https://github.com/globalcitizen/lxc-gentoo
The script creates a root filesystem and config file suitable for initializing a Gentoo guest within an LXC (Linux Containers) environment.
Typical startup time on modern hardware (even without an SSD) is under half a second, and as hardware detection and kernel bootstrapping is not required, the init process is largely IO bound (however adding more network interfaces does increase startup latency).
- Don't treat guests as root safe
- Best practice is to be paranoid:
- Drop most capabilities
- Give each guest a dedicated filesystem (eg. separate LVM2 logical block device, ZFS dataset, or loopback-mounted file)
- Do not use UIDs on the guest that intsersect with the host system
- Make sure you never both (1) mount
proc
in a guest that you don't trust, and (2) haveCONFIG_MAGIC_SYSRQ
'Magic SysRq Key' enabled in your kernel (which creates/proc/sysrq-trigger
) ... as this can be abused for denial of service - If you use DHCP be sure to use the default busybox DHCP daemon as your client (to avoid the bash shellshock issues)
- If applicable to your kernel, ensure
sysctl -w kernel.core_pattern=core
. (see http://www.openwall.com/lists/oss-security/2015/04/14/4 for details)
- Recent Linux kernel (>=3.2.x recommended, >=3.7.x actively tested)
http://www.kernel.org/ (Gentoo:
emerge hardened-sources
/emerge gentoo-sources
/emerge vanilla-sources
)- Relevant kernel options enabled (try
lxc-checkconfig
or review the documentation at http://wiki.gentoo.org/wiki/Lxc)
- Relevant kernel options enabled (try
- Recent lxc userspace utilities
(Gentoo:
emerge lxc
) - fuidshift
go get github.com/lxc/lxd && go get github.com/gorilla/websocket git clone https://github.com/lxc/lxd cd lxd/fuidshift make && cp fuidshift /usr/local/bin
While normally run interactively, the script also accepts input from various environment variables.
- interactive:
lxc-gentoo create
- interactive (with environment):
CACHE=/cache lxc-gentoo create
- automated:
lxc-gentoo create -q
- automated (with environment):
CACHE=/cache lxc-gentoo create -q
- automated:
lxc-gentoo create -n test-lvm -u test-lvm.local -a amd64 -q -s 10000000:10000000 -i 192.168.3.99/24 -g 192.168.3.254 -P /usr/portage/tree -B lvm
Available environment variables are as follows.
Property | Environment Variable |
Default Value |
Notes |
Cache Path | $CACHE |
/var/cache/lxc/gentoo | Stores arch/subarch/variant combo specific stage3 tarballs and the portage snapshot. |
Mirror | $MIRROR |
http://distfiles.gentoo.org | Specifies the location from which the stage3 tarball, portage snapshot and metadata should be fetched. |
Stage 3 tarball | $STAGE3_TARBALL |
Specifies the location of a custom stage3 tarball. When this option is present, fetching will be skipped | |
Portage source | $PORTAGE_SOURCE |
path to a custom portage to use. Can be one of:
|
|
LXC Container Name | $NAME |
gentoo | Used by lxc-start, lxc-stop, etc. |
Hostname | $UTSNAME |
gentoo | May be altered by DHCP. |
IPv4 Address | $IPV4 |
dhcp | dhcp is a special value; normally format is x.x.x.x/y. |
IPv4 Gateway | $IPV4_GATEWAY |
auto | auto is a special value; normally format is x.x.x.x/y. Ignored if $IPV4 is dhcp. |
IPv6 Address | $IPV6 |
dhcp | dhcp is a special value; normally format is x/y. |
IPv6 Gateway | $IPV6_GATEWAY |
auto | auto is a special value; normally format is x/y. Ignored if $IPV6 is dhcp. |
Guest Root Password | $GUESTROOTPASS |
Will be phased out soon. | |
Gentoo Architecture | $ARCH |
amd64 | Gentoo architecture code: alpha, amd64, arm, hppa, ia64, ppc, s390, sh, sparc, x86 |
Gentoo Architecture Variant | $ARCHVARIANT |
(none) | Usually none, hardened or hardened+nomultilib |
PGP (GPG/GNUPG) directory | $PGP_DIR |
$HOME/.gnupg | Preferred key directory, or empty string (disable). See notes below. |
lxc.conf Location | $CONFFILE |
${NAME}.conf | Path at which to generate the lxc.conf file, one of:
|
There are 3 possible setups for PGP/GPG (GNU Privacy Guard) signature checking:
- off
PGP_DIR="" lxc-gentoo ...
- on with $HOME/.gnupg as the keys directory
lxc-gentoo ...
PGP_DIR="$HOME/.gnupg" lxc-gentoo ...
- on with random directory
PGP_DIR="/path/to/random/dir" lxc-gentoo ...
GNUPG key setup:
If you have never used GPG before, first generate a GPG keypair:
gpg --full-generate-key
You then need the 'Gentoo Linux Release Engineering (Automated Weekly Release Key)' that can be found at https://wwwold.gentoo.org/proj/en/releng/index.xml
If you wish to use a custom key storage directory, first create it as follows.
Otherwise, GPG will use your default ($HOME/.gnupg
) directory.
# create and use a temporary GPG key storage dir (optional)
PGP_DIR=/path/to/random/dir
mkdir -p ${PGP_DIR}
chmod 0700 ${PGP_DIR}
alias gpg="gpg --homedir ${PGP_DIR}"
Now continue with the key imports.
# Import stage3 signing key (subkeys.pgp.net is flakey)
gpg --keyserver pool.sks-keyservers.net --recv-keys 0xBB572E0E2D182910
# Check fingerprint (2015/04 = 13EB BDBE DE7A 1277 5DFD B1BA BB57 2E0E 2D18 2910)
gpg --fingerprint 0xBB572E0E2D182910
# Trust it
gpg --edit-key 0xBB572E0E2D182910 trust
# If you do not have the portage tree yet: import portage signing key
gpg --keyserver pool.sks-keyservers.net --recv-keys 0xDB6B8C1F96D8BF6D
# Check fingerprint (2015/04 = DCD0 5B71 EAB9 4199 527F 44AC DB6B 8C1F 96D8 BF6D)
gpg --fingerprint 0xDB6B8C1F96D8BF6D
# Trust it
gpg --edit-key 0xDB6B8C1F96D8BF6D trust
Be sure to verify that the keys are actually the right ones (check fingerprint with
friends, ask on IRC #gentoo-releng
, #gentoo
, #gentoo-containers
, #lxccontainers
,
visit https://wwwold.gentoo.org/proj/en/releng/index.xml )
When you're done running lxc-gentoo
, if you used a custom key storage dir, you may
want to reset your GPG alias and environment.
unalias gpg
unset PGP_DIR
LXC guests can have zero or more network interfaces, which can be of various types, and each of which may be configured with zero or more addresses. They may, regardless of the above, be granted access to zero or more external networks, real or virtual.
As is typical of Unix (and Linux networking in particular), this basically means "you can probably achieve anything you set your mind to, but it's not going to be easy".
The lxc-gentoo
script therefore tries to provide a reasonable default for
normal use cases, ie. by configuring guests to use one veth
-type interface
that can be connected to the outside world via iptables
.
Basic connectivity can be established with the following host-side commands:
lxc-start -n guest -f guest.conf
ifconfig guest x.x.x.x
(You should now be able to ping the guest. If not, check yourguest.conf
network configuration versus the host-side configuration. Make sure that both addresses are in the same range and differ, for example the host may be10.10.10.1
and the guest may be10.10.10.2
)
Once you have established basic connectivity, external network connectivity can be established as follows:
sysctl net.ipv4.ip_forward=1
and/orsysctl net.ipv6.ip_forward=1
(Optionally also set these in/etc/systctl.conf
to persist after reboot)iptables -t nat -A POSTROUTING -o outward-interface -j MASQUERADE
(Whereoutward-interface
is the name of the interface that carries traffic to/from the host and the internet, or other destination that you wish to allow the guest to connect to. Different distributions have different ways to persist theseiptables
rules, but you can useiptables-save >some-ruleset
andiptables-restore <some-ruleset
on any distribution)
Alternatively to a pure iptables
-based approach, you may consider
interface bridging. A bridge interface is a layer 2 software-based bridge
on the host to which guest interfaces may be linked. It requires some
configuration. Here is an IPv4 example:
- install the
bridge-utils
package (gentoo:emerge bridge-utils
) brctl addbr br0
(create a bridge called br0)brctl setfd br0 0
(set forward delay of zero for optimisation)ifconfig br0 172.20.0.1 255.255.255.0
(select an address range)brctl addif br0 <guest-interface>
(add guest to bridge)
For further reading, the following resources are recommended:
lxc.conf
man page, ie.man 5 lxc.conf
.man 8 iptables
man 8 brctl
man 8 ifconfig
; or the modern alternativeman 8 ip
The following notes describe LXC-specific network topology design considerations:
- Guest startup times will be higher if DHCP is used. In addition, DHCP use creates a dependency on a valid guest-external DHCP configuration which can compromise portability or reliability when executing in different environments. As such, if you are planning to use your LXC guests for executing what should be reliable, repeated jobs, consider avoiding DHCP. Basically it is nothing but a potential source of failure (eg. address pool exhaustion, DHCP server configuration expectation differences between multiple guests, etc.) and should be removed if your infrastructure can be configured to facilitate it. (KISS principal)
- VLANs have been observed to sometimes come up with unavoidable delays
(depending upon various factors such as spanning tree configuration and
intermediate hardware/software). For this reason they are perhaps best
left for the host system to establish connectivity with and for any LXC
guest access to be provided via the host
iptables
configuration. The ideal setup will depend upon your particular use case. KISS. - There have been bugs regarding relative MAC addresses in other LXC
interface types in the past, which initially caused us to move toward
veth
-style interface configuration. Bear this in mind if moving back!
-
Enable
BINFMT_MISC
support in your kernel. -
emerge
staticqemu
with the relevant architecture enabled inQEMU_USER_TARGETS=""
. Hint: Do this in a native container so you don't have left over static binaries on your system :) -
Use either the
qemu
-provided/etc/init.d/qemu-binfmt
script to set up the binfmt handlers or something of your own. Note that the ARM handler @qemu-binfmt
is broken and you will probably have to replace it with the line found here: https://bugs.gentoo.org/show_bug.cgi?id=407099 -
Copy the statically-linked
qemu-$ARCH
executable into the rootfs (docat /proc/sys/fs/binfmt_misc/$ARCH
to see where).
May 2020
- Add cgroups 2 support (thanks to @bircoph)
January 2018
- Improve documentation for new GPG users
December 2017
- LVM support
- sub uid/gid
- LXC 2.1 suport
February 2017
- Fix comment typo
January 2016
- Add basic versionless ebuild (thanks to @josch09)
- Add Travis CI automated build testing
October 2015
- Add IPv6 support (largely untested)
- Adopt new defaults ('dhcp' for IPv4/IPv6 addresses, and 'auto' for gateways)
April 2015
- Add GPG signature and checksum validation
- Add major kernel bug to security notes
- Improved documentation
February 2015
- Minor update to mirror information parsing to suit new format
- Additional notes on loopback setup (see below)
September 2014
- Cleanup old init system workarounds
- Remove hushlogin default
July 2014
- Hacky FQDN support within
UTSNAME
June 2014
- Fix to
wget
argument handling to improve reliability and performance on developing world or other real / half-broken internet connections
May 2014
- Remove
kmod-static-nodes
fromsysinit
runlevel to avoidopenrc
related error messages during guest startup.
April 2014
- No longer drop the
sys_boot
capability in containers, as this preventsshutdown
orpoweroff
command within the container from properly closing the container, resulting in a hunginit
process and failure to recognize the container state on the host side.
February 2014
- External networking documentation
- Discourage intra-guest dynamic network configuration for portability
- Add
/etc/rc.conf
line:rc_provide="net"
to fix service start issues
January 2014
- Resolve issues downloading stage3 archives
- Set a default, unicode-enabled locale to silence
perl
whinging - Fallback to local cache when offline
- Silence errors for antique OpenRC fixes
- Minor fixes for recent OpenRC releases
- Working defaults for quiet mode (
lxc-gentoo create -q
) - Minor typo fixes
June 2013
- Bashisms
- Cleaner syntax
- Improved error handling
- Speedups
- Better/improved locking
- More full-featured prompting
- Portage tree bind mount support
- Portage workspace now
tmpfs
mounted - More verbose download
- Compress cache at
/var/cache/lxc/gentoo
January 2013
- Deployment of whizz-bang screenshot eyecandy.
- Up to date OpenRC fixes for fast and minimalist
boot (eg. newer OpenRC
net
dependency handling) - Additional boot verbosity with
agetty --noclear
- Fairly significant updates to error handling, which should now be relatively reliable.
- Improved internal and external documentation.
- Explicit inclusion of GPLv3 license text
- Cancel the following point; we have a fairly large stylistic mismatch in addition to our use of GPLv3 and their use of LGPL. I guess that's a good thing in some ways because we can continue to hack freely without being stylistically constrained ;)
- Development of this script will soon be moving to the main lxc utils repository, which has recently moved to github. While it has not yet been committed (a style review is pending vs. other scripts), you can find that repo over here: https://github.com/lxc/lxc
November 2012
- Comments regarding recent kernel JIT spraying vulnerability: http://bit.ly/T9CkqJ
- Various contributed minor improvements around locking, indentation, shell syntax, etc.
- Don't drop
CAP_FOWNER
capability, as it breaks portage's ability to chown. - Don't create
/etc/init.d/net.eth0
unless DHCP is specified.
October 2012
- Migrate stage3 URL from
ARCH
toSUBARCH
basis, as per Gentoo Release Guidelines.
September 2012
- Default network config has changed. Instead
of assuming a bridge setup, we use simpler
veth
based tunnels direct to the host, which now appear as ''guestname'' in the host's interface list. (Also resolves an apparently outstanding bug related to random MAC assignment, see http://bit.ly/QWAkOy ) - Generated guests now attempt to aggressively
drop capabilities (
man 7 capabilities
) in a bid to plug known security issues, also to pre-mount/proc
and remove/sys
for the same purpose. (See also: http://bit.ly/SSDbY0 ) - Add DHCP support
sshd
setup code dropped as out of scope- More OpenRC related fixes for faster startup.
- Various minor updates
If you do not generate your guest on a dedicated
filesystem and/or block device then you are very
likely to encounter inode exhaustion on many
default extN
-class filesystems. Therefore, do
create a new device. A good modern solution is
to use a new ZFS dataset (man zfs
and/or see
http://zfsonlinux.org) or an LVM2 logical volume
(man lvcreate
and/or see http://sourceware.org/lvm2/).
However, ZFS or LVM2 are not always available. You can achieve a similar, more portable but lower performance and storage efficiency result with the nearly always available Linux loopback device driver.
First, create a 1024MB (for example) virtual block device image file.
# dd if=/dev/zero of=myguest.image bs=1MiB count=1024
Second, manually request the generation of a loopback device in order to facilitate the initial process of filesystem creation.
# losetup --show -f myguest.image
/dev/loop0
Third, create an appropriate filesystem.
# mkfs -t ext4 /dev/loop0
Finally, detach the device.
# losetup --detach /dev/loop0
You can now mount the image in which to store your guest as follows:
# mkdir /mnt/myguest
# mount -o loop myguest.image /mnt/myguest
# cd /mnt/myguest
# /path/to/lxc-gentoo create
The project was originally hosted at... http://sourceforge.net/projects/lxc-gentoo/
It was moved to github by Julien Sanchez at: https://github.com/gentooboontoo/lxc-gentoo
This was then forked again by the original author in order to move project hosting to github. https://github.com/globalcitizen/lxc-gentoo
Since then it has seen contributions from many parties.