This chapter is from the book
This chapter is from the book
This chapter is from the book
3.5 Remote Login
3.5.1 Telnet
Telnet provides simple terminal access to a machine. The protocol includes provisions for handling various terminal settings such as raw mode, character echo, and so on. As a rule, telnet daemons call login to authenticate and initialize the session. The caller supplies an account name and usually a password to login.
Most telnet
sessions come from untrusted machines. Neither the calling program, the
calling operating system, nor the intervening networks can be trusted. The
password and the terminal session are available to prying eyes. The
local telnet program may be compromised to record username and password
combinations or to log the entire session. This is a common hacking trick, and
we have seen it employed often.
In 1994, password sniffers were discovered on a number of well-placed hosts belonging to major Internet service providers (ISPs). These sniffers had access to a significant percent of the Internet traffic flow. They recorded the first 128 characters of each telnet, ftp, and rlogin that passed. This is enough to record the destination host, username, and password.
These sniffers are often discovered when a disk fills up and the system administrator investigates. On the other hand, there are now sniffers available that encrypt their information with public keys, and ship them elsewhere.
Traditional passwords are not reliable when any part of the communications link is tapped. We strongly recommend the use of a one-time password scheme. The best are based on some sort of handheld authenticator (see Chapter 7 for a more complete discussion of this and other options).
The authenticators can secure a login nicely, but they do not protect the rest of a session. Wiretappers can read the text of the session (perhaps proprietary information read during the session), or even hijack the session after authentication is complete (see Section 5.10.) If the telnet command has been tampered with, it could insert unwanted commands into your session or retain the connection after you think you have logged off.
The same could be done by an opponent who plays games with the wires. Since early 1995, the hacking community has had access to TCP hijacking tools, which enable them to commandeer TCP sessions under certain circumstances. Telnet and rlogin sessions are quite attractive targets. Our one-time passwords do not protect us against this kind of attack using standard telnet.
It is possible to encrypt telnet sessions, as discussed in Chapter 18. But encryption is useless if you cannot trust one of the endpoints. Indeed, it can be worse than useless: The untrusted endpoint must be provided with your key, thus compromising it. Several encrypted telnet solutions have appeared. Examples include stel [Vincenzetti et al., 1995], SSLtelnet, stelnet [Blaze and Bellovin, 1995], and especially ssh [Ylonen, 1996 ¨ ].
There is also a standardized version of encrypting telnet [Ts'o, 2000], but it isn't clear how many vendors will implement it. Ssh appears to be the de facto standard.
3.5.2 The "r " Commands
To the first order, every computer in the world is connected to every other computer.
—BOB MORRIS
The "r" commands rely on the BSD authentication mechanism. One can rlogin to a remote machine without entering a password if the authentication's criteria are met. These criteria are as follows:
The call must originate from a privileged TCP port. On other systems (like PCs) there are no such restrictions, nor do they make any sense. A corollary of this is that rlogin and rsh calls should be permitted only from machines on which this restriction is enforced.
The calling user and machine must be listed in the destination machine's list of trusted partners (typically /etc/hosts.equiv) or in a user's .rhostsfile.
The caller's name must correspond to its IP address. (Most current implementations check this. See Section 2.2.2.)
From a user's viewpoint, this scheme works fairly well. Users can bless the machines they want to use, and won't be bothered by passwords when reaching out to more computers.
For the hackers, these routines offer two benefits: a way into a machine, and an entry into even more trusted machines once the first computer is breached. A principal goal of probing hackers is to deposit an appropriate entry into /etc/hosts.equivor some user's .rhostsfile. They may try to use FTP, uucp, TFTP, or some other means. They frequently target the home directory of accounts not usually accessed in this manner, such as root, bin, ftp, or uucp. Be especially wary of the latter two, as they are file transfer accounts that often own their own home directories. We have seen uucp being used to deposit a .rhostsfile in /usr/spool/uucppublic, and FTP used to deposit one in /usr/ftp. The permission and ownership structure of the server machine must be set up to prohibit this, and it frequently is not.
The connection
is validated by the IP address and reverse DNS entry of the caller. Both of
these are suspect: The hackers have the tools needed for IP spoofing attacks
(see Section 2.1.1) and the compromise of DNS (see Section 2.2.2). Address-based
authentication is generally very weak, and only suitable in certain very controlled
situations. It is a poor choice in most situations where the r commands
are currently employed.
When hackers have acquired an account on a computer, their first goals are usually to cover their tracks by erasing logs (not that most versions of the rsh daemon create any), attain root access, and leave trapdoors to get back in, even if the original access route is closed. The /etc/hosts.equivand $HOME/.rhostsfiles are a fine route.
Once an account is penetrated on one machine, many other computers may be accessible. The hacker can get a list of likely trusting machines from /etc/hosts.equiv, files in the user's bindirectory, or by checking the user's shell history file. Other system logs may suggest other trusting machines. With other /etc/passwdfiles available for dictionary attacks, the target site may be facing a major disaster.
Notice that quite of a bit of a machine's security is in the hands of the user, who can bless remote machines in his or her own .rhostsfile and can make the .rhostsfile world-writable. We think these decisions should be made only by the system administrator. Some versions of the rlogin and rsh daemons provide a mechanism to enforce this; if yours do not, a cron job that hunts down rogue .rhostsfiles might be in order.
Given the many weaknesses of this authentication system, we do not recommend that these services be available on computers that are accessible from the Internet, and we do not support them to or through our gateways. Of course, note the quote at the start of this section: You may have more machines at risk than you think. Even if there is no direct access to the Internet, an inside hacker can use these commands to devastate a company.
There is a delicate trade-off here. The usual alternative to rlogin is to use telnet plus a cleartext password, a choice that has its own vulnerabilities. In many situations, the perils of the latter outweigh the risks of the former; your behavior should be adjusted accordingly.
The r commands are a major means by which hackers spread their attack through a trusting community. If host A trusts host B, and B trusts C, then A and C are connected by transitive trust. An attacker only needs to break into a single host, the weakest link, of a group of computers. The rest of the hosts just let them log in. We wonder how interlinked a large corporation's intranet may be based simply on this transitive relation of trust.
There is one more use for rlogind that is worth mentioning. The protocol is capable of carrying extra information that the user supplies on the command line, nominally as the remote login name. This can be overloaded to contain a host name as well, perhaps to supply additional information to an intermediate relay host. This is safe as long as you do not grant any privileges based on the information thus received. Hackers have used this data path to open previously installed back doors in systems.
3.5.3 Ssh
Ssh [Ylonen,¨ 1996] is a replacement for rlogin, rdist, rsh and rcp, written by Tatu Ylonen.¨ It includes replacement programs—ssh and scp—that have the same user interface as rsh and rcp, but use an encrypted protocol. It also includes a mechanism that can tunnel X11 or arbitrary TCP ports.
A variety of encryption and authentication methods are available. Ssh can supplement or replace traditional host and password authentication with RSA- or DSA-keyed and challenge response authentication.
It is a fundamental tool for the modern network administrator, although it takes a bit of study to install it safely. There is much to configure: authentication type, encryption used, host keys, and so on. Each host has a unique key, but users can have their own keys, too. Moreover, the user keys can be passed on to subsequent connections using the ssh-agent. There are two protocols, numbers one and two, and the first has had a number of problems—we stick to protocol two when we can, though we must sometimes support older implementations that only speak protocol one.
We have a number of concerns about ssh and its configuration and protocols:
The original protocol was custom-designed. This is always dangerous—protocol design is a black art, and looks much easier than it is. History has shown that Tatu did a decent job, but there have been problems (c.f. CERT Vulnerability Note VU#596827). On at least two occasions so far, the protocol has been changed in response to security problems. The fixes were prompt, and we have some fair confidence in the protocol. Even with the flaws, ssh has been much safer than the alternatives.
An IETF standards group is working on standardizing version 2 of the protocol.
The server runs as root (this one really needs to) and is complicated, hard to audit, and dangerous (CERT Advisory CA-1999-15, CERT Vulnerability Note VU#40327).
The server cannot specify authentication at the client level. For example, the sshd server is configured with PasswordAuthenticationyesor no, for all clients. The selection of the authentication method should belong to the owner of the machine, and be configured in the owner's server. In addition, the owner should be able to decide that for this host key, no password is needed, and for other hosts, a password or user key is required. The host-specific entries of ssh configshould be implemented in sshd config.
Commercialization of ssh caused a code split. The commercial version now competes with OpenSSH. There are a variety of Windows-based versions of varying capabilities and prices. The freeware putty client is nice, as it requires no installation.
All our eggs are in the ssh basket. A major hole here causes thousands of administrators to drop everything and scramble to repair the problem. Unfortunately, this has happened more than once. It seems to happen when the administrator is traveling. . .
The user can lock an RSA or DSA key in a file with a passphrase. If the host is compromised, that file is subject to dictionary attacks.
One can tunnel other protocols over ssh and thus evade firewalls.
We discuss how to use ssh safely in Section 8.2, and the cryptographic options in Section 18.4.1.