Error "Unable to find the text segment base addr" if space present in binary path

[Angelo942]

Hi,
I played with this error for the past two days and pinpointed it to when the local binary passed to gdb contains a space in it's absolute path.

This happens in native gdb and with pwndbg too, but not GEF. Unfortunately GEF as a similar problem when connecting to a gdbserver (hugsy/gef#901 (comment)) so I have to work around it anyway.

I know that gdb has always had troubles with paths containing spaces, but it's the first time after a year working with this setup that I notice one so I don't know if the problems lies in gdb or this library.

[mahaloz]

Hi @Angelo942, to clarify, you are running a local binary (not remote), and you are getting this error if a space is in the binary path?

[mahaloz]

My immediate hunch is that this line is the cause of this issue. If you wanted to debug it before I can get to it (which will likely be this weekend), you can set a breakpoint here and see what's going on:

decomp2dbg/decomp2dbg/clients/gdb/utils.py

Line 83 in 1073f22

return path_name == get_filepath()

This is if you are running locally.

[Angelo942]

I tested it both debugging a local process and a remote one. The bug is present every time we pass a copy of the binary to gdb.

so gdb ./challenge will never work even if we connect to a gdbserver, but for remote processes just gdb will work fine.

I took a look at the code and we have a problem here, where the path_name used is truncated:

decomp2dbg/decomp2dbg/clients/gdb/utils.py

Line 60 in 1073f22

path = mapping.split(" ")[-1]

I tried changing the parse to:

def vmmap_base_addrs():
    addr_maps = {}
    mappings = gdb.execute("info proc mappings", to_string=True).split("\n")
    for mapping in mappings:    
        try:
            addr = int(re.findall(r"0x[0-9a-fA-F]+", mapping)[0], 16)
            path = "/" + "/".join(mapping.split("/")[1:])
        except IndexError:
            continue

        # always use the lowest addr
        if path in addr_maps or path == "/":
            continue

        if addr and path:
            addr_maps[path] = addr

    return addr_maps

This solves the problem for local processes, but not for remote ones. (I focus on that part because gdb ./challenge is how pwntools' gdb.debug launches gdb before connecting to the gdbserver)

decomp2dbg/decomp2dbg/clients/gdb/utils.py

Line 91 in 1073f22

if other_file_hash == binary_hash:

Now I'm trying to figure out why the hash is right with gdb, but not with gdb ./challenge.
The only think I noticed so far is that is that in the first case the hash changes every time, while in the second both hashes [other_file_hash and binary_hash] are constant but different. I tried with a path that doesn't contains any spaces and in that case the hashes are all constant.

[mahaloz]

Linking this GEF issue which may fix some stuff for us: hugsy/gef#999

[Grazfather]

This one too hugsy/gef#998