xref: /openbmc/linux/net/ipv4/udp_offload.c (revision 0bea2a65)
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 
17 static 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));
147 out:
148 	return segs;
149 }
150 
151 struct 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 
183 out_unlock:
184 	rcu_read_unlock();
185 
186 	return segs;
187 }
188 EXPORT_SYMBOL(skb_udp_tunnel_segment);
189 
190 static 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 (!pskb_may_pull(skb, sizeof(struct udphdr)))
207 		goto out;
208 
209 	mss = skb_shinfo(skb)->gso_size;
210 	if (unlikely(skb->len <= mss))
211 		goto out;
212 
213 	/* Do software UFO. Complete and fill in the UDP checksum as
214 	 * HW cannot do checksum of UDP packets sent as multiple
215 	 * IP fragments.
216 	 */
217 
218 	uh = udp_hdr(skb);
219 	iph = ip_hdr(skb);
220 
221 	uh->check = 0;
222 	csum = skb_checksum(skb, 0, skb->len, 0);
223 	uh->check = udp_v4_check(skb->len, iph->saddr, iph->daddr, csum);
224 	if (uh->check == 0)
225 		uh->check = CSUM_MANGLED_0;
226 
227 	skb->ip_summed = CHECKSUM_UNNECESSARY;
228 
229 	/* If there is no outer header we can fake a checksum offload
230 	 * due to the fact that we have already done the checksum in
231 	 * software prior to segmenting the frame.
232 	 */
233 	if (!skb->encap_hdr_csum)
234 		features |= NETIF_F_HW_CSUM;
235 
236 	/* Fragment the skb. IP headers of the fragments are updated in
237 	 * inet_gso_segment()
238 	 */
239 	segs = skb_segment(skb, features);
240 out:
241 	return segs;
242 }
243 
244 struct sk_buff **udp_gro_receive(struct sk_buff **head, struct sk_buff *skb,
245 				 struct udphdr *uh, udp_lookup_t lookup)
246 {
247 	struct sk_buff *p, **pp = NULL;
248 	struct udphdr *uh2;
249 	unsigned int off = skb_gro_offset(skb);
250 	int flush = 1;
251 	struct sock *sk;
252 
253 	if (NAPI_GRO_CB(skb)->encap_mark ||
254 	    (skb->ip_summed != CHECKSUM_PARTIAL &&
255 	     NAPI_GRO_CB(skb)->csum_cnt == 0 &&
256 	     !NAPI_GRO_CB(skb)->csum_valid))
257 		goto out;
258 
259 	/* mark that this skb passed once through the tunnel gro layer */
260 	NAPI_GRO_CB(skb)->encap_mark = 1;
261 
262 	rcu_read_lock();
263 	sk = (*lookup)(skb, uh->source, uh->dest);
264 
265 	if (sk && udp_sk(sk)->gro_receive)
266 		goto unflush;
267 	goto out_unlock;
268 
269 unflush:
270 	flush = 0;
271 
272 	for (p = *head; p; p = p->next) {
273 		if (!NAPI_GRO_CB(p)->same_flow)
274 			continue;
275 
276 		uh2 = (struct udphdr   *)(p->data + off);
277 
278 		/* Match ports and either checksums are either both zero
279 		 * or nonzero.
280 		 */
281 		if ((*(u32 *)&uh->source != *(u32 *)&uh2->source) ||
282 		    (!uh->check ^ !uh2->check)) {
283 			NAPI_GRO_CB(p)->same_flow = 0;
284 			continue;
285 		}
286 	}
287 
288 	skb_gro_pull(skb, sizeof(struct udphdr)); /* pull encapsulating udp header */
289 	skb_gro_postpull_rcsum(skb, uh, sizeof(struct udphdr));
290 	pp = call_gro_receive_sk(udp_sk(sk)->gro_receive, sk, head, skb);
291 
292 out_unlock:
293 	rcu_read_unlock();
294 out:
295 	NAPI_GRO_CB(skb)->flush |= flush;
296 	return pp;
297 }
298 EXPORT_SYMBOL(udp_gro_receive);
299 
300 static struct sk_buff **udp4_gro_receive(struct sk_buff **head,
301 					 struct sk_buff *skb)
302 {
303 	struct udphdr *uh = udp_gro_udphdr(skb);
304 
305 	if (unlikely(!uh))
306 		goto flush;
307 
308 	/* Don't bother verifying checksum if we're going to flush anyway. */
309 	if (NAPI_GRO_CB(skb)->flush)
310 		goto skip;
311 
312 	if (skb_gro_checksum_validate_zero_check(skb, IPPROTO_UDP, uh->check,
313 						 inet_gro_compute_pseudo))
314 		goto flush;
315 	else if (uh->check)
316 		skb_gro_checksum_try_convert(skb, IPPROTO_UDP, uh->check,
317 					     inet_gro_compute_pseudo);
318 skip:
319 	NAPI_GRO_CB(skb)->is_ipv6 = 0;
320 	return udp_gro_receive(head, skb, uh, udp4_lib_lookup_skb);
321 
322 flush:
323 	NAPI_GRO_CB(skb)->flush = 1;
324 	return NULL;
325 }
326 
327 int udp_gro_complete(struct sk_buff *skb, int nhoff,
328 		     udp_lookup_t lookup)
329 {
330 	__be16 newlen = htons(skb->len - nhoff);
331 	struct udphdr *uh = (struct udphdr *)(skb->data + nhoff);
332 	int err = -ENOSYS;
333 	struct sock *sk;
334 
335 	uh->len = newlen;
336 
337 	/* Set encapsulation before calling into inner gro_complete() functions
338 	 * to make them set up the inner offsets.
339 	 */
340 	skb->encapsulation = 1;
341 
342 	rcu_read_lock();
343 	sk = (*lookup)(skb, uh->source, uh->dest);
344 	if (sk && udp_sk(sk)->gro_complete)
345 		err = udp_sk(sk)->gro_complete(sk, skb,
346 				nhoff + sizeof(struct udphdr));
347 	rcu_read_unlock();
348 
349 	if (skb->remcsum_offload)
350 		skb_shinfo(skb)->gso_type |= SKB_GSO_TUNNEL_REMCSUM;
351 
352 	return err;
353 }
354 EXPORT_SYMBOL(udp_gro_complete);
355 
356 static int udp4_gro_complete(struct sk_buff *skb, int nhoff)
357 {
358 	const struct iphdr *iph = ip_hdr(skb);
359 	struct udphdr *uh = (struct udphdr *)(skb->data + nhoff);
360 
361 	if (uh->check) {
362 		skb_shinfo(skb)->gso_type |= SKB_GSO_UDP_TUNNEL_CSUM;
363 		uh->check = ~udp_v4_check(skb->len - nhoff, iph->saddr,
364 					  iph->daddr, 0);
365 	} else {
366 		skb_shinfo(skb)->gso_type |= SKB_GSO_UDP_TUNNEL;
367 	}
368 
369 	return udp_gro_complete(skb, nhoff, udp4_lib_lookup_skb);
370 }
371 
372 static const struct net_offload udpv4_offload = {
373 	.callbacks = {
374 		.gso_segment = udp4_ufo_fragment,
375 		.gro_receive  =	udp4_gro_receive,
376 		.gro_complete =	udp4_gro_complete,
377 	},
378 };
379 
380 int __init udpv4_offload_init(void)
381 {
382 	return inet_add_offload(&udpv4_offload, IPPROTO_UDP);
383 }
384