1
\documentclass[11pt,twoside,final,openright]{report}
2
\usepackage{a4,graphicx,html,parskip,setspace,times,xspace,url}
5
\renewcommand{\ttdefault}{pcr}
7
\def\Xend{{Xend}\xspace}
8
\def\xend{{xend}\xspace}
10
\latexhtml{\renewcommand{\path}[1]{{\small {\tt #1}}}}{\renewcommand{\path}[1]{{\tt #1}}}
19
\includegraphics{figs/xenlogo.eps}
24
{\Huge \bf Users' Manual} \\[4mm]
25
{\huge Xen v3.3} \\[80mm]
29
{\bf DISCLAIMER: This documentation is always under active development
30
and as such there may be mistakes and omissions --- watch out for
31
these and please report any you find to the developers' mailing list,
32
xen-devel@lists.xensource.com. The latest version is always available
33
on-line. Contributions of material, suggestions and corrections are
45
Xen is Copyright \copyright 2002-2008, Citrix Systems, Inc., University of Cambridge, UK, XenSource Inc., IBM Corp., Hewlett-Packard Co., Intel Corp., AMD Inc., and others. All rights reserved.
47
Xen is an open-source project. Most portions of Xen are licensed for copying
48
under the terms of the GNU General Public License, version 2. Other portions
49
are licensed under the terms of the GNU Lesser General Public License, the
50
Zope Public License 2.0, or under ``BSD-style'' licenses. Please refer to the
51
COPYING file for details.
53
Xen includes software by Christopher Clark. This software is covered by the
57
Copyright (c) 2002, Christopher Clark. All rights reserved.
59
Redistribution and use in source and binary forms, with or without
60
modification, are permitted provided that the following conditions are met:
63
\item Redistributions of source code must retain the above copyright notice,
64
this list of conditions and the following disclaimer.
66
\item Redistributions in binary form must reproduce the above copyright
67
notice, this list of conditions and the following disclaimer in the
68
documentation and/or other materials provided with the distribution.
70
\item Neither the name of the original author; nor the names of any
71
contributors may be used to endorse or promote products derived from this
72
software without specific prior written permission.
75
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
76
AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
77
IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
78
DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE
79
FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
80
DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
81
SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
82
CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
83
OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
84
OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
93
{ \parskip 0pt plus 1pt
98
% PREPARE FOR MAIN TEXT
99
\pagenumbering{arabic}
105
\renewcommand{\topfraction}{.8}
106
\renewcommand{\bottomfraction}{.8}
107
\renewcommand{\textfraction}{.2}
108
\renewcommand{\floatpagefraction}{.8}
112
%% Chapter Introduction moved to introduction.tex
113
\chapter{Introduction}
116
Xen is an open-source \emph{para-virtualizing} virtual machine monitor
117
(VMM), or ``hypervisor'', for a variety of processor architectures including x86. Xen can securely execute multiple virtual machines on a single physical system with near native performance. Xen facilitates enterprise-grade functionality, including:
120
\item Virtual machines with performance close to native hardware.
121
\item Live migration of running virtual machines between physical hosts.
122
\item Up to 32\footnote{IA64 supports up to 64 virtual CPUs per guest virtual machine} virtual CPUs per guest virtual machine, with VCPU hotplug.
123
\item x86/32 with PAE, x86/64, and IA64 platform support.
124
\item Intel and AMD Virtualization Technology for unmodified guest operating systems (including Microsoft Windows).
125
\item Excellent hardware support (supports almost all Linux device
130
\section{Usage Scenarios}
132
Usage scenarios for Xen include:
135
\item [Server Consolidation.] Move multiple servers onto a single
136
physical host with performance and fault isolation provided at the
137
virtual machine boundaries.
138
\item [Hardware Independence.] Allow legacy applications and operating
139
systems to exploit new hardware.
140
\item [Multiple OS configurations.] Run multiple operating systems
141
simultaneously, for development or testing purposes.
142
\item [Kernel Development.] Test and debug kernel modifications in a
143
sand-boxed virtual machine --- no need for a separate test machine.
144
\item [Cluster Computing.] Management at VM granularity provides more
145
flexibility than separately managing each physical host, but better
146
control and isolation than single-system image solutions,
147
particularly by using live migration for load balancing.
148
\item [Hardware support for custom OSes.] Allow development of new
149
OSes while benefiting from the wide-ranging hardware support of
150
existing OSes such as Linux.
154
\section{Operating System Support}
156
Para-virtualization permits very high performance virtualization, even
157
on architectures like x86 that are traditionally very hard to
160
This approach requires operating systems to be \emph{ported} to run on
161
Xen. Porting an OS to run on Xen is similar to supporting a new
162
hardware platform, however the process is simplified because the
163
para-virtual machine architecture is very similar to the underlying
164
native hardware. Even though operating system kernels must explicitly
165
support Xen, a key feature is that user space applications and
166
libraries \emph{do not} require modification.
168
With hardware CPU virtualization as provided by Intel VT and AMD
169
SVM technology, the ability to run an unmodified guest OS kernel
170
is available. No porting of the OS is required, although some
171
additional driver support is necessary within Xen itself. Unlike
172
traditional full virtualization hypervisors, which suffer a tremendous
173
performance overhead, the combination of Xen and VT or Xen and
174
Pacifica technology complement one another to offer superb performance
175
for para-virtualized guest operating systems and full support for
176
unmodified guests running natively on the processor.
178
Paravirtualized Xen support is available for increasingly many
179
operating systems: currently, mature Linux support is available and
180
included in the standard distribution. Other OS ports, including
181
NetBSD, FreeBSD and Solaris are also complete.
184
\section{Hardware Support}
186
Xen currently runs on the IA64 and x86 architectures. Multiprocessor
187
machines are supported, and there is support for HyperThreading (SMT).
189
The default 32-bit Xen requires processor support for Physical
190
Addressing Extensions (PAE), which enables the hypervisor to address
191
up to 16GB of physical memory. Xen also supports x86/64 platforms
192
such as Intel EM64T and AMD Opteron which can currently address up to
193
1TB of physical memory.
195
Xen offloads most of the hardware support issues to the guest OS
196
running in the \emph{Domain~0} management virtual machine. Xen itself
197
contains only the code required to detect and start secondary
198
processors, set up interrupt routing, and perform PCI bus
199
enumeration. Device drivers run within a privileged guest OS rather
200
than within Xen itself. This approach provides compatibility with the
201
majority of device hardware supported by Linux. The default XenLinux
202
build contains support for most server-class network and disk
203
hardware, but you can add support for other hardware by configuring
204
your XenLinux kernel in the normal way.
207
\section{Structure of a Xen-Based System}
209
A Xen system has multiple layers, the lowest and most privileged of
212
Xen may host multiple \emph{guest} operating systems, each of which is
213
executed within a secure virtual machine. In Xen terminology, a
214
\emph{domain}. Domains are scheduled by Xen to make effective use of the
215
available physical CPUs. Each guest OS manages its own applications.
216
This management includes the responsibility of scheduling each
217
application within the time allotted to the VM by Xen.
219
The first domain, \emph{domain~0}, is created automatically when the
220
system boots and has special management privileges. Domain~0 builds
221
other domains and manages their virtual devices. It also performs
222
administrative tasks such as suspending, resuming and migrating other
225
Within domain~0, a process called \emph{xend} runs to manage the system.
226
\Xend\ is responsible for managing virtual machines and providing access
227
to their consoles. Commands are issued to \xend\ over an HTTP interface,
228
via a command-line tool.
233
Xen was originally developed by the Systems Research Group at the
234
University of Cambridge Computer Laboratory as part of the XenoServers
235
project, funded by the UK-EPSRC\@.
237
XenoServers aim to provide a ``public infrastructure for global
238
distributed computing''. Xen plays a key part in that, allowing one to
239
efficiently partition a single machine to enable multiple independent
240
clients to run their operating systems and applications in an
241
environment. This environment provides protection, resource isolation
242
and accounting. The project web page contains further information along
243
with pointers to papers and technical reports:
244
\path{http://www.cl.cam.ac.uk/xeno}
246
Xen has grown into a fully-fledged project in its own right, enabling us
247
to investigate interesting research issues regarding the best techniques
248
for virtualizing resources such as the CPU, memory, disk and network.
249
Project contributors now include Citrix, Intel, IBM, HP, AMD, Novell,
250
RedHat, Sun, Fujitsu, and Samsung.
252
Xen was first described in a paper presented at SOSP in
254
http://www.cl.cam.ac.uk/netos/papers/2003-xensosp.pdf}, and the first
255
public release (1.0) was made that October. Since then, Xen has
256
significantly matured and is now used in production scenarios on many
264
\item IO Emulation (stub domains) for HVM IO performance and scailability
265
\item Replacement of Intel VT vmxassist by new 16b emulation code
266
\item Improved VT-d device pass-through e.g. for graphics devices
267
\item Enhanced C and P state power management
268
\item Exploitation of multi-queue support on modern NICs
269
\item Removal of domain lock for improved PV guest scalability
270
\item 2MB page support for HVM and PV guests
271
\item CPU Portability
274
Xen 3.3 delivers the capabilities needed by enterprise customers and gives computing industry leaders a solid, secure platform to build upon for their virtualization solutions. This latest release establishes Xen as the definitive open source solution for virtualization.
280
%% Chapter Basic Installation
281
\chapter{Basic Installation}
283
The Xen distribution includes three main components: Xen itself, ports
284
of Linux and NetBSD to run on Xen, and the userspace tools required to
285
manage a Xen-based system. This chapter describes how to install the
286
Xen~3.3 distribution from source. Alternatively, there may be pre-built
287
packages available as part of your operating system distribution.
290
\section{Prerequisites}
291
\label{sec:prerequisites}
293
The following is a full list of prerequisites. Items marked `$\dag$' are
294
required by the \xend\ control tools, and hence required if you want to
295
run more than one virtual machine; items marked `$*$' are only required
296
if you wish to build from source.
298
\item A working Linux distribution using the GRUB bootloader and running
299
on a P6-class or newer CPU\@.
300
\item [$\dag$] The \path{iproute2} package.
301
\item [$\dag$] The Linux bridge-utils\footnote{Available from {\tt
302
http://bridge.sourceforge.net}} (e.g., \path{/sbin/brctl})
303
\item [$\dag$] The Linux hotplug system\footnote{Available from {\tt
304
http://linux-hotplug.sourceforge.net/}} (e.g.,
305
\path{/sbin/hotplug} and related scripts). On newer distributions,
306
this is included alongside the Linux udev system\footnote{See {\tt
307
http://www.kernel.org/pub/linux/utils/kernel/hotplug/udev.html/}}.
308
\item [$*$] Build tools (gcc v3.2.x or v3.3.x, binutils, GNU make).
309
\item [$*$] Development installation of zlib (e.g.,\ zlib-dev).
310
\item [$*$] Development installation of Python v2.2 or later (e.g.,\
312
\item [$*$] \LaTeX\ and transfig are required to build the
316
Once you have satisfied these prerequisites, you can now install either
317
a binary or source distribution of Xen.
319
\section{Installing from Binary Tarball}
321
Pre-built tarballs are available for download from the XenSource downloads
323
\begin{quote} {\tt http://www.xensource.com/downloads/}
326
Once you've downloaded the tarball, simply unpack and install:
328
# tar zxvf xen-3.0-install.tgz
333
Once you've installed the binaries you need to configure your system as
334
described in Section~\ref{s:configure}.
336
\section{Installing from RPMs}
337
Pre-built RPMs are available for download from the XenSource downloads
339
\begin{quote} {\tt http://www.xensource.com/downloads/}
342
Once you've downloaded the RPMs, you typically install them via the
345
\verb|# rpm -iv rpmname|
347
See the instructions and the Release Notes for each RPM set referenced at:
349
{\tt http://www.xensource.com/downloads/}.
352
\section{Installing from Source}
354
This section describes how to obtain, build and install Xen from source.
356
\subsection{Obtaining the Source}
358
The Xen source tree is available as either a compressed source tarball
359
or as a clone of our master Mercurial repository.
362
\item[Obtaining the Source Tarball]\mbox{} \\
363
Stable versions and daily snapshots of the Xen source tree are
364
available from the Xen download page:
365
\begin{quote} {\tt \tt http://www.xensource.com/downloads/}
367
\item[Obtaining the source via Mercurial]\mbox{} \\
368
The source tree may also be obtained via the public Mercurial
370
\begin{quote}{\tt http://xenbits.xensource.com}
371
\end{quote} See the instructions and the Getting Started Guide
374
{\tt http://www.xensource.com/downloads/}
378
% \section{The distribution}
380
% The Xen source code repository is structured as follows:
382
% \begin{description}
383
% \item[\path{tools/}] Xen node controller daemon (Xend), command line
384
% tools, control libraries
385
% \item[\path{xen/}] The Xen VMM.
386
% \item[\path{buildconfigs/}] Build configuration files
387
% \item[\path{linux-*-xen-sparse/}] Xen support for Linux.
388
% \item[\path{patches/}] Experimental patches for Linux.
389
% \item[\path{docs/}] Various documentation files for users and
391
% \item[\path{extras/}] Bonus extras.
394
\subsection{Building from Source}
396
The top-level Xen Makefile includes a target ``world'' that will do the
401
\item Build the control tools, including \xend.
402
\item Download (if necessary) and unpack the Linux 2.6 source code, and
403
patch it for use with Xen.
404
\item Build a Linux kernel to use in domain~0 and a smaller unprivileged
405
kernel, which can be used for unprivileged virtual machines.
408
After the build has completed you should have a top-level directory
409
called \path{dist/} in which all resulting targets will be placed. Of
410
particular interest are the two XenLinux kernel images, one with a
411
``-xen0'' extension which contains hardware device drivers and drivers
412
for Xen's virtual devices, and one with a ``-xenU'' extension that
413
just contains the virtual ones. These are found in
414
\path{dist/install/boot/} along with the image for Xen itself and the
415
configuration files used during the build.
417
%The NetBSD port can be built using:
421
%\end{verbatim}\end{quote}
422
%NetBSD port is built using a snapshot of the netbsd-2-0 cvs branch.
423
%The snapshot is downloaded as part of the build process if it is not
424
%yet present in the \path{NETBSD\_SRC\_PATH} search path. The build
425
%process also downloads a toolchain which includes all of the tools
426
%necessary to build the NetBSD kernel under Linux.
428
To customize the set of kernels built you need to edit the top-level
429
Makefile. Look for the line:
432
KERNELS ?= linux-2.6-xen0 linux-2.6-xenU
436
You can edit this line to include any set of operating system kernels
437
which have configurations in the top-level \path{buildconfigs/}
440
%% Inspect the Makefile if you want to see what goes on during a
441
%% build. Building Xen and the tools is straightforward, but XenLinux
442
%% is more complicated. The makefile needs a `pristine' Linux kernel
443
%% tree to which it will then add the Xen architecture files. You can
444
%% tell the makefile the location of the appropriate Linux compressed
446
%% setting the LINUX\_SRC environment variable, e.g. \\
447
%% \verb!# LINUX_SRC=/tmp/linux-2.6.11.tar.bz2 make world! \\ or by
448
%% placing the tar file somewhere in the search path of {\tt
449
%% LINUX\_SRC\_PATH} which defaults to `{\tt .:..}'. If the
450
%% makefile can't find a suitable kernel tar file it attempts to
451
%% download it from kernel.org (this won't work if you're behind a
454
%% After untaring the pristine kernel tree, the makefile uses the {\tt
455
%% mkbuildtree} script to add the Xen patches to the kernel.
457
%% \framebox{\parbox{5in}{
458
%% {\bf Distro specific:} \\
459
%% {\it Gentoo} --- if not using udev (most installations,
460
%% currently), you'll need to enable devfs and devfs mount at boot
461
%% time in the xen0 config. }}
463
\subsection{Custom Kernels}
465
% If you have an SMP machine you may wish to give the {\tt '-j4'}
466
% argument to make to get a parallel build.
468
If you wish to build a customized XenLinux kernel (e.g.\ to support
469
additional devices or enable distribution-required features), you can
470
use the standard Linux configuration mechanisms, specifying that the
471
architecture being built for is \path{xen}, e.g:
474
# cd linux-2.6.12-xen0
475
# make ARCH=xen xconfig
481
You can also copy an existing Linux configuration (\path{.config}) into
482
e.g.\ \path{linux-2.6.12-xen0} and execute:
485
# make ARCH=xen oldconfig
489
You may be prompted with some Xen-specific options. We advise accepting
490
the defaults for these options.
492
Note that the only difference between the two types of Linux kernels
493
that are built is the configuration file used for each. The ``U''
494
suffixed (unprivileged) versions don't contain any of the physical
495
hardware device drivers, leading to a 30\% reduction in size; hence you
496
may prefer these for your non-privileged domains. The ``0'' suffixed
497
privileged versions can be used to boot the system, as well as in driver
498
domains and unprivileged domains.
500
\subsection{Installing Generated Binaries}
502
The files produced by the build process are stored under the
503
\path{dist/install/} directory. To install them in their default
511
Alternatively, users with special installation requirements may wish to
512
install them manually by copying the files to their appropriate
515
%% Files in \path{install/boot/} include:
517
%% \item \path{install/boot/xen-3.0.gz} Link to the Xen 'kernel'
518
%% \item \path{install/boot/vmlinuz-2.6-xen0} Link to domain 0
520
%% \item \path{install/boot/vmlinuz-2.6-xenU} Link to unprivileged
524
The \path{dist/install/boot} directory will also contain the config
525
files used for building the XenLinux kernels, and also versions of Xen
526
and XenLinux kernels that contain debug symbols such as
527
(\path{xen-syms-3.0.0} and \path{vmlinux-syms-2.6.12.6-xen0}) which are
528
essential for interpreting crash dumps. Retain these files as the
529
developers may wish to see them if you post on the mailing list.
532
\section{Configuration}
535
Once you have built and installed the Xen distribution, it is simple to
536
prepare the machine for booting and running Xen.
538
\subsection{GRUB Configuration}
540
An entry should be added to \path{grub.conf} (often found under
541
\path{/boot/} or \path{/boot/grub/}) to allow Xen / XenLinux to boot.
542
This file is sometimes called \path{menu.lst}, depending on your
543
distribution. The entry should look something like the following:
545
%% KMSelf Thu Dec 1 19:06:13 PST 2005 262144 is useful for RHEL/RH and
549
title Xen 3.0 / XenLinux 2.6
550
kernel /boot/xen-3.0.gz dom0_mem=262144
551
module /boot/vmlinuz-2.6-xen0 root=/dev/sda4 ro console=tty0
555
The kernel line tells GRUB where to find Xen itself and what boot
556
parameters should be passed to it (in this case, setting the domain~0
557
memory allocation in kilobytes and the settings for the serial port).
558
For more details on the various Xen boot parameters see
559
Section~\ref{s:xboot}.
561
The module line of the configuration describes the location of the
562
XenLinux kernel that Xen should start and the parameters that should be
563
passed to it. These are standard Linux parameters, identifying the root
564
device and specifying it be initially mounted read only and instructing
565
that console output be sent to the screen. Some distributions such as
566
SuSE do not require the \path{ro} parameter.
568
%% \framebox{\parbox{5in}{
569
%% {\bf Distro specific:} \\
570
%% {\it SuSE} --- Omit the {\tt ro} option from the XenLinux
571
%% kernel command line, since the partition won't be remounted rw
574
To use an initrd, add another \path{module} line to the configuration,
577
module /boot/my_initrd.gz
581
%% KMSelf Thu Dec 1 19:05:30 PST 2005 Other configs as an appendix?
583
When installing a new kernel, it is recommended that you do not delete
584
existing menu options from \path{menu.lst}, as you may wish to boot your
585
old Linux kernel in future, particularly if you have problems.
587
\subsection{Serial Console (optional)}
589
Serial console access allows you to manage, monitor, and interact with
590
your system over a serial console. This can allow access from another
591
nearby system via a null-modem (``LapLink'') cable or remotely via a serial
594
You system's BIOS, bootloader (GRUB), Xen, Linux, and login access must
595
each be individually configured for serial console access. It is
596
\emph{not} strictly necessary to have each component fully functional,
597
but it can be quite useful.
599
For general information on serial console configuration under Linux,
600
refer to the ``Remote Serial Console HOWTO'' at The Linux Documentation
601
Project: \url{http://www.tldp.org}
603
\subsubsection{Serial Console BIOS configuration}
605
Enabling system serial console output neither enables nor disables
606
serial capabilities in GRUB, Xen, or Linux, but may make remote
607
management of your system more convenient by displaying POST and other
608
boot messages over serial port and allowing remote BIOS configuration.
610
Refer to your hardware vendor's documentation for capabilities and
611
procedures to enable BIOS serial redirection.
614
\subsubsection{Serial Console GRUB configuration}
616
Enabling GRUB serial console output neither enables nor disables Xen or
617
Linux serial capabilities, but may made remote management of your system
618
more convenient by displaying GRUB prompts, menus, and actions over
619
serial port and allowing remote GRUB management.
621
Adding the following two lines to your GRUB configuration file,
622
typically either \path{/boot/grub/menu.lst} or \path{/boot/grub/grub.conf}
623
depending on your distro, will enable GRUB serial output.
626
{\small \begin{verbatim}
627
serial --unit=0 --speed=115200 --word=8 --parity=no --stop=1
628
terminal --timeout=10 serial console
632
Note that when both the serial port and the local monitor and keyboard
633
are enabled, the text ``\emph{Press any key to continue}'' will appear
634
at both. Pressing a key on one device will cause GRUB to display to
635
that device. The other device will see no output. If no key is
636
pressed before the timeout period expires, the system will boot to the
637
default GRUB boot entry.
639
Please refer to the GRUB documentation for further information.
642
\subsubsection{Serial Console Xen configuration}
644
Enabling Xen serial console output neither enables nor disables Linux
645
kernel output or logging in to Linux over serial port. It does however
646
allow you to monitor and log the Xen boot process via serial console and
647
can be very useful in debugging.
649
%% kernel /boot/xen-2.0.gz dom0_mem=131072 console=com1,vga com1=115200,8n1
650
%% module /boot/vmlinuz-2.6-xen0 root=/dev/sda4 ro
652
In order to configure Xen serial console output, it is necessary to
653
add a boot option to your GRUB config; e.g.\ replace the previous
654
example kernel line with:
655
\begin{quote} {\small \begin{verbatim}
656
kernel /boot/xen.gz dom0_mem=131072 com1=115200,8n1 console=com1,vga
660
This configures Xen to output on COM1 at 115,200 baud, 8 data bits, no
661
parity and 1 stop bit. Modify these parameters for your environment.
662
See Section~\ref{s:xboot} for an explanation of all boot parameters.
664
One can also configure XenLinux to share the serial console; to achieve
665
this append ``\path{console=ttyS0}'' to your module line.
668
\subsubsection{Serial Console Linux configuration}
670
Enabling Linux serial console output at boot neither enables nor
671
disables logging in to Linux over serial port. It does however allow
672
you to monitor and log the Linux boot process via serial console and can be
673
very useful in debugging.
675
To enable Linux output at boot time, add the parameter
676
\path{console=ttyS0} (or ttyS1, ttyS2, etc.) to your kernel GRUB line.
677
Under Xen, this might be:
679
{\footnotesize \begin{verbatim}
680
module /vmlinuz-2.6-xen0 ro root=/dev/VolGroup00/LogVol00 \
681
console=ttyS0, 115200
684
to enable output over ttyS0 at 115200 baud.
688
\subsubsection{Serial Console Login configuration}
690
Logging in to Linux via serial console, under Xen or otherwise, requires
691
specifying a login prompt be started on the serial port. To permit root
692
logins over serial console, the serial port must be added to
693
\path{/etc/securetty}.
696
To automatically start a login prompt over the serial port,
697
add the line: \begin{quote} {\small {\tt c:2345:respawn:/sbin/mingetty
698
ttyS0}} \end{quote} to \path{/etc/inittab}. Run \path{init q} to force
699
a reload of your inttab and start getty.
701
To enable root logins, add \path{ttyS0} to \path{/etc/securetty} if not
704
Your distribution may use an alternate getty; options include getty,
705
mgetty and agetty. Consult your distribution's documentation
706
for further information.
709
\subsection{TLS Libraries}
711
Users of the XenLinux 2.6 kernel should disable Thread Local Storage
712
(TLS) (e.g.\ by doing a \path{mv /lib/tls /lib/tls.disabled}) before
713
attempting to boot a XenLinux kernel\footnote{If you boot without first
714
disabling TLS, you will get a warning message during the boot process.
715
In this case, simply perform the rename after the machine is up and
716
then run \path{/sbin/ldconfig} to make it take effect.}. You can
717
always reenable TLS by restoring the directory to its original location
718
(i.e.\ \path{mv /lib/tls.disabled /lib/tls}).
720
The reason for this is that the current TLS implementation uses
721
segmentation in a way that is not permissible under Xen. If TLS is not
722
disabled, an emulation mode is used within Xen which reduces performance
723
substantially. To ensure full performance you should install a
724
`Xen-friendly' (nosegneg) version of the library.
727
\section{Booting Xen}
729
It should now be possible to restart the system and use Xen. Reboot and
730
choose the new Xen option when the Grub screen appears.
732
What follows should look much like a conventional Linux boot. The first
733
portion of the output comes from Xen itself, supplying low level
734
information about itself and the underlying hardware. The last portion
735
of the output comes from XenLinux.
737
You may see some error messages during the XenLinux boot. These are not
738
necessarily anything to worry about---they may result from kernel
739
configuration differences between your XenLinux kernel and the one you
742
When the boot completes, you should be able to log into your system as
743
usual. If you are unable to log in, you should still be able to reboot
744
with your normal Linux kernel by selecting it at the GRUB prompt.
748
\chapter{Booting a Xen System}
750
Booting the system into Xen will bring you up into the privileged
751
management domain, Domain0. At that point you are ready to create
752
guest domains and ``boot'' them using the \texttt{xm create} command.
754
\section{Booting Domain0}
756
After installation and configuration is complete, reboot the system
757
and and choose the new Xen option when the Grub screen appears.
759
What follows should look much like a conventional Linux boot. The
760
first portion of the output comes from Xen itself, supplying low level
761
information about itself and the underlying hardware. The last
762
portion of the output comes from XenLinux.
764
%% KMSelf Wed Nov 30 18:09:37 PST 2005: We should specify what these are.
766
When the boot completes, you should be able to log into your system as
767
usual. If you are unable to log in, you should still be able to
768
reboot with your normal Linux kernel by selecting it at the GRUB prompt.
770
The first step in creating a new domain is to prepare a root
771
filesystem for it to boot. Typically, this might be stored in a normal
772
partition, an LVM or other volume manager partition, a disk file or on
773
an NFS server. A simple way to do this is simply to boot from your
774
standard OS install CD and install the distribution into another
775
partition on your hard drive.
777
To start the \xend\ control daemon, type
782
If you wish the daemon to start automatically, see the instructions in
783
Section~\ref{s:xend}. Once the daemon is running, you can use the
784
\path{xm} tool to monitor and maintain the domains running on your
785
system. This chapter provides only a brief tutorial. We provide full
786
details of the \path{xm} tool in the next chapter.
788
% \section{From the web interface}
790
% Boot the Xen machine and start Xensv (see Chapter~\ref{cha:xensv}
791
% for more details) using the command: \\
792
% \verb_# xensv start_ \\
793
% This will also start Xend (see Chapter~\ref{cha:xend} for more
796
% The domain management interface will then be available at {\tt
797
% http://your\_machine:8080/}. This provides a user friendly wizard
798
% for starting domains and functions for managing running domains.
800
% \section{From the command line}
801
\section{Booting Guest Domains}
803
\subsection{Creating a Domain Configuration File}
805
Before you can start an additional domain, you must create a
806
configuration file. We provide two example files which you can use as
809
\item \path{/etc/xen/xmexample1} is a simple template configuration
810
file for describing a single VM\@.
811
\item \path{/etc/xen/xmexample2} file is a template description that
812
is intended to be reused for multiple virtual machines. Setting the
813
value of the \path{vmid} variable on the \path{xm} command line
814
fills in parts of this template.
817
There are also a number of other examples which you may find useful.
818
Copy one of these files and edit it as appropriate. Typical values
819
you may wish to edit include:
823
\item[kernel] Set this to the path of the kernel you compiled for use
824
with Xen (e.g.\ \path{kernel = ``/boot/vmlinuz-2.6-xenU''})
825
\item[memory] Set this to the size of the domain's memory in megabytes
826
(e.g.\ \path{memory = 64})
827
\item[disk] Set the first entry in this list to calculate the offset
828
of the domain's root partition, based on the domain ID\@. Set the
829
second to the location of \path{/usr} if you are sharing it between
830
domains (e.g.\ \path{disk = ['phy:your\_hard\_drive\%d,sda1,w' \%
831
(base\_partition\_number + vmid),
832
'phy:your\_usr\_partition,sda6,r' ]}
833
\item[dhcp] Uncomment the dhcp variable, so that the domain will
834
receive its IP address from a DHCP server (e.g.\ \path{dhcp=``dhcp''})
838
You may also want to edit the {\bf vif} variable in order to choose
839
the MAC address of the virtual ethernet interface yourself. For
843
\verb_vif = ['mac=00:16:3E:F6:BB:B3']_
845
If you do not set this variable, \xend\ will automatically generate a
846
random MAC address from the range 00:16:3E:xx:xx:xx, assigned by IEEE to
847
XenSource as an OUI (organizationally unique identifier). XenSource
848
Inc. gives permission for anyone to use addresses randomly allocated
849
from this range for use by their Xen domains.
851
For a list of IEEE OUI assignments, see
852
\url{http://standards.ieee.org/regauth/oui/oui.txt}
855
\subsection{Booting the Guest Domain}
857
The \path{xm} tool provides a variety of commands for managing
858
domains. Use the \path{create} command to start new domains. Assuming
859
you've created a configuration file \path{myvmconf} based around
860
\path{/etc/xen/xmexample2}, to start a domain with virtual machine
861
ID~1 you should type:
865
# xm create -c myvmconf vmid=1
869
The \path{-c} switch causes \path{xm} to turn into the domain's
870
console after creation. The \path{vmid=1} sets the \path{vmid}
871
variable used in the \path{myvmconf} file.
873
You should see the console boot messages from the new domain appearing
874
in the terminal in which you typed the command, culminating in a login
878
\section{Starting / Stopping Domains Automatically}
880
It is possible to have certain domains start automatically at boot
881
time and to have dom0 wait for all running domains to shutdown before
882
it shuts down the system.
884
To specify a domain is to start at boot-time, place its configuration
885
file (or a link to it) under \path{/etc/xen/auto/}.
887
A Sys-V style init script for Red Hat and LSB-compliant systems is
888
provided and will be automatically copied to \path{/etc/init.d/}
889
during install. You can then enable it in the appropriate way for
892
For instance, on Red Hat:
895
\verb_# chkconfig --add xendomains_
898
By default, this will start the boot-time domains in runlevels 3, 4
901
You can also use the \path{service} command to run this script
905
\verb_# service xendomains start_
907
Starts all the domains with config files under /etc/xen/auto/.
911
\verb_# service xendomains stop_
913
Shuts down all running Xen domains.
918
\part{Configuration and Management}
920
%% Chapter Domain Management Tools and Daemons
921
\chapter{Domain Management Tools}
923
This chapter summarizes the management software and tools available.
930
The \Xend\ node control daemon performs system management functions
931
related to virtual machines. It forms a central point of control of
932
virtualized resources, and must be running in order to start and manage
933
virtual machines. \Xend\ must be run as root because it needs access to
934
privileged system management functions.
936
An initialization script named \texttt{/etc/init.d/xend} is provided to
937
start \Xend\ at boot time. Use the tool appropriate (i.e. chkconfig) for
938
your Linux distribution to specify the runlevels at which this script
939
should be executed, or manually create symbolic links in the correct
940
runlevel directories.
942
\Xend\ can be started on the command line as well, and supports the
943
following set of parameters:
946
\verb!# xend start! & start \xend, if not already running \\
947
\verb!# xend stop! & stop \xend\ if already running \\
948
\verb!# xend restart! & restart \xend\ if running, otherwise start it \\
949
% \verb!# xend trace_start! & start \xend, with very detailed debug logging \\
950
\verb!# xend status! & indicates \xend\ status by its return code
953
A SysV init script called {\tt xend} is provided to start \xend\ at
954
boot time. {\tt make install} installs this script in
955
\path{/etc/init.d}. To enable it, you have to make symbolic links in
956
the appropriate runlevel directories or use the {\tt chkconfig} tool,
957
where available. Once \xend\ is running, administration can be done
958
using the \texttt{xm} tool.
962
As \xend\ runs, events will be logged to \path{/var/log/xen/xend.log} and
963
(less frequently) to \path{/var/log/xen/xend-debug.log}. These, along with
964
the standard syslog files, are useful when troubleshooting problems.
966
\subsection{Configuring \Xend\ }
968
\Xend\ is written in Python. At startup, it reads its configuration
969
information from the file \path{/etc/xen/xend-config.sxp}. The Xen
970
installation places an example \texttt{xend-config.sxp} file in the
971
\texttt{/etc/xen} subdirectory which should work for most installations.
973
See the example configuration file \texttt{xend-debug.sxp} and the
974
section 5 man page \texttt{xend-config.sxp} for a full list of
975
parameters and more detailed information. Some of the most important
976
parameters are discussed below.
978
An HTTP interface and a Unix domain socket API are available to
979
communicate with \Xend. This allows remote users to pass commands to the
980
daemon. By default, \Xend does not start an HTTP server. It does start a
981
Unix domain socket management server, as the low level utility
982
\texttt{xm} requires it. For support of cross-machine migration, \Xend\
983
can start a relocation server. This support is not enabled by default
984
for security reasons.
986
Note: the example \texttt{xend} configuration file modifies the defaults and
987
starts up \Xend\ as an HTTP server as well as a relocation server.
992
#(xend-http-server no)
993
(xend-http-server yes)
994
#(xend-unix-server yes)
995
#(xend-relocation-server no)
996
(xend-relocation-server yes)
999
Comment or uncomment lines in that file to disable or enable features
1002
Connections from remote hosts are disabled by default:
1005
# Address xend should listen on for HTTP connections, if xend-http-server is
1007
# Specifying 'localhost' prevents remote connections.
1008
# Specifying the empty string '' (the default) allows all connections.
1010
(xend-address localhost)
1013
It is recommended that if migration support is not needed, the
1014
\texttt{xend-relocation-server} parameter value be changed to
1015
``\texttt{no}'' or commented out.
1020
The xm tool is the primary tool for managing Xen from the console. The
1021
general format of an xm command line is:
1024
# xm command [switches] [arguments] [variables]
1027
The available \emph{switches} and \emph{arguments} are dependent on the
1028
\emph{command} chosen. The \emph{variables} may be set using
1029
declarations of the form {\tt variable=value} and command line
1030
declarations override any of the values in the configuration file being
1031
used, including the standard variables described above and any custom
1032
variables (for instance, the \path{xmdefconfig} file uses a {\tt vmid}
1035
For online help for the commands available, type:
1043
This will list the most commonly used commands. The full list can be obtained
1044
using \verb_xm help --long_. You can also type \path{xm help $<$command$>$}
1045
for more information on a given command.
1047
\subsection{Basic Management Commands}
1049
One useful command is \verb_# xm list_ which lists all domains running in rows
1050
of the following format:
1051
\begin{center} {\tt name domid memory vcpus state cputime}
1054
The meaning of each field is as follows:
1057
\item[name] The descriptive name of the virtual machine.
1058
\item[domid] The number of the domain ID this virtual machine is
1060
\item[memory] Memory size in megabytes.
1061
\item[vcpus] The number of virtual CPUs this domain has.
1062
\item[state] Domain state consists of 5 fields:
1070
\item[cputime] How much CPU time (in seconds) the domain has used so
1075
The \path{xm list} command also supports a long output format when the
1076
\path{-l} switch is used. This outputs the full details of the
1077
running domains in \xend's SXP configuration format.
1079
If you want to know how long your domains have been running for, then
1080
you can use the \verb_# xm uptime_ command.
1083
You can get access to the console of a particular domain using
1084
the \verb_# xm console_ command (e.g.\ \verb_# xm console myVM_).
1086
\subsection{Domain Scheduling Management Commands}
1088
The credit CPU scheduler automatically load balances guest VCPUs
1089
across all available physical CPUs on an SMP host. The user need
1090
not manually pin VCPUs to load balance the system. However, she
1091
can restrict which CPUs a particular VCPU may run on using
1092
the \path{xm vcpu-pin} command.
1094
Each guest domain is assigned a \path{weight} and a \path{cap}.
1096
A domain with a weight of 512 will get twice as much CPU as a
1097
domain with a weight of 256 on a contended host. Legal weights
1098
range from 1 to 65535 and the default is 256.
1100
The cap optionally fixes the maximum amount of CPU a guest will
1101
be able to consume, even if the host system has idle CPU cycles.
1102
The cap is expressed in percentage of one physical CPU: 100 is
1103
1 physical CPU, 50 is half a CPU, 400 is 4 CPUs, etc... The
1104
default, 0, means there is no upper cap.
1106
When you are running with the credit scheduler, you can check and
1107
modify your domains' weights and caps using the \path{xm sched-credit}
1111
\verb!xm sched-credit -d <domain>! & lists weight and cap \\
1112
\verb!xm sched-credit -d <domain> -w <weight>! & sets the weight \\
1113
\verb!xm sched-credit -d <domain> -c <cap>! & sets the cap
1118
%% Chapter Domain Configuration
1119
\chapter{Domain Configuration}
1122
The following contains the syntax of the domain configuration files
1123
and description of how to further specify networking, driver domain
1124
and general scheduling behavior.
1127
\section{Configuration Files}
1130
Xen configuration files contain the following standard variables.
1131
Unless otherwise stated, configuration items should be enclosed in
1132
quotes: see the configuration scripts in \path{/etc/xen/}
1133
for concrete examples.
1136
\item[kernel] Path to the kernel image.
1137
\item[ramdisk] Path to a ramdisk image (optional).
1138
% \item[builder] The name of the domain build function (e.g.
1139
% {\tt'linux'} or {\tt'netbsd'}.
1140
\item[memory] Memory size in megabytes.
1141
\item[vcpus] The number of virtual CPUs.
1142
\item[console] Port to export the domain console on (default 9600 +
1144
\item[vif] Network interface configuration. This may simply contain
1145
an empty string for each desired interface, or may override various
1148
vif = [ 'mac=00:16:3E:00:00:11, bridge=xen-br0',
1151
to assign a MAC address and bridge to the first interface and assign
1152
a different bridge to the second interface, leaving \xend\ to choose
1153
the MAC address. The settings that may be overridden in this way are
1154
type, mac, bridge, ip, script, backend, and vifname.
1155
\item[disk] List of block devices to export to the domain e.g.
1156
\verb_disk = [ 'phy:hda1,sda1,r' ]_
1157
exports physical device \path{/dev/hda1} to the domain as
1158
\path{/dev/sda1} with read-only access. Exporting a disk read-write
1159
which is currently mounted is dangerous -- if you are \emph{certain}
1160
you wish to do this, you can specify \path{w!} as the mode.
1161
\item[dhcp] Set to {\tt `dhcp'} if you want to use DHCP to configure
1163
\item[netmask] Manually configured IP netmask.
1164
\item[gateway] Manually configured IP gateway.
1165
\item[hostname] Set the hostname for the virtual machine.
1166
\item[root] Specify the root device parameter on the kernel command
1168
\item[nfs\_server] IP address for the NFS server (if any).
1169
\item[nfs\_root] Path of the root filesystem on the NFS server (if
1171
\item[extra] Extra string to append to the kernel command line (if
1175
Additional fields are documented in the example configuration files
1176
(e.g. to configure virtual TPM functionality).
1178
For additional flexibility, it is also possible to include Python
1179
scripting commands in configuration files. An example of this is the
1180
\path{xmexample2} file, which uses Python code to handle the
1181
\path{vmid} variable.
1184
%\part{Advanced Topics}
1187
\section{Network Configuration}
1189
For many users, the default installation should work ``out of the
1190
box''. More complicated network setups, for instance with multiple
1191
Ethernet interfaces and/or existing bridging setups will require some
1192
special configuration.
1194
The purpose of this section is to describe the mechanisms provided by
1195
\xend\ to allow a flexible configuration for Xen's virtual networking.
1197
\subsection{Xen virtual network topology}
1199
Each domain network interface is connected to a virtual network
1200
interface in dom0 by a point to point link (effectively a ``virtual
1201
crossover cable''). These devices are named {\tt
1202
vif$<$domid$>$.$<$vifid$>$} (e.g.\ {\tt vif1.0} for the first
1203
interface in domain~1, {\tt vif3.1} for the second interface in
1206
Traffic on these virtual interfaces is handled in domain~0 using
1207
standard Linux mechanisms for bridging, routing, rate limiting, etc.
1208
Xend calls on two shell scripts to perform initial configuration of
1209
the network and configuration of new virtual interfaces. By default,
1210
these scripts configure a single bridge for all the virtual
1211
interfaces. Arbitrary routing / bridging configurations can be
1212
configured by customizing the scripts, as described in the following
1215
\subsection{Xen networking scripts}
1217
Xen's virtual networking is configured by two shell scripts (by
1218
default \path{network-bridge} and \path{vif-bridge}). These are called
1219
automatically by \xend\ when certain events occur, with arguments to
1220
the scripts providing further contextual information. These scripts
1221
are found by default in \path{/etc/xen/scripts}. The names and
1222
locations of the scripts can be configured in
1223
\path{/etc/xen/xend-config.sxp}.
1226
\item[network-bridge:] This script is called whenever \xend\ is started or
1227
stopped to respectively initialize or tear down the Xen virtual
1228
network. In the default configuration initialization creates the
1229
bridge `xen-br0' and moves eth0 onto that bridge, modifying the
1230
routing accordingly. When \xend\ exits, it deletes the Xen bridge
1231
and removes eth0, restoring the normal IP and routing configuration.
1233
%% In configurations where the bridge already exists, this script
1234
%% could be replaced with a link to \path{/bin/true} (for instance).
1236
\item[vif-bridge:] This script is called for every domain virtual
1237
interface and can configure firewalling rules and add the vif to the
1238
appropriate bridge. By default, this adds and removes VIFs on the
1242
Other example scripts are available (\path{network-route} and
1243
\path{vif-route}, \path{network-nat} and \path{vif-nat}).
1244
For more complex network setups (e.g.\ where routing is required or
1245
integrate with existing bridges) these scripts may be replaced with
1246
customized variants for your site's preferred configuration.
1248
\section{Driver Domain Configuration}
1254
Individual PCI devices can be assigned to a given domain (a PCI driver domain)
1255
to allow that domain direct access to the PCI hardware.
1257
While PCI Driver Domains can increase the stability and security of a system
1258
by addressing a number of security concerns, there are some security issues
1259
that remain that you can read about in Section~\ref{s:ddsecurity}.
1261
\subsubsection{Compile-Time Setup}
1262
To use this functionality, ensure
1263
that the PCI Backend is compiled in to a privileged domain (e.g. domain 0)
1264
and that the domains which will be assigned PCI devices have the PCI Frontend
1265
compiled in. In XenLinux, the PCI Backend is available under the Xen
1266
configuration section while the PCI Frontend is under the
1267
architecture-specific "Bus Options" section. You may compile both the backend
1268
and the frontend into the same kernel; they will not affect each other.
1270
\subsubsection{PCI Backend Configuration - Binding at Boot}
1271
The PCI devices you wish to assign to unprivileged domains must be "hidden"
1272
from your backend domain (usually domain 0) so that it does not load a driver
1273
for them. Use the \path{pciback.hide} kernel parameter which is specified on
1274
the kernel command-line and is configurable through GRUB (see
1275
Section~\ref{s:configure}). Note that devices are not really hidden from the
1276
backend domain. The PCI Backend appears to the Linux kernel as a regular PCI
1277
device driver. The PCI Backend ensures that no other device driver loads
1278
for the devices by binding itself as the device driver for those devices.
1279
PCI devices are identified by hexadecimal slot/function numbers (on Linux,
1280
use \path{lspci} to determine slot/function numbers of your devices) and
1281
can be specified with or without the PCI domain: \\
1282
\centerline{ {\tt ({\em bus}:{\em slot}.{\em func})} example {\tt (02:1d.3)}} \\
1283
\centerline{ {\tt ({\em domain}:{\em bus}:{\em slot}.{\em func})} example {\tt (0000:02:1d.3)}} \\
1285
An example kernel command-line which hides two PCI devices might be: \\
1286
\centerline{ {\tt root=/dev/sda4 ro console=tty0 pciback.hide=(02:01.f)(0000:04:1d.0) } } \\
1288
\subsubsection{PCI Backend Configuration - Late Binding}
1289
PCI devices can also be bound to the PCI Backend after boot through the manual
1290
binding/unbinding facilities provided by the Linux kernel in sysfs (allowing
1291
for a Xen user to give PCI devices to driver domains that were not specified
1292
on the kernel command-line). There are several attributes with the PCI
1293
Backend's sysfs directory (\path{/sys/bus/pci/drivers/pciback}) that can be
1294
used to bind/unbind devices:
1297
\item[slots] lists all of the PCI slots that the PCI Backend will try to seize
1298
(or "hide" from Domain 0). A PCI slot must appear in this list before it can
1299
be bound to the PCI Backend through the \path{bind} attribute.
1300
\item[new\_slot] write the name of a slot here (in 0000:00:00.0 format) to
1301
have the PCI Backend seize the device in this slot.
1302
\item[remove\_slot] write the name of a slot here (same format as
1303
\path{new\_slot}) to have the PCI Backend no longer try to seize devices in
1304
this slot. Note that this does not unbind the driver from a device it has
1306
\item[bind] write the name of a slot here (in 0000:00:00.0 format) to have
1307
the Linux kernel attempt to bind the device in that slot to the PCI Backend
1309
\item[unbind] write the name of a skit here (same format as \path{bind}) to have
1310
the Linux kernel unbind the device from the PCI Backend. DO NOT unbind a
1311
device while it is currently given to a PCI driver domain!
1316
Bind a device to the PCI Backend which is not bound to any other driver.
1318
# # Add a new slot to the PCI Backend's list
1319
# echo -n 0000:01:04.d > /sys/bus/pci/drivers/pciback/new_slot
1320
# # Now that the backend is watching for the slot, bind to it
1321
# echo -n 0000:01:04.d > /sys/bus/pci/drivers/pciback/bind
1324
Unbind a device from its driver and bind to the PCI Backend.
1326
# # Unbind a PCI network card from its network driver
1327
# echo -n 0000:05:02.0 > /sys/bus/pci/drivers/3c905/unbind
1328
# # And now bind it to the PCI Backend
1329
# echo -n 0000:05:02.0 > /sys/bus/pci/drivers/pciback/new_slot
1330
# echo -n 0000:05:02.0 > /sys/bus/pci/drivers/pciback/bind
1333
Note that the "-n" option in the example is important as it causes echo to not
1336
\subsubsection{PCI Backend Configuration - User-space Quirks}
1337
Quirky devices (such as the Broadcom Tigon 3) may need write access to their
1338
configuration space registers. Xen can be instructed to allow specified PCI
1339
devices write access to specific configuration space registers. The policy may
1342
\centerline{ \path{/etc/xen/xend-pci-quirks.sxp} }
1344
The policy file is heavily commented and is intended to provide enough
1345
documentation for developers to extend it.
1347
\subsubsection{PCI Backend Configuration - Permissive Flag}
1348
If the user-space quirks approach doesn't meet your needs you may want to enable
1349
the permissive flag for that device. To do so, first get the PCI domain, bus,
1350
slot, and function information from dom0 via \path{lspci}. Then augment the
1351
user-space policy for permissive devices. The permissive policy can be found
1354
\centerline{ \path{/etc/xen/xend-pci-permissive.sxp} }
1356
Currently, the only way to reset the permissive flag is to unbind the device
1357
from the PCI Backend driver.
1359
\subsubsection{PCI Backend - Checking Status}
1360
There two important sysfs nodes that provide a mechanism to view specifics on
1361
quirks and permissive devices:
1363
\item \path{/sys/bus/drivers/pciback/permissive} \\
1364
Use \path{cat} on this file to view a list of permissive slots.
1365
\item \path{/sys/bus/drivers/pciback/quirks} \\
1366
Use \path{cat} on this file view a hierarchical view of devices bound to the
1367
PCI backend, their PCI vendor/device ID, and any quirks that are associated with
1368
that particular slot.
1371
You may notice that every device bound to the PCI backend has 17 quirks standard
1372
"quirks" regardless of \path{xend-pci-quirks.sxp}. These default entries are
1373
necessary to support interactions between the PCI bus manager and the device bound
1374
to it. Even non-quirky devices should have these standard entries.
1376
In this case, preference was given to accuracy over aesthetics by choosing to
1377
show the standard quirks in the quirks list rather than hide them from the
1380
\subsubsection{PCI Frontend Configuration}
1381
To configure a domU to receive a PCI device:
1384
\item[Command-line:]
1385
Use the {\em pci} command-line flag. For multiple devices, use the option
1387
\centerline{ {\tt xm create netcard-dd pci=01:00.0 pci=02:03.0 }} \\
1389
\item[Flat Format configuration file:]
1390
Specify all of your PCI devices in a python list named {\em pci}. \\
1391
\centerline{ {\tt pci=['01:00.0','02:03.0'] }} \\
1393
\item[SXP Format configuration file:]
1394
Use a single PCI device section for all of your devices (specify the numbers
1395
in hexadecimal with the preceding '0x'). Note that {\em domain} here refers
1396
to the PCI domain, not a virtual machine within Xen.
1400
(dev (domain 0x0)(bus 0x3)(slot 0x1a)(func 0x1)
1401
(dev (domain 0x0)(bus 0x1)(slot 0x5)(func 0x0)
1407
%% There are two possible types of privileges: IO privileges and
1408
%% administration privileges.
1410
\section{Support for virtual Trusted Platform Module (vTPM)}
1413
Paravirtualized domains can be given access to a virtualized version
1414
of a TPM. This enables applications in these domains to use the services
1415
of the TPM device for example through a TSS stack
1416
\footnote{Trousers TSS stack: http://sourceforge.net/projects/trousers}.
1417
The Xen source repository provides the necessary software components to
1418
enable virtual TPM access. Support is provided through several
1419
different pieces. First, a TPM emulator has been modified to provide TPM's
1420
functionality for the virtual TPM subsystem. Second, a virtual TPM Manager
1421
coordinates the virtual TPMs efforts, manages their creation, and provides
1422
protected key storage using the TPM. Third, a device driver pair providing
1423
a TPM front- and backend is available for XenLinux to deliver TPM commands
1424
from the domain to the virtual TPM manager, which dispatches it to a
1425
software TPM. Since the TPM Manager relies on a HW TPM for protected key
1426
storage, therefore this subsystem requires a Linux-supported hardware TPM.
1427
For development purposes, a TPM emulator is available for use on non-TPM
1430
\subsubsection{Compile-Time Setup}
1431
To enable access to the virtual TPM, the virtual TPM backend driver must
1432
be compiled for a privileged domain (e.g. domain 0). Using the XenLinux
1433
configuration, the necessary driver can be selected in the Xen configuration
1434
section. Unless the driver has been compiled into the kernel, its module
1435
must be activated using the following command:
1441
Similarly, the TPM frontend driver must be compiled for the kernel trying
1442
to use TPM functionality. Its driver can be selected in the kernel
1443
configuration section Device Driver / Character Devices / TPM Devices.
1444
Along with that the TPM driver for the built-in TPM must be selected.
1445
If the virtual TPM driver has been compiled as module, it
1446
must be activated using the following command:
1452
Furthermore, it is necessary to build the virtual TPM manager and software
1453
TPM by making changes to entries in Xen build configuration files.
1454
The following entry in the file Config.mk in the Xen root source
1455
directory must be made:
1461
After a build of the Xen tree and a reboot of the machine, the TPM backend
1462
drive must be loaded. Once loaded, the virtual TPM manager daemon
1463
must be started before TPM-enabled guest domains may be launched.
1464
To enable being the destination of a virtual TPM Migration, the virtual TPM
1465
migration daemon must also be loaded.
1474
Once the VTPM manager is running, the VTPM can be accessed by loading the
1475
front end driver in a guest domain.
1477
\subsubsection{Development and Testing TPM Emulator}
1478
For development and testing on non-TPM enabled platforms, a TPM emulator
1479
can be used in replacement of a platform TPM. First, the entry in the file
1480
tools/vtpm/Rules.mk must look as follows:
1486
Second, the entry in the file tool/vtpm\_manager/Rules.mk must be uncommented
1490
# TCS talks to fifo's rather than /dev/tpm. TPM Emulator assumed on fifos
1491
CFLAGS += -DDUMMY_TPM
1494
Before starting the virtual TPM Manager, start the emulator by executing
1495
the following in dom0:
1501
\subsubsection{vTPM Frontend Configuration}
1502
To provide TPM functionality to a user domain, a line must be added to
1503
the virtual TPM configuration file using the following format:
1506
vtpm = ['instance=<instance number>, backend=<domain id>']
1509
The { \it instance number} reflects the preferred virtual TPM instance
1510
to associate with the domain. If the selected instance is
1511
already associated with another domain, the system will automatically
1512
select the next available instance. An instance number greater than
1513
zero must be provided. It is possible to omit the instance
1514
parameter from the configuration file.
1516
The {\it domain id} provides the ID of the domain where the
1517
virtual TPM backend driver and virtual TPM are running in. It should
1518
currently always be set to '0'.
1521
Examples for valid vtpm entries in the configuration file are
1524
vtpm = ['instance=1, backend=0']
1528
vtpm = ['backend=0'].
1531
\subsubsection{Using the virtual TPM}
1533
Access to TPM functionality is provided by the virtual TPM frontend driver.
1534
Similar to existing hardware TPM drivers, this driver provides basic TPM
1535
status information through the {\it sysfs} filesystem. In a Xen user domain
1536
the sysfs entries can be found in /sys/devices/xen/vtpm-0.
1538
Commands can be sent to the virtual TPM instance using the character
1539
device /dev/tpm0 (major 10, minor 224).
1541
% Chapter Storage and FileSytem Management
1542
\chapter{Storage and File System Management}
1544
Storage can be made available to virtual machines in a number of
1545
different ways. This chapter covers some possible configurations.
1547
The most straightforward method is to export a physical block device (a
1548
hard drive or partition) from dom0 directly to the guest domain as a
1549
virtual block device (VBD).
1551
Storage may also be exported from a filesystem image or a partitioned
1552
filesystem image as a \emph{file-backed VBD}.
1554
Finally, standard network storage protocols such as NBD, iSCSI, NFS,
1555
etc., can be used to provide storage to virtual machines.
1558
\section{Exporting Physical Devices as VBDs}
1559
\label{s:exporting-physical-devices-as-vbds}
1561
One of the simplest configurations is to directly export individual
1562
partitions from domain~0 to other domains. To achieve this use the
1563
\path{phy:} specifier in your domain configuration file. For example a
1566
\verb_disk = ['phy:hda3,sda1,w']_
1568
specifies that the partition \path{/dev/hda3} in domain~0 should be
1569
exported read-write to the new domain as \path{/dev/sda1}; one could
1570
equally well export it as \path{/dev/hda} or \path{/dev/sdb5} should
1573
In addition to local disks and partitions, it is possible to export
1574
any device that Linux considers to be ``a disk'' in the same manner.
1575
For example, if you have iSCSI disks or GNBD volumes imported into
1576
domain~0 you can export these to other domains using the \path{phy:}
1579
\verb_disk = ['phy:vg/lvm1,sda2,w']_
1583
\framebox{\bf Warning: Block device sharing}
1586
Block devices should typically only be shared between domains in a
1587
read-only fashion otherwise the Linux kernel's file systems will get
1588
very confused as the file system structure may change underneath
1589
them (having the same ext3 partition mounted \path{rw} twice is a
1590
sure fire way to cause irreparable damage)! \Xend\ will attempt to
1591
prevent you from doing this by checking that the device is not
1592
mounted read-write in domain~0, and hasn't already been exported
1593
read-write to another domain. If you want read-write sharing,
1594
export the directory to other domains via NFS from domain~0 (or use
1595
a cluster file system such as GFS or ocfs2).
1599
\section{Using File-backed VBDs}
1601
It is also possible to use a file in Domain~0 as the primary storage
1602
for a virtual machine. As well as being convenient, this also has the
1603
advantage that the virtual block device will be \emph{sparse} ---
1604
space will only really be allocated as parts of the file are used. So
1605
if a virtual machine uses only half of its disk space then the file
1606
really takes up half of the size allocated.
1608
For example, to create a 2GB sparse file-backed virtual block device
1609
(actually only consumes no disk space at all):
1611
\verb_# dd if=/dev/zero of=vm1disk bs=1k seek=2048k count=0_
1614
Make a file system in the disk file:
1616
\verb_# mkfs -t ext3 vm1disk_
1619
(when the tool asks for confirmation, answer `y')
1621
Populate the file system e.g.\ by copying from the current root:
1624
# mount -o loop vm1disk /mnt
1625
# cp -ax /{root,dev,var,etc,usr,bin,sbin,lib} /mnt
1626
# mkdir /mnt/{proc,sys,home,tmp}
1630
Tailor the file system by editing \path{/etc/fstab},
1631
\path{/etc/hostname}, etc.\ Don't forget to edit the files in the
1632
mounted file system, instead of your domain~0 filesystem, e.g.\ you
1633
would edit \path{/mnt/etc/fstab} instead of \path{/etc/fstab}. For
1634
this example put \path{/dev/sda1} to root in fstab.
1636
Now unmount (this is important!):
1638
\verb_# umount /mnt_
1641
In the configuration file set:
1643
\verb_disk = ['tap:aio:/full/path/to/vm1disk,sda1,w']_
1646
As the virtual machine writes to its `disk', the sparse file will be
1647
filled in and consume more space up to the original 2GB.
1649
{\em{Note:}} Users that have worked with file-backed VBDs on Xen in previous
1650
versions will be interested to know that this support is now provided through
1651
the blktap driver instead of the loopback driver. This change results in
1652
file-based block devices that are higher-performance, more scalable, and which
1653
provide better safety properties for VBD data. All that is required to update
1654
your existing file-backed VM configurations is to change VBD configuration
1657
\verb_disk = ['file:/full/path/to/vm1disk,sda1,w']_
1661
\verb_disk = ['tap:aio:/full/path/to/vm1disk,sda1,w']_
1665
\subsection{Loopback-mounted file-backed VBDs (deprecated)}
1667
{\em{{\bf{Note:}} Loopback mounted VBDs have now been replaced with
1668
blktap-based support for raw image files, as described above. This
1669
section remains to detail a configuration that was used by older Xen
1672
Raw image file-backed VBDs may also be attached to VMs using the
1673
Linux loopback driver. The only required change to the raw file
1674
instructions above are to specify the configuration entry as:
1676
\verb_disk = ['file:/full/path/to/vm1disk,sda1,w']_
1679
{\bf Note that loopback file-backed VBDs may not be appropriate for backing
1680
I/O-intensive domains.} This approach is known to experience
1681
substantial slowdowns under heavy I/O workloads, due to the I/O
1682
handling by the loopback block device used to support file-backed VBDs
1683
in dom0. Loopback support remains for old Xen installations, and users
1684
are strongly encouraged to use the blktap-based file support (using
1685
``{\tt{tap:aio}}'' as described above).
1687
Additionally, Linux supports a maximum of eight loopback file-backed
1688
VBDs across all domains by default. This limit can be statically
1689
increased by using the \emph{max\_loop} module parameter if
1690
CONFIG\_BLK\_DEV\_LOOP is compiled as a module in the dom0 kernel, or
1691
by using the \emph{max\_loop=n} boot option if CONFIG\_BLK\_DEV\_LOOP
1692
is compiled directly into the dom0 kernel. Again, users are encouraged
1693
to use the blktap-based file support described above which scales to much
1694
larger number of active VBDs.
1697
\section{Using LVM-backed VBDs}
1698
\label{s:using-lvm-backed-vbds}
1700
A particularly appealing solution is to use LVM volumes as backing for
1701
domain file-systems since this allows dynamic growing/shrinking of
1702
volumes as well as snapshot and other features.
1704
To initialize a partition to support LVM volumes:
1707
# pvcreate /dev/sda10
1711
Create a volume group named `vg' on the physical partition:
1714
# vgcreate vg /dev/sda10
1718
Create a logical volume of size 4GB named `myvmdisk1':
1721
# lvcreate -L4096M -n myvmdisk1 vg
1725
You should now see that you have a \path{/dev/vg/myvmdisk1} Make a
1726
filesystem, mount it and populate it, e.g.:
1729
# mkfs -t ext3 /dev/vg/myvmdisk1
1730
# mount /dev/vg/myvmdisk1 /mnt
1736
Now configure your VM with the following disk configuration:
1739
disk = [ 'phy:vg/myvmdisk1,sda1,w' ]
1743
LVM enables you to grow the size of logical volumes, but you'll need
1744
to resize the corresponding file system to make use of the new space.
1745
Some file systems (e.g.\ ext3) now support online resize. See the LVM
1746
manuals for more details.
1748
You can also use LVM for creating copy-on-write (CoW) clones of LVM
1749
volumes (known as writable persistent snapshots in LVM terminology).
1750
This facility is new in Linux 2.6.8, so isn't as stable as one might
1751
hope. In particular, using lots of CoW LVM disks consumes a lot of
1752
dom0 memory, and error conditions such as running out of disk space
1753
are not handled well. Hopefully this will improve in future.
1755
To create two copy-on-write clones of the above file system you would
1756
use the following commands:
1760
# lvcreate -s -L1024M -n myclonedisk1 /dev/vg/myvmdisk1
1761
# lvcreate -s -L1024M -n myclonedisk2 /dev/vg/myvmdisk1
1765
Each of these can grow to have 1GB of differences from the master
1766
volume. You can grow the amount of space for storing the differences
1767
using the lvextend command, e.g.:
1770
# lvextend +100M /dev/vg/myclonedisk1
1774
Don't let the `differences volume' ever fill up otherwise LVM gets
1775
rather confused. It may be possible to automate the growing process by
1776
using \path{dmsetup wait} to spot the volume getting full and then
1777
issue an \path{lvextend}.
1779
In principle, it is possible to continue writing to the volume that
1780
has been cloned (the changes will not be visible to the clones), but
1781
we wouldn't recommend this: have the cloned volume as a `pristine'
1782
file system install that isn't mounted directly by any of the virtual
1786
\section{Using NFS Root}
1788
First, populate a root filesystem in a directory on the server
1789
machine. This can be on a distinct physical machine, or simply run
1790
within a virtual machine on the same node.
1792
Now configure the NFS server to export this filesystem over the
1793
network by adding a line to \path{/etc/exports}, for instance:
1798
/export/vm1root 192.0.2.4/24 (rw,sync,no_root_squash)
1803
Finally, configure the domain to use NFS root. In addition to the
1804
normal variables, you should make sure to set the following values in
1805
the domain's configuration file:
1811
nfs_server = '2.3.4.5' # substitute IP address of server
1812
nfs_root = '/path/to/root' # path to root FS on the server
1817
The domain will need network access at boot time, so either statically
1818
configure an IP address using the config variables \path{ip},
1819
\path{netmask}, \path{gateway}, \path{hostname}; or enable DHCP
1820
(\path{dhcp='dhcp'}).
1822
Note that the Linux NFS root implementation is known to have stability
1823
problems under high load (this is not a Xen-specific problem), so this
1824
configuration may not be appropriate for critical servers.
1827
\chapter{CPU Management}
1829
%% KMS Something sage about CPU / processor management.
1831
Xen allows a domain's virtual CPU(s) to be associated with one or more
1832
host CPUs. This can be used to allocate real resources among one or
1833
more guests, or to make optimal use of processor resources when
1834
utilizing dual-core, hyperthreading, or other advanced CPU technologies.
1836
Xen enumerates physical CPUs in a `depth first' fashion. For a system
1837
with both hyperthreading and multiple cores, this would be all the
1838
hyperthreads on a given core, then all the cores on a given socket,
1839
and then all sockets. I.e. if you had a two socket, dual core,
1840
hyperthreaded Xeon the CPU order would be:
1844
\begin{tabular}{l|l|l|l|l|l|l|r}
1845
\multicolumn{4}{c|}{socket0} & \multicolumn{4}{c}{socket1} \\ \hline
1846
\multicolumn{2}{c|}{core0} & \multicolumn{2}{c|}{core1} &
1847
\multicolumn{2}{c|}{core0} & \multicolumn{2}{c}{core1} \\ \hline
1848
ht0 & ht1 & ht0 & ht1 & ht0 & ht1 & ht0 & ht1 \\
1849
\#0 & \#1 & \#2 & \#3 & \#4 & \#5 & \#6 & \#7 \\
1854
Having multiple vcpus belonging to the same domain mapped to the same
1855
physical CPU is very likely to lead to poor performance. It's better to
1856
use `vcpus-set' to hot-unplug one of the vcpus and ensure the others are
1857
pinned on different CPUs.
1859
If you are running IO intensive tasks, its typically better to dedicate
1860
either a hyperthread or whole core to running domain 0, and hence pin
1861
other domains so that they can't use CPU 0. If your workload is mostly
1862
compute intensive, you may want to pin vcpus such that all physical CPU
1863
threads are available for guest domains.
1865
\chapter{Migrating Domains}
1867
\section{Domain Save and Restore}
1869
The administrator of a Xen system may suspend a virtual machine's
1870
current state into a disk file in domain~0, allowing it to be resumed at
1873
For example you can suspend a domain called ``VM1'' to disk using the
1876
# xm save VM1 VM1.chk
1879
This will stop the domain named ``VM1'' and save its current state
1880
into a file called \path{VM1.chk}.
1882
To resume execution of this domain, use the \path{xm restore} command:
1884
# xm restore VM1.chk
1887
This will restore the state of the domain and resume its execution.
1888
The domain will carry on as before and the console may be reconnected
1889
using the \path{xm console} command, as described earlier.
1891
\section{Migration and Live Migration}
1893
Migration is used to transfer a domain between physical hosts. There
1894
are two varieties: regular and live migration. The former moves a
1895
virtual machine from one host to another by pausing it, copying its
1896
memory contents, and then resuming it on the destination. The latter
1897
performs the same logical functionality but without needing to pause
1898
the domain for the duration. In general when performing live migration
1899
the domain continues its usual activities and---from the user's
1900
perspective---the migration should be imperceptible.
1902
To perform a live migration, both hosts must be running Xen / \xend\ and
1903
the destination host must have sufficient resources (e.g.\ memory
1904
capacity) to accommodate the domain after the move. Furthermore we
1905
currently require both source and destination machines to be on the same
1908
Currently, there is no support for providing automatic remote access
1909
to filesystems stored on local disk when a domain is migrated.
1910
Administrators should choose an appropriate storage solution (i.e.\
1911
SAN, NAS, etc.) to ensure that domain filesystems are also available
1912
on their destination node. GNBD is a good method for exporting a
1913
volume from one machine to another. iSCSI can do a similar job, but is
1914
more complex to set up.
1916
When a domain migrates, it's MAC and IP address move with it, thus it is
1917
only possible to migrate VMs within the same layer-2 network and IP
1918
subnet. If the destination node is on a different subnet, the
1919
administrator would need to manually configure a suitable etherip or IP
1920
tunnel in the domain~0 of the remote node.
1922
A domain may be migrated using the \path{xm migrate} command. To live
1923
migrate a domain to another machine, we would use the command:
1926
# xm migrate --live mydomain destination.ournetwork.com
1929
Without the \path{--live} flag, \xend\ simply stops the domain and
1930
copies the memory image over to the new node and restarts it. Since
1931
domains can have large allocations this can be quite time consuming,
1932
even on a Gigabit network. With the \path{--live} flag \xend\ attempts
1933
to keep the domain running while the migration is in progress, resulting
1934
in typical down times of just 60--300ms.
1936
For now it will be necessary to reconnect to the domain's console on the
1937
new machine using the \path{xm console} command. If a migrated domain
1938
has any open network connections then they will be preserved, so SSH
1939
connections do not have this limitation.
1942
%% Chapter Securing Xen
1943
\chapter{Securing Xen}
1945
This chapter describes how to secure a Xen system. It describes a number
1946
of scenarios and provides a corresponding set of best practices. It
1947
begins with a section devoted to understanding the security implications
1951
\section{Xen Security Considerations}
1953
When deploying a Xen system, one must be sure to secure the management
1954
domain (Domain-0) as much as possible. If the management domain is
1955
compromised, all other domains are also vulnerable. The following are a
1956
set of best practices for Domain-0:
1959
\item \textbf{Run the smallest number of necessary services.} The less
1960
things that are present in a management partition, the better.
1961
Remember, a service running as root in the management domain has full
1962
access to all other domains on the system.
1963
\item \textbf{Use a firewall to restrict the traffic to the management
1964
domain.} A firewall with default-reject rules will help prevent
1965
attacks on the management domain.
1966
\item \textbf{Do not allow users to access Domain-0.} The Linux kernel
1967
has been known to have local-user root exploits. If you allow normal
1968
users to access Domain-0 (even as unprivileged users) you run the risk
1969
of a kernel exploit making all of your domains vulnerable.
1972
\section{Driver Domain Security Considerations}
1973
\label{s:ddsecurity}
1975
Driver domains address a range of security problems that exist regarding
1976
the use of device drivers and hardware. On many operating systems in common
1977
use today, device drivers run within the kernel with the same privileges as
1978
the kernel. Few or no mechanisms exist to protect the integrity of the kernel
1979
from a misbehaving (read "buggy") or malicious device driver. Driver
1980
domains exist to aid in isolating a device driver within its own virtual
1981
machine where it cannot affect the stability and integrity of other
1982
domains. If a driver crashes, the driver domain can be restarted rather than
1983
have the entire machine crash (and restart) with it. Drivers written by
1984
unknown or untrusted third-parties can be confined to an isolated space.
1985
Driver domains thus address a number of security and stability issues with
1988
However, due to limitations in current hardware, a number of security
1989
concerns remain that need to be considered when setting up driver domains (it
1990
should be noted that the following list is not intended to be exhaustive).
1993
\item \textbf{Without an IOMMU, a hardware device can DMA to memory regions
1994
outside of its controlling domain.} Architectures which do not have an
1995
IOMMU (e.g. most x86-based platforms) to restrict DMA usage by hardware
1996
are vulnerable. A hardware device which can perform arbitrary memory reads
1997
and writes can read/write outside of the memory of its controlling domain.
1998
A malicious or misbehaving domain could use a hardware device it controls
1999
to send data overwriting memory in another domain or to read arbitrary
2000
regions of memory in another domain.
2001
\item \textbf{Shared buses are vulnerable to sniffing.} Devices that share
2002
a data bus can sniff (and possible spoof) each others' data. Device A that
2003
is assigned to Domain A could eavesdrop on data being transmitted by
2004
Domain B to Device B and then relay that data back to Domain A.
2005
\item \textbf{Devices which share interrupt lines can either prevent the
2006
reception of that interrupt by the driver domain or can trigger the
2007
interrupt service routine of that guest needlessly.} A devices which shares
2008
a level-triggered interrupt (e.g. PCI devices) with another device can
2009
raise an interrupt and never clear it. This effectively blocks other devices
2010
which share that interrupt line from notifying their controlling driver
2011
domains that they need to be serviced. A device which shares an
2012
any type of interrupt line can trigger its interrupt continually which
2013
forces execution time to be spent (in multiple guests) in the interrupt
2014
service routine (potentially denying time to other processes within that
2015
guest). System architectures which allow each device to have its own
2016
interrupt line (e.g. PCI's Message Signaled Interrupts) are less
2017
vulnerable to this denial-of-service problem.
2018
\item \textbf{Devices may share the use of I/O memory address space.} Xen can
2019
only restrict access to a device's physical I/O resources at a certain
2020
granularity. For interrupt lines and I/O port address space, that
2021
granularity is very fine (per interrupt line and per I/O port). However,
2022
Xen can only restrict access to I/O memory address space on a page size
2023
basis. If more than one device shares use of a page in I/O memory address
2024
space, the domains to which those devices are assigned will be able to
2025
access the I/O memory address space of each other's devices.
2029
\section{Security Scenarios}
2032
\subsection{The Isolated Management Network}
2034
In this scenario, each node has two network cards in the cluster. One
2035
network card is connected to the outside world and one network card is a
2036
physically isolated management network specifically for Xen instances to
2039
As long as all of the management partitions are trusted equally, this is
2040
the most secure scenario. No additional configuration is needed other
2041
than forcing Xend to bind to the management interface for relocation.
2044
\subsection{A Subnet Behind a Firewall}
2046
In this scenario, each node has only one network card but the entire
2047
cluster sits behind a firewall. This firewall should do at least the
2051
\item Prevent IP spoofing from outside of the subnet.
2052
\item Prevent access to the relocation port of any of the nodes in the
2053
cluster except from within the cluster.
2056
The following iptables rules can be used on each node to prevent
2057
migrations to that node from outside the subnet assuming the main
2058
firewall does not do this for you:
2061
# this command disables all access to the Xen relocation
2063
iptables -A INPUT -p tcp --destination-port 8002 -j REJECT
2065
# this command enables Xen relocations only from the specific
2067
iptables -I INPUT -p tcp -{}-source 192.0.2.0/24 \
2068
--destination-port 8002 -j ACCEPT
2071
\subsection{Nodes on an Untrusted Subnet}
2073
Migration on an untrusted subnet is not safe in current versions of Xen.
2074
It may be possible to perform migrations through a secure tunnel via an
2075
VPN or SSH. The only safe option in the absence of a secure tunnel is to
2076
disable migration completely. The easiest way to do this is with
2080
# this command disables all access to the Xen relocation port
2081
iptables -A INPUT -p tcp -{}-destination-port 8002 -j REJECT
2084
%% Chapter Xen Mandatory Access Control Framework
2085
\chapter{sHype/Xen Access Control}
2086
The Xen mandatory access control framework is an implementation of the
2087
sHype Hypervisor Security Architecture
2088
(www.research.ibm.com/ssd\_shype). It permits or denies communication
2089
and resource access of domains based on a security policy. The
2090
mandatory access controls are enforced in addition to the Xen core
2091
controls, such as memory protection. They are designed to remain
2092
transparent during normal operation of domains (policy-conform
2093
behavior) but to intervene when domains move outside their intended
2094
sharing behavior. This chapter will describe how the sHype access
2095
controls in Xen can be configured to prevent viruses from spilling
2096
over from one into another workload type and secrets from leaking from
2097
one workload type to another. sHype/Xen depends on the correct
2098
behavior of Domain-0 (cf previous chapter).
2100
Benefits of configuring sHype/ACM in Xen include:
2102
\item robust workload and resource protection effective against rogue
2104
\item simple, platform- and operating system-independent security
2105
policies (ideal for heterogeneous distributed environments)
2106
\item safety net with minimal performance overhead in case operating
2107
system security is missing, does not scale, or fails
2110
These benefits are very valuable because today's operating systems
2111
become increasingly complex and often have no or insufficient
2112
mandatory access controls. (Discretionary access controls, supported
2113
by most operating systems, are not effective against viruses or
2114
misbehaving programs.) Where mandatory access control exists (e.g.,
2115
SELinux), they usually deploy platform-specific, complex, and difficult
2116
to understand security policies. Multi-tier applications in business
2117
environments typically require different operating systems
2118
(e.g., AIX, Windows, Linux) in different tiers. Related distributed
2119
transactions and workloads cannot be easily protected on the OS level.
2120
The Xen access control framework steps in to offer a coarse-grained
2121
but very robust and consistent security layer and safety net across
2122
different platforms and operating systems.
2124
To control sharing between domains, Xen mediates all inter-domain
2125
communication (shared memory, events) as well as the access of domains
2126
to resources such as storage disks. Thus, Xen can confine distributed
2127
workloads (domain payloads) by permitting sharing among domains
2128
running the same type of workload and denying sharing between pairs of
2129
domains that run different workload types. We assume that--from a Xen
2130
perspective--only one workload type is running per user domain. To
2131
enable Xen to associate domains and resources with workload types,
2132
security labels including the workload types are attached to domains
2133
and resources. These labels and the hypervisor sHype controls cannot
2134
be manipulated or bypassed by user domains and are effective even
2135
against compromised or rogue domains.
2138
This section gives an overview of how workloads can be protected using
2139
the sHype mandatory access control framework in Xen.
2140
Figure~\ref{fig:acmoverview} shows the necessary steps in activating
2141
the Xen workload protection. These steps are described in detail in
2142
Section~\ref{section:acmexample}.
2146
\includegraphics[width=13cm]{figs/acm_overview.eps}
2147
\caption{Overview of activating sHype workload protection in Xen.
2148
Section numbers point to representative examples.}
2149
\label{fig:acmoverview}
2152
First, the sHype/ACM access control must be enabled in the Xen
2153
distribution and the distribution must be built and installed (cf
2154
Subsection~\ref{subsection:acmexampleconfigure}). Before we can
2155
enforce security, a Xen security policy must be created (cf
2156
Subsection~\ref{subsection:acmexamplecreate}) and deployed (cf
2157
Subsection~\ref{subsection:acmexampleinstall}). This policy defines
2158
the workload types differentiated during access control. It also
2159
defines the rules that compare workload types of domains and resources
2160
to decide about access requests. Workload types are represented by
2161
security labels that can be securely associated to domains and resources (cf
2162
Subsections~\ref{subsection:acmexamplelabeldomains}
2163
and~\ref{subsection:acmexamplelabelresources}). The functioning of
2164
the active sHype/Xen workload protection is demonstrated using simple
2165
resource assignment, and domain creation tests in
2166
Subsection~\ref{subsection:acmexampletest}.
2167
Section~\ref{section:acmpolicy} describes the syntax and semantics of
2168
the sHype/Xen security policy in detail and introduces briefly the
2169
tools that are available to help you create your own sHype security policies.
2171
The next section describes all the necessary steps to create, deploy,
2172
and test a simple workload protection policy. It is meant to enable
2173
Xen users and developers to quickly try out the sHype/Xen workload
2174
protection. Those readers who are interested in learning more about
2175
how the sHype access control in Xen works and how it is configured
2176
using the XML security policy should read Section~\ref{section:acmpolicy}
2177
as well. Section~\ref{section:acmlimitations} concludes this chapter with
2178
current limitations of the sHype implementation for Xen.
2180
\section{Xen Workload Protection Step-by-Step}
2181
\label{section:acmexample}
2183
You are about to configure and deploy the Xen sHype workload protection
2184
by following 5 simple steps:
2186
\item configure and install sHype/Xen
2187
\item create a simple workload protection security policy
2188
\item deploy the sHype/Xen security policy
2189
\item associate domains and resources with workload labels,
2190
\item test the workload protection
2192
The essential commands to create and deploy an sHype/Xen security
2193
policy are numbered throughout the following sections. If you want a
2194
quick-guide or return at a later time to go quickly through this
2195
demonstration, simply look for the numbered commands and apply them in
2198
\subsection{Configuring/Building sHype Support into Xen}
2199
\label{subsection:acmexampleconfigure}
2200
First, we need to configure the access control module in Xen and
2201
install the ACM-enabled Xen hypervisor. This step installs security
2202
tools and compiles sHype/ACM controls into the Xen hypervisor.
2204
To enable sHype/ACM in Xen, please edit the Config.mk file in the top
2209
Change: XSM_ENABLE ?= n
2212
Change: ACM_SECURITY ?= n
2213
To: ACM_SECURITY ?= y
2216
Then install the security-enabled Xen environment as follows:
2223
Reboot into the security-enabled Xen hypervisor.
2229
Xen will boot into the default security policy. After reboot,
2230
you can explore the simple DEFAULT policy.
2234
Supported security subsystems : ACM
2235
Policy name : DEFAULT
2237
Version of XML policy : 1.0
2238
Policy configuration : loaded
2244
Name ID Mem VCPUs State Time(s) Label
2245
Domain-0 0 941 1 r----- 38.1 ACM:DEFAULT:SystemManagement
2249
In this state, no domains can be started.
2250
Now, a policy can be created and loaded into the hypervisor.
2252
\subsection{Creating A WLP Policy in 3 Simple Steps with ezPolicy}
2253
\label{subsection:acmexamplecreate}
2255
We will use the ezPolicy tool to quickly create a policy that protects
2256
workloads. You will need both the Python and wxPython packages to run
2257
this tool. To run the tool in Domain-0, you can download the wxPython
2258
package from www.wxpython.org or use the command \verb|yum install wxPython|
2259
in Redhat/Fedora. To run the tool on MS Windows, you also need to download
2260
the Python package from www.python.org. After these packages are installed,
2261
start the ezPolicy tool with the following command:
2264
(4) # xensec_ezpolicy
2267
Figure~\ref{fig:acmezpolicy} shows a screen-shot of the tool. The
2268
following steps illustrate how you can create the workload definition
2269
shown in Figure~\ref{fig:acmezpolicy}. You can use \verb|<CTRL>-h| to
2270
pop up a help window at any time. The indicators (a), (b), and (c) in
2271
Figure~\ref{fig:acmezpolicy} show the buttons that are used during the
2272
3 steps of creating a policy:
2274
\item defining workloads
2275
\item defining run-time conflicts
2276
\item translating the workload definition into an sHype/Xen access
2280
\paragraph{Defining workloads.} Workloads are defined for each
2281
organization and department that you enter in the left panel.
2283
To ease the transition from an unlabeled to a fully labeled workload-protection
2284
environment, we have added support to sHype/Xen to run unlabeled domains accessing
2285
unlabeled resources in addition to labeled domains accessing labeled resources.
2287
Support for running unlabeled domains on sHype/Xen is enabled by adding the
2288
predefined workload type and label \verb|__UNLABELED__| to the security
2289
policy. (This is a double underscore
2290
followed by the string ''\verb|UNLABELED|'' followed by a double underscore.)
2291
The ezPolicy tool automatically adds this organization-level workload type
2292
to a new workload definition (cf Figure~\ref{fig:acmezpolicy}). It can simply be
2293
deleted from the workload definition if no such support is desired. If unlabeled domains
2294
are supported in the policy, then any domain or resource that has no label will implicitly
2295
inherit this label when access control decisions are made. In effect, unlabeled
2296
domains and resources define a new workload type \verb|__UNLABELED__|, which is
2297
confined from any other labeled workload.
2299
Please use now the ``New Org'' button to add the organization workload types
2300
``A-Bank'', ``B-Bank'', and ``AutoCorp''.
2302
You can refine an organization to differentiate between multiple
2303
department workloads by right-clicking the organization and selecting
2304
\verb|Add Department| (or selecting an organization and pressing
2305
\verb|<CRTL>-a|). Create department workloads ``SecurityUnderwriting'',
2306
and ``MarketAnalysis'' for the ``A-Bank''. The resulting layout of the
2307
tool should be similar to the left panel shown in
2308
Figure~\ref{fig:acmezpolicy}.
2312
\includegraphics[width=13cm]{figs/acm_ezpolicy_gui.eps}
2313
\caption{Final layout including workload definition and Run-time Exclusion rules.}
2314
\label{fig:acmezpolicy}
2317
\paragraph{Defining run-time conflicts.} Workloads that shall be
2318
prohibited from running concurrently on the same hypervisor platform
2319
are grouped into ``Run-time Exclusion rules'' on the right panel of
2320
the window. Cautious users should include the \verb|__UNLABELED__|
2321
workload type in all run-time exclusion rules because any workload
2322
could run inside unlabeled domains.
2324
To prevent A-Bank and B-Bank workloads (including their
2325
departmental workloads) from running simultaneously on the same
2326
hypervisor system, select the organization ``A-Bank'' and, while
2327
pressing the \verb|<CTRL>|-key, select the organization ``B-Bank''.
2328
Being cautious, we also prevent unlabeled workloads from running with
2329
any of those workloads by pressing the \verb|<CTRL>|-key and selecting
2330
``\_\_UNLABELED\_\_''. Now press the button named ``Create run-time exclusion
2331
rule from selection''. A popup window will ask for the name for this run-time
2332
exclusion rule (enter a name or just hit \verb|<ENTER>|). A rule will
2333
appear on the right panel. The name is used as reference only and does
2334
not affect access control decisions.
2336
Please repeat this process to create another run-time exclusion rule
2337
for the department workloads ``A-Bank.SecurityUnderwriting'',
2338
``A-Bank.MarketAnalysis''. Also add the ``\_\_UNLABELED\_\_''
2339
workload type to this conflict set.
2341
The resulting layout of your window should be similar to
2342
Figure~\ref{fig:acmezpolicy}. Save this workload definition by
2343
selecting ``Save Workload Definition as ...'' in the ``File'' menu.
2344
This workload definition can be later refined if required.
2346
\paragraph{Translating the workload definition into an sHype/Xen access
2347
control policy.} To translate the workload definition into a access
2348
control policy understood by Xen, please select the ``Save as Xen ACM
2349
Security Policy'' in the ``File'' menu. Enter the following policy
2350
name in the popup window: \verb|mytest|. If you are running ezPolicy in
2351
Domain-0, the resulting policy file mytest\_security-policy.xml will
2352
automatically be placed into the right directory (/etc/xen/acm-security/policies/).
2353
If you run the tool on another system, then you need to copy the
2354
resulting policy file into Domain-0 before continuing. See
2355
Section~\ref{subsection:acmnaming} for naming conventions of security
2359
\textbf{Note:} The support for \verb|__UNLABELED__| domains and
2360
resources is meant to help transitioning from an uncontrolled
2361
environment to a workload-protected environment by starting with
2362
unlabeled domains and resources and then step-by-step labeling domains
2363
and resources. Once all workloads are labeled, the \verb|__UNLABELED__|
2364
type can simply be removed from the Domain-0 label or from the policy
2365
through a policy update. Section~\ref{subsection:acmpolicymanagement} will
2366
show how unlabeled domains can be disabled by updating the
2367
\verb|mytest| policy at run-time.
2370
\subsection{Deploying a WLP Policy}
2371
\label{subsection:acmexampleinstall}
2372
To deploy the workload protection policy we created in
2373
Section~\ref{subsection:acmexamplecreate}, we create a policy
2374
representation (mytest.bin), load it into the Xen
2375
hypervisor, and configure Xen to also load this policy during
2378
The following command translates the source policy representation
2379
into a format that can be loaded into Xen with sHype/ACM support,
2380
activates the policy, and configures this policy for future boot
2381
cycles into the boot sequence. Please refer to the \verb|xm|
2382
man page for further details:
2385
(5) # xm setpolicy ACM mytest
2386
Successfully set the new policy.
2387
Supported security subsystems : ACM
2388
Policy name : mytest
2390
Version of XML policy : 1.0
2391
Policy configuration : loaded, activated for boot
2394
Alternatively, if installing the policy fails (e.g., because it cannot
2395
identify the Xen boot entry), you can manually install the policy in 3
2398
(\textit{Alternatively to 5 - step a}) Manually copy the policy binary
2399
file into the boot directory:
2403
# cp /etc/xen/acm-security/policies/mytest.bin /boot/mytest.bin
2407
(\textit{Alternatively to 5 - step b}) Manually add a module line to your
2408
Xen boot entry so that grub loads this policy file during startup:
2412
title XEN Devel with 2.6.18.8
2414
module /vmlinuz-2.6.18.8-xen root=/dev/sda3 ro console=tty0
2415
module /initrd-2.6.18.8-xen.img
2420
(\textit{Alternatively to 5 - step c}) Reboot. Xen will choose the
2421
bootstrap label defined in the policy as Domain-0 label during reboot.
2422
After reboot, you can re-label Domain-0 at run-time,
2423
cf Section~\ref{subsection:acmlabeldom0}.
2425
Assuming that command (5) succeeded or you followed the alternative
2426
instructions above, you should see the new policy and label appear
2427
when listing domains:
2432
Name ID Mem VCPUs State Time(s) Label
2433
Domain-0 0 941 1 r----- 81.5 ACM:mytest:SystemManagement
2437
If the security label at the end of the line says ``INACTIVE'' then the
2438
security is not enabled. Verify the previous steps. Note: Domain-0 is
2439
assigned a default label (see \verb|bootstrap| policy attribute
2440
explained in Section~\ref{section:acmpolicy}). All other domains must
2441
be explicitly labeled, which we describe in detail below.
2443
\subsection{Labeling Unmanaged User Domains}
2444
\label{subsection:acmexamplelabeldomains}
2446
Unmanaged domains are started in Xen by using a configuration
2447
file. Please refer to Section~\ref{subsection:acmlabelmanageddomains}
2448
if you are using managed domains.
2450
The following configuration file defines \verb|domain1|:
2455
kernel= "/boot/vmlinuz-2.6.18.8-xen"
2460
disk = ['file:/home/xen/dom_fc5/fedora.fc5.img,sda1,w', \
2461
'file:/home/xen/dom_fc5/fedora.fc5.swap,sda2,w']
2462
root = "/dev/sda1 ro xencons=tty"
2466
Every domain must be associated with a security label before it can start
2467
on sHype/Xen. Otherwise, sHype/Xen would not be able to enforce the policy
2468
consistently. Our \verb|mytest| policy is configured so that Xen
2469
assigns a default label \verb|__UNLABELED__| to domains and resources that
2470
have no label and supports them in a controlled manner. Since neither the domain,
2471
nor the resources are (yet) labeled, this domain can start under the \verb|mytest|
2476
# xm create domain1.xm
2477
Using config file "./domain1.xm".
2478
Started domain domain1
2481
Name ID Mem VCPUs State Time(s) Label
2482
domain1 1 128 1 -b---- 0.7 ACM:mytest:__UNLABELED__
2483
Domain-0 0 875 1 r----- 84.6 ACM:mytest:SystemManagement
2487
Please shutdown domain1 so that we can move it into the protection
2488
domain of workload \verb|A-Bank|.
2492
# xm shutdown domain1
2493
(wait some seconds until the domain has shut down)
2496
Name ID Mem VCPUs State Time(s) Label
2497
Domain-0 0 875 1 r----- 86.4 ACM:mytest:SystemManagement
2501
We assume that the processing in domain1 contributes to the \verb|A-Bank| workload.
2502
We explore now how to transition this domain into the ``A-Bank'' workload-protection.
2503
The following command prints all domain labels available in the active policy:
2509
A-Bank.MarketAnalysis
2510
A-Bank.SecurityUnderwriting
2518
Now label \verb|domain1| with the A-Bank label and another \verb|domain2|
2519
with the B-Bank label. Please refer to the xm man page for
2520
further information.
2523
(6) # xm addlabel A-Bank dom domain1.xm
2524
# xm addlabel B-Bank dom domain2.xm
2527
Let us try to start the domain again:
2531
# xm create domain1.xm
2532
Using config file "./domain1.xm".
2533
Error: VM's access to block device 'file:/home/xen/dom_fc5/fedora.fc5.img' denied
2537
This error indicates that \verb|domain1|, if started, would not be able to
2538
access its image and swap files because they are not labeled. This
2539
makes sense because to confine workloads, access of domains to
2540
resources must be controlled. Otherwise, domains that are not allowed
2541
to communicate or run simultaneously could share data through storage
2544
\subsection{Labeling Resources}
2545
\label{subsection:acmexamplelabelresources}
2546
You can use the \verb|xm labels type=res| command to list available
2547
resource labels. Let us assign the A-Bank resource label to the
2548
\verb|domain1| image file representing \verb|/dev/sda1| and to its swap file:
2551
(7) # xm addlabel A-Bank res \
2552
file:/home/xen/dom_fc5/fedora.fc5.img
2554
# xm addlabel A-Bank res \
2555
file:/home/xen/dom_fc5/fedora.fc5.swap
2558
The following command lists all labeled resources on the system, e.g.,
2559
to lookup or verify the labeling:
2564
file:/home/xen/dom_fc5/fedora.fc5.swap
2568
file:/home/xen/dom_fc5/fedora.fc5.img
2575
Starting \verb|domain1| will now succeed:
2579
# xm create domain1.xm
2580
Using config file "./domain1.xm".
2581
Started domain domain1
2584
Name ID Mem VCPUs State Time(s) Label
2585
domain1 3 128 1 -b---- 0.8 ACM:mytest:A-Bank
2586
Domain-0 0 875 1 r----- 90.9 ACM:mytest:SystemManagement
2590
Currently, if a labeled resource is moved to another location, the
2591
label must first be manually removed, and after the move re-attached
2592
using the xm commands \verb|rmlabel| and \verb|addlabel|
2593
respectively. Please see Section~\ref{section:acmlimitations} for
2597
(8) Label the resources of domain2 as B-Bank
2598
but please do not start this domain yet.
2601
\subsection{Testing The Xen Workload Protection}
2602
\label{subsection:acmexampletest}
2604
We are about to demonstrate the sHype/Xen workload protection by verifying
2606
\item that user domains with conflicting workloads cannot run
2608
\item that user domains cannot access resources of workloads other than the
2609
one they are associated with
2610
\item that user domains cannot exchange network packets if they are not
2611
associated with the same workload type (not yet supported in Xen)
2614
\paragraph{Test 1: Run-time exclusion rules.} We assume that \verb|domain1|
2615
with the A-Bank label is still running. While \verb|domain1| is running,
2616
the run-time exclusion set of our policy implies that \verb|domain2| cannot
2617
start because the label of \verb|domain1| includes the CHWALL type A-Bank
2618
and the label of \verb|domain2| includes the CHWALL type B-Bank. The
2619
run-time exclusion rule of our policy enforces that A-Bank and
2620
B-Bank cannot run at the same time on the same hypervisor platform.
2621
Once domain1 is stopped, saved, or migrated to another platform,
2622
\verb|domain2| can start. Once \verb|domain2| is started, however,
2623
\verb|domain1| can no longer start or resume on this system. When creating the
2624
Chinese Wall types for the workload labels, the ezPolicy tool policy
2625
translation component ensures that department workloads inherit all the
2626
organization types (and with it any organization exclusions).
2631
Name ID Mem VCPUs State Time(s) Label
2632
domain1 3 128 1 -b---- 0.8 ACM:mytest:A-Bank
2633
Domain-0 0 875 1 r----- 90.9 ACM:mytest:SystemManagement
2635
# xm create domain2.xm
2636
Using config file "./domain2.xm".
2637
Error: 'Domain in conflict set with running domains'
2639
# xm shutdown domain1
2640
(wait some seconds until domain 1 is shut down)
2643
Name ID Mem VCPUs State Time(s) Label
2644
Domain-0 0 873 1 r----- 95.3 ACM:mytest:SystemManagement
2646
# xm create domain2.xm
2647
Using config file "./domain2.xm".
2648
Started domain domain2
2651
Name ID Mem VCPUs State Time(s) Label
2652
domain2 5 164 1 -b---- 0.3 ACM:mytest:B-Bank
2653
Domain-0 0 839 1 r----- 96.4 ACM:mytest:SystemManagement
2655
# xm create domain1.xm
2656
Using config file "domain1.xm".
2657
Error: 'Domain in conflict with running domains'
2659
# xm shutdown domain2
2661
Name ID Mem VCPUs State Time(s) Label
2662
Domain-0 0 839 1 r----- 97.8 ACM:mytest:SystemManagement
2666
You can verify that domains with AutoCorp label can run together with
2667
domains labeled A-Bank or B-Bank.
2669
\paragraph{Test2: Resource access.} In this test, we will re-label the
2670
swap file for \verb|domain1| with the \verb|B-Bank| resource label. In a
2671
real environment, the swap file must be sanitized (scrubbed/zeroed) before
2672
it is reassigned to prevent data leaks from the A-Bank to the B-Bank workload
2673
through the swap file.
2675
We expect that \verb|domain1| will no longer start because it cannot access
2676
this resource. This test checks the sharing abilities of domains, which are
2677
defined by the Simple Type Enforcement Policy component.
2681
# xm rmlabel res file:/home/xen/dom_fc5/fedora.fc5.swap
2683
# xm addlabel B-Bank res file:/home/xen/dom_fc5/fedora.fc5.swap
2686
file:/home/xen/dom_fc5/fedora.fc5.swap
2690
file:/home/xen/dom_fc5/fedora.fc5.img
2695
# xm create domain1.xm
2696
Using config file "./domain1.xm".
2698
VM's access to block device 'file:/home/xen/dom_fc5/fedora.fc5.swap' denied
2702
The resource authorization checks are performed before the domain is actually started
2703
so that failures during the startup are prevented. A domain is only started if all
2704
the resources specified in its configuration are accessible.
2706
\paragraph{Test 3: Communication.} In this test we would verify that
2707
two domains with labels A-Bank and B-Bank cannot exchange network packets
2708
by using the 'ping' connectivity test. It is also related to the STE
2709
policy. {\bf Note:} sHype/Xen does control direct communication between
2710
domains. However, domains associated with different workloads can
2711
currently still communicate through the Domain-0 virtual network. We
2712
are working on the sHype/ACM controls for local and remote network
2713
traffic through Domain-0. Please monitor the xen-devel mailing list
2714
for updated information.
2717
\subsection{Labeling Domain-0 --or-- Restricting System Authorization}
2718
\label{subsection:acmlabeldom0}
2719
The major use case for explicitly labeling or relabeling Domain-0 is to restrict
2720
or extend which workload types can run on a virtualized Xen system. This enables
2721
flexible partitioning of the physical infrastructure as well as the workloads
2722
running on it in a multi-platform environment.
2724
In case no Domain-0 label is explicitly stated, we automatically assigned Domain-0
2725
the \verb|SystemManagement| label, which includes all STE (workload) types that
2726
are known to the policy. In effect, the Domain-0 label authorizes the Xen system
2727
to run only those workload types, whose STE types are included in the Domain-0
2728
label. Hence, choosing the \verb|SystemManagement| label for Domain-0 permits any
2729
labeled domain to run. Resetting the label for Domain-0 at boot or run-time to
2730
a label with a subset of the known STE workload types restricts which user domains
2731
can run on this system. If Domain-0 is relabeled at run-time, then the new label
2732
must at least include all STE types of those domains that are currently running.
2733
The operation fails otherwise. This requirement ensures that the system remains
2734
in a valid security configuration after re-labelling.
2736
Restricting the Domain-0 authorization through the label creates a flexible
2737
policy-driven way to strongly partition the physical infrastructure and the
2738
workloads running on it. This partitioning will be automatically enforced during
2739
migration, start, or resume of domains and simplifies the security management
2740
considerably. Strongly competing workloads can be forced to run on separate physical
2741
infrastructure and become less depend on the domain isolation capabilities
2744
First, we relabel the swap image back to A-Bank and then start up domain1:
2747
# xm rmlabel res file:/home/xen/dom_fc5/fedora.fc5.swap
2749
# xm addlabel A-Bank res file:/home/xen/dom_fc5/fedora.fc5.swap
2751
# xm create domain1.xm
2752
Using config file "./domain1.xm".
2753
Started domain domain1
2756
Name ID Mem VCPUs State Time(s) Label
2757
domain1 7 128 1 -b---- 0.7 ACM:mytest:A-Bank
2758
Domain-0 0 839 1 r----- 103.1 ACM:mytest:SystemManagement
2762
The following command will restrict the Xen system to only run STE types
2763
included in the A-Bank label.
2767
# xm addlabel A-Bank mgt Domain-0
2768
Successfully set the label of domain 'Domain-0' to 'A-Bank'.
2771
Name ID Mem VCPUs State Time(s) Label
2772
Domain-0 0 839 1 r----- 103.7 ACM:mytest:A-Bank
2773
domain1 7 128 1 -b---- 0.7 ACM:mytest:A-Bank
2778
In our example policy in Figure~\ref{fig:acmxmlfileb}, this means that
2779
only \verb|A-Bank| domains and workloads (types) can run after the
2780
successful completion of this command because the \verb|A-Bank| label
2781
includes only a single STE type, namely \verb|A-Bank|. This command
2782
fails if any running domain has an STE type in its label that is not
2783
included in the A-Bank label.
2785
If we now label a domain3 with AutoCorp, it cannot start because Domain-0 is
2786
no longer authorized to run the workload type \verb|AutoCorp|.
2789
# xm addlabel AutoCorp dom domain3.xm
2790
(remember to label its resources, too)
2792
# xm create domain3.xm
2793
Using config file "./domain3.xm".
2794
Error: VM is not authorized to run.
2797
Name ID Mem VCPUs State Time(s) Label
2798
Domain-0 0 839 1 r----- 104.7 ACM:mytest:A-Bank
2799
domain1 7 128 1 -b---- 0.7 ACM:mytest:A-Bank
2803
At this point, unlabeled domains cannot start either. Let domain4.xm
2804
describe an unlabeled domain, then trying to start domain4
2808
# xm getlabel dom domain4.xm
2809
Error: 'Domain not labeled'
2811
# xm create domain4.xm
2812
Using config file "./domain4.xm".
2813
Error: VM is not authorized to run.
2817
Relabeling Domain-0 with the SystemManagement label will enable domain3 to start.
2820
# xm addlabel SystemManagement mgt Domain-0
2821
Successfully set the label of domain 'Domain-0' to 'SystemManagement'.
2824
Name ID Mem VCPUs State Time(s) Label
2825
domain1 7 128 1 -b---- 0.8 ACM:mytest:A-Bank
2826
Domain-0 0 839 1 r----- 106.6 ACM:mytest:SystemManagement
2828
# xm create domain3.xm
2829
Using config file "./domain3.xm".
2830
Started domain domain3
2833
Name ID Mem VCPUs State Time(s) Label
2834
domain1 7 128 1 -b---- 0.8 ACM:mytest:A-Bank
2835
domain3 8 164 1 -b---- 0.3 ACM:mytest:AutoCorp
2836
Domain-0 0 711 1 r----- 107.6 ACM:mytest:SystemManagement
2841
\subsection{Labeling Managed User Domains}
2842
\label{subsection:acmlabelmanageddomains}
2844
Xend has been extended with functionality to manage domains along with their
2845
configuration information. Such domains are configured and started via Xen-API
2846
calls. Since managed domains do not have an associated xm configuration file,
2847
the existing \verb|addlabel| command, which adds the security label into a
2848
domain's configuration file, will not work for such managed domains.
2850
Therefore, we have extended the \verb|xm addlabel| and \verb|xm rmlabel|
2851
subcommands to enable adding security labels to and removing security
2852
labels from managed domain configurations. The following example shows how
2853
the \verb|A-Bank| label can be assigned to the xend-managed
2854
domain configuration of \verb|domain1|. Removing labels from managed user
2855
domain configurations works similarly.
2857
Below, we show a dormant configuration of the managed domain1
2858
with ID \verb|"-1"| and state \verb|"-----"| before labeling:
2862
Name ID Mem VCPUs State Time(s) Label
2863
domain1 -1 128 1 ------ 0.0 ACM:mytest:__UNLABELED__
2864
Domain-0 0 711 1 r----- 128.4 ACM:mytest:SystemManagement
2868
Now we label the managed domain:
2871
# xm addlabel A-Bank mgt domain1
2872
Successfully set the label of the dormant domain 'domain1' to 'A-Bank'.
2876
After labeling, you can see that the security label is part of the
2877
domain configuration:
2881
Name ID Mem VCPUs State Time(s) Label
2882
domain1 -1 128 1 ------ 0.0 ACM:mytest:A-Bank
2883
Domain-0 0 711 1 r----- 129.7 ACM:mytest:SystemManagement
2887
This command extension does not support relabeling of individual running user domains
2888
for several reasons. For one, because of the difficulty to revoke resources
2889
in cases where a running domain's new label does not permit access to resources
2890
that were accessible under the old label. Another reason is that changing the
2891
label of a single domain of a workload is rarely a good choice and will affect
2892
the workload isolation properties of the overall workload.
2894
However, the name and contents of the label associated with running domains can
2895
be indirectly changed through a global policy change, which will update the whole
2896
workload consistently (domains and resources), cf.
2897
Section~\ref{subsection:acmpolicymanagement}.
2899
\section{Xen Access Control Policy}
2900
\label{section:acmpolicy}
2902
This section describes the sHype/Xen access control policy in detail.
2903
It gives enough information to enable the reader to write custom
2904
access control policies and to use the available Xen policy tools. The
2905
policy language is expressive enough to specify most symmetric access
2906
relationships between domains and resources efficiently.
2908
The Xen access control policy consists of two policy components. The
2909
first component, called Simple Type Enforcement (STE) policy, controls
2910
the sharing between running domains, i.e., communication or access to
2911
shared resources. The second component, called Chinese Wall (CHWALL)
2912
policy, controls which domains can run simultaneously on the same
2913
virtualized platform. The CHWALL and STE policy components complement
2914
each other. The XML policy file includes all information
2915
needed by Xen to enforce those policies.
2917
Figures~\ref{fig:acmxmlfilea} and \ref{fig:acmxmlfileb} show the fully
2918
functional but very simple example Xen security policy that is created
2919
by ezPolicy as shown in Figure~\ref{fig:acmezpolicy}. The policy can
2920
distinguish the 6 workload types shown in lines 11-17 in
2921
Fig.~\ref{fig:acmxmlfilea}. The whole XML Security Policy consists of
2924
\item Policy header including the policy name
2925
\item Simple Type Enforcement block
2926
\item Chinese Wall Policy block
2927
\item Label definition block
2933
01 <?xml version="1.0" ?>
2934
02 <!-- Auto-generated by ezPolicy -->
2935
03 <SecurityPolicyDefinition ...">
2937
05 <PolicyName>mytest</PolicyName>
2938
06 <Date>Mon Nov 19 22:51:56 2007</Date>
2939
07 <Version>1.0</Version>
2941
09 <SimpleTypeEnforcement>
2942
10 <SimpleTypeEnforcementTypes>
2943
11 <Type>SystemManagement</Type>
2944
12 <Type>__UNLABELED__</Type>
2945
13 <Type>A-Bank</Type>
2946
14 <Type>A-Bank.SecurityUnderwriting</Type>
2947
15 <Type>A-Bank.MarketAnalysis</Type>
2948
16 <Type>B-Bank</Type>
2949
17 <Type>AutoCorp</Type>
2950
18 </SimpleTypeEnforcementTypes>
2951
19 </SimpleTypeEnforcement>
2952
20 <ChineseWall priority="PrimaryPolicyComponent">
2953
21 <ChineseWallTypes>
2954
22 <Type>SystemManagement</Type>
2955
23 <Type>__UNLABELED__</Type>
2956
24 <Type>A-Bank</Type>
2957
25 <Type>A-Bank.SecurityUnderwriting</Type>
2958
26 <Type>A-Bank.MarketAnalysis</Type>
2959
27 <Type>B-Bank</Type>
2960
28 <Type>AutoCorp</Type>
2961
29 </ChineseWallTypes>
2963
31 <Conflict name="RER">
2964
32 <Type>A-Bank</Type>
2965
33 <Type>B-Bank</Type>
2966
34 <Type>__UNLABELED__</Type>
2968
36 <Conflict name="RER">
2969
37 <Type>A-Bank.MarketAnalysis</Type>
2970
38 <Type>A-Bank.SecurityUnderwriting</Type>
2971
39 <Type>__UNLABELED__</Type>
2977
\caption{Example XML security policy file -- Part I: Types and Rules Definition.}
2978
\label{fig:acmxmlfilea}
2981
\subsection{Policy Header and Policy Name}
2982
\label{subsection:acmnaming}
2983
Lines 1-2 (cf Figure~\ref{fig:acmxmlfilea}) include the usual XML
2984
header. The security policy definition starts in Line 3 and refers to
2985
the policy schema. The XML-Schema definition for the Xen policy can be
2987
\textit{/etc/xen/acm-security/policies/security-policy.xsd}. Examples
2988
for security policies can be found in the example subdirectory. The
2989
acm-security directory is only installed if ACM security is configured
2990
during installation (cf Section~\ref{subsection:acmexampleconfigure}).
2992
The \verb|Policy Header| spans lines 4-8. It includes a date field and
2993
defines the policy name \verb|mytest| as well
2994
as the version of the XML. It can also include optional fields that are
2995
not shown and are for future use (see schema definition).
2997
The policy name serves two purposes: First, it provides a unique name
2998
for the security policy. This name is also exported by the Xen
2999
hypervisor to the Xen management tools in order to ensure that both
3000
the Xen hypervisor and Domain-0 enforce the same policy.
3001
We plan to extend the policy name with a
3002
digital fingerprint of the policy contents to better protect this
3003
correlation. Second, it implicitly points the xm tools to the
3004
location where the XML policy file is stored on the Xen system.
3005
Replacing the colons in the policy name by slashes yields the local
3006
path to the policy file starting from the global policy directory
3007
\verb|/etc/xen/acm-security/policies|. The last part of the policy
3008
name is the prefix for the XML policy file name, completed by
3009
\verb|-security_policy.xml|. Our example policy with the name
3010
\verb|mytest| can be found in the XML policy file named
3011
\verb|mytest-security_policy.xml| that is stored under the global
3012
policy directory. Another, preinstalled example policy named
3013
\verb|example.test| can be found in the \verb|test-security_policy.xml|
3014
under \verb|/etc/xen/acm-security/policies/example|.
3016
\subsection{Simple Type Enforcement Policy Component}
3018
The Simple Type Enforcement (STE) policy controls which domains can
3019
communicate or share resources. This way, Xen can enforce confinement
3020
of workload types by confining the domains running those workload
3021
types and their resources. The mandatory access control framework
3022
enforces its policy when
3023
domains access intended communication or cooperation means (shared
3024
memory, events, shared resources such as block devices). It builds on
3025
top of the core hypervisor isolation, which restricts the ways of
3026
inter-communication to those intended means. STE does not protect or
3027
intend to protect from covert channels in the hypervisor or hardware;
3028
this is an orthogonal problem that can be mitigated by using the
3029
Run-time Exclusion rules described above or by fixing the problem leading
3030
to those covert channels in the core hypervisor or hardware platform.
3032
Xen controls sharing between domains on the resource and domain level
3033
because this is the abstraction the hypervisor and its management
3034
understand naturally. While this is coarse-grained, it is also very
3035
reliable and robust and it requires minimal changes to implement
3036
mandatory access controls in the hypervisor. It enables platform- and
3037
operating system-independent policies as part of a layered security
3040
Lines 11-17 (cf Figure~\ref{fig:acmxmlfilea}) define the Simple Type
3041
Enforcement policy component. Essentially, they define the workload
3042
type names \verb|SystemManagement|, \verb|A-Bank|,
3043
\verb|AutoCorp| etc. that are available in the STE policy component. The
3044
policy rules are implicit: Xen permits two domains to communicate with
3045
each other if and only if their security labels have at least one STE type in
3046
common. Similarly, Xen permits a user domain to access a
3047
resource if and only if the labels of the domain and the resource
3048
have at least one STE workload type in common.
3050
\subsection{Chinese Wall Policy Component}
3052
The Chinese Wall security policy interpretation of sHype enables users
3053
to prevent certain workloads from running simultaneously on the same
3054
hypervisor platform. Run-time Exclusion rules (RER), also called
3055
Conflict Sets or Anti-Collocation rules, define a set of workload types
3056
that are not permitted to run simultaneously on the same virtualized
3057
platform. Of all the workloads specified in a Run-time
3058
Exclusion rule, at most one type can run on the same hypervisor
3059
platform at a time. Run-time Exclusion Rules implement a less
3060
rigorous variant of the original Chinese Wall security component. They
3061
do not implement the *-property of the policy, which would require to
3062
restrict also types that are not part of an exclusion rule once they
3063
are running together with a type in an exclusion rule
3064
(http://www.gammassl.co.uk/topics/chinesewall.html provides more information
3065
on the original Chinese Wall policy).
3067
Xen considers the \verb|ChineseWallTypes| part of the label for the
3068
enforcement of the Run-time Exclusion rules. It is illegal to define
3069
labels including conflicting Chinese Wall types.
3071
Lines 20-41 (cf Figure~\ref{fig:acmxmlfilea}) define the Chinese Wall
3072
policy component. Lines 22-28 define the known Chinese Wall types,
3073
which coincide here with the STE types defined above. This usually
3074
holds if the criteria for sharing among domains and sharing of the
3075
hardware platform are the same. Lines 30-41 define one Run-time
3076
Exclusion rules, the first of which is depicted below:
3080
31 <Conflict name="RER">
3081
32 <Type>A-Bank</Type>
3082
33 <Type>B-Bank</Type>
3083
34 <Type>__UNLABELED__</Type>
3088
Based on this rule, Xen enforces that only one of the types
3089
\verb|A-Bank|, \verb|B-Bank|, or \verb|__UNLABELED__| will run
3090
on a single hypervisor platform at a time. For example, once a domain assigned a
3091
\verb|A-Bank| workload type is started, domains with the
3092
\verb|B-Bank| type or unlabeled domains will be denied to start.
3093
When the former domain stops and no other domains with the \verb|A-Bank|
3094
type are running, then domains with the \verb|B-Bank| type or unlabeled domains
3097
Xen maintains reference counts on each running workload type to keep
3098
track of which workload types are running. Every time a domain starts
3099
or resumes, the reference count on those Chinese Wall types that are
3100
referenced in the domain's label are incremented. Every time a domain
3101
is destroyed or saved, the reference counts of its Chinese Wall types
3102
are decremented. sHype in Xen fully supports migration and live-migration,
3103
which is subject to access control the same way as saving a domain on
3104
the source platform and resuming it on the destination platform.
3106
Here are some reasons why users might want to restrict workloads or domains
3107
from sharing the system hardware simultaneously:
3110
\item Imperfect resource management or control might enable a compromised
3111
user domain to starve other domains and the workload running in them.
3112
\item Redundant user domains might run the same workload to increase
3113
availability; such domains should not run on the same hardware to
3114
avoid single points of failure.
3115
\item Imperfect Xen core domain isolation might enable two rogue
3116
domains running different workload types to use unintended and
3117
unknown ways (covert channels) to exchange some bits of information.
3118
This way, they bypass the policed Xen access control mechanisms. Such
3119
imperfections cannot be completely eliminated and are a result of
3120
trade-offs between security and other design requirements. For a
3121
simple example of a covert channel see
3122
http://www.multicians.org/timing-chn.html. Such covert channels
3123
exist also between workloads running on different platforms if they
3124
are connected through networks. The Xen Chinese Wall policy provides
3125
an approximated ``air-gap'' between selected workload types.
3128
\subsection{Security Labels}
3130
To enable Xen to associate domains with workload types running in
3131
them, each domain is assigned a security label that includes the
3132
workload types of the domain.
3135
\begin{tabular*}{\textwidth}{@{\extracolsep{\fill}}l|l}
3136
\begin{minipage}{0.475\textwidth}
3139
<SecurityLabelTemplate>
3140
<SubjectLabels bootstrap="SystemManagement">
3141
<VirtualMachineLabel>
3142
<Name>SystemManagement</Name>
3143
<SimpleTypeEnforcementTypes>
3144
<Type>SystemManagement</Type>
3145
<Type>__UNLABELED__</Type>
3147
<Type>A-Bank.SecurityUnderwriting</Type>
3148
<Type>A-Bank.MarketAnalysis</Type>
3150
<Type>AutoCorp</Type>
3151
</SimpleTypeEnforcementTypes>
3153
<Type>SystemManagement</Type>
3155
</VirtualMachineLabel>
3156
<VirtualMachineLabel>
3157
<Name>__UNLABELED__</Name>
3158
<SimpleTypeEnforcementTypes>
3159
<Type>__UNLABELED__</Type>
3160
</SimpleTypeEnforcementTypes>
3162
<Type>__UNLABELED__</Type>
3164
</VirtualMachineLabel>
3165
<VirtualMachineLabel>
3167
<SimpleTypeEnforcementTypes>
3169
</SimpleTypeEnforcementTypes>
3173
</VirtualMachineLabel>
3174
<VirtualMachineLabel>
3175
<Name>A-Bank.SecurityUnderwriting</Name>
3176
<SimpleTypeEnforcementTypes>
3177
<Type>A-Bank.SecurityUnderwriting</Type>
3178
</SimpleTypeEnforcementTypes>
3181
<Type>A-Bank.SecurityUnderwriting</Type>
3183
</VirtualMachineLabel>
3184
<VirtualMachineLabel>
3185
<Name>A-Bank.MarketAnalysis</Name>
3186
<SimpleTypeEnforcementTypes>
3187
<Type>A-Bank.MarketAnalysis</Type>
3188
</SimpleTypeEnforcementTypes>
3191
<Type>A-Bank.MarketAnalysis</Type>
3193
</VirtualMachineLabel>
3194
<VirtualMachineLabel>
3196
<SimpleTypeEnforcementTypes>
3198
</SimpleTypeEnforcementTypes>
3202
</VirtualMachineLabel>
3206
\begin{minipage}{0.475\textwidth}
3209
<VirtualMachineLabel>
3210
<Name>AutoCorp</Name>
3211
<SimpleTypeEnforcementTypes>
3212
<Type>AutoCorp</Type>
3213
</SimpleTypeEnforcementTypes>
3215
<Type>AutoCorp</Type>
3217
</VirtualMachineLabel>
3221
<Name>SystemManagement</Name>
3222
<SimpleTypeEnforcementTypes>
3223
<Type>SystemManagement</Type>
3224
</SimpleTypeEnforcementTypes>
3227
<Name>__UNLABELED__</Name>
3228
<SimpleTypeEnforcementTypes>
3229
<Type>__UNLABELED__</Type>
3230
</SimpleTypeEnforcementTypes>
3234
<SimpleTypeEnforcementTypes>
3236
</SimpleTypeEnforcementTypes>
3239
<Name>A-Bank.SecurityUnderwriting</Name>
3240
<SimpleTypeEnforcementTypes>
3241
<Type>A-Bank.SecurityUnderwriting</Type>
3242
</SimpleTypeEnforcementTypes>
3245
<Name>A-Bank.MarketAnalysis</Name>
3246
<SimpleTypeEnforcementTypes>
3247
<Type>A-Bank.MarketAnalysis</Type>
3248
</SimpleTypeEnforcementTypes>
3252
<SimpleTypeEnforcementTypes>
3254
</SimpleTypeEnforcementTypes>
3257
<Name>AutoCorp</Name>
3258
<SimpleTypeEnforcementTypes>
3259
<Type>AutoCorp</Type>
3260
</SimpleTypeEnforcementTypes>
3263
</SecurityLabelTemplate>
3264
</SecurityPolicyDefinition>
3277
\caption{Example XML security policy file -- Part II: Label Definition.}
3278
\label{fig:acmxmlfileb}
3280
% DO NOT MODIFY WHITESPACE ABOVE, it balances the columns
3281
The \verb|SecurityLabelTemplate| (cf Figure~\ref{fig:acmxmlfileb}) defines
3282
the security labels that can be associated with domains and resources when
3283
this policy is active (use the \verb|xm labels type=any| command described in
3284
Section~\ref{subsection:acmexamplelabeldomains} to list all available labels).
3286
The domain labels include
3287
Chinese Wall types while resource labels do not include Chinese Wall types.
3288
The \verb|SubjectLabels| policy section defines the labels that can be
3289
assigned to domains. The VM label
3290
\verb|A-Bank.SecurityUnderwriting| in Figure~\ref{fig:acmxmlfileb})
3291
associates the domain that carries it with the workload STE type
3292
\verb|A-Bank.SecurityUnderwriting| and with the CHWALL types \verb|A-Bank|
3293
and \verb|A-Bank.SecurityUnderwriting|. The ezPolicy tool
3294
assumes that any department workload will inherit any conflict set that
3295
is specified for its organization, i.e., if \verb|B-Bank| is running, not
3296
only \verb|A-Bank| but also all its departmental workloads are prevented
3297
from running by this first run-time exclusion set. The separation of STE
3298
and CHWALL types in the label definition ensures that
3299
all departmental workloads are isolated from each other and from their generic
3300
organization workloads, while they are sharing CHWALL types to
3301
simplify the formulation of run-time exclusion sets.
3303
The \verb|bootstrap| attribute of the \verb|<SubjectLabels>| XML node
3304
in our example policy shown in Figure~\ref{fig:acmxmlfileb} names
3305
the label \verb|SystemManagement| as the label that Xen will assign
3306
to Domain-0 at boot time (if this policy is installed as boot policy). The
3307
label of Domain-0 can be persistently changed at run-time with the
3308
\verb|addlabel| command, which adds an overriding option to the grub.conf
3309
boot entry (cf Section~\ref{subsection:acmlabeldom0}).
3310
All user domains are assigned labels according to their domain configuration
3311
(see Section~\ref{subsection:acmexamplelabeldomains} for examples of
3312
how to label domains).
3314
The \verb|ObjectLabels| depicted in Figure~\ref{fig:acmxmlfileb} can be
3315
assigned to resources when this policy is active.
3317
In general, user domains should be assigned labels that have only a
3318
single SimpleTypeEnforcement workload type. This way, workloads remain
3319
confined even if user domains become rogue. Any domain that is
3320
assigned a label with multiple STE types must be trusted to keep
3321
information belonging to the different STE types separate (confined).
3322
For example, Domain-0 is assigned the bootstrap label
3323
\verb|SystemManagement|, which includes all existing STE types.
3324
Therefore, Domain-0 must take care not to enable unauthorized
3325
information flow (eg. through block devices or virtual networking)
3326
between domains or resources that are assigned different STE types.
3328
Security administrators simply use the name of a label (specified in
3329
the \verb|<Name>| field) to associate a label with a domain (cf.
3330
Section~\ref{subsection:acmexamplelabeldomains}). The types inside the
3331
label are used by the Xen access control enforcement. While the name
3332
can be arbitrarily chosen (as long as it is unique), it is advisable
3333
to choose the label name in accordance to the security types included.
3334
Similarly, the STE and CHWALL types should be named according to the
3335
workloads they represent. While the XML representation of the label
3336
in the above example seems unnecessary flexible, labels in general
3337
must be able to include multiple types.
3339
We assume in the following example, that \verb|A-Bank.SecurityUnderwriting| and
3340
\verb|A-Bank.MarketAnalysis| workloads use virtual disks that are provided
3341
by a virtual I/O domain hosting a physical storage device and carrying
3342
the following label:
3346
<VirtualMachineLabel>
3347
<Name>VIOServer</Name>
3348
<SimpleTypeEnforcementTypes>
3350
<Type>A-Bank.SecurityUnderwriting</Type>
3351
<Type>A-Bank.MarketAnalysis</Type>
3352
<Type>VIOServer</Type>
3353
</SimpleTypeEnforcementTypes>
3355
<Type>VIOServer</Type>
3357
</VirtualMachineLabel>
3361
This Virtual I/O domain (VIO) exports its virtualized disks by
3362
communicating to all domains labeled with the
3363
\verb|A-Bank.SecurityUnderwriting|, the \verb|A-Bank|, or the
3364
\verb|A-Bank.MarketAnalysis| label. This requires the
3365
VIO domain to carry those STE types. In addition, this label includes a
3366
new \verb|VIOServer| type that can be used to restrict direct access to the
3367
physical storage resource to the VIODomain.
3369
In this example, the confinement of these A-Bank workloads depends on the
3370
VIO domain that must keep the data of those different workloads separate.
3371
The virtual disks are labeled as well to keep track of their assignments
3372
to workload types (see Section~\ref{subsection:acmexamplelabelresources}
3373
for labeling resources) and enforcement functions inside the VIO
3374
domain must ensure that the labels of the domain mounting a virtual
3375
disk and the virtual disk label share a common STE type. The VIO label
3376
carrying its own VIOServer CHWALL type introduces the flexibility to
3377
permit the trusted VIO server to run together with \verb|A-Bank.SecurityUnderwriting|
3378
or \verb|A-Bank.MarketAnalysis| workloads.
3380
Alternatively, a system that has two hard-drives does not need a VIO
3381
domain but can directly assign one hardware storage device to each of
3382
the workloads if the platform offers an IO-MMU, cf
3383
Section~\ref{s:ddsecurity}. Sharing hardware through virtualized devices
3384
is a trade-off between the amount of trusted code (size of the trusted
3385
computing base) and the amount of acceptable over-provisioning. This
3386
holds both for peripherals and for system platforms.
3389
\subsection{Managing sHype/Xen Security Policies at Run-time}
3390
\label{subsection:acmpolicymanagement}
3392
\subsubsection{Removing the sHype/Xen Security Policy}
3393
When resetting the policy, no labeled domains can be running.
3394
Please stop or shutdown all running labeled domains. Then you can reset
3395
the policy to the default policy using the \verb|resetpolicy| command:
3400
Supported security subsystems : ACM
3401
Policy name : mytest
3403
Version of XML policy : 1.0
3404
Policy configuration : loaded, activated for boot
3407
Successfully reset the system's policy.
3410
Supported security subsystems : ACM
3411
Policy name : DEFAULT
3413
Version of XML policy : 1.0
3414
Policy configuration : loaded
3417
file:/home/xen/dom_fc5/fedora.fc5.swap
3421
file:/home/xen/dom_fc5/fedora.fc5.img
3428
As the \verb|xm resources| output shows, all resource labels have
3429
invalidated type information but their semantics remain associated
3430
with the resources so that they can later on either be relabeled
3431
with semantically equivalent labels or sanitized and reused
3432
(storage resources).
3434
At this point, the system is in the same initial state as after
3435
configuring XSM and sHype/ACM and rebooting the system without
3436
a specific policy. No user domains can run.
3438
\subsubsection{Changing to a Different sHype/Xen Security Policy}
3439
The easiest way to change to a different, unrelated policy is to reset the system
3440
policy and then set the new policy. Please consider that the existing
3441
domain and resource labels become invalid at this point. Please refer
3442
to the next section for an example of how to seamlessly update an
3443
active policy at run-time without invalidating labels.
3448
Successfully reset the system's policy.
3450
# xm setpolicy ACM example.test
3451
Successfully set the new policy.
3452
Supported security subsystems : ACM
3453
Policy name : example.test
3455
Version of XML policy : 1.0
3456
Policy configuration : loaded, activated for boot
3464
Name ID Mem VCPUs State Time(s) Label
3465
Domain-0 0 873 1 r----- 56.3 ACM:example.test:SystemManagement
3468
Successfully reset the system's policy.
3471
Supported security subsystems : ACM
3472
Policy name : DEFAULT
3474
Version of XML policy : 1.0
3475
Policy configuration : loaded
3478
Name ID Mem VCPUs State Time(s) Label
3479
Domain-0 0 873 1 r----- 57.2 ACM:DEFAULT:SystemManagement
3481
# xm setpolicy ACM mytest
3482
Successfully set the new policy.
3483
Supported security subsystems : ACM
3484
Policy name : mytest
3486
Version of XML policy : 1.0
3487
Policy configuration : loaded, activated for boot
3491
A-Bank.MarketAnalysis
3492
A-Bank.SecurityUnderwriting
3499
Name ID Mem VCPUs State Time(s) Label
3500
Domain-0 0 873 1 r----- 58.0 ACM:mytest:SystemManagement
3504
The described way of changing policies by resetting the existing
3505
policy is useful for testing different policies. For real deployment
3506
environments, a policy update as described in the following section
3507
is more appropriate and can be applied seamlessly at run-time while
3508
user domains are running.
3510
\subsubsection{Update an sHype/Xen Security Policy at Run-time}
3512
Once an ACM security policy is activated (loaded into the Xen
3513
hypervisor), the policy may be updated at run-time without the
3514
need to re-boot the system. The XML update-policy contains several
3515
additional information fields that are required to safely link the
3516
new policy contents to the old policy and ensure a consistent
3517
transformation of the system security state from the old to the
3518
new policy. Those additional fields are required for policies that
3519
are updating an existing policy at run-time.
3521
The major benefit of policy updates is the ability to add, delete,
3522
or rename workload types, labels, and conflict sets (run-time
3523
exclusion rules) to accommodate changes in the managed virtual
3524
environment without the need to reboot the Xen system. When a
3525
new policy renames labels of the current policy, the labels
3526
attached to resources and domains are automatically updated
3527
during a successful policy update.
3529
We have manually crafted an update policy for the \verb|mytest|
3530
security policy and stored it in the file mytest\_update-security\_policy.xml
3531
in the policies directory. We will discuss this policy in detail before
3532
using it to update a running sHype/Xen system. The following figures contain
3533
the whole contents of the update policy file.
3535
Figure~\ref{fig:acmupdateheader} shows the policy
3536
header of an update-policy and the new \verb|FromPolicy| XML
3537
node. For the policy update to succeed, the policy name and the
3538
policy version fields of the \verb|FromPolicy| XML node must
3539
exactly match those of the currently enforced policy. This
3540
ensures a controlled update path of the policy.
3545
<?xml version="1.0" encoding="UTF-8"?>
3546
<!-- Auto-generated by ezPolicy -->
3547
<SecurityPolicyDefinition xmlns="http://www.ibm.com"
3548
xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
3549
xsi:schemaLocation="http://www.ibm.com ../../security_policy.xsd ">
3551
<PolicyName>mytest</PolicyName>
3552
<Date>Tue Nov 27 21:53:45 2007</Date>
3553
<Version>1.1</Version>
3555
<PolicyName>mytest</PolicyName>
3556
<Version>1.0</Version>
3561
\caption{XML security policy update -- Part I: Updated Policy Header.}
3562
\label{fig:acmupdateheader}
3565
The version number of the new policy, which is shown in the
3566
node following the \verb|Date| node, must be a logical increment
3567
to the current policy's version. Therefore at least the minor
3568
number of the policy version must be incremented. This ensures
3569
that a policy update is applied only to exactly the policy for
3570
which this update was created and minimizes unforseen side-effects
3573
\paragraph{Types and Conflic Sets}
3574
The type names and the assignment of types to labels or conflict
3575
sets (run-time exclusion rules) can
3576
simply be changed consistently throughout the policy. Types,
3577
as opposed to labels, are not directly associated or referenced
3578
outside the policy so they do not need to carry their history
3579
in a ``From'' field. The figure below shows the update for the
3580
types and conflict sets. The \verb|__UNLABELED__| type is removed
3581
to disable support for running unlabeled domains. Additionally,
3582
we have renamed the two \verb|A-Bank| department types with
3583
abbreviated names \verb|A-Bank.SU| and \verb|A-Bank.MA|. You
3584
can also see how those type names are
3585
consistently changed within the conflict set definition.
3590
<SimpleTypeEnforcement>
3591
<SimpleTypeEnforcementTypes>
3592
<Type>SystemManagement</Type>
3594
<Type>A-Bank.SU</Type>
3595
<Type>A-Bank.MA</Type>
3597
<Type>AutoCorp</Type>
3598
</SimpleTypeEnforcementTypes>
3599
</SimpleTypeEnforcement>
3601
<ChineseWall priority="PrimaryPolicyComponent">
3603
<Type>SystemManagement</Type>
3605
<Type>A-Bank.SU</Type>
3606
<Type>A-Bank.MA</Type>
3608
<Type>AutoCorp</Type>
3612
<Conflict name="RER">
3616
<Conflict name="RER">
3617
<Type>A-Bank.MA</Type>
3618
<Type>A-Bank.SU</Type>
3624
\caption{XML security policy update -- Part II: Updated Types and Conflict Sets.}
3625
\label{fig:acmupdatetypesnrules}
3628
In the same way, new types can be introduced and new conflict sets
3629
can be defined by simply adding the types or conflict sets to the
3632
\paragraph{Labels} Virtual machine and resource labels of an existing policy can be
3633
deleted through a policy update simply by omitting them in the
3634
update-policy. However, if a currently running virtual machine
3635
or a currently used resource is labeled with a label not stated
3636
in the update-policy, then the policy update is rejected. This
3637
ensures that a policy update leaves the system in a consistent
3640
A policy update also enables the renaming of virtual machine and
3641
resource labels. Linking the old label name with the new label
3642
name is achieved through the \verb|from| attribute in the
3643
\verb|VirtualMachineLabel| or \verb|ResourceLabel| nodes in the
3644
update-policy. Figure~\ref{fig:acmupdatelabels} shown how subject
3646
are updated from their old name \verb|A-Bank.SecurityUnterwriting|
3647
to their new name \verb|A-Bank.SU| using the \verb|from| attribute.
3650
\begin{tabular*}{\textwidth}{@{\extracolsep{\fill}}l|l}
3651
\begin{minipage}{0.475\textwidth}
3654
<SecurityLabelTemplate>
3655
<SubjectLabels bootstrap="SystemManagement">
3656
<VirtualMachineLabel>
3657
<Name>SystemManagement</Name>
3658
<SimpleTypeEnforcementTypes>
3659
<Type>SystemManagement</Type>
3661
<Type>A-Bank.SU</Type>
3662
<Type>A-Bank.MA</Type>
3664
<Type>AutoCorp</Type>
3665
</SimpleTypeEnforcementTypes>
3667
<Type>SystemManagement</Type>
3669
</VirtualMachineLabel>
3670
<VirtualMachineLabel>
3671
<Name>A-Bank-WL</Name>
3672
<SimpleTypeEnforcementTypes>
3673
<Type>SystemManagement</Type>
3675
<Type>A-Bank.SU</Type>
3676
<Type>A-Bank.MA</Type>
3677
</SimpleTypeEnforcementTypes>
3679
<Type>SystemManagement</Type>
3681
</VirtualMachineLabel>
3682
<VirtualMachineLabel>
3684
<SimpleTypeEnforcementTypes>
3686
</SimpleTypeEnforcementTypes>
3690
</VirtualMachineLabel>
3691
<VirtualMachineLabel>
3692
<Name from="A-Bank.SecurityUnderwriting">
3694
<SimpleTypeEnforcementTypes>
3695
<Type>A-Bank.SU</Type>
3696
</SimpleTypeEnforcementTypes>
3699
<Type>A-Bank.SU</Type>
3701
</VirtualMachineLabel>
3702
<VirtualMachineLabel>
3703
<Name from="A-Bank.MarketAnalysis">
3705
<SimpleTypeEnforcementTypes>
3706
<Type>A-Bank.MA</Type>
3707
</SimpleTypeEnforcementTypes>
3710
<Type>A-Bank.MA</Type>
3712
</VirtualMachineLabel>
3716
\begin{minipage}{0.475\textwidth}
3719
<VirtualMachineLabel>
3721
<SimpleTypeEnforcementTypes>
3723
</SimpleTypeEnforcementTypes>
3727
</VirtualMachineLabel>
3728
<VirtualMachineLabel>
3729
<Name>AutoCorp</Name>
3730
<SimpleTypeEnforcementTypes>
3731
<Type>AutoCorp</Type>
3732
</SimpleTypeEnforcementTypes>
3734
<Type>AutoCorp</Type>
3736
</VirtualMachineLabel>
3741
<Name>SystemManagement</Name>
3742
<SimpleTypeEnforcementTypes>
3743
<Type>SystemManagement</Type>
3744
</SimpleTypeEnforcementTypes>
3748
<SimpleTypeEnforcementTypes>
3750
</SimpleTypeEnforcementTypes>
3753
<Name from="A-Bank.SecurityUnderwriting">
3755
<SimpleTypeEnforcementTypes>
3756
<Type>A-Bank.SU</Type>
3757
</SimpleTypeEnforcementTypes>
3760
<Name from="A-Bank.MarketAnalysis">
3762
<SimpleTypeEnforcementTypes>
3763
<Type>A-Bank.MA</Type>
3764
</SimpleTypeEnforcementTypes>
3768
<SimpleTypeEnforcementTypes>
3770
</SimpleTypeEnforcementTypes>
3773
<Name>AutoCorp</Name>
3774
<SimpleTypeEnforcementTypes>
3775
<Type>AutoCorp</Type>
3776
</SimpleTypeEnforcementTypes>
3779
</SecurityLabelTemplate>
3780
</SecurityPolicyDefinition>
3785
\caption{XML security policy update -- Part III: Updated Label Definition.}
3786
\label{fig:acmupdatelabels}
3788
% DO NOT MODIFY WHITESPACE ABOVE, it balances the columns
3790
The updated label definition also includes a new label \verb|A-Bank-WL|
3791
that includes all STE types related to A-Bank. Its CHWALL type
3792
is \verb|SystemManagement|. This indicates that this label is designed
3793
as Domain-0 label. A Xen system can be restricted to only run A-Bank
3794
related workloads by relabeling Domain-0 with the \verb|A-Bank-WL| label.
3796
We assume that the update-policy shown in
3797
Figures~\ref{fig:acmupdateheader}, \ref{fig:acmupdatetypesnrules}
3798
and \ref{fig:acmupdatelabels}
3799
is stored in the XML file mytest\_update-security\_policy.xml located
3800
in the ACM policy directory. See Section~\ref{subsection:acmnaming}
3801
for information about policy names and locations.
3803
The following \verb|xm setpolicy| command updates the active ACM
3804
security policy at run-time.
3809
Name ID Mem VCPUs State Time(s) Label
3810
domain1 2 128 1 -b---- 0.6 ACM:mytest:A-Bank
3811
domain4 3 164 1 -b---- 0.3 ACM:mytest:A-Bank.SecurityUnderwriting
3812
Domain-0 0 711 1 r----- 71.8 ACM:mytest:SystemManagement
3815
file:/home/xen/dom_fc5/fedora.fc5.swap
3819
file:/home/xen/dom_fc5/fedora.fc5.img
3824
# xm setpolicy ACM mytest_update
3825
Successfully set the new policy.
3826
Supported security subsystems : ACM
3827
Policy name : mytest
3829
Version of XML policy : 1.1
3830
Policy configuration : loaded, activated for boot
3833
Name ID Mem VCPUs State Time(s) Label
3834
domain1 2 128 1 -b---- 0.7 ACM:mytest:A-Bank
3835
domain4 3 164 1 -b---- 0.3 ACM:mytest:A-Bank.SU
3836
Domain-0 0 711 1 r----- 72.8 ACM:mytest:SystemManagement
3847
file:/home/xen/dom_fc5/fedora.fc5.swap
3851
file:/home/xen/dom_fc5/fedora.fc5.img
3858
After successful completion of this command, \verb|xm list --label|
3859
shows that the labels of running domains changed to their new names.
3860
\verb|xm labels| shows that new labels \verb|A-Bank.SU| and \verb|A-Bank.AM|
3861
are now available in the policy. The resource labels remain valid after
3862
the successful update as \verb|xm resources| confirms.
3864
The \verb|setpolicy| command fails if the new policy is inconsistent
3865
with the current one or the policy is inconsistent internally (e.g., types
3866
are renamed in the type definition but not in the label definition part of
3867
the policy). In this case, the old policy remains active.
3869
After relabeling Domain-0 with the new \verb|A-Bank-WL| label, we can no
3870
longer run domains labeled \verb|B-Bank| or \verb|AutoCorp| since their
3871
STE types are not a subset of the new Domain-0 label.
3875
# xm addlabel A-Bank-WL mgt Domain-0
3876
Successfully set the label of domain 'Domain-0' to 'A-Bank-WL'.
3879
Name ID Mem VCPUs State Time(s) Label
3880
domain1 2 128 1 -b---- 0.8 ACM:mytest:A-Bank
3881
Domain-0 0 711 1 r----- 74.5 ACM:mytest:A-Bank-WL
3882
domain4 3 164 1 -b---- 0.3 ACM:mytest:A-Bank.SU
3884
# xm getlabel dom domain3.xm
3885
policytype=ACM,policy=mytest,label=AutoCorp
3887
# xm create domain3.xm
3888
Using config file "./domain3.xm".
3889
Error: VM is not authorized to run.
3891
# xm addlabel SystemManagement mgt Domain-0
3892
Successfully set the label of domain 'Domain-0' to 'SystemManagement'.
3895
Name ID Mem VCPUs State Time(s) Label
3896
domain1 2 128 1 -b---- 0.8 ACM:mytest:A-Bank
3897
domain4 3 164 1 -b---- 0.3 ACM:mytest:A-Bank.SU
3898
Domain-0 0 709 1 r----- 76.4 ACM:mytest:SystemManagement
3900
# xm create domain3.xm
3901
Using config file "./domain3.xm".
3902
Started domain domain3
3905
Name ID Mem VCPUs State Time(s) Label
3906
domain1 2 128 1 -b---- 0.8 ACM:mytest:A-Bank
3907
domain4 3 164 1 -b---- 0.3 ACM:mytest:A-Bank.SU
3908
domain3 4 164 1 -b---- 0.3 ACM:mytest:AutoCorp
3909
Domain-0 0 547 1 r----- 77.5 ACM:mytest:SystemManagement
3913
In the same manner, you can add new labels to support new workloads and
3914
add, delete, or rename workload types (STE and/or CHWALL types) simply
3915
by changing the composition of labels. Another use case is to add new
3916
workload types to the current Domain-0 label to enable them to run.
3917
Conflict sets (run-time exclusion rules) can be simply omitted or added.
3918
The policy and label changes become active at once and new workloads
3919
can be run in protected mode without rebooting the Xen system.
3921
In all these cases, if any running user domain would--under the new policy--not
3922
be allowed to run or would not be allowed to access any of the resources
3923
it currently uses, then the policy update is rejected. In this case, you
3924
can stop domains that conflict with the new policy and update the policy
3925
afterwards. The old policy remains active until a policy update succeeds
3926
or Xen is re-booted into a new policy.
3928
\subsection{Tools For Creating sHype/Xen Security Policies}
3929
To create a security policy for Xen, you can use one of the following
3932
\item \verb|ezPolicy| GUI tool -- start writing policies
3933
\item \verb|xensec_gen| tool -- refine policies created with \verb|ezPolicy|
3934
\item text or XML editor
3937
We use the \verb|ezPolicy| tool in
3938
Section~\ref{subsection:acmexamplecreate} to quickly create a workload
3939
protection policy. If desired, the resulting XML policy file can be
3940
loaded into the \verb|xensec_gen| tool to refine it. It can also be
3941
directly edited using an XML editor. Any XML policy file is verified
3942
against the security policy schema when it is translated (see
3943
Subsection~\ref{subsection:acmexampleinstall}).
3945
\section{Current Limitations}
3946
\label{section:acmlimitations}
3948
The sHype/ACM configuration for Xen is work in progress. There is
3949
ongoing work for protecting virtualized resources and planned and
3950
ongoing work for protecting access to remote resources and domains.
3951
The following sections describe limitations of some of the areas into
3952
which access control is being extended.
3954
\subsection{Network Traffic}
3955
Local and remote network traffic is currently not controlled.
3956
Solutions to add sHype/ACM policy enforcement to the virtual network
3957
exist but need to be discussed before they can become part of Xen.
3958
Subjecting external network traffic to the ACM security policy is work
3959
in progress. Manually setting up filters in domain 0 is required for
3960
now but does not scale well.
3962
\subsection{Resource Access and Usage Control}
3964
Enforcing the security policy across multiple hypervisor systems and
3965
on access to remote shared resources is work in progress. Extending
3966
access control to new types of resources is ongoing work (e.g. network
3969
On a single Xen system, information about the association of resources
3970
and security labels is stored in
3971
\verb|/var/lib/xend/security/policies/resource_labels|. This file relates
3972
a full resource path with a security label. This association is weak
3973
and will break if resources are moved or renamed without adapting the
3974
label file. Improving the protection of label-resource relationships
3977
Controlling resource usage and enforcing resource limits in general is
3978
ongoing work in the Xen community.
3980
\subsection{Domain Migration}
3982
Labels on domains are enforced during domain migration and the
3983
destination hypervisor will ensure that the domain label is valid and
3984
the domain is permitted to run (considering the Chinese Wall policy
3985
rules) before it accepts the migration. However, the network between
3986
the source and destination hypervisor as well as both hypervisors must
3987
be trusted. Architectures and prototypes exist that both protect the
3988
network connection and ensure that the hypervisors enforce access
3989
control consistently but patches are not yet available for the main
3992
\subsection{Covert Channels}
3994
The sHype access control aims at system independent security policies.
3995
It builds on top of the core hypervisor isolation. Any covert channels
3996
that exist in the core hypervisor or in the hardware (e.g., shared
3997
processor cache) will be inherited. If those covert channels are not
3998
the result of trade-offs between security and other system properties,
3999
then they are most effectively minimized or eliminated where they are
4000
caused. sHype offers however some means to mitigate their impact, e.g.,
4001
run-time exclusion rules (cf Section~\ref{subsection:acmexamplecreate})
4002
or limiting the system authorization (cf Section~\ref{subsection:acmlabeldom0}).
4007
%% Chapter Build and Boot Options
4008
\chapter{Build and Boot Options}
4010
This chapter describes the build- and boot-time options which may be
4011
used to tailor your Xen system.
4013
\section{Top-level Configuration Options}
4015
Top-level configuration is achieved by editing one of two
4016
files: \path{Config.mk} and \path{Makefile}.
4018
The former allows the overall build target architecture to be
4019
specified. You will typically not need to modify this unless
4020
you are cross-compiling. Additional configuration options are
4021
documented in the \path{Config.mk} file.
4023
The top-level \path{Makefile} is chiefly used to customize the set of
4024
kernels built. Look for the line:
4027
KERNELS ?= linux-2.6-xen0 linux-2.6-xenU
4031
Allowable options here are any kernels which have a corresponding
4032
build configuration file in the \path{buildconfigs/} directory.
4036
\section{Xen Build Options}
4038
Xen provides a number of build-time options which should be set as
4039
environment variables or passed on make's command-line.
4042
\item[verbose=y] Enable debugging messages when Xen detects an
4043
unexpected condition. Also enables console output from all domains.
4044
\item[debug=y] Enable debug assertions. Implies {\bf verbose=y}.
4045
(Primarily useful for tracing bugs in Xen).
4046
\item[debugger=y] Enable the in-Xen debugger. This can be used to
4047
debug Xen, guest OSes, and applications.
4048
\item[perfc=y] Enable performance counters for significant events
4049
within Xen. The counts can be reset or displayed on Xen's console
4050
via console control keys.
4054
\section{Xen Boot Options}
4057
These options are used to configure Xen's behaviour at runtime. They
4058
should be appended to Xen's command line, either manually or by
4059
editing \path{grub.conf}.
4062
\item [ noreboot ] Don't reboot the machine automatically on errors.
4063
This is useful to catch debug output if you aren't catching console
4064
messages via the serial line.
4065
\item [ nosmp ] Disable SMP support. This option is implied by
4067
\item [ watchdog ] Enable NMI watchdog which can report certain
4069
\item [ noirqbalance ] Disable software IRQ balancing and affinity.
4070
This can be used on systems such as Dell 1850/2850 that have
4071
workarounds in hardware for IRQ-routing issues.
4072
\item [ badpage=$<$page number$>$,$<$page number$>$, \ldots ] Specify
4073
a list of pages not to be allocated for use because they contain bad
4074
bytes. For example, if your memory tester says that byte 0x12345678
4075
is bad, you would place `badpage=0x12345' on Xen's command line.
4076
\item [ serial\_tx\_buffer=$<$size$>$ ] Size of serial transmit
4077
buffers. Default is 16kB.
4078
\item [ com1=$<$baud$>$,DPS,$<$io\_base$>$,$<$irq$>$
4079
com2=$<$baud$>$,DPS,$<$io\_base$>$,$<$irq$>$ ] \mbox{}\\
4080
Xen supports up to two 16550-compatible serial ports. For example:
4081
`com1=9600, 8n1, 0x408, 5' maps COM1 to a 9600-baud port, 8 data
4082
bits, no parity, 1 stop bit, I/O port base 0x408, IRQ 5. If some
4083
configuration options are standard (e.g., I/O base and IRQ), then
4084
only a prefix of the full configuration string need be specified. If
4085
the baud rate is pre-configured (e.g., by the bootloader) then you
4086
can specify `auto' in place of a numeric baud rate.
4087
\item [ console=$<$specifier list$>$ ] Specify the destination for Xen
4088
console I/O. This is a comma-separated list of, for example:
4090
\item[ vga ] Use VGA console (until domain 0 boots, unless {\bf
4091
vga=...keep } is specified).
4092
\item[ com1 ] Use serial port com1.
4093
\item[ com2H ] Use serial port com2. Transmitted chars will have the
4094
MSB set. Received chars must have MSB set.
4095
\item[ com2L] Use serial port com2. Transmitted chars will have the
4096
MSB cleared. Received chars must have MSB cleared.
4098
The latter two examples allow a single port to be shared by two
4099
subsystems (e.g.\ console and debugger). Sharing is controlled by
4100
MSB of each transmitted/received character. [NB. Default for this
4101
option is `com1,vga']
4102
\item [ vga=$<$mode$>$(,keep) ] The mode is one of the following options:
4104
\item[ ask ] Display a vga menu allowing manual selection of video
4106
\item[ current ] Use existing vga mode without modification.
4107
\item[ text-$<$mode$>$ ] Select text-mode resolution, where mode is
4108
one of 80x25, 80x28, 80x30, 80x34, 80x43, 80x50, 80x60.
4109
\item[ gfx-$<$mode$>$ ] Select VESA graphics mode
4110
$<$width$>$x$<$height$>$x$<$depth$>$ (e.g., `vga=gfx-1024x768x32').
4111
\item[ mode-$<$mode$>$ ] Specify a mode number as discovered by `vga
4112
ask'. Note that the numbers are displayed in hex and hence must be
4113
prefixed by `0x' here (e.g., `vga=mode-0x0335').
4115
The mode may optionally be followed by `{\bf,keep}' to cause Xen to keep
4116
writing to the VGA console after domain 0 starts booting (e.g., `vga=text-80x50,keep').
4117
\item [ no-real-mode ] (x86 only) Do not execute real-mode bootstrap
4118
code when booting Xen. This option should not be used except for
4119
debugging. It will effectively disable the {\bf vga} option, which
4120
relies on real mode to set the video mode.
4121
\item [ edid=no,force ] (x86 only) Either force retrieval of monitor
4122
EDID information via VESA DDC, or disable it (edid=no). This option
4123
should not normally be required except for debugging purposes.
4124
\item [ edd=off,on,skipmbr ] (x86 only) Control retrieval of Extended
4125
Disc Data (EDD) from the BIOS during boot.
4126
\item [ console\_to\_ring ] Place guest console output into the
4127
hypervisor console ring buffer. This is disabled by default.
4128
When enabled, both hypervisor output and guest console output
4129
is available from the ring buffer. This can be useful for logging
4130
and/or remote presentation of console data.
4131
\item [ sync\_console ] Force synchronous console output. This is
4132
useful if you system fails unexpectedly before it has sent all
4133
available output to the console. In most cases Xen will
4134
automatically enter synchronous mode when an exceptional event
4135
occurs, but this option provides a manual fallback.
4136
\item [ conswitch=$<$switch-char$><$auto-switch-char$>$ ] Specify how
4137
to switch serial-console input between Xen and DOM0. The required
4138
sequence is CTRL-$<$switch-char$>$ pressed three times. Specifying
4139
the backtick character disables switching. The
4140
$<$auto-switch-char$>$ specifies whether Xen should auto-switch
4141
input to DOM0 when it boots --- if it is `x' then auto-switching is
4142
disabled. Any other value, or omitting the character, enables
4143
auto-switching. [NB. Default switch-char is `a'.]
4144
\item [ loglvl=$<$level$>/<$level$>$ ]
4145
Specify logging level. Messages of the specified severity level (and
4146
higher) will be printed to the Xen console. Valid levels are `none',
4147
`error', `warning', `info', `debug', and `all'. The second level
4148
specifier is optional: it is used to specify message severities
4149
which are to be rate limited. Default is `loglvl=warning'.
4150
\item [ guest\_loglvl=$<$level$>/<$level$>$ ] As for loglvl, but
4151
applies to messages relating to guests. Default is
4152
`guest\_loglvl=none/warning'.
4153
\item [ console\_timestamps ]
4154
Adds a timestamp prefix to each line of Xen console output.
4156
Specify what to do with an NMI parity or I/O error. \\
4157
`nmi=fatal': Xen prints a diagnostic and then hangs. \\
4158
`nmi=dom0': Inform DOM0 of the NMI. \\
4159
`nmi=ignore': Ignore the NMI.
4160
\item [ mem=xxx ] Set the physical RAM address limit. Any RAM
4161
appearing beyond this physical address in the memory map will be
4162
ignored. This parameter may be specified with a B, K, M or G suffix,
4163
representing bytes, kilobytes, megabytes and gigabytes respectively.
4164
The default unit, if no suffix is specified, is kilobytes.
4165
\item [ dom0\_mem=$<$specifier list$>$ ] Set the amount of memory to
4166
be allocated to domain 0. This is a comma-separated list containing
4167
the following optional components:
4169
\item[ min:$<$min\_amt$>$ ] Minimum amount to allocate to domain 0
4170
\item[ max:$<$min\_amt$>$ ] Maximum amount to allocate to domain 0
4171
\item[ $<$amt$>$ ] Precise amount to allocate to domain 0
4173
Each numeric parameter may be specified with a B, K, M or
4174
G suffix, representing bytes, kilobytes, megabytes and gigabytes
4175
respectively; if no suffix is specified, the parameter defaults to
4176
kilobytes. Negative values are subtracted from total available
4177
memory. If $<$amt$>$ is not specified, it defaults to all available
4178
memory less a small amount (clamped to 128MB) for uses such as DMA
4180
\item [ dom0\_vcpus\_pin ] Pins domain 0 VCPUs on their respective
4181
physical CPUS (default=false).
4182
\item [ tbuf\_size=xxx ] Set the size of the per-cpu trace buffers, in
4184
\item [ sched=xxx ] Select the CPU scheduler Xen should use. The
4185
current possibilities are `credit' (default), and `sedf'.
4186
\item [ apic\_verbosity=debug,verbose ] Print more detailed
4187
information about local APIC and IOAPIC configuration.
4188
\item [ lapic ] Force use of local APIC even when left disabled by
4190
\item [ nolapic ] Ignore local APIC in a uniprocessor system, even if
4191
enabled by the BIOS.
4192
\item [ apic=bigsmp,default,es7000,summit ] Specify NUMA platform.
4193
This can usually be probed automatically.
4194
\item [ dma\_bits=xxx ] Specify width of DMA addresses in bits. This
4195
is used in NUMA systems to prevent this special DMA memory from
4196
being exhausted in one node when remote nodes have available memory.
4197
\item [ vcpu\_migration\_delay=$<$minimum\_time$>$] Set minimum time of
4198
vcpu migration in microseconds (default 0). This parameter avoids agressive
4199
vcpu migration. For example, the linux kernel uses 0.5ms by default.
4202
In addition, the following options may be specified on the Xen command
4203
line. Since domain 0 shares responsibility for booting the platform,
4204
Xen will automatically propagate these options to its command line.
4205
These options are taken from Linux's command-line syntax with
4206
unchanged semantics.
4209
\item [ acpi=off,force,strict,ht,noirq,\ldots ] Modify how Xen (and
4210
domain 0) parses the BIOS ACPI tables.
4211
\item [ acpi\_skip\_timer\_override ] Instruct Xen (and domain~0) to
4212
ignore timer-interrupt override instructions specified by the BIOS
4214
\item [ noapic ] Instruct Xen (and domain~0) to ignore any IOAPICs
4215
that are present in the system, and instead continue to use the
4220
\section{XenLinux Boot Options}
4222
In addition to the standard Linux kernel boot options, we support:
4224
\item[ xencons=xxx ] Specify the device node to which the Xen virtual
4225
console driver is attached. The following options are supported:
4228
`xencons=off': disable virtual console \\
4229
`xencons=tty': attach console to /dev/tty1 (tty0 at boot-time) \\
4230
`xencons=ttyS': attach console to /dev/ttyS0 \\
4231
`xencons=xvc': attach console to /dev/xvc0
4234
The default is ttyS for dom0 and xvc for all other domains.
4238
%% Chapter Further Support
4239
\chapter{Further Support}
4241
If you have questions that are not answered by this manual, the
4242
sources of information listed below may be of interest to you. Note
4243
that bug reports, suggestions and contributions related to the
4244
software (or the documentation) should be sent to the Xen developers'
4245
mailing list (address below).
4248
\section{Other Documentation}
4250
For developers interested in porting operating systems to Xen, the
4251
\emph{Xen Interface Manual} is distributed in the \path{docs/}
4252
directory of the Xen source distribution.
4255
\section{Online References}
4257
The official Xen web site can be found at:
4258
\begin{quote} {\tt http://www.xen.org}
4261
This contains links to the latest versions of all online
4262
documentation, including the latest version of the FAQ.
4264
Information regarding Xen is also available at the Xen Wiki at
4265
\begin{quote} {\tt http://wiki.xensource.com/xenwiki/}\end{quote}
4266
The Xen project uses Bugzilla as its bug tracking system. You'll find
4267
the Xen Bugzilla at http://bugzilla.xensource.com/bugzilla/.
4270
\section{Mailing Lists}
4272
There are several mailing lists that are used to discuss Xen related
4273
topics. The most widely relevant are listed below. An official page of
4274
mailing lists and subscription information can be found at \begin{quote}
4275
{\tt http://lists.xensource.com/} \end{quote}
4278
\item[xen-devel@lists.xensource.com] Used for development
4279
discussions and bug reports. Subscribe at: \\
4280
{\small {\tt http://lists.xensource.com/xen-devel}}
4281
\item[xen-users@lists.xensource.com] Used for installation and usage
4282
discussions and requests for help. Subscribe at: \\
4283
{\small {\tt http://lists.xensource.com/xen-users}}
4284
\item[xen-announce@lists.xensource.com] Used for announcements only.
4286
{\small {\tt http://lists.xensource.com/xen-announce}}
4287
\item[xen-changelog@lists.xensource.com] Changelog feed
4288
from the unstable and 3.x trees - developer oriented. Subscribe at: \\
4289
{\small {\tt http://lists.xensource.com/xen-changelog}}
4294
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
4298
\chapter{Unmodified (HVM) guest domains in Xen with Hardware support for Virtualization}
4300
Xen supports guest domains running unmodified guest operating systems using
4301
virtualization extensions available on recent processors. Currently processors
4302
featuring the Intel Virtualization Extension (Intel-VT) or the AMD extension
4303
(AMD-V) are supported. The technology covering both implementations is
4304
called HVM (for Hardware Virtual Machine) in Xen. More information about the
4305
virtualization extensions are available on the respective websites:
4306
{\small {\tt http://www.intel.com/technology/computing/vptech}}
4309
{\small {\tt http://www.amd.com/us-en/assets/content\_type/white\_papers\_and\_tech\_docs/24593.pdf}}
4311
\section{Building Xen with HVM support}
4313
The following packages need to be installed in order to build Xen with HVM support. Some Linux distributions do not provide these packages by default.
4315
\begin{tabular}{lp{11.0cm}}
4316
{\bfseries Package} & {\bfseries Description} \\
4318
dev86 & The dev86 package provides an assembler and linker for real mode 80x86 instructions. You need to have this package installed in order to build the BIOS code which runs in (virtual) real mode.
4320
If the dev86 package is not available on the x86\_64 distribution, you can install the i386 version of it. The dev86 rpm package for various distributions can be found at {\scriptsize {\tt http://www.rpmfind.net/linux/rpm2html/search.php?query=dev86\&submit=Search}} \\
4322
SDL-devel, SDL & Simple DirectMedia Layer (SDL) is another way of virtualizing the unmodified guest console. It provides an X window for the guest console.
4324
If the SDL and SDL-devel packages are not installed by default on the build system, they can be obtained from {\scriptsize {\tt http://www.rpmfind.net/linux/rpm2html/search.php?query=SDL\&submit=Search}}
4327
{\scriptsize {\tt http://www.rpmfind.net/linux/rpm2html/search.php?query=SDL-devel\&submit=Search}} \\
4331
\section{Configuration file for unmodified HVM guests}
4333
The Xen installation includes a sample configuration file, {\small {\tt /etc/xen/xmexample.hvm}}. There are comments describing all the options. In addition to the common options that are the same as those for paravirtualized guest configurations, HVM guest configurations have the following settings:
4335
\begin{tabular}{lp{11.0cm}}
4337
{\bfseries Parameter} & {\bfseries Description} \\
4339
kernel & The HVM firmware loader, {\small {\tt /usr/lib/xen/boot/hvmloader}}\\
4341
builder & The domain build function. The HVM domain uses the 'hvm' builder.\\
4343
acpi & Enable HVM guest ACPI, default=1 (enabled)\\
4345
apic & Enable HVM guest APIC, default=1 (enabled)\\
4347
pae & Enable HVM guest PAE, default=1 (enabled)\\
4349
hap & Enable hardware-assisted paging support, such as AMD-V's nested paging
4350
or Intel\textregistered VT's extended paging. If available, Xen will
4351
use hardware-assisted paging instead of shadow paging for this guest's memory
4354
vif & Optionally defines MAC address and/or bridge for the network interfaces. Random MACs are assigned if not given. {\small {\tt type=ioemu}} means ioemu is used to virtualize the HVM NIC. If no type is specified, vbd is used, as with paravirtualized guests.\\
4356
disk & Defines the disk devices you want the domain to have access to, and what you want them accessible as. If using a physical device as the HVM guest's disk, each disk entry is of the form
4358
{\small {\tt phy:UNAME,ioemu:DEV,MODE,}}
4360
where UNAME is the host device file, DEV is the device name the domain will see, and MODE is r for read-only, w for read-write. ioemu means the disk will use ioemu to virtualize the HVM disk. If not adding ioemu, it uses vbd like paravirtualized guests.
4362
If using disk image file, its form should be like
4364
{\small {\tt file:FILEPATH,ioemu:DEV,MODE}}
4366
Optical devices can be emulated by appending cdrom to the device type
4368
{\small {\tt ',hdc:cdrom,r'}}
4370
If using more than one disk, there should be a comma between each disk entry. For example:
4372
{\scriptsize {\tt disk = ['file:/var/images/image1.img,ioemu:hda,w', 'phy:hda1,hdb1,w', 'file:/var/images/install1.iso,hdc:cdrom,r']}}\\
4374
boot & Boot from floppy (a), hard disk (c) or CD-ROM (d). For example, to boot from CD-ROM and fallback to HD, the entry should be:
4378
device\_model & The device emulation tool for HVM guests. This parameter should not be changed.\\
4380
sdl & Enable SDL library for graphics, default = 0 (disabled)\\
4382
vnc & Enable VNC library for graphics, default = 1 (enabled)\\
4384
vncconsole & Enable spawning of the vncviewer (only valid when vnc=1), default = 0 (disabled)
4386
If vnc=1 and vncconsole=0, user can use vncviewer to manually connect HVM from remote. For example:
4388
{\small {\tt vncviewer domain0\_IP\_address:HVM\_domain\_id}} \\
4390
serial & Enable redirection of HVM serial output to pty device\\
4394
\begin{tabular}{lp{10cm}}
4396
usb & Enable USB support without defining a specific USB device.
4397
This option defaults to 0 (disabled) unless the option usbdevice is
4398
specified in which case this option then defaults to 1 (enabled).\\
4400
usbdevice & Enable USB support and also enable support for the given
4401
device. Devices that can be specified are {\small {\tt mouse}} (a PS/2 style
4402
mouse), {\small {\tt tablet}} (an absolute pointing device) and
4403
{\small {\tt host:id1:id2}} (a physical USB device on the host machine whose
4404
ids are {\small {\tt id1}} and {\small {\tt id2}}). The advantage
4405
of {\small {\tt tablet}} is that Windows guests will automatically recognize
4406
and support this device so specifying the config line
4414
will create a mouse that works transparently with Windows guests under VNC.
4415
Linux doesn't recognize the USB tablet yet so Linux guests under VNC will
4416
still need the Summagraphics emulation.
4417
Details about mouse emulation are provided in section \textbf{A.4.3}.\\
4419
localtime & Set the real time clock to local time [default=0, that is, set to UTC].\\
4421
soundhw & Enable sound card support and specify the hardware to emulate. Values can be sb16, es1370 or all. Default is none.\\
4423
full-screen & Start in full screen.\\
4425
nographic & Another way to redirect serial output. If enabled, no 'sdl' or 'vnc' can work. Not recommended.\\
4430
\section{Creating virtual disks from scratch}
4431
\subsection{Using physical disks}
4432
If you are using a physical disk or physical disk partition, you need to install a Linux OS on the disk first. Then the boot loader should be installed in the correct place. For example {\small {\tt dev/sda}} for booting from the whole disk, or {\small {\tt /dev/sda1}} for booting from partition 1.
4434
\subsection{Using disk image files}
4435
You need to create a large empty disk image file first; then, you need to install a Linux OS onto it. There are two methods you can choose. One is directly installing it using a HVM guest while booting from the OS installation CD-ROM. The other is copying an installed OS into it. The boot loader will also need to be installed.
4437
\subsubsection*{To create the image file:}
4438
The image size should be big enough to accommodate the entire OS. This example assumes the size is 1G (which is probably too small for most OSes).
4440
{\small {\tt \# dd if=/dev/zero of=hd.img bs=1M count=0 seek=1024}}
4442
\subsubsection*{To directly install Linux OS into an image file using a HVM guest:}
4444
Install Xen and create HVM with the original image file with booting from CD-ROM. Then it is just like a normal Linux OS installation. The HVM configuration file should have a stanza for the CD-ROM as well as a boot device specification:
4446
{\small {\tt disk=['file:/var/images/your-hd.img,hda,w', ',hdc:cdrom,r' ]
4449
If this method does not succeed, you can choose the following method of copying an installed Linux OS into an image file.
4451
\subsubsection*{To copy a installed OS into an image file:}
4452
Directly installing is an easier way to make partitions and install an OS in a disk image file. But if you want to create a specific OS in your disk image, then you will most likely want to use this method.
4455
\item {\bfseries Install a normal Linux OS on the host machine}\\
4456
You can choose any way to install Linux, such as using yum to install Red Hat Linux or YAST to install Novell SuSE Linux. The rest of this example assumes the Linux OS is installed in {\small {\tt /var/guestos/}}.
4458
\item {\bfseries Make the partition table}\\
4459
The image file will be treated as hard disk, so you should make the partition table in the image file. For example:
4461
{\scriptsize {\tt \# losetup /dev/loop0 hd.img\\
4462
\# fdisk -b 512 -C 4096 -H 16 -S 32 /dev/loop0\\
4463
press 'n' to add new partition\\
4464
press 'p' to choose primary partition\\
4465
press '1' to set partition number\\
4466
press "Enter" keys to choose default value of "First Cylinder" parameter.\\
4467
press "Enter" keys to choose default value of "Last Cylinder" parameter.\\
4468
press 'w' to write partition table and exit\\
4469
\# losetup -d /dev/loop0}}
4471
\item {\bfseries Make the file system and install grub}\\
4472
{\scriptsize {\tt \# ln -s /dev/loop0 /dev/loop\\
4473
\# losetup /dev/loop0 hd.img\\
4474
\# losetup -o 16384 /dev/loop1 hd.img\\
4475
\# mkfs.ext3 /dev/loop1\\
4476
\# mount /dev/loop1 /mnt\\
4477
\# mkdir -p /mnt/boot/grub\\
4478
\# cp /boot/grub/stage* /boot/grub/e2fs\_stage1\_5 /mnt/boot/grub\\
4481
grub> device (hd0) /dev/loop\\
4482
grub> root (hd0,0)\\
4486
\# losetup -d /dev/loop0\\
4487
\# losetup -d /dev/loop1}}
4489
The {\small {\tt losetup}} option {\small {\tt -o 16384}} skips the partition table in the image file. It is the number of sectors times 512. We need {\small {\tt /dev/loop}} because grub is expecting a disk device \emph{name}, where \emph{name} represents the entire disk and \emph{name1} represents the first partition.
4491
\item {\bfseries Copy the OS files to the image}\\
4492
If you have Xen installed, you can easily use {\small {\tt lomount}} instead of {\small {\tt losetup}} and {\small {\tt mount}} when coping files to some partitions. {\small {\tt lomount}} just needs the partition information.
4494
{\scriptsize {\tt \# lomount -t ext3 -diskimage hd.img -partition 1 /mnt/guest\\
4495
\# cp -ax /var/guestos/\{root,dev,var,etc,usr,bin,sbin,lib\} /mnt/guest\\
4496
\# mkdir /mnt/guest/\{proc,sys,home,tmp\}}}
4498
\item {\bfseries Edit the {\small {\tt /etc/fstab}} of the guest image}\\
4499
The fstab should look like this:
4501
{\scriptsize {\tt \# vim /mnt/guest/etc/fstab\\
4502
/dev/hda1 / ext3 defaults 1 1\\
4503
none /dev/pts devpts gid=5,mode=620 0 0\\
4504
none /dev/shm tmpfs defaults 0 0\\
4505
none /proc proc defaults 0 0\\
4506
none /sys sysfs efaults 0 0}}
4508
\item {\bfseries umount the image file}\\
4509
{\small {\tt \# umount /mnt/guest}}
4512
Now, the guest OS image {\small {\tt hd.img}} is ready. You can also reference {\small {\tt http://free.oszoo.org}} for quickstart images. But make sure to install the boot loader.
4514
\section{HVM Guests}
4515
\subsection{Editing the Xen HVM config file}
4516
Make a copy of the example HVM configuration file {\small {\tt /etc/xen/xmexample.hvm}} and edit the line that reads
4518
{\small {\tt disk = [ 'file:/var/images/\emph{min-el3-i386.img},hda,w' ]}}
4520
replacing \emph{min-el3-i386.img} with the name of the guest OS image file you just made.
4522
\subsection{Creating HVM guests}
4523
Simply follow the usual method of creating the guest, providing the filename of your HVM configuration file:\\
4525
{\small {\tt \# xend start\\
4526
\# xm create /etc/xen/hvmguest.hvm}}
4528
In the default configuration, VNC is on and SDL is off. Therefore VNC windows will open when HVM guests are created. If you want to use SDL to create HVM guests, set {\small {\tt sdl=1}} in your HVM configuration file. You can also turn off VNC by setting {\small {\tt vnc=0}}.
4530
\subsection{Mouse issues, especially under VNC}
4531
Mouse handling when using VNC is a little problematic.
4532
The problem is that the VNC viewer provides a virtual pointer which is
4533
located at an absolute location in the VNC window and only absolute
4534
coordinates are provided.
4535
The HVM device model converts these absolute mouse coordinates
4536
into the relative motion deltas that are expected by the PS/2
4537
mouse driver running in the guest.
4539
it is impossible to keep these generated mouse deltas
4540
accurate enough for the guest cursor to exactly match
4542
This can lead to situations where the guest's cursor
4543
is in the center of the screen and there's no way to
4544
move that cursor to the left
4545
(it can happen that the VNC pointer is at the left
4546
edge of the screen and,
4548
there are no longer any left mouse deltas that
4549
can be provided by the device model emulation code.)
4551
To deal with these mouse issues there are 4 different
4552
mouse emulations available from the HVM device model:
4555
\item[PS/2 mouse over the PS/2 port.]
4556
This is the default mouse
4557
that works perfectly well under SDL.
4558
Under VNC the guest cursor will get
4559
out of sync with the VNC pointer.
4560
When this happens you can re-synchronize
4561
the guest cursor to the VNC pointer by
4567
While these keys are down VNC pointer motions
4568
will not be reported to the guest so
4569
that the VNC pointer can be moved
4570
to a place where it is possible
4571
to move the guest cursor again.
4573
\item[Summagraphics mouse over the serial port.]
4574
The device model also provides emulation
4575
for a Summagraphics tablet,
4576
an absolute pointer device.
4577
This emulation is provided over the second
4580
for Linux guests and
4584
neither Linux nor Windows provides
4585
default support for the Summagraphics
4586
tablet so the guest will have to be
4587
manually configured for this mouse.
4589
\textbf{Linux configuration.}
4592
configure the GPM service to use the Summagraphics tablet.
4593
This can vary between distributions but,
4595
all that needs to be done is modify the file
4596
\path{/etc/sysconfig/mouse} to contain the lines:
4606
and then restart the GPM daemon.
4609
modify the X11 config
4610
\path{/etc/X11/xorg.conf}
4611
to support the Summgraphics tablet by replacing
4612
the input device stanza with the following:
4616
Section "InputDevice"
4619
Option "Device" "/dev/ttyS1"
4620
Option "InputFashion" "Tablet"
4621
Option "Mode" "Absolute"
4622
Option "Name" "EasyPen"
4623
Option "Compatible" "True"
4624
Option "Protocol" "Auto"
4625
Option "SendCoreEvents" "on"
4626
Option "Vendor" "GENIUS"
4631
Restart X and the X cursor should now properly
4632
track the VNC pointer.
4635
\textbf{Windows configuration.}
4638
\path{http://www.cad-plan.de/files/download/tw2k.exe}
4639
and execute that file on the guest,
4640
answering the questions as follows:
4643
\item When the program asks for \textbf{model},
4644
scroll down and select \textbf{SummaSketch (MM Compatible)}.
4646
\item When the program asks for \textbf{COM Port} specify \textbf{com2}.
4648
\item When the programs asks for a \textbf{Cursor Type} specify
4649
\textbf{4 button cursor/puck}.
4651
\item The guest system will then reboot and,
4652
when it comes back up,
4653
the guest cursor will now properly track
4657
\item[PS/2 mouse over USB port.]
4658
This is just the same PS/2 emulation except it is
4659
provided over a USB port.
4660
This emulation is enabled by the configuration flag:
4667
\item[USB tablet over USB port.]
4668
The USB tablet is an absolute pointing device
4669
that has the advantage that it is automatically
4670
supported under Windows guests,
4671
although Linux guests still require some
4672
manual configuration.
4673
This mouse emulation is enabled by the
4681
\textbf{Linux configuration.}
4684
there is no GPM support for the
4685
USB tablet at this point in time.
4686
If you intend to use a GPM pointing
4687
device under VNC you should
4688
configure the guest for Summagraphics
4691
Support for X11 is available by following
4692
the instructions at\\
4693
\verb+http://stz-softwaretechnik.com/~ke/touchscreen/evtouch.html+\\
4694
with one minor change.
4697
given in those instructions
4698
uses the wrong values for the X \& Y minimums and maximums,
4699
use the following config stanza instead:
4703
Section "InputDevice"
4706
Option "Device" "/dev/input/event2"
4707
Option "DeviceName" "touchscreen"
4710
Option "MaxX" "32256"
4711
Option "MaxY" "32256"
4712
Option "ReportingMode" "Raw"
4713
Option "Emulate3Buttons"
4714
Option "Emulate3Timeout" "50"
4715
Option "SendCoreEvents" "On"
4720
\textbf{Windows configuration.}
4722
Just enabling the USB tablet in the
4723
guest's configuration file is sufficient,
4724
Windows will automatically recognize and
4725
configure device drivers for this
4730
\subsection{USB Support}
4731
There is support for an emulated USB mouse,
4732
an emulated USB tablet
4733
and physical low speed USB devices
4734
(support for high speed USB 2.0 devices is
4735
still under development).
4738
\item[USB PS/2 style mouse.]
4739
Details on the USB mouse emulation are
4744
Enabling USB PS/2 style mouse emulation
4745
is just a matter of adding the line
4753
to the configuration file.
4755
Details on the USB tablet emulation are
4760
Enabling USB tablet emulation
4761
is just a matter of adding the line
4769
to the configuration file.
4770
\item[USB physical devices.]
4771
Access to a physical (low speed) USB device
4772
is enabled by adding a line of the form
4776
usbdevice='host:vid:pid'
4780
into the the configuration file.\footnote{
4781
There is an alternate
4782
way of specifying a USB device that
4784
\textbf{host:bus.addr}
4785
but this syntax suffers from
4786
a major problem that makes
4787
it effectively useless.
4788
The problem is that the
4790
portion of this address
4791
changes every time the USB device
4792
is plugged into the system.
4793
For this reason this addressing
4794
scheme is not recommended and
4795
will not be documented further.
4803
that uniquely identify
4805
These ids can be identified
4809
\item Through the control window.
4810
As described in section
4813
is activated by pressing
4815
in the guest VGA window.
4816
As long as USB support is
4817
enabled in the guest by including
4818
the config file line
4824
then executing the command
4830
in the control window
4831
will display a list of all
4832
usb devices and their ids.
4837
Device 1.3, speed 1.5 Mb/s
4838
Class 00: USB device 04b3:310b
4841
was created from a USB mouse with
4846
This device could be made available
4847
to the HVM guest by including the
4851
usbdevice='host:04be:310b'
4855
It is also possible to
4856
enable access to a USB
4857
device dynamically through
4859
The control window command
4862
usb_add host:vid:pid
4865
will also allow access to a
4866
USB device with vendor id
4871
\path{/proc} file system.
4872
The contents of the pseudo file
4873
\path{/proc/bus/usb/devices}
4874
can also be used to identify
4875
vendor and product ids.
4876
Looking at this file,
4877
the line starting with
4881
giving the vendor id and
4884
giving the product id.
4886
\path{/proc/bus/usb/devices}
4887
for the example mouse is as
4891
T: Bus=01 Lev=01 Prnt=01 Port=01 Cnt=02 Dev#= 3 Spd=1.5 MxCh= 0
4892
D: Ver= 2.00 Cls=00(>ifc ) Sub=00 Prot=00 MxPS= 8 #Cfgs= 1
4893
P: Vendor=04b3 ProdID=310b Rev= 1.60
4894
C:* #Ifs= 1 Cfg#= 1 Atr=a0 MxPwr=100mA
4895
I: If#= 0 Alt= 0 #EPs= 1 Cls=03(HID ) Sub=01 Prot=02 Driver=(none)
4896
E: Ad=81(I) Atr=03(Int.) MxPS= 4 Ivl=10ms
4901
line correctly identifies the
4902
vendor id and product id
4906
There is one other issue to
4907
be aware of when accessing a
4908
physical USB device from the guest.
4909
The Dom0 kernel must not have
4910
a device driver loaded for
4911
the device that the guest wishes
4913
This means that the Dom0
4914
kernel must not have that
4915
device driver compiled into
4918
that driver module must
4920
Note that this is the device
4921
specific USB driver that must
4927
USB controller driver must
4930
Going back to the USB mouse
4944
then the USB mouse is being
4945
used by the Dom0 kernel and is
4946
not available to the guest.
4947
Executing the command
4948
\textbf{rmmod usbhid}\footnote{
4951
driver is the significant
4952
one for the USB mouse,
4953
the presence or absence of
4956
has no effect on whether or
4957
not the guest can see a USB mouse.}
4958
will remove the USB mouse
4959
driver from the Dom0 kernel
4960
and the mouse will now be
4961
accessible by the HVM guest.
4963
Be aware the the Linux USB
4964
hotplug system will reload
4965
the drivers if a USB device
4966
is removed and plugged back
4968
This means that just unloading
4969
the driver module might not
4970
be sufficient if the USB device
4971
is removed and added back.
4972
A more reliable technique is
4975
the driver and then rename the
4979
just to make sure it doesn't get
4983
\subsection{Destroy HVM guests}
4984
HVM guests can be destroyed in the same way as can paravirtualized guests. We recommend that you shut-down the guest using the guest OS' provided method, for Linux, type the command
4986
{\small {\tt poweroff}}
4988
in the HVM guest's console, for Windows use Start -> Shutdown first to prevent
4989
data loss. Depending on the configuration the guest will be automatically
4990
destroyed, otherwise execute the command
4992
{\small {\tt xm destroy \emph{vmx\_guest\_id} }}
4994
at the Domain0 console.
4996
\subsection{HVM window (X or VNC) Hot Key}
4997
If you are running in the X environment after creating a HVM guest, an X window is created. There are several hot keys for control of the HVM guest that can be used in the window.
4999
{\bfseries Ctrl+Alt+2} switches from guest VGA window to the control window. Typing {\small {\tt help }} shows the control commands help. For example, 'q' is the command to destroy the HVM guest.\\
5000
{\bfseries Ctrl+Alt+1} switches back to HVM guest's VGA.\\
5001
{\bfseries Ctrl+Alt+3} switches to serial port output. It captures serial output from the HVM guest. It works only if the HVM guest was configured to use the serial port. \\
5003
\chapter{Vnets - Domain Virtual Networking}
5005
Xen optionally supports virtual networking for domains using {\em vnets}.
5006
These emulate private LANs that domains can use. Domains on the same
5007
vnet can be hosted on the same machine or on separate machines, and the
5008
vnets remain connected if domains are migrated. Ethernet traffic
5009
on a vnet is tunneled inside IP packets on the physical network. A vnet is a virtual
5010
network and addressing within it need have no relation to addressing on
5011
the underlying physical network. Separate vnets, or vnets and the physical network,
5012
can be connected using domains with more than one network interface and
5013
enabling IP forwarding or bridging in the usual way.
5015
Vnet support is included in \texttt{xm} and \xend:
5017
# xm vnet-create <config>
5019
creates a vnet using the configuration in the file \verb|<config>|.
5020
When a vnet is created its configuration is stored by \xend and the vnet persists until it is
5023
# xm vnet-delete <vnetid>
5025
The vnets \xend knows about are listed by
5029
More vnet management commands are available using the
5030
\texttt{vn} tool included in the vnet distribution.
5032
The format of a vnet configuration file is
5036
(vnetif <vnet interface>)
5039
White space is not significant. The parameters are:
5041
\item \verb|<vnetid>|: vnet id, the 128-bit vnet identifier. This can be given
5042
as 8 4-digit hex numbers separated by colons, or in short form as a single 4-digit hex number.
5043
The short form is the same as the long form with the first 7 fields zero.
5044
Vnet ids must be non-zero and id 1 is reserved.
5046
\item \verb|<bridge>|: the name of a bridge interface to create for the vnet. Domains
5047
are connected to the vnet by connecting their virtual interfaces to the bridge.
5048
Bridge names are limited to 14 characters by the kernel.
5050
\item \verb|<vnetif>|: the name of the virtual interface onto the vnet (optional). The
5051
interface encapsulates and decapsulates vnet traffic for the network and is attached
5052
to the vnet bridge. Interface names are limited to 14 characters by the kernel.
5054
\item \verb|<level>|: security level for the vnet (optional). The level may be one of
5056
\item \verb|none|: no security (default). Vnet traffic is in clear on the network.
5057
\item \verb|auth|: authentication. Vnet traffic is authenticated using IPSEC
5059
\item \verb|conf|: confidentiality. Vnet traffic is authenticated and encrypted
5060
using IPSEC ESP with hmac96 and AES-128.
5062
Authentication and confidentiality are experimental and use hard-wired keys at present.
5064
When a vnet is created its configuration is stored by \xend and the vnet persists until it is
5065
deleted using \texttt{xm vnet-delete <vnetid>}. The interfaces and bridges used by vnets
5066
are visible in the output of \texttt{ifconfig} and \texttt{brctl show}.
5069
If the file \path{vnet97.sxp} contains
5071
(vnet (id 97) (bridge vnet97) (vnetif vnif97)
5074
Then \texttt{xm vnet-create vnet97.sxp} will define a vnet with id 97 and no security.
5075
The bridge for the vnet is called vnet97 and the virtual interface for it is vnif97.
5076
To add an interface on a domain to this vnet set its bridge to vnet97
5077
in its configuration. In Python:
5083
(dev (vif (mac aa:00:00:01:02:03) (bridge vnet97)))
5085
Once the domain is started you should see its interface in the output of \texttt{brctl show}
5086
under the ports for \texttt{vnet97}.
5088
To get best performance it is a good idea to reduce the MTU of a domain's interface
5089
onto a vnet to 1400. For example using \texttt{ifconfig eth0 mtu 1400} or putting
5090
\texttt{MTU=1400} in \texttt{ifcfg-eth0}.
5091
You may also have to change or remove cached config files for eth0 under
5092
\texttt{/etc/sysconfig/networking}. Vnets work anyway, but performance can be reduced
5093
by IP fragmentation caused by the vnet encapsulation exceeding the hardware MTU.
5095
\section{Installing vnet support}
5096
Vnets are implemented using a kernel module, which needs to be loaded before
5097
they can be used. You can either do this manually before starting \xend, using the
5098
command \texttt{vn insmod}, or configure \xend to use the \path{network-vnet}
5099
script in the xend configuration file \texttt{/etc/xend/xend-config.sxp}:
5101
(network-script network-vnet)
5103
This script insmods the module and calls the \path{network-bridge} script.
5105
The vnet code is not compiled and installed by default.
5106
To compile the code and install on the current system
5107
use \texttt{make install} in the root of the vnet source tree,
5108
\path{tools/vnet}. It is also possible to install to an installation
5109
directory using \texttt{make dist}. See the \path{Makefile} in
5110
the source for details.
5112
The vnet module creates vnet interfaces \texttt{vnif0002},
5113
\texttt{vnif0003} and \texttt{vnif0004} by default. You can test that
5114
vnets are working by configuring IP addresses on these interfaces
5115
and trying to ping them across the network. For example, using machines
5118
hostA# ifconfig vnif0004 192.0.2.100 up
5119
hostB# ifconfig vnif0004 192.0.2.101 up
5120
hostB# ping 192.0.2.100
5123
The vnet implementation uses IP multicast to discover vnet interfaces, so
5124
all machines hosting vnets must be reachable by multicast. Network switches
5125
are often configured not to forward multicast packets, so this often
5126
means that all machines using a vnet must be on the same LAN segment,
5127
unless you configure vnet forwarding.
5129
You can test multicast coverage by pinging the vnet multicast address:
5131
# ping -b 224.10.0.1
5133
You should see replies from all machines with the vnet module running.
5134
You can see if vnet packets are being sent or received by dumping traffic
5135
on the vnet UDP port:
5137
# tcpdump udp port 1798
5140
If multicast is not being forwarded between machines you can configure
5141
multicast forwarding using vn. Suppose we have machines hostA on 192.0.2.200
5142
and hostB on 192.0.2.211 and that multicast is not forwarded between them.
5143
We use vn to configure each machine to forward to the other:
5145
hostA# vn peer-add hostB
5146
hostB# vn peer-add hostA
5148
Multicast forwarding needs to be used carefully - you must avoid creating forwarding
5149
loops. Typically only one machine on a subnet needs to be configured to forward,
5150
as it will forward multicasts received from other machines on the subnet.
5152
%% Chapter Glossary of Terms moved to glossary.tex
5153
\chapter{Glossary of Terms}
5157
\item[Domain] A domain is the execution context that contains a
5158
running {\bf virtual machine}. The relationship between virtual
5159
machines and domains on Xen is similar to that between programs and
5160
processes in an operating system: a virtual machine is a persistent
5161
entity that resides on disk (somewhat like a program). When it is
5162
loaded for execution, it runs in a domain. Each domain has a {\bf
5165
\item[Domain 0] The first domain to be started on a Xen machine.
5166
Domain 0 is responsible for managing the system.
5168
\item[Domain ID] A unique identifier for a {\bf domain}, analogous to
5169
a process ID in an operating system.
5171
\item[Full virtualization] An approach to virtualization which
5172
requires no modifications to the hosted operating system, providing
5173
the illusion of a complete system of real hardware devices.
5175
\item[Hypervisor] An alternative term for {\bf VMM}, used because it
5176
means `beyond supervisor', since it is responsible for managing
5177
multiple `supervisor' kernels.
5179
\item[Live migration] A technique for moving a running virtual machine
5180
to another physical host, without stopping it or the services
5183
\item[Paravirtualization] An approach to virtualization which requires
5184
modifications to the operating system in order to run in a virtual
5185
machine. Xen uses paravirtualization but preserves binary
5186
compatibility for user space applications.
5188
\item[Shadow pagetables] A technique for hiding the layout of machine
5189
memory from a virtual machine's operating system. Used in some {\bf
5190
VMMs} to provide the illusion of contiguous physical memory, in
5191
Xen this is used during {\bf live migration}.
5193
\item[Virtual Block Device] Persistent storage available to a virtual
5194
machine, providing the abstraction of an actual block storage device.
5195
{\bf VBD}s may be actual block devices, filesystem images, or
5196
remote/network storage.
5198
\item[Virtual Machine] The environment in which a hosted operating
5199
system runs, providing the abstraction of a dedicated machine. A
5200
virtual machine may be identical to the underlying hardware (as in
5201
{\bf full virtualization}, or it may differ, as in {\bf
5202
paravirtualization}).
5204
\item[VMM] Virtual Machine Monitor - the software that allows multiple
5205
virtual machines to be multiplexed on a single physical machine.
5207
\item[Xen] Xen is a paravirtualizing virtual machine monitor,
5208
developed primarily by the Systems Research Group at the University
5209
of Cambridge Computer Laboratory.
5211
\item[XenLinux] A name for the port of the Linux kernel that
5220
%% Other stuff without a home
5222
%% Instructions Re Python API
5224
%% Other Control Tasks using Python
5225
%% ================================
5227
%% A Python module 'Xc' is installed as part of the tools-install
5228
%% process. This can be imported, and an 'xc object' instantiated, to
5229
%% provide access to privileged command operations:
5234
%% # help(xc.domain_create)
5236
%% In this way you can see that the class 'xc' contains useful
5237
%% documentation for you to consult.
5239
%% A further package of useful routines (xenctl) is also installed:
5241
%% # import xenctl.utils
5242
%% # help(xenctl.utils)
5244
%% You can use these modules to write your own custom scripts or you
5245
%% can customise the scripts supplied in the Xen distribution.
5249
% Explain about AGP GART
5252
%% If you're not intending to configure the new domain with an IP
5253
%% address on your LAN, then you'll probably want to use NAT. The
5254
%% 'xen_nat_enable' installs a few useful iptables rules into domain0
5255
%% to enable NAT. [NB: We plan to support RSIP in future]
5259
%% Installing the file systems from the CD
5260
%% =======================================
5262
%% If you haven't got an existing Linux installation onto which you
5263
%% can just drop down the Xen and Xenlinux images, then the file
5264
%% systems on the CD provide a quick way of doing an install. However,
5265
%% you would be better off in the long run doing a proper install of
5266
%% your preferred distro and installing Xen onto that, rather than
5267
%% just doing the hack described below:
5269
%% Choose one or two partitions, depending on whether you want a
5270
%% separate /usr or not. Make file systems on it/them e.g.:
5271
%% mkfs -t ext3 /dev/hda3
5272
%% [or mkfs -t ext2 /dev/hda3 && tune2fs -j /dev/hda3 if using an old
5275
%% Next, mount the file system(s) e.g.:
5276
%% mkdir /mnt/root && mount /dev/hda3 /mnt/root
5277
%% [mkdir /mnt/usr && mount /dev/hda4 /mnt/usr]
5279
%% To install the root file system, simply untar /usr/XenDemoCD/root.tar.gz:
5280
%% cd /mnt/root && tar -zxpf /usr/XenDemoCD/root.tar.gz
5282
%% You'll need to edit /mnt/root/etc/fstab to reflect your file system
5283
%% configuration. Changing the password file (etc/shadow) is probably a
5286
%% To install the usr file system, copy the file system from CD on
5287
%% /usr, though leaving out the "XenDemoCD" and "boot" directories:
5288
%% cd /usr && cp -a X11R6 etc java libexec root src bin dict kerberos
5289
%% local sbin tmp doc include lib man share /mnt/usr
5291
%% If you intend to boot off these file systems (i.e. use them for
5292
%% domain 0), then you probably want to copy the /usr/boot
5293
%% directory on the cd over the top of the current symlink to /boot
5294
%% on your root filesystem (after deleting the current symlink)
5296
%% cd /mnt/root ; rm boot ; cp -a /usr/boot .