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
__udpv6_gso_segment_csum(struct sk_buff * seg,struct in6_addr * oldip,const struct in6_addr * newip,__be16 * oldport,__be16 newport)250 static void __udpv6_gso_segment_csum(struct sk_buff *seg,
251 struct in6_addr *oldip,
252 const struct in6_addr *newip,
253 __be16 *oldport, __be16 newport)
254 {
255 struct udphdr *uh = udp_hdr(seg);
256
257 if (ipv6_addr_equal(oldip, newip) && *oldport == newport)
258 return;
259
260 if (uh->check) {
261 inet_proto_csum_replace16(&uh->check, seg, oldip->s6_addr32,
262 newip->s6_addr32, true);
263
264 inet_proto_csum_replace2(&uh->check, seg, *oldport, newport,
265 false);
266 if (!uh->check)
267 uh->check = CSUM_MANGLED_0;
268 }
269
270 *oldip = *newip;
271 *oldport = newport;
272 }
273
__udpv6_gso_segment_list_csum(struct sk_buff * segs)274 static struct sk_buff *__udpv6_gso_segment_list_csum(struct sk_buff *segs)
275 {
276 const struct ipv6hdr *iph;
277 const struct udphdr *uh;
278 struct ipv6hdr *iph2;
279 struct sk_buff *seg;
280 struct udphdr *uh2;
281
282 seg = segs;
283 uh = udp_hdr(seg);
284 iph = ipv6_hdr(seg);
285 uh2 = udp_hdr(seg->next);
286 iph2 = ipv6_hdr(seg->next);
287
288 if (!(*(const u32 *)&uh->source ^ *(const u32 *)&uh2->source) &&
289 ipv6_addr_equal(&iph->saddr, &iph2->saddr) &&
290 ipv6_addr_equal(&iph->daddr, &iph2->daddr))
291 return segs;
292
293 while ((seg = seg->next)) {
294 uh2 = udp_hdr(seg);
295 iph2 = ipv6_hdr(seg);
296
297 __udpv6_gso_segment_csum(seg, &iph2->saddr, &iph->saddr,
298 &uh2->source, uh->source);
299 __udpv6_gso_segment_csum(seg, &iph2->daddr, &iph->daddr,
300 &uh2->dest, uh->dest);
301 }
302
303 return segs;
304 }
305
__udp_gso_segment_list(struct sk_buff * skb,netdev_features_t features,bool is_ipv6)306 static struct sk_buff *__udp_gso_segment_list(struct sk_buff *skb,
307 netdev_features_t features,
308 bool is_ipv6)
309 {
310 unsigned int mss = skb_shinfo(skb)->gso_size;
311
312 skb = skb_segment_list(skb, features, skb_mac_header_len(skb));
313 if (IS_ERR(skb))
314 return skb;
315
316 udp_hdr(skb)->len = htons(sizeof(struct udphdr) + mss);
317
318 if (is_ipv6)
319 return __udpv6_gso_segment_list_csum(skb);
320 else
321 return __udpv4_gso_segment_list_csum(skb);
322 }
323
__udp_gso_segment(struct sk_buff * gso_skb,netdev_features_t features,bool is_ipv6)324 struct sk_buff *__udp_gso_segment(struct sk_buff *gso_skb,
325 netdev_features_t features, bool is_ipv6)
326 {
327 struct sock *sk = gso_skb->sk;
328 unsigned int sum_truesize = 0;
329 struct sk_buff *segs, *seg;
330 struct udphdr *uh;
331 unsigned int mss;
332 bool copy_dtor;
333 __sum16 check;
334 __be16 newlen;
335
336 mss = skb_shinfo(gso_skb)->gso_size;
337 if (gso_skb->len <= sizeof(*uh) + mss)
338 return ERR_PTR(-EINVAL);
339
340 if (unlikely(skb_checksum_start(gso_skb) !=
341 skb_transport_header(gso_skb) &&
342 !(skb_shinfo(gso_skb)->gso_type & SKB_GSO_FRAGLIST)))
343 return ERR_PTR(-EINVAL);
344
345 /* We don't know if egress device can segment and checksum the packet
346 * when IPv6 extension headers are present. Fall back to software GSO.
347 */
348 if (gso_skb->ip_summed != CHECKSUM_PARTIAL)
349 features &= ~(NETIF_F_GSO_UDP_L4 | NETIF_F_CSUM_MASK);
350
351 if (skb_gso_ok(gso_skb, features | NETIF_F_GSO_ROBUST)) {
352 /* Packet is from an untrusted source, reset gso_segs. */
353 skb_shinfo(gso_skb)->gso_segs = DIV_ROUND_UP(gso_skb->len - sizeof(*uh),
354 mss);
355 return NULL;
356 }
357
358 if (skb_shinfo(gso_skb)->gso_type & SKB_GSO_FRAGLIST) {
359 /* Detect modified geometry and pass those to skb_segment. */
360 if (skb_pagelen(gso_skb) - sizeof(*uh) == skb_shinfo(gso_skb)->gso_size)
361 return __udp_gso_segment_list(gso_skb, features, is_ipv6);
362
363 /* Setup csum, as fraglist skips this in udp4_gro_receive. */
364 gso_skb->csum_start = skb_transport_header(gso_skb) - gso_skb->head;
365 gso_skb->csum_offset = offsetof(struct udphdr, check);
366 gso_skb->ip_summed = CHECKSUM_PARTIAL;
367
368 uh = udp_hdr(gso_skb);
369 if (is_ipv6)
370 uh->check = ~udp_v6_check(gso_skb->len,
371 &ipv6_hdr(gso_skb)->saddr,
372 &ipv6_hdr(gso_skb)->daddr, 0);
373 else
374 uh->check = ~udp_v4_check(gso_skb->len,
375 ip_hdr(gso_skb)->saddr,
376 ip_hdr(gso_skb)->daddr, 0);
377 }
378
379 skb_pull(gso_skb, sizeof(*uh));
380
381 /* clear destructor to avoid skb_segment assigning it to tail */
382 copy_dtor = gso_skb->destructor == sock_wfree;
383 if (copy_dtor) {
384 gso_skb->destructor = NULL;
385 gso_skb->sk = NULL;
386 }
387
388 segs = skb_segment(gso_skb, features);
389 if (IS_ERR_OR_NULL(segs)) {
390 if (copy_dtor) {
391 gso_skb->destructor = sock_wfree;
392 gso_skb->sk = sk;
393 }
394 return segs;
395 }
396
397 /* GSO partial and frag_list segmentation only requires splitting
398 * the frame into an MSS multiple and possibly a remainder, both
399 * cases return a GSO skb. So update the mss now.
400 */
401 if (skb_is_gso(segs))
402 mss *= skb_shinfo(segs)->gso_segs;
403
404 seg = segs;
405 uh = udp_hdr(seg);
406
407 /* preserve TX timestamp flags and TS key for first segment */
408 skb_shinfo(seg)->tskey = skb_shinfo(gso_skb)->tskey;
409 skb_shinfo(seg)->tx_flags |=
410 (skb_shinfo(gso_skb)->tx_flags & SKBTX_ANY_TSTAMP);
411
412 /* compute checksum adjustment based on old length versus new */
413 newlen = htons(sizeof(*uh) + mss);
414 check = csum16_add(csum16_sub(uh->check, uh->len), newlen);
415
416 for (;;) {
417 if (copy_dtor) {
418 seg->destructor = sock_wfree;
419 seg->sk = sk;
420 sum_truesize += seg->truesize;
421 }
422
423 if (!seg->next)
424 break;
425
426 uh->len = newlen;
427 uh->check = check;
428
429 if (seg->ip_summed == CHECKSUM_PARTIAL)
430 gso_reset_checksum(seg, ~check);
431 else
432 uh->check = gso_make_checksum(seg, ~check) ? :
433 CSUM_MANGLED_0;
434
435 seg = seg->next;
436 uh = udp_hdr(seg);
437 }
438
439 /* last packet can be partial gso_size, account for that in checksum */
440 newlen = htons(skb_tail_pointer(seg) - skb_transport_header(seg) +
441 seg->data_len);
442 check = csum16_add(csum16_sub(uh->check, uh->len), newlen);
443
444 uh->len = newlen;
445 uh->check = check;
446
447 if (seg->ip_summed == CHECKSUM_PARTIAL)
448 gso_reset_checksum(seg, ~check);
449 else
450 uh->check = gso_make_checksum(seg, ~check) ? : CSUM_MANGLED_0;
451
452 /* On the TX path, CHECKSUM_NONE and CHECKSUM_UNNECESSARY have the same
453 * meaning. However, check for bad offloads in the GSO stack expects the
454 * latter, if the checksum was calculated in software. To vouch for the
455 * segment skbs we actually need to set it on the gso_skb.
456 */
457 if (gso_skb->ip_summed == CHECKSUM_NONE)
458 gso_skb->ip_summed = CHECKSUM_UNNECESSARY;
459
460 /* update refcount for the packet */
461 if (copy_dtor) {
462 int delta = sum_truesize - gso_skb->truesize;
463
464 /* In some pathological cases, delta can be negative.
465 * We need to either use refcount_add() or refcount_sub_and_test()
466 */
467 if (likely(delta >= 0))
468 refcount_add(delta, &sk->sk_wmem_alloc);
469 else
470 WARN_ON_ONCE(refcount_sub_and_test(-delta, &sk->sk_wmem_alloc));
471 }
472 return segs;
473 }
474 EXPORT_SYMBOL_GPL(__udp_gso_segment);
475
udp4_ufo_fragment(struct sk_buff * skb,netdev_features_t features)476 static struct sk_buff *udp4_ufo_fragment(struct sk_buff *skb,
477 netdev_features_t features)
478 {
479 struct sk_buff *segs = ERR_PTR(-EINVAL);
480 unsigned int mss;
481 __wsum csum;
482 struct udphdr *uh;
483 struct iphdr *iph;
484
485 if (skb->encapsulation &&
486 (skb_shinfo(skb)->gso_type &
487 (SKB_GSO_UDP_TUNNEL|SKB_GSO_UDP_TUNNEL_CSUM))) {
488 segs = skb_udp_tunnel_segment(skb, features, false);
489 goto out;
490 }
491
492 if (!(skb_shinfo(skb)->gso_type & (SKB_GSO_UDP | SKB_GSO_UDP_L4)))
493 goto out;
494
495 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
496 goto out;
497
498 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4)
499 return __udp_gso_segment(skb, features, false);
500
501 mss = skb_shinfo(skb)->gso_size;
502 if (unlikely(skb->len <= mss))
503 goto out;
504
505 /* Do software UFO. Complete and fill in the UDP checksum as
506 * HW cannot do checksum of UDP packets sent as multiple
507 * IP fragments.
508 */
509
510 uh = udp_hdr(skb);
511 iph = ip_hdr(skb);
512
513 uh->check = 0;
514 csum = skb_checksum(skb, 0, skb->len, 0);
515 uh->check = udp_v4_check(skb->len, iph->saddr, iph->daddr, csum);
516 if (uh->check == 0)
517 uh->check = CSUM_MANGLED_0;
518
519 skb->ip_summed = CHECKSUM_UNNECESSARY;
520
521 /* If there is no outer header we can fake a checksum offload
522 * due to the fact that we have already done the checksum in
523 * software prior to segmenting the frame.
524 */
525 if (!skb->encap_hdr_csum)
526 features |= NETIF_F_HW_CSUM;
527
528 /* Fragment the skb. IP headers of the fragments are updated in
529 * inet_gso_segment()
530 */
531 segs = skb_segment(skb, features);
532 out:
533 return segs;
534 }
535
536
537 #define UDP_GRO_CNT_MAX 64
udp_gro_receive_segment(struct list_head * head,struct sk_buff * skb)538 static struct sk_buff *udp_gro_receive_segment(struct list_head *head,
539 struct sk_buff *skb)
540 {
541 struct udphdr *uh = udp_gro_udphdr(skb);
542 struct sk_buff *pp = NULL;
543 struct udphdr *uh2;
544 struct sk_buff *p;
545 unsigned int ulen;
546 int ret = 0;
547 int flush;
548
549 /* requires non zero csum, for symmetry with GSO */
550 if (!uh->check) {
551 NAPI_GRO_CB(skb)->flush = 1;
552 return NULL;
553 }
554
555 /* Do not deal with padded or malicious packets, sorry ! */
556 ulen = ntohs(uh->len);
557 if (ulen <= sizeof(*uh) || ulen != skb_gro_len(skb)) {
558 NAPI_GRO_CB(skb)->flush = 1;
559 return NULL;
560 }
561 /* pull encapsulating udp header */
562 skb_gro_pull(skb, sizeof(struct udphdr));
563
564 list_for_each_entry(p, head, list) {
565 if (!NAPI_GRO_CB(p)->same_flow)
566 continue;
567
568 uh2 = udp_hdr(p);
569
570 /* Match ports only, as csum is always non zero */
571 if ((*(u32 *)&uh->source != *(u32 *)&uh2->source)) {
572 NAPI_GRO_CB(p)->same_flow = 0;
573 continue;
574 }
575
576 if (NAPI_GRO_CB(skb)->is_flist != NAPI_GRO_CB(p)->is_flist) {
577 NAPI_GRO_CB(skb)->flush = 1;
578 return p;
579 }
580
581 flush = gro_receive_network_flush(uh, uh2, p);
582
583 /* Terminate the flow on len mismatch or if it grow "too much".
584 * Under small packet flood GRO count could elsewhere grow a lot
585 * leading to excessive truesize values.
586 * On len mismatch merge the first packet shorter than gso_size,
587 * otherwise complete the GRO packet.
588 */
589 if (ulen > ntohs(uh2->len) || flush) {
590 pp = p;
591 } else {
592 if (NAPI_GRO_CB(skb)->is_flist) {
593 if (!pskb_may_pull(skb, skb_gro_offset(skb))) {
594 NAPI_GRO_CB(skb)->flush = 1;
595 return NULL;
596 }
597 if ((skb->ip_summed != p->ip_summed) ||
598 (skb->csum_level != p->csum_level)) {
599 NAPI_GRO_CB(skb)->flush = 1;
600 return NULL;
601 }
602 ret = skb_gro_receive_list(p, skb);
603 } else {
604 skb_gro_postpull_rcsum(skb, uh,
605 sizeof(struct udphdr));
606
607 ret = skb_gro_receive(p, skb);
608 }
609 }
610
611 if (ret || ulen != ntohs(uh2->len) ||
612 NAPI_GRO_CB(p)->count >= UDP_GRO_CNT_MAX)
613 pp = p;
614
615 return pp;
616 }
617
618 /* mismatch, but we never need to flush */
619 return NULL;
620 }
621
udp_gro_receive(struct list_head * head,struct sk_buff * skb,struct udphdr * uh,struct sock * sk)622 struct sk_buff *udp_gro_receive(struct list_head *head, struct sk_buff *skb,
623 struct udphdr *uh, struct sock *sk)
624 {
625 struct sk_buff *pp = NULL;
626 struct sk_buff *p;
627 struct udphdr *uh2;
628 unsigned int off = skb_gro_offset(skb);
629 int flush = 1;
630
631 /* We can do L4 aggregation only if the packet can't land in a tunnel
632 * otherwise we could corrupt the inner stream. Detecting such packets
633 * cannot be foolproof and the aggregation might still happen in some
634 * cases. Such packets should be caught in udp_unexpected_gso later.
635 */
636 NAPI_GRO_CB(skb)->is_flist = 0;
637 if (!sk || !udp_sk(sk)->gro_receive) {
638 /* If the packet was locally encapsulated in a UDP tunnel that
639 * wasn't detected above, do not GRO.
640 */
641 if (skb->encapsulation)
642 goto out;
643
644 if (skb->dev->features & NETIF_F_GRO_FRAGLIST)
645 NAPI_GRO_CB(skb)->is_flist = sk ? !udp_test_bit(GRO_ENABLED, sk) : 1;
646
647 if ((!sk && (skb->dev->features & NETIF_F_GRO_UDP_FWD)) ||
648 (sk && udp_test_bit(GRO_ENABLED, sk)) || NAPI_GRO_CB(skb)->is_flist)
649 return call_gro_receive(udp_gro_receive_segment, head, skb);
650
651 /* no GRO, be sure flush the current packet */
652 goto out;
653 }
654
655 if (NAPI_GRO_CB(skb)->encap_mark ||
656 (uh->check && skb->ip_summed != CHECKSUM_PARTIAL &&
657 NAPI_GRO_CB(skb)->csum_cnt == 0 &&
658 !NAPI_GRO_CB(skb)->csum_valid))
659 goto out;
660
661 /* mark that this skb passed once through the tunnel gro layer */
662 NAPI_GRO_CB(skb)->encap_mark = 1;
663
664 flush = 0;
665
666 list_for_each_entry(p, head, list) {
667 if (!NAPI_GRO_CB(p)->same_flow)
668 continue;
669
670 uh2 = (struct udphdr *)(p->data + off);
671
672 /* Match ports and either checksums are either both zero
673 * or nonzero.
674 */
675 if ((*(u32 *)&uh->source != *(u32 *)&uh2->source) ||
676 (!uh->check ^ !uh2->check)) {
677 NAPI_GRO_CB(p)->same_flow = 0;
678 continue;
679 }
680 }
681
682 skb_gro_pull(skb, sizeof(struct udphdr)); /* pull encapsulating udp header */
683 skb_gro_postpull_rcsum(skb, uh, sizeof(struct udphdr));
684 pp = call_gro_receive_sk(udp_sk(sk)->gro_receive, sk, head, skb);
685
686 out:
687 skb_gro_flush_final(skb, pp, flush);
688 return pp;
689 }
690 EXPORT_SYMBOL(udp_gro_receive);
691
udp4_gro_lookup_skb(struct sk_buff * skb,__be16 sport,__be16 dport)692 static struct sock *udp4_gro_lookup_skb(struct sk_buff *skb, __be16 sport,
693 __be16 dport)
694 {
695 const struct iphdr *iph = skb_gro_network_header(skb);
696 struct net *net = dev_net_rcu(skb->dev);
697 int iif, sdif;
698
699 inet_get_iif_sdif(skb, &iif, &sdif);
700
701 return __udp4_lib_lookup(net, iph->saddr, sport,
702 iph->daddr, dport, iif,
703 sdif, net->ipv4.udp_table, NULL);
704 }
705
706 INDIRECT_CALLABLE_SCOPE
udp4_gro_receive(struct list_head * head,struct sk_buff * skb)707 struct sk_buff *udp4_gro_receive(struct list_head *head, struct sk_buff *skb)
708 {
709 struct udphdr *uh = udp_gro_udphdr(skb);
710 struct sock *sk = NULL;
711 struct sk_buff *pp;
712
713 if (unlikely(!uh))
714 goto flush;
715
716 /* Don't bother verifying checksum if we're going to flush anyway. */
717 if (NAPI_GRO_CB(skb)->flush)
718 goto skip;
719
720 if (skb_gro_checksum_validate_zero_check(skb, IPPROTO_UDP, uh->check,
721 inet_gro_compute_pseudo))
722 goto flush;
723 else if (uh->check)
724 skb_gro_checksum_try_convert(skb, IPPROTO_UDP,
725 inet_gro_compute_pseudo);
726 skip:
727 NAPI_GRO_CB(skb)->is_ipv6 = 0;
728
729 if (static_branch_unlikely(&udp_encap_needed_key))
730 sk = udp4_gro_lookup_skb(skb, uh->source, uh->dest);
731
732 pp = udp_gro_receive(head, skb, uh, sk);
733 return pp;
734
735 flush:
736 NAPI_GRO_CB(skb)->flush = 1;
737 return NULL;
738 }
739
udp_gro_complete_segment(struct sk_buff * skb)740 static int udp_gro_complete_segment(struct sk_buff *skb)
741 {
742 struct udphdr *uh = udp_hdr(skb);
743
744 skb->csum_start = (unsigned char *)uh - skb->head;
745 skb->csum_offset = offsetof(struct udphdr, check);
746 skb->ip_summed = CHECKSUM_PARTIAL;
747
748 skb_shinfo(skb)->gso_segs = NAPI_GRO_CB(skb)->count;
749 skb_shinfo(skb)->gso_type |= SKB_GSO_UDP_L4;
750
751 if (skb->encapsulation)
752 skb->inner_transport_header = skb->transport_header;
753
754 return 0;
755 }
756
udp_gro_complete(struct sk_buff * skb,int nhoff,udp_lookup_t lookup)757 int udp_gro_complete(struct sk_buff *skb, int nhoff,
758 udp_lookup_t lookup)
759 {
760 __be16 newlen = htons(skb->len - nhoff);
761 struct udphdr *uh = (struct udphdr *)(skb->data + nhoff);
762 struct sock *sk;
763 int err;
764
765 uh->len = newlen;
766
767 sk = INDIRECT_CALL_INET(lookup, udp6_lib_lookup_skb,
768 udp4_lib_lookup_skb, skb, uh->source, uh->dest);
769 if (sk && udp_sk(sk)->gro_complete) {
770 skb_shinfo(skb)->gso_type = uh->check ? SKB_GSO_UDP_TUNNEL_CSUM
771 : SKB_GSO_UDP_TUNNEL;
772
773 /* clear the encap mark, so that inner frag_list gro_complete
774 * can take place
775 */
776 NAPI_GRO_CB(skb)->encap_mark = 0;
777
778 /* Set encapsulation before calling into inner gro_complete()
779 * functions to make them set up the inner offsets.
780 */
781 skb->encapsulation = 1;
782 err = udp_sk(sk)->gro_complete(sk, skb,
783 nhoff + sizeof(struct udphdr));
784 } else {
785 err = udp_gro_complete_segment(skb);
786 }
787
788 if (skb->remcsum_offload)
789 skb_shinfo(skb)->gso_type |= SKB_GSO_TUNNEL_REMCSUM;
790
791 return err;
792 }
793 EXPORT_SYMBOL(udp_gro_complete);
794
udp4_gro_complete(struct sk_buff * skb,int nhoff)795 INDIRECT_CALLABLE_SCOPE int udp4_gro_complete(struct sk_buff *skb, int nhoff)
796 {
797 const u16 offset = NAPI_GRO_CB(skb)->network_offsets[skb->encapsulation];
798 const struct iphdr *iph = (struct iphdr *)(skb->data + offset);
799 struct udphdr *uh = (struct udphdr *)(skb->data + nhoff);
800
801 /* do fraglist only if there is no outer UDP encap (or we already processed it) */
802 if (NAPI_GRO_CB(skb)->is_flist && !NAPI_GRO_CB(skb)->encap_mark) {
803 uh->len = htons(skb->len - nhoff);
804
805 skb_shinfo(skb)->gso_type |= (SKB_GSO_FRAGLIST|SKB_GSO_UDP_L4);
806 skb_shinfo(skb)->gso_segs = NAPI_GRO_CB(skb)->count;
807
808 __skb_incr_checksum_unnecessary(skb);
809
810 return 0;
811 }
812
813 if (uh->check)
814 uh->check = ~udp_v4_check(skb->len - nhoff, iph->saddr,
815 iph->daddr, 0);
816
817 return udp_gro_complete(skb, nhoff, udp4_lib_lookup_skb);
818 }
819
udpv4_offload_init(void)820 int __init udpv4_offload_init(void)
821 {
822 net_hotdata.udpv4_offload = (struct net_offload) {
823 .callbacks = {
824 .gso_segment = udp4_ufo_fragment,
825 .gro_receive = udp4_gro_receive,
826 .gro_complete = udp4_gro_complete,
827 },
828 };
829 return inet_add_offload(&net_hotdata.udpv4_offload, IPPROTO_UDP);
830 }
831