xref: /openbmc/linux/net/ipv6/ip6_offload.c (revision 99a15348)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *	IPV6 GSO/GRO offload support
4  *	Linux INET6 implementation
5  */
6 
7 #include <linux/kernel.h>
8 #include <linux/socket.h>
9 #include <linux/netdevice.h>
10 #include <linux/skbuff.h>
11 #include <linux/printk.h>
12 
13 #include <net/protocol.h>
14 #include <net/ipv6.h>
15 #include <net/inet_common.h>
16 #include <net/tcp.h>
17 #include <net/udp.h>
18 #include <net/gro.h>
19 
20 #include "ip6_offload.h"
21 
22 /* All GRO functions are always builtin, except UDP over ipv6, which lays in
23  * ipv6 module, as it depends on UDPv6 lookup function, so we need special care
24  * when ipv6 is built as a module
25  */
26 #if IS_BUILTIN(CONFIG_IPV6)
27 #define INDIRECT_CALL_L4(f, f2, f1, ...) INDIRECT_CALL_2(f, f2, f1, __VA_ARGS__)
28 #else
29 #define INDIRECT_CALL_L4(f, f2, f1, ...) INDIRECT_CALL_1(f, f2, __VA_ARGS__)
30 #endif
31 
32 #define indirect_call_gro_receive_l4(f2, f1, cb, head, skb)	\
33 ({								\
34 	unlikely(gro_recursion_inc_test(skb)) ?			\
35 		NAPI_GRO_CB(skb)->flush |= 1, NULL :		\
36 		INDIRECT_CALL_L4(cb, f2, f1, head, skb);	\
37 })
38 
39 static int ipv6_gso_pull_exthdrs(struct sk_buff *skb, int proto)
40 {
41 	const struct net_offload *ops = NULL;
42 
43 	for (;;) {
44 		struct ipv6_opt_hdr *opth;
45 		int len;
46 
47 		if (proto != NEXTHDR_HOP) {
48 			ops = rcu_dereference(inet6_offloads[proto]);
49 
50 			if (unlikely(!ops))
51 				break;
52 
53 			if (!(ops->flags & INET6_PROTO_GSO_EXTHDR))
54 				break;
55 		}
56 
57 		if (unlikely(!pskb_may_pull(skb, 8)))
58 			break;
59 
60 		opth = (void *)skb->data;
61 		len = ipv6_optlen(opth);
62 
63 		if (unlikely(!pskb_may_pull(skb, len)))
64 			break;
65 
66 		opth = (void *)skb->data;
67 		proto = opth->nexthdr;
68 		__skb_pull(skb, len);
69 	}
70 
71 	return proto;
72 }
73 
74 static struct sk_buff *ipv6_gso_segment(struct sk_buff *skb,
75 	netdev_features_t features)
76 {
77 	struct sk_buff *segs = ERR_PTR(-EINVAL);
78 	struct ipv6hdr *ipv6h;
79 	const struct net_offload *ops;
80 	int proto;
81 	struct frag_hdr *fptr;
82 	unsigned int payload_len;
83 	u8 *prevhdr;
84 	int offset = 0;
85 	bool encap, udpfrag;
86 	int nhoff;
87 	bool gso_partial;
88 
89 	skb_reset_network_header(skb);
90 	nhoff = skb_network_header(skb) - skb_mac_header(skb);
91 	if (unlikely(!pskb_may_pull(skb, sizeof(*ipv6h))))
92 		goto out;
93 
94 	encap = SKB_GSO_CB(skb)->encap_level > 0;
95 	if (encap)
96 		features &= skb->dev->hw_enc_features;
97 	SKB_GSO_CB(skb)->encap_level += sizeof(*ipv6h);
98 
99 	ipv6h = ipv6_hdr(skb);
100 	__skb_pull(skb, sizeof(*ipv6h));
101 	segs = ERR_PTR(-EPROTONOSUPPORT);
102 
103 	proto = ipv6_gso_pull_exthdrs(skb, ipv6h->nexthdr);
104 
105 	if (skb->encapsulation &&
106 	    skb_shinfo(skb)->gso_type & (SKB_GSO_IPXIP4 | SKB_GSO_IPXIP6))
107 		udpfrag = proto == IPPROTO_UDP && encap &&
108 			  (skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
109 	else
110 		udpfrag = proto == IPPROTO_UDP && !skb->encapsulation &&
111 			  (skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
112 
113 	ops = rcu_dereference(inet6_offloads[proto]);
114 	if (likely(ops && ops->callbacks.gso_segment)) {
115 		skb_reset_transport_header(skb);
116 		segs = ops->callbacks.gso_segment(skb, features);
117 		if (!segs)
118 			skb->network_header = skb_mac_header(skb) + nhoff - skb->head;
119 	}
120 
121 	if (IS_ERR_OR_NULL(segs))
122 		goto out;
123 
124 	gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
125 
126 	for (skb = segs; skb; skb = skb->next) {
127 		ipv6h = (struct ipv6hdr *)(skb_mac_header(skb) + nhoff);
128 		if (gso_partial && skb_is_gso(skb))
129 			payload_len = skb_shinfo(skb)->gso_size +
130 				      SKB_GSO_CB(skb)->data_offset +
131 				      skb->head - (unsigned char *)(ipv6h + 1);
132 		else
133 			payload_len = skb->len - nhoff - sizeof(*ipv6h);
134 		ipv6h->payload_len = htons(payload_len);
135 		skb->network_header = (u8 *)ipv6h - skb->head;
136 		skb_reset_mac_len(skb);
137 
138 		if (udpfrag) {
139 			int err = ip6_find_1stfragopt(skb, &prevhdr);
140 			if (err < 0) {
141 				kfree_skb_list(segs);
142 				return ERR_PTR(err);
143 			}
144 			fptr = (struct frag_hdr *)((u8 *)ipv6h + err);
145 			fptr->frag_off = htons(offset);
146 			if (skb->next)
147 				fptr->frag_off |= htons(IP6_MF);
148 			offset += (ntohs(ipv6h->payload_len) -
149 				   sizeof(struct frag_hdr));
150 		}
151 		if (encap)
152 			skb_reset_inner_headers(skb);
153 	}
154 
155 out:
156 	return segs;
157 }
158 
159 /* Return the total length of all the extension hdrs, following the same
160  * logic in ipv6_gso_pull_exthdrs() when parsing ext-hdrs.
161  */
162 static int ipv6_exthdrs_len(struct ipv6hdr *iph,
163 			    const struct net_offload **opps)
164 {
165 	struct ipv6_opt_hdr *opth = (void *)iph;
166 	int len = 0, proto, optlen = sizeof(*iph);
167 
168 	proto = iph->nexthdr;
169 	for (;;) {
170 		if (proto != NEXTHDR_HOP) {
171 			*opps = rcu_dereference(inet6_offloads[proto]);
172 			if (unlikely(!(*opps)))
173 				break;
174 			if (!((*opps)->flags & INET6_PROTO_GSO_EXTHDR))
175 				break;
176 		}
177 		opth = (void *)opth + optlen;
178 		optlen = ipv6_optlen(opth);
179 		len += optlen;
180 		proto = opth->nexthdr;
181 	}
182 	return len;
183 }
184 
185 INDIRECT_CALLABLE_SCOPE struct sk_buff *ipv6_gro_receive(struct list_head *head,
186 							 struct sk_buff *skb)
187 {
188 	const struct net_offload *ops;
189 	struct sk_buff *pp = NULL;
190 	struct sk_buff *p;
191 	struct ipv6hdr *iph;
192 	unsigned int nlen;
193 	unsigned int hlen;
194 	unsigned int off;
195 	u16 flush = 1;
196 	int proto;
197 
198 	off = skb_gro_offset(skb);
199 	hlen = off + sizeof(*iph);
200 	iph = skb_gro_header_fast(skb, off);
201 	if (skb_gro_header_hard(skb, hlen)) {
202 		iph = skb_gro_header_slow(skb, hlen, off);
203 		if (unlikely(!iph))
204 			goto out;
205 	}
206 
207 	skb_set_network_header(skb, off);
208 	skb_gro_pull(skb, sizeof(*iph));
209 	skb_set_transport_header(skb, skb_gro_offset(skb));
210 
211 	flush += ntohs(iph->payload_len) != skb_gro_len(skb);
212 
213 	proto = iph->nexthdr;
214 	ops = rcu_dereference(inet6_offloads[proto]);
215 	if (!ops || !ops->callbacks.gro_receive) {
216 		__pskb_pull(skb, skb_gro_offset(skb));
217 		skb_gro_frag0_invalidate(skb);
218 		proto = ipv6_gso_pull_exthdrs(skb, proto);
219 		skb_gro_pull(skb, -skb_transport_offset(skb));
220 		skb_reset_transport_header(skb);
221 		__skb_push(skb, skb_gro_offset(skb));
222 
223 		ops = rcu_dereference(inet6_offloads[proto]);
224 		if (!ops || !ops->callbacks.gro_receive)
225 			goto out;
226 
227 		iph = ipv6_hdr(skb);
228 	}
229 
230 	NAPI_GRO_CB(skb)->proto = proto;
231 
232 	flush--;
233 	nlen = skb_network_header_len(skb);
234 
235 	list_for_each_entry(p, head, list) {
236 		const struct ipv6hdr *iph2;
237 		__be32 first_word; /* <Version:4><Traffic_Class:8><Flow_Label:20> */
238 
239 		if (!NAPI_GRO_CB(p)->same_flow)
240 			continue;
241 
242 		iph2 = (struct ipv6hdr *)(p->data + off);
243 		first_word = *(__be32 *)iph ^ *(__be32 *)iph2;
244 
245 		/* All fields must match except length and Traffic Class.
246 		 * XXX skbs on the gro_list have all been parsed and pulled
247 		 * already so we don't need to compare nlen
248 		 * (nlen != (sizeof(*iph2) + ipv6_exthdrs_len(iph2, &ops)))
249 		 * memcmp() alone below is sufficient, right?
250 		 */
251 		 if ((first_word & htonl(0xF00FFFFF)) ||
252 		     !ipv6_addr_equal(&iph->saddr, &iph2->saddr) ||
253 		     !ipv6_addr_equal(&iph->daddr, &iph2->daddr) ||
254 		     iph->nexthdr != iph2->nexthdr) {
255 not_same_flow:
256 			NAPI_GRO_CB(p)->same_flow = 0;
257 			continue;
258 		}
259 		if (unlikely(nlen > sizeof(struct ipv6hdr))) {
260 			if (memcmp(iph + 1, iph2 + 1,
261 				   nlen - sizeof(struct ipv6hdr)))
262 				goto not_same_flow;
263 		}
264 		/* flush if Traffic Class fields are different */
265 		NAPI_GRO_CB(p)->flush |= !!((first_word & htonl(0x0FF00000)) |
266 			(__force __be32)(iph->hop_limit ^ iph2->hop_limit));
267 		NAPI_GRO_CB(p)->flush |= flush;
268 
269 		/* If the previous IP ID value was based on an atomic
270 		 * datagram we can overwrite the value and ignore it.
271 		 */
272 		if (NAPI_GRO_CB(skb)->is_atomic)
273 			NAPI_GRO_CB(p)->flush_id = 0;
274 	}
275 
276 	NAPI_GRO_CB(skb)->is_atomic = true;
277 	NAPI_GRO_CB(skb)->flush |= flush;
278 
279 	skb_gro_postpull_rcsum(skb, iph, nlen);
280 
281 	pp = indirect_call_gro_receive_l4(tcp6_gro_receive, udp6_gro_receive,
282 					 ops->callbacks.gro_receive, head, skb);
283 
284 out:
285 	skb_gro_flush_final(skb, pp, flush);
286 
287 	return pp;
288 }
289 
290 static struct sk_buff *sit_ip6ip6_gro_receive(struct list_head *head,
291 					      struct sk_buff *skb)
292 {
293 	/* Common GRO receive for SIT and IP6IP6 */
294 
295 	if (NAPI_GRO_CB(skb)->encap_mark) {
296 		NAPI_GRO_CB(skb)->flush = 1;
297 		return NULL;
298 	}
299 
300 	NAPI_GRO_CB(skb)->encap_mark = 1;
301 
302 	return ipv6_gro_receive(head, skb);
303 }
304 
305 static struct sk_buff *ip4ip6_gro_receive(struct list_head *head,
306 					  struct sk_buff *skb)
307 {
308 	/* Common GRO receive for SIT and IP6IP6 */
309 
310 	if (NAPI_GRO_CB(skb)->encap_mark) {
311 		NAPI_GRO_CB(skb)->flush = 1;
312 		return NULL;
313 	}
314 
315 	NAPI_GRO_CB(skb)->encap_mark = 1;
316 
317 	return inet_gro_receive(head, skb);
318 }
319 
320 INDIRECT_CALLABLE_SCOPE int ipv6_gro_complete(struct sk_buff *skb, int nhoff)
321 {
322 	const struct net_offload *ops;
323 	struct ipv6hdr *iph = (struct ipv6hdr *)(skb->data + nhoff);
324 	int err = -ENOSYS;
325 
326 	if (skb->encapsulation) {
327 		skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IPV6));
328 		skb_set_inner_network_header(skb, nhoff);
329 	}
330 
331 	iph->payload_len = htons(skb->len - nhoff - sizeof(*iph));
332 
333 	nhoff += sizeof(*iph) + ipv6_exthdrs_len(iph, &ops);
334 	if (WARN_ON(!ops || !ops->callbacks.gro_complete))
335 		goto out;
336 
337 	err = INDIRECT_CALL_L4(ops->callbacks.gro_complete, tcp6_gro_complete,
338 			       udp6_gro_complete, skb, nhoff);
339 
340 out:
341 	return err;
342 }
343 
344 static int sit_gro_complete(struct sk_buff *skb, int nhoff)
345 {
346 	skb->encapsulation = 1;
347 	skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
348 	return ipv6_gro_complete(skb, nhoff);
349 }
350 
351 static int ip6ip6_gro_complete(struct sk_buff *skb, int nhoff)
352 {
353 	skb->encapsulation = 1;
354 	skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP6;
355 	return ipv6_gro_complete(skb, nhoff);
356 }
357 
358 static int ip4ip6_gro_complete(struct sk_buff *skb, int nhoff)
359 {
360 	skb->encapsulation = 1;
361 	skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP6;
362 	return inet_gro_complete(skb, nhoff);
363 }
364 
365 static struct packet_offload ipv6_packet_offload __read_mostly = {
366 	.type = cpu_to_be16(ETH_P_IPV6),
367 	.callbacks = {
368 		.gso_segment = ipv6_gso_segment,
369 		.gro_receive = ipv6_gro_receive,
370 		.gro_complete = ipv6_gro_complete,
371 	},
372 };
373 
374 static struct sk_buff *sit_gso_segment(struct sk_buff *skb,
375 				       netdev_features_t features)
376 {
377 	if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
378 		return ERR_PTR(-EINVAL);
379 
380 	return ipv6_gso_segment(skb, features);
381 }
382 
383 static struct sk_buff *ip4ip6_gso_segment(struct sk_buff *skb,
384 					  netdev_features_t features)
385 {
386 	if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP6))
387 		return ERR_PTR(-EINVAL);
388 
389 	return inet_gso_segment(skb, features);
390 }
391 
392 static struct sk_buff *ip6ip6_gso_segment(struct sk_buff *skb,
393 					  netdev_features_t features)
394 {
395 	if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP6))
396 		return ERR_PTR(-EINVAL);
397 
398 	return ipv6_gso_segment(skb, features);
399 }
400 
401 static const struct net_offload sit_offload = {
402 	.callbacks = {
403 		.gso_segment	= sit_gso_segment,
404 		.gro_receive    = sit_ip6ip6_gro_receive,
405 		.gro_complete   = sit_gro_complete,
406 	},
407 };
408 
409 static const struct net_offload ip4ip6_offload = {
410 	.callbacks = {
411 		.gso_segment	= ip4ip6_gso_segment,
412 		.gro_receive    = ip4ip6_gro_receive,
413 		.gro_complete   = ip4ip6_gro_complete,
414 	},
415 };
416 
417 static const struct net_offload ip6ip6_offload = {
418 	.callbacks = {
419 		.gso_segment	= ip6ip6_gso_segment,
420 		.gro_receive    = sit_ip6ip6_gro_receive,
421 		.gro_complete   = ip6ip6_gro_complete,
422 	},
423 };
424 static int __init ipv6_offload_init(void)
425 {
426 
427 	if (tcpv6_offload_init() < 0)
428 		pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
429 	if (ipv6_exthdrs_offload_init() < 0)
430 		pr_crit("%s: Cannot add EXTHDRS protocol offload\n", __func__);
431 
432 	dev_add_offload(&ipv6_packet_offload);
433 
434 	inet_add_offload(&sit_offload, IPPROTO_IPV6);
435 	inet6_add_offload(&ip6ip6_offload, IPPROTO_IPV6);
436 	inet6_add_offload(&ip4ip6_offload, IPPROTO_IPIP);
437 
438 	return 0;
439 }
440 
441 fs_initcall(ipv6_offload_init);
442