Chroot

From Gentoo Wiki
(Redirected from Chroots)
Jump to:navigation Jump to:search
This page contains changes which are not marked for translation.
Resources

Chroot (Change root) is a Unix system utility used to change the apparent root directory to create a new environment logically separate from the main system's root directory. This new environment is known as a "chroot jail." A user operating inside the jail cannot see or access files outside of the environment they have been locked into.

One of the main uses for chrooting is to create a separate Linux system on top of a the current one for the purpose of testing or software compatibility. Chroot is often seen as a lightweight alternative to virtualization because it is able to run without the overhead of a hypervisor.

Prerequisites

Setting up the environment

When creating a new chroot setup, the first thing needed is a directory in which the chroot can reside. For example, a chroot could be created in /mnt/mychroot:

root #mkdir /mnt/mychroot
root #cd /mnt/mychroot

To mount an existing installation from a partition the following command can be run. Be sure to replace the <DEVICE> string in the example below with the drive and partition of the existing installation:

root #mkdir /mnt/mychroot
root #mount /dev/<DEVICE> /mnt/mychroot

If an installation has been previously created in a sub directory of the current root file system, the above steps need not be repeated.

Unpacking the system files

When building a new install, the next step is to download the stage3 tarball and unpack it to chroot location. For more information on this process please see Downloading the stage tarball and Unpacking the stage tarball in the Gentoo Handbook.

root #links http://distfiles.gentoo.org/releases/amd64/autobuilds/
root #tar xvjpf stage3-*.tar.bz2 -C /mnt/mychroot

Configuration

Before entering the chroot a number of directories must be mounted:

root #mount --rbind /dev /mnt/mychroot/dev
root #mount --make-rslave /mnt/mychroot/dev
root #mount -t proc /proc /mnt/mychroot/proc
root #mount --rbind /sys /mnt/mychroot/sys
root #mount --make-rslave /mnt/mychroot/sys
root #mount --rbind /tmp /mnt/mychroot/tmp
root #mount --bind /run /mnt/mychroot/run

Some basic configuration files must be copied from the host. Do not copy /etc/portage/make.conf when using an existing installation:

root #cp /etc/portage/make.conf /mnt/mychroot/etc/portage # When using an existing installation, skip this command.
root #cp /etc/resolv.conf /mnt/mychroot/etc

Usage

Once done, enter the chroot environment by executing the following commands:

root #chroot /mnt/mychroot /bin/bash
root #. /etc/profile
root #export PS1="(chroot) $PS1"

When creating a new installation, Portage should be synced to make sure everything is up to date.

(chroot) root #emerge-webrsync
(chroot) root #emerge --sync

The system is now ready. Feel free to install software, mess with settings, test experimental packages and configurations -- all without having any effect on the main system. To leave the chroot simply type exit or press Ctrl+d. Doing so will return the console to the normal environment. Do not forget to umount the directories that have been mounted.

systemd-nspawn

If the system uses systemd, systemd-nspawn can be used, which can automatically handle much of the boilerplate required in administering chroots. For example, to enter a chroot via systemd-nspawn with the same configuration as specified in the Configuration section, simply run:

root #cp /etc/portage/make.conf /mnt/mychroot/etc/portage
root #systemd-nspawn -D /mnt/mychroot --bind=/tmp --resolv-conf=/etc/resolv.conf

Init scripts

If setting up chroots is a task that must be performed often, it is possible to speed up the mounting of the directories by using an init script. The script could be added to the default runlevel and therefore set up automatically on system boot:

FILE /etc/init.d/mychroot
#!/sbin/openrc-run
 
depend() {
   need localmount
   need bootmisc
}
 
start() {
     ebegin "Mounting chroot directories"
     mount -o rbind /dev /mnt/mychroot/dev > /dev/null &
     mount -t proc none /mnt/mychroot/proc > /dev/null &
     mount -o bind /sys /mnt/mychroot/sys > /dev/null &
     mount -o bind /tmp /mnt/mychroot/tmp > /dev/null &
     eend $? "An error occurred while mounting chroot directories"
}
 
stop() {
     ebegin "Unmounting chroot directories"
     umount -f /mnt/mychroot/dev > /dev/null &
     umount -f /mnt/mychroot/proc > /dev/null &
     umount -f /mnt/mychroot/sys > /dev/null &
     umount -f /mnt/mychroot/tmp > /dev/null &
     eend $? "An error occurred while unmounting chroot directories"
}

When using a different directory or partition, add the necessary mounting commands in the start() function and change /mnt/chroot to the appropriate name.

Sound and graphics

The software running inside the chroot will by default not have access to the system sound- and display-server. Fixing this is done by either sharing a socket, or by running the communication with TCP over localhost.

Wayland

Wayland uses a socket to connect clients with the compositor. This socket needs to be shared with the chroot to make graphical applications work. The general procedure for finding this socket is:[1]

  1. If WAYLAND_SOCKET is set, interpret it as a file descriptor number on which the connection is already established, assuming that the parent process configured the connection for us.
  2. If WAYLAND_DISPLAY is set, concat with XDG_RUNTIME_DIR to form the path to the Unix socket.
  3. Assume the socket name is wayland-0 and concat with XDG_RUNTIME_DIR to form the path to the Unix socket.

Using WAYLAND_DISPLAY and XDG_RUNTIME_DIR is fine in most cases and will be used here. By default XDG_RUNTIME_DIR is set to /run/user/$(uid). This directory will not be available in the chroot because the #Configuration instructions bind mounts /run non-recursively. Assuming the user's uid is 1000, this can be solved by either bind-mounting {{Path|/run/user/1000 with:

root #mkdir -p /mnt/mychroot/run/user/1000
root #mount --bind /run/user/1000 /mnt/mychroot/run/user/1000

or by simply recursively bind mounting /run with:

root #mount --rbind /run /mnt/mychroot/run

The Wayland library dev-libs/wayland uses the same procedure for finding out the socket as listed above. So to share the socket with the chroot, the only thing that's needed to do is defining XDG_RUNTIME_DIR and WAYLAND_DISPLAY. Here it is assumed that the Wayland socket name WAYLAND_DISPLAY is wayland-0.

(chroot) root #useradd -m user
(chroot) root #su -l user
(chroot) user $export XDG_RUNTIME_DIR=/run/user/1000
(chroot) user $export WAYLAND_DISPLAY=wayland-0
(chroot) user $MOZ_ENABLE_WAYLAND=1 firefox-bin

Permission errors will occur if the user in the chroot does not have permissions to access the Wayland socket. This can be solved by using user namespace remapping or ACLs. The easiest solution is to just make sure that the user ids match. The useradd -u, --uid UID option can be used when creating a user.

PipeWire

Like Wayland, PipeWire uses a socket to connect clients to the PipeWire daemon.

Applications assume that the PipeWire socket will be located in ${XDG_RUNTIME_DIR}/pipewire-0, so the only thing that's needed to get PipeWire clients to connect to the host's daemon is to expose XDG_RUNTIME_DIR to the chroot. This process is identical to the one described in #Wayland. To expose XDG_RUNTIME_DIR, often /run/user/$(uid), the following commands are used:

root #mkdir -p /mnt/mychroot/run/user/1000
root #mount --bind /run/user/1000 /mnt/mychroot/run/user/1000

XDG_RUNTIME_DIR will not be set when logging in inside the chroot, therefore XDG_RUNTIME_DIR needs to exported so the PipeWire client can find the socket:

(chroot) user $export XDG_RUNTIME_DIR=/run/user/1000
(chroot) user $pw-cli

Xorg

Xorg by default listens on a socket located in /tmp/.X11-unix/X${DISPLAY}, as well as on localhost TCP port 6000 + ${DISPLAY}[2]. The instructions in #Configuration bind mounts /tmp, and therefore no additional configuration is needed except setting the DISPLAY variable before running a graphical application:

(chroot) user $DISPLAY=:0 firefox-bin

If the uid of the user inside the chroot does not match the uid outside the chroot, then setting permissions with xhost will be needed. To allow all local connections, run outside the chroot:

user $xhost +local:

See also

External resources

References

  1. https://wayland-book.com/protocol-design/wire-protocol.html
  2. So if DISPLAY=:12, then Xorg will listen on localhost TCP port 6012