[edit]Background
It is possible to start a diskless machine remotely using Open-iSCSI on the client side. You have to take care of these things:
- create an initrd which will connect the iSCSI node
- configure your tftp server, so that your machine starts properly via PXE/tftp
This HOWTO explains how to set up your system so that it boots a diskless station using PXE/tftp and Open-iSCSI.
[edit]Creating initrd
Below some theory; for practical examples, see
downloads section.
Your initrd has to do the following things:
- load all needed modules, like:
- network card modules (e1000, tg3, mii, 8139too, e100, depending on the card you use etc.)
- filesystem modules (jbd, ext3, depending on the filesystem you use)
- SCSI modules (scsi_mod, sd_mod)
- Open-iSCSI modules (scsi_transport_iscsi, libiscsi, iscsi_tcp)
- mount /proc and /sys
- bring a network interface up
- initiate a iSCSI session with
iscsistart
- mount your filesystem
- switch to a new root
- execute
/sbin/init
from a new root
[edit]Preparing initrd
Note that the below instruction use common tools from your favourite distribution. You may create a similar initrd with busybox/uClibc.
- create a directory for your Open-ISCSI-aware initrd:
mkdir initrd
- create all needed directories:
cd initrd
mkdir -p bin dev etc lib/modules proc sys sysroot
ln -s bin sbin
- copy the following binaries to
bin/
:
bash cat chroot echo ifconfig insmod iscsistart mount sleep
You can do it with a following command:
which bash cat chroot echo ifconfig insmod iscsistart mount sleep | xargs -i{} cp {} bin/
If your distribution doesn't ship iscsistart
(most distros don't), you have to compile open-iscsi yourself - you'll find there iscsistart
.
- copy shared libraries to
lib/
; you will find the shared libraries with ldd
tool, for example:
# ldd bash
linux-gate.so.1 => (0xb7fa6000)
libtermcap.so.2 => /lib/libtermcap.so.2 (0xb7f9d000)
libdl.so.2 => /lib/libdl.so.2 (0xb7f99000)
libc.so.6 => /lib/i686/libc.so.6 (0xb7e6a000)
/lib/ld-linux.so.2 (0xb7fa7000)
- after all binaries and libraries are copied to correct places, verify if you didn't forget anything with
chroot
command - you may get a similar "no name" prompt (it's OK, since there are no users in that initrd), and all binaries should be able to start:
# cd initrd
# chroot .
[I have no name!@localhost /]# mount --help
(...)
- create needed nodes in
dev/
(with mknod
tool, or just copy them from your /dev):
console sda sda1 sda2
- create the
init
file (don't forget to chmod 755 init
); place it in the root of your initrd directory; note we load some modules here - you have to copy the modules to /lib/modules/
in your initrd; here's the example init
file which detects kernel command line parameters like initiator name, target IP etc.:
#!/bin/bash
# Load modules
# You may only need to load the network card modules you will use
# To find module dependencies, you can use "modinfo" tool
echo "Loading jbd.ko module"
insmod /lib/modules/jbd.ko
# On some kernels, you will need mbcache module if it is not in the kernel
echo "Loading mbcache.ko module"
insmod /lib/modules/mbcache.ko
echo "Loading ext3.ko module"
insmod /lib/modules/ext3.ko
echo "Loading libcrc32c.ko module"
insmod /lib/modules/libcrc32c.ko
echo "Loading crc32c.ko module"
insmod /lib/modules/crc32c.ko
# Network card modules - start
echo "Loading mii.ko module"
insmod /lib/modules/mii.ko
echo "Loading 8139too.ko module"
insmod /lib/modules/8139too.ko
echo "Loading e100.ko module"
insmod /lib/modules/e100.ko
echo "Loading e1000.ko module"
insmod /lib/modules/e1000.ko
echo "Loading tulip.ko module"
insmod /lib/modules/tulip.ko
echo "Loading tg3.ko module"
insmod /lib/modules/tg3.ko
# Network card modules - end
echo "Loading scsi_mod.ko module"
insmod /lib/modules/scsi_mod.ko
echo "Loading sd_mod.ko module"
insmod /lib/modules/sd_mod.ko
echo "Loading scsi_transport_iscsi.ko module"
insmod /lib/modules/scsi_transport_iscsi.ko
echo "Loading libiscsi.ko module"
insmod /lib/modules/libiscsi.ko
echo "Loading iscsi_tcp.ko module"
insmod /lib/modules/iscsi_tcp.ko
# Mount /proc and /sys
echo Mounting /proc filesystem
mount -t proc /proc /proc
echo Mounting sysfs
mount -t sysfs /sys /sys
# We need to extract four command line parameters form /proc/cmdline:
# iscsi_i_ip - local initiator IP address/netmask (i.e. 192.168.111.168/255.255.255.192)
# iscsi_i - local initiator name (InitiatorName)
# iscsi_t - remote target name (TargetName)
# iscsi_a - iSCSI target IP address
# Read the kernel cmdline
CMDLINE=$(cat /proc/cmdline)
# Find out IP/NETMASK first
TEMPVAR=${CMDLINE#*iscsi_i_ip=}
ISCSI_I_IP=${TEMPVAR%% *}
# Find InitiatorName
TEMPVAR=${CMDLINE#*iscsi_i=}
ISCSI_I=${TEMPVAR%% *}
# Find TargetName
TEMPVAR=${CMDLINE#*iscsi_t=}
ISCSI_T=${TEMPVAR%% *}
# Find iSCSI target IP address
TEMPVAR=${CMDLINE#*iscsi_a=}
ISCSI_A=${TEMPVAR%% *}
# Bring the network interface up
ifconfig eth0 ${ISCSI_I_IP%/*} netmask ${ISCSI_I_IP#*/}
# Connect the iSCSI drive
iscsistart -i "$ISCSI_I" -t "$ISCSI_T" -g 1 -a $ISCSI_A
mount -o ro /dev/sda2 /sysroot
echo Switching to new root
cd /sysroot
# Uncomment "bash" below, and comment out "exec ..." if you are having booting problems.
# It will give you a bash shell with limited tools.
exec chroot . sh -c 'exec /sbin/init'
#bash
[edit]Creating initrd with cpio
Now that you have all the files ready, create initrd image:
find ./ | cpio -H newc -o > ../initrd.cpio
cd ..
gzip -9 initrd.cpio
mv initrd.cpio.gz initrd.img
Next, copy initrd.img
to your tftp server.
[edit]Configuring tftp server
Add something like that to a pxelinux.cfg/default
file to initiate a remote boot by hand (not recommended):
LABEL remote1
KERNEL remote/vmlinuz-2.6.17-5mdv
APPEND initrd=remote/initrd.img iscsi_i=iqn.2007-01.com.example:server.remote1 iscsi_i_ip=10.1.1.165/255.255.255.192 iscsi_t=iqn.2007-01.com.example:storage.remote1 iscsi_a=10.1.1.180
You have to add these parameters to APPEND
line appropriately:
- initrd - initrd image, for example,
initrd=remote/initrd.img
- iscsi_i_ip - local initiator IP address/netmask (i.e. 192.168.111.168/255.255.255.192), for example,
iscsi_i_ip=10.1.1.165/255.255.255.192
- iscsi_i - local initiator name (InitiatorName), for example,
iscsi_i=iqn.2007-01.com.example:server.remote1
- iscsi_t - remote target name (TargetName), for example,
iscsi_t=iqn.2007-01.com.example:storage.remote1
- iscsi_a - iSCSI target IP address, for example,
iscsi_a=10.1.1.180
It is recommended that you add the above entry to a pxelinux.cfg/aa-bb-cc-dd-ee-ff
file; aa-bb-cc-dd-ee-ff being the MAC address of the PC you want to boot remotely with Open-iSCSI / PXE.
[edit]Downloads
[edit]Additional info / troubleshooting
[edit]Contact
For general Open-iSCSI info, please ask at
open-iscsi mailing list.
If you wish to contact me directly, write me an email: Tomasz Chmielewski <tch (at) wpkg (dot) org>.
Keywords: Open-iSCSI remote boot, diskless, iSCSI, Xen, tftp, PXE, network boot