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