Re: [mpls-linux-devel] A few problems with ldpd

[James R. L. <jl...@mi...>]

I've submitted an alternate patch to solve this issue.  I make
a call checking if_is_loopback() on the ifp and skip accordingly.

In addition the quagga-mpls tree in my development depot has been
update to solve a couple of nagging bugs.

Anyone working with quagga-mpls will want to pull the latest
updates.

On Tue, Mar 25, 2008 at 03:33:03PM -0400, Scott A. Yoder wrote:
> Yeah that did solve the problem but isn't there some way to control which addresses
> are announced to peers? It seems like the address-mode lsr-id doesn't cause ldpd to only announce
> the configured lsr-id.
> 
> Here's a little patch to stop registering loopback addresses. Does Quagga have a LOOPBACK() macro
> like the kernel? I could only find an INADDR_LOOPBACK define.
> 
> --- ldp_zebra.c Wed Mar 26 00:05:22 2008
> +++ ldp_zebra-new.c     Tue Mar 25 23:44:43 2008
> @@ -201,16 +201,20 @@
>      c = zebra_interface_address_read(command, zclient->ibuf);
>      if (c == NULL || c->address->family != AF_INET) {
>         return 0;
>      }
> 
>      ifp = c->ifp;
>      p = c->address;
> 
> +    /* Don't register loopback addresses */
> +    if (((p->u.prefix4.s_addr) & htonl(0xff000000)) == htonl(0x7f000000))
> +        return 0;
> +
>      zlog_info("address add %s to interface %s",inet_ntoa(p->u.prefix4),
>         ifp->name);
> 
>      if (ldp) {
>         prefix2mpls_inet_addr(p, &addr.address);
>         iff.handle = ifp;
>         ldp_cfg_if_addr_set(ldp->h, &iff, &addr, LDP_CFG_ADD);
> 
> 
> Scott Yoder
> Support Engineer
> ImageStream Internet Solutions, Inc.
> E-mail: sy...@im...
> 
> James R. Leu wrote:
> > Good stuff.  Thank you for working with ldp-portable and taking the
> > time to dig into the code.  I've incorporated your fixes for the memory
> > leaks and I'm double checking the byte order fix you suggest.
> > 
> > As far as the multiple session issue, I think the fix is to never
> > send any 127.0.0.0/8 addresses in the address message.  One way to
> > accomplish this is to never register 127.0.0.0/8 addresses with
> > LDP by modifing ldp_interface_address_add() in ldp_zebra.c.
> > 
> > On Mon, Mar 24, 2008 at 04:58:20PM -0400, Scott A. Yoder wrote:
> >> Hi James. I've been able to get ldpd working quite well with 3 routers in my testbed. Even
> >> the targeted ldp sessions now appear to be working. I have a left LSR, middle, LSR and right LSR. The
> >> targeted session is between the left and right routers.
> >>
> >> I've found a few problems and want to get your take on how best to fix them.
> >>
> >> 1) The xconnect command needs to have the destination IP address stored in host byte order.
> >>
> >> --- ldp_vty.c   Tue Mar 25 01:19:59 2008
> >> +++ ldp_vty-new.c       Wed Mar 19 01:55:44 2008
> >> @@ -1618,6 +1618,7 @@
> >>    dest.port = 646;
> >>
> >>    VTY_GET_UINT32_RANGE("VCID",vcid,argv[1],0,255);
> >> +  dest.addr.u.ipv4 = ntohl(dest.addr.u.ipv4);
> >>
> >>    if (ldp_remote_peer_find(ldp, &dest)) {
> >>      return CMD_WARNING;
> >>
> >>
> >> 2) I found a couple of memory leaks with valgrind:
> >>
> >> --- ldp_session.c       Fri Jan  4 23:58:59 2008
> >> +++ ldp_session-new.c       Tue Mar 25 01:25:32 2008
> >> @@ -252,6 +252,8 @@
> >>  {
> >>    LDP_PRINT(NULL, "session delete");
> >>    MPLS_REFCNT_ASSERT(s, 0);
> >> +  ldp_buf_delete(s->tx_buffer);
> >> +  ldp_mesg_delete(s->tx_message);
> >>    mpls_free(s);
> >>  }
> >>
> >> --- ldp_buf.c   Tue Mar 25 01:29:01 2008
> >> +++ ldp_buf-new.c       Fri Mar 21 21:18:06 2008
> >> @@ -49,6 +49,8 @@
> >>  void ldp_buf_delete(ldp_buf * b)
> >>  {
> >>    MPLS_ASSERT(b);
> >> +  if (b->buffer)
> >> +    mpls_free(b->buffer);
> >>    mpls_free(b);
> >>  }
> >>
> >>
> >> These 2 leaks caused ldpd to grow several megabytes per hour on my system.
> >>
> >>
> >> 3) ldpd cannot establish more than 1 session at a time.
> >>
> >> I don't know if I have a config error but I keep getting a "ldp_addr_process: session X already advertised this address"
> >> error when the 2nd router attempts to start up a session. I think the address is 127.0.0.1. I tried changing the lsr-id and
> >> transport address but that caused ldpd to crash. The neighbor's status is stuck in the discovery state.
> >>
> >> Just to get around this problem and keep going I stuck a return MPLS_SUCCESS in ldpd/ldp_addr.c line 315:
> >>
> >>       /* the addr is in the tree */
> >>       if (addr->session) {
> >>         LDP_PRINT(g->user_data,
> >>           "ldp_addr_process: session (%d) already advertised this address\n",
> >>           addr->session->index);
> >>   return MPLS_SUCCESS;
> >>         return MPLS_FAILURE;
> >>       }
> >>
> >> Is there something I'm missing with the config to support multiple peers?
> >>
> >> Thanks,
> >> Scott Yoder
> >> Support Engineer
> >> ImageStream Internet Solutions, Inc.
> >> E-mail: sy...@im...
> >>
> >> -------------------------------------------------------------------------
> >> This SF.net email is sponsored by: Microsoft
> >> Defy all challenges. Microsoft(R) Visual Studio 2008.
> >> http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/
> >> _______________________________________________
> >> mpls-linux-devel mailing list
> >> mpl...@li...
> >> https://lists.sourceforge.net/lists/listinfo/mpls-linux-devel
> > 

-- 
James R. Leu
jl...@mi...