2 * IPV4 GSO/GRO offload support
3 * Linux INET implementation
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License
7 * as published by the Free Software Foundation; either version
8 * 2 of the License, or (at your option) any later version.
13 #include <linux/skbuff.h>
15 #include <net/protocol.h>
17 static struct sk_buff *__skb_udp_tunnel_segment(struct sk_buff *skb,
18 netdev_features_t features,
19 struct sk_buff *(*gso_inner_segment)(struct sk_buff *skb,
20 netdev_features_t features),
21 __be16 new_protocol, bool is_ipv6)
23 int tnl_hlen = skb_inner_mac_header(skb) - skb_transport_header(skb);
24 bool remcsum, need_csum, offload_csum, gso_partial;
25 struct sk_buff *segs = ERR_PTR(-EINVAL);
26 struct udphdr *uh = udp_hdr(skb);
27 u16 mac_offset = skb->mac_header;
28 __be16 protocol = skb->protocol;
29 u16 mac_len = skb->mac_len;
30 int udp_offset, outer_hlen;
34 if (unlikely(!pskb_may_pull(skb, tnl_hlen)))
37 /* Adjust partial header checksum to negate old length.
38 * We cannot rely on the value contained in uh->len as it is
39 * possible that the actual value exceeds the boundaries of the
40 * 16 bit length field due to the header being added outside of an
41 * IP or IPv6 frame that was already limited to 64K - 1.
43 if (skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)
44 partial = (__force __wsum)uh->len;
46 partial = (__force __wsum)htonl(skb->len);
47 partial = csum_sub(csum_unfold(uh->check), partial);
49 /* setup inner skb. */
50 skb->encapsulation = 0;
51 SKB_GSO_CB(skb)->encap_level = 0;
52 __skb_pull(skb, tnl_hlen);
53 skb_reset_mac_header(skb);
54 skb_set_network_header(skb, skb_inner_network_offset(skb));
55 skb->mac_len = skb_inner_network_offset(skb);
56 skb->protocol = new_protocol;
58 need_csum = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_TUNNEL_CSUM);
59 skb->encap_hdr_csum = need_csum;
61 remcsum = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TUNNEL_REMCSUM);
62 skb->remcsum_offload = remcsum;
64 need_ipsec = skb_dst(skb) && dst_xfrm(skb_dst(skb));
65 /* Try to offload checksum if possible */
66 offload_csum = !!(need_csum &&
69 (is_ipv6 ? (NETIF_F_HW_CSUM | NETIF_F_IPV6_CSUM) :
70 (NETIF_F_HW_CSUM | NETIF_F_IP_CSUM))));
72 features &= skb->dev->hw_enc_features;
74 /* The only checksum offload we care about from here on out is the
75 * outer one so strip the existing checksum feature flags and
76 * instead set the flag based on our outer checksum offload value.
79 features &= ~NETIF_F_CSUM_MASK;
80 if (!need_csum || offload_csum)
81 features |= NETIF_F_HW_CSUM;
84 /* segment inner packet. */
85 segs = gso_inner_segment(skb, features);
86 if (IS_ERR_OR_NULL(segs)) {
87 skb_gso_error_unwind(skb, protocol, tnl_hlen, mac_offset,
92 gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
94 outer_hlen = skb_tnl_header_len(skb);
95 udp_offset = outer_hlen - tnl_hlen;
101 skb->ip_summed = CHECKSUM_NONE;
103 /* Set up inner headers if we are offloading inner checksum */
104 if (skb->ip_summed == CHECKSUM_PARTIAL) {
105 skb_reset_inner_headers(skb);
106 skb->encapsulation = 1;
109 skb->mac_len = mac_len;
110 skb->protocol = protocol;
112 __skb_push(skb, outer_hlen);
113 skb_reset_mac_header(skb);
114 skb_set_network_header(skb, mac_len);
115 skb_set_transport_header(skb, udp_offset);
116 len = skb->len - udp_offset;
119 /* If we are only performing partial GSO the inner header
120 * will be using a length value equal to only one MSS sized
121 * segment instead of the entire frame.
123 if (gso_partial && skb_is_gso(skb)) {
124 uh->len = htons(skb_shinfo(skb)->gso_size +
125 SKB_GSO_CB(skb)->data_offset +
126 skb->head - (unsigned char *)uh);
128 uh->len = htons(len);
134 uh->check = ~csum_fold(csum_add(partial,
135 (__force __wsum)htonl(len)));
137 if (skb->encapsulation || !offload_csum) {
138 uh->check = gso_make_checksum(skb, ~uh->check);
140 uh->check = CSUM_MANGLED_0;
142 skb->ip_summed = CHECKSUM_PARTIAL;
143 skb->csum_start = skb_transport_header(skb) - skb->head;
144 skb->csum_offset = offsetof(struct udphdr, check);
146 } while ((skb = skb->next));
151 struct sk_buff *skb_udp_tunnel_segment(struct sk_buff *skb,
152 netdev_features_t features,
155 __be16 protocol = skb->protocol;
156 const struct net_offload **offloads;
157 const struct net_offload *ops;
158 struct sk_buff *segs = ERR_PTR(-EINVAL);
159 struct sk_buff *(*gso_inner_segment)(struct sk_buff *skb,
160 netdev_features_t features);
164 switch (skb->inner_protocol_type) {
165 case ENCAP_TYPE_ETHER:
166 protocol = skb->inner_protocol;
167 gso_inner_segment = skb_mac_gso_segment;
169 case ENCAP_TYPE_IPPROTO:
170 offloads = is_ipv6 ? inet6_offloads : inet_offloads;
171 ops = rcu_dereference(offloads[skb->inner_ipproto]);
172 if (!ops || !ops->callbacks.gso_segment)
174 gso_inner_segment = ops->callbacks.gso_segment;
180 segs = __skb_udp_tunnel_segment(skb, features, gso_inner_segment,
188 EXPORT_SYMBOL(skb_udp_tunnel_segment);
190 struct sk_buff *__udp_gso_segment(struct sk_buff *gso_skb,
191 netdev_features_t features)
193 struct sock *sk = gso_skb->sk;
194 unsigned int sum_truesize = 0;
195 struct sk_buff *segs, *seg;
202 mss = skb_shinfo(gso_skb)->gso_size;
203 if (gso_skb->len <= sizeof(*uh) + mss)
204 return ERR_PTR(-EINVAL);
206 hdrlen = gso_skb->data - skb_mac_header(gso_skb);
207 skb_pull(gso_skb, sizeof(*uh));
209 /* clear destructor to avoid skb_segment assigning it to tail */
210 WARN_ON_ONCE(gso_skb->destructor != sock_wfree);
211 gso_skb->destructor = NULL;
213 segs = skb_segment(gso_skb, features);
214 if (unlikely(IS_ERR_OR_NULL(segs))) {
215 gso_skb->destructor = sock_wfree;
221 /* compute checksum adjustment based on old length versus new */
222 newlen = htons(sizeof(*uh) + mss);
223 check = csum16_add(csum16_sub(uh->check, uh->len), newlen);
225 for (seg = segs; seg; seg = seg->next) {
228 /* last packet can be partial gso_size */
230 newlen = htons(seg->len - hdrlen);
231 check = csum16_add(csum16_sub(uh->check, uh->len),
238 seg->destructor = sock_wfree;
240 sum_truesize += seg->truesize;
243 refcount_add(sum_truesize - gso_skb->truesize, &sk->sk_wmem_alloc);
247 EXPORT_SYMBOL_GPL(__udp_gso_segment);
249 static struct sk_buff *__udp4_gso_segment(struct sk_buff *gso_skb,
250 netdev_features_t features)
252 if (!can_checksum_protocol(features, htons(ETH_P_IP)))
253 return ERR_PTR(-EIO);
255 return __udp_gso_segment(gso_skb, features);
258 static struct sk_buff *udp4_ufo_fragment(struct sk_buff *skb,
259 netdev_features_t features)
261 struct sk_buff *segs = ERR_PTR(-EINVAL);
267 if (skb->encapsulation &&
268 (skb_shinfo(skb)->gso_type &
269 (SKB_GSO_UDP_TUNNEL|SKB_GSO_UDP_TUNNEL_CSUM))) {
270 segs = skb_udp_tunnel_segment(skb, features, false);
274 if (!(skb_shinfo(skb)->gso_type & (SKB_GSO_UDP | SKB_GSO_UDP_L4)))
277 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
280 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4)
281 return __udp4_gso_segment(skb, features);
283 mss = skb_shinfo(skb)->gso_size;
284 if (unlikely(skb->len <= mss))
287 /* Do software UFO. Complete and fill in the UDP checksum as
288 * HW cannot do checksum of UDP packets sent as multiple
296 csum = skb_checksum(skb, 0, skb->len, 0);
297 uh->check = udp_v4_check(skb->len, iph->saddr, iph->daddr, csum);
299 uh->check = CSUM_MANGLED_0;
301 skb->ip_summed = CHECKSUM_UNNECESSARY;
303 /* If there is no outer header we can fake a checksum offload
304 * due to the fact that we have already done the checksum in
305 * software prior to segmenting the frame.
307 if (!skb->encap_hdr_csum)
308 features |= NETIF_F_HW_CSUM;
310 /* Fragment the skb. IP headers of the fragments are updated in
313 segs = skb_segment(skb, features);
318 struct sk_buff **udp_gro_receive(struct sk_buff **head, struct sk_buff *skb,
319 struct udphdr *uh, udp_lookup_t lookup)
321 struct sk_buff *p, **pp = NULL;
323 unsigned int off = skb_gro_offset(skb);
327 if (NAPI_GRO_CB(skb)->encap_mark ||
328 (skb->ip_summed != CHECKSUM_PARTIAL &&
329 NAPI_GRO_CB(skb)->csum_cnt == 0 &&
330 !NAPI_GRO_CB(skb)->csum_valid))
333 /* mark that this skb passed once through the tunnel gro layer */
334 NAPI_GRO_CB(skb)->encap_mark = 1;
337 sk = (*lookup)(skb, uh->source, uh->dest);
339 if (sk && udp_sk(sk)->gro_receive)
346 for (p = *head; p; p = p->next) {
347 if (!NAPI_GRO_CB(p)->same_flow)
350 uh2 = (struct udphdr *)(p->data + off);
352 /* Match ports and either checksums are either both zero
355 if ((*(u32 *)&uh->source != *(u32 *)&uh2->source) ||
356 (!uh->check ^ !uh2->check)) {
357 NAPI_GRO_CB(p)->same_flow = 0;
362 skb_gro_pull(skb, sizeof(struct udphdr)); /* pull encapsulating udp header */
363 skb_gro_postpull_rcsum(skb, uh, sizeof(struct udphdr));
364 pp = call_gro_receive_sk(udp_sk(sk)->gro_receive, sk, head, skb);
369 NAPI_GRO_CB(skb)->flush |= flush;
372 EXPORT_SYMBOL(udp_gro_receive);
374 static struct sk_buff **udp4_gro_receive(struct sk_buff **head,
377 struct udphdr *uh = udp_gro_udphdr(skb);
382 /* Don't bother verifying checksum if we're going to flush anyway. */
383 if (NAPI_GRO_CB(skb)->flush)
386 if (skb_gro_checksum_validate_zero_check(skb, IPPROTO_UDP, uh->check,
387 inet_gro_compute_pseudo))
390 skb_gro_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
391 inet_gro_compute_pseudo);
393 NAPI_GRO_CB(skb)->is_ipv6 = 0;
394 return udp_gro_receive(head, skb, uh, udp4_lib_lookup_skb);
397 NAPI_GRO_CB(skb)->flush = 1;
401 int udp_gro_complete(struct sk_buff *skb, int nhoff,
404 __be16 newlen = htons(skb->len - nhoff);
405 struct udphdr *uh = (struct udphdr *)(skb->data + nhoff);
411 /* Set encapsulation before calling into inner gro_complete() functions
412 * to make them set up the inner offsets.
414 skb->encapsulation = 1;
417 sk = (*lookup)(skb, uh->source, uh->dest);
418 if (sk && udp_sk(sk)->gro_complete)
419 err = udp_sk(sk)->gro_complete(sk, skb,
420 nhoff + sizeof(struct udphdr));
423 if (skb->remcsum_offload)
424 skb_shinfo(skb)->gso_type |= SKB_GSO_TUNNEL_REMCSUM;
428 EXPORT_SYMBOL(udp_gro_complete);
430 static int udp4_gro_complete(struct sk_buff *skb, int nhoff)
432 const struct iphdr *iph = ip_hdr(skb);
433 struct udphdr *uh = (struct udphdr *)(skb->data + nhoff);
436 skb_shinfo(skb)->gso_type |= SKB_GSO_UDP_TUNNEL_CSUM;
437 uh->check = ~udp_v4_check(skb->len - nhoff, iph->saddr,
440 skb_shinfo(skb)->gso_type |= SKB_GSO_UDP_TUNNEL;
443 return udp_gro_complete(skb, nhoff, udp4_lib_lookup_skb);
446 static const struct net_offload udpv4_offload = {
448 .gso_segment = udp4_ufo_fragment,
449 .gro_receive = udp4_gro_receive,
450 .gro_complete = udp4_gro_complete,
454 int __init udpv4_offload_init(void)
456 return inet_add_offload(&udpv4_offload, IPPROTO_UDP);