxxZPnnD4 at 50 MHz: 4 kB I-Cache 4 kB D-Cache FEC present Board: TQM860LDB0A3-T50.202 DRAM: 16 MB FLASH: 8 MB In: serial Out: serial Err: serial Net: SCC ETHERNET, FEC ETHERNET [PRIME] PCMCIA: 3.3V card found: Transcend 256M Fixed Disk Card IDE interface [silicon] [unique] [single] [sleep] [standby] [idle] [low power] Bus 0: OK Device 0: Model: Transcend 256M Firm: 1.1 Ser#: SSSC256M04Z27A25906T Type: Removable Hard Disk Capacity: 244.5 MB = 0.2 GB (500736 x 512) Type "run flash_nfs" to mount root filesystem over NFS Hit any key to stop autoboot: 0 reading linux.img 1025657 bytes read ## Booting image at 00200000 ... Image Name: Linux with Pivot Root Helper Created: 2005-06-13 0:32:41 UTC Image Type: PowerPC Linux Multi-File Image (gzip compressed) Data Size: 1025593 Bytes = 1001.6 kB Load Address: 00000000 Entry Point: 00000000 Contents: Image 0: 787146 Bytes = 768.7 kB Image 1: 238433 Bytes = 232.8 kB Verifying Checksum ... OK Uncompressing Multi-File Image ... OK Loading Ramdisk to 00f3d000, end 00f77361 ... OK Linux version 2.4.25 (wd@xpert) (gcc version 3.3.3 (DENX ELDK 3.1.1 3.3.3-9)) #1 Mon Jun 13 02:32:10 MEST 2005 On node 0 totalpages: 4096 zone(0): 4096 pages. zone(1): 0 pages. zone(2): 0 pages. Kernel command line: rw ip=192.168.3.80:192.168.3.1::255.255.255.0:tqm860l:eth1:off panic=1 Decrementer Frequency = 187500000/60 Calibrating delay loop... 49.86 BogoMIPS ... NET4: Unix domain sockets 1.0/SMP for Linux NET4.0. RAMDISK: Compressed image found at block 0 Freeing initrd memory: 232k freed VFS: Mounted root (ext2 filesystem). Red Hat nash version 4.1.18 starting Mounting /proc filesystem Creating block devices Creating root device Mounting flash card hda: hda1 hda: hda1 losetup for filesystem image Mounting root filesystem image Running pivot_root Freeing unused kernel memory: 60k init BusyBox v0.60.5 (2005.03.07-06:54+0000) Built-in shell (msh) Enter 'help' for a list of built-in commands. # ### Application running ...
Kernel with a Flattened Device Tree Blob
When booting anarch/powerpc
kernel that requires a flattened device tree blob, the above procedure must be slightly modified. Namely, the multi-image file has to include the blob as the thrid image. Here's an example of the
mkimage
command to create it:
mkimage -A ppc -O Linux -T multi -C gzip -n 'Kernel + Pivot Root Helper initrd + FDT blob' -d vmlinux.bin.gz:ramdisk_image.gz:canyonlands.dtb kernel+initrd+blob.img Image Name: Kernel + Pivot Root Helper initrd + FDT blob Created: Fri Sep 14 18:24:29 2007 Image Type: PowerPC Linux Multi-File Image (gzip compressed) Data Size: 2894576 Bytes = 2826.73 kB = 2.76 MB Load Address: 0x00000000 Entry Point: 0x00000000 Contents: Image 0: 1351205 Bytes = 1319 kB = 1 MB Image 1: 1531063 Bytes = 1495 kB = 1 MB Image 2: 12288 Bytes = 12 kB = 0 MB
The newly created file kernel+initrd+blob.img
needs to be copied to the CF card.
9.6. Root File System Selection
Now we know several options for file systems we can use,and know how to create the corresponding images.But how can we decide which one to chose?
For practical purposes in embedded systemsthe following criteria are often essential:
- boot time (i. e. time needed from power on until application code is running)
- flash memory footprint
- RAM memory footprint
- effects on software updates
The following data was measured for the different configurations.All measurements were performed on the same TQM860L board (MPC860 CPU at 50 MHz, 16 MB RAM, 8 MB flash, 256 MB CompactFlash card):
File System Type | Boot Time | Free Mem | Updates | while running |
---|---|---|---|---|
ramdisk | 16.3 sec | 6.58 MB | whole image | yes |
JFFS2 | 21.4 sec | 10.3 MB | per file | only non-active files |
cramfs | 10.8 sec | 10.3 MB | whole image | no |
ext2 (ro) | 9.1 sec | 10.8 MB | whole image | no |
ext2 on CF (ro) | 9.3 sec | 10.9 MB | whole image | no |
File on FAT fs | 11.4 sec | 7.8 MB | whole image | yes |
As you can see, the ramdisk solution is the worst of allin terms of RAM memory footprint;also it takes a pretty long time to boot.However, it is one of the few solutions that allow an in-situupdate while the system is running.
JFFS2 is easy to use as it's a writable file systembut it takes a long time to boot.
A read-only ext2 file system shines when boot time and RAM memoryfootprint are important; you pay for this with an increased flash memoryfootprint.
External flash memory devices like CompactFlash cards or USB memorysticks can be cheap and efficient solutions especially whenlots of data need to be stored or when easy update procedures are required.
Other resources http://man.chinaunix.net/linux/how/Bootdisk-HOWTO-4.html