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 int ret = 0;
336
337 mss = skb_shinfo(gso_skb)->gso_size;
338 if (gso_skb->len <= sizeof(*uh) + mss)
339 return ERR_PTR(-EINVAL);
340
341 if (unlikely(skb_checksum_start(gso_skb) !=
342 skb_transport_header(gso_skb) &&
343 !(skb_shinfo(gso_skb)->gso_type & SKB_GSO_FRAGLIST)))
344 return ERR_PTR(-EINVAL);
345
346 if (skb_gso_ok(gso_skb, features | NETIF_F_GSO_ROBUST)) {
347 /* Packet is from an untrusted source, reset gso_segs. */
348 skb_shinfo(gso_skb)->gso_segs = DIV_ROUND_UP(gso_skb->len - sizeof(*uh),
349 mss);
350 return NULL;
351 }
352
353 if (skb_shinfo(gso_skb)->gso_type & SKB_GSO_FRAGLIST) {
354 /* Detect modified geometry and pass those to skb_segment. */
355 if (skb_pagelen(gso_skb) - sizeof(*uh) == skb_shinfo(gso_skb)->gso_size)
356 return __udp_gso_segment_list(gso_skb, features, is_ipv6);
357
358 ret = __skb_linearize(gso_skb);
359 if (ret)
360 return ERR_PTR(ret);
361
362 /* Setup csum, as fraglist skips this in udp4_gro_receive. */
363 gso_skb->csum_start = skb_transport_header(gso_skb) - gso_skb->head;
364 gso_skb->csum_offset = offsetof(struct udphdr, check);
365 gso_skb->ip_summed = CHECKSUM_PARTIAL;
366
367 uh = udp_hdr(gso_skb);
368 if (is_ipv6)
369 uh->check = ~udp_v6_check(gso_skb->len,
370 &ipv6_hdr(gso_skb)->saddr,
371 &ipv6_hdr(gso_skb)->daddr, 0);
372 else
373 uh->check = ~udp_v4_check(gso_skb->len,
374 ip_hdr(gso_skb)->saddr,
375 ip_hdr(gso_skb)->daddr, 0);
376 }
377
378 skb_pull(gso_skb, sizeof(*uh));
379
380 /* clear destructor to avoid skb_segment assigning it to tail */
381 copy_dtor = gso_skb->destructor == sock_wfree;
382 if (copy_dtor) {
383 gso_skb->destructor = NULL;
384 gso_skb->sk = NULL;
385 }
386
387 segs = skb_segment(gso_skb, features);
388 if (IS_ERR_OR_NULL(segs)) {
389 if (copy_dtor) {
390 gso_skb->destructor = sock_wfree;
391 gso_skb->sk = sk;
392 }
393 return segs;
394 }
395
396 /* GSO partial and frag_list segmentation only requires splitting
397 * the frame into an MSS multiple and possibly a remainder, both
398 * cases return a GSO skb. So update the mss now.
399 */
400 if (skb_is_gso(segs))
401 mss *= skb_shinfo(segs)->gso_segs;
402
403 seg = segs;
404 uh = udp_hdr(seg);
405
406 /* preserve TX timestamp flags and TS key for first segment */
407 skb_shinfo(seg)->tskey = skb_shinfo(gso_skb)->tskey;
408 skb_shinfo(seg)->tx_flags |=
409 (skb_shinfo(gso_skb)->tx_flags & SKBTX_ANY_TSTAMP);
410
411 /* compute checksum adjustment based on old length versus new */
412 newlen = htons(sizeof(*uh) + mss);
413 check = csum16_add(csum16_sub(uh->check, uh->len), newlen);
414
415 for (;;) {
416 if (copy_dtor) {
417 seg->destructor = sock_wfree;
418 seg->sk = sk;
419 sum_truesize += seg->truesize;
420 }
421
422 if (!seg->next)
423 break;
424
425 uh->len = newlen;
426 uh->check = check;
427
428 if (seg->ip_summed == CHECKSUM_PARTIAL)
429 gso_reset_checksum(seg, ~check);
430 else
431 uh->check = gso_make_checksum(seg, ~check) ? :
432 CSUM_MANGLED_0;
433
434 seg = seg->next;
435 uh = udp_hdr(seg);
436 }
437
438 /* last packet can be partial gso_size, account for that in checksum */
439 newlen = htons(skb_tail_pointer(seg) - skb_transport_header(seg) +
440 seg->data_len);
441 check = csum16_add(csum16_sub(uh->check, uh->len), newlen);
442
443 uh->len = newlen;
444 uh->check = check;
445
446 if (seg->ip_summed == CHECKSUM_PARTIAL)
447 gso_reset_checksum(seg, ~check);
448 else
449 uh->check = gso_make_checksum(seg, ~check) ? : CSUM_MANGLED_0;
450
451 /* update refcount for the packet */
452 if (copy_dtor) {
453 int delta = sum_truesize - gso_skb->truesize;
454
455 /* In some pathological cases, delta can be negative.
456 * We need to either use refcount_add() or refcount_sub_and_test()
457 */
458 if (likely(delta >= 0))
459 refcount_add(delta, &sk->sk_wmem_alloc);
460 else
461 WARN_ON_ONCE(refcount_sub_and_test(-delta, &sk->sk_wmem_alloc));
462 }
463 return segs;
464 }
465 EXPORT_SYMBOL_GPL(__udp_gso_segment);
466
udp4_ufo_fragment(struct sk_buff * skb,netdev_features_t features)467 static struct sk_buff *udp4_ufo_fragment(struct sk_buff *skb,
468 netdev_features_t features)
469 {
470 struct sk_buff *segs = ERR_PTR(-EINVAL);
471 unsigned int mss;
472 __wsum csum;
473 struct udphdr *uh;
474 struct iphdr *iph;
475
476 if (skb->encapsulation &&
477 (skb_shinfo(skb)->gso_type &
478 (SKB_GSO_UDP_TUNNEL|SKB_GSO_UDP_TUNNEL_CSUM))) {
479 segs = skb_udp_tunnel_segment(skb, features, false);
480 goto out;
481 }
482
483 if (!(skb_shinfo(skb)->gso_type & (SKB_GSO_UDP | SKB_GSO_UDP_L4)))
484 goto out;
485
486 if (!pskb_may_pull(skb, sizeof(struct udphdr)))
487 goto out;
488
489 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4)
490 return __udp_gso_segment(skb, features, false);
491
492 mss = skb_shinfo(skb)->gso_size;
493 if (unlikely(skb->len <= mss))
494 goto out;
495
496 /* Do software UFO. Complete and fill in the UDP checksum as
497 * HW cannot do checksum of UDP packets sent as multiple
498 * IP fragments.
499 */
500
501 uh = udp_hdr(skb);
502 iph = ip_hdr(skb);
503
504 uh->check = 0;
505 csum = skb_checksum(skb, 0, skb->len, 0);
506 uh->check = udp_v4_check(skb->len, iph->saddr, iph->daddr, csum);
507 if (uh->check == 0)
508 uh->check = CSUM_MANGLED_0;
509
510 skb->ip_summed = CHECKSUM_UNNECESSARY;
511
512 /* If there is no outer header we can fake a checksum offload
513 * due to the fact that we have already done the checksum in
514 * software prior to segmenting the frame.
515 */
516 if (!skb->encap_hdr_csum)
517 features |= NETIF_F_HW_CSUM;
518
519 /* Fragment the skb. IP headers of the fragments are updated in
520 * inet_gso_segment()
521 */
522 segs = skb_segment(skb, features);
523 out:
524 return segs;
525 }
526
skb_gro_receive_list(struct sk_buff * p,struct sk_buff * skb)527 static int skb_gro_receive_list(struct sk_buff *p, struct sk_buff *skb)
528 {
529 if (unlikely(p->len + skb->len >= 65536))
530 return -E2BIG;
531
532 if (NAPI_GRO_CB(p)->last == p)
533 skb_shinfo(p)->frag_list = skb;
534 else
535 NAPI_GRO_CB(p)->last->next = skb;
536
537 skb_pull(skb, skb_gro_offset(skb));
538
539 NAPI_GRO_CB(p)->last = skb;
540 NAPI_GRO_CB(p)->count++;
541 p->data_len += skb->len;
542
543 /* sk ownership - if any - completely transferred to the aggregated packet */
544 skb->destructor = NULL;
545 skb->sk = NULL;
546 p->truesize += skb->truesize;
547 p->len += skb->len;
548
549 NAPI_GRO_CB(skb)->same_flow = 1;
550
551 return 0;
552 }
553
554
555 #define UDP_GRO_CNT_MAX 64
udp_gro_receive_segment(struct list_head * head,struct sk_buff * skb)556 static struct sk_buff *udp_gro_receive_segment(struct list_head *head,
557 struct sk_buff *skb)
558 {
559 struct udphdr *uh = udp_gro_udphdr(skb);
560 struct sk_buff *pp = NULL;
561 struct udphdr *uh2;
562 struct sk_buff *p;
563 unsigned int ulen;
564 int ret = 0;
565 int flush;
566
567 /* requires non zero csum, for symmetry with GSO */
568 if (!uh->check) {
569 NAPI_GRO_CB(skb)->flush = 1;
570 return NULL;
571 }
572
573 /* Do not deal with padded or malicious packets, sorry ! */
574 ulen = ntohs(uh->len);
575 if (ulen <= sizeof(*uh) || ulen != skb_gro_len(skb)) {
576 NAPI_GRO_CB(skb)->flush = 1;
577 return NULL;
578 }
579 /* pull encapsulating udp header */
580 skb_gro_pull(skb, sizeof(struct udphdr));
581
582 list_for_each_entry(p, head, list) {
583 if (!NAPI_GRO_CB(p)->same_flow)
584 continue;
585
586 uh2 = udp_hdr(p);
587
588 /* Match ports only, as csum is always non zero */
589 if ((*(u32 *)&uh->source != *(u32 *)&uh2->source)) {
590 NAPI_GRO_CB(p)->same_flow = 0;
591 continue;
592 }
593
594 if (NAPI_GRO_CB(skb)->is_flist != NAPI_GRO_CB(p)->is_flist) {
595 NAPI_GRO_CB(skb)->flush = 1;
596 return p;
597 }
598
599 flush = NAPI_GRO_CB(p)->flush;
600
601 if (NAPI_GRO_CB(p)->flush_id != 1 ||
602 NAPI_GRO_CB(p)->count != 1 ||
603 !NAPI_GRO_CB(p)->is_atomic)
604 flush |= NAPI_GRO_CB(p)->flush_id;
605 else
606 NAPI_GRO_CB(p)->is_atomic = false;
607
608 /* Terminate the flow on len mismatch or if it grow "too much".
609 * Under small packet flood GRO count could elsewhere grow a lot
610 * leading to excessive truesize values.
611 * On len mismatch merge the first packet shorter than gso_size,
612 * otherwise complete the GRO packet.
613 */
614 if (ulen > ntohs(uh2->len) || flush) {
615 pp = p;
616 } else {
617 if (NAPI_GRO_CB(skb)->is_flist) {
618 if (!pskb_may_pull(skb, skb_gro_offset(skb))) {
619 NAPI_GRO_CB(skb)->flush = 1;
620 return NULL;
621 }
622 if ((skb->ip_summed != p->ip_summed) ||
623 (skb->csum_level != p->csum_level)) {
624 NAPI_GRO_CB(skb)->flush = 1;
625 return NULL;
626 }
627 ret = skb_gro_receive_list(p, skb);
628 } else {
629 skb_gro_postpull_rcsum(skb, uh,
630 sizeof(struct udphdr));
631
632 ret = skb_gro_receive(p, skb);
633 }
634 }
635
636 if (ret || ulen != ntohs(uh2->len) ||
637 NAPI_GRO_CB(p)->count >= UDP_GRO_CNT_MAX)
638 pp = p;
639
640 return pp;
641 }
642
643 /* mismatch, but we never need to flush */
644 return NULL;
645 }
646
udp_gro_receive(struct list_head * head,struct sk_buff * skb,struct udphdr * uh,struct sock * sk)647 struct sk_buff *udp_gro_receive(struct list_head *head, struct sk_buff *skb,
648 struct udphdr *uh, struct sock *sk)
649 {
650 struct sk_buff *pp = NULL;
651 struct sk_buff *p;
652 struct udphdr *uh2;
653 unsigned int off = skb_gro_offset(skb);
654 int flush = 1;
655
656 /* We can do L4 aggregation only if the packet can't land in a tunnel
657 * otherwise we could corrupt the inner stream. Detecting such packets
658 * cannot be foolproof and the aggregation might still happen in some
659 * cases. Such packets should be caught in udp_unexpected_gso later.
660 */
661 NAPI_GRO_CB(skb)->is_flist = 0;
662 if (!sk || !udp_sk(sk)->gro_receive) {
663 /* If the packet was locally encapsulated in a UDP tunnel that
664 * wasn't detected above, do not GRO.
665 */
666 if (skb->encapsulation)
667 goto out;
668
669 if (skb->dev->features & NETIF_F_GRO_FRAGLIST)
670 NAPI_GRO_CB(skb)->is_flist = sk ? !udp_test_bit(GRO_ENABLED, sk) : 1;
671
672 if ((!sk && (skb->dev->features & NETIF_F_GRO_UDP_FWD)) ||
673 (sk && udp_test_bit(GRO_ENABLED, sk)) || NAPI_GRO_CB(skb)->is_flist)
674 return call_gro_receive(udp_gro_receive_segment, head, skb);
675
676 /* no GRO, be sure flush the current packet */
677 goto out;
678 }
679
680 if (NAPI_GRO_CB(skb)->encap_mark ||
681 (uh->check && skb->ip_summed != CHECKSUM_PARTIAL &&
682 NAPI_GRO_CB(skb)->csum_cnt == 0 &&
683 !NAPI_GRO_CB(skb)->csum_valid))
684 goto out;
685
686 /* mark that this skb passed once through the tunnel gro layer */
687 NAPI_GRO_CB(skb)->encap_mark = 1;
688
689 flush = 0;
690
691 list_for_each_entry(p, head, list) {
692 if (!NAPI_GRO_CB(p)->same_flow)
693 continue;
694
695 uh2 = (struct udphdr *)(p->data + off);
696
697 /* Match ports and either checksums are either both zero
698 * or nonzero.
699 */
700 if ((*(u32 *)&uh->source != *(u32 *)&uh2->source) ||
701 (!uh->check ^ !uh2->check)) {
702 NAPI_GRO_CB(p)->same_flow = 0;
703 continue;
704 }
705 }
706
707 skb_gro_pull(skb, sizeof(struct udphdr)); /* pull encapsulating udp header */
708 skb_gro_postpull_rcsum(skb, uh, sizeof(struct udphdr));
709 pp = call_gro_receive_sk(udp_sk(sk)->gro_receive, sk, head, skb);
710
711 out:
712 skb_gro_flush_final(skb, pp, flush);
713 return pp;
714 }
715 EXPORT_SYMBOL(udp_gro_receive);
716
udp4_gro_lookup_skb(struct sk_buff * skb,__be16 sport,__be16 dport)717 static struct sock *udp4_gro_lookup_skb(struct sk_buff *skb, __be16 sport,
718 __be16 dport)
719 {
720 const struct iphdr *iph = skb_gro_network_header(skb);
721 struct net *net = dev_net(skb->dev);
722 int iif, sdif;
723
724 inet_get_iif_sdif(skb, &iif, &sdif);
725
726 return __udp4_lib_lookup(net, iph->saddr, sport,
727 iph->daddr, dport, iif,
728 sdif, net->ipv4.udp_table, NULL);
729 }
730
731 INDIRECT_CALLABLE_SCOPE
udp4_gro_receive(struct list_head * head,struct sk_buff * skb)732 struct sk_buff *udp4_gro_receive(struct list_head *head, struct sk_buff *skb)
733 {
734 struct udphdr *uh = udp_gro_udphdr(skb);
735 struct sock *sk = NULL;
736 struct sk_buff *pp;
737
738 if (unlikely(!uh))
739 goto flush;
740
741 /* Don't bother verifying checksum if we're going to flush anyway. */
742 if (NAPI_GRO_CB(skb)->flush)
743 goto skip;
744
745 if (skb_gro_checksum_validate_zero_check(skb, IPPROTO_UDP, uh->check,
746 inet_gro_compute_pseudo))
747 goto flush;
748 else if (uh->check)
749 skb_gro_checksum_try_convert(skb, IPPROTO_UDP,
750 inet_gro_compute_pseudo);
751 skip:
752 NAPI_GRO_CB(skb)->is_ipv6 = 0;
753
754 if (static_branch_unlikely(&udp_encap_needed_key))
755 sk = udp4_gro_lookup_skb(skb, uh->source, uh->dest);
756
757 pp = udp_gro_receive(head, skb, uh, sk);
758 return pp;
759
760 flush:
761 NAPI_GRO_CB(skb)->flush = 1;
762 return NULL;
763 }
764
udp_gro_complete_segment(struct sk_buff * skb)765 static int udp_gro_complete_segment(struct sk_buff *skb)
766 {
767 struct udphdr *uh = udp_hdr(skb);
768
769 skb->csum_start = (unsigned char *)uh - skb->head;
770 skb->csum_offset = offsetof(struct udphdr, check);
771 skb->ip_summed = CHECKSUM_PARTIAL;
772
773 skb_shinfo(skb)->gso_segs = NAPI_GRO_CB(skb)->count;
774 skb_shinfo(skb)->gso_type |= SKB_GSO_UDP_L4;
775
776 if (skb->encapsulation)
777 skb->inner_transport_header = skb->transport_header;
778
779 return 0;
780 }
781
udp_gro_complete(struct sk_buff * skb,int nhoff,udp_lookup_t lookup)782 int udp_gro_complete(struct sk_buff *skb, int nhoff,
783 udp_lookup_t lookup)
784 {
785 __be16 newlen = htons(skb->len - nhoff);
786 struct udphdr *uh = (struct udphdr *)(skb->data + nhoff);
787 struct sock *sk;
788 int err;
789
790 uh->len = newlen;
791
792 sk = INDIRECT_CALL_INET(lookup, udp6_lib_lookup_skb,
793 udp4_lib_lookup_skb, skb, uh->source, uh->dest);
794 if (sk && udp_sk(sk)->gro_complete) {
795 skb_shinfo(skb)->gso_type = uh->check ? SKB_GSO_UDP_TUNNEL_CSUM
796 : SKB_GSO_UDP_TUNNEL;
797
798 /* clear the encap mark, so that inner frag_list gro_complete
799 * can take place
800 */
801 NAPI_GRO_CB(skb)->encap_mark = 0;
802
803 /* Set encapsulation before calling into inner gro_complete()
804 * functions to make them set up the inner offsets.
805 */
806 skb->encapsulation = 1;
807 err = udp_sk(sk)->gro_complete(sk, skb,
808 nhoff + sizeof(struct udphdr));
809 } else {
810 err = udp_gro_complete_segment(skb);
811 }
812
813 if (skb->remcsum_offload)
814 skb_shinfo(skb)->gso_type |= SKB_GSO_TUNNEL_REMCSUM;
815
816 return err;
817 }
818 EXPORT_SYMBOL(udp_gro_complete);
819
udp4_gro_complete(struct sk_buff * skb,int nhoff)820 INDIRECT_CALLABLE_SCOPE int udp4_gro_complete(struct sk_buff *skb, int nhoff)
821 {
822 const u16 offset = NAPI_GRO_CB(skb)->network_offsets[skb->encapsulation];
823 const struct iphdr *iph = (struct iphdr *)(skb->data + offset);
824 struct udphdr *uh = (struct udphdr *)(skb->data + nhoff);
825
826 /* do fraglist only if there is no outer UDP encap (or we already processed it) */
827 if (NAPI_GRO_CB(skb)->is_flist && !NAPI_GRO_CB(skb)->encap_mark) {
828 uh->len = htons(skb->len - nhoff);
829
830 skb_shinfo(skb)->gso_type |= (SKB_GSO_FRAGLIST|SKB_GSO_UDP_L4);
831 skb_shinfo(skb)->gso_segs = NAPI_GRO_CB(skb)->count;
832
833 __skb_incr_checksum_unnecessary(skb);
834
835 return 0;
836 }
837
838 if (uh->check)
839 uh->check = ~udp_v4_check(skb->len - nhoff, iph->saddr,
840 iph->daddr, 0);
841
842 return udp_gro_complete(skb, nhoff, udp4_lib_lookup_skb);
843 }
844
845 static const struct net_offload udpv4_offload = {
846 .callbacks = {
847 .gso_segment = udp4_ufo_fragment,
848 .gro_receive = udp4_gro_receive,
849 .gro_complete = udp4_gro_complete,
850 },
851 };
852
udpv4_offload_init(void)853 int __init udpv4_offload_init(void)
854 {
855 return inet_add_offload(&udpv4_offload, IPPROTO_UDP);
856 }
857