notParsec / 문서전체 문서

생성 코드·템플릿

extract_ui_evidence.py

design/extract_ui_evidence.py

출처: 현재 작업 폴더 · 원문 다운로드

"""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)<addr<int(p['address'],16)+len(p['text'])
            and p['text'][addr-int(p['address'],16):]==r['text']]
        if parents:
            parent=parents[0];r['fragmentOf']=parent['id']
            r['functionGroups']=sorted(set(r['functionGroups'])|set(parent['functionGroups']))
    manifest=[]
    for address in sorted(ui_funcs):
        f=funcs.get(address)
        if not f:continue
        filename='functions/'+address+'.c'
        (OUT/filename).write_text(f['body'],encoding='utf-8')
        manifest.append(dict(address=address,sourceFile=f['file'],sourceLine=f['line'],extractedFile=filename,
            groups=groups_by_func[address],callees=sorted({r['callee_address'] for r in calls if r['caller_address']==address})))
    catalogue=[]
    for g in GROUPS:
        examples=[]
        for anchor in g['anchors']:
            matched=[r for r in rows if anchor.lower() in r['text'].lower() and
                any(ref['function'] in g['functions'] for ref in r['references'])]
            if matched:
                r=min(matched,key=lambda x:len(x['text']))
                examples.append(dict(id=r['id'],text=r['text'],address=r['address']))
        catalogue.append({**g,'examples':examples})
    # Audit all existing component string reports, preserving versions separately.
    report_inventory=[]
    for path in sorted((BASE/'reports').glob('*-strings.json')):
        records=json.loads(path.read_text(encoding='utf-8'))
        report_inventory.append(dict(file=str(path.relative_to(ROOT.parent)).replace('\\','/'),
            strings=len(records),sha256=hashlib.sha256(path.read_bytes()).hexdigest()))
    components=[]
    for program,report in [('parsecd.exe','app__parsecd.exe'),('pservice.exe','app__pservice.exe'),('teams.exe','app__teams.exe'),
            ('app__vdd__parsec-vdd.exe','app__vdd__parsec-vdd.exe'),('app__vusb__parsec-vud.exe','app__vusb__parsec-vud.exe')]:
        strings=json.loads((BASE/'reports'/(report+'-strings.json')).read_text(encoding='utf-8'))
        cr=read_tsv(BASE/'decompiled'/program/'string-xrefs.tsv')
        selected=[]
        for s in strings:
            text=s['text']
            if re.search(r'(?i)cannot load parsec|out of date or corrupt|File System Error -|incompatible with your machine|please contact support|Teams Computer Error:|Please wait while Setup|Installer integrity check|Error launching installer|Error writing temporary',text):
                addr=(s.get('va') or '').removeprefix('0x').lstrip('0')
                refs=[dict(function=r['function_address'],site=r['reference_address']) for r in cr if r['string_address'].lstrip('0')==addr]
                selected.append(dict(text=text,address=s.get('va'),references=refs,
                    kind='Teams CLI 진단' if program=='teams.exe' else 'loader/installer 문구 후보'))
        components.append(dict(program=program,totalStrings=len(strings),candidates=selected,
            handling='F70 loader / F69 setup / Teams CLI는 개인용 제외; service 로그는 사용자 UI 증거로 간주하지 않음'))
    # RT_STRING and RT_DIALOG resource texts. NSIS compressed language tables
    # are outside these PE resources and remain an explicit coverage gap.
    resources=[]
    for path in sorted((BASE/'binaries').rglob('*')):
        if path.suffix.lower() not in ('.exe','.dll','.sys'):continue
        p=pefile.PE(str(path));entries=[]
        for typ in getattr(getattr(p,'DIRECTORY_ENTRY_RESOURCE',None),'entries',[]):
            if typ.id not in (5,6):continue
            for name in typ.directory.entries:
                for lang in name.directory.entries:
                    d=lang.data.struct;blob=p.get_data(d.OffsetToData,d.Size)
                    strings=[]
                    if typ.id==6:
                        pos=0
                        for i in range(16):
                            if pos+2>len(blob):break
                            n=struct.unpack_from('<H',blob,pos)[0];pos+=2
                            s=blob[pos:pos+2*n].decode('utf-16-le',errors='replace');pos+=2*n
                            if s:strings.append(dict(id=(name.id-1)*16+i,text=s))
                    else:
                        for m in re.finditer(rb'(?:[\x20-\x7e]\x00){2,}',blob):strings.append(dict(offset=m.start(),text=m[0].decode('utf-16-le')))
                    entries.append(dict(type='RT_STRING' if typ.id==6 else 'RT_DIALOG UTF-16 snippets (not full template parser)',name=str(name.name or name.id),language=lang.id,strings=strings))
        resources.append(dict(binary=str(path.relative_to(BASE)).replace('\\','/'),sha256=hashlib.sha256(path.read_bytes()).hexdigest(),resources=entries))
        p.close()
    stats=dict(rawCoreStrings=len(raw),xrefRecords=len(xrefs),candidateStrings=len(rows),
        referencedCandidates=sum(r['evidence']=='함수 참조' for r in rows),dataReferencedCandidates=sum(r['evidence']=='데이터 참조' for r in rows),textOnlyCandidates=sum(r['evidence']=='문자열만' for r in rows),
        overlappingFragments=sum('fragmentOf' in r for r in rows),
        functions=len(manifest),flows=len(catalogue),datDecoded=short_count,
        componentReports=len(report_inventory),componentStrings=sum(r['strings'] for r in report_inventory),
        resourceBinaries=len(resources),resourceTextSnippets=sum(len(e['strings']) for r in resources for e in r['resources']),
        kinds=dict(collections.Counter(r['kind'] for r in rows)),dispositions=dict(collections.Counter(g['result'] for g in catalogue)))
    evidence=dict(schemaVersion=1,program=PROGRAM,sha256=HASH,stats=stats,candidates=rows,functions=manifest,
        componentReports=report_inventory,components=components,resources=resources,
        limitations=['함수별 범위/anchor 추적이며 모든 string의 렌더링 도달성을 증명한 결과가 아님',
            '분기 조건을 실제 Parsec UI에서 실행하지 않음; Windows 설치본의 macOS 문자열도 runtime 지원 증거가 아님',
            '서버 오류 사전·changelog·웹 UI 번들·동적 장치명/사용자명은 로컬 텍스트만으로 완전 복원 불가',
            'NSIS 압축 language table과 installer 페이지 전체는 미복원; PE resource/stub 문구만 확인',
            '난독화/압축/실행 중 조합된 문구나 무참조 data table은 추가 검증 필요; UTF-16 dialog 추출은 snippet 방식',
            '전체 componentStrings는 복사본/버전 중복을 포함한 검사량이지 고유 UI 문구 수가 아님'])
    write(OUT/'ui-text-evidence.json',evidence)
    write(ROOT/'ui-text-catalog.json',dict(version='2026-09-20.3',program=PROGRAM,sha256=HASH,stats=stats,flows=catalogue,
        limitations=evidence['limitations'],evidencePath='../analysis/parsec/research/ui/ui-text-evidence.json'))
    with (OUT/'ui-texts.tsv').open('w',encoding='utf-8-sig',newline='') as f:
        writer=csv.writer(f,delimiter='\t');writer.writerow(['id','address','text','kind','evidence','function_groups','functions'])
        for r in rows:writer.writerow([r['id'],r['address'],r['text'],r['kind'],r['evidence'],','.join(r['functionGroups']),','.join(sorted({ref['function'] for ref in r['references'] if ref['function']}))])
    print(json.dumps(stats,ensure_ascii=False,indent=2))
    empty=[g['id'] for g in catalogue if not g['examples']]
    if empty:raise RuntimeError('Groups with no matching anchor evidence: '+str(empty))

if __name__=='__main__':run()