How to use this tool to retrieve global variables

[Plyees]

I am looking to retrieve all global variables from debuginfo that is generated for C application under Ubuntu. Could anyone please kindly give me a hint?

[sevaa]

Use https://github.com/eliben/pyelftools/blob/master/examples/dwarf_die_tree.py as the starting point. The example scrolls through all CUs and DIEs in the file; you'd have to write a condition that would identify a variable (start with die.tag = "DW_TAG_variable", but that's probably not sufficient), then dump whatever variable properties you are interested in.

Variables inside functions are not globals. I don't know if function scoped static variables qualify for your case.

[Plyees]

Thank you for your hint. I have tried to explore into DIE to assemble variables, but blocked by some variables with a mix of structure and array involving multiple recursion.

[Plyees]

function scoped variables are not my interest. My debuginfo seems containing only global variables defined outside the functions

[sevaa]

I take it, you are interested in datatypes, too.

In general, spelling out the datatype of a C++ variable from DWARF data is a very tricky task. Between classes, templates, namespaces, typedefs, const-ness, and function pointers (where you have to spell out the type of each argument), it's a major pain. I have a semi-working piece of Python code that sort of spells the datatype for the subset of DWARF that is relevant to me, but it's a work in progress and it isn't public.

[sevaa]

Update: a version of that logic found its way into pyelftools proper. See describe_cpp_datatype in elftools/dwarf/datatype_cpp.

It was tested on the small corpus of binaries that we have. Needless say, not nearly exhaustive. Also, it stops on typedefs (in line with what llvm-dwarfdump does) rather than parsing down to the basic types.

[sevaa]

@Plyees is still an issue?