8000 [3.10] gh-121285: Remove backtracking when parsing tarfile headers (G… · python/cpython@743acbe · GitHub
[go: up one dir, main page]
Skip to content

Commit 743acbe

Browse files
sethmlarsonEclips4gpshead
authored
[3.10] gh-121285: Remove backtracking when parsing tarfile headers (GH-121286) (#123640)
* Remove backtracking when parsing tarfile headers * Rewrite PAX header parsing to be stricter * Optimize parsing of GNU extended sparse headers v0.0 (cherry picked from commit 34ddb64) Co-authored-by: Kirill Podoprigora <kirill.bast9@mail.ru> Co-authored-by: Gregory P. Smith <greg@krypto.org>
1 parent e0264a6 commit 743acbe

File tree

3 files changed

+111
-38
lines changed

3 files changed

+111
-38
lines changed

‎Lib/tarfile.py

Lines changed: 67 additions & 38 deletions
Original file line numberDiff line numberDiff line change
@@ -841,6 +841,9 @@ def data_filter(member, dest_path):
841841
# Sentinel for replace() defaults, meaning "don't change the attribute"
842842
_KEEP = object()
843843

844+
# Header length is digits followed by a space.
845+
_header_length_prefix_re = re.compile(br"([0-9]{1,20}) ")
846+
844847
class TarInfo(object):
845848
"""Informational class which holds the details about an
846849
archive member given by a tar header block.
@@ -1410,59 +1413,76 @@ def _proc_pax(self, tarfile):
14101413
else:
14111414
pax_headers = tarfile.pax_headers.copy()
14121415

1413-
# Check if the pax header contains a hdrcharset field. This tells us
1414-
# the encoding of the path, linkpath, uname and gname fields. Normally,
1415-
# these fields are UTF-8 encoded but since POSIX.1-2008 tar
1416-
# implementations are allowed to store them as raw binary strings if
1417-
# the translation to UTF-8 fails.
1418-
match = re.search(br"\d+ hdrcharset=([^\n]+)\n", buf)
1419-
if match is not None:
1420-
pax_headers["hdrcharset"] = match.group(1).decode("utf-8")
1421-
1422-
# For the time being, we don't care about anything other than "BINARY".
1423-
# The only other value that is currently allowed by the standard is
1424-
# "ISO-IR 10646 2000 UTF-8" in other words UTF-8.
1425-
hdrcharset = pax_headers.get("hdrcharset")
1426-
if hdrcharset == "BINARY":
1427-
encoding = tarfile.encoding
1428-
else:
1429-
encoding = "utf-8"
1430-
14311416
# Parse pax header information. A record looks like that:
14321417
# "%d %s=%s\n" % (length, keyword, value). length is the size
14331418
# of the complete record including the length field itself and
1434-
# the newline. keyword and value are both UTF-8 encoded strings.
1435-
regex = re.compile(br"(\d+) ([^=]+)=")
1419+
# the newline.
14361420
pos = 0
1437-
while True:
1438-
match = regex.match(buf, pos)
1439-
if not match:
1440-
break
1421+
encoding = None
1422+
raw_headers = []
1423+
while len(buf) > pos and buf[pos] != 0x00:
1424+
if not (match := _header_length_prefix_re.match(buf, pos)):
1425+
raise InvalidHeaderError("invalid header")
1426+
try:
1427+
length = int(match.group(1))
1428+
except ValueError:
1429+
raise InvalidHeaderError("invalid header")
1430+
# Headers must be at least 5 bytes, shortest being '5 x=\n'.
1431+
# Value is allowed to be empty.
1432+
if length < 5:
1433+
raise InvalidHeaderError("invalid header")
1434+
if pos + length > len(buf):
1435+
raise InvalidHeaderError("invalid header")
14411436

1442-
length, keyword = match.groups()
1443-
length = int(length)
1444-
if length == 0:
1437+
header_value_end_offset = match.start(1) + length - 1 # Last byte of the header
1438+
keyword_and_value = buf[match.end(1) + 1:header_value_end_offset]
1439+
raw_keyword, equals, raw_value = keyword_and_value.partition(b"=")
1440+
1441+
# Check the framing of the header. The last character must be '\n' (0x0A)
1442+
if not raw_keyword or equals != b"=" or buf[header_value_end_offset] != 0x0A:
14451443
raise InvalidHeaderError("invalid header")
1446-
value = buf[match.end(2) + 1:match.start(1) + length - 1]
1444+
raw_headers.append((length, raw_keyword, raw_value))
1445+
1446+
# Check if the pax header contains a hdrcharset field. This tells us
1447+
# the encoding of the path, linkpath, uname and gname fields. Normally,
1448+
# these fields are UTF-8 encoded but since POSIX.1-2008 tar
1449+
# implementations are allowed to store them as raw binary strings if
1450+
# the translation to UTF-8 fails. For the time being, we don't care about
1451+
# anything other than "BINARY". The only other value that is currently
1452+
# allowed by the standard is "ISO-IR 10646 2000 UTF-8" in other words UTF-8.
1453+
# Note that we only follow the initial 'hdrcharset' setting to preserve
1454+
# the initial behavior of the 'tarfile' module.
1455+
if raw_keyword == b"hdrcharset" and encoding is None:
1456+
if raw_value == b"BINARY":
1457+
encoding = tarfile.encoding
1458+
else: # This branch ensures only the first 'hdrcharset' header is used.
1459+
encoding = "utf-8"
1460+
1461+
pos += length
14471462

1463+
# If no explicit hdrcharset is set, we use UTF-8 as a default.
1464+
if encoding is None:
1465+
encoding = "utf-8"
1466+
1467+
# After parsing the raw headers we can decode them to text.
1468+
for length, raw_keyword, raw_value in raw_headers:
14481469
# Normally, we could just use "utf-8" as the encoding and "strict"
14491470
# as the error handler, but we better not take the risk. For
14501471
# example, GNU tar <= 1.23 is known to store filenames it cannot
14511472
# translate to UTF-8 as raw strings (unfortunately without a
14521473
# hdrcharset=BINARY header).
14531474
# We first try the strict standard encoding, and if that fails we
14541475
# fall back on the user's encoding and error handler.
1455-
keyword = self._decode_pax_field(keyword, "utf-8", "utf-8",
1476+
keyword = self._decode_pax_field(raw_keyword, "utf-8", "utf-8",
14561477
tarfile.errors)
14571478
if keyword in PAX_NAME_FIELDS:
1458-
value = self._decode_pax_field(value, encoding, tarfile.encoding,
1479+ D966
value = self._decode_pax_field(raw_value, encoding, tarfile.encoding,
14591480
tarfile.errors)
14601481
else:
1461-
value = self._decode_pax_field(value, "utf-8", "utf-8",
1482+
value = self._decode_pax_field(raw_value, "utf-8", "utf-8",
14621483
tarfile.errors)
14631484

14641485
pax_headers[keyword] = value
1465-
pos += length
14661486

14671487
# Fetch the next header.
14681488
try:
@@ -1477,7 +1497,7 @@ def _proc_pax(self, tarfile):
14771497

14781498
elif "GNU.sparse.size" in pax_headers:
14791499
# GNU extended sparse format version 0.0.
1480-
self._proc_gnusparse_00(next, pax_headers, buf)
1500+
self._proc_gnusparse_00(next, raw_headers)
14811501

14821502
elif pax_headers.get("GNU.sparse.major") == "1" and pax_headers.get("GNU.sparse.minor") == "0":
14831503
# GNU extended sparse format version 1.0.
@@ -1499,15 +1519,24 @@ def _proc_pax(self, tarfile):
14991519

15001520
return next
15011521

1502-
def _proc_gnusparse_00(self, next, pax_headers, buf):
1522+
def _proc_gnusparse_00(self, next, raw_headers):
15031523
"""Process a GNU tar extended sparse header, version 0.0.
15041524
"""
15051525
offsets = []
1506-
for match in re.finditer(br"\d+ GNU.sparse.offset=(\d+)\n", buf):
1507-
offsets.append(int(match.group(1)))
15081526
numbytes = []
1509-
for match in re.finditer(br"\d+ GNU.sparse.numbytes=(\d+)\n", buf):
1510-
numbytes.append(int(match.group(1)))
1527+
for _, keyword, value in raw_headers:
1528+
if keyword == b"GNU.sparse.offset":
1529+
try:
1530+
offsets.append(int(value.decode()))
1531+
except ValueError:
1532+
raise InvalidHeaderError("invalid header")
1533+
1534+
elif keyword == b"GNU.sparse.numbytes":
1535+
try:
1536+
numbytes.append(int(value.decode()))
1537+
except ValueError:
1538+
raise InvalidHeaderError("invalid header")
1539+
15111540
next.sparse = list(zip(offsets, numbytes))
15121541

15131542
def _proc_gnusparse_01(self, next, pax_headers):

‎Lib/test/test_tarfile.py

Lines changed: 42 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1139,6 +1139,48 @@ def test_pax_number_fields(self):
11391139
finally:
11401140
tar.close()
11411141

1142+
def test_pax_header_bad_formats(self):
1143+
# The fields from the pax header have priority over the
1144+
# TarInfo.
1145+
pax_header_replacements = (
1146+
b" foo=bar\n",
1147+
b"0 \n",
1148+
b"1 \n",
1149+
b"2 \n",
1150+
b"3 =\n",
1151+
b"4 =a\n",
1152+
b"1000000 foo=bar\n",
1153+
b"0 foo=bar\n",
1154+
b"-12 foo=bar\n",
1155+
b"000000000000000000000000036 foo=bar\n",
1156+
)
1157+
pax_headers = {"foo": "bar"}
1158+
1159+
for replacement in pax_header_replacements:
1160+
with self.subTest(header=replacement):
1161+
tar = tarfile.open(tmpname, "w", format=tarfile.PAX_FORMAT,
1162+
encoding="iso8859-1")
1163+
try:
1164+
t = tarfile.TarInfo()
1165+
t.name = "pax" # non-ASCII
1166+
t.uid = 1
1167+
t.pax_headers = pax_headers
1168+
tar.addfile(t)
1169+
finally:
1170+
tar.close()
1171+
1172+
with open(tmpname, "rb") as f:
1173+
data = f.read()
1174+
self.assertIn(b"11 foo=bar\n", data)
1175+
data = data.replace(b"11 foo=bar\n", replacement)
1176+
1177+
with open(tmpname, "wb") as f:
1178+
f.truncate()
1179+
f.write(data)
1180+
1181+
with self.assertRaisesRegex(tarfile.ReadError, r"method tar: ReadError\('invalid header'\)"):
1182+
tarfile.open(tmpname, encoding="iso8859-1")
1183+
11421184

11431185
class WriteTestBase(TarTest):
11441186
# Put all write tests in here that are supposed to be tested

‎Misc/NEWS.d/next/Security/2024-07-02-13-39-20.gh-issue-121285.hrl-yI.rst

Lines changed: 2 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,2 @@
1+
Remove backtracking from tarfile header parsing for ``hdrcharset``, PAX, and
2+
GNU sparse headers.

0 commit comments

Comments
 (0)
0