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
19 #include "ip6_offload.h"
20
21 /* All GRO functions are always builtin, except UDP over ipv6, which lays in
22 * ipv6 module, as it depends on UDPv6 lookup function, so we need special care
23 * when ipv6 is built as a module
24 */
25 #if IS_BUILTIN(CONFIG_IPV6)
26 #define INDIRECT_CALL_L4(f, f2, f1, ...) INDIRECT_CALL_2(f, f2, f1, __VA_ARGS__)
27 #else
28 #define INDIRECT_CALL_L4(f, f2, f1, ...) INDIRECT_CALL_1(f, f2, __VA_ARGS__)
29 #endif
30
31 #define indirect_call_gro_receive_l4(f2, f1, cb, head, skb) \
32 ({ \
33 unlikely(gro_recursion_inc_test(skb)) ? \
34 NAPI_GRO_CB(skb)->flush |= 1, NULL : \
35 INDIRECT_CALL_L4(cb, f2, f1, head, skb); \
36 })
37
ipv6_gso_pull_exthdrs(struct sk_buff * skb,int proto)38 static int ipv6_gso_pull_exthdrs(struct sk_buff *skb, int proto)
39 {
40 const struct net_offload *ops = NULL;
41
42 for (;;) {
43 struct ipv6_opt_hdr *opth;
44 int len;
45
46 if (proto != NEXTHDR_HOP) {
47 ops = rcu_dereference(inet6_offloads[proto]);
48
49 if (unlikely(!ops))
50 break;
51
52 if (!(ops->flags & INET6_PROTO_GSO_EXTHDR))
53 break;
54 }
55
56 if (unlikely(!pskb_may_pull(skb, 8)))
57 break;
58
59 opth = (void *)skb->data;
60 len = ipv6_optlen(opth);
61
62 if (unlikely(!pskb_may_pull(skb, len)))
63 break;
64
65 opth = (void *)skb->data;
66 proto = opth->nexthdr;
67 __skb_pull(skb, len);
68 }
69
70 return proto;
71 }
72
ipv6_gso_segment(struct sk_buff * skb,netdev_features_t features)73 static struct sk_buff *ipv6_gso_segment(struct sk_buff *skb,
74 netdev_features_t features)
75 {
76 struct sk_buff *segs = ERR_PTR(-EINVAL);
77 struct ipv6hdr *ipv6h;
78 const struct net_offload *ops;
79 int proto;
80 struct frag_hdr *fptr;
81 unsigned int payload_len;
82 u8 *prevhdr;
83 int offset = 0;
84 bool encap, udpfrag;
85 int nhoff;
86 bool gso_partial;
87
88 skb_reset_network_header(skb);
89 nhoff = skb_network_header(skb) - skb_mac_header(skb);
90 if (unlikely(!pskb_may_pull(skb, sizeof(*ipv6h))))
91 goto out;
92
93 encap = SKB_GSO_CB(skb)->encap_level > 0;
94 if (encap)
95 features &= skb->dev->hw_enc_features;
96 SKB_GSO_CB(skb)->encap_level += sizeof(*ipv6h);
97
98 ipv6h = ipv6_hdr(skb);
99 __skb_pull(skb, sizeof(*ipv6h));
100 segs = ERR_PTR(-EPROTONOSUPPORT);
101
102 proto = ipv6_gso_pull_exthdrs(skb, ipv6h->nexthdr);
103
104 if (skb->encapsulation &&
105 skb_shinfo(skb)->gso_type & (SKB_GSO_IPXIP4 | SKB_GSO_IPXIP6))
106 udpfrag = proto == IPPROTO_UDP && encap &&
107 (skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
108 else
109 udpfrag = proto == IPPROTO_UDP && !skb->encapsulation &&
110 (skb_shinfo(skb)->gso_type & SKB_GSO_UDP);
111
112 ops = rcu_dereference(inet6_offloads[proto]);
113 if (likely(ops && ops->callbacks.gso_segment)) {
114 skb_reset_transport_header(skb);
115 segs = ops->callbacks.gso_segment(skb, features);
116 if (!segs)
117 skb->network_header = skb_mac_header(skb) + nhoff - skb->head;
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 */
ipv6_exthdrs_len(struct ipv6hdr * iph,const struct net_offload ** opps)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
ipv6_gro_receive(struct list_head * head,struct sk_buff * skb)184 INDIRECT_CALLABLE_SCOPE struct sk_buff *ipv6_gro_receive(struct list_head *head,
185 struct sk_buff *skb)
186 {
187 const struct net_offload *ops;
188 struct sk_buff *pp = NULL;
189 struct sk_buff *p;
190 struct ipv6hdr *iph;
191 unsigned int nlen;
192 unsigned int hlen;
193 unsigned int off;
194 u16 flush = 1;
195 int proto;
196
197 off = skb_gro_offset(skb);
198 hlen = off + sizeof(*iph);
199 iph = skb_gro_header_fast(skb, off);
200 if (skb_gro_header_hard(skb, hlen)) {
201 iph = skb_gro_header_slow(skb, hlen, off);
202 if (unlikely(!iph))
203 goto out;
204 }
205
206 skb_set_network_header(skb, off);
207 skb_gro_pull(skb, sizeof(*iph));
208 skb_set_transport_header(skb, skb_gro_offset(skb));
209
210 flush += ntohs(iph->payload_len) != skb_gro_len(skb);
211
212 rcu_read_lock();
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_unlock;
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 *(u16 *)&iph->nexthdr != *(u16 *)&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 NAPI_GRO_CB(p)->flush |= flush;
267
268 /* If the previous IP ID value was based on an atomic
269 * datagram we can overwrite the value and ignore it.
270 */
271 if (NAPI_GRO_CB(skb)->is_atomic)
272 NAPI_GRO_CB(p)->flush_id = 0;
273 }
274
275 NAPI_GRO_CB(skb)->is_atomic = true;
276 NAPI_GRO_CB(skb)->flush |= flush;
277
278 skb_gro_postpull_rcsum(skb, iph, nlen);
279
280 pp = indirect_call_gro_receive_l4(tcp6_gro_receive, udp6_gro_receive,
281 ops->callbacks.gro_receive, head, skb);
282
283 out_unlock:
284 rcu_read_unlock();
285
286 out:
287 skb_gro_flush_final(skb, pp, flush);
288
289 return pp;
290 }
291
sit_ip6ip6_gro_receive(struct list_head * head,struct sk_buff * skb)292 static struct sk_buff *sit_ip6ip6_gro_receive(struct list_head *head,
293 struct sk_buff *skb)
294 {
295 /* Common GRO receive for SIT and IP6IP6 */
296
297 if (NAPI_GRO_CB(skb)->encap_mark) {
298 NAPI_GRO_CB(skb)->flush = 1;
299 return NULL;
300 }
301
302 NAPI_GRO_CB(skb)->encap_mark = 1;
303
304 return ipv6_gro_receive(head, skb);
305 }
306
ip4ip6_gro_receive(struct list_head * head,struct sk_buff * skb)307 static struct sk_buff *ip4ip6_gro_receive(struct list_head *head,
308 struct sk_buff *skb)
309 {
310 /* Common GRO receive for SIT and IP6IP6 */
311
312 if (NAPI_GRO_CB(skb)->encap_mark) {
313 NAPI_GRO_CB(skb)->flush = 1;
314 return NULL;
315 }
316
317 NAPI_GRO_CB(skb)->encap_mark = 1;
318
319 return inet_gro_receive(head, skb);
320 }
321
ipv6_gro_complete(struct sk_buff * skb,int nhoff)322 INDIRECT_CALLABLE_SCOPE int ipv6_gro_complete(struct sk_buff *skb, int nhoff)
323 {
324 const struct net_offload *ops;
325 struct ipv6hdr *iph = (struct ipv6hdr *)(skb->data + nhoff);
326 int err = -ENOSYS;
327
328 if (skb->encapsulation) {
329 skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IPV6));
330 skb_set_inner_network_header(skb, nhoff);
331 }
332
333 iph->payload_len = htons(skb->len - nhoff - sizeof(*iph));
334
335 rcu_read_lock();
336
337 nhoff += sizeof(*iph) + ipv6_exthdrs_len(iph, &ops);
338 if (WARN_ON(!ops || !ops->callbacks.gro_complete))
339 goto out_unlock;
340
341 err = INDIRECT_CALL_L4(ops->callbacks.gro_complete, tcp6_gro_complete,
342 udp6_gro_complete, skb, nhoff);
343
344 out_unlock:
345 rcu_read_unlock();
346
347 return err;
348 }
349
sit_gro_complete(struct sk_buff * skb,int nhoff)350 static int sit_gro_complete(struct sk_buff *skb, int nhoff)
351 {
352 skb->encapsulation = 1;
353 skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4;
354 return ipv6_gro_complete(skb, nhoff);
355 }
356
ip6ip6_gro_complete(struct sk_buff * skb,int nhoff)357 static int ip6ip6_gro_complete(struct sk_buff *skb, int nhoff)
358 {
359 skb->encapsulation = 1;
360 skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP6;
361 return ipv6_gro_complete(skb, nhoff);
362 }
363
ip4ip6_gro_complete(struct sk_buff * skb,int nhoff)364 static int ip4ip6_gro_complete(struct sk_buff *skb, int nhoff)
365 {
366 skb->encapsulation = 1;
367 skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP6;
368 return inet_gro_complete(skb, nhoff);
369 }
370
371 static struct packet_offload ipv6_packet_offload __read_mostly = {
372 .type = cpu_to_be16(ETH_P_IPV6),
373 .callbacks = {
374 .gso_segment = ipv6_gso_segment,
375 .gro_receive = ipv6_gro_receive,
376 .gro_complete = ipv6_gro_complete,
377 },
378 };
379
sit_gso_segment(struct sk_buff * skb,netdev_features_t features)380 static struct sk_buff *sit_gso_segment(struct sk_buff *skb,
381 netdev_features_t features)
382 {
383 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4))
384 return ERR_PTR(-EINVAL);
385
386 return ipv6_gso_segment(skb, features);
387 }
388
ip4ip6_gso_segment(struct sk_buff * skb,netdev_features_t features)389 static struct sk_buff *ip4ip6_gso_segment(struct sk_buff *skb,
390 netdev_features_t features)
391 {
392 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP6))
393 return ERR_PTR(-EINVAL);
394
395 return inet_gso_segment(skb, features);
396 }
397
ip6ip6_gso_segment(struct sk_buff * skb,netdev_features_t features)398 static struct sk_buff *ip6ip6_gso_segment(struct sk_buff *skb,
399 netdev_features_t features)
400 {
401 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP6))
402 return ERR_PTR(-EINVAL);
403
404 return ipv6_gso_segment(skb, features);
405 }
406
407 static const struct net_offload sit_offload = {
408 .callbacks = {
409 .gso_segment = sit_gso_segment,
410 .gro_receive = sit_ip6ip6_gro_receive,
411 .gro_complete = sit_gro_complete,
412 },
413 };
414
415 static const struct net_offload ip4ip6_offload = {
416 .callbacks = {
417 .gso_segment = ip4ip6_gso_segment,
418 .gro_receive = ip4ip6_gro_receive,
419 .gro_complete = ip4ip6_gro_complete,
420 },
421 };
422
423 static const struct net_offload ip6ip6_offload = {
424 .callbacks = {
425 .gso_segment = ip6ip6_gso_segment,
426 .gro_receive = sit_ip6ip6_gro_receive,
427 .gro_complete = ip6ip6_gro_complete,
428 },
429 };
ipv6_offload_init(void)430 static int __init ipv6_offload_init(void)
431 {
432
433 if (tcpv6_offload_init() < 0)
434 pr_crit("%s: Cannot add TCP protocol offload\n", __func__);
435 if (ipv6_exthdrs_offload_init() < 0)
436 pr_crit("%s: Cannot add EXTHDRS protocol offload\n", __func__);
437
438 dev_add_offload(&ipv6_packet_offload);
439
440 inet_add_offload(&sit_offload, IPPROTO_IPV6);
441 inet6_add_offload(&ip6ip6_offload, IPPROTO_IPV6);
442 inet6_add_offload(&ip4ip6_offload, IPPROTO_IPIP);
443
444 return 0;
445 }
446
447 fs_initcall(ipv6_offload_init);
448