ip6_offload.c

  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
 40static 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_gro_receive_network_offset(skb));
 71	return proto;
 72}
 73
 74static 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		ops = rcu_dereference(inet6_offloads[proto]);
 83
 84		if (unlikely(!ops))
 85			break;
 86
 87		if (!(ops->flags & INET6_PROTO_GSO_EXTHDR))
 88			break;
 89
 90		if (unlikely(!pskb_may_pull(skb, 8)))
 91			break;
 92
 93		opth = (void *)skb->data;
 94		len = ipv6_optlen(opth);
 95
 96		if (unlikely(!pskb_may_pull(skb, len)))
 97			break;
 98
 99		opth = (void *)skb->data;
100		proto = opth->nexthdr;
101		__skb_pull(skb, len);
102	}
103
104	return proto;
105}
106
107static struct sk_buff *ipv6_gso_segment(struct sk_buff *skb,
108	netdev_features_t features)
109{
110	struct sk_buff *segs = ERR_PTR(-EINVAL);
111	struct ipv6hdr *ipv6h;
112	const struct net_offload *ops;
113	int proto, err;
114	struct frag_hdr *fptr;
115	unsigned int payload_len;
116	u8 *prevhdr;
117	int offset = 0;
118	bool encap, udpfrag;
119	int nhoff;
120	bool gso_partial;
121
122	skb_reset_network_header(skb);
123	err = ipv6_hopopt_jumbo_remove(skb);
124	if (err)
125		return ERR_PTR(err);
126	nhoff = skb_network_header(skb) - skb_mac_header(skb);
127	if (unlikely(!pskb_may_pull(skb, sizeof(*ipv6h))))
128		goto out;
129
130	encap = SKB_GSO_CB(skb)->encap_level > 0;
131	if (encap)
132		features &= skb->dev->hw_enc_features;
133	SKB_GSO_CB(skb)->encap_level += sizeof(*ipv6h);
134
135	ipv6h = ipv6_hdr(skb);
136	__skb_pull(skb, sizeof(*ipv6h));
137	segs = ERR_PTR(-EPROTONOSUPPORT);
138
139	proto = ipv6_gso_pull_exthdrs(skb, ipv6h->nexthdr);
140
141	if (skb->encapsulation &&
142	    skb_shinfo(skb)->gso_type & (SKB_GSO_IPXIP4 | SKB_GSO_IPXIP6))
143		udpfrag = proto == IPPROTO_UDP && encap &&
144			  (skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
145	else
146		udpfrag = proto == IPPROTO_UDP && !skb->encapsulation &&
147			  (skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
148
149	ops = rcu_dereference(inet6_offloads[proto]);
150	if (likely(ops && ops->callbacks.gso_segment)) {
151		if (!skb_reset_transport_header_careful(skb))
152			goto out;
153
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
193out:
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 */
200static 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		*opps = rcu_dereference(inet6_offloads[proto]);
209		if (unlikely(!(*opps)))
210			break;
211		if (!((*opps)->flags & INET6_PROTO_GSO_EXTHDR))
212			break;
213
214		opth = (void *)opth + optlen;
215		optlen = ipv6_optlen(opth);
216		len += optlen;
217		proto = opth->nexthdr;
218	}
219	return len;
220}
221
222INDIRECT_CALLABLE_SCOPE struct sk_buff *ipv6_gro_receive(struct list_head *head,
223							 struct sk_buff *skb)
224{
225	const struct net_offload *ops;
226	struct sk_buff *pp = NULL;
227	struct sk_buff *p;
228	struct ipv6hdr *iph;
229	unsigned int nlen;
230	unsigned int hlen;
231	unsigned int off;
232	u16 flush = 1;
233	int proto;
234
235	off = skb_gro_offset(skb);
236	hlen = off + sizeof(*iph);
237	iph = skb_gro_header(skb, hlen, off);
238	if (unlikely(!iph))
239		goto out;
240
241	NAPI_GRO_CB(skb)->network_offsets[NAPI_GRO_CB(skb)->encap_mark] = off;
242
243	flush += ntohs(iph->payload_len) != skb->len - hlen;
244
245	proto = iph->nexthdr;
246	ops = rcu_dereference(inet6_offloads[proto]);
247	if (!ops || !ops->callbacks.gro_receive) {
248		proto = ipv6_gro_pull_exthdrs(skb, hlen, proto);
249
250		ops = rcu_dereference(inet6_offloads[proto]);
251		if (!ops || !ops->callbacks.gro_receive)
252			goto out;
253
254		iph = skb_gro_network_header(skb);
255	} else {
256		skb_gro_pull(skb, sizeof(*iph));
257	}
258
259	skb_set_transport_header(skb, skb_gro_offset(skb));
260
261	NAPI_GRO_CB(skb)->proto = proto;
262
263	flush--;
264	nlen = skb_gro_offset(skb) - off;
265
266	list_for_each_entry(p, head, list) {
267		const struct ipv6hdr *iph2;
268		__be32 first_word; /* <Version:4><Traffic_Class:8><Flow_Label:20> */
269
270		if (!NAPI_GRO_CB(p)->same_flow)
271			continue;
272
273		iph2 = (struct ipv6hdr *)(p->data + off);
274		first_word = *(__be32 *)iph ^ *(__be32 *)iph2;
275
276		/* All fields must match except length and Traffic Class.
277		 * XXX skbs on the gro_list have all been parsed and pulled
278		 * already so we don't need to compare nlen
279		 * (nlen != (sizeof(*iph2) + ipv6_exthdrs_len(iph2, &ops)))
280		 * memcmp() alone below is sufficient, right?
281		 */
282		 if ((first_word & htonl(0xF00FFFFF)) ||
283		     !ipv6_addr_equal(&iph->saddr, &iph2->saddr) ||
284		     !ipv6_addr_equal(&iph->daddr, &iph2->daddr) ||
285		     iph->nexthdr != iph2->nexthdr) {
286not_same_flow:
287			NAPI_GRO_CB(p)->same_flow = 0;
288			continue;
289		}
290		if (unlikely(nlen > sizeof(struct ipv6hdr))) {
291			if (memcmp(iph + 1, iph2 + 1,
292				   nlen - sizeof(struct ipv6hdr)))
293				goto not_same_flow;
294		}
295	}
296
297	NAPI_GRO_CB(skb)->flush |= flush;
298
299	skb_gro_postpull_rcsum(skb, iph, nlen);
300
301	pp = indirect_call_gro_receive_l4(tcp6_gro_receive, udp6_gro_receive,
302					 ops->callbacks.gro_receive, head, skb);
303
304out:
305	skb_gro_flush_final(skb, pp, flush);
306
307	return pp;
308}
309
310static struct sk_buff *sit_ip6ip6_gro_receive(struct list_head *head,
311					      struct sk_buff *skb)
312{
313	/* Common GRO receive for SIT and IP6IP6 */
314
315	if (NAPI_GRO_CB(skb)->encap_mark) {
316		NAPI_GRO_CB(skb)->flush = 1;
317		return NULL;
318	}
319
320	NAPI_GRO_CB(skb)->encap_mark = 1;
321
322	return ipv6_gro_receive(head, skb);
323}
324
325static struct sk_buff *ip4ip6_gro_receive(struct list_head *head,
326					  struct sk_buff *skb)
327{
328	/* Common GRO receive for SIT and IP6IP6 */
329
330	if (NAPI_GRO_CB(skb)->encap_mark) {
331		NAPI_GRO_CB(skb)->flush = 1;
332		return NULL;
333	}
334
335	NAPI_GRO_CB(skb)->encap_mark = 1;
336
337	return inet_gro_receive(head, skb);
338}
339
340INDIRECT_CALLABLE_SCOPE int ipv6_gro_complete(struct sk_buff *skb, int nhoff)
341{
342	const struct net_offload *ops;
343	struct ipv6hdr *iph;
344	int err = -ENOSYS;
345	u32 payload_len;
346
347	if (skb->encapsulation) {
348		skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IPV6));
349		skb_set_inner_network_header(skb, nhoff);
350	}
351
352	payload_len = skb->len - nhoff - sizeof(*iph);
353	if (unlikely(payload_len > IPV6_MAXPLEN)) {
354		struct hop_jumbo_hdr *hop_jumbo;
355		int hoplen = sizeof(*hop_jumbo);
356
357		/* Move network header left */
358		memmove(skb_mac_header(skb) - hoplen, skb_mac_header(skb),
359			skb->transport_header - skb->mac_header);
360		skb->data -= hoplen;
361		skb->len += hoplen;
362		skb->mac_header -= hoplen;
363		skb->network_header -= hoplen;
364		iph = (struct ipv6hdr *)(skb->data + nhoff);
365		hop_jumbo = (struct hop_jumbo_hdr *)(iph + 1);
366
367		/* Build hop-by-hop options */
368		hop_jumbo->nexthdr = iph->nexthdr;
369		hop_jumbo->hdrlen = 0;
370		hop_jumbo->tlv_type = IPV6_TLV_JUMBO;
371		hop_jumbo->tlv_len = 4;
372		hop_jumbo->jumbo_payload_len = htonl(payload_len + hoplen);
373
374		iph->nexthdr = NEXTHDR_HOP;
375		iph->payload_len = 0;
376	} else {
377		iph = (struct ipv6hdr *)(skb->data + nhoff);
378		iph->payload_len = htons(payload_len);
379	}
380
381	nhoff += sizeof(*iph) + ipv6_exthdrs_len(iph, &ops);
382	if (WARN_ON(!ops || !ops->callbacks.gro_complete))
383		goto out;
384
385	err = INDIRECT_CALL_L4(ops->callbacks.gro_complete, tcp6_gro_complete,
386			       udp6_gro_complete, skb, nhoff);
387
388out:
389	return err;
390}
391
392static int sit_gro_complete(struct sk_buff *skb, int nhoff)
393{
394	skb->encapsulation = 1;
395	skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
396	return ipv6_gro_complete(skb, nhoff);
397}
398
399static int ip6ip6_gro_complete(struct sk_buff *skb, int nhoff)
400{
401	skb->encapsulation = 1;
402	skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP6;
403	return ipv6_gro_complete(skb, nhoff);
404}
405
406static int ip4ip6_gro_complete(struct sk_buff *skb, int nhoff)
407{
408	skb->encapsulation = 1;
409	skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP6;
410	return inet_gro_complete(skb, nhoff);
411}
412
413
414static struct sk_buff *sit_gso_segment(struct sk_buff *skb,
415				       netdev_features_t features)
416{
417	if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
418		return ERR_PTR(-EINVAL);
419
420	return ipv6_gso_segment(skb, features);
421}
422
423static struct sk_buff *ip4ip6_gso_segment(struct sk_buff *skb,
424					  netdev_features_t features)
425{
426	if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP6))
427		return ERR_PTR(-EINVAL);
428
429	return inet_gso_segment(skb, features);
430}
431
432static struct sk_buff *ip6ip6_gso_segment(struct sk_buff *skb,
433					  netdev_features_t features)
434{
435	if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP6))
436		return ERR_PTR(-EINVAL);
437
438	return ipv6_gso_segment(skb, features);
439}
440
441static const struct net_offload sit_offload = {
442	.callbacks = {
443		.gso_segment	= sit_gso_segment,
444		.gro_receive    = sit_ip6ip6_gro_receive,
445		.gro_complete   = sit_gro_complete,
446	},
447};
448
449static const struct net_offload ip4ip6_offload = {
450	.callbacks = {
451		.gso_segment	= ip4ip6_gso_segment,
452		.gro_receive    = ip4ip6_gro_receive,
453		.gro_complete   = ip4ip6_gro_complete,
454	},
455};
456
457static const struct net_offload ip6ip6_offload = {
458	.callbacks = {
459		.gso_segment	= ip6ip6_gso_segment,
460		.gro_receive    = sit_ip6ip6_gro_receive,
461		.gro_complete   = ip6ip6_gro_complete,
462	},
463};
464static int __init ipv6_offload_init(void)
465{
466
467	if (tcpv6_offload_init() < 0)
468		pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
469	if (ipv6_exthdrs_offload_init() < 0)
470		pr_crit("%s: Cannot add EXTHDRS protocol offload\n", __func__);
471
472	net_hotdata.ipv6_packet_offload = (struct packet_offload) {
473		.type = cpu_to_be16(ETH_P_IPV6),
474		.callbacks = {
475			.gso_segment = ipv6_gso_segment,
476			.gro_receive = ipv6_gro_receive,
477			.gro_complete = ipv6_gro_complete,
478		},
479	};
480	dev_add_offload(&net_hotdata.ipv6_packet_offload);
481
482	inet_add_offload(&sit_offload, IPPROTO_IPV6);
483	inet6_add_offload(&ip6ip6_offload, IPPROTO_IPV6);
484	inet6_add_offload(&ip4ip6_offload, IPPROTO_IPIP);
485
486	return 0;
487}
488
489fs_initcall(ipv6_offload_init);