tomcat-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From jean-frederic clere <jfrederic.cl...@fujitsu-siemens.com>
Subject Re: jk_connect and multi-threading
Date Tue, 13 Jul 2004 11:20:04 GMT
Henri Gomez wrote:
> jean-frederic clere wrote:
> 
>> Henri Gomez wrote:
>>
>>> Henri Gomez wrote:
>>>
>>>> Bill Barker wrote:
>>>>
>>>>> ----- Original Message -----
>>>>> From: "Henri Gomez" <hgomez@apache.org>
>>>>> To: "Tomcat Developers List" <tomcat-dev@jakarta.apache.org>
>>>>> Sent: Monday, July 12, 2004 3:07 AM
>>>>> Subject: jk_connect and multi-threading
>>>>>
>>>>>
>>>>>
>>>>>> Hi to all,
>>>>>>
>>>>>> I'm looking for a strange problem under iSeries (AS/400) and wonder
>>>>>> about this look in jk_open_socket() (jk_connect.c) :
>>>>>>
>>>>>>         do {
>>>>>>             jk_log(l, JK_LOG_DEBUG, "jk_open_socket, try to connect
>>>>>> socket = %d to %s\n",
>>>>>>                    sock, jk_dump_hinfo(addr, buf));
>>>>>>
>>>>>>             ret = connect(sock,
>>>>>>                           (struct sockaddr *)addr,
>>>>>>                           sizeof(struct sockaddr_in));
>>>>>> #if defined(WIN32) || (defined(NETWARE) && defined(__NOVELL_LIBC__))
>>>>>>             if(SOCKET_ERROR == ret) {
>>>>>>                 errno = WSAGetLastError() - WSABASEERR;
>>>>>>             }
>>>>>> #endif /* WIN32 */
>>>>>>             jk_log(l, JK_LOG_DEBUG, "jk_open_socket, after connect

>>>>>> ret
>>>>>> = %d\n", ret);
>>>>>>         } while (-1 == ret && EINTR == errno);
>>>>>>
>>>>>>
>>>>>> What's the status on errno in multi-threaded environnement ?
>>>>>>
>>>>>
>>>>>
>>>>> On older *nix boxes it wasn't safe to use it with threads.  Don't 
>>>>> know about
>>>>> iSeries.  On Solaris it's per-thread (assuming that you've compiled
>>>>> with -D_REENTRANT).
>>>>>
>>>>>
>>>>>> Shouldn't we clear errno before the connect() call ?
>>>>>>
>>>>>
>>>>>
>>>>> It's not supposed to matter.
>>>>>
>>>>>
>>>>>> BTW, I wonder why we check the errno in such case...
>>>>
>>>>
>>>>
>>>>
>>>>
>>>> Ok, I take a look at IBM errno list and error 3021 is for EINVAL 
>>>> (invalid argument).
>>>>
>>>> I'll investigate this error farther...
>>>
>>>
>>>
>>>
>>> The problem could be related to BSD 4.4/Unix 98 where the sa_len 
>>> should be defined before connect call. May be one of the various 
>>> reason where
>>> jk_connect failed strangely on many boxes.
>>
>>
>>
>> Yes, APR uses:
>> +++
>> apr_status_t apr_socket_connect(apr_socket_t *sock, apr_sockaddr_t *sa)
>> {
>>     int rc;
>>
>>     do {
>>         rc = connect(sock->socketdes,
>>                      (const struct sockaddr *)&sa->sa.sin,
>>                      sa->salen);
>>     } while (rc == -1 && errno == EINTR);
>> +++
>> So setting  sa_len and USING sa_len instead sizeof(sockaddr_in) in the 
>> socket() should fix most of the problems.
> 
> 
> Ok, but how could we get the correct sa_len ?

That should be done in jk_resolve().

> 
> Patch welcome
> 
> 
> 
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: tomcat-dev-unsubscribe@jakarta.apache.org
> For additional commands, e-mail: tomcat-dev-help@jakarta.apache.org
> 
> 


---------------------------------------------------------------------
To unsubscribe, e-mail: tomcat-dev-unsubscribe@jakarta.apache.org
For additional commands, e-mail: tomcat-dev-help@jakarta.apache.org


Mime
View raw message