[ECOS] DHCP over multiple interfaces

[ECOS] DHCP over multiple interfaces

[Dave Lawrence]

I can't seem to get DHCP work over multiple interfaces.

The problem is that packets sent to 255.255.255.255 always
come out on eth1.

According to "TCP/IP Illustrated Volume 1" (Stevens)
section 12.2:
"Most BSD systems treat 255.255.255.255 as an alias
for the broadcast address of the first interface that
was configured"
and
"The Host Requirements RFC takes no stand on the issue of
whether a multihomed host should send a limited broadcast
out all its interfaces"

Not that sending the packets out to both interfaces would
be a solution either, because you could accept a lease from
the wrong network.  Currently if I have a DHCP server on the
eth1 network then eth0 will obtain its address from there.

Is it possible to configure more than one interface to use
DHCP?  If not, then how can I specify which interface
broadcasts 255.255.255.255 packets?


-- 
Before posting, please read the FAQ: http://ecos.sourceware.org/fom/ecos
and search the list archive: http://ecos.sourceware.org/ml/ecos-discuss

Re: [ECOS] DHCP over multiple interfaces

[Andrew Lunn]

On Wed, Jan 23, 2008 at 04:11:11PM +0000, Dave Lawrence wrote:
> I can't seem to get DHCP work over multiple interfaces.

The eCos DHCP client used to work with multiple interfaces. This is
why there is the half up states, so that it can force the packet out
the correct interface and the clearing of routes when performing DHCP
transactions.

If it does not work, i suggest you work your way back and see which
patch broke it.

      Andrew

-- 
Before posting, please read the FAQ: http://ecos.sourceware.org/fom/ecos
and search the list archive: http://ecos.sourceware.org/ml/ecos-discuss

[ECOS] Re: DHCP over multiple interfaces

[Dave Lawrence]

[-- Attachment #1: Type: text/plain, Size: 1466 bytes --]

Andrew Lunn wrote:
> On Wed, Jan 23, 2008 at 04:11:11PM +0000, Dave Lawrence wrote:
>> I can't seem to get DHCP work over multiple interfaces.
> 
> The eCos DHCP client used to work with multiple interfaces. This is
> why there is the half up states, so that it can force the packet out
> the correct interface and the clearing of routes when performing DHCP
> transactions.
> 
> If it does not work, i suggest you work your way back and see which
> patch broke it.
> 
>       Andrew
> 

I had a look on an older platform that uses an older ecos tree.  It does 
work but there were some problems:
  * Packets could come out on the wrong interface if two threads were 
simultaneously calling do_dhcp (on different interfaces)
  * Our application was also interfering with the routing tables in 
response to detecting devices going up / down

So I don't have any opinion about whether my fixes are of use to the 
wider ecos world but I've attached a patch in case.  It does the following:

* Provide dhcp_get_lock and dhcp_release_lock as global functions which 
the application can use to ensure interference is avoided
* Use these functions at the entry and exit points of do_dhcp to prevent 
it functioning reetrantly (is that a word?)

Unfortunately I'm no closer to fixing the original problem - on my new 
platform packets for 255.255.255.255 *always* come out on eth1.

Was I correct to continue with this thread or should I have switched to 
the patches list?

[-- Attachment #2: dhcp.patch --]
[-- Type: text/plain, Size: 4376 bytes --]

Index: net/common/current/include/dhcp.h
===================================================================
RCS file: /cvs/ecos/ecos-opt/net/net/common/current/include/dhcp.h,v
retrieving revision 1.2
diff -u -5 -r1.2 dhcp.h
--- net/common/current/include/dhcp.h	16 Oct 2003 08:13:46 -0000	1.2
+++ net/common/current/include/dhcp.h	25 Jan 2008 17:02:34 -0000
@@ -241,9 +241,15 @@
 
 extern int
 do_dhcp_release(const char *intf, struct bootp *res,
         cyg_uint8 *pstate, struct dhcp_lease *lease);
 
+extern void
+    dhcp_get_lock (void);
+
+extern void
+    dhcp_release_lock (void);
+
 #endif // CYGPKG_NET_DHCP
 
 #endif // CYGONCE_NET_TCPIP_DHCP_H
 // EOF dhcp.h
Index: net/common/current/src/dhcp_prot.c
===================================================================
RCS file: /cvs/ecos/ecos-opt/net/net/common/current/src/dhcp_prot.c,v
retrieving revision 1.20
diff -u -5 -r1.20 dhcp_prot.c
--- net/common/current/src/dhcp_prot.c	5 Mar 2006 12:48:37 -0000	1.20
+++ net/common/current/src/dhcp_prot.c	25 Jan 2008 17:02:35 -0000
@@ -88,10 +88,12 @@
     CYG_ASSERT( (strlen(hostname)<=CYGNUM_NET_DHCP_OPTION_HOST_NAME_LEN), "dhcp hostname too long" );
     strncpy(dhcp_hostname, hostname, CYGNUM_NET_DHCP_OPTION_HOST_NAME_LEN);
 }
 #endif
 
+static cyg_sem_t dhcp_sem;
+
 /* Forward reference prototypes. */
 static int unset_tag( struct bootp *ppkt, unsigned char tag );
 
 // ------------------------------------------------------------------------
 // Returns a pointer to the end of dhcp message (or NULL if invalid)
@@ -691,10 +693,12 @@
     struct bootp *received = &rx_local;
     struct bootp *xmit = res;
     struct bootp xmit2;
     int xlen;
 
+    dhcp_get_lock ();
+
     // First, get a socket on the interface in question.  But Zeroth, if
     // needs be, bring it to the half-up broadcast only state if needs be.
     
     if ( DHCPSTATE_INIT      == oldstate
          || DHCPSTATE_FAILED == oldstate
@@ -1044,10 +1048,11 @@
                     }
                 }
             }
 
             // Otherwise, nothing whatsoever to do...
+            dhcp_release_lock ();
             return true;
 
         case DHCPSTATE_RENEWING:
             // Just send what you got with a DHCPREQUEST in the message
             // type UNICAST straight to the server.  Then wait for an ACK.
@@ -1295,27 +1300,30 @@
                     }
                 }
             }
 
             // Otherwise, nothing whatsoever to do...
+            dhcp_release_lock ();
             return true;
 
         case DHCPSTATE_NOTBOUND:
             // All done with socket
             close(s);
             // No lease active
             no_lease( lease );
             // Leave interface up so app can tidy.
+            dhcp_release_lock ();
             return true;
 
         case DHCPSTATE_FAILED:
             // All done with socket
             close(s);
             // No lease active
             no_lease( lease );
             // Unconditionally down the interface.
             do_dhcp_down_net( intf, res, &oldstate, lease );
+            dhcp_release_lock ();
             return false;
 
         case DHCPSTATE_DO_RELEASE:
             // We have been forced here by external means, to release the
             // lease for graceful shutdown.
@@ -1362,17 +1370,19 @@
             break;
 
         default:
             no_lease( lease );
             close(s);
+            dhcp_release_lock ();
             return false;
         }
     }
 out:
     if (s != -1) 
       close (s);
     
+    dhcp_release_lock ();
     return false;
 }
 
 // ------------------------------------------------------------------------
 // Bring an interface down, failed to initialize it or lease is expired
@@ -1484,10 +1494,33 @@
         cyg_thread_delay( 100 );             // to let it leave the building
     }
     return true;
 }
 
+void dhcp_get_lock (void)
+{
+    static int dhcp_sem_initialised;
+
+    cyg_scheduler_lock ();
+
+    if (!dhcp_sem_initialised)
+    {
+        cyg_semaphore_init (&dhcp_sem, 1);
+        dhcp_sem_initialised = true;
+    }
+
+   cyg_scheduler_unlock ();
+
+   cyg_semaphore_wait (&dhcp_sem);
+}
+
+void dhcp_release_lock (void)
+{
+   cyg_semaphore_post (&dhcp_sem);
+}
+
+
 // ------------------------------------------------------------------------
 
 #endif // CYGPKG_NET_DHCP
 
 // EOF dhcp_prot.c


[-- Attachment #3: Type: text/plain, Size: 148 bytes --]

-- 
Before posting, please read the FAQ: http://ecos.sourceware.org/fom/ecos
and search the list archive: http://ecos.sourceware.org/ml/ecos-discuss

RE: [ECOS] DHCP over multiple interfaces

[Jay Foster]

I don't know about eCos, but I have solved a similar issue on Linux using
the SIOCGIFCONF, SIOCGIFINDEX ioctls, and the setsockopt(s, SOL_SOCKET,
SO_BINDTODEVICE) option to bind a socket to a specific interface for the
purposes of sending broadcasts.

Jay

-----Original Message-----
From: Dave Lawrence [mailto:dlawrence@ad-holdings.co.uk]
Sent: Wednesday, January 23, 2008 8:11 AM
To: ecos-discuss@sources.redhat.com
Subject: [ECOS] DHCP over multiple interfaces


I can't seem to get DHCP work over multiple interfaces.

The problem is that packets sent to 255.255.255.255 always
come out on eth1.

According to "TCP/IP Illustrated Volume 1" (Stevens)
section 12.2:
"Most BSD systems treat 255.255.255.255 as an alias
for the broadcast address of the first interface that
was configured"
and
"The Host Requirements RFC takes no stand on the issue of
whether a multihomed host should send a limited broadcast
out all its interfaces"

Not that sending the packets out to both interfaces would
be a solution either, because you could accept a lease from
the wrong network.  Currently if I have a DHCP server on the
eth1 network then eth0 will obtain its address from there.

Is it possible to configure more than one interface to use
DHCP?  If not, then how can I specify which interface
broadcasts 255.255.255.255 packets?


-- 
Before posting, please read the FAQ: http://ecos.sourceware.org/fom/ecos
and search the list archive: http://ecos.sourceware.org/ml/ecos-discuss

-- 
Before posting, please read the FAQ: http://ecos.sourceware.org/fom/ecos
and search the list archive: http://ecos.sourceware.org/ml/ecos-discuss