[3.12] gh-156002: Bound zipfile decompression for bzip2/LZMA/Zstandard (GH-156003) (GH-156362) (#156739)

* gh-156002: Bound zipfile decompression for bzip2/LZMA/Zstandard (GH-156003) (GH-156362)

Patch by @tonghuaroot.

zipfile.ZipExtFile._read1() bounds the output of each decompress() call
for DEFLATE members by passing a max_length to zlib, but for bzip2, LZMA,
and Zstandard members it called decompress() with no bound. A whole
compressed chunk was therefore expanded into a single allocation before
the data[:self._left] clip ran, so a consumer that deliberately reads in
small chunks to limit memory (for example zf.open(name).read(8192)) was
silently unprotected for non-DEFLATE members. A small, spec-conformant
archive member declaring a large uncompressed size could drive multi-GB
peak memory.

_read1() now passes a per-call bound to the non-DEFLATE decompress()
(mirroring the DEFLATE branch) and drains the decompressor's internal
buffer across calls by checking needs_input before reading more
compressed input. zipfile's LZMADecompressor wrapper forwards max_length
and exposes needs_input so the bound also holds for LZMA members.

(cherry picked from commit f897dbf)
(cherry picked from commit 1b424c0)

* gh-156002: Keep reading through monkey-patched zipfile decompressors (GH-157180) (GH-157557)

(cherry picked from commit f507e69)

Co-authored-by: Petr Viktorin <encukou@gmail.com>
Co-authored-by: tonghuaroot <tonghuaroot@gmail.com>
Co-authored-by: rasmusfaber <rfaber@gmail.com>

Author: 4 people

Lib/test/test_zipfile/test_core.py

@@ -2590,6 +2590,110 @@ def tearDown(self):
         unlink(TESTFN2)
 
 
+class AbstractBoundedDecompressTests:
+    # ZipExtFile._read1() bounds the output of each decompress() call so that a
+    # small member declaring a large uncompressed size cannot expand into one
+    # unbounded read.
+    def test_read1_output_is_bounded(self):
+        buf = io.BytesIO()
+        with zipfile.ZipFile(buf, "w", compression=self.compression) as zf:
+            zf.writestr("big", b"\0" * (4 * 1024 * 1024))
+        with zipfile.ZipFile(io.BytesIO(buf.getvalue())) as zf:
+            with zf.open("big") as f:
+                self.assertLessEqual(len(f._read1(100)), f.MIN_READ_SIZE)
+
+
+class StoredBoundedDecompressTests(AbstractBoundedDecompressTests,
+                                   unittest.TestCase):
+    compression = zipfile.ZIP_STORED
+
+
+@requires_zlib()
+class DeflateBoundedDecompressTests(AbstractBoundedDecompressTests,
+                                    unittest.TestCase):
+    compression = zipfile.ZIP_DEFLATED
+
+
+@requires_bz2()
+class Bzip2BoundedDecompressTests(AbstractBoundedDecompressTests,
+                                  unittest.TestCase):
+    compression = zipfile.ZIP_BZIP2
+
+
+@requires_lzma()
+class LzmaBoundedDecompressTests(AbstractBoundedDecompressTests,
+                                 unittest.TestCase):
+    compression = zipfile.ZIP_LZMA
+
+
+
+class MonkeypatchedDecompressorTests(unittest.TestCase):
+    # Some third-party projects monkey-patch _get_decompressor() to add
+    # additional compression schemes. This can break at any time as the
+    # internal compressor objects change.
+    # To protect users, we try to keep this case working.
+    # See also: GH-156002 and GH-113767.
+    COMPRESSION = 99
+
+    class Compressor:
+        """Compressor with only the original BZ2Compressor API"""
+        def compress(self, data):
+            return data.swapcase()
+
+        def flush(self):
+            return b''
+
+    class Decompressor:
+        """Decompressor with only the 3.3+ BZ2Decompressor API"""
+        eof = False
+
+        def decompress(self, data):
+            return data.swapcase()
+
+    def setUp(self):
+        orig_check_compression = zipfile._check_compression
+        orig_get_compressor = zipfile._get_compressor
+        orig_get_decompressor = zipfile._get_decompressor
+
+        def check_compression(compression):
+            if compression != self.COMPRESSION:
+                orig_check_compression(compression)
+
+        def get_compressor(compress_type, compresslevel=None):
+            if compress_type == self.COMPRESSION:
+                return self.Compressor()
+            return orig_get_compressor(compress_type, compresslevel)
+
+        def get_decompressor(compress_type):
+            if compress_type == self.COMPRESSION:
+                return self.Decompressor()
+            return orig_get_decompressor(compress_type)
+
+        self.enterContext(mock.patch.object(
+            zipfile, '_check_compression', check_compression))
+        self.enterContext(mock.patch.object(
+            zipfile, '_get_compressor', get_compressor))
+        self.enterContext(mock.patch.object(
+            zipfile, '_get_decompressor', get_decompressor))
+
+    def test_roundtrip_monkeypatched_decompressor(self):
+        data = bytes(range(256)) * 8
+        buf = io.BytesIO()
+        with zipfile.ZipFile(buf, "w", compression=self.COMPRESSION) as zf:
+            zf.writestr("member", data)
+        self.assertIn(data.swapcase(), buf.getvalue())
+        with zipfile.ZipFile(io.BytesIO(buf.getvalue())) as zf:
+            self.assertEqual(zf.read("member"), data)
+            with zf.open("member") as f:
+                self.assertEqual(f.read(100), data[:100])
+                self.assertEqual(f.read1(100), data[100:200])
+                f.seek(-100, os.SEEK_END)
+                self.assertEqual(f.read(), data[-100:])
+                # Rewinding past the read buffer re-creates the decompressor.
+                f.seek(0)
+                self.assertEqual(f.read(), data)
+
+
 class AbstractBadCrcTests:
     def test_testzip_with_bad_crc(self):
         """Tests that files with bad CRCs return their name from testzip."""

Lib/zipfile/__init__.py

@@ -698,7 +698,16 @@ def __init__(self):
         self._unconsumed = b''
         self.eof = False
 
-    def decompress(self, data):
+    @property
+    def needs_input(self):
+        # While the LZMA properties header is still being buffered, more input
+        # is required; afterwards defer to the wrapped decompressor so a bounded
+        # decompress() call can be drained across reads.
+        if self._decomp is None:
+            return True
+        return self._decomp.needs_input
+
+    def decompress(self, data, max_length=-1):
         if self._decomp is None:
             self._unconsumed += data
             if len(self._unconsumed) <= 4:
@@ -714,7 +723,7 @@ def decompress(self, data):
             data = self._unconsumed[4 + psize:]
             del self._unconsumed
 
-        result = self._decomp.decompress(data)
+        result = self._decomp.decompress(data, max_length)
         self.eof = self._decomp.eof
         return result
 
@@ -1074,8 +1083,15 @@ def _read1(self, n):
             data = self._decompressor.unconsumed_tail
             if n > len(data):
                 data += self._read2(n - len(data))
-        else:
+        elif self._compress_type == ZIP_STORED:
             data = self._read2(n)
+        else:
+            # bzip2/lzma/zstd: a bounded decompress() call may leave input
+            # buffered inside the decompressor; drain that before reading more.
+            if getattr(self._decompressor, "needs_input", True):
+                data = self._read2(n)
+            else:
+                data = b''
 
         if self._compress_type == ZIP_STORED:
             self._eof = self._compress_left <= 0
@@ -1088,8 +1104,17 @@ def _read1(self, n):
             if self._eof:
                 data += self._decompressor.flush()
         else:
-            data = self._decompressor.decompress(data)
-            self._eof = self._decompressor.eof or self._compress_left <= 0
+            # Bound the output of a single decompress() call (mirroring the
+            # DEFLATE path above) so that a small compressed member cannot
+            # expand into one unbounded read.
+            try:
+                data = self._decompressor.decompress(data, max(n, self.MIN_READ_SIZE))
+            except TypeError:
+                # See MonkeypatchedDecompressorTests in test_core.py
+                data = self._decompressor.decompress(data)
+            self._eof = (self._decompressor.eof or
+                         self._compress_left <= 0 and
+                         getattr(self._decompressor, "needs_input", True))
 
         data = data[:self._left]
         self._left -= len(data)

Misc/NEWS.d/next/Library/2026-09-08-13-06-29.gh-issue-156002.vmOC8T.rst

@@ -0,0 +1,5 @@
+:mod:`zipfile` again reads members through a third-party decompressor
+installed by monkey-patching the private ``_get_decompressor()`` to return an
+object that only implements old BZ2Decompressor API from Python 3.3.
+Note that decompressors without ``needs_input`` and two-argument
+``decompress()`` are vulnerable to :cve:`2026-15310`.

Misc/NEWS.d/next/Security/2026-08-18-13-54-05.gh-issue-156002.CcWXPP.rst

@@ -0,0 +1,4 @@
+Bound the amount of data :mod:`zipfile` decompresses per read for members
+compressed with bzip2, LZMA, or Zstandard, matching the existing limit for
+deflate. A small archive member could previously expand into an unbounded
+allocation even when read in small chunks.