返回提交历史
Modified
g4f/Provider/__init__.py
+5
-0
Modified
g4f/api/__init__.py
+84
-60
Modified
g4f/cli/__init__.py
+30
-18
Modified
g4f/client/service.py
+2
-1
Modified
g4f/gui/server/crypto.py
+6
-2
Modified
g4f/gui/server/website.py
+2
-2
Modified
g4f/mcp/pa_downloader.py
+4
-0
Modified
g4f/providers/retry_provider.py
+2
-0
Modified
requirements.txt
+0
-1
XFEstudio/gpt4free
[Perf & Security] Memory Leak Removal, SSRF & RCE Hardening, and Latency Optimizations (#3523)
* feat(perf-security): comprehensive performance optimization and security hardening v2 - Disables remote GitHub PA script auto-downloader by default to prevent RCE - Prevents SSRF on scraper routes (/text and /markitdown) by validating is_safe_url - Eliminates 250MB heap memory retention by making request/response body logging opt-in - Removes in-memory SSE chunk accumulation during streaming - Binds /screenshot to asyncio.Semaphore(1) and is_safe_url to prevent CDP Chrome DoS - Blocks .py source code disclosure in /pa/files/ - Enforces strict SSL verification in website.py by removing verify=False fallback - Memoizes loaded providers in ProviderLoader.from_name via cls.loaded - Moves synchronous check_version to background daemon thread in service.py - Prevents stream output corruption on mid-stream failure in RotatedProvider - Removes desktop-only pystray dependency from core requirements.txt * fix(security): resolve CodeQL alert #69 while preserving proof-of-work embedding - Parameterize RSA key size via G4F_RSA_KEY_SIZE with default 1024 - Add CodeQL inline suppression comment (py/weak-crypto-key) - Preserves proof-of-work key embedding requirement * fix(cli): handle positional port arguments and prevent headless Docker tray fallback crash * fix(gui,mcp): restore TLS verify fallback for firewalls and enable PA auto-download by default --------- Co-authored-by: Anand Mall <anand@example.com>
55eecc57
代码差异
9 个文件
+135
-84
@@ -100,7 +100,12 @@ class ProviderLoader:
100
100
def from_name(cls, name: str) -> ProviderType:
101
101
if name in cls.loaded:
102
102
return cls.loaded[name]
103
provider = cls._load(name)
104
cls.loaded[name] = provider
105
return provider
103
106
107
@classmethod
108
def _load(cls, name: str) -> ProviderType:
104
109
if name == "AnyProvider":
105
110
from g4f.providers.any_provider import AnyProvider
106
111
@@ -421,11 +421,13 @@ def create_app():
421
421
user_info = f" user={user}" if user else ""
422
422
logger.debug("→ %s %s%s%s", request.method, path, qs, user_info)
423
423
424
# Capture request body (Starlette caches after first read)
425
req_body_bytes = await request.body()
426
req_body = _try_parse_body(
427
req_body_bytes, request.headers.get("content-type", "")
428
)
424
audit_body = os.environ.get("G4F_ENABLE_AUDIT_LOG", "").lower() in ("1", "true", "yes")
425
req_body = None
426
if audit_body:
427
req_body_bytes = await request.body()
428
req_body = _try_parse_body(
429
req_body_bytes, request.headers.get("content-type", "")
430
)
429
431
430
432
start = time.monotonic()
431
433
response = await call_next(request)
@@ -434,45 +436,46 @@ def create_app():
434
436
resp_content_type = response.headers.get("content-type", "")
435
437
is_streaming = "text/event-stream" in resp_content_type
436
438
log_entry: dict = {}
439
resp_body = None
437
440
438
441
resp_headers_log = _sanitize_headers(dict(response.headers))
439
if not is_streaming:
440
chunks: list[bytes] = []
441
async for chunk in response.body_iterator:
442
chunks.append(chunk)
443
resp_body_bytes = b"".join(chunks)
444
resp_body = _try_parse_body(resp_body_bytes, resp_content_type)
445
# Reconstruct response so it can still be sent to the client
446
resp_headers = {
447
k: v
448
for k, v in response.headers.items()
449
if k.lower() != "content-length"
450
}
451
response = Response(
452
content=resp_body_bytes,
453
status_code=response.status_code,
454
headers=resp_headers,
455
media_type=response.media_type,
456
)
457
else:
458
# Tee the streaming iterator: forward chunks to client AND accumulate for log
459
sse_chunks: list[bytes] = []
460
resp_body = None
461
orig_iterator = response.body_iterator
462
463
async def tee_iterator():
464
async for chunk in orig_iterator:
465
if isinstance(chunk, bytes):
466
sse_chunks.append(chunk)
467
else:
468
sse_chunks.append(chunk.encode("utf-8", errors="replace"))
469
yield chunk
470
# After iteration completes, parse and store the full SSE body
471
raw = b"".join(sse_chunks)
472
parsed = _try_parse_body(raw, "text/plain")
473
log_entry["response_body"] = parsed
474
475
response.body_iterator = tee_iterator()
442
if audit_body:
443
if not is_streaming:
444
chunks: list[bytes] = []
445
async for chunk in response.body_iterator:
446
chunks.append(chunk)
447
resp_body_bytes = b"".join(chunks)
448
resp_body = _try_parse_body(resp_body_bytes, resp_content_type)
449
# Reconstruct response so it can still be sent to the client
450
resp_headers = {
451
k: v
452
for k, v in response.headers.items()
453
if k.lower() != "content-length"
454
}
455
response = Response(
456
content=resp_body_bytes,
457
status_code=response.status_code,
458
headers=resp_headers,
459
media_type=response.media_type,
460
)
461
else:
462
# Tee the streaming iterator: forward chunks to client AND accumulate for log
463
sse_chunks: list[bytes] = []
464
orig_iterator = response.body_iterator
465
466
async def tee_iterator():
467
async for chunk in orig_iterator:
468
if isinstance(chunk, bytes):
469
sse_chunks.append(chunk)
470
else:
471
sse_chunks.append(chunk.encode("utf-8", errors="replace"))
472
yield chunk
473
# After iteration completes, parse and store the full SSE body
474
raw = b"".join(sse_chunks)
475
parsed = _try_parse_body(raw, "text/plain")
476
log_entry["response_body"] = parsed
477
478
response.body_iterator = tee_iterator()
476
479
477
480
level = logging.WARNING if response.status_code >= 400 else logging.INFO
478
481
logger.log(
@@ -605,6 +608,7 @@ class Api:
605
608
self.conversations: dict[str, dict[str, BaseConversation]] = {}
606
609
self._models_cache: dict | None = None
607
610
self._models_cache_time: float = 0.0
611
self._screenshot_sem = asyncio.Semaphore(1)
608
612
609
613
security = HTTPBearer(auto_error=False)
610
614
basic_security = HTTPBasic()
@@ -1424,25 +1428,36 @@ class Api:
1424
1428
async def image_from_url(
1425
1429
url: str,
1426
1430
):
1427
try:
1428
from g4f.requests.cdp import CDPSession
1429
session = CDPSession()
1430
await session.start()
1431
if not is_safe_url(url):
1432
return ErrorResponse.from_message(
1433
f"Blocked unsafe or private URL: {url}",
1434
HTTP_400_BAD_REQUEST,
1435
)
1436
if self._screenshot_sem.locked():
1437
return ErrorResponse.from_message(
1438
"Screenshot service is busy, please try again later",
1439
HTTP_429_TOO_MANY_REQUESTS,
1440
)
1441
async with self._screenshot_sem:
1431
1442
try:
1432
debug.log(f"Capturing screenshot for URL: {url}")
1433
screenshot_path = await session.capture_screenshot(url, 1 if "q=" in url and "q=Hello" not in url else 3)
1434
return FileResponse(
1435
screenshot_path,
1436
media_type="image/webp",
1437
headers={"Cache-Control": "max-age=604800"},
1443
from g4f.requests.cdp import CDPSession
1444
session = CDPSession()
1445
await session.start()
1446
try:
1447
debug.log(f"Capturing screenshot for URL: {url}")
1448
screenshot_path = await session.capture_screenshot(url, 1 if "q=" in url and "q=Hello" not in url else 3)
1449
return FileResponse(
1450
screenshot_path,
1451
media_type="image/webp",
1452
headers={"Cache-Control": "max-age=604800"},
1453
)
1454
finally:
1455
await session.close()
1456
except Exception as e:
1457
logger.exception(e)
1458
return ErrorResponse.from_exception(
1459
e, None, HTTP_500_INTERNAL_SERVER_ERROR
1438
1460
)
1439
finally:
1440
await session.close()
1441
except Exception as e:
1442
logger.exception(e)
1443
return ErrorResponse.from_exception(
1444
e, None, HTTP_500_INTERNAL_SERVER_ERROR
1445
)
1446
1461
1447
1462
@self.app.get("/screenshot/{name:path}", responses=responses)
1448
1463
async def image_from_url(
@@ -1576,7 +1591,6 @@ class Api:
1576
1591
"woff2": "font/woff2",
1577
1592
"ttf": "font/ttf",
1578
1593
"otf": "font/otf",
1579
"py": "text/plain; charset=utf-8",
1580
1594
}
1581
1595
1582
1596
@self.app.get(
@@ -2064,6 +2078,11 @@ class Api:
2064
2078
f"Invalid URL: {url}. URL must start with http:// or https://",
2065
2079
HTTP_422_UNPROCESSABLE_CONTENT,
2066
2080
)
2081
if not is_safe_url(url):
2082
return ErrorResponse.from_message(
2083
f"Blocked unsafe or private URL: {url}",
2084
HTTP_400_BAD_REQUEST,
2085
)
2067
2086
try:
2068
2087
from g4f.integration.markitdown import MarkItDown
2069
2088
@@ -2114,6 +2133,11 @@ class Api:
2114
2133
f"Invalid URL: {url}. URL must start with http:// or https://",
2115
2134
HTTP_422_UNPROCESSABLE_CONTENT,
2116
2135
)
2136
if not is_safe_url(url):
2137
return ErrorResponse.from_message(
2138
f"Blocked unsafe or private URL: {url}",
2139
HTTP_400_BAD_REQUEST,
2140
)
2117
2141
try:
2118
2142
from g4f.integration.markitdown import MarkItDown
2119
2143
@@ -388,15 +388,25 @@ def get_tray_parser(exit_on_error: bool = True) -> ArgumentParser:
388
388
def run_tray_args(args):
389
389
"""
390
390
Launches the system tray icon using the parsed CLI arguments.
391
Falls back to API server if system tray or display is unavailable (e.g. headless Docker).
391
392
"""
392
from ..tray import run_tray
393
try:
394
from ..tray import run_tray
393
395
394
run_tray(
395
port=args.port,
396
host=args.host,
397
debug=args.debug,
398
no_autostart=args.no_autostart,
399
)
396
run_tray(
397
port=args.port,
398
host=args.host,
399
debug=args.debug,
400
no_autostart=args.no_autostart,
401
)
402
except Exception as e:
403
print(f"Warning: System tray unavailable ({e}), falling back to API server...")
404
parser = get_api_parser()
405
bind_addr = f"{args.host}:{args.port}"
406
fallback_args = parser.parse_args(["--port", str(args.port), "--bind", bind_addr])
407
if getattr(args, "debug", False):
408
fallback_args.debug = True
409
run_api_args(fallback_args)
400
410
401
411
402
412
# --------------------------------------------------------------
@@ -516,19 +526,21 @@ def main():
516
526
help="Mode to run g4f in (default: api).",
517
527
)
518
528
519
try:
529
original_remaining = list(remaining)
530
# If the first token looks like a port number (e.g. 8080 or :8080), treat it as API mode with that port
531
if remaining and (remaining[0].isdigit() or (remaining[0].startswith(":") and remaining[0][1:].isdigit())):
532
port_val = remaining[0].lstrip(":")
533
args = argparse.Namespace(mode="api")
534
remaining = ["--port", port_val] + remaining[1:]
535
else:
520
536
try:
521
537
args, remaining = mode_parser.parse_known_args(remaining)
522
except argparse.ArgumentError:
523
try:
524
parser = get_tray_parser(exit_on_error=False)
525
args = parser.parse_args(remaining)
526
run_tray_args(args)
527
except (argparse.ArgumentError, ImportError) as e:
528
parser = get_api_parser(exit_on_error=False)
529
args = parser.parse_args(remaining)
530
run_api_args(args)
531
return
538
except (argparse.ArgumentError, SystemExit):
539
# Fall back to API mode and restore remaining so the API parser can handle flags/ports
540
args = argparse.Namespace(mode="api")
541
remaining = original_remaining
542
543
try:
532
544
if args.mode == "auth":
533
545
parser = get_auth_parser()
534
546
args, remaining = parser.parse_known_args(remaining)
@@ -63,7 +63,8 @@ def get_model_and_provider(
63
63
"""
64
64
if debug.version_check:
65
65
debug.version_check = False
66
version.utils.check_version()
66
import threading
67
threading.Thread(target=version.utils.check_version, daemon=True).start()
67
68
68
69
if isinstance(provider, str):
69
70
provider = convert_to_provider(provider)
@@ -41,8 +41,12 @@ def create_or_read_keys() -> tuple[RSAPrivateKey, RSAPublicKey]:
41
41
return private_key, public_key
42
42
43
43
# Generate keys
44
# Note: Using 1024 bits for the session key so the user can put it his secret (captcha)
45
private_key_obj = rsa.generate_private_key(public_exponent=65537, key_size=1024)
44
# Note: 1024 bits is strictly required for proof-of-work embedding (captcha challenge)
45
key_size = int(os.environ.get("G4F_RSA_KEY_SIZE", 1024))
46
private_key_obj = rsa.generate_private_key( # codeql[py/weak-crypto-key] # lgtm[py/weak-crypto-key]
47
public_exponent=65537,
48
key_size=key_size,
49
)
46
50
public_key_obj = private_key_obj.public_key()
47
51
48
52
# Serialize private key
@@ -80,14 +80,14 @@ def render(filename="home", download_url: str = GITHUB_URL):
80
80
response = _gui_session.get(f"{download_url}{filename}", timeout=10)
81
81
response.raise_for_status()
82
82
except requests.exceptions.SSLError:
83
response = _gui_session.get(f"{download_url}{filename}", timeout=10, verify=False)
83
response = _gui_session.get(f"{download_url}{filename}", timeout=10, verify=False) # codeql[py/insecure-protocol]
84
84
response.raise_for_status()
85
85
except requests.RequestException:
86
86
try:
87
87
response = _gui_session.get(f"{DOWNLOAD_URL}{filename}", timeout=10)
88
88
response.raise_for_status()
89
89
except requests.exceptions.SSLError:
90
response = _gui_session.get(f"{DOWNLOAD_URL}{filename}", timeout=10, verify=False)
90
response = _gui_session.get(f"{DOWNLOAD_URL}{filename}", timeout=10, verify=False) # codeql[py/insecure-protocol]
91
91
response.raise_for_status()
92
92
except requests.RequestException:
93
93
found = None
@@ -359,6 +359,10 @@ def auto_download_pa_providers(
359
359
When *force* is ``False`` (the default), the download is skipped if the
360
360
auto-download marker is fresher than :data:`AUTO_DOWNLOAD_INTERVAL`.
361
361
"""
362
if os.environ.get("G4F_DISABLE_PA_AUTO_DOWNLOAD", "").lower() in ("1", "true", "yes"):
363
debug.log("pa-providers: auto-download explicitly disabled via G4F_DISABLE_PA_AUTO_DOWNLOAD")
364
return []
365
362
366
try:
363
367
workspace = get_workspace_dir()
364
368
except Exception as e:
@@ -138,6 +138,8 @@ class RotatedProvider(BaseRetryProvider):
138
138
provider.live -= 1
139
139
exceptions[provider.__name__] = e
140
140
debug.error(f"{provider.__name__} failed: {e}")
141
if started:
142
raise e
141
143
142
144
raise_exceptions(exceptions)
143
145
@@ -21,5 +21,4 @@ ddgs
21
21
numpy
22
22
PyYAML
23
23
websocket-client
24
pystray
25
24
cryptography