| lxc-attach(1) | lxc-attach(1) |
lxc-attach - start a process inside a running container.
lxc-attach
{-n, --name name} [-f, --rcfile config_file] [-a, --arch
arch] [-e, --elevated-privileges privileges] [-s, --namespaces
namespaces] [-R, --remount-sys-proc] [--keep-env] [--clear-env] [-v,
--set-var variable] [--keep-var variable] [-- command]
lxc-attach runs the specified command inside the container specified by name. The container has to be running already.
If no command is specified, the current default shell of the user running lxc-attach will be looked up inside the container and executed. This will fail if no such user exists inside the container or the container does not have a working nsswitch mechanism.
Previous versions of lxc-attach simply attached to the specified namespaces of a container and ran a shell or the specified command without first allocating a pseudo terminal. This made them vulnerable to input faking via a TIOCSTI ioctl call after switching between userspace execution contexts with different privilege levels. Newer versions of lxc-attach will try to allocate a pseudo terminal file descriptor pair on the host and attach any standard file descriptors which refer to a terminal to the container side of the pseudo terminal before executing a shell or command. Note, that if none of the standard file descriptors refer to a terminal lxc-attach will not try to allocate a pseudo terminal. Instead it will simply attach to the containers namespaces and run a shell or the specified command.
This configuration file if present will be used even if there is already a configuration file present in the previously created container (via lxc-create).
You may specify privileges, in case you do not want to elevate all of them, as a pipe-separated list, e.g. CGROUP|LSM. Allowed values are CGROUP, CAP and LSM representing cgroup, capabilities and restriction privileges respectively. (The pipe symbol needs to be escaped, e.g. CGROUP\|LSM or quoted, e.g. "CGROUP|LSM".)
Warning: This may leak privileges into the container if the command starts subprocesses that remain active after the main process that was attached is terminated. The (re-)starting of daemons inside the container is problematic, especially if the daemon starts a lot of subprocesses such as cron or sshd. Use with great care.
Important: This option implies -e.
Please see the Notes section for more details.
This option will be ignored if one tries to attach to the mount namespace anyway.
These options are common to most of lxc commands.
Note that this option is setting the priority of the events log in the alternate log file. It do not have effect on the ERROR events log on stderr.
This configuration file if present will be used even if there is already a configuration file present in the previously created container (via lxc-create).
To spawn a new shell running inside an existing container, use
lxc-attach -n container
To restart the cron service of a running Debian container, use
lxc-attach -n container -- /etc/init.d/cron restart
To deactivate the network link eth1 of a running container that does not have the NET_ADMIN capability, use either the -e option to use increased capabilities, assuming the ip tool is installed:
lxc-attach -n container -e -- /sbin/ip link delete eth1
Or, alternatively, use the -s to use the tools installed on the host outside the container:
lxc-attach -n container -s NETWORK -- /sbin/ip link delete eth1
Attaching completely (including the pid and mount namespaces) to a container requires a kernel of version 3.8 or higher, or a patched kernel, please see the lxc website for details. lxc-attach will fail in that case if used with an unpatched kernel of version 3.7 and prior.
Nevertheless, it will succeed on an unpatched kernel of version 3.0 or higher if the -s option is used to restrict the namespaces that the process is to be attached to to one or more of NETWORK, IPC and UTSNAME.
Attaching to user namespaces is supported by kernel 3.8 or higher with enabling user namespace.
The Linux /proc and /sys filesystems contain information about some quantities that are affected by namespaces, such as the directories named after process ids in /proc or the network interface information in /sys/class/net. The namespace of the process mounting the pseudo-filesystems determines what information is shown, not the namespace of the process accessing /proc or /sys.
If one uses the -s option to only attach to the pid namespace of a container, but not its mount namespace (which will contain the /proc of the container and not the host), the contents of /proc will reflect that of the host and not the container. Analogously, the same issue occurs when reading the contents of /sys/class/net and attaching to just the network namespace.
To work around this problem, the -R flag provides the option to remount /proc and /sys in order for them to reflect the network/pid namespace context of the attached process. In order not to interfere with the host's actual filesystem, the mount namespace will be unshared (like lxc-unshare does) before this is done, essentially giving the process a new mount namespace, which is identical to the hosts's mount namespace except for the /proc and /sys filesystems.
Previous versions of lxc-attach suffered a bug whereby a user could attach to a containers namespace without being placed in a writeable cgroup for some critical subsystems. Newer versions of lxc-attach will check whether a user is in a writeable cgroup for those critical subsystems. lxc-attach might thus fail unexpectedly for some users (E.g. on systems where an unprivileged user is not placed in a writeable cgroup in critical subsystems on login.). However, this behavior is correct and more secure.
The -e and -s options should be used with care, as it may break the isolation of the containers if used improperly.
lxc(7), lxc-create(1), lxc-copy(1), lxc-destroy(1), lxc-start(1), lxc-stop(1), lxc-execute(1), lxc-console(1), lxc-monitor(1), lxc-wait(1), lxc-cgroup(1), lxc-ls(1), lxc-info(1), lxc-freeze(1), lxc-unfreeze(1), lxc-attach(1), lxc.conf(5)
Daniel Lezcano <daniel.lezcano@free.fr>
| 2019-04-14 |