isle/tools/reccmp/reccmp.py
MS ce68a7b1f4
Reccmp comparison engine refactor (#405)
* Reccmp comparison engine refactor

* Remove redundant references to 'entry' symbol
2024-01-04 18:12:55 -05:00

456 lines
15 KiB
Python
Executable file

#!/usr/bin/env python3
import argparse
import base64
import difflib
import json
import logging
import os
import re
from isledecomp import (
Bin,
get_file_in_script_dir,
OffsetPlaceholderGenerator,
print_diff,
)
from isledecomp.compare import Compare as IsleCompare
from capstone import Cs, CS_ARCH_X86, CS_MODE_32
import colorama
from pystache import Renderer
REGISTER_LIST = set(
[
"ax",
"bp",
"bx",
"cx",
"di",
"dx",
"eax",
"ebp",
"ebx",
"ecx",
"edi",
"edx",
"esi",
"esp",
"si",
"sp",
]
)
WORDS = re.compile(r"\w+")
def sanitize(file, placeholder_generator, mnemonic, op_str):
op_str_is_number = False
try:
int(op_str, 16)
op_str_is_number = True
except ValueError:
pass
if (mnemonic in ["call", "jmp"]) and op_str_is_number:
# Filter out "calls" because the offsets we're not currently trying to
# match offsets. As long as there's a call in the right place, it's
# probably accurate.
op_str = placeholder_generator.get(int(op_str, 16))
else:
def filter_out_ptr(ptype, op_str):
try:
ptrstr = ptype + " ptr ["
start = op_str.index(ptrstr) + len(ptrstr)
end = op_str.index("]", start)
# This will throw ValueError if not hex
inttest = int(op_str[start:end], 16)
return (
op_str[0:start] + placeholder_generator.get(inttest) + op_str[end:]
)
except ValueError:
return op_str
# Filter out dword ptrs where the pointer is to an offset
op_str = filter_out_ptr("dword", op_str)
op_str = filter_out_ptr("word", op_str)
op_str = filter_out_ptr("byte", op_str)
# Use heuristics to filter out any args that look like offsets
words = op_str.split(" ")
for i, word in enumerate(words):
try:
inttest = int(word, 16)
if file.is_relocated_addr(inttest):
words[i] = placeholder_generator.get(inttest)
except ValueError:
pass
op_str = " ".join(words)
return mnemonic, op_str
def parse_asm(disassembler, file, asm_addr, size):
asm = []
data = file.read(asm_addr, size)
placeholder_generator = OffsetPlaceholderGenerator()
for i in disassembler.disasm(data, 0):
# Use heuristics to disregard some differences that aren't representative
# of the accuracy of a function (e.g. global offsets)
mnemonic, op_str = sanitize(file, placeholder_generator, i.mnemonic, i.op_str)
if op_str is None:
asm.append(mnemonic)
else:
asm.append(f"{mnemonic} {op_str}")
return asm
def get_registers(line: str):
to_replace = []
# use words regex to find all matching positions:
for match in WORDS.finditer(line):
reg = match.group(0)
if reg in REGISTER_LIST:
to_replace.append((reg, match.start()))
return to_replace
def replace_register(
lines: list[str], start_line: int, reg: str, replacement: str
) -> list[str]:
return [
line.replace(reg, replacement) if i >= start_line else line
for i, line in enumerate(lines)
]
# Is it possible to make new_asm the same as original_asm by swapping registers?
def can_resolve_register_differences(original_asm, new_asm):
# Split the ASM on spaces to get more granularity, and so
# that we don't modify the original arrays passed in.
original_asm = [part for line in original_asm for part in line.split()]
new_asm = [part for line in new_asm for part in line.split()]
# Swapping ain't gonna help if the lengths are different
if len(original_asm) != len(new_asm):
return False
# Look for the mismatching lines
for i, original_line in enumerate(original_asm):
new_line = new_asm[i]
if new_line != original_line:
# Find all the registers to replace
to_replace = get_registers(original_line)
for replace in to_replace:
(reg, reg_index) = replace
replacing_reg = new_line[reg_index : reg_index + len(reg)]
if replacing_reg in REGISTER_LIST:
if replacing_reg != reg:
# Do a three-way swap replacing in all the subsequent lines
temp_reg = "&" * len(reg)
new_asm = replace_register(new_asm, i, replacing_reg, temp_reg)
new_asm = replace_register(new_asm, i, reg, replacing_reg)
new_asm = replace_register(new_asm, i, temp_reg, reg)
else:
# No replacement to do, different code, bail out
return False
# Check if the lines are now the same
for i, original_line in enumerate(original_asm):
if new_asm[i] != original_line:
return False
return True
def gen_html(html_file, data):
output_data = Renderer().render_path(
get_file_in_script_dir("template.html"), {"data": data}
)
with open(html_file, "w", encoding="utf-8") as htmlfile:
htmlfile.write(output_data)
def gen_svg(svg_file, name_svg, icon, svg_implemented_funcs, total_funcs, raw_accuracy):
icon_data = None
if icon:
with open(icon, "rb") as iconfile:
icon_data = base64.b64encode(iconfile.read()).decode("utf-8")
total_statistic = raw_accuracy / total_funcs
full_percentbar_width = 127.18422
output_data = Renderer().render_path(
get_file_in_script_dir("template.svg"),
{
"name": name_svg,
"icon": icon_data,
"implemented": f"{(svg_implemented_funcs / total_funcs * 100):.2f}% ({svg_implemented_funcs}/{total_funcs})",
"accuracy": f"{(raw_accuracy / svg_implemented_funcs * 100):.2f}%",
"progbar": total_statistic * full_percentbar_width,
"percent": f"{(total_statistic * 100):.2f}%",
},
)
with open(svg_file, "w", encoding="utf-8") as svgfile:
svgfile.write(output_data)
# Do the actual work
def main():
# pylint: disable=too-many-locals, too-many-nested-blocks, too-many-branches, too-many-statements
parser = argparse.ArgumentParser(
allow_abbrev=False,
description="Recompilation Compare: compare an original EXE with a recompiled EXE + PDB.",
)
parser.add_argument(
"original", metavar="original-binary", help="The original binary"
)
parser.add_argument(
"recompiled", metavar="recompiled-binary", help="The recompiled binary"
)
parser.add_argument(
"pdb", metavar="recompiled-pdb", help="The PDB of the recompiled binary"
)
parser.add_argument(
"decomp_dir", metavar="decomp-dir", help="The decompiled source tree"
)
parser.add_argument(
"--total",
"-T",
metavar="<count>",
help="Total number of expected functions (improves total accuracy statistic)",
)
parser.add_argument(
"--verbose",
"-v",
metavar="<offset>",
help="Print assembly diff for specific function (original file's offset)",
)
parser.add_argument(
"--html",
"-H",
metavar="<file>",
help="Generate searchable HTML summary of status and diffs",
)
parser.add_argument(
"--no-color", "-n", action="store_true", help="Do not color the output"
)
parser.add_argument(
"--svg", "-S", metavar="<file>", help="Generate SVG graphic of progress"
)
parser.add_argument("--svg-icon", metavar="icon", help="Icon to use in SVG (PNG)")
parser.add_argument(
"--print-rec-addr",
action="store_true",
help="Print addresses of recompiled functions too",
)
parser.set_defaults(loglevel=logging.INFO)
parser.add_argument(
"--debug",
action="store_const",
const=logging.DEBUG,
dest="loglevel",
help="Print script debug information",
)
args = parser.parse_args()
logging.basicConfig(level=args.loglevel, format="[%(levelname)s] %(message)s")
colorama.init()
verbose = None
found_verbose_target = False
if args.verbose:
try:
verbose = int(args.verbose, 16)
except ValueError:
parser.error("invalid verbose argument")
html_path = args.html
plain = args.no_color
original = args.original
if not os.path.isfile(original):
parser.error(f"Original binary {original} does not exist")
recomp = args.recompiled
if not os.path.isfile(recomp):
parser.error(f"Recompiled binary {recomp} does not exist")
syms = args.pdb
if not os.path.isfile(syms):
parser.error(f"Symbols PDB {syms} does not exist")
source = args.decomp_dir
if not os.path.isdir(source):
parser.error(f"Source directory {source} does not exist")
svg = args.svg
with Bin(original, find_str=True) as origfile, Bin(recomp) as recompfile:
if verbose is not None:
# Mute logger events from compare engine
logging.getLogger("isledecomp.compare.db").setLevel(logging.CRITICAL)
logging.getLogger("isledecomp.compare.lines").setLevel(logging.CRITICAL)
isle_compare = IsleCompare(origfile, recompfile, syms, source)
print()
capstone_disassembler = Cs(CS_ARCH_X86, CS_MODE_32)
function_count = 0
total_accuracy = 0
total_effective_accuracy = 0
htmlinsert = []
matches = []
if verbose is not None:
match = isle_compare.get_one_function(verbose)
if match is not None:
found_verbose_target = True
matches = [match]
else:
matches = isle_compare.get_functions()
for match in matches:
# The effective_ratio is the ratio when ignoring differing register
# allocation vs the ratio is the true ratio.
ratio = 0.0
effective_ratio = 0.0
if match.size:
origasm = parse_asm(
capstone_disassembler,
origfile,
match.orig_addr,
match.size,
)
recompasm = parse_asm(
capstone_disassembler,
recompfile,
match.recomp_addr,
match.size,
)
diff = difflib.SequenceMatcher(None, origasm, recompasm)
ratio = diff.ratio()
effective_ratio = ratio
if ratio != 1.0:
# Check whether we can resolve register swaps which are actually
# perfect matches modulo compiler entropy.
if can_resolve_register_differences(origasm, recompasm):
effective_ratio = 1.0
else:
ratio = 0
percenttext = f"{(effective_ratio * 100):.2f}%"
if not plain:
if effective_ratio == 1.0:
percenttext = (
colorama.Fore.GREEN + percenttext + colorama.Style.RESET_ALL
)
elif effective_ratio > 0.8:
percenttext = (
colorama.Fore.YELLOW + percenttext + colorama.Style.RESET_ALL
)
else:
percenttext = (
colorama.Fore.RED + percenttext + colorama.Style.RESET_ALL
)
if effective_ratio == 1.0 and ratio != 1.0:
if plain:
percenttext += "*"
else:
percenttext += colorama.Fore.RED + "*" + colorama.Style.RESET_ALL
if args.print_rec_addr:
addrs = f"0x{match.orig_addr:x} / 0x{match.recomp_addr:x}"
else:
addrs = hex(match.orig_addr)
if not verbose:
print(
f" {match.name} ({addrs}) is {percenttext} similar to the original"
)
function_count += 1
total_accuracy += ratio
total_effective_accuracy += effective_ratio
if match.size:
udiff = difflib.unified_diff(origasm, recompasm, n=10)
# If verbose, print the diff for that function to the output
if verbose:
if effective_ratio == 1.0:
ok_text = (
"OK!"
if plain
else (
colorama.Fore.GREEN
+ "✨ OK! ✨"
+ colorama.Style.RESET_ALL
)
)
if ratio == 1.0:
print(f"{addrs}: {match.name} 100% match.\n\n{ok_text}\n\n")
else:
print(
f"{addrs}: {match.name} Effective 100%% match. (Differs in register allocation only)\n\n{ok_text} (still differs in register allocation)\n\n"
)
else:
print_diff(udiff, plain)
print(
f"\n{match.name} is only {percenttext} similar to the original, diff above"
)
# If html, record the diffs to an HTML file
if html_path:
htmlinsert.append(
{
"address": f"0x{match.orig_addr:x}",
"name": match.name,
"matching": effective_ratio,
"diff": "\n".join(udiff),
}
)
if html_path:
gen_html(html_path, json.dumps(htmlinsert))
if verbose:
if not found_verbose_target:
print(f"Failed to find the function with address 0x{verbose:x}")
else:
implemented_funcs = function_count
if args.total:
function_count = int(args.total)
if function_count > 0:
effective_accuracy = total_effective_accuracy / function_count * 100
actual_accuracy = total_accuracy / function_count * 100
print(
f"\nTotal effective accuracy {effective_accuracy:.2f}% across {function_count} functions ({actual_accuracy:.2f}% actual accuracy)"
)
if svg:
gen_svg(
svg,
os.path.basename(original),
args.svg_icon,
implemented_funcs,
function_count,
total_effective_accuracy,
)
if __name__ == "__main__":
raise SystemExit(main())