Submitted by: steve@openssl.org

More robust fix and workaround for PR#1949. Don't try to work out if there
is any write pending data as this can be unreliable: always flush.
This commit is contained in:
Dr. Stephen Henson
2010-01-26 19:40:36 +00:00
parent 9413788571
commit cc62974182
5 changed files with 38 additions and 48 deletions

View File

@@ -166,7 +166,6 @@ int ssl3_accept(SSL *s)
BUF_MEM *buf;
unsigned long l,Time=(unsigned long)time(NULL);
void (*cb)(const SSL *ssl,int type,int val)=NULL;
long num1;
int ret= -1;
int new_state,state,skip=0;
@@ -447,29 +446,24 @@ int ssl3_accept(SSL *s)
break;
case SSL3_ST_SW_FLUSH:
/* number of bytes to be flushed */
/* This originally and incorrectly called BIO_CTRL_INFO
* The reason why this is wrong is mentioned in PR#1949.
* Unfortunately, as suggested in that bug some
* versions of Apache unconditionally return 0
* for BIO_CTRL_WPENDING meaning we don't correctly
* flush data and some operations, like renegotiation,
* don't work. Other software may also be affected so
* call BIO_CTRL_INFO to retain compatibility with
* previous behaviour and BIO_CTRL_WPENDING if we
* get zero to address the PR#1949 case.
/* This code originally checked to see if
* any data was pending using BIO_CTRL_INFO
* and then flushed. This caused problems
* as documented in PR#1939. The proposed
* fix doesn't completely resolve this issue
* as buggy implementations of BIO_CTRL_PENDING
* still exist. So instead we just flush
* unconditionally.
*/
num1=BIO_ctrl(s->wbio,BIO_CTRL_INFO,0,NULL);
if (num1 == 0)
num1=BIO_ctrl(s->wbio,BIO_CTRL_WPENDING,0,NULL);
if (num1 > 0)
s->rwstate=SSL_WRITING;
if (BIO_flush(s->wbio) <= 0)
{
s->rwstate=SSL_WRITING;
num1=BIO_flush(s->wbio);
if (num1 <= 0) { ret= -1; goto end; }
s->rwstate=SSL_NOTHING;
ret= -1;
goto end;
}
s->rwstate=SSL_NOTHING;
s->state=s->s3->tmp.next_state;
break;