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/gso.h>
12 #include <net/udp.h>
13 #include <net/protocol.h>
14 #include <net/inet_common.h>
15
__skb_udp_tunnel_segment(struct sk_buff * skb,netdev_features_t features,struct sk_buff * (* gso_inner_segment)(struct sk_buff * skb,netdev_features_t features),__be16 new_protocol,bool is_ipv6)16 static 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),
20 __be16 new_protocol, bool is_ipv6)
21 {
22 int tnl_hlen = skb_inner_mac_header(skb) - skb_transport_header(skb);
23 bool remcsum, need_csum, offload_csum, gso_partial;
24 struct sk_buff *segs = ERR_PTR(-EINVAL);
25 struct udphdr *uh = udp_hdr(skb);
26 u16 mac_offset = skb->mac_header;
27 __be16 protocol = skb->protocol;
28 u16 mac_len = skb->mac_len;
29 int udp_offset, outer_hlen;
30 __wsum partial;
31 bool need_ipsec;
32
33 if (unlikely(!pskb_may_pull(skb, tnl_hlen)))
34 goto out;
35
36 /* Adjust partial header checksum to negate old length.
37 * We cannot rely on the value contained in uh->len as it is
38 * possible that the actual value exceeds the boundaries of the
39 * 16 bit length field due to the header being added outside of an
40 * IP or IPv6 frame that was already limited to 64K - 1.
41 */
42 if (skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)
43 partial = (__force __wsum)uh->len;
44 else
45 partial = (__force __wsum)htonl(skb->len);
46 partial = csum_sub(csum_unfold(uh->check), partial);
47
48 /* setup inner skb. */
49 skb->encapsulation = 0;
50 SKB_GSO_CB(skb)->encap_level = 0;
51 __skb_pull(skb, tnl_hlen);
52 skb_reset_mac_header(skb);
53 skb_set_network_header(skb, skb_inner_network_offset(skb));
54 skb_set_transport_header(skb, skb_inner_transport_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 if (need_csum)
74 features &= ~NETIF_F_SCTP_CRC;
75
76 /* 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 */
80 if (remcsum) {
81 features &= ~NETIF_F_CSUM_MASK;
82 if (!need_csum || offload_csum)
83 features |= NETIF_F_HW_CSUM;
84 }
85
86 /* segment inner packet. */
87 segs = gso_inner_segment(skb, features);
88 if (IS_ERR_OR_NULL(segs)) {
89 skb_gso_error_unwind(skb, protocol, tnl_hlen, mac_offset,
90 mac_len);
91 goto out;
92 }
93
94 gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
95
96 outer_hlen = skb_tnl_header_len(skb);
97 udp_offset = outer_hlen - tnl_hlen;
98 skb = segs;
99 do {
100 unsigned int len;
101
102 if (remcsum)
103 skb->ip_summed = CHECKSUM_NONE;
104
105 /* Set up inner headers if we are offloading inner checksum */
106 if (skb->ip_summed == CHECKSUM_PARTIAL) {
107 skb_reset_inner_headers(skb);
108 skb->encapsulation = 1;
109 }
110
111 skb->mac_len = mac_len;
112 skb->protocol = protocol;
113
114 __skb_push(skb, outer_hlen);
115 skb_reset_mac_header(skb);
116 skb_set_network_header(skb, mac_len);
117 skb_set_transport_header(skb, udp_offset);
118 len = skb->len - udp_offset;
119 uh = udp_hdr(skb);
120
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 */
125 if (gso_partial && skb_is_gso(skb)) {
126 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 }
132
133 if (!need_csum)
134 continue;
135
136 uh->check = ~csum_fold(csum_add(partial,
137 (__force __wsum)htonl(len)));
138
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 {
144 skb->ip_summed = CHECKSUM_PARTIAL;
145 skb->csum_start = skb_transport_header(skb) - skb->head;
146 skb->csum_offset = offsetof(struct udphdr, check);
147 }
148 } while ((skb = skb->next));
149 out:
150 return segs;
151 }
152
skb_udp_tunnel_segment(struct sk_buff * skb,netdev_features_t features,bool is_ipv6)153 struct sk_buff *skb_udp_tunnel_segment(struct sk_buff *skb,
154 netdev_features_t features,
155 bool is_ipv6)
156 {
157 const struct net_offload __rcu **offloads;
158 __be16 protocol = skb->protocol;
159 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,
183 protocol, is_ipv6);
184
185 out_unlock:
186 rcu_read_unlock();
187
188 return segs;
189 }
190 EXPORT_SYMBOL(skb_udp_tunnel_segment);
191
__udpv4_gso_segment_csum(struct sk_buff * seg,__be32 * oldip,__be32 * newip,__be16 * oldport,__be16 * newport)192 static 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
__udpv4_gso_segment_list_csum(struct sk_buff * segs)219 static 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
__udp_gso_segment_list(struct sk_buff * skb,netdev_features_t features,bool is_ipv6)250 static struct sk_buff *__udp_gso_segment_list(struct sk_buff *skb,
251 netdev_features_t features,
252 bool is_ipv6)
253 {
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
262 return is_ipv6 ? skb : __udpv4_gso_segment_list_csum(skb);
263 }
264
__udp_gso_segment(struct sk_buff * gso_skb,netdev_features_t features,bool is_ipv6)265 struct sk_buff *__udp_gso_segment(struct sk_buff *gso_skb,
266 netdev_features_t features, bool is_ipv6)
267 {
268 struct sock *sk = gso_skb->sk;
269 unsigned int sum_truesize = 0;
270 struct sk_buff *segs, *seg;
271 struct udphdr *uh;
272 unsigned int mss;
273 bool copy_dtor;
274 __sum16 check;
275 __be16 newlen;
276
277 mss = skb_shinfo(gso_skb)->gso_size;
278 if (gso_skb->len <= sizeof(*uh) + mss)
279 return ERR_PTR(-EINVAL);
280
281 if (unlikely(skb_checksum_start(gso_skb) !=
282 skb_transport_header(gso_skb) &&
283 !(skb_shinfo(gso_skb)->gso_type & SKB_GSO_FRAGLIST)))
284 return ERR_PTR(-EINVAL);
285
286 if (skb_gso_ok(gso_skb, features | NETIF_F_GSO_ROBUST)) {
287 /* Packet is from an untrusted source, reset gso_segs. */
288 skb_shinfo(gso_skb)->gso_segs = DIV_ROUND_UP(gso_skb->len - sizeof(*uh),
289 mss);
290 return NULL;
291 }
292
293 if (skb_shinfo(gso_skb)->gso_type & SKB_GSO_FRAGLIST)
294 return __udp_gso_segment_list(gso_skb, features, is_ipv6);
295
296 skb_pull(gso_skb, sizeof(*uh));
297
298 /* clear destructor to avoid skb_segment assigning it to tail */
299 copy_dtor = gso_skb->destructor == sock_wfree;
300 if (copy_dtor)
301 gso_skb->destructor = NULL;
302
303 segs = skb_segment(gso_skb, features);
304 if (IS_ERR_OR_NULL(segs)) {
305 if (copy_dtor)
306 gso_skb->destructor = sock_wfree;
307 return segs;
308 }
309
310 /* GSO partial and frag_list segmentation only requires splitting
311 * the frame into an MSS multiple and possibly a remainder, both
312 * cases return a GSO skb. So update the mss now.
313 */
314 if (skb_is_gso(segs))
315 mss *= skb_shinfo(segs)->gso_segs;
316
317 seg = segs;
318 uh = udp_hdr(seg);
319
320 /* preserve TX timestamp flags and TS key for first segment */
321 skb_shinfo(seg)->tskey = skb_shinfo(gso_skb)->tskey;
322 skb_shinfo(seg)->tx_flags |=
323 (skb_shinfo(gso_skb)->tx_flags & SKBTX_ANY_TSTAMP);
324
325 /* compute checksum adjustment based on old length versus new */
326 newlen = htons(sizeof(*uh) + mss);
327 check = csum16_add(csum16_sub(uh->check, uh->len), newlen);
328
329 for (;;) {
330 if (copy_dtor) {
331 seg->destructor = sock_wfree;
332 seg->sk = sk;
333 sum_truesize += seg->truesize;
334 }
335
336 if (!seg->next)
337 break;
338
339 uh->len = newlen;
340 uh->check = check;
341
342 if (seg->ip_summed == CHECKSUM_PARTIAL)
343 gso_reset_checksum(seg, ~check);
344 else
345 uh->check = gso_make_checksum(seg, ~check) ? :
346 CSUM_MANGLED_0;
347
348 seg = seg->next;
349 uh = udp_hdr(seg);
350 }
351
352 /* last packet can be partial gso_size, account for that in checksum */
353 newlen = htons(skb_tail_pointer(seg) - skb_transport_header(seg) +
354 seg->data_len);
355 check = csum16_add(csum16_sub(uh->check, uh->len), newlen);
356
357 uh->len = newlen;
358 uh->check = check;
359
360 if (seg->ip_summed == CHECKSUM_PARTIAL)
361 gso_reset_checksum(seg, ~check);
362 else
363 uh->check = gso_make_checksum(seg, ~check) ? : CSUM_MANGLED_0;
364
365 /* update refcount for the packet */
366 if (copy_dtor) {
367 int delta = sum_truesize - gso_skb->truesize;
368
369 /* In some pathological cases, delta can be negative.
370 * We need to either use refcount_add() or refcount_sub_and_test()
371 */
372 if (likely(delta >= 0))
373 refcount_add(delta, &sk->sk_wmem_alloc);
374 else
375 WARN_ON_ONCE(refcount_sub_and_test(-delta, &sk->sk_wmem_alloc));
376 }
377 return segs;
378 }
379 EXPORT_SYMBOL_GPL(__udp_gso_segment);
380
udp4_ufo_fragment(struct sk_buff * skb,netdev_features_t features)381 static struct sk_buff *udp4_ufo_fragment(struct sk_buff *skb,
382 netdev_features_t features)
383 {
384 struct sk_buff *segs = ERR_PTR(-EINVAL);
385 unsigned int mss;
386 __wsum csum;
387 struct udphdr *uh;
388 struct iphdr *iph;
389
390 if (skb->encapsulation &&
391 (skb_shinfo(skb)->gso_type &
392 (SKB_GSO_UDP_TUNNEL|SKB_GSO_UDP_TUNNEL_CSUM))) {
393 segs = skb_udp_tunnel_segment(skb, features, false);
394 goto out;
395 }
396
397 if (!(skb_shinfo(skb)->gso_type & (SKB_GSO_UDP | SKB_GSO_UDP_L4)))
398 goto out;
399
400 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
401 goto out;
402
403 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4)
404 return __udp_gso_segment(skb, features, false);
405
406 mss = skb_shinfo(skb)->gso_size;
407 if (unlikely(skb->len <= mss))
408 goto out;
409
410 /* Do software UFO. Complete and fill in the UDP checksum as
411 * HW cannot do checksum of UDP packets sent as multiple
412 * IP fragments.
413 */
414
415 uh = udp_hdr(skb);
416 iph = ip_hdr(skb);
417
418 uh->check = 0;
419 csum = skb_checksum(skb, 0, skb->len, 0);
420 uh->check = udp_v4_check(skb->len, iph->saddr, iph->daddr, csum);
421 if (uh->check == 0)
422 uh->check = CSUM_MANGLED_0;
423
424 skb->ip_summed = CHECKSUM_UNNECESSARY;
425
426 /* If there is no outer header we can fake a checksum offload
427 * due to the fact that we have already done the checksum in
428 * software prior to segmenting the frame.
429 */
430 if (!skb->encap_hdr_csum)
431 features |= NETIF_F_HW_CSUM;
432
433 /* Fragment the skb. IP headers of the fragments are updated in
434 * inet_gso_segment()
435 */
436 segs = skb_segment(skb, features);
437 out:
438 return segs;
439 }
440
skb_gro_receive_list(struct sk_buff * p,struct sk_buff * skb)441 static int skb_gro_receive_list(struct sk_buff *p, struct sk_buff *skb)
442 {
443 if (unlikely(p->len + skb->len >= 65536))
444 return -E2BIG;
445
446 if (NAPI_GRO_CB(p)->last == p)
447 skb_shinfo(p)->frag_list = skb;
448 else
449 NAPI_GRO_CB(p)->last->next = skb;
450
451 skb_pull(skb, skb_gro_offset(skb));
452
453 NAPI_GRO_CB(p)->last = skb;
454 NAPI_GRO_CB(p)->count++;
455 p->data_len += skb->len;
456
457 /* sk ownership - if any - completely transferred to the aggregated packet */
458 skb->destructor = NULL;
459 skb->sk = NULL;
460 p->truesize += skb->truesize;
461 p->len += skb->len;
462
463 NAPI_GRO_CB(skb)->same_flow = 1;
464
465 return 0;
466 }
467
468
469 #define UDP_GRO_CNT_MAX 64
udp_gro_receive_segment(struct list_head * head,struct sk_buff * skb)470 static struct sk_buff *udp_gro_receive_segment(struct list_head *head,
471 struct sk_buff *skb)
472 {
473 struct udphdr *uh = udp_gro_udphdr(skb);
474 struct sk_buff *pp = NULL;
475 struct udphdr *uh2;
476 struct sk_buff *p;
477 unsigned int ulen;
478 int ret = 0;
479 int flush;
480
481 /* requires non zero csum, for symmetry with GSO */
482 if (!uh->check) {
483 NAPI_GRO_CB(skb)->flush = 1;
484 return NULL;
485 }
486
487 /* Do not deal with padded or malicious packets, sorry ! */
488 ulen = ntohs(uh->len);
489 if (ulen <= sizeof(*uh) || ulen != skb_gro_len(skb)) {
490 NAPI_GRO_CB(skb)->flush = 1;
491 return NULL;
492 }
493 /* pull encapsulating udp header */
494 skb_gro_pull(skb, sizeof(struct udphdr));
495
496 list_for_each_entry(p, head, list) {
497 if (!NAPI_GRO_CB(p)->same_flow)
498 continue;
499
500 uh2 = udp_hdr(p);
501
502 /* Match ports only, as csum is always non zero */
503 if ((*(u32 *)&uh->source != *(u32 *)&uh2->source)) {
504 NAPI_GRO_CB(p)->same_flow = 0;
505 continue;
506 }
507
508 if (NAPI_GRO_CB(skb)->is_flist != NAPI_GRO_CB(p)->is_flist) {
509 NAPI_GRO_CB(skb)->flush = 1;
510 return p;
511 }
512
513 flush = NAPI_GRO_CB(p)->flush;
514
515 if (NAPI_GRO_CB(p)->flush_id != 1 ||
516 NAPI_GRO_CB(p)->count != 1 ||
517 !NAPI_GRO_CB(p)->is_atomic)
518 flush |= NAPI_GRO_CB(p)->flush_id;
519 else
520 NAPI_GRO_CB(p)->is_atomic = false;
521
522 /* Terminate the flow on len mismatch or if it grow "too much".
523 * Under small packet flood GRO count could elsewhere grow a lot
524 * leading to excessive truesize values.
525 * On len mismatch merge the first packet shorter than gso_size,
526 * otherwise complete the GRO packet.
527 */
528 if (ulen > ntohs(uh2->len) || flush) {
529 pp = p;
530 } else {
531 if (NAPI_GRO_CB(skb)->is_flist) {
532 if (!pskb_may_pull(skb, skb_gro_offset(skb))) {
533 NAPI_GRO_CB(skb)->flush = 1;
534 return NULL;
535 }
536 if ((skb->ip_summed != p->ip_summed) ||
537 (skb->csum_level != p->csum_level)) {
538 NAPI_GRO_CB(skb)->flush = 1;
539 return NULL;
540 }
541 ret = skb_gro_receive_list(p, skb);
542 } else {
543 skb_gro_postpull_rcsum(skb, uh,
544 sizeof(struct udphdr));
545
546 ret = skb_gro_receive(p, skb);
547 }
548 }
549
550 if (ret || ulen != ntohs(uh2->len) ||
551 NAPI_GRO_CB(p)->count >= UDP_GRO_CNT_MAX)
552 pp = p;
553
554 return pp;
555 }
556
557 /* mismatch, but we never need to flush */
558 return NULL;
559 }
560
udp_gro_receive(struct list_head * head,struct sk_buff * skb,struct udphdr * uh,struct sock * sk)561 struct sk_buff *udp_gro_receive(struct list_head *head, struct sk_buff *skb,
562 struct udphdr *uh, struct sock *sk)
563 {
564 struct sk_buff *pp = NULL;
565 struct sk_buff *p;
566 struct udphdr *uh2;
567 unsigned int off = skb_gro_offset(skb);
568 int flush = 1;
569
570 /* We can do L4 aggregation only if the packet can't land in a tunnel
571 * otherwise we could corrupt the inner stream. Detecting such packets
572 * cannot be foolproof and the aggregation might still happen in some
573 * cases. Such packets should be caught in udp_unexpected_gso later.
574 */
575 NAPI_GRO_CB(skb)->is_flist = 0;
576 if (!sk || !udp_sk(sk)->gro_receive) {
577 /* If the packet was locally encapsulated in a UDP tunnel that
578 * wasn't detected above, do not GRO.
579 */
580 if (skb->encapsulation)
581 goto out;
582
583 if (skb->dev->features & NETIF_F_GRO_FRAGLIST)
584 NAPI_GRO_CB(skb)->is_flist = sk ? !udp_test_bit(GRO_ENABLED, sk) : 1;
585
586 if ((!sk && (skb->dev->features & NETIF_F_GRO_UDP_FWD)) ||
587 (sk && udp_test_bit(GRO_ENABLED, sk)) || NAPI_GRO_CB(skb)->is_flist)
588 return call_gro_receive(udp_gro_receive_segment, head, skb);
589
590 /* no GRO, be sure flush the current packet */
591 goto out;
592 }
593
594 if (NAPI_GRO_CB(skb)->encap_mark ||
595 (uh->check && skb->ip_summed != CHECKSUM_PARTIAL &&
596 NAPI_GRO_CB(skb)->csum_cnt == 0 &&
597 !NAPI_GRO_CB(skb)->csum_valid))
598 goto out;
599
600 /* mark that this skb passed once through the tunnel gro layer */
601 NAPI_GRO_CB(skb)->encap_mark = 1;
602
603 flush = 0;
604
605 list_for_each_entry(p, head, list) {
606 if (!NAPI_GRO_CB(p)->same_flow)
607 continue;
608
609 uh2 = (struct udphdr *)(p->data + off);
610
611 /* Match ports and either checksums are either both zero
612 * or nonzero.
613 */
614 if ((*(u32 *)&uh->source != *(u32 *)&uh2->source) ||
615 (!uh->check ^ !uh2->check)) {
616 NAPI_GRO_CB(p)->same_flow = 0;
617 continue;
618 }
619 }
620
621 skb_gro_pull(skb, sizeof(struct udphdr)); /* pull encapsulating udp header */
622 skb_gro_postpull_rcsum(skb, uh, sizeof(struct udphdr));
623 pp = call_gro_receive_sk(udp_sk(sk)->gro_receive, sk, head, skb);
624
625 out:
626 skb_gro_flush_final(skb, pp, flush);
627 return pp;
628 }
629 EXPORT_SYMBOL(udp_gro_receive);
630
udp4_gro_lookup_skb(struct sk_buff * skb,__be16 sport,__be16 dport)631 static struct sock *udp4_gro_lookup_skb(struct sk_buff *skb, __be16 sport,
632 __be16 dport)
633 {
634 const struct iphdr *iph = skb_gro_network_header(skb);
635 struct net *net = dev_net(skb->dev);
636 int iif, sdif;
637
638 inet_get_iif_sdif(skb, &iif, &sdif);
639
640 return __udp4_lib_lookup(net, iph->saddr, sport,
641 iph->daddr, dport, iif,
642 sdif, net->ipv4.udp_table, NULL);
643 }
644
645 INDIRECT_CALLABLE_SCOPE
udp4_gro_receive(struct list_head * head,struct sk_buff * skb)646 struct sk_buff *udp4_gro_receive(struct list_head *head, struct sk_buff *skb)
647 {
648 struct udphdr *uh = udp_gro_udphdr(skb);
649 struct sock *sk = NULL;
650 struct sk_buff *pp;
651
652 if (unlikely(!uh))
653 goto flush;
654
655 /* Don't bother verifying checksum if we're going to flush anyway. */
656 if (NAPI_GRO_CB(skb)->flush)
657 goto skip;
658
659 if (skb_gro_checksum_validate_zero_check(skb, IPPROTO_UDP, uh->check,
660 inet_gro_compute_pseudo))
661 goto flush;
662 else if (uh->check)
663 skb_gro_checksum_try_convert(skb, IPPROTO_UDP,
664 inet_gro_compute_pseudo);
665 skip:
666 NAPI_GRO_CB(skb)->is_ipv6 = 0;
667
668 if (static_branch_unlikely(&udp_encap_needed_key))
669 sk = udp4_gro_lookup_skb(skb, uh->source, uh->dest);
670
671 pp = udp_gro_receive(head, skb, uh, sk);
672 return pp;
673
674 flush:
675 NAPI_GRO_CB(skb)->flush = 1;
676 return NULL;
677 }
678
udp_gro_complete_segment(struct sk_buff * skb)679 static int udp_gro_complete_segment(struct sk_buff *skb)
680 {
681 struct udphdr *uh = udp_hdr(skb);
682
683 skb->csum_start = (unsigned char *)uh - skb->head;
684 skb->csum_offset = offsetof(struct udphdr, check);
685 skb->ip_summed = CHECKSUM_PARTIAL;
686
687 skb_shinfo(skb)->gso_segs = NAPI_GRO_CB(skb)->count;
688 skb_shinfo(skb)->gso_type |= SKB_GSO_UDP_L4;
689
690 if (skb->encapsulation)
691 skb->inner_transport_header = skb->transport_header;
692
693 return 0;
694 }
695
udp_gro_complete(struct sk_buff * skb,int nhoff,udp_lookup_t lookup)696 int udp_gro_complete(struct sk_buff *skb, int nhoff,
697 udp_lookup_t lookup)
698 {
699 __be16 newlen = htons(skb->len - nhoff);
700 struct udphdr *uh = (struct udphdr *)(skb->data + nhoff);
701 struct sock *sk;
702 int err;
703
704 uh->len = newlen;
705
706 sk = INDIRECT_CALL_INET(lookup, udp6_lib_lookup_skb,
707 udp4_lib_lookup_skb, skb, uh->source, uh->dest);
708 if (sk && udp_sk(sk)->gro_complete) {
709 skb_shinfo(skb)->gso_type = uh->check ? SKB_GSO_UDP_TUNNEL_CSUM
710 : SKB_GSO_UDP_TUNNEL;
711
712 /* clear the encap mark, so that inner frag_list gro_complete
713 * can take place
714 */
715 NAPI_GRO_CB(skb)->encap_mark = 0;
716
717 /* Set encapsulation before calling into inner gro_complete()
718 * functions to make them set up the inner offsets.
719 */
720 skb->encapsulation = 1;
721 err = udp_sk(sk)->gro_complete(sk, skb,
722 nhoff + sizeof(struct udphdr));
723 } else {
724 err = udp_gro_complete_segment(skb);
725 }
726
727 if (skb->remcsum_offload)
728 skb_shinfo(skb)->gso_type |= SKB_GSO_TUNNEL_REMCSUM;
729
730 return err;
731 }
732 EXPORT_SYMBOL(udp_gro_complete);
733
udp4_gro_complete(struct sk_buff * skb,int nhoff)734 INDIRECT_CALLABLE_SCOPE int udp4_gro_complete(struct sk_buff *skb, int nhoff)
735 {
736 const u16 offset = NAPI_GRO_CB(skb)->network_offsets[skb->encapsulation];
737 const struct iphdr *iph = (struct iphdr *)(skb->data + offset);
738 struct udphdr *uh = (struct udphdr *)(skb->data + nhoff);
739
740 /* do fraglist only if there is no outer UDP encap (or we already processed it) */
741 if (NAPI_GRO_CB(skb)->is_flist && !NAPI_GRO_CB(skb)->encap_mark) {
742 uh->len = htons(skb->len - nhoff);
743
744 skb_shinfo(skb)->gso_type |= (SKB_GSO_FRAGLIST|SKB_GSO_UDP_L4);
745 skb_shinfo(skb)->gso_segs = NAPI_GRO_CB(skb)->count;
746
747 __skb_incr_checksum_unnecessary(skb);
748
749 return 0;
750 }
751
752 if (uh->check)
753 uh->check = ~udp_v4_check(skb->len - nhoff, iph->saddr,
754 iph->daddr, 0);
755
756 return udp_gro_complete(skb, nhoff, udp4_lib_lookup_skb);
757 }
758
759 static const struct net_offload udpv4_offload = {
760 .callbacks = {
761 .gso_segment = udp4_ufo_fragment,
762 .gro_receive = udp4_gro_receive,
763 .gro_complete = udp4_gro_complete,
764 },
765 };
766
udpv4_offload_init(void)767 int __init udpv4_offload_init(void)
768 {
769 return inet_add_offload(&udpv4_offload, IPPROTO_UDP);
770 }
771