This document explains what you see in firewall logs, especially what port numbers means. You can use this information to help figure out what hackers/worms are up to.
This document is intended for both security-experts maintaining corporate firewalls as well as home users of personal firewalls.
Copyright 1998-2003 by Robert Graham (email@example.com.
All rights reserved. Note that I care that people properly attribute this work but I don't care much of others profit by copying it.
All the traffic going through the firewall is part of a connection. A connection consists of the pair of IP addresses that are talking to each other, as well a pair of port numbers that identify the protocol or service. The destination port number of the first packet often indicates the type of service being connected to. When a firewall blocks a connection, it will save the destination port number to its logfile. This section describes some of the meanings of these port numbers.
Port numbers are divided into three ranges:
Where to get a more complete list of port info:
/etc/servicescontains a list of commonly used UNIX port number assignments. On Windows NT, this file is located in
|0||Commonly used to help determine the operating system. This works because on some systems, port 0 is "invalid" and will generate a different response when you connect to it vs. a normal closed port. One typical scan uses a destination IP address of 0.0.0.0 and sets the ACK bit, with broadcast at the Ethernet layer.|
|1||tcpmux||Indicates someone searching for SGI Irix machines. Irix is the only major vendor that has implemented tcpmux, and it is enabled by default on Irix machines. Irix machines ship with several default passwordless accounts, such as lp, guest, uucp, nuucp, demos, tutor, diag, EZsetup, OutOfBox, and 4Dgifts. Many administrators forget to close these accounts after installation. Therefore, hackers scan the Internet looking first for tcpmux, then these accounts. [ CA-1995-15 RFC 1078 ]|
You will see lots of these from people looking for fraggle amplifiers sent
to addresses of x.x.x.0 and x.x.x.255.
A common DoS attack is an echo-loop, where the attacker forges a UDP from one machine and sends it to the other, then both machines bounce packets off each other as fast as they can (see also chargen). [CA-96.01]
Another common thing seen is TCP connections to this port by DoubleClick. They use a product called "Resonate Global Dispatch" that connects to this port on DNS servers in order to locate the closest one.
Harvest/squid caches will send tbese UDP echoes from port 3130. To quote their document:
If the cache is configured with
This is a UNIX service that will list all the running processes on a machine
and who started them. This gives an intruder a huge amount of information
that might be used to compromise the machine, such as indicating programs
with known vulnerabilities or user accounts. It is similar the contents
that can be displayed with the UNIX "ps" command. This service is usually disabled,
scans for this don't expect to actually succeed most of the time.
Some people come here looking for ICMP port 11. To repeat: firewall logs are confusing, ICMP doesn't have ports; if you see something that says "ICMP port 11", you probably want ICMP type=11.
|19||chargen||This is a service that simply spits out characters for testing purposes. The UDP version will respond with a packet containing garbage characters whenever a UDP packet is received. On a TCP connection, it spits out a stream of garbage characters until the connection is closed. Hackers can take advantage of IP spoofing for denial of service attacks. Forging UDP packets between two chargen servers, or a chargen and echo can overload links as the two servers attempt to infinitely bounce the traffic back and forth. Likewise, the "fraggle" DoS attack broadcasts a packet destined to this port with a forged victim address, and the victim gets overloaded with all the responses. [CA-96.01]|
The most common attack you will see are hackers/crackers looking
for "open anonymous" FTP servers. These are servers with directories that
can be written to and read from. Hackers/crackers use these machines as way-points
for transferring warez (pirated programs) and pr0n (intentionally misspelled
word to avoid search engines classifying this document).
In early 2003, I occasionally see people trying to exploit the FTP server using a wide sprectrum of vulnerabilities. For example, I see them try several kinds of buffer-overflows.
SSH is a popular
way to remotely run a command-prompt on systems, primarily UNIX systems. It provides secure authentication
and encryption, so it is especially popular among security professionals. There is a commercial version
by the company that originally created it, a popular open-source OpenSSH alternative, and many other
In 2002, numerous vulnerabilities in most all versions were discovered, exploited, and routinely scanned for. Many security professionals had their boxes compromised through SSH -- in many cases, SSH was the only service they had remotely reachable.
Also note that the ssh package comes with a program called make-ssh-known-hosts that will scan a domain for ssh hosts. You will sometimes be scanned from innocent people running this utility.
UDP (rather than TCP) packets directed at this port along with port 5632 indicate a scan for pcAnywhere. The number 5632 is (hex) 0x1600, which byte-swapped is 0x0016, which is 22 decimal.
Telnet is the most popular protocol for getting a remote command line.
The most common use by scanners is to get the "banner" that prompts the user for a login name. The banner tells a lot about system -- often the attacker isn't interested in actually exploiting Telnet as to figure out more about the system when attacking other ports.
As of 2002, most attackers are interested in finding network equipment such as switches and routers, especially Cisco equipment. When my honeypot gives them a command prompt, they spend more time trying out Cisco commands than they do things like "uname" to figure out what system they are running on.
Historically (and still common as of 2002), hackers look for Unix systems with default accounts. They will try a series of logon names and empty passwords. Since Unix systems have largely fixed this problem of default accounts, this has become a less popular attack.
SMTP (Simple Mail Transfer Protocol) is the protocol that transfers virtuall all the world's e-mail.
Scans against this port are almost certain coming from spammers (and occasionally anti-spammers) looking for "open relays". An open relay is a mail server that will accept e-mail from anyone and forward it on. This allows the spammer to hide behind the relay, as well as take advantage of the fact that they can submit one e-mail with 20 recipients -- and the relay will do the job of sending copies to each recipient. This lowers the spammer's bandwidth costs.
Note that there continue to be vulnerabilities in mail servers themselves.
DNS (Domain Name Service) is a core Internet protocol; it translates names into Internet addresses (like a phonebook translates
names into phone numbers).
It is so important that when DNS servers go down, users usually think the Internet
itself has gone down.
Ways of breaking into DNS servers are frequently discovered, such as the BIND exploit in 2002. The BIND (Berkeley Internet Name Daemon) is the most popular DNS server. Many UDP packets you see rejected by the firewall are looking for the name "version.bind", which will tell the hacker what version of BIND you are (hopefully) running, and therefore which exploits they can run to break into your service. If you put a vulnerable version of BIND on the Internet, it will likely be compromised in a few days.
DNS information tells the hacker a lot of about the intended victim. Rejected TCP attempts probably reflect a desire by the hacker to do a "zone transfer", which will list all the computers in your domain. Victims often name systems in ways that help hackers figure out what is going on, such as "cisco-rtr.example.com" or "payroll.example.com".
Since DNS is such an important protocol to the Internet, firewall administrators often allow port 53 when they shouldn't. They sacrifice security in order to get ease-of-use and reliability. This allows hackers to use port 53 for protocols other than DNS. An important thing to note is that you will frequently see port 53 used as the source UDP port. Stateless firewalls frequently allow such traffic on the assumption that it is a response to a DNS query. Hackers are increasingly exploiting this to pierce firewalls.
|67 and 68||bootp|
DHCP (and the older version, BOOTP) are the protocols that assign your desktop
computer an IP address.
Firewalls will see (and reject) a lot of DHCP requests from your local network. This is an interesting problem with cable and DSL modems, because they create "virtual" local networks including people in your nearby physical neighborhood. You can identify these local requests because they are not sent to you, but are are instead to what's called the "local broadcast" address: 255.255.255.255. These machines are asking to for an address assignment from a DHCP server. You could probably hack into them by giving them such an assignment and specifying yourself as the local router, then execute a wide range of man-in-the-middle attacks. The client requests configuration on a broadcast to port 68 (bootps). The server broadcasts back the response to port 67 (bootpc). The response uses some type of broadcast because the client doesn't yet have an IP address that can be sent to.
You rarely see attackers from remote parts of the Internet trying to exploit DHCP vulnerabilities.
As of 2003, an important exploit has been found in a DHCP service, so remote hackers may start scanning for this. [CA-2003-01]
|69||TFTP||(over UDP). Many servers support this protocol in conjunction with BOOTP in order to download boot code to the system. However, they are frequently misconfigured to provide any file from the system, such as password files. They can also be used to write files to the system.|
|79||finger||Hackers are trying to:|
|80||http||Prior to 2003, I did not include an entry for this port. Presumably, you would know what port 80 meant without this guide having to tell you. However, a enormous number of worms infecting Windows and Unix systems are now using this port, so I am including it for worm discussion.|
|98||linuxconf||The utility "linuxconf" provide easy administration of Linux boxen. It includes a web-enabled interface at port 98 through an integrated HTTP server. It has had a number of security issues. Some versions are setuid root, trust the local network, create world-accessible files in /tmp, and a buffer overflow in the LANG environment variable. Also, because it contains an integrated web server, it may be vulnerable to many of the typical HTTP exploits (buffer overruns, directory traversal using ../.., etc.).|
|109||POP2||POP2 is not nearly as popular as POP3 (see below), but many servers support both (for backwards compatibility). Many of the holes that can be exploited on POP3 can also be exploited via the POP2 port on the same server.|
|110||POP3||POP3 is used by clients accessing e-mail on their servers. POP3 services have many well-known vulnerabilities. At least 20 implementations are vulnerable to a buffer overflow in the username or password exchange (meaning that hackers can break in at this stage before really logging in). There are other buffer overflows that can be executed after successfully logging in.|
|Sun RPC PortMapper/RPCBIND.
Access to portmapper is the first step in scanning a system looking
for all the RPC services enabled, such as rpc.mountd, NFS, rpc.statd,
rpc.csmd, rpc.ttybd, amd, etc. If the intruder finds the appropriate
service enabled, s/he will then run an exploit against the port
where the service is running.
Note that by putting a logging daemon, IDS, or sniffer on the wire, you can find out what programs the intruder is attempting to access in order to figure out exactly what is going on.
|This is a protocol that runs on many machines that identifies the user of a TCP connection. In standard usage this reveals a LOT of information about a machine that hackers can exploit. However, it used by a lot of services by loggers, especially FTP, POP, IMAP, SMTP, and IRC servers. In general, if you have any clients accessing these services through a firewall, you will see incoming connection attempts on this port. Note that if you block this port, clients will perceive slow connections to e-mail servers on the other side of the firewall. Many firewalls support sending back a RST on the TCP connection as part of the blocking procedure, which will stop these slow connections.|
|Network News Transfer Protocol, carries USENET
traffic. This is the port used when you have a URL like news://comp.security.firewalls.
Attempts on this port are usually by people hunting for open USENET servers.
Most ISPs restrict access to their news servers to only their customers.
Open news servers allow posting and reading from anybody, and are used
to access newsgroups blocked by someone's ISP, to post anonymously, or
to post spam.
Update: @Home has started scanning their subscribers to see if they are running USENET servers. They are doing this in order to find these servers and close them before spammers can take advantage of them.
MS RPC end-point mapper
As of 2003, the most common reason you see port 135/udp is because of WinPopup/Messenger
spam. This is a feature in Windows that allows system administrators to notify employees
of unusual events, such as the network or file servers about to be rebooted. However, spammers
have found a way to subvert this and use this mechanism to send popup messages on the victim's
Microsoft runs its DCE RPC end-point mapper for its DCOM services at this port. This has much the same functionality as port 111 for UNIX systems. Services that use DCOM and/or RPC register their location with the end-point mapper on the machine. When clients remotely connect to the machine, they query the end-point mapper to find out where the service is. Likewise, hackers can scan the machine on this port in order to find out such things as "is Exchange Server running on this machine, and which version?".
This port is often hit in order to scan for services (for example, using the "epdump" utility), but this port may also be attacked directly. Currently, there are a few denial-of-service attacks that can be directed at this port.
No RPC service except the endpoint mapper runs on this port, except that "broadcast" messages intended for other RPC services can be forwarded through this port.
|(UDP) This is the most common item seen by firewall administrators and is perfectly normal. Please read the NetBIOS section below for more details.|
File and Print Sharing
Incoming connections to this port are trying to reach
NetBIOS/SMB, the protocols used for Windows "File and Print Sharing"
as well as SAMBA. People sharing their hard disks on this port are
probably the most common vulnerability on the Internet.
Same security idea as POP3 above, numerous IMAP servers
have buffer overflows that allow compromise during the login.
Note that for awhile,
there was a Linux worm (admw0rm) that would spread by compromising port 143, so
a lot of scans on this port are actually from innocent people who have already
been compromised. IMAP exploits became popular when RedHat enabled the
service by default on its distributions. In fact, this may have been
the first widely scanned for exploit since the Morris Worm.
This port is also used for IMAP2, but that version wasn't very popular.
Several people have noted attacks from port 0 to port 143, which appears to be from some attack script.
(UDP) A very common port that intruders probe for. SNMP allows for remote management
of devices. All the configuration and performance information is stored in a database
that can be retrieved or set via SNMP. Many managers mistakeningly leave this
available on the Internet. Crackers will first attempt to use the default
passwords "public" and "private" to access the system; they may then attempt to
"crack" the password by trying all combinations.
SNMP packets may be mistakenly directed at your network. Windows machines running HP JetDirect remote management software uses SNMP, and misconfigured machines are frequent. HP OBJECT IDENTIFIERs will be seen in the packets. Newer versions of Win98 will use SNMP for name resolution; you will see packets broadcast on local subnets (cable modem, DSL) looking up sysName and other info.
In early 2002, a university in Finland released its "PROTOS" tool that demonstrated many flaws in popular SNMP implementations. These flaws had been known for more than a decade, but this was the first time security implications were shown for these flaws.
|162||SNMP trap||Probably a misconfiguration.|
|177||xdmcp||Numerous hacks may allow access to an X-Window console; it needs port 6000 open as well in order to really succeed.|
File and Print Sharing
See port 139 for more information.
In Windows 2000 and Windows XP, port 445 is essentially a duplicate of port 139. These ports are used for Micrsoft's file and printer sharing, remote registry access, named pipes services, and many MS-RPC services. The difference is that port 139 supports these services on top of NetBIOS, whereas port 445 gets rid of this middleman, supporting these services directly over TCP/IP.
Whereas many ISPs now filter port 139, many do not filter port 445. As of mid-2002, we are seeing more scans for port 445 as hackers learn to get around port 139 filters. In late 2002, we are seeing worms propogate via this port.
|513||rwho||Probably from UNIX machines on your DSL/cable-modem segment broadcasting who is logged into their servers. These people are kindly giving you really interesting information that you can use to hack into their systems.|
This is the standard protocol for remote printing on UNIX systems. Virtually
every UNIX system from Sun Solaris to Linux will listen on this port. In addition,
most laster printers support this protocol as well. There are widespread vulnerabilities
on this port, due either to vulnerabilities in the protocol itself, or vulnerabilities
in printer-specific drivers behind this port. The RedHat 7 LPRng bug was exploited by
the Ramen worm in early 2001.
As of late 2002, this is one of the more common ports probed, both because of Linux worms propogating, but also from hackers looking for well-know vulnerabilities.
|(UDP) If you are on a cable-modem or DSL VLAN, then you may see broadcasts to this port. CORBA is an object-oriented remote procedure call (RPC) system. It is highly likely that when you see these broadcasts, you can use the information to hack back into the systems generating these broadcasts. There are many exploits possible against this port, but as of August 2002, they haven't been reported to Bugtraq yet.|
See port 1524 for more info.
Some script kiddies feel they're contributing substantially to the exploit programs by making a minor change from ingreslock to pcserver in constant text... -- Alan J. Rosenthal.
|635||mountd||Linux mountd bug. This is a popular bug that people are scanning for. Most scans on this port are UDP-based, but they are increasingly TCP-based (mountd runs on both ports simultaneously). Note that mountd can run at any port (for which you must first do a portmap lookup at port 111), it's just that Linux defaulted to port 635 in much the same way that NFS universally runs at port 2049.|
|1024||-----||Many people ask the question what this port is used for. The answer is that this is the first port number in the dynamic range of ports. Many applications don't care what port they use for a network connection, so they ask the operating system to assign the "next freely available port". In point of fact, they as for port 0, but are assigned one starting with port 1024. This means the first application on your system that requests a dynamic port will be assigned port 1024. You can test this fact by booting your computer, then in one window open a Telnet session, and in another window run "netstat -a". You will see that the Telnet application has been assigned port 1024 for its end of the connection. As more applications request more and more dynamic ports, the operating system will assign increasingly higher port numbers. Again, you can watch this effect with 'netstat' as your browse the Internet with your web browser, as each web-page requires a new connection.|
|1025||-----||See port 1024.|
|1026||-----||See port 1024.|
|1027||-----||See port 1024.|
This protocol tunnels traffic through firewalls, allowing many people
behind the firewall access to the Internet through a single IP address.
In theory, it should only tunnel inside traffic out towards the
Internet. However, it is frequently misconfigured and allows
hackers/crackers to tunnel their attacks inwards, or simply bounce
through the system to other Internet machines, masking
their attacks as if they were coming from you. WinGate, a popular
Windows personal firewall, is frequently misconfigured this
In the year 2000, much activity on this port was for the purpose of connecting to IRC chatrooms. Usually the goal was DoS the chatroom. For this reason, most IRC servers will not scan your machine for SOCKS out of self-defense: they want to make sure that you are a legitimate user and now somebody who left the SOCKS service running that a hacker is tunneling through.
In the year 2003, most of this activity is now by spammers. They are looking for SOCKS servers in order to funnel spam through. This hides the original source of the spam.
There are several websites that maintain lists of open SOCKS servers. In 2002, most of the scans I see were from people who maintain these lists.
|1114||SQL||This is rarely probed by itself, but is almost always seen as part of the sscan script.|
|1243||Sub-7||Trojan Horse (TCP). See the section on SubSeven for more details.|
Microsoft runs its SQL database server on this port.
In the year 2002, several worms started exploiting this port. See section 11.2 for more information.
|1434||MS SQL Service Discovery Protocol worm||
Microsoft's SQL server uses this port for discovery of SQL services on the local LAN.
On January 26, 2003, the SQLslammer worm took down parts of the Internet in the early hours of the morning. It took advantage of a buffer overflow on this service. Administrators quickly respond by widely configuring packet filters throughout the Internet, so by the time many people woke up in the morning in the U.S., much of the problem had gone away. See section 11.3 for more info.
|Many attack scripts install a backdoor shell at this port (especially those against Sun systems via holes in sendmail and RPC services like statd, ttdbserver, and cmsd). If you've just installed your firewall and are seeing connection attempts on this port, then this may be the cause. Try telnetting to the attempted machine in order to see if it indeed comes up with a shell. Connections to port 600/pcserver also have this problem. [IN-99-04]|
|2049||NFS||The NFS program usually runs at this port. Normally, access to portmapper is needed to find which port this service runs on, but since most installations run NFS on this port, hackers/crackers can bypass portmapper and try this port directly.|
|Used by Sun Solaris boxes as a printer service, alternative to the standard printer on port 515. Exploit scripts against Solaris machines will frequently bind a shell to this port, similar to the ingreslock port. In particular, a well-known exploit against the snmpXdmid vulnerability left behind a shell on this port.|
|3128||squid||This is the default port for the "squid" HTTP proxy. An attacker scanning for this port is likely searching for a proxy server they can use to surf the Internet anonymously. You may see scans for other proxies at the same time, such as at port 8000/8001/8080/8888. Another cause of scans at this port, for a similar reason, is when users enter chatrooms. Others users (or the servers themselves) will attempt to check this port to see if the user's machines supports proxying. See section 5.3 for more info.|
|5632||pcAnywhere||You may see lots of these, depending on the sort of segment you are on. When a user opens pcAnywhere, it scans the local Class C range looking for potential agents. Hackers/crackers also scan looking for open machines, so look at the source address to see which it is. Some scans for pcAnywhere frequently also include a UDP packet to port 22. See dialup probes for more info.|
|6776||Sub7 artifact||This port is used separately from the SubSeven main port to transfer data. One example where you might see this is when a master is controling a slave on a dialup line, then the slave machine hangs up. Therefore, when someone else dials-in at that IP address, they will see a continuous stream of connection attempts at this port. more on dialups|
|6970||RealAudio||Clients receive incoming audio streams from servers on UDP ports in the range 6970-7170. This is setup by the outgoing control connection on TCP port 7070.|
|13223||PowWow||The "PowWow" chat program from Tribal Voice. It allows users to open up private chat connections with each other on this port. The program is very aggressive at trying to establish the connection and will "camp" on the TCP port waiting for a response. This causes a connection attempt at regular intervals like a heartbeat. This can be seen by dial-up users who inherit IP addresses from somebody who was chatting with other people: it will appear as if many different people are probing that port. The protocol uses the letters "OPNG" as the first four bytes of its connection attempt. more|
Outbound: This is seen on outbound connections. It is caused
by users inside the corporation who have installed shareware
programs using the Conducent "adbot" wrapper. This wrapper
shows advertisements to users of the shareware. A popular shareware
program that uses this is PKware.
Bill Royds mentions that in his experience, you can block this
outbound connection with no problem, but if you block the
IP addresses themselves, then the adbots can overload the link
trying to reach the servers by continually connecting many times
The machines will attempt to resolve the DNS name "ads.conducent.com", which resolve to the IP addresses:
|27374||Sub-7||Trojan Horse (TCP).
See the section on SubSeven for more details.
Also used as a backdoor port left behind by exploit scripts, such as those in the Ramen worm. While some scans for this port may be due to SubSeven, others may be looking for a remote shell.
|30100||NetSphere||Trojan Horse (TCP). This is a commonly seen scan looking for systems compromised by this trojan.|
|This number means "elite" in hacker/cracker spelling (3=E, 1=L, 7=T). Lots of hacker/cracker backdoors run at this port, but the most important is Back Orifice. At one time, this was by far the most popular scan on the Internet. These days, it's popularity is waning and other remote access trojans are becoming popular.|
|31789||Hack-a-tack||UDP traffic on this port is currently being seen due to the "Hack-a-tack" RAT (Remote Access Trojan). This trojan includes a built-in scanner that scans from port 31790, so any packets FROM 31789 TO 317890 indicate a possible intrusion. (Port 31789 is the control connection; port 31790 is the file transfer connection).|
|32770 ~ 32900||RPC services||Sun Solaris puts most of its RPC services in this range. In particular, older versions of Solaris (pre-2.5.1) put a portmapper in this range, allowing hackers access to this even when low ports are blocked by a firewall. Probes in this range might either be for this portmapper, or for known RPC services that can be exploited.|
|33434 - 33600||traceroute||If you see a series of UDP packets within this port range (and only within thisrange), then it is probably indicative of traceroute. See traceroute for more info.|
|41508||Inoculan||Inoculan on UDP. Older versions of Inoculan apparently generate huge quantities of UDP traffic directed at subnets in order to discover each other. More info can be found at http://www.circlemud.org/~jelson/software/udpsend.html and http://www.ccd.bnl.gov/nss/tips/inoculan/index.html. Thanks to Jerry Leslie, NeoNET < leslie at clio dot rice dot edu>|
Ports 1-1024 are for reserved services, and almost never appear as the source. There are some exceptions, such as when connections come from NAT machines. See section 1.9 for some more details.
Ports closely after 1024 (i.e. 1024-5000) are the ones most commonly seen. These are the "dynamic" range that are assigned to applications that don't care what port they use for their connection.
|1-5/tcp||dynamic||FTP||Ports 1-5 are indicative of a script called 'sscan'|
|20/tcp||dynamic||FTP||FTP servers usually transfer files from this port.|
|53||dynamic||FTP||DNS servers will send UDP responses from this port. You may also see TCP connections with source/destination ports of 53.|
|123||dynamic||S/NTP||The (Simple) Network Time Protocol (S/NTP) servers run at this port. They will also send broadcasts to this port.|
|27910-27961/udp||dynamic||Quake games||Quake (and Quake-derived games) usually run servers at these ports. Therefore, UDP packet from this range (and to this range) will usually be games.|
|61000+||dynamic||FTP||Ports above 61000 might come from machines behind a Linux NAT server called "IP Masquerade".|
This is due to a "decoy" scan, such as in 'nmap'. One of them is the attacker; the others are not.
Forensics and protocol analysis can be used to track down who this is. For example, if you ping each of the systems, you can match up the TTL fields in those responses with the connection attempts. This will at least point a finger at a decoy scan. (The TTLs should match; if not, then they are being spoofed). [Newer versions of scanner now randomize the attackers own TTL, making it harder to weed them out].
You can also attempt to go back further in your logs, looking for all the decoy addresses or people from the same subnets. You will often see that the attacker has actually connected to you recently, while the decoyed addresses haven't.
The first stage of a Trojan Horse attack is to get the program on a user's machine. Typical techniques are:
The next stage of the attack is to scan the Internet looking for machines that might be compromised. The problem is that most of the techniques outlined above don't tell the cracker/hacker where their victim machine is. Therefore, the cracker/hacker must scan the Internet looking for the machines they might have compromised.
This leads the condition where owners of firewalls (including personal firewalls) regularly see "probes" directed at their machines from crackers/hackers looking for these machines. However, if the machine hasn't been compromised, then these probes are not a problem. The probes cannot compromise the machine by themselves. Administrators can usually ignore these "attacks".
Typical ports used by these probes are listed below. In order to tell if your machine might be running one of these trojans, run the program "netstat -an" on your machine. Look for the ports that might be "listening" for incoming connections.
|31337||BackOrifice, and many others|
|50505||Sockets de Troie|
In short, it not only is an excellent hacking tool, the little "magic" tricks are designed to scare the <bleep> out of victims.
Sub7 is written by a hacker who calls himself "Mobman". His site can be reached at http://subseven.slak.org/.
Sub7 might use the following ports:
Q: My filters reject incoming packets with source ports below 1024, so the DNS lookups are failing.
A: Don't filter that way. Lots of firewalls have similar rules, but this is somewhat "misguided" since hackers/crackers can forge whatever ports they want.
Q: Are these NAT firewalls doing it incorrectly?
A: Not in theory, but in practice it will result in failures. The "correct" way would be more strictly control DNS traffic in any case (such as essentially "proxying" DNS and forcing out through port 53).
Q: I thought DNS lookup was supposed to use a random source port above 1024?
A: In practice, your average DNS client will use a non-reserved port. However, a lot of implementations use a source port of 53. In any case, the NAT issue is completely separate because it completely changes the entire 'socket' (IP address + port combo).
One of the most popular applications is "chat", like IRC. One feature of chat programs is that they reveal the IP address of the people you are chatting with. One problem with chatrooms is that people enter the rooms "anonymously" and play around, either by disrupting conversations with offtopic comments and flamebait, or by "flooding" the servers or other clients in an attempt to kicked them off.
Therefore, both servers and clients are implementing measures to stop "anonymous" use of chatrooms. In particular, they check people entering chatrooms in order to see if they are "proxying" through some other connection. The most popular of such probes is SOCKS. The assumption is that if the IP address of where you are coming from supports SOCKS, then it is possible that you have a completely separate machine and are only going through the indicated machine in order to hide your true identity. Undernet's policy on this can be found at http://help.undernet.org/proxyscan.
At the same time, crackers/hackers will scan people's machines in order to determine if they are running some sort of server that can be bounced through. Again, by checking for SOCKS, the attacker hopes to find somebody that has left SOCKS open, such as a home user implementing connection sharing using SOCKS, but accidentally configured it so that anybody on the Internet has access to it.
Remapping is done under the theory that making the port harder to find will make it more difficult for a hacker to exploit. Instead of simply exploiting a well-known service at a well-known port, the hacker will have to port scan the machine.
Most port remapping is done at some variation of the original port. Therefore, most HTTP ports are based upon a variation of the theme "80": 81, 88, 8000, 8080, 8888, and so forth. POP, which is originally at port 110 can often be found at port 1100.
There are other statistically significant chosen numbers, like 12345, 23456, 34567, etc. Many people also choose numbers that are well known for other reasons; 42, 69, 666, 31337, and so on. The recent proliferation of Remote Access Trojans (RATs) has resulted in hackers/crackers choosing the same defaults for their programs. For example, NetBus defaults to port 12345.
Blake R. Swopes points out that remapping is also done because on UNIX machines, your server needs root privileges to listen on ports below 1024. If you don't have root level access and want to run a web service, you will need to install it on a high-numbered port. Likewise, some ISPs might firewall low-numbered ports, forcing you to remap even when you own the entire machine.
netcat -L -p 1234A lot of protocols will send data as the first part of the connection. By setting up netcat listening on the port, you might be able to figure out what protocol that are using. If you are lucky, the protocol in question will be HTTP, which will give you a wealth of information that you can use to track down what is happening.
The "-L" option means to listen continuously. Normally, netcat would accept a single connection, dump the contents, then exit. By adding this option, it will remain running for multiple connections.
Some firewalls (inaccurately) label ICMP fields as "ports". ICMP has no ports like TCP or UDP, but it does have two fields called "type" and "code". While these fields serve completely unrelated purposes, the fact that there are two of them have led to firewalls mislabeling them. Thus, you will see firewall logs that look like:
REJECT proto=1 from=10.2.3.4:8 to=192.0.2.123:0In reality, it should be represented like:
REJECT proto=1 icmptype=8 icmpcode=0 from=10.2.3.4 to=192.0.2.123
For more on ICMP, please read my Infosec Lexicon entry on ICMP .
The official reference for what ICMP Type/Code fields mean is found at http://www.isi.edu/in-notes/iana/assignments/icmp-parameters. While that document describes the official meanings, this section describes what hackers are trying to do. This section contains a brief summary at top, then more details descriptions down below.
A response to a ping.
An indication back from a host or router that some packet did not reach its destination.
Route configuration problem or incorrectly specified IP address.
|1||Host Unreachable||It means that the router one hop before the desired host could not ARP the host.|
The server tells the client that nobody is listening at the port the
client attempted to contact.
|4||Fragmentation Needed but DF set||
Important: If you are seeing these in your firewall reject logs, then you've
misconfigured your firewall. You should allow this packet to pass through,
otherwise your clients will see their TCP connections mysteriously hang.
Congestion on the Internet.
|5||*||Redirect||Somebody is trying to redirect your default router. This could be from a hacker trying to execute a man-in-the-middle against you by causing you to route through their own machine.|
|9||*||Router Advertisement||There is exists a hack against Win9x and Solaris such that a hacker can DoS you by redirecting your default router. A neighboring hacker can also do a man-in-the-middle attack by directing you through his/her router.|
|11||*||Time Exceeded In Transit||It means that a packet never reached its target because something timed out.|
Router dropped the packet either because of a routing loop or maybe because of a traceroute.
|1||Fragment reassembly timeout||
The host dropped the packet because it didn't receive all the fragments.
Something unusual is going on, and probably indicates an attack.
Note that Unreachables sometimes play a part in defeating SYN floods. This means that if a host you are talking to is under SYN flood attack, you will not be able to reach them if you block incoming Unreachables.
In some cases, you will receive destination unreachable packets from hosts you have never heard of. The most common cause of this is a "decoy scan". An attacker is sending spoofed packets a target using possibly hundreds of source addresses, including one that is the real address. The hacker's theory is that the victim won't wade through all the decoys in order to pin them down.
The best way to solve this is to examine the actual packets as described below. Try to discover is the pattern looks like what one would see in a decoy scan. For example, look for alternating port numbers in TCP or UDP headers contained within the ICMP portion of the packet.
This packet is sent by a SERVER when a CLIENT tries to connect to a UDP port that isn't running. For example, if you try to send an SNMP packet to port 161, but the machine doesn't support the SNMP service, you will get back an ICMP Destination Port Unreachable packet.
The first thing to debug this problem is to check the port numbers within the packet. You probably need to use a sniffing utility as firewalls tend not to log the information. This technique relies upon the fact that ICMP messages include the IP and UDP headers of the original packet. Here is a hex dump of an ICMP unreachable:
00 00 BA 5E BA 11 00 60 97 07 C0 FF 08 00 45 00 00 38 6F DF 00 00 80 01 B4 12 0A 00 01 0B 0A 00 01 C9 03 03 C2 D2 00 00 00 00 45 00 00 47 07 F0 00 00 80 11 1B E3 0A 00 01 C9 0A 00 01 0B 08 A7 79 19 00 33 B8 36Where the bytes 03 03 are the type/code for the ICMP packet. The last 8 bytes of the packet are the original UDP header, which decodes as:
Here are some reasons why you may be seeing this:
Why? Both IP and TCP fragment data, but in different ways. TCP is vastly more efficient at fragmentation than IP. Therefore, stacks attempt to find the "Path MTU (Maximum Transmission Unit)". This ICMP message is sent during that process.
Let's consider ALICE talking to BOB. Both are on Ethernets (max frame size = 1500 bytes), but some intervening link limits the maximum IP packet size to 600 bytes. This means all IP packets sent will be fragmented by the routers on that link into 3 fragments. Since it is much more efficient to fragment at the TCP layer, the TCP stack will attempt to discover the MTU. It does this by setting the "DF" (Don't Fragment) bit in all its packets. As soon as it hits a router than cannot forward a packet that large, the router will send back this ICMP error message. From that, the TCP stack will know how to fragment correctly.
You should probably let these packets through the firewall. Otherwise, the intended recipient will have a hung connection as small packets get through to set up the connection, but the large packets are mysteriously dropped. A common result from this are people who see web pages that are only halfway returned.
Path MTU Discovery is becoming more and more integrated into communication. For example, IPsec needs this functionality.
In general, the rules for source quenches are now (RFC 1122):
However, hosts still react to Source Quenches by slowing communication, so they can be used as a denial of service. Firewalls should filter these out. If a DoS is suspected, the source address of the packets will be meaningless, because the IP addresses are spoofed.
Source quenches have been known to be sent by some SMTP servers.
These are ping request packets. They are used all over the place; it may indicate hostile intent of someone trying to scan your computer, but it may be part of the normal network functionality. See Type = 0 (Echo Response) above for more info.
Lots of network management "scanners" will precede a scan using a special ping packet. These include ISS scanner, WhatsUp monitor, and others. This will be visible in the payload of the scanner. Most firewalls don't log this payload, so you may need to use some sort of sniffer to capture them or some time of Intrusion Detection System to flag them.
Note that blocking incoming PINGs does not mean a hacker can't scan the network. There are many other ways of doing this. For example, TCP ACK scanning becoming popular -- they usually get through the firewall, and they illicit a response from the target system.
Pings sent to broadcast IP addresses like x.x.x.0 or x.x.x.255 are probably attempts to use your network as a smurf amplifier.
Another common reason firewall administrators see this is due to routing loops developing in the Internet. Route flapping (constant route changes) is a common problem, and will often briefly result in a loop. This means that while a IP packet is heading towards it destination, the packet gets misrouted to a router that it previously visited it. The packet then gets routed in a circle infinitely -- or it would be, if the routers didn't decrement the TTL field each time and discard the packet once that value hit zero.
Another cause of this is distance. Many machines start with a default TTL of 127 (Windows) or even lower. Routers will often decrement the TTL more than by one in order to reflect slow lines like dialups or transcontinental links. Therefore, a site might not be reachable with a low initial TTL. In addition, some hackers/crackers like to make their site unreachable through this method.
There are a couple of network management uses of this packet, such as testing to see if two computers can talk to each other. A network manager at point A may send a packet to B through point C. This tells A if B & C can talk to each other.
The same technique can be used to evade firewalls, subvert trust relationships, and communicate with machines using "private" address (10.x.x.x, 192.168.x.x, 172.[16-31].x.x).
Let's say you are a hacker/cracker on the Internet and you want to talk to some machines behind a firewall who use 10.x.x.x as their IP addresses. Since the routers on the Internet do not know where this subnet is located, they will drop your packets. However, you put a loose source route option in the IP packet and tell all the Internet routers to first forward to the firewall. Since the firewall straddles both the Internet and the private network, it will know how to forward the packet appropriately. Thus, you can carry on a conversation with the victim by bouncing all packets through the firewall.
This can be used with IP spoofing. You pretend to be a router (like the firewall mentioned above) and pretend that somebody else is bouncing packets through you. Thus, pick some random machine on the Internet (ALICE) as the spoofee, then send packets from ALICE to your victim BOB. BOB will think the packets are coming from ALICE, but in reality they are coming from you. This masks the real source of the attack.
This is even better if you know that BOB trusts ALICE. IP addresses are often used as part of authentication. Let's say the firewall has a rule allowing all traffic from ALICE into the network. By forging all IP packets to be from ALICE (but being source routed through your own machine), then you get free access to the victim network.
More and more core Internet routers are disabling source routed packets. They slow down routing anyway, but they are a huge security risk. There is also no real need for them. Managers should do the same and disable source routing everywhere: on firewalls, on routers, and even on end-nodes so that they won't even accept incoming source routed packets.
See Microsoft Knowledge Base article Q217336 for setting the "DisableIPSourceRouting" on WinNT SP5 systems
This is happening a lot these days as more and more people use DSL or cable-modem connections. The reason is that unlike point-to-point connections (like T-1, frame relay, etc.), these new high-speed technologies drop you onto an ATM VLAN, which is a single broadcast domains. In fact, many cable-modem users are seeing multiple megabytes of traffic per day simply from such broadcasts.
You must remember that such packets MUST be local. Routers (generally) refuse to forward packets with the IP address of 255.255.255.255. This address is known as a "local broadcast" for this reason: it never travels past the local segment (or these days, the local "virtual" segment).
What are these packets for?
Check the list of ports at the top of this document. If it is not listed there, then the only way to figure this out is to capture them with a sniffer and view their contents.
For example, a common service that runs with a random port number is CORBA IIOP packets. Many services run at port 535, but it is frequently reconfigured to broadcast on other ports. If you look at the hex dump in the sniffer, you will see the letters "IIOP" somewhere in the contents.
In any case, this is rarely something to be concerned about. In fact, it advertises something about the person sending the traffic that can be used to hack them. Hackers rarely attack their own neighborhoods (because it is easy to detect), so it probably is accidental, not malicious.
It should be noted that with today's ATM networks, the source of the broadcast may not even be in the same state as you are; they may be hundreds of miles away. The word "local" means in terms of the network topology, not distance.
Remember that IP addresses can be spoofed, so that the "owner" of an IP address may be innocent. Increasingly, attacks are coming from compromised machines. The owner of the IP may actually be grateful! Both of these statements come to the same conclusion: be polite and professional.
Many companies have established the e-mail address "firstname.lastname@example.org" (replace "example" with the proper company). This e-mail role is for both e-mail abuse (such as spam) as well as for network abuse. When you find the owner of the IP address, you should probably compose a message including the evidence of the attack.
In the past, all the IP address owners were kept by the Internic. A database built from that information is at http://ipindex.dragonstar.net/. There are now 3 official registrars for North America, Asia, and Europe. Unfortunately, you will have to query each individual database. However, if you start with the North America registrar, it will tell you if the address belongs to one of the other three. Warning: The returned information is fragile; so don't send flames to these people because you have only about 30% chance of reaching the right people.
ARIN (American Registry for Internet Numbers)
RIPE (Reseaux IP Europeens)
|Asia and Pacific||http://www.apnic.net/apnic-bin/whois.pl
APNIC (Asia Pacific Network Information Centre)
Running traceroute will often find at least the ISP who is hosting the IP address. A reverse DNS lookup on the actual IP address is easy to spoof, but the route to the machine will reveal who is hosting the possible intruder.
Common IP addresses
Many attacks are now coming from cable-modem subscribers in the 24.x.x.x range. These are probably from machines who have been compromised by a Remote Access Trojan (RAT). (While hackers/crackers frequently use dial-up lines because they don't care if their account gets canceled, few users want to have their cable-modem accounts canceled).
Another important range is the "private address" ranges of 10.x.x.x, 192.168.x.x, and 172.16.x.x-172.31.x.x. See 3.4 below.
Addresses like 127.x.x.x indicate "localhost" and should never be seen on the Internet.
The address range 192.0.2.x has been designated for "examples", like "example.com".
The "private address" ranges are 10.x.x.x, 192.168.x.x, and 172.16.x.x-172.31.x.x. In theory, these cannot be routed on the Internet, which means that while you can use them internally to your network, they cannot leave your network. This further means that you shouldn't see them enter your network from the outside world -- but what does it mean when they do?
I've been seeing these in these cases:
The most common reason you get these is because hackers are scanning your network for "smurf amplifiers". The "smurf" (and related "fraggle") takes advantage of the amplification effect when everyone on a subnet responds to a directed broadcast. The attacker spoofs a victim's IP source, and when your subnet receives that packet, all of you respond to the victim. This allows the attacker to send one packet outbound, but cause his victim to receive many packets inbound.
This may also just be normal scans and worms because they don't care about the difference between a directed broadcast and normal address.
Note that while directed broadcast addresses usually end in .255, this is not guaranteed. It depends upon how you've structured your subnet with your address mask.
Once a DHCP Client has determined it must auto-configure an IP address, it chooses an address. The algorithm for choosing an address is implementation dependant. The address range to use MUST be "169.254/16", which is registered with the IANA as the LINKLOCAL net.
This only happens when the normal DHCP process fails.
This new technique was introduced with Microsoft Win98 and Apple MacOS 8.5.
Also see: http://www.unixreview.com/archives/articles/1999/july/9907dd.shtml
This is because the POP and SMTP servers are trying to establish an identd/AUTH connection back to the client. These reverse-connections are blocked, and it takes a while before the servers timeout and continue.
The identd/AUTH service identifies the user of the TCP connection (user name, process id, etc.). When the e-mail server accepts the incoming TCP connection, before sending the greetings, it will first attempt to gather information via the identd protocol. This consists of a TCP connection in the reverse direction. In other words, when I connect to my e-mail server, my e-mail server attempts to connect back to me on port 113, the identd port. My e-mail connection just sits there until the e-mail server resolves the identd information.
The problem comes about because the firewall silently drops the SYN packet. The e-mail server is expecting an immediate SYN-ACK (identd supported) or RST (identd not supported), but when the firewall drops the packet it keeps trying until the connection times out.
Note that the e-mail server doesn't care if I don't support identd, and indeed most people don't on their clients. It just wants an immediate response one way or the other. The firewall blocks that. This is why some personal firewalls for Windows (like BlackICE Defender from my company) contain default rules that allow identd/AUTH to pass through. Windows doesn't reveal the information that UNIX does, and opening it up gives the immediate response these servers are looking for.
To solve this problem:
Note that this means you should be seeing lots of dropped incoming connection attempts at port 113 in your log files because of this.
The program "traceroute" is based upon a very intelligent hack by Van Jacobson (also famous for other nifty kludges). Every IP packet has a time-to-live (TTL) field that indicates how many hops the packet can travel before being dropped. This field is needed because routers sometimes get misconfigured and will forward packets in a continuous: i.e. Alice forwards the packet to Bob who forwards it to Charlene who mistakenly forwards it back to Alice.
Therefore, each router decrements (subtracts 1) from the TTL field. When each reaches zero, the router who currently has the packet will simply "drop" it (not forward it on). When a router drops a packet, it sends a message back to the sender informing for this. This message is called an ICMP "TLL Exceeded in Transit".
The nifty thing about this is that the router uses its own IP address as the source address of the ICMP message. Therefore, if you send a packet to a target but with a TTL of only 1, the first router will receive the packet, decrement the field to 0, drop it, then send back the ICMP notification. This informs you of the first router along the route (which you probably knew anyway).
The same goes for an initial TTL of 2. The first router gets it, decrements to 1, then forwards to the second router along the route. This router then decrements to 0, drops the packet, and sends back and error ICMP message.
By continuing this process, you eventually end up with the list of routers between yourself and the target.
Versions of traceroute
There are various versions of the traceroute program. In particular, the Windows program "tracert.exe" uses pings as the packet it sends to the target. Therefore, you might see ICMP Echoes on your firewall.
The most popular "traceroute" program for UNIX programs sends UDP datagrams to port 33434 for the first packet sent, then increases this port number by one for each successive packet. This means that you will never see port 33434 on your firewall, but you will start to see successive ones starting at higher port numbers. Traceroute programs typically send 3 packets for each hop (in case some get dropped). Therefore, if somebody is 10 hops away, the first port you will see is 33434 + 3*10 = 33464.
Firewall administrators should learn the symptoms of traceroute activity.
Some traceroutes are designed to bypass firewalls. See http://www.packetfactory.net/Projects/Firewalk/firewalk-final.html for more information.
The 'sscan' tool has become a popular scanning tool on the Internet. It not only "port scans" but attempts to discover some common vulnerabilities. There are several versions of sscan, and it is very configurable, so matching an exact signature to this program may be difficult. The 'sscan' program is derived from the older 'mscan' tool.
A sscan goes through several phases:
The following is a record pulled from an intrusion detection system.
ports=1 22 23 25 53 79 110 111 143 1114 2766 6000 31337
Unfortunately, the system consolidates alerts, discards duplicates, and keeps the port numbers in sort order. In a real scan, several of the ports would have duplicate connection attempts, and port 1/tcpmux would be one of the last probes, not one of the first.
In late summer of 1999, probes for ports 80/8080/3128 were particularly noticed. These came from all over the Internet and were fairly disjoint. These came from a Trojan Horse called "Ring0" (RingZero). It would infect PCs, then scan random IP addresses for proxy servers. The SANS Institute (a security training/conference organization) coordinated an effort to track down exactly what was happening from reports from many of their customers. A common symptom of this Trojan is 3 probes spaced within a minute from the same IP address from this Trojan. More information can be found at: http://www.sans.org/newlook/resources/ringzero.htm. A news article by CMP can be found at: http://www.techweb.com/wire/story/TWB19991013S0018
A list of open proxies can be found at: http://freebooks.hypermart.net/proxy/proxies.htm
Ports with variations of the "80" them (81, 88, 8000, 8080, 8888, etc) are most commonly used for proxies. In addition, a popular free proxy server called "squid" runs at port 3128.
A smurf is a ping (ICMP Echo Request) whereas a fraggle is a UDP port 7/echo. These are named after the programs/scripts that first implemented them.
These packets are sent to broadcast addresses. In IP, a directed broadcast has all the "host" bits set to either one or zero. This means an address that looks something like 192.0.2.0 or 192.0.2.255 is likely a broadcast. The key thing to remember is that such addresses are only broadcasts if the router on that subnet chooses to interpret it as a broadcast. If that router has this configured as a broadcast in its routing tables, it will forward the single IP packet as broadcast on that (Ethernet) segment, causing all systems on that (Ethernet) segment to receive the packet.
Therefore, there are two configuration problems:
Q: Why are these only aimed at strategic points like broadcast addresses?
A: Because if a single packet is sent to a broadcast, then it generates lots of responses to the spoofed address of the victim.
Q: I monitor multiple networks. Why is only this network being attacked this way?
A: Your network isn't being attacked; instead it is the third party in a fraggle attack. Your network is being used to attack somebody else (the source address of the packets, which is spoofed). Either your other networks aren't nearly as effective as fraggle amplifiers, or they have been registered in smurf/fraggle registries yet. Hackers rarely look for their own amplifiers, but instead simply look up good amplifiers in such directories. If you get registered, then multiple hackers will use/abuse your network.
Q: Why port UDP 7 only?
A: There are a number of reasons. The first is that script-kiddies aren't too bright. If they only scripts available use port 7, then that is all they can use. Secondly, the service has to respond to broadcast requests. Therefore, you cannot use TCP (which will only respond to directed queries). Many other UDP services only respond to directed queries. Finally, when fraggle was first developed, many firewalls allowed Echos to pass through (because they were used for performance monitoring). More dangerous protocols like NetBIOS (port 137) are already blocked by firewalls.
Remember that DNS servers will "recursively" send out queries when resolving names on behalf of clients. Each outgoing request is given a unique transaction identifier; incoming responses contain the same transaction identifier.
Therefore, if a server sends request #45689 to server 192.0.2.131, but gets response #45689 back from server 192.0.2.3, then it triggers this alert.
The most common cause of this is due to proxying, caching, and dual-homed hosts. For example, the DNS server might have two IP addresses: [192.0.2.131] and [192.0.2.3]. The typical way of writing a DNS server is to not bind the sockets to individual IP addresses. What this means is that the DNS server does not know which IP address the request was received on, nor does it tell the underlying TCP/IP stack which IP address to use when sending the response. Therefore, when the DNS server sends the response, the underlying stack uses one of the IP addresses at random (which can be the wrong one).
May 12 04:33:01 ns1 named: unapproved query from [192.0.2.71].35687 for "VERSION.BIND"
14:03:00 192.0.2.243 GET /index.html - 200 Mozilla/4.0 - - 14:03:03 192.0.2.243 GET http://www.example.com/ - 200 - - -The first is a normal line, but what is that complete URL starting with "HTTP"? This is an attempt to see if the machine supports proxying. This is how pretty much all HTTP proxies work -- they receive a complete URL, then fetch that URL for the user.
See section 5.3 for more info.
As always, these attempts are usually from scans against thousands/millions of machines rather than against you in particular. Every few months, a new exploit script is published for Linux or Solaris services, and script kiddies start scanning the Internet for that service. Most of the vulnerabilities in the services listed are buffer overflows.
Note that on Sun Solaris machines, these services usually have port numbers in the range starting at port 32770. Many other times, RPC services will have ports below 1024, on the assumption that it provides a little better security because
More info on RPC can be found in RFC1833.txt.
|0||NULL||This is a "ping" style command -- it just verifies that the service is running. You see these almost never.|
|1||SET||If you see this go across the wire, then it is an intrusion attempt. This should be used only internally as RPC-based programs register themselves with portmapper.|
|2||UNSET||If you see this go across the wire, then it is an intrusion attempt. This should be used only internally as RPC-based programs unregister themselves with portmapper. It is sometimes used as a DoS attack in order to kill your services. Such attacks are frequently spoofed.|
|3||GETPORT||This is the normal use of portmapper that you should see 99.9% of the time going across the wire. An external client looks up the corresponding port number for the desired service. When reviewing logs, if you see requests to strange services, you can lookup the program number in the table below.|
|4||DUMP||This dumps all the mappings in the portmapper database. The UNIX command "rpcinfo -p" carries out this command. This is a common reconnaissance technique for hackers.|
|5||CALLIT||This may be an attempt to compromise the system. The callit feature was created for RPC broadcasts. Because a desired service runs on different ports on different systems, one cannot simply broadcast to it. Therefore, portmapper will accept incoming broadcasts on port 111, then forward them to the appropriate program. However, some even protocols that don't support broadcasts can be compromised by sending the requests through this service.|
I've put an astrisk * next to the ones that have been seen to use the callit feature.
|Allows CPU, network traffic, and disk statistics to be remotely monitored. Hackers may use this as part of recon.|
|Lists the users on a machine, which reveals lots of info to hackers.|
|100005||NFS mountd||In late 1998, the RedHat Linux distribution contained a buffer overflow bug in the mountd service running at port 635. The popularity of RedHat and the fact that the service ran at a common port number resulting in popularity among hackers. Not only did hackers scour the Internet for such machines, but a worm was created to spread via this service. [CA-1998-12]|
|The program walld, which sends messages to users from the system administrator (such as notifying them the system is about to be rebooted, so they had better save their work). Messages are frequently sent via callit broadcasts.|
Solaris Calender Messaging Service
In the middle of 1999, a buffer-overflow was found in this service. Immediately after this discovering, hackers started doing extensive scans for this service, resulting in thousands of hacks against web-sites using Solaris. [CA-1999-08]
|100083||ToolTalk||ToolTalk (rpc.ttdbserverd), a CDE service allowing inter-application communication on a Unix desktop. This is enabled by default on Unix workstations (Solaris, HP-UX, AIX, SGI, etc.), and new holes in this services are constantly being discovered. [ CA-1998-11 CA-1999-11 CA-2001-27 CA-2002-29 CA-2002-26 ]|
|100232||rpc.sadmind||Sun Solstice Adminsuite, installed by default on Solaris systems 2.5 and above (2.4 and below installed a similar service called rpc.admind). [CA-1999-16]|
|100249||snmpXdmi||The snmpXdmi is an "SNMP" to "DMI" (Desktop Management Interface) translator. [CA-2001-05]|
In late 1999, a buffer overflow bug was found in the logging service. While any code based upon the original BSD sources is vulnerable, hackers are probably scanning for the Linux implementation includes in many distros. [CA-1999-12]
|I'm not sure what this service is, but UnixWare sends callit broadcasts across this program number.|
|This number has been assigned to FrameMaker for UNIX. You can download an evaluation copy of this program at: http://www.adobe.com/support/downloads/fmunix.htm. Apparently, the license manager supports callit broadcasts. This license manager supports a "roving" license whereby many people can have it installed, but only a few can use the product.|
|Legato NetWorker Server Remote Status. This is a backup service (also OEMed as Solstice Backup). Status updates are broadcast via callit.|
|Part of NeXTstep replacement for YP. When a child netinfod process starts up, it searches for a parent by broadcasting a NIBIND_BIND procedure (function=8) on the local subnet.|
The problem is that many administrators simply install servers without taking these simple precautions. Spammers take advantage of this fact. They give a single e-mail to the mail server and a recipient list containing hundreds of unrelated recipients. This allows them to send huge quantities of e-mail using a slow dialup connection. This is important because once the ISPs get enough complaints, they will terminate the user's account, so they must continual get new dialup connections. It also has the effect of partially hiding the true source of the spam.
If you get error messages about relaying, that is a good thing: you've configured your server correctly. If you don't get such messages, this is a bad thing. This means that you are probably not rejecting relayed messages. Has your server seemed slow lately?
Not only do spammers hunt for open relays, anti-spam organizations do the same in an attempt to "blacklist" open relays. Some of the good guys are:
Not only do you receive relay attempts from spammers, you also get attempts from anti-spam organizations. There are several organizations that regularly scan the Internet looking for open relays. The most common is from "manawatu.co.nz"; don't get too upset -- they
If you do the command "VRFY root", you might be able to find out the postmaster's e-mail address. This is good reconnaissance technique.
By doing a "VRFY decode" or "VRFY uudecode", you might be able to find out some security holes in the system related to these subsystems. Other commonly scanned user names are "bbs", "lp", "demo", "guest", and "debug".
Some systems have buffer overflows in this command, either in the command itself or in the logging system behind the command. You might see entries for very long strings like "xxxxxxxxxxxxxxxxxxxxxxxx".
If you see a bunch of these in a row, you are probably being scanned by a vulnerability scanner (ISS/CyberCop/Nessus). They will generate a bunch of other junk in your logs as well.
The absolute minimum ICMP traffic to allow is the packets dealing with TCP path MTU discovery. Fragmenting a stream is more efficient at the TCP layer rather than the IP layer, so the TCP layer will try to discover when IP packets are being inadvertently fragmented. They do this by setting the "DF" (Don't Fragment) on all outgoing packets. When a router cannot forward the packet because it is too big, rather than fragmenting it, it sends back a "fragmentation needed" ICMP packet (type=3/code=4). The TCP stack then starts sending smaller IP packets, segmenting the data at the TCP layer rather than allow routers to fragment at the IP layer. Therefore, firewalls must be configured to allow incoming ICMP type=3, code=4 packets.
Another issue is Host unreachable and Destination Unreachable packets. Allowing these to come in through your firewall will allow connections to timeout faster, but they can also be used as a denial of service attack (by disconnecting clients from servers).
Users will constantly ask for the ability to ping and traceroute machines on the Internet. Most firewall adminsitrators will eventually give into these demands. Nobody really needs to ping/traceroute, but they really want to. It should be remembered, however, that ICMP ping responses are often used as a covert-cahnnel. (The massive DDoS attacks against Internet portals used this as a covert channel).
For more information on this, you may want to consult "Protect and Survive Using IBM Firewall 3.1 for AIX", IBM publication SG24-2577-02. See http://www.redbooks.ibm.com for more info. I disagree with it, though.
Another good document is http://www.worldgate.com/~marcs/mtu/.
There is a little twist to this scenario. A little-endian machine (like Intel processors) stores numbers in reverse byte-order than how numbers are represented on the wire. This means that a monotonically increasing integer from a Wintel box will increment the high-order byte first, whereas a Sun SPARC box will increment the low-order byte first. Therefore, lets say that you are being pinged steadily from both a Sun SPARC and a Wintel, you will see the following sort of progression in the IP ID field:
The above numbers are shown in hexadecimal, which shows the byte-order problem. However, many firewall logs (stupidly) show these numbers in decimal. If a firewall system assumes the number is big-endian but the incoming packets are little endian, then the progression of the numbers is hidden. For example:
This entire issue is complicated by the fact that a firewall running on a platform doesn't have to base its decimal calculation of the IP ID field on the underlying CPU. What I mean by this is that the C code that interprets the IP ID could be written in two ways;
/* ID field is a 2-byte number at offset 4 within the IP header */ int ipid_cpu = *(unsigned short*)(iphdr+4); int ipid_be = iphdr * 256 + iphdr;
The first example is CPU dependent: x86 CPUs will pull it out as a little-endian number, but SPARC CPUs will pull it out as a big-endian number. The second form is CPU independent: it tells all CPUs to interpret the field as a big-endian number. Note: ntohs(*(unsigned short*)(iphdr+4)) will crash a SPARC CPU and is not a good solution
Therefore, if you are running a Linux-based x86 firewall that interprets the IP ID field as a little-endian number, then a string of packets from a Wintel box will demonstrate a monotonically increasing number. However, a stream from a SPARC box will show skipping numbers. Conversely, if the Linux-based firewall uses the (correct) field parsing method, you'll see the reverse.
Moral of the story: Find out the byte order interpretation of the IP ID field used within your firewall. Also send your firewall vendor flames telling them to get with the program and represent the field in hex in the first place.
It also means that you can tell how far away a person is from the TTL field, and sometimes what kind of platform they are running. Most Windows machines send packets with a starting TTL of 128. This means that if your firewall log shows a TTL=112, then you can make the guess that the sender is 16 hops away, and that they are using a Windows machine.
Conversely, UNIX machines typically choose 64 as the starting TTL, so a packet when the TTL is 51, then it probably isn't Windows, but it is probably 13 hops away.
This technique was once used to find the source of nmap decoy scans. The decoys where given random TTLs, but the real scans were give normal TTLs. This allowed the astute observer the ability to sift through the incoming decoys and find the real scan. The nmap program was soon fixed to randomize the TTL of the real scan as well.
Note, as of January 2003, the vast majority of packets on this port are from Windows File Sharing worms
NetBIOS requests to UDP port 137 are the most common item you will see in your firewall reject logs. This comes about from a feature in Microsoft's Windows: when a program resolves an IP address into a name, it may send a NetBIOS query to IP address. This is part of the background radiation of the Internet, and is rarely something to be concerned about.
The discussion of these NetBIOS packets crops up over and over again on firewall/incident mailing lists. In this section, I've tried to come up with the "definitive" answer to this question.
While most NetBIOS packets are NON-HOSTILE, some are. Hackers regularly scan the Internet looking for information. Worms that infect port 139 will first need a NetBIOS lookup on por 137.
In the year 2002, port 137 probes have increased dramatically due to various worms that spread themselves via Windows File Sharing. These worms are not
We call DNS a directory service, where the word directory has the same meaning as in phone networks. In the U.S., we can dial directory assistance at 411 rather than looking up a name in the phone book. Either way, the goal is to lookup a name, and receive a number.
In a similar manner, sometimes you have a number, and you want to find the name. Let's say that you have caller ID and somebody calls you with the phone number (212) 555-1038. This doesn't tell you who this is, so you want to do the reverse lookup and discover the person's name.
In much the same fashion, the Internet provides a number of capabilities to resolve an IP address into a name.
The key thing to remember about gethostbyaddr() is that it is virtual. It doesn't specify how it resolves an address into a name. In practice, it will use all available mechanisms.
If we look at UNIX, Windows, and Macintosh systems, we see the following techniques:
Windows systems do the /etc/hosts, DNS, WINS, and NodeStatus techniques.
In more excruciating detail, Microsoft has a generic system component called a naming service. All the protocol stacks in the system (NetBIOS, TCP/IP, Novel IPX, AppleTalk, Banyan, etc.) register the kinds of name resolutions they can perform. Some RPC products will likewise register an NIS naming service. When a program requests to resolve an address, this address gets passed onto the generic naming service. Windows will try each registered name resolution subsystem sequentially until it gets an answer. (Side note: User's sometimes complained that accessing Windows servers is slow. This is caused by installing unneeded protocol stacks that must timeout first before the real protocol stack is queried for the server name.).
The order in which it performs these resolution steps for IP addresses can be configured under the Windows registry key HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\Tcpip\ServiceProvider. Of course, that doesn't help you the firewall admin.
The process is simply that a program requests the name for an IP address, and sends this request to all the protocol stacks. If the NetBIOS stack receives such a request, it always sends a NetBIOS query to the IP address. It doesn't matter if you have (or haven't) an existing NetBIOS connection to the machine.
In other words, the only requirement necessary in order to receive such packets is that you have an IP address.
Note that starting in late 1999, desktop security tools like personal firewalls have exploded. This means that the number of NetBIOS queries have dramatically risen.
Also, see section 10.6 for an explanation of how a simple configuration error in DNS can cause you to be suddenly flooded with such requests.
If the Windows box is trying to find the name for the IP address 192.0.2.21, it will do the following steps:
The personal firewall BlackICE Defender will may do its own NetBIOS queries separate from the underlying Windows OS. These will look like UNIX queries from dynamic ports, and have longer, progressive timeouts of 15-seconds, 30-seconds, and 1-minute.
The packet looks something like the example below. For more information about interpreting this, please see my sniffing FAQ at http://www.robertgraham.com/pubs/sniffing-faq.html.
ETHERNET: ETYPE = 0x0800 : Protocol = IP: DOD Internet Protocol IP: ID = 0x3E16; Proto = UDP; Len: 78 UDP: Src Port: NETBIOS (137); Dst Port: NETBIOS (137); Length = 58 NBT: NS: Query req. for *<00...(15)> NBT: Transaction ID = 57032 (0xDEC8) + NBT: Flags Summary = 0x0000 - Req.; Query; Success NBT: Question Count = 1 (0x1) NBT: Answer Count = 0 (0x0) NBT: Name Service Count = 0 (0x0) NBT: Additional Record Count = 0 (0x0) NBT: Question Name = *<00...(15)> NBT: Question Type = Node Status Request NBT: Question Class = Internet Class 00000: 00 E0 18 E0 0C E7 00 40 05 A4 79 32 08 00 45 00 .......@..y2..E. 00010: 00 4E 3E 16 00 00 80 11 2F CE 0A 0A 00 09 C0 00 .N>...../....... 00020: 02 A8 00 89 00 89 00 3A 14 AB DE C8 00 00 00 01 .......:........ 00030: 00 00 00 00 00 00 20 43 4B 41 41 41 41 41 41 41 ...... CKAAAAAAA 00040: 41 41 41 41 41 41 41 41 41 41 41 41 41 41 41 41 AAAAAAAAAAAAAAAA 00050: 41 41 41 41 41 41 41 00 00 21 00 01 AAAAAAA..!..
If you are seeing a lot of these requests, it probably means you have one of the following DNS issues.
Note that in this day/age with CIDR and address blocks smaller than 255 members, a lot of ISPs don't know how to forward DNS PTR requests to your server.
No matter what you do, you will still get requests because of configuration errors on the client's ISP. However, making sure the issues above are resolved on your own DNS servers will be an important first step.
The thing to remember is that A and PTR queries are unrelated.
When you register your domain name (example.com) you go to the owner of .com (Network Solutions) and purchase the address. As part of your registration, you tell Network Solutions something to effect "Please pass any DNS queries for the domain example.com to my DNS server ns1.example.com which is located at the IP address 192.0.2.168".
Thus, when resolving www.example.com, you first ask .com for the DNS server for example.com, which is ns1.example.com/192.0.2.168. You then ask that server for the A record for www.
Now going the reverse direction is a bit tougher. When trying to figure out who owns the IP address 192.0.2.3, you've got a problem. What is the first step? The solution was to query for a PTR record with the pseudo-name that looks something like "220.127.116.11.in-addr.arpa". Like the .com domain, the .arpa domain is run by a special company. It forwards the requests to the backbone ISPs, which then forward the request to the smaller ISPs and customers.
This forwarding mechanism is easily broken due to CIDR addresses. An ISP may assign 192.0.2.[0.127] to one customer, and 192.0.2.[128-255] to another customer. In order to fix this issue, the ISP must support special CNAME records that delegate lookups. For the network 192.0.2.128/25, then the CNAME record would look like 128/18.104.22.168.IN-ADDR.ARPA. This is kinda complex, easy to get wrong, and the administrators at ISPs often don't know how to do it right.
Please see CNAME -> PTR indirection described in RFC 2317 for more details on this. Also see http://www.dns.net/dnsrd/ for extensive DNS resources.
The upshot is that you probably have a CIDR allocation that breaks PTR queries causing NetBIOS queries. Harangue your ISP until they fix this.
This section analyzes some common worms.
Windows contains a protocol called "SMB/CIFS" that allows you to store files on servers in corporate environments. It also works in the home, so you can store all your files on one machine and access them from any other machine in the home. Unfortunately, many people make their desktop machines into "file servers". This opens their machines to attack.
Worms can spread themselves in many ways.
Older Win9x (Win95/Win98/WinME) systems have a different Windows File Sharing subsystem than the newer versions of Windows. One aspect of this system is that the worm must know the NetBIOS name of the target system. Therefore, the worm first does a NetBIOS NodeStatus query as explained in section 10. If that query fails, then the worm can't infect the system. Note that worm will really go over port 139/tcp, but what you see in your logs are failed 137/udp attempts.
Another aspect of Win9x is that it uses "share-level access". The user sharing files usually puts just a password on a share. Rather than logging in with an "account" containing your own user name and password, you log on anonymously with a different password specific to that "share". All unpatched Win9x systems have a bug such that an attacker or worm only needs to know the first character of a password -- and they can easily try all characters until they guess the correct one. At least one worm uses this technique.
When attacking Win9x systems this way, most worms will attempt to simply get to a share called "C", which is frequently the name of the "C:\" drive. They will then try to copy themselves to places on the system that will run on the next bootup.
The worm will usually connect directly to port 139. Most often, the worm will attempt to log on as the user "Administrator" with no password or some common passwords. Sometimes it will do a NetBIOS NodeStatus query as discussed in section 10 in order to find somebody's login name.
Some worms try to connect to "default" shares. WinNT systems will conveniently share your "C:\" drive as a share calld "C$". Other worms try to connect to a drive called "C". Many worms will first list all your shares, and then connect to each one.
The worm hopes to drop copies of itself in places that will run next time you log in or boot the system.
When attached to a gigabit link, a single Pentium 4 2 GHz machine can generate several hundred megabits-per-second, pumping out over 100,000 UDP packets-per-second. Alone, such a machine could infect the entire Internet (4-billion addresses) in 12-hours. A hundred such machines would infect the entire Internet in less than 10-minutes. Since more than 100 machines on gigabit links were infected, the outbreak of the worm was nearly instantaneous. The slow machines helped spread the worm, too.
At the same time, it was relatively easy to for owners of the high-bandwidth machines to quickly turn the systems off, not fast enough to stop infection, but fast enough to enough to stop much of the disruption. More importantly, UDP port 1434 isn't used for anything important, so ISPs could quickly put packet filters in their firewalls to contain the outbreak. This is important: the same thing was not an option for CodeRed and Nimda because they used port 80 -- the most commonly used port on the Internet.
Most of the "damage" caused by the worm weren't high levels of incoming packets coming into your network. Instead, if your network was taken down, it was because you forgot to patch one of your hundreds of machines, and that one machine by itself took down your outbound link. As soon as you turned off that machine, your network connectivity was restored.
Maybe it wasn't one of your machines, but one of your next-hop ISP's customers. The effect was the same: as soon as they turned off the high-bandwidth machine (or firewalled it on the router immediately upstream), connectivity was restored.
Note the distorting aspects of the traffic. Most of the infected machines were on low bandwidth machines, usually home users running MSDE (Microsoft's desktop SQL service bundled with products like Visio). However, since these guys were on low bandwidth machines, you probably never saw packets from them. Most of the packets you saw were from the infected machines on high bandwidth connections from major corporations and universities -- a single machine on a 100-mbps link produces 1000 times as much traffic as one on a DSL link. A hundred thousand machines on DSL links would not produce as many attacks as 100 machines on gigabit links.
I pose the above poll because what you see in your firewall logs says more about you than the attacker. You are seeing just the normal background radiation of the Internet, your firewall logs probably do not indicate that somebody is attacking you personally.
While the number of paranoid people is small, they are vocal. I often get e-mail from people trying to prove that their is a conspiracy, that one or more people are out to get them. This is almost never the case, and the firewall logs they show me are your normal everyday sort.
This is common. When you dialup the Internet (including non-static cable-modem/DSL connections), you will be assigned an IP address that was recently used by somebody else. That person may have had open chat sessions or been using a peer-to-peer (P2P) application. It becomes a case of mistaken identity: they don't know that there is a new person at the IP address that is no longer running that application. Furthermore, many applications are poorly written such that they will continue to bombard you with connection attempts.
Today, the most common cause of this is peer-to-peer applications like Gnutella and Napster-clones. Unfortunately, your IP address was given out to lots of people, and it takes awhile for it to be taken off their list. You can see connection attempts many hours later.
Another example is chat programs. If someone simply hangs up, then everyone who was chatting with that person will attempt to still send traffic to them. Some programs take a long time to timeout. Typical programs that show this behavior are PowWow and ICQ.
Another example is on-line, multiple games. You might see such traffic from gaming providers like MPlayer, or maybe from unknown servers (Quake servers litter the Internet). These games are typically UDP based, so there is no concept of a connection that can be dropped. They also are quite aggressive at maintaining connections, in order to make a good user experience. Some game ports that you might see are:
|7777||Unreal, Klingon Honor Guard|
|27015||Half-life, Team Fortress Classic (TFC)|
|28000-28008||Starsiege TRIBES (TRIBES.DYNAMIX.COM)|
Another example is multimedia audio/visual. For example, RealAudio uses UDP ports in the range of 6970-7170 for clients to receive audio streams.
Make sure that you carefully figure out the correct side of the connection. For example, an ICQ server runs on port 4000, and the client chooses a random high-numbered port. That means you will see UDP packets from port 4000 going to the random port. In other words, don't go looking in a port database trying to figure what that random, high-numbered port means. The significant port is the source.
The SubSeven trojan has a similar problem. It uses separate TCP connections for different services. If the slave agent goes away, it will continue to create connection attempts to the slave ports, especially at port 6776.
Look closely at the sentance above. I didn't say "firewalls stop hackers", I said that "you stop hackers, using a firewall". Firewalls work by blocking traffic. You have to tell them what traffic you want to block. Firewalls cannot stop attacks in traffic that you want to allow.
Note that any program can choose any port. Just because a program chooses a particular port doesn't mean it provides the service described in this guide.