The original FreeBSD developers felt the need for a method to restrict a
processes access to the host system resources so if it becomes
compromised the host system is protected from also being compromised.
They achieved this goal with the "chroot" command which was in the
original 4.4BSD system, from which the current FreeBSD RELEASE is a
direct descendant. This first generation "chroot" environment, made it
look like the named directory was the "root" IE starting point; of a
system directory tree. Just like "/" is to FreeBSD. In this basic
incarnation, the directory tree would just have the binaries necessary to
form a environment for a single application such as apache web server.
You could have multiple such "chroot" environments. They all shared the
hosts network and disk space. This trait continues into todays jail
systems. As you can imagine, occupants of these basic "chroots"
influenced users to stay at the RELEASE they were at because of the size
of the task to redevelop them under a new RELEASE mixture of binaries.
Jail deployments greater than two were uncommon.
The jail utility appeared in FreeBSD 4.0. With this second generation
"chroot" enhancement came the renaming of a "chroot" environment to a
"jail, the ability to assign IP address to a jail, auto starting jails
at boot time, and a general shift in thought about the occupant of the
jail. The customized streamline apache web server jail that had no way
too be easily configured, progressed into a complete clone of the
operating system with all the customizing options one is familiar with on
the host. The major shortcoming of this type jail system is each jail
has its own copy of the running system binaries. FreeBSD reserves a
limited number of control structures for storing files and directories,
called inodes. Creating a few jails consumes many of those valuable
inodes, eventually preventing the creation of new jails. Worse yet is
each jail loads its own copy of the running binaries into memory, which
causes thrashing on the swap device as memory pages are swapped in and
out as the limited memory is shared between the host and the jails. Besides
consuming resources and creating performance degradation, this also
causes a major administration headache when wanting to update the host
running system, because the host and the jails have to be running the
same version of the binaries. Jail deployments greater that four were
Then about RELEASE 5.4 the creative use of the nullfs command added the
ability for jails to share a single set of the running binaries between
all the jails. This third generation solution solved the performance
problems of the second generation, but had its own problems. Setting up
a nullfs running binaries environment to support multiple jails was a
undocumented manual one. Plus a second type of jail became available
called an "image". The image jail introduced the ability to predefine the
amount of disk space a jail could consume. This was accomplished with the
mdconfig command, which mounts a flat file as a directory tree. Jail
deployments greater than 10 were uncommon. The administration of this jail
system type became increasing difficult with each newly added jail.
During FreeBSD RELEASE 8.0, "qjail" was introduced which is a wrapper
that camouflages the underlying "jail" commands and automates those
manual setup steps into a single command. Mounting a directory tree
containing the running binaries as read-only files using "nullfs" became
the method forming the basic design of the "qjail" jail system. The
functions necessary to manage jails were condensed into the following
commands, "install" for installing an pristine copy of the RELEASE
version of FreeBSD, the "create" command to create both directory tree
type jails and sparse image type jails. This includes the ability to
assign IP address with their network device name, so aliases are auto
created on jail start and auto removed on jail stop. An archived seed
jail pre-configured with ports can be used as the template to form new
jails. To make the deployment of many jails with the same configuration,
jails can be auto duplicated while at the same time incrementing the last
octal of the IP address. The archive, restore and delete commands are
commonly understood functions. The "update" command for using the
portsnap command to populate a complete ports tree, and the ability to
copy the hosts running binaries after a host RELEASE upgrade. A "list"
command to display the qjail jail status. The "config" command can flag a
jail as "norun" to exclude it from being auto started at boot time. The
"norun" status can be toggled back and forth on a single jail or all
jails at once. Jails can be renamed and their IP address changed.
Qjail deploys two different jail types. The first type is based on a
Directory tree. This type has unlimited disk space growth potential,it
shares the hosts disk space. The jail will never run out of space until
the host does. The second type is based on a sparse image file. A sparse
file is one that occupies only the sum size of its contents, not its
allocation size. IE; a sparse file allocated size of 5M, but only having
7 files, each 1k in size, only occupies 7k of physical disk space. As
content is added, additional physical disk space is occupied up to the 5M
allocation ceiling. The sparse file is mounted as a memory disk using the
mdconfig command and populated with the directory tree content of a jail.
This configuration is called a sparse image jail. Its major benefits is
it provides a way to put a hard limit on the maximum amount of disk space
a jail can consume. This provides an addition level of protection to the
host from intentional or unintentional run-a-way processes inside of a
jail consuming disk space until the host system dies.
But by far "qjail" greatest achievement to the advancement of jailed
systems, is the addition of "user-friendliness" that simplifies the
management of large deployments of hundreds of jails. This enhancement
adds the ability to designate a portion of the jail name as a group
prefix so the command being executed will apply to only those jail names
matching that prefix. A simple jail naming convention allows the grouping
of like function jails together. The other advancement is the ability to
create different "zones" consisting of identical jail systems each with
their own groups of jails.
Qjail reduces the complexities of large jail deployments to the novice
level. Qjail has a fully documented manpage, which is a rarity in the
FreeBSD world. Details are given to facilitate the use of qjails
capabilities to the fullest extent possible.
The qjail system is comprised of four components, qjail, qjail2,
qjail.conf, and qjail-jail2.
qjail is the main workhorse utility. Its located at
/usr/local/bin/qjail.It can install the qjail environment, create new
jails, archive, restore, delete and update jails, open a jail console,
and list the status of all the jails. See qjail(8) for complete usage
qjail.conf contains qjail environment defaults. In most cases the
defaults do not need changing. Its recommended to use the defaults. The
defaults are hard coded in the qjail code. The qjail.conf file as
delivered is located at /usr/local/etc/qjail.conf.sample and is not
required for the qjail system to run. To make a permanent override to the
defaults, you first must remove the .sample suffix.
qjail2 script is located at /usr/local/etc/rc.d/qjail2. Its main purpose
is to start and stop jails when called by qjail. In addition it can auto
start all the jails at boot time. Adding qjail_enable="YES" to
/etc/rc.conf will activate it.
qjail-jail2 script is located at /usr/local/etc/rc.d/qjail-qjail2. Its
a clone of /etc/rc.d/jail thats been modified to display the
"started/stopped" console messages for large number of jails. Its called
by the qjail2 script.