xref: /openbmc/linux/net/ipv4/udp_offload.c (revision af3122f5)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *	IPV4 GSO/GRO offload support
4  *	Linux INET implementation
5  *
6  *	UDPv4 GSO support
7  */
8 
9 #include <linux/skbuff.h>
10 #include <net/gro.h>
11 #include <net/gso.h>
12 #include <net/udp.h>
13 #include <net/protocol.h>
14 #include <net/inet_common.h>
15 
__skb_udp_tunnel_segment(struct sk_buff * skb,netdev_features_t features,struct sk_buff * (* gso_inner_segment)(struct sk_buff * skb,netdev_features_t features),__be16 new_protocol,bool is_ipv6)16 static struct sk_buff *__skb_udp_tunnel_segment(struct sk_buff *skb,
17 	netdev_features_t features,
18 	struct sk_buff *(*gso_inner_segment)(struct sk_buff *skb,
19 					     netdev_features_t features),
20 	__be16 new_protocol, bool is_ipv6)
21 {
22 	int tnl_hlen = skb_inner_mac_header(skb) - skb_transport_header(skb);
23 	bool remcsum, need_csum, offload_csum, gso_partial;
24 	struct sk_buff *segs = ERR_PTR(-EINVAL);
25 	struct udphdr *uh = udp_hdr(skb);
26 	u16 mac_offset = skb->mac_header;
27 	__be16 protocol = skb->protocol;
28 	u16 mac_len = skb->mac_len;
29 	int udp_offset, outer_hlen;
30 	__wsum partial;
31 	bool need_ipsec;
32 
33 	if (unlikely(!pskb_may_pull(skb, tnl_hlen)))
34 		goto out;
35 
36 	/* Adjust partial header checksum to negate old length.
37 	 * We cannot rely on the value contained in uh->len as it is
38 	 * possible that the actual value exceeds the boundaries of the
39 	 * 16 bit length field due to the header being added outside of an
40 	 * IP or IPv6 frame that was already limited to 64K - 1.
41 	 */
42 	if (skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)
43 		partial = (__force __wsum)uh->len;
44 	else
45 		partial = (__force __wsum)htonl(skb->len);
46 	partial = csum_sub(csum_unfold(uh->check), partial);
47 
48 	/* setup inner skb. */
49 	skb->encapsulation = 0;
50 	SKB_GSO_CB(skb)->encap_level = 0;
51 	__skb_pull(skb, tnl_hlen);
52 	skb_reset_mac_header(skb);
53 	skb_set_network_header(skb, skb_inner_network_offset(skb));
54 	skb_set_transport_header(skb, skb_inner_transport_offset(skb));
55 	skb->mac_len = skb_inner_network_offset(skb);
56 	skb->protocol = new_protocol;
57 
58 	need_csum = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_TUNNEL_CSUM);
59 	skb->encap_hdr_csum = need_csum;
60 
61 	remcsum = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TUNNEL_REMCSUM);
62 	skb->remcsum_offload = remcsum;
63 
64 	need_ipsec = skb_dst(skb) && dst_xfrm(skb_dst(skb));
65 	/* Try to offload checksum if possible */
66 	offload_csum = !!(need_csum &&
67 			  !need_ipsec &&
68 			  (skb->dev->features &
69 			   (is_ipv6 ? (NETIF_F_HW_CSUM | NETIF_F_IPV6_CSUM) :
70 				      (NETIF_F_HW_CSUM | NETIF_F_IP_CSUM))));
71 
72 	features &= skb->dev->hw_enc_features;
73 	if (need_csum)
74 		features &= ~NETIF_F_SCTP_CRC;
75 
76 	/* The only checksum offload we care about from here on out is the
77 	 * outer one so strip the existing checksum feature flags and
78 	 * instead set the flag based on our outer checksum offload value.
79 	 */
80 	if (remcsum) {
81 		features &= ~NETIF_F_CSUM_MASK;
82 		if (!need_csum || offload_csum)
83 			features |= NETIF_F_HW_CSUM;
84 	}
85 
86 	/* segment inner packet. */
87 	segs = gso_inner_segment(skb, features);
88 	if (IS_ERR_OR_NULL(segs)) {
89 		skb_gso_error_unwind(skb, protocol, tnl_hlen, mac_offset,
90 				     mac_len);
91 		goto out;
92 	}
93 
94 	gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
95 
96 	outer_hlen = skb_tnl_header_len(skb);
97 	udp_offset = outer_hlen - tnl_hlen;
98 	skb = segs;
99 	do {
100 		unsigned int len;
101 
102 		if (remcsum)
103 			skb->ip_summed = CHECKSUM_NONE;
104 
105 		/* Set up inner headers if we are offloading inner checksum */
106 		if (skb->ip_summed == CHECKSUM_PARTIAL) {
107 			skb_reset_inner_headers(skb);
108 			skb->encapsulation = 1;
109 		}
110 
111 		skb->mac_len = mac_len;
112 		skb->protocol = protocol;
113 
114 		__skb_push(skb, outer_hlen);
115 		skb_reset_mac_header(skb);
116 		skb_set_network_header(skb, mac_len);
117 		skb_set_transport_header(skb, udp_offset);
118 		len = skb->len - udp_offset;
119 		uh = udp_hdr(skb);
120 
121 		/* If we are only performing partial GSO the inner header
122 		 * will be using a length value equal to only one MSS sized
123 		 * segment instead of the entire frame.
124 		 */
125 		if (gso_partial && skb_is_gso(skb)) {
126 			uh->len = htons(skb_shinfo(skb)->gso_size +
127 					SKB_GSO_CB(skb)->data_offset +
128 					skb->head - (unsigned char *)uh);
129 		} else {
130 			uh->len = htons(len);
131 		}
132 
133 		if (!need_csum)
134 			continue;
135 
136 		uh->check = ~csum_fold(csum_add(partial,
137 				       (__force __wsum)htonl(len)));
138 
139 		if (skb->encapsulation || !offload_csum) {
140 			uh->check = gso_make_checksum(skb, ~uh->check);
141 			if (uh->check == 0)
142 				uh->check = CSUM_MANGLED_0;
143 		} else {
144 			skb->ip_summed = CHECKSUM_PARTIAL;
145 			skb->csum_start = skb_transport_header(skb) - skb->head;
146 			skb->csum_offset = offsetof(struct udphdr, check);
147 		}
148 	} while ((skb = skb->next));
149 out:
150 	return segs;
151 }
152 
skb_udp_tunnel_segment(struct sk_buff * skb,netdev_features_t features,bool is_ipv6)153 struct sk_buff *skb_udp_tunnel_segment(struct sk_buff *skb,
154 				       netdev_features_t features,
155 				       bool is_ipv6)
156 {
157 	const struct net_offload __rcu **offloads;
158 	__be16 protocol = skb->protocol;
159 	const struct net_offload *ops;
160 	struct sk_buff *segs = ERR_PTR(-EINVAL);
161 	struct sk_buff *(*gso_inner_segment)(struct sk_buff *skb,
162 					     netdev_features_t features);
163 
164 	rcu_read_lock();
165 
166 	switch (skb->inner_protocol_type) {
167 	case ENCAP_TYPE_ETHER:
168 		protocol = skb->inner_protocol;
169 		gso_inner_segment = skb_mac_gso_segment;
170 		break;
171 	case ENCAP_TYPE_IPPROTO:
172 		offloads = is_ipv6 ? inet6_offloads : inet_offloads;
173 		ops = rcu_dereference(offloads[skb->inner_ipproto]);
174 		if (!ops || !ops->callbacks.gso_segment)
175 			goto out_unlock;
176 		gso_inner_segment = ops->callbacks.gso_segment;
177 		break;
178 	default:
179 		goto out_unlock;
180 	}
181 
182 	segs = __skb_udp_tunnel_segment(skb, features, gso_inner_segment,
183 					protocol, is_ipv6);
184 
185 out_unlock:
186 	rcu_read_unlock();
187 
188 	return segs;
189 }
190 EXPORT_SYMBOL(skb_udp_tunnel_segment);
191 
__udpv4_gso_segment_csum(struct sk_buff * seg,__be32 * oldip,__be32 * newip,__be16 * oldport,__be16 * newport)192 static void __udpv4_gso_segment_csum(struct sk_buff *seg,
193 				     __be32 *oldip, __be32 *newip,
194 				     __be16 *oldport, __be16 *newport)
195 {
196 	struct udphdr *uh;
197 	struct iphdr *iph;
198 
199 	if (*oldip == *newip && *oldport == *newport)
200 		return;
201 
202 	uh = udp_hdr(seg);
203 	iph = ip_hdr(seg);
204 
205 	if (uh->check) {
206 		inet_proto_csum_replace4(&uh->check, seg, *oldip, *newip,
207 					 true);
208 		inet_proto_csum_replace2(&uh->check, seg, *oldport, *newport,
209 					 false);
210 		if (!uh->check)
211 			uh->check = CSUM_MANGLED_0;
212 	}
213 	*oldport = *newport;
214 
215 	csum_replace4(&iph->check, *oldip, *newip);
216 	*oldip = *newip;
217 }
218 
__udpv4_gso_segment_list_csum(struct sk_buff * segs)219 static struct sk_buff *__udpv4_gso_segment_list_csum(struct sk_buff *segs)
220 {
221 	struct sk_buff *seg;
222 	struct udphdr *uh, *uh2;
223 	struct iphdr *iph, *iph2;
224 
225 	seg = segs;
226 	uh = udp_hdr(seg);
227 	iph = ip_hdr(seg);
228 
229 	if ((udp_hdr(seg)->dest == udp_hdr(seg->next)->dest) &&
230 	    (udp_hdr(seg)->source == udp_hdr(seg->next)->source) &&
231 	    (ip_hdr(seg)->daddr == ip_hdr(seg->next)->daddr) &&
232 	    (ip_hdr(seg)->saddr == ip_hdr(seg->next)->saddr))
233 		return segs;
234 
235 	while ((seg = seg->next)) {
236 		uh2 = udp_hdr(seg);
237 		iph2 = ip_hdr(seg);
238 
239 		__udpv4_gso_segment_csum(seg,
240 					 &iph2->saddr, &iph->saddr,
241 					 &uh2->source, &uh->source);
242 		__udpv4_gso_segment_csum(seg,
243 					 &iph2->daddr, &iph->daddr,
244 					 &uh2->dest, &uh->dest);
245 	}
246 
247 	return segs;
248 }
249 
__udp_gso_segment_list(struct sk_buff * skb,netdev_features_t features,bool is_ipv6)250 static struct sk_buff *__udp_gso_segment_list(struct sk_buff *skb,
251 					      netdev_features_t features,
252 					      bool is_ipv6)
253 {
254 	unsigned int mss = skb_shinfo(skb)->gso_size;
255 
256 	skb = skb_segment_list(skb, features, skb_mac_header_len(skb));
257 	if (IS_ERR(skb))
258 		return skb;
259 
260 	udp_hdr(skb)->len = htons(sizeof(struct udphdr) + mss);
261 
262 	return is_ipv6 ? skb : __udpv4_gso_segment_list_csum(skb);
263 }
264 
__udp_gso_segment(struct sk_buff * gso_skb,netdev_features_t features,bool is_ipv6)265 struct sk_buff *__udp_gso_segment(struct sk_buff *gso_skb,
266 				  netdev_features_t features, bool is_ipv6)
267 {
268 	struct sock *sk = gso_skb->sk;
269 	unsigned int sum_truesize = 0;
270 	struct sk_buff *segs, *seg;
271 	struct udphdr *uh;
272 	unsigned int mss;
273 	bool copy_dtor;
274 	__sum16 check;
275 	__be16 newlen;
276 
277 	mss = skb_shinfo(gso_skb)->gso_size;
278 	if (gso_skb->len <= sizeof(*uh) + mss)
279 		return ERR_PTR(-EINVAL);
280 
281 	if (unlikely(skb_checksum_start(gso_skb) !=
282 		     skb_transport_header(gso_skb) &&
283 		     !(skb_shinfo(gso_skb)->gso_type & SKB_GSO_FRAGLIST)))
284 		return ERR_PTR(-EINVAL);
285 
286 	if (skb_gso_ok(gso_skb, features | NETIF_F_GSO_ROBUST)) {
287 		/* Packet is from an untrusted source, reset gso_segs. */
288 		skb_shinfo(gso_skb)->gso_segs = DIV_ROUND_UP(gso_skb->len - sizeof(*uh),
289 							     mss);
290 		return NULL;
291 	}
292 
293 	if (skb_shinfo(gso_skb)->gso_type & SKB_GSO_FRAGLIST) {
294 		 /* Detect modified geometry and pass those to skb_segment. */
295 		if (skb_pagelen(gso_skb) - sizeof(*uh) == skb_shinfo(gso_skb)->gso_size)
296 			return __udp_gso_segment_list(gso_skb, features, is_ipv6);
297 
298 		 /* Setup csum, as fraglist skips this in udp4_gro_receive. */
299 		gso_skb->csum_start = skb_transport_header(gso_skb) - gso_skb->head;
300 		gso_skb->csum_offset = offsetof(struct udphdr, check);
301 		gso_skb->ip_summed = CHECKSUM_PARTIAL;
302 
303 		uh = udp_hdr(gso_skb);
304 		if (is_ipv6)
305 			uh->check = ~udp_v6_check(gso_skb->len,
306 						  &ipv6_hdr(gso_skb)->saddr,
307 						  &ipv6_hdr(gso_skb)->daddr, 0);
308 		else
309 			uh->check = ~udp_v4_check(gso_skb->len,
310 						  ip_hdr(gso_skb)->saddr,
311 						  ip_hdr(gso_skb)->daddr, 0);
312 	}
313 
314 	skb_pull(gso_skb, sizeof(*uh));
315 
316 	/* clear destructor to avoid skb_segment assigning it to tail */
317 	copy_dtor = gso_skb->destructor == sock_wfree;
318 	if (copy_dtor)
319 		gso_skb->destructor = NULL;
320 
321 	segs = skb_segment(gso_skb, features);
322 	if (IS_ERR_OR_NULL(segs)) {
323 		if (copy_dtor)
324 			gso_skb->destructor = sock_wfree;
325 		return segs;
326 	}
327 
328 	/* GSO partial and frag_list segmentation only requires splitting
329 	 * the frame into an MSS multiple and possibly a remainder, both
330 	 * cases return a GSO skb. So update the mss now.
331 	 */
332 	if (skb_is_gso(segs))
333 		mss *= skb_shinfo(segs)->gso_segs;
334 
335 	seg = segs;
336 	uh = udp_hdr(seg);
337 
338 	/* preserve TX timestamp flags and TS key for first segment */
339 	skb_shinfo(seg)->tskey = skb_shinfo(gso_skb)->tskey;
340 	skb_shinfo(seg)->tx_flags |=
341 			(skb_shinfo(gso_skb)->tx_flags & SKBTX_ANY_TSTAMP);
342 
343 	/* compute checksum adjustment based on old length versus new */
344 	newlen = htons(sizeof(*uh) + mss);
345 	check = csum16_add(csum16_sub(uh->check, uh->len), newlen);
346 
347 	for (;;) {
348 		if (copy_dtor) {
349 			seg->destructor = sock_wfree;
350 			seg->sk = sk;
351 			sum_truesize += seg->truesize;
352 		}
353 
354 		if (!seg->next)
355 			break;
356 
357 		uh->len = newlen;
358 		uh->check = check;
359 
360 		if (seg->ip_summed == CHECKSUM_PARTIAL)
361 			gso_reset_checksum(seg, ~check);
362 		else
363 			uh->check = gso_make_checksum(seg, ~check) ? :
364 				    CSUM_MANGLED_0;
365 
366 		seg = seg->next;
367 		uh = udp_hdr(seg);
368 	}
369 
370 	/* last packet can be partial gso_size, account for that in checksum */
371 	newlen = htons(skb_tail_pointer(seg) - skb_transport_header(seg) +
372 		       seg->data_len);
373 	check = csum16_add(csum16_sub(uh->check, uh->len), newlen);
374 
375 	uh->len = newlen;
376 	uh->check = check;
377 
378 	if (seg->ip_summed == CHECKSUM_PARTIAL)
379 		gso_reset_checksum(seg, ~check);
380 	else
381 		uh->check = gso_make_checksum(seg, ~check) ? : CSUM_MANGLED_0;
382 
383 	/* update refcount for the packet */
384 	if (copy_dtor) {
385 		int delta = sum_truesize - gso_skb->truesize;
386 
387 		/* In some pathological cases, delta can be negative.
388 		 * We need to either use refcount_add() or refcount_sub_and_test()
389 		 */
390 		if (likely(delta >= 0))
391 			refcount_add(delta, &sk->sk_wmem_alloc);
392 		else
393 			WARN_ON_ONCE(refcount_sub_and_test(-delta, &sk->sk_wmem_alloc));
394 	}
395 	return segs;
396 }
397 EXPORT_SYMBOL_GPL(__udp_gso_segment);
398 
udp4_ufo_fragment(struct sk_buff * skb,netdev_features_t features)399 static struct sk_buff *udp4_ufo_fragment(struct sk_buff *skb,
400 					 netdev_features_t features)
401 {
402 	struct sk_buff *segs = ERR_PTR(-EINVAL);
403 	unsigned int mss;
404 	__wsum csum;
405 	struct udphdr *uh;
406 	struct iphdr *iph;
407 
408 	if (skb->encapsulation &&
409 	    (skb_shinfo(skb)->gso_type &
410 	     (SKB_GSO_UDP_TUNNEL|SKB_GSO_UDP_TUNNEL_CSUM))) {
411 		segs = skb_udp_tunnel_segment(skb, features, false);
412 		goto out;
413 	}
414 
415 	if (!(skb_shinfo(skb)->gso_type & (SKB_GSO_UDP | SKB_GSO_UDP_L4)))
416 		goto out;
417 
418 	if (!pskb_may_pull(skb, sizeof(struct udphdr)))
419 		goto out;
420 
421 	if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4)
422 		return __udp_gso_segment(skb, features, false);
423 
424 	mss = skb_shinfo(skb)->gso_size;
425 	if (unlikely(skb->len <= mss))
426 		goto out;
427 
428 	/* Do software UFO. Complete and fill in the UDP checksum as
429 	 * HW cannot do checksum of UDP packets sent as multiple
430 	 * IP fragments.
431 	 */
432 
433 	uh = udp_hdr(skb);
434 	iph = ip_hdr(skb);
435 
436 	uh->check = 0;
437 	csum = skb_checksum(skb, 0, skb->len, 0);
438 	uh->check = udp_v4_check(skb->len, iph->saddr, iph->daddr, csum);
439 	if (uh->check == 0)
440 		uh->check = CSUM_MANGLED_0;
441 
442 	skb->ip_summed = CHECKSUM_UNNECESSARY;
443 
444 	/* If there is no outer header we can fake a checksum offload
445 	 * due to the fact that we have already done the checksum in
446 	 * software prior to segmenting the frame.
447 	 */
448 	if (!skb->encap_hdr_csum)
449 		features |= NETIF_F_HW_CSUM;
450 
451 	/* Fragment the skb. IP headers of the fragments are updated in
452 	 * inet_gso_segment()
453 	 */
454 	segs = skb_segment(skb, features);
455 out:
456 	return segs;
457 }
458 
skb_gro_receive_list(struct sk_buff * p,struct sk_buff * skb)459 static int skb_gro_receive_list(struct sk_buff *p, struct sk_buff *skb)
460 {
461 	if (unlikely(p->len + skb->len >= 65536))
462 		return -E2BIG;
463 
464 	if (NAPI_GRO_CB(p)->last == p)
465 		skb_shinfo(p)->frag_list = skb;
466 	else
467 		NAPI_GRO_CB(p)->last->next = skb;
468 
469 	skb_pull(skb, skb_gro_offset(skb));
470 
471 	NAPI_GRO_CB(p)->last = skb;
472 	NAPI_GRO_CB(p)->count++;
473 	p->data_len += skb->len;
474 
475 	/* sk ownership - if any - completely transferred to the aggregated packet */
476 	skb->destructor = NULL;
477 	skb->sk = NULL;
478 	p->truesize += skb->truesize;
479 	p->len += skb->len;
480 
481 	NAPI_GRO_CB(skb)->same_flow = 1;
482 
483 	return 0;
484 }
485 
486 
487 #define UDP_GRO_CNT_MAX 64
udp_gro_receive_segment(struct list_head * head,struct sk_buff * skb)488 static struct sk_buff *udp_gro_receive_segment(struct list_head *head,
489 					       struct sk_buff *skb)
490 {
491 	struct udphdr *uh = udp_gro_udphdr(skb);
492 	struct sk_buff *pp = NULL;
493 	struct udphdr *uh2;
494 	struct sk_buff *p;
495 	unsigned int ulen;
496 	int ret = 0;
497 	int flush;
498 
499 	/* requires non zero csum, for symmetry with GSO */
500 	if (!uh->check) {
501 		NAPI_GRO_CB(skb)->flush = 1;
502 		return NULL;
503 	}
504 
505 	/* Do not deal with padded or malicious packets, sorry ! */
506 	ulen = ntohs(uh->len);
507 	if (ulen <= sizeof(*uh) || ulen != skb_gro_len(skb)) {
508 		NAPI_GRO_CB(skb)->flush = 1;
509 		return NULL;
510 	}
511 	/* pull encapsulating udp header */
512 	skb_gro_pull(skb, sizeof(struct udphdr));
513 
514 	list_for_each_entry(p, head, list) {
515 		if (!NAPI_GRO_CB(p)->same_flow)
516 			continue;
517 
518 		uh2 = udp_hdr(p);
519 
520 		/* Match ports only, as csum is always non zero */
521 		if ((*(u32 *)&uh->source != *(u32 *)&uh2->source)) {
522 			NAPI_GRO_CB(p)->same_flow = 0;
523 			continue;
524 		}
525 
526 		if (NAPI_GRO_CB(skb)->is_flist != NAPI_GRO_CB(p)->is_flist) {
527 			NAPI_GRO_CB(skb)->flush = 1;
528 			return p;
529 		}
530 
531 		flush = NAPI_GRO_CB(p)->flush;
532 
533 		if (NAPI_GRO_CB(p)->flush_id != 1 ||
534 		    NAPI_GRO_CB(p)->count != 1 ||
535 		    !NAPI_GRO_CB(p)->is_atomic)
536 			flush |= NAPI_GRO_CB(p)->flush_id;
537 		else
538 			NAPI_GRO_CB(p)->is_atomic = false;
539 
540 		/* Terminate the flow on len mismatch or if it grow "too much".
541 		 * Under small packet flood GRO count could elsewhere grow a lot
542 		 * leading to excessive truesize values.
543 		 * On len mismatch merge the first packet shorter than gso_size,
544 		 * otherwise complete the GRO packet.
545 		 */
546 		if (ulen > ntohs(uh2->len) || flush) {
547 			pp = p;
548 		} else {
549 			if (NAPI_GRO_CB(skb)->is_flist) {
550 				if (!pskb_may_pull(skb, skb_gro_offset(skb))) {
551 					NAPI_GRO_CB(skb)->flush = 1;
552 					return NULL;
553 				}
554 				if ((skb->ip_summed != p->ip_summed) ||
555 				    (skb->csum_level != p->csum_level)) {
556 					NAPI_GRO_CB(skb)->flush = 1;
557 					return NULL;
558 				}
559 				ret = skb_gro_receive_list(p, skb);
560 			} else {
561 				skb_gro_postpull_rcsum(skb, uh,
562 						       sizeof(struct udphdr));
563 
564 				ret = skb_gro_receive(p, skb);
565 			}
566 		}
567 
568 		if (ret || ulen != ntohs(uh2->len) ||
569 		    NAPI_GRO_CB(p)->count >= UDP_GRO_CNT_MAX)
570 			pp = p;
571 
572 		return pp;
573 	}
574 
575 	/* mismatch, but we never need to flush */
576 	return NULL;
577 }
578 
udp_gro_receive(struct list_head * head,struct sk_buff * skb,struct udphdr * uh,struct sock * sk)579 struct sk_buff *udp_gro_receive(struct list_head *head, struct sk_buff *skb,
580 				struct udphdr *uh, struct sock *sk)
581 {
582 	struct sk_buff *pp = NULL;
583 	struct sk_buff *p;
584 	struct udphdr *uh2;
585 	unsigned int off = skb_gro_offset(skb);
586 	int flush = 1;
587 
588 	/* We can do L4 aggregation only if the packet can't land in a tunnel
589 	 * otherwise we could corrupt the inner stream. Detecting such packets
590 	 * cannot be foolproof and the aggregation might still happen in some
591 	 * cases. Such packets should be caught in udp_unexpected_gso later.
592 	 */
593 	NAPI_GRO_CB(skb)->is_flist = 0;
594 	if (!sk || !udp_sk(sk)->gro_receive) {
595 		/* If the packet was locally encapsulated in a UDP tunnel that
596 		 * wasn't detected above, do not GRO.
597 		 */
598 		if (skb->encapsulation)
599 			goto out;
600 
601 		if (skb->dev->features & NETIF_F_GRO_FRAGLIST)
602 			NAPI_GRO_CB(skb)->is_flist = sk ? !udp_test_bit(GRO_ENABLED, sk) : 1;
603 
604 		if ((!sk && (skb->dev->features & NETIF_F_GRO_UDP_FWD)) ||
605 		    (sk && udp_test_bit(GRO_ENABLED, sk)) || NAPI_GRO_CB(skb)->is_flist)
606 			return call_gro_receive(udp_gro_receive_segment, head, skb);
607 
608 		/* no GRO, be sure flush the current packet */
609 		goto out;
610 	}
611 
612 	if (NAPI_GRO_CB(skb)->encap_mark ||
613 	    (uh->check && skb->ip_summed != CHECKSUM_PARTIAL &&
614 	     NAPI_GRO_CB(skb)->csum_cnt == 0 &&
615 	     !NAPI_GRO_CB(skb)->csum_valid))
616 		goto out;
617 
618 	/* mark that this skb passed once through the tunnel gro layer */
619 	NAPI_GRO_CB(skb)->encap_mark = 1;
620 
621 	flush = 0;
622 
623 	list_for_each_entry(p, head, list) {
624 		if (!NAPI_GRO_CB(p)->same_flow)
625 			continue;
626 
627 		uh2 = (struct udphdr   *)(p->data + off);
628 
629 		/* Match ports and either checksums are either both zero
630 		 * or nonzero.
631 		 */
632 		if ((*(u32 *)&uh->source != *(u32 *)&uh2->source) ||
633 		    (!uh->check ^ !uh2->check)) {
634 			NAPI_GRO_CB(p)->same_flow = 0;
635 			continue;
636 		}
637 	}
638 
639 	skb_gro_pull(skb, sizeof(struct udphdr)); /* pull encapsulating udp header */
640 	skb_gro_postpull_rcsum(skb, uh, sizeof(struct udphdr));
641 	pp = call_gro_receive_sk(udp_sk(sk)->gro_receive, sk, head, skb);
642 
643 out:
644 	skb_gro_flush_final(skb, pp, flush);
645 	return pp;
646 }
647 EXPORT_SYMBOL(udp_gro_receive);
648 
udp4_gro_lookup_skb(struct sk_buff * skb,__be16 sport,__be16 dport)649 static struct sock *udp4_gro_lookup_skb(struct sk_buff *skb, __be16 sport,
650 					__be16 dport)
651 {
652 	const struct iphdr *iph = skb_gro_network_header(skb);
653 	struct net *net = dev_net(skb->dev);
654 	int iif, sdif;
655 
656 	inet_get_iif_sdif(skb, &iif, &sdif);
657 
658 	return __udp4_lib_lookup(net, iph->saddr, sport,
659 				 iph->daddr, dport, iif,
660 				 sdif, net->ipv4.udp_table, NULL);
661 }
662 
663 INDIRECT_CALLABLE_SCOPE
udp4_gro_receive(struct list_head * head,struct sk_buff * skb)664 struct sk_buff *udp4_gro_receive(struct list_head *head, struct sk_buff *skb)
665 {
666 	struct udphdr *uh = udp_gro_udphdr(skb);
667 	struct sock *sk = NULL;
668 	struct sk_buff *pp;
669 
670 	if (unlikely(!uh))
671 		goto flush;
672 
673 	/* Don't bother verifying checksum if we're going to flush anyway. */
674 	if (NAPI_GRO_CB(skb)->flush)
675 		goto skip;
676 
677 	if (skb_gro_checksum_validate_zero_check(skb, IPPROTO_UDP, uh->check,
678 						 inet_gro_compute_pseudo))
679 		goto flush;
680 	else if (uh->check)
681 		skb_gro_checksum_try_convert(skb, IPPROTO_UDP,
682 					     inet_gro_compute_pseudo);
683 skip:
684 	NAPI_GRO_CB(skb)->is_ipv6 = 0;
685 
686 	if (static_branch_unlikely(&udp_encap_needed_key))
687 		sk = udp4_gro_lookup_skb(skb, uh->source, uh->dest);
688 
689 	pp = udp_gro_receive(head, skb, uh, sk);
690 	return pp;
691 
692 flush:
693 	NAPI_GRO_CB(skb)->flush = 1;
694 	return NULL;
695 }
696 
udp_gro_complete_segment(struct sk_buff * skb)697 static int udp_gro_complete_segment(struct sk_buff *skb)
698 {
699 	struct udphdr *uh = udp_hdr(skb);
700 
701 	skb->csum_start = (unsigned char *)uh - skb->head;
702 	skb->csum_offset = offsetof(struct udphdr, check);
703 	skb->ip_summed = CHECKSUM_PARTIAL;
704 
705 	skb_shinfo(skb)->gso_segs = NAPI_GRO_CB(skb)->count;
706 	skb_shinfo(skb)->gso_type |= SKB_GSO_UDP_L4;
707 
708 	if (skb->encapsulation)
709 		skb->inner_transport_header = skb->transport_header;
710 
711 	return 0;
712 }
713 
udp_gro_complete(struct sk_buff * skb,int nhoff,udp_lookup_t lookup)714 int udp_gro_complete(struct sk_buff *skb, int nhoff,
715 		     udp_lookup_t lookup)
716 {
717 	__be16 newlen = htons(skb->len - nhoff);
718 	struct udphdr *uh = (struct udphdr *)(skb->data + nhoff);
719 	struct sock *sk;
720 	int err;
721 
722 	uh->len = newlen;
723 
724 	sk = INDIRECT_CALL_INET(lookup, udp6_lib_lookup_skb,
725 				udp4_lib_lookup_skb, skb, uh->source, uh->dest);
726 	if (sk && udp_sk(sk)->gro_complete) {
727 		skb_shinfo(skb)->gso_type = uh->check ? SKB_GSO_UDP_TUNNEL_CSUM
728 					: SKB_GSO_UDP_TUNNEL;
729 
730 		/* clear the encap mark, so that inner frag_list gro_complete
731 		 * can take place
732 		 */
733 		NAPI_GRO_CB(skb)->encap_mark = 0;
734 
735 		/* Set encapsulation before calling into inner gro_complete()
736 		 * functions to make them set up the inner offsets.
737 		 */
738 		skb->encapsulation = 1;
739 		err = udp_sk(sk)->gro_complete(sk, skb,
740 				nhoff + sizeof(struct udphdr));
741 	} else {
742 		err = udp_gro_complete_segment(skb);
743 	}
744 
745 	if (skb->remcsum_offload)
746 		skb_shinfo(skb)->gso_type |= SKB_GSO_TUNNEL_REMCSUM;
747 
748 	return err;
749 }
750 EXPORT_SYMBOL(udp_gro_complete);
751 
udp4_gro_complete(struct sk_buff * skb,int nhoff)752 INDIRECT_CALLABLE_SCOPE int udp4_gro_complete(struct sk_buff *skb, int nhoff)
753 {
754 	const u16 offset = NAPI_GRO_CB(skb)->network_offsets[skb->encapsulation];
755 	const struct iphdr *iph = (struct iphdr *)(skb->data + offset);
756 	struct udphdr *uh = (struct udphdr *)(skb->data + nhoff);
757 
758 	/* do fraglist only if there is no outer UDP encap (or we already processed it) */
759 	if (NAPI_GRO_CB(skb)->is_flist && !NAPI_GRO_CB(skb)->encap_mark) {
760 		uh->len = htons(skb->len - nhoff);
761 
762 		skb_shinfo(skb)->gso_type |= (SKB_GSO_FRAGLIST|SKB_GSO_UDP_L4);
763 		skb_shinfo(skb)->gso_segs = NAPI_GRO_CB(skb)->count;
764 
765 		__skb_incr_checksum_unnecessary(skb);
766 
767 		return 0;
768 	}
769 
770 	if (uh->check)
771 		uh->check = ~udp_v4_check(skb->len - nhoff, iph->saddr,
772 					  iph->daddr, 0);
773 
774 	return udp_gro_complete(skb, nhoff, udp4_lib_lookup_skb);
775 }
776 
777 static const struct net_offload udpv4_offload = {
778 	.callbacks = {
779 		.gso_segment = udp4_ufo_fragment,
780 		.gro_receive  =	udp4_gro_receive,
781 		.gro_complete =	udp4_gro_complete,
782 	},
783 };
784 
udpv4_offload_init(void)785 int __init udpv4_offload_init(void)
786 {
787 	return inet_add_offload(&udpv4_offload, IPPROTO_UDP);
788 }
789