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