Breaks when running under Python 3.10
mgperkow opened this issue · 6 comments
sstp-server segfaults immediately after accepting and setting up a connection when running under Python 3.10. The segfault seems to happen within the context of codecmodule.c.
Can confirm.
I have also been able to track this down to a segfault in "codecmodule.c":
2022-03-12 11:02:35,636 INFO: Listening on 0.0.0.0:4433...
2022-03-12 11:02:47,270 INFO: SSTP control packet (CALL_CONNECT_REQUEST) received.
2022-03-12 11:02:47,270 INFO: Registered address 192.168.123.234
2022-03-12 11:02:47,287 VERBOSE: Raw data: 7eff7d23c0217d217d217d207d387d227d267d207d207d207d207d237d24c2277d257d2664e17d2f8a7d277d227d287d22d0e47
Next that happens in PPPDProtocol.out_received is frames = self.decoder.unescape(data) where the segfault occurs, and the following line of self.sstp.write_ppp_frames(frames) is never reached.
Server here is a current Arch Linux with CPython 3.10.2, tested with MikroTik ROS 7.1.3 client.
EDIT: Works fine when using CPython 3.9.10.
...well, since it's also working again with CPython 3.11, I guess it's nothing truly related to this package. 🤷♂️
Is there any workaround we can apply beyond creating a new environment for python 3.11 since this one is not GA yet?
I am running Ubuntu 22.04 which comes with 3.10 by default and I don't think it's possible for me to downport to 3.9.
Ideally there would be a patch to allow compatibility with 3.10 as well.
Thanks in advance
I happened to come back to this and decided to try to analyze it a little more in-depth myself. I'm not really a Python programmer, but I'm pretty experienced with other languages, so I did some breakpointing and debugging and found the exact instruction where the program was crashing. After a little trial and error (and perusing the Python developer docs a bit), I came up with this patch, which seems to correct the problem when running on Python 3.10.8 (which I'm currently using) after rebuilding the sstp-server package from source:
--- sstpd/codecmodule.c.orig 2022-10-17 15:06:17.238786534 -0400
+++ sstpd/codecmodule.c 2022-10-17 15:04:21.130988032 -0400
@@ -1,3 +1,4 @@
+#define PY_SSIZE_T_CLEAN
#include <Python.h>
#include <stdbool.h>
@@ -147,7 +148,7 @@ PppDecoder_unescape(PppDecoder *self, Py
PyObject* frame = Py_BuildValue("y#",
self->frame_buf, self->frame_buf_pos - 2);
if (PyList_Append(frames, frame) == -1) {
- Py_DECREF(frame);
+ if (frame) Py_DECREF(frame);
self->frame_buf_pos = 0;
PyBuffer_Release(&buf_in);
return NULL;
I happened to come back to this and decided to try to analyze it a little more in-depth myself. I'm not really a Python programmer, but I'm pretty experienced with other languages, so I did some breakpointing and debugging and found the exact instruction where the program was crashing. After a little trial and error (and perusing the Python developer docs a bit), I came up with this patch, which seems to correct the problem when running on Python 3.10.8 (which I'm currently using) after rebuilding the sstp-server package from source:
--- sstpd/codecmodule.c.orig 2022-10-17 15:06:17.238786534 -0400 +++ sstpd/codecmodule.c 2022-10-17 15:04:21.130988032 -0400 @@ -1,3 +1,4 @@ +#define PY_SSIZE_T_CLEAN #include <Python.h> #include <stdbool.h> @@ -147,7 +148,7 @@ PppDecoder_unescape(PppDecoder *self, Py PyObject* frame = Py_BuildValue("y#", self->frame_buf, self->frame_buf_pos - 2); if (PyList_Append(frames, frame) == -1) { - Py_DECREF(frame); + if (frame) Py_DECREF(frame); self->frame_buf_pos = 0; PyBuffer_Release(&buf_in); return NULL;
then why not opening a pull request ?
Oh, I suppose I could. I'm kinda new to that end of the process, but I'll read through the documentation on pull requests and see if I can get one going.
Any tips for installing python 3.11 in a way that it works with the sstpd command on Arch?
Been struggling to set it up