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/udp.h>
12 #include <net/protocol.h>
13 #include <net/inet_common.h>
14 
__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)15 static struct sk_buff *__skb_udp_tunnel_segment(struct sk_buff *skb,
16 	netdev_features_t features,
17 	struct sk_buff *(*gso_inner_segment)(struct sk_buff *skb,
18 					     netdev_features_t features),
19 	__be16 new_protocol, bool is_ipv6)
20 {
21 	int tnl_hlen = skb_inner_mac_header(skb) - skb_transport_header(skb);
22 	bool remcsum, need_csum, offload_csum, gso_partial;
23 	struct sk_buff *segs = ERR_PTR(-EINVAL);
24 	struct udphdr *uh = udp_hdr(skb);
25 	u16 mac_offset = skb->mac_header;
26 	__be16 protocol = skb->protocol;
27 	u16 mac_len = skb->mac_len;
28 	int udp_offset, outer_hlen;
29 	__wsum partial;
30 	bool need_ipsec;
31 
32 	if (unlikely(!pskb_may_pull(skb, tnl_hlen)))
33 		goto out;
34 
35 	/* Adjust partial header checksum to negate old length.
36 	 * We cannot rely on the value contained in uh->len as it is
37 	 * possible that the actual value exceeds the boundaries of the
38 	 * 16 bit length field due to the header being added outside of an
39 	 * IP or IPv6 frame that was already limited to 64K - 1.
40 	 */
41 	if (skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)
42 		partial = (__force __wsum)uh->len;
43 	else
44 		partial = (__force __wsum)htonl(skb->len);
45 	partial = csum_sub(csum_unfold(uh->check), partial);
46 
47 	/* setup inner skb. */
48 	skb->encapsulation = 0;
49 	SKB_GSO_CB(skb)->encap_level = 0;
50 	__skb_pull(skb, tnl_hlen);
51 	skb_reset_mac_header(skb);
52 	skb_set_network_header(skb, skb_inner_network_offset(skb));
53 	skb_set_transport_header(skb, skb_inner_transport_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 	need_ipsec = skb_dst(skb) && dst_xfrm(skb_dst(skb));
64 	/* Try to offload checksum if possible */
65 	offload_csum = !!(need_csum &&
66 			  !need_ipsec &&
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 	if (need_csum)
73 		features &= ~NETIF_F_SCTP_CRC;
74 
75 	/* The only checksum offload we care about from here on out is the
76 	 * outer one so strip the existing checksum feature flags and
77 	 * instead set the flag based on our outer checksum offload value.
78 	 */
79 	if (remcsum) {
80 		features &= ~NETIF_F_CSUM_MASK;
81 		if (!need_csum || offload_csum)
82 			features |= NETIF_F_HW_CSUM;
83 	}
84 
85 	/* segment inner packet. */
86 	segs = gso_inner_segment(skb, features);
87 	if (IS_ERR_OR_NULL(segs)) {
88 		skb_gso_error_unwind(skb, protocol, tnl_hlen, mac_offset,
89 				     mac_len);
90 		goto out;
91 	}
92 
93 	gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
94 
95 	outer_hlen = skb_tnl_header_len(skb);
96 	udp_offset = outer_hlen - tnl_hlen;
97 	skb = segs;
98 	do {
99 		unsigned int len;
100 
101 		if (remcsum)
102 			skb->ip_summed = CHECKSUM_NONE;
103 
104 		/* Set up inner headers if we are offloading inner checksum */
105 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
106 			skb_reset_inner_headers(skb);
107 			skb->encapsulation = 1;
108 		}
109 
110 		skb->mac_len = mac_len;
111 		skb->protocol = protocol;
112 
113 		__skb_push(skb, outer_hlen);
114 		skb_reset_mac_header(skb);
115 		skb_set_network_header(skb, mac_len);
116 		skb_set_transport_header(skb, udp_offset);
117 		len = skb->len - udp_offset;
118 		uh = udp_hdr(skb);
119 
120 		/* If we are only performing partial GSO the inner header
121 		 * will be using a length value equal to only one MSS sized
122 		 * segment instead of the entire frame.
123 		 */
124 		if (gso_partial && skb_is_gso(skb)) {
125 			uh->len = htons(skb_shinfo(skb)->gso_size +
126 					SKB_GSO_CB(skb)->data_offset +
127 					skb->head - (unsigned char *)uh);
128 		} else {
129 			uh->len = htons(len);
130 		}
131 
132 		if (!need_csum)
133 			continue;
134 
135 		uh->check = ~csum_fold(csum_add(partial,
136 				       (__force __wsum)htonl(len)));
137 
138 		if (skb->encapsulation || !offload_csum) {
139 			uh->check = gso_make_checksum(skb, ~uh->check);
140 			if (uh->check == 0)
141 				uh->check = CSUM_MANGLED_0;
142 		} else {
143 			skb->ip_summed = CHECKSUM_PARTIAL;
144 			skb->csum_start = skb_transport_header(skb) - skb->head;
145 			skb->csum_offset = offsetof(struct udphdr, check);
146 		}
147 	} while ((skb = skb->next));
148 out:
149 	return segs;
150 }
151 
skb_udp_tunnel_segment(struct sk_buff * skb,netdev_features_t features,bool is_ipv6)152 struct sk_buff *skb_udp_tunnel_segment(struct sk_buff *skb,
153 				       netdev_features_t features,
154 				       bool is_ipv6)
155 {
156 	const struct net_offload __rcu **offloads;
157 	__be16 protocol = skb->protocol;
158 	const struct net_offload *ops;
159 	struct sk_buff *segs = ERR_PTR(-EINVAL);
160 	struct sk_buff *(*gso_inner_segment)(struct sk_buff *skb,
161 					     netdev_features_t features);
162 
163 	rcu_read_lock();
164 
165 	switch (skb->inner_protocol_type) {
166 	case ENCAP_TYPE_ETHER:
167 		protocol = skb->inner_protocol;
168 		gso_inner_segment = skb_mac_gso_segment;
169 		break;
170 	case ENCAP_TYPE_IPPROTO:
171 		offloads = is_ipv6 ? inet6_offloads : inet_offloads;
172 		ops = rcu_dereference(offloads[skb->inner_ipproto]);
173 		if (!ops || !ops->callbacks.gso_segment)
174 			goto out_unlock;
175 		gso_inner_segment = ops->callbacks.gso_segment;
176 		break;
177 	default:
178 		goto out_unlock;
179 	}
180 
181 	segs = __skb_udp_tunnel_segment(skb, features, gso_inner_segment,
182 					protocol, is_ipv6);
183 
184 out_unlock:
185 	rcu_read_unlock();
186 
187 	return segs;
188 }
189 EXPORT_SYMBOL(skb_udp_tunnel_segment);
190 
__udpv4_gso_segment_csum(struct sk_buff * seg,__be32 * oldip,__be32 * newip,__be16 * oldport,__be16 * newport)191 static void __udpv4_gso_segment_csum(struct sk_buff *seg,
192 				     __be32 *oldip, __be32 *newip,
193 				     __be16 *oldport, __be16 *newport)
194 {
195 	struct udphdr *uh;
196 	struct iphdr *iph;
197 
198 	if (*oldip == *newip && *oldport == *newport)
199 		return;
200 
201 	uh = udp_hdr(seg);
202 	iph = ip_hdr(seg);
203 
204 	if (uh->check) {
205 		inet_proto_csum_replace4(&uh->check, seg, *oldip, *newip,
206 					 true);
207 		inet_proto_csum_replace2(&uh->check, seg, *oldport, *newport,
208 					 false);
209 		if (!uh->check)
210 			uh->check = CSUM_MANGLED_0;
211 	}
212 	*oldport = *newport;
213 
214 	csum_replace4(&iph->check, *oldip, *newip);
215 	*oldip = *newip;
216 }
217 
__udpv4_gso_segment_list_csum(struct sk_buff * segs)218 static struct sk_buff *__udpv4_gso_segment_list_csum(struct sk_buff *segs)
219 {
220 	struct sk_buff *seg;
221 	struct udphdr *uh, *uh2;
222 	struct iphdr *iph, *iph2;
223 
224 	seg = segs;
225 	uh = udp_hdr(seg);
226 	iph = ip_hdr(seg);
227 
228 	if ((udp_hdr(seg)->dest == udp_hdr(seg->next)->dest) &&
229 	    (udp_hdr(seg)->source == udp_hdr(seg->next)->source) &&
230 	    (ip_hdr(seg)->daddr == ip_hdr(seg->next)->daddr) &&
231 	    (ip_hdr(seg)->saddr == ip_hdr(seg->next)->saddr))
232 		return segs;
233 
234 	while ((seg = seg->next)) {
235 		uh2 = udp_hdr(seg);
236 		iph2 = ip_hdr(seg);
237 
238 		__udpv4_gso_segment_csum(seg,
239 					 &iph2->saddr, &iph->saddr,
240 					 &uh2->source, &uh->source);
241 		__udpv4_gso_segment_csum(seg,
242 					 &iph2->daddr, &iph->daddr,
243 					 &uh2->dest, &uh->dest);
244 	}
245 
246 	return segs;
247 }
248 
__udp_gso_segment_list(struct sk_buff * skb,netdev_features_t features,bool is_ipv6)249 static struct sk_buff *__udp_gso_segment_list(struct sk_buff *skb,
250 					      netdev_features_t features,
251 					      bool is_ipv6)
252 {
253 	unsigned int mss = skb_shinfo(skb)->gso_size;
254 
255 	skb = skb_segment_list(skb, features, skb_mac_header_len(skb));
256 	if (IS_ERR(skb))
257 		return skb;
258 
259 	udp_hdr(skb)->len = htons(sizeof(struct udphdr) + mss);
260 
261 	return is_ipv6 ? skb : __udpv4_gso_segment_list_csum(skb);
262 }
263 
__udp_gso_segment(struct sk_buff * gso_skb,netdev_features_t features,bool is_ipv6)264 struct sk_buff *__udp_gso_segment(struct sk_buff *gso_skb,
265 				  netdev_features_t features, bool is_ipv6)
266 {
267 	struct sock *sk = gso_skb->sk;
268 	unsigned int sum_truesize = 0;
269 	struct sk_buff *segs, *seg;
270 	struct udphdr *uh;
271 	unsigned int mss;
272 	bool copy_dtor;
273 	__sum16 check;
274 	__be16 newlen;
275 
276 	if (skb_shinfo(gso_skb)->gso_type & SKB_GSO_FRAGLIST)
277 		return __udp_gso_segment_list(gso_skb, features, is_ipv6);
278 
279 	mss = skb_shinfo(gso_skb)->gso_size;
280 	if (gso_skb->len <= sizeof(*uh) + mss)
281 		return ERR_PTR(-EINVAL);
282 
283 	skb_pull(gso_skb, sizeof(*uh));
284 
285 	/* clear destructor to avoid skb_segment assigning it to tail */
286 	copy_dtor = gso_skb->destructor == sock_wfree;
287 	if (copy_dtor)
288 		gso_skb->destructor = NULL;
289 
290 	segs = skb_segment(gso_skb, features);
291 	if (IS_ERR_OR_NULL(segs)) {
292 		if (copy_dtor)
293 			gso_skb->destructor = sock_wfree;
294 		return segs;
295 	}
296 
297 	/* GSO partial and frag_list segmentation only requires splitting
298 	 * the frame into an MSS multiple and possibly a remainder, both
299 	 * cases return a GSO skb. So update the mss now.
300 	 */
301 	if (skb_is_gso(segs))
302 		mss *= skb_shinfo(segs)->gso_segs;
303 
304 	seg = segs;
305 	uh = udp_hdr(seg);
306 
307 	/* preserve TX timestamp flags and TS key for first segment */
308 	skb_shinfo(seg)->tskey = skb_shinfo(gso_skb)->tskey;
309 	skb_shinfo(seg)->tx_flags |=
310 			(skb_shinfo(gso_skb)->tx_flags & SKBTX_ANY_TSTAMP);
311 
312 	/* compute checksum adjustment based on old length versus new */
313 	newlen = htons(sizeof(*uh) + mss);
314 	check = csum16_add(csum16_sub(uh->check, uh->len), newlen);
315 
316 	for (;;) {
317 		if (copy_dtor) {
318 			seg->destructor = sock_wfree;
319 			seg->sk = sk;
320 			sum_truesize += seg->truesize;
321 		}
322 
323 		if (!seg->next)
324 			break;
325 
326 		uh->len = newlen;
327 		uh->check = check;
328 
329 		if (seg->ip_summed == CHECKSUM_PARTIAL)
330 			gso_reset_checksum(seg, ~check);
331 		else
332 			uh->check = gso_make_checksum(seg, ~check) ? :
333 				    CSUM_MANGLED_0;
334 
335 		seg = seg->next;
336 		uh = udp_hdr(seg);
337 	}
338 
339 	/* last packet can be partial gso_size, account for that in checksum */
340 	newlen = htons(skb_tail_pointer(seg) - skb_transport_header(seg) +
341 		       seg->data_len);
342 	check = csum16_add(csum16_sub(uh->check, uh->len), newlen);
343 
344 	uh->len = newlen;
345 	uh->check = check;
346 
347 	if (seg->ip_summed == CHECKSUM_PARTIAL)
348 		gso_reset_checksum(seg, ~check);
349 	else
350 		uh->check = gso_make_checksum(seg, ~check) ? : CSUM_MANGLED_0;
351 
352 	/* update refcount for the packet */
353 	if (copy_dtor) {
354 		int delta = sum_truesize - gso_skb->truesize;
355 
356 		/* In some pathological cases, delta can be negative.
357 		 * We need to either use refcount_add() or refcount_sub_and_test()
358 		 */
359 		if (likely(delta >= 0))
360 			refcount_add(delta, &sk->sk_wmem_alloc);
361 		else
362 			WARN_ON_ONCE(refcount_sub_and_test(-delta, &sk->sk_wmem_alloc));
363 	}
364 	return segs;
365 }
366 EXPORT_SYMBOL_GPL(__udp_gso_segment);
367 
udp4_ufo_fragment(struct sk_buff * skb,netdev_features_t features)368 static struct sk_buff *udp4_ufo_fragment(struct sk_buff *skb,
369 					 netdev_features_t features)
370 {
371 	struct sk_buff *segs = ERR_PTR(-EINVAL);
372 	unsigned int mss;
373 	__wsum csum;
374 	struct udphdr *uh;
375 	struct iphdr *iph;
376 
377 	if (skb->encapsulation &&
378 	    (skb_shinfo(skb)->gso_type &
379 	     (SKB_GSO_UDP_TUNNEL|SKB_GSO_UDP_TUNNEL_CSUM))) {
380 		segs = skb_udp_tunnel_segment(skb, features, false);
381 		goto out;
382 	}
383 
384 	if (!(skb_shinfo(skb)->gso_type & (SKB_GSO_UDP | SKB_GSO_UDP_L4)))
385 		goto out;
386 
387 	if (!pskb_may_pull(skb, sizeof(struct udphdr)))
388 		goto out;
389 
390 	if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4 &&
391 	    !skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST))
392 		return __udp_gso_segment(skb, features, false);
393 
394 	mss = skb_shinfo(skb)->gso_size;
395 	if (unlikely(skb->len <= mss))
396 		goto out;
397 
398 	/* Do software UFO. Complete and fill in the UDP checksum as
399 	 * HW cannot do checksum of UDP packets sent as multiple
400 	 * IP fragments.
401 	 */
402 
403 	uh = udp_hdr(skb);
404 	iph = ip_hdr(skb);
405 
406 	uh->check = 0;
407 	csum = skb_checksum(skb, 0, skb->len, 0);
408 	uh->check = udp_v4_check(skb->len, iph->saddr, iph->daddr, csum);
409 	if (uh->check == 0)
410 		uh->check = CSUM_MANGLED_0;
411 
412 	skb->ip_summed = CHECKSUM_UNNECESSARY;
413 
414 	/* If there is no outer header we can fake a checksum offload
415 	 * due to the fact that we have already done the checksum in
416 	 * software prior to segmenting the frame.
417 	 */
418 	if (!skb->encap_hdr_csum)
419 		features |= NETIF_F_HW_CSUM;
420 
421 	/* Fragment the skb. IP headers of the fragments are updated in
422 	 * inet_gso_segment()
423 	 */
424 	segs = skb_segment(skb, features);
425 out:
426 	return segs;
427 }
428 
skb_gro_receive_list(struct sk_buff * p,struct sk_buff * skb)429 static int skb_gro_receive_list(struct sk_buff *p, struct sk_buff *skb)
430 {
431 	if (unlikely(p->len + skb->len >= 65536))
432 		return -E2BIG;
433 
434 	if (NAPI_GRO_CB(p)->last == p)
435 		skb_shinfo(p)->frag_list = skb;
436 	else
437 		NAPI_GRO_CB(p)->last->next = skb;
438 
439 	skb_pull(skb, skb_gro_offset(skb));
440 
441 	NAPI_GRO_CB(p)->last = skb;
442 	NAPI_GRO_CB(p)->count++;
443 	p->data_len += skb->len;
444 
445 	/* sk owenrship - if any - completely transferred to the aggregated packet */
446 	skb->destructor = NULL;
447 	p->truesize += skb->truesize;
448 	p->len += skb->len;
449 
450 	NAPI_GRO_CB(skb)->same_flow = 1;
451 
452 	return 0;
453 }
454 
455 
456 #define UDP_GRO_CNT_MAX 64
udp_gro_receive_segment(struct list_head * head,struct sk_buff * skb)457 static struct sk_buff *udp_gro_receive_segment(struct list_head *head,
458 					       struct sk_buff *skb)
459 {
460 	struct udphdr *uh = udp_gro_udphdr(skb);
461 	struct sk_buff *pp = NULL;
462 	struct udphdr *uh2;
463 	struct sk_buff *p;
464 	unsigned int ulen;
465 	int ret = 0;
466 
467 	/* requires non zero csum, for symmetry with GSO */
468 	if (!uh->check) {
469 		NAPI_GRO_CB(skb)->flush = 1;
470 		return NULL;
471 	}
472 
473 	/* Do not deal with padded or malicious packets, sorry ! */
474 	ulen = ntohs(uh->len);
475 	if (ulen <= sizeof(*uh) || ulen != skb_gro_len(skb)) {
476 		NAPI_GRO_CB(skb)->flush = 1;
477 		return NULL;
478 	}
479 	/* pull encapsulating udp header */
480 	skb_gro_pull(skb, sizeof(struct udphdr));
481 
482 	list_for_each_entry(p, head, list) {
483 		if (!NAPI_GRO_CB(p)->same_flow)
484 			continue;
485 
486 		uh2 = udp_hdr(p);
487 
488 		/* Match ports only, as csum is always non zero */
489 		if ((*(u32 *)&uh->source != *(u32 *)&uh2->source)) {
490 			NAPI_GRO_CB(p)->same_flow = 0;
491 			continue;
492 		}
493 
494 		if (NAPI_GRO_CB(skb)->is_flist != NAPI_GRO_CB(p)->is_flist) {
495 			NAPI_GRO_CB(skb)->flush = 1;
496 			return p;
497 		}
498 
499 		/* Terminate the flow on len mismatch or if it grow "too much".
500 		 * Under small packet flood GRO count could elsewhere grow a lot
501 		 * leading to excessive truesize values.
502 		 * On len mismatch merge the first packet shorter than gso_size,
503 		 * otherwise complete the GRO packet.
504 		 */
505 		if (ulen > ntohs(uh2->len)) {
506 			pp = p;
507 		} else {
508 			if (NAPI_GRO_CB(skb)->is_flist) {
509 				if (!pskb_may_pull(skb, skb_gro_offset(skb))) {
510 					NAPI_GRO_CB(skb)->flush = 1;
511 					return NULL;
512 				}
513 				if ((skb->ip_summed != p->ip_summed) ||
514 				    (skb->csum_level != p->csum_level)) {
515 					NAPI_GRO_CB(skb)->flush = 1;
516 					return NULL;
517 				}
518 				ret = skb_gro_receive_list(p, skb);
519 			} else {
520 				skb_gro_postpull_rcsum(skb, uh,
521 						       sizeof(struct udphdr));
522 
523 				ret = skb_gro_receive(p, skb);
524 			}
525 		}
526 
527 		if (ret || ulen != ntohs(uh2->len) ||
528 		    NAPI_GRO_CB(p)->count >= UDP_GRO_CNT_MAX)
529 			pp = p;
530 
531 		return pp;
532 	}
533 
534 	/* mismatch, but we never need to flush */
535 	return NULL;
536 }
537 
udp_gro_receive(struct list_head * head,struct sk_buff * skb,struct udphdr * uh,struct sock * sk)538 struct sk_buff *udp_gro_receive(struct list_head *head, struct sk_buff *skb,
539 				struct udphdr *uh, struct sock *sk)
540 {
541 	struct sk_buff *pp = NULL;
542 	struct sk_buff *p;
543 	struct udphdr *uh2;
544 	unsigned int off = skb_gro_offset(skb);
545 	int flush = 1;
546 
547 	/* we can do L4 aggregation only if the packet can't land in a tunnel
548 	 * otherwise we could corrupt the inner stream
549 	 */
550 	NAPI_GRO_CB(skb)->is_flist = 0;
551 	if (!sk || !udp_sk(sk)->gro_receive) {
552 		if (skb->dev->features & NETIF_F_GRO_FRAGLIST)
553 			NAPI_GRO_CB(skb)->is_flist = sk ? !udp_sk(sk)->gro_enabled : 1;
554 
555 		if ((!sk && (skb->dev->features & NETIF_F_GRO_UDP_FWD)) ||
556 		    (sk && udp_sk(sk)->gro_enabled) || NAPI_GRO_CB(skb)->is_flist)
557 			return call_gro_receive(udp_gro_receive_segment, head, skb);
558 
559 		/* no GRO, be sure flush the current packet */
560 		goto out;
561 	}
562 
563 	if (NAPI_GRO_CB(skb)->encap_mark ||
564 	    (uh->check && skb->ip_summed != CHECKSUM_PARTIAL &&
565 	     NAPI_GRO_CB(skb)->csum_cnt == 0 &&
566 	     !NAPI_GRO_CB(skb)->csum_valid))
567 		goto out;
568 
569 	/* mark that this skb passed once through the tunnel gro layer */
570 	NAPI_GRO_CB(skb)->encap_mark = 1;
571 
572 	flush = 0;
573 
574 	list_for_each_entry(p, head, list) {
575 		if (!NAPI_GRO_CB(p)->same_flow)
576 			continue;
577 
578 		uh2 = (struct udphdr   *)(p->data + off);
579 
580 		/* Match ports and either checksums are either both zero
581 		 * or nonzero.
582 		 */
583 		if ((*(u32 *)&uh->source != *(u32 *)&uh2->source) ||
584 		    (!uh->check ^ !uh2->check)) {
585 			NAPI_GRO_CB(p)->same_flow = 0;
586 			continue;
587 		}
588 	}
589 
590 	skb_gro_pull(skb, sizeof(struct udphdr)); /* pull encapsulating udp header */
591 	skb_gro_postpull_rcsum(skb, uh, sizeof(struct udphdr));
592 	pp = call_gro_receive_sk(udp_sk(sk)->gro_receive, sk, head, skb);
593 
594 out:
595 	skb_gro_flush_final(skb, pp, flush);
596 	return pp;
597 }
598 EXPORT_SYMBOL(udp_gro_receive);
599 
udp4_gro_lookup_skb(struct sk_buff * skb,__be16 sport,__be16 dport)600 static struct sock *udp4_gro_lookup_skb(struct sk_buff *skb, __be16 sport,
601 					__be16 dport)
602 {
603 	const struct iphdr *iph = skb_gro_network_header(skb);
604 	struct net *net = dev_net(skb->dev);
605 
606 	return __udp4_lib_lookup(net, iph->saddr, sport,
607 				 iph->daddr, dport, inet_iif(skb),
608 				 inet_sdif(skb), net->ipv4.udp_table, NULL);
609 }
610 
611 INDIRECT_CALLABLE_SCOPE
udp4_gro_receive(struct list_head * head,struct sk_buff * skb)612 struct sk_buff *udp4_gro_receive(struct list_head *head, struct sk_buff *skb)
613 {
614 	struct udphdr *uh = udp_gro_udphdr(skb);
615 	struct sock *sk = NULL;
616 	struct sk_buff *pp;
617 
618 	if (unlikely(!uh))
619 		goto flush;
620 
621 	/* Don't bother verifying checksum if we're going to flush anyway. */
622 	if (NAPI_GRO_CB(skb)->flush)
623 		goto skip;
624 
625 	if (skb_gro_checksum_validate_zero_check(skb, IPPROTO_UDP, uh->check,
626 						 inet_gro_compute_pseudo))
627 		goto flush;
628 	else if (uh->check)
629 		skb_gro_checksum_try_convert(skb, IPPROTO_UDP,
630 					     inet_gro_compute_pseudo);
631 skip:
632 	NAPI_GRO_CB(skb)->is_ipv6 = 0;
633 
634 	if (static_branch_unlikely(&udp_encap_needed_key))
635 		sk = udp4_gro_lookup_skb(skb, uh->source, uh->dest);
636 
637 	pp = udp_gro_receive(head, skb, uh, sk);
638 	return pp;
639 
640 flush:
641 	NAPI_GRO_CB(skb)->flush = 1;
642 	return NULL;
643 }
644 
udp_gro_complete_segment(struct sk_buff * skb)645 static int udp_gro_complete_segment(struct sk_buff *skb)
646 {
647 	struct udphdr *uh = udp_hdr(skb);
648 
649 	skb->csum_start = (unsigned char *)uh - skb->head;
650 	skb->csum_offset = offsetof(struct udphdr, check);
651 	skb->ip_summed = CHECKSUM_PARTIAL;
652 
653 	skb_shinfo(skb)->gso_segs = NAPI_GRO_CB(skb)->count;
654 	skb_shinfo(skb)->gso_type |= SKB_GSO_UDP_L4;
655 
656 	if (skb->encapsulation)
657 		skb->inner_transport_header = skb->transport_header;
658 
659 	return 0;
660 }
661 
udp_gro_complete(struct sk_buff * skb,int nhoff,udp_lookup_t lookup)662 int udp_gro_complete(struct sk_buff *skb, int nhoff,
663 		     udp_lookup_t lookup)
664 {
665 	__be16 newlen = htons(skb->len - nhoff);
666 	struct udphdr *uh = (struct udphdr *)(skb->data + nhoff);
667 	struct sock *sk;
668 	int err;
669 
670 	uh->len = newlen;
671 
672 	sk = INDIRECT_CALL_INET(lookup, udp6_lib_lookup_skb,
673 				udp4_lib_lookup_skb, skb, uh->source, uh->dest);
674 	if (sk && udp_sk(sk)->gro_complete) {
675 		skb_shinfo(skb)->gso_type = uh->check ? SKB_GSO_UDP_TUNNEL_CSUM
676 					: SKB_GSO_UDP_TUNNEL;
677 
678 		/* clear the encap mark, so that inner frag_list gro_complete
679 		 * can take place
680 		 */
681 		NAPI_GRO_CB(skb)->encap_mark = 0;
682 
683 		/* Set encapsulation before calling into inner gro_complete()
684 		 * functions to make them set up the inner offsets.
685 		 */
686 		skb->encapsulation = 1;
687 		err = udp_sk(sk)->gro_complete(sk, skb,
688 				nhoff + sizeof(struct udphdr));
689 	} else {
690 		err = udp_gro_complete_segment(skb);
691 	}
692 
693 	if (skb->remcsum_offload)
694 		skb_shinfo(skb)->gso_type |= SKB_GSO_TUNNEL_REMCSUM;
695 
696 	return err;
697 }
698 EXPORT_SYMBOL(udp_gro_complete);
699 
udp4_gro_complete(struct sk_buff * skb,int nhoff)700 INDIRECT_CALLABLE_SCOPE int udp4_gro_complete(struct sk_buff *skb, int nhoff)
701 {
702 	const struct iphdr *iph = ip_hdr(skb);
703 	struct udphdr *uh = (struct udphdr *)(skb->data + nhoff);
704 
705 	/* do fraglist only if there is no outer UDP encap (or we already processed it) */
706 	if (NAPI_GRO_CB(skb)->is_flist && !NAPI_GRO_CB(skb)->encap_mark) {
707 		uh->len = htons(skb->len - nhoff);
708 
709 		skb_shinfo(skb)->gso_type |= (SKB_GSO_FRAGLIST|SKB_GSO_UDP_L4);
710 		skb_shinfo(skb)->gso_segs = NAPI_GRO_CB(skb)->count;
711 
712 		if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
713 			if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
714 				skb->csum_level++;
715 		} else {
716 			skb->ip_summed = CHECKSUM_UNNECESSARY;
717 			skb->csum_level = 0;
718 		}
719 
720 		return 0;
721 	}
722 
723 	if (uh->check)
724 		uh->check = ~udp_v4_check(skb->len - nhoff, iph->saddr,
725 					  iph->daddr, 0);
726 
727 	return udp_gro_complete(skb, nhoff, udp4_lib_lookup_skb);
728 }
729 
730 static const struct net_offload udpv4_offload = {
731 	.callbacks = {
732 		.gso_segment = udp4_ufo_fragment,
733 		.gro_receive  =	udp4_gro_receive,
734 		.gro_complete =	udp4_gro_complete,
735 	},
736 };
737 
udpv4_offload_init(void)738 int __init udpv4_offload_init(void)
739 {
740 	return inet_add_offload(&udpv4_offload, IPPROTO_UDP);
741 }
742