pygdbmi 0.8.3.0

Parse gdb machine interface output with Python

pygdbmi - Get Structured Output from GDB's Machine Interface

API Documentation

What's in the box?

1. A function to parse gdb machine interface string output and return structured data types (Python dicts) that are JSON serializable. Useful for writing the backend to a gdb frontend. For example, gdbgui uses pygdbmi on the backend.
2. A Python class to control and interact with gdb as a subprocess

To get machine interface output from gdb, run gdb with the --interpreter=mi2 flag like so:

gdb --interpreter=mi2

Installation

pip install pygdbmi

Compatibility

Operating Systems

Cross platform support for Linux, macOS and Windows

• Linux/Unix

  Ubuntu 14.04 and 16.04 have been tested to work. Other versions likely work as well.

• macOS

  Note: the error please check gdb is codesigned - see taskgated(8) can be fixed by codesigning gdb with these instructions. If the error is not fixed, please create an issue in github.

• Windows

  Windows 10 has been tested to work with MinGW and cygwin.

gdb versions

• gdb 7.6+ has been tested. Older versions may work as well.

Examples

gdb mi defines a syntax for its output that is suitable for machine readability and scripting: example output:

-> -break-insert main
<- ^done,bkpt={number="1",type="breakpoint",disp="keep",
enabled="y",addr="0x08048564",func="main",file="myprog.c",
fullname="/home/myprog.c",line="68",thread-groups=["i1"],
times="0"}
<- (gdb)

Use pygdbmi.gdbmiparser.parse_response to turn that string output into a JSON serializable dictionary

from pygdbmi import gdbmiparser
from pprint import pprint
response = gdbmiparser.parse_response('^done,bkpt={number="1",type="breakpoint",disp="keep", enabled="y",addr="0x08048564",func="main",file="myprog.c",fullname="/home/myprog.c",line="68",thread-groups=["i1"],times="0"')
pprint(response)
> {'message': 'done',
'payload': {'bkpt': {'addr': '0x08048564',
'disp': 'keep',
'enabled': 'y',
'file': 'myprog.c',
'fullname': '/home/myprog.c',
'func': 'main',
'line': '68',
'number': '1',
'thread-groups': ['i1'],
'times': '0',
'type': 'breakpoint'}},
'type': 'result'}

Programmatic Control Over gdb

But how do you get the gdb output into Python in the first place? If you want, pygdbmi also has a class to control gdb as subprocess. You can write commands, and get structured output back:

from pygdbmi.gdbcontroller import GdbController
from pprint import pprint

# Start gdb process
gdbmi = GdbController()

# Load binary a.out and get structured response
response = gdbmi.write('-file-exec-file a.out')
pprint(response)
[{'message': u'thread-group-added',
'payload': {u'id': u'i1'},
'type': 'notify'},
{'message': u'done', 'payload': None, 'type': 'result'}]

Now do whatever you want with gdb. All gdb commands, as well as gdb machine interface commands are acceptable. gdb mi commands give better structured output that is machine readable, rather than gdb console output. mi commands begin with a -.

response = gdbmi.write('-break-insert main')  # machine interface (MI) commands start with a '-'
response = gdbmi.write('break main')  # normal gdb commands work too, but the return value is slightly different
response = gdbmi.write('-exec-run')
response = gdbmi.write('run')
response = gdbmi.write('-exec-next', timeout_sec=0.1)  # the wait time can be modified from the default of 1 second
response = gdbmi.write('next')
response = gdbmi.write('next', raise_error_on_timeout=False)
response = gdbmi.write('next', raise_error_on_timeout=True, timeout_sec=0.01)
response = gdbmi.write('-exec-continue')
response = gdbmi.send_signal_to_gdb('SIGKILL')  # name of signal is okay
response = gdbmi.send_signal_to_gdb(2)  # value of signal is okay too
response = gdbmi.interrupt_gdb()  # sends SIGINT to gdb
response = gdbmi.write('continue')
response = gdbmi.exit()

Parsed Output Format

Each parsed gdb response consists of a list of dictionaries. Each dictionary has keys message, payload, token, and type.

• message contains a textual message from gdb, which is not always present. When missing, this is None.
• payload contains the content of gdb's output, which can contain any of the following: dictionary, list, string. This too is not always present, and can be None depending on the response.
• token If an input command was prefixed with a (optional) token then the corresponding output for that command will also be prefixed by that same token. This field is only present for pygdbmi output types nofity and result. When missing, this is None.

The type is defined based on gdb's various mi output record types, and can be

• result - the result of a gdb command, such as done, running, error, etc.
• notify - additional async changes that have occurred, such as breakpoint modified
• console - textual responses to cli commands
• log - debugging messages from gdb's internals
• output - output from target
• target - output from remote target
• done - when gdb has finished its output

Contributing

Documentation fixes, bug fixes, performance improvements, and functional improvements are welcome. You may want to create an issue before beginning work to make sure I am interested in merging it to the master branch.

To develop, set up a new virtual environment, then clone this repo and run pip install -r requirements.txt and pip install -r dev_requirements.txt.

Confirm unit tests are working with make test, then begin development.

Update unit tests as necessary at pygdbmi/tests/test_app.py.

Projects Using pygdbmi

• gdbgui implements a browser-based frontend to gdb, using pygdbmi on the backend
• PINCE is a gdb frontend that aims to provide a reverse engineering tool and a reusable library focused on games. It uses pygdbmi to parse gdb/mi based output for some functions
• avatar² is an orchestration framework for reversing and analysing firmware of embedded devices. It utilizes pygdbmi for internal communication to different analysis targets.
• Know of another project? Create a PR and add it here.

Authors

pygdbmi was written by Chad Smith with contributions from the community for which the author is very grateful. Thanks especially to @mariusmue, @bobthekingofegypt, @mouuff, and @felipesere.