test_crypt segfaults when using libxcrypt instead of libcrypt

[stratakis]

BPO	32635
Nosy	@vstinner, @tiran, @benjaminp, @stratakis, @geofft
PRs	bpo-32635: Fix a segfault when importing the crypt module with libxcrypt. #5284 [2.7] bpo-32635: Fix a segfault when importing the crypt module with libxcrypt. (GH-5284) #5295 [3.6] bpo-32635: Fix a segfault when importing the crypt module with libxcrypt. (GH-5284) #5296
Files	00290-cryptmodule-Include-crypt.h-for-declaration-of-crypt.patch

^{Note: these values reflect the state of the issue at the time it was migrated and might not reflect the current state.}

Show more details

GitHub fields:

assignee = None
closed_at = <Date 2018-01-24.13:30:26.894>
created_at = <Date 2018-01-23.10:22:06.979>
labels = ['extension-modules', '3.7']
title = 'test_crypt segfaults when using libxcrypt instead of libcrypt'
updated_at = <Date 2021-07-27.21:33:23.914>
user = 'https://github.com/stratakis'

bugs.python.org fields:

activity = <Date 2021-07-27.21:33:23.914>
actor = 'geofft'
assignee = 'none'
closed = True
closed_date = <Date 2018-01-24.13:30:26.894>
closer = 'cstratak'
components = ['Extension Modules']
creation = <Date 2018-01-23.10:22:06.979>
creator = 'cstratak'
dependencies = []
files = ['47402']
hgrepos = []
issue_num = 32635
keywords = ['patch']
message_count = 10.0
messages = ['310490', '310503', '310563', '310573', '310575', '310579', '310581', '397959', '398254', '398322']
nosy_count = 6.0
nosy_names = ['vstinner', 'christian.heimes', 'benjamin.peterson', 'Arfrever', 'cstratak', 'geofft']
pr_nums = ['5284', '5295', '5296']
priority = 'normal'
resolution = 'fixed'
stage = 'resolved'
status = 'closed'
superseder = None
type = None
url = 'https://bugs.python.org/issue32635'
versions = ['Python 2.7', 'Python 3.6', 'Python 3.7']

[stratakis]

Currently in Fedora glibc stopped providing libcrypt[0] a change which is slowly being upstreamed as well[1] in favor of the libxcrypt project[2].

This causes a segfault when importing the crypt module as python assumes that crypt.h is always available.

Providing a working patch from the libxcrypt maintainer. I'll convert the patch into a PR.

[0] https://fedoraproject.org/wiki/Changes/Replace_glibc_libcrypt_with_libxcrypt
[1] https://sourceware.org/ml/libc-alpha/2017-08/msg01257.html
[2] https://github.com/besser82/libxcrypt

[vstinner]

New changeset e768c86 by Victor Stinner (stratakis) in branch 'master':
bpo-32635: Fix a segfault when importing the crypt module with libxcrypt. (bpo-5284)
e768c86

[benjaminp]

Incidentally, why did this manifest as a segfault rather than a compilation error?

[vstinner]

Charalampos, Christian: are you ok to backport the fix to Python 2.7 and 3.6? (PR 5295 and PR 5296)

[tiran]

yes, go ahead

[vstinner]

New changeset 6ccdad7 by Victor Stinner (Miss Islington (bot)) in branch '2.7':
bpo-32635: Fix a segfault when importing the crypt module with libxcrypt. (GH-5284) (bpo-5295)
6ccdad7

[vstinner]

New changeset 8e230e1 by Victor Stinner (Miss Islington (bot)) in branch '3.6':
bpo-32635: Fix a segfault when importing the crypt module with libxcrypt. (GH-5284) (bpo-5296)
8e230e1

[geofft]

Sorry to bump an old thread, but - does <crypt.h> need to be in the public Python.h header? Or would it be enough to be in _cryptmodule.c, which is the code that uses crypt_r()?

I can provide a quick patch to change that if that seems sound.

[Arfrever]

Geoffrey Thomas:
I agree that including crypt.h only where required is better.
Can you file new bug with patch?

Benjamin Peterson:
I suspect that with crypt.h not included in _cryptmodule.c, crypt() was treated by compiler as implicitly declared function (returning 'int'), however real crypt() function returns 'char*'.

[geofft]

Opened bpo-44751 and #27394 .