xref: /openbmc/linux/net/ipv6/ip6_offload.c (revision ae213c44)
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 #include <net/inet_common.h>
20 
21 #include "ip6_offload.h"
22 
23 /* All GRO functions are always builtin, except UDP over ipv6, which lays in
24  * ipv6 module, as it depends on UDPv6 lookup function, so we need special care
25  * when ipv6 is built as a module
26  */
27 #if IS_BUILTIN(CONFIG_IPV6)
28 #define INDIRECT_CALL_L4(f, f2, f1, ...) INDIRECT_CALL_2(f, f2, f1, __VA_ARGS__)
29 #else
30 #define INDIRECT_CALL_L4(f, f2, f1, ...) INDIRECT_CALL_1(f, f2, __VA_ARGS__)
31 #endif
32 
33 #define indirect_call_gro_receive_l4(f2, f1, cb, head, skb)	\
34 ({								\
35 	unlikely(gro_recursion_inc_test(skb)) ?			\
36 		NAPI_GRO_CB(skb)->flush |= 1, NULL :		\
37 		INDIRECT_CALL_L4(cb, f2, f1, head, skb);	\
38 })
39 
40 static int ipv6_gso_pull_exthdrs(struct sk_buff *skb, int proto)
41 {
42 	const struct net_offload *ops = NULL;
43 
44 	for (;;) {
45 		struct ipv6_opt_hdr *opth;
46 		int len;
47 
48 		if (proto != NEXTHDR_HOP) {
49 			ops = rcu_dereference(inet6_offloads[proto]);
50 
51 			if (unlikely(!ops))
52 				break;
53 
54 			if (!(ops->flags & INET6_PROTO_GSO_EXTHDR))
55 				break;
56 		}
57 
58 		if (unlikely(!pskb_may_pull(skb, 8)))
59 			break;
60 
61 		opth = (void *)skb->data;
62 		len = ipv6_optlen(opth);
63 
64 		if (unlikely(!pskb_may_pull(skb, len)))
65 			break;
66 
67 		opth = (void *)skb->data;
68 		proto = opth->nexthdr;
69 		__skb_pull(skb, len);
70 	}
71 
72 	return proto;
73 }
74 
75 static struct sk_buff *ipv6_gso_segment(struct sk_buff *skb,
76 	netdev_features_t features)
77 {
78 	struct sk_buff *segs = ERR_PTR(-EINVAL);
79 	struct ipv6hdr *ipv6h;
80 	const struct net_offload *ops;
81 	int proto;
82 	struct frag_hdr *fptr;
83 	unsigned int payload_len;
84 	u8 *prevhdr;
85 	int offset = 0;
86 	bool encap, udpfrag;
87 	int nhoff;
88 	bool gso_partial;
89 
90 	skb_reset_network_header(skb);
91 	nhoff = skb_network_header(skb) - skb_mac_header(skb);
92 	if (unlikely(!pskb_may_pull(skb, sizeof(*ipv6h))))
93 		goto out;
94 
95 	encap = SKB_GSO_CB(skb)->encap_level > 0;
96 	if (encap)
97 		features &= skb->dev->hw_enc_features;
98 	SKB_GSO_CB(skb)->encap_level += sizeof(*ipv6h);
99 
100 	ipv6h = ipv6_hdr(skb);
101 	__skb_pull(skb, sizeof(*ipv6h));
102 	segs = ERR_PTR(-EPROTONOSUPPORT);
103 
104 	proto = ipv6_gso_pull_exthdrs(skb, ipv6h->nexthdr);
105 
106 	if (skb->encapsulation &&
107 	    skb_shinfo(skb)->gso_type & (SKB_GSO_IPXIP4 | SKB_GSO_IPXIP6))
108 		udpfrag = proto == IPPROTO_UDP && encap &&
109 			  (skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
110 	else
111 		udpfrag = proto == IPPROTO_UDP && !skb->encapsulation &&
112 			  (skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
113 
114 	ops = rcu_dereference(inet6_offloads[proto]);
115 	if (likely(ops && ops->callbacks.gso_segment)) {
116 		skb_reset_transport_header(skb);
117 		segs = ops->callbacks.gso_segment(skb, features);
118 	}
119 
120 	if (IS_ERR_OR_NULL(segs))
121 		goto out;
122 
123 	gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
124 
125 	for (skb = segs; skb; skb = skb->next) {
126 		ipv6h = (struct ipv6hdr *)(skb_mac_header(skb) + nhoff);
127 		if (gso_partial && skb_is_gso(skb))
128 			payload_len = skb_shinfo(skb)->gso_size +
129 				      SKB_GSO_CB(skb)->data_offset +
130 				      skb->head - (unsigned char *)(ipv6h + 1);
131 		else
132 			payload_len = skb->len - nhoff - sizeof(*ipv6h);
133 		ipv6h->payload_len = htons(payload_len);
134 		skb->network_header = (u8 *)ipv6h - skb->head;
135 		skb_reset_mac_len(skb);
136 
137 		if (udpfrag) {
138 			int err = ip6_find_1stfragopt(skb, &prevhdr);
139 			if (err < 0) {
140 				kfree_skb_list(segs);
141 				return ERR_PTR(err);
142 			}
143 			fptr = (struct frag_hdr *)((u8 *)ipv6h + err);
144 			fptr->frag_off = htons(offset);
145 			if (skb->next)
146 				fptr->frag_off |= htons(IP6_MF);
147 			offset += (ntohs(ipv6h->payload_len) -
148 				   sizeof(struct frag_hdr));
149 		}
150 		if (encap)
151 			skb_reset_inner_headers(skb);
152 	}
153 
154 out:
155 	return segs;
156 }
157 
158 /* Return the total length of all the extension hdrs, following the same
159  * logic in ipv6_gso_pull_exthdrs() when parsing ext-hdrs.
160  */
161 static int ipv6_exthdrs_len(struct ipv6hdr *iph,
162 			    const struct net_offload **opps)
163 {
164 	struct ipv6_opt_hdr *opth = (void *)iph;
165 	int len = 0, proto, optlen = sizeof(*iph);
166 
167 	proto = iph->nexthdr;
168 	for (;;) {
169 		if (proto != NEXTHDR_HOP) {
170 			*opps = rcu_dereference(inet6_offloads[proto]);
171 			if (unlikely(!(*opps)))
172 				break;
173 			if (!((*opps)->flags & INET6_PROTO_GSO_EXTHDR))
174 				break;
175 		}
176 		opth = (void *)opth + optlen;
177 		optlen = ipv6_optlen(opth);
178 		len += optlen;
179 		proto = opth->nexthdr;
180 	}
181 	return len;
182 }
183 
184 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp6_gro_receive(struct list_head *,
185 							   struct sk_buff *));
186 INDIRECT_CALLABLE_DECLARE(struct sk_buff *udp6_gro_receive(struct list_head *,
187 							   struct sk_buff *));
188 INDIRECT_CALLABLE_SCOPE struct sk_buff *ipv6_gro_receive(struct list_head *head,
189 							 struct sk_buff *skb)
190 {
191 	const struct net_offload *ops;
192 	struct sk_buff *pp = NULL;
193 	struct sk_buff *p;
194 	struct ipv6hdr *iph;
195 	unsigned int nlen;
196 	unsigned int hlen;
197 	unsigned int off;
198 	u16 flush = 1;
199 	int proto;
200 
201 	off = skb_gro_offset(skb);
202 	hlen = off + sizeof(*iph);
203 	iph = skb_gro_header_fast(skb, off);
204 	if (skb_gro_header_hard(skb, hlen)) {
205 		iph = skb_gro_header_slow(skb, hlen, off);
206 		if (unlikely(!iph))
207 			goto out;
208 	}
209 
210 	skb_set_network_header(skb, off);
211 	skb_gro_pull(skb, sizeof(*iph));
212 	skb_set_transport_header(skb, skb_gro_offset(skb));
213 
214 	flush += ntohs(iph->payload_len) != skb_gro_len(skb);
215 
216 	rcu_read_lock();
217 	proto = iph->nexthdr;
218 	ops = rcu_dereference(inet6_offloads[proto]);
219 	if (!ops || !ops->callbacks.gro_receive) {
220 		__pskb_pull(skb, skb_gro_offset(skb));
221 		skb_gro_frag0_invalidate(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 	list_for_each_entry(p, head, list) {
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 sufficient, right?
254 		 */
255 		 if ((first_word & htonl(0xF00FFFFF)) ||
256 		    !ipv6_addr_equal(&iph->saddr, &iph2->saddr) ||
257 		    !ipv6_addr_equal(&iph->daddr, &iph2->daddr) ||
258 		    *(u16 *)&iph->nexthdr != *(u16 *)&iph2->nexthdr) {
259 not_same_flow:
260 			NAPI_GRO_CB(p)->same_flow = 0;
261 			continue;
262 		}
263 		if (unlikely(nlen > sizeof(struct ipv6hdr))) {
264 			if (memcmp(iph + 1, iph2 + 1,
265 				   nlen - sizeof(struct ipv6hdr)))
266 				goto not_same_flow;
267 		}
268 		/* flush if Traffic Class fields are different */
269 		NAPI_GRO_CB(p)->flush |= !!(first_word & htonl(0x0FF00000));
270 		NAPI_GRO_CB(p)->flush |= flush;
271 
272 		/* If the previous IP ID value was based on an atomic
273 		 * datagram we can overwrite the value and ignore it.
274 		 */
275 		if (NAPI_GRO_CB(skb)->is_atomic)
276 			NAPI_GRO_CB(p)->flush_id = 0;
277 	}
278 
279 	NAPI_GRO_CB(skb)->is_atomic = true;
280 	NAPI_GRO_CB(skb)->flush |= flush;
281 
282 	skb_gro_postpull_rcsum(skb, iph, nlen);
283 
284 	pp = indirect_call_gro_receive_l4(tcp6_gro_receive, udp6_gro_receive,
285 					 ops->callbacks.gro_receive, head, skb);
286 
287 out_unlock:
288 	rcu_read_unlock();
289 
290 out:
291 	skb_gro_flush_final(skb, pp, flush);
292 
293 	return pp;
294 }
295 
296 static struct sk_buff *sit_ip6ip6_gro_receive(struct list_head *head,
297 					      struct sk_buff *skb)
298 {
299 	/* Common GRO receive for SIT and IP6IP6 */
300 
301 	if (NAPI_GRO_CB(skb)->encap_mark) {
302 		NAPI_GRO_CB(skb)->flush = 1;
303 		return NULL;
304 	}
305 
306 	NAPI_GRO_CB(skb)->encap_mark = 1;
307 
308 	return ipv6_gro_receive(head, skb);
309 }
310 
311 static struct sk_buff *ip4ip6_gro_receive(struct list_head *head,
312 					  struct sk_buff *skb)
313 {
314 	/* Common GRO receive for SIT and IP6IP6 */
315 
316 	if (NAPI_GRO_CB(skb)->encap_mark) {
317 		NAPI_GRO_CB(skb)->flush = 1;
318 		return NULL;
319 	}
320 
321 	NAPI_GRO_CB(skb)->encap_mark = 1;
322 
323 	return inet_gro_receive(head, skb);
324 }
325 
326 INDIRECT_CALLABLE_DECLARE(int tcp6_gro_complete(struct sk_buff *, int));
327 INDIRECT_CALLABLE_DECLARE(int udp6_gro_complete(struct sk_buff *, int));
328 INDIRECT_CALLABLE_SCOPE int ipv6_gro_complete(struct sk_buff *skb, int nhoff)
329 {
330 	const struct net_offload *ops;
331 	struct ipv6hdr *iph = (struct ipv6hdr *)(skb->data + nhoff);
332 	int err = -ENOSYS;
333 
334 	if (skb->encapsulation) {
335 		skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IPV6));
336 		skb_set_inner_network_header(skb, nhoff);
337 	}
338 
339 	iph->payload_len = htons(skb->len - nhoff - sizeof(*iph));
340 
341 	rcu_read_lock();
342 
343 	nhoff += sizeof(*iph) + ipv6_exthdrs_len(iph, &ops);
344 	if (WARN_ON(!ops || !ops->callbacks.gro_complete))
345 		goto out_unlock;
346 
347 	err = INDIRECT_CALL_L4(ops->callbacks.gro_complete, tcp6_gro_complete,
348 			       udp6_gro_complete, skb, nhoff);
349 
350 out_unlock:
351 	rcu_read_unlock();
352 
353 	return err;
354 }
355 
356 static int sit_gro_complete(struct sk_buff *skb, int nhoff)
357 {
358 	skb->encapsulation = 1;
359 	skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
360 	return ipv6_gro_complete(skb, nhoff);
361 }
362 
363 static int ip6ip6_gro_complete(struct sk_buff *skb, int nhoff)
364 {
365 	skb->encapsulation = 1;
366 	skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP6;
367 	return ipv6_gro_complete(skb, nhoff);
368 }
369 
370 static int ip4ip6_gro_complete(struct sk_buff *skb, int nhoff)
371 {
372 	skb->encapsulation = 1;
373 	skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP6;
374 	return inet_gro_complete(skb, nhoff);
375 }
376 
377 static struct packet_offload ipv6_packet_offload __read_mostly = {
378 	.type = cpu_to_be16(ETH_P_IPV6),
379 	.callbacks = {
380 		.gso_segment = ipv6_gso_segment,
381 		.gro_receive = ipv6_gro_receive,
382 		.gro_complete = ipv6_gro_complete,
383 	},
384 };
385 
386 static struct sk_buff *sit_gso_segment(struct sk_buff *skb,
387 				       netdev_features_t features)
388 {
389 	if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
390 		return ERR_PTR(-EINVAL);
391 
392 	return ipv6_gso_segment(skb, features);
393 }
394 
395 static struct sk_buff *ip4ip6_gso_segment(struct sk_buff *skb,
396 					  netdev_features_t features)
397 {
398 	if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP6))
399 		return ERR_PTR(-EINVAL);
400 
401 	return inet_gso_segment(skb, features);
402 }
403 
404 static struct sk_buff *ip6ip6_gso_segment(struct sk_buff *skb,
405 					  netdev_features_t features)
406 {
407 	if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP6))
408 		return ERR_PTR(-EINVAL);
409 
410 	return ipv6_gso_segment(skb, features);
411 }
412 
413 static const struct net_offload sit_offload = {
414 	.callbacks = {
415 		.gso_segment	= sit_gso_segment,
416 		.gro_receive    = sit_ip6ip6_gro_receive,
417 		.gro_complete   = sit_gro_complete,
418 	},
419 };
420 
421 static const struct net_offload ip4ip6_offload = {
422 	.callbacks = {
423 		.gso_segment	= ip4ip6_gso_segment,
424 		.gro_receive    = ip4ip6_gro_receive,
425 		.gro_complete   = ip4ip6_gro_complete,
426 	},
427 };
428 
429 static const struct net_offload ip6ip6_offload = {
430 	.callbacks = {
431 		.gso_segment	= ip6ip6_gso_segment,
432 		.gro_receive    = sit_ip6ip6_gro_receive,
433 		.gro_complete   = ip6ip6_gro_complete,
434 	},
435 };
436 static int __init ipv6_offload_init(void)
437 {
438 
439 	if (tcpv6_offload_init() < 0)
440 		pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
441 	if (ipv6_exthdrs_offload_init() < 0)
442 		pr_crit("%s: Cannot add EXTHDRS protocol offload\n", __func__);
443 
444 	dev_add_offload(&ipv6_packet_offload);
445 
446 	inet_add_offload(&sit_offload, IPPROTO_IPV6);
447 	inet6_add_offload(&ip6ip6_offload, IPPROTO_IPV6);
448 	inet6_add_offload(&ip4ip6_offload, IPPROTO_IPIP);
449 
450 	return 0;
451 }
452 
453 fs_initcall(ipv6_offload_init);
454