Difference between revisions of "Virtual Private Server"

Revision as of 19:45, 24 January 2014

From Wikipedia:Virtual private server:

Virtual private server (VPS) is a term used by Internet hosting services to refer to a virtual machine. The term is used for emphasizing that the virtual machine, although running in software on the same physical computer as other customers' virtual machines, is in many respects functionally equivalent to a separate physical computer, is dedicated to the individual customer's needs, has the privacy of a separate physical computer, and can be configured to run server software.

This article discusses the use of Arch Linux on Virtual Private Servers, and includes some fixes and installation instructions specific to VPSes.

Providers that offer Arch Linux

Installation

KVM

See QEMU#Preparing an (Arch) Linux guest.

OpenVZ

Updating a 2010.05 installation image

These instructions assume you have a 2010.05 image from your VPS provider and you would like to get it updated. The biggest work involves preparing /lib for the symlink upgrade (glibc 2.16, and later filesystem 2013.01).

To start, grab the latest BusyBox from http://busybox.net/downloads/binaries/latest/. This allows you to force glibc (losing /lib temporarily) without losing your OS (BusyBox comes with its own GNU tools which are statically linked).

# wget http://busybox.net/downloads/binaries/latest/busybox-i686
# chmod +x busybox-i686

First, you can get a list of packages that own files in /lib with the following command:

$ pacman -Qo /lib/* | cut -d' ' -f 5 | egrep -v 'glibc' | uniq | xargs

For the current 2010.05 that comes from ibiru's page, these are the packages that were required to be removed for me:

pacman -S acl attr util-linux-ng bzip2 libcap e2fsprogs libgcrypt libgpg-error udev readline ncurses pam pcre popt procps readline shadow e2fsprogs sysfsutils udev util-linux-ng sysvinit coreutils

You may have to remove /lib/udev/devices/loop0 (a simple rm works).

After the upgrade finishes, you must remove any extra empty directories in /lib (/lib/modules is the common offender):

# rm -rf /lib/modules

Install tzdata to fix some dependencies, and remove /etc/profile.d/locale.sh:

# pacman -S tzdata
# rm /etc/profile.d/locale.sh

Remove /var/run (you should have nothing running that matters):

# rm -rf /var/run

Force glibc, which will pull in the latest filesystem package, but will BREAK everything (other than BusyBox):

# pacman -S --force glibc}}

Now, you will have a broken system, so symlink /usr/lib to /lib with BusyBox's ln program:

# ./busybox-i686 ln -s /usr/lib /lib

And you should have a fully functional system where you can now update pacman.

# pacman -S pacman; pacman-key --init; pacman-key --populate archlinux; pacman-db-upgrade; pacman -Syy

Now, update initscripts to get the iproute2 package:

# pacman -S initscripts

Install the makedev package:

pacman -S makedev

Add the following to your /etc/rc.local:

/usr/sbin/MAKEDEV tty
/usr/sbin/MAKEDEV pty

Comment the following lines in /etc/inittab:

#c1:2345:respawn:/sbin/agetty -8 -s 38400 tty1 linux
#c2:2345:respawn:/sbin/agetty -8 -s 38400 tty2 linux
#c3:2345:respawn:/sbin/agetty -8 -s 38400 tty3 linux
#c4:2345:respawn:/sbin/agetty -8 -s 38400 tty4 linux
#c5:2345:respawn:/sbin/agetty -8 -s 38400 tty5 linux
#c6:2345:respawn:/sbin/agetty -8 -s 38400 tty6 linux

Finally, you should be able to upgrade the whole system:

# pacman -Syu

You may run into some issues with krb5 and heimdal, as krb5 no longer provides+conflicts+replaces heimdal (https://projects.archlinux.org/svntogit/packages.git/commit/trunk/PKGBUILD?h=packages/krb5&id=f5e6d77fd14ced15ebf5b6a78a7c76e0db0625f7). The old openssh depends on heimdal (and the new openssh depends on krb5), so force install krb5, then upgrade openssh, then remove heimdal and reinstall krb5.

# pacman -S --force krb5
# pacman -S openssh openssl
# pacman -R heimdal
# pacman -S krb5

Fix syslog-ng. Set the src to unix-dgram("/dev/log") and add --no-caps to both check and run args in /etc/conf.d/syslog-ng.

Make sure your /etc/rc.conf is not messed up with broken network definitions, or else be sure serial access works on your VPS before you reboot.

Moving your VPS from network configuration in rc.conf to netcfg (tested with OpenVZ)

1) Install netcfg

pacman -S netcfg

2) Create a netcfg configuration file /etc/network.d/venet

CONNECTION='ethernet'
DESCRIPTION='VPS venet connection'
INTERFACE='venet0'
IP='static'
IPCFG=(
#default
'addr add 127.0.0.1/32 broadcast 0.0.0.0 dev venet0'
#IPv4 address
'addr add xxx.xxx.xxx.xxx/32 broadcast 0.0.0.0 dev venet0'
#IPv4 route
'route add default dev venet0'
#IPv6 address
'addr add xxxx:xx:xx::x/128 dev venet0'
#IPv6 route
'-6 route add default dev venet0'
)
DNS=('xxx.xxx.xxx.xxx' 'xxx.xxx.xxx.xxx')

3) Edit your netcfg main conf file /etc/conf.d/netcfg

NETWORKS=(venet)
WIRED_INTERFACE="venet0"

4) Try your new setup

rc.d stop network && ip addr flush venet0 && netcfg venet

Your VPS should still be connected and have its IP addresses set correctly. (Check with ip a)

DO NOT proceed to next step if this isn't the case.

5) Make your new setup survive reboots

In the DAEMONS array in /etc/rc.conf, replace network with net-profiles.

Remove all networking information that is in /etc/rc.conf.

reboot

Moving your VPS from initscripts to systemd

This is very similar to a regular arch system, except you probably don't have access to your kernel line.

1) Move from network in rc.conf to netcfg (see above).

2) Install systemd

pacman -S systemd

2 bonus for OpenVZ) Remove kernel core dump pattern since this is blocked by OpenVZ and causes errors

Edit /usr/lib/sysctl.d/coredump.conf, comment out the following line:

#kernel.core_pattern=|/usr/lib/systemd/systemd-coredump %p %u %g %s %t %e

3) Move all configuration from /etc/rc.conf (except the DAEMONS array) to its appropriate location.

See Native configuration and rc.conf for details.

Now your /etc/rc.conf should only contain the DAEMONS array.

4) Install systemd-sysvcompat

pacman -S systemd-sysvcompat

It will ask to replace sysvinit, say yes.

reboot

5) Move daemons from the DAEMONS array in /etc/rc.conf to systemd

See the guide and the daemons list.

If your DAEMONS array is now empty, skip next step.

6) Moving rc.d daemons with no systemd support, example: vzquota

Create a custom systemd service file for vzquota: /etc/systemd/system/newvzquota.service:

[Unit]
Description=Setup vzquota on VPS
ConditionFileIsExecutable=/etc/rc.d/vzquota

[Service]
Type=oneshot
ExecStart=/etc/rc.d/vzquota start
ExecStop=/etc/rc.d/vzquota stop
TimeoutSec=0
StandardInput=tty
RemainAfterExit=yes

[Install]
WantedBy=multi-user.target

Enable this service:

systemctl enable newvzquota.service

Remove vzquota from the DAEMONS array in /etc/rc.conf

Repeat this step to remove all daemons from /etc/rc.conf.

7) Removing /etc/rc.local and /etc/rc.local.shutdown

Write custom .service files to replace functionality in /etc/rc.local and /etc/rc.local.shutdown. You can take a look at /usr/lib/systemd/system/rc-local.service and /usr/lib/systemd/system/rc-local-shutdown.service for inspiration.

8) Removing initscripts

Your /etc/rc.conf file must look like this:

DAEMONS=()

and /etc/rc.local and /etc/rc.local.shutdown must now be empty.

Uninstall initscripts

pacman -R initscripts

reboot

Xen

See Xen#Arch as Xen guest (PVHVM mode) and/or Xen#Arch as Xen guest (PV mode).

Converting OpenStack and Xen components to systemd

There are three components that need to be enabled in systemd when using a VPS based on OpenStack/Xen, such as Rackspace NextGen Cloud. The current version of xe-guest-utilities contains two of these: xe-linux-distribution and xe-daemon.

You will need to create a custom service file for the OpenStack nova-agent, as the current version 0.0.1.37 only comes with a sysvinit start-up script.

/etc/systemd/system/nova-agent.service

[Unit]
Description=nova-agent service
After=xe-daemon.service

[Service]
Environment=LD_LIBRARY_PATH=/usr/share/nova-agent/0.0.1.37/lib
ExecStart=usr/bin/nova-agent -n -l info /usr/share/nova-agent/nova-agent.py

[Install]
WantedBy=multi-user.target

Once these steps are done, you can continue with converting the server from sysvinit to systemd.

Make sure to enable the following services:

# systemctl enable xe-linux-distribution
# systemctl enable xe-daemon
# systemctl enable nova-agent

Troubleshooting

OpenVZ: kernel too old for glibc

Are you on a virtual private server (VPS) with an old kernel & broke your system? Are you using OpenVZ?

Check your kernel version with:

uname -r

If your kernel is older than 2.6.32, you will need a custom version of glibc (because of dependencies in glibc).

Arch Template Used: https://dev.archlinux.org/~ibiru/openvz/2010.05/arch-2010.05-i686-minimal.tar.gz

pacman -U https://dev.archlinux.org/~ibiru/openvz/glibc-vps/i686/glibc-2.16.0-101-i686.pkg.tar.xz

pacman -U https://dev.archlinux.org/~ibiru/openvz/glibc-vps/x86_64/glibc-2.16.0-101-x86_64.pkg.tar.xz

Following similar instructions from DeveloperWiki:usrlib.

Try doing the following to fix it:

1) Edit /etc/pacman.conf and add the following repository ABOVE [core]:

For 32-bit:

[glibc-vps]
Server = https://dev.archlinux.org/~ibiru/openvz/glibc-vps/i686

For 64-bit:

[glibc-vps]
Server = https://dev.archlinux.org/~ibiru/openvz/glibc-vps/x86_64

2) Then run pacman -Syy followed by pacman -Syu. You will be notified to upgrade pacman first.

3) Upgrade the pacman database by running pacman-db-upgrade as root.

4) Edit /etc/pacman.conf.pacnew (new pacman config file) and add the following repository ABOVE [core]:

[glibc-vps]
Server = https://dev.archlinux.org/~ibiru/openvz/glibc-vps/$arch

5) Replace /etc/pacman.conf with /etc/pacman.conf.pacnew (run as root):

# mv /etc/pacman.conf.pacnew /etc/pacman.conf

6) Upgrade your whole system with new packages again pacman -Syu

If you get the following error or a similar error:

initscripts: /etc/profile.d/locale.sh exists in filesystem

Simply delete that file (e.g., rm -f /etc/profile.d/locale.sh), then run pacman -Syu again.

If you get the following error or a similar error:

filesystem: /etc/mtab exists in filesystem

Run pacman -S filesystem --force

If you get the following error or a similar error:

libusb-compat: /usr/bin/libusb-config exists in filesystem

Run pacman -S libusb and then pacman -S libusb-compat

7) Before rebooting, you need to install the makedev package from the official repositories by running pacman -S makedev.

8) Add MAKEDEV to /etc/rc.local:

/usr/sbin/MAKEDEV tty
/usr/sbin/MAKEDEV pty

9) Edit /etc/inittab, comment out the following lines; otherwise, you will see errors in /var/log/errors.log):

#c1:2345:respawn:/sbin/agetty -8 -s 38400 tty1 linux
#c2:2345:respawn:/sbin/agetty -8 -s 38400 tty2 linux
#c3:2345:respawn:/sbin/agetty -8 -s 38400 tty3 linux
#c4:2345:respawn:/sbin/agetty -8 -s 38400 tty4 linux
#c5:2345:respawn:/sbin/agetty -8 -s 38400 tty5 linux
#c6:2345:respawn:/sbin/agetty -8 -s 38400 tty6 linux

10) To enable use of the hostname command, install the package inetutils from the official repositories.

11) Remove disabling of the SysRq key and setup of core dump pattern because this is blocked by OpenVZ and causes errors.

Because sysctl does not use /etc/sysctl.conf any more[1], you must transfer all settings to /etc/sysctl.d/99-sysctl.conf (or any other file in /etc/sysctl.d/; however, do not transfer the following line:

kernel.sysrq = 0

Edit /usr/lib/sysctl.d/coredump.conf and comment out the following line:

#kernel.core_pattern=|/usr/lib/systemd/systemd-coredump %p %u %g %s %t %e

12) Save and reboot.

Enjoy & thank ioni if you happen to be in #archlinux

SSH fails: PTY allocation request failed on channel 0

Some VPSes have an outdated /etc/rc.sysinit. You may be able to log in via serial console or with the following command:

$ ssh root@broken.server '/bin/bash -i'

Then run the following:

# mv /etc/rc.sysinit.pacnew /etc/rc.sysinit
# reboot

Once it is working, you should be able to comment out the udevd_modprobe line in /etc/rc.sysinit to save a bit of RAM the next time you reboot.

If the above does not work, take a look at this guide.