"""Reproducible, offline UI-text audit of copied Parsec binaries. Never executes a Parsec binary. Candidate collection is intentionally broader than visible UI: logs, config keys, formats and widget IDs remain distinguishable. """ import bisect import collections import csv import hashlib import json import re import struct from pathlib import Path import pefile from ui_audit_catalog import GROUPS ROOT=Path(__file__).resolve().parent BASE=ROOT.parent/'analysis/parsec' OUT=BASE/'research/ui' PROGRAM='parsecd-150-104a.dll' HASH='8c8c4299ddf503b2c25c0905b0a0820f3e8238c252e9ace37bf25964ef4a8f6f' def read_tsv(path): with path.open(encoding='utf-8',newline='') as f: return list(csv.DictReader(f,delimiter='\t')) def write(path,value): path.write_text(json.dumps(value,ensure_ascii=False,indent=2)+'\n',encoding='utf-8') def functions(program): result={} for filename in ('decompiled.c','retry-decompiled.c'): path=BASE/'decompiled'/program/filename if not path.exists():continue src=path.read_text(encoding='utf-8') newlines=[m.start() for m in re.finditer('\n',src)] for m in re.finditer(r'/\* ===== ([0-9a-f]+) .*? ===== \*/([\s\S]*?)(?=/\* ===== |\Z)',src): address=m[1] item=dict(address=address,file=str(path.relative_to(ROOT.parent)).replace('\\','/'), line=bisect.bisect_left(newlines,m.start())+1,body=m[0]) if len(m[0])>len(result.get(address,{}).get('body','')):result[address]=item return result def classify(text): if re.match(r'(https?://|/data/|/auth/|/v2/|/me/|/friend)',text):return 'URL・API・외부 데이터' if re.match(r'(client_|host_|app_|encoder_|decoder_|network_|server_|echo_|ignore_)\w+$',text):return '설정 키' if re.search(r'(POPUP|ROOT_WINDOW|BUTTON_MENU|THE_MODAL)',text):return 'widget 내부 ID' if text.startswith(('BG:','Auth success:','CONFIG','Attempted to','A bad ','X-Parsec-', 'Authorization:', 'Content-Type:', 'D:\\','MTY_')):return '진단・구현 문자열' if re.fullmatch(r'[\W\d%sufxc.#+:/ \-]+',text):return '서식・기호' return '화면 문구 후보' def run(): binary=BASE/'binaries/runtime'/PROGRAM assert hashlib.sha256(binary.read_bytes()).hexdigest()==HASH,'Unexpected binary; review address map first' OUT.mkdir(parents=True,exist_ok=True) (OUT/'functions').mkdir(exist_ok=True) xrefs=read_tsv(BASE/'decompiled'/PROGRAM/'string-xrefs.tsv') calls=read_tsv(BASE/'decompiled'/PROGRAM/'calls.tsv') funcs=functions(PROGRAM) declared={f for g in GROUPS for f in g['functions']} # Native app UI component range discovered from the contiguous UI call graph. # This includes dispatch/config code; it is NOT a renderer reachability proof. ui_funcs=declared|{r['function_address'] for r in xrefs if r['function_address'] and 0x180009000<=int(r['function_address'],16)<0x180038000} groups_by_func=collections.defaultdict(list) for g in GROUPS: for f in g['functions']:groups_by_func[f].append(g['id']) by_address=collections.defaultdict(list) for r in xrefs:by_address[r['string_address']].append(r) raw=json.loads((BASE/'reports/runtime__parsecd-150-104a.dll-strings.json').read_text(encoding='utf-8')) candidates={} for r in raw: va=r.get('va') if not va:continue a=int(va,16) if 0x180347600<=a<=0x18034eeb0 or 0x180330000<=a<0x180332da0: candidates[format(a,'x')]=dict(address=format(a,'x'),text=r['text'],encoding=r['encoding'],reason='UI 인접 문자열/선택값 table — 도달 여부 별도 확인') for r in xrefs: if r['function_address'] in ui_funcs: a=r['string_address'] candidates.setdefault(a,dict(address=a,text=r['value'],encoding='ghidra-string',reason='UI/상태 함수의 직접 문자열 참조')) # Decode short DAT_ C strings that a minimum-length strings pass can miss. pe=pefile.PE(str(binary)) short_count=0 for f in sorted(ui_funcs): for a in set(re.findall(r'\bDAT_([0-9a-f]+)\b',funcs.get(f,{}).get('body',''))): if a in candidates:continue va=int(a,16) if not (0x18027f000<=va<0x180380000):continue data=pe.get_data(va-pe.OPTIONAL_HEADER.ImageBase,256).split(b'\0',1)[0] if not data or len(data)>120:continue if all(32<=b<127 for b in data): value=data.decode('ascii') candidates[a]=dict(address=a,text=value,encoding='PE-ascii',reason='DAT 포인터 C 문자열; UI 표시 여부 별도 확인',datFunctions=[f]) short_count+=1 # Prefer complete NUL-terminated text over line fragments in strings reports. for a,c in candidates.items(): if c['encoding'] in ('ascii','PE-ascii','ghidra-string'): data=pe.get_data(int(a,16)-pe.OPTIONAL_HEADER.ImageBase,8192).split(b'\0',1)[0] if data and all(b in (9,10,13) or 32<=b<127 for b in data): c['text']=data.decode('ascii') rows=[] for a,c in sorted(candidates.items()): refs=by_address.get(a,[]) addresses=sorted({r['function_address'] for r in refs if r['function_address']}|set(c.get('datFunctions',[]))) c['kind']=classify(c['text']) c['references']=[dict(function=r['function_address'] or None,site=r['reference_address']) for r in refs] c['functionGroups']=sorted({g for f in addresses for g in groups_by_func[f]}) c['evidence']='함수 참조' if addresses else '데이터 참조' if refs else '문자열만' c['id']='T-'+a rows.append(c) # A line-based strings report can start in the middle of a longer tooltip. # Retain the exact address but identify that overlap explicitly. for index,r in enumerate(rows): addr=int(r['address'],16) parents=[p for p in rows[:index] if int(p['address'],16)len(blob):break n=struct.unpack_from('