wilhelm is a reversing tool I’ve been working on for a long while, which I’ve finally decided to release. It’s a Python API that provides (in my opinion) a better interface for working with Hex-Rays. In particular, I designed it with the IDAPython REPL (aka console) in mind; while it works fine in scripts, it’s meant to be used interactively to quickly automate some analysis while reversing.
Currently, wilhelm is very much a work in progress, and many features that I plan to add haven’t been written yet. However, the parts that already do exist have proven quite useful to me while I do my work, so I decided that other people might find it useful too. In particular, I think the API for accessing the abstract syntax tree (AST) of a decompiled function comes in really handy when reversing large applications, and the AST node selection feature (known as WilPath) allows one to automate a lot of otherwise menial tasks.
Besides for AST access, wilhelm also provides an event system (necessary when working interactively as the IDA database can change), a qualified name management system, and will in the future include a higher-level type system. You can try wilhelm out by cloning the repository on GitHub; I haven’t gotten down to packaging it yet, but it should work if you just append the right directory to IDAPython’s
Check out wilhelm on GitHub here.
The best way to show what wilhelm can do is through a simple example. I decided to randomly choose some binary and start analyzing it in IDA. I ended up looking at a copy of
libstagefright.so , taken from some old Android firmware.
Quick introduction to libstagefright: libstagefright is a media processing library that is part of the Android Open Source Project (AOSP). A function of interest in libstagefright is
android::MPEG4Extractor::parseChunk() : it is part of the MPEG4 decoder, and parses chunks (aka boxes) in an MPEG4 stream. Each chunk begins with 4 bytes indicating its size, and then 4 bytes indicating the type of the chunk, often referred to as the chunk’s fourcc.
parseChunk() checks the chunk’s fourcc and performs different actions based on it. Most of the time, fourcc values consist of four printable ASCII characters, such as
moov, which make them easy to identify.
Here’s what some of the fourcc parsing code looks like when freshly decompiled by Hex-Rays (the fourcc value is stored in the local variable
As you can see, Hex-Rays represents the literal fourcc values as 32-bit integers. This makes the code harder to read if you’re familiar with the fourcc values represented as ASCII strings. We can manually convert the integers to ASCII strings by pressing the “R” hotkey, but this would take a while as
parseChunk() is a pretty huge function (the pseudocode is 4,937 lines long). I’m feeling lazy; could we write a few lines of Python code to automate this?
Accessing the AST
We can, with wilhelm’s help. The first thing we want to do is identify the components of the AST that we wish to manipulate. These would be the literal numeric expressions that appear as part of a binary operation expression (binop) where the left-hand-side (LHS) operand is the local variable “
We can access a decompiled function’s AST using wilhelm:
func object is the function at the current cursor position (screen_ea), and its AST can be accessed via
func.body, which is the top-level block statement of the function. For example, we could access the first statement in the function using
func.body, and (because this is an if-statement), access its LHS expression with the following:
wilhelm allows nodes within the AST to be filtered and navigated, in a fashion similar to how you can filter and navigate through elements in an HTML/XML document. The optional WilPath feature (which we enabled by passing
wilhelm.initialize()) allows us to use an XPath-inspired DSL to quickly return a list of nodes that meet a specific criteria.
(Note that if you don’t want to use WilPath, you can still filter and navigate through nodes using methods on the
The WilPath DSL consists of a sequence of selectors, which get applied to a list of AST nodes. A selector is either a navigator selector, which maps nodes in the list to other nodes, or a filter selector, which returns a subset of nodes in the list meeting some criteria. See the docstring in
path.py for more information.
When WilPath is loaded, lists of nodes (
NodeList objects) have a
select() function can be used to apply a WilPath to the list. Let’s start with the function’s top-level block:
This returns all if-statements that are direct children of the block, i.e. at the top level.
select() returns a
NodeList, which is a lazy container that does not compute its contents until necessary. By creating a list out of the
NodeList, we can force it to evaluate and see it contains two statements. To make the rest of the code easier to read, we define a function to call select and evaluate the returned
Now, what we are looking for are binops in the entire function. We can search for all binops using the following WilPath:
Using the all navigator (
*/) allows us to search for all descendants of the function body, and not just immediate children.
We don’t want just any binary operation though; only those that involve
fourcc2. We can select for that using the following:
In this expression, we use the attribute navigator
.e_lhs to select the LHS operands of all found binops. We then filter these operands to include only those that are local variable expressions, and further filter those to only include those where the local variable is
This brings us closer to our goal. However, what we have is a list of local variable expressions referencing the local variable
fourcc2, but not the numeric expressions we want, which are actually the right-hand side (RHS) operands of the parent binops. We can select these by doing the following:
We use the subpath filter (
) to hold on to our reference to the parent binary operation expression (the filter will return only nodes for which the WilPath in the brackets successfully selects something). Then, we select the RHS operand using the attribute navigator, and filter to include only literal number expressions.
Converting to ASCII Strings
We now have a list of literal numeric expressions which should all be fourcc values. The final thing to do is to ask Hex-Rays to display these numbers as ASCII strings instead.
Unfortunately, this is not as straightforward as it should be, because I can’t seem to find a function that does it automatically (when working with the disassembly, you can just use the
op_chr() function). I had to write a helper function to do this:
The function uses the Hex-Rays API to modify the user-defined number formats associated with the function. I’ll probably include the code to do this in a future version of wilhelm. (If anybody knows of a better way of doing this, please let me know!)
Once this helper function is defined, we can then complete our task and convert all fourcc values to ASCII with (almost) a single line of Python:
I couldn’t seem to find a way to get Hex-Rays to update the pseudocode view to reflect the new number formats, but re-decompiling the function (e.g. pressing F5) appears to do the trick. The final result, with a total of 99 fourcc literals converted:
There are of course other ways of automating this particular task: it might be more efficient to search the entire function for number literals, and convert them if they appear to be made up of printable ASCII literals. But I thought this would make a good example of how wilhelm can help you make your way around the AST.
Why Use the AST?
You might also wonder why one should bother with the AST at all; why not just work with the disassembly, or with the Hex-Rays microcode? In this particular case, we’re modifying the AST to make things easier to read. In general, though, there are advantages to working at the AST level. The AST is the result of complex program analysis, and using it allows you to benefit from what was already analyzed. You can, for example, directly work with the arguments to function calls, which you might have had to figure out by yourself if you were working at a disassembly level.
I use this a lot when I find logging functions in stripped binaries. If calls to the logging function happen to include the caller’s function name as as argument, I can quickly write some code to rename all functions that call the logger.
So that’s an example of what you can do using wilhelm’s AST and WilPath APIs. Hopefully you’ll find this useful in your work, and hopefully the other features I’ll be adding to wilhelm will come in handy as well. Please feel free to try it out, and submit plenty of bug reports as it’s not been heavily tested yet!
PS: wilhelm is named after Wilhelm Röntgen, discoverer of X-rays.
PPS: The IDAPython REPL is a lot nicer to use with a proper console. Try out ipyida to get a modern Qt console with better autocomplete, history, syntax highlighting and more.