* bpo-31499, xml.etree: Fix xmlparser_gc_clear() crash
xml.etree: xmlparser_gc_clear() now sets self.parser to NULL to prevent a
crash in xmlparser_dealloc() if xmlparser_gc_clear() was called previously
by the garbage collector, because the parser was part of a reference cycle.
Co-Authored-By: Serhiy Storchaka <storchaka@gmail.com>
(cherry picked from commit e727d41ffc)
* Avoid calling "PyObject_GetAttrString()" (and potentially executing user code) with a live exception set.
* Ignore only AttributeError on attribute lookups in ElementTree.XMLParser() and propagate all other exceptions.
(cherry picked from commit c8d8e15bfc)
Cast Py_buffer.len (Py_ssize_t, signed) to size_t (unsigned) to
prevent the following warning:
Modules/_ssl.c:3089:21: warning: comparison between signed and
unsigned integer expressions [-Wsign-compare].
(cherry picked from commit 5a61559fb0)
* bpo-30923: Disable warning that has been part of -Wextra since gcc-7.0. (#3142)
(cherry picked from commit d73a960c57)
* bpo-30923: Silence fall-through warnings included in -Wextra since gcc-7.0. (#3157)
(cherry picked from commit f432a3234f)
* bpo-31275: Small refactoring to silence a fall-through warning. (#3206)
(cherry picked from commit 138753c1b9)
* bpo-29136: Add TLS 1.3 support
TLS 1.3 introduces a new, distinct set of cipher suites. The TLS 1.3
cipher suites don't overlap with cipher suites from TLS 1.2 and earlier.
Since Python sets its own set of permitted ciphers, TLS 1.3 handshake
will fail as soon as OpenSSL 1.1.1 is released. Let's enable the common
AES-GCM and ChaCha20 suites.
Additionally the flag OP_NO_TLSv1_3 is added. It defaults to 0 (no op) with
OpenSSL prior to 1.1.1. This allows applications to opt-out from TLS 1.3
now.
Signed-off-by: Christian Heimes <christian@python.org>.
(cherry picked from commit cb5b68abde)
SSLObject.version() now correctly returns None when handshake over BIO has
not been performed yet.
Signed-off-by: Christian Heimes <christian@python.org>
(cherry picked from commit 6877111)
Drop handshake_done and peer_cert members from PySSLSocket struct. The
peer certificate can be acquired from *SSL directly.
SSL_get_peer_certificate() does not trigger any network activity.
Instead of manually tracking the handshake state, simply use
SSL_is_init_finished().
In combination these changes fix auto-handshake for non-blocking
MemoryBIO connections.
Signed-off-by: Christian Heimes <christian@python.org>.
(cherry picked from commit 66dc33b682)
The ssl and hashlib modules now call OPENSSL_add_all_algorithms_noconf() on
OpenSSL < 1.1.0. The function detects CPU features and enables optimizations
on some CPU architectures such as POWER8. Patch is based on research from
Gustavo Serra Scalet.
Signed-off-by: Christian Heimes <christian@python.org>
(cherry picked from commit c941e62)
Include sys/sysmacros.h for major(), minor(), and makedev(). GNU C libray
plans to remove the functions from sys/types.h.
Signed-off-by: Christian Heimes <christian@python.org>
(cherry picked from commit 75b9618)
* Change NPN detection:
Version breakdown, support disabled (pre-patch/post-patch):
- pre-1.0.1: OPENSSL_NPN_NEGOTIATED will not be defined -> False/False
- 1.0.1 and 1.0.2: OPENSSL_NPN_NEGOTIATED will not be defined ->
False/False
- 1.1.0+: OPENSSL_NPN_NEGOTIATED will be defined and
OPENSSL_NO_NEXTPROTONEG will be defined -> True/False
Version breakdown support enabled (pre-patch/post-patch):
- pre-1.0.1: OPENSSL_NPN_NEGOTIATED will not be defined -> False/False
- 1.0.1 and 1.0.2: OPENSSL_NPN_NEGOTIATED will be defined and
OPENSSL_NO_NEXTPROTONEG will not be defined -> True/True
- 1.1.0+: OPENSSL_NPN_NEGOTIATED will be defined and
OPENSSL_NO_NEXTPROTONEG will not be defined -> True/True
* Refine NPN guard:
- If NPN is disabled, but ALPN is available we need our callback
- Make clinic's ssl behave the same way
This created a working ssl module for me, with NPN disabled and ALPN
enabled for OpenSSL 1.1.0f.
Concerns to address:
The initial commit for NPN support into OpenSSL [1], had the
OPENSSL_NPN_* variables defined inside the OPENSSL_NO_NEXTPROTONEG
guard. The question is if that ever made it into a release.
This would need an ugly hack, something like:
GH-if defined(OPENSSL_NO_NEXTPROTONEG) && \
!defined(OPENSSL_NPN_NEGOTIATED)
GH- define OPENSSL_NPN_UNSUPPORTED 0
GH- define OPENSSL_NPN_NEGOTIATED 1
GH- define OPENSSL_NPN_NO_OVERLAP 2
GH-endif
[1] https://github.com/openssl/openssl/commit/68b33cc5c7
(cherry picked from commit b2d096b)
* [3.6] bpo-9146: Raise a ValueError if OpenSSL fails to init a hash func. (GH-1777)
This helps people in weird FIPS mode environments where common things
like MD5 are not available in the binary as a matter of policy.
(cherry picked from commit 07244a8301)
* Include a NEWS entry.
* Fixes#30581 by adding a path to use newer GetMaximumProcessorCount API on Windows calls to os.cpu_count()
* Add NEWS.d entry for bpo-30581, os.cpu_count on Windows.
* Tweak NEWS entry
Ctypes currently produces wrong pep3118 type codes for several types.
E.g. memoryview(ctypes.c_long()).format gives "<l" on 64-bit platforms,
but it should be "<q" instead for sizeof(c_long) == 8
The problem is that the '<>' endian specification in the struct syntax
also turns on the "standard size" mode, which makes type characters have
a platform-independent meaning, which does not match with the codes used
internally in ctypes. The struct module format syntax also does not
allow specifying native-size non-native-endian items.
This commit adds a converter function that maps the internal ctypes
codes to appropriate struct module standard-size codes in the pep3118
format strings. The tests are modified to check for this.
(cherry picked from commit 07f1658aa0)
The current test_child_terminated_in_stopped_state() function test
creates a child process which calls ptrace(PTRACE_TRACEME, 0, 0) and
then crash (SIGSEGV). The problem is that calling os.waitpid() in the
parent process is not enough to close the process: the child process
remains alive and so the unit test leaks a child process in a
strange state. Closing the child process requires non-trivial code,
maybe platform specific.
Remove the functional test and replaces it with an unit test which
mocks os.waitpid() using a new _testcapi.W_STOPCODE() function to
test the WIFSTOPPED() path.
(cherry picked from commit 7b7c6dcfff)
* Improve signal delivery
Avoid using Py_AddPendingCall from signal handler, to avoid calling signal-unsafe functions.
* Remove unused function
* Improve comments
* Use _Py_atomic API for concurrency-sensitive signal state
* Add blurb
(cherry picked from commit 2c8a5e4c96)
If history-length is set in .inputrc, and the history file is double the
history size (or more), history_get(N) returns NULL, and python
segfaults. Fix that by checking for NULL return value.
It seems that the root cause is incorrect handling of bigger history in
readline, but Python should not segfault even if readline returns
unexpected value.
This issue affects only GNU readline. When using libedit emulation
system history size option does not work.
* [3.6] bpo-30703: Improve signal delivery (GH-2415)
* Improve signal delivery
Avoid using Py_AddPendingCall from signal handler, to avoid calling signal-unsafe functions.
* Remove unused function
* Improve comments
* Add stress test
* Adapt for --without-threads
* Add second stress test
* Add NEWS blurb
* Address comments @haypo.
(cherry picked from commit c08177a1cc)
* bpo-30796: Fix failures in signal delivery stress test (#2488)
* bpo-30796: Fix failures in signal delivery stress test
setitimer() can have a poor minimum resolution on some machines,
this would make the test reach its deadline (and a stray signal
could then kill a subsequent test).
* Make sure to clear the itimer after the test
When os.spawnv() fails while handling arguments, free correctly
argvlist: pass lastarg+1 rather than lastarg to free_string_array()
to also free the first item.
(cherry picked from commit 8acb4cf2b3)
Based on patch by Victor Stinner.
Add private C API function _PyUnicode_AsUnicode() which is similar to
PyUnicode_AsUnicode(), but checks for null characters..
(cherry picked from commit f7eae0adfc)
New error condition paths were introduced, which did not decrement
`key2` and `val2` objects. Therefore, decrement references before
jumping to the error label.
Signed-off-by: Eric N. Vander Weele <ericvw@gmail.com>
(cherry picked from commit a7874c73c0)
* bpo-29591: Upgrade Modules/expat to libexpat 2.2
* bpo-29591: Restore Python changes on expat
* bpo-29591: Remove expat config of unsupported platforms
Remove the configuration (Modules/expat/*config.h) of unsupported
platforms:
* Amiga
* MacOS Classic on PPC32
* Open Watcom
* bpo-29591: Remove useless XML_HAS_SET_HASH_SALT
The XML_HAS_SET_HASH_SALT define of Modules/expat/expat.h became
useless since our local expat copy was upgrade to expat 2.1 (it's now
expat 2.2.0).
(cherry picked from commit 23ec4b57e1)
Before, it was possible to get the following sequence of
events (especially on Windows, where the C-level signal handler for
SIGINT is run in a separate thread):
- SIGINT arrives
- trip_signal is called
- trip_signal writes to the wakeup fd
- the main thread wakes up from select()-or-equivalent
- the main thread checks for pending signals, but doesn't see any
- the main thread drains the wakeup fd
- the main thread goes back to sleep
- trip_signal sets is_tripped=1 and calls Py_AddPendingCall to notify
the main thread the it should run the Python-level signal handler
- the main thread doesn't notice because it's asleep
This has been causing repeated failures in the Trio test suite:
https://github.com/python-trio/trio/issues/119
(cherry picked from commit 4ae0149697)
If we have a chain of generators/coroutines that are 'yield from'ing
each other, then resuming the stack works like:
- call send() on the outermost generator
- this enters _PyEval_EvalFrameDefault, which re-executes the
YIELD_FROM opcode
- which calls send() on the next generator
- which enters _PyEval_EvalFrameDefault, which re-executes the
YIELD_FROM opcode
- ...etc.
However, every time we enter _PyEval_EvalFrameDefault, the first thing
we do is to check for pending signals, and if there are any then we
run the signal handler. And if it raises an exception, then we
immediately propagate that exception *instead* of starting to execute
bytecode. This means that e.g. a SIGINT at the wrong moment can "break
the chain" – it can be raised in the middle of our yield from chain,
with the bottom part of the stack abandoned for the garbage collector.
The fix is pretty simple: there's already a special case in
_PyEval_EvalFrameEx where it skips running signal handlers if the next
opcode is SETUP_FINALLY. (I don't see how this accomplishes anything
useful, but that's another story.) If we extend this check to also
skip running signal handlers when the next opcode is YIELD_FROM, then
that closes the hole – now the exception can only be raised at the
innermost stack frame.
This shouldn't have any performance implications, because the opcode
check happens inside the "slow path" after we've already determined
that there's a pending signal or something similar for us to process;
the vast majority of the time this isn't true and the new check
doesn't run at all..
(cherry picked from commit ab4413a7e9)
* RFC 1750 has been been obsoleted by RFC 4086.
* RFC 3280 has been obsoleted by RFC 5280.
* RFC 4366 has been obsoleted by RFC 6066.
(cherry picked from commit 63c2c8ac17)
Fix a reference leak in _io._WindowsConsoleIO: PyUnicode_FSDecoder()
always initialize decodedname when it succeed and it doesn't clear
input decodedname object.
(cherry picked from commit 29adc13bd7)
If pass a server_hostname= that fails IDNA decoding to SSLContext.wrap_socket or SSLContext.wrap_bio, then the SSLContext object had a spurious Py_DECREF called on it, eventually leading to segfaults.
(cherry picked from commit 65ece7ca23)
* bpo-30557: faulthandler now correctly filters and displays exception codes on Windows (#1924)
* bpo-30557: faulthandler now correctly filters and displays exception codes on Windows
* Adds test for non-fatal exceptions.
* Adds bpo number to comment.
* bpo-30557: Fix test_faulthandler (#1969)
On Windows 8, 8.1 and 10 at least, the exit code is the exception
code (no bit is cleared).
* bpo-30544: _io._WindowsConsoleIO.write raises the wrong error when WriteConsoleW fails
* bpo-30544: _io._WindowsConsoleIO.write raises the wrong error when WriteConsoleW fails
timegm() return type is time_t, not int. Use time_t to prevent the
following compiler warning on Windows:
timemodule.c: warning C4244: '=': conversion from 'time_t' to 'int',
possible loss of data
(cherry picked from commit 0d659e5614)
when there are no more `await` or `yield (from)` before return in coroutine,
cancel was ignored.
example:
async def coro():
asyncio.Task.current_task().cancel()
return 42
...
res = await coro() # should raise CancelledError
(cherry picked from commit 991adca012)
It was possible to get a core dump by using uninitialized
_json objects. Now __new__ methods create initialized objects.
__init__ methods are removed..
(cherry picked from commit 76a3e51a40)
* Remove conditional on free of `dps`, since `dps` is now allocated for
all versions of OpenSSL
* Remove call to `x509_check_ca` since it was only used to cache
the `crldp` field of the certificate
CRL_DIST_POINTS_free is available in all supported versions of OpenSSL
(recent 0.9.8+) and LibreSSL.
(cherry picked from commit 2849cc34a8)
bpo-29939: Changed test code to suppress a compiler warning, while taking care to avoid the code being optimized out by the compiler.
(cherry picked from commit 164d30eb1e)
* bpo-29942: Fix the use of recursion in itertools.chain.from_iterable.
Fix the use of recursion in itertools.chain.from_iterable. Using recursion
is unnecessary, and can easily cause stack overflows, especially when
building in low optimization modes or with Py_DEBUG enabled.
(cherry picked from commit 5466d4af5f)
Directory and zipfile execution previously added
the parent directory of the directory or zipfile
as sys.path[0] and then subsequently overwrote
it with the directory or zipfile itself.
This caused problems in isolated mode, as it
overwrote the "stdlib as a zip archive" entry
in sys.path, as the parent directory was
never added.
The attempted fix to that issue in bpo-29319
created the opposite problem in *non*-isolated
mode, by potentially leaving the parent
directory on sys.path instead of overwriting it.
This change fixes the root cause of the problem
by removing the whole "add-and-overwrite" dance
for sys.path[0], and instead simply never adds
the parent directory to sys.path in the first
place.
(cherry picked from commit d2977a3ae2)
bpo-29619: os.stat() and os.DirEntry.inodeo() now convert inode
(st_ino) using unsigned integers.
(cherry picked from commit 0f6d73343d)
(Misc/NEWS conflict handled manually.)
Miss Islington (bot)
Christian Heimes
Xiang Zhang
Victor Stinner
Serhiy Storchaka
Segev Finer
INADA Naoki
Gregory P. Smith
Christopher Wilcox
Oren Milman
Pauli Virtanen
Antoine Pitrou
Nir Soffer
Ned Deily
Yury Selivanov
Emily Morehouse
Zachary Ware
Nick Coghlan
Nathaniel J. Smith
Steve Dower
Mariatta
Benjamin Peterson
Vinay Sajip
T. Wouters
Christophe Zeitouny
Michael Seifert