projects/linux·blob master · refs

466 lines · 11.0 KiB

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#include <net/ip_tunnels.h>
21
22#include "ip6_offload.h"
23#include "tcpv6_offload.c"
24
25static int ipv6_gro_pull_exthdrs(struct sk_buff *skb, int off, int proto)
26{
27 const struct net_offload *ops = NULL;
28 struct ipv6_opt_hdr *opth;
29
30 for (;;) {
31 int len;
32
33 ops = rcu_dereference(inet6_offloads[proto]);
34
35 if (unlikely(!ops))
36 break;
37
38 if (!(ops->flags & INET6_PROTO_GSO_EXTHDR))
39 break;
40
41 opth = skb_gro_header(skb, off + sizeof(*opth), off);
42 if (unlikely(!opth))
43 break;
44
45 len = ipv6_optlen(opth);
46
47 opth = skb_gro_header(skb, off + len, off);
48 if (unlikely(!opth))
49 break;
50 proto = opth->nexthdr;
51
52 off += len;
53 }
54
55 skb_gro_pull(skb, off - skb_gro_receive_network_offset(skb));
56 return proto;
57}
58
59static int ipv6_gso_pull_exthdrs(struct sk_buff *skb, int proto)
60{
61 const struct net_offload *ops = NULL;
62
63 for (;;) {
64 struct ipv6_opt_hdr *opth;
65 int len;
66
67 ops = rcu_dereference(inet6_offloads[proto]);
68
69 if (unlikely(!ops))
70 break;
71
72 if (!(ops->flags & INET6_PROTO_GSO_EXTHDR))
73 break;
74
75 if (unlikely(!pskb_may_pull(skb, 8)))
76 break;
77
78 opth = (void *)skb->data;
79 len = ipv6_optlen(opth);
80
81 if (unlikely(!pskb_may_pull(skb, len)))
82 break;
83
84 opth = (void *)skb->data;
85 proto = opth->nexthdr;
86 __skb_pull(skb, len);
87 }
88
89 return proto;
90}
91
92static struct sk_buff *ipv6_gso_segment(struct sk_buff *skb,
93 netdev_features_t features)
94{
95 struct sk_buff *segs = ERR_PTR(-EINVAL);
96 struct ipv6hdr *ipv6h;
97 const struct net_offload *ops;
98 int proto;
99 struct frag_hdr *fptr;
100 unsigned int payload_len;
101 u8 *prevhdr;
102 int offset = 0;
103 bool encap, udpfrag;
104 int nhoff;
105 bool gso_partial;
106
107 skb_reset_network_header(skb);
108 if (unlikely(gso_recursion_inc_test(skb,
109 IP_TUNNEL_RECURSION_LIMIT)))
110 goto out;
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;
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_IPXIP4 | SKB_GSO_IPXIP6))
128 udpfrag = proto == IPPROTO_UDP && encap &&
129 (skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
130 else
131 udpfrag = proto == IPPROTO_UDP && !skb->encapsulation &&
132 (skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
133
134 ops = rcu_dereference(inet6_offloads[proto]);
135 if (likely(ops && ops->callbacks.gso_segment)) {
136 if (!skb_reset_transport_header_careful(skb))
137 goto out;
138
139 segs = ops->callbacks.gso_segment(skb, features);
140 if (!segs)
141 skb->network_header = skb_mac_header(skb) + nhoff - skb->head;
142 }
143
144 if (IS_ERR_OR_NULL(segs))
145 goto out;
146
147 gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL);
148
149 for (skb = segs; skb; skb = skb->next) {
150 ipv6h = (struct ipv6hdr *)(skb_mac_header(skb) + nhoff);
151 if (gso_partial && skb_is_gso(skb))
152 payload_len = skb_shinfo(skb)->gso_size +
153 SKB_GSO_CB(skb)->data_offset +
154 skb->head - (unsigned char *)(ipv6h + 1);
155 else
156 payload_len = skb->len - nhoff - sizeof(*ipv6h);
157 ipv6h->payload_len = htons(payload_len);
158 skb->network_header = (u8 *)ipv6h - skb->head;
159 skb_reset_mac_len(skb);
160
161 if (udpfrag) {
162 int err = ip6_find_1stfragopt(skb, &prevhdr);
163 if (err < 0) {
164 kfree_skb_list(segs);
165 return ERR_PTR(err);
166 }
167 fptr = (struct frag_hdr *)((u8 *)ipv6h + err);
168 fptr->frag_off = htons(offset);
169 if (skb->next)
170 fptr->frag_off |= htons(IP6_MF);
171 offset += (ntohs(ipv6h->payload_len) -
172 sizeof(struct frag_hdr));
173 }
174 if (encap)
175 skb_reset_inner_headers(skb);
176 }
177
178out:
179 return segs;
180}
181
182/* Return the total length of all the extension hdrs, following the same
183 * logic in ipv6_gso_pull_exthdrs() when parsing ext-hdrs.
184 */
185static int ipv6_exthdrs_len(struct ipv6hdr *iph,
186 const struct net_offload **opps)
187{
188 struct ipv6_opt_hdr *opth = (void *)iph;
189 int len = 0, proto, optlen = sizeof(*iph);
190
191 proto = iph->nexthdr;
192 for (;;) {
193 *opps = rcu_dereference(inet6_offloads[proto]);
194 if (unlikely(!(*opps)))
195 break;
196 if (!((*opps)->flags & INET6_PROTO_GSO_EXTHDR))
197 break;
198
199 opth = (void *)opth + optlen;
200 optlen = ipv6_optlen(opth);
201 len += optlen;
202 proto = opth->nexthdr;
203 }
204 return len;
205}
206
207INDIRECT_CALLABLE_SCOPE struct sk_buff *ipv6_gro_receive(struct list_head *head,
208 struct sk_buff *skb)
209{
210 const struct net_offload *ops;
211 struct sk_buff *pp = NULL;
212 struct sk_buff *p;
213 struct ipv6hdr *iph;
214 unsigned int nlen;
215 unsigned int hlen;
216 unsigned int off;
217 u16 flush = 1;
218 int proto;
219
220 off = skb_gro_offset(skb);
221 hlen = off + sizeof(*iph);
222 iph = skb_gro_header(skb, hlen, off);
223 if (unlikely(!iph))
224 goto out;
225
226 NAPI_GRO_CB(skb)->network_offsets[NAPI_GRO_CB(skb)->encap_mark] = off;
227
228 flush += ntohs(iph->payload_len) != skb->len - hlen;
229
230 proto = iph->nexthdr;
231 ops = rcu_dereference(inet6_offloads[proto]);
232 if (!ops || !ops->callbacks.gro_receive) {
233 proto = ipv6_gro_pull_exthdrs(skb, hlen, proto);
234
235 ops = rcu_dereference(inet6_offloads[proto]);
236 if (!ops || !ops->callbacks.gro_receive)
237 goto out;
238
239 iph = skb_gro_network_header(skb);
240 } else {
241 skb_gro_pull(skb, sizeof(*iph));
242 }
243
244 skb_set_transport_header(skb, skb_gro_offset(skb));
245
246 NAPI_GRO_CB(skb)->proto = proto;
247
248 flush--;
249 nlen = skb_gro_offset(skb) - off;
250
251 list_for_each_entry(p, head, list) {
252 const struct ipv6hdr *iph2;
253 __be32 first_word; /* <Version:4><Traffic_Class:8><Flow_Label:20> */
254
255 if (!NAPI_GRO_CB(p)->same_flow)
256 continue;
257
258 iph2 = (struct ipv6hdr *)(p->data + off);
259 first_word = *(__be32 *)iph ^ *(__be32 *)iph2;
260
261 /* All fields must match except length and Traffic Class.
262 * XXX skbs on the gro_list have all been parsed and pulled
263 * already so we don't need to compare nlen
264 * (nlen != (sizeof(*iph2) + ipv6_exthdrs_len(iph2, &ops)))
265 * memcmp() alone below is sufficient, right?
266 */
267 if ((first_word & htonl(0xF00FFFFF)) ||
268 !ipv6_addr_equal(&iph->saddr, &iph2->saddr) ||
269 !ipv6_addr_equal(&iph->daddr, &iph2->daddr) ||
270 iph->nexthdr != iph2->nexthdr) {
271not_same_flow:
272 NAPI_GRO_CB(p)->same_flow = 0;
273 continue;
274 }
275 if (unlikely(nlen > sizeof(struct ipv6hdr))) {
276 if (memcmp(iph + 1, iph2 + 1,
277 nlen - sizeof(struct ipv6hdr)))
278 goto not_same_flow;
279 }
280 }
281
282 NAPI_GRO_CB(skb)->flush |= flush;
283
284 skb_gro_postpull_rcsum(skb, iph, nlen);
285
286 if (unlikely(gro_recursion_inc_test(skb))) {
287 flush = 1;
288 goto out;
289 }
290
291 if (likely(proto == IPPROTO_TCP))
292 pp = tcp6_gro_receive(head, skb);
293#if IS_ENABLED(CONFIG_IPV6)
294 else if (likely(proto == IPPROTO_UDP))
295 pp = udp6_gro_receive(head, skb);
296#endif
297 else
298 pp = ops->callbacks.gro_receive(head, skb);
299out:
300 skb_gro_flush_final(skb, pp, flush);
301
302 return pp;
303}
304EXPORT_INDIRECT_CALLABLE(ipv6_gro_receive);
305
306static struct sk_buff *sit_ip6ip6_gro_receive(struct list_head *head,
307 struct sk_buff *skb)
308{
309 /* Common GRO receive for SIT and IP6IP6 */
310
311 if (NAPI_GRO_CB(skb)->encap_mark) {
312 NAPI_GRO_CB(skb)->flush = 1;
313 return NULL;
314 }
315
316 NAPI_GRO_CB(skb)->encap_mark = 1;
317
318 return ipv6_gro_receive(head, skb);
319}
320
321static struct sk_buff *ip4ip6_gro_receive(struct list_head *head,
322 struct sk_buff *skb)
323{
324 /* Common GRO receive for SIT and IP6IP6 */
325
326 if (NAPI_GRO_CB(skb)->encap_mark) {
327 NAPI_GRO_CB(skb)->flush = 1;
328 return NULL;
329 }
330
331 NAPI_GRO_CB(skb)->encap_mark = 1;
332
333 return inet_gro_receive(head, skb);
334}
335
336INDIRECT_CALLABLE_SCOPE int ipv6_gro_complete(struct sk_buff *skb, int nhoff)
337{
338 const struct net_offload *ops;
339 struct ipv6hdr *iph;
340 int err = -ENOSYS;
341
342 if (skb->encapsulation) {
343 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IPV6));
344 skb_set_inner_network_header(skb, nhoff);
345 }
346
347 iph = (struct ipv6hdr *)(skb->data + nhoff);
348 ipv6_set_payload_len(iph, skb->len - nhoff - sizeof(*iph));
349
350 nhoff += sizeof(*iph) + ipv6_exthdrs_len(iph, &ops);
351
352 if (likely(ops == &net_hotdata.tcpv6_offload))
353 return tcp6_gro_complete(skb, nhoff);
354#if IS_ENABLED(CONFIG_IPV6)
355 if (ops == &net_hotdata.udpv6_offload)
356 return udp6_gro_complete(skb, nhoff);
357#endif
358
359 if (WARN_ON(!ops || !ops->callbacks.gro_complete))
360 goto out;
361
362 err = ops->callbacks.gro_complete(skb, nhoff);
363
364out:
365 return err;
366}
367EXPORT_INDIRECT_CALLABLE(ipv6_gro_complete);
368
369static int sit_gro_complete(struct sk_buff *skb, int nhoff)
370{
371 skb->encapsulation = 1;
372 skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
373 return ipv6_gro_complete(skb, nhoff);
374}
375
376static int ip6ip6_gro_complete(struct sk_buff *skb, int nhoff)
377{
378 skb->encapsulation = 1;
379 skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP6;
380 return ipv6_gro_complete(skb, nhoff);
381}
382
383static int ip4ip6_gro_complete(struct sk_buff *skb, int nhoff)
384{
385 skb->encapsulation = 1;
386 skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP6;
387 return inet_gro_complete(skb, nhoff);
388}
389
390
391static struct sk_buff *sit_gso_segment(struct sk_buff *skb,
392 netdev_features_t features)
393{
394 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
395 return ERR_PTR(-EINVAL);
396
397 return ipv6_gso_segment(skb, features);
398}
399
400static struct sk_buff *ip4ip6_gso_segment(struct sk_buff *skb,
401 netdev_features_t features)
402{
403 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP6))
404 return ERR_PTR(-EINVAL);
405
406 return inet_gso_segment(skb, features);
407}
408
409static struct sk_buff *ip6ip6_gso_segment(struct sk_buff *skb,
410 netdev_features_t features)
411{
412 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP6))
413 return ERR_PTR(-EINVAL);
414
415 return ipv6_gso_segment(skb, features);
416}
417
418static const struct net_offload sit_offload = {
419 .callbacks = {
420 .gso_segment = sit_gso_segment,
421 .gro_receive = sit_ip6ip6_gro_receive,
422 .gro_complete = sit_gro_complete,
423 },
424};
425
426static const struct net_offload ip4ip6_offload = {
427 .callbacks = {
428 .gso_segment = ip4ip6_gso_segment,
429 .gro_receive = ip4ip6_gro_receive,
430 .gro_complete = ip4ip6_gro_complete,
431 },
432};
433
434static const struct net_offload ip6ip6_offload = {
435 .callbacks = {
436 .gso_segment = ip6ip6_gso_segment,
437 .gro_receive = sit_ip6ip6_gro_receive,
438 .gro_complete = ip6ip6_gro_complete,
439 },
440};
441static int __init ipv6_offload_init(void)
442{
443
444 if (tcpv6_offload_init() < 0)
445 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
446 if (ipv6_exthdrs_offload_init() < 0)
447 pr_crit("%s: Cannot add EXTHDRS protocol offload\n", __func__);
448
449 net_hotdata.ipv6_packet_offload = (struct packet_offload) {
450 .type = cpu_to_be16(ETH_P_IPV6),
451 .callbacks = {
452 .gso_segment = ipv6_gso_segment,
453 .gro_receive = ipv6_gro_receive,
454 .gro_complete = ipv6_gro_complete,
455 },
456 };
457 dev_add_offload(&net_hotdata.ipv6_packet_offload);
458
459 inet_add_offload(&sit_offload, IPPROTO_IPV6);
460 inet6_add_offload(&ip6ip6_offload, IPPROTO_IPV6);
461 inet6_add_offload(&ip4ip6_offload, IPPROTO_IPIP);
462
463 return 0;
464}
465
466fs_initcall(ipv6_offload_init);