Terminal Access (Telnet)
Remote Command Execution
Network News Transfer Protocol (NNTP)
World Wide Web (WWW) and HTTP
Other Information Services
Information Lookup Services
Real-Time Conferencing Services
Domain Name System (DNS)
Network Management Services
Network Time Protocol (NTP)
Network File System (NFS)
Network Information Service/Yellow Pages (NIS/YP)
X11 Window System
Printing Protocols (lpr and lp)
Analyzing Other Protocols
This chapter describes the major Internet services: how they work, what their packet filtering and proxying characteristics are, what their security implications are with respect to firewalls, and how to make them work with a firewall. The purpose of this chapter is to give you the information that will help you decide which services to offer at your site and to help you configure these services so they are as safe and as functional as possible in your firewall environment.
This chapter is intended primarily as a reference; it's not necessarily intended to be read in depth from start to finish, though you might learn a lot of interesting stuff by skimming the whole thing.
This chapter assumes that you are familiar with what the various Internet services are used for, and concentrates on explaining how to provide those services through a firewall. For introductory information about what particular services are used for, see Chapter 2, Internet Services.
Where we discuss the packet filtering characteristics of particular services, we use the same abstract tabular form we used to show filtering rules in Chapter 6, Packet Filtering. You'll need to translate various abstractions like "internal," "external," and so on to appropriate values for your own configuration. See Chapter 6 for an explanation of how you can translate abstract rules to rules for particular products and packages, as well as more information on packet filtering in general.
Where we discuss the proxy characteristics of particular services, we rely on concepts and terminology discussed in Chapter 7, Proxy Systems.
Throughout this chapter, for each service, we'll show how its packets flow through a firewall. The following figures show the basic packet flow: when a service runs directly (Figure 8.1) and when a proxy service is used (Figure 8.2). The other figures in this chapter show variations of these figures for individual services. If there are no specific figures for a particular service, you can assume that these generic figures are appropriate for that service.
NOTE: We frequently characterize client port numbers as "a random port number above 1023." Some protocols specify this as a requirement, and on others it is merely a convention (spread to other platforms from UNIX, where ports below 1024 cannot be opened by regular users). Although it is theoretically allowable for clients to use ports below 1024 on non-UNIX platforms, it is extraordinarily rare: rare enough that many firewalls, including ones on major public sites that handle clients of all types, rely on this distinction and report never having rejected a connection because of it.
From a user's point of view, electronic mail is perhaps the most fundamental Internet service. Unfortunately, it's also one of the most vulnerable. Mail servers make extremely tempting targets, because they accept arbitrary data from arbitrary external hosts.
A mail system has three parts, which may be implemented by different programs or may be implemented by the same program, in any combination.
A server accepts mail from external hosts or sends it to external hosts
A user agent lets the recipient read the mail and compose outgoing mail
Each of these parts is vulnerable for a different reason:
The server directly accepts commands (related to delivering mail) from external hosts; for this reason, if the server isn't secure, it may end up immediately giving an attacker all the access it has itself.
The delivery agent needs special permissions because it needs to be able to write into every user's mailbox. Although the delivery agent doesn't need to talk to the external world, if it can be subverted somehow, the intruder obtains very broad access.
The user agent runs as one user, and it doesn't talk to the external world, which limits its power and accessibility; however, it can often run arbitrary other programs in response to the data received.
Because it talks to the external world, the server is vulnerable to attacks in the commands it gets from the outside world; these are called command channel attacks. The delivery agent and the user agent don't receive the commands directly, but they may be vulnerable to dangers in the content of the mail message; these are called data-driven attacks. In addition, any program may be vulnerable to misuse by somebody who can control how it is executed (what arguments are used, what user is running it, what its data files are), through command-line bugs.
Data-driven attacks can exploit the delivery agent, the user agent, or the user. (The server generally pays no attention to the data.) For example, on UNIX machines, most versions of Sendmail use /bin/mail as a local delivery agent. /bin/mail is also a user agent and therefore has extensions to allow it to execute arbitrary commands. (If you have only one terminal, you're typing a mail message, and you want to see the output of some other command, it's very useful to be able to get your user agent to run the command for you.) Unfortunately, these extensions are not always disabled when /bin/mail is running as a delivery agent, and the data in the mail message is sometimes capable of triggering them, allowing an outsider to run arbitrary commands on your system merely by sending appropriately formatted mail to you.
User agents may also be subverted even when they're acting as designed, as user agents. Today's multimedia mail readers are particularly subject to attacks of this kind, because they need to execute other programs to "display" a message for you (for example, a graphics program to display a picture or an audio program to play an audio attachment), but older mail readers were vulnerable as well. One Christmas, most of IBM's corporate network was put out of commission by a data-driven attack. Users received a message, usually from someone they knew and trusted, that said simply "run me". Because it appeared to be coming from a friend, most people did run it; when they did, it displayed a Christmas tree, and then resent itself to all of the entries in the reader's address book (personal alias file). It propagated very fast, and the network melted down under the force of trying to send millions of copies.
Finally, even if the server, the delivery agent, and the user agent are bug-free, some data-driven attacks work by subverting the user, and getting the user to perform some action the attacker wants. Sometimes that action involves running the program the attacker sent in the mail. Sometimes it's something more direct; for example, it's common for attackers to send mail to users that appears to come from their system's managers and that directs them to change their password to something specific. If the users do so, the attacker can now access their account.
How well does a firewall protect against these different types of attacks?
There isn't much a firewall can do about data-driven attacks; the data has to be allowed through, or you won't actually be able to receive mail. In some cases it's possible to filter out "dangerous" characters in the mail addresses if you can somehow recognize them. Your best bet, though, is to run up-to-date delivery and user agents and to educate your users. Doing so will protect against most data-driven attacks. In any case, because data-driven attacks tend to be complicated and difficult to get information back from, they are relatively rare.
Command-line bugs are outside the scope of a firewall, because they can be exploited only by someone who is already able to execute commands on your system. One purpose of a firewall is to keep attackers from getting that ability.
The following sections describe the two protocols commonly used for electronic mail: SMTP and POP, as well as the MIME extension.
On the Internet, electronic mail exchange between mail servers is handled with SMTP. A host's SMTP server accepts mail and examines the destination address to decide whether to deliver the mail locally or to forward it on to some other machine. If it decides to deliver the mail locally, it recodes the mail headers and delivery address into the proper form for the local delivery program, and it then hands the mail to that program. If it decides to forward the mail to another machine, it modifies the headers, and contacts that machine (usually via SMTP, but sometimes via UUCP or another protocol), and forwards the mail.
SMTP is a store-and-forward system, and such systems are well-suited to firewall applications, particularly those using proxy services. In Chapter 7, Figure 7.2 shows how mail sent from an individual user's workstation is directed initially to a gateway system before leaving the user's own network. Mail entering a network goes to a gateway system on that network before being distributed to individual users on other hosts.
The mailer most commonly used on UNIX systems is Sendmail. Sendmail is very powerful, but it also has a long and troubling history of major and minor security problems. Other mailers are available, including smail 3, MMDF, and Z-Mail, but none of them appears to be particularly more secure than a modern version of Sendmail.
Sendmail's security problems have been widely discussed, while the problems of these other mailers have not. However, the lack of public discussion about other mailers should not lead you to assume these mailers are any more secure than Sendmail. While some of them are intended to be more secure than Sendmail, there is nothing to guarantee they are. Most of these alternative mailers are apparently intended to address Sendmail's incredibly baroque configuration files and its occasional failure to comply with standards, or to provide features Sendmail doesn't have. Additional security is often an afterthought, not a design feature. These mailers are simply not as widely used as Sendmail, and therefore, they have fewer people - with both good intentions and bad - who are examining them for security problems. This fact, more than any inherent superiority, is probably the reason for the apparently greater security of these alternative mailers; it's not that the problems aren't there, it's just that not as many people know what they are.
Sendmail is the devil that everybody knows; this is both an advantage and a disadvantage. On the one hand, problems are going to be found in Sendmail because that's where lots of people are looking for them (because lots of people use Sendmail). On the other hand, what problems are found are likely to be fixed very quickly (again, because lots of people use Sendmail). Sendmail is very actively supported on security issues.
One of the reasons Sendmail has security problems is that it's a very complex program. It performs several different functions, and it requires the collection of permissions necessary to perform all of those functions. Sendmail needs root privileges for a number of reasons; for example, these privileges allow Sendmail to:
Listen on port 25 (a privileged port) for incoming SMTP connections.
Operate as a particular user to read .forward files and :include: alias files owned by that user, and to run programs specified by those files.
Execute certain kernel system calls that (in some versions of UNIX) are restricted to programs running as root, for example, to determine the amount of free disk space available to accept incoming messages.
Protect files in the mail queue (i.e., messages in transit) from snooping by unprivileged users
These root permissions can be a liability, though, when Sendmail acts as an SMTP server; an attacker who manages to exploit a bug over an SMTP connection is now talking to a process that is running as root. The process can do essentially anything on the target machine at the attacker's bidding. Sendmail tries to be careful to give up its privileges whenever it doesn't really need them, but there have still been quite a number of privilege-related bugs over the years.
On a bastion host, it should be possible to make Sendmail run setuid to something other than root. You can use an alternative SMTP server (the smap package, discussed below) for incoming SMTP connections, so that Sendmail doesn't need to listen on port 25. You shouldn't have any users on the bastion host, so you shouldn't need the ability to operate as particular users to read protected .forward and :include: files. There probably aren't any privileged system calls on your system that are critical to Sendmail's operation (though you may need to recompile Sendmail from source to prevent it from attempting to use those calls). All you're left with is the need to keep ownership of files in the mail queue consistent, and to keep nonprivileged users (which the bastion host shouldn't have anyway) from snooping on messages in transit. Creating a uid just for Sendmail, and making that uid the owner of the queue directory should solve that problem.
Each of these tasks could probably be done in more secure ways, but this would require a major redesign and reimplementation of Sendmail, and nobody has yet stepped up to accept this challenge: among other reasons, out of fear that doing so would probably introduce new problems. Instead, we keep getting patch after patch for problem after problem, so that "the current Sendmail patch" has become something of a running joke in the network security community.
Sendmail has exhibited all of the types of general mailer vulnerabilities we discussed above. Patching has eliminated or reduced most of them; for example, it used to be easy to exploit command-line bugs in Sendmail as an unprivileged user, but modern versions strictly limit the options available to unprivileged users. Given the program's past history, however, there are sure to be more problems yet to be discovered. Also, patches for old problems have sometimes introduced new problems.
An important step a firewall can take to improve security is to prevent attackers from speaking SMTP directly to Sendmail and, instead, to use a substitute server. Fortunately, this is feasible. SMTP stands for "Simple Mail Transport Protocol," and it really is simple. There are only about a half-dozen or so commands in the protocol that an SMTP server needs to implement in order to accept incoming mail.
You should consider adopting the smap package that is part of the TIS FWTK as a "wrapper" for your SMTP server (whether it is Sendmail or something else). The package includes a pair of programs called smap and smapd.
smap is a very short, simple program intended solely to handle incoming SMTP connections; unlike Sendmail, which contains about 30,000 lines of code, smap contains only about 700 lines. The relative simplicity of smap means that, unlike Sendmail, it can be easily be examined and considered in its entirety for security problems. Furthermore, it's designed with least privilege and compartmentalization in mind. The smap program runs without root privileges. It is started by inetd, which takes care of binding it to port 25 before starting it, so that smap doesn't need to run as root to do that. It runs chroot'd to a particular queue directory, and thus can't access anything outside that directory. All it does is accept incoming messages from the Internet via SMTP. It speaks the very minimum necessary set of SMTP commands, and it stores each message it receives in a separate file in the queue directory.
The result of using this substitute SMTP server is that an attacker never has a direct SMTP connection to Sendmail or any other complex SMTP server. Such a system does not protect against data-driven security holes, but such holes would be extremely hard for any firewall system to guard against. Fortunately, data-driven holes in Sendmail seem to be very rare anyway; there has only been one instance to date.
 This is covered in CERT Advisory 93:16. For information on obtaining CERT Advisories, see Appendix A.
You do give up certain capabilities by using the smap package, because smap quite intentionally handles only the minimum possible set of SMTP commands. In particular, you give up the ability to do Extended SMTP (ESMTP). ESMTP supports a number of enhancements to basic SMTP, such as better handling of MIME messages (discussed below) and messages containing eight-bit data. The basic SMTP service supports only seven-bit data, and requires that eight-bit data be converted using something like uuencode before being transmitted, which leaves the recipient with the problem of unconverting the data. This isn't a big problem right now, because only a few clients and servers currently support ESMTP and those that do have compatibility modes that let them talk regular SMTP. ESMTP is becoming more and more common, however, and this will become more of a problem as time goes by. Of course, it's always possible that the smap package will be updated to support ESMTP at some point, if it becomes a critical issue. Like many other situations involving security, by using smap you're trading off functionality for security.
Incoming mail, sender to recipient
Incoming mail, recipient to sender
Outgoing mail, sender to recipient
Outgoing mail, recipient to sender
 ACK is not set on the first packet of this type (establishing connection) but will be set on the rest.
Normally, you want to configure your packet filters to allow incoming and outgoing SMTP only between external hosts and the bastion host, and between the bastion host and your internal mail servers.
Do not allow external hosts to contact random internal hosts via SMTP. As we've discussed above, only specially configured hosts can safely accept SMTP connections.
If you cannot filter on the ACK bit, you cannot safely allow outgoing SMTP connections directly from random internal hosts, as we demonstrate in the final example in Chapter 6. If you can filter on the ACK bit, you allow internal hosts to send mail to external hosts, but there isn't much advantage in doing so. Although it shouldn't increase your vulnerability, it increases the likelihood that you're going to send misformatted mail, because the mail (mis)configurations of all your machines would be visible to the external world, and the chances that all your internal machines do all the right things with mail headers (particularly in adding fully qualified domain names to addresses and "Message-ID:" lines) are low. Sending outgoing mail via the bastion host allows the bastion host the opportunity to clean up the headers before the mail is loosed upon the world.
Because SMTP is a store-and-forward protocol, it's inherently suited to proxying. Because any SMTP server can be a proxy, it's rare to set up separate proxying for it. Instead, most sites direct SMTP connections to a bastion host running a secure SMTP server that is the proxy.
Dedicated firewall products that provide proxying may proxy SMTP (they can't reasonably be expected to run a full SMTP server). This is straightforward to configure, because SMTP uses a single connection. In this configuration, it's not unreasonable to continue to direct the proxied SMTP connections to a single secured SMTP server on a bastion host that acts as a second proxy. Proxying protects you from unwanted connections, but not from misuses of connections; you don't want to let external hosts talk to a standard unsecured SMTP server, even through a proxy.
Use DNS Mail Exchange (MX) records to specify that all your incoming mail should be directed to your bastion host(s).
 For a detailed discussion of MX records, how they work, and how to use them, see the books TCP/IP Network Administration by Craig Hunt (O'Reilly & Associates, 1992) and DNS and BIND by Paul Albitz and Cricket Liu (O'Reilly & Associates, 1992).
Configure the mailer on the bastion host to check the destination address on mail it receives. If the mail is being sent to an external host, the bastion host should process the mail as usual; if the mail is to an internal host, the bastion host should simply pass the mail to an internal mail server for processing, rather than attempt to deliver the mail itself. By passing the incoming mail to a single internal server for processing, the bastion host is relieved of having to keep track of internal aliases and internal mail configuration; this means you don't have to update the mailer configuration on the bastion host nearly as often. If the bastion host passes the incoming mail to a single internal server or small list of internal servers, the filtering system can restrict SMTP connections from the bastion host to just that host or hosts, reducing the number of internal systems that can be attacked via SMTP from the bastion host if the bastion host itself is compromised.
Configure your internal systems to send all outgoing mail to the bastion host.
You may also want to configure your mail system so that mail is sent out with a central address, instead of with the name of an individual host, as its return address. For example, you might want mail from your users to appear as email@example.com not as firstname.lastname@example.org. Because all of the incoming mail (replies to the above addresses in outgoing mail) will be going to the bastion host in any case, this doesn't remove any necessary information. It helps to guarantee that mail will go to the bastion host correctly, even if there are problems with the MX records for individual machines, and it gives more consistent information to the recipients of the mail.
Use the normal store-and-forward features of SMTP to send all incoming and outgoing mail through the bastion host.
Use packet filtering to restrict SMTP connections from external hosts to just the bastion host.
Use packet filtering to restrict SMTP connections from the bastion host to a specific internal server or set of servers.
Allow any internal system to send outgoing mail to the bastion host.
Use smap instead of Sendmail as the SMTP server on your bastion host and probably on your internal mail server as well.
Keep up to date with patches on delivery agents and user agents.
Educate your users concerning mail-based scams, such as instructions to run particular programs or to change their passwords to some specified string.
SMTP is used to exchange mail between servers. Normally, users access their mail as a file (directly, or using NFS or something similar) on the machine where it is delivered; however, there are sometimes reasons to use a separate protocol to distribute mail from a server to an individual user.
POP is a client-server protocol for handling user electronic mailboxes. With POP, a user's mailbox (the actual file where that user's incoming email is held for his later access) is kept on a server, rather than on the user's personal machine. The server is probably available to accept incoming mail more consistently than the user's personal machine is (particularly if the user's "personal machine" is a portable that he carries with him and that is only sometimes connected to the network). When the user wants his email, he accesses his mailbox using a client program (such as Eudora or Z-Mail) on his own machine using the POP protocol.
There are two major security issues involved in using POP across the Internet. First, be aware that standard POP clients and servers send the user's POP password over the Internet in the clear, so that anyone snooping on the connection can capture and reuse it later. In most cases, the POP password is the same as the user's login password, so that someone who snoops on it can get all of the user's privileges - not just the user's electronic mail. There are more secure variants of POP that support Kerberos (often called KPOP) and nonreusable passwords for authentication (often called APOP), but these secure variants are not widely available or widely supported. You may have trouble finding a combination of clients and servers that support these variants and that works for your site.
Second, regardless of the authentication issues, be sure to also consider the sensitivity of the email your users will be accessing over the Internet via POP. Whatever email your users access will be visible to anyone snooping on their POP sessions; you need to think about how sensitive email might be in your own environment. Many sites decide that, regardless of the authentication issues, their users' internal email is often too sensitive to risk being snooped on by someone monitoring their POP sessions. These sites decide to provide alternative access methods, such as dial-ups, that aren't as susceptible to snooping. If you provide your users with the ability to reach your network on the inside of the firewall (for example, with modems and PPP or SLIP), you can give them POP access while they're traveling without allowing it across the Internet.
POP is a TCP-based service. POP servers for the current version of the POP protocol (which is known as POP3 and is by far the most common version in use) use port 110. Servers for the older POP2 protocol use port 109. (POP1 was never in widespread use.) POP clients use ports above 1023.
Incoming POP connection, client to server
Incoming POP connection, server to client
Outgoing POP connection, client to server
Outgoing POP connection, server to client
 Modern POP (POP3) servers use port 110; older POP2 servers use port 109.
 ACK is not set on the first packet of this type (establishing connection) but will be set on the rest.
An outgoing POP connection would allow your users to download their mail from other sites. This is no more dangerous than allowing outgoing Telnet, and you will probably want to allow such a POP connection if there is any demand. On the other hand, POP over the Internet is rare enough that there is unlikely to be anyone interested in outgoing POP at your site, and there is no point in its use.
Incoming POP connections are those that allow people at other sites to read mail delivered for them at your site. As discussed in the previous section, you probably don't want to allow incoming POP. If you do, you should certainly limit POP connections to a POP server running on a single host. That will limit the number of vulnerable accounts and the amount of information that's being passed across the Internet. Although there are no known vulnerabilities in POP servers, if any are found, you should patch them on just one host, without worrying about all your internal hosts. Because POP requires user accounts, you don't want to run it on your normal bastion host. Although you can prevent users from logging in on POP accounts, they represent a maintenance hassle.
Figure 8.5 shows how POP works with a firewall.
POP is straightforward to proxy, because it uses a single connection. Precompiled proxy-enabled POP clients (those that work with SOCKS, for example) are not widely available. This is mostly because, although POP is used extensively within LANs, POP across the Internet is rare. UNIX POP clients are available in source form and trivial to adapt for modified-client proxying systems. It is generally more difficult to locate the source for personal-computer implementations but no more difficult to modify them.
There is no simple way to do modified-procedure proxying for connections between internal clients and external servers, or external clients and internal servers, unless all your clients are connecting to the same server. If that's the case, then you could run a generic TCP proxy program (such as the plug-gw program from the TIS FWTK) on the POP3 port on your bastion host, configured to relay all connections to the single POP server; you would then configure your clients to access the "POP server" (really the proxy program) on the bastion host.
If you do need to provide access to multiple POP servers and can ensure that all client connections from a given IP address or domain should be directed to a particular server, you could set up a more complex configuration with the plug-gw program to direct connections to the appropriate server based on where the connection request is coming from. If multiple users on the same client machine or machines need to access different POP servers across the firewall, however, there's no simple way to do it with the code that's currently available. It is possible to write a custom POP proxy server to run on the bastion host that authenticated the user, decided which server that user needed to talk to, opened a connection and authenticated the user to that server, and then sat back and played the traditional proxy server "pass-through" role; it would be difficult to make this work with nonreusable passwords, though.
Do not allow your users to transfer your site's mail over the Internet via POP, unless you can do so without revealing reusable passwords, and unless either you aren't concerned about the sensitivity of the mail itself or you have an encrypted channel to transfer it over.
If you have users who wish to transfer mail from other sites via POP, allow it via packet filtering, perhaps restricted to connections from specific sites or to specific hosts on your end.
Proxying would be an effective solution for some POP problems, but you might need to do at least minor code modifications.
Non-ASCII character sets
Nontext data such as pictures and audio segments
So-called "rich text" messages (messages containing formatted text, with different fonts and so on, rather than simple single-font unformatted text)
Multipart messages (messages containing multiple pieces, each piece in its own format).
MIME support is mostly a client issue; to mail servers and transport systems, MIME messages are generally just another message. The question is whether or not a given client can generate outgoing MIME messages, and whether or not it can recognize and cope with incoming MIME messages.
The MIME standards define certain basic data types, such as plain text, formatted text, standard audio, and so on. MIME is designed to be extensible, so that new data types can be added as necessary. MIME-capable mail clients generally understand certain data types (often only multipart messages and plain text), and rely on other programs to handle other data types (for example, graphics programs to display images, and sound programs to play audio clips). The clients generally have a list of external programs to run for particular types of data; this list can be extended or modified by the user.
Ironically, the biggest use of MIME hasn't been in mail systems, but in World Wide Web browsers. Most mail systems have little or no MIME support, but MIME is a key service upon which the World Wide Web is built. Every WWW server uses MIME to describe the format of every WWW page it hands to a client; every WWW client uses MIME to determine how to display or otherwise process the data it receives.
Because MIME is used more extensively for WWW support than for electronic mail, we discuss it below in the section on WWW, even though it is theoretically email-related. If you do have email clients that support MIME, they will be subject to the same vulnerabilities discussed in "What can a malicious server do to your clients?" in the section on the World Wide Web later in this chapter.
One difference between MIME support in email clients and WWW clients is how data is obtained. With a WWW client, the user chooses what data to access; with email, the user accesses whatever anybody sends them. In theory, email clients are more vulnerable, because you can't control what other people send you by email. In practice, however, the difference isn't that important, because it's fairly easy to lure a WWW user into accessing whatever you want them to access. Either way, you need to carefully control what data types your clients understand and how they process that data. See the full discussion of this issue in the WWW section.