xref: /openbmc/linux/net/ipv6/ip6_offload.c (revision 275876e2)
1 /*
2  *	IPV6 GSO/GRO offload support
3  *	Linux INET6 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 
11 #include <linux/kernel.h>
12 #include <linux/socket.h>
13 #include <linux/netdevice.h>
14 #include <linux/skbuff.h>
15 #include <linux/printk.h>
16 
17 #include <net/protocol.h>
18 #include <net/ipv6.h>
19 
20 #include "ip6_offload.h"
21 
22 static int ipv6_gso_pull_exthdrs(struct sk_buff *skb, int proto)
23 {
24 	const struct net_offload *ops = NULL;
25 
26 	for (;;) {
27 		struct ipv6_opt_hdr *opth;
28 		int len;
29 
30 		if (proto != NEXTHDR_HOP) {
31 			ops = rcu_dereference(inet6_offloads[proto]);
32 
33 			if (unlikely(!ops))
34 				break;
35 
36 			if (!(ops->flags & INET6_PROTO_GSO_EXTHDR))
37 				break;
38 		}
39 
40 		if (unlikely(!pskb_may_pull(skb, 8)))
41 			break;
42 
43 		opth = (void *)skb->data;
44 		len = ipv6_optlen(opth);
45 
46 		if (unlikely(!pskb_may_pull(skb, len)))
47 			break;
48 
49 		proto = opth->nexthdr;
50 		__skb_pull(skb, len);
51 	}
52 
53 	return proto;
54 }
55 
56 static int ipv6_gso_send_check(struct sk_buff *skb)
57 {
58 	const struct ipv6hdr *ipv6h;
59 	const struct net_offload *ops;
60 	int err = -EINVAL;
61 
62 	if (unlikely(!pskb_may_pull(skb, sizeof(*ipv6h))))
63 		goto out;
64 
65 	ipv6h = ipv6_hdr(skb);
66 	__skb_pull(skb, sizeof(*ipv6h));
67 	err = -EPROTONOSUPPORT;
68 
69 	ops = rcu_dereference(inet6_offloads[
70 		ipv6_gso_pull_exthdrs(skb, ipv6h->nexthdr)]);
71 
72 	if (likely(ops && ops->callbacks.gso_send_check)) {
73 		skb_reset_transport_header(skb);
74 		err = ops->callbacks.gso_send_check(skb);
75 	}
76 
77 out:
78 	return err;
79 }
80 
81 static struct sk_buff *ipv6_gso_segment(struct sk_buff *skb,
82 	netdev_features_t features)
83 {
84 	struct sk_buff *segs = ERR_PTR(-EINVAL);
85 	struct ipv6hdr *ipv6h;
86 	const struct net_offload *ops;
87 	int proto;
88 	struct frag_hdr *fptr;
89 	unsigned int unfrag_ip6hlen;
90 	u8 *prevhdr;
91 	int offset = 0;
92 	bool encap, udpfrag;
93 	int nhoff;
94 
95 	if (unlikely(skb_shinfo(skb)->gso_type &
96 		     ~(SKB_GSO_UDP |
97 		       SKB_GSO_DODGY |
98 		       SKB_GSO_TCP_ECN |
99 		       SKB_GSO_GRE |
100 		       SKB_GSO_GRE_CSUM |
101 		       SKB_GSO_IPIP |
102 		       SKB_GSO_SIT |
103 		       SKB_GSO_UDP_TUNNEL |
104 		       SKB_GSO_UDP_TUNNEL_CSUM |
105 		       SKB_GSO_MPLS |
106 		       SKB_GSO_TCPV6 |
107 		       0)))
108 		goto out;
109 
110 	skb_reset_network_header(skb);
111 	nhoff = skb_network_header(skb) - skb_mac_header(skb);
112 	if (unlikely(!pskb_may_pull(skb, sizeof(*ipv6h))))
113 		goto out;
114 
115 	encap = SKB_GSO_CB(skb)->encap_level > 0;
116 	if (encap)
117 		features = skb->dev->hw_enc_features & netif_skb_features(skb);
118 	SKB_GSO_CB(skb)->encap_level += sizeof(*ipv6h);
119 
120 	ipv6h = ipv6_hdr(skb);
121 	__skb_pull(skb, sizeof(*ipv6h));
122 	segs = ERR_PTR(-EPROTONOSUPPORT);
123 
124 	proto = ipv6_gso_pull_exthdrs(skb, ipv6h->nexthdr);
125 
126 	if (skb->encapsulation &&
127 	    skb_shinfo(skb)->gso_type & (SKB_GSO_SIT|SKB_GSO_IPIP))
128 		udpfrag = proto == IPPROTO_UDP && encap;
129 	else
130 		udpfrag = proto == IPPROTO_UDP && !skb->encapsulation;
131 
132 	ops = rcu_dereference(inet6_offloads[proto]);
133 	if (likely(ops && ops->callbacks.gso_segment)) {
134 		skb_reset_transport_header(skb);
135 		segs = ops->callbacks.gso_segment(skb, features);
136 	}
137 
138 	if (IS_ERR(segs))
139 		goto out;
140 
141 	for (skb = segs; skb; skb = skb->next) {
142 		ipv6h = (struct ipv6hdr *)(skb_mac_header(skb) + nhoff);
143 		ipv6h->payload_len = htons(skb->len - nhoff - sizeof(*ipv6h));
144 		skb->network_header = (u8 *)ipv6h - skb->head;
145 
146 		if (udpfrag) {
147 			unfrag_ip6hlen = ip6_find_1stfragopt(skb, &prevhdr);
148 			fptr = (struct frag_hdr *)((u8 *)ipv6h + unfrag_ip6hlen);
149 			fptr->frag_off = htons(offset);
150 			if (skb->next != NULL)
151 				fptr->frag_off |= htons(IP6_MF);
152 			offset += (ntohs(ipv6h->payload_len) -
153 				   sizeof(struct frag_hdr));
154 		}
155 		if (encap)
156 			skb_reset_inner_headers(skb);
157 	}
158 
159 out:
160 	return segs;
161 }
162 
163 /* Return the total length of all the extension hdrs, following the same
164  * logic in ipv6_gso_pull_exthdrs() when parsing ext-hdrs.
165  */
166 static int ipv6_exthdrs_len(struct ipv6hdr *iph,
167 			    const struct net_offload **opps)
168 {
169 	struct ipv6_opt_hdr *opth = (void *)iph;
170 	int len = 0, proto, optlen = sizeof(*iph);
171 
172 	proto = iph->nexthdr;
173 	for (;;) {
174 		if (proto != NEXTHDR_HOP) {
175 			*opps = rcu_dereference(inet6_offloads[proto]);
176 			if (unlikely(!(*opps)))
177 				break;
178 			if (!((*opps)->flags & INET6_PROTO_GSO_EXTHDR))
179 				break;
180 		}
181 		opth = (void *)opth + optlen;
182 		optlen = ipv6_optlen(opth);
183 		len += optlen;
184 		proto = opth->nexthdr;
185 	}
186 	return len;
187 }
188 
189 static struct sk_buff **ipv6_gro_receive(struct sk_buff **head,
190 					 struct sk_buff *skb)
191 {
192 	const struct net_offload *ops;
193 	struct sk_buff **pp = NULL;
194 	struct sk_buff *p;
195 	struct ipv6hdr *iph;
196 	unsigned int nlen;
197 	unsigned int hlen;
198 	unsigned int off;
199 	u16 flush = 1;
200 	int proto;
201 
202 	off = skb_gro_offset(skb);
203 	hlen = off + sizeof(*iph);
204 	iph = skb_gro_header_fast(skb, off);
205 	if (skb_gro_header_hard(skb, hlen)) {
206 		iph = skb_gro_header_slow(skb, hlen, off);
207 		if (unlikely(!iph))
208 			goto out;
209 	}
210 
211 	skb_set_network_header(skb, off);
212 	skb_gro_pull(skb, sizeof(*iph));
213 	skb_set_transport_header(skb, skb_gro_offset(skb));
214 
215 	flush += ntohs(iph->payload_len) != skb_gro_len(skb);
216 
217 	rcu_read_lock();
218 	proto = iph->nexthdr;
219 	ops = rcu_dereference(inet6_offloads[proto]);
220 	if (!ops || !ops->callbacks.gro_receive) {
221 		__pskb_pull(skb, skb_gro_offset(skb));
222 		proto = ipv6_gso_pull_exthdrs(skb, proto);
223 		skb_gro_pull(skb, -skb_transport_offset(skb));
224 		skb_reset_transport_header(skb);
225 		__skb_push(skb, skb_gro_offset(skb));
226 
227 		ops = rcu_dereference(inet6_offloads[proto]);
228 		if (!ops || !ops->callbacks.gro_receive)
229 			goto out_unlock;
230 
231 		iph = ipv6_hdr(skb);
232 	}
233 
234 	NAPI_GRO_CB(skb)->proto = proto;
235 
236 	flush--;
237 	nlen = skb_network_header_len(skb);
238 
239 	for (p = *head; p; p = p->next) {
240 		const struct ipv6hdr *iph2;
241 		__be32 first_word; /* <Version:4><Traffic_Class:8><Flow_Label:20> */
242 
243 		if (!NAPI_GRO_CB(p)->same_flow)
244 			continue;
245 
246 		iph2 = (struct ipv6hdr *)(p->data + off);
247 		first_word = *(__be32 *)iph ^ *(__be32 *)iph2 ;
248 
249 		/* All fields must match except length and Traffic Class.
250 		 * XXX skbs on the gro_list have all been parsed and pulled
251 		 * already so we don't need to compare nlen
252 		 * (nlen != (sizeof(*iph2) + ipv6_exthdrs_len(iph2, &ops)))
253 		 * memcmp() alone below is suffcient, right?
254 		 */
255 		 if ((first_word & htonl(0xF00FFFFF)) ||
256 		    memcmp(&iph->nexthdr, &iph2->nexthdr,
257 			   nlen - offsetof(struct ipv6hdr, nexthdr))) {
258 			NAPI_GRO_CB(p)->same_flow = 0;
259 			continue;
260 		}
261 		/* flush if Traffic Class fields are different */
262 		NAPI_GRO_CB(p)->flush |= !!(first_word & htonl(0x0FF00000));
263 		NAPI_GRO_CB(p)->flush |= flush;
264 	}
265 
266 	NAPI_GRO_CB(skb)->flush |= flush;
267 
268 	skb_gro_postpull_rcsum(skb, iph, nlen);
269 
270 	pp = ops->callbacks.gro_receive(head, skb);
271 
272 out_unlock:
273 	rcu_read_unlock();
274 
275 out:
276 	NAPI_GRO_CB(skb)->flush |= flush;
277 
278 	return pp;
279 }
280 
281 static int ipv6_gro_complete(struct sk_buff *skb, int nhoff)
282 {
283 	const struct net_offload *ops;
284 	struct ipv6hdr *iph = (struct ipv6hdr *)(skb->data + nhoff);
285 	int err = -ENOSYS;
286 
287 	iph->payload_len = htons(skb->len - nhoff - sizeof(*iph));
288 
289 	rcu_read_lock();
290 
291 	nhoff += sizeof(*iph) + ipv6_exthdrs_len(iph, &ops);
292 	if (WARN_ON(!ops || !ops->callbacks.gro_complete))
293 		goto out_unlock;
294 
295 	err = ops->callbacks.gro_complete(skb, nhoff);
296 
297 out_unlock:
298 	rcu_read_unlock();
299 
300 	return err;
301 }
302 
303 static struct packet_offload ipv6_packet_offload __read_mostly = {
304 	.type = cpu_to_be16(ETH_P_IPV6),
305 	.callbacks = {
306 		.gso_send_check = ipv6_gso_send_check,
307 		.gso_segment = ipv6_gso_segment,
308 		.gro_receive = ipv6_gro_receive,
309 		.gro_complete = ipv6_gro_complete,
310 	},
311 };
312 
313 static const struct net_offload sit_offload = {
314 	.callbacks = {
315 		.gso_send_check = ipv6_gso_send_check,
316 		.gso_segment	= ipv6_gso_segment,
317 	},
318 };
319 
320 static int __init ipv6_offload_init(void)
321 {
322 
323 	if (tcpv6_offload_init() < 0)
324 		pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
325 	if (udp_offload_init() < 0)
326 		pr_crit("%s: Cannot add UDP protocol offload\n", __func__);
327 	if (ipv6_exthdrs_offload_init() < 0)
328 		pr_crit("%s: Cannot add EXTHDRS protocol offload\n", __func__);
329 
330 	dev_add_offload(&ipv6_packet_offload);
331 
332 	inet_add_offload(&sit_offload, IPPROTO_IPV6);
333 
334 	return 0;
335 }
336 
337 fs_initcall(ipv6_offload_init);
338