summaryrefslogtreecommitdiffstats
path: root/meta-python/recipes-devtools/python/python3-cbor2/CVE-2026-26209.patch
diff options
context:
space:
mode:
authorHitendra Prajapati <hprajapati@mvista.com>2026-03-31 11:45:06 +0530
committerAnuj Mittal <anuj.mittal@oss.qualcomm.com>2026-04-03 15:00:47 +0530
commit4810cd8c5bbc0b4349a78eac85a6a882bc0b03a2 (patch)
tree772d54d051e1374ce4d971d0673a17a39bcf91e2 /meta-python/recipes-devtools/python/python3-cbor2/CVE-2026-26209.patch
parentb13ae5a8eb286f8f4e53c290d5b09e15c303289c (diff)
downloadmeta-openembedded-4810cd8c5bbc0b4349a78eac85a6a882bc0b03a2.tar.gz
python3-cbor2: patch CVE-2026-26209
Backport the patch[1] which fixes this vulnerability as mentioned in the comment[3]. Details: https://nvd.nist.gov/vuln/detail/CVE-2026-26209 [1] https://github.com/agronholm/cbor2/commit/e61a5f365ba610d5907a0ae1bc72769bba34294b [2] https://github.com/agronholm/cbor2/commit/fb4ee1612a8a1ac0dbd8cf2f2f6f931a4e06d824 (pre patch) [3] https://github.com/agronholm/cbor2/pull/275 Dropped changes to the changelog from the original commit. Signed-off-by: Hitendra Prajapati <hprajapati@mvista.com> Signed-off-by: Anuj Mittal <anuj.mittal@oss.qualcomm.com>
Diffstat (limited to 'meta-python/recipes-devtools/python/python3-cbor2/CVE-2026-26209.patch')
-rw-r--r--meta-python/recipes-devtools/python/python3-cbor2/CVE-2026-26209.patch415
1 files changed, 415 insertions, 0 deletions
diff --git a/meta-python/recipes-devtools/python/python3-cbor2/CVE-2026-26209.patch b/meta-python/recipes-devtools/python/python3-cbor2/CVE-2026-26209.patch
new file mode 100644
index 0000000000..1a9c5a3995
--- /dev/null
+++ b/meta-python/recipes-devtools/python/python3-cbor2/CVE-2026-26209.patch
@@ -0,0 +1,415 @@
1From e61a5f365ba610d5907a0ae1bc72769bba34294b Mon Sep 17 00:00:00 2001
2From: Andreas Eriksen <andreer@vespa.ai>
3Date: Sat, 28 Feb 2026 22:21:06 +0100
4Subject: [PATCH] Set default read_size to 1 for backwards compatibility (#275)
5MIME-Version: 1.0
6Content-Type: text/plain; charset=UTF-8
7Content-Transfer-Encoding: 8bit
8
9The buffered reads introduced in 5.8.0 could cause issues when code needs to access the stream position after decoding. This changes the default back to 1 (matching 5.7.1 behavior) while allowing users to opt-in to faster decoding by passing read_size=4096.
10
11Implementation details:
12- Use function pointer dispatch to eliminate runtime checks for read_size=1
13- Skip buffer allocation entirely for unbuffered path
14- Add read_size parameter to load() and loads() for API completeness
15
16CVE: CVE-2026-26209
17Upstream-Status: Backport [https://github.com/agronholm/cbor2/commit/e61a5f365ba610d5907a0ae1bc72769bba34294b]
18Signed-off-by: Hitendra Prajapati <hprajapati@mvista.com>
19---
20 cbor2/_decoder.py | 33 ++++++++++++++++--
21 docs/usage.rst | 11 ++++++
22 source/decoder.c | 78 +++++++++++++++++++++++++++++--------------
23 source/decoder.h | 16 +++++++--
24 tests/test_decoder.py | 15 +++++++++
25 5 files changed, 123 insertions(+), 30 deletions(-)
26
27diff --git a/cbor2/_decoder.py b/cbor2/_decoder.py
28index 4aeadcf..5a1f65b 100644
29--- a/cbor2/_decoder.py
30+++ b/cbor2/_decoder.py
31@@ -72,6 +72,7 @@ class CBORDecoder:
32 tag_hook: Callable[[CBORDecoder, CBORTag], Any] | None = None,
33 object_hook: Callable[[CBORDecoder, dict[Any, Any]], Any] | None = None,
34 str_errors: Literal["strict", "error", "replace"] = "strict",
35+ read_size: int = 1,
36 ):
37 """
38 :param fp:
39@@ -90,6 +91,13 @@ class CBORDecoder:
40 :param str_errors:
41 determines how to handle unicode decoding errors (see the `Error Handlers`_
42 section in the standard library documentation for details)
43+ :param read_size:
44+ the minimum number of bytes to read at a time.
45+ Setting this to a higher value like 4096 improves performance,
46+ but is likely to read past the end of the CBOR value, advancing the stream
47+ position beyond the decoded data. This only matters if you need to reuse the
48+ stream after decoding.
49+ Ignored in the pure Python implementation, but included for API compatibility.
50
51 .. _Error Handlers: https://docs.python.org/3/library/codecs.html#error-handlers
52
53@@ -813,6 +821,7 @@ def loads(
54 tag_hook: Callable[[CBORDecoder, CBORTag], Any] | None = None,
55 object_hook: Callable[[CBORDecoder, dict[Any, Any]], Any] | None = None,
56 str_errors: Literal["strict", "error", "replace"] = "strict",
57+ read_size: int = 1,
58 ) -> Any:
59 """
60 Deserialize an object from a bytestring.
61@@ -831,6 +840,10 @@ def loads(
62 :param str_errors:
63 determines how to handle unicode decoding errors (see the `Error Handlers`_
64 section in the standard library documentation for details)
65+ :param read_size:
66+ the minimum number of bytes to read at a time.
67+ Setting this to a higher value like 4096 improves performance.
68+ Ignored in the pure Python implementation, but included for API compatibility.
69 :return:
70 the deserialized object
71
72@@ -839,7 +852,11 @@ def loads(
73 """
74 with BytesIO(s) as fp:
75 return CBORDecoder(
76- fp, tag_hook=tag_hook, object_hook=object_hook, str_errors=str_errors
77+ fp,
78+ tag_hook=tag_hook,
79+ object_hook=object_hook,
80+ str_errors=str_errors,
81+ read_size=read_size,
82 ).decode()
83
84
85@@ -848,6 +865,7 @@ def load(
86 tag_hook: Callable[[CBORDecoder, CBORTag], Any] | None = None,
87 object_hook: Callable[[CBORDecoder, dict[Any, Any]], Any] | None = None,
88 str_errors: Literal["strict", "error", "replace"] = "strict",
89+ read_size: int = 1,
90 ) -> Any:
91 """
92 Deserialize an object from an open file.
93@@ -866,6 +884,13 @@ def load(
94 :param str_errors:
95 determines how to handle unicode decoding errors (see the `Error Handlers`_
96 section in the standard library documentation for details)
97+ :param read_size:
98+ the minimum number of bytes to read at a time.
99+ Setting this to a higher value like 4096 improves performance,
100+ but is likely to read past the end of the CBOR value, advancing the stream
101+ position beyond the decoded data. This only matters if you need to reuse the
102+ stream after decoding.
103+ Ignored in the pure Python implementation, but included for API compatibility.
104 :return:
105 the deserialized object
106
107@@ -873,5 +898,9 @@ def load(
108
109 """
110 return CBORDecoder(
111- fp, tag_hook=tag_hook, object_hook=object_hook, str_errors=str_errors
112+ fp,
113+ tag_hook=tag_hook,
114+ object_hook=object_hook,
115+ str_errors=str_errors,
116+ read_size=read_size,
117 ).decode()
118diff --git a/docs/usage.rst b/docs/usage.rst
119index 797db59..6f53174 100644
120--- a/docs/usage.rst
121+++ b/docs/usage.rst
122@@ -74,6 +74,17 @@ instead encodes a reference to the nth sufficiently long string already encoded.
123 .. warning:: Support for string referencing is rare in other CBOR implementations, so think carefully
124 whether you want to enable it.
125
126+Performance tuning
127+------------------
128+
129+By default, the decoder only reads the exact amount of bytes it needs. But this can negatively
130+impact the performance due to the potentially large number of individual read operations.
131+To make it faster, you can pass a different ``read_size`` parameter (say, 4096), to :func:`load`,
132+:func:`loads` or :class:`CBORDecoder`.
133+
134+.. warning:: If the input stream contains data other than the CBOR stream, that data (or parts of)
135+ may be lost.
136+
137 Tag support
138 -----------
139
140diff --git a/source/decoder.c b/source/decoder.c
141index 9cd1596..f8adc93 100644
142--- a/source/decoder.c
143+++ b/source/decoder.c
144@@ -47,6 +47,10 @@ static int _CBORDecoder_set_tag_hook(CBORDecoderObject *, PyObject *, void *);
145 static int _CBORDecoder_set_object_hook(CBORDecoderObject *, PyObject *, void *);
146 static int _CBORDecoder_set_str_errors(CBORDecoderObject *, PyObject *, void *);
147
148+// Forward declarations for read dispatch functions
149+static int fp_read_unbuffered(CBORDecoderObject *, char *, Py_ssize_t);
150+static int fp_read_buffered(CBORDecoderObject *, char *, Py_ssize_t);
151+
152 static PyObject * decode(CBORDecoderObject *, DecodeOptions);
153 static PyObject * decode_bytestring(CBORDecoderObject *, uint8_t);
154 static PyObject * decode_string(CBORDecoderObject *, uint8_t);
155@@ -155,6 +159,7 @@ CBORDecoder_new(PyTypeObject *type, PyObject *args, PyObject *kwargs)
156 self->readahead_size = 0;
157 self->read_pos = 0;
158 self->read_len = 0;
159+ self->fp_read = fp_read_unbuffered; // default, will be set properly in init
160 }
161 return (PyObject *) self;
162 error:
163@@ -164,7 +169,7 @@ error:
164
165
166 // CBORDecoder.__init__(self, fp=None, tag_hook=None, object_hook=None,
167-// str_errors='strict', read_size=4096)
168+// str_errors='strict', read_size=1)
169 int
170 CBORDecoder_init(CBORDecoderObject *self, PyObject *args, PyObject *kwargs)
171 {
172@@ -233,7 +238,8 @@ _CBORDecoder_set_fp_with_read_size(CBORDecoderObject *self, PyObject *value, Py_
173 return -1;
174 }
175
176- if (self->readahead == NULL || self->readahead_size != read_size) {
177+ // Skip buffer allocation for read_size=1 (direct read path doesn't use buffer)
178+ if (read_size > 1 && (self->readahead == NULL || self->readahead_size != read_size)) {
179 new_buffer = (char *)PyMem_Malloc(read_size);
180 if (!new_buffer) {
181 Py_DECREF(read);
182@@ -254,8 +260,15 @@ _CBORDecoder_set_fp_with_read_size(CBORDecoderObject *self, PyObject *value, Py_
183 if (new_buffer) {
184 PyMem_Free(self->readahead);
185 self->readahead = new_buffer;
186- self->readahead_size = read_size;
187+ } else if (read_size == 1 && self->readahead != NULL) {
188+ // Free existing buffer when switching to direct read path (read_size=1)
189+ PyMem_Free(self->readahead);
190+ self->readahead = NULL;
191 }
192+ self->readahead_size = read_size;
193+
194+ // Set read dispatch function - eliminates runtime check on every read
195+ self->fp_read = (read_size == 1) ? fp_read_unbuffered : fp_read_buffered;
196
197 return 0;
198 }
199@@ -447,9 +460,25 @@ fp_read_bytes(CBORDecoderObject *self, char *buf, Py_ssize_t size)
200 return bytes_read;
201 }
202
203-// Read into caller's buffer using the readahead buffer
204+// Unbuffered read - used when read_size=1 (backwards compatible mode)
205+// This matches the 5.7.1 behavior with no runtime overhead
206+static int
207+fp_read_unbuffered(CBORDecoderObject *self, char *buf, Py_ssize_t size)
208+{
209+ Py_ssize_t bytes_read = fp_read_bytes(self, buf, size);
210+ if (bytes_read == size)
211+ return 0;
212+ if (bytes_read >= 0)
213+ PyErr_Format(
214+ _CBOR2_CBORDecodeEOF,
215+ "premature end of stream (expected to read %zd bytes, "
216+ "got %zd instead)", size, bytes_read);
217+ return -1;
218+}
219+
220+// Buffered read - used when read_size > 1 for improved performance
221 static int
222-fp_read(CBORDecoderObject *self, char *buf, const Py_ssize_t size)
223+fp_read_buffered(CBORDecoderObject *self, char *buf, Py_ssize_t size)
224 {
225 Py_ssize_t available, to_copy, remaining, total_copied;
226
227@@ -507,7 +536,7 @@ fp_read_object(CBORDecoderObject *self, const Py_ssize_t size)
228 if (!ret)
229 return NULL;
230
231- if (fp_read(self, PyBytes_AS_STRING(ret), size) == -1) {
232+ if (self->fp_read(self, PyBytes_AS_STRING(ret), size) == -1) {
233 Py_DECREF(ret);
234 return NULL;
235 }
236@@ -528,7 +557,7 @@ CBORDecoder_read(CBORDecoderObject *self, PyObject *length)
237 return NULL;
238 ret = PyBytes_FromStringAndSize(NULL, len);
239 if (ret) {
240- if (fp_read(self, PyBytes_AS_STRING(ret), len) == -1) {
241+ if (self->fp_read(self, PyBytes_AS_STRING(ret), len) == -1) {
242 Py_DECREF(ret);
243 ret = NULL;
244 }
245@@ -576,19 +605,19 @@ decode_length(CBORDecoderObject *self, uint8_t subtype,
246 if (subtype < 24) {
247 *length = subtype;
248 } else if (subtype == 24) {
249- if (fp_read(self, value.u8.buf, sizeof(uint8_t)) == -1)
250+ if (self->fp_read(self, value.u8.buf, sizeof(uint8_t)) == -1)
251 return -1;
252 *length = value.u8.value;
253 } else if (subtype == 25) {
254- if (fp_read(self, value.u16.buf, sizeof(uint16_t)) == -1)
255+ if (self->fp_read(self, value.u16.buf, sizeof(uint16_t)) == -1)
256 return -1;
257 *length = be16toh(value.u16.value);
258 } else if (subtype == 26) {
259- if (fp_read(self, value.u32.buf, sizeof(uint32_t)) == -1)
260+ if (self->fp_read(self, value.u32.buf, sizeof(uint32_t)) == -1)
261 return -1;
262 *length = be32toh(value.u32.value);
263 } else {
264- if (fp_read(self, value.u64.buf, sizeof(uint64_t)) == -1)
265+ if (self->fp_read(self, value.u64.buf, sizeof(uint64_t)) == -1)
266 return -1;
267 *length = be64toh(value.u64.value);
268 }
269@@ -752,7 +781,7 @@ decode_indefinite_bytestrings(CBORDecoderObject *self)
270 list = PyList_New(0);
271 if (list) {
272 while (1) {
273- if (fp_read(self, &lead.byte, 1) == -1)
274+ if (self->fp_read(self, &lead.byte, 1) == -1)
275 break;
276 if (lead.major == 2 && lead.subtype != 31) {
277 ret = decode_bytestring(self, lead.subtype);
278@@ -959,7 +988,7 @@ decode_indefinite_strings(CBORDecoderObject *self)
279 list = PyList_New(0);
280 if (list) {
281 while (1) {
282- if (fp_read(self, &lead.byte, 1) == -1)
283+ if (self->fp_read(self, &lead.byte, 1) == -1)
284 break;
285 if (lead.major == 3 && lead.subtype != 31) {
286 ret = decode_string(self, lead.subtype);
287@@ -2040,7 +2069,7 @@ CBORDecoder_decode_simple_value(CBORDecoderObject *self)
288 PyObject *tag, *ret = NULL;
289 uint8_t buf;
290
291- if (fp_read(self, (char*)&buf, sizeof(uint8_t)) == 0) {
292+ if (self->fp_read(self, (char*)&buf, sizeof(uint8_t)) == 0) {
293 tag = PyStructSequence_New(&CBORSimpleValueType);
294 if (tag) {
295 PyStructSequence_SET_ITEM(tag, 0, PyLong_FromLong(buf));
296@@ -2066,7 +2095,7 @@ CBORDecoder_decode_float16(CBORDecoderObject *self)
297 char buf[sizeof(uint16_t)];
298 } u;
299
300- if (fp_read(self, u.buf, sizeof(uint16_t)) == 0)
301+ if (self->fp_read(self, u.buf, sizeof(uint16_t)) == 0)
302 ret = PyFloat_FromDouble(unpack_float16(u.i));
303 set_shareable(self, ret);
304 return ret;
305@@ -2084,7 +2113,7 @@ CBORDecoder_decode_float32(CBORDecoderObject *self)
306 char buf[sizeof(float)];
307 } u;
308
309- if (fp_read(self, u.buf, sizeof(float)) == 0) {
310+ if (self->fp_read(self, u.buf, sizeof(float)) == 0) {
311 u.i = be32toh(u.i);
312 ret = PyFloat_FromDouble(u.f);
313 }
314@@ -2104,7 +2133,7 @@ CBORDecoder_decode_float64(CBORDecoderObject *self)
315 char buf[sizeof(double)];
316 } u;
317
318- if (fp_read(self, u.buf, sizeof(double)) == 0) {
319+ if (self->fp_read(self, u.buf, sizeof(double)) == 0) {
320 u.i = be64toh(u.i);
321 ret = PyFloat_FromDouble(u.f);
322 }
323@@ -2133,7 +2162,7 @@ decode(CBORDecoderObject *self, DecodeOptions options)
324 if (Py_EnterRecursiveCall(" in CBORDecoder.decode"))
325 return NULL;
326
327- if (fp_read(self, &lead.byte, 1) == 0) {
328+ if (self->fp_read(self, &lead.byte, 1) == 0) {
329 switch (lead.major) {
330 case 0: ret = decode_uint(self, lead.subtype); break;
331 case 1: ret = decode_negint(self, lead.subtype); break;
332@@ -2387,13 +2416,12 @@ PyDoc_STRVAR(CBORDecoder__doc__,
333 " :class:`dict` object. The return value is substituted for the dict\n"
334 " in the deserialized output.\n"
335 ":param read_size:\n"
336-" the size of the read buffer (default 4096). The decoder reads from\n"
337-" the stream in chunks of this size for performance. This means the\n"
338-" stream position may advance beyond the bytes actually decoded. For\n"
339-" large values (bytestrings, text strings), reads may be larger than\n"
340-" ``read_size``. Code that needs to read from the stream after\n"
341-" decoding should use :meth:`decode_from_bytes` instead, or set\n"
342-" ``read_size=1`` to disable buffering (at a performance cost).\n"
343+" the minimum number of bytes to read at a time.\n"
344+" Setting this to a higher value like 4096 improves performance,\n"
345+" but is likely to read past the end of the CBOR value, advancing the stream\n"
346+" position beyond the decoded data. This only matters if you need to reuse the\n"
347+" stream after decoding.\n"
348+" Ignored in the pure Python implementation, but included for API compatibility.\n"
349 "\n"
350 ".. _CBOR: https://cbor.io/\n"
351 );
352diff --git a/source/decoder.h b/source/decoder.h
353index a2f4bf1..3efff8b 100644
354--- a/source/decoder.h
355+++ b/source/decoder.h
356@@ -3,10 +3,17 @@
357 #include <stdbool.h>
358 #include <stdint.h>
359
360-// Default readahead buffer size for streaming reads
361-#define CBOR2_DEFAULT_READ_SIZE 4096
362+// Default readahead buffer size for streaming reads.
363+// Set to 1 for backwards compatibility (no buffering).
364+#define CBOR2_DEFAULT_READ_SIZE 1
365
366-typedef struct {
367+// Forward declaration for function pointer typedef
368+struct CBORDecoderObject_;
369+
370+// Function pointer type for read dispatch (eliminates runtime check)
371+typedef int (*fp_read_fn)(struct CBORDecoderObject_ *, char *, Py_ssize_t);
372+
373+typedef struct CBORDecoderObject_ {
374 PyObject_HEAD
375 PyObject *read; // cached read() method of fp
376 PyObject *tag_hook;
377@@ -23,6 +30,9 @@ typedef struct {
378 Py_ssize_t readahead_size; // size of allocated buffer
379 Py_ssize_t read_pos; // current position in buffer
380 Py_ssize_t read_len; // valid bytes in buffer
381+
382+ // Read dispatch - points to unbuffered or buffered implementation
383+ fp_read_fn fp_read;
384 } CBORDecoderObject;
385
386 extern PyTypeObject CBORDecoderType;
387diff --git a/tests/test_decoder.py b/tests/test_decoder.py
388index 9bf5a10..c5d1a9c 100644
389--- a/tests/test_decoder.py
390+++ b/tests/test_decoder.py
391@@ -123,6 +123,21 @@ def test_load(impl):
392 assert impl.load(fp=stream) == 1
393
394
395+def test_stream_position_after_decode(impl):
396+ """Test that stream position is exactly at end of decoded CBOR value."""
397+ # CBOR: integer 1 (1 byte: 0x01) followed by extra data
398+ cbor_data = b"\x01"
399+ extra_data = b"extra"
400+ with BytesIO(cbor_data + extra_data) as stream:
401+ decoder = impl.CBORDecoder(stream)
402+ result = decoder.decode()
403+ assert result == 1
404+ # Stream position should be exactly at end of CBOR data
405+ assert stream.tell() == len(cbor_data)
406+ # Should be able to read the extra data
407+ assert stream.read() == extra_data
408+
409+
410 @pytest.mark.parametrize(
411 "payload, expected",
412 [
413--
4142.50.1
415