Dytan is a dynamic taint analysis framework that allows users to
implement different kinds of dynamic taint analyses. This file
discusses the features of Dytan that can be accessed through the
high-level XML interface. The use of Dytan's core without leveraging
the interface is described in file internals.txt
====================================
Prerequisites
====================================
Dytan is based on PIN version 2.2-13635-gcc.4.0.0-ia32-linux. To
download and install the right version of PIN, users should execute
script "setup-pin.sh", which is provided with the distribution.
Because of recent changes in the Linux kernel, this version of PIN
does not work correctly on the most recent Linux distributions. We use
Linux kernel version 2.6.15. More precisely, we recommend users to
install Dytan on a machine running Ubuntu 6.06, which is the platform
on which it was tested.
To be able to compile Dytan, it is necessary to install the libxml2
development library (package libxml2-dev for Debian/Ubuntu users).
Control-flow based taint propagation leverages post-dominance
information computed by one of Dytan's modules. In the presence of
indirect jumps in the code, postdominance information is not computed
and a special basic block is added in the position where the jump is.
To alleviate this issue, it is recommended to pre-process the source
code of the programs under analysis using CIL, whenever possible. CIL
eliminates some of the constructs that cause indirect jumps in the
code, such as "switch" statements.
====================================
Dytan's features and configuration
====================================
The implementation of Dytan's interface is still work in progress. As
of now, Dytan allows users to:
- Taint specific arguments of a function.
- Taint arbitrary memory ranges (from within a program).
- Taint data originating from a file
- Taint data originating from a network connection
Dytan's interface does not support yet function return values.
Dytan's configuration
---------------------
Most of Dytan's options can be configured by modifying a configuration
file called "config.xml. Examples of configuration files are provided
with the distribution (in the "root" and in the "sample" folders). The
config file contains information about propagation policies, sources,
and sinks.
The propagation policy is defined in the <propagation> section. By
specifying the <controlflow> tag as "false" or "true", users can
define whether propagation should occur only through data dependences
or through both data and control dependences. (Note that our current
implementation accounts for information flows due to direct control
dependences only, that is, flows due to the fact that something is not
executed are not considered.)
Taint sources are specified in the <sources> section. Within the
section, tag <taint-marks> is used to specify the number of unique
taint marks Dytan should use. For example, the following config
fragment would tell Dytan to use 32 unique taint marks (after all
marks have been used, Dytan starts reassigning them):
<sources>
<taint-marks>32</taint-marks>
...
</sources>
Currently, users can specify two kinds of taint sources in the config
file: files and network connections. To specify specific arguments of
a function or arbitrary memory ranges, users need to directly modify
the program or Dytan, respectively, as discussed below.
Files can be specified as sources in the config file as follows:
<sources>
...
<source type="path">
<file>path/to/file1/filename1</file>
<file>path/to/file2/filename2</file>
<granularity>PerRead</granularity>
</source>
...
</sources>
Note that, due to a current limitation of the implementation, Dytan
identifies file sources by matching the path used by the program with
the one specified in the <source> section. Therefore, the file path
specified in the config file should match what the program uses.
The <granularity> tag specifies the level of granularity at which
taint marks are assigned. Value "PerRead" tells Dytan to taint all the
bytes read by "read" operation with a single taint mark. conversely,
value "PerByte" specifies that each byte read should be tainted with a
different taint mark.
To taint data coming from a network connection, the user must add to
the config file a section in the following format:
<sources>
...
<source type="network">
<host>127.0.0.1</host>
<port>80</port>
<granularity>PerRead</granularity>
</source>
...
</sources>
There can be multiple <source> entries in the configuration file.
How to taint specific arguments of a function
---------------------------------------------
To taint function arguments, users must modify the functions in file
taint_func_args.cpp. After modifying the file, Dytan must be
recompiled by invoking "make" in the root directory of the
distribution. As an example, the version of the file in the
distribution contains code to taint arguments to the "main" function.
Please refer to that file for a better understanding of the next
paragraph, which describes how to modify file taint_func_args.cpp.
The string array "taint_function" should contain the names of the
routines whose parameters are to be tainted. In function
"taint_routines(RTN, void *)", there should be an if block for each of
the functions in the "taint_function" array. Each of these blocks
should use PIN API to insert a call to a suitably defined wrapper
before the original function is executed. This is accomplished by
using PIN API function "RTN_InsertCall". This function takes the
following arguments:
- RTN rtn: the original function.
- IPOINT_BEFORE: tells PIN to insert the call before rtn's entry.
- AFUNPTR(<wrapper name>): specifies the wrapper function's name.
- Assuming that rtn has n arguments:
- IARG_FUNCARG_ENTRYPOINT_REFERENCE, 0,
- IARG_FUNCARG_ENTRYPOINT_REFERENCE, 1,
- ...
- IARG_FUNCARG_ENTRYPOINT_REFERENCE, n,
- IARG_END: tells RTN_InsertCall that there are no more parameters.
The wrapper function is the place where the tainting actually occurs.
It takes n arguments of type "ADDRINT *", where n is the number of
arguments of the original function. For each parameter to be tainted,
the wrapper calls Dytan's memory tainting function, which takes three
parameters: the starting address of the parameter's memory location,
the size of the parameter in bytes, and the numeric value of the taint
mark to be associated with the parameter. See function
"main_wrapper_func" in file taint_func_args.cpp for an example of a
wrapper.
How to taint arbitrary memory ranges
------------------------------------
To taint arbitrary memory ranges from within your program, make a call
to function
DYTAN_tag(ADDRINT start_address, size_t size, char * name)
where size is the size of the memory to be tainted and name is the
string to be associated with this taint mark.
Analogously, to display taint marks at a particular memory location
from within your program, make a call to function
DYTAN_display(ADDRINT start_address, size_t size, char *fmt)
where fmt is the format in which the taint marks should be displayed.
The sample program provided with this distribution, available in
directory "sample/wc", makes the above calls for tainting memory and
displaying memory taint marks. See file internals.txt for
instructions on how to compile the sample program.
====================================
Running Dytan
====================================
After configuring and recompiling Dytan, it can be run by executing:
./run-dytan <program_name>
For example, to run Dytan on the provided example, go to the root
directory of the distribution and run
./run-dytan sample/wc/wc sample/wc/a.txt
====================================
Internals
====================================
See internals.txt for details on the internals of Dytan.