9.5.4. Root File System on a Read-Only ext2 File System

When storing the root file system in on-board flash memory it seems only natural to look for special flash filesystems like JFFS2, or for other file system types that are designed for such environments like cramfs. It seems to be a bad idea to use a standard ext2 file system because it contains neither any type of wear leveling which is 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 mount read-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 image that can be used as a read-only root file system the following steps are necessary:

1. 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
   

2. 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 the tmpfs 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/*
   

3. Like we did for the ramdisk, we now create an ext2 file system image using the genext2fs 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}
   

4. 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
   ...