LXC is the well-known and heavily tested low-level Linux
container runtime. It is in active development since 2008 and has
proven itself in critical production environments world-wide.
Some of its core contributors are the same people that helped to
implement various well-known containerization features inside the
Linux kernel.
LXC's main focus is system containers. That is, containers which
offer an environment as close as possible as the one you'd get
from a VM but without the overhead that comes with running a
separate kernel and simulating all the hardware.
This is achieved through a combination of kernel security
features such as namespaces, mandatory access control and control
groups.
LXC has support for unprivileged containers. Unprivileged
containers are containers that are run without any privilege.
This requires support for user namespaces in the kernel that the
container is run on. LXC was the first runtime to support
unprivileged containers after user namespaces were merged into
the mainline kernel.
In essence, user namespaces isolate given sets of UIDs and GIDs.
This is achieved by establishing a mapping between a range of
UIDs and GIDs on the host to a different (unprivileged) range of
UIDs and GIDs in the container. The kernel will translate this
mapping in such a way that inside the container all UIDs and GIDs
appear as you would expect from the host whereas on the host
these UIDs and GIDs are in fact unprivileged. For example, a
process running as UID and GID 0 inside the container might
appear as UID and GID 100000 on the host. The implementation and
working details can be gathered from the corresponding user
namespace man page. UID and GID mappings can be defined with the
lxc.idmap key.
Linux containers are defined with a simple configuration file.
Each option in the configuration file has the form key = value
fitting in one line. The "#" character means the line is a
comment. List options, like capabilities and cgroups options, can
be used with no value to clear any previously defined values of
that option.
LXC namespaces configuration keys use single dots. This means
complex configuration keys such as lxc.net.0 expose various
subkeys such as lxc.net.0.type, lxc.net.0.link,
lxc.net.0.ipv6.address, and others for even more fine-grained
configuration.
CONFIGURATION
In order to ease administration of multiple related containers,
it is possible to have a container configuration file cause
another file to be loaded. For instance, network configuration
can be defined in one common file which is included by multiple
containers. Then, if the containers are moved to another host,
only one file may need to be updated.
lxc.include
Specify the file to be included. The included file must be
in the same valid lxc configuration file format.
ARCHITECTURE
Allows one to set the architecture for the container. For
example, set a 32bits architecture for a container running 32bits
binaries on a 64bits host. This fixes the container scripts which
rely on the architecture to do some work like downloading the
packages.
lxc.arch
Specify the architecture for the container.
Some valid options are x86, i686, x86_64, amd64
HOSTNAME
The utsname section defines the hostname to be set for the
container. That means the container can set its own hostname
without changing the one from the system. That makes the hostname
private for the container.
lxc.uts.name
specify the hostname for the container
HALT SIGNAL
Allows one to specify signal name or number sent to the
container's init process to cleanly shutdown the container.
Different init systems could use different signals to perform
clean shutdown sequence. This option allows the signal to be
specified in kill(1) fashion, e.g. SIGPWR, SIGRTMIN+14,
SIGRTMAX-10 or plain number. The default signal is SIGPWR.
lxc.signal.halt
specify the signal used to halt the container
REBOOT SIGNAL
Allows one to specify signal name or number to reboot the
container. This option allows signal to be specified in kill(1)
fashion, e.g. SIGTERM, SIGRTMIN+14, SIGRTMAX-10 or plain number.
The default signal is SIGINT.
lxc.signal.reboot
specify the signal used to reboot the container
STOP SIGNAL
Allows one to specify signal name or number to forcibly shutdown
the container. This option allows signal to be specified in
kill(1) fashion, e.g. SIGKILL, SIGRTMIN+14, SIGRTMAX-10 or plain
number. The default signal is SIGKILL.
lxc.signal.stop
specify the signal used to stop the container
INIT COMMAND
Sets the command to use as the init system for the containers.
lxc.execute.cmd
Absolute path from container rootfs to the binary to run
by default. This mostly makes sense for lxc-execute.
lxc.init.cmd
Absolute path from container rootfs to the binary to use
as init. This mostly makes sense for lxc-start. Default is
/sbin/init.
INIT WORKING DIRECTORY
Sets the absolute path inside the container as the working
directory for the containers. LXC will switch to this directory
before executing init.
lxc.init.cwd
Absolute path inside the container to use as the working
directory.
INIT ID
Sets the UID/GID to use for the init system, and subsequent
commands. Note that using a non-root UID when booting a system
container will likely not work due to missing privileges. Setting
the UID/GID is mostly useful when running application containers.
Defaults to: UID(0), GID(0)
lxc.init.uid
UID to use for init.
lxc.init.gid
GID to use for init.
PROC
Configure proc filesystem for the container.
lxc.proc.[proc file name]
Specify the proc file name to be set. The file names
available are those listed under /proc/PID/. Example:
lxc.proc.oom_score_adj = 10
EPHEMERAL
Allows one to specify whether a container will be destroyed on
shutdown.
lxc.ephemeral
The only allowed values are 0 and 1. Set this to 1 to
destroy a container on shutdown.
NETWORK
The network section defines how the network is virtualized in the
container. The network virtualization acts at layer two. In order
to use the network virtualization, parameters must be specified
to define the network interfaces of the container. Several
virtual interfaces can be assigned and used in a container even
if the system has only one physical network interface.
lxc.net
may be used without a value to clear all previous network
options.
lxc.net.[i].type
specify what kind of network virtualization to be used for
the container. Must be specified before any other
option(s) on the net device. Multiple networks can be
specified by using an additional index i after all
lxc.net.* keys. For example, lxc.net.0.type = veth and
lxc.net.1.type = veth specify two different networks of
the same type. All keys sharing the same index i will be
treated as belonging to the same network. For example,
lxc.net.0.link = br0 will belong to lxc.net.0.type.
Currently, the different virtualization types can be:
none: will cause the container to share the host's network
namespace. This means the host network devices are usable
in the container. It also means that if both the container
and host have upstart as init, 'halt' in a container (for
instance) will shut down the host. Note that unprivileged
containers do not work with this setting due to an
inability to mount sysfs. An unsafe workaround would be to
bind mount the host's sysfs.
empty: will create only the loopback interface.
veth: a virtual ethernet pair device is created with one
side assigned to the container and the other side on the
host. lxc.net.[i].veth.mode specifies the mode the veth
parent will use on the host. The accepted modes are
bridge and router. The mode defaults to bridge if not
specified. In bridge mode the host side is attached to a
bridge specified by the lxc.net.[i].link option. If the
bridge link is not specified, then the veth pair device
will be created but not attached to any bridge.
Otherwise, the bridge has to be created on the system
before starting the container. lxc won't handle any
configuration outside of the container. In router mode
static routes are created on the host for the container's
IP addresses pointing to the host side veth interface.
Additionally Proxy ARP and Proxy NDP entries are added on
the host side veth interface for the gateway IPs defined
in the container to allow the container to reach the host.
By default, lxc chooses a name for the network device
belonging to the outside of the container, but if you wish
to handle this name yourselves, you can tell lxc to set a
specific name with the lxc.net.[i].veth.pair option
(except for unprivileged containers where this option is
ignored for security reasons). Static routes can be added
on the host pointing to the container using the
lxc.net.[i].veth.ipv4.route and
lxc.net.[i].veth.ipv6.route options. Several lines
specify several routes. The route is in format x.y.z.t/m,
eg. 192.168.1.0/24. In bridge mode untagged VLAN
membership can be set with the lxc.net.[i].veth.vlan.id
option. It accepts a special value of 'none' indicating
that the container port should be removed from the
bridge's default untagged VLAN. The
lxc.net.[i].veth.vlan.tagged.id option can be specified
multiple times to set the container's bridge port
membership to one or more tagged VLANs.
vlan: a vlan interface is linked with the interface
specified by the lxc.net.[i].link and assigned to the
container. The vlan identifier is specified with the
option lxc.net.[i].vlan.id.
macvlan: a macvlan interface is linked with the interface
specified by the lxc.net.[i].link and assigned to the
container. lxc.net.[i].macvlan.mode specifies the mode
the macvlan will use to communicate between different
macvlan on the same upper device. The accepted modes are
private, vepa, bridge and passthru. In private mode, the
device never communicates with any other device on the
same upper_dev (default). In vepa mode, the new Virtual
Ethernet Port Aggregator (VEPA) mode, it assumes that the
adjacent bridge returns all frames where both source and
destination are local to the macvlan port, i.e. the bridge
is set up as a reflective relay. Broadcast frames coming
in from the upper_dev get flooded to all macvlan
interfaces in VEPA mode, local frames are not delivered
locally. In bridge mode, it provides the behavior of a
simple bridge between different macvlan interfaces on the
same port. Frames from one interface to another one get
delivered directly and are not sent out externally.
Broadcast frames get flooded to all other bridge ports and
to the external interface, but when they come back from a
reflective relay, we don't deliver them again. Since we
know all the MAC addresses, the macvlan bridge mode does
not require learning or STP like the bridge module does.
In passthru mode, all frames received by the physical
interface are forwarded to the macvlan interface. Only one
macvlan interface in passthru mode is possible for one
physical interface.
ipvlan: an ipvlan interface is linked with the interface
specified by the lxc.net.[i].link and assigned to the
container. lxc.net.[i].ipvlan.mode specifies the mode the
ipvlan will use to communicate between different ipvlan on
the same upper device. The accepted modes are l3, l3s and
l2. It defaults to l3 mode. In l3 mode TX processing up
to L3 happens on the stack instance attached to the
dependent device and packets are switched to the stack
instance of the parent device for the L2 processing and
routing from that instance will be used before packets are
queued on the outbound device. In this mode the dependent
devices will not receive nor can send multicast /
broadcast traffic. In l3s mode TX processing is very
similar to the L3 mode except that iptables (conn-
tracking) works in this mode and hence it is L3-symmetric
(L3s). This will have slightly less performance but that
shouldn't matter since you are choosing this mode over
plain-L3 mode to make conn-tracking work. In l2 mode TX
processing happens on the stack instance attached to the
dependent device and packets are switched and queued to
the parent device to send devices out. In this mode the
dependent devices will RX/TX multicast and broadcast (if
applicable) as well. lxc.net.[i].ipvlan.isolation
specifies the isolation mode. The accepted isolation
values are bridge, private and vepa. It defaults to
bridge. In bridge isolation mode dependent devices can
cross-talk among themselves apart from talking through the
parent device. In private isolation mode the port is set
in private mode. i.e. port won't allow cross
communication between dependent devices. In vepa
isolation mode the port is set in VEPA mode. i.e. port
will offload switching functionality to the external
entity as described in 802.1Qbg.
phys: an already existing interface specified by the
lxc.net.[i].link is assigned to the container.
lxc.net.[i].flags
Specify an action to do for the network.
up: activates the interface.
lxc.net.[i].link
Specify the interface to be used for real network traffic.
lxc.net.[i].l2proxy
Controls whether layer 2 IP neighbour proxy entries will
be added to the lxc.net.[i].link interface for the IP
addresses of the container. Can be set to 0 or 1.
Defaults to 0. When used with IPv4 addresses, the
following sysctl values need to be set:
net.ipv4.conf.[link].forwarding=1 When used with IPv6
addresses, the following sysctl values need to be set:
net.ipv6.conf.[link].proxy_ndp=1
net.ipv6.conf.[link].forwarding=1
lxc.net.[i].mtu
Specify the maximum transfer unit for this interface.
lxc.net.[i].name
The interface name is dynamically allocated, but if
another name is needed because the configuration files
being used by the container use a generic name, eg. eth0,
this option will rename the interface in the container.
lxc.net.[i].hwaddr
The interface mac address is dynamically allocated by
default to the virtual interface, but in some cases, this
is needed to resolve a mac address conflict or to always
have the same link-local ipv6 address. Any "x" in address
will be replaced by random value, this allows setting
hwaddr templates.
lxc.net.[i].ipv4.address
Specify the ipv4 address to assign to the virtualized
interface. Several lines specify several ipv4 addresses.
The address is in format x.y.z.t/m, eg. 192.168.1.123/24.
You can optionally specify the broadcast address after the
IP adress, e.g. 192.168.1.123/24 255.255.255.255.
Otherwise it is automatically calculated from the IP
address.
lxc.net.[i].ipv4.gateway
Specify the ipv4 address to use as the gateway inside the
container. The address is in format x.y.z.t, eg.
192.168.1.123. Can also have the special value auto,
which means to take the primary address from the bridge
interface (as specified by the lxc.net.[i].link option)
and use that as the gateway. auto is only available when
using the veth, macvlan and ipvlan network types. Can
also have the special value of dev, which means to set the
default gateway as a device route. This is primarily for
use with layer 3 network modes, such as IPVLAN.
lxc.net.[i].ipv6.address
Specify the ipv6 address to assign to the virtualized
interface. Several lines specify several ipv6 addresses.
The address is in format x::y/m, eg.
2003:db8:1:0:214:1234:fe0b:3596/64
lxc.net.[i].ipv6.gateway
Specify the ipv6 address to use as the gateway inside the
container. The address is in format x::y, eg.
2003:db8:1:0::1 Can also have the special value auto,
which means to take the primary address from the bridge
interface (as specified by the lxc.net.[i].link option)
and use that as the gateway. auto is only available when
using the veth, macvlan and ipvlan network types. Can
also have the special value of dev, which means to set the
default gateway as a device route. This is primarily for
use with layer 3 network modes, such as IPVLAN.
lxc.net.[i].script.up
Add a configuration option to specify a script to be
executed after creating and configuring the network used
from the host side.
In addition to the information available to all hooks. The
following information is provided to the script:
• LXC_HOOK_TYPE: the hook type. This is either 'up' or
'down'.
• LXC_HOOK_SECTION: the section type 'net'.
• LXC_NET_TYPE: the network type. This is one of the valid
network types listed here (e.g. 'vlan', 'macvlan',
'ipvlan', 'veth').
• LXC_NET_PARENT: the parent device on the host. This is
only set for network types 'mavclan', 'veth', 'phys'.
• LXC_NET_PEER: the name of the peer device on the host.
This is only set for 'veth' network types. Note that
this information is only available when lxc.hook.version
is set to 1.
Whether this information is provided in the form of environment
variables or as arguments to the script depends on the value of
lxc.hook.version. If set to 1 then information is provided in the
form of environment variables. If set to 0 information is
provided as arguments to the script.
Standard output from the script is logged at debug level.
Standard error is not logged, but can be captured by the hook
redirecting its standard error to standard output.
lxc.net.[i].script.down
Add a configuration option to specify a script to be
executed before destroying the network used from the host
side.
In addition to the information available to all hooks. The
following information is provided to the script:
• LXC_HOOK_TYPE: the hook type. This is either 'up' or
'down'.
• LXC_HOOK_SECTION: the section type 'net'.
• LXC_NET_TYPE: the network type. This is one of the valid
network types listed here (e.g. 'vlan', 'macvlan',
'ipvlan', 'veth').
• LXC_NET_PARENT: the parent device on the host. This is
only set for network types 'mavclan', 'veth', 'phys'.
• LXC_NET_PEER: the name of the peer device on the host.
This is only set for 'veth' network types. Note that
this information is only available when lxc.hook.version
is set to 1.
Whether this information is provided in the form of environment
variables or as arguments to the script depends on the value of
lxc.hook.version. If set to 1 then information is provided in the
form of environment variables. If set to 0 information is
provided as arguments to the script.
Standard output from the script is logged at debug level.
Standard error is not logged, but can be captured by the hook
redirecting its standard error to standard output.
NEW PSEUDO TTY INSTANCE (DEVPTS)
For stricter isolation the container can have its own private
instance of the pseudo tty.
lxc.pty.max
If set, the container will have a new pseudo tty instance,
making this private to it. The value specifies the maximum
number of pseudo ttys allowed for a pty instance (this
limitation is not implemented yet).
CONTAINER SYSTEM CONSOLE
If the container is configured with a root filesystem and the
inittab file is setup to use the console, you may want to specify
where the output of this console goes.
lxc.console.buffer.size
Setting this option instructs liblxc to allocate an in-
memory ringbuffer. The container's console output will be
written to the ringbuffer. Note that ringbuffer must be at
least as big as a standard page size. When passed a value
smaller than a single page size liblxc will allocate a
ringbuffer of a single page size. A page size is usually
4KB. The keyword 'auto' will cause liblxc to allocate a
ringbuffer of 128KB. When manually specifying a size for
the ringbuffer the value should be a power of 2 when
converted to bytes. Valid size prefixes are 'KB', 'MB',
'GB'. (Note that all conversions are based on multiples of
1024. That means 'KB' == 'KiB', 'MB' == 'MiB', 'GB' ==
'GiB'. Additionally, the case of the suffix is ignored,
i.e. 'kB', 'KB' and 'Kb' are treated equally.)
lxc.console.size
Setting this option instructs liblxc to place a limit on
the size of the console log file specified in
lxc.console.logfile. Note that size of the log file must
be at least as big as a standard page size. When passed a
value smaller than a single page size liblxc will set the
size of log file to a single page size. A page size is
usually 4KB. The keyword 'auto' will cause liblxc to
place a limit of 128KB on the log file. When manually
specifying a size for the log file the value should be a
power of 2 when converted to bytes. Valid size prefixes
are 'KB', 'MB', 'GB'. (Note that all conversions are based
on multiples of 1024. That means 'KB' == 'KiB', 'MB' ==
'MiB', 'GB' == 'GiB'. Additionally, the case of the
suffix is ignored, i.e. 'kB', 'KB' and 'Kb' are treated
equally.) If users want to mirror the console ringbuffer
on disk they should set lxc.console.size equal to
lxc.console.buffer.size.
lxc.console.logfile
Specify a path to a file where the console output will be
written. Note that in contrast to the on-disk ringbuffer
logfile this file will keep growing potentially filling up
the users disks if not rotated and deleted. This problem
can also be avoided by using the in-memory ringbuffer
options lxc.console.buffer.size and
lxc.console.buffer.logfile.
lxc.console.rotate
Whether to rotate the console logfile specified in
lxc.console.logfile. Users can send an API request to
rotate the logfile. Note that the old logfile will have
the same name as the original with the suffix ".1"
appended. Users wishing to prevent the console log file
from filling the disk should rotate the logfile and delete
it if unneeded. This problem can also be avoided by using
the in-memory ringbuffer options lxc.console.buffer.size
and lxc.console.buffer.logfile.
lxc.console.path
Specify a path to a device to which the console will be
attached. The keyword 'none' will simply disable the
console. Note, when specifying 'none' and creating a
device node for the console in the container at
/dev/console or bind-mounting the hosts's /dev/console
into the container at /dev/console the container will have
direct access to the hosts's /dev/console. This is
dangerous when the container has write access to the
device and should thus be used with caution.
CONSOLE THROUGH THE TTYS
This option is useful if the container is configured with a root
filesystem and the inittab file is setup to launch a getty on the
ttys. The option specifies the number of ttys to be available for
the container. The number of gettys in the inittab file of the
container should not be greater than the number of ttys specified
in this option, otherwise the excess getty sessions will die and
respawn indefinitely giving annoying messages on the console or
in /var/log/messages.
lxc.tty.max
Specify the number of tty to make available to the
container.
CONSOLE DEVICES LOCATION
LXC consoles are provided through Unix98 PTYs created on the host
and bind-mounted over the expected devices in the container. By
default, they are bind-mounted over /dev/console and /dev/ttyN.
This can prevent package upgrades in the guest. Therefore you can
specify a directory location (under /dev under which LXC will
create the files and bind-mount over them. These will then be
symbolically linked to /dev/console and /dev/ttyN. A package
upgrade can then succeed as it is able to remove and replace the
symbolic links.
lxc.tty.dir
Specify a directory under /dev under which to create the
container console devices. Note that LXC will move any
bind-mounts or device nodes for /dev/console into this
directory.
/DEV DIRECTORY
By default, lxc creates a few symbolic links
(fd,stdin,stdout,stderr) in the container's /dev directory but
does not automatically create device node entries. This allows
the container's /dev to be set up as needed in the container
rootfs. If lxc.autodev is set to 1, then after mounting the
container's rootfs LXC will mount a fresh tmpfs under /dev
(limited to 500K by default, unless defined in
lxc.autodev.tmpfs.size) and fill in a minimal set of initial
devices. This is generally required when starting a container
containing a "systemd" based "init" but may be optional at other
times. Additional devices in the containers /dev directory may be
created through the use of the lxc.hook.autodev hook.
lxc.autodev
Set this to 0 to stop LXC from mounting and populating a
minimal /dev when starting the container.
lxc.autodev.tmpfs.size
Set this to define the size of the /dev tmpfs. The
default value is 500000 (500K). If the parameter is used
but without value, the default value is used.
MOUNT POINTS
The mount points section specifies the different places to be
mounted. These mount points will be private to the container and
won't be visible by the processes running outside of the
container. This is useful to mount /etc, /var or /home for
examples.
NOTE - LXC will generally ensure that mount targets and relative
bind-mount sources are properly confined under the container
root, to avoid attacks involving over-mounting host directories
and files. (Symbolic links in absolute mount sources are ignored)
However, if the container configuration first mounts a directory
which is under the control of the container user, such as
/home/joe, into the container at some path, and then mounts under
path, then a TOCTTOU attack would be possible where the container
user modifies a symbolic link under their home directory at just
the right time.
lxc.mount.fstab
specify a file location in the fstab format, containing
the mount information. The mount target location can and
in most cases should be a relative path, which will become
relative to the mounted container root. For instance,
proc proc proc nodev,noexec,nosuid 0 0
Will mount a proc filesystem under the container's /proc,
regardless of where the root filesystem comes from. This
is resilient to block device backed filesystems as well as
container cloning.
Note that when mounting a filesystem from an image file or
block device the third field (fs_vfstype) cannot be auto
as with mount(8) but must be explicitly specified.
lxc.mount.entry
Specify a mount point corresponding to a line in the fstab
format. Moreover lxc supports mount propagation, such as
rshared or rprivate, and adds three additional mount
options. optional don't fail if mount does not work.
create=dir or create=file to create dir (or file) when the
point will be mounted. relative source path is taken to
be relative to the mounted container root. For instance,
dev/null proc/kcore none bind,relative 0 0
Will expand dev/null to ${LXC_ROOTFS_MOUNT}/dev/null, and
mount it to proc/kcore inside the container.
lxc.mount.auto
specify which standard kernel file systems should be
automatically mounted. This may dramatically simplify the
configuration. The file systems are:
• proc:mixed (or proc): mount /proc as read-write, but
remount /proc/sys and /proc/sysrq-trigger read-only for
security / container isolation purposes.
• proc:rw: mount /proc as read-write
• sys:mixed (or sys): mount /sys as read-only but with
/sys/devices/virtual/net writable.
• sys:ro: mount /sys as read-only for security / container
isolation purposes.
• sys:rw: mount /sys as read-write
• cgroup:mixed: Mount a tmpfs to /sys/fs/cgroup, create
directories for all hierarchies to which the container
is added, create subdirectories in those hierarchies
with the name of the cgroup, and bind-mount the
container's own cgroup into that directory. The
container will be able to write to its own cgroup
directory, but not the parents, since they will be
remounted read-only.
• cgroup:mixed:force: The force option will cause LXC to
perform the cgroup mounts for the container under all
circumstances. Otherwise it is similar to cgroup:mixed.
This is mainly useful when the cgroup namespaces are
enabled where LXC will normally leave mounting cgroups
to the init binary of the container since it is
perfectly safe to do so.
• cgroup:ro: similar to cgroup:mixed, but everything will
be mounted read-only.
• cgroup:ro:force: The force option will cause LXC to
perform the cgroup mounts for the container under all
circumstances. Otherwise it is similar to cgroup:ro.
This is mainly useful when the cgroup namespaces are
enabled where LXC will normally leave mounting cgroups
to the init binary of the container since it is
perfectly safe to do so.
• cgroup:rw: similar to cgroup:mixed, but everything will
be mounted read-write. Note that the paths leading up to
the container's own cgroup will be writable, but will
not be a cgroup filesystem but just part of the tmpfs of
/sys/fs/cgroup
• cgroup:rw:force: The force option will cause LXC to
perform the cgroup mounts for the container under all
circumstances. Otherwise it is similar to cgroup:rw.
This is mainly useful when the cgroup namespaces are
enabled where LXC will normally leave mounting cgroups
to the init binary of the container since it is
perfectly safe to do so.
• cgroup (without specifier): defaults to cgroup:rw if the
container retains the CAP_SYS_ADMIN capability,
cgroup:mixed otherwise.
• cgroup-full:mixed: mount a tmpfs to /sys/fs/cgroup,
create directories for all hierarchies to which the
container is added, bind-mount the hierarchies from the
host to the container and make everything read-only
except the container's own cgroup. Note that compared to
cgroup, where all paths leading up to the container's
own cgroup are just simple directories in the underlying
tmpfs, here /sys/fs/cgroup/$hierarchy will contain the
host's full cgroup hierarchy, albeit read-only outside
the container's own cgroup. This may leak quite a bit
of information into the container.
• cgroup-full:mixed:force: The force option will cause LXC
to perform the cgroup mounts for the container under all
circumstances. Otherwise it is similar to cgroup-
full:mixed. This is mainly useful when the cgroup
namespaces are enabled where LXC will normally leave
mounting cgroups to the init binary of the container
since it is perfectly safe to do so.
• cgroup-full:ro: similar to cgroup-full:mixed, but
everything will be mounted read-only.
• cgroup-full:ro:force: The force option will cause LXC to
perform the cgroup mounts for the container under all
circumstances. Otherwise it is similar to cgroup-
full:ro. This is mainly useful when the cgroup
namespaces are enabled where LXC will normally leave
mounting cgroups to the init binary of the container
since it is perfectly safe to do so.
• cgroup-full:rw: similar to cgroup-full:mixed, but
everything will be mounted read-write. Note that in this
case, the container may escape its own cgroup. (Note
also that if the container has CAP_SYS_ADMIN support and
can mount the cgroup filesystem itself, it may do so
anyway.)
• cgroup-full:rw:force: The force option will cause LXC to
perform the cgroup mounts for the container under all
circumstances. Otherwise it is similar to cgroup-
full:rw. This is mainly useful when the cgroup
namespaces are enabled where LXC will normally leave
mounting cgroups to the init binary of the container
since it is perfectly safe to do so.
• cgroup-full (without specifier): defaults to cgroup-
full:rw if the container retains the CAP_SYS_ADMIN
capability, cgroup-full:mixed otherwise.
If cgroup namespaces are enabled, then any cgroup auto-mounting
request will be ignored, since the container can mount the
filesystems itself, and automounting can confuse the container
init.
Note that if automatic mounting of the cgroup filesystem is
enabled, the tmpfs under /sys/fs/cgroup will always be mounted
read-write (but for the :mixed and :ro cases, the individual
hierarchies, /sys/fs/cgroup/$hierarchy, will be read-only). This
is in order to work around a quirk in Ubuntu's mountall(8)
command that will cause containers to wait for user input at boot
if /sys/fs/cgroup is mounted read-only and the container can't
remount it read-write due to a lack of CAP_SYS_ADMIN.
Examples:
lxc.mount.auto = proc sys cgroup
lxc.mount.auto = proc:rw sys:rw cgroup-full:rw
ROOT FILE SYSTEM
The root file system of the container can be different than that
of the host system.
lxc.rootfs.path
specify the root file system for the container. It can be
an image file, a directory or a block device. If not
specified, the container shares its root file system with
the host.
For directory or simple block-device backed containers, a
pathname can be used. If the rootfs is backed by a nbd
device, then nbd:file:1 specifies that file should be
attached to a nbd device, and partition 1 should be
mounted as the rootfs. nbd:file specifies that the nbd
device itself should be mounted. overlayfs:/lower:/upper
specifies that the rootfs should be an overlay with /upper
being mounted read-write over a read-only mount of /lower.
For overlay multiple /lower directories can be specified.
loop:/file tells lxc to attach /file to a loop device and
mount the loop device.
lxc.rootfs.mount
where to recursively bind lxc.rootfs.path before pivoting.
This is to ensure success of the pivot_root(8) syscall.
Any directory suffices, the default should generally work.
lxc.rootfs.options
Specify extra mount options to use when mounting the
rootfs. The format of the mount options corresponds to
the format used in fstab. In addition, LXC supports the
custom idmap= mount option. This option can be used to
tell LXC to create an idmapped mount for the container's
rootfs. This is useful when the user doesn't want to
recursively chown the rootfs of the container to match the
idmapping of the user namespace the container is going to
use. Instead an idmapped mount can be used to handle this.
The argument for idmap= can either be a path pointing to a
user namespace file that LXC will open and use to idmap
the rootfs or the special value "container" which will
instruct LXC to use the container's user namespace to
idmap the rootfs.
lxc.rootfs.managed
Set this to 0 to indicate that LXC is not managing the
container storage, then LXC will not modify the container
storage. The default is 1.
CONTROL GROUPS ("CGROUPS")
The control group section contains the configuration for the
different subsystem. lxc does not check the correctness of the
subsystem name. This has the disadvantage of not detecting
configuration errors until the container is started, but has the
advantage of permitting any future subsystem.
The kernel implementation of cgroups has changed significantly
over the years. With Linux 4.5 support for a new cgroup
filesystem was added usually referred to as "cgroup2" or "unified
hierarchy". Since then the old cgroup filesystem is usually
referred to as "cgroup1" or the "legacy hierarchies". Please see
the cgroups manual page for a detailed explanation of the
differences between the two versions.
LXC distinguishes settings for the legacy and the unified
hierarchy by using different configuration key prefixes. To alter
settings for controllers in a legacy hierarchy the key prefix
lxc.cgroup. must be used and in order to alter the settings for a
controller in the unified hierarchy the lxc.cgroup2. key must be
used. Note that LXC will ignore lxc.cgroup. settings on systems
that only use the unified hierarchy. Conversely, it will ignore
lxc.cgroup2. options on systems that only use legacy hierarchies.
At its core a cgroup hierarchy is a way to hierarchically
organize processes. Usually a cgroup hierarchy will have one or
more "controllers" enabled. A "controller" in a cgroup hierarchy
is usually responsible for distributing a specific type of system
resource along the hierarchy. Controllers include the "pids"
controller, the "cpu" controller, the "memory" controller and
others. Some controllers however do not fall into the category of
distributing a system resource, instead they are often referred
to as "utility" controllers. One utility controller is the
device controller. Instead of distributing a system resource it
allows one to manage device access.
In the legacy hierarchy the device controller was implemented
like most other controllers as a set of files that could be
written to. These files where named "devices.allow" and
"devices.deny". The legacy device controller allowed the
implementation of both "allowlists" and "denylists".
An allowlist is a device program that by default blocks access to
all devices. In order to access specific devices "allow rules"
for particular devices or device classes must be specified. In
contrast, a denylist is a device program that by default allows
access to all devices. In order to restrict access to specific
devices "deny rules" for particular devices or device classes
must be specified.
In the unified cgroup hierarchy the implementation of the device
controller has completely changed. Instead of files to read from
and write to a eBPF program of BPF_PROG_TYPE_CGROUP_DEVICE can be
attached to a cgroup. Even though the kernel implementation has
changed completely LXC tries to allow for the same semantics to
be followed in the legacy device cgroup and the unified eBPF-
based device controller. The following paragraphs explain the
semantics for the unified eBPF-based device controller.
As mentioned the format for specifying device rules for the
unified eBPF-based device controller is the same as for the
legacy cgroup device controller; only the configuration key
prefix has changed. Specifically, device rules for the legacy
cgroup device controller are specified via
lxc.cgroup.devices.allow and lxc.cgroup.devices.deny whereas for
the cgroup2 eBPF-based device controller
lxc.cgroup2.devices.allow and lxc.cgroup2.devices.deny must be
used.
• A allowlist device rule
lxc.cgroup2.devices.deny = a
will cause LXC to instruct the kernel to block access to all
devices by default. To grant access to devices allow device
rules must be added via the lxc.cgroup2.devices.allow key. This
is referred to as a "allowlist" device program.
• A denylist device rule
lxc.cgroup2.devices.allow = a
will cause LXC to instruct the kernel to allow access to all
devices by default. To deny access to devices deny device rules
must be added via lxc.cgroup2.devices.deny key. This is
referred to as a "denylist" device program.
• Specifying any of the aforementioned two rules will cause all
previous rules to be cleared, i.e. the device list will be
reset.
• When an allowlist program is requested, i.e. access to all
devices is blocked by default, specific deny rules for
individual devices or device classes are ignored.
• When a denylist program is requested, i.e. access to all
devices is allowed by default, specific allow rules for
individual devices or device classes are ignored.
For example the set of rules:
lxc.cgroup2.devices.deny = a
lxc.cgroup2.devices.allow = c *:* m
lxc.cgroup2.devices.allow = b *:* m
lxc.cgroup2.devices.allow = c 1:3 rwm
implements an allowlist device program, i.e. the kernel will
block access to all devices not specifically allowed in this
list. This particular program states that all character and block
devices may be created but only /dev/null might be read or
written.
If we instead switch to the following set of rules:
lxc.cgroup2.devices.allow = a
lxc.cgroup2.devices.deny = c *:* m
lxc.cgroup2.devices.deny = b *:* m
lxc.cgroup2.devices.deny = c 1:3 rwm
then LXC would instruct the kernel to implement a denylist, i.e.
the kernel will allow access to all devices not specifically
denied in this list. This particular program states that no
character devices or block devices might be created and that
/dev/null is not allow allowed to be read, written, or created.
Now consider the same program but followed by a "global rule"
which determines the type of device program (allowlist or
denylist) as explained above:
lxc.cgroup2.devices.allow = a
lxc.cgroup2.devices.deny = c *:* m
lxc.cgroup2.devices.deny = b *:* m
lxc.cgroup2.devices.deny = c 1:3 rwm
lxc.cgroup2.devices.allow = a
The last line will cause LXC to reset the device list without
changing the type of device program.
If we specify:
lxc.cgroup2.devices.allow = a
lxc.cgroup2.devices.deny = c *:* m
lxc.cgroup2.devices.deny = b *:* m
lxc.cgroup2.devices.deny = c 1:3 rwm
lxc.cgroup2.devices.deny = a
instead then the last line will cause LXC to reset the device
list and switch from a allowlist program to a denylist program.
lxc.cgroup.[controller name].[controller file]
Specify the control group value to be set on a legacy
cgroup hierarchy. The controller name is the literal name
of the control group. The permitted names and the syntax
of their values is not dictated by LXC, instead it depends
on the features of the Linux kernel running at the time
the container is started, eg. lxc.cgroup.cpuset.cpus
lxc.cgroup2.[controller name].[controller file]
Specify the control group value to be set on the unified
cgroup hierarchy. The controller name is the literal name
of the control group. The permitted names and the syntax
of their values is not dictated by LXC, instead it depends
on the features of the Linux kernel running at the time
the container is started, eg. lxc.cgroup2.memory.high
lxc.cgroup.dir
specify a directory or path in which the container's
cgroup will be created. For example, setting
lxc.cgroup.dir = my-cgroup/first for a container named
"c1" will create the container's cgroup as a sub-cgroup of
"my-cgroup". For example, if the user's current cgroup
"my-user" is located in the root cgroup of the cpuset
controller in a cgroup v1 hierarchy this would create the
cgroup "/sys/fs/cgroup/cpuset/my-user/my-cgroup/first/c1"
for the container. Any missing cgroups will be created by
LXC. This presupposes that the user has write access to
its current cgroup.
lxc.cgroup.dir.container
This is similar to lxc.cgroup.dir, but must be used
together with lxc.cgroup.dir.monitor and affects only the
container's cgroup path. This option is mutually exclusive
with lxc.cgroup.dir. Note that the final path the
container attaches to may be extended further by the
lxc.cgroup.dir.container.inner option.
lxc.cgroup.dir.monitor
This is the monitor process counterpart to
lxc.cgroup.dir.container.
lxc.cgroup.dir.monitor.pivot
On container termination the PID of the monitor process is
attached to this cgroup. This path should not be a
subpath of any other configured cgroup dir to ensure
proper removal of other cgroup paths on container
termination.
lxc.cgroup.dir.container.inner
Specify an additional subdirectory where the cgroup
namespace will be created. With this option, the cgroup
limits will be applied to the outer path specified in
lxc.cgroup.dir.container, which is not accessible from
within the container, making it possible to better enforce
limits for privileged containers in a way they cannot
override them. This only works in conjunction with the
lxc.cgroup.dir.container and lxc.cgroup.dir.monitor
options and has otherwise no effect.
lxc.cgroup.relative
Set this to 1 to instruct LXC to never escape to the root
cgroup. This makes it easy for users to adhere to
restrictions enforced by cgroup2 and systemd.
Specifically, this makes it possible to run LXC containers
as systemd services.
CAPABILITIES
The capabilities can be dropped in the container if this one is
run as root.
lxc.cap.drop
Specify the capability to be dropped in the container. A
single line defining several capabilities with a space
separation is allowed. The format is the lower case of the
capability definition without the "CAP_" prefix, eg.
CAP_SYS_MODULE should be specified as sys_module. See
capabilities(7). If used with no value, lxc will clear
any drop capabilities specified up to this point.
lxc.cap.keep
Specify the capability to be kept in the container. All
other capabilities will be dropped. When a special value
of "none" is encountered, lxc will clear any keep
capabilities specified up to this point. A value of "none"
alone can be used to drop all capabilities.
NAMESPACES
A namespace can be cloned (lxc.namespace.clone), kept
(lxc.namespace.keep) or shared (lxc.namespace.share.[namespace
identifier]).
lxc.namespace.clone
Specify namespaces which the container is supposed to be
created with. The namespaces to create are specified as a
space separated list. Each namespace must correspond to
one of the standard namespace identifiers as seen in the
/proc/PID/ns directory. When lxc.namespace.clone is not
explicitly set all namespaces supported by the kernel and
the current configuration will be used.
To create a new mount, net and ipc namespace set
lxc.namespace.clone=mount net ipc.
lxc.namespace.keep
Specify namespaces which the container is supposed to
inherit from the process that created it. The namespaces
to keep are specified as a space separated list. Each
namespace must correspond to one of the standard namespace
identifiers as seen in the /proc/PID/ns directory. The
lxc.namespace.keep is a denylist option, i.e. it is useful
when enforcing that containers must keep a specific set of
namespaces.
To keep the network, user and ipc namespace set
lxc.namespace.keep=user net ipc.
Note that sharing pid namespaces will likely not work with
most init systems.
Note that if the container requests a new user namespace
and the container wants to inherit the network namespace
it needs to inherit the user namespace as well.
lxc.namespace.share.[namespace identifier]
Specify a namespace to inherit from another container or
process. The [namespace identifier] suffix needs to be
replaced with one of the namespaces that appear in the
/proc/PID/ns directory.
To inherit the namespace from another process set the
lxc.namespace.share.[namespace identifier] to the PID of
the process, e.g. lxc.namespace.share.net=42.
To inherit the namespace from another container set the
lxc.namespace.share.[namespace identifier] to the name of
the container, e.g. lxc.namespace.share.pid=c3.
To inherit the namespace from another container located in
a different path than the standard liblxc path set the
lxc.namespace.share.[namespace identifier] to the full
path to the container, e.g.
lxc.namespace.share.user=/opt/c3.
In order to inherit namespaces the caller needs to have
sufficient privilege over the process or container.
Note that sharing pid namespaces between system containers
will likely not work with most init systems.
Note that if two processes are in different user
namespaces and one process wants to inherit the other's
network namespace it usually needs to inherit the user
namespace as well.
Note that without careful additional configuration of an
LSM, sharing user+pid namespaces with a task may allow
that task to escalate privileges to that of the task
calling liblxc.
lxc.time.offset.boot
Specify a positive or negative offset for the boottime
clock. The format accepts hours (h), minutes (m), seconds
(s), milliseconds (ms), microseconds (us), and nanoseconds
(ns).
lxc.time.offset.monotonic
Specify a positive or negative offset for the monotonic
clock. The format accepts hours (h), minutes (m), seconds
(s), milliseconds (ms), microseconds (us), and nanoseconds
(ns).
RESOURCE LIMITS
The soft and hard resource limits for the container can be
changed. Unprivileged containers can only lower them. Resources
which are not explicitly specified will be inherited.
lxc.prlimit.[limit name]
Specify the resource limit to be set. A limit is specified
as two colon separated values which are either numeric or
the word 'unlimited'. A single value can be used as a
shortcut to set both soft and hard limit to the same
value. The permitted names the "RLIMIT_" resource names in
lowercase without the "RLIMIT_" prefix, eg. RLIMIT_NOFILE
should be specified as "nofile". See setrlimit(2). If
used with no value, lxc will clear the resource limit
specified up to this point. A resource with no explicitly
configured limitation will be inherited from the process
starting up the container.
SYSCTL
Configure kernel parameters for the container.
lxc.sysctl.[kernel parameters name]
Specify the kernel parameters to be set. The parameters
available are those listed under /proc/sys/. Note that
not all sysctls are namespaced. Changing Non-namespaced
sysctls will cause the system-wide setting to be modified.
sysctl(8). If used with no value, lxc will clear the
parameters specified up to this point.
APPARMOR PROFILE
If lxc was compiled and installed with apparmor support, and the
host system has apparmor enabled, then the apparmor profile under
which the container should be run can be specified in the
container configuration. The default is lxc-container-default-
cgns if the host kernel is cgroup namespace aware, or lxc-
container-default otherwise.
lxc.apparmor.profile
Specify the apparmor profile under which the container
should be run. To specify that the container should be
unconfined, use
lxc.apparmor.profile = unconfined
If the apparmor profile should remain unchanged (i.e. if
you are nesting containers and are already confined), then
use
lxc.apparmor.profile = unchanged
If you instruct LXC to generate the apparmor profile, then
use
lxc.apparmor.profile = generated
lxc.apparmor.allow_incomplete
Apparmor profiles are pathname based. Therefore many file
restrictions require mount restrictions to be effective
against a determined attacker. However, these mount
restrictions are not yet implemented in the upstream
kernel. Without the mount restrictions, the apparmor
profiles still protect against accidental damager.
If this flag is 0 (default), then the container will not
be started if the kernel lacks the apparmor mount
features, so that a regression after a kernel upgrade will
be detected. To start the container under partial apparmor
protection, set this flag to 1.
lxc.apparmor.allow_nesting
If set this to 1, causes the following changes. When
generated apparmor profiles are used, they will contain
the necessary changes to allow creating a nested
container. In addition to the usual mount points,
/dev/.lxc/proc and /dev/.lxc/sys will contain procfs and
sysfs mount points without the lxcfs overlays, which, if
generated apparmor profiles are being used, will not be
read/writable directly.
lxc.apparmor.raw
A list of raw AppArmor profile lines to append to the
profile. Only valid when using generated profiles.
SELINUX CONTEXT
If lxc was compiled and installed with SELinux support, and the
host system has SELinux enabled, then the SELinux context under
which the container should be run can be specified in the
container configuration. The default is unconfined_t, which means
that lxc will not attempt to change contexts. See
/usr/local/share/lxc/selinux/lxc.te for an example policy and
more information.
lxc.selinux.context
Specify the SELinux context under which the container
should be run or unconfined_t. For example
lxc.selinux.context = system_u:system_r:lxc_t:s0:c22
lxc.selinux.context.keyring
Specify the SELinux context under which the container's
keyring should be created. By default this the same as
lxc.selinux.context, or the context lxc is executed under
if lxc.selinux.context has not been set.
lxc.selinux.context.keyring = system_u:system_r:lxc_t:s0:c22
KERNEL KEYRING
The Linux Keyring facility is primarily a way for various kernel
components to retain or cache security data, authentication keys,
encryption keys, and other data in the kernel. By default lxc
will create a new session keyring for the started application.
lxc.keyring.session
Disable the creation of new session keyring by lxc. The
started application will then inherit the current session
keyring. By default, or when passing the value 1, a new
keyring will be created.
lxc.keyring.session = 0
SECCOMP CONFIGURATION
A container can be started with a reduced set of available system
calls by loading a seccomp profile at startup. The seccomp
configuration file must begin with a version number on the first
line, a policy type on the second line, followed by the
configuration.
Versions 1 and 2 are currently supported. In version 1, the
policy is a simple allowlist. The second line therefore must read
"allowlist", with the rest of the file containing one (numeric)
syscall number per line. Each syscall number is allowlisted,
while every unlisted number is denylisted for use in the
container
In version 2, the policy may be denylist or allowlist, supports
per-rule and per-policy default actions, and supports per-
architecture system call resolution from textual names.
An example denylist policy, in which all system calls are allowed
except for mknod, which will simply do nothing and return 0
(success), looks like:
2
denylist
mknod errno 0
ioctl notify
Specifying "errno" as action will cause LXC to register a seccomp
filter that will cause a specific errno to be returned to the
caller. The errno value can be specified after the "errno" action
word.
Specifying "notify" as action will cause LXC to register a
seccomp listener and retrieve a listener file descriptor from the
kernel. When a syscall is made that is registered as "notify" the
kernel will generate a poll event and send a message over the
file descriptor. The caller can read this message, inspect the
syscalls including its arguments. Based on this information the
caller is expected to send back a message informing the kernel
which action to take. Until that message is sent the kernel will
block the calling process. The format of the messages to read and
sent is documented in seccomp itself.
lxc.seccomp.profile
Specify a file containing the seccomp configuration to
load before the container starts.
lxc.seccomp.allow_nesting
If this flag is set to 1, then seccomp filters will be
stacked regardless of whether a seccomp profile is already
loaded. This allows nested containers to load their own
seccomp profile. The default setting is 0.
lxc.seccomp.notify.proxy
Specify a unix socket to which LXC will connect and
forward seccomp events to. The path must be in the form
unix:/path/to/socket or unix:@socket. The former specifies
a path-bound unix domain socket while the latter specifies
an abstract unix domain socket.
lxc.seccomp.notify.cookie
An additional string sent along with proxied seccomp
notification requests.
PR_SET_NO_NEW_PRIVS
With PR_SET_NO_NEW_PRIVS active execve() promises not to grant
privileges to do anything that could not have been done without
the execve() call (for example, rendering the set-user-ID and
set-group-ID mode bits, and file capabilities non-functional).
Once set, this bit cannot be unset. The setting of this bit is
inherited by children created by fork() and clone(), and
preserved across execve(). Note that PR_SET_NO_NEW_PRIVS is
applied after the container has changed into its intended
AppArmor profile or SElinux context.
lxc.no_new_privs
Specify whether the PR_SET_NO_NEW_PRIVS flag should be set
for the container. Set to 1 to activate.
UID MAPPINGS
A container can be started in a private user namespace with user
and group id mappings. For instance, you can map userid 0 in the
container to userid 200000 on the host. The root user in the
container will be privileged in the container, but unprivileged
on the host. Normally a system container will want a range of
ids, so you would map, for instance, user and group ids 0 through
20,000 in the container to the ids 200,000 through 220,000.
lxc.idmap
Four values must be provided. First a character, either
'u', or 'g', to specify whether user or group ids are
being mapped. Next is the first userid as seen in the user
namespace of the container. Next is the userid as seen on
the host. Finally, a range indicating the number of
consecutive ids to map.
CONTAINER HOOKS
Container hooks are programs or scripts which can be executed at
various times in a container's lifetime.
When a container hook is executed, additional information is
passed along. The lxc.hook.version argument can be used to
determine if the following arguments are passed as command line
arguments or through environment variables. The arguments are:
• Container name.
• Section (always 'lxc').
• The hook type (i.e. 'clone' or 'pre-mount').
• Additional arguments. In the case of the clone hook, any extra
arguments passed will appear as further arguments to the hook.
In the case of the stop hook, paths to filedescriptors for each
of the container's namespaces along with their types are
passed.
The following environment variables are set:
• LXC_CGNS_AWARE: indicator whether the container is cgroup
namespace aware.
• LXC_CONFIG_FILE: the path to the container configuration file.
• LXC_HOOK_TYPE: the hook type (e.g. 'clone', 'mount', 'pre-
mount'). Note that the existence of this environment variable
is conditional on the value of lxc.hook.version. If it is set
to 1 then LXC_HOOK_TYPE will be set.
• LXC_HOOK_SECTION: the section type (e.g. 'lxc', 'net'). Note
that the existence of this environment variable is conditional
on the value of lxc.hook.version. If it is set to 1 then
LXC_HOOK_SECTION will be set.
• LXC_HOOK_VERSION: the version of the hooks. This value is
identical to the value of the container's lxc.hook.version
config item. If it is set to 0 then old-style hooks are used.
If it is set to 1 then new-style hooks are used.
• LXC_LOG_LEVEL: the container's log level.
• LXC_NAME: is the container's name.
• LXC_[NAMESPACE IDENTIFIER]_NS: path under /proc/PID/fd/ to a
file descriptor referring to the container's namespace. For
each preserved namespace type there will be a separate
environment variable. These environment variables will only be
set if lxc.hook.version is set to 1.
• LXC_ROOTFS_MOUNT: the path to the mounted root filesystem.
• LXC_ROOTFS_PATH: this is the lxc.rootfs.path entry for the
container. Note this is likely not where the mounted rootfs is
to be found, use LXC_ROOTFS_MOUNT for that.
• LXC_SRC_NAME: in the case of the clone hook, this is the
original container's name.
Standard output from the hooks is logged at debug level.
Standard error is not logged, but can be captured by the hook
redirecting its standard error to standard output.
lxc.hook.version
To pass the arguments in new style via environment
variables set to 1 otherwise set to 0 to pass them as
arguments. This setting affects all hooks arguments that
were traditionally passed as arguments to the script.
Specifically, it affects the container name, section (e.g.
'lxc', 'net') and hook type (e.g. 'clone', 'mount', 'pre-
mount') arguments. If new-style hooks are used then the
arguments will be available as environment variables. The
container name will be set in LXC_NAME. (This is set
independently of the value used for this config item.) The
section will be set in LXC_HOOK_SECTION and the hook type
will be set in LXC_HOOK_TYPE. It also affects how the
paths to file descriptors referring to the container's
namespaces are passed. If set to 1 then for each namespace
a separate environment variable LXC_[NAMESPACE
IDENTIFIER]_NS will be set. If set to 0 then the paths
will be passed as arguments to the stop hook.
lxc.hook.pre-start
A hook to be run in the host's namespace before the
container ttys, consoles, or mounts are up.
lxc.hook.pre-mount
A hook to be run in the container's fs namespace but
before the rootfs has been set up. This allows for
manipulation of the rootfs, i.e. to mount an encrypted
filesystem. Mounts done in this hook will not be reflected
on the host (apart from mounts propagation), so they will
be automatically cleaned up when the container shuts down.
lxc.hook.mount
A hook to be run in the container's namespace after
mounting has been done, but before the pivot_root.
lxc.hook.autodev
A hook to be run in the container's namespace after
mounting has been done and after any mount hooks have run,
but before the pivot_root, if lxc.autodev == 1. The
purpose of this hook is to assist in populating the /dev
directory of the container when using the autodev option
for systemd based containers. The container's /dev
directory is relative to the ${LXC_ROOTFS_MOUNT}
environment variable available when the hook is run.
lxc.hook.start-host
A hook to be run in the host's namespace after the
container has been setup, and immediately before starting
the container init.
lxc.hook.start
A hook to be run in the container's namespace immediately
before executing the container's init. This requires the
program to be available in the container.
lxc.hook.stop
A hook to be run in the host's namespace with references
to the container's namespaces after the container has been
shut down. For each namespace an extra argument is passed
to the hook containing the namespace's type and a filename
that can be used to obtain a file descriptor to the
corresponding namespace, separated by a colon. The type is
the name as it would appear in the /proc/PID/ns directory.
For instance for the mount namespace the argument usually
looks like mnt:/proc/PID/fd/12.
lxc.hook.post-stop
A hook to be run in the host's namespace after the
container has been shut down.
lxc.hook.clone
A hook to be run when the container is cloned to a new
one. See lxc-clone(1) for more information.
lxc.hook.destroy
A hook to be run when the container is destroyed.
CONTAINER HOOKS ENVIRONMENT VARIABLES
A number of environment variables are made available to the
startup hooks to provide configuration information and assist in
the functioning of the hooks. Not all variables are valid in all
contexts. In particular, all paths are relative to the host
system and, as such, not valid during the lxc.hook.start hook.
LXC_NAME
The LXC name of the container. Useful for logging messages
in common log environments. [-n]
LXC_CONFIG_FILE
Host relative path to the container configuration file.
This gives the container to reference the original, top
level, configuration file for the container in order to
locate any additional configuration information not
otherwise made available. [-f]
LXC_CONSOLE
The path to the console output of the container if not
NULL. [-c] [lxc.console.path]
LXC_CONSOLE_LOGPATH
The path to the console log output of the container if not
NULL. [-L]
LXC_ROOTFS_MOUNT
The mount location to which the container is initially
bound. This will be the host relative path to the
container rootfs for the container instance being started
and is where changes should be made for that instance.
[lxc.rootfs.mount]
LXC_ROOTFS_PATH
The host relative path to the container root which has
been mounted to the rootfs.mount location.
[lxc.rootfs.path]
LXC_SRC_NAME
Only for the clone hook. Is set to the original container
name.
LXC_TARGET
Only for the stop hook. Is set to "stop" for a container
shutdown or "reboot" for a container reboot.
LXC_CGNS_AWARE
If unset, then this version of lxc is not aware of cgroup
namespaces. If set, it will be set to 1, and lxc is aware
of cgroup namespaces. Note this does not guarantee that
cgroup namespaces are enabled in the kernel. This is used
by the lxcfs mount hook.
LOGGING
Logging can be configured on a per-container basis. By default,
depending upon how the lxc package was compiled, container
startup is logged only at the ERROR level, and logged to a file
named after the container (with '.log' appended) either under the
container path, or under /usr/local/var/log/lxc.
Both the default log level and the log file can be specified in
the container configuration file, overriding the default
behavior. Note that the configuration file entries can in turn be
overridden by the command line options to lxc-start.
lxc.log.level
The level at which to log. The log level is an integer in
the range of 0..8 inclusive, where a lower number means
more verbose debugging. In particular 0 = trace, 1 =
debug, 2 = info, 3 = notice, 4 = warn, 5 = error, 6 =
critical, 7 = alert, and 8 = fatal. If unspecified, the
level defaults to 5 (error), so that only errors and above
are logged.
Note that when a script (such as either a hook script or a
network interface up or down script) is called, the
script's standard output is logged at level 1, debug.
lxc.log.file
The file to which logging info should be written.
lxc.log.syslog
Send logging info to syslog. It respects the log level
defined in lxc.log.level. The argument should be the
syslog facility to use, valid ones are: daemon, local0,
local1, local2, local3, local4, local5, local5, local6,
local7.
AUTOSTART
The autostart options support marking which containers should be
auto-started and in what order. These options may be used by LXC
tools directly or by external tooling provided by the
distributions.
lxc.start.auto
Whether the container should be auto-started. Valid
values are 0 (off) and 1 (on).
lxc.start.delay
How long to wait (in seconds) after the container is
started before starting the next one.
lxc.start.order
An integer used to sort the containers when auto-starting
a series of containers at once. A lower value means an
earlier start.
lxc.monitor.unshare
If not zero the mount namespace will be unshared from the
host before initializing the container (before running any
pre-start hooks). This requires the CAP_SYS_ADMIN
capability at startup. Default is 0.
lxc.monitor.signal.pdeath
Set the signal to be sent to the container's init when the
lxc monitor exits. By default it is set to SIGKILL which
will cause all container processes to be killed when the
lxc monitor process dies. To ensure that containers stay
alive even if lxc monitor dies set this to 0.
lxc.group
A multi-value key (can be used multiple times) to put the
container in a container group. Those groups can then be
used (amongst other things) to start a series of related
containers.
AUTOSTART AND SYSTEM BOOT
Each container can be part of any number of groups or no group at
all. Two groups are special. One is the NULL group, i.e. the
container does not belong to any group. The other group is the
"onboot" group.
When the system boots with the LXC service enabled, it will first
attempt to boot any containers with lxc.start.auto == 1 that is a
member of the "onboot" group. The startup will be in order of
lxc.start.order. If an lxc.start.delay has been specified, that
delay will be honored before attempting to start the next
container to give the current container time to begin
initialization and reduce overloading the host system. After
starting the members of the "onboot" group, the LXC system will
proceed to boot containers with lxc.start.auto == 1 which are not
members of any group (the NULL group) and proceed as with the
onboot group.
CONTAINER ENVIRONMENT
If you want to pass environment variables into the container
(that is, environment variables which will be available to init
and all of its descendents), you can use lxc.environment
parameters to do so. Be careful that you do not pass in anything
sensitive; any process in the container which doesn't have its
environment scrubbed will have these variables available to it,
and environment variables are always available via
/proc/PID/environ.
This configuration parameter can be specified multiple times;
once for each environment variable you wish to configure.
lxc.environment
Specify an environment variable to pass into the
container. Example:
lxc.environment = APP_ENV=production
lxc.environment = SYSLOG_SERVER=192.0.2.42
It is possible to inherit host environment variables by
setting the name of the variable without a "=" sign. For
example:
lxc.environment = PATH
