Re: Fwd: Re: pf: BAD state happens often with portsnap fetch update

Colin Percival wrote:
> Chuck Swiger wrote:
>> FYI, if you pass a:
>>
>>   Connection: keep-alive
>>
>> ...header in the request, Squid will not close the client->proxy
>> connection and you ought to be able to re-use it to make additional
>> requests.
> 
> In HTTP/1.1, connections are assumed to be persistent unless declared otherwise.

Yeah.

>> PS: Squid "supports HTTP/1.0 persistent connections", from which the HTTP/1.1
>> style keepalives derive-- and quick testing suggests these persistent
>> connections work with either HTTP/1.0 or 1.1 specified in the request.
> 
> Grr.  I hate adding workarounds to improve compatibility with hopelessly
> antique code (seriously, why doesn't squid support http/1.1 yet?),

See src/HttpMsg.c and/or grep for proxy_keepalive:

/* returns true if connection should be "persistent"
  * after processing this message */
int
httpMsgIsPersistent(http_version_t http_ver, const HttpHeader * hdr)
{
#if WHEN_SQUID_IS_HTTP1_1
     if ((http_ver.major >= 1) && (http_ver.minor >= 1)) {
         /*
          * for modern versions of HTTP: persistent unless there is
          * a "Connection: close" header.
          */
         return !httpHeaderHasConnDir(hdr, "close");
     } else {
#else
[ ... ]

> but I guess I'll make phttpget emit a completely bogus "Connection: Keep-Alive"
> header to go along with its HTTP/1.1 requests.

There are lots of proxies and firewalls which restrict connections to HTTP/1.0 
behavior besides Squid.  It's not that horrible to have to request persistent 
connections explicitly if it solves the problem...and I hope it does.

-- 
-Chuck