udhcpc: use more correct, and more importantly, more understandable BPF code
authorVladislav Grishenko <themiron@mail.ru>
Mon, 14 Feb 2011 03:50:30 +0000 (04:50 +0100)
committerDenys Vlasenko <vda.linux@googlemail.com>
Mon, 14 Feb 2011 03:50:30 +0000 (04:50 +0100)
Signed-off-by: Vladislav Grishenko <themiron@mail.ru>
Signed-off-by: Denys Vlasenko <vda.linux@googlemail.com>
networking/udhcp/dhcpc.c

index 7e5ab61fd64639db1c925a1689e58f1f4253ea89..d97a404fa0f3109fce2eb9eafa516e1f1298ec46 100644 (file)
@@ -714,22 +714,25 @@ static int udhcp_raw_socket(int ifindex)
         *
         * TODO: make conditional?
         */
-#define SERVER_AND_CLIENT_PORTS  ((67 << 16) + 68)
        static const struct sock_filter filter_instr[] = {
-               /* check for udp */
+               /* load 9th byte (protocol) */
                BPF_STMT(BPF_LD|BPF_B|BPF_ABS, 9),
-               BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, IPPROTO_UDP, 2, 0),     /* L5, L1, is UDP? */
-               /* ugly check for arp on ethernet-like and IPv4 */
-               BPF_STMT(BPF_LD|BPF_W|BPF_ABS, 2),                      /* L1: */
-               BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, 0x08000604, 3, 4),      /* L3, L4 */
-               /* skip IP header */
-               BPF_STMT(BPF_LDX|BPF_B|BPF_MSH, 0),                     /* L5: */
-               /* check udp source and destination ports */
-               BPF_STMT(BPF_LD|BPF_W|BPF_IND, 0),
-               BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, SERVER_AND_CLIENT_PORTS, 0, 1), /* L3, L4 */
-               /* returns */
-               BPF_STMT(BPF_RET|BPF_K, 0x0fffffff ),                   /* L3: pass */
-               BPF_STMT(BPF_RET|BPF_K, 0),                             /* L4: reject */
+               /* jump to L1 if it is IPPROTO_UDP, else to L4 */
+               BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, IPPROTO_UDP, 0, 6),
+               /* L1: load halfword from offset 6 (flags and frag offset) */
+               BPF_STMT(BPF_LD|BPF_H|BPF_ABS, 6),
+               /* jump to L4 if any bits in frag offset field are set, else to L2 */
+               BPF_JUMP(BPF_JMP|BPF_JSET|BPF_K, 0x1fff, 4, 0),
+               /* L2: skip IP header (load index reg with header len) */
+               BPF_STMT(BPF_LDX|BPF_B|BPF_MSH, 0),
+               /* load udp destination port from halfword[header_len + 2] */
+               BPF_STMT(BPF_LD|BPF_H|BPF_IND, 2),
+               /* jump to L3 if udp dport is CLIENT_PORT, else to L4 */
+               BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, 68, 0, 1),
+               /* L3: accept packet */
+               BPF_STMT(BPF_RET|BPF_K, 0xffffffff),
+               /* L4: discard packet */
+               BPF_STMT(BPF_RET|BPF_K, 0),
        };
        static const struct sock_fprog filter_prog = {
                .len = sizeof(filter_instr) / sizeof(filter_instr[0]),
@@ -742,18 +745,19 @@ static int udhcp_raw_socket(int ifindex)
        fd = xsocket(PF_PACKET, SOCK_DGRAM, htons(ETH_P_IP));
        log1("Got raw socket fd %d", fd); //log2?
 
-       if (SERVER_PORT == 67 && CLIENT_PORT == 68) {
-               /* Use only if standard ports are in use */
+       sock.sll_family = AF_PACKET;
+       sock.sll_protocol = htons(ETH_P_IP);
+       sock.sll_ifindex = ifindex;
+       xbind(fd, (struct sockaddr *) &sock, sizeof(sock));
+
+       if (CLIENT_PORT == 68) {
+               /* Use only if standard port is in use */
                /* Ignoring error (kernel may lack support for this) */
                if (setsockopt(fd, SOL_SOCKET, SO_ATTACH_FILTER, &filter_prog,
                                sizeof(filter_prog)) >= 0)
                        log1("Attached filter to raw socket fd %d", fd); // log?
        }
 
-       sock.sll_family = AF_PACKET;
-       sock.sll_protocol = htons(ETH_P_IP);
-       sock.sll_ifindex = ifindex;
-       xbind(fd, (struct sockaddr *) &sock, sizeof(sock));
        log1("Created raw socket");
 
        return fd;