vxlan: Add support for UDP checksums (v4 sending, v6 zero csums)

Added VXLAN link configuration for sending UDP checksums, and allowing TX and RX of UDP6 checksums. Also, call common iptunnel_handle_offloads and added GSO support for checksums.

Tom Herbert committed 12 years ago

Commit 359a0ea1 parent 4749c094 files · +74 −63Browse files

Unified · Split

4 files

drivers/net/vxlan.c +59 −61
include/net/vxlan.h +11 −1
include/uapi/linux/if_link.h +3 −0
net/openvswitch/vport-vxlan.c +1 −1

drivers/net/vxlan.c+59−61

@@ -135,7 +135,7 @@ struct vxlan_dev { · 134 unmodified lines
135	135	__u16 port_max;
136	136	__u8 tos; /* TOS override */
137	137	__u8 ttl;
138		− u32 flags; /* VXLAN_F_* below */
	138	+ u32 flags; /* VXLAN_F_* in vxlan.h */
139	139
140	140	struct work_struct sock_work;
141	141	struct work_struct igmp_join;
@@ -150,13 +150,6 @@ struct vxlan_dev { · 8 unmodified lines
150	150	struct hlist_head fdb_head[FDB_HASH_SIZE];
151	151	};
152	152
153		−#define VXLAN_F_LEARN 0x01
154		−#define VXLAN_F_PROXY 0x02
155		−#define VXLAN_F_RSC 0x04
156		−#define VXLAN_F_L2MISS 0x08
157		−#define VXLAN_F_L3MISS 0x10
158		−#define VXLAN_F_IPV6 0x20 /* internal flag */
159		−
160	153	/* salt for hash table */
161	154	static u32 vxlan_salt __read_mostly;
162	155	static struct workqueue_struct *vxlan_wq;
@@ -1601,18 +1594,11 @@ __be16 vxlan_src_port(__u16 port_min, __u16 port_max, struct sk_buff *skb) · 1438 unmodified lines
1601	1594	}
1602	1595	EXPORT_SYMBOL_GPL(vxlan_src_port);
1603	1596
1604		−static int handle_offloads(struct sk_buff *skb)
	1597	+static inline struct sk_buff vxlan_handle_offloads(struct sk_buff skb,
	1598	+ bool udp_csum)
1605	1599	{
1606		− if (skb_is_gso(skb)) {
1607		− int err = skb_unclone(skb, GFP_ATOMIC);
1608		− if (unlikely(err))
1609		− return err;
1610		−
1611		− skb_shinfo(skb)->gso_type \|= SKB_GSO_UDP_TUNNEL;
1612		− } else if (skb->ip_summed != CHECKSUM_PARTIAL)
1613		− skb->ip_summed = CHECKSUM_NONE;
1614		−
1615		− return 0;
	1600	+ int type = udp_csum ? SKB_GSO_UDP_TUNNEL_CSUM : SKB_GSO_UDP_TUNNEL;
	1601	+ return iptunnel_handle_offloads(skb, udp_csum, type);
1616	1602	}
1617	1603
1618	1604	#if IS_ENABLED(CONFIG_IPV6)
@@ -1629,10 +1615,9 @@ static int vxlan6_xmit_skb(struct vxlan_sock *vs, · 10 unmodified lines
1629	1615	int min_headroom;
1630	1616	int err;
1631	1617
1632		− if (!skb->encapsulation) {
1633		− skb_reset_inner_headers(skb);
1634		− skb->encapsulation = 1;
1635		− }
	1618	+ skb = vxlan_handle_offloads(skb, !udp_get_no_check6_tx(vs->sock->sk));
	1619	+ if (IS_ERR(skb))
	1620	+ return -EINVAL;
1636	1621
1637	1622	skb_scrub_packet(skb, xnet);
1638	1623
@@ -1666,27 +1651,14 @@ static int vxlan6_xmit_skb(struct vxlan_sock *vs, · 27 unmodified lines
1666	1651	uh->source = src_port;
1667	1652
1668	1653	uh->len = htons(skb->len);
1669		− uh->check = 0;
1670	1654
1671	1655	memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1672	1656	IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE \| IPSKB_XFRM_TRANSFORMED \|
1673	1657	IPSKB_REROUTED);
1674	1658	skb_dst_set(skb, dst);
1675	1659
1676		− if (!skb_is_gso(skb) && !(dst->dev->features & NETIF_F_IPV6_CSUM)) {
1677		− __wsum csum = skb_checksum(skb, 0, skb->len, 0);
1678		− skb->ip_summed = CHECKSUM_UNNECESSARY;
1679		− uh->check = csum_ipv6_magic(saddr, daddr, skb->len,
1680		− IPPROTO_UDP, csum);
1681		− if (uh->check == 0)
1682		− uh->check = CSUM_MANGLED_0;
1683		− } else {
1684		− skb->ip_summed = CHECKSUM_PARTIAL;
1685		− skb->csum_start = skb_transport_header(skb) - skb->head;
1686		− skb->csum_offset = offsetof(struct udphdr, check);
1687		− uh->check = ~csum_ipv6_magic(saddr, daddr,
1688		− skb->len, IPPROTO_UDP, 0);
1689		− }
	1660	+ udp6_set_csum(udp_get_no_check6_tx(vs->sock->sk), skb,
	1661	+ saddr, daddr, skb->len);
1690	1662
1691	1663	__skb_push(skb, sizeof(*ip6h));
1692	1664	skb_reset_network_header(skb);
@@ -1702,10 +1674,6 @@ static int vxlan6_xmit_skb(struct vxlan_sock *vs, · 9 unmodified lines
1702	1674	ip6h->daddr = *daddr;
1703	1675	ip6h->saddr = *saddr;
1704	1676
1705		− err = handle_offloads(skb);
1706		− if (err)
1707		− return err;
1708		−
1709	1677	ip6tunnel_xmit(skb, dev);
1710	1678	return 0;
1711	1679	}
@@ -1721,10 +1689,9 @@ int vxlan_xmit_skb(struct vxlan_sock *vs, · 9 unmodified lines
1721	1689	int min_headroom;
1722	1690	int err;
1723	1691
1724		− if (!skb->encapsulation) {
1725		− skb_reset_inner_headers(skb);
1726		− skb->encapsulation = 1;
1727		− }
	1692	+ skb = vxlan_handle_offloads(skb, !vs->sock->sk->sk_no_check_tx);
	1693	+ if (IS_ERR(skb))
	1694	+ return -EINVAL;
1728	1695
1729	1696	min_headroom = LL_RESERVED_SPACE(rt->dst.dev) + rt->dst.header_len
1730	1697	+ VXLAN_HLEN + sizeof(struct iphdr)
@@ -1756,11 +1723,9 @@ int vxlan_xmit_skb(struct vxlan_sock *vs, · 25 unmodified lines
1756	1723	uh->source = src_port;
1757	1724
1758	1725	uh->len = htons(skb->len);
1759		− uh->check = 0;
1760	1726
1761		− err = handle_offloads(skb);
1762		− if (err)
1763		− return err;
	1727	+ udp_set_csum(vs->sock->sk->sk_no_check_tx, skb,
	1728	+ src, dst, skb->len);
1764	1729
1765	1730	return iptunnel_xmit(vs->sock->sk, rt, skb, src, dst, IPPROTO_UDP,
1766	1731	tos, ttl, df, xnet);
@@ -2405,7 +2370,7 @@ static void vxlan_del_work(struct work_struct *work) · 638 unmodified lines
2405	2370	* could be used for both IPv4 and IPv6 communications, but
2406	2371	* users may set bindv6only=1.
2407	2372	*/
2408		−static struct socket create_v6_sock(struct net net, __be16 port)
	2373	+static struct socket create_v6_sock(struct net net, __be16 port, u32 flags)
2409	2374	{
2410	2375	struct sock *sk;
2411	2376	struct socket *sock;
@@ -2442,18 +2407,25 @@ static struct socket create_v6_sock(struct net net, __be16 port) · 30 unmodified lines
2442	2407
2443	2408	/* Disable multicast loopback */
2444	2409	inet_sk(sk)->mc_loop = 0;
	2410	+
	2411	+ if (flags & VXLAN_F_UDP_ZERO_CSUM6_TX)
	2412	+ udp_set_no_check6_tx(sk, true);
	2413	+
	2414	+ if (flags & VXLAN_F_UDP_ZERO_CSUM6_RX)
	2415	+ udp_set_no_check6_rx(sk, true);
	2416	+
2445	2417	return sock;
2446	2418	}
2447	2419
2448	2420	#else
2449	2421
2450		−static struct socket create_v6_sock(struct net net, __be16 port)
	2422	+static struct socket create_v6_sock(struct net net, __be16 port, u32 flags)
2451	2423	{
2452	2424	return ERR_PTR(-EPFNOSUPPORT);
2453	2425	}
2454	2426	#endif
2455	2427
2456		−static struct socket create_v4_sock(struct net net, __be16 port)
	2428	+static struct socket create_v4_sock(struct net net, __be16 port, u32 flags)
2457	2429	{
2458	2430	struct sock *sk;
2459	2431	struct socket *sock;
@@ -2486,18 +2458,24 @@ static struct socket create_v4_sock(struct net net, __be16 port) · 26 unmodified lines
2486	2458
2487	2459	/* Disable multicast loopback */
2488	2460	inet_sk(sk)->mc_loop = 0;
	2461	+
	2462	+ if (!(flags & VXLAN_F_UDP_CSUM))
	2463	+ sock->sk->sk_no_check_tx = 1;
	2464	+
2489	2465	return sock;
2490	2466	}
2491	2467
2492	2468	/* Create new listen socket if needed */
2493	2469	static struct vxlan_sock vxlan_socket_create(struct net net, __be16 port,
2494		− vxlan_rcv_t rcv, void data, bool ipv6)
	2470	+ vxlan_rcv_t rcv, void data,
	2471	+ u32 flags)
2495	2472	{
2496	2473	struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2497	2474	struct vxlan_sock *vs;
2498	2475	struct socket *sock;
2499	2476	struct sock *sk;
2500	2477	unsigned int h;
	2478	+ bool ipv6 = !!(flags & VXLAN_F_IPV6);
2501	2479
2502	2480	vs = kzalloc(sizeof(*vs), GFP_KERNEL);
2503	2481	if (!vs)
@@ -2509,9 +2487,9 @@ static struct vxlan_sock vxlan_socket_create(struct net net, __be16 port, · 5 unmodified lines
2509	2487	INIT_WORK(&vs->del_work, vxlan_del_work);
2510	2488
2511	2489	if (ipv6)
2512		− sock = create_v6_sock(net, port);
	2490	+ sock = create_v6_sock(net, port, flags);
2513	2491	else
2514		− sock = create_v4_sock(net, port);
	2492	+ sock = create_v4_sock(net, port, flags);
2515	2493	if (IS_ERR(sock)) {
2516	2494	kfree(vs);
2517	2495	return ERR_CAST(sock);
@@ -2549,12 +2527,12 @@ static struct vxlan_sock vxlan_socket_create(struct net net, __be16 port, · 31 unmodified lines
2549	2527
2550	2528	struct vxlan_sock vxlan_sock_add(struct net net, __be16 port,
2551	2529	vxlan_rcv_t rcv, void data,
2552		− bool no_share, bool ipv6)
	2530	+ bool no_share, u32 flags)
2553	2531	{
2554	2532	struct vxlan_net *vn = net_generic(net, vxlan_net_id);
2555	2533	struct vxlan_sock *vs;
2556	2534
2557		− vs = vxlan_socket_create(net, port, rcv, data, ipv6);
	2535	+ vs = vxlan_socket_create(net, port, rcv, data, flags);
2558	2536	if (!IS_ERR(vs))
2559	2537	return vs;
2560	2538
@@ -2587,7 +2565,7 @@ static void vxlan_sock_work(struct work_struct *work) · 26 unmodified lines
2587	2565	__be16 port = vxlan->dst_port;
2588	2566	struct vxlan_sock *nvs;
2589	2567
2590		− nvs = vxlan_sock_add(net, port, vxlan_rcv, NULL, false, vxlan->flags & VXLAN_F_IPV6);
	2568	+ nvs = vxlan_sock_add(net, port, vxlan_rcv, NULL, false, vxlan->flags);
2591	2569	spin_lock(&vn->sock_lock);
2592	2570	if (!IS_ERR(nvs))
2593	2571	vxlan_vs_add_dev(nvs, vxlan);
@@ -2711,6 +2689,17 @@ static int vxlan_newlink(struct net net, struct net_device dev, · 117 unmodified lines
2711	2689	if (data[IFLA_VXLAN_PORT])
2712	2690	vxlan->dst_port = nla_get_be16(data[IFLA_VXLAN_PORT]);
2713	2691
	2692	+ if (data[IFLA_VXLAN_UDP_CSUM] && nla_get_u8(data[IFLA_VXLAN_UDP_CSUM]))
	2693	+ vxlan->flags \|= VXLAN_F_UDP_CSUM;
	2694	+
	2695	+ if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX] &&
	2696	+ nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_TX]))
	2697	+ vxlan->flags \|= VXLAN_F_UDP_ZERO_CSUM6_TX;
	2698	+
	2699	+ if (data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX] &&
	2700	+ nla_get_u8(data[IFLA_VXLAN_UDP_ZERO_CSUM6_RX]))
	2701	+ vxlan->flags \|= VXLAN_F_UDP_ZERO_CSUM6_RX;
	2702	+
2714	2703	if (vxlan_find_vni(net, vni, vxlan->dst_port)) {
2715	2704	pr_info("duplicate VNI %u\n", vni);
2716	2705	return -EEXIST;
@@ -2774,7 +2763,10 @@ static size_t vxlan_get_size(const struct net_device *dev) · 57 unmodified lines
2774	2763	nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_AGEING */
2775	2764	nla_total_size(sizeof(__u32)) + /* IFLA_VXLAN_LIMIT */
2776	2765	nla_total_size(sizeof(struct ifla_vxlan_port_range)) +
2777		− nla_total_size(sizeof(__be16))+ /* IFLA_VXLAN_PORT */
	2766	+ nla_total_size(sizeof(__be16)) + /* IFLA_VXLAN_PORT */
	2767	+ nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_CSUM */
	2768	+ nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_TX */
	2769	+ nla_total_size(sizeof(__u8)) + /* IFLA_VXLAN_UDP_ZERO_CSUM6_RX */
2778	2770	0;
2779	2771	}
2780	2772
@@ -2834,7 +2826,13 @@ static int vxlan_fill_info(struct sk_buff skb, const struct net_device dev) · 53 unmodified lines
2834	2826	!!(vxlan->flags & VXLAN_F_L3MISS)) \|\|
2835	2827	nla_put_u32(skb, IFLA_VXLAN_AGEING, vxlan->age_interval) \|\|
2836	2828	nla_put_u32(skb, IFLA_VXLAN_LIMIT, vxlan->addrmax) \|\|
2837		− nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->dst_port))
	2829	+ nla_put_be16(skb, IFLA_VXLAN_PORT, vxlan->dst_port) \|\|
	2830	+ nla_put_u8(skb, IFLA_VXLAN_UDP_CSUM,
	2831	+ !!(vxlan->flags & VXLAN_F_UDP_CSUM)) \|\|
	2832	+ nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
	2833	+ !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_TX)) \|\|
	2834	+ nla_put_u8(skb, IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
	2835	+ !!(vxlan->flags & VXLAN_F_UDP_ZERO_CSUM6_RX)))
2838	2836	goto nla_put_failure;
2839	2837
2840	2838	if (nla_put(skb, IFLA_VXLAN_PORT_RANGE, sizeof(ports), &ports))

include/net/vxlan.h+11−1

@@ -24,9 +24,19 @@ struct vxlan_sock { · 23 unmodified lines
24	24	struct udp_offload udp_offloads;
25	25	};
26	26
	27	+#define VXLAN_F_LEARN 0x01
	28	+#define VXLAN_F_PROXY 0x02
	29	+#define VXLAN_F_RSC 0x04
	30	+#define VXLAN_F_L2MISS 0x08
	31	+#define VXLAN_F_L3MISS 0x10
	32	+#define VXLAN_F_IPV6 0x20
	33	+#define VXLAN_F_UDP_CSUM 0x40
	34	+#define VXLAN_F_UDP_ZERO_CSUM6_TX 0x80
	35	+#define VXLAN_F_UDP_ZERO_CSUM6_RX 0x100
	36	+
27	37	struct vxlan_sock vxlan_sock_add(struct net net, __be16 port,
28	38	vxlan_rcv_t rcv, void data,
29		− bool no_share, bool ipv6);
	39	+ bool no_share, u32 flags);
30	40
31	41	void vxlan_sock_release(struct vxlan_sock *vs);
32	42

include/uapi/linux/if_link.h+3−0

@@ -319,6 +319,9 @@ enum { · 318 unmodified lines
319	319	IFLA_VXLAN_PORT, /* destination port */
320	320	IFLA_VXLAN_GROUP6,
321	321	IFLA_VXLAN_LOCAL6,
	322	+ IFLA_VXLAN_UDP_CSUM,
	323	+ IFLA_VXLAN_UDP_ZERO_CSUM6_TX,
	324	+ IFLA_VXLAN_UDP_ZERO_CSUM6_RX,
322	325	__IFLA_VXLAN_MAX
323	326	};
324	327	#define IFLA_VXLAN_MAX (__IFLA_VXLAN_MAX - 1)

net/openvswitch/vport-vxlan.c+1−1

@@ -122,7 +122,7 @@ static struct vport vxlan_tnl_create(const struct vport_parms parms) · 121 unmodified lines
122	122	vxlan_port = vxlan_vport(vport);
123	123	strncpy(vxlan_port->name, parms->name, IFNAMSIZ);
124	124
125		− vs = vxlan_sock_add(net, htons(dst_port), vxlan_rcv, vport, true, false);
	125	+ vs = vxlan_sock_add(net, htons(dst_port), vxlan_rcv, vport, true, 0);
126	126	if (IS_ERR(vs)) {
127	127	ovs_vport_free(vport);
128	128	return (void *)vs;