systemd-nspawn systemd Developer Lennart Poettering lennart@poettering.net systemd-nspawn 1 systemd-nspawn Spawn a namespace container for debugging, testing and building systemd-nspawn OPTIONS COMMAND ARGS systemd-nspawn --boot OPTIONS ARGS Description systemd-nspawn may be used to run a command or OS in a light-weight namespace container. In many ways it is similar to chroot1, but more powerful since it fully virtualizes the file system hierarchy, as well as the process tree, the various IPC subsystems and the host and domain name. systemd-nspawn limits access to various kernel interfaces in the container to read-only, such as /sys, /proc/sys or /sys/fs/selinux. Network interfaces and the system clock may not be changed from within the container. Device nodes may not be created. The host system cannot be rebooted and kernel modules may not be loaded from within the container. Note that even though these security precautions are taken systemd-nspawn is not suitable for fully secure container setups. Many of the security features may be circumvented and are hence primarily useful to avoid accidental changes to the host system from the container. In contrast to chroot1 systemd-nspawn may be used to boot full Linux-based operating systems in a container. Use a tool like dnf8, debootstrap8, or pacman8 to set up an OS directory tree suitable as file system hierarchy for systemd-nspawn containers. Note that systemd-nspawn will mount file systems private to the container to /dev, /run and similar. These will not be visible outside of the container, and their contents will be lost when the container exits. Note that running two systemd-nspawn containers from the same directory tree will not make processes in them see each other. The PID namespace separation of the two containers is complete and the containers will share very few runtime objects except for the underlying file system. Use machinectl1's login command to request an additional login prompt in a running container. systemd-nspawn implements the Container Interface specification. As a safety check systemd-nspawn will verify the existence of /usr/lib/os-release or /etc/os-release in the container tree before starting the container (see os-release5). It might be necessary to add this file to the container tree manually if the OS of the container is too old to contain this file out-of-the-box. Options If option is specified, the arguments are used as arguments for the init binary. Otherwise, COMMAND specifies the program to launch in the container, and the remaining arguments are used as arguments for this program. If is not used and no arguments are specified, a shell is launched in the container. The following options are understood: Directory to use as file system root for the container. If neither , nor is specified the directory is determined by searching for a directory named the same as the machine name specified with . See machinectl1 section "Files and Directories" for the precise search path. If neither , , nor are specified, the current directory will be used. May not be specified together with . Directory or btrfs subvolume to use as template for the container's root directory. If this is specified and the container's root directory (as configured by ) does not yet exist it is created as btrfs subvolume and populated from this template tree. Ideally, the specified template path refers to the root of a btrfs subvolume, in which case a simple copy-on-write snapshot is taken, and populating the root directory is instant. If the specified template path does not refer to the root of a btrfs subvolume (or not even to a btrfs file system at all), the tree is copied, which can be substantially more time-consuming. Note that if this option is used the container's root directory (in contrast to the template directory!) must be located on a btrfs file system, so that the btrfs subvolume may be created. May not be specified together with or . Note that this switch leaves host name, machine ID and all other settings that could identify the instance unmodified. If specified, the container is run with a temporary btrfs snapshot of its root directory (as configured with ), that is removed immediately when the container terminates. This option is only supported if the root file system is btrfs. May not be specified together with or . Note that this switch leaves host name, machine ID and all other settings that could identify the instance unmodified. Disk image to mount the root directory for the container from. Takes a path to a regular file or to a block device node. The file or block device must contain either: An MBR partition table with a single partition of type 0x83 that is marked bootable. A GUID partition table (GPT) with a single partition of type 0fc63daf-8483-4772-8e79-3d69d8477de4. A GUID partition table (GPT) with a marked root partition which is mounted as the root directory of the container. Optionally, GPT images may contain a home and/or a server data partition which are mounted to the appropriate places in the container. All these partitions must be identified by the partition types defined by the Discoverable Partitions Specification. Any other partitions, such as foreign partitions, swap partitions or EFI system partitions are not mounted. May not be specified together with , or . Invoke the shell or specified program as process ID (PID) 2 instead of PID 1 (init). By default, if neither this option nor is used, the selected binary is run as process with PID 1, a mode only suitable for programs that are aware of the special semantics that the process with PID 1 has on UNIX. For example, it needs to reap all processes reparented to it, and should implement sysvinit compatible signal handling (specifically: it needs to reboot on SIGINT, reexecute on SIGTERM, reload configuration on SIGHUP, and so on). With a minimal stub init process is run as PID 1 and the selected binary is executed as PID 2 (and hence does not need to implement any special semantics). The stub init process will reap processes as necessary and react appropriately to signals. It is recommended to use this mode to invoke arbitrary commands in containers, unless they have been modified to run correctly as PID 1. Or in other words: this switch should be used for pretty much all commands, except when the command refers to an init or shell implementation, as these are generally capable of running correctly as PID 1. This option may not be combined with or . Automatically search for an init binary and invoke it as PID 1, instead of a shell or a user supplied program. If this option is used, arguments specified on the command line are used as arguments for the init binary. This option may not be combined with or . The following table explains the different modes of invocation and relationship to (see above): Invocation Mode Switch Explanation Neither nor specified The passed parameters are interpreted as the command line, which is executed as PID 1 in the container. specified The passed parameters are interpreted as the command line, which is executed as PID 2 in the container. A stub init process is run as PID 1. specified An init binary as automatically searched and run as PID 1 in the container. The passed parameters are used as invocation parameters for this process.
Change to the specified working directory before invoking the process in the container. Expects an absolute path in the container's file system namespace. After transitioning into the container, change to the specified user-defined in the container's user database. Like all other systemd-nspawn features, this is not a security feature and provides protection against accidental destructive operations only. Sets the machine name for this container. This name may be used to identify this container during its runtime (for example in tools like machinectl1 and similar), and is used to initialize the container's hostname (which the container can choose to override, however). If not specified, the last component of the root directory path of the container is used, possibly suffixed with a random identifier in case mode is selected. If the root directory selected is the host's root directory the host's hostname is used as default instead. Set the specified UUID for the container. The init system will initialize /etc/machine-id from this if this file is not set yet. Note that this option takes effect only if /etc/machine-id in the container is unpopulated. Make the container part of the specified slice, instead of the default machine.slice. This is only applies if the machine is run in its own scope unit, i.e. if is not used. Set a unit property on the scope unit to register for the machine. This only applies if the machine is run in its own scope unit, i.e. if is not used. Takes unit property assignments in the same format as systemctl set-property. This is useful to set memory limits and similar for machines. Controls user namespacing. If enabled, the container will run with its own private set of UNIX user and group ids (UIDs and GIDs). This involves mapping the private UIDs/GIDs used in the container (starting with the container's root user 0 and up) to a range of UIDs/GIDs on the host that are not used for other purposes (usually in the range beyond the host's UID/GID 65536). The parameter may be specified as follows: The value no turns off user namespacing. This is the default. The value yes (or the omission of a parameter) turns on user namespacing. The UID/GID range to use is determined automatically from the file ownership of the root directory of the container's directory tree. To use this option, make sure to prepare the directory tree in advance, and ensure that all files and directories in it are owned by UIDs/GIDs in the range you'd like to use. Also, make sure that used file ACLs exclusively reference UIDs/GIDs in the appropriate range. If this mode is used the number of UIDs/GIDs assigned to the container for use is 65536, and the UID/GID of the root directory must be a multiple of 65536. The value "pick" turns on user namespacing. In this case the UID/GID range is automatically chosen. As first step, the file owner of the root directory of the container's directory tree is read, and it is checked that it is currently not used by the system otherwise (in particular, that no other container is using it). If this check is successful, the UID/GID range determined this way is used, similar to the behaviour if "yes" is specified. If the check is not successful (and thus the UID/GID range indicated in the root directory's file owner is already used elsewhere) a new – currently unused – UID/GID range of 65536 UIDs/GIDs is randomly chosen between the host UID/GIDs of 524288 and 1878982656, always starting at a multiple of 65536. This setting implies (see below), which has the effect that the files and directories in the container's directory tree will be owned by the appropriate users of the range picked. Using this option makes user namespace behaviour fully automatic. Note that the first invocation of a previously unused container image might result in picking a new UID/GID range for it, and thus in the (possibly expensive) file ownership adjustment operation. However, subsequent invocations of the container will be cheap (unless of course the picked UID/GID range is assigned to a different use by then). Finally if one or two colon-separated numeric parameters are specified, user namespacing is turned on, too. The first parameter specifies the first host UID/GID to assign to the container, the second parameter specifies the number of host UIDs/GIDs to assign to the container. If the second parameter is omitted, 65536 UIDs/GIDs are assigned. It is recommended to assign at least 65536 UIDs/GIDs to each container, so that the usable UID/GID range in the container covers 16 bit. For best security, do not assign overlapping UID/GID ranges to multiple containers. It is hence a good idea to use the upper 16 bit of the host 32-bit UIDs/GIDs as container identifier, while the lower 16 bit encode the container UID/GID used. This is in fact the behaviour enforced by the option. When user namespaces are used, the GID range assigned to each container is always chosen identical to the UID range. In most cases, using is the recommended option as it enhances container security massively and operates fully automatically in most cases. Note that the picked UID/GID range is not written to /etc/passwd or /etc/group. In fact, the allocation of the range is not stored persistently anywhere, except in the file ownership of the files and directories of the container. If the kernel supports the user namespaces feature, equivalent to , otherwise equivalent to . If specified, all files and directories in the container's directory tree will adjusted so that they are owned to the appropriate UIDs/GIDs selected for the container (see above). This operation is potentially expensive, as it involves descending and iterating through the full directory tree of the container. Besides actual file ownership, file ACLs are adjusted as well. This option is implied if is used. This option has no effect if user namespacing is not used. Disconnect networking of the container from the host. This makes all network interfaces unavailable in the container, with the exception of the loopback device and those specified with and configured with . If this option is specified, the CAP_NET_ADMIN capability will be added to the set of capabilities the container retains. The latter may be disabled by using . Assign the specified network interface to the container. This will remove the specified interface from the calling namespace and place it in the container. When the container terminates, it is moved back to the host namespace. Note that implies . This option may be used more than once to add multiple network interfaces to the container. Create a macvlan interface of the specified Ethernet network interface and add it to the container. A macvlan interface is a virtual interface that adds a second MAC address to an existing physical Ethernet link. The interface in the container will be named after the interface on the host, prefixed with mv-. Note that implies . This option may be used more than once to add multiple network interfaces to the container. Create an ipvlan interface of the specified Ethernet network interface and add it to the container. An ipvlan interface is a virtual interface, similar to a macvlan interface, which uses the same MAC address as the underlying interface. The interface in the container will be named after the interface on the host, prefixed with iv-. Note that implies . This option may be used more than once to add multiple network interfaces to the container. Create a virtual Ethernet link (veth) between host and container. The host side of the Ethernet link will be available as a network interface named after the container's name (as specified with ), prefixed with ve-. The container side of the Ethernet link will be named host0. The option implies . Note that systemd-networkd.service8 includes by default a network file /usr/lib/systemd/network/80-container-ve.network matching the host-side interfaces created this way, which contains settings to enable automatic address provisioning on the created virtual link via DHCP, as well as automatic IP routing onto the host's external network interfaces. It also contains /usr/lib/systemd/network/80-container-host0.network matching the container-side interface created this way, containing settings to enable client side address assignment via DHCP. In case systemd-networkd is running on both the host and inside the container, automatic IP communication from the container to the host is thus available, with further connectivity to the external network. Adds an additional virtual Ethernet link between host and container. Takes a colon-separated pair of host interface name and container interface name. The latter may be omitted in which case the container and host sides will be assigned the same name. This switch is independent of , and — in contrast — may be used multiple times, and allows configuration of the network interface names. Note that has no effect on interfaces created with . Adds the host side of the Ethernet link created with to the specified Ethernet bridge interface. Expects a valid network interface name of a bridge device as argument. Note that implies . If this option is used, the host side of the Ethernet link will use the vb- prefix instead of ve-. Creates a virtual Ethernet link (veth) to the container and adds it to an automatically managed Ethernet bridge interface. The bridge interface is named after the passed argument, prefixed with vz-. The bridge interface is automatically created when the first container configured for its name is started, and is automatically removed when the last container configured for its name exits. Hence, each bridge interface configured this way exists only as long as there's at least one container referencing it running. This option is very similar to , besides this automatic creation/removal of the bridge device. This setting makes it easy to place multiple related containers on a common, virtual Ethernet-based broadcast domain, here called a "zone". Each container may only be part of one zone, but each zone may contain any number of containers. Each zone is referenced by its name. Names may be chosen freely (as long as they form valid network interface names when prefixed with vz-), and it is sufficient to pass the same name to the switch of the various concurrently running containers to join them in one zone. Note that systemd-networkd.service8 includes by default a network file /usr/lib/systemd/network/80-container-vz.network matching the bridge interfaces created this way, which contains settings to enable automatic address provisioning on the created virtual network via DHCP, as well as automatic IP routing onto the host's external network interfaces. Using is hence in most cases fully automatic and sufficient to connect multiple local containers in a joined broadcast domain to the host, with further connectivity to the external network. If private networking is enabled, maps an IP port on the host onto an IP port on the container. Takes a protocol specifier (either tcp or udp), separated by a colon from a host port number in the range 1 to 65535, separated by a colon from a container port number in the range from 1 to 65535. The protocol specifier and its separating colon may be omitted, in which case tcp is assumed. The container port number and its colon may be omitted, in which case the same port as the host port is implied. This option is only supported if private networking is used, such as with , . Sets the SELinux security context to be used to label processes in the container. Sets the SELinux security context to be used to label files in the virtual API file systems in the container. List one or more additional capabilities to grant the container. Takes a comma-separated list of capability names, see capabilities7 for more information. Note that the following capabilities will be granted in any way: CAP_CHOWN, CAP_DAC_OVERRIDE, CAP_DAC_READ_SEARCH, CAP_FOWNER, CAP_FSETID, CAP_IPC_OWNER, CAP_KILL, CAP_LEASE, CAP_LINUX_IMMUTABLE, CAP_NET_BIND_SERVICE, CAP_NET_BROADCAST, CAP_NET_RAW, CAP_SETGID, CAP_SETFCAP, CAP_SETPCAP, CAP_SETUID, CAP_SYS_ADMIN, CAP_SYS_CHROOT, CAP_SYS_NICE, CAP_SYS_PTRACE, CAP_SYS_TTY_CONFIG, CAP_SYS_RESOURCE, CAP_SYS_BOOT, CAP_AUDIT_WRITE, CAP_AUDIT_CONTROL. Also CAP_NET_ADMIN is retained if is specified. If the special value all is passed, all capabilities are retained. Specify one or more additional capabilities to drop for the container. This allows running the container with fewer capabilities than the default (see above). Specify the process signal to send to the container's PID 1 when nspawn itself receives SIGTERM, in order to trigger an orderly shutdown of the container. Defaults to SIGRTMIN+3 if is used (on systemd-compatible init systems SIGRTMIN+3 triggers an orderly shutdown). For a list of valid signals, see signal7. Control whether the container's journal shall be made visible to the host system. If enabled, allows viewing the container's journal files from the host (but not vice versa). Takes one of no, host, try-host, guest, try-guest, auto. If no, the journal is not linked. If host, the journal files are stored on the host file system (beneath /var/log/journal/machine-id) and the subdirectory is bind-mounted into the container at the same location. If guest, the journal files are stored on the guest file system (beneath /var/log/journal/machine-id) and the subdirectory is symlinked into the host at the same location. try-host and try-guest do the same but do not fail if the host does not have persistent journalling enabled. If auto (the default), and the right subdirectory of /var/log/journal exists, it will be bind mounted into the container. If the subdirectory does not exist, no linking is performed. Effectively, booting a container once with guest or host will link the journal persistently if further on the default of auto is used. Equivalent to . Mount the root file system read-only for the container. Bind mount a file or directory from the host into the container. Takes one of: a path argument — in which case the specified path will be mounted from the host to the same path in the container —, or a colon-separated pair of paths — in which case the first specified path is the source in the host, and the second path is the destination in the container —, or a colon-separated triple of source path, destination path and mount options. Mount options are comma-separated and currently, only "rbind" and "norbind" are allowed. Defaults to "rbind". Backslash escapes are interpreted, so \: may be used to embed colons in either path. This option may be specified multiple times for creating multiple independent bind mount points. The option creates read-only bind mounts. Mount a tmpfs file system into the container. Takes a single absolute path argument that specifies where to mount the tmpfs instance to (in which case the directory access mode will be chosen as 0755, owned by root/root), or optionally a colon-separated pair of path and mount option string that is used for mounting (in which case the kernel default for access mode and owner will be chosen, unless otherwise specified). This option is particularly useful for mounting directories such as /var as tmpfs, to allow state-less systems, in particular when combined with . Backslash escapes are interpreted in the path, so \: may be used to embed colons in the path. Combine multiple directory trees into one overlay file system and mount it into the container. Takes a list of colon-separated paths to the directory trees to combine and the destination mount point. Backslash escapes are interpreted in the paths, so \: may be used to embed colons in the paths. If three or more paths are specified, then the last specified path is the destination mount point in the container, all paths specified before refer to directory trees on the host and are combined in the specified order into one overlay file system. The left-most path is hence the lowest directory tree, the second-to-last path the highest directory tree in the stacking order. If is used instead of , a read-only overlay file system is created. If a writable overlay file system is created, all changes made to it are written to the highest directory tree in the stacking order, i.e. the second-to-last specified. If only two paths are specified, then the second specified path is used both as the top-level directory tree in the stacking order as seen from the host, as well as the mount point for the overlay file system in the container. At least two paths have to be specified. For details about overlay file systems, see overlayfs.txt. Note that the semantics of overlay file systems are substantially different from normal file systems, in particular regarding reported device and inode information. Device and inode information may change for a file while it is being written to, and processes might see out-of-date versions of files at times. Note that this switch automatically derives the workdir= mount option for the overlay file system from the top-level directory tree, making it a sibling of it. It is hence essential that the top-level directory tree is not a mount point itself (since the working directory must be on the same file system as the top-most directory tree). Also note that the lowerdir= mount option receives the paths to stack in the opposite order of this switch. Specifies an environment variable assignment to pass to the init process in the container, in the format NAME=VALUE. This may be used to override the default variables or to set additional variables. This parameter may be used more than once. Allows the container to share certain system facilities with the host. More specifically, this turns off PID namespacing, UTS namespacing and IPC namespacing, and thus allows the guest to see and interact more easily with processes outside of the container. Note that using this option makes it impossible to start up a full Operating System in the container, as an init system cannot operate in this mode. It is only useful to run specific programs or applications this way, without involving an init system in the container. This option implies . This option may not be combined with . Controls whether the container is registered with systemd-machined8. Takes a boolean argument, which defaults to yes. This option should be enabled when the container runs a full Operating System (more specifically: an init system), and is useful to ensure that the container is accessible via machinectl1 and shown by tools such as ps1. If the container does not run an init system, it is recommended to set this option to no. Note that implies . Instead of creating a transient scope unit to run the container in, simply register the service or scope unit systemd-nspawn has been invoked in with systemd-machined8. This has no effect if is used. This switch should be used if systemd-nspawn is invoked from within a service unit, and the service unit's sole purpose is to run a single systemd-nspawn container. This option is not available if run from a user session. Control the architecture ("personality") reported by uname2 in the container. Currently, only x86 and x86-64 are supported. This is useful when running a 32-bit container on a 64-bit host. If this setting is not used, the personality reported in the container is the same as the one reported on the host. Turns off any status output by the tool itself. When this switch is used, the only output from nspawn will be the console output of the container OS itself. MODE Boots the container in volatile mode. When no mode parameter is passed or when mode is specified as , full volatile mode is enabled. This means the root directory is mounted as a mostly unpopulated tmpfs instance, and /usr from the OS tree is mounted into it in read-only mode (the system thus starts up with read-only OS resources, but pristine state and configuration, any changes to the either are lost on shutdown). When the mode parameter is specified as , the OS tree is mounted read-only, but /var is mounted as a tmpfs instance into it (the system thus starts up with read-only OS resources and configuration, but pristine state, and any changes to the latter are lost on shutdown). When the mode parameter is specified as (the default), the whole OS tree is made available writable. Note that setting this to or will only work correctly with operating systems in the container that can boot up with only /usr mounted, and are able to populate /var automatically, as needed. MODE Controls whether systemd-nspawn shall search for and use additional per-container settings from .nspawn files. Takes a boolean or the special values or . If enabled (the default), a settings file named after the machine (as specified with the setting, or derived from the directory or image file name) with the suffix .nspawn is searched in /etc/systemd/nspawn/ and /run/systemd/nspawn/. If it is found there, its settings are read and used. If it is not found there, it is subsequently searched in the same directory as the image file or in the immediate parent of the root directory of the container. In this case, if the file is found, its settings will be also read and used, but potentially unsafe settings are ignored. Note that in both these cases, settings on the command line take precedence over the corresponding settings from loaded .nspawn files, if both are specified. Unsafe settings are considered all settings that elevate the container's privileges or grant access to additional resources such as files or directories of the host. For details about the format and contents of .nspawn files, consult systemd.nspawn5. If this option is set to , the file is searched, read and used the same way, however, the order of precedence is reversed: settings read from the .nspawn file will take precedence over the corresponding command line options, if both are specified. If this option is set to , the file is searched, read and used the same way, but regardless of being found in /etc/systemd/nspawn/, /run/systemd/nspawn/ or next to the image file or container root directory, all settings will take effect, however, command line arguments still take precedence over corresponding settings. If disabled, no .nspawn file is read and no settings except the ones on the command line are in effect. --notify-ready= Configures support for notifications from the container's init process. --notify-ready= takes a boolean ( and ). With option systemd-nspawn notifies systemd with a READY=1 message when the init process is created. With option systemd-nspawn waits for the READY=1 message from the init process in the container before sending its own to systemd. For more details about notifications see sd_notify3).
Examples Download a Fedora image and start a shell in it # machinectl pull-raw --verify=no http://ftp.halifax.rwth-aachen.de/fedora/linux/releases/21/Cloud/Images/x86_64/Fedora-Cloud-Base-20141203-21.x86_64.raw.xz # systemd-nspawn -M Fedora-Cloud-Base-20141203-21 This downloads an image using machinectl1 and opens a shell in it. Build and boot a minimal Fedora distribution in a container # dnf -y --releasever=23 --installroot=/srv/mycontainer --disablerepo='*' --enablerepo=fedora --enablerepo=updates install systemd passwd dnf fedora-release vim-minimal # systemd-nspawn -bD /srv/mycontainer This installs a minimal Fedora distribution into the directory /srv/mycontainer/ and then boots an OS in a namespace container in it. Spawn a shell in a container of a minimal Debian unstable distribution # debootstrap --arch=amd64 unstable ~/debian-tree/ # systemd-nspawn -D ~/debian-tree/ This installs a minimal Debian unstable distribution into the directory ~/debian-tree/ and then spawns a shell in a namespace container in it. Boot a minimal Arch Linux distribution in a container # pacstrap -c -d ~/arch-tree/ base # systemd-nspawn -bD ~/arch-tree/ This installs a minimal Arch Linux distribution into the directory ~/arch-tree/ and then boots an OS in a namespace container in it. Boot into an ephemeral <literal>btrfs</literal> snapshot of the host system # systemd-nspawn -D / -xb This runs a copy of the host system in a btrfs snapshot which is removed immediately when the container exits. All file system changes made during runtime will be lost on shutdown, hence. Run a container with SELinux sandbox security contexts # chcon system_u:object_r:svirt_sandbox_file_t:s0:c0,c1 -R /srv/container # systemd-nspawn -L system_u:object_r:svirt_sandbox_file_t:s0:c0,c1 -Z system_u:system_r:svirt_lxc_net_t:s0:c0,c1 -D /srv/container /bin/sh Exit status The exit code of the program executed in the container is returned. See Also systemd1, systemd.nspawn5, chroot1, dnf8, debootstrap8, pacman8, systemd.slice5, machinectl1, btrfs8