blob: aa2e0a28ca61388637616b2860da12e6d144e6dd [file] [log] [blame]
Thomas Gleixner2874c5f2019-05-27 08:55:01 +02001// SPDX-License-Identifier: GPL-2.0-or-later
Daniel Borkmannda5bab02013-06-08 12:56:03 +02002/*
3 * IPV4 GSO/GRO offload support
4 * Linux INET implementation
5 *
Daniel Borkmannda5bab02013-06-08 12:56:03 +02006 * UDPv4 GSO support
7 */
8
9#include <linux/skbuff.h>
Eric Dumazet47210312021-11-15 09:05:51 -080010#include <net/gro.h>
Eric Dumazetd457a0e2023-06-08 19:17:37 +000011#include <net/gso.h>
Daniel Borkmannda5bab02013-06-08 12:56:03 +020012#include <net/udp.h>
13#include <net/protocol.h>
Paolo Abeni028e0a42018-12-14 11:51:59 +010014#include <net/inet_common.h>
Daniel Borkmannda5bab02013-06-08 12:56:03 +020015
Tom Herbert8bce6d72014-09-29 20:22:29 -070016static struct sk_buff *__skb_udp_tunnel_segment(struct sk_buff *skb,
17 netdev_features_t features,
18 struct sk_buff *(*gso_inner_segment)(struct sk_buff *skb,
19 netdev_features_t features),
Tom Herbert4bcb8772014-11-04 09:06:52 -080020 __be16 new_protocol, bool is_ipv6)
Tom Herbert155e0102014-07-13 19:49:56 -070021{
Alexander Duyckdbef4912016-02-05 15:28:20 -080022 int tnl_hlen = skb_inner_mac_header(skb) - skb_transport_header(skb);
David S. Miller880388a2017-07-03 07:29:12 -070023 bool remcsum, need_csum, offload_csum, gso_partial;
Tom Herbert155e0102014-07-13 19:49:56 -070024 struct sk_buff *segs = ERR_PTR(-EINVAL);
Alexander Duyckdbef4912016-02-05 15:28:20 -080025 struct udphdr *uh = udp_hdr(skb);
Tom Herbert155e0102014-07-13 19:49:56 -070026 u16 mac_offset = skb->mac_header;
Tom Herbert155e0102014-07-13 19:49:56 -070027 __be16 protocol = skb->protocol;
Alexander Duyckdbef4912016-02-05 15:28:20 -080028 u16 mac_len = skb->mac_len;
Tom Herbert155e0102014-07-13 19:49:56 -070029 int udp_offset, outer_hlen;
Alexander Duyck08334822016-03-11 14:05:47 -080030 __wsum partial;
Ansis Attekab40c5f42017-04-21 15:23:05 -070031 bool need_ipsec;
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));
Xin Long527beb82020-10-29 15:04:57 +080054 skb_set_transport_header(skb, skb_inner_transport_offset(skb));
Tom Herbert155e0102014-07-13 19:49:56 -070055 skb->mac_len = skb_inner_network_offset(skb);
Tom Herbert8bce6d72014-09-29 20:22:29 -070056 skb->protocol = new_protocol;
Alexander Duyckfdaefd62016-02-05 15:28:14 -080057
58 need_csum = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_TUNNEL_CSUM);
Tom Herbert4bcb8772014-11-04 09:06:52 -080059 skb->encap_hdr_csum = need_csum;
Alexander Duyckfdaefd62016-02-05 15:28:14 -080060
61 remcsum = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TUNNEL_REMCSUM);
Tom Herberte585f232014-11-04 09:06:54 -080062 skb->remcsum_offload = remcsum;
Tom Herbert155e0102014-07-13 19:49:56 -070063
Ansis Attekab40c5f42017-04-21 15:23:05 -070064 need_ipsec = skb_dst(skb) && dst_xfrm(skb_dst(skb));
Tom Herbert4bcb8772014-11-04 09:06:52 -080065 /* Try to offload checksum if possible */
66 offload_csum = !!(need_csum &&
Ansis Attekab40c5f42017-04-21 15:23:05 -070067 !need_ipsec &&
Alexander Duyckfdaefd62016-02-05 15:28:14 -080068 (skb->dev->features &
69 (is_ipv6 ? (NETIF_F_HW_CSUM | NETIF_F_IPV6_CSUM) :
70 (NETIF_F_HW_CSUM | NETIF_F_IP_CSUM))));
Tom Herbert155e0102014-07-13 19:49:56 -070071
Alexander Duyckbef3c6c2016-02-05 15:27:31 -080072 features &= skb->dev->hw_enc_features;
Xin Long4eb5d4a2021-01-16 13:59:17 +080073 if (need_csum)
74 features &= ~NETIF_F_SCTP_CRC;
Alexander Duyckbef3c6c2016-02-05 15:27:31 -080075
Alexander Duyck7fbeffe2016-02-05 15:27:43 -080076 /* The only checksum offload we care about from here on out is the
77 * outer one so strip the existing checksum feature flags and
78 * instead set the flag based on our outer checksum offload value.
79 */
David S. Miller880388a2017-07-03 07:29:12 -070080 if (remcsum) {
Alexander Duyck7fbeffe2016-02-05 15:27:43 -080081 features &= ~NETIF_F_CSUM_MASK;
Alexander Duyck22463872016-02-24 16:46:21 -080082 if (!need_csum || offload_csum)
Alexander Duyck7fbeffe2016-02-05 15:27:43 -080083 features |= NETIF_F_HW_CSUM;
84 }
85
Tom Herbert155e0102014-07-13 19:49:56 -070086 /* segment inner packet. */
Alexander Duyckbef3c6c2016-02-05 15:27:31 -080087 segs = gso_inner_segment(skb, features);
Himangi Saraogi27446442014-07-27 12:38:38 +053088 if (IS_ERR_OR_NULL(segs)) {
Tom Herbert155e0102014-07-13 19:49:56 -070089 skb_gso_error_unwind(skb, protocol, tnl_hlen, mac_offset,
90 mac_len);
91 goto out;
92 }
93
Steffen Klassert07b26c92016-09-19 12:58:47 +020094 gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
95
Tom Herbert155e0102014-07-13 19:49:56 -070096 outer_hlen = skb_tnl_header_len(skb);
97 udp_offset = outer_hlen - tnl_hlen;
98 skb = segs;
99 do {
Alexander Duyck802ab552016-04-10 21:45:03 -0400100 unsigned int len;
Tom Herbert155e0102014-07-13 19:49:56 -0700101
Alexander Duyckfdaefd62016-02-05 15:28:14 -0800102 if (remcsum)
Tom Herbert4bcb8772014-11-04 09:06:52 -0800103 skb->ip_summed = CHECKSUM_NONE;
Alexander Duyckfdaefd62016-02-05 15:28:14 -0800104
105 /* Set up inner headers if we are offloading inner checksum */
106 if (skb->ip_summed == CHECKSUM_PARTIAL) {
Tom Herbert4bcb8772014-11-04 09:06:52 -0800107 skb_reset_inner_headers(skb);
108 skb->encapsulation = 1;
109 }
Tom Herbert155e0102014-07-13 19:49:56 -0700110
111 skb->mac_len = mac_len;
Tom Herbert4bcb8772014-11-04 09:06:52 -0800112 skb->protocol = protocol;
Tom Herbert155e0102014-07-13 19:49:56 -0700113
Alexander Duyckdbef4912016-02-05 15:28:20 -0800114 __skb_push(skb, outer_hlen);
Tom Herbert155e0102014-07-13 19:49:56 -0700115 skb_reset_mac_header(skb);
116 skb_set_network_header(skb, mac_len);
117 skb_set_transport_header(skb, udp_offset);
Alexander Duyck802ab552016-04-10 21:45:03 -0400118 len = skb->len - udp_offset;
Tom Herbert155e0102014-07-13 19:49:56 -0700119 uh = udp_hdr(skb);
Alexander Duyck802ab552016-04-10 21:45:03 -0400120
121 /* If we are only performing partial GSO the inner header
122 * will be using a length value equal to only one MSS sized
123 * segment instead of the entire frame.
124 */
Alexey Kodanev3d0241d2017-10-06 19:02:35 +0300125 if (gso_partial && skb_is_gso(skb)) {
Alexander Duyck802ab552016-04-10 21:45:03 -0400126 uh->len = htons(skb_shinfo(skb)->gso_size +
127 SKB_GSO_CB(skb)->data_offset +
128 skb->head - (unsigned char *)uh);
129 } else {
130 uh->len = htons(len);
131 }
Tom Herbert155e0102014-07-13 19:49:56 -0700132
Tom Herbert4bcb8772014-11-04 09:06:52 -0800133 if (!need_csum)
134 continue;
Tom Herbert155e0102014-07-13 19:49:56 -0700135
Alexander Duyck802ab552016-04-10 21:45:03 -0400136 uh->check = ~csum_fold(csum_add(partial,
137 (__force __wsum)htonl(len)));
Alexander Duyckfdaefd62016-02-05 15:28:14 -0800138
139 if (skb->encapsulation || !offload_csum) {
140 uh->check = gso_make_checksum(skb, ~uh->check);
141 if (uh->check == 0)
142 uh->check = CSUM_MANGLED_0;
143 } else {
Tom Herbert4bcb8772014-11-04 09:06:52 -0800144 skb->ip_summed = CHECKSUM_PARTIAL;
145 skb->csum_start = skb_transport_header(skb) - skb->head;
146 skb->csum_offset = offsetof(struct udphdr, check);
Tom Herbert155e0102014-07-13 19:49:56 -0700147 }
Tom Herbert155e0102014-07-13 19:49:56 -0700148 } while ((skb = skb->next));
149out:
150 return segs;
151}
152
Tom Herbert8bce6d72014-09-29 20:22:29 -0700153struct sk_buff *skb_udp_tunnel_segment(struct sk_buff *skb,
154 netdev_features_t features,
155 bool is_ipv6)
156{
Eric Dumazet8d65cd82021-08-30 20:26:08 -0700157 const struct net_offload __rcu **offloads;
Tom Herbert8bce6d72014-09-29 20:22:29 -0700158 __be16 protocol = skb->protocol;
Tom Herbert8bce6d72014-09-29 20:22:29 -0700159 const struct net_offload *ops;
160 struct sk_buff *segs = ERR_PTR(-EINVAL);
161 struct sk_buff *(*gso_inner_segment)(struct sk_buff *skb,
162 netdev_features_t features);
163
164 rcu_read_lock();
165
166 switch (skb->inner_protocol_type) {
167 case ENCAP_TYPE_ETHER:
168 protocol = skb->inner_protocol;
169 gso_inner_segment = skb_mac_gso_segment;
170 break;
171 case ENCAP_TYPE_IPPROTO:
172 offloads = is_ipv6 ? inet6_offloads : inet_offloads;
173 ops = rcu_dereference(offloads[skb->inner_ipproto]);
174 if (!ops || !ops->callbacks.gso_segment)
175 goto out_unlock;
176 gso_inner_segment = ops->callbacks.gso_segment;
177 break;
178 default:
179 goto out_unlock;
180 }
181
182 segs = __skb_udp_tunnel_segment(skb, features, gso_inner_segment,
Tom Herbert4bcb8772014-11-04 09:06:52 -0800183 protocol, is_ipv6);
Tom Herbert8bce6d72014-09-29 20:22:29 -0700184
185out_unlock:
186 rcu_read_unlock();
187
188 return segs;
189}
Tom Herberta6024562016-04-05 08:22:51 -0700190EXPORT_SYMBOL(skb_udp_tunnel_segment);
Tom Herbert8bce6d72014-09-29 20:22:29 -0700191
Dongseok Yic3df39a2021-01-30 08:13:27 +0900192static void __udpv4_gso_segment_csum(struct sk_buff *seg,
193 __be32 *oldip, __be32 *newip,
194 __be16 *oldport, __be16 *newport)
195{
196 struct udphdr *uh;
197 struct iphdr *iph;
198
199 if (*oldip == *newip && *oldport == *newport)
200 return;
201
202 uh = udp_hdr(seg);
203 iph = ip_hdr(seg);
204
205 if (uh->check) {
206 inet_proto_csum_replace4(&uh->check, seg, *oldip, *newip,
207 true);
208 inet_proto_csum_replace2(&uh->check, seg, *oldport, *newport,
209 false);
210 if (!uh->check)
211 uh->check = CSUM_MANGLED_0;
212 }
213 *oldport = *newport;
214
215 csum_replace4(&iph->check, *oldip, *newip);
216 *oldip = *newip;
217}
218
219static struct sk_buff *__udpv4_gso_segment_list_csum(struct sk_buff *segs)
220{
221 struct sk_buff *seg;
222 struct udphdr *uh, *uh2;
223 struct iphdr *iph, *iph2;
224
225 seg = segs;
226 uh = udp_hdr(seg);
227 iph = ip_hdr(seg);
228
229 if ((udp_hdr(seg)->dest == udp_hdr(seg->next)->dest) &&
230 (udp_hdr(seg)->source == udp_hdr(seg->next)->source) &&
231 (ip_hdr(seg)->daddr == ip_hdr(seg->next)->daddr) &&
232 (ip_hdr(seg)->saddr == ip_hdr(seg->next)->saddr))
233 return segs;
234
235 while ((seg = seg->next)) {
236 uh2 = udp_hdr(seg);
237 iph2 = ip_hdr(seg);
238
239 __udpv4_gso_segment_csum(seg,
240 &iph2->saddr, &iph->saddr,
241 &uh2->source, &uh->source);
242 __udpv4_gso_segment_csum(seg,
243 &iph2->daddr, &iph->daddr,
244 &uh2->dest, &uh->dest);
245 }
246
247 return segs;
248}
249
Steffen Klassert9fd1ff52020-01-25 11:26:45 +0100250static struct sk_buff *__udp_gso_segment_list(struct sk_buff *skb,
Dongseok Yic3df39a2021-01-30 08:13:27 +0900251 netdev_features_t features,
252 bool is_ipv6)
Steffen Klassert9fd1ff52020-01-25 11:26:45 +0100253{
254 unsigned int mss = skb_shinfo(skb)->gso_size;
255
256 skb = skb_segment_list(skb, features, skb_mac_header_len(skb));
257 if (IS_ERR(skb))
258 return skb;
259
260 udp_hdr(skb)->len = htons(sizeof(struct udphdr) + mss);
261
Dongseok Yic3df39a2021-01-30 08:13:27 +0900262 return is_ipv6 ? skb : __udpv4_gso_segment_list_csum(skb);
Steffen Klassert9fd1ff52020-01-25 11:26:45 +0100263}
264
Willem de Bruijnee80d1e2018-04-26 13:42:16 -0400265struct sk_buff *__udp_gso_segment(struct sk_buff *gso_skb,
Dongseok Yic3df39a2021-01-30 08:13:27 +0900266 netdev_features_t features, bool is_ipv6)
Willem de Bruijnee80d1e2018-04-26 13:42:16 -0400267{
Willem de Bruijnad405852018-04-26 13:42:18 -0400268 struct sock *sk = gso_skb->sk;
269 unsigned int sum_truesize = 0;
Willem de Bruijnee80d1e2018-04-26 13:42:16 -0400270 struct sk_buff *segs, *seg;
Willem de Bruijnee80d1e2018-04-26 13:42:16 -0400271 struct udphdr *uh;
Alexander Duyckb21c0342018-05-07 11:08:28 -0700272 unsigned int mss;
Alexander Duyck04d55b22018-05-07 11:08:52 -0700273 bool copy_dtor;
Alexander Duyck9a0d41b2018-05-07 11:08:34 -0700274 __sum16 check;
275 __be16 newlen;
Willem de Bruijnee80d1e2018-04-26 13:42:16 -0400276
Alexander Duyckb21c0342018-05-07 11:08:28 -0700277 mss = skb_shinfo(gso_skb)->gso_size;
Willem de Bruijnee80d1e2018-04-26 13:42:16 -0400278 if (gso_skb->len <= sizeof(*uh) + mss)
279 return ERR_PTR(-EINVAL);
280
Yan Zhai98400362023-07-13 10:28:00 -0700281 if (skb_gso_ok(gso_skb, features | NETIF_F_GSO_ROBUST)) {
282 /* Packet is from an untrusted source, reset gso_segs. */
283 skb_shinfo(gso_skb)->gso_segs = DIV_ROUND_UP(gso_skb->len - sizeof(*uh),
284 mss);
285 return NULL;
286 }
287
288 if (skb_shinfo(gso_skb)->gso_type & SKB_GSO_FRAGLIST)
289 return __udp_gso_segment_list(gso_skb, features, is_ipv6);
290
Willem de Bruijnee80d1e2018-04-26 13:42:16 -0400291 skb_pull(gso_skb, sizeof(*uh));
292
Willem de Bruijnad405852018-04-26 13:42:18 -0400293 /* clear destructor to avoid skb_segment assigning it to tail */
Alexander Duyck04d55b22018-05-07 11:08:52 -0700294 copy_dtor = gso_skb->destructor == sock_wfree;
295 if (copy_dtor)
296 gso_skb->destructor = NULL;
Willem de Bruijnad405852018-04-26 13:42:18 -0400297
Willem de Bruijnee80d1e2018-04-26 13:42:16 -0400298 segs = skb_segment(gso_skb, features);
Enrico Weigelt88e235b2019-06-05 23:09:05 +0200299 if (IS_ERR_OR_NULL(segs)) {
Alexander Duyck04d55b22018-05-07 11:08:52 -0700300 if (copy_dtor)
301 gso_skb->destructor = sock_wfree;
Willem de Bruijnee80d1e2018-04-26 13:42:16 -0400302 return segs;
Willem de Bruijnad405852018-04-26 13:42:18 -0400303 }
Willem de Bruijnee80d1e2018-04-26 13:42:16 -0400304
Alexander Duyck6053d0f2018-05-07 11:08:46 -0700305 /* GSO partial and frag_list segmentation only requires splitting
306 * the frame into an MSS multiple and possibly a remainder, both
307 * cases return a GSO skb. So update the mss now.
308 */
309 if (skb_is_gso(segs))
310 mss *= skb_shinfo(segs)->gso_segs;
311
Alexander Duyck0ad65092018-05-07 11:08:40 -0700312 seg = segs;
313 uh = udp_hdr(seg);
Alexander Duyck9a0d41b2018-05-07 11:08:34 -0700314
Fred Klassen76e21532019-06-17 12:05:07 -0700315 /* preserve TX timestamp flags and TS key for first segment */
316 skb_shinfo(seg)->tskey = skb_shinfo(gso_skb)->tskey;
317 skb_shinfo(seg)->tx_flags |=
318 (skb_shinfo(gso_skb)->tx_flags & SKBTX_ANY_TSTAMP);
319
Alexander Duyck9a0d41b2018-05-07 11:08:34 -0700320 /* compute checksum adjustment based on old length versus new */
321 newlen = htons(sizeof(*uh) + mss);
322 check = csum16_add(csum16_sub(uh->check, uh->len), newlen);
323
Alexander Duyck0ad65092018-05-07 11:08:40 -0700324 for (;;) {
Alexander Duyck04d55b22018-05-07 11:08:52 -0700325 if (copy_dtor) {
326 seg->destructor = sock_wfree;
327 seg->sk = sk;
328 sum_truesize += seg->truesize;
329 }
Willem de Bruijnee80d1e2018-04-26 13:42:16 -0400330
Alexander Duyck0ad65092018-05-07 11:08:40 -0700331 if (!seg->next)
332 break;
Willem de Bruijnad405852018-04-26 13:42:18 -0400333
Alexander Duyck9a0d41b2018-05-07 11:08:34 -0700334 uh->len = newlen;
335 uh->check = check;
336
Alexander Duyck6053d0f2018-05-07 11:08:46 -0700337 if (seg->ip_summed == CHECKSUM_PARTIAL)
338 gso_reset_checksum(seg, ~check);
339 else
340 uh->check = gso_make_checksum(seg, ~check) ? :
341 CSUM_MANGLED_0;
342
Alexander Duyck0ad65092018-05-07 11:08:40 -0700343 seg = seg->next;
344 uh = udp_hdr(seg);
Willem de Bruijnee80d1e2018-04-26 13:42:16 -0400345 }
346
Alexander Duyck0ad65092018-05-07 11:08:40 -0700347 /* last packet can be partial gso_size, account for that in checksum */
348 newlen = htons(skb_tail_pointer(seg) - skb_transport_header(seg) +
349 seg->data_len);
350 check = csum16_add(csum16_sub(uh->check, uh->len), newlen);
351
352 uh->len = newlen;
353 uh->check = check;
354
Alexander Duyck6053d0f2018-05-07 11:08:46 -0700355 if (seg->ip_summed == CHECKSUM_PARTIAL)
356 gso_reset_checksum(seg, ~check);
357 else
358 uh->check = gso_make_checksum(seg, ~check) ? : CSUM_MANGLED_0;
359
Jakub Sitnicki10154db2024-06-26 19:51:26 +0200360 /* On the TX path, CHECKSUM_NONE and CHECKSUM_UNNECESSARY have the same
361 * meaning. However, check for bad offloads in the GSO stack expects the
362 * latter, if the checksum was calculated in software. To vouch for the
363 * segment skbs we actually need to set it on the gso_skb.
364 */
365 if (gso_skb->ip_summed == CHECKSUM_NONE)
366 gso_skb->ip_summed = CHECKSUM_UNNECESSARY;
367
Alexander Duyck0ad65092018-05-07 11:08:40 -0700368 /* update refcount for the packet */
Eric Dumazet575b65b2018-05-10 19:07:13 -0700369 if (copy_dtor) {
370 int delta = sum_truesize - gso_skb->truesize;
371
372 /* In some pathological cases, delta can be negative.
373 * We need to either use refcount_add() or refcount_sub_and_test()
374 */
375 if (likely(delta >= 0))
376 refcount_add(delta, &sk->sk_wmem_alloc);
377 else
378 WARN_ON_ONCE(refcount_sub_and_test(-delta, &sk->sk_wmem_alloc));
379 }
Willem de Bruijnee80d1e2018-04-26 13:42:16 -0400380 return segs;
381}
382EXPORT_SYMBOL_GPL(__udp_gso_segment);
383
Willem de Bruijn0c19f8462017-11-21 10:22:25 -0500384static struct sk_buff *udp4_ufo_fragment(struct sk_buff *skb,
385 netdev_features_t features)
Daniel Borkmannda5bab02013-06-08 12:56:03 +0200386{
387 struct sk_buff *segs = ERR_PTR(-EINVAL);
Willem de Bruijn0c19f8462017-11-21 10:22:25 -0500388 unsigned int mss;
389 __wsum csum;
390 struct udphdr *uh;
391 struct iphdr *iph;
Wei-Chun Chao7a7ffba2013-12-26 13:10:22 -0800392
393 if (skb->encapsulation &&
Tom Herbert0f4f4ff2014-06-04 17:20:16 -0700394 (skb_shinfo(skb)->gso_type &
Willem de Bruijn0c19f8462017-11-21 10:22:25 -0500395 (SKB_GSO_UDP_TUNNEL|SKB_GSO_UDP_TUNNEL_CSUM))) {
Tom Herbert8bce6d72014-09-29 20:22:29 -0700396 segs = skb_udp_tunnel_segment(skb, features, false);
Willem de Bruijn0c19f8462017-11-21 10:22:25 -0500397 goto out;
398 }
Daniel Borkmannda5bab02013-06-08 12:56:03 +0200399
Willem de Bruijnee80d1e2018-04-26 13:42:16 -0400400 if (!(skb_shinfo(skb)->gso_type & (SKB_GSO_UDP | SKB_GSO_UDP_L4)))
Willem de Bruijn121d57a2018-01-19 09:29:18 -0500401 goto out;
402
Willem de Bruijn0c19f8462017-11-21 10:22:25 -0500403 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
404 goto out;
405
Yan Zhai98400362023-07-13 10:28:00 -0700406 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4)
Dongseok Yic3df39a2021-01-30 08:13:27 +0900407 return __udp_gso_segment(skb, features, false);
Willem de Bruijnee80d1e2018-04-26 13:42:16 -0400408
Willem de Bruijn0c19f8462017-11-21 10:22:25 -0500409 mss = skb_shinfo(skb)->gso_size;
410 if (unlikely(skb->len <= mss))
411 goto out;
412
413 /* Do software UFO. Complete and fill in the UDP checksum as
414 * HW cannot do checksum of UDP packets sent as multiple
415 * IP fragments.
416 */
417
418 uh = udp_hdr(skb);
419 iph = ip_hdr(skb);
420
421 uh->check = 0;
422 csum = skb_checksum(skb, 0, skb->len, 0);
423 uh->check = udp_v4_check(skb->len, iph->saddr, iph->daddr, csum);
424 if (uh->check == 0)
425 uh->check = CSUM_MANGLED_0;
426
427 skb->ip_summed = CHECKSUM_UNNECESSARY;
428
429 /* If there is no outer header we can fake a checksum offload
430 * due to the fact that we have already done the checksum in
431 * software prior to segmenting the frame.
432 */
433 if (!skb->encap_hdr_csum)
434 features |= NETIF_F_HW_CSUM;
435
436 /* Fragment the skb. IP headers of the fragments are updated in
437 * inet_gso_segment()
438 */
439 segs = skb_segment(skb, features);
440out:
Daniel Borkmannda5bab02013-06-08 12:56:03 +0200441 return segs;
442}
443
Eric Dumazet0b935d72021-11-15 09:05:52 -0800444
Paolo Abenie20cf8d2018-11-07 12:38:29 +0100445#define UDP_GRO_CNT_MAX 64
446static struct sk_buff *udp_gro_receive_segment(struct list_head *head,
447 struct sk_buff *skb)
448{
Alexander Lobakin4b1a8622020-11-11 20:45:25 +0000449 struct udphdr *uh = udp_gro_udphdr(skb);
Paolo Abenie20cf8d2018-11-07 12:38:29 +0100450 struct sk_buff *pp = NULL;
451 struct udphdr *uh2;
452 struct sk_buff *p;
Eric Dumazet4dd2b822019-05-01 18:56:28 -0700453 unsigned int ulen;
Steffen Klassert9fd1ff52020-01-25 11:26:45 +0100454 int ret = 0;
Richard Gobert5babae72024-04-30 16:35:55 +0200455 int flush;
Paolo Abenie20cf8d2018-11-07 12:38:29 +0100456
457 /* requires non zero csum, for symmetry with GSO */
458 if (!uh->check) {
459 NAPI_GRO_CB(skb)->flush = 1;
460 return NULL;
461 }
462
Eric Dumazet4dd2b822019-05-01 18:56:28 -0700463 /* Do not deal with padded or malicious packets, sorry ! */
464 ulen = ntohs(uh->len);
465 if (ulen <= sizeof(*uh) || ulen != skb_gro_len(skb)) {
466 NAPI_GRO_CB(skb)->flush = 1;
467 return NULL;
468 }
Paolo Abenie20cf8d2018-11-07 12:38:29 +0100469 /* pull encapsulating udp header */
470 skb_gro_pull(skb, sizeof(struct udphdr));
Paolo Abenie20cf8d2018-11-07 12:38:29 +0100471
472 list_for_each_entry(p, head, list) {
473 if (!NAPI_GRO_CB(p)->same_flow)
474 continue;
475
476 uh2 = udp_hdr(p);
477
478 /* Match ports only, as csum is always non zero */
479 if ((*(u32 *)&uh->source != *(u32 *)&uh2->source)) {
480 NAPI_GRO_CB(p)->same_flow = 0;
481 continue;
482 }
483
Steffen Klassert9fd1ff52020-01-25 11:26:45 +0100484 if (NAPI_GRO_CB(skb)->is_flist != NAPI_GRO_CB(p)->is_flist) {
485 NAPI_GRO_CB(skb)->flush = 1;
486 return p;
487 }
488
Richard Gobert4b0ebbc2024-05-09 21:08:18 +0200489 flush = gro_receive_network_flush(uh, uh2, p);
Richard Gobert5babae72024-04-30 16:35:55 +0200490
Paolo Abenie20cf8d2018-11-07 12:38:29 +0100491 /* Terminate the flow on len mismatch or if it grow "too much".
492 * Under small packet flood GRO count could elsewhere grow a lot
Eric Dumazet4dd2b822019-05-01 18:56:28 -0700493 * leading to excessive truesize values.
Paolo Abeni21f1b8a2019-04-26 12:50:44 +0200494 * On len mismatch merge the first packet shorter than gso_size,
495 * otherwise complete the GRO packet.
Paolo Abenie20cf8d2018-11-07 12:38:29 +0100496 */
Richard Gobert5babae72024-04-30 16:35:55 +0200497 if (ulen > ntohs(uh2->len) || flush) {
Steffen Klassert9fd1ff52020-01-25 11:26:45 +0100498 pp = p;
499 } else {
500 if (NAPI_GRO_CB(skb)->is_flist) {
501 if (!pskb_may_pull(skb, skb_gro_offset(skb))) {
502 NAPI_GRO_CB(skb)->flush = 1;
503 return NULL;
504 }
505 if ((skb->ip_summed != p->ip_summed) ||
506 (skb->csum_level != p->csum_level)) {
507 NAPI_GRO_CB(skb)->flush = 1;
508 return NULL;
509 }
510 ret = skb_gro_receive_list(p, skb);
511 } else {
512 skb_gro_postpull_rcsum(skb, uh,
513 sizeof(struct udphdr));
514
515 ret = skb_gro_receive(p, skb);
516 }
517 }
518
519 if (ret || ulen != ntohs(uh2->len) ||
Paolo Abenie20cf8d2018-11-07 12:38:29 +0100520 NAPI_GRO_CB(p)->count >= UDP_GRO_CNT_MAX)
521 pp = p;
Paolo Abenie20cf8d2018-11-07 12:38:29 +0100522
523 return pp;
524 }
525
526 /* mismatch, but we never need to flush */
527 return NULL;
528}
529
David Millerd4546c22018-06-24 14:13:49 +0900530struct sk_buff *udp_gro_receive(struct list_head *head, struct sk_buff *skb,
Steffen Klassert9fd1ff52020-01-25 11:26:45 +0100531 struct udphdr *uh, struct sock *sk)
Or Gerlitzb582ef02014-01-20 13:59:19 +0200532{
David Millerd4546c22018-06-24 14:13:49 +0900533 struct sk_buff *pp = NULL;
534 struct sk_buff *p;
Tom Herbert57c67ff2014-08-22 13:34:44 -0700535 struct udphdr *uh2;
536 unsigned int off = skb_gro_offset(skb);
Or Gerlitzb582ef02014-01-20 13:59:19 +0200537 int flush = 1;
538
Antoine Tenart3d010c82024-03-26 12:33:58 +0100539 /* We can do L4 aggregation only if the packet can't land in a tunnel
540 * otherwise we could corrupt the inner stream. Detecting such packets
541 * cannot be foolproof and the aggregation might still happen in some
542 * cases. Such packets should be caught in udp_unexpected_gso later.
Paolo Abeni18f25dc2021-03-30 12:28:50 +0200543 */
Xin Longbde1b562020-03-30 23:31:45 +0800544 NAPI_GRO_CB(skb)->is_flist = 0;
Paolo Abeni18f25dc2021-03-30 12:28:50 +0200545 if (!sk || !udp_sk(sk)->gro_receive) {
Antoine Tenart64235ea2024-03-26 12:34:01 +0100546 /* If the packet was locally encapsulated in a UDP tunnel that
547 * wasn't detected above, do not GRO.
548 */
549 if (skb->encapsulation)
550 goto out;
551
Paolo Abeni18f25dc2021-03-30 12:28:50 +0200552 if (skb->dev->features & NETIF_F_GRO_FRAGLIST)
Eric Dumazete1dc06152023-09-12 09:17:24 +0000553 NAPI_GRO_CB(skb)->is_flist = sk ? !udp_test_bit(GRO_ENABLED, sk) : 1;
Paolo Abenie20cf8d2018-11-07 12:38:29 +0100554
Paolo Abeni18f25dc2021-03-30 12:28:50 +0200555 if ((!sk && (skb->dev->features & NETIF_F_GRO_UDP_FWD)) ||
Eric Dumazete1dc06152023-09-12 09:17:24 +0000556 (sk && udp_test_bit(GRO_ENABLED, sk)) || NAPI_GRO_CB(skb)->is_flist)
Paolo Abenib43c8902021-07-03 00:38:43 +0200557 return call_gro_receive(udp_gro_receive_segment, head, skb);
558
559 /* no GRO, be sure flush the current packet */
560 goto out;
Paolo Abenie20cf8d2018-11-07 12:38:29 +0100561 }
562
Paolo Abeni18f25dc2021-03-30 12:28:50 +0200563 if (NAPI_GRO_CB(skb)->encap_mark ||
Daniel Borkmann89e5c582021-02-26 22:22:48 +0100564 (uh->check && skb->ip_summed != CHECKSUM_PARTIAL &&
Tom Herbert662880f2014-08-27 21:26:56 -0700565 NAPI_GRO_CB(skb)->csum_cnt == 0 &&
Paolo Abeni18f25dc2021-03-30 12:28:50 +0200566 !NAPI_GRO_CB(skb)->csum_valid))
Steffen Klassert9fd1ff52020-01-25 11:26:45 +0100567 goto out;
Or Gerlitzb582ef02014-01-20 13:59:19 +0200568
Jesse Grossfac8e0f2016-03-19 09:32:01 -0700569 /* mark that this skb passed once through the tunnel gro layer */
570 NAPI_GRO_CB(skb)->encap_mark = 1;
Or Gerlitzb582ef02014-01-20 13:59:19 +0200571
Or Gerlitzb582ef02014-01-20 13:59:19 +0200572 flush = 0;
573
David Millerd4546c22018-06-24 14:13:49 +0900574 list_for_each_entry(p, head, list) {
Or Gerlitzb582ef02014-01-20 13:59:19 +0200575 if (!NAPI_GRO_CB(p)->same_flow)
576 continue;
577
578 uh2 = (struct udphdr *)(p->data + off);
Tom Herbert57c67ff2014-08-22 13:34:44 -0700579
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)) {
Or Gerlitzb582ef02014-01-20 13:59:19 +0200585 NAPI_GRO_CB(p)->same_flow = 0;
586 continue;
587 }
588 }
589
590 skb_gro_pull(skb, sizeof(struct udphdr)); /* pull encapsulating udp header */
Tom Herbert6bae1d42014-06-10 18:54:26 -0700591 skb_gro_postpull_rcsum(skb, uh, sizeof(struct udphdr));
Sabrina Dubrocafcd91dd2016-10-20 15:58:02 +0200592 pp = call_gro_receive_sk(udp_sk(sk)->gro_receive, sk, head, skb);
Or Gerlitzb582ef02014-01-20 13:59:19 +0200593
Steffen Klassert9fd1ff52020-01-25 11:26:45 +0100594out:
Sabrina Dubroca603d4cf2018-06-30 17:38:55 +0200595 skb_gro_flush_final(skb, pp, flush);
Or Gerlitzb582ef02014-01-20 13:59:19 +0200596 return pp;
597}
Tom Herberta6024562016-04-05 08:22:51 -0700598EXPORT_SYMBOL(udp_gro_receive);
Or Gerlitzb582ef02014-01-20 13:59:19 +0200599
Alexander Lobakin55e72982020-11-11 20:45:38 +0000600static 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);
Kuniyuki Iwashimaba6aac12022-11-14 13:57:56 -0800604 struct net *net = dev_net(skb->dev);
Richard Gobert7938cd12023-07-27 17:33:56 +0200605 int iif, sdif;
606
607 inet_get_iif_sdif(skb, &iif, &sdif);
Alexander Lobakin55e72982020-11-11 20:45:38 +0000608
Kuniyuki Iwashimaba6aac12022-11-14 13:57:56 -0800609 return __udp4_lib_lookup(net, iph->saddr, sport,
Richard Gobert7938cd12023-07-27 17:33:56 +0200610 iph->daddr, dport, iif,
611 sdif, net->ipv4.udp_table, NULL);
Alexander Lobakin55e72982020-11-11 20:45:38 +0000612}
613
Paolo Abeni028e0a42018-12-14 11:51:59 +0100614INDIRECT_CALLABLE_SCOPE
615struct sk_buff *udp4_gro_receive(struct list_head *head, struct sk_buff *skb)
Tom Herbert57c67ff2014-08-22 13:34:44 -0700616{
617 struct udphdr *uh = udp_gro_udphdr(skb);
Alexander Lobakin55e72982020-11-11 20:45:38 +0000618 struct sock *sk = NULL;
Steffen Klassert9fd1ff52020-01-25 11:26:45 +0100619 struct sk_buff *pp;
Tom Herbert57c67ff2014-08-22 13:34:44 -0700620
Steffen Klassert9fd1ff52020-01-25 11:26:45 +0100621 if (unlikely(!uh))
Tom Herbert2abb7cd2014-08-31 15:12:43 -0700622 goto flush;
Tom Herbert57c67ff2014-08-22 13:34:44 -0700623
Tom Herbert2abb7cd2014-08-31 15:12:43 -0700624 /* Don't bother verifying checksum if we're going to flush anyway. */
Scott Wood2d8f7e22014-09-10 21:23:18 -0500625 if (NAPI_GRO_CB(skb)->flush)
Tom Herbert2abb7cd2014-08-31 15:12:43 -0700626 goto skip;
627
628 if (skb_gro_checksum_validate_zero_check(skb, IPPROTO_UDP, uh->check,
629 inet_gro_compute_pseudo))
630 goto flush;
631 else if (uh->check)
Li RongQingb39c78b2020-01-03 11:51:00 +0800632 skb_gro_checksum_try_convert(skb, IPPROTO_UDP,
Tom Herbert2abb7cd2014-08-31 15:12:43 -0700633 inet_gro_compute_pseudo);
634skip:
Tom Herbertefc98d02014-10-03 15:48:08 -0700635 NAPI_GRO_CB(skb)->is_ipv6 = 0;
Alexander Lobakin55e72982020-11-11 20:45:38 +0000636
637 if (static_branch_unlikely(&udp_encap_needed_key))
638 sk = udp4_gro_lookup_skb(skb, uh->source, uh->dest);
639
Steffen Klassert9fd1ff52020-01-25 11:26:45 +0100640 pp = udp_gro_receive(head, skb, uh, sk);
Steffen Klassert9fd1ff52020-01-25 11:26:45 +0100641 return pp;
Tom Herbert2abb7cd2014-08-31 15:12:43 -0700642
643flush:
644 NAPI_GRO_CB(skb)->flush = 1;
645 return NULL;
Tom Herbert57c67ff2014-08-22 13:34:44 -0700646}
647
Paolo Abenie20cf8d2018-11-07 12:38:29 +0100648static int udp_gro_complete_segment(struct sk_buff *skb)
649{
650 struct udphdr *uh = udp_hdr(skb);
651
652 skb->csum_start = (unsigned char *)uh - skb->head;
653 skb->csum_offset = offsetof(struct udphdr, check);
654 skb->ip_summed = CHECKSUM_PARTIAL;
655
656 skb_shinfo(skb)->gso_segs = NAPI_GRO_CB(skb)->count;
657 skb_shinfo(skb)->gso_type |= SKB_GSO_UDP_L4;
Jakub Sitnickid51c5902021-07-29 15:48:20 +0200658
659 if (skb->encapsulation)
660 skb->inner_transport_header = skb->transport_header;
661
Paolo Abenie20cf8d2018-11-07 12:38:29 +0100662 return 0;
663}
664
Tom Herberta6024562016-04-05 08:22:51 -0700665int udp_gro_complete(struct sk_buff *skb, int nhoff,
666 udp_lookup_t lookup)
Or Gerlitzb582ef02014-01-20 13:59:19 +0200667{
Or Gerlitzb582ef02014-01-20 13:59:19 +0200668 __be16 newlen = htons(skb->len - nhoff);
669 struct udphdr *uh = (struct udphdr *)(skb->data + nhoff);
Tom Herberta6024562016-04-05 08:22:51 -0700670 struct sock *sk;
Menglong Dongcef21192020-11-09 21:57:58 -0500671 int err;
Or Gerlitzb582ef02014-01-20 13:59:19 +0200672
673 uh->len = newlen;
674
Paolo Abeni4f24ed72018-12-14 11:52:00 +0100675 sk = INDIRECT_CALL_INET(lookup, udp6_lib_lookup_skb,
676 udp4_lib_lookup_skb, skb, uh->source, uh->dest);
Steffen Klassert9fd1ff52020-01-25 11:26:45 +0100677 if (sk && udp_sk(sk)->gro_complete) {
Paolo Abenie20cf8d2018-11-07 12:38:29 +0100678 skb_shinfo(skb)->gso_type = uh->check ? SKB_GSO_UDP_TUNNEL_CSUM
679 : SKB_GSO_UDP_TUNNEL;
680
Paolo Abenie0e30702021-03-30 12:28:51 +0200681 /* clear the encap mark, so that inner frag_list gro_complete
682 * can take place
683 */
684 NAPI_GRO_CB(skb)->encap_mark = 0;
685
Paolo Abenie20cf8d2018-11-07 12:38:29 +0100686 /* Set encapsulation before calling into inner gro_complete()
687 * functions to make them set up the inner offsets.
688 */
689 skb->encapsulation = 1;
Tom Herberta6024562016-04-05 08:22:51 -0700690 err = udp_sk(sk)->gro_complete(sk, skb,
691 nhoff + sizeof(struct udphdr));
Steffen Klassert9fd1ff52020-01-25 11:26:45 +0100692 } else {
693 err = udp_gro_complete_segment(skb);
Paolo Abenie20cf8d2018-11-07 12:38:29 +0100694 }
Tom Herbert6db93ea2015-02-10 16:30:29 -0800695
696 if (skb->remcsum_offload)
697 skb_shinfo(skb)->gso_type |= SKB_GSO_TUNNEL_REMCSUM;
698
Or Gerlitzb582ef02014-01-20 13:59:19 +0200699 return err;
700}
Tom Herberta6024562016-04-05 08:22:51 -0700701EXPORT_SYMBOL(udp_gro_complete);
Or Gerlitzb582ef02014-01-20 13:59:19 +0200702
Paolo Abeni028e0a42018-12-14 11:51:59 +0100703INDIRECT_CALLABLE_SCOPE int udp4_gro_complete(struct sk_buff *skb, int nhoff)
Tom Herbert57c67ff2014-08-22 13:34:44 -0700704{
Richard Gobert5ef31ea2024-04-30 16:35:54 +0200705 const u16 offset = NAPI_GRO_CB(skb)->network_offsets[skb->encapsulation];
706 const struct iphdr *iph = (struct iphdr *)(skb->data + offset);
Tom Herbert57c67ff2014-08-22 13:34:44 -0700707 struct udphdr *uh = (struct udphdr *)(skb->data + nhoff);
708
Paolo Abenie0e30702021-03-30 12:28:51 +0200709 /* do fraglist only if there is no outer UDP encap (or we already processed it) */
710 if (NAPI_GRO_CB(skb)->is_flist && !NAPI_GRO_CB(skb)->encap_mark) {
Steffen Klassert9fd1ff52020-01-25 11:26:45 +0100711 uh->len = htons(skb->len - nhoff);
712
713 skb_shinfo(skb)->gso_type |= (SKB_GSO_FRAGLIST|SKB_GSO_UDP_L4);
714 skb_shinfo(skb)->gso_segs = NAPI_GRO_CB(skb)->count;
715
Antoine Tenartf0b8c302024-03-26 12:34:00 +0100716 __skb_incr_checksum_unnecessary(skb);
Steffen Klassert9fd1ff52020-01-25 11:26:45 +0100717
718 return 0;
719 }
720
Paolo Abenie20cf8d2018-11-07 12:38:29 +0100721 if (uh->check)
Tom Herbert57c67ff2014-08-22 13:34:44 -0700722 uh->check = ~udp_v4_check(skb->len - nhoff, iph->saddr,
723 iph->daddr, 0);
724
Tom Herberta6024562016-04-05 08:22:51 -0700725 return udp_gro_complete(skb, nhoff, udp4_lib_lookup_skb);
Tom Herbert57c67ff2014-08-22 13:34:44 -0700726}
727
Daniel Borkmannda5bab02013-06-08 12:56:03 +0200728int __init udpv4_offload_init(void)
729{
Eric Dumazet6a55ca62024-03-06 16:00:24 +0000730 net_hotdata.udpv4_offload = (struct net_offload) {
731 .callbacks = {
732 .gso_segment = udp4_ufo_fragment,
733 .gro_receive = udp4_gro_receive,
734 .gro_complete = udp4_gro_complete,
735 },
736 };
737 return inet_add_offload(&net_hotdata.udpv4_offload, IPPROTO_UDP);
Daniel Borkmannda5bab02013-06-08 12:56:03 +0200738}