IPv6: Complete IPV6_DONTFRAG support

Finally add support to detect a local IPV6_DONTFRAG event and return the relevant data to the user if they've enabled IPV6_RECVPATHMTU on the socket. The next recvmsg() will return no data, but have an IPV6_PATHMTU as ancillary data. Signed-off-by: Brian Haley <brian.haley@hp.com> Signed-off-by: David S. Miller <davem@davemloft.net>

IPv6: Complete IPV6_DONTFRAG support
Finally add support to detect a local IPV6_DONTFRAG event and return the relevant data to the user if they've enabled IPV6_RECVPATHMTU on the socket. The next recvmsg() will return no data, but have an IPV6_PATHMTU as ancillary data. Signed-off-by: Brian Haley <brian.haley@hp.com> Signed-off-by: David S. Miller <davem@davemloft.net>
Brian Haley · David S. Miller
1 parent 13b52cd446
Showing 7 changed files with 116 additions and 8 deletions Side-by-side Diff
include/linux/ipv6.h
include/net/ipv6.h
net/ipv6/af_inet6.c
net/ipv6/datagram.c
net/ipv6/ip6_output.c
net/ipv6/raw.c
net/ipv6/udp.c
@@ -257,6 +257,7 @@
 };
  
 #define IP6CB(skb)	((struct inet6_skb_parm*)((skb)->cb))
+#define IP6CBMTU(skb)	((struct ip6_mtuinfo *)((skb)->cb))
  
 static inline int inet6_iif(const struct sk_buff *skb)
 {
@@ -366,6 +367,7 @@
  
 	struct ipv6_txoptions	*opt;
 	struct sk_buff		*pktoptions;
+	struct sk_buff		*rxpmtu;
 	struct {
 		struct ipv6_txoptions *opt;
 		u8 hop_limit;
@@ -578,9 +578,11 @@
 						     struct sockaddr *addr, int addr_len);
  
 extern int 			ipv6_recv_error(struct sock *sk, struct msghdr *msg, int len);
+extern int 			ipv6_recv_rxpmtu(struct sock *sk, struct msghdr *msg, int len);
 extern void			ipv6_icmp_error(struct sock *sk, struct sk_buff *skb, int err, __be16 port,
 						u32 info, u8 *payload);
 extern void			ipv6_local_error(struct sock *sk, int err, struct flowi *fl, u32 info);
+extern void			ipv6_local_rxpmtu(struct sock *sk, struct flowi *fl, u32 mtu);
  
 extern int inet6_release(struct socket *sock);
 extern int inet6_bind(struct socket *sock, struct sockaddr *uaddr, 
@@ -417,6 +417,9 @@
 	if ((skb = xchg(&np->pktoptions, NULL)) != NULL)
 		kfree_skb(skb);
  
+	if ((skb = xchg(&np->rxpmtu, NULL)) != NULL)
+		kfree_skb(skb);
+
 	/* Free flowlabels */
 	fl6_free_socklist(sk);
  
@@ -278,6 +278,45 @@
 		kfree_skb(skb);
 }
  
+void ipv6_local_rxpmtu(struct sock *sk, struct flowi *fl, u32 mtu)
+{
+	struct ipv6_pinfo *np = inet6_sk(sk);
+	struct ipv6hdr *iph;
+	struct sk_buff *skb;
+	struct ip6_mtuinfo *mtu_info;
+
+	if (!np->rxopt.bits.rxpmtu)
+		return;
+
+	skb = alloc_skb(sizeof(struct ipv6hdr), GFP_ATOMIC);
+	if (!skb)
+		return;
+
+	skb_put(skb, sizeof(struct ipv6hdr));
+	skb_reset_network_header(skb);
+	iph = ipv6_hdr(skb);
+	ipv6_addr_copy(&iph->daddr, &fl->fl6_dst);
+
+	mtu_info = IP6CBMTU(skb);
+	if (!mtu_info) {
+		kfree_skb(skb);
+		return;
+	}
+
+	mtu_info->ip6m_mtu = mtu;
+	mtu_info->ip6m_addr.sin6_family = AF_INET6;
+	mtu_info->ip6m_addr.sin6_port = 0;
+	mtu_info->ip6m_addr.sin6_flowinfo = 0;
+	mtu_info->ip6m_addr.sin6_scope_id = fl->oif;
+	ipv6_addr_copy(&mtu_info->ip6m_addr.sin6_addr, &ipv6_hdr(skb)->daddr);
+
+	__skb_pull(skb, skb_tail_pointer(skb) - skb->data);
+	skb_reset_transport_header(skb);
+
+	skb = xchg(&np->rxpmtu, skb);
+	kfree_skb(skb);
+}
+
 /*
  *	Handle MSG_ERRQUEUE
  */
@@ -381,6 +420,54 @@
 	return err;
 }
  
+/*
+ *	Handle IPV6_RECVPATHMTU
+ */
+int ipv6_recv_rxpmtu(struct sock *sk, struct msghdr *msg, int len)
+{
+	struct ipv6_pinfo *np = inet6_sk(sk);
+	struct sk_buff *skb;
+	struct sockaddr_in6 *sin;
+	struct ip6_mtuinfo mtu_info;
+	int err;
+	int copied;
+
+	err = -EAGAIN;
+	skb = xchg(&np->rxpmtu, NULL);
+	if (skb == NULL)
+		goto out;
+
+	copied = skb->len;
+	if (copied > len) {
+		msg->msg_flags |= MSG_TRUNC;
+		copied = len;
+	}
+	err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
+	if (err)
+		goto out_free_skb;
+
+	sock_recv_timestamp(msg, sk, skb);
+
+	memcpy(&mtu_info, IP6CBMTU(skb), sizeof(mtu_info));
+
+	sin = (struct sockaddr_in6 *)msg->msg_name;
+	if (sin) {
+		sin->sin6_family = AF_INET6;
+		sin->sin6_flowinfo = 0;
+		sin->sin6_port = 0;
+		sin->sin6_scope_id = mtu_info.ip6m_addr.sin6_scope_id;
+		ipv6_addr_copy(&sin->sin6_addr, &mtu_info.ip6m_addr.sin6_addr);
+	}
+
+	put_cmsg(msg, SOL_IPV6, IPV6_PATHMTU, sizeof(mtu_info), &mtu_info);
+
+	err = copied;
+
+out_free_skb:
+	kfree_skb(skb);
+out:
+	return err;
+}
  
  
 int datagram_recv_ctl(struct sock *sk, struct msghdr *msg, struct sk_buff *skb)
@@ -1219,15 +1219,23 @@
 	 */
  
 	inet->cork.length += length;
-	if (((length > mtu) && (sk->sk_protocol == IPPROTO_UDP)) &&
-	    (rt->u.dst.dev->features & NETIF_F_UFO)) {
+	if (length > mtu) {
+		int proto = sk->sk_protocol;
+		if (dontfrag && (proto == IPPROTO_UDP || proto == IPPROTO_RAW)){
+			ipv6_local_rxpmtu(sk, fl, mtu-exthdrlen);
+			return -EMSGSIZE;
+		}
  
-		err = ip6_ufo_append_data(sk, getfrag, from, length, hh_len,
-					  fragheaderlen, transhdrlen, mtu,
-					  flags);
-		if (err)
-			goto error;
-		return 0;
+		if (proto == IPPROTO_UDP &&
+		    (rt->u.dst.dev->features & NETIF_F_UFO)) {
+
+			err = ip6_ufo_append_data(sk, getfrag, from, length,
+						  hh_len, fragheaderlen,
+						  transhdrlen, mtu, flags);
+			if (err)
+				goto error;
+			return 0;
+		}
 	}
  
 	if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
@@ -461,6 +461,9 @@
 	if (flags & MSG_ERRQUEUE)
 		return ipv6_recv_error(sk, msg, len);
  
+	if (np->rxpmtu && np->rxopt.bits.rxpmtu)
+		return ipv6_recv_rxpmtu(sk, msg, len);
+
 	skb = skb_recv_datagram(sk, flags, noblock, &err);
 	if (!skb)
 		goto out;
@@ -335,6 +335,9 @@
 	if (flags & MSG_ERRQUEUE)
 		return ipv6_recv_error(sk, msg, len);
  
+	if (np->rxpmtu && np->rxopt.bits.rxpmtu)
+		return ipv6_recv_rxpmtu(sk, msg, len);
+
 try_again:
 	skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
 				  &peeked, &err);
...	...	@@ -257,6 +257,7 @@
257	257	};
258	258
259	259	#define IP6CB(skb) ((struct inet6_skb_parm*)((skb)->cb))
	260	+#define IP6CBMTU(skb) ((struct ip6_mtuinfo *)((skb)->cb))
260	261
261	262	static inline int inet6_iif(const struct sk_buff *skb)
262	263	{
...	...	@@ -366,6 +367,7 @@
366	367
367	368	struct ipv6_txoptions *opt;
368	369	struct sk_buff *pktoptions;
	370	+ struct sk_buff *rxpmtu;
369	371	struct {
370	372	struct ipv6_txoptions *opt;
371	373	u8 hop_limit;
...	...	@@ -578,9 +578,11 @@
578	578	struct sockaddr *addr, int addr_len);
579	579
580	580	extern int ipv6_recv_error(struct sock sk, struct msghdr msg, int len);
	581	+extern int ipv6_recv_rxpmtu(struct sock sk, struct msghdr msg, int len);
581	582	extern void ipv6_icmp_error(struct sock sk, struct sk_buff skb, int err, __be16 port,
582	583	u32 info, u8 *payload);
583	584	extern void ipv6_local_error(struct sock sk, int err, struct flowi fl, u32 info);
	585	+extern void ipv6_local_rxpmtu(struct sock sk, struct flowi fl, u32 mtu);
584	586
585	587	extern int inet6_release(struct socket *sock);
586	588	extern int inet6_bind(struct socket sock, struct sockaddr uaddr,
...	...	@@ -417,6 +417,9 @@
417	417	if ((skb = xchg(&np->pktoptions, NULL)) != NULL)
418	418	kfree_skb(skb);
419	419
	420	+ if ((skb = xchg(&np->rxpmtu, NULL)) != NULL)
	421	+ kfree_skb(skb);
	422	+
420	423	/* Free flowlabels */
421	424	fl6_free_socklist(sk);
422	425
...	...	@@ -278,6 +278,45 @@
278	278	kfree_skb(skb);
279	279	}
280	280
	281	+void ipv6_local_rxpmtu(struct sock sk, struct flowi fl, u32 mtu)
	282	+{
	283	+ struct ipv6_pinfo *np = inet6_sk(sk);
	284	+ struct ipv6hdr *iph;
	285	+ struct sk_buff *skb;
	286	+ struct ip6_mtuinfo *mtu_info;
	287	+
	288	+ if (!np->rxopt.bits.rxpmtu)
	289	+ return;
	290	+
	291	+ skb = alloc_skb(sizeof(struct ipv6hdr), GFP_ATOMIC);
	292	+ if (!skb)
	293	+ return;
	294	+
	295	+ skb_put(skb, sizeof(struct ipv6hdr));
	296	+ skb_reset_network_header(skb);
	297	+ iph = ipv6_hdr(skb);
	298	+ ipv6_addr_copy(&iph->daddr, &fl->fl6_dst);
	299	+
	300	+ mtu_info = IP6CBMTU(skb);
	301	+ if (!mtu_info) {
	302	+ kfree_skb(skb);
	303	+ return;
	304	+ }
	305	+
	306	+ mtu_info->ip6m_mtu = mtu;
	307	+ mtu_info->ip6m_addr.sin6_family = AF_INET6;
	308	+ mtu_info->ip6m_addr.sin6_port = 0;
	309	+ mtu_info->ip6m_addr.sin6_flowinfo = 0;
	310	+ mtu_info->ip6m_addr.sin6_scope_id = fl->oif;
	311	+ ipv6_addr_copy(&mtu_info->ip6m_addr.sin6_addr, &ipv6_hdr(skb)->daddr);
	312	+
	313	+ __skb_pull(skb, skb_tail_pointer(skb) - skb->data);
	314	+ skb_reset_transport_header(skb);
	315	+
	316	+ skb = xchg(&np->rxpmtu, skb);
	317	+ kfree_skb(skb);
	318	+}
	319	+
281	320	/*
282	321	* Handle MSG_ERRQUEUE
283	322	*/
...	...	@@ -381,6 +420,54 @@
381	420	return err;
382	421	}
383	422
	423	+/*
	424	+ * Handle IPV6_RECVPATHMTU
	425	+ */
	426	+int ipv6_recv_rxpmtu(struct sock sk, struct msghdr msg, int len)
	427	+{
	428	+ struct ipv6_pinfo *np = inet6_sk(sk);
	429	+ struct sk_buff *skb;
	430	+ struct sockaddr_in6 *sin;
	431	+ struct ip6_mtuinfo mtu_info;
	432	+ int err;
	433	+ int copied;
	434	+
	435	+ err = -EAGAIN;
	436	+ skb = xchg(&np->rxpmtu, NULL);
	437	+ if (skb == NULL)
	438	+ goto out;
	439	+
	440	+ copied = skb->len;
	441	+ if (copied > len) {
	442	+ msg->msg_flags \|= MSG_TRUNC;
	443	+ copied = len;
	444	+ }
	445	+ err = skb_copy_datagram_iovec(skb, 0, msg->msg_iov, copied);
	446	+ if (err)
	447	+ goto out_free_skb;
	448	+
	449	+ sock_recv_timestamp(msg, sk, skb);
	450	+
	451	+ memcpy(&mtu_info, IP6CBMTU(skb), sizeof(mtu_info));
	452	+
	453	+ sin = (struct sockaddr_in6 *)msg->msg_name;
	454	+ if (sin) {
	455	+ sin->sin6_family = AF_INET6;
	456	+ sin->sin6_flowinfo = 0;
	457	+ sin->sin6_port = 0;
	458	+ sin->sin6_scope_id = mtu_info.ip6m_addr.sin6_scope_id;
	459	+ ipv6_addr_copy(&sin->sin6_addr, &mtu_info.ip6m_addr.sin6_addr);
	460	+ }
	461	+
	462	+ put_cmsg(msg, SOL_IPV6, IPV6_PATHMTU, sizeof(mtu_info), &mtu_info);
	463	+
	464	+ err = copied;
	465	+
	466	+out_free_skb:
	467	+ kfree_skb(skb);
	468	+out:
	469	+ return err;
	470	+}
384	471
385	472
386	473	int datagram_recv_ctl(struct sock sk, struct msghdr msg, struct sk_buff *skb)
...	...	@@ -1219,15 +1219,23 @@
1219	1219	*/
1220	1220
1221	1221	inet->cork.length += length;
1222		- if (((length > mtu) && (sk->sk_protocol == IPPROTO_UDP)) &&
1223		- (rt->u.dst.dev->features & NETIF_F_UFO)) {
	1222	+ if (length > mtu) {
	1223	+ int proto = sk->sk_protocol;
	1224	+ if (dontfrag && (proto == IPPROTO_UDP \|\| proto == IPPROTO_RAW)){
	1225	+ ipv6_local_rxpmtu(sk, fl, mtu-exthdrlen);
	1226	+ return -EMSGSIZE;
	1227	+ }
1224	1228
1225		- err = ip6_ufo_append_data(sk, getfrag, from, length, hh_len,
1226		- fragheaderlen, transhdrlen, mtu,
1227		- flags);
1228		- if (err)
1229		- goto error;
1230		- return 0;
	1229	+ if (proto == IPPROTO_UDP &&
	1230	+ (rt->u.dst.dev->features & NETIF_F_UFO)) {
	1231	+
	1232	+ err = ip6_ufo_append_data(sk, getfrag, from, length,
	1233	+ hh_len, fragheaderlen,
	1234	+ transhdrlen, mtu, flags);
	1235	+ if (err)
	1236	+ goto error;
	1237	+ return 0;
	1238	+ }
1231	1239	}
1232	1240
1233	1241	if ((skb = skb_peek_tail(&sk->sk_write_queue)) == NULL)
...	...	@@ -461,6 +461,9 @@
461	461	if (flags & MSG_ERRQUEUE)
462	462	return ipv6_recv_error(sk, msg, len);
463	463
	464	+ if (np->rxpmtu && np->rxopt.bits.rxpmtu)
	465	+ return ipv6_recv_rxpmtu(sk, msg, len);
	466	+
464	467	skb = skb_recv_datagram(sk, flags, noblock, &err);
465	468	if (!skb)
466	469	goto out;
...	...	@@ -335,6 +335,9 @@
335	335	if (flags & MSG_ERRQUEUE)
336	336	return ipv6_recv_error(sk, msg, len);
337	337
	338	+ if (np->rxpmtu && np->rxopt.bits.rxpmtu)
	339	+ return ipv6_recv_rxpmtu(sk, msg, len);
	340	+
338	341	try_again:
339	342	skb = __skb_recv_datagram(sk, flags \| (noblock ? MSG_DONTWAIT : 0),
340	343	&peeked, &err);