This month's challenge is to analyze a successful compromise and the attacker's actions after it. All submissions are due no later then 23:00 GMT, Friday, May 23rd. Results will be released Friday, May 30.
Skill Level: Intermediate
Members of the AT&T Mexico Honeynet captured a unique attack. As common, what is interesting is not how the attackers broke in, but what they did afterwards. Your mission is to analyze the network capture of the attacker's activity and decode the attacker's actions. There are two binary log files. Day1 captured the break in, Day3 captures some unique activity following the compromise. The honeypot in question is IP 192.168.100.28. Make sure you review the challenge criteria before submitting your writeup.
I did the analysis on a NetBSD 1.6P running on vmware virtual machine. The reason why I decided to use the vmware virtual machine are the risks of analysing unknown files. At first the day1.log and day3.log files were downloaded from the SOTM 28 page and their integrity checked by checking the md5 checksums.
bash-2.05b$ ftp http://project.honeynet.org/scans/scan28/day1.log.gz Requesting http://project.honeynet.org/scans/scan28/day1.log.gz 100% |**************************************************************************************| 2818 KB 33.42 KB/s 00:00 ETA 2885991 bytes retrieved in 01:24 (33.42 KB/s) bash-2.05b$ ftp http://project.honeynet.org/scans/scan28/day3.log.gz Requesting http://project.honeynet.org/scans/scan28/day3.log.gz 100% |**************************************************************************************| 3393 KB 30.92 KB/s 00:00 ETA 3474770 bytes retrieved in 01:49 (30.91 KB/s) bash-2.05b$ md5 -n day.log.gz 79e5871791542c8f38dd9cee2b2bc317 day1.log.gz af8ab95f41530fe3561b506b422ed636 day3.log.gz bash-2.05b$
After that the day1.log was loaded to ethereal
bash-2.05b$ ethereal -r day1.log&and quickly scanned to get an impression what to expect and what kind of tools would be useful in more detailed and systematic analysis. This task is quite fast because we have a honeypot, so all activity is at suspect. This also means that at first we can skip scans and only pay attention to successfull tcp connection establishments (and the connections) and udp-packets. There were also lots of nameserver queries originating from the honeypot, since they are most likely benign they were filtered out using command:
bash-2.05$ tcpdump -r day1.log -w day1.nodns.log '!udp and port 53'
If the capture would be from a real server, then in most cases we would have so much data the this approach would not be feasible. In that case we might want to feed the capture to some intrusion detection system like snort to quickly find if there are any know exploits.
Also at the beginning of the analysis the different tcp connecttion flow were saved to files, for easy reference during further analysis. This task was done using tcpflow-tool.
bash-2.05b$ mkdir day1 bash-2.05b$ (cd day1 && tcpflow -r ../day1.nodns.log ) bash-2.05b$
The resulting file are listed here.
Before digging deeper into the dump file these flow files were investigated
using standard unix tools like
less. This was done to see
what to expect to find later during the analysis. It was also
quite easy to spot the flow used for the exploit.
To get the timeline of the tcp connections argus and ragator tools were used.
bash-2.05b$ argus -r day1.nodns.log -w day1.argus bash-2.05b$ ragator -c -G -nn - tcp < day1.argus| sort > day1.ragator
After these preparations we can start to go trought the cleaned day1.log, connection by connection. Using the port numbers and ethereal's follow tcp stream tool to identify the connections. Once connection had been identified, we can make notes what other network traffic it should generate and use the ethereal colorize display feature to mark the packets belonging to the connection so that it can be easily identified on the display.
Do this until all the packets have been colorized. Now you should have
quite good idea what is happening. Here is an example screen dump how
the network capture looks like once the colorizing has been done.
The complete ethereal color filter file
colorfilterscan be found from the .ethereal directory
in the users home directory.
One way to try to learn to know the attaker and to find answers to questions 6 and 7, is to follow the IRC conversations he is having. In this case they we in italian, so Google language tools provided some help.
The table below summarizes the tools used for the analysis.
|ftp||Internet file transfer program|
|md5||display file checksums and block counts|
|ethereal||Interactively browse network traffi|
|tcpdump||dump traffic on a network|
|tcpflow||TCP flow recorder|
|file||determine file type|
|strings||print the strings of printable characters in files|
|less||opposite of more (a filter for browsing text files)|
|argus||audit record generation and utilization system|
|ragator||aggregate argus(8) data file entries|
Tools used for making this report are not listed.
The the honeypot is running SunOS 5.8 operating system, which mean it is a Solaris 8 machine (see Solaris 2 FAQ for mapping SunOS revisions to Solaris versions). The analysis below describes the steps used to determine this information.
The answer to this question can be found from the scan the attaker did before the exploit. The scan consists of 3 steps:
Frame 571 (136 bytes on wire, 136 bytes captured) Ethernet II, Src: 08:00:20:d1:76:19, Dst: 00:07:ec:b2:d0:0a Internet Protocol, Src Addr: 192.168.100.28 (192.168.100.28), Dst Addr: 188.8.131.52 (184.108.40.206) Transmission Control Protocol, Src Port: 6112 (6112), Dst Port: 56710 (56710), Seq: 3124564266, Ack: 2140233551, Len: 70 Data (70 bytes) 0000 30 30 30 30 30 30 30 30 31 34 30 30 33 32 30 30 0000000014003200 0010 30 31 20 20 33 20 00 2f 2f 2e 53 50 43 5f 41 41 01 3 .//.SPC_AA 0020 41 56 54 61 71 44 64 00 31 30 30 30 00 7a 6f 62 AVTaqDd.1000.zob 0030 65 72 69 75 73 3a 53 75 6e 4f 53 3a 35 2e 38 3a erius:SunOS:5.8: 0040 73 75 6e 34 75 00 sun4u.
Determining the operating system is also one of the very first things
the attaker does after connecting to his rootshell by giving '
uname -a' command.
To be more certain (maybe the server was running application level honeypot on some other operating system, and only emulating vulnerabile dtspc-service) we can use passive finger printing verify this information. In this case if we compare the
Frame 104 (62 bytes on wire, 62 bytes captured) Ethernet II, Src: 08:00:20:d1:76:19, Dst: 00:07:ec:b2:d0:0a Internet Protocol, Src Addr: 192.168.100.28 (192.168.100.28), Dst Addr: 220.127.116.11 (18.104.22.168) Version: 4 Header length: 20 bytes Differentiated Services Field: 0x00 (DSCP 0x00: Default; ECN: 0x00) Total Length: 48 Identification: 0x8891 (34961) Flags: 0x04 .1.. = Don't fragment: Set ..0. = More fragments: Not set Fragment offset: 0 Time to live: 64 Protocol: TCP (0x06) Header checksum: 0x813e (incorrect, should be 0x0c8f) Source: 192.168.100.28 (192.168.100.28) Destination: 22.214.171.124 (126.96.36.199) Transmission Control Protocol, Src Port: 32783 (32783), Dst Port: ftp (21), Seq: 3216911659, Ack: 0, Len: 0 Source port: 32783 (32783) Destination port: ftp (21) Sequence number: 3216911659 Header length: 28 bytes Flags: 0x0002 (SYN) Window size: 24820 Checksum: 0x8424 (incorrect, should be 0x0f75) Options: (8 bytes)
In this case the values are
TTL is 64,
window size is 24820,
DF bit is
TOS is set to 0.
If we compare this with the values for Solaris 8,
we see that they match. So we got one more piece of evidence
suggesting that the kernel is SunOS 5.8 (which is used by Solaris 8).
The attaker used the dtspc vulnerability, see CERT advisory. and Scan of The Month 20 for more information of this attack.
There are several ways to catch this attack. One can feed the network capture log file to some IDS for example snort. Snort has at least two rules that could match and generate an alert:
alert ip $EXTERNAL_NET any -> $HOME_NET $SHELLCODE_PORTS (msg:"SHELLCODE sparc NOOP"; content:"|801c 4011 801c 4011 801c 4011 801c 4011|"; reference:arachnids,353; classtype:shellcode-detect; sid:645; rev:3;)and
alert tcp $EXTERNAL_NET any -> $HOME_NET 6112 (msg:"EXPLOIT CDE dtspcd exploit attempt"; flow:to_server,established; content:"1"; offset:10; depth:1; content:!"000"; offset:11; depth:3; reference:cve,CAN-2001-0803; reference:url,www.cert.org/advisories/CA-2002-01.html; classtype:misc-attack; sid:1398; rev:5;)
If we would be dealing with real system (i.e. not honeypot) using some IDS (like snort) would be good way to catch the attack.
In our case we know that none of the traffic is legitimate so we can do the scanning manually (especially since this is SOTM challenge we know there will be something interesting for us to find :-) and the size of the day1 log file is quite small...
The manual scanning can be done described earlier in the analysis overview chapter.The suspected exploit packet is below:
Frame 580 (1514 bytes on wire, 1514 bytes captured) Ethernet II, Src: 00:07:ec:b2:d0:0a, Dst: 08:00:20:d1:76:19 Internet Protocol, Src Addr: 188.8.131.52 (184.108.40.206), Dst Addr: 192.168.100.28 (192.168.100.28) Transmission Control Protocol, Src Port: 56711 (56711), Dst Port: 6112 (6112), Seq: 2143411080, Ack: 3124882182, Len: 1448 Data (1448 bytes) 0000 30 30 30 30 30 30 30 32 30 34 31 30 33 65 30 30 0000000204103e00 0010 30 33 20 20 34 20 00 00 00 31 30 00 80 1c 40 11 03 4 ...10...@. 0020 80 1c 40 11 10 80 01 01 80 1c 40 11 80 1c 40 11 ..@.......@...@. 0030 80 1c 40 11 80 1c 40 11 80 1c 40 11 80 1c 40 11 ..@...@...@...@. <removed lines> 04b0 80 1c 40 11 80 1c 40 11 80 1c 40 11 20 bf ff ff ..@...@...@. ... 04c0 20 bf ff ff 7f ff ff ff 90 03 e0 34 92 23 e0 20 ..........4.#. 04d0 a2 02 20 0c a4 02 20 10 c0 2a 20 08 c0 2a 20 0e .. ... ..* ..* . 04e0 d0 23 ff e0 e2 23 ff e4 e4 23 ff e8 c0 23 ff ec .#...#...#...#.. 04f0 82 10 20 0b 91 d0 20 08 2f 62 69 6e 2f 6b 73 68 .. ... ./bin/ksh 0500 20 20 20 20 2d 63 20 20 65 63 68 6f 20 22 69 6e -c echo "in 0510 67 72 65 73 6c 6f 63 6b 20 73 74 72 65 61 6d 20 greslock stream 0520 74 63 70 20 6e 6f 77 61 69 74 20 72 6f 6f 74 20 tcp nowait root 0530 2f 62 69 6e 2f 73 68 20 73 68 20 2d 69 22 3e 2f /bin/sh sh -i">/ 0540 74 6d 70 2f 78 3b 2f 75 73 72 2f 73 62 69 6e 2f tmp/x;/usr/sbin/ 0550 69 6e 65 74 64 20 2d 73 20 2f 74 6d 70 2f 78 3b inetd -s /tmp/x; 0560 73 6c 65 65 70 20 31 30 3b 2f 62 69 6e 2f 72 6d sleep 10;/bin/rm 0570 20 2d 66 20 2f 74 6d 70 2f 78 20 41 41 41 41 41 -f /tmp/x AAAAA 0580 41 41 41 41 41 41 41 41 41 41 41 41 41 41 41 41 AAAAAAAAAAAAAAAA 0590 41 41 41 41 41 41 41 41 41 41 41 41 41 41 41 41 AAAAAAAAAAAAAAAA 05a0 41 41 41 41 41 41 41 41 AAAAAAAA
Once we have found were the exploit takes place we can try to indentify it. So far we know the target operating system, the TCP port (6112), and the attack packet. Armed with these a search on Google turns out a link to the advisory of the dtspc-attack, which we see fits our case.
How do we then know that this was really the way the attaker got in, and not just some unsuccesfull attempt. For that we have to look what happened before and after this packet. Well we did part of that analysis already while answering question 1, namely wee saw the scan of the port that will be the root shell, it was not active. But shortly after this attack packet we see successfull connection establishment to the very same port (packets #588 and #590)..
Note that if we would have used an IDS to check the capture file, it would have told us this information. Since we are analysing a honeypot (and trying to learn), we should not trust the information the IDS provided to us, but to really check the exploit and make sure that it is not a false positive or some new variant of the old exploit.
Here is the list of systems used and what they were used for:
The systems that are directly connected to the attack are easily found
by checking the commands from the exploit and comparing them to the
network capture. The exploit created a listening
socket which will have an intereactive shell connected to it.
The attaker then connects to this rootshell and by following the commands the attaker gives,
we will find the FTP commands, the wget and start
of the setup shell script. The setup shell
script goes out to ftp.sunsolve.com to fetch some SunOS
patches using FTP. The rootkit contains SSH server at port 5001
and PsyBNC server at port 7000. So we might expect at some point
later activity targeted also to these ports.
Finally the attaker gives
./solbnc command to start
a distributed denial of service (DDOS) agent. This binary then
tries to contact the DDOS handlers.
And while doing this keep an eye on the packet capture and mark the connections that the actions of the intruder create. If there is a packet that one can not associate with an action - look harder :-)
How we can be sure that we have the complete list? The answer is easy we can not be sure. There can be any number of systems that are used in the attack that we have no idea, take for example, the machine the root shell was originated, do we know how the attaker was using the machine - directly or over the net ?
|2002-11-29 17:36:25||Scan of the vulnerability of the honeypot|
|17:36:26||Commense dtspc exploit against the honeypot|
|17:36:37||connection to the root shell on the honeypoit created by the expoloit|
|17:42.42||FTP connection to 220.127.116.11 to get wget, dlp, and solbnc|
|17:45:29||HTTP connection to 18.104.22.168 to get sol.tar.gz (the rootkit) from http://22.214.171.124/bobzz.sol.tar.gz using wget|
|17:53:56||FTP connection to ftp.sunsolve.com to get sun patches|
|17:59:52||first stacheldraht agent communication|
|18:04:21||connection to the spyBNC IRC bouncer running on the honeypot|
|18:11:02||successfully connected to IRC network irc.stealth.net|
The main source for this information is the root shell commands given and the TCP connection information obtained with argus and ragator. Ragator report with some annotations is here.
The ICMP packets with 'skillz' in them seems to be related to some distributed denial of service tool. The tool could be some variant of stacheldraht.
From the packet capture we can notice that the first ICMP packet with 'skillz' in it appears after the solbnc binary has been started from the root shell. This gives us two possibilities to try to find out the purpose of the packets, we can study the binary or try to find information based on the packet itself.
If we only look the sent ICMP packet and search from Google with keywords icmp, skill, and 6666 we will find references to stacheldraht DDOS tool. Further study of the solbnc binary show that there are some similarities but there is no perfect match (e.g. strings output from the reported stacheldraht agents are different). Unfortunately I ran out of time in trying to analyze the solbnc-binary further. The day3.log shows what I believe is the agent in action (DDOS attack against ) and also some more ICMP echo reply control packets.
The attaker used the Configured tunneling of IPv6 over IPv4 mechanisms as defined in RFC 2898 to carry IPv6 traffic over IPv4.
The motivation for doing this is not clear to me. In the day3.log network dump file this protocol was only used for IRC. No features of the IPv6 that could not be accomplished with IPv4 were used. This suggest that IPv6 was used purely for 'hack value', i.e. because the attaker was able to do it and wanted to connect what might have been IPv6 only IRC-network. There might be some clues of the motivation in the IRC-conversation the attaker had, but with my non-existant italian skills (even with the help of google-translations), I could not find the reason.
An easy way to detect and identify this protocol is to use the 'protocol hierarchy statistics' from ethereal. Here is a screen dump of the protocol hierarchy statistics run over the cleaned day3.log network capture. These statistics show that on top of internet protocol we have some ICMP, UDP, TCP and IPv6 packets. We can identify the protocal by looking one of these packets:
Frame 116258 (102 bytes on wire, 102 bytes captured) Ethernet II, Src: 08:00:20:d1:76:19, Dst: 00:07:ec:b2:d0:0a Internet Protocol, Src Addr: 192.168.100.28 (192.168.100.28), Dst Addr: 126.96.36.199 (188.8.131.52) Version: 4 Header length: 20 bytes Differentiated Services Field: 0x00 (DSCP 0x00: Default; ECN: 0x00) Total Length: 88 Identification: 0x7b83 (31619) Flags: 0x04 Fragment offset: 0 Time to live: 60 Protocol: IPv6 (0x29) Header checksum: 0xc494 (incorrect, should be 0x4fe5) Source: 192.168.100.28 (192.168.100.28) Destination: 184.108.40.206 (220.127.116.11) Internet Protocol Version 6 Version: 6 Traffic class: 0x00 Flowlabel: 0x00000 Payload length: 28 Next header: TCP (0x06) Hop limit: 60 Source address: 2001:6b8:0:400::5d0e (2001:6b8:0:400::5d0e) Destination address: 2001:750:2:0:202:a5ff:fef0:aac7 (2001:750:2:0:202:a5ff:fef0:aac7) Transmission Control Protocol, Src Port: 32780 (32780), Dst Port: ircd (6667), Seq: 53523086, Ack: 0, Len: 0
We see that the IP protocol number is 41 (hex 0x29), which is used to denote IPv6 over IPv4.
The nationality of the attaker is most likely Italian.
Even the attack was originated from Taiwan, the attaker was most likely Italian. I come to this conclusion mainly because the language used in the IRC discussions and also the locations from which the attaker got his tools.
First I must disagree IPv6 beeing unusual! It has been around quite a while and it is gaining popularity, altought slower than some people might want (me myself included:-).
Th main implication is that ID-industry must follow its time, i.e. keep the products current.
The most important tools used in this analysis are able to decode this protocol: ethereal, tcpdump, (and native tools on IPv6 aware operating systems) . On the tools used for this challenge the tcpflow-tool is not IPv6 compatiple. However, the same function can performed with ethereal, with a bit more manual work.