I'm working on multicast and encountered this hash in linux code(ipmr.c). As far I understand, the hash table size is 64. And MFC_HASH takes highest 8 bits of ip destination address and highest 6 bits of source ip address. It XORs both and AND it with 63. Hence the result is bound to be between 0-63.

Please find the below code for more information.

struct mr_table {
          struct list_head        list;
          possible_net_t          net;
          u32                     id;
          struct sock __rcu       *mroute_sk;
          struct timer_list       ipmr_expire_timer;
          struct list_head        mfc_unres_queue;
          struct list_head        mfc_cache_array[MFC_LINES];
...
...
}

 #define MFC_LINES               64

 #ifdef __BIG_ENDIAN
 #define MFC_HASH(a,b)   ((((a)>>24)^((b)>>26))&(MFC_LINES-1))
 #else
 #define MFC_HASH(a,b)   (((a)^((b)>>2))&(MFC_LINES-1))
 #endif

What is special about this hash? How is it better than just adding the two ip addresses and doing a modulo by 63. Or just doing a modulo on the group ip address? Also if I want to increase the hash table size to 128 from 64 then is it enough if I change the MFC_LINE to 128? Or do I need to change the number of bits MFC_HASH uses from group-ip and source-ip? Can someone please help? My machine is big endian.

Thank you.