Loading...
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
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),
21 __be16 new_protocol, bool is_ipv6)
22{
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;
31 __wsum partial;
32 bool need_ipsec;
33
34 if (unlikely(!pskb_may_pull(skb, tnl_hlen)))
35 goto out;
36
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.
42 */
43 if (skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)
44 partial = (__force __wsum)uh->len;
45 else
46 partial = (__force __wsum)htonl(skb->len);
47 partial = csum_sub(csum_unfold(uh->check), partial);
48
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;
57
58 need_csum = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_TUNNEL_CSUM);
59 skb->encap_hdr_csum = need_csum;
60
61 remcsum = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TUNNEL_REMCSUM);
62 skb->remcsum_offload = remcsum;
63
64 need_ipsec = skb_dst(skb) && dst_xfrm(skb_dst(skb));
65 /* Try to offload checksum if possible */
66 offload_csum = !!(need_csum &&
67 !need_ipsec &&
68 (skb->dev->features &
69 (is_ipv6 ? (NETIF_F_HW_CSUM | NETIF_F_IPV6_CSUM) :
70 (NETIF_F_HW_CSUM | NETIF_F_IP_CSUM))));
71
72 features &= skb->dev->hw_enc_features;
73
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.
77 */
78 if (remcsum) {
79 features &= ~NETIF_F_CSUM_MASK;
80 if (!need_csum || offload_csum)
81 features |= NETIF_F_HW_CSUM;
82 }
83
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,
88 mac_len);
89 goto out;
90 }
91
92 gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
93
94 outer_hlen = skb_tnl_header_len(skb);
95 udp_offset = outer_hlen - tnl_hlen;
96 skb = segs;
97 do {
98 unsigned int len;
99
100 if (remcsum)
101 skb->ip_summed = CHECKSUM_NONE;
102
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;
107 }
108
109 skb->mac_len = mac_len;
110 skb->protocol = protocol;
111
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;
117 uh = udp_hdr(skb);
118
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.
122 */
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);
127 } else {
128 uh->len = htons(len);
129 }
130
131 if (!need_csum)
132 continue;
133
134 uh->check = ~csum_fold(csum_add(partial,
135 (__force __wsum)htonl(len)));
136
137 if (skb->encapsulation || !offload_csum) {
138 uh->check = gso_make_checksum(skb, ~uh->check);
139 if (uh->check == 0)
140 uh->check = CSUM_MANGLED_0;
141 } else {
142 skb->ip_summed = CHECKSUM_PARTIAL;
143 skb->csum_start = skb_transport_header(skb) - skb->head;
144 skb->csum_offset = offsetof(struct udphdr, check);
145 }
146 } while ((skb = skb->next));
147out:
148 return segs;
149}
150
151struct sk_buff *skb_udp_tunnel_segment(struct sk_buff *skb,
152 netdev_features_t features,
153 bool is_ipv6)
154{
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);
161
162 rcu_read_lock();
163
164 switch (skb->inner_protocol_type) {
165 case ENCAP_TYPE_ETHER:
166 protocol = skb->inner_protocol;
167 gso_inner_segment = skb_mac_gso_segment;
168 break;
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)
173 goto out_unlock;
174 gso_inner_segment = ops->callbacks.gso_segment;
175 break;
176 default:
177 goto out_unlock;
178 }
179
180 segs = __skb_udp_tunnel_segment(skb, features, gso_inner_segment,
181 protocol, is_ipv6);
182
183out_unlock:
184 rcu_read_unlock();
185
186 return segs;
187}
188EXPORT_SYMBOL(skb_udp_tunnel_segment);
189
190static struct sk_buff *udp4_ufo_fragment(struct sk_buff *skb,
191 netdev_features_t features)
192{
193 struct sk_buff *segs = ERR_PTR(-EINVAL);
194 unsigned int mss;
195 __wsum csum;
196 struct udphdr *uh;
197 struct iphdr *iph;
198
199 if (skb->encapsulation &&
200 (skb_shinfo(skb)->gso_type &
201 (SKB_GSO_UDP_TUNNEL|SKB_GSO_UDP_TUNNEL_CSUM))) {
202 segs = skb_udp_tunnel_segment(skb, features, false);
203 goto out;
204 }
205
206 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP))
207 goto out;
208
209 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
210 goto out;
211
212 mss = skb_shinfo(skb)->gso_size;
213 if (unlikely(skb->len <= mss))
214 goto out;
215
216 /* Do software UFO. Complete and fill in the UDP checksum as
217 * HW cannot do checksum of UDP packets sent as multiple
218 * IP fragments.
219 */
220
221 uh = udp_hdr(skb);
222 iph = ip_hdr(skb);
223
224 uh->check = 0;
225 csum = skb_checksum(skb, 0, skb->len, 0);
226 uh->check = udp_v4_check(skb->len, iph->saddr, iph->daddr, csum);
227 if (uh->check == 0)
228 uh->check = CSUM_MANGLED_0;
229
230 skb->ip_summed = CHECKSUM_UNNECESSARY;
231
232 /* If there is no outer header we can fake a checksum offload
233 * due to the fact that we have already done the checksum in
234 * software prior to segmenting the frame.
235 */
236 if (!skb->encap_hdr_csum)
237 features |= NETIF_F_HW_CSUM;
238
239 /* Fragment the skb. IP headers of the fragments are updated in
240 * inet_gso_segment()
241 */
242 segs = skb_segment(skb, features);
243out:
244 return segs;
245}
246
247struct sk_buff **udp_gro_receive(struct sk_buff **head, struct sk_buff *skb,
248 struct udphdr *uh, udp_lookup_t lookup)
249{
250 struct sk_buff *p, **pp = NULL;
251 struct udphdr *uh2;
252 unsigned int off = skb_gro_offset(skb);
253 int flush = 1;
254 struct sock *sk;
255
256 if (NAPI_GRO_CB(skb)->encap_mark ||
257 (skb->ip_summed != CHECKSUM_PARTIAL &&
258 NAPI_GRO_CB(skb)->csum_cnt == 0 &&
259 !NAPI_GRO_CB(skb)->csum_valid))
260 goto out;
261
262 /* mark that this skb passed once through the tunnel gro layer */
263 NAPI_GRO_CB(skb)->encap_mark = 1;
264
265 rcu_read_lock();
266 sk = (*lookup)(skb, uh->source, uh->dest);
267
268 if (sk && udp_sk(sk)->gro_receive)
269 goto unflush;
270 goto out_unlock;
271
272unflush:
273 flush = 0;
274
275 for (p = *head; p; p = p->next) {
276 if (!NAPI_GRO_CB(p)->same_flow)
277 continue;
278
279 uh2 = (struct udphdr *)(p->data + off);
280
281 /* Match ports and either checksums are either both zero
282 * or nonzero.
283 */
284 if ((*(u32 *)&uh->source != *(u32 *)&uh2->source) ||
285 (!uh->check ^ !uh2->check)) {
286 NAPI_GRO_CB(p)->same_flow = 0;
287 continue;
288 }
289 }
290
291 skb_gro_pull(skb, sizeof(struct udphdr)); /* pull encapsulating udp header */
292 skb_gro_postpull_rcsum(skb, uh, sizeof(struct udphdr));
293 pp = call_gro_receive_sk(udp_sk(sk)->gro_receive, sk, head, skb);
294
295out_unlock:
296 rcu_read_unlock();
297out:
298 NAPI_GRO_CB(skb)->flush |= flush;
299 return pp;
300}
301EXPORT_SYMBOL(udp_gro_receive);
302
303static struct sk_buff **udp4_gro_receive(struct sk_buff **head,
304 struct sk_buff *skb)
305{
306 struct udphdr *uh = udp_gro_udphdr(skb);
307
308 if (unlikely(!uh))
309 goto flush;
310
311 /* Don't bother verifying checksum if we're going to flush anyway. */
312 if (NAPI_GRO_CB(skb)->flush)
313 goto skip;
314
315 if (skb_gro_checksum_validate_zero_check(skb, IPPROTO_UDP, uh->check,
316 inet_gro_compute_pseudo))
317 goto flush;
318 else if (uh->check)
319 skb_gro_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
320 inet_gro_compute_pseudo);
321skip:
322 NAPI_GRO_CB(skb)->is_ipv6 = 0;
323 return udp_gro_receive(head, skb, uh, udp4_lib_lookup_skb);
324
325flush:
326 NAPI_GRO_CB(skb)->flush = 1;
327 return NULL;
328}
329
330int udp_gro_complete(struct sk_buff *skb, int nhoff,
331 udp_lookup_t lookup)
332{
333 __be16 newlen = htons(skb->len - nhoff);
334 struct udphdr *uh = (struct udphdr *)(skb->data + nhoff);
335 int err = -ENOSYS;
336 struct sock *sk;
337
338 uh->len = newlen;
339
340 /* Set encapsulation before calling into inner gro_complete() functions
341 * to make them set up the inner offsets.
342 */
343 skb->encapsulation = 1;
344
345 rcu_read_lock();
346 sk = (*lookup)(skb, uh->source, uh->dest);
347 if (sk && udp_sk(sk)->gro_complete)
348 err = udp_sk(sk)->gro_complete(sk, skb,
349 nhoff + sizeof(struct udphdr));
350 rcu_read_unlock();
351
352 if (skb->remcsum_offload)
353 skb_shinfo(skb)->gso_type |= SKB_GSO_TUNNEL_REMCSUM;
354
355 return err;
356}
357EXPORT_SYMBOL(udp_gro_complete);
358
359static int udp4_gro_complete(struct sk_buff *skb, int nhoff)
360{
361 const struct iphdr *iph = ip_hdr(skb);
362 struct udphdr *uh = (struct udphdr *)(skb->data + nhoff);
363
364 if (uh->check) {
365 skb_shinfo(skb)->gso_type |= SKB_GSO_UDP_TUNNEL_CSUM;
366 uh->check = ~udp_v4_check(skb->len - nhoff, iph->saddr,
367 iph->daddr, 0);
368 } else {
369 skb_shinfo(skb)->gso_type |= SKB_GSO_UDP_TUNNEL;
370 }
371
372 return udp_gro_complete(skb, nhoff, udp4_lib_lookup_skb);
373}
374
375static const struct net_offload udpv4_offload = {
376 .callbacks = {
377 .gso_segment = udp4_ufo_fragment,
378 .gro_receive = udp4_gro_receive,
379 .gro_complete = udp4_gro_complete,
380 },
381};
382
383int __init udpv4_offload_init(void)
384{
385 return inet_add_offload(&udpv4_offload, IPPROTO_UDP);
386}
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * IPV4 GSO/GRO offload support
4 * Linux INET implementation
5 *
6 * UDPv4 GSO support
7 */
8
9#include <linux/skbuff.h>
10#include <net/gro.h>
11#include <net/udp.h>
12#include <net/protocol.h>
13#include <net/inet_common.h>
14
15static struct sk_buff *__skb_udp_tunnel_segment(struct sk_buff *skb,
16 netdev_features_t features,
17 struct sk_buff *(*gso_inner_segment)(struct sk_buff *skb,
18 netdev_features_t features),
19 __be16 new_protocol, bool is_ipv6)
20{
21 int tnl_hlen = skb_inner_mac_header(skb) - skb_transport_header(skb);
22 bool remcsum, need_csum, offload_csum, gso_partial;
23 struct sk_buff *segs = ERR_PTR(-EINVAL);
24 struct udphdr *uh = udp_hdr(skb);
25 u16 mac_offset = skb->mac_header;
26 __be16 protocol = skb->protocol;
27 u16 mac_len = skb->mac_len;
28 int udp_offset, outer_hlen;
29 __wsum partial;
30 bool need_ipsec;
31
32 if (unlikely(!pskb_may_pull(skb, tnl_hlen)))
33 goto out;
34
35 /* Adjust partial header checksum to negate old length.
36 * We cannot rely on the value contained in uh->len as it is
37 * possible that the actual value exceeds the boundaries of the
38 * 16 bit length field due to the header being added outside of an
39 * IP or IPv6 frame that was already limited to 64K - 1.
40 */
41 if (skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)
42 partial = (__force __wsum)uh->len;
43 else
44 partial = (__force __wsum)htonl(skb->len);
45 partial = csum_sub(csum_unfold(uh->check), partial);
46
47 /* setup inner skb. */
48 skb->encapsulation = 0;
49 SKB_GSO_CB(skb)->encap_level = 0;
50 __skb_pull(skb, tnl_hlen);
51 skb_reset_mac_header(skb);
52 skb_set_network_header(skb, skb_inner_network_offset(skb));
53 skb_set_transport_header(skb, skb_inner_transport_offset(skb));
54 skb->mac_len = skb_inner_network_offset(skb);
55 skb->protocol = new_protocol;
56
57 need_csum = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_TUNNEL_CSUM);
58 skb->encap_hdr_csum = need_csum;
59
60 remcsum = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TUNNEL_REMCSUM);
61 skb->remcsum_offload = remcsum;
62
63 need_ipsec = skb_dst(skb) && dst_xfrm(skb_dst(skb));
64 /* Try to offload checksum if possible */
65 offload_csum = !!(need_csum &&
66 !need_ipsec &&
67 (skb->dev->features &
68 (is_ipv6 ? (NETIF_F_HW_CSUM | NETIF_F_IPV6_CSUM) :
69 (NETIF_F_HW_CSUM | NETIF_F_IP_CSUM))));
70
71 features &= skb->dev->hw_enc_features;
72 if (need_csum)
73 features &= ~NETIF_F_SCTP_CRC;
74
75 /* The only checksum offload we care about from here on out is the
76 * outer one so strip the existing checksum feature flags and
77 * instead set the flag based on our outer checksum offload value.
78 */
79 if (remcsum) {
80 features &= ~NETIF_F_CSUM_MASK;
81 if (!need_csum || offload_csum)
82 features |= NETIF_F_HW_CSUM;
83 }
84
85 /* segment inner packet. */
86 segs = gso_inner_segment(skb, features);
87 if (IS_ERR_OR_NULL(segs)) {
88 skb_gso_error_unwind(skb, protocol, tnl_hlen, mac_offset,
89 mac_len);
90 goto out;
91 }
92
93 gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
94
95 outer_hlen = skb_tnl_header_len(skb);
96 udp_offset = outer_hlen - tnl_hlen;
97 skb = segs;
98 do {
99 unsigned int len;
100
101 if (remcsum)
102 skb->ip_summed = CHECKSUM_NONE;
103
104 /* Set up inner headers if we are offloading inner checksum */
105 if (skb->ip_summed == CHECKSUM_PARTIAL) {
106 skb_reset_inner_headers(skb);
107 skb->encapsulation = 1;
108 }
109
110 skb->mac_len = mac_len;
111 skb->protocol = protocol;
112
113 __skb_push(skb, outer_hlen);
114 skb_reset_mac_header(skb);
115 skb_set_network_header(skb, mac_len);
116 skb_set_transport_header(skb, udp_offset);
117 len = skb->len - udp_offset;
118 uh = udp_hdr(skb);
119
120 /* If we are only performing partial GSO the inner header
121 * will be using a length value equal to only one MSS sized
122 * segment instead of the entire frame.
123 */
124 if (gso_partial && skb_is_gso(skb)) {
125 uh->len = htons(skb_shinfo(skb)->gso_size +
126 SKB_GSO_CB(skb)->data_offset +
127 skb->head - (unsigned char *)uh);
128 } else {
129 uh->len = htons(len);
130 }
131
132 if (!need_csum)
133 continue;
134
135 uh->check = ~csum_fold(csum_add(partial,
136 (__force __wsum)htonl(len)));
137
138 if (skb->encapsulation || !offload_csum) {
139 uh->check = gso_make_checksum(skb, ~uh->check);
140 if (uh->check == 0)
141 uh->check = CSUM_MANGLED_0;
142 } else {
143 skb->ip_summed = CHECKSUM_PARTIAL;
144 skb->csum_start = skb_transport_header(skb) - skb->head;
145 skb->csum_offset = offsetof(struct udphdr, check);
146 }
147 } while ((skb = skb->next));
148out:
149 return segs;
150}
151
152struct sk_buff *skb_udp_tunnel_segment(struct sk_buff *skb,
153 netdev_features_t features,
154 bool is_ipv6)
155{
156 const struct net_offload __rcu **offloads;
157 __be16 protocol = skb->protocol;
158 const struct net_offload *ops;
159 struct sk_buff *segs = ERR_PTR(-EINVAL);
160 struct sk_buff *(*gso_inner_segment)(struct sk_buff *skb,
161 netdev_features_t features);
162
163 rcu_read_lock();
164
165 switch (skb->inner_protocol_type) {
166 case ENCAP_TYPE_ETHER:
167 protocol = skb->inner_protocol;
168 gso_inner_segment = skb_mac_gso_segment;
169 break;
170 case ENCAP_TYPE_IPPROTO:
171 offloads = is_ipv6 ? inet6_offloads : inet_offloads;
172 ops = rcu_dereference(offloads[skb->inner_ipproto]);
173 if (!ops || !ops->callbacks.gso_segment)
174 goto out_unlock;
175 gso_inner_segment = ops->callbacks.gso_segment;
176 break;
177 default:
178 goto out_unlock;
179 }
180
181 segs = __skb_udp_tunnel_segment(skb, features, gso_inner_segment,
182 protocol, is_ipv6);
183
184out_unlock:
185 rcu_read_unlock();
186
187 return segs;
188}
189EXPORT_SYMBOL(skb_udp_tunnel_segment);
190
191static void __udpv4_gso_segment_csum(struct sk_buff *seg,
192 __be32 *oldip, __be32 *newip,
193 __be16 *oldport, __be16 *newport)
194{
195 struct udphdr *uh;
196 struct iphdr *iph;
197
198 if (*oldip == *newip && *oldport == *newport)
199 return;
200
201 uh = udp_hdr(seg);
202 iph = ip_hdr(seg);
203
204 if (uh->check) {
205 inet_proto_csum_replace4(&uh->check, seg, *oldip, *newip,
206 true);
207 inet_proto_csum_replace2(&uh->check, seg, *oldport, *newport,
208 false);
209 if (!uh->check)
210 uh->check = CSUM_MANGLED_0;
211 }
212 *oldport = *newport;
213
214 csum_replace4(&iph->check, *oldip, *newip);
215 *oldip = *newip;
216}
217
218static struct sk_buff *__udpv4_gso_segment_list_csum(struct sk_buff *segs)
219{
220 struct sk_buff *seg;
221 struct udphdr *uh, *uh2;
222 struct iphdr *iph, *iph2;
223
224 seg = segs;
225 uh = udp_hdr(seg);
226 iph = ip_hdr(seg);
227
228 if ((udp_hdr(seg)->dest == udp_hdr(seg->next)->dest) &&
229 (udp_hdr(seg)->source == udp_hdr(seg->next)->source) &&
230 (ip_hdr(seg)->daddr == ip_hdr(seg->next)->daddr) &&
231 (ip_hdr(seg)->saddr == ip_hdr(seg->next)->saddr))
232 return segs;
233
234 while ((seg = seg->next)) {
235 uh2 = udp_hdr(seg);
236 iph2 = ip_hdr(seg);
237
238 __udpv4_gso_segment_csum(seg,
239 &iph2->saddr, &iph->saddr,
240 &uh2->source, &uh->source);
241 __udpv4_gso_segment_csum(seg,
242 &iph2->daddr, &iph->daddr,
243 &uh2->dest, &uh->dest);
244 }
245
246 return segs;
247}
248
249static struct sk_buff *__udp_gso_segment_list(struct sk_buff *skb,
250 netdev_features_t features,
251 bool is_ipv6)
252{
253 unsigned int mss = skb_shinfo(skb)->gso_size;
254
255 skb = skb_segment_list(skb, features, skb_mac_header_len(skb));
256 if (IS_ERR(skb))
257 return skb;
258
259 udp_hdr(skb)->len = htons(sizeof(struct udphdr) + mss);
260
261 return is_ipv6 ? skb : __udpv4_gso_segment_list_csum(skb);
262}
263
264struct sk_buff *__udp_gso_segment(struct sk_buff *gso_skb,
265 netdev_features_t features, bool is_ipv6)
266{
267 struct sock *sk = gso_skb->sk;
268 unsigned int sum_truesize = 0;
269 struct sk_buff *segs, *seg;
270 struct udphdr *uh;
271 unsigned int mss;
272 bool copy_dtor;
273 __sum16 check;
274 __be16 newlen;
275
276 if (skb_shinfo(gso_skb)->gso_type & SKB_GSO_FRAGLIST)
277 return __udp_gso_segment_list(gso_skb, features, is_ipv6);
278
279 mss = skb_shinfo(gso_skb)->gso_size;
280 if (gso_skb->len <= sizeof(*uh) + mss)
281 return ERR_PTR(-EINVAL);
282
283 skb_pull(gso_skb, sizeof(*uh));
284
285 /* clear destructor to avoid skb_segment assigning it to tail */
286 copy_dtor = gso_skb->destructor == sock_wfree;
287 if (copy_dtor)
288 gso_skb->destructor = NULL;
289
290 segs = skb_segment(gso_skb, features);
291 if (IS_ERR_OR_NULL(segs)) {
292 if (copy_dtor)
293 gso_skb->destructor = sock_wfree;
294 return segs;
295 }
296
297 /* GSO partial and frag_list segmentation only requires splitting
298 * the frame into an MSS multiple and possibly a remainder, both
299 * cases return a GSO skb. So update the mss now.
300 */
301 if (skb_is_gso(segs))
302 mss *= skb_shinfo(segs)->gso_segs;
303
304 seg = segs;
305 uh = udp_hdr(seg);
306
307 /* preserve TX timestamp flags and TS key for first segment */
308 skb_shinfo(seg)->tskey = skb_shinfo(gso_skb)->tskey;
309 skb_shinfo(seg)->tx_flags |=
310 (skb_shinfo(gso_skb)->tx_flags & SKBTX_ANY_TSTAMP);
311
312 /* compute checksum adjustment based on old length versus new */
313 newlen = htons(sizeof(*uh) + mss);
314 check = csum16_add(csum16_sub(uh->check, uh->len), newlen);
315
316 for (;;) {
317 if (copy_dtor) {
318 seg->destructor = sock_wfree;
319 seg->sk = sk;
320 sum_truesize += seg->truesize;
321 }
322
323 if (!seg->next)
324 break;
325
326 uh->len = newlen;
327 uh->check = check;
328
329 if (seg->ip_summed == CHECKSUM_PARTIAL)
330 gso_reset_checksum(seg, ~check);
331 else
332 uh->check = gso_make_checksum(seg, ~check) ? :
333 CSUM_MANGLED_0;
334
335 seg = seg->next;
336 uh = udp_hdr(seg);
337 }
338
339 /* last packet can be partial gso_size, account for that in checksum */
340 newlen = htons(skb_tail_pointer(seg) - skb_transport_header(seg) +
341 seg->data_len);
342 check = csum16_add(csum16_sub(uh->check, uh->len), newlen);
343
344 uh->len = newlen;
345 uh->check = check;
346
347 if (seg->ip_summed == CHECKSUM_PARTIAL)
348 gso_reset_checksum(seg, ~check);
349 else
350 uh->check = gso_make_checksum(seg, ~check) ? : CSUM_MANGLED_0;
351
352 /* update refcount for the packet */
353 if (copy_dtor) {
354 int delta = sum_truesize - gso_skb->truesize;
355
356 /* In some pathological cases, delta can be negative.
357 * We need to either use refcount_add() or refcount_sub_and_test()
358 */
359 if (likely(delta >= 0))
360 refcount_add(delta, &sk->sk_wmem_alloc);
361 else
362 WARN_ON_ONCE(refcount_sub_and_test(-delta, &sk->sk_wmem_alloc));
363 }
364 return segs;
365}
366EXPORT_SYMBOL_GPL(__udp_gso_segment);
367
368static struct sk_buff *udp4_ufo_fragment(struct sk_buff *skb,
369 netdev_features_t features)
370{
371 struct sk_buff *segs = ERR_PTR(-EINVAL);
372 unsigned int mss;
373 __wsum csum;
374 struct udphdr *uh;
375 struct iphdr *iph;
376
377 if (skb->encapsulation &&
378 (skb_shinfo(skb)->gso_type &
379 (SKB_GSO_UDP_TUNNEL|SKB_GSO_UDP_TUNNEL_CSUM))) {
380 segs = skb_udp_tunnel_segment(skb, features, false);
381 goto out;
382 }
383
384 if (!(skb_shinfo(skb)->gso_type & (SKB_GSO_UDP | SKB_GSO_UDP_L4)))
385 goto out;
386
387 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
388 goto out;
389
390 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4 &&
391 !skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST))
392 return __udp_gso_segment(skb, features, false);
393
394 mss = skb_shinfo(skb)->gso_size;
395 if (unlikely(skb->len <= mss))
396 goto out;
397
398 /* Do software UFO. Complete and fill in the UDP checksum as
399 * HW cannot do checksum of UDP packets sent as multiple
400 * IP fragments.
401 */
402
403 uh = udp_hdr(skb);
404 iph = ip_hdr(skb);
405
406 uh->check = 0;
407 csum = skb_checksum(skb, 0, skb->len, 0);
408 uh->check = udp_v4_check(skb->len, iph->saddr, iph->daddr, csum);
409 if (uh->check == 0)
410 uh->check = CSUM_MANGLED_0;
411
412 skb->ip_summed = CHECKSUM_UNNECESSARY;
413
414 /* If there is no outer header we can fake a checksum offload
415 * due to the fact that we have already done the checksum in
416 * software prior to segmenting the frame.
417 */
418 if (!skb->encap_hdr_csum)
419 features |= NETIF_F_HW_CSUM;
420
421 /* Fragment the skb. IP headers of the fragments are updated in
422 * inet_gso_segment()
423 */
424 segs = skb_segment(skb, features);
425out:
426 return segs;
427}
428
429static int skb_gro_receive_list(struct sk_buff *p, struct sk_buff *skb)
430{
431 if (unlikely(p->len + skb->len >= 65536))
432 return -E2BIG;
433
434 if (NAPI_GRO_CB(p)->last == p)
435 skb_shinfo(p)->frag_list = skb;
436 else
437 NAPI_GRO_CB(p)->last->next = skb;
438
439 skb_pull(skb, skb_gro_offset(skb));
440
441 NAPI_GRO_CB(p)->last = skb;
442 NAPI_GRO_CB(p)->count++;
443 p->data_len += skb->len;
444
445 /* sk owenrship - if any - completely transferred to the aggregated packet */
446 skb->destructor = NULL;
447 p->truesize += skb->truesize;
448 p->len += skb->len;
449
450 NAPI_GRO_CB(skb)->same_flow = 1;
451
452 return 0;
453}
454
455
456#define UDP_GRO_CNT_MAX 64
457static struct sk_buff *udp_gro_receive_segment(struct list_head *head,
458 struct sk_buff *skb)
459{
460 struct udphdr *uh = udp_gro_udphdr(skb);
461 struct sk_buff *pp = NULL;
462 struct udphdr *uh2;
463 struct sk_buff *p;
464 unsigned int ulen;
465 int ret = 0;
466
467 /* requires non zero csum, for symmetry with GSO */
468 if (!uh->check) {
469 NAPI_GRO_CB(skb)->flush = 1;
470 return NULL;
471 }
472
473 /* Do not deal with padded or malicious packets, sorry ! */
474 ulen = ntohs(uh->len);
475 if (ulen <= sizeof(*uh) || ulen != skb_gro_len(skb)) {
476 NAPI_GRO_CB(skb)->flush = 1;
477 return NULL;
478 }
479 /* pull encapsulating udp header */
480 skb_gro_pull(skb, sizeof(struct udphdr));
481
482 list_for_each_entry(p, head, list) {
483 if (!NAPI_GRO_CB(p)->same_flow)
484 continue;
485
486 uh2 = udp_hdr(p);
487
488 /* Match ports only, as csum is always non zero */
489 if ((*(u32 *)&uh->source != *(u32 *)&uh2->source)) {
490 NAPI_GRO_CB(p)->same_flow = 0;
491 continue;
492 }
493
494 if (NAPI_GRO_CB(skb)->is_flist != NAPI_GRO_CB(p)->is_flist) {
495 NAPI_GRO_CB(skb)->flush = 1;
496 return p;
497 }
498
499 /* Terminate the flow on len mismatch or if it grow "too much".
500 * Under small packet flood GRO count could elsewhere grow a lot
501 * leading to excessive truesize values.
502 * On len mismatch merge the first packet shorter than gso_size,
503 * otherwise complete the GRO packet.
504 */
505 if (ulen > ntohs(uh2->len)) {
506 pp = p;
507 } else {
508 if (NAPI_GRO_CB(skb)->is_flist) {
509 if (!pskb_may_pull(skb, skb_gro_offset(skb))) {
510 NAPI_GRO_CB(skb)->flush = 1;
511 return NULL;
512 }
513 if ((skb->ip_summed != p->ip_summed) ||
514 (skb->csum_level != p->csum_level)) {
515 NAPI_GRO_CB(skb)->flush = 1;
516 return NULL;
517 }
518 ret = skb_gro_receive_list(p, skb);
519 } else {
520 skb_gro_postpull_rcsum(skb, uh,
521 sizeof(struct udphdr));
522
523 ret = skb_gro_receive(p, skb);
524 }
525 }
526
527 if (ret || ulen != ntohs(uh2->len) ||
528 NAPI_GRO_CB(p)->count >= UDP_GRO_CNT_MAX)
529 pp = p;
530
531 return pp;
532 }
533
534 /* mismatch, but we never need to flush */
535 return NULL;
536}
537
538struct sk_buff *udp_gro_receive(struct list_head *head, struct sk_buff *skb,
539 struct udphdr *uh, struct sock *sk)
540{
541 struct sk_buff *pp = NULL;
542 struct sk_buff *p;
543 struct udphdr *uh2;
544 unsigned int off = skb_gro_offset(skb);
545 int flush = 1;
546
547 /* we can do L4 aggregation only if the packet can't land in a tunnel
548 * otherwise we could corrupt the inner stream
549 */
550 NAPI_GRO_CB(skb)->is_flist = 0;
551 if (!sk || !udp_sk(sk)->gro_receive) {
552 if (skb->dev->features & NETIF_F_GRO_FRAGLIST)
553 NAPI_GRO_CB(skb)->is_flist = sk ? !udp_sk(sk)->gro_enabled : 1;
554
555 if ((!sk && (skb->dev->features & NETIF_F_GRO_UDP_FWD)) ||
556 (sk && udp_sk(sk)->gro_enabled) || NAPI_GRO_CB(skb)->is_flist)
557 return call_gro_receive(udp_gro_receive_segment, head, skb);
558
559 /* no GRO, be sure flush the current packet */
560 goto out;
561 }
562
563 if (NAPI_GRO_CB(skb)->encap_mark ||
564 (uh->check && skb->ip_summed != CHECKSUM_PARTIAL &&
565 NAPI_GRO_CB(skb)->csum_cnt == 0 &&
566 !NAPI_GRO_CB(skb)->csum_valid))
567 goto out;
568
569 /* mark that this skb passed once through the tunnel gro layer */
570 NAPI_GRO_CB(skb)->encap_mark = 1;
571
572 flush = 0;
573
574 list_for_each_entry(p, head, list) {
575 if (!NAPI_GRO_CB(p)->same_flow)
576 continue;
577
578 uh2 = (struct udphdr *)(p->data + off);
579
580 /* Match ports and either checksums are either both zero
581 * or nonzero.
582 */
583 if ((*(u32 *)&uh->source != *(u32 *)&uh2->source) ||
584 (!uh->check ^ !uh2->check)) {
585 NAPI_GRO_CB(p)->same_flow = 0;
586 continue;
587 }
588 }
589
590 skb_gro_pull(skb, sizeof(struct udphdr)); /* pull encapsulating udp header */
591 skb_gro_postpull_rcsum(skb, uh, sizeof(struct udphdr));
592 pp = call_gro_receive_sk(udp_sk(sk)->gro_receive, sk, head, skb);
593
594out:
595 skb_gro_flush_final(skb, pp, flush);
596 return pp;
597}
598EXPORT_SYMBOL(udp_gro_receive);
599
600static struct sock *udp4_gro_lookup_skb(struct sk_buff *skb, __be16 sport,
601 __be16 dport)
602{
603 const struct iphdr *iph = skb_gro_network_header(skb);
604 struct net *net = dev_net(skb->dev);
605
606 return __udp4_lib_lookup(net, iph->saddr, sport,
607 iph->daddr, dport, inet_iif(skb),
608 inet_sdif(skb), net->ipv4.udp_table, NULL);
609}
610
611INDIRECT_CALLABLE_SCOPE
612struct sk_buff *udp4_gro_receive(struct list_head *head, struct sk_buff *skb)
613{
614 struct udphdr *uh = udp_gro_udphdr(skb);
615 struct sock *sk = NULL;
616 struct sk_buff *pp;
617
618 if (unlikely(!uh))
619 goto flush;
620
621 /* Don't bother verifying checksum if we're going to flush anyway. */
622 if (NAPI_GRO_CB(skb)->flush)
623 goto skip;
624
625 if (skb_gro_checksum_validate_zero_check(skb, IPPROTO_UDP, uh->check,
626 inet_gro_compute_pseudo))
627 goto flush;
628 else if (uh->check)
629 skb_gro_checksum_try_convert(skb, IPPROTO_UDP,
630 inet_gro_compute_pseudo);
631skip:
632 NAPI_GRO_CB(skb)->is_ipv6 = 0;
633
634 if (static_branch_unlikely(&udp_encap_needed_key))
635 sk = udp4_gro_lookup_skb(skb, uh->source, uh->dest);
636
637 pp = udp_gro_receive(head, skb, uh, sk);
638 return pp;
639
640flush:
641 NAPI_GRO_CB(skb)->flush = 1;
642 return NULL;
643}
644
645static int udp_gro_complete_segment(struct sk_buff *skb)
646{
647 struct udphdr *uh = udp_hdr(skb);
648
649 skb->csum_start = (unsigned char *)uh - skb->head;
650 skb->csum_offset = offsetof(struct udphdr, check);
651 skb->ip_summed = CHECKSUM_PARTIAL;
652
653 skb_shinfo(skb)->gso_segs = NAPI_GRO_CB(skb)->count;
654 skb_shinfo(skb)->gso_type |= SKB_GSO_UDP_L4;
655
656 if (skb->encapsulation)
657 skb->inner_transport_header = skb->transport_header;
658
659 return 0;
660}
661
662int udp_gro_complete(struct sk_buff *skb, int nhoff,
663 udp_lookup_t lookup)
664{
665 __be16 newlen = htons(skb->len - nhoff);
666 struct udphdr *uh = (struct udphdr *)(skb->data + nhoff);
667 struct sock *sk;
668 int err;
669
670 uh->len = newlen;
671
672 sk = INDIRECT_CALL_INET(lookup, udp6_lib_lookup_skb,
673 udp4_lib_lookup_skb, skb, uh->source, uh->dest);
674 if (sk && udp_sk(sk)->gro_complete) {
675 skb_shinfo(skb)->gso_type = uh->check ? SKB_GSO_UDP_TUNNEL_CSUM
676 : SKB_GSO_UDP_TUNNEL;
677
678 /* clear the encap mark, so that inner frag_list gro_complete
679 * can take place
680 */
681 NAPI_GRO_CB(skb)->encap_mark = 0;
682
683 /* Set encapsulation before calling into inner gro_complete()
684 * functions to make them set up the inner offsets.
685 */
686 skb->encapsulation = 1;
687 err = udp_sk(sk)->gro_complete(sk, skb,
688 nhoff + sizeof(struct udphdr));
689 } else {
690 err = udp_gro_complete_segment(skb);
691 }
692
693 if (skb->remcsum_offload)
694 skb_shinfo(skb)->gso_type |= SKB_GSO_TUNNEL_REMCSUM;
695
696 return err;
697}
698EXPORT_SYMBOL(udp_gro_complete);
699
700INDIRECT_CALLABLE_SCOPE int udp4_gro_complete(struct sk_buff *skb, int nhoff)
701{
702 const struct iphdr *iph = ip_hdr(skb);
703 struct udphdr *uh = (struct udphdr *)(skb->data + nhoff);
704
705 /* do fraglist only if there is no outer UDP encap (or we already processed it) */
706 if (NAPI_GRO_CB(skb)->is_flist && !NAPI_GRO_CB(skb)->encap_mark) {
707 uh->len = htons(skb->len - nhoff);
708
709 skb_shinfo(skb)->gso_type |= (SKB_GSO_FRAGLIST|SKB_GSO_UDP_L4);
710 skb_shinfo(skb)->gso_segs = NAPI_GRO_CB(skb)->count;
711
712 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
713 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
714 skb->csum_level++;
715 } else {
716 skb->ip_summed = CHECKSUM_UNNECESSARY;
717 skb->csum_level = 0;
718 }
719
720 return 0;
721 }
722
723 if (uh->check)
724 uh->check = ~udp_v4_check(skb->len - nhoff, iph->saddr,
725 iph->daddr, 0);
726
727 return udp_gro_complete(skb, nhoff, udp4_lib_lookup_skb);
728}
729
730static const struct net_offload udpv4_offload = {
731 .callbacks = {
732 .gso_segment = udp4_ufo_fragment,
733 .gro_receive = udp4_gro_receive,
734 .gro_complete = udp4_gro_complete,
735 },
736};
737
738int __init udpv4_offload_init(void)
739{
740 return inet_add_offload(&udpv4_offload, IPPROTO_UDP);
741}