gdb-plus 6.1.2

Python library to automate gdb debugging

GDB+

Python library to automate gdb debugging

GDB+ is a wrapper around gdb powered by pwntools. The goal is automate your interactions with gdb and add some extra features.

Main features

• Include a python function as callback when you set a breakpoint.
• Read the canary instead of bruteforcing it every time you need it while testing your exploit.
• Test a specific function of a binary with the parameters you want at any time.
• Log the code of a self modifying function.
• Backup parts of your memory and restore it during future executions.
• Don't waste time commenting your code. The arguments NOPTRACE and REMOTE make the exploit skip any action related to gdb.

Installation

stable

pip3 install gdb_plus

or dev branch

pip3 install git+https://github.com/Angelo942/gdb_plus.git@dev

Warning for pwndbg users:
Previous bugs in Pwndbg used to break the api for python. While most of GDB+ should work with the current version of pwndbg [19/12/2022], pwndbg can not debug both processes after a fork. you are strongly advised to use GEF instead.

Debugging

You can debug a program using the path to the binary.
If you really have to you can also use a process or even just his pid. For pwn challenges set the remote address with Debugger().remote(<host>, <port>) and use the argument REMOTE once you want to exploit the server

from gdb_plus import *

gdbinit = """
handle SIGALRM nopass
"""

dbg = Debugger("./rev_challenge", script=gdbinit, aslr=False)
dbg = Debugger("./pwn_challenge", script=gdbinit).remote("10.10.0.1", 1337)


p = process("./challenge", aslr=False)
dbg = Debugger(p, script=gdbinit)

#pidof process : 3134
dbg = Debugger(3134, script=gdbinit, binary="./challenge")

By default your process will be analysed from the first instruction of the loader. If the process you are debugging has some checks you want to avoid use Debugger(..., debug_from=<address>) to attach with gdb at a specific address.
This will block you script until you reach you address. If you need to execute code in between you can use the non blocking alternative:

Im_done = Event()
dbg = Debugger("./challenge").debug_from("main+0x34", event=Im_done)
dbg.p.sendline("this is my input")
Im_done.set()
dbg.debug_from_done.wait()

This should pass any check for a debugger, even searches for INT3, exept a full check that the memory hasn't been edited.

Calling your script with pwntools arguments NOPTRACE or REMOTE will allow you to disable all actions related to gdb and test your exploit localy or attack the remote target without having to comment anything. If you want a finer control you can use ìf dbg.debugging to discriminate the code that should be executed when gdb is opened or not.

Note
Debugger can also take as parameter a dictionary for the environment variables. You CAN use it to preload libraries, but if you want to do it for the libc I would advise to patch the rpath of the binary instead (if you don't know how take a look at spwn or pwninit. This will prevent problems when running system("/bin/sh") that will fail due to LD_PRELOAD and may hide other problems in your exploit.

Warning
Old versions of gdbserver (< 11.0.50) have problems launching 32bit binaries. If you see a crash trying to find the canary use from_start=False as parameter for the debugger. This will launch the process and then attach to it once the memory has been correctly mapped

Control Flow

The main actions for gdb are already wrapped in individual methods. For all commands not present you can reconstruct them by calling dbg.execute(<command>) as if you where using gdb. Just make sure to use dbg.execute_action(<command>) if your command will require calling dbg.wait().

dbg.step() # Single instruction (will enter function calls)
dbg.next() # Next instruction (will jump over function calls)
dbg.cont() # Continue execution
dbg.finish() # Finish current function
dbg.interrupt() # Stop the execution of your process
dbg.wait() # Wait until you have control of gdb

Note

• dbg.cont(until=ADDRESS) allows you to block your script until you reach a specific address (or symbol). This means that you can debug manually with gdb while your script is waiting without problems

Warning

• finish can only work if the stack frame hasn't been corrupted
• you should specify if you want continue to wait or not with dbg.cont(wait=True/False). We aren't sure about what should be the default behaviour and may set it to wait=True in a future version
• Try avoiding interrupt as much as possible.

If the function modifies itself you may find yourself unable to set breakpoints where you want. To analyse these function we can run them step by step

def MyCallback(dbg):
    if dbg.next_inst.mnemonic == "int3":
        dbg.step()
        dbg.signal("SIGINT")
    print(dbg.next_inst.toString())

dbg.step_until_ret(MyCallback)

In this example at each step the callback will be executed decrypting the next function chunk by chunk and logging the instructions. See this example for more details solving a real challenge.

You could also use dbg.step_until_address(<address>, <callback=None>) if you just want to execute a limited area of code or dbg.step_until_condition(<check>) if you are not sure where to stop.

Breakpoints

Breakpoints have three main features:

• if the address is smaller than 0x10000 the address will immediately be interpreted as relative for PIE binaries
• you can use a symbol name instead of an address such as "main" or "main+0x12"
• you can set callbacks to be executed when the breakpoint is reach and may choose to let the process continue after the execution.

The callback is a function that takes the debugger as parameter and returns a boolean to tell gdb if it should stop or not. The callbacks shouldn't be limited in what you can code. If you find problems raise an issue. If you want to pass data from your callback function to your exploit you can use pointers (lists, dictionaries or queues)

Note
Setting a breakpoint requires the process to be interrupted.

from gdb_plus import *
from queue import Queue

# I let the process run in this example to reinforce the need for the interrupt later
gdbinit = """
handle SIGALRM nopass
"""

dbg = Debugger("./challenge", script=gdbinit).remote("leet.pwn.com", 31337)
pointer = Queue()

# step over an hypothetical call to a function, overwrite the return value and save it in a queue
def MyCallback(dbg):
    dbg.next()
    log.info(f"I tried to return {dbg.rax}")
    pointer.put(dbg.rax)
    dbg.rax = 0
    # Return False to let the process run after executing the callback
    return False

def SecondCallback(dbg):
    log.info("now you can play around with gdb")
    log.info("once you are done execute continue in gdb and your script will resume once we reach main+0x534")
    # Return True to interrupt the process
    return True

# You can not set a breakpoint while the process is running
dbg.breakpoint("main+0x42", callback=MyCallback, temporary = True)
dbg.breakpoint("main+0x124", callback=SecondCallback)
dbg.cont(until="main+0x534")

# Read the data set by the breakpoint
print(pointer_to_secret.get())
dbg.delete_breakpoint("main+0x124")

Memory access

All registers are accessible as properties

dbg.rax = 0xdeadbeefdeadbeef
dbg.eax = 0xfafafafa
dbg.ax  = 0xbabe
dbg.ah = 0x90
assert dbg.rax == 0xdeadbeeffafa90be

You can allocate chunks on the heap (or the bss if you don't have the libc), write and read in the ram and read the canary from anywhere.

pointer = dbg.alloc(8)
dbg.write(pointer, p64(0xdeadbeef))
secret = dbg.read(dbg.elf.symbols["secret"], 0x10)
canary = dbg.canary

Note
While you can access the registers only when the process is at a stop, remember that you can read and write on the memory at any time

Pwntools let you access the address where each library is loaded with p.libs()[<path_to_library>] We have two wrapper for the main ones:

• dbg.base_elf
• dbg.base_libc

from gdb_plus >= 5.4.0 dbg.elf.address is already set to the correct address even with ASLR on, so you may need dbg.base_elf only if you debug a process for wich you don't have the binary

We can also use capstone to know what is the next instruction that will be executed

print(dbg.next_inst.toString()) # "mov rax, r12"
print(dbg.next_inst.mnemonic)   # "mov"

Fork

You can set the debugger to spwn a new instance of gdb every time the process calls fork with dbg.set_split_on_fork().

dbg = Debugger("http_server").set_split_on_fork()
dbg.cont(wait=False)
dbg.p.sendline("input")
pid_child = dbg.wait_split()

# all children are saved in dbg.children and can be accessed by the pid
child = dbg.children[pid_child]

If the program traces its child to make sure you aren't debugging it or to unpack a region of code, you should be able to emulate the calls to ptrace.

child.emulate_ptrace_slave(dbg)
dbg.emulate_ptrace_master(child)

This will interrupt the process at every call to waitpid for the master and SIGSTOP or INT3 for the child. You have to handle yourself when to let each one of them continue while you debug them.

Warning pwndbg can not handle multi-process applications and this section is only possible in native gdb or with GEF

Call functions

If you want to test the effects of a specific function you can directly call it with the parameters you want

pointer = dbg.alloc(100)
# Initialize data
dbg.write(pointer, bytes([i for i in range(100)]))
dbg.call(dbg.p.symbols["obfuscated_pbox"], [pointer, "user_1", 1])
dbg.read(pointer, 100)

See this example for more details

Note
You can pass parameters as strings or byte_arrays. By default they will be saved on the heap with a null terminator in the case of a string. If you can't use the heap set heap=False

Warning
If the stack frame has been corrupted finish() may not work. If this is the case set last address of your function in call(..., end_pointer= ...).

Alternatives

from version 6.0 gdb_plus should be able to script anything you can imagine, but you it can be slow as hell for some uses. This tool is meant to help debugging during challenges, if you only want to automate exploit development you may prefer something like libdebug which doesn't has to communicate with gdb for each command.

TODO

• Add option to use libdebug instead of gdb
• Migrate all wait=False to non blocking functions with events set when finished
• Identify actions performed manually in gdb (overwrite finish and ni)
• Handle fork and ptrace from syscall instead of libc
• Improve ptrace emulation
• Maybe nop split_on_fork&co while debugging is False