Linux Audio

Check our new training course

Loading...
v6.9.4
  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
 16static 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));
149out:
150	return segs;
151}
152
153struct 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
185out_unlock:
186	rcu_read_unlock();
187
188	return segs;
189}
190EXPORT_SYMBOL(skb_udp_tunnel_segment);
191
192static 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
219static 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
250static 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
265struct 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 (skb_gso_ok(gso_skb, features | NETIF_F_GSO_ROBUST)) {
282		/* Packet is from an untrusted source, reset gso_segs. */
283		skb_shinfo(gso_skb)->gso_segs = DIV_ROUND_UP(gso_skb->len - sizeof(*uh),
284							     mss);
285		return NULL;
286	}
287
288	if (skb_shinfo(gso_skb)->gso_type & SKB_GSO_FRAGLIST)
289		return __udp_gso_segment_list(gso_skb, features, is_ipv6);
290
291	skb_pull(gso_skb, sizeof(*uh));
292
293	/* clear destructor to avoid skb_segment assigning it to tail */
294	copy_dtor = gso_skb->destructor == sock_wfree;
295	if (copy_dtor)
296		gso_skb->destructor = NULL;
297
298	segs = skb_segment(gso_skb, features);
299	if (IS_ERR_OR_NULL(segs)) {
300		if (copy_dtor)
301			gso_skb->destructor = sock_wfree;
302		return segs;
303	}
304
305	/* GSO partial and frag_list segmentation only requires splitting
306	 * the frame into an MSS multiple and possibly a remainder, both
307	 * cases return a GSO skb. So update the mss now.
308	 */
309	if (skb_is_gso(segs))
310		mss *= skb_shinfo(segs)->gso_segs;
311
312	seg = segs;
313	uh = udp_hdr(seg);
314
315	/* preserve TX timestamp flags and TS key for first segment */
316	skb_shinfo(seg)->tskey = skb_shinfo(gso_skb)->tskey;
317	skb_shinfo(seg)->tx_flags |=
318			(skb_shinfo(gso_skb)->tx_flags & SKBTX_ANY_TSTAMP);
319
320	/* compute checksum adjustment based on old length versus new */
321	newlen = htons(sizeof(*uh) + mss);
322	check = csum16_add(csum16_sub(uh->check, uh->len), newlen);
323
324	for (;;) {
325		if (copy_dtor) {
326			seg->destructor = sock_wfree;
327			seg->sk = sk;
328			sum_truesize += seg->truesize;
329		}
330
331		if (!seg->next)
332			break;
333
334		uh->len = newlen;
335		uh->check = check;
336
337		if (seg->ip_summed == CHECKSUM_PARTIAL)
338			gso_reset_checksum(seg, ~check);
339		else
340			uh->check = gso_make_checksum(seg, ~check) ? :
341				    CSUM_MANGLED_0;
342
343		seg = seg->next;
344		uh = udp_hdr(seg);
345	}
346
347	/* last packet can be partial gso_size, account for that in checksum */
348	newlen = htons(skb_tail_pointer(seg) - skb_transport_header(seg) +
349		       seg->data_len);
350	check = csum16_add(csum16_sub(uh->check, uh->len), newlen);
351
352	uh->len = newlen;
353	uh->check = check;
354
355	if (seg->ip_summed == CHECKSUM_PARTIAL)
356		gso_reset_checksum(seg, ~check);
357	else
358		uh->check = gso_make_checksum(seg, ~check) ? : CSUM_MANGLED_0;
359
360	/* update refcount for the packet */
361	if (copy_dtor) {
362		int delta = sum_truesize - gso_skb->truesize;
363
364		/* In some pathological cases, delta can be negative.
365		 * We need to either use refcount_add() or refcount_sub_and_test()
366		 */
367		if (likely(delta >= 0))
368			refcount_add(delta, &sk->sk_wmem_alloc);
369		else
370			WARN_ON_ONCE(refcount_sub_and_test(-delta, &sk->sk_wmem_alloc));
371	}
372	return segs;
373}
374EXPORT_SYMBOL_GPL(__udp_gso_segment);
375
376static struct sk_buff *udp4_ufo_fragment(struct sk_buff *skb,
377					 netdev_features_t features)
378{
379	struct sk_buff *segs = ERR_PTR(-EINVAL);
380	unsigned int mss;
381	__wsum csum;
382	struct udphdr *uh;
383	struct iphdr *iph;
384
385	if (skb->encapsulation &&
386	    (skb_shinfo(skb)->gso_type &
387	     (SKB_GSO_UDP_TUNNEL|SKB_GSO_UDP_TUNNEL_CSUM))) {
388		segs = skb_udp_tunnel_segment(skb, features, false);
389		goto out;
390	}
391
392	if (!(skb_shinfo(skb)->gso_type & (SKB_GSO_UDP | SKB_GSO_UDP_L4)))
393		goto out;
394
395	if (!pskb_may_pull(skb, sizeof(struct udphdr)))
396		goto out;
397
398	if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4)
399		return __udp_gso_segment(skb, features, false);
400
401	mss = skb_shinfo(skb)->gso_size;
402	if (unlikely(skb->len <= mss))
403		goto out;
404
405	/* Do software UFO. Complete and fill in the UDP checksum as
406	 * HW cannot do checksum of UDP packets sent as multiple
407	 * IP fragments.
408	 */
409
410	uh = udp_hdr(skb);
411	iph = ip_hdr(skb);
412
413	uh->check = 0;
414	csum = skb_checksum(skb, 0, skb->len, 0);
415	uh->check = udp_v4_check(skb->len, iph->saddr, iph->daddr, csum);
416	if (uh->check == 0)
417		uh->check = CSUM_MANGLED_0;
418
419	skb->ip_summed = CHECKSUM_UNNECESSARY;
420
421	/* If there is no outer header we can fake a checksum offload
422	 * due to the fact that we have already done the checksum in
423	 * software prior to segmenting the frame.
424	 */
425	if (!skb->encap_hdr_csum)
426		features |= NETIF_F_HW_CSUM;
427
428	/* Fragment the skb. IP headers of the fragments are updated in
429	 * inet_gso_segment()
430	 */
431	segs = skb_segment(skb, features);
432out:
433	return segs;
434}
435
436static int skb_gro_receive_list(struct sk_buff *p, struct sk_buff *skb)
437{
438	if (unlikely(p->len + skb->len >= 65536))
439		return -E2BIG;
440
441	if (NAPI_GRO_CB(p)->last == p)
442		skb_shinfo(p)->frag_list = skb;
443	else
444		NAPI_GRO_CB(p)->last->next = skb;
445
446	skb_pull(skb, skb_gro_offset(skb));
447
448	NAPI_GRO_CB(p)->last = skb;
449	NAPI_GRO_CB(p)->count++;
450	p->data_len += skb->len;
451
452	/* sk ownership - if any - completely transferred to the aggregated packet */
453	skb->destructor = NULL;
454	skb->sk = NULL;
455	p->truesize += skb->truesize;
456	p->len += skb->len;
457
458	NAPI_GRO_CB(skb)->same_flow = 1;
459
460	return 0;
461}
462
463
464#define UDP_GRO_CNT_MAX 64
465static struct sk_buff *udp_gro_receive_segment(struct list_head *head,
466					       struct sk_buff *skb)
467{
468	struct udphdr *uh = udp_gro_udphdr(skb);
469	struct sk_buff *pp = NULL;
470	struct udphdr *uh2;
471	struct sk_buff *p;
472	unsigned int ulen;
473	int ret = 0;
474	int flush;
475
476	/* requires non zero csum, for symmetry with GSO */
477	if (!uh->check) {
478		NAPI_GRO_CB(skb)->flush = 1;
479		return NULL;
480	}
481
482	/* Do not deal with padded or malicious packets, sorry ! */
483	ulen = ntohs(uh->len);
484	if (ulen <= sizeof(*uh) || ulen != skb_gro_len(skb)) {
485		NAPI_GRO_CB(skb)->flush = 1;
486		return NULL;
487	}
488	/* pull encapsulating udp header */
489	skb_gro_pull(skb, sizeof(struct udphdr));
490
491	list_for_each_entry(p, head, list) {
492		if (!NAPI_GRO_CB(p)->same_flow)
493			continue;
494
495		uh2 = udp_hdr(p);
496
497		/* Match ports only, as csum is always non zero */
498		if ((*(u32 *)&uh->source != *(u32 *)&uh2->source)) {
499			NAPI_GRO_CB(p)->same_flow = 0;
500			continue;
501		}
502
503		if (NAPI_GRO_CB(skb)->is_flist != NAPI_GRO_CB(p)->is_flist) {
504			NAPI_GRO_CB(skb)->flush = 1;
505			return p;
506		}
507
508		flush = NAPI_GRO_CB(p)->flush;
509
510		if (NAPI_GRO_CB(p)->flush_id != 1 ||
511		    NAPI_GRO_CB(p)->count != 1 ||
512		    !NAPI_GRO_CB(p)->is_atomic)
513			flush |= NAPI_GRO_CB(p)->flush_id;
514		else
515			NAPI_GRO_CB(p)->is_atomic = false;
516
517		/* Terminate the flow on len mismatch or if it grow "too much".
518		 * Under small packet flood GRO count could elsewhere grow a lot
519		 * leading to excessive truesize values.
520		 * On len mismatch merge the first packet shorter than gso_size,
521		 * otherwise complete the GRO packet.
522		 */
523		if (ulen > ntohs(uh2->len) || flush) {
524			pp = p;
525		} else {
526			if (NAPI_GRO_CB(skb)->is_flist) {
527				if (!pskb_may_pull(skb, skb_gro_offset(skb))) {
528					NAPI_GRO_CB(skb)->flush = 1;
529					return NULL;
530				}
531				if ((skb->ip_summed != p->ip_summed) ||
532				    (skb->csum_level != p->csum_level)) {
533					NAPI_GRO_CB(skb)->flush = 1;
534					return NULL;
535				}
536				ret = skb_gro_receive_list(p, skb);
537			} else {
538				skb_gro_postpull_rcsum(skb, uh,
539						       sizeof(struct udphdr));
540
541				ret = skb_gro_receive(p, skb);
542			}
543		}
544
545		if (ret || ulen != ntohs(uh2->len) ||
546		    NAPI_GRO_CB(p)->count >= UDP_GRO_CNT_MAX)
547			pp = p;
548
549		return pp;
550	}
551
552	/* mismatch, but we never need to flush */
553	return NULL;
554}
555
556struct sk_buff *udp_gro_receive(struct list_head *head, struct sk_buff *skb,
557				struct udphdr *uh, struct sock *sk)
558{
559	struct sk_buff *pp = NULL;
560	struct sk_buff *p;
561	struct udphdr *uh2;
562	unsigned int off = skb_gro_offset(skb);
563	int flush = 1;
564
565	/* We can do L4 aggregation only if the packet can't land in a tunnel
566	 * otherwise we could corrupt the inner stream. Detecting such packets
567	 * cannot be foolproof and the aggregation might still happen in some
568	 * cases. Such packets should be caught in udp_unexpected_gso later.
569	 */
570	NAPI_GRO_CB(skb)->is_flist = 0;
571	if (!sk || !udp_sk(sk)->gro_receive) {
572		/* If the packet was locally encapsulated in a UDP tunnel that
573		 * wasn't detected above, do not GRO.
574		 */
575		if (skb->encapsulation)
576			goto out;
577
578		if (skb->dev->features & NETIF_F_GRO_FRAGLIST)
579			NAPI_GRO_CB(skb)->is_flist = sk ? !udp_test_bit(GRO_ENABLED, sk) : 1;
580
581		if ((!sk && (skb->dev->features & NETIF_F_GRO_UDP_FWD)) ||
582		    (sk && udp_test_bit(GRO_ENABLED, sk)) || NAPI_GRO_CB(skb)->is_flist)
583			return call_gro_receive(udp_gro_receive_segment, head, skb);
584
585		/* no GRO, be sure flush the current packet */
586		goto out;
587	}
588
589	if (NAPI_GRO_CB(skb)->encap_mark ||
590	    (uh->check && skb->ip_summed != CHECKSUM_PARTIAL &&
591	     NAPI_GRO_CB(skb)->csum_cnt == 0 &&
592	     !NAPI_GRO_CB(skb)->csum_valid))
593		goto out;
594
595	/* mark that this skb passed once through the tunnel gro layer */
596	NAPI_GRO_CB(skb)->encap_mark = 1;
597
 
 
 
 
 
 
 
 
598	flush = 0;
599
600	list_for_each_entry(p, head, list) {
601		if (!NAPI_GRO_CB(p)->same_flow)
602			continue;
603
604		uh2 = (struct udphdr   *)(p->data + off);
605
606		/* Match ports and either checksums are either both zero
607		 * or nonzero.
608		 */
609		if ((*(u32 *)&uh->source != *(u32 *)&uh2->source) ||
610		    (!uh->check ^ !uh2->check)) {
611			NAPI_GRO_CB(p)->same_flow = 0;
612			continue;
613		}
614	}
615
616	skb_gro_pull(skb, sizeof(struct udphdr)); /* pull encapsulating udp header */
617	skb_gro_postpull_rcsum(skb, uh, sizeof(struct udphdr));
618	pp = call_gro_receive_sk(udp_sk(sk)->gro_receive, sk, head, skb);
619
 
 
620out:
621	skb_gro_flush_final(skb, pp, flush);
622	return pp;
623}
624EXPORT_SYMBOL(udp_gro_receive);
625
626static struct sock *udp4_gro_lookup_skb(struct sk_buff *skb, __be16 sport,
627					__be16 dport)
628{
629	const struct iphdr *iph = skb_gro_network_header(skb);
630	struct net *net = dev_net(skb->dev);
631	int iif, sdif;
632
633	inet_get_iif_sdif(skb, &iif, &sdif);
634
635	return __udp4_lib_lookup(net, iph->saddr, sport,
636				 iph->daddr, dport, iif,
637				 sdif, net->ipv4.udp_table, NULL);
638}
639
640INDIRECT_CALLABLE_SCOPE
641struct sk_buff *udp4_gro_receive(struct list_head *head, struct sk_buff *skb)
642{
643	struct udphdr *uh = udp_gro_udphdr(skb);
644	struct sock *sk = NULL;
645	struct sk_buff *pp;
646
647	if (unlikely(!uh))
648		goto flush;
649
650	/* Don't bother verifying checksum if we're going to flush anyway. */
651	if (NAPI_GRO_CB(skb)->flush)
652		goto skip;
653
654	if (skb_gro_checksum_validate_zero_check(skb, IPPROTO_UDP, uh->check,
655						 inet_gro_compute_pseudo))
656		goto flush;
657	else if (uh->check)
658		skb_gro_checksum_try_convert(skb, IPPROTO_UDP,
659					     inet_gro_compute_pseudo);
660skip:
661	NAPI_GRO_CB(skb)->is_ipv6 = 0;
662
663	if (static_branch_unlikely(&udp_encap_needed_key))
664		sk = udp4_gro_lookup_skb(skb, uh->source, uh->dest);
665
666	pp = udp_gro_receive(head, skb, uh, sk);
667	return pp;
668
669flush:
670	NAPI_GRO_CB(skb)->flush = 1;
671	return NULL;
672}
673
674static int udp_gro_complete_segment(struct sk_buff *skb)
675{
676	struct udphdr *uh = udp_hdr(skb);
677
678	skb->csum_start = (unsigned char *)uh - skb->head;
679	skb->csum_offset = offsetof(struct udphdr, check);
680	skb->ip_summed = CHECKSUM_PARTIAL;
681
682	skb_shinfo(skb)->gso_segs = NAPI_GRO_CB(skb)->count;
683	skb_shinfo(skb)->gso_type |= SKB_GSO_UDP_L4;
684
685	if (skb->encapsulation)
686		skb->inner_transport_header = skb->transport_header;
687
688	return 0;
689}
690
691int udp_gro_complete(struct sk_buff *skb, int nhoff,
692		     udp_lookup_t lookup)
693{
694	__be16 newlen = htons(skb->len - nhoff);
695	struct udphdr *uh = (struct udphdr *)(skb->data + nhoff);
 
696	struct sock *sk;
697	int err;
698
699	uh->len = newlen;
700
701	sk = INDIRECT_CALL_INET(lookup, udp6_lib_lookup_skb,
702				udp4_lib_lookup_skb, skb, uh->source, uh->dest);
703	if (sk && udp_sk(sk)->gro_complete) {
704		skb_shinfo(skb)->gso_type = uh->check ? SKB_GSO_UDP_TUNNEL_CSUM
705					: SKB_GSO_UDP_TUNNEL;
706
707		/* clear the encap mark, so that inner frag_list gro_complete
708		 * can take place
709		 */
710		NAPI_GRO_CB(skb)->encap_mark = 0;
711
712		/* Set encapsulation before calling into inner gro_complete()
713		 * functions to make them set up the inner offsets.
714		 */
715		skb->encapsulation = 1;
716		err = udp_sk(sk)->gro_complete(sk, skb,
717				nhoff + sizeof(struct udphdr));
718	} else {
719		err = udp_gro_complete_segment(skb);
720	}
721
722	if (skb->remcsum_offload)
723		skb_shinfo(skb)->gso_type |= SKB_GSO_TUNNEL_REMCSUM;
724
725	return err;
726}
727EXPORT_SYMBOL(udp_gro_complete);
728
729INDIRECT_CALLABLE_SCOPE int udp4_gro_complete(struct sk_buff *skb, int nhoff)
730{
731	const u16 offset = NAPI_GRO_CB(skb)->network_offsets[skb->encapsulation];
732	const struct iphdr *iph = (struct iphdr *)(skb->data + offset);
733	struct udphdr *uh = (struct udphdr *)(skb->data + nhoff);
734
735	/* do fraglist only if there is no outer UDP encap (or we already processed it) */
736	if (NAPI_GRO_CB(skb)->is_flist && !NAPI_GRO_CB(skb)->encap_mark) {
737		uh->len = htons(skb->len - nhoff);
738
739		skb_shinfo(skb)->gso_type |= (SKB_GSO_FRAGLIST|SKB_GSO_UDP_L4);
740		skb_shinfo(skb)->gso_segs = NAPI_GRO_CB(skb)->count;
741
742		__skb_incr_checksum_unnecessary(skb);
743
744		return 0;
745	}
746
747	if (uh->check)
748		uh->check = ~udp_v4_check(skb->len - nhoff, iph->saddr,
749					  iph->daddr, 0);
 
 
 
750
751	return udp_gro_complete(skb, nhoff, udp4_lib_lookup_skb);
752}
753
 
 
 
 
 
 
 
 
754int __init udpv4_offload_init(void)
755{
756	net_hotdata.udpv4_offload = (struct net_offload) {
757		.callbacks = {
758			.gso_segment = udp4_ufo_fragment,
759			.gro_receive  =	udp4_gro_receive,
760			.gro_complete =	udp4_gro_complete,
761		},
762	};
763	return inet_add_offload(&net_hotdata.udpv4_offload, IPPROTO_UDP);
764}
v4.17
 
  1/*
  2 *	IPV4 GSO/GRO offload support
  3 *	Linux INET implementation
  4 *
  5 *	This program is free software; you can redistribute it and/or
  6 *	modify it under the terms of the GNU General Public License
  7 *	as published by the Free Software Foundation; either version
  8 *	2 of the License, or (at your option) any later version.
  9 *
 10 *	UDPv4 GSO support
 11 */
 12
 13#include <linux/skbuff.h>
 
 
 14#include <net/udp.h>
 15#include <net/protocol.h>
 
 16
 17static struct sk_buff *__skb_udp_tunnel_segment(struct sk_buff *skb,
 18	netdev_features_t features,
 19	struct sk_buff *(*gso_inner_segment)(struct sk_buff *skb,
 20					     netdev_features_t features),
 21	__be16 new_protocol, bool is_ipv6)
 22{
 23	int tnl_hlen = skb_inner_mac_header(skb) - skb_transport_header(skb);
 24	bool remcsum, need_csum, offload_csum, gso_partial;
 25	struct sk_buff *segs = ERR_PTR(-EINVAL);
 26	struct udphdr *uh = udp_hdr(skb);
 27	u16 mac_offset = skb->mac_header;
 28	__be16 protocol = skb->protocol;
 29	u16 mac_len = skb->mac_len;
 30	int udp_offset, outer_hlen;
 31	__wsum partial;
 32	bool need_ipsec;
 33
 34	if (unlikely(!pskb_may_pull(skb, tnl_hlen)))
 35		goto out;
 36
 37	/* Adjust partial header checksum to negate old length.
 38	 * We cannot rely on the value contained in uh->len as it is
 39	 * possible that the actual value exceeds the boundaries of the
 40	 * 16 bit length field due to the header being added outside of an
 41	 * IP or IPv6 frame that was already limited to 64K - 1.
 42	 */
 43	if (skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL)
 44		partial = (__force __wsum)uh->len;
 45	else
 46		partial = (__force __wsum)htonl(skb->len);
 47	partial = csum_sub(csum_unfold(uh->check), partial);
 48
 49	/* setup inner skb. */
 50	skb->encapsulation = 0;
 51	SKB_GSO_CB(skb)->encap_level = 0;
 52	__skb_pull(skb, tnl_hlen);
 53	skb_reset_mac_header(skb);
 54	skb_set_network_header(skb, skb_inner_network_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
 74	/* The only checksum offload we care about from here on out is the
 75	 * outer one so strip the existing checksum feature flags and
 76	 * instead set the flag based on our outer checksum offload value.
 77	 */
 78	if (remcsum) {
 79		features &= ~NETIF_F_CSUM_MASK;
 80		if (!need_csum || offload_csum)
 81			features |= NETIF_F_HW_CSUM;
 82	}
 83
 84	/* segment inner packet. */
 85	segs = gso_inner_segment(skb, features);
 86	if (IS_ERR_OR_NULL(segs)) {
 87		skb_gso_error_unwind(skb, protocol, tnl_hlen, mac_offset,
 88				     mac_len);
 89		goto out;
 90	}
 91
 92	gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
 93
 94	outer_hlen = skb_tnl_header_len(skb);
 95	udp_offset = outer_hlen - tnl_hlen;
 96	skb = segs;
 97	do {
 98		unsigned int len;
 99
100		if (remcsum)
101			skb->ip_summed = CHECKSUM_NONE;
102
103		/* Set up inner headers if we are offloading inner checksum */
104		if (skb->ip_summed == CHECKSUM_PARTIAL) {
105			skb_reset_inner_headers(skb);
106			skb->encapsulation = 1;
107		}
108
109		skb->mac_len = mac_len;
110		skb->protocol = protocol;
111
112		__skb_push(skb, outer_hlen);
113		skb_reset_mac_header(skb);
114		skb_set_network_header(skb, mac_len);
115		skb_set_transport_header(skb, udp_offset);
116		len = skb->len - udp_offset;
117		uh = udp_hdr(skb);
118
119		/* If we are only performing partial GSO the inner header
120		 * will be using a length value equal to only one MSS sized
121		 * segment instead of the entire frame.
122		 */
123		if (gso_partial && skb_is_gso(skb)) {
124			uh->len = htons(skb_shinfo(skb)->gso_size +
125					SKB_GSO_CB(skb)->data_offset +
126					skb->head - (unsigned char *)uh);
127		} else {
128			uh->len = htons(len);
129		}
130
131		if (!need_csum)
132			continue;
133
134		uh->check = ~csum_fold(csum_add(partial,
135				       (__force __wsum)htonl(len)));
136
137		if (skb->encapsulation || !offload_csum) {
138			uh->check = gso_make_checksum(skb, ~uh->check);
139			if (uh->check == 0)
140				uh->check = CSUM_MANGLED_0;
141		} else {
142			skb->ip_summed = CHECKSUM_PARTIAL;
143			skb->csum_start = skb_transport_header(skb) - skb->head;
144			skb->csum_offset = offsetof(struct udphdr, check);
145		}
146	} while ((skb = skb->next));
147out:
148	return segs;
149}
150
151struct sk_buff *skb_udp_tunnel_segment(struct sk_buff *skb,
152				       netdev_features_t features,
153				       bool is_ipv6)
154{
 
155	__be16 protocol = skb->protocol;
156	const struct net_offload **offloads;
157	const struct net_offload *ops;
158	struct sk_buff *segs = ERR_PTR(-EINVAL);
159	struct sk_buff *(*gso_inner_segment)(struct sk_buff *skb,
160					     netdev_features_t features);
161
162	rcu_read_lock();
163
164	switch (skb->inner_protocol_type) {
165	case ENCAP_TYPE_ETHER:
166		protocol = skb->inner_protocol;
167		gso_inner_segment = skb_mac_gso_segment;
168		break;
169	case ENCAP_TYPE_IPPROTO:
170		offloads = is_ipv6 ? inet6_offloads : inet_offloads;
171		ops = rcu_dereference(offloads[skb->inner_ipproto]);
172		if (!ops || !ops->callbacks.gso_segment)
173			goto out_unlock;
174		gso_inner_segment = ops->callbacks.gso_segment;
175		break;
176	default:
177		goto out_unlock;
178	}
179
180	segs = __skb_udp_tunnel_segment(skb, features, gso_inner_segment,
181					protocol, is_ipv6);
182
183out_unlock:
184	rcu_read_unlock();
185
186	return segs;
187}
188EXPORT_SYMBOL(skb_udp_tunnel_segment);
189
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
190static struct sk_buff *udp4_ufo_fragment(struct sk_buff *skb,
191					 netdev_features_t features)
192{
193	struct sk_buff *segs = ERR_PTR(-EINVAL);
194	unsigned int mss;
195	__wsum csum;
196	struct udphdr *uh;
197	struct iphdr *iph;
198
199	if (skb->encapsulation &&
200	    (skb_shinfo(skb)->gso_type &
201	     (SKB_GSO_UDP_TUNNEL|SKB_GSO_UDP_TUNNEL_CSUM))) {
202		segs = skb_udp_tunnel_segment(skb, features, false);
203		goto out;
204	}
205
206	if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP))
207		goto out;
208
209	if (!pskb_may_pull(skb, sizeof(struct udphdr)))
210		goto out;
211
 
 
 
212	mss = skb_shinfo(skb)->gso_size;
213	if (unlikely(skb->len <= mss))
214		goto out;
215
216	/* Do software UFO. Complete and fill in the UDP checksum as
217	 * HW cannot do checksum of UDP packets sent as multiple
218	 * IP fragments.
219	 */
220
221	uh = udp_hdr(skb);
222	iph = ip_hdr(skb);
223
224	uh->check = 0;
225	csum = skb_checksum(skb, 0, skb->len, 0);
226	uh->check = udp_v4_check(skb->len, iph->saddr, iph->daddr, csum);
227	if (uh->check == 0)
228		uh->check = CSUM_MANGLED_0;
229
230	skb->ip_summed = CHECKSUM_UNNECESSARY;
231
232	/* If there is no outer header we can fake a checksum offload
233	 * due to the fact that we have already done the checksum in
234	 * software prior to segmenting the frame.
235	 */
236	if (!skb->encap_hdr_csum)
237		features |= NETIF_F_HW_CSUM;
238
239	/* Fragment the skb. IP headers of the fragments are updated in
240	 * inet_gso_segment()
241	 */
242	segs = skb_segment(skb, features);
243out:
244	return segs;
245}
246
247struct sk_buff **udp_gro_receive(struct sk_buff **head, struct sk_buff *skb,
248				 struct udphdr *uh, udp_lookup_t lookup)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
249{
250	struct sk_buff *p, **pp = NULL;
 
251	struct udphdr *uh2;
252	unsigned int off = skb_gro_offset(skb);
253	int flush = 1;
254	struct sock *sk;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
255
256	if (NAPI_GRO_CB(skb)->encap_mark ||
257	    (skb->ip_summed != CHECKSUM_PARTIAL &&
258	     NAPI_GRO_CB(skb)->csum_cnt == 0 &&
259	     !NAPI_GRO_CB(skb)->csum_valid))
260		goto out;
261
262	/* mark that this skb passed once through the tunnel gro layer */
263	NAPI_GRO_CB(skb)->encap_mark = 1;
264
265	rcu_read_lock();
266	sk = (*lookup)(skb, uh->source, uh->dest);
267
268	if (sk && udp_sk(sk)->gro_receive)
269		goto unflush;
270	goto out_unlock;
271
272unflush:
273	flush = 0;
274
275	for (p = *head; p; p = p->next) {
276		if (!NAPI_GRO_CB(p)->same_flow)
277			continue;
278
279		uh2 = (struct udphdr   *)(p->data + off);
280
281		/* Match ports and either checksums are either both zero
282		 * or nonzero.
283		 */
284		if ((*(u32 *)&uh->source != *(u32 *)&uh2->source) ||
285		    (!uh->check ^ !uh2->check)) {
286			NAPI_GRO_CB(p)->same_flow = 0;
287			continue;
288		}
289	}
290
291	skb_gro_pull(skb, sizeof(struct udphdr)); /* pull encapsulating udp header */
292	skb_gro_postpull_rcsum(skb, uh, sizeof(struct udphdr));
293	pp = call_gro_receive_sk(udp_sk(sk)->gro_receive, sk, head, skb);
294
295out_unlock:
296	rcu_read_unlock();
297out:
298	NAPI_GRO_CB(skb)->flush |= flush;
299	return pp;
300}
301EXPORT_SYMBOL(udp_gro_receive);
302
303static struct sk_buff **udp4_gro_receive(struct sk_buff **head,
304					 struct sk_buff *skb)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
305{
306	struct udphdr *uh = udp_gro_udphdr(skb);
 
 
307
308	if (unlikely(!uh))
309		goto flush;
310
311	/* Don't bother verifying checksum if we're going to flush anyway. */
312	if (NAPI_GRO_CB(skb)->flush)
313		goto skip;
314
315	if (skb_gro_checksum_validate_zero_check(skb, IPPROTO_UDP, uh->check,
316						 inet_gro_compute_pseudo))
317		goto flush;
318	else if (uh->check)
319		skb_gro_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
320					     inet_gro_compute_pseudo);
321skip:
322	NAPI_GRO_CB(skb)->is_ipv6 = 0;
323	return udp_gro_receive(head, skb, uh, udp4_lib_lookup_skb);
 
 
 
 
 
324
325flush:
326	NAPI_GRO_CB(skb)->flush = 1;
327	return NULL;
328}
329
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
330int udp_gro_complete(struct sk_buff *skb, int nhoff,
331		     udp_lookup_t lookup)
332{
333	__be16 newlen = htons(skb->len - nhoff);
334	struct udphdr *uh = (struct udphdr *)(skb->data + nhoff);
335	int err = -ENOSYS;
336	struct sock *sk;
 
337
338	uh->len = newlen;
339
340	/* Set encapsulation before calling into inner gro_complete() functions
341	 * to make them set up the inner offsets.
342	 */
343	skb->encapsulation = 1;
 
 
 
 
 
 
344
345	rcu_read_lock();
346	sk = (*lookup)(skb, uh->source, uh->dest);
347	if (sk && udp_sk(sk)->gro_complete)
 
348		err = udp_sk(sk)->gro_complete(sk, skb,
349				nhoff + sizeof(struct udphdr));
350	rcu_read_unlock();
 
 
351
352	if (skb->remcsum_offload)
353		skb_shinfo(skb)->gso_type |= SKB_GSO_TUNNEL_REMCSUM;
354
355	return err;
356}
357EXPORT_SYMBOL(udp_gro_complete);
358
359static int udp4_gro_complete(struct sk_buff *skb, int nhoff)
360{
361	const struct iphdr *iph = ip_hdr(skb);
 
362	struct udphdr *uh = (struct udphdr *)(skb->data + nhoff);
363
364	if (uh->check) {
365		skb_shinfo(skb)->gso_type |= SKB_GSO_UDP_TUNNEL_CSUM;
 
 
 
 
 
 
 
 
 
 
 
366		uh->check = ~udp_v4_check(skb->len - nhoff, iph->saddr,
367					  iph->daddr, 0);
368	} else {
369		skb_shinfo(skb)->gso_type |= SKB_GSO_UDP_TUNNEL;
370	}
371
372	return udp_gro_complete(skb, nhoff, udp4_lib_lookup_skb);
373}
374
375static const struct net_offload udpv4_offload = {
376	.callbacks = {
377		.gso_segment = udp4_ufo_fragment,
378		.gro_receive  =	udp4_gro_receive,
379		.gro_complete =	udp4_gro_complete,
380	},
381};
382
383int __init udpv4_offload_init(void)
384{
385	return inet_add_offload(&udpv4_offload, IPPROTO_UDP);
 
 
 
 
 
 
 
386}