Re: [PATCH] apr_os_sock_put() fixup

[Jeff Trawick <tr...@attglobal.net>]

weigelt@metux.de wrote:
> hi folks, 
> 
> here's an little patch for fixing address problems when using
> apr_os_sock_put(). it fetches the socket addresses w/ getsockname()
> and getpeername() and puts them into the apr-socket structure.

1) apr_os_sock_put() should not be changed to do those syscalls.  Some 
callers use apr_os_sock_put() to make sockets and never need the local 
or remote names, and they shouldn't suffer the syscalls.

2) apr_os_sock_put() remembers that it doesn't know the remote and local 
names, so if the app ever queries for such info apr will do the 
getsockname() or getpeername() then.  So I wonder why you need to modify 
apr_os_sock_put() in the first place.  Maybe you found a hole in the 
existing logic to make the syscall if/when such info is queried?  If so, 
maybe that is what needs some changes.

3) Check out apr_os_sock_make(), which allows the caller to specify the 
remote and/or local address if for some reason you need the info known 
at the time the apr_socket_t is created.  The app would have to make the 
call(s), but if you're an Apache MPM don't you already know that 
information?

Re: [PATCH] apr_os_sock_put() fixup

[Jeff Trawick <tr...@attglobal.net>]

weigelt@metux.de wrote:
> On Sun, May 25, 2003 at 07:18:29PM -0400, Jeff Trawick wrote:
> 
> hi,
> 
> <snip>
> 
>>2) apr_os_sock_put() remembers that it doesn't know the remote and local 
>>names, so if the app ever queries for such info apr will do the 
>>getsockname() or getpeername() then.  So I wonder why you need to modify 
>>apr_os_sock_put() in the first place.  
> 
> We had the problem that requests coming from sockets created w/ 
> apr_os_sock_put() are not logged correctly. 

Here is why I think it should work okay...  perhaps some debugging on 
your side will show which assumption is incorrect.

1) apr_os_sock_put() sets the local_interface_unknown flag to 1

2) Apache's core_create_conn() hook is called as a create_connection() 
hook.  It fills out c->local_addr by calling apr_socket_addr_get().

3) apr_socket_addr_get() sees that the local_interface_unknown flag in 
the socket is 1, so it calls get_local_addr() which calls getsockname() 
and all is well.

4) Apache's core_create_conn() calls apr_sockaddr_ip_get() passing in 
c->local_addr, and fills in c->local_ip which is what the logging code uses.

So see where we deviate with muxmpm and we should be able to figure out 
what to fix.

> (please have a look at our muxmpm code)

sorry, can't find it now, don't have time anyway :(

Re: [PATCH] apr_os_sock_put() fixup

[we...@metux.de]

On Sun, May 25, 2003 at 07:18:29PM -0400, Jeff Trawick wrote:

hi,

<snip>
> 2) apr_os_sock_put() remembers that it doesn't know the remote and local 
> names, so if the app ever queries for such info apr will do the 
> getsockname() or getpeername() then.  So I wonder why you need to modify 
> apr_os_sock_put() in the first place.  
We had the problem that requests coming from sockets created w/ 
apr_os_sock_put() are not logged correctly. 
(please have a look at our muxmpm code)

cu
-- 
---------------------------------------------------------------------
 Enrico Weigelt    ==   metux ITS 
 Webhosting ab 5 EUR/Monat.          UUCP, rawIP und vieles mehr.

 phone:     +49 36207 519931         www:       http://www.metux.de/     
 fax:       +49 36207 519932         email:     contact@metux.de
 cellphone: +49 174 7066481	     smsgate:   sms.weigelt@metux.de
---------------------------------------------------------------------
 Diese Mail wurde mit UUCP versandt.      http://www.metux.de/uucp/