blob: 097060def7f0c9ce655250d7c25bd809e93e392f [file] [log] [blame]
Daniel Borkmannda5bab02013-06-08 12:56:03 +02001/*
2 * IPV4 GSO/GRO offload support
3 * Linux INET implementation
4 *
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.
9 *
10 * UDPv4 GSO support
11 */
12
13#include <linux/skbuff.h>
14#include <net/udp.h>
15#include <net/protocol.h>
16
Tom Herbert8bce6d72014-09-29 20:22:29 -070017static 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),
Tom Herbert4bcb8772014-11-04 09:06:52 -080021 __be16 new_protocol, bool is_ipv6)
Tom Herbert155e0102014-07-13 19:49:56 -070022{
Alexander Duyckdbef4912016-02-05 15:28:20 -080023 int tnl_hlen = skb_inner_mac_header(skb) - skb_transport_header(skb);
Alexander Duyck22463872016-02-24 16:46:21 -080024 bool remcsum, need_csum, offload_csum, ufo;
Tom Herbert155e0102014-07-13 19:49:56 -070025 struct sk_buff *segs = ERR_PTR(-EINVAL);
Alexander Duyckdbef4912016-02-05 15:28:20 -080026 struct udphdr *uh = udp_hdr(skb);
Tom Herbert155e0102014-07-13 19:49:56 -070027 u16 mac_offset = skb->mac_header;
Tom Herbert155e0102014-07-13 19:49:56 -070028 __be16 protocol = skb->protocol;
Alexander Duyckdbef4912016-02-05 15:28:20 -080029 u16 mac_len = skb->mac_len;
Tom Herbert155e0102014-07-13 19:49:56 -070030 int udp_offset, outer_hlen;
Alexander Duyck08334822016-03-11 14:05:47 -080031 __wsum partial;
Tom Herbert155e0102014-07-13 19:49:56 -070032
33 if (unlikely(!pskb_may_pull(skb, tnl_hlen)))
34 goto out;
35
Alexander Duyck08334822016-03-11 14:05:47 -080036 /* Adjust partial header checksum to negate old length.
37 * We cannot rely on the value contained in uh->len as it is
38 * possible that the actual value exceeds the boundaries of the
39 * 16 bit length field due to the header being added outside of an
40 * IP or IPv6 frame that was already limited to 64K - 1.
41 */
Alexander Duyck802ab552016-04-10 21:45:03 -040042 if (skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)
43 partial = (__force __wsum)uh->len;
44 else
45 partial = (__force __wsum)htonl(skb->len);
46 partial = csum_sub(csum_unfold(uh->check), partial);
Alexander Duyckdbef4912016-02-05 15:28:20 -080047
48 /* setup inner skb. */
Tom Herbert155e0102014-07-13 19:49:56 -070049 skb->encapsulation = 0;
Alexander Duyck5197f342016-03-22 16:18:07 -070050 SKB_GSO_CB(skb)->encap_level = 0;
Tom Herbert155e0102014-07-13 19:49:56 -070051 __skb_pull(skb, tnl_hlen);
52 skb_reset_mac_header(skb);
53 skb_set_network_header(skb, skb_inner_network_offset(skb));
54 skb->mac_len = skb_inner_network_offset(skb);
Tom Herbert8bce6d72014-09-29 20:22:29 -070055 skb->protocol = new_protocol;
Alexander Duyckfdaefd62016-02-05 15:28:14 -080056
57 need_csum = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_TUNNEL_CSUM);
Tom Herbert4bcb8772014-11-04 09:06:52 -080058 skb->encap_hdr_csum = need_csum;
Alexander Duyckfdaefd62016-02-05 15:28:14 -080059
60 remcsum = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TUNNEL_REMCSUM);
Tom Herberte585f232014-11-04 09:06:54 -080061 skb->remcsum_offload = remcsum;
Tom Herbert155e0102014-07-13 19:49:56 -070062
Alexander Duyck22463872016-02-24 16:46:21 -080063 ufo = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
64
Tom Herbert4bcb8772014-11-04 09:06:52 -080065 /* Try to offload checksum if possible */
66 offload_csum = !!(need_csum &&
Alexander Duyckfdaefd62016-02-05 15:28:14 -080067 (skb->dev->features &
68 (is_ipv6 ? (NETIF_F_HW_CSUM | NETIF_F_IPV6_CSUM) :
69 (NETIF_F_HW_CSUM | NETIF_F_IP_CSUM))));
Tom Herbert155e0102014-07-13 19:49:56 -070070
Alexander Duyckbef3c6c2016-02-05 15:27:31 -080071 features &= skb->dev->hw_enc_features;
72
Alexander Duyck7fbeffe2016-02-05 15:27:43 -080073 /* The only checksum offload we care about from here on out is the
74 * outer one so strip the existing checksum feature flags and
75 * instead set the flag based on our outer checksum offload value.
76 */
Alexander Duyck22463872016-02-24 16:46:21 -080077 if (remcsum || ufo) {
Alexander Duyck7fbeffe2016-02-05 15:27:43 -080078 features &= ~NETIF_F_CSUM_MASK;
Alexander Duyck22463872016-02-24 16:46:21 -080079 if (!need_csum || offload_csum)
Alexander Duyck7fbeffe2016-02-05 15:27:43 -080080 features |= NETIF_F_HW_CSUM;
81 }
82
Tom Herbert155e0102014-07-13 19:49:56 -070083 /* segment inner packet. */
Alexander Duyckbef3c6c2016-02-05 15:27:31 -080084 segs = gso_inner_segment(skb, features);
Himangi Saraogi27446442014-07-27 12:38:38 +053085 if (IS_ERR_OR_NULL(segs)) {
Tom Herbert155e0102014-07-13 19:49:56 -070086 skb_gso_error_unwind(skb, protocol, tnl_hlen, mac_offset,
87 mac_len);
88 goto out;
89 }
90
91 outer_hlen = skb_tnl_header_len(skb);
92 udp_offset = outer_hlen - tnl_hlen;
93 skb = segs;
94 do {
Alexander Duyck802ab552016-04-10 21:45:03 -040095 unsigned int len;
Tom Herbert155e0102014-07-13 19:49:56 -070096
Alexander Duyckfdaefd62016-02-05 15:28:14 -080097 if (remcsum)
Tom Herbert4bcb8772014-11-04 09:06:52 -080098 skb->ip_summed = CHECKSUM_NONE;
Alexander Duyckfdaefd62016-02-05 15:28:14 -080099
100 /* Set up inner headers if we are offloading inner checksum */
101 if (skb->ip_summed == CHECKSUM_PARTIAL) {
Tom Herbert4bcb8772014-11-04 09:06:52 -0800102 skb_reset_inner_headers(skb);
103 skb->encapsulation = 1;
104 }
Tom Herbert155e0102014-07-13 19:49:56 -0700105
106 skb->mac_len = mac_len;
Tom Herbert4bcb8772014-11-04 09:06:52 -0800107 skb->protocol = protocol;
Tom Herbert155e0102014-07-13 19:49:56 -0700108
Alexander Duyckdbef4912016-02-05 15:28:20 -0800109 __skb_push(skb, outer_hlen);
Tom Herbert155e0102014-07-13 19:49:56 -0700110 skb_reset_mac_header(skb);
111 skb_set_network_header(skb, mac_len);
112 skb_set_transport_header(skb, udp_offset);
Alexander Duyck802ab552016-04-10 21:45:03 -0400113 len = skb->len - udp_offset;
Tom Herbert155e0102014-07-13 19:49:56 -0700114 uh = udp_hdr(skb);
Alexander Duyck802ab552016-04-10 21:45:03 -0400115
116 /* If we are only performing partial GSO the inner header
117 * will be using a length value equal to only one MSS sized
118 * segment instead of the entire frame.
119 */
120 if (skb_is_gso(skb)) {
121 uh->len = htons(skb_shinfo(skb)->gso_size +
122 SKB_GSO_CB(skb)->data_offset +
123 skb->head - (unsigned char *)uh);
124 } else {
125 uh->len = htons(len);
126 }
Tom Herbert155e0102014-07-13 19:49:56 -0700127
Tom Herbert4bcb8772014-11-04 09:06:52 -0800128 if (!need_csum)
129 continue;
Tom Herbert155e0102014-07-13 19:49:56 -0700130
Alexander Duyck802ab552016-04-10 21:45:03 -0400131 uh->check = ~csum_fold(csum_add(partial,
132 (__force __wsum)htonl(len)));
Alexander Duyckfdaefd62016-02-05 15:28:14 -0800133
134 if (skb->encapsulation || !offload_csum) {
135 uh->check = gso_make_checksum(skb, ~uh->check);
136 if (uh->check == 0)
137 uh->check = CSUM_MANGLED_0;
138 } else {
Tom Herbert4bcb8772014-11-04 09:06:52 -0800139 skb->ip_summed = CHECKSUM_PARTIAL;
140 skb->csum_start = skb_transport_header(skb) - skb->head;
141 skb->csum_offset = offsetof(struct udphdr, check);
Tom Herbert155e0102014-07-13 19:49:56 -0700142 }
Tom Herbert155e0102014-07-13 19:49:56 -0700143 } while ((skb = skb->next));
144out:
145 return segs;
146}
147
Tom Herbert8bce6d72014-09-29 20:22:29 -0700148struct sk_buff *skb_udp_tunnel_segment(struct sk_buff *skb,
149 netdev_features_t features,
150 bool is_ipv6)
151{
152 __be16 protocol = skb->protocol;
153 const struct net_offload **offloads;
154 const struct net_offload *ops;
155 struct sk_buff *segs = ERR_PTR(-EINVAL);
156 struct sk_buff *(*gso_inner_segment)(struct sk_buff *skb,
157 netdev_features_t features);
158
159 rcu_read_lock();
160
161 switch (skb->inner_protocol_type) {
162 case ENCAP_TYPE_ETHER:
163 protocol = skb->inner_protocol;
164 gso_inner_segment = skb_mac_gso_segment;
165 break;
166 case ENCAP_TYPE_IPPROTO:
167 offloads = is_ipv6 ? inet6_offloads : inet_offloads;
168 ops = rcu_dereference(offloads[skb->inner_ipproto]);
169 if (!ops || !ops->callbacks.gso_segment)
170 goto out_unlock;
171 gso_inner_segment = ops->callbacks.gso_segment;
172 break;
173 default:
174 goto out_unlock;
175 }
176
177 segs = __skb_udp_tunnel_segment(skb, features, gso_inner_segment,
Tom Herbert4bcb8772014-11-04 09:06:52 -0800178 protocol, is_ipv6);
Tom Herbert8bce6d72014-09-29 20:22:29 -0700179
180out_unlock:
181 rcu_read_unlock();
182
183 return segs;
184}
Tom Herberta6024562016-04-05 08:22:51 -0700185EXPORT_SYMBOL(skb_udp_tunnel_segment);
Tom Herbert8bce6d72014-09-29 20:22:29 -0700186
Daniel Borkmannda5bab02013-06-08 12:56:03 +0200187static struct sk_buff *udp4_ufo_fragment(struct sk_buff *skb,
188 netdev_features_t features)
189{
190 struct sk_buff *segs = ERR_PTR(-EINVAL);
191 unsigned int mss;
Wei-Chun Chao7a7ffba2013-12-26 13:10:22 -0800192 __wsum csum;
Tom Herbertf71470b2014-09-20 14:52:29 -0700193 struct udphdr *uh;
194 struct iphdr *iph;
Wei-Chun Chao7a7ffba2013-12-26 13:10:22 -0800195
196 if (skb->encapsulation &&
Tom Herbert0f4f4ff2014-06-04 17:20:16 -0700197 (skb_shinfo(skb)->gso_type &
198 (SKB_GSO_UDP_TUNNEL|SKB_GSO_UDP_TUNNEL_CSUM))) {
Tom Herbert8bce6d72014-09-29 20:22:29 -0700199 segs = skb_udp_tunnel_segment(skb, features, false);
Wei-Chun Chao7a7ffba2013-12-26 13:10:22 -0800200 goto out;
201 }
Daniel Borkmannda5bab02013-06-08 12:56:03 +0200202
Tom Herbertf71470b2014-09-20 14:52:29 -0700203 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
204 goto out;
205
Daniel Borkmannda5bab02013-06-08 12:56:03 +0200206 mss = skb_shinfo(skb)->gso_size;
207 if (unlikely(skb->len <= mss))
208 goto out;
209
210 if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) {
211 /* Packet is from an untrusted source, reset gso_segs. */
212 int type = skb_shinfo(skb)->gso_type;
213
214 if (unlikely(type & ~(SKB_GSO_UDP | SKB_GSO_DODGY |
215 SKB_GSO_UDP_TUNNEL |
Tom Herbert0f4f4ff2014-06-04 17:20:16 -0700216 SKB_GSO_UDP_TUNNEL_CSUM |
Tom Herberte585f232014-11-04 09:06:54 -0800217 SKB_GSO_TUNNEL_REMCSUM |
Eric Dumazetcb32f512013-10-19 11:42:57 -0700218 SKB_GSO_IPIP |
Pravin B Shelar59b93b42014-11-05 15:27:48 -0800219 SKB_GSO_GRE | SKB_GSO_GRE_CSUM) ||
Daniel Borkmannda5bab02013-06-08 12:56:03 +0200220 !(type & (SKB_GSO_UDP))))
221 goto out;
222
223 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss);
224
225 segs = NULL;
226 goto out;
227 }
228
Wei-Chun Chao7a7ffba2013-12-26 13:10:22 -0800229 /* Do software UFO. Complete and fill in the UDP checksum as
230 * HW cannot do checksum of UDP packets sent as multiple
231 * IP fragments.
232 */
Tom Herbertf71470b2014-09-20 14:52:29 -0700233
234 uh = udp_hdr(skb);
235 iph = ip_hdr(skb);
236
237 uh->check = 0;
238 csum = skb_checksum(skb, 0, skb->len, 0);
239 uh->check = udp_v4_check(skb->len, iph->saddr, iph->daddr, csum);
240 if (uh->check == 0)
241 uh->check = CSUM_MANGLED_0;
242
Wei-Chun Chao7a7ffba2013-12-26 13:10:22 -0800243 skb->ip_summed = CHECKSUM_NONE;
244
Alexander Duyck22463872016-02-24 16:46:21 -0800245 /* If there is no outer header we can fake a checksum offload
246 * due to the fact that we have already done the checksum in
247 * software prior to segmenting the frame.
248 */
249 if (!skb->encap_hdr_csum)
250 features |= NETIF_F_HW_CSUM;
251
Daniel Borkmannda5bab02013-06-08 12:56:03 +0200252 /* Fragment the skb. IP headers of the fragments are updated in
253 * inet_gso_segment()
254 */
Wei-Chun Chao7a7ffba2013-12-26 13:10:22 -0800255 segs = skb_segment(skb, features);
Daniel Borkmannda5bab02013-06-08 12:56:03 +0200256out:
257 return segs;
258}
259
Tom Herbert57c67ff2014-08-22 13:34:44 -0700260struct sk_buff **udp_gro_receive(struct sk_buff **head, struct sk_buff *skb,
Tom Herberta6024562016-04-05 08:22:51 -0700261 struct udphdr *uh, udp_lookup_t lookup)
Or Gerlitzb582ef02014-01-20 13:59:19 +0200262{
Or Gerlitzb582ef02014-01-20 13:59:19 +0200263 struct sk_buff *p, **pp = NULL;
Tom Herbert57c67ff2014-08-22 13:34:44 -0700264 struct udphdr *uh2;
265 unsigned int off = skb_gro_offset(skb);
Or Gerlitzb582ef02014-01-20 13:59:19 +0200266 int flush = 1;
Tom Herberta6024562016-04-05 08:22:51 -0700267 struct sock *sk;
Or Gerlitzb582ef02014-01-20 13:59:19 +0200268
Jesse Grossfac8e0f2016-03-19 09:32:01 -0700269 if (NAPI_GRO_CB(skb)->encap_mark ||
Tom Herbert662880f2014-08-27 21:26:56 -0700270 (skb->ip_summed != CHECKSUM_PARTIAL &&
271 NAPI_GRO_CB(skb)->csum_cnt == 0 &&
272 !NAPI_GRO_CB(skb)->csum_valid))
Or Gerlitzb582ef02014-01-20 13:59:19 +0200273 goto out;
274
Jesse Grossfac8e0f2016-03-19 09:32:01 -0700275 /* mark that this skb passed once through the tunnel gro layer */
276 NAPI_GRO_CB(skb)->encap_mark = 1;
Or Gerlitzb582ef02014-01-20 13:59:19 +0200277
278 rcu_read_lock();
Tom Herberta6024562016-04-05 08:22:51 -0700279 sk = (*lookup)(skb, uh->source, uh->dest);
280
281 if (sk && udp_sk(sk)->gro_receive)
282 goto unflush;
Or Gerlitzb582ef02014-01-20 13:59:19 +0200283 goto out_unlock;
284
285unflush:
286 flush = 0;
287
288 for (p = *head; p; p = p->next) {
289 if (!NAPI_GRO_CB(p)->same_flow)
290 continue;
291
292 uh2 = (struct udphdr *)(p->data + off);
Tom Herbert57c67ff2014-08-22 13:34:44 -0700293
294 /* Match ports and either checksums are either both zero
295 * or nonzero.
296 */
297 if ((*(u32 *)&uh->source != *(u32 *)&uh2->source) ||
298 (!uh->check ^ !uh2->check)) {
Or Gerlitzb582ef02014-01-20 13:59:19 +0200299 NAPI_GRO_CB(p)->same_flow = 0;
300 continue;
301 }
302 }
303
304 skb_gro_pull(skb, sizeof(struct udphdr)); /* pull encapsulating udp header */
Tom Herbert6bae1d42014-06-10 18:54:26 -0700305 skb_gro_postpull_rcsum(skb, uh, sizeof(struct udphdr));
Tom Herberta6024562016-04-05 08:22:51 -0700306 pp = udp_sk(sk)->gro_receive(sk, head, skb);
Or Gerlitzb582ef02014-01-20 13:59:19 +0200307
308out_unlock:
309 rcu_read_unlock();
310out:
311 NAPI_GRO_CB(skb)->flush |= flush;
312 return pp;
313}
Tom Herberta6024562016-04-05 08:22:51 -0700314EXPORT_SYMBOL(udp_gro_receive);
Or Gerlitzb582ef02014-01-20 13:59:19 +0200315
Tom Herbert57c67ff2014-08-22 13:34:44 -0700316static struct sk_buff **udp4_gro_receive(struct sk_buff **head,
317 struct sk_buff *skb)
318{
319 struct udphdr *uh = udp_gro_udphdr(skb);
320
Tom Herbert2abb7cd2014-08-31 15:12:43 -0700321 if (unlikely(!uh))
322 goto flush;
Tom Herbert57c67ff2014-08-22 13:34:44 -0700323
Tom Herbert2abb7cd2014-08-31 15:12:43 -0700324 /* Don't bother verifying checksum if we're going to flush anyway. */
Scott Wood2d8f7e22014-09-10 21:23:18 -0500325 if (NAPI_GRO_CB(skb)->flush)
Tom Herbert2abb7cd2014-08-31 15:12:43 -0700326 goto skip;
327
328 if (skb_gro_checksum_validate_zero_check(skb, IPPROTO_UDP, uh->check,
329 inet_gro_compute_pseudo))
330 goto flush;
331 else if (uh->check)
332 skb_gro_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
333 inet_gro_compute_pseudo);
334skip:
Tom Herbertefc98d02014-10-03 15:48:08 -0700335 NAPI_GRO_CB(skb)->is_ipv6 = 0;
Tom Herberta6024562016-04-05 08:22:51 -0700336 return udp_gro_receive(head, skb, uh, udp4_lib_lookup_skb);
Tom Herbert2abb7cd2014-08-31 15:12:43 -0700337
338flush:
339 NAPI_GRO_CB(skb)->flush = 1;
340 return NULL;
Tom Herbert57c67ff2014-08-22 13:34:44 -0700341}
342
Tom Herberta6024562016-04-05 08:22:51 -0700343int udp_gro_complete(struct sk_buff *skb, int nhoff,
344 udp_lookup_t lookup)
Or Gerlitzb582ef02014-01-20 13:59:19 +0200345{
Or Gerlitzb582ef02014-01-20 13:59:19 +0200346 __be16 newlen = htons(skb->len - nhoff);
347 struct udphdr *uh = (struct udphdr *)(skb->data + nhoff);
348 int err = -ENOSYS;
Tom Herberta6024562016-04-05 08:22:51 -0700349 struct sock *sk;
Or Gerlitzb582ef02014-01-20 13:59:19 +0200350
351 uh->len = newlen;
352
353 rcu_read_lock();
Tom Herberta6024562016-04-05 08:22:51 -0700354 sk = (*lookup)(skb, uh->source, uh->dest);
355 if (sk && udp_sk(sk)->gro_complete)
356 err = udp_sk(sk)->gro_complete(sk, skb,
357 nhoff + sizeof(struct udphdr));
Or Gerlitzb582ef02014-01-20 13:59:19 +0200358 rcu_read_unlock();
Tom Herbert6db93ea2015-02-10 16:30:29 -0800359
360 if (skb->remcsum_offload)
361 skb_shinfo(skb)->gso_type |= SKB_GSO_TUNNEL_REMCSUM;
362
363 skb->encapsulation = 1;
364 skb_set_inner_mac_header(skb, nhoff + sizeof(struct udphdr));
365
Or Gerlitzb582ef02014-01-20 13:59:19 +0200366 return err;
367}
Tom Herberta6024562016-04-05 08:22:51 -0700368EXPORT_SYMBOL(udp_gro_complete);
Or Gerlitzb582ef02014-01-20 13:59:19 +0200369
Eric Dumazet72bb17b2014-09-09 08:29:12 -0700370static int udp4_gro_complete(struct sk_buff *skb, int nhoff)
Tom Herbert57c67ff2014-08-22 13:34:44 -0700371{
372 const struct iphdr *iph = ip_hdr(skb);
373 struct udphdr *uh = (struct udphdr *)(skb->data + nhoff);
374
Tom Herbert6db93ea2015-02-10 16:30:29 -0800375 if (uh->check) {
376 skb_shinfo(skb)->gso_type |= SKB_GSO_UDP_TUNNEL_CSUM;
Tom Herbert57c67ff2014-08-22 13:34:44 -0700377 uh->check = ~udp_v4_check(skb->len - nhoff, iph->saddr,
378 iph->daddr, 0);
Tom Herbert6db93ea2015-02-10 16:30:29 -0800379 } else {
380 skb_shinfo(skb)->gso_type |= SKB_GSO_UDP_TUNNEL;
381 }
Tom Herbert57c67ff2014-08-22 13:34:44 -0700382
Tom Herberta6024562016-04-05 08:22:51 -0700383 return udp_gro_complete(skb, nhoff, udp4_lib_lookup_skb);
Tom Herbert57c67ff2014-08-22 13:34:44 -0700384}
385
Daniel Borkmannda5bab02013-06-08 12:56:03 +0200386static const struct net_offload udpv4_offload = {
387 .callbacks = {
Daniel Borkmannda5bab02013-06-08 12:56:03 +0200388 .gso_segment = udp4_ufo_fragment,
Tom Herbert57c67ff2014-08-22 13:34:44 -0700389 .gro_receive = udp4_gro_receive,
390 .gro_complete = udp4_gro_complete,
Daniel Borkmannda5bab02013-06-08 12:56:03 +0200391 },
392};
393
394int __init udpv4_offload_init(void)
395{
396 return inet_add_offload(&udpv4_offload, IPPROTO_UDP);
397}