[lwip-users] Re: [lwip] ARP bug (compare MAC instead of IP)

[John C. Toman]

Hi Leon,

With any luck, one of the two of us will get something with arp cache 
timeouts working in the near future, and this will cease to be a 
problem. Is your ARP working with lwip already? I was going to extend 
and/or complete Adam's implementation at my first opportunity, which may 
be soon (my 0.5.0 port is now close to the testing phase)...

Thanks,

John

address@hidden wrote:

>Hello John,
>
>you are correct.
>
>Currently, without timeouts, however, it is not appropriate to store more
>than one
>(IP, MAC) mapping for each MAC, as the cache grows dirty over time. Which
>results in miscommunication, or communication starvation.
>
>Something I experienced myself. My own "solution" was to compare MAC
>addresses,
>as I (falsy) reasoned that any interface needs only one (IP,MAC) mapping in
>order to
>be reachable on the physical network. Later I added timeouts, without
>rethinking the
>stuff.
>
>In short, both IP address comparison AND timeouts are necessary to ensure
>correct
>operation.
>
>Regards,
>
>Leon Woestenberg.
>
>
>
>                                                                               
>                                           
>                    "John C. Toman"                                            
>                                           
>                    <address@hidden       To:     address@hidden               
>                                         
>                    -tech.com>                cc:                              
>                                           
>                    Sent by:                  Subject:     Re: [lwip] ARP bug 
> (compare MAC instead of IP)                
>                    address@hidden                                             
>                                       
>                                                                               
>                                           
>                                                                               
>                                           
>                    17-12-01 19:05                                             
>                                           
>                    Please respond to                                          
>                                           
>                    lwip                                                       
>                                           
>                                                                               
>                                           
>                                                                               
>                                           
>
>
>
>
>Hi Leon,
>
>It is very possible for one MAC address to be configured for two or more
>IP addresses! In my own office environment, this is done for various
>reasons. Proxy ARP (often used for PPP and firewalls) is another case
>where this is done. Thus it is appropriate to use IP addresses as a
>comparison point, not MAC addresses.
>
>With an ARP cache, and the scenario you mentioned (where an IP address
>is changed), the old entry should quickly time out when it isn't used.
>Your scenario is one of the key arguments for quick ARP cache timeouts.
>Timeouts aren't in the current lwIP ARP implementation but will be at
>some point. Another stack I've used sets ARP cache timeouts for five
>minutes. Other than not implementing timeouts, I believe Adam's
>implementation of this is correct.
>
>Regards,
>
>John
>
>address@hidden wrote:
>
>>Hello Adam,
>>
>>in the ARP code, when adding an (MAC, IP) entry, you use IP
>>addresses to compare. You should, however, use the MAC
>>addresses to find an old entry.
>>
>>----
>>To see what goes wrong, using IP address compare, an example:
>>
>>Suppose client with MAC 01:02:03:04:05:06 has obtained
>>IP address 192.168.0.2, and by whatever means, is configured
>>to use 192.168.0.3 now.
>>
>>Adding this (MAC, IP) mapping will not find the old entry, and create
>>a new entry in the ARP cache.  The ARP cache will now hold two
>>mappings with the same MAC address.
>>----
>>
>>static void
>>add_arp_entry(struct ip_addr *ipaddr, struct eth_addr *ethaddr)
>>{
>> u8_t i, j, k;
>> u8_t maxage;
>>
>> /* Walk through the ARP mapping table and try to find an entry to
>>    update. If none is found, the IP -> MAC address mapping is
>>    inserted in the ARP table. */
>> for(i = 0; i < ARPTABLE_SIZE; ++i) {
>>
>>   /* Only check those entries that are actually in use. */
>>   if(!ip_addr_isany(&arp_table[i].ipaddr)) {
>>     /* Check if the source IP address of the incoming packet matches
>>        the IP address in this ARP table entry. */
>>     if(ip_addr_cmp(ipaddr, &arp_table[i].ipaddr)) {
>><--------------------------- should test MAC bytes here, not IP address
>>    /* An old entry found, update this and return. */
>>
>>[This message was sent through the lwip discussion list.]
>>
>
>
>
>[This message was sent through the lwip discussion list.]
>
>
>
>
>[This message was sent through the lwip discussion list.]
>



[This message was sent through the lwip discussion list.]