Re: lingering problem: route_dst_netlink: can't find interface

[David Fifield <david () bamsoftware com>]

On Tue, Aug 28, 2012 at 12:44:12PM +0100, Djalal Harouni wrote:
> Hi,
>
> Thanks for sending all that info.
>
> Please see below (I'm not familiar with IP-over-InfiniBand interfaces)
>
> On Thu, Aug 23, 2012 at 08:51:27PM -0400, starlight.2012q3 () binnacle cx wrote:
> > So it seems to be a problem with 'libdnet'
> > failing when it sees an Infiniband
> > interface.  Built latest and ran test:
> >
> > $ dnet intf show
> > lo: flags=0x3<UP,LOOPBACK> mtu 16436
> >         inet 127.0.0.1/8
> >         alias ::1
> > sit0: flags=0x8<NOARP> mtu 1480
> > lt-dnet: intf_loop: Invalid argument
> >
> > Runs ok on an nearly identical CentOS 5.8
> > system that has no IB interfaces.
> >
> > Reported this to 'libdnet' owner/maintainer.
> Yes libdnet does not handle this. Plus these IPoIB (IP-over-InfiniBand)
> use 20bytes for hardware address, and current libdnet code can store only
> upto 16bytes, and I'm not sure how other parts of libdnet will deal
> with this.
>
>
> Another point is can we send ethernet frames in this case ? or only IP
> packets ? which I suppose can be handled by the '--send-ip' Nmap switch.
>
> Anyway you can start with the attached patch, and see if it will work,
> only 6bytes of hardware address will be reported.
>
> Please try to see if you can send IP packets with Nmap on that interface.

> Index: libdnet-stripped/include/dnet/arp.h
> ===================================================================
> --- libdnet-stripped/include/dnet/arp.h       (revision 29679)
> +++ libdnet-stripped/include/dnet/arp.h       (working copy)
> @@ -39,6 +39,7 @@
>  #define ARP_HRD_ETH  0x0001  /* ethernet hardware */
>  #define ARP_HRD_IEEE802      0x0006  /* IEEE 802 hardware */
>  
> +#define ARP_HRD_INFINIBAND 0x0020 /* InfiniBand */
>  #define ARP_HRD_IEEE80211_RADIOTAP 0x0323  /* IEEE 802.11 + radiotap header */
>  #define ARP_HRD_VOID 0xFFFF                  /* Void type, nothing is known */
>  
> Index: libdnet-stripped/src/addr.c
> ===================================================================
> --- libdnet-stripped/src/addr.c       (revision 29679)
> +++ libdnet-stripped/src/addr.c       (working copy)
> @@ -319,6 +319,7 @@
>       case AF_UNSPEC:
>       case ARP_HRD_ETH:       /* XXX- Linux arp(7) */
>       case ARP_HRD_IEEE80211_RADIOTAP: /* IEEE 802.11 + radiotap header */
> +     case ARP_HRD_INFINIBAND: /* InfiniBand */
>               a->addr_type = ADDR_TYPE_ETH;
>               a->addr_bits = ETH_ADDR_BITS;
>               memcpy(&a->addr_eth, sa->sa_data, ETH_ADDR_LEN);

This looks okay to me. Please apply it, Djalal. Also please add a
comment stating that only 6 of the 20 bytes of the hardware address are
captured, and link to this thread.

David Fifield
_______________________________________________
Sent through the nmap-dev mailing list
http://cgi.insecure.org/mailman/listinfo/nmap-dev
Archived at http://seclists.org/nmap-dev/