2024-02-21, 10:33:30 -0800, Jakub Kicinski wrote:
On Wed, 21 Feb 2024 14:59:40 +0100 Sabrina Dubroca wrote:
It's not exactly enough, since tls_record_content_type will return 0 on a content type mismatch. We'll have to translate that into an "error".
Ugh, that's unpleasant.
I think it would be a bit nicer to set err=1 and then check err != 0 in tls_sw_recvmsg (we can document that in a comment above process_rx_list) rather than making up a fake errno. See diff [1].
Or we could swap the 0/1 returns from tls_record_content_type and switch the err <= 0 tests to err != 0 after the existing calls, then process_rx_list doesn't have a weird special case [2].
What do you think?
I missed the error = 1 case, sorry. No strong preference, then. Checking for error = 1 will be as special as the new rx_more flag. Should I apply this version as is, then?
If you're ok with that version, sure. Thanks.