Q-1: I can't get x11vnc to start up. It says "XOpenDisplay failed (null)" or "Xlib: connection to ":0.0" refused by server Xlib: No protocol specified" and then exits. What do I need to do?
For the former error, you need to specify the X display to connect to (it also needs to be on the same machine the x11vnc process is to run on.) Set your DISPLAY environment variable (or use the -display option) to specify it. Nearly always the correct value will be ":0" (in fact, x11vnc will now assume :0 if given no other information.)
For the latter error, you need to set up the X11 permissions correctly.
To make sure X11 permissions are the problem do this simple test: while sitting at the physical X display open a terminal window (gnome-terminal, xterm, etc.) You should be able to run x11vnc successfully without any need for special steps or command line options in that terminal (i.e. just type "x11vnc".) If that works OK then you know X11 permissions are the only thing preventing it from working when you try to start x11vnc via, say, a remote shell.
How to Solve: See the xauth(1), Xsecurity(7), and xhost(1) man pages or this Howto for much info on X11 permissions. For example, you may need to set your XAUTHORITY environment variable (or use the -auth option) to point to the correct MIT-MAGIC-COOKIE file (e.g. /home/joe/.Xauthority or /var/gdm/:0.Xauth or /var/lib/kdm/A:0-crWk72K or /tmp/.gdmzndVlR, etc, etc.), or simply be sure you run x11vnc as the correct user (i.e. the user who is logged into the X session you wish to view.)
Note: The MIT cookie file contains the secret key that allows x11vnc to connect to the desired X display.
If, say, sshd has set XAUTHORITY to point to a random file it has created for X forwarding that will cause problems. (Under some circumstances even su(1) and telnet(1) can set XAUTHORITY. See also the gdm parameter NeverPlaceCookiesOnNFS that sets XAUTHORITY to a random filename in /tmp for the whole X session.)
Running x11vnc as root is often not enough: you need to know where the MIT-MAGIC-COOKIE file for the desired X display is.
Example solution:
x11vnc -display :0 -auth /var/gdm/:0.Xauth
(this is for the display manager gdm and requires root permission to read the gdm cookie file, see this faq for other display manager cookie file names.)
Note as of Feb/2007 you can also try the -find option instead of "-display ..." and see if that finds your display and Xauthority.
Less safe, but to avoid figuring out where the correct XAUTHORITY file is, if the person sitting at the physical X session types "xhost +localhost" then one should be able to attach x11vnc to the session (from the same machine.) The person could then type "xhost -localhost" after x11vnc has connected to go back to the default permissions. Also, for some situations the "-users lurk=" option may soon be of use (please read the documentation on the -users option.)
To test out your X11 permissions from a remote shell, set DISPLAY and possibly XAUTHORITY (see your shell's man page, bash(1), tcsh(1), on how to set environment variables) and type xdpyinfo in the same place you will be typing (or otherwise running) x11vnc. If information is printed out about the X display (screen sizes, supported extensions, color visuals info) that means the X11 permissions are set up properly: xdpyinfo successfully connected to DISPLAY! You could also type xclock and make sure no errors are reported (a clock should appear on the X display, press Ctrl-C to stop it.) If these work, then typing "x11vnc" in the same environment should also work.
Important: if you cannot get your X11 permissions so that the xdpyinfo or xclock tests work, x11vnc also will not work (all of these X clients must be allowed to connect to the X server to function properly.)
Firewalls: Speaking of permissions, it should go without saying that the host-level firewall will need to be configured to allow connections in on a port. E.g. 5900 (default VNC port) or 22 (default SSH port for tunnelling VNC.) Most systems these days have firewalls turned on by default, so you will actively have to do something to poke a hole in the firewall at the desired port number. See your system administration tool for Firewall settings (Yast, Firestarter, etc.)
Make sure you have gcc (or other C compiler) and all of the required libraries and the corresponding -dev/-devel packages installed. These include Xorg/XFree86, libX11, libjpeg, libz, libssl, ... and don't forget the devs: libjpeg-dev, libssl-dev ...
The most common build problem that people encounter is that the necessary X11 libraries are installed on their system however it does not have the corresponding -dev/-devel packages installed. These dev packages include C header files and build-time .so symlink. It is a shame the current trend in distros is to not install the dev package by default when the the library runtime package is installed... (it diminishes the power of open source)
As of Nov/2006 here is a list of libraries that x11vnc usually likes to use:
libc.so libX11.so libXtst.so libXext.so
libXfixes.so libXdamage.so libXinerama.so libXrandr.so
libz.so libjpeg.so libpthread.so
libssl.so libcrypto.so libcrypt.so
although x11vnc will be pretty usable with the subset: libc.so, libX11.so, libXtst.so, libXext.so, libz.so, and libjpeg.so.
After running the libvncserver configure, carefully examine the output and the messages in the config.log file looking for missing components. For example, if the configure output looks like:
checking how to run the C preprocessor... gcc -E
checking for X... no
checking for XkbSelectEvents in -lX11... no
checking for XineramaQueryScreens in -lXinerama... no
checking for XTestFakeKeyEvent in -lXtst... no
or even worse:
checking for C compiler default output file name... configure: error:
C compiler cannot create executables
See `config.log' for more details.
there is quite a bit wrong with the build environment. Hopefully simply adding -dev packages and/or gcc or make will fix it.
For Debian the list seems to be:
gcc
make
libc6-dev
libjpeg8-dev (formerly libjpeg62-dev)
libx11-dev
x11proto-core-dev (formerly x-dev)
libxext-dev
libxtst-dev
libxdamage-dev
libxfixes-dev
libxrandr-dev
libxinerama-dev
libxss-dev (formerly xlibs-static-dev)
zlib1g-dev
libssl-dev
libavahi-client-dev
linux-libc-dev (only needed for linux console rawfb support)
Note that depending on your OS version the above names may have been changed and/or additional packages may be needed.
For Redhat the list seems to be:
gcc
make
glibc-devel
libjpeg-devel
libX11-devel
xorg-x11-proto-devel
libXdamage-devel
libXfixes-devel
libXrandr-devel
zlib-devel
openssl-devel
avahi-devel
kernel-headers (only needed for linux console rawfb support)
For other distros or OS's the package names may not be the same but will look similar. Also, distros tend to rename packages as well so the above list may be out of date. So only use the above lists as hints for the package names that are needed.
Have a look at Misc. Build Problems for additional fixes.
Note: there is growing trend in Linux and other distros to slice up core X11 software into more and smaller packages. So be prepared for more headaches compiling software...
Q-3: I just built x11vnc successfully, but when I use it my keystrokes and mouse button clicks are ignored (I am able to move the mouse though.)
This is most likely due to you not having a working build environment for the XTEST client library libXtst.so. The library is probably present on your system, but the package installing the build header file is missing.
If you were watching carefully while configure was running you would have seen: checking for XTestFakeKeyEvent in -lXtst... no
The solution is to add the necessary build environment package (and the library package if that is missing too.) On Debian the build package is libxtst-dev. Other distros/OS's may have it in another package.
x11vnc will build without support for this library (e.g. perhaps one wants a view-only x11vnc on a stripped down or embedded system...) And at runtime it will also continue to run even if the X server it connects to does not support XTEST. In both cases it cannot inject keystrokes or button clicks since XTEST is needed for that (it can still move the mouse pointer using the X API XWarpPointer().)
You will see a warning message something like this at run time:
20/03/2005 22:33:09 WARNING: XTEST extension not available (either missing from
20/03/2005 22:33:09 display or client library libXtst missing at build time.)
20/03/2005 22:33:09 MOST user input (pointer and keyboard) will be DISCARDED.
20/03/2005 22:33:09 If display does have XTEST, be sure to build x11vnc with
20/03/2005 22:33:09 a working libXtst build environment (e.g. libxtst-dev,
20/03/2005 22:33:09 or other packages.)
20/03/2005 22:33:09 No XTEST extension, switching to -xwarppointer mode for
20/03/2005 22:33:09 pointer motion input.
Also, as of Nov/2006 there will be a configure build time warning as well:
...
checking for XFixesGetCursorImage in -lXfixes... yes
checking for XDamageQueryExtension in -lXdamage... yes
configure: WARNING:
==========================================================================
A working build environment for the XTEST extension was not found (libXtst).
An x11vnc built this way will be only barely usable. You will be able to
move the mouse but not click or type. There can also be deadlocks if an
application grabs the X server.
It is recommended that you install the necessary development packages
for XTEST (perhaps it is named something like libxtst-dev) and run
configure again.
==========================================================================
We apologize that x11vnc does not build cleanly on older versions of Solaris, Linux, etc.: very few users are on these old releases.
We have heard that since Dec/2004 a Solaris 2.6 built x11vnc will run on Solaris Solaris 2.5 and 2.5.1 (since a workaround for XConvertCase is provided.)
In any event, here is a workaround for Solaris 2.5.1 (and perhaps earlier and perhaps non-Solaris):
First use the environment settings (CPPFLAGS, LDFLAGS, etc.) in the above Solaris build script to run the configure command. That should succeed without failure. Then you have to hand edit the autogenerated rfb/rfbconfig.h file in the source tree, and just before the last #endif at the bottom of that file insert these workaround lines:
struct timeval _tmp_usleep_tv;
#define usleep(x) \
_tmp_usleep_tv.tv_sec = (x) / 1000000; \
_tmp_usleep_tv.tv_usec = (x) % 1000000; \
select(0, NULL, NULL, NULL, &_tmp_usleep_tv);
int gethostname(char *name, int namelen);
long random();
int srandom(unsigned int seed);
#undef LIBVNCSERVER_HAVE_LIBPTHREAD
#define SHUT_RDWR 2
typedef unsigned int in_addr_t;
#define snprintf(a, n, args...) sprintf((a), ## args)
Then run make with the Solaris build script environment, everything should compile without problems, and the resulting x11vnc binary should work OK. If some non-x11vnc related programs fail (e.g. test programs) and the x11vnc binary is not created try "make -k" to have it keep going. Similar sorts of kludges in rfb/rfbconfig.h can be done on other older OS (Solaris, Linux, ...) releases.
Here are some notes for similar steps that need to be done to build on SunOS 4.x
Please let us know if you had to use the above workaround (and whether it worked or not.) If there is enough demand we will try to push clean compilations back to earlier Solaris, Linux, etc, releases.
Hopefully the build steps above and FAQ provide enough info for a painless compile for most environments. Please report problems with the x11vnc configure, make, etc. on your system (if your system is known to compile other GNU packages successfully.)
There are precompiled x11vnc binaries built by other groups that are available at the following locations:
- Slackware: (.tgz) http://www.linuxpackages.net/
- SuSE: (.rpm) http:/software.opensuse.org/
- Gentoo: (info) http://gentoo-wiki.com/ and http://gentoo-portage.com/
- FreeBSD: (.tbz) http://www.freebsd.org/ http://www.freshports.org/net/x11vnc
- NetBSD: (src) http://pkgsrc.se/x11/x11vnc
- OpenBSD: (.tgz) http://openports.se/
- Arch Linux: (.tgz) http://www.archlinux.org/
- Nokia 770 (.deb) http://mike.saunby.googlepages.com/x11vncfornokia7702
- Sharp Zaurus: http://www.focv.com/
- Debian: (.deb) http://packages.debian.org/x11vnc
- Redhat/Fedora: (.rpm) http://packages.sw.be/x11vnc RPMforge http://dag.wieers.com/rpm/packages/x11vnc/ (N.B.: unmaintained after 0.9.3)
- Solaris: (pkg) http://www.sunfreeware.com/
If the above binaries don't work and building x11vnc on your OS fails (and all else fails!) you can try one of My Collection of x11vnc Binaries for various OS's and x11vnc releases.
As a general note, the x11vnc program is simple enough you don't really need to install a package: the binary will in most cases work as is and from any location (as long as your system libraries are not too old, etc.) So, for Linux distributions that are not one of the above, the x11vnc binary from the above packages has a good chance of working. You can "install" it by just copying the x11vnc binary to the desired directory in your PATH. Tip on extracting files from a Debian package: extract the archive via a command like: "ar x x11vnc_0.6-2_i386.deb" and then you can find the binary in the resulting data.tar.gz tar file. Also, rpm2cpio(1) is useful in extracting files from rpm packages.
If you use a standalone binary like this and also want x11vnc to serve up the Java VNC Viewer jar file (either SSL enabled or regular one), then you will need to extract the classes subdirectory from the source tarball and point x11vnc to it via the -httpdir option. E.g.: x11vnc -httpdir /path/to/x11vnc-0.9.9/classes/ssl ...
Alternatively, you can also go for a more up-to-date approach and use the very decent noVNC viewer (https://kanaka.github.io/noVNC/) that is purely HTML5 and does not need any plugins at all.
Q-6: Where can I get a VNC Viewer binary (or source code) for the Operating System I will be viewing from?
To obtain VNC viewers for the viewing side (Windows, Mac OS, or Unix) try here:
- http://www.tightvnc.com/download.html
- http://www.realvnc.com/download-free.html
- http://sourceforge.net/projects/cotvnc/
- http://www.ultravnc.com/
- Our Enhanced TightVNC Viewer (SSVNC)[ssvnc.gif]
Run: x11vnc -opts to list just the option names or run: x11vnc -help for long descriptions about each option. The output is listed here as well. Yes, x11vnc does have a lot of options, doesn't it...
Q-8: I don't like typing arcane command line options every time I start x11vnc. What can I do? Is there a config file? Or a GUI?
You could create a shell script that calls x11vnc with your options:
#!/bin/sh
#
# filename: X11vnc (i.e. not "x11vnc")
# It resides in a directory in $PATH. "chmod 755 X11vnc" has been run on it.
#
x11vnc -wait 50 -localhost -rfbauth $HOME/.vnc/passwd -display :0 $*
a similar thing can be done via aliases in your shell (bash, tcsh, csh, etc..)
Or as of Jun/2004 you can use the simple $HOME/.x11vncrc config file support. If that file exists, each line is taken as a command line option. E.g. the above would be:
# this is a comment in my ~/.x11vncrc file
wait 50 # this is a comment to the end of the line.
-localhost # note: the leading "-" is optional.
rfbauth /home/fred/.vnc/passwd
display :0
As of Dec/2004 there is now a simple Tcl/Tk GUI based on the remote-control functionality ("-R") that was added. The /usr/bin/wish program is needed for operation. The gui is not particularly user-friendly, it just provides a point and click mode to set all the many x11vnc parameters and obtain help on them. It is also very useful for testing. See the -gui option for more info. Examples: "x11vnc ... -gui" and "x11vnc ... -gui other:0" in the latter case the gui is displayed on other:0, not the X display x11vnc is polling. There is also a "-gui tray" system tray mode.
[tkx11vnc.gif]
NOTE: You may need to install the "wish" or "tk" or "tk8.4" package for the gui mode to work (the package name depends on your OS/distro.) The tcl/tk "wish" interpreter is used. In debian (and so ubuntu too) one would run "apt-get install tk" or perhaps "apt-get install tk8.4"
As of Jul/2005 the gui can run in a more friendly small icon mode "-gui icon" or in the system tray: "-gui tray". It has balloon status, a simple menu, and a Properities dialog. The full, complicated, gui is only available under "Advanced". Other improvements were added as well. Try "Misc -> simple_gui" for a gui with fewer esoteric menu items.
If the gui fails to embed itself in the system tray, do a retry via "Window View -> icon" followed by "Window View -> tray" with the popup menu.
For inexperienced users starting up x11vnc and the GUI while sitting at the physical X display (not remotely), using something like "x11vnc -display :0 -gui tray=setpass" might be something for them that they are accustomed to in a Desktop environment (it prompts for an initial password, etc.) This is a basic "Share My Desktop" usage mode.
As of Nov/2008 in x11vnc 0.9.6 there is a desktop menu item (x11vnc.desktop) that runs this command:
x11vnc -gui tray=setpass -rfbport PROMPT -logfile %HOME/.x11vnc.log.%VNCDISPLAY
which also prompts for which VNC port to use and a couple other parameters.
Use something like, e.g., "x11vnc -rfbport 5901" to force it to use port 5901 (this is VNC display :1.) If something else is using that port x11vnc will exit immediately. If you do not supply the -rfbport option, it will autoprobe starting at 5900 and work its way up to 5999 looking for a free port to listen on. In that case, watch for the PORT=59xx line to see which port it found, then subtract 5900 from it for the VNC display number to enter into the VNC Viewer(s).
The "-N" option will try to match the VNC display number to the X display (e.g. X11 DISPLAY of :5 (port 6005) will have VNC display :5 (port 5905).)
Also see the "-autoport n" option to indicated at which value the auto probing should start at.
See the Firewalls/Routers discussion.
Q-12: Is it possible for a VNC Viewer and a VNC Server to connect to each other even though both are behind Firewalls that block all incoming connections?
This is very difficult or impossible to do unless a third machine, reachable by both, is used as a relay. So we assume a third machine is somehow being used as a relay.
(Update: It may be possible to do "NAT-2-NAT" without a relay machine by using a UDP tunnel such as http://samy.pl/pwnat/. All that is required is that both NAT firewalls allow in UDP packets from an IP address to which a UDP packet has recently been sent to. If you try it out let us know how it went.)
In the following discussion, we will suppose port 5950 is being used on the relay machine as the VNC port for the rendezvous.
A way to rendezvous is to have the VNC Server start a reverse connection to the relay machine: x11vnc -connect third-machine.net:5950 ...
and the VNC viewer forward connects as usual: vncviewer third-machine.net:50
Or maybe two ports would be involved, e.g. the viewer goes to display :51 (5951.) It depends on the relay software being used.
What software to run on third-machine? A TCP relay of some sort could be used... Try a google search on "tcp relay" or "ip relay". However, note that this isn't a simple redirection because it hooks up two incoming connections. You can look at our UltraVNC repeater implementation ultravnc_repeater.pl for ideas and possibly to customize.
Also, if you are not the admin of third-machine you'd have to convince the owner to allow you to install this software (and he would likely need to open his server's firewall to allow the port through.)
It is recommended that SSL is used for encryption (e.g. "-ssl SAVE") when going over the internet.
We have a prototype for performing a rendezvous via a Web Server acting as the relay machine. Download the vncxfer CGI script and see the instructions at the top.
Once that CGI script is set up on the website, both users go to, say, http://somesite.com/vncxfer (or maybe a "/cgi-bin" directory or ".cgi" suffix must be used.) Previously, both have agreed on the same session name (say by phone or email) , e.g. "5cows", and put that into the entry form on the vncxfer starting page (hopefully separated by a few seconds, so the relay helper can fully start up at the first request.)
The page returned tells them the hostname and port number and possible command to use for forward (VNC Viewer) and reverse (VNC Server, i.e. x11vnc) connections as described above.
Also since Oct/2007, x11vnc can connect directly (no web browser), like this:
x11vnc ... -connect localhost:0 -proxy 'http://somesite.com/vncxfer?session=5cows&'
Unfortunately the prototype requires that the Web server's firewall allow in the port (e.g. 5950) used for the rendezvous. Most web servers are not configured to do this, so you would need to ask the admin to do this for you. Nearly all free webspace sites, e.g. www.zendurl.com, will not allow your CGI script to be an open relay like this. (If you find one that does allow this, let me know!)
Maybe someday a clever trick will be thought up to relax the listening port requirement (e.g. use HTTP/CGI itself for the transfer... it is difficult to emulate a full-duplex TCP connection with them.)
See also the Firewalls/Routers discussion and Reverse Connection Proxy discussion.
SSH method: If both users (i.e. one on Viewer-side and the other on x11vnc server side) have SSH access to a common machine on the internet (or otherwise mutually reachable), then SSH plumbing can be used to solve this problem. The users create SSH tunnels going through the SSH login machine.
Instead of assuming port 5900 is free on the SSH machine, we will assume both users agreed to use 5933. This will illustrate how to use a different port for the redir. It could be any port, what matters is that both parties refer to the same one.
Set up the Tunnel from the VNC Server side:
ssh -t -R 5933:localhost:5900 [email protected]
Set up the Tunnel from the VNC Viewer side:
ssh -t -L 5900:localhost:5933 [email protected]
Run Server on the VNC Server side:
x11vnc -rfbport 5900 -localhost ...
Run Viewer on the VNC Viewer side:
vncviewer -encodings "copyrect tight zrle hextile" localhost:0
(we assume the old-style -encodings option needs to be used. See here for details.)
If the SSH machine has been configured (see sshd_config(5)) with the option GatewayPorts=yes, then the tunnel set up by the VNC Server will be reachable directly by the VNC viewer (as long as the SSH machine's firewall does not block the port, 5933 in this example.) So in that case the Viewer side does not need to run any ssh command, but rather only runs: vncviewer third-machine.net:33
In this case we recommend SSL be used for encryption.
The creation of both tunnels can be automated. As of Oct/2007 the -ssh x11vnc option is available and so only this command needs to be run on the VNC Server side: x11vnc -ssh [email protected]:33 ...
(the SSH passphrase may need to be supplied.)
To automate on the VNC Viewer side, the user can use the Enhanced TightVNC Viewer (SSVNC) by:
- Clicking on 'Use SSH'
- Entering [email protected]:33 into 'VNC Host:Display' entry box
- Clicking on 'Connect'
As above, if the SSH GatewayPorts=yes setting is configured the Viewer side doesn't need to create a SSH tunnel. In SSVNC the Viewer user could instead select 'Use SSL' and then, e.g., on the Server side supply "-ssl SAVE" to x11vnc. Then end-to-end SSL encryption would be used (in addition to the SSH encryption on the Server-side leg.)
Use the -q and -bg options, respectively. (also: -quiet is an alias for -q)
Note that under -bg the stderr messages will be lost unless you use the "-o logfile" option.
Q-14: Sometimes when a VNC viewer dies abruptly, x11vnc also dies with the error message like: "Broken pipe". I'm using the -forever mode and I want x11vnc to keep running.
As of Jan/2004 the SIGPIPE signal is ignored. So if a viewer client terminates abruptly, libvncserver will notice on the next I/O operation and will close the connection and continue on.
Up until of Apr/2004 the above fix only works for BSD signal systems (Linux, FreeBSD, ...) For SYSV systems there is a workaround in place since about Jun/2004.
This appears to be fixed in x11vnc version 0.9 and later. If you need to use an earlier version of x11vnc, try using the "-rfbversion 3.7" option. In general sometimes one can get a misbehaving viewer to work by supplying rfb versions 3.7 or 3.3.
This has been fixed in x11vnc version 0.8.4. More info here, here, and here.
This is a bug in the X server obviously; an X client should never be able to crash it.
The problem seems to be with the RECORD X extension and so a workaround is to use the "-noxrecord" x11vnc command line option.
One user reports when running x11vnc on AIX 5.3 in his CDE session after a few minutes or seconds x11vnc will "freeze" (no more updates being sent, etc.) The freezing appeared to be worse for versions later than 0.9.2.
The problem seems to be with the RECORD X extension on AIX and so a workaround is to use the "-noxrecord" x11vnc command line option. The user found no freezes occurred when using that option.
Q-19: Are there any build-time customizations possible, e.g. change defaults, create a smaller binary, etc?
There are some options. They are enabled by adding something like -Dxxxx=1 to the CPPFLAGS environment variable before running configure (see the build notes for general background.)
/*
* Mar/2006
* Build-time customization via CPPFLAGS.
*
* Summary of options to include in CPPFLAGS for custom builds:
*
* -DVNCSHARED to have the vnc display shared by default.
* -DFOREVER to have -forever on by default.
* -DNOREPEAT=0 to have -repeat on by default.
* -DADDKEYSYMS=0 to have -noadd_keysyms the default.
*
* -DREMOTE_DEFAULT=0 to disable remote-control on by default (-yesremote.)
* -DREMOTE_CONTROL=0 to disable remote-control mechanism completely.
* -DEXTERNAL_COMMANDS=0 to disable the running of all external commands.
* -DFILEXFER=0 disable filexfer.
*
* -DHARDWIRE_PASSWD=... hardwired passwords, quoting necessary.
* -DHARDWIRE_VIEWPASSWD=...
* -DNOPW=1 make -nopw the default (skip warning)
* -DUSEPW=1 make -usepw the default
* -DPASSWD_REQUIRED=1 exit unless a password is supplied.
* -DPASSWD_UNLESS_NOPW=1 exit unless a password is supplied and no -nopw.
*
* -DWIREFRAME=0 to have -nowireframe as the default.
* -DWIREFRAME_COPYRECT=0 to have -nowirecopyrect as the default.
* -DWIREFRAME_PARMS=... set default -wirecopyrect parameters.
* -DSCROLL_COPYRECT=0 to have -noscrollcopyrect as the default.
* -DSCROLL_COPYRECT_PARMS=... set default -scrollcopyrect parameters.
* -DSCALING_COPYRECT=0
* -DXDAMAGE=0 to have -noxdamage as the default.
* -DSKIPDUPS=0 to have -noskip_dups as the default or vice versa.
*
* -DPOINTER_MODE_DEFAULT={0,1,2,3,4} set default -pointer_mode.
* -DBOLDLY_CLOSE_DISPLAY=0 to not close X DISPLAY under -rawfb.
* -DSMALL_FOOTPRINT=1 for smaller binary size (no help, no gui, etc)
* use 2 or 3 for even smaller footprint.
* -DNOGUI do not include the gui tkx11vnc.
* -DPOLL_8TO24_DELAY=N
* -DDEBUG_XEVENTS=1 enable printout for X events.
*
* Set these in CPPFLAGS before running configure. E.g.:
*
* % env CPPFLAGS="-DFOREVER -DREMOTE_CONTROL=0" ./configure
* % make
*/
If other things (e.g. "-I ...") are needed in CPPFLAGS add them as well.
On some systems is seems you need to set LC_ALL=C for configure to work properly...
Be careful the following two variables: HARDWIRE_PASSWD and HARDWIRE_VIEWPASSWD. If set (remember to include the double quotes around the string), they will be used as default values for the -passwd and -viewpasswd options. Of course the strings will exist unobscured in the x11vnc binary: it better not be readable by unintendeds. Perhaps this is of use in remote access for an embedded application, etc...
Let us know if more build-time customizations would be useful.
Q-20: I have two separate machine displays in front of me, one Windows the other X11: can I use x11vnc in combination with Win2VNC in dual-screen mode to pass the keystrokes and mouse motions to the X11 display?
Yes, for best response start up x11vnc with the "-nofb" option (disables framebuffer polling, and does other optimizations) on the secondary display (X11) machine. Then start up Win2VNC on the primary display (Windows) referring it to the secondary display.
This will also work X11 to X11 using x2vnc, however you would probably just want to avoid VNC and use x2x for that.
For reference, here are some links to Win2VNC-like programs for multiple monitor setups:
- Original Win2VNC
- Enhanced Win2VNC (broken?) and sourceforge link
- x2vnc
- x2x
- zvnc (MorphOS)
All of them will work with x11vnc (except x2x where it is not needed.)
Q-21: I am running Win2VNC on my Windows machine and "x11vnc -nofb" on Unix to pass keyboard and mouse to the Unix monitor. Whenever I start Win2VNC it quickly disconnects and x11vnc says: rfbProcessClientNormalMessage: read: Connection reset by peer
Is the default visual of the X display you run x11vnc on low color (e.g. 8 bit per pixel PseudoColor)? (you can run xdpyinfo to check, look in the "screen" section.) There seems to be a bug in Win2VNC in that it cannot deal correctly with colormaps (PseudoColor is the most common example of a visual with a colormap.)
If so, there are a couple options. 1) Can you set the default visual on your display to be depth 24 TrueColor? Sun machines often have 8+24 overlay/multi-depth visuals, and you can make the default visual depth 24 TrueColor (see fbconfig(1) and Xsun(1).) 2) As of Feb/2004 x11vnc has the -visual option to allow you to force the framebuffer visual to whatever you want (this usually messes up the colors unless you are very clever.) In this case, the option provides a convenient workaround for the Win2VNC bug:
x11vnc -nofb -visual TrueColor -display :0 ...
So the visual will be set to 8bpp TrueColor and Win2VNC can handle this. Since Win2VNC does not use the framebuffer data there should be no problems in doing this.
Q-22: Can I run "x11vnc -nofb" on a Mac OS X machine to redirect mouse and keyboard input to it from Windows and X11 machines via Win2VNC and x2vnc, respectively?
Yes, as of Nov/2006 you can. There may be a trick or two you'll need to do to get the Clipboard exchange between the machines to work.
Q-23: The X display I run x11vnc on is only 8 bits per pixel (bpp) PseudoColor (i.e. only 256 distinct colors.) The x11vnc colors may start out OK, but after a while they are incorrect in certain windows.
Use the -flashcmap option to have x11vnc watch for changes in the colormap, and propagate those changes back to connected clients. This can be slow (since the whole screen must be updated over the network whenever the colormap changes.) This flashing colormap behavior often happens if an application installs its own private colormap when the mouse is in its window. "netscape -install" is a well-known historical example of this. Consider reconfiguring the system to 16 bpp or depth 24 TrueColor if at all possible.
Also note the option -8to24 (Jan/2006) can often remove the need for flashing the colormap. Everything is dynamically transformed to depth 24 at 32 bpp using the colormaps. There may be painting errors however (see the following FAQ for tips on reducing and correcting them.)
In some rare cases (SCO unixware) the -notruecolor option has corrected colors on 8bpp displays. The red, green, and blue masks were non-zero in 8bpp PseudoColor on an obscure setup, and this option corrected the problems.
Q-24: Color problems: Why are the colors for some windows incorrect in x11vnc? BTW, my X display has nice overlay/multi-depth visuals of different color depths: e.g. there are both depth 8 and 24 visuals available at the same time.
You may want to review the previous question regarding 8 bpp PseudoColor.
On some hardware (Sun/SPARC and SGI), the -overlay option discussed a couple paragraphs down may solve this for you (you may want to skip to it directly.) On other hardware the less robust -8to24 option may help (also discussed below.)
Run xdpyinfo(1) to see what the default visual is and what the depths of the other visuals are. Does the default visual have a depth of 8 but there are other visuals of depth 24? If it does, can you possibly re-configure your X server to make a depth 24 visual the default? If you can do it, this will save you a lot of grief WRT colors and x11vnc (and for general usage too!) Here is how I do this on an old Sparcstation 20 running Solaris 9 with SX graphics
xinit -- -dev /dev/fb defclass TrueColor defdepth 24
and it works nicely (note: to log into console from the dtlogin window, select "Options -> Command Line Login", then login and enter the above command.) See the -dev section of the Xsun(1) manpage for a description of the above arguments. If you have root permission, a more permanent and convenient thing to do is to record the arguments in a line like:
:0 Local local_uid@console root /usr/openwin/bin/Xsun -dev /dev/fb defclass TrueColor defdepth 24
in /etc/dt/config/Xservers (copy /usr/dt/config/Xservers.) Also look at the fbconfig(1) and related manpages (e.g. ffbconfig, m64config, pgxconfig, SUNWjfb_config, etc ...) for hardware framebuffer settings that may achieve the same effect.
In general for non-Sun machines, look at the "-cc class" and related options in your X server manpage (perhaps Xserver(1)), it may allow modifying the default visual (e.g. "-cc 4", see <X11/X.h> for the visual class numbers.) On XFree86 some video card drivers (e.g. Matrox mga) have settings like Option "Overlay" "24,8" to support multi-depth overlays. For these, use the "-cc 4" X server command line option to get a depth 24 default visual.
The -overlay mode: Another option is if the system with overlay visuals is a Sun system running Solaris or SGI running IRIX you can use the -overlay x11vnc option (Aug/2004) to have x11vnc use the Solaris XReadScreen(3X11) function to poll the "true view" of the whole screen at depth 24 TrueColor. XReadDisplay(3X11) is used on IRIX. This is useful for Legacy applications (older versions of Cadence CAD apps are mentioned by x11vnc users) that require the default depth be 8bpp, or the app will use a 8bpp visual even if depth 24 visuals are available, and so the default depth workaround described in the previous paragraph is not sufficient for these apps.
It seems that Xorg is working toward supporting XReadDisplay(3X11) as part of the RENDER extension work. When it does support it and provides a library API x11vnc will be modified to take advantage of the feature to support -overlay on Linux, *BSD, etc. Until then see the -8to24 mode below.
Misc. notes on -overlay mode: An amusing by-product of -overlay mode is that the mouse cursor shape is correct! (i.e. XFIXES is not needed.) The -overlay mode may be somewhat slower than normal mode due to the extra framebuffer manipulations that must be performed. Also, on Solaris there is a bug in that for some popup menus, the windows they overlap will have painting errors (flashing colors) while the popup is up (a workaround is to disable SaveUnders by passing -su to Xsun, e.g. in your /etc/dt/config/Xservers file.)
The -8to24 mode: The -8to24 x11vnc option (Jan/2006) is a kludge to try to dynamically rewrite the pixel values so that the 8bpp part of the screen is mapped onto depth 24 TrueColor. This is less robust than the -overlay mode because it is done by x11vnc outside of the X server. So only use it on OS's that do not support -overlay. The -8to24 mode will work if the default visual is depth 24 or depth 8. It scans for any windows within 3 levels of the root window that are 8bpp (i.e. legacy application), or in general ones that are not using the default visual. For the windows it finds it uses XGetSubImage() to retrieve the pixels values and uses the correct indexed colormap to create a depth 24 TrueColor view of the whole screen. This depth 24, 32bpp view is exported via VNC.
Even on pure 8bpp displays it can be used as an alternative to -flashcmap to avoid color flashing completely.
This scheme is approximate and can often lead to painting errors. You can manually correct most painting errors by pressing 3 Alt_L's in a row, or by using something like: -fixscreen V=3.0 to automatically refresh the screen every 3 seconds. Also -fixscreen 8=3.0 has been added to just refresh the non-default visual parts of the screen.
In general the scheme uses many resources and may give rise to sluggish behavior. If multiple windows are using different 8bpp indexed colormaps all but one window may need to be iconified for the colors to be correct. There are a number of tunable parameters to try to adjust performance and painting accuracy. The option -8to24 nogetimage can give a nice speedup if the default depth 24 X server supports hiding the 8bpp bits in bits 25-32 of the framebuffer data. On very slow machines -8to24 poll=0.2,cachewin=5.0 gives an useful speedup. See the -8to24 help description for information on tunable parameters, etc.
Colors still not working correctly? Run xwininfo on the application with the incorrect colors to verify that the depth of its visual is different from the default visual depth (gotten from xdpyinfo.) One possible workaround in this case is to use the -id option to point x11vnc at the application window itself. If the application is complicated (lots of toplevel windows and popup menus) this may not be acceptable, and may even crash x11vnc (but not the application.) See also -appshare.
It is theoretically possible to solve this problem in general (see xwd(1) for example), but it does not seem trivial or sufficiently fast for x11vnc to be able to do so in real time. The -8to24 method does this approximately and is somewhat usable. Fortunately the -overlay option works for Solaris machines with overlay visuals where most of this problem occurs.
Q-25: I am on a high color system (depth >= 24) but I seem to have colormap problems. They either flash or everything is very dark.
This can happen if the default Visual (use xdpyinfo to list them) is DirectColor instead of TrueColor. These are both usually used in high color modes, but whereas TrueColor uses static ramps for the Red, Green, and Blue components, DirectColor has arbitrary colormaps for the Red, Green, and Blue Components. Currently x11vnc cannot decode these colormaps and treats them just like TrueColor.
The only workaround so far is to restart the X server with the "-cc 4" option to force TrueColor as the default visual (DirectColor is "-cc 5"; see /usr/include/X11/X.h.) The only place we have seen this is with the virtual framebuffer server Xvfb on Xorg 7.2. So in that case you probably should restart it with something like this: "Xvfb :1 -cc 4 -screen 0 1280x1024x24". It should be possible for x11vnc to handle DirectColor, but this hasn't been implemented due to its rare usage.
You may also see this problem on an X display with a TrueColor default visual where an application chooses a DirectColor visual for its window(s). It seems the application also needs to install its own colormap for the visual for the colors to be messed up in x11vnc. One can make xwud do this for example.
Run the xwininfo program in a terminal. It will ask you to click on the desired application window. After clicking, it will print out much information, including the window id (e.g. 0x6000010.) Also, the visual and depth of the window printed out is often useful in debugging x11vnc color problems.
Also, as of Dec/2004 you can use "-id pick" to have x11vnc run xwininfo(1) for you and after you click the window it extracts the windowid. Besides "pick" there is also "id:root" to allow you to go back to root window when doing remote-control.
Q-27: Why don't menus or other transient windows come up when I am using the -id windowid option to view a single application window?
This is related to the behavior of the XGetImage(3X11) and XShmGetImage() interfaces regarding backingstore, saveunders, etc. The way the image is retrieved depends on some aspects of how the X server maintains the display image data and whether other windows are clipping or obscuring it. See the XGetImage(3X11) man page for more details. If you disable BackingStore and SaveUnders in the X server you should be able to see these transient windows.
If things are not working and you still want to do the single window polling, try the -sid windowid option ("shifted" windowid.)
Update: as of Nov/2009 in the 0.9.9 x11vnc development tarball, there is an experimental Application Sharing mode that improves upon the -id/-sid single window sharing: -appshare (run "x11vnc -appshare -help" for more info.) It is still very primitive and approximate, but at least it displays multiple top-level windows.
Q-28: My X display is depth 24 at 24bpp (instead of the normal depth 24 at 32bpp.) I'm having lots of color and visual problems with x11vnc and/or vncviewer. What's up?
First off, depth 24 at 24bpp (bpp=bits-per-pixel) is fairly uncommon and can cause problems in general. It also can be slower than depth 24 at 32bpp. You might want to switch to 32bpp (for XFree86 see the "-fbbpp 32", DefaultFbBpp, FbBpp and related options.) Perhaps you have 24bpp because the video memory of the machine is low and the screen wouldn't fit in video RAM at 32bpp. For this case depth 16 at 16bpp might be an acceptable option.
In any event x11vnc should handle depth 24 at 24bpp (although performance may be slower, and you may need to use the ZRLE encoding instead of Tight.) There are some caveats involving the viewer however:
The RealVNC Unix viewer cannot handle 24bpp from the server, it will say: "main: setPF: not 8, 16 or 32 bpp?" and exit. I have not checked the RealVNC Windows viewer.
So you need to use the TightVNC Unix viewer. However there are some problems with that too. It seems libvncserver does not do 24bpp correctly with the Tight encoding. The colors and screen ultimately get messed up. So you have to use a different encoding with the TightVNC vncviewer, try "zlib", "hextile", or one of the other encodings (e.g. vncviewer -encodings "zlib hextile" ....) I have not checked the TightVNC or UltraVNC Windows viewers.
It appears the older RealVNC Unix viewers (e.g. 3.3.3 and 3.3.7) can handle 24bpp from the server, so you may want to use those. They evidently request 32 bpp and libvncserver obliges.
Update: as of Apr/2006 you can use the -24to32 option to have x11vnc dynamically transform the 24bpp pixel data to 32bpp. This extra transformation could slow things down further however.
Now coming the opposite direction if you are running the vncviewer on the 24bpp display, TightVNC will fail with "Can't cope with 24 bits-per-pixel. Sorry." and RealVNC will fail with "main: Error: couldn't find suitable pixmap format" so evidently you cannot use 24bpp for the vncviewers to work on that X display.
Note, however, that the Unix viewer in the Enhanced TightVNC Viewer (SSVNC) project can handle 24bpp X displays. It does this by requesting a 16bpp pixel format (or 8bpp if the -bgr233 option has been supplied) from the VNC server, and translates that to 24bpp locally.
Q-29: Can I use x11vnc to view and interact with an Xterminal (e.g. NCD) that is not running UNIX and so x11vnc cannot be run on it directly?
You can, but it will likely be very wasteful of network bandwidth since you will be polling the X display over the network as opposed to over the local hardware. To do this, run x11vnc on a UNIX machine as close as possible network-wise (e.g. same switch) to the Xterminal machine. Use the -display option to point the display to that of the Xterminal (you'll of course need basic X11 permission to do that) and finally supply the -noshm option (this enables the polling over the network.)
If the Xterminal's X display is open to the network for connections, you might use something like "-display xterm123:0". If you are trying to do this via an SSH tunnel (assuming you can actually ssh into the Xterminal) it will be a little tricky (either use the ssh "-R" option or consider ssh-ing in the other direction.) In all cases the X11 permissions need to allow the connection.
The response will likely be sluggish (maybe only one "frame" per second.) This mode is not recommended except for "quick checks" of hard to get to X servers. Use something like "-wait 150" to cut down on the polling rate. You may also need -flipbyteorder if the colors get messed up due to endian byte order differences.
Q-30: How do I get my X permissions (MIT-MAGIC-COOKIE file) correct for a Unix/Linux machine acting as an Xterminal?
If the X display machine is a traditional Xterminal (where the X server process runs on the Xterminal box, but all of the X client applications (firefox, etc) run on a central server (aka "terminal server")), you will need to log into the Xterminal machine (i.e. get a shell running there) and then start the x11vnc program. If the Xterminal Linux/Unix machine is stripped down (e.g. no users besides root) that may be difficult.
The next problem is the login Display Manager (e.g. gdm, kdm), and hence the MIT-MAGIC-COOKIE auth files, are on the central server and not on the Xterminal box where the X server and x11vnc processes are.
So unless X permissions are completely turned off (e.g. "xhost +"), to run the x11vnc process on the Xterminal box the MIT-MAGIC-COOKIE auth file data (XAUTHORITY or $HOME/.Xauthority) must be accessible by or copied to the Xterminal. If $HOME/.Xauthority is exported via NFS (this is insecure of course, but has been going on for decades), then x11vnc can simply pick it up via NFS (you may need to use the -auth option to point to the correct file.) Other options include copying the auth file using scp, or something like:
central-server> xauth nextract - xterm123:0 | ssh xterm123 xauth nmerge -
and then, say, ssh from central-server to xterm123 to start x11vnc. Here "xterm123" refers to the computer acting as the Xterminal and "central-server" is the terminal server. You can use "xauth -f /path/to/cookie-file list" to examine the contents of the cookie(s) in a file "/path/to/cookie-file". See the xauth(1) manpage for more details.
If the display name in the cookie file needs to be changed between the two hosts, see this note on the "xauth add ..." command.
A less secure option is to run something like "xhost +127.0.0.1" while sitting at the Xterminal box to allow cookie-free local access for x11vnc. You can run "xhost -127.0.0.1" after x11vnc connects if you want to go back to the original permissions.
If the Xterminal is really stripped down and doesn't have any user accounts, NFS, etc. you'll need to contact your system administrator to set something up. It can be done!!! Some Xterminal projects have actually enabled "run locally" facilities for the running of an occasional app more efficiently locally on the Xterminal box (e.g. realplayer.)
Not recommended, but as a last resort, you could have x11vnc poll the Xterminal Display over the network. For this you would run a "x11vnc -noshm ..." process on the central-server (and hope the network admin doesn't get angry...)
Note: use of Display Manager (gdm, kdm, ...) auth cookie files (i.e. from /var/..., /tmp/..., or elsewhere) may require modification via xauth(1) to correctly include the display x11vnc refers to (e.g. "xauth -f cookie-file add :0 . 45be51ae2ce9dfbacd882ab3ef8e96b1", where the "45be51..." cookie value was found from an "xauth -f /path/to/original/cookie-file list") or other reasons. See xauth(1) manpage for full details on how to transfer an MIT-MAGIC-COOKIE between machines and displays.
VNCviewer performance on Xterminals: This isn't related to x11vnc on Xterminals, but we mention it here anyway because of the similar issues. If you are on an Xterminal and want to use vncviewer to connect to a VNC server somewhere, then performance would be best if you ran the viewer on the Xterminal box. Otherwise, (i.e. running the viewer process on the central-server) all of the vncviewer screen drawing is done more inefficiently over the network. Something to consider, especially on a busy network. (BTW, this has all of the above permission, etc, problems: both vncviewer and x11vnc are X client apps desired to be run on the Xterminal box.)
The Sun Ray technology is a bit like "VNC done in hardware" (the Sun Ray terminal device, DTU, playing the role of the vncviewer.) Completely independent of that, the SunRay user's session is still an X server that speaks the X11 protocol and so x11vnc simply talks to the X server part to export the SunRay desktop to any place in the world (i.e. not only to a Sun Ray terminal device), creating a sort of "Soft Ray". Please see this discussion of Sun Ray issues for solutions to problems.
Also see the Sun Ray Remote Control Toolkit that uses x11vnc.
As of Dec/2004 there is a remote control feature. It can change a huge number of parameters on the fly: see the -remote and -query options. To shut down the running x11vnc server just type "x11vnc -R stop". To disconnect all clients do "x11vnc -R disconnect:all", etc.
If the -forever option has not been supplied, x11vnc will automatically exit after the first client disconnects. In general if you cannot use the remote control, then you will have to kill the x11vnc process This can be done via: "kill NNNNN" (where NNNNN is the x11vnc process id number found from ps(1)), or "pkill x11vnc", or "killall x11vnc" (Linux only.)
If you have not put x11vnc in the background via the -bg option or shell & operator, then simply press Ctrl-C in the shell where x11vnc is running to stop it.
Potential Gotcha: If somehow your Keypress of Ctrl-C went through x11vnc to the Xserver that then delivered it to x11vnc it is possible one or both of the Ctrl or C keys will be left stuck in the pressed down state in the Xserver. Tapping the stuck key (either via a new x11vnc or at the physical console) will release it from the stuck state. If the keyboard seems to be acting strangely it is often fixed by tapping Ctrl, Shift, and Alt. Alternatively, the -clear_mods option and -clear_keys option can be used to release pressed keys at startup and exit. The option -clear_all will also try to unset Caps_Lock, Num_Lock, etc.
Look at the -remote (an alias is -R) and -query (an alias is -Q) options added in Dec/2004. They allow nearly everything to be changed dynamically and settings to be queried. Examples: "x11vnc -R shared", "x11vnc -R forever", "x11vnc -R scale:3/4", "x11vnc -Q modtweak", "x11vnc -R stop", "x11vnc -R disconnect:all", etc..
These commands do not start a x11vnc server, but rather communicate with one that is already running. The X display (X11VNC_REMOTE property) is used as the communication channel, so the X permissions and DISPLAY must be set up correctly for communication to be possible.
There is also a simple Tcl/Tk gui based on this remote control mechanism. See the -gui option for more info. You will need to have Tcl/Tk (i.e. /usr/bin/wish) installed for it to work. It can also run in the system tray: "-gui tray" or as a standalone small icon window: "-gui icon". Use "-gui tray=setpass" for a naive user "Share My Desktop" mode.
You may already have one in $HOME/.vnc/passwd if you have used, say, the vncserver program from the regular RealVNC or TightVNC packages (i.e. launching the Xvnc server.) Otherwise, you could use the vncpasswd(1) program from those packages.
As of Jun/2004 x11vnc supports the -storepasswd "pass" "file" option, which is the same functionality of storepasswd. Be sure to quote the "pass" if it contains shell meta characters, spaces, etc. Example:
x11vnc -storepasswd 'sword*fish' $HOME/myvncpasswd
You then use the password via the x11vnc option: "-rfbauth $HOME/myvncpasswd"
As of Jan/2006 if you do not supply any arguments:
x11vnc -storepasswd
you will be prompted for a password to save to ~/.vnc/passwd (your keystrokes when entering the password will not be echoed to the screen.) If you supply one argument, e.g. "x11vnc -storepasswd ~/.mypass", the password you are prompted for will be stored in that file.
x11vnc also has the -passwdfile and -passwd/-viewpasswd plain text (i.e. not obscured like the -rfbauth VNC passwords) password options.
You can use the -usepw option to automatically use any password file you have in ~/.vnc/passwd or ~/.vnc/passwdfile (the latter is used with the -passwdfile option.)
x11vnc -usepw -display :0 ...
If neither file exists you are prompted to store a password in ~/.vnc/passwd. If a password file cannot be found or created x11vnc exits immediately. An admin may want to set it up this way for users who do not know better.
You can use the vncpasswd program from RealVNC or TightVNC mentioned above. As of Jan/2006 the -storepasswd option without any arguments will not echo your password as you type it and save the file to ~/.vnc/passwd:
# x11vnc -storepasswd
Enter VNC password:
Verify password:
Write password to /home/myname/.vnc/passwd? [y]/n
Password written to: /home/myname/.vnc/passwd
You can also give it an alternate filename, e.g. "x11vnc -storepasswd ~/.mypass"
Q-36: Can I have two passwords for VNC viewers, one for full access and the other for view-only access to the display?
Yes, as of May/2004 there is the -viewpasswd option to supply the view-only password. Note the full-access password option -passwd must be supplied at the same time. E.g.: -passwd sword -viewpasswd fish.
To avoid specifying the passwords on the command line (where they could be observed via the ps(1) command by any user) you can use the -passwdfile option to specify a file containing plain text passwords. Presumably this file is readable only by you, and ideally it is located on the machine x11vnc is run on (to avoid being snooped on over the network.) The first line of this file is the full-access password. If there is a second line in the file and it is non-blank, it is taken as the view-only password. (use "EMPTY" to supply an empty one.)
View-only passwords currently do not work for the -rfbauth password option (standard VNC password storing mechanism.) FWIW, note that although the output (usually placed in $HOME/.vnc/passwd) by the vncpasswd or storepasswd programs (or from x11vnc -storepasswd) looks encrypted they are really just obscured to avoid "casual" password stealing. It takes almost no skill to figure out how to extract the plain text passwords from $HOME/.vnc/passwd since it is very straight-forward to work out what to do from the VNC source code.
Yes, as of Jan/2006 in the libvncserver CVS the -passwdfile option has been extended to handle as many passwords as you like. You put the view-only passwords after a line BEGIN_VIEWONLY.
You can also easily annotate and comment out passwords in the file. You can have x11vnc re-read the file dynamically when it is modified.
Q-38: Does x11vnc support Unix usernames and passwords? Can I further limit the set of Unix usernames who can connect to the VNC desktop?
Update: as of Feb/2006 x11vnc has the -unixpw option that does this outside of the VNC protocol and libvncserver. The standard su(1) program is used to validate the user's password. A familiar "login:" and "Password:" dialog is presented to the user on a black screen inside the vncviewer. The connection is dropped if the user fails to supply the correct password in 3 tries or does not send one before a 25 second timeout. Existing clients are view-only during this period. A list of allowed Unix usernames may also be supplied along with per-user settings.
There is also the -unixpw_nis option for non-shadow-password (typically NIS environments, hence the name) systems where the traditional getpwnam() and crypt() functions are used instead of su(1). The encrypted user passwords must be accessible to the user running x11vnc in -unixpw_nis mode, otherwise the logins will always fail even when the correct password is supplied. See ypcat(1) and shadow(5).
Two settings are enforced in the -unixpw and -unixpw_nis modes to provide extra security: the 1) -localhost and 2) -stunnel or -ssl options. Without these one might send the Unix username and password data in clear text over the network which is a very bad idea. They can be relaxed if you want to provide encryption other than stunnel or -ssl (the constraint is automatically relaxed if SSH_CONNECTION is set and indicates you have ssh-ed in, however the -localhost requirement is still enforced.)
The two -unixpw modes have been tested on Linux, Solaris, Mac OS X, HP-UX, AIX, Tru64, FreeBSD, OpenBSD, and NetBSD. Additional testing is appreciated. For the last 4 it appears that su(1) will not prompt for a password if su-ing to oneself. Since x11vnc requires a password prompt from su, x11vnc forces those logins to fail even when the correct password is supplied. On *BSD it appears this can be corrected by removing the pam_self.so entry in /etc/pam.d/su.
Previous older discussion (prior to the -unixpw option):
Until the VNC protocol and libvncserver support this things will be approximate at best.
One approximate method involves starting x11vnc with the -localhost option. This basically requires the viewer user to log into the workstation where x11vnc is running via their Unix username and password, and then somehow set up a port redirection of his vncviewer connection to make it appear to emanate from the local machine. As discussed above, ssh is useful for this: "ssh -L 5900:localhost:5900 user@hostname ..." See the ssh wrapper scripts mentioned elsewhere on this page. stunnel does this as well.
Of course a malicious user could allow other users to get in through his channel, but that is a problem with every method. Another thing to watch out for is a malicious user on the viewer side (where ssh is running) trying to sneak in through the ssh port redirection there.
Regarding limiting the set of Unix usernames who can connect, the traditional way would be to further require a VNC password to supplied (-rfbauth, -passwd, etc) and only tell the people allowed in what the VNC password is. A scheme that avoids a second password involves using the -accept option that runs a program to examine the connection information to determine which user is connecting from the local machine. That may be difficult to do, but, for example, the program could use the ident service on the local machine (normally ident should not be trusted over the network, but on the local machine it should be accurate: otherwise root has been compromised and so there are more serious problems! Unfortunately recent Linux distros seem to provide a random string (MD5 hash?) instead of the username.) An example script passed in via -accept scriptname that deduces the Unix username and limits who can be accepted might look something like this:
#!/bin/sh
if [ "$RFB_CLIENT_IP" != "127.0.0.1" -o "$RFB_SERVER_IP" != "127.0.0.1" ]; then
exit 1 # something fishy... reject it.
fi
user=`echo "$RFB_CLIENT_PORT, $RFB_SERVER_PORT" | nc -w 1 $RFB_CLIENT_IP 113 \
| grep 'USERID.*UNIX' | head -n 1 | sed -e 's/[\r ]//g' | awk -F: '{pri
nt $4}'`
for okuser in fred barney wilma betty
do
if [ "X$user" = "X$okuser" ]; then
exit 0 # accept it
fi
done
exit 1 # reject it
For this to work with ssh port redirection, the ssh option
UsePrivilegeSeparation must be enabled otherwise the userid will
always be "root".
Here is a similar example based on Linux netstat(1) output:
#!/bin/sh
#
# accept_local_netstat: x11vnc -accept command to accept a local
# vncviewer connection from acceptable users. Linux netstat -nte is used.
PATH=/bin:/usr/bin:$PATH; export PATH; # set to get system utils
allowed="`id -u fred`"; # add more user numbers if desired.
# check required settings
ok=1
if [ "X$allowed" = "X" ]; then
ok=0; # something wrong with allowed list
fi
if [ "X$RFB_CLIENT_IP" != "X127.0.0.1" -o "X$RFB_SERVER_IP" != "X127.0.0.1" ];
then
ok=0; # connection not over localhost
fi
if [ "$RFB_CLIENT_PORT" -le 0 -o "$RFB_SERVER_PORT" -le 0 ]; then
ok=0; # something wrong with tcp port numbers
fi
if [ "$ok" = 0 ]; then
echo "$0: invalid setting:" 1>&2
env | grep ^RFB | sort 1>&2
exit 1
fi
# Linux netstat -nte:
# Proto Recv-Q Send-Q Local Address Foreign Address State
User Inode
# 0 0 0 RFB_CLIENT RFB_SERVER ESTABLISHED
nnnn ....
#
user=`netstat -nte | grep ESTABLISHED \
| grep " $RFB_CLIENT_IP:$RFB_CLIENT_PORT *$RFB_SERVER_IP:$RFB_SERVER_P
ORT "`
echo "netstat match: $user" 1>&2
user=`echo "$user" | head -n 1 | sed -e 's/^.*ESTABLISHED/ /' | awk '{print $1}
'`
ok=0
for u in $allowed
do
if [ "X$user" = "X$u" ]; then
ok=1
break
fi
done
if [ "X$ok" = "X1" ]; then
echo "$0: user accepted: '$user'" 1>&2
exit 0
else
echo "$0: user '$user' invalid:" 1>&2
echo "$0: allowed: $allowed" 1>&2
env | grep ^RFB | sort 1>&2
exit 1
fi
Q-39: Can I supply an external program to provide my own custom login method (e.g. Dynamic/One-time passwords or non-Unix (LDAP) usernames and passwords)?
Yes, there are several possibilities. For background see the FAQ on the -accept where an external program may be run to decide if a VNC client should be allowed to try to connect and log in. If the program (or local user prompted by a popup) answers "yes", then -accept proceeds to the normal VNC and x11vnc authentication methods, otherwise the connection is dropped.
To provide more direct coupling to the VNC client's username and/or supplied password the following options were added in Sep/2006:
- -unixpw_cmd command
- -passwdfile cmd:command
- -passwdfile custom:command
In each case "command" is an external command run by x11vnc. You supply it. For example, it may couple to your LDAP system or other servers you set up.
For -unixpw_cmd the normal -unixpw Login: and Password: prompts are supplied to the VNC viewer and the strings the client returns are then piped into "command" as the first two lines of its standard input. If the command returns success, i.e. exit(0), the VNC client is accepted, otherwise it is rejected.
For "-passwdfile cmd:command" the command is run and it returns a password list (like a password file, see the -passwdfile read:filename mode.) Perhaps a dynamic, one-time password is retrieved from a server this way.
For "-passwdfile custom:command" one gets complete control over the VNC challenge-response dialog with the VNC client. x11vnc sends out a string of random bytes (16 by the VNC spec) and the client returns the same number of bytes in a way the server can verify only the authorized user could have created. The VNC protocol specifies DES encryption with a password. If you are willing to modify the VNC viewers, you can have it be anything you want, perhaps a less crackable MD5 hash scheme or one-time pad. Your program will read from its standard input the size of the challenge-response followed by a newline, then the challenge bytes followed by the response bytes. If your command then returns success, i.e. exit(0), the VNC client is accepted, otherwise it is rejected.
In all cases the "RFB_*" environment variables are set as under -accept. These variables can provide useful information for the externally supplied program to use.
Q-40: Why does x11vnc exit as soon as the VNC viewer disconnects? And why doesn't it allow more than one VNC viewer to connect at the same time?
These defaults are simple safety measures to avoid someone unknowingly leaving his X11 desktop exposed (to the internet, say) for long periods of time. Use the -forever option (aka -many) to have x11vnc wait for more connections after the first client disconnects. Use the -shared option to have x11vnc allow multiple clients to connect simultaneously.
Recommended additional safety measures include using ssh (see above), stunnel, -ssl, or a VPN to authenticate and encrypt the viewer connections or to at least use the -rfbauth passwd-file option to use VNC password protection (or -passwdfile) It is up to YOU to apply these security measures, they will not be done for you automatically.
Yes, look at the -allow and -localhost options to limit connections by hostname or IP address. E.g.
x11vnc -allow 192.168.0.1,192.168.0.2
for those two hosts or
x11vnc -allow 192.168.0.
for a subnet. For individual hosts you can use the hostname instead of the IP number, e.g.: "-allow snoopy", and "-allow darkstar,wombat". Note that -localhost achieves the same thing as "-allow 127.0.0.1"
For more control, build libvncserver with libwrap support (tcp_wrappers) and then use /etc/hosts.allow See hosts_access(5) for complete details.
Here is one way to pass this information to the configure script:
env CPPFLAGS=-DUSE_LIBWRAP LDFLAGS=-lwrap ./configure
then run make as usual. This requires libwrap and its development package (tcpd.h) to be installed on the build machine. If additional CPPFLAGS or LDFLAGS options are needed supply them as well using quotes.
The resulting x11vnc then uses libwrap/tcp_wrappers for connections. The service name you will use in /etc/hosts.allow and /etc/hosts.deny is "vnc", e.g.:
vnc: 192.168.100.3 .example.com
Note that if you run x11vnc out of inetd you do not need to build x11vnc with libwrap support because the /usr/sbin/tcpd reference in /etc/inetd.conf handles the tcp_wrappers stuff.
Q-43: Can I have x11vnc only listen on one network interface (e.g. internal LAN) rather than having it listen on all network interfaces and relying on -allow to filter unwanted connections out?
As of Mar/2005 there is the "-listen ipaddr" option that enables this. For ipaddr either supply the desired network interface's IP address (or use a hostname that resolves to it) or use the string "localhost". For additional filtering simultaneously use the "-allow host1,..." option to allow only specific hosts in.
This option is useful if you want to insure that no one can even begin a dialog with x11vnc from untrusted network interfaces (e.g. ppp0.) The option -localhost now implies "-listen localhost" since that is what most people expect it to do.
Q-44: Now that -localhost implies listening only on the loopback interface, how I can occasionally allow in a non-localhost via the -R allowonce remote control command?
To do this specify "-allow localhost". Unlike -localhost this will leave x11vnc listening on all interfaces (but of course only allowing in local connections, e.g. ssh redirs.) Then you can later run "x11vnc -R allowonce:somehost" or use to gui to permit a one-shot connection from a remote host.
Q-45: Can I fine tune what types of user input are allowed? E.g. have some users just be able to move the mouse, but not click or type anything?
As of Feb/2005, the -input option allows you to do this. "K", "M", "B", "C", and "F" stand for Keystroke, Mouse-motion, Button-clicks, Clipboard, and File-Transfer, respectively. The setting: "-input M" makes attached viewers only able to move the mouse. "-input KMBC,M" lets normal clients do everything and enables view-only clients to move the mouse.
These settings can also be applied on a per-viewer basis via the remote control mechanism or the GUI. E.g. x11vnc -R input:hostname:M
Q-46: Can I prompt the user at the local X display whether the incoming VNC client should be accepted or not? Can I decide to make some clients view-only? How about running an arbitrary program to make the decisions?
Yes, look at the "-accept command" option, it allows you to specify an external command that is run for each new client. (use quotes around the command if it contains spaces, etc.) If the external command returns 0 (success) the client is accepted, otherwise with any other return code the client is rejected. See below how to also accept clients view-only.
The external command will have the RFB_CLIENT_IP environment variable set to the client's numerical IP address, RFB_CLIENT_PORT its port number. Similarly for RFB_SERVER_IP and RFB_SERVER_PORT to allow identification of the tcp virtual circuit. DISPLAY will be set to that of the X11 display being polled. Also, RFB_X11VNC_PID is set to the x11vnc process id (e.g. in case you decided to kill it), RFB_CLIENT_ID will be an id number, and RFB_CLIENT_COUNT the number of other clients currently connected. RFB_MODE will be "accept".
Built-in Popup Window: As a special case, "-accept popup" will instruct x11vnc to create its own simple popup window. To accept the client press "y" or click mouse on the "Yes" button. To reject the client press "n" or click mouse on the "No" button. To accept the client View-only, press "v" or click mouse on the "View" button. If the -viewonly option has been supplied, the "View" action will not be present: the whole display is view only in that case.
The popup window times out after 120 seconds, to change this behavior use "-accept popup:N" where N is the number of seconds (use 0 for no timeout.) More tricks: "-accept popupmouse" will only take mouse click responses, while "-accept popupkey" will only take keystroke responses (popup takes both.) After any of the 3 popup keywords you can supply a position of the window: +N+M, (the default is to center the window) e.g. -accept popupmouse+10+10.
Also as a special case "-accept xmessage" will run the xmessage(1) program to prompt the user whether the client should be accepted or not. This requires that you have xmessage installed and available via PATH. In case it is not already on your system, the xmessage program is available at ftp://ftp.x.org/ (End of Built-in Popup Window:)
To include view-only decisions for the external commands, prefix the command something like this: "yes:0,no:,view:3 mycommand ..." This associates the three actions: yes(accept), no(reject), and view(accept-view-only), with the numerical return (i.e. exit()) codes. Use "" instead of a number to set the default action (e.g. in case the external command returns an unexpected return code.)
Here is an example -accept script called accept_or_lock. It uses xmessage and xlock (replace with your screen lock command, maybe it is "xscreensaver-command -lock", or kdesktop_lock, or "dtaction LockDisplay".) It will prompt the user at the X display whether to accept, reject, or accept view-only the client, but if the prompt times out after 60 seconds the screen is locked and the VNC client is accepted. This allows the remote access when no one is at the display.
#!/bin/sh
#
# accept_or_lock: prompt user at X display whether to accept an incoming
# VNC connection. If timeout expires, screen is locked
# and the VNC viewer is accepted (allows remote access
# when no one is sitting at the display.)
#
# usage: x11vnc ... -forever -accept 'yes:0,no:*,view:4 accept_or_lock'
#
xmessage -buttons yes:2,no:3,view-only:4 -center \
-timeout 60 "x11vnc: accept connection from $RFB_CLIENT_IP?"
rc=$?
if [ $rc = 0 ]; then
xlock & # or "xlock -mode blank" for no animations.
sleep 5
exit 0
elif [ $rc = 2 ]; then
exit 0
elif [ $rc = 4 ]; then
exit 4
fi
exit 1
Stefan Radman has written a nice dtksh script dtVncPopup for use in CDE environments to do the same sort of thing. Information on how to use it is found at the top of the file. He encourages you to provide feedback to him to help improve the script.
Note that in all cases x11vnc will block while the external command or popup is being run, so attached clients will not receive screen updates, etc during this period.
To run a command when a client disconnects, use the "-gone command" option. This is for the user's convenience only: the return code of the command is not interpreted by x11vnc. The same environment variables are set as in "-accept command" (except that RFB_MODE will be "gone".)
As of Jan/2006 the "-afteraccept command" option will run the command only after the VNC client has been accepted and authenticated. Like -gone the return code is not interpreted. RFB_MODE will be "afteraccept".)
Q-47: I start x11vnc as root because it is launched via inetd(8) or a display manager like gdm(1). Can I have x11vnc later switch to a different user?
As of Feb/2005 x11vnc has the -users option that allows things like this. Please read the documentation on it (also in the x11vnc -help output) carefully for features and caveats. It's use can often decrease security unless care is taken.
BTW, a nice use of it is "-users +nobody" that switches to the Unix user nobody right after connections to the X display are established.
In any event, while running x11vnc as root, remember it comes with no warranty ;-).
Q-48: I use a screen-lock when I leave my workstation (e.g. xscreensaver or xlock.) When I remotely access my workstation desktop via x11vnc I can unlock the desktop fine, but I am worried people will see my activities on the physical monitor. What can I do to prevent this, or at least make it more difficult?
Probably most work environments would respect your privacy if you powered off the monitor. Also remember if people have physical access to your workstation they basically can do anything they want with it (e.g. install a backdoor for later use, etc.)
In any event, as of Jun/2004 there is an experimental utility to make it more difficult for nosey people to see your x11vnc activities. The source for it is blockdpy.c The idea behind it is simple (but obviously not bulletproof): when a VNC client attaches to x11vnc put the display monitor in the DPMS "off" state, if the DPMS state ever changes immediately start up the screen-lock program. The x11vnc user will notice something is happening and think about what to do next (while the screen is in a locked state.)
This works (or at least has a chance of working) because if the intruder moves the mouse or presses a key on the keyboard, the monitor wakes up out of the DPMS off state, and this induces the screen lock program to activate as soon as possible. Of course there are cracks in this, the eavesdropper could detach your monitor and insert a non-DPMS one, and there are race conditions. As mentioned above this is not bulletproof. A really robust solution would likely require X server and perhaps even video hardware support.
The blockdpy utility is launched by the -accept option and told to exit via the -gone option (the vnc client user should obviously re-lock the screen before disconnecting!) Instructions can be found in the source code for the utility at the above link. Roughly it is something like this:
x11vnc ... -accept "blockdpy -bg -f $HOME/.bdpy" -gone "touch $HOME/.bdpy"
but please read the top of the file.
Update: As of Feb/2007 there is some builtin support for this: -forcedpms and -clientdpms however, they are probably less robust than the above blockdpy.c scheme, since if the person floods the physical machine with mouse or pointer input he can usually see flashes of the screen before the monitor is powered off again. See also the -grabkbd, -grabptr, and -grabalways options.
Yes, a user mentions he uses the -gone option under CDE to run a screen lock program:
x11vnc -display :0 -forever -gone 'dtaction LockDisplay'
Other possibilities are:
x11vnc -display :0 -forever -gone 'xscreensaver-command -lock'
x11vnc -display :0 -forever -gone 'kdesktop_lock'
x11vnc -display :0 -forever -gone 'xlock &'
x11vnc -display :0 -forever -gone 'xlock -mode blank &'
Here is a scheme using the -afteraccept option (in version 0.8) to unlock the screen after the first valid VNC login and to lock the screen after the last valid VNC login disconnects:
x11vnc -display :0 -forever -shared -afteraccept ./myxlocker -gone ./myxlocker
Where the script ./myxlocker is:
#!/bin/sh
#/usr/bin/env | grep RFB_ | sort # for viewing RFB_* settings.
if [ "X$RFB_MODE" = "Xafteraccept" ]; then
if [ "X$RFB_STATE" = "XNORMAL" ]; then # require valid login
if [ "X$RFB_CLIENT_COUNT" = "X1" ]; then
killall xlock # Linux only.
fi
fi
elif [ "X$RFB_MODE" = "Xgone" ]; then
if [ "X$RFB_STATE" = "XNORMAL" ]; then # require valid login
if [ "X$RFB_CLIENT_COUNT" = "X0" ]; then
xlock -mode blank &
fi
fi
fi
Note the xlock option "-mode blank" to avoid animations.
There is a problem if you have x11vnc running this way in -forever mode and you hit Ctrl-C to stop it. The xlock (or other program) will get killed too. To work around this make a little script called setpgrp that looks like:
#!/usr/bin/perl
setpgrp(0, 0);
exec @ARGV;
then use -gone "setpgrp xlock &", etc.
Q-50: How can I tunnel my connection to x11vnc via an encrypted SSH channel between two Unix machines?
See the description earlier on this page on how to tunnel VNC via SSH from Unix to Unix. A number of ways are described along with some issues you may encounter.
Other secure encrypted methods exists, e.g. stunnel, IPSEC, various VPNs, etc.
See also the Enhanced TightVNC Viewer (SSVNC) page where much of this is now automated.
Q-51: How can I tunnel my connection to x11vnc via an encrypted SSH channel from Windows using an SSH client like Putty?
Above we described how to tunnel VNC via SSH from Unix to Unix, you may want to review it. To do this from Windows using Putty it would go something like this:
- In the Putty dialog window under 'Session' enter the hostname or IP number of the Unix machine with display to be viewed.
- Make sure the SSH protocol is selected and the server port is correct.
- Under 'Connections/SSH/Tunnels' Add a Local connection with 'Source port: 5900' and 'Destination: localhost:5900'
- Log into the remote machine by pressing 'Open' and supplying username, password, etc.
- In that SSH shell, start up x11vnc by typing the command: x11vnc -display :0 plus any other desired options (e.g. -localhost.)
- Finally, start up your VNC Viewer in Windows and enter 'localhost:0' as the VNC server.
You can keep all of the settings in a Putty 'Saved Session'. Also, once everything is working, you can consider putting x11vnc -display :0 (plus other cmdline options) in the 'Remote command' Putty setting under 'Connections/SSH'.
See also the Enhanced TightVNC Viewer (SSVNC) page where much of this is now automated via the Putty plink utility.
For extra protection feel free to run x11vnc with the -localhost and -rfbauth/-passwdfile options.
If the machine you SSH into via Putty is not the same machine with the X display you wish to view (e.g. your company provides incoming SSH access to a gateway machine), then you need to change the above Putty dialog setting to: 'Destination: otherhost:5900', Once logged in, you'll need to do a second login (ssh or rsh) to the workstation machine 'otherhost' and then start up x11vnc on it. This can also be automated by Chaining SSH's.
As discussed above another option is to first start the VNC viewer in "listen" mode, and then launch x11vnc with the "-connect localhost" option to establish the reverse connection. In this case a Remote port redirection (not Local) is needed for port 5500 instead of 5900 (i.e. 'Source port: 5500' and 'Destination: localhost:5500' for a Remote connection.)
Q-52: How can I tunnel my connection to x11vnc via an encrypted SSL channel using an external tool like stunnel?
It is possible to use a "lighter weight" encryption setup than SSH or IPSEC. SSL tunnels such as stunnel (also stunnel.org) provide an encrypted channel without the need for Unix users, passwords, and key passphrases required for ssh (and at the other extreme SSL can also provide a complete signed certificate chain of trust.) On the other hand, since SSH is usually installed everywhere and firewalls often let its port through, ssh is frequently the path of least resistance (it also nicely manages public keys for you.)
Update: As of Feb/2006 x11vnc has the options -ssl, -stunnel, and -sslverify to provide integrated SSL schemes. They are discussed in the Next FAQ (you probably want to skip to it now.)
We include these non-built-in method descriptions below for historical reference. They are handy because can be used to create SSL tunnels to any VNC (or other type of) server.
Here are some basic examples using stunnel but the general idea for any SSL tunnel utility is the same:
- Start up x11vnc and constrain it to listen on localhost.
- Then start up the SSL tunnel running on the same machine to forward incoming connections to that x11vnc.
- Set up and run a similar SSL tunnel for the outgoing connection on the VNC viewer machine pointing it to the SSL/x11vnc server.
- Optionally, set up server (or even client) public/private keys for use in authenticating one side to the other.
- Finally, start the VNC Viewer and tell it to connect to the local port (e.g. a vnc display localhost:0) where its outgoing SSL tunnel is listening.
We'll first use the stunnel version 3 syntax since it is the most concise and Unixy.
Start up x11vnc listening on port 5900:
x11vnc -display :0 -rfbport 5900 -localhost -bg -passwdfile ~/mypass
Then start stunnel (version 3, not 4) with this command:
stunnel -d 5901 -r 5900 -p /path/to/stunnel.pem
The above two commands are run on host "far-away.east". The stunnel.pem is the self-signed PEM file certificate created when stunnel is built. One can also create certificates signed by Certificate Authorities or self-signed if desired using the x11vnc utilities described there.
SSL Viewers: Next, on the VNC viewer side we need an SSL tunnel to encrypt the outgoing connection. The nice thing is any SSL tunnel can be used because the protocol is a standard. For this example we'll also use stunnel on the viewer side on Unix. First start up the client-side stunnel (version 3, not 4):
stunnel -c -d localhost:5902 -r far-away.east:5901
Then point the viewer to the local tunnel on port 5902:
vncviewer -encodings "copyrect tight zrle hextile" localhost:2
That's it. Note that the ss_vncviewer script can automate this easily, and so can the Enhanced TightVNC Viewer (SSVNC) package.
Be sure to use a VNC password because unlike ssh by default the encrypted SSL channel provides no authentication (only privacy.) With some extra configuration one could also set up certificates to provide authentication of either or both sides as well (and hence avoid man-in-the-middle attacks.) See the stunnel and openssl documentation and also the key management section for details.
stunnel has also been ported to Windows, and there are likely others to choose from for that OS. Much info for using it on Windows can be found at the stunnel site and in this article The article also shows the detailed steps to set up all the authentication certificates. (for both server and clients, see also the x11vnc utilities that do this.) The default Windows client setup (no certs) is simpler and only 4 files are needed in a folder: stunnel.exe, stunnel.conf, libssl32.dll, libeay32.dll. We used an stunnel.conf containing:
# stunnel.conf:
client = yes
options = ALL
[myvncssl]
accept = localhost:5902
connect = far-away.east:5901
then double click on the stunnel.exe icon to launch it (followed by pointing the VNC viewer to localhost:2).
stunnel inetd-like mode:
As an aside, if you don't like the little "gap" of unencrypted TCP traffic (and a localhost listening socket) on the local machine between stunnel and x11vnc it can actually be closed by having stunnel start up x11vnc in -inetd mode:
stunnel -p /path/to/stunnel.pem -P none -d 5900 -l ./x11vnc_sh
Where the script x11vnc_sh starts up x11vnc:
#!/bin/sh
x11vnc -q -inetd -display :0 -passwdfile ~/mypass
Note that this creates a separate x11vnc process for each incoming connection (as any inetd x11vnc usage would), but for the case of normally just one viewer at a time it should not be a big problem.
stunnel 4 syntax:
Somewhat sadly, the stunnel version 4 syntax is not so amenable to the command line or scripts. You need to create a config file with the parameters. E.g.:
stunnel x11vnc.cfg
Where the file x11vnc.cfg contains:
foreground = yes
pid =
cert = /path/to/stunnel.pem
[x11vnc_stunnel]
accept = 5901
connect = 5900
One nice thing about version 4 is often the PEM file does not need to be specified because stunnel finds it in its installed area. One other gotcha the PEM file is usually only readable by root (it has the private key afterall), so you'll need to relax the permissions or make a copy that the user running x11vnc/stunnel can read.
SSL VNC Viewers:
Regarding VNC viewers that "natively" do SSL unfortunately there do not seem to be many. The SingleClick UltraVNC Java Viewer is SSL and is compatible with x11vnc's -ssl option and stunnel.) Commercial versions of VNC seem to have some SSL-like encryption built in, but we haven't tried those either and they probably wouldn't work since their (proprietary) SSL-like negotiation is likely embedded in the VNC protocol unlike our case where it is external.
Note: as of Mar/2006 libvncserver/x11vnc provides a SSL-enabled Java applet that can be served up via the -httpdir or -http options when -ssl is enabled. It will also be served via HTTPS via either the VNC port (e.g. https://host:5900/) or a 2nd port via the -https option.
In general current SSL VNC solutions are not particularly "seemless". But it can be done, and with a wrapper script on the viewer side and the -stunnel or -ssl option on the server side it works well and is convenient. Here is a simple script ss_vncviewer that automates running stunnel on the VNC viewer side on Unix a little more carefully than the commands printed above. (One could probably do a similar thing with a .BAT file on Windows in the stunnel folder.)
Update Jul/2006: we now provide an Enhanced TightVNC Viewer (SSVNC) package that starts up STUNNEL automatically along with some other features. All binaries (stunnel, vncviewer, and some utilities) are provided in the package. It works on Unix, Mac OS X, and Windows.
You can read about non-built-in methods in the Previous FAQ for background.
SSL tunnels provide an encrypted channel without the need for Unix users, passwords, and key passphrases required for ssh (and at the other extreme SSL can also provide a complete signed certificate chain of trust.) On the other hand, since SSH is usually installed everywhere and firewalls often let its port through, ssh is frequently the path of least resistance.
Built-in SSL x11vnc options:
As of Feb/2006 the x11vnc -ssl option automates the SSL tunnel creation on the x11vnc server side. An SSL-enabled Java Viewer applet is also provided that can be served via HTTP or HTTPS to automate SSL on the client side.
The -ssl mode uses the www.openssl.org library if available at build time.
The mode requires an SSL certificate and key (i.e. .pem file.) These are usually created via the openssl(1) program (in fact in for "-ssl" (same as "-ssl SAVE") it will run openssl for you automatically.) So the SSL is not completely "built-in" since this external tool needs to be installed, but at least x11vnc runs it for you automatically.
An -ssl example:
x11vnc -display :0 -ssl -passwdfile ~/mypass
You'll get output like this:
09/04/2006 19:27:35 Creating a self-signed PEM certificate...
09/04/2006 19:27:35
...
The SSL VNC desktop is: far-away.east:0
PORT=5900
SSLPORT=5900
In this case openssl(1) was used to create a PEM automatically. It will prompt you if you want to protect it with with a passphrase. Use "-ssl SAVE_NOPROMPT" to not be prompted. Use "-ssl TMP" to create a temporary self-signed cert that will be discarded when x11vnc exits.
Update: As of Nov/2008 x11vnc also supports the VeNCrypt SSL/TLS tunnel extension to the VNC protocol. The older ANONTLS method (vino) is also supported. This support is on by default when the -ssl option is in use and can be fine-tuned using these options: -vencrypt, -anontls, and -sslonly.
The normal x11vnc -ssl operation is somewhat like a URL method vncs://hostname if vnc://hostname indicates a standard unencrypted VNC connection. Just as https://hostname is an SSL encrypted version of http://hostname. The entire VNC session goes through the SSL tunnel. VeNCrypt, on the other hand, switches to SSL/TLS early in the VNC protocol handshake. x11vnc 0.9.6 supports both simultaneously when -ssl is active.
Note: With the advent of OpenSSL 1.1.0, SSLv2 is dropped and SSLv3 deactivated per default. A couple broken ciphers have also gone, most importantly though is that clients trying to connect to x11vnc will now have to support TLS if encryption is to be used. You can of course always cook up your own build and run time OpenSSL 1.1.x if SSLv3 is absolutely required, but it isn't wise from a security point of view.
SSL VNC Viewers:. Viewer-side will need to use SSL as well. See the next FAQ and here for SSL enabled VNC Viewers, including SSVNC, to connect to the above x11vnc via SSL.
As seen above, the PEM (privacy enhanced mail) file does not need to be supplied if the openssl(1) command is available in PATH, in that case a self-signed, certificate good the current and subsequent x11vnc sessions is created (this may take a while on very slow machines.)
In general, the PEM file contains both the Certificate (i.e. public key) and the Private Key. Because of the latter, the file should be protected from being read by untrusted users. The best way to do this is to encrypt the key with a passphrase (note however this requires supplying the passphrase each time x11vnc is started up.)
See the discussion on x11vnc Key Management for some utilities provided for creating and managing certificates and keys and even for creating your own Certificate Authority (CA) for signing VNC server and client certificates. This may be done by importing the certificate into Web Browser or Java plugin keystores, or pointing stunnel to it. The wrapper script ss_vncviewer provides an example on unix (see the -verify option.)
Here are some notes on the simpler default (non-CA) operation. To have x11vnc save the generated certificate and key, use the "SAVE" keyword like this:
x11vnc -ssl SAVE -display :0 ...
(this is the same as the default: "-ssl".) This way it will be saved in the default directory ~/.vnc/certs/ as server.crt (the certificate only) and server.pem (both certificate and private key.) This opens up the possibility of copying the server.crt to machines where the VNC Viewer will be run to enable authenticating the x11vnc SSL VNC server to the clients. When authentication takes place this way (or via the more sophisticated CA signing described here), then Man-In-The-Middle-Attacks are prevented. Otherwise, the SSL encryption only provides protection against passive network traffic "sniffing" (i.e. you are not protected against M-I-T-M attacks.) Nowadays, most people seem mostly concerned mainly about passive sniffing (and the default x11vnc SSL mode protects against it.) Note that there are hacker tools like dsniff/webmitm and cain that implement SSL Man-In-The-Middle attacks. They rely on the client not bothering to check the cert.
One can test to some degree that SSL is working after starting x11vnc with the -stunnel or -ssl option. From another machine one can use the openssl command something like this:
openssl s_client -debug -msg -showcerts -connect far-away.east:5900
After all of the debugging output and informational messages you'll see the string "RFB 003.008" that came from x11vnc. Pointing a web browser connecting to: https://far-away.east:5900/ and then viewing the SSL certificate information about the connection in the panels will also work.
Note: If you serve up the SSL enabled Java VNC Viewer via something like:
x11vnc -ssl -httpdir /usr/local/share/x11vnc/classes/ssl
(or just the -http option), you can test it out completely using that, including using https to download it into the browser and connect to x11vnc.
The older -stunnel option: Before the -ssl option there was a convenience option -stunnel that would start an external SSL tunnel for you using stunnel. The -ssl method is the preferred way, but for historical reference we keep the -stunnel info here.
The -stunnel mode requires the stunnel.mirt.net command stunnel(8) to be installed on the system.
Some -stunnel examples:
x11vnc -display :0 -stunnel /path/to/stunnel.pem -passwdfile ~/mypass
x11vnc -display :0 -stunnel SAVE ...
You'll get output like this:
The VNC desktop is: localhost:50
The SSL VNC desktop is: far-away.east:0
PORT=5950
SSLPORT=5900
That indicates stunnel is listening on port 5900 for incoming SSL-wrapped VNC connections from viewers. x11vnc is listening for local connections on port 5950 in this case (remote viewers cannot connect to it directly.) For -stunnel to work the stunnel command must be installed on the machine and available in PATH (note stunnel is often installed in sbin directories rather than bin.) Note that the default "-stunnel" by itself creates a temporary cert (as in "-ssl TMP".)
Notes on using "native" VNC Viewers with SSL:
There aren't any native VNC Viewers that do SSL (ask your VNC viewer developer to add the feature.) So a tunnel must be setup that you point the VNC Viewer to. This is often STUNNEL. You can do this manually, or use the ss_vncviewer script on Unix, or our Enhanced TightVNC Viewer (SSVNC) package on Unix, Windows, or MacOSX. See the next section for Java Web browser SSL VNC Viewers (you only need a Java-enabled Web browser for it to work.)
Notes on the SSL enabled Java VNC Viewer provided in x11vnc classes/ssl/VncViewer.jar:
A Java applet VNC Viewer allows you to connect to a VNC Server from a Java-enabled Web browser.
The SSL enabled Java VNC Viewer (VncViewer.jar) in the x11vnc package supports only SSL based connections by default. As mentioned above the -httpdir can be used to specify the path to .../classes/ssl. A typical location might be /usr/local/share/x11vnc/classes/ssl. Or -http can be used to try to have it find the directory automatically.
Also note that the SingleClick UltraVNC Java Viewer is compatible with x11vnc's -ssl SSL mode. (We tested it this way: "java -cp ./VncViewer.jar VncViewer HOST far-away.east PORT 5900 USESSL 1 TRUSTALL 1")
The Java viewer uses SSL to communicate securely with x11vnc. Note that the applet can optionally also be downloaded into your web browser via HTTPS (which is HTTP over SSL.) This way the HTML page and the Java applet itself are also delivered securely with SSL (as opposed to only the VNC traffic being encrypted with SSL.)
For this case the output will be something like this:
x11vnc -ssl SAVE -http
...
The SSL VNC desktop is: far-away.east:0
Java SSL viewer URL: https://far-away.east:5900/
Java SSL viewer URL: http://far-away.east:5800/
PORT=5900
SSLPORT=5900
Indicating the two URLs (the first one encrypted, the second not) one could point the web browser at to get the VNC viewer applet. E.g. put this
http://far-away.east:5800/
or:
https://far-away.east:5900/
into your Java-enabled Web browser.
Note that KDE's Konqueror web browser seems to have problems with https Java applets, so you'll have to use the http/5800 with it (if you get https/5900 working let us know how you did it.)
If you are using a router/firewall with port-redirection, and you are redirecting ports other than the default ones (5800, 5900) listed above see here.
The https service provided thru the actual VNC port (5900 in the above example) can occasionally be slow or unreliable (it has to read some input and try to guess if the connection is VNC or HTTP.) If it is unreliable for you and you still want to serve the Java applet via https, use the -https option to get an additional port dedicated to https (its URL will also be printed in the output.)
Another possibility is to add the GET applet parameter:
https://far-away.east:5900/?GET=1
This will have the VNC Viewer send a special HTTP GET string "GET /request.https.vnc.connection HTTP/1.0" that x11vnc will notice more quickly as a request for a VNC connection. Otherwise it must wait for a timeout to expire before it assumes a VNC connection.
You may also use "urlPrefix=somestring" to have /somestring prepended to /request.https.vnc.connection". Perhaps you are using a web server proxy scheme to enter a firewall or otherwise have rules applied to the URL. If you need to have any slashes "/" in "somestring" use "2F" (a deficiency in libvncserver prevents using the more natural "%2F".)
You apply multiple applet parameters in the regular URL way, e.g.:
https://far-away.east:5900/?GET=1&urlPrefix=mysubdir&...
All of the x11vnc Java Viewer applet parameters are described in the file classes/ssl/README
Tips on Getting the SSL Java Applet Working the First Time: Unfortunately, it can be a little tricky getting the SSL VNC Java Viewer working with x11vnc. Here are some tips to getting working the first time (afterwards you can incrementally customize with more complex settings.)
- First try it on the LAN: Do NOT try to have it work the first time going through firewalls, Web proxies, home router port redirections, or Apache portal. Just try a direct connection over your LAN first (if you only have 1 machine and no LAN, just do a direct connection to the same machine: localhost.) If the LAN machine you run x11vnc on has its own host-level firewall (most linux machine come with that on by default), disable it or at least let tcp ports 5800-6000 through.
- First try HTTP to download the Java Applet: x11vnc can serve both the Java Applet jar file and VNC out of the same port (both tunneled through SSL, see below.) But it can lead to timing and other problems. So first try HTTP instead of HTTPS to download the Applet jar file (VncViewer.jar.) That is to say try http://hostname:5800 in your web browser first before trying https://hostname:5900. x11vnc will print out the ports and URLs it is using, so use the HTTP one it prints out.
- Always Restart the Browser: If you are having failures and have to repeatedly retry things ALWAYS restart the browser (i.e. completely exit it and then start a new browser process) each time. Otherwise as you are changing things the browser may "remember" failed applet downloads, etc. and just add to the confusion and irreproducibility. If you see it trying to download VncViewer.class (instead of VncViewer.jar) you know it is really confused and needs to be restarted.
- Step Lively: If you get Browser or Java VM or VNC Viewer applet dialog boxes saying things like "Do you want to trust this certificate?" or "The hostname does not match the one on the certificate", etc. just go through them as quickly as possible. x11vnc cannot wait forever for each SSL connection, and so if you dawdle too long inspecting the certs, etc it can lead to problems. Get it working first before taking your time to read the details in the dialogs, etc.
- No inetd, Please: Even if you intend to deploy via inetd or xinetd eventually, get that working later (and remember do not use something like "-ssl TMP" that creates a new temporary SSL certificate for every new socket connection.)
- Nothing Fancy: Do not try fancy stuff like -svc, -create, -unixpw, "-users unixpw=", "-users sslpeer=", -sslverify, etc. Just get the simplest connection working first and then incrementally add what you need.
So the recommended test command lines are:
x11vnc -ssl SAVE -http
x11vnc -ssl SAVE -httpdir /path/to/x11vnc/classes/ssl
Use the latter if x11vnc cannot automatically find the classes/ssl directory (this what the -http option instructs it to do.) Then point your browser to the HTTP (not HTTPS) URL it prints out.
Following the above guidelines, did it work? If so, Congratulations!! you created an SSL encrypted connection between the SSL Java applet running in your web browser and x11vnc. The fact that you used HTTP instead of HTTPS to download the applet is not the end of the world (some users do it this way), the main thing is that the VNC traffic is encrypted with SSL. If you are having trouble even with the above baseline test case feel free to contact me (please send the Full x11vnc output, not just part of it; the complete x11vnc command line; the URL(s) entered in the browser; the full Java Console output; and anything else you can think of.)
Next, you can add the features you want one by one testing it still works each time. I suggest first turning on the HTTPS applet download (https://hostname:5900) if that is what you intend to use. That one gives the most trouble because of the ambiguity of passing two different protocols (HTTP and VNC) through the same SSL service port.
Next, turn on inetd if you intend to use that (this can be tricky too, be sure to use -oa logfile and inspect it carefully if there are problems.) If you are going to use non-standard ports (e.g. "-rfbport 443" as root), work on that next. Then enable the firewall, router port redirection channel (you will somehow need to be outside to do that, maybe test that through another VNC session.)
Then, if you plan to use them, enable "fancy stuff" like "-svc" or "-unixpw", etc, etc. Be sure to add a password either "-rfbauth" or "-unixpw" or both. If you need to have the web browser use a corporate Web Proxy (i.e. it cannot connect directly) work on that last. Ditto for the Apache portal.
Router/Firewall port redirs: If you are doing port redirection at your router to an internal machine running x11vnc AND the internet facing port is different from the internal machine's VNC port, you will need to apply the PORT applet parameter to indicate to the applet the Internet facing port number (otherwise by default the internal machine's port, say 5900, is sent and that of course is rejected at the firewall/router.) For example:
https://far-away.east:443/?GET=1&PORT=443
So in this example the user configures his router to redirect connections to port 443 on his Internet side to, say, port 5900 on the internal machine running x11vnc. See also the -httpsredir option that will try to automate this for you.
To configure your router to do port redirection, see its instructions. Typically, from the inside you point a web browser to a special URL (e.g. http://192.168.1.1) and you get a web interface to configure it. Look for something like "Port Redirection" or "Port Forwarding", probably under "Advanced" or something like that. If you have a Linux or Unix system acting as your firewall/router, see its firewall configuration.
You can also use x11vnc options -rfbport NNNNN and -httpport NNNNN to match the ports that your firewall will be redirecting to the machine where x11vnc is run.
Tedious Dialogs: If you do serve the SSL enabled Java viewer via https be prepared for quite a number of "are you sure you trust this site?" dialogs:
- First from the Web browser that cannot verify the self-signed certificate when it downloads index.vnc.
- From the Web browser again noting that the common name on the certificate does not match the hostname of the remote machine.
- Next from the Java VM that cannot verify the self-signed certificate when it downloads VncViewer.jar.
- And also from the Java VM again noting that the common name on the certificate does not match the hostname of the remote machine.
- Finally from the Java VncViewer applet itself saying it cannot verify the certificate! (or a popup asking you if you want to see the certificate.)
Note that sometimes if you pause too long at one of the above dialogs then x11vnc may exceed a timeout and assume the current socket connection is VNC instead of the HTTPS it actually is (but since you have paused too long at the dialog the GET request comes too late.) Often hitting Reload and going through the dialogs more quickly will let you connect. The Java VM dialogs are the most important ones to NOT linger at. If you see in the x11vnc output a request for VncViewer.class instead of VncViewer.jar it is too late... you will need to completely restart the Web browser to get it to try for the jar again. You can use the -https option if you want a dedicated port for HTTPS connections instead of sharing the VNC port.
To see example x11vnc output for a successful https://host:5900/ connection with the Java Applet see This Page. And here is a newer example including the Java Console output.
All of the x11vnc Java Viewer applet parameters are described in the file classes/ssl/README
Notes on the VNC Viewer ss_vncviewer wrapper script:
If you want to use a native VNC Viewer with the SSL enabled x11vnc you will need to run an external SSL tunnel on the Viewer side. There do not seem to be any native SSL VNC Viewers outside of our x11vnc and SSVNC packages. The basic ideas of doing this were discussed for external tunnel utilities here.
The ss_vncviewer script provided with x11vnc and SSVNC can set up the stunnel tunnel automatically on unix as long as the stunnel command is installed on the Viewer machine and available in PATH (and vncviewer too of course.) Note that on a Debian based system you will need to install the package stunnel4 not stunnel. You can set the environment variables STUNNEL and VNCVIEWERCMD to point to the correct programs if you want to override the defaults.
Here are some examples:
1) ss_vncviewer far-away.east:0
2) ss_vncviewer far-away.east:0 -encodings "copyrect tight zrle hextile"
3) ss_vncviewer -verify ./server.crt far-away.east:0
4) ss_vncviewer -mycert ./client.pem far-away.east:0
5) ss_vncviewer -proxy far-away.east:8080 myworkstation:0
The first one is the default mode and accepts the x11vnc certificate without question. The second one is as the first, but adds the -encodings options to the vncviewer command line.
The third one requires that the x11vnc server authenticate itself to the client against the certificate in the file ./server.crt (e.g. one created by "x11vnc -ssl SAVE" and safely copied to the VNC viewer machine.)
The fourth one is for VNC Viewer authentication, it uses ./client.pem to authenticate itself to x11vnc. One can supply both -verify and -mycert simultaneously.
The fifth one shows that Web proxies can be used if that is the only way to get out of the firewall. If the "double proxy" situation arises separate the two by commas. See this page for more information on how Web proxies come into play.
If one uses a Certificate Authority (CA) scheme described here, the wrapper script would use the CA cert instead of the server cert:
3') ss_vncviewer -verify ./cacert.crt far-away.east:0
Update Jul/2006: we now provide an Enhanced TightVNC Viewer (SSVNC) package that starts up STUNNEL automatically along with some other features. All binaries (stunnel, vncviewer, and some utilities) are provided in the package. It works on Unix, Mac OS X, and Windows.
Q-55: How do I use the Java applet VNC Viewer with built-in SSL tunneling when going through a Web Proxy?
The SSL enabled Java VNC Viewer and firewall Proxies:
SSL and HTTPS aside, there is a general problem with Firewall Proxies and Java Applets that open sockets. The applet is downloaded successfully (through the browser) using HTTP and the proxy, but when the applet tries to reconnect to the originating host (the only one allowed by security) it does not use the proxy channel. So it cannot reconnect to the server the applet came from!
We have found a convenient workaround: in the directory where VncViewer.jar resides there is a digitally signed version of the same applet called SignedVncViewer.jar. Since the applet is digitally signed, there will be an additional dialog from the Java VM plugin asking you if you want to trust the applet fully.
You should say "Yes". If you do, the applet will be run in a mode where it can try to determine the firewall proxy host name and port (it will ask you for them if it cannot find them.) This way it can connect directly to the Proxy and then request the CONNECT method to be redirected to the originating host (the x11vnc VNC Server.) SSL is then layered over this socket.
To do this you should use the proxy.vnc HTML file like via this URL in your browser:
https://yourmachine.com:5900/proxy.vnc
(instead of the unsigned one in https://yourmachine.com:5900/ that gives the default index.vnc)
Proxies that limit CONNECT to ports 443 and 563:
Things become trickier if the Web proxy restricts which CONNECT ports can be redirected to. For security, some (most?) proxies only allow port 443 (HTTPS) and 563 (SNEWS) by default. In this case, the only thing to do is run x11vnc on that low port, e.g. "-rfbport 443", (or use a port redirection on, say, a firewall or router port 443 to the internal machine.)
If you do such a redirection to an internal machine and x11vnc is not listening on port 443, you will probably need to edit proxy.vnc. Suppose the SSL x11vnc server was listening on port 5901. You should change the line in proxy.vnc from:
<param name=PORT value=$PORT>
to:
<param name=PORT value=443>
Since otherwise $PORT will be expanded to 5901 by x11vnc and the viewer applet will fail to connect to that port on the firewall.
Another way to achieve the same thing is to use the applet PORT parameter:
https://yourmachine.com/proxy.vnc?PORT=443
this is cleaner because it avoids editing the file, but requires more parameters in the URL. See also the -httpsredir x11vnc option that will try to automate this for you. To use the GET trick discussed above, do:
https://yourmachine.com/proxy.vnc?GET=1&PORT=443
All of the x11vnc Java Viewer applet parameters are described in the file classes/ssl/README
Here is an example of Java Console and x11vnc output for the Web proxy case.
Note that both the ss_vncviewer stunnel Unix wrapper script and Enhanced TightVNC Viewer (SSVNC) can use Web proxies as well even though they do not involve a Web browser.
Q-56: Can Apache web server act as a gateway for users to connect via SSL from the Internet with a Web browser to x11vnc running on their workstations behind a firewall?
Yes. You will need to configure apache to forward these connections. It is discussed here. This SSL VNC portal provides a clean alternative to the traditional method where the user uses SSH to log in through the gateway to create the encrypted port redirection to x11vnc running on her desktop.
Also see the desktop.cgi CGI script method that achieves much of what this Apache VNC SSL portal method does (as long as desktop.cgi's 'port redirection' mode is enabled.)
Yes, see this page for how to do this and the utility commands x11vnc provides to create and manage many types of certificates and private keys.
There are a number of ways to do this. The primary thing you need to decide is whether you want x11vnc to connect to the X session on the machine 1) regardless of who (or if anyone) has the X session, or 2) only if a certain user has the X session. Because X sessions are protected by X permissions (MIT-MAGIC-COOKIE files XAUTHORITY and $HOME/.Xauthority) the automatically started x11vnc will of course need to have sufficient permissions to connect to the X display.
Here are some ideas:
- Use the description under "Continuously" in the FAQ on x11vnc and Display Managers
- Use the description in the FAQ on x11vnc and inetd(8)
- Use the description in the FAQ on Unix user logins and inetd(8)
- Start x11vnc from your $HOME/.xsession (or $HOME/.xinitrc or autostart script or ...)
- Although less reliable, see the x11vnc_loop rc.local hack below.
The display manager scheme will not be specific to which user has the X session unless a test is specifically put into the display startup script (often named Xsetup.) The inetd(8) scheme may or may not be specific to which user has the X session (and it may not be able to do all users via the XAUTHORITY permission issues.)
The .xsession/.xinitrc scheme is obviously is specific to a particular user and only when they are logged into X. If you do not know what a $HOME/.xsession script is or how to use one, perhaps your desktop has a "session startup commands" configuration option. The command to be run in the .xsession or .xinitrc file may look like this:
x11vnc -logfile $HOME/.x11vnc.log -rfbauth $HOME/.vnc/passwd -forever -bg
plus any other options you desire.
Depending on your desktop and/or OS/distribution the automatically run X startup scripts (traditionally .xsession/.xinitrc) may have to be in a different directory or have a different basename. One user recommends the description under 'Running Scripts Automatically' at this link.
Firewalls: note all methods will require the host-level firewall to be configured to allow connections in on a port. E.g. 5900 (default VNC port) or 22 (default SSH port for tunnelling VNC.) Most systems these days have firewalls turned on by default, so you will actively have to do something to poke a hole in the firewall at the desired port number. See your system administration tool for Firewall settings (Yast, Firestarter, etc.)
Q-59: How can I use x11vnc to connect to an X login screen like xdm, GNOME gdm, KDE kdm, or CDE dtlogin? (i.e. nobody is logged into an X session yet.)
We describe two scenarios here. The first is called 'One time only' meaning you just need to do it quickly once and don't want to repeat; and the second is called 'Continuously' meaning you want the access to be available after every reboot and after every desktop logout. ---
One time only: If the X login screen is running and you just want to connect to it once (i.e. a one-shot):
It is usually possible to do this by just adjusting the XAUTHORITY environment variable to point to the correct MIT-COOKIE auth file while running x11vnc as root, e.g. for the gnome display manager, GDM:
x11vnc -auth /var/gdm/:0.Xauth -display :0
(the -auth option sets the XAUTHORITY variable for you.)
There will be a similar thing to do for xdm using however a different auth directory path (perhaps something like /var/lib/xdm/authdir/authfiles/A:0-XQvaJk) for the xdm greeter or /var/lib/kdm/A:0-crWk72 (or /var/run/xauth/A:0-qQPftr, etc. etc) for the kdm greeter. Of course, the random characters in the file basename will vary and you will need to use the actual filename on your system. Read your system docs to find out where the display manager cookie files are kept.
Trick: sometimes ps(1) can reveal the X server process -auth argument (e.g. "ps wwaux | grep auth") and hence the path to the auth file.
x11vnc must be run as root for this because the /var/gdm/:0.Xauth, /var/lib/kdm/A:0-crWk72, etc. auth files are only readable by root. If you do not want to run x11vnc as root, you can copy (as root or sudo) the auth file to some location and make it readable by your userid. Then run x11vnc as your userid with -auth pointed to the copied file.
Update Dec/2009: use "-auth guess" to have x11vnc try to guess the location of the auth file for you.
You next connect to x11vnc with a VNC viewer, give your username and password to the X login prompt to start your session.
Note: GDM: gdm seems to have an annoying setting that causes x11vnc (and any other X clients) to be killed after the user logs in. Setting KillInitClients=false in the [daemon] section of /etc/X11/gdm/gdm.conf (or /etc/gdm/gdm.conf, etc.) avoids this. Otherwise, just restart x11vnc and then reconnect your viewer. Other display managers (kdm, etc) may also have a similar problem. One user reports having to alter "gdm.conf-custom" as well.
Note: Solaris: For dtlogin in addition to the above sort of trick (BTW, the auth file should be in /var/dt), you'll also need to add something like Dtlogin*grabServer:False to the Xconfig file (/etc/dt/config/Xconfig or /usr/dt/config/Xconfig on Solaris, see the example at the end of this FAQ.) Then restart dtlogin, e.g.: /etc/init.d/dtlogin stop; /etc/init.d/dtlogin start or reboot.
Update Nov/2008: Regarding GDM KillInitClients: see the -reopen option for another possible workaround.
Update Oct/2009: Regarding GDM KillInitClients: starting with x11vnc 0.9.9 it will try to apply heuristics to detect if a window manager is not running (i.e. whether the Display Manager Greeter Login panel is still up.) If it thinks the display manager login is still up it will delay creating windows or using XFIXES. The former is what GDM uses to kill the initial clients, use of the latter can cause a different problem: an Xorg server crash. So with 0.9.9 and later it should all work without needing to set KillInitClients=false (which is a good because recent GDM, v2.24, has removed this option) or use -noxfixes. To disable the heuristics and delaying set X11VNC_AVOID_WINDOWS=never; to set the delay time explicitly use, e.g., X11VNC_AVOID_WINDOWS=120 (delays for 120 seconds after the VNC connection; you have that long to log in.)
Continuously: Have x11vnc reattach each time the X server is restarted (i.e. after each logout and reboot):
To make x11vnc always attached to the X server including the login screen you will need to add a command to a display manager startup script.
Please consider the security implications of this! The VNC display for the X session always accessible (but hopefully password protected.) Add -localhost if you only plan to access via a SSH tunnel.
The name of the display manager startup script file depends on desktop used and seem to be:
GDM (GNOME) /etc/X11/gdm/Init/Default
/etc/gdm/Init/Default
KDM (KDE) /etc/kde*/kdm/Xsetup
XDM /etc/X11/xdm/Xsetup (or sometimes xdm/Xsetup_0)
CDE /etc/dt/config/Xsetup
although the exact location can be operating system, distribution, and time dependent. See the documentation for your display manager: gdm(1), kdm(1), xdm(1), dtlogin(1) for additional details. There may also be display number specific scripts: e.g. Xsetup_0 vs. Xsetup, you need to watch out for.
Note: You should read and understand all of the Note's and Update's in the 'One time only' section above. All of the GDM topics apply here as well:
Note: GDM: The above (in 'One time only') gdm setting of KillInitClients=false in /etc/X11/gdm/gdm.conf (or /etc/gdm/gdm.conf, etc.) for GDM is needed here as well. Other display managers (KDM, etc) may also have a similar problem.
Also see the Update Oct/2009 above where x11vnc 0.9.9 and later automatically avoids being killed.
Note: DtLogin: The above (in 'One time only') Dtlogin*grabServer:False step for Solaris will be needed for dtlogin here as well.
In any event, the line you will add to the display manager script (Xsetup, Default, or whatever) will look something like:
/usr/local/bin/x11vnc -rfbauth /path/to/the/vnc/passwd -o /var/log/x11vnc.log -forever -bg
where you should customize the exact command to your needs (e.g. -localhost for SSH tunnel-only access; -ssl SAVE for SSL access; etc.)
Happy, happy, joy, joy: Note that we do not need to specify -display or -auth because happily they are already set for us in the DISPLAY and XAUTHORITY environment variables for the Xsetup script!!!
You may also want to force the VNC port with something like "-rfbport 5900" (or -N) to avoid autoselecting one if 5900 is already taken.
Fedora/gdm: Here is an example of what we did on a vanilla install of Fedora-C3 (seems to use gdm by default.) Add a line like this to /etc/X11/gdm/Init/:0
/usr/local/bin/x11vnc -rfbauth /etc/x11vnc.passwd -forever -bg -o /var/log/x11vnc.log
And then add this line to /etc/X11/gdm/gdm.conf (or /etc/gdm/gdm.conf, etc.) in the [daemon] section:
KillInitClients=false
Then restart: /usr/sbin/gdm-restart (or reboot.) The KillInitClients=false setting is important: without it x11vnc will be killed immediately after the user logs in. Here are full details on how to configure gdm ---
Solaris/dtlogin: Here is an example of what we did on a vanilla install of Solaris: Make the directory /etc/dt/config:
mkdir -p /etc/dt/config
Copy over the Xconfig file for customization:
cp /usr/dt/config/Xconfig /etc/dt/config/Xconfig
Edit /etc/dt/config/Xconfig and uncomment the line:
Dtlogin*grabServer: False
Next, copy over Xsetup for customization:
cp /usr/dt/config/Xsetup /etc/dt/config/Xsetup
Edit /etc/dt/config/Xsetup and at the bottom put a line like:
/usr/local/bin/x11vnc -forever -o /var/log/x11vnc.log -bg
(tweaked to your local setup and preferences, a password via -rfbauth, etc. would be a very good idea.)
Restart the X server and dtlogin:
/etc/init.d/dtlogin stop
/etc/init.d/dtlogin start
(or reboot or maybe just restart the X session.)
---
KDM: One user running the kdm display manager reports putting this line:
x11vnc -forever -rfbauth /home/xyz/.vnc/passwd -bg -o /var/log/x11vnc.log
in /etc/kde/kdm/Xsetup. After rebooting the system it all seemed to work fine.
---
If you do not want to deal with any display manager startup scripts, here is a kludgey script that can be run manually or out of a boot file like rc.local: x11vnc_loop It will need some local customization before running. Because the XAUTHORITY auth file must be guessed by this script, use of the display manager script method described above is greatly preferred. There is also the -loop option that does something similar.
If the machine is a traditional Xterminal you may want to read this FAQ.
Firewalls: note all methods will require the host-level firewall to be configured to allow connections in on a port. E.g. 5900 (default VNC port) or 22 (default SSH port for tunnelling VNC.) Most systems these days have firewalls turned on by default, so you will actively have to do something to poke a hole in the firewall at the desired port number. See your system administration tool for Firewall settings (Yast, Firestarter, etc.)
Yes, perhaps a line something like this in /etc/inetd.conf will do it for you:
5900 stream tcp nowait root /usr/sbin/tcpd /usr/local/bin/x11vnc_sh
where the shell script /usr/local/bin/x11vnc_sh uses the -inetd option and looks something like (you'll need to customize to your settings.)
#!/bin/sh
/usr/local/bin/x11vnc -inetd -display :0 -auth /home/fred/.Xauthority \
-rfbauth /home/fred/.vnc/passwd -o /var/log/x11vnc_sh.log
Important: Note that you must redirect the standard error output to a log file (e.g. -o logfile) or "2>/dev/null" for proper operation via inetd (otherwise the standard error also goes to the VNC vncviewer, and that confuses it greatly, causing it to abort.) If you do not use a wrapper script as above but rather call x11vnc directly in /etc/inetd.conf and do not redirect stderr to a file, then you must specify the -q (aka -quiet) option: "/usr/local/bin/x11vnc -q -inetd ...". When you supply both -q and -inet and no "-o logfile" then stderr will automatically be closed (to prevent, e.g. library stderr messages leaking out to the viewer.) The recommended practice is to use "-o logfile" to collect the output in a file or wrapper script with "2>logfile" redirection because the errors and warnings printed out are very useful in troubleshooting problems.
Note also the need to set XAUTHORITY via -auth to point to the MIT-COOKIE auth file to get permission to connect to the X display (setting and exporting the XAUTHORITY variable accomplishes the same thing.) See the x11vnc_loop file in the previous question for more ideas on what that auth file may be, etc. The scheme described in the FAQ on Unix user logins and inetd(8) works around the XAUTHORITY issue nicely.
Note: On Solaris you cannot have the bare number 5900 in /etc/inetd.conf, you'll need to replace it with a word like x11vnc an then put something like "x11vnc 5900/tcp" in /etc/services.
Since the process runs as root, it might be a bad idea to have the logfile in a world-writable area like /tmp if there are untrustworthy users on the machine. Perhaps /var/log is a better place.
Be sure to look at your /etc/hosts.allow and /etc/hosts.deny settings to limit the machines that can connect to this service (your desktop!) For the above example with /etc/hosts.allow:
x11vnc_sh : 123.45.67.89
A really safe way to do things is to limit the above inetd to localhost only (via /etc/hosts.allow) and use ssh to tunnel the incoming connection. Using inetd for this prevents there being a tiny window of opportunity between x11vnc starting up and your vncviewer connecting to it. Always use a VNC password to further protect against unwanted access.
For xinetd(8), one user reports he created the file /etc/xinetd.d/x11vncservice containing the following:
# default: off
# description:
service x11vncservice
{
flags = REUSE NAMEINARGS
port = 5900
type = UNLISTED
socket_type = stream
protocol = tcp
wait = no
user = root
server = /usr/sbin/tcpd
server_args = /usr/local/bin/x11vnc_sh
disable = no
}
With the contents of /usr/local/bin/x11vnc_sh similar to the example given above. One user reports this works with avoiding the wrapper script:
service x11vncservice
{
port = 5900
type = UNLISTED
socket_type = stream
protocol = tcp
wait = no
user = root
server = /usr/local/bin/x11vnc
server_args = -inetd -q -display :0 -auth /var/gdm/:0.Xauth
disable = no
}
(or one can replace the -q with say "-o /var/log/x11vnc.log" to capture a log)
The above works nicely for GDM because the -auth file is a fixed name. For KDM or XDM the filename varies. Here is one idea for a x11vnc_sh wrapper to try to guess the name:
#!/bin/sh
COLUMNS=256
export COLUMNS
authfile=`ps wwaux | grep '/X.*-auth' | grep -v grep | sed -e 's/^.*-auth *//'
-e 's/ .*$//' | head -n 1`
if [ -r "$authfile" ]; then
exec /usr/local/bin/x11vnc -inetd -o /var/log/x11vnc.log -display :0 -a
uth "$authfile"
fi
exit 1
Starting with x11vnc 0.9.3 this can be automated by:
#!/bin/sh
exec /usr/local/bin/x11vnc -inetd -o /var/log/x11vnc.log -find -env FD_XDM=1
Q-61: Can I have x11vnc advertise its VNC service and port via mDNS/Zeroconf (e.g. Avahi) so VNC viewers on the local network can detect it automatically?
Yes, as of Feb/2007 x11vnc supports mDNS / Zeroconf advertising of its service via the Avahi client library. Use the option -avahi (same as -mdns or -zeroconf) to enable it. Depending on your setup you may need to install Avahi (including the development/build packages), enable the server: avahi-daemon and avahi-dnsconfd, and possibly open up UDP port 5353 on your firewall.
If the Avahi client library or build environment is not available at build-time, then at run-time x11vnc will try to look for external helper programs, avahi-browse(1) or dns-sd(1), to do the work.
The service was tested with Chicken of the VNC ("Use Bonjour" selected) on a Mac on the same network and the service was noted and listed in the servers list. Clicking on it and then "Connect" connected automatically w/o having to enter any hostnames or port numbers.
It appears SuSE 10.1 comes with avahi (or you can add packages, e.g. avahi-0.6.5-27) but not the development package (you can use the OpenSuSE avahi-devel rpm.) Unfortunately, you may need to disable another Zeroconf daemon "/etc/init.d/mdnsd stop", before doing "/etc/init.d/avahi-daemon start" and "/etc/init.d/avahi-dnsconfd start". We also had to comment out the browse-domains line in /etc/avahi/avahi-daemon.conf. Hopefully there is "LessConf" to do on other distros/OS's...
Q-62: Can I have x11vnc allow a user to log in with her UNIX username and password and then have it find her X session display on that machine and then attach to it? How about starting an X session if one cannot be found?
The easiest way to do this is via inetd(8) using the -unixpw and -display WAIT options. The reason inetd(8) makes this easier is that it starts a new x11vnc process for each new user connection. Otherwise a wrapper would have to listen for connections and spawn new x11vnc's (see this example and also the -loopbg option.) inetd(8) is not required for this, but it makes some aspects more general.
Also with inetd(8) users always connect to a fixed VNC display, say hostname:0, and do not need to memorize a special VNC display number just for their personal use, etc.
Update: Use the -find, -create, -svc, and -xdmsvc options that are shorthand for common FINDCREATEDISPLAY usage modes (e.g. terminal services) described below. (i.e. simply use "-svc" instead of the cumbersome "-display WAIT:cmd=FINDCREATEDISPLAY-Xvfb -unixpw -users unixpw= -ssl SAVE")
The -display WAIT option makes x11vnc wait until a VNC viewer is connected before attaching to the X display.
Additionally it can be used to run an external command that returns the DISPLAY and XAUTHORITY data. We provide some useful builtin ones (FINDDISPLAY and FINDCREATEDISPLAY below), but in principle one could supply his own script: "-display WAIT:cmd=/path/to/find_display" where the script find_display might look something like this.
A default script somewhat like the above is used under "-display WAIT:cmd=FINDDISPLAY" (same as -find) The format for any such command is that it returns DISPLAY=:disp as the first line and any remaining lines are either XAUTHORITY=file or raw xauth data (the above example does the latter.) If applicable (-unixpw mode), the program is run as the Unix user name who logged in.
On Linux if the virtual terminal is known the program appends ",VT=n" to the DISPLAY line; a chvt n will be attempted automatically. Or if only the X server process ID is known it appends ",XPID=n" (a chvt will be attempted by x11vnc.)
Tip: Note that the -find option is an alias for "-display WAIT:cmd=FINDDISPLAY". Use it!
The -unixpw option allows UNIX password logins. It conveniently knows the Unix username whose X display should be found. Here are a couple /etc/inetd.conf examples of this usage:
5900 stream tcp nowait nobody /usr/sbin/tcpd /usr/local/bin/x11vnc -inetd -unixpw \
-find -o /var/log/x11vnc.log -ssl SAVE -ssldir /usr/local/certs
5900 stream tcp nowait root /usr/sbin/tcpd /usr/local/bin/x11vnc -inetd -unixpw \
-find -o /var/log/x11vnc.log -ssl SAVE -users unixpw=
Note we have used the -find alias and the very long lines have been split. An alternative is to use a wrapper script, e.g. /usr/local/bin/x11vnc.sh that has all of the options. (see also the -svc alias.)
In the first inetd line x11vnc is run as user "nobody" and stays user nobody during the whole session. The permissions of the log files and certs directory will need to be set up to allow "nobody" to use them.
In the second one x11vnc is run as root and switches to the user that logs in due to the "-users unixpw=" option.
Note that SSL is required for this mode because otherwise the Unix password would be passed in clear text over the network. In general -unixpw is not required for this sort of scheme, but it is convenient because it determines exactly who the Unix user is whose display should be sought. Otherwise the find_display script would have to use some method to work out DISPLAY, XAUTHORITY, etc (perhaps you use multiple inetd ports and hardwire usernames for different ports.)
If you really want to disable the SSL or SSH -localhost constraints (this is not recommended unless you really know what you are doing: Unix passwords sent in clear text is a very bad idea...) read the -unixpw documentation.
A inetd(8) scheme for a fixed user that doesn't use SSL or unix passwds could be:
/usr/local/bin/x11vnc -inetd -users =fred -find -rfbauth /home/fred/.vnc/pass
wd -o /var/log/x11vnc.log
The "-users =fred" option will cause x11vnc to switch to user fred and then find his X display. The VNC password (-rfbauth) as opposed to Unix password (-unixpw) is used to authenticate the VNC client.
Similar looking commands to the above examples can be run directly and do not use inetd (just remove the -inetd option and run from the cmdline, etc.)
X Session Creation: An added (Nov/2006) extension to FINDDISPLAY is FINDCREATEDISPLAY where if it does not find an X display via the FINDDISPLAY method it will create an X server session for the user (i.e. desktop/terminal server.) This is the only time x11vnc actually tries to start up an X server (normally it just attaches to an existing one.)
For virtual sessions you will need to install the Xvfb program (e.g. apt-get install xvfb) or our Xdummy program (see below.)
By default it will only try to start up virtual (non-hardware) X servers: first Xvfb and if that is not available then Xdummy (included in the x11vnc source code.) Note that Xdummy only works on Linux whereas Xvfb works just about everywhere (and in some situations Xdummy must be run as root.) An advantage of Xdummy over Xvfb is that Xdummy supports RANDR dynamic screen resizing, which can be handy if the user accesses the desktop from different sized screens (e.g. workstation and laptop.)
So an inetd(8) example might look like:
5900 stream tcp nowait root /usr/sbin/tcpd /usr/local/bin/x11vnc -inetd \
-o /var/log/x11vnc.log -http -prog /usr/local/bin/x11vnc \
-ssl SAVE -unixpw -users unixpw= -display WAIT:cmd=FINDCREATEDISPLAY
Where the very long lines have been split. See below where that long and cumbersome last line is replaced by the -svc alias.
The above mode will allow direct SSL (e.g. ss_vncviewer or SSVNC) access and also Java Web browers access via: https://hostname:5900/.
Tip: Note that the -create option is an alias for "-display WAIT:cmd=FINDCREATEDISPLAY-Xvfb".
Tip: Note that -svc is a short hand for the long "-ssl SAVE -unixpw -users unixpw= -display WAIT:cmd=FINDCREATEDISPLAY" part. Unlike -create, this alias also sets up SSL encryption and Unix password login.
The above inetd example then simplifies to:
5900 stream tcp nowait root /usr/sbin/tcpd /usr/local/bin/x11vnc -inetd \
-o /var/log/x11vnc.log -http -prog /usr/local/bin/x11vnc \
-svc
Tip: In addition to the usual unixpw parameters, inside the VNC viewer the user can specify after his username (following a ":" see -display WAIT for details) for FINDCREATEDISPLAY they can add "geom=WxH" or "geom=WxHxD" to specify the width, height, and optionally the color depth. E.g. "fred:geom=800x600" at the login: prompt. Also if the env. var X11VNC_CREATE_GEOM is set to the desired WxH or WxHxD that will be used by x11vnc.
You can set the env. var X11VNC_SKIP_DISPLAY to a comma separated list of displays to ignore in the FINDDISPLAY process (to force creation of new displays in some cases.) The user logging in via the vncviewer can also set this via username:nodisplay=...)
If you do not plan on using the Java Web browser applet you can remove the -http (and -prog) option since this will speed up logging-in by a few seconds (x11vnc will not have to wait to see if a connection is HTTPS or VNC.)
For reference, xinetd format in the file, say, /etc/xinetd.d/x11vnc:
service x11vnc
{
type = UNLISTED
port = 5900
socket_type = stream
protocol = tcp
wait = no
user = root
server = /usr/local/bin/x11vnc
server_args = -inetd -o /var/log/x11vnc.log -http -prog /usr/local/
bin/x11vnc -svc
disable = no
}
To print out the script in this case use "-display WAIT:cmd=FINDCREATEDISPLAY-print". To change the preference of Xservers and which to try list them, e.g.: "-display WAIT:cmd=FINDCREATEDISPLAY-X,Xvfb,Xdummy" or use "-create_xsrv X,Xvfb,Xdummy". The "X" one means to try to start up a real, hardware X server, e.g. startx(1) (if there is already a real X server running this may only work on Linux and the chvt program may need to be run to switch to the correct Linux virtual terminal.) x11vnc will try to run chvt automatically if it can determine which VT should be switched to.
XDM/GDM/KDM Login Greeter Panel: If you want to present the user with a xdm/gdm/kdm display manager "greeter" login you can use Xvfb.xdmcp instead of Xvfb, etc in the above list. However, you need to configure xdm/gdm/kdm to accept localhost XDMCP messages, this can be done by (from -help output):
If you want the FINDCREATEDISPLAY session to contact an XDMCP login
manager (xdm/gdm/kdm) on the same machine, then use "Xvfb.xdmcp"
instead of "Xvfb", etc. The user will have to supply his username
and password one more time (but he gets to select his desktop
type so that can be useful.) For this to work, you will need to
enable localhost XDMCP (udp port 177) for the display manager.
This seems to be:
for gdm in gdm.conf: Enable=true in section [xdmcp]
for kdm in kdmrc: Enable=true in section [Xdmcp]
for xdm in xdm-config: DisplayManager.requestPort: 177
Unless you are also providing XDMCP service to xterminals or other machines, make sure that the host access list only allows local connections (the name of this file is often Xaccess and it is usually setup by default to do just that.) Nowadays, host level firewalling will also typically block UDP (port 177 for XDMCP) by default effectively limiting the UDP connections to localhost.
Tip: Note that -xdmsvc is a short hand alias for the long "-ssl SAVE -unixpw -users unixpw= -display WAIT:cmd=FINDCREATEDISPLAY-Xvfb.xdmcp". So we simply use:
service x11vnc
{
type = UNLISTED
port = 5900
socket_type = stream
protocol = tcp
wait = no
user = root
server = /usr/local/bin/x11vnc
server_args = -inetd -o /var/log/x11vnc.log -xdmsvc
disable = no
}
(Note: use "-svc" instead of "-xdmsvc" for no XDMCP login greeter.)
Local access (VNC Server and VNC Viewer on the same machine): To access your virtual X display session locally (i.e. while sitting at the same machine it is running on) one can perhaps have something like this in their $HOME/.xinitrc
#!/bin/sh
x11vnc -create -rfbport 5905 -env WAITBG=1
vncviewer -geometry +0+0 -encodings raw -passwd $HOME/.vnc/passwd localhost:5
You may not need the -passwd. Recent RealVNC viewers might be this:
#!/bin/sh
x11vnc -create -rfbport 5905 -env WAITBG=1
vncviewer -FullScreen -PreferredEncoding raw -passwd $HOME/.vnc/passwd localhost:5
This way a bare X server is run with no window manager or desktop; it simply runs only the VNC Viewer on the real X server. The Viewer then draws the virtual X session on to the real one. On your system it might not be $HOME/.xinitrc, but rather .xsession, .Xclients, or something else. You will need to figure out what it is for your system and configuration.
There may be a problem if the resolution (WxH) of the virtual X display does not match that of the physical X display.
If you do not want to or cannot figure out the X startup script name (.xinitrc, etc) you could save the above commands to a shell script, say "vnclocal", and the log in via the normal KDM or GDM greeter program using the "Failsafe" option. Then in the lone xterm that comes up type "vnclocal" to connect to your virtual X display via x11vnc and vncviewer.
Summary: The "-display WAIT:cmd=FINDCREATEDISPLAY" scheme can be used to provide a "desktop service" (i.e. terminal service) on the server machine: you always get some desktop there, either a real hardware X server or a virtual one (depending on how you set things up.)
So it provides simple "terminal services" based on Unix username and password. The created X server sessions (virtual or real hardware) will remain running after you disconnect the VNC viewer and will be found again on reconnecting via VNC and logging in. To terminate them use the normal way to Exit/LogOut from inside your X session. The user does not have to memorize which VNC display number is his. They all go the same one (e.g. hostname:0) and it switches based on username.
One could do this in a shell script, but now there is an option -loop that makes it easier. Of course when x11vnc restarts it needs to have permissions to connect to the (potentially new) X display. This mode could be useful if the X server restarts often. Use e.g. "-loop5000" to sleep 5000 ms between restarts. Also "-loop2000,5" to sleep 2000 ms and only restart 5 times.
One can also use the -loopbg to emulate inetd(8) to some degree, where each connected process runs in the background. It could be combined, say, with the -svc option to provide simple terminal services without using inetd(8).
To have x11vnc serve up a Java VNC viewer applet to any web browsers that connect to it, run x11vnc with this option:
-httpdir /path/to/the/java/classes/dir
(this directory will contain the files index.vnc and, for example, VncViewer.jar) Note that libvncserver contains the TightVNC Java classes jar file for your convenience. (it is the file classes/VncViewer.jar in the source tree.)
You will see output something like this:
14/05/2004 11:13:56 Autoprobing selected port 5900
14/05/2004 11:13:56 Listening for HTTP connections on TCP port 5800
14/05/2004 11:13:56 URL http://walnut:5800
14/05/2004 11:13:56 screen setup finished.
14/05/2004 11:13:56 The VNC desktop is walnut:0
PORT=5900
then you can connect to that URL with any Java enabled browser. Feel free to customize the default index.vnc file in the classes directory.
As of May/2005 the -http option will try to guess where the Java classes jar file is by looking in expected locations and ones relative to the x11vnc binary.
Also note that if you wanted to, you could also start the Java viewer entirely from the viewer-side by having the jar file there and using either the java or appletviewer commands to run the program.
java -cp ./VncViewer.jar VncViewer HOST far-away.east PORT 5900
Proxies: See the discussion here if the web browser must use a web proxy to connect to the internet. It is tricky to get Java applets to work in this case: a signed applet must be used so it can connect to the proxy and ask for the redirection to the VNC server. One way to do this is to use the signed SSL one referred to in classes/ssl/proxy.vnc and set disableSSL=yes (note that this has no encryption; please use SSL or SSH as discuss elsewhere on this page) in the URL or the file.
Q-65: Are reverse connections (i.e. the VNC server connecting to the VNC viewer) using "vncviewer -listen" and vncconnect(1) supported?
As of Mar/2004 x11vnc supports reverse connections. On Unix one starts the VNC viewer in listen mode: "vncviewer -listen" (see your documentation for Windows, etc), and then starts up x11vnc with the -connect option. To connect immediately at x11vnc startup time use the "-connect host:port" option (use commas for a list of hosts to connect to.) The ":port" is optional (default is VNC listening port is 5500.)
If a file is specified instead: -connect /path/to/some/file then that file is checked periodically (about once a second) for new hosts to connect to.
The -remote control option (aka -R) can also be used to do this during an active x11vnc session, e.g.:
x11vnc -display :0 -R connect:hostname.domain
Use the "-connect_or_exit" option to have x11vnc exit if the reverse connection fails. Also, note the "-rfbport 0" option disables TCP listening for connections (potentially useful for reverse connection mode, assuming you do not want any "forward" connections.)
Note that as of Mar/2006 x11vnc requires password authentication for reverse connections as well as for forward ones (assuming password auth has been enabled, e.g. via -rfbauth, -passwdfile, etc.) Many VNC servers do not require any password for reverse connections. To regain the old behavior supply this option "-env X11VNC_REVERSE_CONNECTION_NO_AUTH=1" to x11vnc.
Vncconnect command: To use the vncconnect(1) program (from the core VNC package at www.realvnc.com) specify the -vncconnect option to x11vnc (Note: as of Dec/2004 -vncconnect is now the default.) vncconnect(1) must be pointed to the same X11 DISPLAY as x11vnc (since it uses X properties to communicate with x11vnc.) If you do not have or do not want to get the vncconnect(1) program, the following script (named "Vncconnect") may work if your xprop(1) supports the -set option:
#!/bin/sh
# usage: Vncconnect <host>
# Vncconnect <host:port>
# note: not all xprop(1) support -set.
#
xprop -root -f VNC_CONNECT 8s -set VNC_CONNECT "$1"
Yes, as of Oct/2007 x11vnc supports reverse connections through proxies: use the "-proxy host:port" option. The default is to assume the proxy is a Web proxy. Note that most Web proxies only allow proxy destination connections to ports 443 (HTTPS) and 563 (SNEWS) and so this might not be too useful unless the proxy has been modified (AllowCONNECT apache setting) or the VNC viewer listens on one of those ports (or the router does a port redir.) A web proxy may also be specified via "-proxy http://host:port"
For SOCKS4 and SOCKS4a proxies use this format "-proxy socks://host:port". If the reverse connection hostname is a numerical IP or "localhost" then SOCKS4 (no host lookup) is used, otherwise SOCKS4a will be used. For SOCKS5 (proxy will do lookup and many other things) use "-proxy socks5://host:port". Note that the SSH builtin SOCKS proxy "ssh -D port" only does SOCKS4 or SOCKS5, so use socks5:// for a ssh -D proxy.
The proxying works for both SSL encrypted and normal reverse connections.
An experimental mode is "-proxy http://host:port/..." where the URL (e.g. a CGI script) is retrieved via the GET method. See -proxy for more info.
Another experimental mode is "-proxy ssh://user@host" in which case a SSH tunnel is used for the proxying. See -proxy for more info.
Up to 3 proxies may be chained together by listing them by commas e.g.: "-proxy http://host1:port1,socks5://host2:port2" in case one needs to ricochet off of several machines to ultimately reach the listening viewer.
Yes. See the example script desktop.cgi for ideas. It is in the source tree in the directory x11vnc/misc. It serves x11vnc's SSL enabled Java Applet to the web browser with the correct connection information for the user's virtual desktop (an Xvfb session via -create; be sure to add the Xvfb package.) HTTPS/SSL enabled Apache should be used to serve the script to avoid unix and vnc passwords from being sent in cleartext and sniffed.
By default it uses a separate VNC port for each user desktop (either by autoprobing in a range of ports or using a port based on the userid number.) The web server's firewall must allow incoming connections to these ports.
It is somewhat difficult to do all of this with x11vnc listening on a single port, however there is also a 'fixed port' scheme described in the script based on -loopbg that works fairly well (but more experience is needed to see what problems contention for the same port causes; however at worst one user may need to re-login.)
There is also an optional 'port redirection' mode for desktop.cgi that allows redirection to other machines inside the firewall already running SSL enabled VNC servers. This provides much of the functionality as the SSL Portal and is easier to set up.
Q-68: Can I use x11vnc as a replacement for Xvnc? (i.e. not for a real display, but for a virtual one I keep around.)
You can, but you would not be doing this for performance reasons (for virtual X sessions via VNC, Xvnc should give the fastest response.) You may want to do this because Xvnc is buggy and crashes, does not support an X server extension you desire, or you want to take advantage of one of x11vnc's unending number of options and features.
One way to achieve this is to have a Xvfb(1) virtual framebuffer X server running in the background and have x11vnc attached to it. Another method, faster and more accurate, is to use the "dummy" Device Driver in XFree86/Xorg (see below.)
For these virtual sessions you will need to install the Xvfb program (e.g. apt-get install xvfb) or our Xdummy program (see below.)
In either case, one can view this desktop both remotely and also locally using vncviewer. Make sure vncviewer's "-encodings raw" is in effect for local viewing (compression seems to slow things down locally.) For local viewing you set up a "bare" window manager that just starts up vncviewer and nothing else (See how below.)
Here is one way to start up Xvfb:
xinit -- /usr/bin/Xvfb :1 -cc 4 -screen 0 1024x768x16
This starts up a 16bpp virtual display. To export it via VNC use
x11vnc -display :1 ...
Then have the remote vncviewer attach to x11vnc's VNC display (e.g. :0 which is port 5900.)
The "-cc 4" Xvfb option is to force it to use a TrueColor visual instead of DirectColor (this works around a recent bug in the Xorg Xvfb server.)
One good thing about Xvfb is that the virtual framebuffer exists in main memory (rather than in the video hardware), and so x11vnc can "screen scrape" it very efficiently (more than, say, 100X faster than normal video hardware.)
Update Nov/2006: See the FINDCREATEDISPLAY discussion of the "-display WAIT:cmd=FINDDISPLAY" option where virtual (Xvfb or Xdummy, or even real ones by changing an option) X servers are started automatically for new users connecting. This provides a "desktop service" for the machine. You either get your real X session or your virtual (Xvfb/Xdummy) one whenever you connect to the machine (inetd(8) is a nice way to provide this service.) The -find, -create, -svc, and -xdmsvc aliases can also come in handy here.
There are some annoyances WRT Xvfb however. The default keyboard mapping seems to be very poor. One should run x11vnc with -add_keysyms option to have keysyms added automatically. Also, to add the Shift_R and Control_R modifiers something like this is needed:
#!/bin/sh
xmodmap -e "keycode any = Shift_R"
xmodmap -e "add Shift = Shift_L Shift_R"
xmodmap -e "keycode any = Control_R"
xmodmap -e "add Control = Control_L Control_R"
xmodmap -e "keycode any = Alt_L"
xmodmap -e "keycode any = Alt_R"
xmodmap -e "keycode any = Meta_L"
xmodmap -e "add Mod1 = Alt_L Alt_R Meta_L"
(note: these are applied automatically in the FINDCREATEDISPLAY mode of x11vnc.) Perhaps the Xvfb options -xkbdb or -xkbmap could be used to get a better default keyboard mapping...
Dummy Driver: A user points out a faster and more accurate method is to use the "dummy" Device Driver of XFree86/Xorg instead of Xvfb. He uses this to create a persistent and resizable desktop accessible from anywhere. In the Device Section of the config file set Driver "dummy". You may also need to set VideoRam NNN to be large enough to hold the framebuffer. The framebuffer is kept in main memory like Xvfb except that the server code is closely correlated with the real XFree86/Xorg Xserver unlike Xvfb.
The main drawback to this method (besides requiring extra configuration and possibly root permission) is that it also does the Linux Virtual Console/Terminal (VC/VT) switching even though it does not need to (since it doesn't use a real framebuffer.) There are some "dual headed" (actually multi-headed/multi-user) patches to the X server that turn off the VT usage in the X server. Update: As of Jul/2005 we have an LD_PRELOAD script Xdummy that allows you to use a stock (i.e. unpatched) Xorg or XFree86 server with the "dummy" driver and not have any VT switching problems! An advantage of Xdummy over Xvfb is that Xdummy supports RANDR dynamic screen resizing.
The standard way to start the "dummy" driver would be:
startx -- :1 -config /etc/X11/xorg.conf.dummy
where the file /etc/X11/xorg.conf.dummy has its Device Section modified as described above. To use the LD_PRELOAD wrapper script:
startx -- /path/to/Xdummy :1
An xdm(1) example is also provided.
In general, one can use these sorts of schemes to use x11vnc to export other virtual X sessions, say Xnest or even Xvnc itself (useful for testing x11vnc.)
Local access (VNC Server and VNC Viewer on the same machine): You use a VNC viewer to access the display remotely; to access your virtual X display locally (i.e. while sitting at the same machine it is running on) one can perhaps have something like this in their $HOME/.xinitrc
#!/bin/sh
x11vnc -display :5 -rfbport 5905 -bg
vncviewer -geometry +0+0 -encodings raw -passwd $HOME/.vnc/passwd localhost:5
The display numbers (VNC and X) will likely be different (you could also try -find), and you may not need the -passwd. Recent RealVNC viewers might be this:
#!/bin/sh
x11vnc -display :5 -rfbport 5905 -bg
vncviewer -FullScreen -PreferredEncoding raw -passwd $HOME/.vnc/passwd localhost:5
This way a bare X server is run with no window manager or desktop; it simply runs only the VNC Viewer on the real X server. The Viewer then draws the virtual X session on to the real one. On your system it might not be $HOME/.xinitrc, but rather .xsession, .Xclients, or something else. You will need to figure out what it is for your system and configuration.
XDM/GDM/KDM One-Shot X sessions: For the general replacement of Xvnc by Xvfb+x11vnc, one user describes a similar setup he created where the X sessions are one-shot's (destroyed after the vncviewer disconnects) and it uses the XDM/GDM/KDM login greeter here.
An interesting application of x11vnc is to let it export displays of "headless" machines. For example, you may have some lab or server machines with no keyboard, mouse, or monitor, but each one still has a video card. One can use x11vnc to provide a simple "desktop service" from these server machines.
An X server can be started on the headless machine (sometimes this requires configuring the X server to not fail if it cannot detect a keyboard or mouse, see the next paragraph.) Then you can export that X display via x11vnc (e.g. see this FAQ) and access it from anywhere on the network via a VNC viewer.
Some tips on getting X servers to start on machines without keyboard or mouse: For XFree86/Xorg the Option "AllowMouseOpenFail" "true" "ServerFlags" config file option is useful. On Solaris Xsun the +nkeyboard and +nmouse options are useful (put them in the server command line args in /etc/dt/config/Xservers.) There are patches available for Xsun at lease back to Solaris 8 that support this. See Xserver(1) for more info.
Although this usage may sound strange it can be quite useful for a GUI (or other) testing or QA setups: the engineers do not need to walk to lab machines running different hardware, OS's, versions, etc (or have many different machines in their office.) They just connect to the various test machines over the network via VNC. The advantage to testing this way instead of using Xvnc or even Xvfb is that the test is done using the real X server, fonts, video hardware, etc. that will be used in the field.
One can imagine a single server machine crammed with as many video cards as it can hold to provide multiple simultaneous access or testing on different kinds of video hardware.
See also the FINDCREATEDISPLAY discussion of the "-display WAIT:cmd=FINDDISPLAY" option where virtual Xvfb or Xdummy, or real X servers are started automatically for new users connecting. The -find, -create, -svc, and -xdmsvc aliases can also come in handy here.
Q-70: I have lots of memory, but why does x11vnc fail with shmget: No space left on device or Minor opcode of failed request: 1 (X_ShmAttach)?
It is not a matter of free memory, but rather free shared memory (shm) slots, also known as shm segments. This often occurs on a public Solaris machine using the default of only 100 slots. You (or the owner or root) can clean them out with ipcrm(1). x11vnc tries hard to release its slots, but it, and other programs, are not always able to (e.g. if kill -9'd.)
Sometimes x11vnc will notice the problem with shm segments and tries to get by with fewer, only giving a warning like this:
19/03/2004 10:10:58 shmat(tile_row) failed.
shmat: Too many open files
19/03/2004 10:10:58 error creating tile-row shm for len=4
19/03/2004 10:10:58 reverting to single_copytile mode
Here is a shell script shm_clear to list and prompt for removal of your unattached shm segments (attached ones are skipped.) I use it while debugging x11vnc (I use "shm_clear -y" to assume "yes" for each prompt.) If x11vnc is regularly not cleaning up its shm segments, please contact me so we can work to improve the situation.
Longer term, on Solaris you can put something like this in /etc/system:
set shmsys:shminfo_shmmax = 0x2000000
set shmsys:shminfo_shmmni = 0x1000
to sweep the problem under the rug (4096 slots.) On Linux, examine /proc/sys/kernel/shmmni; you can modify the value by writing to that file.
Things are even more tight on Solaris 8 and earlier, there is a default maximum number of shm segments per process of 6. The error is the X server (not x11vnc) being unable to attach to the segments, and looks something like this:
30/04/2004 14:04:26 Got connection from client 192.168.1.23
30/04/2004 14:04:26 other clients:
X Error of failed request: BadAccess (attempt to access private resource denied)
Major opcode of failed request: 131 (MIT-SHM)
Minor opcode of failed request: 1 (X_ShmAttach)
Serial number of failed request: 14
Current serial number in output stream: 17
This tight limit on Solaris 8 can be increased via:
set shmsys:shminfo_shmseg = 100
in /etc/system. See the next paragraph for more workarounds.
To minimize the number of shm segments used by x11vnc try using the -onetile option (corresponds to only 3 shm segments used, and adding -fs 1.0 knocks it down to 2.) If you are having much trouble with shm segments, consider disabling shm completely via the -noshm option. Performance will be somewhat degraded but when done over local machine sockets it should be acceptable (see an earlier question discussing -noshm.)
The -nap (now on by default; use -nonap to disable) and "-wait n" (where n is the sleep between polls in milliseconds, the default is 30 or so) option are good places to start. In addition, something like "-sb 15" will cause x11vnc to go into a deep-sleep mode after 15 seconds of no activity (instead of the default 60.)
Reducing the X server bits per pixel depth (e.g. to 16bpp or even 8bpp) will further decrease memory I/O and network I/O. The ShadowFB X server setting will make x11vnc's screen polling less severe. Using the -onetile option will use less memory and use fewer shared memory slots (add -fs 1.0 for one less slot.)
You can try -threads (note this mode can be unstable and/or crash; and as of May/2008 is strongly discouraged, see the option description) or dial down the wait time (e.g. -wait 1) and possibly dial down -defer as well. Note that if you try to increase the "frame rate" too much you can bog down the server end with the extra work it needs to do compressing the framebuffer data, etc.
That said, it is possible to "stream" video via x11vnc if the video window is small enough. E.g. a 256x192 xawtv TV capture window (using the x11vnc -id option) can be streamed over a LAN or wireless at a reasonable frame rate. If the graphics card's framebuffer read rate is faster than normal then the video window size and frame rate can be much higher. The use of TurboVNC and/or TurboJPEG can make the frame rate somewhat higher still (but most of this hinges on the graphics card's read rate.)
Q-73: I use x11vnc over a slow link with high latency (e.g. dialup modem or broadband), is there anything I can do to speed things up?
Some things you might want to experiment with (many of which will help performance on faster links as well):
X server/session parameters:
- Configure the X server bits per pixel to be 16bpp or even 8bpp. (reduces amount of data needed to be polled, compressed, and sent)
- Use a smaller desktop size (e.g. 1024x768 instead of 1280x1024)
- Make sure the desktop background is a solid color (the background is resent every time it is re-exposed.) Consider using the -solid [color] option to try to do this automatically.
- Configure your window manager or desktop "theme" to not use fancy images, shading, and gradients for the window decorations, etc. Disable window animations, etc. Maybe your desktop has a "low bandwidth" theme you can easily switch into and out of. Also in Firefox disable eye-candy, e.g.: Edit -> Preferences -> Advanced -> Use Smooth Scrolling (deselect it.)
- Avoid small scrolls of large windows using the Arrow keys or scrollbar. Try to use PageUp/PageDown instead. (not so much of a problem in x11vnc 0.7.2 if -scrollcopyrect is active and detecting scrolls for the application.)
- If the -wireframe option is not available (earlier than x11vnc 0.7.2 or you have disabled it via -nowireframe) then Disable Opaque Moves and Resizes in the window manager/desktop.
- However if -wireframe is active (on by default in x11vnc 0.7.2) then you should Enable Opaque Moves and Resizes in the window manager! This seems counter-intuitive, but because x11vnc detects the move/resize events early there is a huge speedup over a slow link when Opaque Moves and Resizes are enabled. (e.g. CopyRect encoding will be used.)
- Turn off Anti-aliased fonts on your system, web browser, terminal windows, etc. AA fonts do not compress as well as traditional fonts (sometimes 10X less.)
- On Firefox/Mozilla (and anything else) turn off "Smooth Scroll" animations. In Firefox put in the URL "about:config" and set general.smoothScroll to false.
- On Xorg/XFree86 turn on the Shadow Framebuffer to speed up reading. (Option "ShadowFB" "true" in the Device section of /etc/X11/XF86Config) This disables 2D acceleration on the physical display and so may not be worth it (if you play games, etc), but could be of use in some situations. Note: If the network link is very slow, this speedup may not be noticed.
VNC viewer parameters:
- Use a TightVNC enabled viewer! (Actually, RealVNC 4.x viewer with ZRLE encoding is not too bad either; some claim it is faster.)
- Make sure the tight (or zrle) encoding is being used (look at vncviewer and x11vnc outputs)
- Request 8 bits per pixel using -bgr233 (up to 4X speedup over depth 24 TrueColor (32bpp), but colors will be off)
- RealVNC 4.x viewer has some extremely low color modes (only 64 and even 8 colors.) SSVNC does too. The colors are poor, but it is usually noticeably faster than bgr233 (256 colors.)
- Try increasing the TightVNC -compresslevel (compresses more on server side before sending, but uses more CPU)
- Try reducing the TightVNC -quality (increases JPEG compression, but is lossy with painting artifacts)
- Try other VNC encodings via -encodings (tight may be the fastest, but you should compare it to zrle and maybe some of the others)
- On the machine where vncviewer is run, make sure Backing Store is enabled (Xorg/XFree86 disables it by default causing re-exposures of vncviewer to be very slow) Option "backingstore" in config file.
x11vnc parameters:
- Make sure the -wireframe option is active (it should be on by default) and you have Opaque Moves/Resizes Enabled in the window manager.
- Make sure the -scrollcopyrect option is active (it should be on by default.) This detects scrolls in many (but not all) applications an applies the CopyRect encoding for a big speedup.
- Enforce a solid background when VNC viewers are connected via -solid
- Try x11vnc's client-side caching client-side caching scheme: -ncache
- Specify -speeds modem to force the wireframe and scrollcopyrect heuristic parameters (and any future ones) to those of a dialup modem connection (or supply the rd,bw,lat numerical values that characterize your link.)
- If wireframe and scrollcopyrect aren't working, try using the more drastic -nodragging (no screen updates when dragging mouse, but sometimes you miss visual feedback)
- Set -fs 1.0 (disables fullscreen updates)
- Try increasing -wait or -defer (reduces the maximum "frame rate", but won't help much for large screen changes)
- Try the -progressive pixelheight mode with the block pixelheight 100 or so (delays sending vertical blocks since they may change while viewer is receiving earlier ones)
- If you just want to watch one (simple) window use -id or -appshare (cuts down extraneous polling and updates, but can be buggy or insufficient)
- Set -nosel (disables all clipboard selection exchange)
- Use -nocursor and -nocursorpos (repainting the remote cursor position and shape takes resources and round trips)
- On very slow links (e.g. <= 28.8) you may need to increase the -readtimeout n setting if it sometimes takes more than 20sec to paint the full screen, etc.
- Do not use -fixscreen to automatically refresh the whole screen, tap three Alt_L's then the screen has painting errors (rare problem.)
Example for the KDE desktop:
Launch the "KDE Control Center" utility. Sometimes this is called "Personal Settings".
Select "Desktop".
Then Select "Window Behavior". In the "Moving" Tab set these:
- YES - Display content in moving windows
- YES - Display content in resizing windows
- NO - Display window geometry when moving or resizing
- NO - Animate minimize and restore
In the "Translucency" Tab set:
- NO - Use translucency/shadows
Next hit "Back" and then select "Panels".
In the "Appearance" Tab set:
- NO - Enable icon mouseover effects
- NO - Enable transparency
Now go all the way back up to the top and Select "Appearance & Themes".
Select "Background" and set:
- YES - No picture
- Colors: Single Color
Select "Fonts" and disable anti-aliased fonts if you are bold enough.
Select "Launch Feedback" and set:
- Busy Cursor: No Busy Cursor
- NO - Enable taskbar notification
Select "Screen Saver" and set:
- Screen Saver: Blank Screen
Select "Style" and in the "Effects" Tab set:
- NO - Enable GUI effects
Example for the GNOME desktop:
- TBD.
Q-74: Does x11vnc support the X DAMAGE Xserver extension to find modified regions of the screen quickly and efficiently?
Yes, as of Mar/2005 x11vnc will use the X DAMAGE extension by default if it is available on the display. This requires libXdamage to be available in the build environment as well (recent Linux distros and Solaris 10 have it.)
The DAMAGE extension enables the X server to report changed regions of the screen back to x11vnc. So x11vnc doesn't have to guess where the changes are (by polling every pixel of the entire screen every 2-4 seconds.) The use of X DAMAGE dramatically reduces the load when the screen is not changing very much (i.e. most of the time.) It also noticeably improves updates, especially for very small changed areas (e.g. clock ticking, cursor flashing, typing, etc.)
Note that the DAMAGE extension does not speed up the actual reading of pixels from the video card framebuffer memory, by, say, mirroring them in main memory. So reading the fb is still painfully slow (e.g. 5MB/sec), and so even using X DAMAGE when large changes occur on the screen the bulk of the time is still spent retrieving them. Not ideal, but use of the ShadowFB XFree86/Xorg option speeds up the reading considerably (at the cost of h/w acceleration.)
Unfortunately the current Xorg DAMAGE extension implementation can at times be overly conservative and report very large rectangles as "damaged" even though only a small portion of the pixels have actually been modified. This behavior is often the fault of the window manager (e.g. it redraws the entire, unseen, frame window underneath the application window when it gains focus), or the application itself (e.g. does large, unnecessary repaints.)
To work around this deficiency, x11vnc currently only trusts small DAMAGE rectangles to contain real damage. The larger rectangles are only used as hints to focus the traditional scanline polling (i.e. if a scanline doesn't intersect a recent DAMAGE rectangle, the scan is skipped.) You can use the "-xd_area A" option to adjust the size of the trusted DAMAGE rectangles. The default is 20000 pixels (e.g. a 140x140 square, etc.) Use "-xd_area 0" to disable the cutoff and trust all DAMAGE rectangles.
The option "-xd_mem f" may also be of use in tuning the algorithm. To disable using DAMAGE entirely use "-noxdamage".
Q-75: My OpenGL application shows no screen updates unless I supply the -noxdamage option to x11vnc.
One user reports in his environment (MythTV using the NVIDIA OpenGL drivers) he gets no updates after the initial screen is drawn unless he uses the "-noxdamage" option.
This seems to be a bug in the X DAMAGE implementation of that driver. You may have to use -noxdamage as well. A way to autodetect this will be tried, probably the best it will do is automatically stop using X DAMAGE.
A developer for MiniMyth reports that the 'alphapulse' tag of the theme G.A.N.T. can also cause problems, and should be avoided when using VNC.
Update: see this FAQ too.
Q-76: When I drag windows around with the mouse or scroll up and down things really bog down (unless I do the drag in a single, quick motion.) Is there anything to do to improve things?
This problem is primarily due to slow hardware read rates from video cards: as you scroll or move a large window around the screen changes are much too rapid for x11vnc to keep up them (it can usually only read the video card at about 5-10 MB/sec, so it can take a good fraction of a second to read the changes induce from moving a large window, if this to be done a number of times in succession the window or scroll appears to "lurch" forward.) See the description in the -pointer_mode option for more info. The next bottleneck is compressing all of these changes and sending them out to connected viewers, however the VNC protocol is pretty much self-adapting with respect to that (updates are only packaged and sent when viewers ask for them.)
As of Jan/2004 there are some improvements to libvncserver. The default should now be much better than before and dragging small windows around should no longer be a huge pain. If for some reason these changes make matters worse, you can go back to the old way via the "-pointer_mode 1" option.
Also added was the -nodragging option that disables all screen updates while dragging with the mouse (i.e. mouse motion with a button held down.) This gives the snappiest response, but might be undesired in some circumstances when you want to see the visual feedback while dragging (e.g. menu traversal or text selection.)
As of Dec/2004 the -pointer_mode n option was introduced. n=1 is the original mode, n=2 an improvement, etc.. See the -pointer_mode n help for more info.
Also, in some circumstances the -threads option can improve response considerably. Be forewarned that if more than one vncviewer is connected at the same time then libvncserver may not be thread safe (try to get the viewers to use different VNC encodings, e.g. tight and ZRLE.) This option can be unstable and so as of Feb/2008 it is disabled by default. Set env. X11VNC_THREADED=1 to re-enable.
As of Apr/2005 two new options (see the wireframe FAQ and scrollcopyrect FAQ below) provide schemes to sweep this problem under the rug for window moves or resizes and for some (but not all) window scrolls. These are the preferred way of avoiding the "lurching" problem, contact me if they are not working. Note on SuSE and some other distros the RECORD X extension used by scrollcopyrect is not enabled by default, turn it on in xorg.conf:
Section "Module"
...
Load "record"
...
EndSection
Q-77: Why not do something like wireframe animations to avoid the windows "lurching" when being moved or resized?
Nice idea for a hack! As of Apr/2005 x11vnc by default will apply heuristics to try to guess if a window is being (opaquely) moved or resized. If such a change is detected framebuffer polling and updates will be suspended and only an animated "wireframe" (a rectangle outline drawn where the moved/resized window would be) is shown. When the window move/resize stops, it returns to normal processing: you should only see the window appear in the new position. This spares you from interacting with a "lurching" window between all of the intermediate steps. BTW the lurching is due to slow video card read rates (see here too.) A displacement, even a small one, of a large window requires a non-negligible amount of time, a good fraction of a second, to read in from the hardware framebuffer.
Note that Opaque Moves/Resizes must be Enabled by your window manager for -wireframe to do any good.
The mode is currently on by default because most people are afflicted with the problem. It can be disabled with the -nowireframe option (aka -nowf.) Why might one want to turn off the wireframing? Since x11vnc is merely guessing when windows are being moved/resized, it may guess poorly for your window-manager or desktop, or even for the way you move the pointer. If your window-manager or desktop already does its own wireframing then this mode is a waste of time and could do the wrong thing occasionally. There may be other reasons the new mode feels unnatural. If you have very expensive video hardware (SGI, well now even proprietary Xorg drivers are fast at reading) or are using an in-RAM video framebuffer (SunRay, ShadowFB, Xvfb), the read rate from that framebuffer may be very fast (100's of MB/sec) and so you don't really see much lurching (at least over a fast LAN): opaque moves look smooth in x11vnc. Note: ShadowFB is often turned on when you are using the vesafb or fbdev XFree86 video driver instead of a native one so you might be using it already and not know.
The heuristics used to guess window motion or resizing are simple, but are not fool proof: x11vnc is sometimes tricked and so you'll occasionally see the lurching opaque move and rarely something even worse.
First it assumes that the move/resize will occur with a mouse button pressed, held down and dragged (of course this is only mostly true.) Next it will only consider a window for wireframing if the mouse pointer is initially "close enough" to the edges of the window frame, e.g. you have grabbed the title bar or a resizer edge (this requirement can be disabled and it also not applied if a modifier key, e.g. Alt, is pressed.) If these are true, it will wait an amount of time to see if the window starts moving or resizing. If it does, it starts drawing the wireframe "outline" of where the window would be. When the mouse button is released, or a timeout occurs, it goes back to the standard mode to allow the actual framebuffer changes to propagate to the viewers.
These parameters can be tweaked:
- Color/Shade of the wireframe.
- Linewidth of the outline frame.
- Cutoff size of windows to not apply wireframing to.
- Cutoffs for closeness to Top, Bottom, Left, and Right edges of window.
- Modifier keys to enable interior window grabbing.
- Maximum time to wait for dragging pointer events.
- Maximum time to wait for the window to start moving/resizing.
- Maximum time to show a wireframe animation.
- Minimum time between sending wireframe outlines.
See the "-wireframe tweaks" option for more details. On a slow link, e.g. dialup modem, the parameters may be automatically adjusted for better response.
CopyRect encoding: In addition to the above there is the "-wirecopyrect mode" option. It is also on by default. This instructs x11vnc to not only show the wireframe animation, but to also instruct all connected VNC viewers to locally translate the window image data from the original position to the new position on the screen when the animation is done. This speedup is the VNC CopyRect encoding: the framebuffer update doesn't need to send the actual new image data. This is nice in general, and very convenient over a slow link, but since it is based on heuristics you may need to disable it with the -nowirecopyrect option (aka -nowcr) if it works incorrectly or unnaturally for you.
The -wirecopyrect modes are: "never" (same as -nowirecopyrect); "top", only apply the CopyRect if the window is appears to be on the top of the window stack and is not obstructed by other windows; and "always" to always try to apply the CopyRect (obstructed regions are usually clipped off and not translated.)
Note that some desktops (KDE and xfce) appear to mess with the window stacking in ways that are not yet clear. In these cases x11vnc works around the problem by applying the CopyRect even if obscuring windows' data is translated! Use -nowirecopyrect if this yields undesirable effects for your desktop.
Also, the CopyRect encoding may give incorrect results under -scale (depending on the scale factor the CopyRect operation is often only approximate: the correctly scaled framebuffer will be slightly different from the translated one.) x11vnc will try to push a "cleanup" update after the CopyRect if -scale is in effect. Use -nowirecopyrect if this or other painting errors are unacceptable.
Q-78: Can x11vnc try to apply heuristics to detect when a window is scrolling its contents and use the CopyRect encoding for a speedup?
Another nice idea for a hack! As of May/2005 x11vnc will by default apply heuristics to try to detect if the window that has the input focus is scrolling its contents (but only when x11vnc is feeding user input, keystroke or pointer, to the X server.) So, when detected, scrolls induced by dragging on a scrollbar or by typing (e.g. Up or Down arrows, hitting Return in a terminal window, etc), will show up much more quickly than via the standard x11vnc screen polling update mechanism.
There will be a speedup for both slow and fast links to viewers. For slow links the speedup is mostly due to the CopyRect encoding not requiring the image data to be transmitted over the network. For fast links the speedup is primarily due to x11vnc not having to read the scrolled framebuffer data from the X server (recall that reading from the hardware framebuffer is slow.)
To do this x11vnc uses the RECORD X extension to snoop the X11 protocol between the X client with the focus window and the X server. This extension is usually present on most X servers (but SuSE disables it for some reason.) On XFree86/Xorg it can be enabled via Load "record" in the Module section of the config file if it isn't already:
Section "Module"
...
Load "record"
...
EndSection
Currently the RECORD extension is used as little as possible so as to not slow down regular use. Only simple heuristics are applied to detect XCopyArea and XConfigureWindow calls from the application. These catch a lot of scrolls, e.g. in mozilla/firefox and in terminal windows like gnome-terminal and xterm. Unfortunately the toolkits KDE applications use make scroll detection less effective (only rarely are they detected: i.e. Konqueror and Konsole don't work.) An interesting project, that may be the direction x11vnc takes, is to record all of the X11 protocol from all clients and try to "tee" the stream into a modified Xvfb watching for CopyRect and other VNC speedups. A potential issue is the RECORD stream is delayed from actual view on the X server display: if one falls too far behind it could become a mess...
The initial implementation of -scrollcopyrect option is useful in that it detects many scrolls and thus gives a much nicer working environment (especially when combined with the -wireframe -wirecopyrect options, which are also on by default; and if you are willing to enable the ShadowFB things are very fast.) The fact that there aren't long delays or lurches during scrolling is the primary improvement.
But there are some drawbacks:
- Not all scrolls are detected. Some apps scroll windows in ways that cannot currently be detected, and other times x11vnc "misses" the scroll due to timeouts, etc. Sometimes it is more distracting that a speedup occasionally doesn't work as opposed to being consistently slow!
- For rapid scrolling (i.e. sequence of many scrolls over a short period) there can be painting errors (tearing, bunching up, etc.) during the scroll. These will repair themselves after the scroll is over, but when they are severe it can be distracting. Try to think of the approximate window contents as a quicker and more useful "animation" compared to the slower polling scheme...
- Scrolling inside shells in terminal windows (gnome-terminal, xterm), can lead to odd painting errors. This is because x11vnc did not have time to detect a screen change just before the scroll (most common is the terminal undraws the block cursor before scrolling the text up: in the viewer you temporarily see multiple block cursors.) Another issue is with things like more(1): scroll detection for 5-6 lines happens nicely, but then it can't keep up and so there is a long pause for the standard polling method to deliver the remaining updates.
- More rarely sometimes painting errors are not repaired after the scroll is over. This may be a bug in x11vnc or libvncserver, or it may be an inescapable fact of the CopyRect encoding and the delay between RECORD callbacks and what is actually on the X display. One can tap the Alt_L key (Left "Alt" key) 3 times in a row to signal x11vnc to refresh the screen to all viewers. Your VNC-viewer may have its own screen refresh hot-key or button. See also: -fixscreen
- Some applications, notably OpenOffice, do XCopyArea scrolls in weird ways that assume ancestor window clipping is taking place. See the -scr_skip option for ways to tweak this on a per-application basis.
- Selecting text while dragging the mouse may be slower, especially if the Button-down event happens near the window's edge. This is because the scrollcopyrect scheme is watching for scrolls via RECORD and has to wait for a timeout to occur before it does the update.
- For reasons not yet understood the RECORD extension can stop responding (and hence scrolls are missed.) As a workaround x11vnc attempts to reset the RECORD connection every 60 seconds or so. Another workaround is to type 4 Super_L (Left Super/Windows-Flag key) in a row to reset RECORD. Work is in progress to try to fix this bug.
- Sometimes you need to "retrain" x11vnc for a certain window because it fails to detect scrolls in it. Sometimes clicking inside the application window or selecting some text in it to force the focus helps.
- When using the -scale option there will be a quick CopyRect scroll, but it needs to be followed by a slower "cleanup" update. This is because for a fixed finite screen resolution (e.g. 75 dpi) scaling and copyrect-ing are not exactly independent. Scaling involves a blending of nearby pixels and if you translate a pixel the neighbor pixel weighting may be different. So you have to wait a bit for the cleanup update to finish. On slow links x11vnc may automatically decide to not detect scrolls when -scale is in effect. In general it will also try to defer the cleanup update if possible.
If you find the -scrollcopyrect behavior too approximate or distracting you can go back to the standard polling-only update method with the -noscrollcopyrect (or -noscr for short.) If you find some extremely bad and repeatable behavior for -scrollcopyrect please report a bug.
Alternatively, as with -wireframe, there are many tuning parameters to try to improve the situation. You can also access these parameters inside the gui under "Tuning". These parameters can be tweaked:
- The minimum pixel area of a rectangle to be watched for scrolls.
- A list if application names to skip scroll detection.
- Which keystrokes should trigger scroll detection.
- Which applications should have a "terminal" tweak applied to them.
- When repeating keys (e.g. Up arrow) should be discarded to preserve a scroll.
- Cutoffs for closeness to Top, Bottom, Left, and Right edges of window for mouse induced scrolls.
- Set timeout parameters for keystroke induced scrolls.
- Set timeout parameters for mouse pointer induced scrolls.
- Have the full screen be periodically refreshed to fix painting errors.
Q-79: Can x11vnc do client-side caching of pixel data? I.e. so when that pixel data is needed again it does not have to be retransmitted over the network.
As of Dec/2006 in the 0.9 development tarball there is an experimental client-side caching implementation enabled by the "-ncache n" option. In fact, during the test period it was on by default with n set to 10. To disable it use "-noncache".
It is a simple scheme where a (very large) lower portion of the framebuffer (i.e. starting just below the user's actual desktop display) is used for storing pixel data. CopyRect; a fast, essentially local viewer-side VNC encoding; is used to swap the pixel data in and out of the actual display area. It gives an excellent speedup for iconifying/deiconifying and moving windows and re-posting of menus (often it doesn't feel like VNC at all: there is no delay waiting for the pixel data to fill in.)
This scheme is nice because it does all of this within the existing VNC protocol, and so it works with all VNC viewers.
A challenge to doing more sophisticated (e.g. compressed and/or shared) client-side caching is that one needs to extend the VNC protocol, modify a viewer and then also convince users to adopt your modified VNC Viewer (or get the new features to be folded into the main VNC viewers, patches accepted, etc... likely takes many years before they might be deployed in the field.) So it is convenient that the "-ncache n" works with any unaltered VNC viewer.
A drawback of the "-ncache n" method is that in the VNC Viewer you can scroll down and actually see the cached pixel data. So it looks like there is a bug: you can scroll down in your viewer and see a strange "history" of windows on your desktop. This is working as intended. One will need to try to adjust the size of his VNC Viewer window so the cache area cannot be seen. SSVNC (see below) can do this automatically.
At some point LibVNCServer may implement a "rfbFBCrop" pseudoencoding that viewers can use to learn which portion of the framebuffer to actually show to the users (with the hidden part used for caching, or perhaps something else, maybe double buffering or other offscreen rendering...)
The Enhanced TightVNC Viewer (SSVNC) Unix viewer has a nice -ycrop option to help hide the pixel cache area from view. It will turn on automatically if the framebuffer appears to be very tall (height more than twice the width), or you can supply the actual value for the height. If the screen is resized by scaling, etc, the ycrop value is scaled as well. In fullscreen mode you cannot scroll past the end of the actual screen, and in non-fullscreen mode the window manager frame is adjusted to fit the actual display (so you don't see the pixel cache region) and the scrollbars are very thin to avoid distraction and trouble fitting inside your display. Use the "-sbwidth n" viewer option to make the scrollbars thicker if you like.
Another drawback of the scheme is that it is VERY memory intensive, the n in "-ncache n" is the factor of increase over the base framebuffer size to use for caching. It is an even integer and should be fairly large, 6-12, to achieve good response. This usually requires about 50-100MB of additional RAM on both the client and server sides. For example with n=6 a 1280x1024 display will use a framebuffer that is 1280x7168: everything below row 1024 is the pixel buffer cache. If you are running on low memory machines or memory is tight because of other running applications you should not use -ncache.
The reason for so much memory is because the pixel data is not compressed and so the whole window to be saved must be stored "offscreen". E.g. for a large web browser window this can be nearly 1 million pixels, and that is only for a single window! One typically wants to cycle between 5-10 large active windows. Also because both backing-store (the window's actual contents) and save-unders (the pixels covered up by the window) are cached offscreen that introduces an additional factor of 2 in memory use.
However, even in the smallest usage mode with n equal 2 and -ncache_no_rootpixmap set (this requires only 2X additional framebuffer memory) there is still a noticable improvement for many activities, although it is not as dramatic as with, say n equal 12 and rootpixmap (desktop background) caching enabled.
The large memory consumption of the current implementation can be thought of as a tradeoff to providing caching and being compatible with all VNC viewers and also ease of implementing. Hopefully it can be tuned to use less, or the VNC community will extend the protocol to allow caching and replaying of compressed blobs of data.
Another option to experiment with is "-ncache_cr". By specifying it, x11vnc will try to do smooth opaque window moves instead of its wireframe. This can give a very nice effect (note: on Unix the realvnc viewer seems to be smoother than the tightvnc viewer), but can lead to some painting problems, and can be jerky in some circumstances.
Surprisingly, for very slow connections, e.g. modem, the -ncache_cr option can actually improve window drags. This is probably because no pixel data (only CopyRect instructions) are sent when dragging a window. Normally, the wireframe must be sent and this involves compressing and sending the lines that give rise to the moving box effect (note that real framebuffer data is sent to "erase" the white lines of the box.)
If you experience painting errors you can can tap the Alt_L key (Left "Alt" key) 3 times in a row to signal x11vnc to refresh the screen to all viewers. You may also need to iconify and then deiconify any damaged windows to correct their cache data as well. Note that if you change color viewer depth (e.g. 8bpp to full color) dynamically that will usually lead to the entire extended framebuffer being resent which can take a long time over very slow links: it may be better to reconnect and reset the format right after doing so. x11vnc will try to detect the format change and clear (make completely black) the cache region.
Gotcha for older Unix VNC Viewers: The older Unix VNC viewers (e.g. current TightVNC Unix Viewer) require X server backingstore to keep off-viewer screen data local. If the viewer-side X server has backingstore disabled (sadly, currently the default on Linux, etc), then to get the offscreen pixels the viewer has to ask for a refresh over the network, thereby defeating the caching. Use something like this in your viewer-side /etc/X11/xorg.conf file (or otherwise get your viewer-side system to do it)
Section "Device"
...
Option "backingstore"
...
EndSection
No problems like this have been observed with Windows VNC Viewers: they all seem to keep their entire framebuffer in local memory.
Gotcha for KDE krdc VNC Viewer: One user found that KDE's krdc viewer has some sort of hardwired limit on the maximum size of the framebuffer (64MB?). It fails quickly saying "The connection to the host has been interrupted." The workaround for his 1280x1024 x11vnc-side display was to run with "-ncache 10", i.e. a smaller value to be under the krdc threshold.
Although this scheme is not as quick (nor as compressed) as nx/nomachine, say, it does provide a good step in the direction of improving VNC performance by client side caching.
As of Feb/2009 (development tarball) there is an experimental kludge to let you build x11vnc using TurboVNC's modified TightVNC encoding. TurboVNC is part of the VirtualGL project. It does two main things to speed up the TightVNC encoding:
- It eliminates bottlenecks, overheads, wait-times in the TightVNC encoding implementation and instead only worries about sending very well (and quickly) compressed JPEG data.
- A fast proprietary JPEG implemention is used (Intel IPP on x86) instead of the usual libjpeg implementation. TurboJPEG is an interface library, libturbojpeg, provided by the project that achieves this.
TurboVNC works very well over LAN and evidently fast Broadband too. When using it with x11vnc in such a situation you may want to dial down the delays, e.g. "-wait 5" and "-defer 5" (or even a smaller setting) to poll and pump things out more quickly.
See the instructions in "x11vnc/misc/turbovnc/README" for how to build x11vnc with TurboVNC support. You will also need to download the TurboJPEG software.
In brief, the steps look like this:
cd x11vnc-x.y.z/x11vnc/misc/turbovnc
./apply_turbovnc
cd ../../..
env LDFLAGS='-L/DIR -Xlinker --rpath=/DIR' ./configure
make AM_LDFLAGS='-lturbojpeg'
where you replace "/DIR" with the directory containing libturbojpeg.so you downloaded separately. If it works out well enough TurboVNC support will be integrated into x11vnc and more of its tuning features will be implemented. Support for TurboVNC in SSVNC viewer has been added as an experiment as well. If you try either one, let us know how it went.
There also may be some Linux.i686 and Darwin.i386 x11vnc binaries with TurboVNC support in the misc. bins directory. For other platforms you will need to compile yourself.
On relatively cheap and old hardware (Althon64 X2 5000+ / GeForce 6200) x11vnc and SSVNC, both TurboVNC enabled, were able to sustain 13.5 frames/sec (fps) and 15 Megapixels/sec using the VirtualGL supplied OpenGL benchmark program glxspheres. VirtualGL on higher-end hardware can sustain 20-30 fps with the glxspheres benchmark.
Potential Slowdown: As we describe elsewhere, unless you use x11vnc with an X server using, say, NVidia proprietary drivers (or a virtual X server like Xvfb or Xdummy, or in ShadowFB mode), then the read rate from the graphics card can be rather slow (e.g. 10 MB/sec) and becomes the bottleneck when using x11vnc over fast networks. Note that all of Xorg's drivers currently (2009) have slow read rates (only proprietary drivers appear to have optimized reads.)
So under these (more or less typical) conditions, the speed improvement provided by TurboVNC may only be marginal. Look for this output to see your read rate:
28/02/2009 11:11:07 Autoprobing TCP port
28/02/2009 11:11:07 Autoprobing selected port 5900
28/02/2009 11:11:08 fb read rate: 10 MB/sec
28/02/2009 11:11:08 screen setup finished.
A rate of 10 MB/sec means a 1280x1024x24 screen takes 0.5 seconds to read in. TurboVNC compresses that to JPEG in a much shorter time. On the other hand, an NVidia driver may have a read rate of 250 MB/sec and so only takes 0.02 seconds to read the entire screen in.
Q-81: Why isn't the mouse cursor shape (the little icon shape where the mouse pointer is) correct as I move from window to window?
On X servers supporting XFIXES or Solaris/IRIX Overlay extensions it is possible for x11vnc to do this correctly. See a few paragraphs down for the answer.
Historically, the X11 mouse cursor shape (i.e. little picture: an arrow, X, I-beam, resizer, etc) is one of the few WRITE-only objects in X11. That is, an application can tell the X server what the cursor shape should be when the pointer is in a given window, but a program (like x11vnc) unfortunately cannot read this information. I believe this is because the cursor shape is often downloaded to the graphics hardware (video card), but I could be mistaken.
A simple kludge is provided by the "-cursor X" option that changes the cursor when the mouse is on the root background (or any window has the same cursor as the root background.) Note that desktops like GNOME or KDE often cover up the root background, so this won't work for those cases. Also see the "-cursor some" option for additional kludges.
Note that as of Aug/2004 on Solaris using the SUN_OVL overlay extension and IRIX, x11vnc can show the correct mouse cursor when the -overlay option is supplied. See this FAQ for more info.
Also as of Dec/2004 XFIXES X extension support has been added to allow exact extraction of the mouse cursor shape. XFIXES fixes the problem of the cursor-shape being write-only: x11vnc can now query the X server for the current shape and send it back to the connected viewers. XFIXES is available on recent Linux Xorg based distros and Solaris 10.
The only XFIXES issue is the handling of alpha channel transparency in cursors. If a cursor has any translucency then in general it must be approximated to opaque RGB values for use in VNC. There are some situations where the cursor transparency can also handled exactly: when the VNC Viewer requires the cursor shape be drawn into the VNC framebuffer or if you apply a patch to your VNC Viewer to extract hidden alpha channel data under 32bpp. Details can be found here.
Q-82: When using XFIXES cursorshape mode, some of the cursors look really bad with extra black borders around the cursor and other cruft. How can I improve their appearance?
This happens for cursors with transparency ("alpha channel"); regular X cursors (bitmaps) should be correct. Unfortunately x11vnc 0.7 was released with a very poor algorithm for approximating the transparency, which led to the ugly black borders.
The problem is as follows: XFIXES allows x11vnc to retrieve the current X server cursor shape, including the alpha channel for transparency. For traditional bitmap cursors the alpha value will be 0 for completely transparent pixels and 255 for completely opaque pixels; whereas for modern, eye-candy cursors an alpha value between 0 and 255 means to blend in the background colors to that degree with the cursor colors. The pixel color blending formula is something like this: Red = Red_cursor * a + Red_background * (1 - a), (where here 0 =< a =< 1), with similar for Green and Blue. The VNC protocol does not currently support an alpha channel in cursors: it only supports regular X bitmap cursors and Rich Cursors that have RGB (Red, Green, Blue) color data, but no "A" = alpha data. So in general x11vnc has to approximate a cursor with transparency to create a Rich Cursor. This is easier said than done: some cursor themes have cursors with complicated drop shadows and other forms of translucency.
Anyway, for the x11vnc 0.7.1 release the algorithm for approximating transparency is much improved and hopefully gives decent cursor shapes for most cursor themes and you don't have to worry about it.
In case it still looks bad for your cursor theme, there are (of course!) some tunable parameters. The "-alphacut n" option lets you set the threshold "n" (between 0 and 255): cursor pixels with alpha values below n will be considered completely transparent while values equal to or above n will be completely opaque. The default is 240. The "-alphafrac f" option tries to correct individual cursors that did not fare well with the default -alphacut value: if a cursor has less than fraction f (between 0.0 and 1.0) of its pixels selected by the default -alphacut, the threshold is lowered until f of its pixels are selected. The default fraction is 0.33.
Finally, there is an option -alpharemove that is useful for themes where many cursors are light colored (e.g. "whiteglass".) XFIXES returns the cursor data with the RGB values pre-multiplied by the alpha value. If the white cursors look too grey, specify -alpharemove to brighten them by having x11vnc divide out the alpha value.
One user played with these parameters and reported back:
Of the cursor themes present on my system:
gentoo and gentoo-blue: alphacut:192 - noalpharemove
gentoo-silver: alphacut:127 and alpharemove
whiteglass and redglass (presumably also handhelds, which is based
heavily on redglass) look fine with the apparent default of alphacut:255.
As of Jan/2005 libvncserver has been modified to allow an alpha channel (i.e. RGBA data) for Rich Cursors. So x11vnc can now send the alpha channel data to libvncserver. However, this data will only be used for VNC clients that do not support the CursorShapeUpdates VNC extension (or have disabled it.) It can be disabled for all clients with the -nocursorshape x11vnc option. In this case the cursor is drawn, correctly blended with the background, into the VNC framebuffer before being sent out to the client. So the alpha blending is done on the x11vnc side. Use the -noalphablend option to disable this behavior (always approximate transparent cursors with opaque RGB values.)
The CursorShapeUpdates VNC extension complicates matters because the cursor shape is sent to the VNC viewers supporting it, and the viewers draw the cursor locally. This improves response over slow links. Alpha channel data for these locally drawn cursors is not supported by the VNC protocol.
However, in the libvncserver CVS there is a patch to the TightVNC viewer to make this work for CursorShapeUpdates under some circumstances. This hack is outside of the VNC protocol. It requires the screens on both sides to be depth 24 at 32bpp (it uses the extra 8 bits to secretly hide the cursor alpha channel data.) Not only does it require depth 24 at 32bpp, but it also currently requires the client and server to be of the same endianness (otherwise the hidden alpha data gets reset to zero by a libvncserver translation function; we can fix this at some point if there is interest.) The patch is for the TightVNC 1.3dev5 Unix vncviewer and it enables the TightVNC viewer to do the cursor alpha blending locally. The patch code should give an example on how to change the Windows TightVNC viewer to achieve the same thing (send me the patch if you get that working.)
This patch is applied to the Enhanced TightVNC Viewer (SSVNC) package we provide.
Q-84: Why does the mouse arrow just stay in one corner in my vncviewer, whereas my cursor (that does move) is just a dot?
This default takes advantage of a tightvnc extension (CursorShapeUpdates) that allows specifying a cursor image shape for the local VNC viewer. You may disable it with the -nocursor option to x11vnc if your viewer does not have this extension.
Note: as of Aug/2004 this should be fixed: the default for non-tightvnc viewers (or ones that do not support CursorShapeUpdates) will be to draw the moving cursor into the x11vnc framebuffer. This can also be disabled via -nocursor.
Q-85: Can I take advantage of the TightVNC extension to the VNC protocol where Cursor Positions Updates are sent back to all connected clients (i.e. passive viewers can see the mouse cursor being moved around by another viewer)?
Use the -cursorpos option when starting x11vnc. A VNC viewer must support the Cursor Positions Updates for the user to see the mouse motions (the TightVNC viewers support this.) As of Aug/2004 -cursorpos is the default. See also -nocursorpos and -nocursorshape.
Q-86: Is it possible to swap the mouse buttons (e.g. left-handed operation), or arbitrarily remap them? How about mapping button clicks to keystrokes, e.g. to partially emulate Mouse wheel scrolling?
You can remap the mouse buttons via something like: -buttonmap 13-31 (or perhaps 12-21.) Also, note that xmodmap(1) lets you directly adjust the X server's button mappings, but in some circumstances it might be more desirable to have x11vnc do it.
One user had an X server with only one mouse button(!) and was able to map all of the VNC client mouse buttons to it via: -buttonmap 123-111.
Note that the -debug_pointer option prints out much info for every mouse/pointer event and is handy in solving problems.
To map mouse button clicks to keystrokes you can use the alternate format where the keystrokes are enclosed between colons like this :: in place of the mouse button digit. For a sequence of keysyms separate them with "+" signs. Look in the include file <X11/keysymdef.h>, or use xev(1), or -debug_keyboard to find the keysym names. Button clicks can also be included in the sequence via the fake keysyms Button1, etc.
As an example, suppose the VNC viewer machine has a mouse wheel (these generate button 4 and 5 events), but the machine that x11vnc is run on only has the 3 regular buttons. In normal operation x11vnc will discard the button 4 and 5 events. However, either of the following button maps could possibly be of use emulating the mouse wheel events in this case:
-buttonmap 12345-123:Prior::Next:
-buttonmap 12345-123:Up+Up+Up::Down+Down+Down:
Exactly what keystroke "scrolling" events they should be bound to depends on one's taste. If this method is too approximate, one could consider not using -buttonmap but rather configuring the X server to think it has a mouse with 5 buttons even though the physical mouse does not. (e.g. 'Option "ZAxisMapping" "4 5"'.)
Note that when a keysym-mapped mouse button is clicked down this immediately generates the key-press and key-release events (for each keysym in turn if the mapping has a sequence of keysyms.) When the mouse button goes back up nothing is generated.
If you include modifier keys like Shift_L instead of key-press immediately followed by key-release the state of the modifier key is toggled (however the initial state of the modifier key is ignored.) So to map the right button to type my name 'Karl Runge' I could use this:
-buttonmap 3-:Shift_L+k+Shift_L+a+r+l+space+Shift_L+r+Shift_L+u+n+g+e:
(yes, this is getting a little silly.)
BTW, Coming the other way around, if the machine you are sitting at does not have a mouse wheel, but the remote machine does (or at least has 5 buttons configured), this key remapping can be useful:
-remap Super_R-Button4,Menu-Button5
you just tap those two keys to get the mouse wheel scrolls (this is more useful than the Up and Down arrow keys because a mouse wheel "click" usually gives a multi-line scroll.)
The option -modtweak should help here. It is a mode that monitors the state of the Shift and AltGr Modifiers and tries to deduce the correct keycode to send, possibly by sending fake modifier key presses and releases in addition to the actual keystroke.
Update: As of Jul/2004 -modtweak is now the default (use -nomodtweak to get the old behavior.) This was done because it was noticed on newer XFree86 setups even on bland "us" keyboards like "pc104 us" XFree86 included a "ghost" key with both "<" and ">" it. This key does not exist on the keyboard (see this FAQ for more info.) Without -modtweak there was then an ambiguity in the reverse map keysym => keycode, making it so the "<" symbol could not be typed.
Also see the FAQ about the -xkb option for a more powerful method of modifier tweaking for use on X servers with the XKEYBOARD extension.
When trying to resolve keyboard mapping problems, note that the -debug_keyboard option prints out much info for every keystroke and so can be useful debugging things.
Note that for some users, the solution is to disable all of the above, and use -nomodtweak. This is the simplest form of keystroke insertion and it actually solved the problem. Try it if the other options don't help.
Q-88: When I try to type a "<" (i.e. less than) instead I get ">" (i.e. greater than)! Strangely, typing ">" works OK!!
Does your keyboard have a single key with both "<" and ">" on it? Even if it doesn't, your X server may think your keyboard has such a key (e.g. pc105 in the XF86Config file when it should be something else, say pc104.)
Short Cut: Try the -xkb or -sloppy_keys options and see if that helps the situation. The discussion below is a bit outdated (e.g. -modtweak is now the default) but it is useful reference for various tricks and so is kept.
The problem here is that on the Xserver where x11vnc is run there are two keycodes that correspond to the "<" keysym. Run something like this to see:
xmodmap -pk | egrep -i 'KeyCode|less|greater'
There are 4 KeySyms per KeyCode; KeyCodes range from 8 to 255.
KeyCode Keysym (Keysym) ...
59 0x002c (comma) 0x003c (less)
60 0x002e (period) 0x003e (greater)
94 0x003c (less) 0x003e (greater)
That keycode 94 is the special key with both "<" and ">". When x11vnc receives the "<" keysym over the wire from the remote VNC client, it unfortunately maps it to keycode 94 instead of 59, and sends 94 to the X server. Since Shift is down (i.e. you are Shifting the comma key), the X server interprets this as Shifted-94, which is ">".
A workaround in the X server configuration is to "deaden" that special key:
xmodmap -e "keycode 94 = "
However, one user said he had to do this:
xmodmap -e "keycode 94 = 0x002c 0x003c"
(If the numerical values are different for your setup, substitute the ones that correspond to your display. The above xmodmap scheme can often be used to work around other ambiguous keysym to keycode mappings.)
Alternatively, here are some x11vnc options to try to work around the problem:
-modtweak
and
-remap less-comma
These are convenient in that they do not modify the actual X server settings. The former (-modtweak) is a mode that monitors the state of the Shift and AltGr modifiers and tries to deduce the correct keycode sequence to send. Since Jul/2004 -modtweak is now the default. The latter (-remap less-comma) is an immediate remapping of the keysym less to the keysym comma when it comes in from a client (so when Shift is down the comma press will yield "<".)
See also the FAQ about the -xkb option as a possible workaround using the XKEYBOARD extension.
Note that the -debug_keyboard option prints out much info for every keystroke to aid debugging keyboard problems.
Q-89: Extra Character Inserted, E.g.: When I try to type a "<" (i.e. less than) instead I get "<," (i.e. an extra comma.)
This is likely because you press "Shift" then "<" but then released the Shift key before releasing the "<". Because of a keymapping ambiguity the last event "< up" is interpreted as "," because that key unshifted is the comma.
This extra character insertion will happen for other combinations of characters: in general it can happen whenever the Shift key is released early.
This should not happen in -xkb mode, because it works hard to resolve the ambiguities. If you do not want to use -xkb, try the option -sloppy_keys to attempt a similar type of algorithm.
One user had this problem for Italian and German keyboards with the key containing ":" and "." When he typed ":" he would get an extra "." inserted after the ":". The solution was -sloppy_keys.
Q-90: I'm using an "international" keyboard (e.g. German "de", or Danish "dk") and the -modtweak mode works well if the VNC viewer is run on a Unix/Linux machine with a similar keyboard. But if I run the VNC viewer on Unix/Linux with a different keyboard (e.g. "us") or Windows with any keyboard, I can't type some keys like: "@", "$", "<", ">", etc. How can I fix this?
The problem with Windows is it does not seem to handle AltGr well. It seems to fake it up by sending Control_L+Alt_R to applications. The Windows VNC viewer sends those two down keystrokes out on the wire to the VNC server, but when the user types the next key to get, e.g., "@" the Windows VNC viewer sends events bringing the up the Control_L+Alt_R keys, and then sends the "@" keysym by itself.
The Unix/Linux VNC viewer on a "us" keyboard does a similar thing since "@" is the Shift of the "2" key. The keysyms Shift and "@" are sent to the VNC server.
In both cases no AltGr is sent to the VNC server, but we know AltGr is needed on the physical international keyboard to type a "@".
This all worked fine with x11vnc running with the -modtweak option (it figures out how to adjust the Modifier keys (Shift or AltGr) to get the "@".) However it fails under recent versions of XFree86 (and the X.org fork.) These run the XKEYBOARD extension by default and make heavy use of it to handle international keyboards.
To make a long story short, on these newer XFree86 setups the traditional X keymap lookup x11vnc uses is no longer accurate. x11vnc can't find the keysym "@" anywhere in the keymapping! (even though it is in the XKEYBOARD extended keymapping.)
How to Solve: As of Jul/2004 x11vnc has two changes:
- -modtweak (tweak Modifier keys) is now the default (use -nomodtweak to go back to the old way)
- there is a new option -xkb to use the XKEYBOARD extension API to do the Modifier key tweaking.
The -xkb option seems to fix all of the missing keys: "@", "<", ">", etc.: it is recommended that you try it if you have this sort of problem. Let us know if there are any remaining problems (see the next paragraph for some known problems.) If you specify the -debug_keyboard (aka -dk) option twice you will get a huge amount of keystroke debugging output (send it along with any problems you report.)
Update: as of Jun/2005 x11vnc will try to automatically enable -xkb if it appears that would be beneficial (e.g. if it sees any of "@", "<", ">", "[" and similar keys are mapped in a way that needs the -xkb to access them.) To disable this automatic check use -noxkb.
Known problems:
-
One user had to disable a "ghost" Mode_switch key that was causing problems under -xkb. His physical AltGr key was bound to ISO_Level3_Shift (which seems to be the XKEYBOARD way of doing things), while there was a ghost key Mode_switch (which seems to be obsolete) in the mapping as well. Both of these keysyms were bound to Mod5 and x11vnc was unfortunately choosing Mode_switch. From the x11vnc -xkb -dk -dk output it was noted that Mode_switch was attached to keycode 93 (no physical key generates this keycode) while ISO_Level3_Shift was attached to keycode 113. The keycode skipping option was used to disable the ghost key: -skip_keycodes 93
-
In implementing -xkb we noticed that some characters were still not getting through, e.g. "
" and "^". This is not really an XKEYBOARD problem. What was happening was the VNC viewer was sending the keysyms asciitilde and asciicircum to x11vnc, but on the X server with the international keyboard those keysyms were not mapped to any keys. So x11vnc had to skip them (Note: as of May/2005 they are added by default see -add_keysyms below.) The way these characters are typically entered on international keyboards is by "dead" (aka "mute") keys. E.g. to enter "" at the physical display the keysym dead_tilde is pressed and released (this usually involves holding AltGr down while another key is pressed) and then space is pressed. (this can also be used get characters with the "~" symbol on top, e.g. "\E3" by typing "a" instead of space.) What to do? In general the VNC protocol has not really solved this problem: what should be done if the VNC viewer sends a keysym not recognized by the VNC server side? Workarounds can possibly be created using the -remap x11vnc option:-remap asciitilde-dead_tilde,asciicircum-dead_circumflexetc. Use -remap filename if the list is long. Please send us your workarounds for this problem on your keyboard. Perhaps we can have x11vnc adjust automatically at some point. Also see the -add_keysyms option in the next paragraph. Update: for convenience "-remap DEAD" does many of these mappings at once.
-
To complement the above workaround using the -remap, an option -add_keysyms was added. This option instructs x11vnc to bind any unknown Keysyms coming in from VNC viewers to unused Keycodes in the X server. This modifies the global state of the X server. When x11vnc exits it removes the extra keymappings it created. Note that the -remap mappings are applied first, right when the Keysym is received from a VNC viewer, and only after that would -add_keysyms, or anything else, come into play. Update: -add_keysyms is now on by default. Use -noadd_keysyms to disable.
Q-91: When typing I sometimes get double, triple, or more of my keystrokes repeated. I'm sure I only typed them once, what can I do?
This may be due to an interplay between your X server's key autorepeat delay and the extra time delays caused by x11vnc processing.
Short answer: disable key autorepeating by running the command "xset r off" on the Xserver where x11vnc is run (restore via "xset r on") or use the new (Jul/2004) -norepeat x11vnc option. You will still have autorepeating because that is taken care of on your VNC viewer side.
Update: as of Dec/2004 -norepeat is now the default. Use -repeat to disable it.
Details: suppose you press a key DOWN and it generates changes in large regions of the screen. The CPU and I/O work x11vnc does for the large screen change could be longer than your X server's key autorepeat delay. x11vnc may not get to processing the key UP event until after the screen work is completed. The X server believes the key has been held down all this time, and applies its autorepeat rules.
Even without inducing changes in large regions of the screen, this problem could arise when accessing x11vnc via a dialup modem or otherwise high latency link (e.g. > 250 ms latency.)
Look at the output of "xset q" for the "auto repeat delay" setting. Is it low (e.g. < 300 ms)? If you turn off autorepeat completely: "xset r off", does the problem go away?
The workaround is to manually apply "xset r off" and "xset r on" as needed, or to use the -norepeat (which has since Dec/2004 been made the default.) Note that with X server autorepeat turned off the VNC viewer side of the connection will (nearly always) do its own autorepeating so there is no big loss here, unless someone is also working at the physical display and misses his autorepeating.
Are you using x11vnc to log in to an X session via display manager? (as described in this FAQ) If so, x11vnc is starting before your session and it disables autorepeat when you connect, but then after you log in your session startup (GNOME, KDE, ...) could be resetting the autorepeat to be on. Or it could be something inside your desktop trying to be helpful that decides to turn it back on.
x11vnc in -norepeat mode will by default reset autorepeat to off 2 times (to help get thru the session startup problem), but it will not continue to battle with things turning autorepeat back on. It will also turn autorepeat off whenever it goes from a state of zero clients to one client. You can adjust the number of resets via "-norepeat N", or use "-norepeat -1" to have it keep resetting it whenever autorepeat gets turned back on when clients are connected.
In general you can manually turn autorepeating off by typing "xset r off", or a using desktop utility/menu, or "x11vnc -R norepeat". If something in your desktop is automatically turning it back on you should figure out how to disable that somehow.
Q-93: After using x11vnc for a while, I find that I cannot type some (or any) characters or my mouse clicks and drags no longer have any effect, or they lead to strange effects. What happened?
Probably a modifier key, e.g. Control or Alt is "stuck" in a pressed down state.
This happens for VNC in general by the following mechanism. Suppose on the Viewer side desktop there is some hot-key to switch desktops/rooms/spaces, etc. E.g. suppose Alt+LeftArrow moves to the left desktop/room/space. Or suppose an Alt+hotkey combination iconifies a window. This can leave the Alt key pressed down on the remote side.
Consider the sequence that happens. The Alt_L key and then the LeftArrow key go down. Since you are inside the viewer the Alt_L key press is sent to the other side (x11vnc) and so it is pressed down in the remote desktop as well. (by "Alt_L" we mean the Alt key on the left-hand side of the keyboard.) Your local desktop (where the VNC Viewer is running) then warps to the new desktop/room/space: Leaving the Alt_L key still pressed down in the remote desktop.
If someone is sitting at the desktop, or when you return in the viewer it may be very confusing because the Alt_L is still pressed down but you (or the person sitting at the desktop) do not realize this. Depending on which remote desktop (x11vnc side) is used, it can act very strangely.
A quick workaround when you notice this is to press and release all of the Alt, Shift, Control, Windows-Flag, modifier keys to free the pressed one. You need to do this for both the left and right Shift, Alt, Control, etc. keys to be sure.
Note that many VNC Viewers try to guard against this when they are notified by the window system that the viewer app has "lost focus". When it receives the "lost focus" event, the viewer sends VNC Key-Release events for all modifier keys that are currently pressed down. This does not always work, however, since it depends on how the desktop manages these "warps". If the viewer is not notified it cannot know it needs to release the modifiers.
You can also use the -clear_mods option to try to clear all of the modifier keys at x11vnc startup. You will still have to be careful that you do not leave the modifier key pressed down during your session. It is difficult to prevent this problem from occurring (short of using -remap to prevent sending all of the problem modifier keys, which would make the destkop pretty unusable.)
During a session these x11vnc remote control commands can also help:
x11vnc -R clear_mods
x11vnc -R clear_keys
x11vnc -R clear_locks
x11vnc -R clear_all
A similar problem can occur if you accidentally press the Caps_Lock or Num_Lock down. When these are locked on the remote side it can sometimes lead to strange desktop behavior (e.g. cannot drag or click on windows.) As above you may not notice this because the lock isn't down on the local (Viewer) side. See this FAQ on lock keys problem. These options may help avoid the problem: -skip_lockkeys and -capslock. See also -clear_all.
Q-94: The machine where I run x11vnc has an AltGr key, but the local machine where I run the VNC viewer does not. Is there a way I can map a local unused key to send an AltGr? How about a Compose key as well?
Something like "-remap Super_R-Mode_switch" x11vnc option may work. Note that Super_R is the "Right Windoze(tm) Flaggie" key; you may want to choose another. The -debug_keyboard option comes in handy in finding keysym names (so does xev(1).)
For Compose how about "-remap Menu-Multi_key" (note that Multi_key is the official name for Compose.) To do both at the same time: "-remap Super_R-Mode_switch,Menu-Multi_key" or use "-remap filename" to specify remappings from a file.
Q-95: I have a Sun machine I run x11vnc on. Its Sun keyboard has just one Alt key labelled "Alt" and two Meta keys labelled with little diamonds. The machine where I run the VNC viewer only has Alt keys. How can I send a Meta keypress? (e.g. emacs needs this)
Here are a couple ideas. The first one is to simply use xmodmap(1) to adjust the Sun X server. Perhaps xmodmap -e "keysym Alt_L = Meta_L Alt_L" will do the trick. (there are other ways to do it, one user used: xmodmap -e "keycode 26 = Meta_L" for his setup.)
Since xmodmap(1) modifies the X server mappings you may not want to do this (because it affects local work on that machine.) Something like the -remap Alt_L-Meta_L to x11vnc may be sufficient for ones needs, and does not modify the X server environment. Note that you cannot send Alt_L in this case, maybe -remap Super_L-Meta_L would be a better choice if the Super_L key is typically unused in Unix.
One user reports this problem on HP-UX Rel_B.11.23. The problem was traced to a strange keyboard mapping for the machine (e.g. xmodmap -pk output) that looked like:
...
039 2 at at at
...
047 3 numbersign numbersign numbersign
and similar triple mappings (with two in the AltGr/Mode_switch group) of a keysum to a single keycode.
Use the -nomodtweak option as a workaround. You can also use xmodmap to correct these mappings in the server, e.g.:
xmodmap -e "keycode 47 = 3 numbersign"
Also, as of Feb/2007, set the environment variable MODTWEAK_LOWEST=1 (either in your shell or via "-env MODTWEAK_LOWEST=1" option) to handle these mappings better.
This can be done directly in some X servers using AccessX and Pointer_EnableKeys, but is a bit awkward. It may be more convenient to have x11vnc do the remapping. This can be done via the -remap option using the fake "keysyms" Button1, Button2, etc. as the "to" keys (i.e. the ones after the "-")
As an example, consider a laptop where the VNC viewer is run that has a touchpad with only two buttons. It is difficult to do a middle button "paste" because (using XFree86/Xorg Emulate3Buttons) you have to click both buttons on the touch pad at the same time. This remapping: -remap Super_R-Button2
maps the Super_R "flag" key press to the Button2 click, thereby making X pasting a bit easier.
Note that once the key goes down, the button down and button up events are generated immediately on the x11vnc side. When the key is released (i.e. goes up) no events are generated.
This is a little tricky because it is possible to get the Caps_Lock state out of sync between your viewer-side machine and the x11vnc-side X server. For best results, we recommend not ever letting the Caps_Lock keypresses be processed by x11vnc. That way when you press Caps_Lock in the viewer your local machine goes into the Caps_Lock on state and sends keysym "A" say when you press "a". x11vnc will then fake things up so that Shift is held down to generate "A". The -skip_lockkeys option should help to accomplish this. For finer grain control use something like: "-remap Caps_Lock-None".
Also try the -nomodtweak and -capslock options.
Another useful option that turns off any Lock keys on the remote side at startup and end is the -clear_all option. During a session you can run these remote control commands to modify the Lock keys: x11vnc -R clear_locks x11vnc -R clear_all
the former will try to unset any Lock keys, the latter will do same and also try to make it so no key is pressed down (e.g. "stuck" Alt_L, etc.)
Q-99: The remote display is larger (in number of pixels) than the local display I am running the vncviewer on. I don't like the vncviewer scrollbars, what I can do?
vncviewer has a option (usually accessible via F8 key or -fullscreen option) for vncviewer to run in full screen, where it will automatically scroll when the mouse is near the edge of the current view. For quick scrolling, also make sure Backing Store is enabled on the machine vncviewer is run on. (XFree86/Xorg disables it by default for some reason, add Option "backingstore" to XF86Config on the vncviewer side.)
BTW, contact me if you are having problems with vncviewer in fullscreen mode with your window manager (i.e. no keyboard response.) I have a workaround for vncviewer using XGrabServer().
There may also be scaling viewers out there (e.g. TightVNC or UltraVNC on Windows) that automatically shrink or expand the remote framebuffer to fit the local display. Especially for hand-held devices. See also the next FAQ on x11vnc scaling.
As of Jun/2004 x11vnc provides basic server-side scaling. It is a global scaling of the desktop, not a per-client setting. To enable it use the "-scale fraction" option. "fraction" can either be a floating point number (e.g. -scale 0.75) or the alternative m/n fraction notation (e.g. -scale 3/4.) Note that if fraction is greater than one the display is magnified.
Extra resources (CPU, memory I/O, and memory) are required to do the scaling. If the machine is slow where x11vnc is run with scaling enabled, the interactive response can be unacceptable. OTOH, if run with scaling on a fast machine the performance degradation is usually not a big issue or even noticeable.
It may help to compile x11vnc with compiler option -O3 or -O4 to speed up the scaling code. Set the CFLAGS env. var. before running configure.
Also, if you just want a quick, rough "thumbnail" of the display you can append ":nb" to the fraction to turn on "no blending" mode. E.g.: "-scale 1/3:nb" Fonts will be difficult to read, but the larger features will be recognizable. BTW, "no blending" mode is forced on when scaling 8bpp PseudoColor displays (because blending an indexed colormap is a bad idea and leads to random colors, use :fb to force it on.)
One can also use the ":nb" with an integer scale factor (say "-scale 2:nb") to use x11vnc as a screen magnifier for vision impaired applications. Since with integer scale factors the framebuffers become huge and scaling operations time consuming, be sure to use ":nb" for the fastest response.
In general for a scaled display if you are using a TightVNC viewer you may want to turn off jpeg encoding (e.g. vncviewer -nojpeg host:0.) There appears to be a noise enhancement effect, especially for regions containing font/text: the scaling can introduce some pixel artifacts that evidently causes the tight encoding algorithm to incorrectly detect the regions as image data and thereby introduce additional pixel artifacts due to the lossiness of the jpeg compression algorithm. Experiment to see if -nojpeg vncviewer option improves the readability of text when using -scale to shrink the display size. Also note that scaling may actually slow down the transfer of text regions because after being scaled they do not compress as well. (this can often be a significant slowdown, e.g. 10X.)
Another issue is that it appears VNC viewers require the screen width to be a multiple of 4. When scaling x11vnc will round the width to the nearest multiple of 4. To disable this use the ":n4" sub option (like ":nb" in the previous paragraph; to specify both use a comma: ":nb,n4", etc.)
If one desires per-client scaling for something like 1:1 from a workstation and 1:2 from a smaller device (e.g. handheld), currently the only option is to run two (or more) x11vnc processes with different scalings listening on separate ports (-rfbport option, etc.)
Update: As of May/2006 x11vnc also supports the UltraVNC server-side scaling. This is a per-client scaling by factors 1/2, 1/3, ... and so may be useful for PDA's ("-scale 1/2", etc. will give similar results except that it applies to all clients.) You may need to supply "-rfbversion 3.6" for this to be recognized by UltraVNC viewers.
BTW, whenever you run two or more x11vnc's on the same X display and use the GUI, then to avoid all of the x11vnc's simultaneously answering the gui you will need to use something like "-connect file1 -gui ..." with different connect files for each x11vnc you want to control via the gui (or remote-control.) The "-connect file1" usage gives separate communication channels between a x11vnc process and the gui process. Otherwise they all share the same X property channels: VNC_CONNECT and X11VNC_REMOTE.
Update: As of Mar/2005 x11vnc now scales the mouse cursor with the same scale factor as the screen. If you don't want that, use the "-scale_cursor frac" option to set the cursor scaling to a different factor (e.g. use "-scale_cursor 1" to keep the cursor at its natural unscaled size.)
Q-101: Does x11vnc work with Xinerama? (i.e. multiple monitors joined together to form one big, single screen.)
Yes, it should generally work because it simply polls the big effective screen.
If the viewing-end monitor is not as big as the remote Xinerama display, then the vncviewer scrollbars, etc, will have to be used to pan across the large area. However one user started two x11vnc's, one with "-clip 1280x1024+0+0" and the other with "-clip 1280x1024+1280+0" to split the big screen into two and used two VNC viewers to access them.
As of Jun/2008: Use "-clip xinerama0" to clip to the first xinerama sub-screen (if xinerama is active.) xinerama1 for the 2nd sub-screen, etc. This way you don't need to figure out the WxH+X+Y of the desired xinerama sub-screen. screens are sorted in increasing distance from the (0,0) origin (I.e. not the Xserver's order.)
There are a couple potential issues with Xinerama however. If the screen is not rectangular (e.g. 1280x1024 and 1024x768 monitors joined together), then there will be "non-existent" areas on the screen. The X server will return "garbage" image data for these areas and so they may be distracting to the viewer. The -blackout x11vnc option allows you to blacken-out rectangles by manually specifying their WxH+X+Y geometries. If your system has the libXinerama library, the -xinerama x11vnc option can be used to have it automatically determine the rectangles to be blackened out. (Note on 8bpp PseudoColor displays the fill color may not be black.) Update: -xinerama is now on by default.
Some users have reported that the mouse does not behave properly for their Xinerama display: i.e. the mouse cannot be moved to all regions of the large display. If this happens try using the -xwarppointer option. This instructs x11vnc to fake mouse pointer motions using the XWarpPointer function instead of the XTestFakeMotionEvent XTEST function. (This may be due to a bug in the X server for XTEST when Xinerama is enabled.) Update: As of Dec/2006 -xwarppointer will be applied automatically if Xinerama is detected. To disable use: -noxwarppointer
Q-102: Can I use x11vnc on a multi-headed display that is not Xinerama (i.e. separate screens :0.0, :0.1, ... for each monitor)?
You can, but it is a little bit awkward: you must start separate x11vnc processes for each screen, and on the viewing end start up separate VNC viewer processes connecting to them. e.g. on the remote end:
x11vnc -display :0.0 -bg -q -rfbport 5900
x11vnc -display :0.1 -bg -q -rfbport 5901
(this could be automated in the display manager Xsetup for example) and then on the local machine where you are sitting:
vncviewer somehost:0 &
vncviewer somehost:1 &
Note: if you are running on Solaris 8 or earlier you can easily hit up against the maximum of 6 shm segments per process (for Xsun in this case) from running multiple x11vnc processes. You should modify /etc/system as mentioned in another FAQ to increase the limit. It is probably also a good idea to run with the -onetile option in this case (to limit each x11vnc to 3 shm segments), or even -noshm to use no shm segments.
Q-103: Can x11vnc show only a portion of the display? (E.g. for a special purpose application or a very large screen.)
As of Mar/2005 x11vnc has the "-clip WxH+X+Y" option to select a rectangle of width W, height H and offset (X, Y). Thus the VNC screen will be the clipped sub-region of the display and be only WxH in size. One user used -clip to split up a large Xinerama screen into two more managable smaller screens.
This also works to view a sub-region of a single application window if the -id or -sid options are used. The offset is measured from the upper left corner of the selected window.
Q-104: Does x11vnc support the XRANDR (X Resize, Rotate and Reflection) extension? Whenever I rotate or resize the screen x11vnc just seems to crash.
As of Dec/2004 x11vnc supports XRANDR. You enable it with the -xrandr option to make x11vnc monitor XRANDR events and also trap X server errors if the screen change occurred in the middle of an X call like XGetImage. Once it traps the screen change it will create a new framebuffer using the new screen.
If the connected vnc viewers support the NewFBSize VNC extension (Windows TightVNC viewer and RealVNC 4.0 windows and Unix viewers do) then the viewer will automatically resize. Otherwise, the new framebuffer is fit as best as possible into the original viewer size (portions of the screen may be clipped, unused, etc.) For these viewers you can try the -padgeom option to make the region big enough to hold all resizes and rotations. We have fixed this problem for the TightVNC Viewer on Unix: SSVNC
If you specify "-xrandr newfbsize" then vnc viewers that do not support NewFBSize will be disconnected before the resize. If you specify "-xrandr exit" then all will be disconnected and x11vnc will terminate.
Q-105: Independent of any XRANDR, can I have x11vnc rotate and/or reflect the screen that the VNC viewers see? (e.g. for a handheld whose screen is rotated 90 degrees.)
As of Jul/2006 there is the -rotate option allow this. E.g's: "-rotate +90", "-rotate -90", "-rotate x", etc.
Q-106: Why is the view in my VNC viewer completely black? Or why is everything flashing around randomly?
See the next FAQ for a possible explanation.
Q-107: I use Linux Virtual Terminals (VT's) to implement 'Fast User Switching' between users' sessions (e.g. Betty is on Ctrl-Alt-F7, Bobby is on Ctrl-Alt-F8, and Sid is on Ctrl-Alt-F1: they use those keystrokes to switch between their sessions.) How come the view in a VNC viewer connecting to x11vnc is either completely black or otherwise all messed up unless the X session x11vnc is attached to is in the active VT?
This seems to have to do with how applications (the X server processes in this case) must "play nicely" if they are not on the active VT (sometimes called VC for virtual console.) That is, they should not read from the keyboard or mouse or manage the video display unless they have the active VT. Given that it appears the XGetImage() call must ultimately retrieve the framebuffer data from the video hardware itself, it would make sense x11vnc's polling wouldn't work unless the X session had active control of the VT.
There does not seem to be an easy way to work around this. Even xwd(1) doesn't work in this case (try it.) Something would need to be done at a lower level, say in the XFree86/Xorg X server. Also, using the Shadow Framebuffer (a copy of the video framebuffer is kept in main memory) does not appear to fix the problem.
If no one is sitting at the workstation and you just want to remotely switch the VT over to the one associated with your X session (so x11vnc can poll it correctly), one can use the chvt(1) command, e.g. "chvt 7" for VT #7.
Q-108: I am using x11vnc where my local machine has "popup/hidden taskbars" and the remote display where x11vnc runs also has "popup/hidden taskbars" and they interfere and fight with each other. What can I do?
When you move the mouse to the edge of the screen where the popups happen, the taskbars interfere with each other in strange ways. This sometimes happens where the local machine is GNOME or Mac OS X and the remote machine is GNOME. Is there a way to temporarily disable one or both of these magic desktop taskbars?
One x11vnc user suggests: it should be straightforward to right mouse click on the task bar panel, and uncheck "enable auto-hide" from the panel properties dialog box. This will make the panel always visible.
This is a new (Jul/2006) problem seen, say, on the version of KDE that is shipped with SuSE 10.1. It is not yet clear what is causing this... If you move the mouse through x11vnc the screensaver shuts off like it should but then a second or two after you stop moving the mouse the screensaver snaps back on.
This may be a bug in kdesktop_lock. For now the only workaround is to disable the screensaver. You can try using another one such as straight xscreensaver (see the instructions here for how to disable kdesktop_lock.) If you have more info on this or see it outside of KDE please let us know.
Update: It appears this is due to kdesktop_lock enabling the screen saver when the Monitor is in DPMS low-power state (e.g. standby, suspend, or off.) In Nov/2006 the x11vnc -nodpms option was added as a workaround. Normally it is a good thing that the monitor powers down (since x11vnc can still poll the framebuffer in this state), but if you experience the kdesktop_lock problem you can specify the "-nodpms" option to keep the Monitor out of low power state while VNC clients are connected. This is basically the same as typing "xset dpms force on" periodically. (if you don't want to do these things just disable the screensaver.) Feel free to file a bug against kdesktop_lock with KDE.
Q-110: I am running the compiz 3D window manager (or beryl, MythTv, Google Earth, or some other OpenGL app) and I do not get screen updates in x11vnc.
This appears to be because the 3D OpenGL/GLX hardware screen updates do not get reported via the XDAMAGE mechanism. So this is a bug in compiz/beryl or XDAMAGE/Xorg or the (possibly 3rd party) video card driver.
As a workaround apply the -noxdamage option. As of Feb/2007 x11vnc will try to autodetect the problem and disable XDAMAGE if is appears to be missing a lot of updates. But if you know you are using compiz you might as well always supply -noxdamage. Thanks to this user who reported the problem and discovered the workaround.
A developer for MiniMyth reports that the 'alphapulse' tag of the theme G.A.N.T. can also cause problems, and should be avoided when using VNC.
Please report a bug or complaint to Beryl/Compiz and/or Xorg about this: running x11vnc with -noxdamage disables a nice improvement in responsiveness (especially for typing) and also leads to unnecessary CPU and memory I/O load due to the extra polling.
Update: as of May/2010 NVIDIA may have fixed this problem in their proprietary drivers. See the NVIDIA Release Notes. (look for 'x11vnc'.)
Yes, since VMWare usually runs as an X application you can view it via x11vnc in the normal way.
Note that VMWare has several viewing modes:
- Normal X application window (with window manager frame)
- Quick-Switch mode (with no window manager frame)
- Fullscreen mode
The way VMWare does Fullscreen mode on Linux is to display the Guest desktop in a separate Virtual Terminal (e.g. VT 8) (see this FAQ on VT's for background.) Unfortunately, this Fullscreen VT is not an X server. So x11vnc cannot access it (however, see this discussion of -rawfb for a possible workaround.) x11vnc works fine with "Normal X application window" and "Quick-Switch mode" because these use X.
Update: It appears the in VMWare 5.x the Fullscreen mode is X, so x11vnc access does work.
One user reports he left his machine with VMWare in the Fullscreen mode, and even though his X session wasn't in the active VT, he could still connect x11vnc to the X session and pass the keystrokes Ctrl-Alt (typing "blind") to the VMWare X app. This induced VMWare to switch out of Fullscreen into Normal X mode and he could continue working in the Guest desktop remotely.
Aside: Sometimes it is convenient (for performance, etc.) to start VMWare in its own X session using startx(1). This can be used to have a minimal window manger (e.g. twm or even no window manager), to improve response. One can also cut the display depth (e.g. to 16bpp) in this 2nd X session to improve video performance. This 2nd X session emulates Fullscreen mode to some degree and can be viewed via x11vnc as long as the VMWare X session is in the active VT.
Also note that with a little bit of playing with "xwininfo -all -children" output one can extract the (non-toplevel) window-id of the of the Guest desktop only when VMWare is running as a normal X application. Then one can export just the guest desktop (i.e. without the VMWare menu buttons) by use of the -id windowid option. The caveats are the X session VMWare is in must be in the active VT and the window must be fully visible, so this mode is not terribly convenient, but could be useful in some circumstances (e.g. running VMWare on a very powerful server machine in a server room that happens to have a video card, (but need not have a monitor, Keyboard or mouse).)
Q-112: Can non-X devices (e.g. a raw framebuffer) be viewed (and even controlled) via VNC with x11vnc?
As of Apr/2005 there is support for this. Two options were added: "-rawfb string" (to indicate the raw frame buffer device, file, etc. and its parameters) and "-pipeinput command" (to provide an external program that will inject or otherwise process mouse and keystroke input.) Some useful -pipeinput schemes, VID, CONSOLE, and UINPUT, have since been built into x11vnc for convenience.
This non-X mode for x11vnc is somewhat experimental because it is so removed in scope from the intended usage of the tool. Incomplete attempt is made to make all of the other options consistent with non-X framebuffer polling. So all of the X-related options (e.g. -add_keysyms, -xkb) are just ignored or may cause an error if used. Be careful applying such an option via remote control.
The format for the -rawfb string is:
-rawfb <type>:<object>@<W>x<H>x<bpp>[-<BPL>][:<R>/<G>/<B>][+<offset>]
There are also some useful aliases (e.g. "console".) Some examples:
-rawfb shm:210337933@800x600x32:ff/ff00/ff0000
-rawfb map:/dev/fb0@1024x768x16
-rawfb map:/tmp/Xvfb_screen0@640x480x8+3232
-rawfb file:/tmp/my.pnm@250x200x24+37
-rawfb file:/dev/urandom@128x128x8
-rawfb snap:/dev/video0@320x240x24 -24to32
-rawfb console
-rawfb vt2
-rawfb video
-rawfb setup:mycmd.sh
So the type can be "shm" for shared memory objects, and "map" or "file" for file objects. "map" uses mmap(2) to map the file into memory and is preferred over "file" (that uses the slower lseek(2) access method.) Only use file if map isn't working. BTW, "mmap" is an alias for "map" and if you do not supply a type and the file exists, map is assumed (see the -help output and below for some exceptions to this.) The "snap:" setting applies the -snapfb option with "file:" type reading (this is useful for exporting webcams or TV tuner video; see the next FAQ for more info.)
Also, if the string is of the form "setup:cmd" then cmd is run and the first line of its output retrieved and used as the rawfb string. This allows initializing the device, determining WxHxB, etc.
The object will be the numerical shared memory id for the case of shm. The idea here is some other program has created this shared memory segment and periodically updates it with new framebuffer data. x11vnc polls the area for changes. See shmat(2) and ipcs(8) for more info. The ipcs command will list current shared memory segments on the system. Sometimes you can snoop on a program's framebuffer it did not expect you would be polling!
The object will be the path to the regular or character special file for the cases of map and file. The idea here is that in the case of a regular file some other program is writing/updating framebuffer image data to it. In the case of a character special (e.g. /dev/fb0) it is the kernel that is "updating" the framebuffer data.
In most cases x11vnc needs to be told the width, height, and number of bits per pixel (bpp) of the framebuffer. This is the @WxHxB field. For the case of the Linux framebuffer device, /dev/fb0, the fbset(8) may be of use (but may not always be accurate for what is currently viewable.) In general some guessing may be required, especially for the bpp. Update: in "-rawfb console" mode x11vnc will use the linuxfb API to try to guess (it is still not always accurate.) Also try "-rawfb vtN" (on x11vnc 0.9.7 and later) for the N-th Linux text console (aka virtual terminal.) If the number of Bytes Per Line is not WxHxB/8 (i.e. the framebuffer lines are padded) you can specify this information after WxHxB via "-BPL", e.g. @800x600x16-2048
Based on the bpp x11vnc will try to guess the red, green, and blue masks (these indicate which bits correspond to each color.) It if gets it wrong you can specify them manually via the optional ":R/G/B" field. E.g. ":0xff0000/0x00ff00/0x0000ff" (this is the default for 32bpp.)
Finally, the framebuffer may not begin at the beginning of the memory object, so use the optional "+offset" parameter to indicate where the framebuffer information starts. So as an example, the Xvfb virtual framebuffer has options -shmem and -fbdir for exporting its virtual screen to either shm or a mapped file. The format of these is XWD and so the initial header should be skipped. BTW, since XWD is not strictly RGB the view will only be approximate, but usable. Of course for the case of Xvfb x11vnc can poll it much better via the X API, but you get the idea.
By default in -rawfb mode x11vnc will actually close any X display it happened to open. This is basically to shake out bugs (e.g it will crash rather than mysteriously interacting with the X display.) If you want x11vnc to keep the X display open while polling the raw framebuffer prefix a "+" sign at the beginning of the string (e.g. +file:/dev/urandom@64x64x8) This could be convenient for keeping the remote control channel active (it uses X properties.) The "-connect /path/to/file" mechanism could also be used for remote control to avoid the X property channel. Rare usage, but if you also supply -noviewonly in this "+" mode then the mouse and keyboard input are still sent to the X display, presumably for doing something amusing with /dev/fb...
Interesting Devices:. Here are some aliases for interesting device files that can be polled via -rawfb:
-rawfb console /dev/fb0 Linux Console
-rawfb vt2 /dev/vcsa2 Linux Console (e.g. virtual terminal #2)
-rawfb video /dev/video0 Video4Linux Capture device
-rawfb rand /dev/urandom Random Bytes
-rawfb null /dev/zero Zero Bytes (black screen)
The Linux console, /dev/fb0, etc needs to have its driver enabled in the kernel. Some of the drivers are video card specific and accelerated. The console is either the Text consoles (usually tty1-tty6), or X graphical display (usually starting at tty7.) In addition to the text console other graphical ones may be viewed and interacted with as well, e.g. DirectFB or SVGAlib apps, VMWare non-X fullscreen, or Qt-embedded apps (PDAs/Handhelds.) By default the pipeinput mechanisms UINPUT and CONSOLE (keystrokes only) are automatically attempted in this mode under "-rawfb console".
The Video4Linux Capture device, /dev/video0, etc is either a Webcam or a TV capture device and needs to have its driver enabled in the kernel. See this FAQ for details. If specified via "-rawfb Video" then the pipeinput method "VID" is applied (it lets you change video parameters dynamically via keystrokes.)
The last two, /dev/urandom and /dev/zero are just for fun, but are also useful in testing.
All of the above -rawfb options are just for viewing the raw framebuffer (although some of the aliases do imply keystroke and mouse pipeinput methods.) That may be enough for certain applications of this feature (e.g. suppose a video camera mapped its framebuffer into memory and you just wanted to look at it via VNC.) To handle the pointer and keyboard input from the viewer users the "-pipeinput cmd" option was added to indicate a helper program to process the user input. The input is streamed to it and looks something like this:
Pointer 1 205 257 0 None
Pointer 1 198 253 0 None
Pointer 1 198 253 1 ButtonPress-1
Pointer 1 198 253 0 ButtonRelease-1
Pointer 1 198 252 0 None
Keysym 1 1 119 w KeyPress
Keysym 1 0 119 w KeyRelease
Keysym 1 1 65288 BackSpace KeyPress
Keysym 1 0 65288 BackSpace KeyRelease
Keysym 1 1 112 p KeyPress
Keysym 1 0 112 p KeyRelease
Run "-pipeinput tee:/bin/cat" to get a description of the format. Note that the -pipeinput option is independent of -rawfb mode and so may have some other interesting uses. The "tee:" prefix means x11vnc will both process the user input and pipe it to the command. The default is to just pipe it to the -pipeinput command.
Note the -pipeinput helper program could actually control the raw framebuffer. In the libvncserver CVS a simple example program x11vnc/misc/slide.pl is provided that demonstrates a simple jpeg "slideshow" application. Also the builtin "-pipeinput VID" mode does this for webcams and TV capture devices (/dev/video0.)
The -pipeinput program is run with these environment variables set: X11VNC_PID, X11VNC_PROG, X11VNC_CMDLINE, X11VNC_RAWFB_STR to aid its knowing what is up.
Another example provided in libvncserver CVS is a script to inject keystrokes into the Linux console (e.g. the virtual consoles: /dev/tty1, /dev/tty2, etc) in x11vnc/misc/vcinject.pl. It is based on the vncterm/LinuxVNC.c program also in the libvncserver CVS. So to view and interact with VT #2 (assuming it is the active VT) one can run something like:
x11vnc -rawfb map:/dev/fb0@1024x768x16 -pipeinput './vcinject.pl 2'
This assumes your Linux framebuffer device (/dev/fb0) is properly configured. See fbset(8) and other documentation. Try "file:/dev/fb0@WxHxB" as a last resort. Starting with x11vnc 0.8.1, the above VT injection is built in, as well as WxHxB determination. Just use something like:
x11vnc -rawfb console
this will try to guess the active virtual console (via /dev/tty0) and also the /dev/fb0 WxHxB and rgb masks automatically. Use, e.g., "-rawfb console3" to force the VT number. This input method can be used generally via "-pipeinput CONSOLE". Also starting with x11vnc 0.8.2 the "-pipeinput UINPUT" mode is tried first (it does both keyboard and mouse input) and then falls back to CONSOLE mode if it is not available. Here is the -help output for this mode:
If the rawfb string begins with "console" the framebuffer device
/dev/fb0 is opened (this requires the appropriate kernel modules to
be installed) and so is /dev/tty0. The latter is used to inject
keystrokes (not all are supported, but the basic ones are.) You
will need to be root to inject keystrokes. /dev/tty0 refers to the
active VT, to indicate one explicitly, use "console2", etc. using
the VT number.
If the Linux version seems to be 2.6 or later and the "uinput"
module appears to be present, then the uinput method will be used
instead of /dev/ttyN. uinput allows insertion of BOTH keystrokes
and mouse input and so it preferred when accessing graphical (e.g.
Qt-embedded) linux console apps. See -pipeinput UINPUT below for
more information on this mode (you may want to also use the
-nodragging and -cursor none options.) Use "console0", etc or
-pipeinput CONSOLE to force the /dev/ttyN method.
Note you can change VT remotely using the chvt(1) command.
Sometimes switching out and back corrects the framebuffer state.
To skip input injecting entirely use "consolex".
The string "/dev/fb0" (1, etc) can be used instead of "console".
This can be used to specify a different framebuffer device, e.g.
/dev/fb1. As a shortcut the "/dev/" can be dropped. If the name is
something nonstandard, use "console:/dev/foofb"
If you do not want x11vnc to guess the framebuffer's WxHxB and
masks automatically (sometimes the kernel gives inaccurate
information), specify them with a @WxHxB at the end of the string.
The above is just an example of what can be done. Note that if you really want to view and interact with the Linux Text console it is better to use the more accurate and faster LinuxVNC program. The advantage x11vnc -rawfb might have is that it can allow interaction with a non-text application, e.g. one based on SVGAlib or Qt-embedded Also, for example the VMWare Fullscreen mode is actually viewable under -rawfb and can be interacted with if uinput is enabled.
If the Linux uinput driver is available then full keystroke and mouse input into the Linux console can be performed. You may be able to enable uinput via commands like these:
modprobe uinput
mknod /dev/input/uinput c 10 223
The -rawfb and -pipeinput features are intended to help one creatively "get out of a jam" (say on a legacy or embedded device) where X is absent or doesn't work properly. Feedback and bug reports are welcome. For more control and less overhead use libvncserver in your own C program that passes the framebuffer to libvncserver.
Yes, you may need to be root to access the devices that make up the linux console.
To access the active Linux console via the computer's framebuffer try something like:
x11vnc -rawfb console
x11vnc -rawfb console2
These will try to access the framebuffer through /dev/fb (or /dev/fb0, etc.) and if it succeeds it will show any text or graphics that is currently displayed. Keystrokes will be injected via the device /dev/tty0 (to force an explicit virtual terminal append a number, e.g. "console2" to select /dev/tty2.)
If your Linux system does not have a framebuffer device (/dev/fb) you can get one by adding, e.g., vga=0x31B boot parameter. This enables the VGA framebuffer device at 1280x1024x24. 0x317 gives 1024x768x16, etc. You can also enable a Linux framebuffer device by modprobing a framebuffer driver specific to your video card.
Note that this "-rawfb console" mode shows the contents of the hardware framebuffer, and so will show whatever is on the screen. It has no concept of Virtual Terminals WRT what there is to view, it always shows the active virtual terminal.
Another mode is specific to the Linux text Virtual Terminals, it shows their text and colors (but no graphics) regardless of whether it is the active VT or not. It is available on x11vnc 0.9.7 and later. Enable this mode like this:
x11vnc -rawfb vt
x11vnc -rawfb vt2
The former will select the active one, the latter the 2nd VT. x11vnc implements this mode by opening the current console text file "/dev/vcsa2" instead of "/dev/fb". In this way it provides the basic functionality of the LibVNCServer LinuxVNC program.
The vt mode can be a useful way to try to get a machine's X server working remotely, e.g. you edit /etc/X11/xorg.conf and then type startx (or similar, e.g. gdm) in the virtual terminal. A 2nd x11vnc could be used to see if the X server is now working correctly.
Yes, this is possible to some degree with the -rawfb option. There is no X11 involved: snapshots from the video capture device are used for the screen image data. See the previous FAQ on -rawfb for background. For best results, use x11vnc version 0.8.1 or later.
Roughly, one would do something like this:
x11vnc -rawfb snap:/dev/video@320x240x32
This requires that the system allows simple read(2) access to the video device. This is true for video4Linux on Linux kernel 2.6 and later (it won't work for 2.4, you'll need a separate program to snapshot to a file that you point -rawfb to; ask me if it is not clear what to do.)
The "snap:" enforces -snapfb mode which appears to be necessary. The read pointer for video capture devices cannot be repositioned (which would be needed for scanline polling), but you can read a full frame of data from the device.
On Linux, if the Video4Linux API is present or the v4l-info(1) program (related to xawtv) exists in in PATH, then x11vnc can be instructed to try it to determine the -rawfb WxHxB parameters for you automatically. In this case one would just type:
x11vnc -rawfb video
or "-rawfb video1" for the 2nd video device, etc.
x11vnc has also been extended to use the Video4Linux API over v4l-info if it is available at build time. This enables setting parameters (e.g. size and brightness) via x11vnc. See the description below. Without Video4Linux you will need to initialize the settings of the video device using something like xawtv or spcaview (and then hope the settings persist until x11vnc reopens the device.)
Many video4linux drivers tend to set the framebuffer to be 24bpp (as opposed to 32bpp.) Since this can cause problems with VNC viewers, etc, the -24to32 option will be automatically imposed when in 24bpp.
Note that by its very nature, video capture involves rapid change in the framebuffer. This is especially true for cameras where slight wavering in brightness is always happening. This can lead to much network bandwidth consumption for the VNC traffic and also local CPU and I/O resource usage. You may want to experiment with "dialing down" the framerate via the -wait, -slow_fb, or -defer options. Decreasing the window size and bpp also helps.
Setting Camera/Tuner parameters via x11vnc:
There is also some support for setting parameters of the capture device. This is done via "-rawfb video:". This could be useful for unattended startup at boottime, etc. Here is the -help description:
A more sophisticated video device scheme allows initializing the
device's settings using:
-rawfb video:<settings>
The prefix could also be, as above, e.g. "video1:" to specify the
device file. The v4l API must be available for this to work.
Otherwise, you will need to try to initialize the device with an
external program, e.g. xawtv, spcaview, and hope they persist when
x11vnc re-opens the device.
<settings> is a comma separated list of key=value pairs. The
device's brightness, color, contrast, and hue can be set to
percentages, e.g. br=80,co=50,cn=44,hu=60.
The device filename can be set too if needed (if it does not start
with "video"), e.g. fn=/dev/qcam.
The width, height and bpp of the framebuffer can be set via, e.g.,
w=160,h=120,bpp=16.
Related to the bpp above, the pixel format can be set via the
fmt=XXX, where XXX can be one of: GREY, HI240, RGB555, RGB565,
RGB24, and RGB32 (with bpp 8, 8, 16, 16, 24, and 32 respectively.)
See http://www.linuxtv.org for more info (V4L api.)
For TV/rf tuner cards one can set the tuning mode via tun=XXX where
XXX can be one of PAL, NTSC, SECAM, or AUTO.
One can switch the input channel by the inp=XXX setting, where XXX
is the name of the input channel (Television, Composite1, S-Video,
etc.) Use the name that is in the information about the device that
is printed at startup.
For input channels with tuners (e.g. Television) one can change
which station is selected by the sta=XXX setting. XXX is the
station number. Currently only the ntsc-cable-us (US cable)
channels are built into x11vnc. See the -freqtab option below to
supply one from xawtv. If XXX is greater than 500, then it is
interpreted as a raw frequency in KHz.
Example:
-rawfb video:br=80,w=320,h=240,fmt=RGB32,tun=NTSC,sta=47
one might need to add inp=Television too for the input channel to
be TV if the card doesn't come up by default in that one.
Note that not all video capture devices will support all of the
above settings.
See the -pipeinput VID option below for a way to control the
settings through the VNC Viewer via keystrokes.
As above, if you specify a "@WxHxB..." after the <settings> string
they are used verbatim: the device is not queried for the current
values. Otherwise the device will be queried.
Also, if you supply the "-pipeinput VID" (or use "-rawfb Video") option you can control the settings to some degree via keystroke mappings, e.g. B to increase the brightness or Up arrow to change the TV station:
For "-pipeinput VID" and you are using the -rawfb for a video
capture device, then an internal list of keyboard mappings is used
to set parameters of the video. The mappings are:
"B" and "b" adjust the brightness up and down.
"H" and "h" adjust the hue.
"C" and "c" adjust the colour.
"N" and "n" adjust the contrast.
"S" and "s" adjust the size of the capture screen.
"I" and "i" cycle through input channels.
Up and Down arrows adjust the station (if a tuner)
F1, F2, ..., F6 will switch the video capture pixel
format to HI240, RGB565, RGB24, RGB32, RGB555, and
GREY respectively. See -rawfb video for details.
See also the -freqtab option to supply your own xawtv channel to frequency mappings for your country (only ntsc-cable-us is built into x11vnc.)
Q-115: Can I connect via VNC to a Qt-embedded/Qt-enhanced/Qtopia application running on my handheld, cell phone, or PC using the Linux console framebuffer (i.e. not X11)?
Yes, the basic method for this is the -rawfb scheme where the Linux console framebuffer (usually /dev/fb0) is polled and the uinput driver is used to inject keystrokes and mouse input. Often you will just have to type:
x11vnc -rawfb console
(you may need to enable the uinput driver on the system via "modprobe uinput; mknod /dev/input/uinput c 10 223") If this does not find the correct frame buffer properties figure them out or guess them and use something like:
x11vnc -rawfb /dev/fb0@640x480x16
Also, to force usage of the uinput injection method use "-pipeinput UINPUT". See the -pipeinput description for tunable parameters, etc.
One problem with the x11vnc uinput scheme is that it cannot guess the mouse motion "acceleration" used by the windowing application (e.g. QWS or X11.) For X11 and Qt-embedded the acceleration is usually 2 (i.e. a dx of 1 from the mouse yields a 2 pixel displacement of the mouse cursor.) The default x11vnc uses is 2, since that is often used. However for one Qt-embedded system we needed to do:
x11vnc -rawfb console -pipeinput UINPUT:accel=4.0
to get reasonable positioning of the mouse.
Even with the correct acceleration setting there is still some drift (probably because of the mouse threshold where the acceleration kicks in) and so x11vnc needs to reposition the cursor from 0,0 about 5 times a second. See the -pipeinput UINPUT option for tuning parameters that can be set (there are some experimental thresh=N tuning parameters as well)
Currently, one can expect mouse input to be a little flakey. All in all, the Linux framebuffer input mechanism for Qt-embedded framebuffer apps is not perfect, but it is usable.
If you need to create a smaller x11vnc binary for a handheld environment be sure to run strip(1) on it and also consider configuring with, e.g. "env CPPFLAGS='-DSMALL_FOOTPRINT=1' ./configure ..." to remove rarely used features and large texts (use 2 or 3 instead of 1 to remove more.) Currently (Jul/2006) this can lower the size of the x11vnc from 1.1MB to 0.6-0.7MB.
The x11vnc uinput method applies to nearly anything on the Linux framebuffer console, not just Qt-embedded/Qtopia. DirectFB, SDL using fbcon driver, SVGAlib applications can also be viewed and interacted with. Even a Linux X session can be viewed and interacted with without using X11 (and x11vnc does not have to terminate when the X server restarts!) The Linux Text consoles (F1-F6) also work.
Note that Qt-embedded supplies its own VNC graphics driver, but it cannot do both the Linux console framebuffer and VNC at the same time, which is often what is desired from VNC.
Update: We are finding some setups like Qtopia on the IPAQ do not allow mouse input via uinput. Please help us debug this problem by trying x11vnc on your device and letting us know what does and does not work. See the next FAQ for a possible workaround for touchscreens.
Q-116: How do I inject touch screen input into an Qt-embedded/Qt-enhanced/Qtopia cell phone such as openmoko/qtmoko Neo Freerunner?
The qtmoko project does not use X11 for the graphical display. Unfortunately the Linux uinput method described in the previous FAQ does not work because Qt is using TSLIB (touch screen library) to process the input and it only reads from one device (often /dev/input/event1) and not from the new UINPUT device that x11vnc creates (under -pipeinput UINPUT)
So something else needs to be done. It was discovered that by simply writing the touchscreen events directly to /dev/input/event1 then input can be injected into the system. There is no x11vnc builtin mode for this yet (until we understand it better), but there is a working script provided in x11vnc/misc/qt_tslib_inject.pl. So one could use it this way for example:
x11vnc ... -rawfb console -pipeinput path/to/qt_tslib_inject.pl -env INJECT_OPTIONS=clickonly,cal=/etc/pointercal
Read the script for how to enable other options and what the above options mean (e.g. /etc/pointercal contains TSLIB's calibration parameters and are necessary to achieve accurate pointing.)
The x11vnc/misc/qt_tslib_inject.pl script can potentially be modified to handle other devices where the uinput method fails. It could also be modified to create 'hot keys', etc.
Please let us know how things go if you try this out; there is much to learn about synthetic input injection in handhelds and cell phones. As we learn more we can develop a builtin x11vnc mode for this sort of injection.
Update Dec/2010: There is experimental built-in UINPUT support in the x11vnc development tarball for qtmoko with touchpad managed by tslib. See -pipeinput UINPUT for more info. Here is an example:
x11vnc -rawfb console -pipeinput UINPUT:touch,tslib_cal=/etc/pointercal,direct_abs=/dev/input/event1,nouinput,dragskip=3
Q-117: Now that non-X11 devices can be exported via VNC using x11vnc, can I build it with no dependencies on X11 header files and libraries?
Yes, as of Jul/2006 x11vnc enables building for -rawfb only support. Just do something like when building:
./configure --without-x (plus any other flags)
make
You can then test via "ldd x11vnc" that the binary does not depend on libX11.so, etc. See the previous FAQ's for non-X11 framebuffer usage. If you use this for an interesting non-X11 application please let us know what you did.
You will need a cross-compiling toolchain. Perhaps your distro provides these or you can find a HOWTO for your distro. We found a nice one at qtmoko.org for building armel binaries on Debian Linux i386 machines. It includes most of the libraries that x11vnc needs. We use that example here.
We ran this script to set PATH, configure, and build:
#!/bin/sh
# toolchain from: qtmoko-debian-toolchain-armv4t-eabi.tar.gz
export PATH=/opt/toolchains/arm920t-eabi/bin:$PATH
env CC=arm-linux-gcc ./configure --host=arm-linux --without-avahi
make
arm-linux-strip ./x11vnc/x11vnc
ls -l ./x11vnc/x11vnc
Note we had to include --without-avahi due to lack of libavahi-client.so.3 supplied by the toolchain we used. One would need to add it if it was desired on the target machine. We also stripped the binary to make it smaller.
For an embedded system one may also want to add --without-x if the embedded system does not use X11 and the -rawfb mechanism must be used.
Yes, since Nov/2006 in the development tree (x11vnc-0.8.4 tarball) there is support for native Mac OS X Aqua/Quartz displays using the -rawfb mechanism described above. The mouse and keyboard input is achieved via Mac OS X API's.
So you can use x11vnc as an alternative to OSXvnc (aka Vine Server), or Apple Remote Desktop (ARD). Perhaps there is some x11vnc feature you'd like to use on Mac OS X, etc. For a number of activities (e.g. window drags) it seems to be faster than OSXvnc.
Notes:
X11: x11vnc will also work (as it has for years) with a X11 server (XDarwin) running on Mac OS X (people often install this software to display remote X11 apps on their Mac OS X system, or use some old favorites locally such as xterm.) However in this case x11vnc will only work reasonably in single window -id windowid mode (and the window may need to have mouse focus.)
If you do not have the DISPLAY env. variable set, x11vnc will assume native Aqua/Quartz on Mac OS X, however if DISPLAY is set it will assume an X11 connection. Use "-rawfb console" to force the native display (or unset DISPLAY.)
Update: Leopard sets DISPLAY by default in all sessions. Since it starts with the string "/tmp/" x11vnc will use that to know if it should ignore it. Use "-display :0.0" to force it.
Building: If you don't have the X11 build and runtime packages installed you will need to build it like this: (cd to the e.g. x11vnc-0.9, source directory)
./configure --without-x
make
Win2VNC/x2vnc: One handy use is to use the -nofb mode to redirect mouse and keyboard input to a nearby Mac (i.e. one to the side of your desk) via x2vnc or Win2VNC. See this FAQ for more info.
Options: Here are the Mac OS X specific x11vnc options:
-macnodim For the native Mac OS X server, disable dimming.
-macnosleep For the native Mac OS X server, disable display sleep.
-macnosaver For the native Mac OS X server, disable screensaver.
-macnowait For the native Mac OS X server, do not wait for the
user to switch back to his display.
-macwheel n For the native Mac OS X server, set the mouse wheel
speed to n (default 5.)
-macnoswap For the native Mac OS X server, do not swap mouse
buttons 2 and 3.
-macnoresize For the native Mac OS X server, do not resize or reset
the framebuffer even if it is detected that the screen
resolution or depth has changed.
-maciconanim n For the native Mac OS X server, set n to the number
of milliseconds that the window iconify/deiconify
animation takes. In -ncache mode this value will be
used to skip the animation if possible. (default 400)
-macmenu For the native Mac OS X server, in -ncache client-side
caching mode, try to cache pull down menus (not perfect
because they have animated fades, etc.)
PasteBoard/Clipboard: There is a bug that the Clipboard (called PasteBoard on Mac it appears) exchange will not take place unless x11vnc was started from inside the Aqua display (e.g. started inside a Terminal app window.) Otherwise it cannot connect to the PasteBoard server. So Clipboard exchange won't work for our standard "ssh in" startup scheme.
Hopefully this deficiency can be removed, but until then for Clipboard exchange to work you will need to start x11vnc inside the desktop session (i.e. either start it running before you leave, or start up a 2nd x11vnc inside from a 1st one started outside, or use the apple script below)
Here also is a osascript trick that seems to work (it opens the Terminal app and instructs it to start x11vnc):
#!/bin/sh
#
# start_x11vnc: start x11vnc in a Terminal window
# (this will allow Clipboard/Pasteboard exchange to work)
tmp=/tmp/start_x11vnc.$$
cat > $tmp <<END
tell application "Terminal"
activate
do script with command "$HOME/x11vnc -rfbauth .vnc/passwd -ssl SAVE"
end tell
END
osascript $tmp
rm -f $tmp
where you should customize the x11vnc command line to your needs and the full path to the binary. Save it in a file e.g. "start_x11vnc" and then after you SSH in just type "./start_x11vnc" (or have ssh run the command for you.) Then once you are connected via VNC, iconify the Terminal windows (you can't delete them since that will kill x11vnc.)
Update Aug/2010: A user reports the following useful information:
This is not a problem on Mac OS X 10.6.x (Snow Leopard) when connecting
via ssh to start x11vnc. And, on Mac OS X 10.5.x (Leopard), the problem
can be permanently eliminated by doing this:
sudo /usr/libexec/PlistBuddy -c 'delete :LimitLoadToSessionType' \
-c 'add :LimitLoadToSessionType string Background' \
/System/Library/LaunchAgents/com.apple.pboard.plist
# ignore any 'Delete: Entry, ":LimitLoadToSessionType", Does Not Exist' message
and then restarting (yes, you must restart not just log off). But
ONLY do that for Mac OS X 10.5.x and NOT for 10.6.x (which doesn't
need it anyway).
We recently got access to a MacOSX 10.6.4 (Snow Leopard) macbook and have confirmed that the above is correct.
Q-120: Can x11vnc be used as a VNC reflector/repeater to improve performance for the case of a large number of simultaneous VNC viewers (e.g. classroom broadcasting or a large demo)?
Yes, as of Feb/2007 there is the "-reflect host:N" option to connect to the VNC server "host:N" (either another x11vnc or any other VNC server) and re-export it. VNC viewers then connect to the x11vnc(s) running -reflect.
The -reflect option is the same as: "-rawfb vnc:host:N". See the -rawfb description under "VNC HOST" for more details.
You can replace "host:N" with "listen" or "listen:port" for reverse connections.
One can set up a number of such reflectors/repeaters to spread the resource usage around, e.g.:
C -------<-------|
C -------<-------|
C -------<-------|---- R -----|
C -------<-------| |
C -------<-------| |
|
C -------<-------| |
C -------<-------| |
C -------<-------|---- R -----|
C -------<-------| |
C -------<-------| |
|====== S
C -------<-------| |
C -------<-------| |
C -------<-------|---- R -----|
C -------<-------| |
C -------<-------| |
|
C -------<-------| |
C -------<-------| |
C -------<-------|---- R -----|
C -------<-------|
C -------<-------|
Where "S" is the original VNC Server, "C" denote VNC viewer clients, and "R" denotes an x11vnc running -reflect to "S".
Ideally, a client "C" will be fairly close network-wise to its "R". It is fine to run the "R" on the same machine as one of its "C's". A nice setup for a large, (e.g. 64-128) viewer classroom broadcast case would be to run R's on areas isolated by network switches, e.g. one R per switch.
In an extreme case (e.g. 1000 viewers) one might actually need a 2nd layer of R's in the tree. If you try something like that let us know!
There are many resource savings in doing something like the above. The first obvious one is network bandwidth savings. Another is less CPU load on "S" since it handles many fewer simultaneous connections. Also, if there are a few clients C on very slow links, their presence does not slow down every other client, just the clients on their "R". One way a slow client affects things is if there are some large framebuffer writes (e.g. jpeg image region) then the repeater may block waiting for that large write to finish before going onto the next client (however, if the write is small enough, the kernel will buffer it and the server can go on to service the next client.)
The x11vnc -reflect implementation uses the libvncclient library in the LibVNCServer project to handle the connection to "S". It is not currently very efficient since it simply does its normal framebuffer polling scheme on the libvncclient framebuffer (which it then re-exports via VNC to its clients C.) However, CopyRect and CursorShape encodings are preserved in the reflection and that helps. Dragging windows with the mouse can be a problem (especially if S is not doing wireframing somehow, consider -nodragging if the problem is severe) For a really fast reflector/repeater it would have to be implemented from scratch with performance in mind. See these other projects:
- http://sourceforge.net/projects/vnc-reflector/,
- http://www.tightvnc.com/projector/ (closed source?),
Automation via Reverse Connections: Instead of having the R's connect directly to S and then the C's connect directly to the R they should use, some convenience can be achieved by using reverse connections (the x11vnc ""-connect host1,host2,..." option.) Suppose all the clients "C" are started up in Listen mode:
client1> vncviewer -listen
client2> vncviewer -listen
client3> vncviewer -listen
...
client64> vncviewer -listen
(e.g. client1> is the cmdline prompt on machine client1 ... etc) and all the repeaters R are started like this:
repeater1> x11vnc -reflect listen -connect client1,client2,...client8
repeater2> x11vnc -reflect listen -connect client9,client10,...client16
...
repeater8> x11vnc -reflect listen -connect client57,client58,...client64
and finally the main server is started to kick the whole thing into motion:
vncserver> x11vnc -display :0 -connect repeater1,repeater2,...repeater8
(or instruct a non-x11vnc VNC server to reverse connect to the repeaters.) For a classroom broadcasting setup one might have the first two sets of commands start automatically at bootup or when someone logs in, and then start everything up with the S server. One may even be able to script the forward connection bootstrap case, let us know what you did. A really nice thing would be some sort of auto-discovery of your repeater, etc...
Q-121: Can x11vnc be used during a Linux, Solaris, etc. system installation so the installation can be done remotely?
This can be done, but it doesn't always work because it depends on how the OS does its install. We have to "sneak in" somehow. Note that some OS's have a remote install (ssh etc.) built in and so you might want to use that instead.
Usually the OS install will have to be a network-install in order to have networking up during the install. Otherwise, you may have a (slim) chance to configure the networking manually (ifconfig(8) and route(8).)
To avoid library dependencies problems in the typical minimal (e.g. busybox) installation OS it is a good idea to build a statically linked x11vnc binary. A way that often works is to do a normal build and then paste the final x11vnc link line into a shell script. Then change the "gcc" to "gcc -static" and run the shell script. You may need to disable features (e.g. "--without-xfixes") if there is not a static library for the feature available. You may also need to add extra link options (e.g. "-lXrender") to complete library dependencies manually.
Let's call the binary x11vnc.static. Place it on a webserver somewhere. It may be possible to retrieve it via scp(1) too.
During the install you need to get a shell to retreive x11vnc.static and run it.
If the Solaris install is an older X-based one, there will be a menu for you to get a terminal window. From that window you might be able to retrieve x11vnc.static via wget, scp, or ftp. Remember to do "chmod 755 ./x11vnc.static" and then find the -auth file as in this FAQ.
If it is a Linux install that uses an X server (e.g. SuSE and probably Fedora), then you can often get a shell by pressing Ctrl-Alt-F2 or similar. Then get the x11vnc binary via something like this:
cd /tmp
wget http://192.168.0.22/x11vnc.static
chmod 755 ./x11vnc.static
Find the name of the auth file as in this FAQ. (maybe run "ps wwaux | grep auth".) Then run it like this: ./x11vnc.static -forever -nopw -display :0 -auth /tmp/wherever/the/authfile
then press Alt-F7 to go back to the X install. You should now be able to connect via a vnc viewer and continue the install. Watch out for the display being :1, etc.
If there is a firewall blocking incoming connections during the install, use the "-connect hostname" option option for a reverse connection to the hostname running the VNC viewer in listen mode.
Debian based installs are either console-text or console-framebuffer based. These are install (or expert) and installgui (or expertgui) boot lines, respectively. For the console-text based installs you probably need to add a boot cmd line option like vga=0x314 (which is 800x600x16) to get the console-text to use the linux framebuffer device properly.
For a Debian console-text based install after the network is configured press Ctrl-Alt-F2 to get a shell. Retrieve the binary via wget as above and chmod 755 it. Then run it something like this: sleep 10; ./x11vnc.static -forever -nopw -rawfb console
then before the sleep is over press Alt-F1 to get back to the install virtual console. You should be able to connect via a VNC viewer and continue with the install.
For a recent (2009) Debian install we booted with "expert vga=0x301" and "expert vga=0x311" to get console text based installs at 640x480x8 and 640x480x16, respectively (replace "expert" with "install" if you like.) Otherwise it was giving a 16 color 640x480x4 (4 bit per pixel) display which x11vnc could not handle.
For Debian console-framebuffer GUI based installs (installgui or expertgui) we have not be able to enter keystrokes or mouse motions. This may be resolved if the install had the Linux kernel module uinput, but it doesn't; one can wget uinput.ko and then run insmod on it, but the module must match the installation kernel. So, failing that, you can only do the GUI view-only, which can be handy to watch a long network install from your desk instead of in front of the machine being installed. For these, after the network is configured press Ctrl-Alt-F2 to get a shell. Retrieve the binary via wget as above and chmod 755 it. Then run it something like this: sleep 10; ./x11vnc.static -forever -nopw -rawfb console
then before the sleep is over press Alt-F5 to get back to the GUI install console. You should be able to connect via a VNC viewer and watch the install. [Misc: Clipboard, File Transfer/Sharing, Printing, Sound, Beeps, Thanks, etc.]
As of Jan/2004 x11vnc supports the "CutText" part of the RFB (aka VNC) protocol. When text is selected/copied in the X session that x11vnc is polling it will be sent to connected VNC viewers. And when CutText is received from a VNC viewer then x11vnc will set the X11 selections PRIMARY, CLIPBOARD, and CUTBUFFER0 to it. x11vnc is able to hold the PRIMARY and CLIPBOARD selections (Xvnc does not seem to do this.)
The X11 selections can be confusing, especially to those coming from Windows or MacOSX where there is just a single 'Clipboard'. The X11 CLIPBOARD selection is a lot like that of Windows and MacOSX, e.g. highlighted text is sent to the clipboard when the user activates "Edit -> Copy" or presses "Control+C" (and pasting it via "Edit -> Paste" or "Control+V".) The X11 PRIMARY selection has been described as 'for power users' or 'an Easter Egg'. As soon as text is highlighted it is set to the PRIMARY selection and so it is immediately ready for pasting, usually via the Middle Mouse Button or "Shift+Insert". See this jwz link for more information.
x11vnc's default behavior is to watch both CLIPBOARD and PRIMARY and whenever one of them changes, it sends the new text to connected viewers. Note that since the RFB protocol only has a single "CutText" then both selections are "merged" to some degree (and this can lead to confusing results.) One user was confused why x11vnc was "forgetting" his CLIPBOARD selection and the reason was he also changed PRIMARY some time after he copied text to the clipboard. Usually an app will set PRIMARY as soon as any text is highlighted so it easy to see how CLIPBOARD was forgotten. Use the -noprimary described below as a workaround. Similarly, by default when x11vnc receives CutText it sets both CLIPBOARD and PRIMARY to it (this is probably less confusing, but could possibly lead to some failure modes as well.)
You may not like these defaults. Here are ways to change the behavior:
- If you don't want the Clipboard/Selection exchanged at all use the -nosel option.
- If you want changes in PRIMARY to be ignored use the -noprimary option.
- If you want changes in CLIPBOARD to be ignored use the -noclipboard option.
- If you don't want x11vnc to set PRIMARY to the "CutText" received from viewers use the -nosetprimary option.
- If you don't want x11vnc to set CLIPBOARD to the "CutText" received from viewers use the -nosetclipboard option.
You can also fine-tune it a bit with the -seldir dir option and also -input.
You may need to watch out for desktop utilities such as KDE's "Klipper" that do odd things with the selection, clipboard, and cutbuffers.
Q-123: Can I use x11vnc to record a Shock Wave Flash (or other format) video of my desktop, e.g. to record a tutorial or demo?
Yes, it is possible with a number of tools that record VNC and transform it to swf format or others. One such popular tool is pyvnc2swf. There are a number of tutorials (broken link?) on how to do this. Another option is to use the vnc2mpg that comes in the LibVNCServer package. An important thing to remember when doing this is that tuning parameters should be applied to x11vnc to speed up its polling for this sort of application, e.g. "-wait 10 -defer 10".
As of Oct/2005 and May/2006 x11vnc enables, respectively, the TightVNC and UltraVNC file transfer implementations that were added to libvncserver. This currently works with TightVNC and UltraVNC viewers (and Windows viewers only support filetransfer it appears... but they do work to some degree under Wine on Linux.)
The SSVNC Unix VNC viewer supports UltraVNC file transfer by use of a Java helper program.
TightVNC file transfer is off by default, if you want to enable it use the -tightfilexfer option.
UltraVNC file transfer is off by default, to enable it use something like "-rfbversion 3.6 -permitfiletransfer" options (UltraVNC incorrectly uses the RFB protocol version to determine if its features are available, so x11vnc has to pretend to be version 3.6.) As of Sep/2006 "-ultrafilexfer" is an alias for these two options. Note that running as RFB version 3.6 may confuse other VNC Viewers.
Sadly you cannot do both -tightfilexfer and -ultrafilexfer at the same time because the latter requires setting the version to 3.6 and tightvnc will not do filetransfer when it sees that version number.
Also, because of the way the LibVNCServer TightVNC file transfer is implemented, you cannot do Tightvnc file transfer in -unixpw mode. However, UltraVNC file transfer does work in -unixpw (but if a client tries it do a filetransfer during the login process it will be disconnected.)
IMPORTANT: please understand if -ultrafilexfer or -tightfilexfer is specified and you run x11vnc as root for, say, inetd or display manager (gdm, kdm, ...) access and you do not have it switch users via the -users option, then VNC Viewers that connect are able to do filetransfer reads and writes as root.
The UltraVNC and TightVNC settings can be toggled on and off inside the gui or by -R remote control. However for TightVNC the changed setting only applies for NEW clients, current clients retain their TightVNC file transfer ability. For UltraVNC it works better, however if an UltraVNC client has initiated a file transfer dialog it will remain in effect until the dialog is closed. If you want to switch between UltraVNC and TightVNC file transfer in the gui or by remote control you will probably be foiled by the "-rfbversion 3.6" issue.
Some of them are supported. To get UltraVNC Viewers to attempt to use these extensions you will need to supply this option to x11vnc: -rfbversion 3.6
Or use -ultrafilexfer which is an alias for the above option and "-permitfiletransfer". UltraVNC evidently treats any other RFB version number as non-UltraVNC.
Here are a list of the UltraVNC extensions supported by x11vnc: * ServerInput: "Toggle Remote Input and Remote Blank Monitor" * FileTransfer: "Open File Transfer..." * SingleWindow: "Select Single Window..." * TextChat: "Open Chat..." * 1/n Server Scaling
The SSVNC Unix VNC viewer supports these UltraVNC extensions.
To disable SingleWindow and ServerInput use -noultraext (the others are managed by LibVNCServer.) See this option too: -noserverdpms.
Also, the UltraVNC repeater proxy is supported for use with reverse connections: "-connect repeater://host:port+ID:NNNN". Use it for both plaintext and SSL connections. This mode can send any string before switching to the VNC protocol, and so could be used with other proxy/gateway tools. Also, a perl repeater implemention is here: ultravnc_repeater.pl
Q-126: Can x11vnc emulate UltraVNC's Single Click helpdesk mode for Unix? I.e. something very simple for a naive user to initiate a reverse vnc connection from their Unix desktop to a helpdesk operator's VNC Viewer.
Yes, UltraVNC's Single Click (SC) mode can be emulated fairly well on Unix.
We use the term "helpdesk" below, but it could be any sort of remote assistance you want to set up, e.g. something for Unix-using friends or family to use. This includes Mac OS X.
Assume you create a helpdesk directory "hd" on your website: http://www.mysite.com/hd (any website that you can upload files to should work, although remember the user will be running the programs you place there.)
In that "hd" subdirectory copy an x11vnc binary to be run on the Unix user's machine (e.g. Linux, etc) and also create a file named "vnc" containing the following:
#!/bin/sh
webhost="http://www.mysite.com/hd" # Your helpdesk dir URL.
vnchost="ip.someplace.net" # Your host running 'vncviewer -listen'
# It could also be your IP number. If it is
# a router/firewall, you will need to
# configure it to redirect port 5500 to your
# workstation running 'vncviewer -listen'
dir=/tmp/vnc_helpdesk.$$ # Make a temporary working dir.
mkdir $dir || exit 1
cd $dir || exit 1
trap "cd /tmp; rm -rf $dir" 0 2 15 # Cleans up on exit.
wget $webhost/x11vnc # Fetch x11vnc binary. If multi-
chmod 755 ./x11vnc # platform, use $webhost/`uname`/x11vnc
# or similar.
./x11vnc -connect_or_exit $vnchost -rfbport 0 -nopw
with the hostnames / IP addresses customized to your case.
On the helpdesk VNC viewer machine (ip.someplace.net in this example) you have the helpdesk operator running VNC viewer in listen mode: vncviewer -listen
or if on Windows, etc. somehow have the VNC viewer be in "listen" mode.
Then, when the naive user needs assistance you instruct him to open up a terminal window on his Unix desktop and paste the following into the shell:
wget -qO - http://www.mysite.com/hd/vnc | sh -
and then press Enter. You could have this instruction on a web page or in an email you send him, etc. This requires that the wget is installed on the user's Unix machine (he might only have curl or lynx, see below for more info.)
So I guess this is about 3-4 clicks (start a terminal and paste) and pressing "Enter" instead of "single click"...
See this page for some variations on this method, e.g. how to add a password, SSL Certificates, etc.
If you don't have a website (there are many free ones) or don't want to use one you will have to email him all of the ingredients (x11vnc binary and a launcher script) and tell him how to run it. This could be easy or challenging depending on the skill of the naive unix user...
A bit of obscurity security could be put in with a -passwd, -rfbauth options, etc. (note that x11vnc will require a password even for reverse connections.) More info here.
Firewalls: If the helpdesk (you) with the vncviewer is behind a NAT/Firewall/Router the router will have to be configured to redirect a port (i.e. 5500 or maybe different one if you like) to the vncviewer machine. If the vncviewer machine also has its own host-level firewall, you will have to open up the port there as well.
NAT-2-NAT: There is currently no way to go "NAT-2-NAT", i.e. both User and Helpdesk workstations behind NAT'ing Firewall/Routers without configuring a router to do a port redirection (i.e. on your side, the HelpDesk.) To avoid modifying either firewall/router, one would need some public (IP address reachable on the internet) redirection/proxy service. Perhaps such a thing exists. http://sc.uvnc.com provides this service for their UltraVNC Single Click users.
Update: It may be possible to do "NAT-2-NAT" with a UDP tunnel such as http://samy.pl/pwnat/. All that is required is that both NAT firewalls allow in UDP packets from an IP address to which a UDP packet has recently been sent to. If you try it out let us know how it went.
Very Naive Users:
If it is beyond the user how to open a terminal window and paste in a command (you have my condolences...) you would have to somehow setup his Web browser to download the "vnc" file (or a script containing the above wget line) and prompt the user if he wants to run it. This may be tricky to set up (which is probably a good thing to not have the web browser readily run arbitrary programs downloaded from the internet...)
One command-line free way, tested with KDE, is to name the file vnc.sh and then instruct the user to right-click on the link and do "Save Link As" to his Desktop. It will appear as an icon, probably one that looks like a terminal or a command line prompt. He next should right-click on the icon and select "Properties" and go to the "Permissions" tab. Then in that dialog select the checkbox "Is executable". He should then be able to click on the icon to launch it. Another option is to right-click on the icon and select "Open With -> Other ..." and for the name of the application type in "/bin/sh". Unfortunately in both cases the command output is lost and so errors cannot be debugged as easily. A similar thing appears to work in GNOME if under "Properties -> Permissions" they click on "Execute" checkbox for "Owner". Then when they click on the icon, they will get a dialog where they can select "Run in Terminal". In general for such cases, if it is feasible, it might be easier to ssh to his machine and set things up yourself...
SSL Encrypted Helpdesk Connections:
As of Apr/2007 x11vnc supports reverse connections in SSL and so we can do this. On the Helpdesk side (Viewer) you will need STUNNEL or better use the Enhanced TightVNC Viewer (SSVNC) package we provide that automates all of the SSL for you.
To do this create a file named "vncs" in the website "hd" directory containing the following:
#!/bin/sh
webhost="http://www.mysite.com/hd" # Your helpdesk dir URL.
vnchost="ip.someplace.net" # Your host running 'vncviewer -listen'
# It could also be your IP number. If it is
# a router/firewall, you will need to
# configure it to redirect port 5500 to your
# workstation running 'vncviewer -listen'
dir=/tmp/vnc_helpdesk.$$ # Make a temporary working dir.
mkdir $dir || exit 1
cd $dir || exit 1
trap "cd /tmp; rm -rf $dir" 0 2 15 # Cleans up on exit.
wget $webhost/x11vnc # Fetch x11vnc binary. If multi-
chmod 755 ./x11vnc # platform, use $webhost/`uname`/x11vnc
# or similar.
./x11vnc -connect_or_exit $vnchost -rfbport 0 -nopw -ssl # Note -ssl option.
with the hostnames or IP addresses customized to your case.
The only change from the "vnc" above is the addition of the -ssl option to x11vnc. This will create a temporary SSL cert: openssl(1) will need to be installed on the user's end. A fixed SSL cert file could be used to avoid this (and provide some authentication; more info here.)
The naive user will be doing this:
wget -qO - http://www.mysite.com/hd/vncs | sh -
(or perhaps even use https:// if available.)
But before that, the helpdesk operator needs to have "vncviewer -listen" running as before, however he needs an SSL tunnel at his end. The easiest way to do this is use Enhanced TightVNC Viewer (SSVNC). Start it, and select Options -> 'Reverse VNC Connection (-listen)'. Then UN-select 'Verify All Certs' (this can be enabled later if you want; you'll need the x11vnc SSL certificate), and click 'Listen'.
If you don't want to use SSVNC for the viewer, but rather set up STUNNEL manually instead, make a file "stunnel.cfg" containing:
foreground = yes
pid =
[vnc]
accept = 5500
connect = localhost:5501
and run:
stunnel ./stunnel.cfg
and then start the "vncviewer -listen 1" (i.e. 1 to correspond to the 5501 port.) Note that this assumes the stunnel install created a Server SSL cert+key, usually /etc/stunnel/stunnel.pem (not all distros will do this.) Also, that file is by default only readable by root, so stunnel needs to be run as root. If your system does not have a key installed or you do not want to run stunnel as root (or change the permissions on the file), you can use x11vnc to create one for you for example:
x11vnc -sslGenCert server self:mystunnel
answer the prompts with whatever you want; you can take the default for all of them if you like. The openssl(1) package must be installed. See this link and this one too for more info on SSL certs. This creates $HOME/.vnc/certs/server-self:mystunnel.pem, then you would change the "stunnel.cfg" to look something like:
foreground = yes
pid =
cert = /home/myusername/.vnc/certs/server-self:mystunnel.pem
[vnc]
accept = 5500
connect = localhost:5501
In any event, with stunnel having been setup, the naive user is instructed to paste in and run: wget -qO - http://www.mysite.com/hd/vncs | sh -
to pick up the vncs script this time.
Of course if a man-in-the-middle can alter what the user downloads then all bets are off!.
More SSL variations and info about certificates can be found here.
OpenSSL libssl.so.0.9.7 problems:
If you build your own stunnel or x11vnc for deployment, you may want to statically link libssl.a and libcrypto.a into it because Linux distros are currently a bit of a mess regarding which version of libssl is installed.
You will find the details here.
Q-127: Can I (temporarily) mount my local (viewer-side) Windows/Samba File share on the machine where x11vnc is running?
You will have to use an external network redirection for this. Filesystem mounting is not part of the VNC protocol.
We show a simple Samba example here.
First you will need a tunnel to redirect the SMB requests from the remote machine to the one you sitting at. We use an ssh tunnel: sitting-here> ssh -C -R 1139:localhost:139 far-away.east
Or one could combine this with the VNC tunnel at the same time, e.g.:
sitting-here> ssh -C -R 1139:localhost:139 -t -L 5900:localhost:5900 far-away.east 'x11vnc -localhost -display :0'
Port 139 is the Windows Service port. For Windows systems instead of Samba, you may need to use the actual IP address of the Window machine instead of "localhost" in the -R option (since the Windows service does not listen on localhost by default.)
Note that we use 1139 instead of 139 on the remote side because 139 would require root permission to listen on (and you may have a samba server running on it already.)
The ssh -C is to enable compression, which might speed up the data transfers.
Depending on the remote system side configuration, it may or may not be possible to mount the SMB share as a non-root user. Try it first as a non-root user and if that fails you will have to become root.
We will assume the user name is "fred" and we will try to mount the viewer-side Windows SMB share "//haystack/pub" in /home/fred/smb-haystack-pub.
far-away> mkdir -p /home/fred/smb-haystack-pub
far-away> smbmount //haystack/pub /home/fred/smb-haystack-pub -o username=fred,ip=127.0.0.1,port=1139
(The 2nd command may need to be run as root.) Then run "df" or "ls -l /home/fred/smb-haystack-pub" to see if it is mounted properly. Consult the smbmount(8) and related documentation (it may require some fiddling to get write permissions correct, etc.) To unmount:
far-away> smbumount /home/fred/smb-haystack-pub
At some point we hope to fold some automation for SMB ssh redir setup into the Enhanced TightVNC Viewer (SSVNC) package we provide (as of Sep 2006 it is there for testing.)
Q-128: Can I redirect CUPS print jobs from the remote desktop where x11vnc is running to a printer on my local (viewer-side) machine?
You will have to use an external network redirection for this. Printing is not part of the VNC protocol.
We show a simple Unix to Unix CUPS example here. Non-CUPS port redirections (e.g. LPD) should also be possible, but may be a bit more tricky. If you are viewing on Windows SMB and don't have a local cups server it may be trickier still (see below.)
First you will need a tunnel to redirect the print requests from the remote machine to the one you sitting at. We use an ssh tunnel: sitting-here> ssh -C -R 6631:localhost:631 far-away.east
Or one could combine this with the VNC tunnel at the same time, e.g.:
sitting-here> ssh -C -R 6631:localhost:631 -t -L 5900:localhost:5900 far-away.east 'x11vnc -localhost -display :0'
Port 631 is the default CUPS port. The above assumes you have a Cups server running on your viewer machine (localhost:631), if not, use something like my-cups-srv:631 (the viewer-side Cups server) in the -R instead.
Note that we use 6631 instead of 631 on the remote side because 631 would require root permission to listen on (and you likely have a cups server running on it already.)
Now the tricky part: to get applications to notice your cups server/printer on localhost:6631.
If you have administrative privilege (i.e. root password) on the x11vnc side where the desktop is running, it should be easy to add the printer through some configuration utility (e.g. in KDE: Utilities -> Printing -> Printing Manager, and then supply admin password, and then Add Printer/Class, and then fill in the inquisitive wizard. Most important is the "Remote IPP server" panel where you put in localhost for Host and 6631 for Port.) The main setting you want to convey is the host is localhost and the port is non-standard (e.g. 6631.) Some configuration utilities will take an Internet Printing Protocol (IPP) URI, e.g. http://localhost:6631/printers/, ipp://localhost:6631/printers/printer-name, ipp://localhost:6631/ipp/printer-name, etc. Check your CUPS documentation and admin interfaces to find what the syntax is and what the "printer name" is.
If you do not have root or print admin privileges, but are running a recent (version 1.2 or greater) of the Cups client software, then an easy way to temporarily switch Cups servers is to create the directory and file: $HOME/.cups/client.conf on the remote side with a line like:
ServerName localhost:6631
When not using x11vnc for remote access you can comment the above line out with a '#' (or rename the client.conf file), to have normal cups operation.
Unfortunately, running applications may need to be restarted to notice the new printers (libcups does not track changes in client.conf.) Depending on circumstances, a running application may actually notice the new printers without restarting (e.g. no print dialog has taken place yet, or there are no CUPS printers configured on the remote side.)
Cups client software that is older (1.1) does not support appending the port number, and for newer ones there is a bug preventing it from always working (fixed in 1.2.3.) Kludges like these at the command line will work:
far-away> env CUPS_SERVER=localhost IPP_PORT=6631 lpstat -p -d
far-away> env CUPS_SERVER=localhost IPP_PORT=6631 lpr -P myprinter file.ps
far-away> env CUPS_SERVER=localhost IPP_PORT=6631 firefox
but are somewhat awkward since you have to retroactively set the env. var IPP_PORT. Its value cannot be broadcast to already running apps (like the $HOME/.cups/client.conf trick sometimes does.) A common workaround for an already running app is to somehow get it to "Print To File", e.g. file.ps and then use something like the lpr example above. Also, the option "-h host:port" works with CUPS lp(1) and lpr(1).
You can also print to Windows shares printers in principle. You may do this with the smbspool(8) command, or configure the remote CUPS via lpadmin(8), etc, to use a printer URI something like smb://machine:port/printer (this may have some name resolution problems WRT localhost.) Also, as with SMB mounting, the port redir (-R) to the Windows machine must use the actual IP address instead of "localhost".
At some point we hope to fold some automation for CUPS ssh redir setup into the Enhanced TightVNC Viewer (SSVNC) package we provide (as of Sep 2006 it is there for testing.)
Q-129: How can I hear the sound (audio) from the remote applications on the desktop I am viewing via x11vnc?
You will have to use an external network audio mechanism for this. Audio is not part of the VNC protocol.
We show a simple Unix to Unix esd example here (artsd should be possible too, we have also verified the esd Windows port works for the method described below.)
First you will need a tunnel to redirect the audio from the remote machine to the one you sitting at. We use an ssh tunnel: sitting-here> ssh -C -R 16001:localhost:16001 far-away.east
Or one could combine this with the VNC tunnel at the same time, e.g.:
sitting-here> ssh -C -R 16001:localhost:16001 -t -L 5900:localhost:5900 far-away.east 'x11vnc -localhost -display :0'
Port 16001 is the default ESD uses. So when an application on the remote desktop makes a sound it will connect to this tunnel and be redirected to port 16001 on the local machine (sitting-here in this example.) The -C option is an attempt to compress the audio a little bit.
So we next need a local (sitting-here) esd daemon running that will receive those requests and play them on the local sound device:
sitting-here> esd -promiscuous -port 16001 -tcp -bind 127.0.0.1
See the esd(1) man page for the meaning of the options (the above are not very secure.) (This method also works with the EsounD windows port esd.exe)
To test this sound tunnel, we use the esdplay program to play a simple .wav file:
far-away> esdplay -s localhost:16001 im_so_happy.wav
If you hear the sound (Captain Kirk in this example), that means you are in great shape.
To run individual audio applications you can use the esddsp(1) command:
far-away> esddsp -s localhost:16001 xmms
Then you could try playing some sounds inside xmms. You could also set the environment variable ESPEAKER=localhost:16001 to not need to supply the -s option all the time. (for reasons not clear, sometimes esddsp can figure it out on its own.) All the script esddsp does is to set ESPEAKER and LD_PRELOAD for you so that when the application opens the sound device (usually /dev/dsp) its interactions with the device will be intercepted and sent to the esd daemon running on sitting-here (that in turn writes them to the real, local /dev/dsp.)
Redirecting All sound:
It does not seem to be possible to switch all of the sound of the remote machine from its sound device to the above esd+ssh tunnel without some preparation. But it can be done reasonably well if you prepare (i.e. restart) the desktop with this in mind.
Here is one way to redirect all sound. The idea is we run the entire desktop with sound directed to localhost:16001. When we are sitting at far-away.east we run "esd -promiscuous -port 16001 -tcp -bind 127.0.0.1" on far-away.east (to be able to hear the sound.) However, when we are sitting at sitting-here.west we kill that esd process and run that same esd command on sitting-here.west and start up the above ssh tunnel. This is a little awkward, but with some scripts one would probably kill and restart the esd processes automatically when x11vnc is used.
So next we have to run the whole desktop pointing toward our esd. Here is a simple way to test. Log in to the machine via the "FailSafe" desktop. Then in the lone terminal type something like:
esddsp -s localhost:16001 gnome-session
or:
esddsp -s localhost:16001 startkde
where the last part is whatever command starts your desktop (even fvwm2.) This causes the environment variables ESPEAKER and LD_PRELOAD to be set appropriately and every application (processes with the desktop as an ancestor) will use them. If this scheme works well you can make it less klunky by adding the command to your ~/.xsession, etc. file that starts your default desktop. Or you may be able to configure your desktop to use localhost:16001, or whatever is needed, via a gui configuration panel. Some Notes:
- Not all audio applications are compatible with the esd and artsd mechanisms, but many are.
- The audio is not compressed so you probably need a broadband or faster connection. Listening to music may not be very pleasant... (Although we found streaming music from across the US over cable modem worked OK, but took 200 KB/sec, to use less bandwidth consider something like "ssh far-away.east 'cat favorite.mp3' | mpg123 -b 4000 -")
- Linux does not seem to have the concept of LD_PRELOAD_64 so if you run on a mixed 64- and 32-bit ABI system (e.g. AMD x86_64) some of the applications will fail to run because LD_PRELOAD will point to libraries of the wrong wordsize.
- At some point we hope to fold some automation for esd or artsd ssh redir setup into the Enhanced TightVNC Viewer (SSVNC) package we provide (as of Sep/2006 it is there for testing.)
As of Dec/2003 "Beep" XBell events are tracked by default. The X server must support the XKEYBOARD extension (this is not on by default in Solaris, see Xserver(1) for how to turn it on via +kb), and so you won't hear them if the extension is not present.
If you don't want to hear the beeps use the -nobell option. If you want to hear the audio from the remote applications, consider trying a redirector such as esd.
Update: as of Apr/2010 in the 0.9.10 x11vnc development tarball, there is now built-in support for IPv6 (128 bit internet addresses.) See the -6 and -connect options for details.
The remainder of this FAQ entry shows how to do with this with pre 0.9.10 x11vnc using IPv6 helper tools.
Using an external IPv6 helper: A way to do this is via a separate helper program such as inetd (or for encrypted connections: ssh or stunnel.) For example, you configure x11vnc to be run from inetd or xinetd and instruct it to listen on an IPv6 address. For xinetd the setting "flags = IPv6" will be needed. For inetd.conf, an example is:
5900 stream tcp6 nowait root /usr/sbin/tcpd /usr/local/bin/x11vnc_wrapper.sh
We also provide a transitional tool in "x11vnc/misc/inet6to4" that acts as a relay for any IPv4 application to allow connections over IPv6. For example:
inet6to4 5900 localhost:5900
where x11vnc is listening on IPv4 port 5900.
Also note that not all VNC Viewers are IPv6 enabled, so a redirector may also be needed for them. The tool "inet6to4 -r ..." can do this as well. SSVNC (see below) supports IPv6 without need for the helper.
# ./inet6to4 -help
inet6to4: Act as an ipv6-to-ipv4 relay for tcp applications that
do not support ipv6.
Usage: inet6to4
inet6to4 -r
Examples: inet6to4 5900 localhost:5900
inet6to4 8080 web1:80
inet6to4 -r 5900 fe80::217:f2ff:fee6:6f5a%eth0:5900
The -r option reverses the direction of translation (e.g. for ipv4
clients that need to connect to ipv6 servers.) Reversing is the default
if this script is named 'inet4to6' (e.g. by a symlink.)
Use Ctrl-C to stop this program.
You can also set env. vars INET6TO4_LOOP=1 or INET6TO4_LOOP=BG
to have an outer loop restarting this program (BG means do that
in the background), and INET6TO4_LOGFILE for a log file.
Also set INET6TO4_VERBOSE to verbosity level and INET6TO4_WAITTIME
and INET6TO4_PIDFILE (see below.)
The "INET6TO4_LOOP=BG" and "INET6TO4_LOGFILE=..." env. variables make the tool run reliably as a daemon for very long periods. Read the top part of the script for more information.
Encrypted Tunnels with IPv6 Support: For SSH tunnelled encrypted VNC connections, one can of course use the IPv6 support in ssh(1).
For SSL encrypted VNC connections, one possibility is to use the IPv6 support in stunnel(1). This includes the built-in support via the -stunnel option. For example:
x11vnc -stunnel SAVE -env STUNNEL_LISTEN=:: -env STUNNEL_DEBUG=1 ...
SSH IPv6 Tricks: It is interesting to note that ssh(1) can do basically the same thing as inet6to4 above by:
ssh -g -L 5900:localhost:5901 localhost "printf 'Press Enter to Exit: '; read x"
(where we have x11vnc running via "-rfbport 5901" in this case.)
Note that one can also make a home-brew SOCKS5 ipv4-to-ipv6 gateway proxy using ssh like this:
ssh -D '*:1080' localhost "printf 'Press Enter to Exit: '; read x"
then specify a proxy like socks://hostname:1080 where hostname is the machine running the above ssh command (add -v to ssh for connection logging info.)
IPv6 SSVNC Viewer: Our SSVNC VNC Viewer is basically a wrapper for ssh(1) and stunnel(1), and so it already has good IPv6 support because these two commands do. On Unix, MacOSX, and Windows nearly all of the the remaining parts of SSVNC (e.g. the built-in proxying and un-encrypted connections) have been modified to support IPv6 in SSVNC 1.0.26.
Contributions:
Please do (any amount is appreciated; very few have donated) and thank you for your support! Click on the PayPal button below for more info.
[x-click-but04.gif]-Submit
Q-133: I am experiencing extreme framebuffer update lags and am using an Intel graphics card. What to do?
Depending on your graphics card model and x.org driver used, several approaches might work out for you:
Some users have reported that when using the 'intel' x.org driver, switching the driver's option named "AccelMethod" to "UXA" helped.
Others reported that using the 'modesetting' x.org driver caused the lags and switching to the 'intel' x.org driver helped.
Some reported that lags only occured with xfwm as window manager and with x11vnc started with monitors off, no matter what the x.org driver. In their case, one of disabling vsync, using 'modesetting' with "AccelMethod" "none" or disabling compositing all helped.