[NETFILTER]: Fix ip6_tables protocol bypass bug

As reported by Mark Dowd <Mark_Dowd@McAfee.com>, ip6_tables is susceptible
to a fragmentation attack causing false negatives on protocol matches.

When the protocol header doesn't follow the fragment header immediately,
the fragment header contains the protocol number of the next extension
header. When the extension header and the protocol header are sent in
a second fragment a rule like "ip6tables .. -p udp -j DROP" will never
match.

Drop fragments that are at offset 0 and don't contain the final protocol
header regardless of the ruleset, since this should not happen normally.

With help from Yasuyuki KOZAKAI <yasuyuki.kozakai@toshiba.co.jp>.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

net/ipv6/netfilter/ip6_tables.c

@@ -111,7 +111,7 @@ ip6_packet_match(const struct sk_buff *skb,
 		 const char *outdev,
 		 const struct ip6t_ip6 *ip6info,
 		 unsigned int *protoff,
-		 int *fragoff)
+		 int *fragoff, int *hotdrop)
 {
 	size_t i;
 	unsigned long ret;
@@ -169,9 +169,11 @@ ip6_packet_match(const struct sk_buff *skb,
 		unsigned short _frag_off;
 
 		protohdr = ipv6_find_hdr(skb, protoff, -1, &_frag_off);
-		if (protohdr < 0)
+		if (protohdr < 0) {
+			if (_frag_off == 0)
+				*hotdrop = 1;
 			return 0;
-
+		}
 		*fragoff = _frag_off;
 
 		dprintf("Packet protocol %hi ?= %s%hi.\n",
@@ -290,7 +292,7 @@ ip6t_do_table(struct sk_buff **pskb,
 		IP_NF_ASSERT(e);
 		IP_NF_ASSERT(back);
 		if (ip6_packet_match(*pskb, indev, outdev, &e->ipv6,
-			&protoff, &offset)) {
+			&protoff, &offset, &hotdrop)) {
 			struct ip6t_entry_target *t;
 
 			if (IP6T_MATCH_ITERATE(e, do_match,