man.bsd.lv manual page server

Setting Up a Jail Directory Tree

This example shows how to setup a jail directory tree containing an entire DragonFly distribution, provided that you built world before already:

D=/here/is/the/jail
cd /usr/src
mkdir -p $D
make installworld DESTDIR=$D
cd etc
make distribution DESTDIR=$D
cd $D
ln -sf dev/null boot/kernel

In many cases this example would put far more stuff in the jail than is needed. In the other extreme case a jail might contain only one single file: the executable to be run in the jail.

We recommend experimentation and caution that it is a lot easier to start with a “fat” jail and remove things until it stops working, than it is to start with a “thin” jail and add things until it works.

Setting Up a Jail

Do what was described in Setting Up a Jail Directory Tree to build the jail directory tree. For the sake of this example, we will assume you built it in /data/jail/192.168.11.100, named for the external jail IP address. Substitute below as needed with your own directory, IP addresses, and hostname.

First, you will want to set up your real system's environment to be “jail-friendly”. For consistency, we will refer to the parent box as the “host environment”, and to the jailed virtual machine as the “jail environment”. Generally speaking making a system jail-friendly does not require much work if you have jail.defaults.allow_listen_override set to 1. This will allow all listen sockets inside the jail to overload and override wildcard listen sockets on the host. This methodology works extremely well as long as you restrict the IPs you pass into the jail to avoid any security concerns, which typically means making at least two IP aliases for each jail that you create (an external IP and a localhost IP).

If you do not want to allow listener socket overloading, you must modify services you intend to leave running on the host system to listen only on specific IPs and not all IPs, or generally only run services inside your jails that do not conflict with services on the host system. This can be difficult and is not recommended.

Common services include: inetd(8), sendmail(8), named(8), rpcbind(8), mountd(8), nfsd(8), sendmail(1), and named(8).

For the purposes of our examples below, 192.168.11.1 is the host IP and we create aliases 192.168.11.X for each jail. In addition, we leave 127.0.0.1 on the host only and create aliases 127.0.0.X for each jail.

Start any jails for the first time without configuring the network interface so that you can clean it up a little and set up accounts. As with any machine (virtual or not) you will need to set a root password, time zone, etc.

To set the jail up for real you need to ifconfig some interface aliases. and set jail defaults. Here is an example:

From within your jail you can generally run services normally. Just use “localhost” normally. Do not inform services that localhost is 127.0.0.2. The kernel will automatically remap 'localhost' (e.g. 127.0.0.1) to the localhost IP you specify for the jail. The kernel will also map the jail's localhost IP back to 127.0.0.1 in the accept(2), getsockname(2), and getpeername(2) system calls.

Listen sockets can be overloaded between jails and between the host and its jails. You can continue to use services that listen on the wildcard *.* socket normally on both the host and its jails if you set the allow_listen_override flag to 1, and jailed listen sockets will override any host listen sockets listening on the wildcard address for the allowed jail IPs.

DragonFly also allows you to overload specific ports, but the jailed service will not receive any connections if the host is also specifically listening on the addr/port pair that the jail is using. If multiple jails are listening on the same addr/port pair (as specified by the jail's IP list), and the host is not, then only one jail will receive connections on that pair.

In otherwords, it is still a good idea to give each jail its own local and non-local IP address rather than have jails share.

Now start the jail:

You will end up with a shell prompt, assuming no errors, within the jail. You can now do the post-install configuration to set various configuration options by editing /etc/rc.conf, etc.

• Disable the port mapper (/etc/rc.conf: rpcbind_enable="NO")
• Run newaliases(1) to quell sendmail(8) warnings.
• Disable interface configuration to quell startup warnings about ifconfig(8) (network_interfaces="")
• Configure /etc/resolv.conf so that name resolution within the jail will work correctly
• Set a root password, probably different from the real host system
• Set the timezone with tzsetup(8)
• Add accounts for users in the jail environment
• Install any packages that you think the environment requires

You may also want to perform any package-specific configuration (web servers, SSH servers, etc), patch up /etc/syslog.conf so it logs as you would like, etc.

Exit from the shell, and the jail will be shut down.

Starting the Jail

You are now ready to restart the jail and bring up the environment with all of its daemons and other programs. To do this, first bring up the virtual host interface, and then start the jail's /etc/rc script from within the jail.

NOTE: If you plan to allow untrusted users to have root access inside the jail, you may wish to consider setting the jail.defaults.set_hostname_allowed to 0. Please see the management reasons why this is a good idea. If you do decide to set this variable, it must be set before starting any jails, and once each boot.

sysctl jail.defaults.net_raw_sockets=1
sysctl jail.defaults.allow_listen_override=1
ifconfig em0 inet alias 192.168.11.100/32
ifconfig lo0 inet alias 127.0.0.2
mount -t procfs proc /data/jail/192.168.11.100/proc
jail /data/jail/192.168.11.100 testhostname 127.0.0.2,192.168.11.100 \
	/bin/sh /etc/rc

A few warnings will be produced, because most sysctl(8) configuration variables cannot be set from within the jail, as they are global across all jails and the host environment. However, it should all work properly. You should be able to see inetd(8), syslogd(8), and other processes running within the jail using ps(1), with the ‘J’ flag appearing beside jailed processes. You should also be able to telnet(1) to the hostname or IP address of the jailed environment, and log in using the accounts you created previously.

Managing the Jail

Normal machine shutdown commands, such as halt(8), reboot(8), and shutdown(8), cannot be used successfully within the jail. To kill all processes in a jail, you may log into the jail and, as root, use one of the following commands, depending on what you want to accomplish:

kill -TERM -1
kill -KILL -1

This will send the SIGTERM or SIGKILL signals to all processes in the jail from within the jail. Depending on the intended use of the jail, you may also want to run /etc/rc.shutdown from within the jail. Currently there is no way to insert new processes into a jail, so you must first log into the jail before performing these actions.

To kill processes from outside the jail, you must individually identify the PID of each process to be killed. The /proc/pid/status file contains, as its last field, the hostname of the jail in which the process runs, or “-” to indicate that the process is not running within a jail. The ps(1) command also shows a ‘J’ flag for processes in a jail. However, the hostname for a jail may be, by default, modified from within the jail, so the /proc status entry is unreliable by default. To disable the setting of the hostname from within a jail, set the jail.set_hostname_allowed sysctl variable in the host environment to 0, which will affect all jails. You can have this sysctl set on each boot using sysctl.conf(5). Just add the following line to /etc/sysctl.conf:

In a future version of DragonFly, the mechanisms for managing jails may be more refined.

Sysctl MIB Entries

Certain aspects of the jail containments environment may be modified from the host environment using sysctl(8) MIB variables. For each jail there will be the same set of MIB variables as shown below but under jail.<id> which allows control of every jail individually. The values of the variables under jail.defaults will be copied to the per-jail MIB variables upon creation thus serving as a kind of system-wide template.

jail.defaults.allow_raw_sockets

This MIB entry determines whether or not prison root is allowed to create raw sockets. Setting this MIB to 1 allows utilities like ping(8) and traceroute(8) to operate inside the prison. If this MIB is set, the source IP addresses are enforced to comply with the IP address bound to the jail, regardless of whether or not the IP_HDRINCL flag has been set on the socket. Because raw sockets can be used to configure and interact with various network subsystems, extra caution should be used where privileged access to jails is given out to untrusted parties. As such, this option is disabled by default.

jail.defaults.chflags_allowed

This MIB entry determines how a privileged user inside a jail will be treated by chflags(2). If zero, such users are treated as unprivileged, and are unable to set or clear system file flags; if non-zero, such users are treated as privileged, and may manipulate system file flags subject to the usual constraints on kern.securelevel.

jail.jailed

This read-only MIB entry can be used to determine if a process is running inside a jail (value is 1) or not (value is 0).

jail.defaults.set_hostname_allowed

This MIB entry determines whether or not processes within a jail are allowed to change their hostname via hostname(1) or sethostname(3). In the current jail implementation, the ability to set the hostname from within the jail can impact management tools relying on the accuracy of jail information in /proc. As such, this should be disabled in environments where privileged access to jails is given out to untrusted parties.

jail.defaults.allow_listen_override

This feature allows both the host and your jails to overload services on the same ports. If enabled, the services in the jails will override wildcarded services on the host for the jail's IP list. As a safety mechanism, any services the host specifically binds to an IP will not be overridden. The host has visibility to all jail IPs but jails only have visibility to their specific IPs.

jail.defaults.socket_unixiproute_only

The jail functionality binds IPv4 and IPv6 addresses to each jail, and limits access to other network addresses in the IPv4 and IPv6 space that may be available in the host environment. However, jail is not currently able to limit access to other network protocol stacks that have not had jail functionality added to them. As such, by default, processes within jails may only access protocols in the following domains: PF_LOCAL, PF_INET, PF_INET6, and PF_ROUTE, permitting them access to UNIX domain sockets, IPv4 and IPv6 addresses, and routing sockets. To enable access to other domains, this MIB variable may be set to 0.

jail.defaults.sysvipc_allowed

This MIB entry determines whether or not processes within a jail have access to System V IPC primitives. In the current jail implementation, System V primitives share a single namespace across the host and jail environments, meaning that processes within a jail would be able to communicate with (and potentially interfere with) processes outside of the jail, and in other jails. As such, this functionality is disabled by default, but can be enabled by setting this MIB entry to 1.