]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - net/ipv4/udp_offload.c
Merge remote-tracking branches 'asoc/topic/inntel', 'asoc/topic/input', 'asoc/topic...
[mirror_ubuntu-bionic-kernel.git] / net / ipv4 / udp_offload.c
CommitLineData
da5bab07
DB
1/*
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
8bce6d7d
TH
17static 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),
4bcb877d 21 __be16 new_protocol, bool is_ipv6)
155e010e 22{
dbef491e 23 int tnl_hlen = skb_inner_mac_header(skb) - skb_transport_header(skb);
07b26c94 24 bool remcsum, need_csum, offload_csum, ufo, gso_partial;
155e010e 25 struct sk_buff *segs = ERR_PTR(-EINVAL);
dbef491e 26 struct udphdr *uh = udp_hdr(skb);
155e010e 27 u16 mac_offset = skb->mac_header;
155e010e 28 __be16 protocol = skb->protocol;
dbef491e 29 u16 mac_len = skb->mac_len;
155e010e 30 int udp_offset, outer_hlen;
08334824 31 __wsum partial;
155e010e
TH
32
33 if (unlikely(!pskb_may_pull(skb, tnl_hlen)))
34 goto out;
35
08334824
AD
36 /* 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 */
802ab55a
AD
42 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);
dbef491e
AD
47
48 /* setup inner skb. */
155e010e 49 skb->encapsulation = 0;
5197f349 50 SKB_GSO_CB(skb)->encap_level = 0;
155e010e
TH
51 __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);
8bce6d7d 55 skb->protocol = new_protocol;
fdaefd62
AD
56
57 need_csum = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_TUNNEL_CSUM);
4bcb877d 58 skb->encap_hdr_csum = need_csum;
fdaefd62
AD
59
60 remcsum = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TUNNEL_REMCSUM);
e585f236 61 skb->remcsum_offload = remcsum;
155e010e 62
22463876
AD
63 ufo = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
64
4bcb877d
TH
65 /* Try to offload checksum if possible */
66 offload_csum = !!(need_csum &&
fdaefd62
AD
67 (skb->dev->features &
68 (is_ipv6 ? (NETIF_F_HW_CSUM | NETIF_F_IPV6_CSUM) :
69 (NETIF_F_HW_CSUM | NETIF_F_IP_CSUM))));
155e010e 70
bef3c6c9
AD
71 features &= skb->dev->hw_enc_features;
72
7fbeffed
AD
73 /* 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 */
22463876 77 if (remcsum || ufo) {
7fbeffed 78 features &= ~NETIF_F_CSUM_MASK;
22463876 79 if (!need_csum || offload_csum)
7fbeffed
AD
80 features |= NETIF_F_HW_CSUM;
81 }
82
155e010e 83 /* segment inner packet. */
bef3c6c9 84 segs = gso_inner_segment(skb, features);
27446442 85 if (IS_ERR_OR_NULL(segs)) {
155e010e
TH
86 skb_gso_error_unwind(skb, protocol, tnl_hlen, mac_offset,
87 mac_len);
88 goto out;
89 }
90
07b26c94
SK
91 gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
92
155e010e
TH
93 outer_hlen = skb_tnl_header_len(skb);
94 udp_offset = outer_hlen - tnl_hlen;
95 skb = segs;
96 do {
802ab55a 97 unsigned int len;
4bcb877d 98
fdaefd62 99 if (remcsum)
4bcb877d 100 skb->ip_summed = CHECKSUM_NONE;
fdaefd62
AD
101
102 /* Set up inner headers if we are offloading inner checksum */
103 if (skb->ip_summed == CHECKSUM_PARTIAL) {
4bcb877d
TH
104 skb_reset_inner_headers(skb);
105 skb->encapsulation = 1;
106 }
155e010e
TH
107
108 skb->mac_len = mac_len;
4bcb877d 109 skb->protocol = protocol;
155e010e 110
dbef491e 111 __skb_push(skb, outer_hlen);
155e010e
TH
112 skb_reset_mac_header(skb);
113 skb_set_network_header(skb, mac_len);
114 skb_set_transport_header(skb, udp_offset);
802ab55a 115 len = skb->len - udp_offset;
155e010e 116 uh = udp_hdr(skb);
802ab55a
AD
117
118 /* If we are only performing partial GSO the inner header
119 * will be using a length value equal to only one MSS sized
120 * segment instead of the entire frame.
121 */
07b26c94 122 if (gso_partial) {
802ab55a
AD
123 uh->len = htons(skb_shinfo(skb)->gso_size +
124 SKB_GSO_CB(skb)->data_offset +
125 skb->head - (unsigned char *)uh);
126 } else {
127 uh->len = htons(len);
128 }
155e010e 129
4bcb877d
TH
130 if (!need_csum)
131 continue;
132
802ab55a
AD
133 uh->check = ~csum_fold(csum_add(partial,
134 (__force __wsum)htonl(len)));
155e010e 135
fdaefd62
AD
136 if (skb->encapsulation || !offload_csum) {
137 uh->check = gso_make_checksum(skb, ~uh->check);
155e010e
TH
138 if (uh->check == 0)
139 uh->check = CSUM_MANGLED_0;
fdaefd62
AD
140 } else {
141 skb->ip_summed = CHECKSUM_PARTIAL;
142 skb->csum_start = skb_transport_header(skb) - skb->head;
143 skb->csum_offset = offsetof(struct udphdr, check);
155e010e 144 }
155e010e
TH
145 } while ((skb = skb->next));
146out:
147 return segs;
148}
149
8bce6d7d
TH
150struct sk_buff *skb_udp_tunnel_segment(struct sk_buff *skb,
151 netdev_features_t features,
152 bool is_ipv6)
153{
154 __be16 protocol = skb->protocol;
155 const struct net_offload **offloads;
156 const struct net_offload *ops;
157 struct sk_buff *segs = ERR_PTR(-EINVAL);
158 struct sk_buff *(*gso_inner_segment)(struct sk_buff *skb,
159 netdev_features_t features);
160
161 rcu_read_lock();
162
163 switch (skb->inner_protocol_type) {
164 case ENCAP_TYPE_ETHER:
165 protocol = skb->inner_protocol;
166 gso_inner_segment = skb_mac_gso_segment;
167 break;
168 case ENCAP_TYPE_IPPROTO:
169 offloads = is_ipv6 ? inet6_offloads : inet_offloads;
170 ops = rcu_dereference(offloads[skb->inner_ipproto]);
171 if (!ops || !ops->callbacks.gso_segment)
172 goto out_unlock;
173 gso_inner_segment = ops->callbacks.gso_segment;
174 break;
175 default:
176 goto out_unlock;
177 }
178
179 segs = __skb_udp_tunnel_segment(skb, features, gso_inner_segment,
4bcb877d 180 protocol, is_ipv6);
8bce6d7d
TH
181
182out_unlock:
183 rcu_read_unlock();
184
185 return segs;
186}
a6024562 187EXPORT_SYMBOL(skb_udp_tunnel_segment);
8bce6d7d 188
da5bab07
DB
189static struct sk_buff *udp4_ufo_fragment(struct sk_buff *skb,
190 netdev_features_t features)
191{
192 struct sk_buff *segs = ERR_PTR(-EINVAL);
193 unsigned int mss;
7a7ffbab 194 __wsum csum;
f71470b3
TH
195 struct udphdr *uh;
196 struct iphdr *iph;
7a7ffbab
WCC
197
198 if (skb->encapsulation &&
0f4f4ffa
TH
199 (skb_shinfo(skb)->gso_type &
200 (SKB_GSO_UDP_TUNNEL|SKB_GSO_UDP_TUNNEL_CSUM))) {
8bce6d7d 201 segs = skb_udp_tunnel_segment(skb, features, false);
7a7ffbab
WCC
202 goto out;
203 }
da5bab07 204
f71470b3
TH
205 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
206 goto out;
207
da5bab07
DB
208 mss = skb_shinfo(skb)->gso_size;
209 if (unlikely(skb->len <= mss))
210 goto out;
211
212 if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) {
213 /* Packet is from an untrusted source, reset gso_segs. */
da5bab07
DB
214
215 skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss);
216
217 segs = NULL;
218 goto out;
219 }
220
7a7ffbab
WCC
221 /* Do software UFO. Complete and fill in the UDP checksum as
222 * HW cannot do checksum of UDP packets sent as multiple
223 * IP fragments.
224 */
f71470b3
TH
225
226 uh = udp_hdr(skb);
227 iph = ip_hdr(skb);
228
229 uh->check = 0;
230 csum = skb_checksum(skb, 0, skb->len, 0);
231 uh->check = udp_v4_check(skb->len, iph->saddr, iph->daddr, csum);
232 if (uh->check == 0)
233 uh->check = CSUM_MANGLED_0;
234
7a7ffbab
WCC
235 skb->ip_summed = CHECKSUM_NONE;
236
22463876
AD
237 /* If there is no outer header we can fake a checksum offload
238 * due to the fact that we have already done the checksum in
239 * software prior to segmenting the frame.
240 */
241 if (!skb->encap_hdr_csum)
242 features |= NETIF_F_HW_CSUM;
243
da5bab07
DB
244 /* Fragment the skb. IP headers of the fragments are updated in
245 * inet_gso_segment()
246 */
7a7ffbab 247 segs = skb_segment(skb, features);
da5bab07
DB
248out:
249 return segs;
250}
251
57c67ff4 252struct sk_buff **udp_gro_receive(struct sk_buff **head, struct sk_buff *skb,
a6024562 253 struct udphdr *uh, udp_lookup_t lookup)
b582ef09 254{
b582ef09 255 struct sk_buff *p, **pp = NULL;
57c67ff4
TH
256 struct udphdr *uh2;
257 unsigned int off = skb_gro_offset(skb);
b582ef09 258 int flush = 1;
a6024562 259 struct sock *sk;
b582ef09 260
fac8e0f5 261 if (NAPI_GRO_CB(skb)->encap_mark ||
662880f4
TH
262 (skb->ip_summed != CHECKSUM_PARTIAL &&
263 NAPI_GRO_CB(skb)->csum_cnt == 0 &&
264 !NAPI_GRO_CB(skb)->csum_valid))
b582ef09
OG
265 goto out;
266
fac8e0f5
JG
267 /* mark that this skb passed once through the tunnel gro layer */
268 NAPI_GRO_CB(skb)->encap_mark = 1;
b582ef09
OG
269
270 rcu_read_lock();
a6024562
TH
271 sk = (*lookup)(skb, uh->source, uh->dest);
272
273 if (sk && udp_sk(sk)->gro_receive)
274 goto unflush;
b582ef09
OG
275 goto out_unlock;
276
277unflush:
278 flush = 0;
279
280 for (p = *head; p; p = p->next) {
281 if (!NAPI_GRO_CB(p)->same_flow)
282 continue;
283
284 uh2 = (struct udphdr *)(p->data + off);
57c67ff4
TH
285
286 /* Match ports and either checksums are either both zero
287 * or nonzero.
288 */
289 if ((*(u32 *)&uh->source != *(u32 *)&uh2->source) ||
290 (!uh->check ^ !uh2->check)) {
b582ef09
OG
291 NAPI_GRO_CB(p)->same_flow = 0;
292 continue;
293 }
294 }
295
296 skb_gro_pull(skb, sizeof(struct udphdr)); /* pull encapsulating udp header */
6bae1d4c 297 skb_gro_postpull_rcsum(skb, uh, sizeof(struct udphdr));
fcd91dd4 298 pp = call_gro_receive_sk(udp_sk(sk)->gro_receive, sk, head, skb);
b582ef09
OG
299
300out_unlock:
301 rcu_read_unlock();
302out:
303 NAPI_GRO_CB(skb)->flush |= flush;
304 return pp;
305}
a6024562 306EXPORT_SYMBOL(udp_gro_receive);
b582ef09 307
57c67ff4
TH
308static struct sk_buff **udp4_gro_receive(struct sk_buff **head,
309 struct sk_buff *skb)
310{
311 struct udphdr *uh = udp_gro_udphdr(skb);
312
2abb7cdc
TH
313 if (unlikely(!uh))
314 goto flush;
57c67ff4 315
2abb7cdc 316 /* Don't bother verifying checksum if we're going to flush anyway. */
2d8f7e2c 317 if (NAPI_GRO_CB(skb)->flush)
2abb7cdc
TH
318 goto skip;
319
320 if (skb_gro_checksum_validate_zero_check(skb, IPPROTO_UDP, uh->check,
321 inet_gro_compute_pseudo))
322 goto flush;
323 else if (uh->check)
324 skb_gro_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
325 inet_gro_compute_pseudo);
326skip:
efc98d08 327 NAPI_GRO_CB(skb)->is_ipv6 = 0;
a6024562 328 return udp_gro_receive(head, skb, uh, udp4_lib_lookup_skb);
2abb7cdc
TH
329
330flush:
331 NAPI_GRO_CB(skb)->flush = 1;
332 return NULL;
57c67ff4
TH
333}
334
a6024562
TH
335int udp_gro_complete(struct sk_buff *skb, int nhoff,
336 udp_lookup_t lookup)
b582ef09 337{
b582ef09
OG
338 __be16 newlen = htons(skb->len - nhoff);
339 struct udphdr *uh = (struct udphdr *)(skb->data + nhoff);
340 int err = -ENOSYS;
a6024562 341 struct sock *sk;
b582ef09
OG
342
343 uh->len = newlen;
344
229740c6
JR
345 /* Set encapsulation before calling into inner gro_complete() functions
346 * to make them set up the inner offsets.
347 */
348 skb->encapsulation = 1;
349
b582ef09 350 rcu_read_lock();
a6024562
TH
351 sk = (*lookup)(skb, uh->source, uh->dest);
352 if (sk && udp_sk(sk)->gro_complete)
353 err = udp_sk(sk)->gro_complete(sk, skb,
354 nhoff + sizeof(struct udphdr));
b582ef09 355 rcu_read_unlock();
6db93ea1
TH
356
357 if (skb->remcsum_offload)
358 skb_shinfo(skb)->gso_type |= SKB_GSO_TUNNEL_REMCSUM;
359
b582ef09
OG
360 return err;
361}
a6024562 362EXPORT_SYMBOL(udp_gro_complete);
b582ef09 363
72bb17b3 364static int udp4_gro_complete(struct sk_buff *skb, int nhoff)
57c67ff4
TH
365{
366 const struct iphdr *iph = ip_hdr(skb);
367 struct udphdr *uh = (struct udphdr *)(skb->data + nhoff);
368
6db93ea1
TH
369 if (uh->check) {
370 skb_shinfo(skb)->gso_type |= SKB_GSO_UDP_TUNNEL_CSUM;
57c67ff4
TH
371 uh->check = ~udp_v4_check(skb->len - nhoff, iph->saddr,
372 iph->daddr, 0);
6db93ea1
TH
373 } else {
374 skb_shinfo(skb)->gso_type |= SKB_GSO_UDP_TUNNEL;
375 }
57c67ff4 376
a6024562 377 return udp_gro_complete(skb, nhoff, udp4_lib_lookup_skb);
57c67ff4
TH
378}
379
da5bab07
DB
380static const struct net_offload udpv4_offload = {
381 .callbacks = {
da5bab07 382 .gso_segment = udp4_ufo_fragment,
57c67ff4
TH
383 .gro_receive = udp4_gro_receive,
384 .gro_complete = udp4_gro_complete,
da5bab07
DB
385 },
386};
387
388int __init udpv4_offload_init(void)
389{
390 return inet_add_offload(&udpv4_offload, IPPROTO_UDP);
391}