Re: [lwip-devel] Bug in ip_route() ?

[Christian Sasso]

Hi,

I didn't see any responses yet, so please let me make a correction.
The last line of the patch should have read:

-  if ((netif_default == NULL) || (!netif_is_up(netif_default))) {
+  if ((netif_default == NULL) || (!netif_is_up(netif_default)) ||
(ip_addr_isany(dest))) {

i.e. roughly netif should have been dest. I introduced this typo while
manually copying the patch (since for some reason cut'n'paste didn't
work between my VMs.) My apologies. Now since I had the opportunity to
give a second look to this last line, I realized what a pile of
useless non-sense it is, and I removed it. So, the new and shorter
patch is:

diff --git a/production/sw/tps/lwip/lwip/
src/core/ipv4/ip4.c
b/production/sw/tps/lwip/lwip/src/core/ipv4/ip4.c
index 322a3d4..fc139a2 100644
--- a/production/sw/tps/lwip/lwip/src/core/ipv4/ip4.c
+++ b/production/sw/tps/lwip/lwip/src/core/ipv4/ip4.c
@@ -178,14 +178,18 @@ ip_route(ip_addr_t *dest)
   /* iterate through netifs */
   for (netif = netif_list; netif != NULL; netif = netif->next) {
     /* network mask matches? */
-    if (netif_is_up(netif)) {
+    if ((netif_is_up(netif))
+#if LWIP_IPV6
+      && (!ip_addr_isany(&(netif->ip_addr)))
+#endif /* LWIP_IPV6 */
+    ) {
       if (ip_addr_netcmp(dest, &(netif->ip_addr), &(netif->netmask))) {
         /* return netif on which to forward IP packet */
         return netif;
       }
     }
   }

i.e. it is the same as yesterday's patch minus the final non-sense.
Sorry again for the typo. Do you think this makes some sense? Thank
you.

Ciao, Chris


On Sun, Jul 14, 2013 at 12:26 PM, Christian Sasso
<address@hidden> wrote:
> Hi,
>
> I'm Christian Sasso and about 4 months ago I started using LwIP on an
> embedded project.
>
> Yesterday I realized that ip_route() can return a non-NULL pointer to
> an interface that is up, but doesn't belong to any IPv4 subnet.
> In fact, when an interface netif doesn't belong to any IPv4 subnet
> both netif->netmask and netif->ip_addr are always zero, causing the
> condition
>     (ip_addr_netcmp(dest, &(netif->ip_addr), &(netif->netmask)))
> to always be true irrespective of the value of dest (the destination address.)
>
> This happened to me with a 6lowpan interface that is IPv6 only: it was
> returned by ip_route() and later an attempt was made to call
> netif->output() on it.
> I wrote a patch to fix the issue that simply requires netif->ip_addr
> to be non-zero when LWIP_IPV6 is enable, and so far it seems to work
> as expected:
>
> diff --git a/production/sw/tps/lwip/lwip/src/core/ipv4/ip4.c
> b/production/sw/tps/lwip/lwip/src/core/ipv4/ip4.c
> index 322a3d4..fc139a2 100644
> --- a/production/sw/tps/lwip/lwip/src/core/ipv4/ip4.c
> +++ b/production/sw/tps/lwip/lwip/src/core/ipv4/ip4.c
> @@ -178,14 +178,18 @@ ip_route(ip_addr_t *dest)
>    /* iterate through netifs */
>    for (netif = netif_list; netif != NULL; netif = netif->next) {
>      /* network mask matches? */
> -    if (netif_is_up(netif)) {
> +    if ((netif_is_up(netif))
> +#if LWIP_IPV6
> +      && (!ip_addr_isany(&(netif->ip_addr)))
> +#endif /* LWIP_IPV6 */
> +    ) {
>        if (ip_addr_netcmp(dest, &(netif->ip_addr), &(netif->netmask))) {
>          /* return netif on which to forward IP packet */
>          return netif;
>        }
>      }
>    }
> -  if ((netif_default == NULL) || (!netif_is_up(netif_default))) {
> +  if ((netif_default == NULL) || (!netif_is_up(netif_default)) ||
> (ip_addr_isany(&(netif->ip_addr)))) {
>      LWIP_DEBUGF(IP_DEBUG | LWIP_DBG_LEVEL_SERIOUS, ("ip_route: No
> route to %"U16_F".%"U16_F".%"U16_F".%"U16_F"\n",
>        ip4_addr1_16(dest), ip4_addr2_16(dest), ip4_addr3_16(dest),
> ip4_addr4_16(dest)));
>      IP_STATS_INC(ip.rterr);
>
> Without this, LwIP crashes on my system trying to execute
> netif->output(). Notice that even if there were an interface netif
> with a defined IPv4 subnet that satisfies
>     (ip_addr_netcmp(dest, &(netif->ip_addr), &(netif->netmask)))
> an IPv6 only interface would still risk to be returned from ip_route()
> if it happened to come first in netif_list. Is this a known problem,
> and if so, is the fix right, or I am missing something important?
> Thank you for your attention.
>
> Ciao, Chris