xref: /openbmc/linux/net/ipv6/ip6_offload.c (revision fe4549b1)
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 #include <net/gso.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_gro_pull_exthdrs(struct sk_buff *skb, int off, int proto)
41 {
42 	const struct net_offload *ops = NULL;
43 	struct ipv6_opt_hdr *opth;
44 
45 	for (;;) {
46 		int len;
47 
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 		opth = skb_gro_header(skb, off + sizeof(*opth), off);
57 		if (unlikely(!opth))
58 			break;
59 
60 		len = ipv6_optlen(opth);
61 
62 		opth = skb_gro_header(skb, off + len, off);
63 		if (unlikely(!opth))
64 			break;
65 		proto = opth->nexthdr;
66 
67 		off += len;
68 	}
69 
70 	skb_gro_pull(skb, off - skb_network_offset(skb));
71 	return proto;
72 }
73 
74 static int ipv6_gso_pull_exthdrs(struct sk_buff *skb, int proto)
75 {
76 	const struct net_offload *ops = NULL;
77 
78 	for (;;) {
79 		struct ipv6_opt_hdr *opth;
80 		int len;
81 
82 		if (proto != NEXTHDR_HOP) {
83 			ops = rcu_dereference(inet6_offloads[proto]);
84 
85 			if (unlikely(!ops))
86 				break;
87 
88 			if (!(ops->flags & INET6_PROTO_GSO_EXTHDR))
89 				break;
90 		}
91 
92 		if (unlikely(!pskb_may_pull(skb, 8)))
93 			break;
94 
95 		opth = (void *)skb->data;
96 		len = ipv6_optlen(opth);
97 
98 		if (unlikely(!pskb_may_pull(skb, len)))
99 			break;
100 
101 		opth = (void *)skb->data;
102 		proto = opth->nexthdr;
103 		__skb_pull(skb, len);
104 	}
105 
106 	return proto;
107 }
108 
109 static struct sk_buff *ipv6_gso_segment(struct sk_buff *skb,
110 	netdev_features_t features)
111 {
112 	struct sk_buff *segs = ERR_PTR(-EINVAL);
113 	struct ipv6hdr *ipv6h;
114 	const struct net_offload *ops;
115 	int proto, err;
116 	struct frag_hdr *fptr;
117 	unsigned int payload_len;
118 	u8 *prevhdr;
119 	int offset = 0;
120 	bool encap, udpfrag;
121 	int nhoff;
122 	bool gso_partial;
123 
124 	skb_reset_network_header(skb);
125 	err = ipv6_hopopt_jumbo_remove(skb);
126 	if (err)
127 		return ERR_PTR(err);
128 	nhoff = skb_network_header(skb) - skb_mac_header(skb);
129 	if (unlikely(!pskb_may_pull(skb, sizeof(*ipv6h))))
130 		goto out;
131 
132 	encap = SKB_GSO_CB(skb)->encap_level > 0;
133 	if (encap)
134 		features &= skb->dev->hw_enc_features;
135 	SKB_GSO_CB(skb)->encap_level += sizeof(*ipv6h);
136 
137 	ipv6h = ipv6_hdr(skb);
138 	__skb_pull(skb, sizeof(*ipv6h));
139 	segs = ERR_PTR(-EPROTONOSUPPORT);
140 
141 	proto = ipv6_gso_pull_exthdrs(skb, ipv6h->nexthdr);
142 
143 	if (skb->encapsulation &&
144 	    skb_shinfo(skb)->gso_type & (SKB_GSO_IPXIP4 | SKB_GSO_IPXIP6))
145 		udpfrag = proto == IPPROTO_UDP && encap &&
146 			  (skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
147 	else
148 		udpfrag = proto == IPPROTO_UDP && !skb->encapsulation &&
149 			  (skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
150 
151 	ops = rcu_dereference(inet6_offloads[proto]);
152 	if (likely(ops && ops->callbacks.gso_segment)) {
153 		skb_reset_transport_header(skb);
154 		segs = ops->callbacks.gso_segment(skb, features);
155 		if (!segs)
156 			skb->network_header = skb_mac_header(skb) + nhoff - skb->head;
157 	}
158 
159 	if (IS_ERR_OR_NULL(segs))
160 		goto out;
161 
162 	gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
163 
164 	for (skb = segs; skb; skb = skb->next) {
165 		ipv6h = (struct ipv6hdr *)(skb_mac_header(skb) + nhoff);
166 		if (gso_partial && skb_is_gso(skb))
167 			payload_len = skb_shinfo(skb)->gso_size +
168 				      SKB_GSO_CB(skb)->data_offset +
169 				      skb->head - (unsigned char *)(ipv6h + 1);
170 		else
171 			payload_len = skb->len - nhoff - sizeof(*ipv6h);
172 		ipv6h->payload_len = htons(payload_len);
173 		skb->network_header = (u8 *)ipv6h - skb->head;
174 		skb_reset_mac_len(skb);
175 
176 		if (udpfrag) {
177 			int err = ip6_find_1stfragopt(skb, &prevhdr);
178 			if (err < 0) {
179 				kfree_skb_list(segs);
180 				return ERR_PTR(err);
181 			}
182 			fptr = (struct frag_hdr *)((u8 *)ipv6h + err);
183 			fptr->frag_off = htons(offset);
184 			if (skb->next)
185 				fptr->frag_off |= htons(IP6_MF);
186 			offset += (ntohs(ipv6h->payload_len) -
187 				   sizeof(struct frag_hdr));
188 		}
189 		if (encap)
190 			skb_reset_inner_headers(skb);
191 	}
192 
193 out:
194 	return segs;
195 }
196 
197 /* Return the total length of all the extension hdrs, following the same
198  * logic in ipv6_gso_pull_exthdrs() when parsing ext-hdrs.
199  */
200 static int ipv6_exthdrs_len(struct ipv6hdr *iph,
201 			    const struct net_offload **opps)
202 {
203 	struct ipv6_opt_hdr *opth = (void *)iph;
204 	int len = 0, proto, optlen = sizeof(*iph);
205 
206 	proto = iph->nexthdr;
207 	for (;;) {
208 		if (proto != NEXTHDR_HOP) {
209 			*opps = rcu_dereference(inet6_offloads[proto]);
210 			if (unlikely(!(*opps)))
211 				break;
212 			if (!((*opps)->flags & INET6_PROTO_GSO_EXTHDR))
213 				break;
214 		}
215 		opth = (void *)opth + optlen;
216 		optlen = ipv6_optlen(opth);
217 		len += optlen;
218 		proto = opth->nexthdr;
219 	}
220 	return len;
221 }
222 
223 INDIRECT_CALLABLE_SCOPE struct sk_buff *ipv6_gro_receive(struct list_head *head,
224 							 struct sk_buff *skb)
225 {
226 	const struct net_offload *ops;
227 	struct sk_buff *pp = NULL;
228 	struct sk_buff *p;
229 	struct ipv6hdr *iph;
230 	unsigned int nlen;
231 	unsigned int hlen;
232 	unsigned int off;
233 	u16 flush = 1;
234 	int proto;
235 
236 	off = skb_gro_offset(skb);
237 	hlen = off + sizeof(*iph);
238 	iph = skb_gro_header(skb, hlen, off);
239 	if (unlikely(!iph))
240 		goto out;
241 
242 	skb_set_network_header(skb, off);
243 	NAPI_GRO_CB(skb)->inner_network_offset = off;
244 
245 	flush += ntohs(iph->payload_len) != skb->len - hlen;
246 
247 	proto = iph->nexthdr;
248 	ops = rcu_dereference(inet6_offloads[proto]);
249 	if (!ops || !ops->callbacks.gro_receive) {
250 		proto = ipv6_gro_pull_exthdrs(skb, hlen, proto);
251 
252 		ops = rcu_dereference(inet6_offloads[proto]);
253 		if (!ops || !ops->callbacks.gro_receive)
254 			goto out;
255 
256 		iph = skb_gro_network_header(skb);
257 	} else {
258 		skb_gro_pull(skb, sizeof(*iph));
259 	}
260 
261 	skb_set_transport_header(skb, skb_gro_offset(skb));
262 
263 	NAPI_GRO_CB(skb)->proto = proto;
264 
265 	flush--;
266 	nlen = skb_network_header_len(skb);
267 
268 	list_for_each_entry(p, head, list) {
269 		const struct ipv6hdr *iph2;
270 		__be32 first_word; /* <Version:4><Traffic_Class:8><Flow_Label:20> */
271 
272 		if (!NAPI_GRO_CB(p)->same_flow)
273 			continue;
274 
275 		iph2 = (struct ipv6hdr *)(p->data + off);
276 		first_word = *(__be32 *)iph ^ *(__be32 *)iph2;
277 
278 		/* All fields must match except length and Traffic Class.
279 		 * XXX skbs on the gro_list have all been parsed and pulled
280 		 * already so we don't need to compare nlen
281 		 * (nlen != (sizeof(*iph2) + ipv6_exthdrs_len(iph2, &ops)))
282 		 * memcmp() alone below is sufficient, right?
283 		 */
284 		 if ((first_word & htonl(0xF00FFFFF)) ||
285 		     !ipv6_addr_equal(&iph->saddr, &iph2->saddr) ||
286 		     !ipv6_addr_equal(&iph->daddr, &iph2->daddr) ||
287 		     iph->nexthdr != iph2->nexthdr) {
288 not_same_flow:
289 			NAPI_GRO_CB(p)->same_flow = 0;
290 			continue;
291 		}
292 		if (unlikely(nlen > sizeof(struct ipv6hdr))) {
293 			if (memcmp(iph + 1, iph2 + 1,
294 				   nlen - sizeof(struct ipv6hdr)))
295 				goto not_same_flow;
296 		}
297 		/* flush if Traffic Class fields are different */
298 		NAPI_GRO_CB(p)->flush |= !!((first_word & htonl(0x0FF00000)) |
299 			(__force __be32)(iph->hop_limit ^ iph2->hop_limit));
300 		NAPI_GRO_CB(p)->flush |= flush;
301 
302 		/* If the previous IP ID value was based on an atomic
303 		 * datagram we can overwrite the value and ignore it.
304 		 */
305 		if (NAPI_GRO_CB(skb)->is_atomic)
306 			NAPI_GRO_CB(p)->flush_id = 0;
307 	}
308 
309 	NAPI_GRO_CB(skb)->is_atomic = true;
310 	NAPI_GRO_CB(skb)->flush |= flush;
311 
312 	skb_gro_postpull_rcsum(skb, iph, nlen);
313 
314 	pp = indirect_call_gro_receive_l4(tcp6_gro_receive, udp6_gro_receive,
315 					 ops->callbacks.gro_receive, head, skb);
316 
317 out:
318 	skb_gro_flush_final(skb, pp, flush);
319 
320 	return pp;
321 }
322 
323 static struct sk_buff *sit_ip6ip6_gro_receive(struct list_head *head,
324 					      struct sk_buff *skb)
325 {
326 	/* Common GRO receive for SIT and IP6IP6 */
327 
328 	if (NAPI_GRO_CB(skb)->encap_mark) {
329 		NAPI_GRO_CB(skb)->flush = 1;
330 		return NULL;
331 	}
332 
333 	NAPI_GRO_CB(skb)->encap_mark = 1;
334 
335 	return ipv6_gro_receive(head, skb);
336 }
337 
338 static struct sk_buff *ip4ip6_gro_receive(struct list_head *head,
339 					  struct sk_buff *skb)
340 {
341 	/* Common GRO receive for SIT and IP6IP6 */
342 
343 	if (NAPI_GRO_CB(skb)->encap_mark) {
344 		NAPI_GRO_CB(skb)->flush = 1;
345 		return NULL;
346 	}
347 
348 	NAPI_GRO_CB(skb)->encap_mark = 1;
349 
350 	return inet_gro_receive(head, skb);
351 }
352 
353 INDIRECT_CALLABLE_SCOPE int ipv6_gro_complete(struct sk_buff *skb, int nhoff)
354 {
355 	const struct net_offload *ops;
356 	struct ipv6hdr *iph;
357 	int err = -ENOSYS;
358 	u32 payload_len;
359 
360 	if (skb->encapsulation) {
361 		skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IPV6));
362 		skb_set_inner_network_header(skb, nhoff);
363 	}
364 
365 	payload_len = skb->len - nhoff - sizeof(*iph);
366 	if (unlikely(payload_len > IPV6_MAXPLEN)) {
367 		struct hop_jumbo_hdr *hop_jumbo;
368 		int hoplen = sizeof(*hop_jumbo);
369 
370 		/* Move network header left */
371 		memmove(skb_mac_header(skb) - hoplen, skb_mac_header(skb),
372 			skb->transport_header - skb->mac_header);
373 		skb->data -= hoplen;
374 		skb->len += hoplen;
375 		skb->mac_header -= hoplen;
376 		skb->network_header -= hoplen;
377 		iph = (struct ipv6hdr *)(skb->data + nhoff);
378 		hop_jumbo = (struct hop_jumbo_hdr *)(iph + 1);
379 
380 		/* Build hop-by-hop options */
381 		hop_jumbo->nexthdr = iph->nexthdr;
382 		hop_jumbo->hdrlen = 0;
383 		hop_jumbo->tlv_type = IPV6_TLV_JUMBO;
384 		hop_jumbo->tlv_len = 4;
385 		hop_jumbo->jumbo_payload_len = htonl(payload_len + hoplen);
386 
387 		iph->nexthdr = NEXTHDR_HOP;
388 		iph->payload_len = 0;
389 	} else {
390 		iph = (struct ipv6hdr *)(skb->data + nhoff);
391 		iph->payload_len = htons(payload_len);
392 	}
393 
394 	nhoff += sizeof(*iph) + ipv6_exthdrs_len(iph, &ops);
395 	if (WARN_ON(!ops || !ops->callbacks.gro_complete))
396 		goto out;
397 
398 	err = INDIRECT_CALL_L4(ops->callbacks.gro_complete, tcp6_gro_complete,
399 			       udp6_gro_complete, skb, nhoff);
400 
401 out:
402 	return err;
403 }
404 
405 static int sit_gro_complete(struct sk_buff *skb, int nhoff)
406 {
407 	skb->encapsulation = 1;
408 	skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
409 	return ipv6_gro_complete(skb, nhoff);
410 }
411 
412 static int ip6ip6_gro_complete(struct sk_buff *skb, int nhoff)
413 {
414 	skb->encapsulation = 1;
415 	skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP6;
416 	return ipv6_gro_complete(skb, nhoff);
417 }
418 
419 static int ip4ip6_gro_complete(struct sk_buff *skb, int nhoff)
420 {
421 	skb->encapsulation = 1;
422 	skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP6;
423 	return inet_gro_complete(skb, nhoff);
424 }
425 
426 static struct packet_offload ipv6_packet_offload __read_mostly = {
427 	.type = cpu_to_be16(ETH_P_IPV6),
428 	.callbacks = {
429 		.gso_segment = ipv6_gso_segment,
430 		.gro_receive = ipv6_gro_receive,
431 		.gro_complete = ipv6_gro_complete,
432 	},
433 };
434 
435 static struct sk_buff *sit_gso_segment(struct sk_buff *skb,
436 				       netdev_features_t features)
437 {
438 	if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
439 		return ERR_PTR(-EINVAL);
440 
441 	return ipv6_gso_segment(skb, features);
442 }
443 
444 static struct sk_buff *ip4ip6_gso_segment(struct sk_buff *skb,
445 					  netdev_features_t features)
446 {
447 	if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP6))
448 		return ERR_PTR(-EINVAL);
449 
450 	return inet_gso_segment(skb, features);
451 }
452 
453 static struct sk_buff *ip6ip6_gso_segment(struct sk_buff *skb,
454 					  netdev_features_t features)
455 {
456 	if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP6))
457 		return ERR_PTR(-EINVAL);
458 
459 	return ipv6_gso_segment(skb, features);
460 }
461 
462 static const struct net_offload sit_offload = {
463 	.callbacks = {
464 		.gso_segment	= sit_gso_segment,
465 		.gro_receive    = sit_ip6ip6_gro_receive,
466 		.gro_complete   = sit_gro_complete,
467 	},
468 };
469 
470 static const struct net_offload ip4ip6_offload = {
471 	.callbacks = {
472 		.gso_segment	= ip4ip6_gso_segment,
473 		.gro_receive    = ip4ip6_gro_receive,
474 		.gro_complete   = ip4ip6_gro_complete,
475 	},
476 };
477 
478 static const struct net_offload ip6ip6_offload = {
479 	.callbacks = {
480 		.gso_segment	= ip6ip6_gso_segment,
481 		.gro_receive    = sit_ip6ip6_gro_receive,
482 		.gro_complete   = ip6ip6_gro_complete,
483 	},
484 };
485 static int __init ipv6_offload_init(void)
486 {
487 
488 	if (tcpv6_offload_init() < 0)
489 		pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
490 	if (ipv6_exthdrs_offload_init() < 0)
491 		pr_crit("%s: Cannot add EXTHDRS protocol offload\n", __func__);
492 
493 	dev_add_offload(&ipv6_packet_offload);
494 
495 	inet_add_offload(&sit_offload, IPPROTO_IPV6);
496 	inet6_add_offload(&ip6ip6_offload, IPPROTO_IPV6);
497 	inet6_add_offload(&ip4ip6_offload, IPPROTO_IPIP);
498 
499 	return 0;
500 }
501 
502 fs_initcall(ipv6_offload_init);
503