ref: 28347e249511cca8718261c13dc7cdb7954a828e
parent: 43129a4c129b47dcfff95bf91abe9408b5164a41
author: Timothy B. Terriberry <tterribe@xiph.org>
date: Tue Aug 6 09:56:30 EDT 2013
Minor comment updates. We've been able to differentiate between a read error and EOF since e2d7b266.
--- a/src/http.c
+++ b/src/http.c
@@ -888,14 +888,8 @@
However, we will not wait if this would block (it's not worth the savings
from session resumption to do so).
Clients (that's us) MAY resume a TLS session that ended with an incomplete
- close, according to RFC 2818, so that's no reason to make sure the server
- shut things down gracefully.
- It also says "client implementations MUST treat any premature closes as
- errors and the data received as potentially truncated," but libopusfile
- treats errors and potentially truncated data in unseekable streams just
- like a normal EOF.
- We warn about this in the docs, and give some suggestions if you truly want
- to avoid truncation attacks.*/
+ close, according to RFC 2818, so there's no reason to make sure the server
+ shut things down gracefully.*/
if(_gracefully&&_conn->ssl_conn!=NULL)SSL_shutdown(_conn->ssl_conn);
op_http_conn_clear(_conn);
_conn->next_pos=-1;
@@ -1025,6 +1019,9 @@
fd.fd=_conn->fd;
ssl_conn=_conn->ssl_conn;
nread=nread_unblocked=0;
+ /*RFC 2818 says "client implementations MUST treat any premature closes as
+ errors and the data received as potentially truncated," so we make very
+ sure to report read errors upwards.*/
do{
int err;
if(ssl_conn!=NULL){