This document describes lsh and related programs. The lsh suite of programs is intended as a free replacement for the ssh suite of programs. In turn, ssh was intended as a secure replacement for the rsh and rlogin programs for remote login over the Internet.
lsh is a component of the GNU system.
This manual explains how to use and hack lsh; it corresponds to lsh version 2.0.
--- The Detailed Node Listing ---
Introduction
Related programs and techniques
Getting started
Invoking lsh
What is this thing called computer security anyway? Why would you want to use a program like lsh?
This chapter explains the threats lsh tries to protect you from, and some of the threats that remain. It also describes some of the technologies used in lsh.
From time to time in this manual, I will speak about the enemy. This means anybody who is trying to eavesdrop or disturb your private communication. This usage is technical, and it does not imply that the enemy is somehow morally inferior to you: The enemy may be some awful criminals trying to eavesdrop on you, or it may be the police trying to eavesdrop on the same criminals.
The enemy can be a criminal, or a competitor, or your boss who's trying to find out how much you tell collegues at competing firms. It may be your own or somebody else's national security officials. Or your ex-boyfriend who happens to be too curious.
So what can the enemy do to your communications and your privacy? Remember that just because you're paranoid that doesn't mean that nobody is trying to get you...
When logging in to some other machine via the Internet, either in the same building or a few continents away, there are several things that may be under enemy attack.
And even without routing anomalies, it is possible that the enemy has been able to take control of some nearby machine, and can listen in from there. Of course, passive eavesdropping is most dangerous if you transmit cleartext passwords. This is the main reason not to use vanilla telnet to login to remote systems. Use a telnet with support for SSL or Kerberos, or use a program like lsh or ssh.
A passive eavesdropper is assumed not to do anything nasty with your
packets beyond listening to them.
lsh makes no attempt to protect you from local attacks. You have to trust the endpoint machines. It seems really difficult to uphold any security if the local machine is compromised. This is important to keep in mind in the “visitor”-scenario, where you visit a friend or perhaps an Internet café and want to connect to some of the machines at home or at work. If the enemy has been able to compromize your friend's or the café's equipment, you may well be in trouble.
Protection from denial-of-service attacks is also a very difficult problem, and lsh makes no attempt to protect you from that.
Instead, the aim of lsh, and most serious tools for cryptographic protection of communications across the net, is to isolate the vulnerabilities to the communication endpoints. If you know that the endpoints are safe, the enemy should not be able to compromize your privacy or communications. Except for denial-of-service attacks (which at least can't be performed without you noticing it).
First of all, lsh provides protection against passive eavesdropping. In addition, if you take the appropriate steps to make sure that hostkeys are properly authenticated, lsh also protects against man-in-the-middle attacks and in particular against attacks on the name resolution. In short, you need only trust the security at the end points: Even if the enemy controls all other network equipment, name resolution and routing infrastructure, etc, he can't do anything beyond the denial-of-service attack.
And at last, remember that there is no such thing as absolute security. You have to estimate the value of that which you are protecting, and adjust the security measures so that your enemies will not find it worth the effort to break them.
lsh does not only provide more secure replacements for telnet, rsh and rlogin, it also provides some other features to make it convenient to communicate securely. This section is expected to grow with time, as more features from the wish-list are added to lsh. One goal for lsh is to make it reasonable easy to extend it, without messing with the core security functionality.
lsh can also be used in something called gateway mode, in which you can authenticate once and set up a connection that can later be used for quickly setting up new sessions with lshg (see Invoking lshg).
lsh can be configured to allow login based on a personal key-pair consisting of a private and a public key, so that you can execute remote commands without typing your password every time. There is also experimental support for Thomas Wu's Secure Remote Password Protocol (SRP). Kerberos support is on the wish list but not yet supported (see Kerberos).
The public-key authentication methods should also be extended to support Simple Public Key Infrastructure (SPKI) certificates, including some mechanism to delegate restricted logins.
Forwarding of arbitrary TCP/IP connections is provided. This is useful for tunneling otherwise insecure protocols, like telnet and pop, through an encrypted lsh connection.
lsh also features a SOCKS-proxy which also provides tunneling of TCP/IP connections, but can be easily used, e.g. from within popular web browsers like Mozilla and Firefox for tunneling web traffic. There are also programs like tsocks that performs transparent redirection of network access through a SOCKS proxy.
Convenient tunneling of X was one of the most impressive features of the original ssh programs. Both lsh and lshd support X-forwarding, although lshg does not.
Whan X forwarding is in effect, the remote process is started in an environment where the DISPLAY variable in the environment points to a fake X server, connections to which are forwarded to the X server in your local environment. lsh also creates a new “fake” ‘MIT-MAGIC-COOKIE-1’ for controlling access control. Your real X authentication data is never sent to the remote machine.
Other kinds of tunneling that may turn out to be useful include authentication (i.e. ssh-agent), general forwarding of UDP, and why not also general IP-tunneling.
This sections describes some other programs and techniques related to lsh. The ssh family of programs use mostly the same kind of security as lsh. Kerberos and IPSEC operate quite differently, in particular when it comes to protection against man-in-the-middle attacks.
ssh-1.xThe first of the Secure shell programs was Tatu Ylönen's ssh.
The latest of the version 1 series is ssh-1.33 which speaks
version 1.5 of the protocol. The “free” version of ssh-1.33
does not allow commercial use without additional licensing, which makes
ssh-1.33 non-free software according to Debian's Free Software
Guidelines and the Open Source Definition.
The version 1 protocol has some subtle weaknesses, in particular, all support for using stream ciphers was disabled by default a few versions back, for security reasons.
There also exists free implementations of ssh-1, for both Unix
and Windows. ossh and later OpenSSH are derived from earlier
version av Tatu Ylönen's ssh, and are free software.
ssh-2.xssh2 implements the next generation of the Secure Shell protocol, the development of which is supervised by the IETF secsh Working Group. Besides lsh, some well known implementations of this protocol includes
There a numerous other implementations, both free and proprietary. The above list is far from complete.
Kerberos is a key distribution system originally developed in the late 1980:s as a part of Project Athena at MIT. Recent development have been done at The Royal Institute of Technology, Stockholm (KTH).
Kerberos uses a central trusted ticket-granting server, and requires less trust on the local machines in the system. It does not use public-key technology.
Usually, Kerberos support is compiled into applications such as telnet, ftp and X-clients. The ssh family of programs, on the other hand, tries to do all needed magic, for instance to forward X securely, and then provides general TCP/IP forwarding as a kitchen sink.
I believe Kerberos' and lsh's protection against passive eavesdropping are mostly equivalent. The difference is in the set of machines and assumptions you have to trust in order to be safe from a man-in-the-middle attack.
I think the main advantage of lsh over Kerberos is that it is easier to install and use for on ordinary mortal user. In order to set up key exchange between two different Kerberos systems (or Kerberos realms), the respective system operators need to exchange keys. In the case of two random users at two random sites, setting up lsh or some other program in the ssh family is likely easier than to get the operators to spend time and attention. So lsh should be easier to use in an anarchistic grass-roots environment.
Another perspective is to combine ssh features like X and TCP/IP forwarding with authentication based on Kerberos. Such an arrangement may provide the best of two worlds for those who happen to have an account at a suitable ticket-granting server.
IPSEC is a set of protocols for protecting general IP traffic. It is developed by another IETF working group, and is also a required part of IP version 6.
Again, the main difference between IPSEC, Kerberos and ssh is the set of machines that have to be secure and the keys that have to be exchanged in order to avoid man-in-the-middle attacks.
Current protocols and implementations of IPSEC only provide authentication of machines; there's nothing analogous to the user authentication in ssh or Kerberos.
On the other hand, IPSEC provides one distinct advantage over application level encryption. Because IP and TCP headers are authenticated, it provides protection against some denial-of-service attacks. In particular, it makes attacks that cause hangup of a TCP connection considerably more difficult.
So it makes sense to use both IPSEC and some application level cryptographic protocol.
Also note that it is possible to use the Point-to-Point Protocol (PPP) to tunnel arbitrary IP traffic accross an ssh connection. This arrangement provides some of the functionality of IPSEC, and is sometimes referred to as “a poor man's Virtual Private Network”.
You install lsh with the usual ./configure && make &&
make install. For a full listing of the options you can give to
configure, use ./configure --help. For example, use
--without-pty to disable pty-support.
The most commonly used option is --prefix, which tells configure where lsh should be installed. Default prefix is /usr/local. The lshd server is installed in $prefix/sbin, all other programs and scripts are installed in $prefix/bin.
The configure script tries to figure out if the linker needs any special flags specifying where to find dynamically linked libraries at run time (one case where this matters is if you have a dynamic libz.so installed in a non-standard place). Usually, you can use
./configure --with-lib-path=/opt/lib:/other/place
to specify extra library directories, and the configure script should do the right thing. If this doesn't work, or you believe that you know your system better than ./configure, just set LDFLAGS and/or LD_LIBRARY_PATH to the right values instead.
This section tells you how to perform some common tasks using the lsh suite of programs, without covering all options and possibilities.
Several of the lsh programs requires a good pseudorandomness generator for secure operation. The first thing you need to do is to create a seed file for the generator. To create a personal seed file, stored as ~/.lsh/yarrow-seed-file, run
lsh-make-seed
To create a seed file for use by lshd, run
lsh-make-seed --server
as root. The seed file is stored as /var/spool/lsh/yarrow-seed-file.
lsh is the program you use for connection to a remote machine. A few examples are:
lsh sara.lysator.liu.se
Connects to ‘sara.lysator.liu.se’ and starts an interactive shell. In this example, and in the rest of the examples in this section, lsh will ask for your password, unless you have public-key user authentication set up.
The first time you try to connect to a new machine, lsh typically complains about an “unknown host key”. This is because it has no reason to believe that it was the right machine that answered, and not a machine controlled by the enemy (see Threats). The default behaviour is to never ever accept a server that is not properly authenticated. A machine is considered authentic if it follows the protocol and has an acl-entry for its public hostkey listed in ~/.lsh/host-acls.
To make lsh less paranoid, use
lsh --sloppy-host-authentication sara.lysator.liu.se
Then lsh will display a fingerprint of the host key of the remote machine, and ask you if it is correct. If so, the machine is considered authentic and a corresponding acl-entry is appended to the file ~/.lsh/captured_keys. You can copy acl-entries you have verified to ~/.lsh/host-acls.
You can even use
lsh --sloppy-host-authentication --capture-to ~/.lsh/host-acls
to get lsh to behave more like the traditional ssh program.
lsh -l omar sara.lysator.liu.se
Connects, like above, but tries to log in as the user “omar”.
lsh sara.lysator.liu.se tar cf - some/dir | (cd /target/dir && tar -xf -)
Copies a directory from the remote machine, by executing one remote and one local tar process and piping them together.
CVS_RSH=lsh cvs -d cvs.lysator.liu.se:/cvsroot/lsh co lsh
Checks out the lsh source code from the CVS repository.
lsh -G -B sara.lysator.liu.se
Opens an ssh connection, creates a “gateway socket”, and forks into the background.
lshg sara.lysator.liu.se
creates a new session using an existing gateway socket, without the overhead for a new key exchange and without asking for any passwords.
One useful feature of lsh and other ssh-like programs is the ability to forward arbitrary connections inside the encrypted connection. There are two flavors: “local” and “remote” forwarding.
An example of local forwarding is
lsh -L 4000:kom.lysator.liu.se:4894 sara.lysator.liu.se
This makes lsh listen on port 4000 on the local machine. When someone connects, lsh asks the server to open a connection from the remote machine (i.e. ‘sara’) to port 4894 on another machine (i.e. ‘kom’). The two connections are piped together using an encrypted channel.
There are a few things that should be noted here:
Only the middle part is protected by lsh: all data flowing through the tunnel is sent across the first and last part in the clear. So forwarding doesn't offer much protection unless the tunnel endpoint and the ultimate target machine are close to each other. They should usually be either the same machine, or two machines connected by a local network that is trusted.
Remote forwarding is similar, but asks the remote machine to listen on a port. An example of remote forwarding is
lsh -g -R 8080:localhost:80 sara.lysator.liu.se
This asks the remote machine to listen on port 8080 (note that you are probably not authorized to listen on port 80). Whenever someone connects, the connection is tunnelled to your local machine, and directed to port 80 on the same machine. Note the use of -g; the effect is to allow anybody in the world to use the tunnel to connect to your local webserver.
The same considerations that apply to forwarded local ports apply also to forwarded remote ports.
At last, you can use any number of -L and -R options on the same command line.
There are no global configuration files for lshd; all configuration is done with command line options (see Invoking lshd).
To run lshd, you must first create a hostkey, usually stored in /etc/lsh_host_key. To do this, run
lsh-keygen --server | lsh-writekey --server
This will also create a file /etc/lsh_host_key.pub, containing the corresponding public key.
A typical command line for starting lshd in daemon mode is simply
lshd --daemonic
You can find init script for lshd tailored for Debian's and RedHat's GNU/Linux systems in the contrib directory.
It is also possible to let init start lshd, by adding it in /etc/inittab.
Public-key user authentication is a way to authenticate for login, without having to type any passwords. There are two steps: Creating a key pair, and authorizing the public key to the systems where you want to log in.
To create a keypair, run
lsh-keygen | lsh-writekey
This can take some time, but in the end it creates two files ~/.lsh/identity and ~/.lsh/identity.pub.
If you want to use the key to login to some other machine, say ‘sara’, you can do that by first copying the key,
lsh sara.lysator.liu.se '>my-key.pub' < ~/.lsh/identity.pub
then authorizing it by executing, on ‘sara’,
lsh-authorize my-key.pub
By default, lsh-writekey encrypts the private key using a passphrase. This gives you some protection if a backup tape gets into the wrong hands, or you use NFS to access the key file in your home directory. If you want an unencrypted key, pass the flag -c none to lsh-writekey.
For security reasons, you should keep the private key ~/.lsh/identity secret. This is of course particularly important if the key is unencrypted; in that case, anybody who can read the file will be able to login in your name to any machine where the corresponding public key is registered as an authorized key.
Naturally, you should also make sure not to authorize any keys but your
own. For instance, it is inappropriate to use an insecure mechanism such
as unauthenticated email, ftp or http to transfer your
public key to the machines where you want to authorize it.
If you have accounts on several systems, you usually create one keypair on each of the systems, and on each system you authorize some or all of your other public keys for login.
The Secure Remote Password protocol is a fairly new protocol that provides mutual authentication based on a password. To use it, you must first choose a secret password. Next, you create a password verifier that is derived from the password. The verifier is stored on the target machine (i.e. the machine you want to log in to).
To create a verifier, you run the srp-gen program and type your new password. You have to do it on either the target machine, redirecting the output to ~/.lsh/srp-verifier, or you can generate it on some other machine and copy it to the target.
The main advantage of using SRP is that you use the password not only to get access to the remote machine, but you also use it to authenticate the remote machine. I.e. you can use it to connect securely, without having to know any hostkeys or fingerprints beforehand!
For instance, you could connect using SRP to fetch the hostkey fingerprint for the remote machine, as a kind of bootstrapping procedure, and then use traditional authentication methods for further connections.
For this to work, the verifier must be kept secret. If the enemy gets your verifier, he can mount some attacks:
If you use SRP to get the hostkey or fingerprint for the remote machine, as outlined above, the impersonation attack destroys security, you could just as well connect the hostkey presented by the remote server without verifying it at all.
If you use SRP exclusively, the situation seems somewhat different. As far as I can see, an attacker knowing your verifier can not mount a traditional man-in-the-middle-attack: He can play the server's part when talking to you, but in order to play your part when talking to the real server, he needs to know your password as well.
SRP support is disabled by default, but can be enabled by the --srp-keyexchange option to lshd and lsh (naturally, it won't be used unless enabled on both sides). At the time of this writing, SRP is too new to be trusted by conservative cryptographers (and remember that conservatism is a virtue when it comes to security).
And even if SRP in itself is secure, the way lsh
integrates it into the ssh protocol has not had much peer review.
The bottom line of this disclaimer is that the SRP support in
lsh should be considered experimental.
As far as I know, using SRP as a host authentication mechanism
is not supported by any other ssh implementation. The protocol
lsh uses is described in the doc/srp-spec.txt.
Implementations that use SRP only as a user authentication
mechanism are not compatible with lsh.
Keys and most other objects lsh needs to store on disk are represented as so called S-expressions or sexps for short. S-expressions have their roots in the Lisp world, and a variant of them in used in the Simple Public Key Infrastructure (SPKI). Currently, lsh's support for SPKI is quite limited, but it uses SPKI's formats for keys and Access Control Lists (ACL:s).
There are several flavours of the sexp syntax:
To see what your ~/.lsh/identity.pub file really contains, try
sexp-conv < ~/.lsh/identity.pub
The sexp-conv program can also be used to computes fingerprints. The fingerprint of a key (or any sexp, for that matter) is simply the hash of its canonical representation. For example,
sexp-conv --hash </etc/lsh_host_key.pub
This flavour of fingerprints is different from the ssh fingerprint convention, which is based on a hash of the key expressed in ssh wire format. To produce ssh standard fingerprints, use ‘lsh-export-key --fingerprint’.
If you are already using ssh2 or OpenSSH, and have created one or more personal keypairs, you need to convert the public keys to lsh's format before you can authorize them. Use the supplied ssh-conv script,
ssh-conv <openssh-key.pub >new-key.pub
You can then use the usual lsh-authorize on the converted keys. ssh-conv supports both DSA and RSA keys.
Conversion of keys the other way is also possible, by using the lsh-export-key program. It reads a public key in the SPKI format used by lsh on stdin, and writes the key in ssh2/OpenSSH format on stdout.
If you want to use your lsh key to log in to another system running and OpenSSH server, you can do like this:
lsh-export-key --openssh < .lsh/identity.pub >sshkey
And on the other machine, after having somehow copied the sshkey file, just add it to the end of your authorized_keys file:
cat sshkey >> ~/.ssh/authorized_keys
lsh-export-key can also be used to check the fingerprint of keys (just like ssh-keygen).
lsh-export-key --fingerprint < /etc/lsh_host_key.pub
show the MD5 and Bubble babble fingerprint of the server public key.
There are currently no tools for converting private keys.
You use lsh to login to a remote machine. Basic usage is
‘lsh [-p port number] sara.lysator.liu.se’
which attempts to connect, login, and start an interactive shell on the remote machine. Default port number is whatever your system's /etc/services lists for ssh. Usually, that is port 22.
There is a plethora of options to lsh, to let you configure where and how to connect, how to authenticate, and what you want to do once properly logged in to the remote host. Many options have both long and short forms. This manual does not list all variants; for a full listing of supported options, use ‘lsh --help’.
Note that for many of the options to lsh, the ordering of the options on the command line is important.
Before a packet is sent, each packet can be compressed, authenticated, and encrypted, in that order. When the packet is received, it is first decrypted, next it is checked that it is authenticated properly, and finally it is decompressed. The algorithms used for this are negotiated with the peer at the other end of the connection, as a part of the initial handshake and key exchange.
Each party provides a list of supported algorithms, and the first algorithm listed by the client, which is also found on the server's list, is selected. Note that this implies that order in which algorithms are listed on the server's list doesn't matter: if several algorithms are present on both the server's and the client's lists, it's the client's order that determines which algorithm is selected.
Algorithms of different types, e.g. data compression and message authentication, are negotiated independently. Furthermore, algorithms used for transmission from the client to the server are independent of the algorithms used for transmission from the server to the client. There are therefore no less than six different lists that could be configured at each end.
The command line options for lsh and lshd don't let you specify arbitrary lists. For instance, you can't specify different preferences for sending and receiving.
There is a set of default algorithm preferences. When you use a command
line option to say that you want to use algorithm for one of the
algorithms, the default list is replaced with a list containing the
single element algorithm. For example, if you use -c
arcfour to say that you want to use arcfour as the encryption
algorithm, the connection will either end up using arcfour, or
algorithm negotiation will fail because the peer doesn't support
arcfour.
| Option | Algorithm type | Default |
|
| -z | Data compression | none, zlib
| The default preference list supports zlib compression, but
prefers not to use it.
|
| -c | Encryption | aes256-cbs, 3dec-cbc, blowfish-cbc, arcfour
| The default encryption algorithm is aes256. The default list
includes only quite old and well studied algorithms. There is a special
algorithm name all to enable all supported encryption algorithms
(except none).
|
| -m | Message Authentication | hmac-sha1, hmac-md5
| Both supported message authentication algorithms are of the
HMAC family.
|
As a special case, -z with no argument changes the compression
algorithm list to zlib, none, which means that you want to
use zlib if the other end supports it. This is different from
-zzlib which causes the negotiation to fail if the other end
doesn't support zlib. A somewhat unobvious consequence of
-z having an optional argument is that if you provide an
argument, it must follow directly after the option letter, no spaces
allowed.
As described earlier (see Threats), proper authentication of the remote host is crucial to protect the connection against man-in-the-middle attacks. By default, lsh verifies the server's claimed host key against the Access Control Lists in ~/.lsh/host-acls. If the remote host cannot be authenticated, the connection is dropped.
The options that change this behaviour are
lsh --sloppy-host-authentication --capture-to ~/.lsh/host-acls
makes lsh behave more like the ssh program.
There are many things lsh can do once you are logged in. There are two types of options that control this: actions and action modifiers. For short options, actions use uppercase letters and modifiers use lowercase.
For each modifier --foo there's also a negated form --no-foo. Options can also be negated by preceding it with the special option -n. This is mainly useful for negating short options. For instance, use -nt to tell lsh not to request a remote pseudo terminal. Each modifier and its negation can be used several times on the command line. For each action, the latest previous modifier of each pair apply.
First, the actions:
If there are trailing arguments after the name of the remote system, this is equivalent to a -E option, with a command string constructed by catenating all the remaining arguments, separated by spaces. This implies that the arguments are usually expanded first by the local shell, and then the resulting command string is interpreted again by the remote system.
If there are no trailing arguments after the name of the remote system, and the -N option is not given, the default action is to start a shell on the remote machine. I.e. this is equivalent to the -S option.
There are a few supported modifiers:
Currently, this option is ignored if there is no local terminal.
These options determines what messages lsh writes on its stderr.
SSH_MSG_USERAUTH_REQUEST
messages, but you should still use it with care.
Note that all these options are orthogonal. If you use --trace, you usually want to add -v as well; --trace does not do that automatically.
You use lshg to login to a remote machine to which you have previously used lsh to set up a gateway (see Action options). Its usage is very similar to that of lsh (see Invoking lsh), except that some options are not available.
Basic usage is
‘lshg [-l username] host’
which attempts to connect to the gateway that should previously have been established by running ‘lsh [-l username] -G host’)
The username and host are used to locate the gateway. The default value for username is determined in the same way as for lsh (see Invoking lsh).
As lshg uses almost the same options as lsh (see Invoking lsh), only options that are not available or have a different meaning in lshg are listed here.
The algorithm options (see Algorithm options) as well as most of the userauth (see Userauth options) and hostauth (see Hostauth options) are not available in lshg as they are only used by session setup, which is already handled by lsh.
Due to technical reasons, X11-forwarding cannot be performed by lshg, thus the --x11-forward option (see Action options) is not available.
To summarize, these are the options that are new, not available or that have different meanings:
lshd is a server that accepts connections from clients speaking the Secure Shell Protocol. It is usually started automatically when the systems boots, and runs with root privileges. However, it is also possible to start lshd manually, and with user privileges.
There are currently no configuration files. Instead, command line options are used to tell lshd what to do. Many options have --foo and --no-foo variants. Options specifying the default behaviour are not listed here.
Some of the options are the shared with lsh. In particular, see Algorithm options and Verbosity options.
Options specific to the lshd server are:
It should also be possible to use several -p options as a convenient way to make lshd listen on several ports on each specified (or default) interface, but that is not yet implemented.
Note that if you use both -p and --interface, the
order matters.
Some examples: --interface=localhost,
--interface=1.2.3.4:443, --interface=[aaaa::bbbb]:4711. To
make lshd listen on several ports and interfaces at the same
time, just use several --interface options on the command line.
lshd should be able to deal with the environment it inherits
if it is started by init or inetd, but this is not
really tested.
This chapters describes all files and all environment variables that are used by lsh, lshd, and related programs.
There are a few environment variables that modifies the behaviour of the lsh programs. And there are also a handful of variables that are setup by lshd when starting user processes.
Files used by the lsh client, stored in the ~/lsh directory:
Files used by lshd, some of which are read from user home directories: