HTML Flow Parser

This software is fed a network capture, retrieves the HTML payloads, and parses them for symptoms of unethical search engine optimization. This seeks to avoid the embarrassing scenario of a home institution being notified by a user or outside entity that their server appears to be hacked (e.g. viagra is advertized). By taking an active approach, we hope the uptime of these problems is cut drastically.

Instructions

Download a copy of the repository by running the command git clone https://github.com/DoctorSher/html-flow-parser.git. Navigate to the created folder, and type make to compile the project. To run, type ./tcpflow pcap-file, where pcap-file is the packet capture file you'd like to search.

Design Components

The first component is a modified tcpflow; the original version is maintained by Simson Garfinkel at https://github.com/simsong/tcpflow. I slimmed it down for my narrow purposes (~4500 lines of code reduction). My version of tcpflow receives a packet capture file, and produces the flow files as the original did. However, once an inbound/outbound flow pair is ready (client to server and server to client per tcp connection), it sends both of these as arguments to the second and last component.

This component is the HTML Flow Parser, and the focus of the project. It receives the flow files, and does the following steps:

1. Determines from the port numbers on the ends of the flow pair which one the outbound flow is. This is a simple check to see which file's remote port is 80.
2. Retrieves the host domain from the outbound flow, which will come into play later.
3. Parses the HTTP options of the inbound flow (containing the HTTP response(s) that we are concerned with). Deals with chunked and compressed sending options. Does whatever recipe of instrutions it needs to retrieve the HTML payload.
4. Parses the HTML payload for symptoms of unethical SEO. We have two currently implemented tests.
   1. The first is a remote URL frequency counter. This collects all of the remote URLs from the page (e.g. using http). With these URLs, it checks to see if a host domain was linked to with immediate proximity of another link to the same domain. The reason we extract the hostname from the inbound flow, is that we don't want to penalize a website for linking remotely to another website in the same domain. If the frequency of this links is higher than a user-configured threshold, a warning is raised. The reason for the frequency counter is that many times malicious posts are very repetitious, in order to maximize the SEO optimization for their malicious site.
   2. The second is a simple keyword search. Just as it sounds, it scans the page for certain keywords. If a keyword appears more than a user-configured threshold, a warning is raised. The keywords used were garnered from a Stefan Savage et al. paper entitled "Judging a site by its content: learning the textual, structural, and visual features of malicious Web pages." In this paper, they used Machine Learning techniques to illuminate the most common keywords for malicious websites. This seemed like a great starting point to us.
5. With all of these warnings, a simple textual report is generated. Ideally, one would run this and then check out the warnings immediately.

Results

The program isn't anywhere close to being polished, but it works quite well for what it does. There are false positives, but part of that can lie in the user-configured thresholds chosen. Future considerations are also looking to reduce the number.

Future Work

• Implementing more and better tests. Open to ideas!
• Making the parser more robust (e.g. with regular expressions instead of string searches).
• Allow for user-uploadable keywords. Currently they can pick which categories they expect to not appear on their site, but cannot upload their own.
• Parse Javascript and CSS, as URLs and other black hat techniques can appear in there.
• Make tests more pluggable. Instead of being hardcoded into the parser, they could be loaded much like kernel modules.

Licenses

Copyright 2012 Washington University in St Louis

Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at

http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.

Please see PARSER-LICENSE for more details.

The original license of tcpflow is also included.

Please see TCPFLOW-LICENSE for more details.