blob: d842303587af93fc58454c0a2da1b6f70c55eec8 [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
Willem de Bruijn89add402024-07-29 16:10:12 -0400281 if (unlikely(skb_checksum_start(gso_skb) !=
Felix Fietkaub128ed52024-08-19 17:06:21 +0200282 skb_transport_header(gso_skb) &&
283 !(skb_shinfo(gso_skb)->gso_type & SKB_GSO_FRAGLIST)))
Willem de Bruijn89add402024-07-29 16:10:12 -0400284 return ERR_PTR(-EINVAL);
285
Jakub Sitnicki30b03f22024-08-08 11:56:22 +0200286 /* We don't know if egress device can segment and checksum the packet
287 * when IPv6 extension headers are present. Fall back to software GSO.
288 */
289 if (gso_skb->ip_summed != CHECKSUM_PARTIAL)
290 features &= ~(NETIF_F_GSO_UDP_L4 | NETIF_F_CSUM_MASK);
291
Yan Zhai98400362023-07-13 10:28:00 -0700292 if (skb_gso_ok(gso_skb, features | NETIF_F_GSO_ROBUST)) {
293 /* Packet is from an untrusted source, reset gso_segs. */
294 skb_shinfo(gso_skb)->gso_segs = DIV_ROUND_UP(gso_skb->len - sizeof(*uh),
295 mss);
296 return NULL;
297 }
298
299 if (skb_shinfo(gso_skb)->gso_type & SKB_GSO_FRAGLIST)
300 return __udp_gso_segment_list(gso_skb, features, is_ipv6);
301
Willem de Bruijnee80d1e2018-04-26 13:42:16 -0400302 skb_pull(gso_skb, sizeof(*uh));
303
Willem de Bruijnad405852018-04-26 13:42:18 -0400304 /* clear destructor to avoid skb_segment assigning it to tail */
Alexander Duyck04d55b22018-05-07 11:08:52 -0700305 copy_dtor = gso_skb->destructor == sock_wfree;
306 if (copy_dtor)
307 gso_skb->destructor = NULL;
Willem de Bruijnad405852018-04-26 13:42:18 -0400308
Willem de Bruijnee80d1e2018-04-26 13:42:16 -0400309 segs = skb_segment(gso_skb, features);
Enrico Weigelt88e235b2019-06-05 23:09:05 +0200310 if (IS_ERR_OR_NULL(segs)) {
Alexander Duyck04d55b22018-05-07 11:08:52 -0700311 if (copy_dtor)
312 gso_skb->destructor = sock_wfree;
Willem de Bruijnee80d1e2018-04-26 13:42:16 -0400313 return segs;
Willem de Bruijnad405852018-04-26 13:42:18 -0400314 }
Willem de Bruijnee80d1e2018-04-26 13:42:16 -0400315
Alexander Duyck6053d0f2018-05-07 11:08:46 -0700316 /* GSO partial and frag_list segmentation only requires splitting
317 * the frame into an MSS multiple and possibly a remainder, both
318 * cases return a GSO skb. So update the mss now.
319 */
320 if (skb_is_gso(segs))
321 mss *= skb_shinfo(segs)->gso_segs;
322
Alexander Duyck0ad65092018-05-07 11:08:40 -0700323 seg = segs;
324 uh = udp_hdr(seg);
Alexander Duyck9a0d41b2018-05-07 11:08:34 -0700325
Fred Klassen76e21532019-06-17 12:05:07 -0700326 /* preserve TX timestamp flags and TS key for first segment */
327 skb_shinfo(seg)->tskey = skb_shinfo(gso_skb)->tskey;
328 skb_shinfo(seg)->tx_flags |=
329 (skb_shinfo(gso_skb)->tx_flags & SKBTX_ANY_TSTAMP);
330
Alexander Duyck9a0d41b2018-05-07 11:08:34 -0700331 /* compute checksum adjustment based on old length versus new */
332 newlen = htons(sizeof(*uh) + mss);
333 check = csum16_add(csum16_sub(uh->check, uh->len), newlen);
334
Alexander Duyck0ad65092018-05-07 11:08:40 -0700335 for (;;) {
Alexander Duyck04d55b22018-05-07 11:08:52 -0700336 if (copy_dtor) {
337 seg->destructor = sock_wfree;
338 seg->sk = sk;
339 sum_truesize += seg->truesize;
340 }
Willem de Bruijnee80d1e2018-04-26 13:42:16 -0400341
Alexander Duyck0ad65092018-05-07 11:08:40 -0700342 if (!seg->next)
343 break;
Willem de Bruijnad405852018-04-26 13:42:18 -0400344
Alexander Duyck9a0d41b2018-05-07 11:08:34 -0700345 uh->len = newlen;
346 uh->check = check;
347
Alexander Duyck6053d0f2018-05-07 11:08:46 -0700348 if (seg->ip_summed == CHECKSUM_PARTIAL)
349 gso_reset_checksum(seg, ~check);
350 else
351 uh->check = gso_make_checksum(seg, ~check) ? :
352 CSUM_MANGLED_0;
353
Alexander Duyck0ad65092018-05-07 11:08:40 -0700354 seg = seg->next;
355 uh = udp_hdr(seg);
Willem de Bruijnee80d1e2018-04-26 13:42:16 -0400356 }
357
Alexander Duyck0ad65092018-05-07 11:08:40 -0700358 /* last packet can be partial gso_size, account for that in checksum */
359 newlen = htons(skb_tail_pointer(seg) - skb_transport_header(seg) +
360 seg->data_len);
361 check = csum16_add(csum16_sub(uh->check, uh->len), newlen);
362
363 uh->len = newlen;
364 uh->check = check;
365
Alexander Duyck6053d0f2018-05-07 11:08:46 -0700366 if (seg->ip_summed == CHECKSUM_PARTIAL)
367 gso_reset_checksum(seg, ~check);
368 else
369 uh->check = gso_make_checksum(seg, ~check) ? : CSUM_MANGLED_0;
370
Jakub Sitnicki10154db2024-06-26 19:51:26 +0200371 /* On the TX path, CHECKSUM_NONE and CHECKSUM_UNNECESSARY have the same
372 * meaning. However, check for bad offloads in the GSO stack expects the
373 * latter, if the checksum was calculated in software. To vouch for the
374 * segment skbs we actually need to set it on the gso_skb.
375 */
376 if (gso_skb->ip_summed == CHECKSUM_NONE)
377 gso_skb->ip_summed = CHECKSUM_UNNECESSARY;
378
Alexander Duyck0ad65092018-05-07 11:08:40 -0700379 /* update refcount for the packet */
Eric Dumazet575b65b2018-05-10 19:07:13 -0700380 if (copy_dtor) {
381 int delta = sum_truesize - gso_skb->truesize;
382
383 /* In some pathological cases, delta can be negative.
384 * We need to either use refcount_add() or refcount_sub_and_test()
385 */
386 if (likely(delta >= 0))
387 refcount_add(delta, &sk->sk_wmem_alloc);
388 else
389 WARN_ON_ONCE(refcount_sub_and_test(-delta, &sk->sk_wmem_alloc));
390 }
Willem de Bruijnee80d1e2018-04-26 13:42:16 -0400391 return segs;
392}
393EXPORT_SYMBOL_GPL(__udp_gso_segment);
394
Willem de Bruijn0c19f8462017-11-21 10:22:25 -0500395static struct sk_buff *udp4_ufo_fragment(struct sk_buff *skb,
396 netdev_features_t features)
Daniel Borkmannda5bab02013-06-08 12:56:03 +0200397{
398 struct sk_buff *segs = ERR_PTR(-EINVAL);
Willem de Bruijn0c19f8462017-11-21 10:22:25 -0500399 unsigned int mss;
400 __wsum csum;
401 struct udphdr *uh;
402 struct iphdr *iph;
Wei-Chun Chao7a7ffba2013-12-26 13:10:22 -0800403
404 if (skb->encapsulation &&
Tom Herbert0f4f4ff2014-06-04 17:20:16 -0700405 (skb_shinfo(skb)->gso_type &
Willem de Bruijn0c19f8462017-11-21 10:22:25 -0500406 (SKB_GSO_UDP_TUNNEL|SKB_GSO_UDP_TUNNEL_CSUM))) {
Tom Herbert8bce6d72014-09-29 20:22:29 -0700407 segs = skb_udp_tunnel_segment(skb, features, false);
Willem de Bruijn0c19f8462017-11-21 10:22:25 -0500408 goto out;
409 }
Daniel Borkmannda5bab02013-06-08 12:56:03 +0200410
Willem de Bruijnee80d1e2018-04-26 13:42:16 -0400411 if (!(skb_shinfo(skb)->gso_type & (SKB_GSO_UDP | SKB_GSO_UDP_L4)))
Willem de Bruijn121d57a2018-01-19 09:29:18 -0500412 goto out;
413
Willem de Bruijn0c19f8462017-11-21 10:22:25 -0500414 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
415 goto out;
416
Yan Zhai98400362023-07-13 10:28:00 -0700417 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4)
Dongseok Yic3df39a2021-01-30 08:13:27 +0900418 return __udp_gso_segment(skb, features, false);
Willem de Bruijnee80d1e2018-04-26 13:42:16 -0400419
Willem de Bruijn0c19f8462017-11-21 10:22:25 -0500420 mss = skb_shinfo(skb)->gso_size;
421 if (unlikely(skb->len <= mss))
422 goto out;
423
424 /* Do software UFO. Complete and fill in the UDP checksum as
425 * HW cannot do checksum of UDP packets sent as multiple
426 * IP fragments.
427 */
428
429 uh = udp_hdr(skb);
430 iph = ip_hdr(skb);
431
432 uh->check = 0;
433 csum = skb_checksum(skb, 0, skb->len, 0);
434 uh->check = udp_v4_check(skb->len, iph->saddr, iph->daddr, csum);
435 if (uh->check == 0)
436 uh->check = CSUM_MANGLED_0;
437
438 skb->ip_summed = CHECKSUM_UNNECESSARY;
439
440 /* If there is no outer header we can fake a checksum offload
441 * due to the fact that we have already done the checksum in
442 * software prior to segmenting the frame.
443 */
444 if (!skb->encap_hdr_csum)
445 features |= NETIF_F_HW_CSUM;
446
447 /* Fragment the skb. IP headers of the fragments are updated in
448 * inet_gso_segment()
449 */
450 segs = skb_segment(skb, features);
451out:
Daniel Borkmannda5bab02013-06-08 12:56:03 +0200452 return segs;
453}
454
Eric Dumazet0b935d72021-11-15 09:05:52 -0800455
Paolo Abenie20cf8d2018-11-07 12:38:29 +0100456#define UDP_GRO_CNT_MAX 64
457static struct sk_buff *udp_gro_receive_segment(struct list_head *head,
458 struct sk_buff *skb)
459{
Alexander Lobakin4b1a8622020-11-11 20:45:25 +0000460 struct udphdr *uh = udp_gro_udphdr(skb);
Paolo Abenie20cf8d2018-11-07 12:38:29 +0100461 struct sk_buff *pp = NULL;
462 struct udphdr *uh2;
463 struct sk_buff *p;
Eric Dumazet4dd2b822019-05-01 18:56:28 -0700464 unsigned int ulen;
Steffen Klassert9fd1ff52020-01-25 11:26:45 +0100465 int ret = 0;
Richard Gobert5babae72024-04-30 16:35:55 +0200466 int flush;
Paolo Abenie20cf8d2018-11-07 12:38:29 +0100467
468 /* requires non zero csum, for symmetry with GSO */
469 if (!uh->check) {
470 NAPI_GRO_CB(skb)->flush = 1;
471 return NULL;
472 }
473
Eric Dumazet4dd2b822019-05-01 18:56:28 -0700474 /* Do not deal with padded or malicious packets, sorry ! */
475 ulen = ntohs(uh->len);
476 if (ulen <= sizeof(*uh) || ulen != skb_gro_len(skb)) {
477 NAPI_GRO_CB(skb)->flush = 1;
478 return NULL;
479 }
Paolo Abenie20cf8d2018-11-07 12:38:29 +0100480 /* pull encapsulating udp header */
481 skb_gro_pull(skb, sizeof(struct udphdr));
Paolo Abenie20cf8d2018-11-07 12:38:29 +0100482
483 list_for_each_entry(p, head, list) {
484 if (!NAPI_GRO_CB(p)->same_flow)
485 continue;
486
487 uh2 = udp_hdr(p);
488
489 /* Match ports only, as csum is always non zero */
490 if ((*(u32 *)&uh->source != *(u32 *)&uh2->source)) {
491 NAPI_GRO_CB(p)->same_flow = 0;
492 continue;
493 }
494
Steffen Klassert9fd1ff52020-01-25 11:26:45 +0100495 if (NAPI_GRO_CB(skb)->is_flist != NAPI_GRO_CB(p)->is_flist) {
496 NAPI_GRO_CB(skb)->flush = 1;
497 return p;
498 }
499
Richard Gobert4b0ebbc2024-05-09 21:08:18 +0200500 flush = gro_receive_network_flush(uh, uh2, p);
Richard Gobert5babae72024-04-30 16:35:55 +0200501
Paolo Abenie20cf8d2018-11-07 12:38:29 +0100502 /* Terminate the flow on len mismatch or if it grow "too much".
503 * Under small packet flood GRO count could elsewhere grow a lot
Eric Dumazet4dd2b822019-05-01 18:56:28 -0700504 * leading to excessive truesize values.
Paolo Abeni21f1b8a2019-04-26 12:50:44 +0200505 * On len mismatch merge the first packet shorter than gso_size,
506 * otherwise complete the GRO packet.
Paolo Abenie20cf8d2018-11-07 12:38:29 +0100507 */
Richard Gobert5babae72024-04-30 16:35:55 +0200508 if (ulen > ntohs(uh2->len) || flush) {
Steffen Klassert9fd1ff52020-01-25 11:26:45 +0100509 pp = p;
510 } else {
511 if (NAPI_GRO_CB(skb)->is_flist) {
512 if (!pskb_may_pull(skb, skb_gro_offset(skb))) {
513 NAPI_GRO_CB(skb)->flush = 1;
514 return NULL;
515 }
516 if ((skb->ip_summed != p->ip_summed) ||
517 (skb->csum_level != p->csum_level)) {
518 NAPI_GRO_CB(skb)->flush = 1;
519 return NULL;
520 }
521 ret = skb_gro_receive_list(p, skb);
522 } else {
523 skb_gro_postpull_rcsum(skb, uh,
524 sizeof(struct udphdr));
525
526 ret = skb_gro_receive(p, skb);
527 }
528 }
529
530 if (ret || ulen != ntohs(uh2->len) ||
Paolo Abenie20cf8d2018-11-07 12:38:29 +0100531 NAPI_GRO_CB(p)->count >= UDP_GRO_CNT_MAX)
532 pp = p;
Paolo Abenie20cf8d2018-11-07 12:38:29 +0100533
534 return pp;
535 }
536
537 /* mismatch, but we never need to flush */
538 return NULL;
539}
540
David Millerd4546c22018-06-24 14:13:49 +0900541struct sk_buff *udp_gro_receive(struct list_head *head, struct sk_buff *skb,
Steffen Klassert9fd1ff52020-01-25 11:26:45 +0100542 struct udphdr *uh, struct sock *sk)
Or Gerlitzb582ef02014-01-20 13:59:19 +0200543{
David Millerd4546c22018-06-24 14:13:49 +0900544 struct sk_buff *pp = NULL;
545 struct sk_buff *p;
Tom Herbert57c67ff2014-08-22 13:34:44 -0700546 struct udphdr *uh2;
547 unsigned int off = skb_gro_offset(skb);
Or Gerlitzb582ef02014-01-20 13:59:19 +0200548 int flush = 1;
549
Antoine Tenart3d010c82024-03-26 12:33:58 +0100550 /* We can do L4 aggregation only if the packet can't land in a tunnel
551 * otherwise we could corrupt the inner stream. Detecting such packets
552 * cannot be foolproof and the aggregation might still happen in some
553 * cases. Such packets should be caught in udp_unexpected_gso later.
Paolo Abeni18f25dc2021-03-30 12:28:50 +0200554 */
Xin Longbde1b562020-03-30 23:31:45 +0800555 NAPI_GRO_CB(skb)->is_flist = 0;
Paolo Abeni18f25dc2021-03-30 12:28:50 +0200556 if (!sk || !udp_sk(sk)->gro_receive) {
Antoine Tenart64235ea2024-03-26 12:34:01 +0100557 /* If the packet was locally encapsulated in a UDP tunnel that
558 * wasn't detected above, do not GRO.
559 */
560 if (skb->encapsulation)
561 goto out;
562
Paolo Abeni18f25dc2021-03-30 12:28:50 +0200563 if (skb->dev->features & NETIF_F_GRO_FRAGLIST)
Eric Dumazete1dc06152023-09-12 09:17:24 +0000564 NAPI_GRO_CB(skb)->is_flist = sk ? !udp_test_bit(GRO_ENABLED, sk) : 1;
Paolo Abenie20cf8d2018-11-07 12:38:29 +0100565
Paolo Abeni18f25dc2021-03-30 12:28:50 +0200566 if ((!sk && (skb->dev->features & NETIF_F_GRO_UDP_FWD)) ||
Eric Dumazete1dc06152023-09-12 09:17:24 +0000567 (sk && udp_test_bit(GRO_ENABLED, sk)) || NAPI_GRO_CB(skb)->is_flist)
Paolo Abenib43c8902021-07-03 00:38:43 +0200568 return call_gro_receive(udp_gro_receive_segment, head, skb);
569
570 /* no GRO, be sure flush the current packet */
571 goto out;
Paolo Abenie20cf8d2018-11-07 12:38:29 +0100572 }
573
Paolo Abeni18f25dc2021-03-30 12:28:50 +0200574 if (NAPI_GRO_CB(skb)->encap_mark ||
Daniel Borkmann89e5c582021-02-26 22:22:48 +0100575 (uh->check && skb->ip_summed != CHECKSUM_PARTIAL &&
Tom Herbert662880f2014-08-27 21:26:56 -0700576 NAPI_GRO_CB(skb)->csum_cnt == 0 &&
Paolo Abeni18f25dc2021-03-30 12:28:50 +0200577 !NAPI_GRO_CB(skb)->csum_valid))
Steffen Klassert9fd1ff52020-01-25 11:26:45 +0100578 goto out;
Or Gerlitzb582ef02014-01-20 13:59:19 +0200579
Jesse Grossfac8e0f2016-03-19 09:32:01 -0700580 /* mark that this skb passed once through the tunnel gro layer */
581 NAPI_GRO_CB(skb)->encap_mark = 1;
Or Gerlitzb582ef02014-01-20 13:59:19 +0200582
Or Gerlitzb582ef02014-01-20 13:59:19 +0200583 flush = 0;
584
David Millerd4546c22018-06-24 14:13:49 +0900585 list_for_each_entry(p, head, list) {
Or Gerlitzb582ef02014-01-20 13:59:19 +0200586 if (!NAPI_GRO_CB(p)->same_flow)
587 continue;
588
589 uh2 = (struct udphdr *)(p->data + off);
Tom Herbert57c67ff2014-08-22 13:34:44 -0700590
591 /* Match ports and either checksums are either both zero
592 * or nonzero.
593 */
594 if ((*(u32 *)&uh->source != *(u32 *)&uh2->source) ||
595 (!uh->check ^ !uh2->check)) {
Or Gerlitzb582ef02014-01-20 13:59:19 +0200596 NAPI_GRO_CB(p)->same_flow = 0;
597 continue;
598 }
599 }
600
601 skb_gro_pull(skb, sizeof(struct udphdr)); /* pull encapsulating udp header */
Tom Herbert6bae1d42014-06-10 18:54:26 -0700602 skb_gro_postpull_rcsum(skb, uh, sizeof(struct udphdr));
Sabrina Dubrocafcd91dd2016-10-20 15:58:02 +0200603 pp = call_gro_receive_sk(udp_sk(sk)->gro_receive, sk, head, skb);
Or Gerlitzb582ef02014-01-20 13:59:19 +0200604
Steffen Klassert9fd1ff52020-01-25 11:26:45 +0100605out:
Sabrina Dubroca603d4cf2018-06-30 17:38:55 +0200606 skb_gro_flush_final(skb, pp, flush);
Or Gerlitzb582ef02014-01-20 13:59:19 +0200607 return pp;
608}
Tom Herberta6024562016-04-05 08:22:51 -0700609EXPORT_SYMBOL(udp_gro_receive);
Or Gerlitzb582ef02014-01-20 13:59:19 +0200610
Alexander Lobakin55e72982020-11-11 20:45:38 +0000611static struct sock *udp4_gro_lookup_skb(struct sk_buff *skb, __be16 sport,
612 __be16 dport)
613{
614 const struct iphdr *iph = skb_gro_network_header(skb);
Kuniyuki Iwashimaba6aac12022-11-14 13:57:56 -0800615 struct net *net = dev_net(skb->dev);
Richard Gobert7938cd12023-07-27 17:33:56 +0200616 int iif, sdif;
617
618 inet_get_iif_sdif(skb, &iif, &sdif);
Alexander Lobakin55e72982020-11-11 20:45:38 +0000619
Kuniyuki Iwashimaba6aac12022-11-14 13:57:56 -0800620 return __udp4_lib_lookup(net, iph->saddr, sport,
Richard Gobert7938cd12023-07-27 17:33:56 +0200621 iph->daddr, dport, iif,
622 sdif, net->ipv4.udp_table, NULL);
Alexander Lobakin55e72982020-11-11 20:45:38 +0000623}
624
Paolo Abeni028e0a42018-12-14 11:51:59 +0100625INDIRECT_CALLABLE_SCOPE
626struct sk_buff *udp4_gro_receive(struct list_head *head, struct sk_buff *skb)
Tom Herbert57c67ff2014-08-22 13:34:44 -0700627{
628 struct udphdr *uh = udp_gro_udphdr(skb);
Alexander Lobakin55e72982020-11-11 20:45:38 +0000629 struct sock *sk = NULL;
Steffen Klassert9fd1ff52020-01-25 11:26:45 +0100630 struct sk_buff *pp;
Tom Herbert57c67ff2014-08-22 13:34:44 -0700631
Steffen Klassert9fd1ff52020-01-25 11:26:45 +0100632 if (unlikely(!uh))
Tom Herbert2abb7cd2014-08-31 15:12:43 -0700633 goto flush;
Tom Herbert57c67ff2014-08-22 13:34:44 -0700634
Tom Herbert2abb7cd2014-08-31 15:12:43 -0700635 /* Don't bother verifying checksum if we're going to flush anyway. */
Scott Wood2d8f7e22014-09-10 21:23:18 -0500636 if (NAPI_GRO_CB(skb)->flush)
Tom Herbert2abb7cd2014-08-31 15:12:43 -0700637 goto skip;
638
639 if (skb_gro_checksum_validate_zero_check(skb, IPPROTO_UDP, uh->check,
640 inet_gro_compute_pseudo))
641 goto flush;
642 else if (uh->check)
Li RongQingb39c78b2020-01-03 11:51:00 +0800643 skb_gro_checksum_try_convert(skb, IPPROTO_UDP,
Tom Herbert2abb7cd2014-08-31 15:12:43 -0700644 inet_gro_compute_pseudo);
645skip:
Tom Herbertefc98d02014-10-03 15:48:08 -0700646 NAPI_GRO_CB(skb)->is_ipv6 = 0;
Alexander Lobakin55e72982020-11-11 20:45:38 +0000647
648 if (static_branch_unlikely(&udp_encap_needed_key))
649 sk = udp4_gro_lookup_skb(skb, uh->source, uh->dest);
650
Steffen Klassert9fd1ff52020-01-25 11:26:45 +0100651 pp = udp_gro_receive(head, skb, uh, sk);
Steffen Klassert9fd1ff52020-01-25 11:26:45 +0100652 return pp;
Tom Herbert2abb7cd2014-08-31 15:12:43 -0700653
654flush:
655 NAPI_GRO_CB(skb)->flush = 1;
656 return NULL;
Tom Herbert57c67ff2014-08-22 13:34:44 -0700657}
658
Paolo Abenie20cf8d2018-11-07 12:38:29 +0100659static int udp_gro_complete_segment(struct sk_buff *skb)
660{
661 struct udphdr *uh = udp_hdr(skb);
662
663 skb->csum_start = (unsigned char *)uh - skb->head;
664 skb->csum_offset = offsetof(struct udphdr, check);
665 skb->ip_summed = CHECKSUM_PARTIAL;
666
667 skb_shinfo(skb)->gso_segs = NAPI_GRO_CB(skb)->count;
668 skb_shinfo(skb)->gso_type |= SKB_GSO_UDP_L4;
Jakub Sitnickid51c5902021-07-29 15:48:20 +0200669
670 if (skb->encapsulation)
671 skb->inner_transport_header = skb->transport_header;
672
Paolo Abenie20cf8d2018-11-07 12:38:29 +0100673 return 0;
674}
675
Tom Herberta6024562016-04-05 08:22:51 -0700676int udp_gro_complete(struct sk_buff *skb, int nhoff,
677 udp_lookup_t lookup)
Or Gerlitzb582ef02014-01-20 13:59:19 +0200678{
Or Gerlitzb582ef02014-01-20 13:59:19 +0200679 __be16 newlen = htons(skb->len - nhoff);
680 struct udphdr *uh = (struct udphdr *)(skb->data + nhoff);
Tom Herberta6024562016-04-05 08:22:51 -0700681 struct sock *sk;
Menglong Dongcef21192020-11-09 21:57:58 -0500682 int err;
Or Gerlitzb582ef02014-01-20 13:59:19 +0200683
684 uh->len = newlen;
685
Paolo Abeni4f24ed72018-12-14 11:52:00 +0100686 sk = INDIRECT_CALL_INET(lookup, udp6_lib_lookup_skb,
687 udp4_lib_lookup_skb, skb, uh->source, uh->dest);
Steffen Klassert9fd1ff52020-01-25 11:26:45 +0100688 if (sk && udp_sk(sk)->gro_complete) {
Paolo Abenie20cf8d2018-11-07 12:38:29 +0100689 skb_shinfo(skb)->gso_type = uh->check ? SKB_GSO_UDP_TUNNEL_CSUM
690 : SKB_GSO_UDP_TUNNEL;
691
Paolo Abenie0e30702021-03-30 12:28:51 +0200692 /* clear the encap mark, so that inner frag_list gro_complete
693 * can take place
694 */
695 NAPI_GRO_CB(skb)->encap_mark = 0;
696
Paolo Abenie20cf8d2018-11-07 12:38:29 +0100697 /* Set encapsulation before calling into inner gro_complete()
698 * functions to make them set up the inner offsets.
699 */
700 skb->encapsulation = 1;
Tom Herberta6024562016-04-05 08:22:51 -0700701 err = udp_sk(sk)->gro_complete(sk, skb,
702 nhoff + sizeof(struct udphdr));
Steffen Klassert9fd1ff52020-01-25 11:26:45 +0100703 } else {
704 err = udp_gro_complete_segment(skb);
Paolo Abenie20cf8d2018-11-07 12:38:29 +0100705 }
Tom Herbert6db93ea2015-02-10 16:30:29 -0800706
707 if (skb->remcsum_offload)
708 skb_shinfo(skb)->gso_type |= SKB_GSO_TUNNEL_REMCSUM;
709
Or Gerlitzb582ef02014-01-20 13:59:19 +0200710 return err;
711}
Tom Herberta6024562016-04-05 08:22:51 -0700712EXPORT_SYMBOL(udp_gro_complete);
Or Gerlitzb582ef02014-01-20 13:59:19 +0200713
Paolo Abeni028e0a42018-12-14 11:51:59 +0100714INDIRECT_CALLABLE_SCOPE int udp4_gro_complete(struct sk_buff *skb, int nhoff)
Tom Herbert57c67ff2014-08-22 13:34:44 -0700715{
Richard Gobert5ef31ea2024-04-30 16:35:54 +0200716 const u16 offset = NAPI_GRO_CB(skb)->network_offsets[skb->encapsulation];
717 const struct iphdr *iph = (struct iphdr *)(skb->data + offset);
Tom Herbert57c67ff2014-08-22 13:34:44 -0700718 struct udphdr *uh = (struct udphdr *)(skb->data + nhoff);
719
Paolo Abenie0e30702021-03-30 12:28:51 +0200720 /* do fraglist only if there is no outer UDP encap (or we already processed it) */
721 if (NAPI_GRO_CB(skb)->is_flist && !NAPI_GRO_CB(skb)->encap_mark) {
Steffen Klassert9fd1ff52020-01-25 11:26:45 +0100722 uh->len = htons(skb->len - nhoff);
723
724 skb_shinfo(skb)->gso_type |= (SKB_GSO_FRAGLIST|SKB_GSO_UDP_L4);
725 skb_shinfo(skb)->gso_segs = NAPI_GRO_CB(skb)->count;
726
Antoine Tenartf0b8c302024-03-26 12:34:00 +0100727 __skb_incr_checksum_unnecessary(skb);
Steffen Klassert9fd1ff52020-01-25 11:26:45 +0100728
729 return 0;
730 }
731
Paolo Abenie20cf8d2018-11-07 12:38:29 +0100732 if (uh->check)
Tom Herbert57c67ff2014-08-22 13:34:44 -0700733 uh->check = ~udp_v4_check(skb->len - nhoff, iph->saddr,
734 iph->daddr, 0);
735
Tom Herberta6024562016-04-05 08:22:51 -0700736 return udp_gro_complete(skb, nhoff, udp4_lib_lookup_skb);
Tom Herbert57c67ff2014-08-22 13:34:44 -0700737}
738
Daniel Borkmannda5bab02013-06-08 12:56:03 +0200739int __init udpv4_offload_init(void)
740{
Eric Dumazet6a55ca62024-03-06 16:00:24 +0000741 net_hotdata.udpv4_offload = (struct net_offload) {
742 .callbacks = {
743 .gso_segment = udp4_ufo_fragment,
744 .gro_receive = udp4_gro_receive,
745 .gro_complete = udp4_gro_complete,
746 },
747 };
748 return inet_add_offload(&net_hotdata.udpv4_offload, IPPROTO_UDP);
Daniel Borkmannda5bab02013-06-08 12:56:03 +0200749}