]> git.ipfire.org Git - thirdparty/Python/cpython.git/commit
[3.10] GH-95494: Fix transport EOF handling in OpenSSL 3.0 (GH-95495) (#103007)
authorMiss Islington (bot) <31488909+miss-islington@users.noreply.github.com>
Mon, 27 Mar 2023 14:14:24 +0000 (07:14 -0700)
committerGitHub <noreply@github.com>
Mon, 27 Mar 2023 14:14:24 +0000 (16:14 +0200)
commitb5bf6c1b2275bfa93e77989bec0711c3e97932a6
treec53b879781e31cc86c8768938cf804b12fec2be7
parentae8a721c2ba2a48de0cb69fa868fa8fc207b129d
[3.10] GH-95494: Fix transport EOF handling in OpenSSL 3.0 (GH-95495) (#103007)

GH-25309 enabled SSL_OP_IGNORE_UNEXPECTED_EOF by default, with a comment
that it restores OpenSSL 1.1.1 behavior, but this wasn't quite right.
That option causes OpenSSL to treat transport EOF as the same as
close_notify (i.e. SSL_ERROR_ZERO_RETURN), whereas Python actually has
distinct SSLEOFError and SSLZeroReturnError exceptions. (The latter is
usually mapped to a zero return from read.) In OpenSSL 1.1.1, the ssl
module would raise them for transport EOF and close_notify,
respectively. In OpenSSL 3.0, both act like close_notify.

Fix this by, instead, just detecting SSL_R_UNEXPECTED_EOF_WHILE_READING
and mapping that to the other exception type.

There doesn't seem to have been any unit test of this error, so fill in
the missing one. This had to be done with the BIO path because it's
actually slightly tricky to simulate a transport EOF with Python's fd
based APIs. (If you instruct the server to close the socket, it gets
confused, probably because the server's SSL object is still referencing
the now dead fd?)
(cherry picked from commit 420bbb783b43216cc897dc8914851899db37a31d)

Co-authored-by: David Benjamin <davidben@google.com>
Lib/test/test_ssl.py
Misc/NEWS.d/next/Library/2022-07-30-23-01-43.gh-issue-95495.RA-q1d.rst [new file with mode: 0644]
Modules/_ssl.c