1// SPDX-License-Identifier: GPL-2.0-only
2#include <linux/kernel.h>
3#include <linux/init.h>
4#include <linux/module.h>
5#include <linux/netfilter.h>
6#include <linux/rhashtable.h>
7#include <linux/ip.h>
8#include <linux/ipv6.h>
9#include <linux/netdevice.h>
10#include <linux/if_ether.h>
11#include <net/gso.h>
12#include <net/ip.h>
13#include <net/ipv6.h>
14#include <net/ip6_route.h>
15#include <net/neighbour.h>
16#include <net/netfilter/nf_flow_table.h>
17#include <net/netfilter/nf_conntrack_acct.h>
18/* For layer 4 checksum field offset. */
19#include <linux/tcp.h>
20#include <linux/udp.h>
21
22static int nf_flow_state_check(struct flow_offload *flow, int proto,
23 struct sk_buff *skb, unsigned int thoff)
24{
25 struct tcphdr *tcph;
26
27 if (proto != IPPROTO_TCP)
28 return 0;
29
30 tcph = (void *)(skb_network_header(skb) + thoff);
31 if (unlikely(tcph->fin || tcph->rst)) {
32 flow_offload_teardown(flow);
33 return -1;
34 }
35
36 return 0;
37}
38
39static void nf_flow_nat_ip_tcp(struct sk_buff *skb, unsigned int thoff,
40 __be32 addr, __be32 new_addr)
41{
42 struct tcphdr *tcph;
43
44 tcph = (void *)(skb_network_header(skb) + thoff);
45 inet_proto_csum_replace4(sum: &tcph->check, skb, from: addr, to: new_addr, pseudohdr: true);
46}
47
48static void nf_flow_nat_ip_udp(struct sk_buff *skb, unsigned int thoff,
49 __be32 addr, __be32 new_addr)
50{
51 struct udphdr *udph;
52
53 udph = (void *)(skb_network_header(skb) + thoff);
54 if (udph->check || skb->ip_summed == CHECKSUM_PARTIAL) {
55 inet_proto_csum_replace4(sum: &udph->check, skb, from: addr,
56 to: new_addr, pseudohdr: true);
57 if (!udph->check)
58 udph->check = CSUM_MANGLED_0;
59 }
60}
61
62static void nf_flow_nat_ip_l4proto(struct sk_buff *skb, struct iphdr *iph,
63 unsigned int thoff, __be32 addr,
64 __be32 new_addr)
65{
66 switch (iph->protocol) {
67 case IPPROTO_TCP:
68 nf_flow_nat_ip_tcp(skb, thoff, addr, new_addr);
69 break;
70 case IPPROTO_UDP:
71 nf_flow_nat_ip_udp(skb, thoff, addr, new_addr);
72 break;
73 }
74}
75
76static void nf_flow_snat_ip(const struct flow_offload *flow,
77 struct sk_buff *skb, struct iphdr *iph,
78 unsigned int thoff, enum flow_offload_tuple_dir dir)
79{
80 __be32 addr, new_addr;
81
82 switch (dir) {
83 case FLOW_OFFLOAD_DIR_ORIGINAL:
84 addr = iph->saddr;
85 new_addr = flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.dst_v4.s_addr;
86 iph->saddr = new_addr;
87 break;
88 case FLOW_OFFLOAD_DIR_REPLY:
89 addr = iph->daddr;
90 new_addr = flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.src_v4.s_addr;
91 iph->daddr = new_addr;
92 break;
93 }
94 csum_replace4(sum: &iph->check, from: addr, to: new_addr);
95
96 nf_flow_nat_ip_l4proto(skb, iph, thoff, addr, new_addr);
97}
98
99static void nf_flow_dnat_ip(const struct flow_offload *flow,
100 struct sk_buff *skb, struct iphdr *iph,
101 unsigned int thoff, enum flow_offload_tuple_dir dir)
102{
103 __be32 addr, new_addr;
104
105 switch (dir) {
106 case FLOW_OFFLOAD_DIR_ORIGINAL:
107 addr = iph->daddr;
108 new_addr = flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.src_v4.s_addr;
109 iph->daddr = new_addr;
110 break;
111 case FLOW_OFFLOAD_DIR_REPLY:
112 addr = iph->saddr;
113 new_addr = flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.dst_v4.s_addr;
114 iph->saddr = new_addr;
115 break;
116 }
117 csum_replace4(sum: &iph->check, from: addr, to: new_addr);
118
119 nf_flow_nat_ip_l4proto(skb, iph, thoff, addr, new_addr);
120}
121
122static void nf_flow_nat_ip(const struct flow_offload *flow, struct sk_buff *skb,
123 unsigned int thoff, enum flow_offload_tuple_dir dir,
124 struct iphdr *iph)
125{
126 if (test_bit(NF_FLOW_SNAT, &flow->flags)) {
127 nf_flow_snat_port(flow, skb, thoff, protocol: iph->protocol, dir);
128 nf_flow_snat_ip(flow, skb, iph, thoff, dir);
129 }
130 if (test_bit(NF_FLOW_DNAT, &flow->flags)) {
131 nf_flow_dnat_port(flow, skb, thoff, protocol: iph->protocol, dir);
132 nf_flow_dnat_ip(flow, skb, iph, thoff, dir);
133 }
134}
135
136static bool ip_has_options(unsigned int thoff)
137{
138 return thoff != sizeof(struct iphdr);
139}
140
141static void nf_flow_tuple_encap(struct sk_buff *skb,
142 struct flow_offload_tuple *tuple)
143{
144 struct vlan_ethhdr *veth;
145 struct pppoe_hdr *phdr;
146 int i = 0;
147
148 if (skb_vlan_tag_present(skb)) {
149 tuple->encap[i].id = skb_vlan_tag_get(skb);
150 tuple->encap[i].proto = skb->vlan_proto;
151 i++;
152 }
153 switch (skb->protocol) {
154 case htons(ETH_P_8021Q):
155 veth = (struct vlan_ethhdr *)skb_mac_header(skb);
156 tuple->encap[i].id = ntohs(veth->h_vlan_TCI);
157 tuple->encap[i].proto = skb->protocol;
158 break;
159 case htons(ETH_P_PPP_SES):
160 phdr = (struct pppoe_hdr *)skb_network_header(skb);
161 tuple->encap[i].id = ntohs(phdr->sid);
162 tuple->encap[i].proto = skb->protocol;
163 break;
164 }
165}
166
167struct nf_flowtable_ctx {
168 const struct net_device *in;
169 u32 offset;
170 u32 hdrsize;
171};
172
173static int nf_flow_tuple_ip(struct nf_flowtable_ctx *ctx, struct sk_buff *skb,
174 struct flow_offload_tuple *tuple)
175{
176 struct flow_ports *ports;
177 unsigned int thoff;
178 struct iphdr *iph;
179 u8 ipproto;
180
181 if (!pskb_may_pull(skb, len: sizeof(*iph) + ctx->offset))
182 return -1;
183
184 iph = (struct iphdr *)(skb_network_header(skb) + ctx->offset);
185 thoff = (iph->ihl * 4);
186
187 if (ip_is_fragment(iph) ||
188 unlikely(ip_has_options(thoff)))
189 return -1;
190
191 thoff += ctx->offset;
192
193 ipproto = iph->protocol;
194 switch (ipproto) {
195 case IPPROTO_TCP:
196 ctx->hdrsize = sizeof(struct tcphdr);
197 break;
198 case IPPROTO_UDP:
199 ctx->hdrsize = sizeof(struct udphdr);
200 break;
201#ifdef CONFIG_NF_CT_PROTO_GRE
202 case IPPROTO_GRE:
203 ctx->hdrsize = sizeof(struct gre_base_hdr);
204 break;
205#endif
206 default:
207 return -1;
208 }
209
210 if (iph->ttl <= 1)
211 return -1;
212
213 if (!pskb_may_pull(skb, len: thoff + ctx->hdrsize))
214 return -1;
215
216 switch (ipproto) {
217 case IPPROTO_TCP:
218 case IPPROTO_UDP:
219 ports = (struct flow_ports *)(skb_network_header(skb) + thoff);
220 tuple->src_port = ports->source;
221 tuple->dst_port = ports->dest;
222 break;
223 case IPPROTO_GRE: {
224 struct gre_base_hdr *greh;
225
226 greh = (struct gre_base_hdr *)(skb_network_header(skb) + thoff);
227 if ((greh->flags & GRE_VERSION) != GRE_VERSION_0)
228 return -1;
229 break;
230 }
231 }
232
233 iph = (struct iphdr *)(skb_network_header(skb) + ctx->offset);
234
235 tuple->src_v4.s_addr = iph->saddr;
236 tuple->dst_v4.s_addr = iph->daddr;
237 tuple->l3proto = AF_INET;
238 tuple->l4proto = ipproto;
239 tuple->iifidx = ctx->in->ifindex;
240 nf_flow_tuple_encap(skb, tuple);
241
242 return 0;
243}
244
245/* Based on ip_exceeds_mtu(). */
246static bool nf_flow_exceeds_mtu(const struct sk_buff *skb, unsigned int mtu)
247{
248 if (skb->len <= mtu)
249 return false;
250
251 if (skb_is_gso(skb) && skb_gso_validate_network_len(skb, mtu))
252 return false;
253
254 return true;
255}
256
257static inline bool nf_flow_dst_check(struct flow_offload_tuple *tuple)
258{
259 if (tuple->xmit_type != FLOW_OFFLOAD_XMIT_NEIGH &&
260 tuple->xmit_type != FLOW_OFFLOAD_XMIT_XFRM)
261 return true;
262
263 return dst_check(dst: tuple->dst_cache, cookie: tuple->dst_cookie);
264}
265
266static unsigned int nf_flow_xmit_xfrm(struct sk_buff *skb,
267 const struct nf_hook_state *state,
268 struct dst_entry *dst)
269{
270 skb_orphan(skb);
271 skb_dst_set_noref(skb, dst);
272 dst_output(net: state->net, sk: state->sk, skb);
273 return NF_STOLEN;
274}
275
276static bool nf_flow_skb_encap_protocol(struct sk_buff *skb, __be16 proto,
277 u32 *offset)
278{
279 struct vlan_ethhdr *veth;
280 __be16 inner_proto;
281
282 switch (skb->protocol) {
283 case htons(ETH_P_8021Q):
284 veth = (struct vlan_ethhdr *)skb_mac_header(skb);
285 if (veth->h_vlan_encapsulated_proto == proto) {
286 *offset += VLAN_HLEN;
287 return true;
288 }
289 break;
290 case htons(ETH_P_PPP_SES):
291 if (nf_flow_pppoe_proto(skb, inner_proto: &inner_proto) &&
292 inner_proto == proto) {
293 *offset += PPPOE_SES_HLEN;
294 return true;
295 }
296 break;
297 }
298
299 return false;
300}
301
302static void nf_flow_encap_pop(struct sk_buff *skb,
303 struct flow_offload_tuple_rhash *tuplehash)
304{
305 struct vlan_hdr *vlan_hdr;
306 int i;
307
308 for (i = 0; i < tuplehash->tuple.encap_num; i++) {
309 if (skb_vlan_tag_present(skb)) {
310 __vlan_hwaccel_clear_tag(skb);
311 continue;
312 }
313 switch (skb->protocol) {
314 case htons(ETH_P_8021Q):
315 vlan_hdr = (struct vlan_hdr *)skb->data;
316 __skb_pull(skb, VLAN_HLEN);
317 vlan_set_encap_proto(skb, vhdr: vlan_hdr);
318 skb_reset_network_header(skb);
319 break;
320 case htons(ETH_P_PPP_SES):
321 skb->protocol = __nf_flow_pppoe_proto(skb);
322 skb_pull(skb, PPPOE_SES_HLEN);
323 skb_reset_network_header(skb);
324 break;
325 }
326 }
327}
328
329static unsigned int nf_flow_queue_xmit(struct net *net, struct sk_buff *skb,
330 const struct flow_offload_tuple_rhash *tuplehash,
331 unsigned short type)
332{
333 struct net_device *outdev;
334
335 outdev = dev_get_by_index_rcu(net, ifindex: tuplehash->tuple.out.ifidx);
336 if (!outdev)
337 return NF_DROP;
338
339 skb->dev = outdev;
340 dev_hard_header(skb, dev: skb->dev, type, daddr: tuplehash->tuple.out.h_dest,
341 saddr: tuplehash->tuple.out.h_source, len: skb->len);
342 dev_queue_xmit(skb);
343
344 return NF_STOLEN;
345}
346
347static struct flow_offload_tuple_rhash *
348nf_flow_offload_lookup(struct nf_flowtable_ctx *ctx,
349 struct nf_flowtable *flow_table, struct sk_buff *skb)
350{
351 struct flow_offload_tuple tuple = {};
352
353 if (skb->protocol != htons(ETH_P_IP) &&
354 !nf_flow_skb_encap_protocol(skb, htons(ETH_P_IP), offset: &ctx->offset))
355 return NULL;
356
357 if (nf_flow_tuple_ip(ctx, skb, tuple: &tuple) < 0)
358 return NULL;
359
360 return flow_offload_lookup(flow_table, tuple: &tuple);
361}
362
363static int nf_flow_offload_forward(struct nf_flowtable_ctx *ctx,
364 struct nf_flowtable *flow_table,
365 struct flow_offload_tuple_rhash *tuplehash,
366 struct sk_buff *skb)
367{
368 enum flow_offload_tuple_dir dir;
369 struct flow_offload *flow;
370 unsigned int thoff, mtu;
371 struct iphdr *iph;
372
373 dir = tuplehash->tuple.dir;
374 flow = container_of(tuplehash, struct flow_offload, tuplehash[dir]);
375
376 mtu = flow->tuplehash[dir].tuple.mtu + ctx->offset;
377 if (unlikely(nf_flow_exceeds_mtu(skb, mtu)))
378 return 0;
379
380 iph = (struct iphdr *)(skb_network_header(skb) + ctx->offset);
381 thoff = (iph->ihl * 4) + ctx->offset;
382 if (nf_flow_state_check(flow, proto: iph->protocol, skb, thoff))
383 return 0;
384
385 if (!nf_flow_dst_check(tuple: &tuplehash->tuple)) {
386 flow_offload_teardown(flow);
387 return 0;
388 }
389
390 if (skb_try_make_writable(skb, write_len: thoff + ctx->hdrsize))
391 return -1;
392
393 flow_offload_refresh(flow_table, flow, force: false);
394
395 nf_flow_encap_pop(skb, tuplehash);
396 thoff -= ctx->offset;
397
398 iph = ip_hdr(skb);
399 nf_flow_nat_ip(flow, skb, thoff, dir, iph);
400
401 ip_decrease_ttl(iph);
402 skb_clear_tstamp(skb);
403
404 if (flow_table->flags & NF_FLOWTABLE_COUNTER)
405 nf_ct_acct_update(ct: flow->ct, dir: tuplehash->tuple.dir, bytes: skb->len);
406
407 return 1;
408}
409
410unsigned int
411nf_flow_offload_ip_hook(void *priv, struct sk_buff *skb,
412 const struct nf_hook_state *state)
413{
414 struct flow_offload_tuple_rhash *tuplehash;
415 struct nf_flowtable *flow_table = priv;
416 enum flow_offload_tuple_dir dir;
417 struct nf_flowtable_ctx ctx = {
418 .in = state->in,
419 };
420 struct flow_offload *flow;
421 struct net_device *outdev;
422 struct rtable *rt;
423 __be32 nexthop;
424 int ret;
425
426 tuplehash = nf_flow_offload_lookup(ctx: &ctx, flow_table, skb);
427 if (!tuplehash)
428 return NF_ACCEPT;
429
430 ret = nf_flow_offload_forward(ctx: &ctx, flow_table, tuplehash, skb);
431 if (ret < 0)
432 return NF_DROP;
433 else if (ret == 0)
434 return NF_ACCEPT;
435
436 if (unlikely(tuplehash->tuple.xmit_type == FLOW_OFFLOAD_XMIT_XFRM)) {
437 rt = (struct rtable *)tuplehash->tuple.dst_cache;
438 memset(skb->cb, 0, sizeof(struct inet_skb_parm));
439 IPCB(skb)->iif = skb->dev->ifindex;
440 IPCB(skb)->flags = IPSKB_FORWARDED;
441 return nf_flow_xmit_xfrm(skb, state, dst: &rt->dst);
442 }
443
444 dir = tuplehash->tuple.dir;
445 flow = container_of(tuplehash, struct flow_offload, tuplehash[dir]);
446
447 switch (tuplehash->tuple.xmit_type) {
448 case FLOW_OFFLOAD_XMIT_NEIGH:
449 rt = (struct rtable *)tuplehash->tuple.dst_cache;
450 outdev = rt->dst.dev;
451 skb->dev = outdev;
452 nexthop = rt_nexthop(rt, daddr: flow->tuplehash[!dir].tuple.src_v4.s_addr);
453 skb_dst_set_noref(skb, dst: &rt->dst);
454 neigh_xmit(fam: NEIGH_ARP_TABLE, outdev, &nexthop, skb);
455 ret = NF_STOLEN;
456 break;
457 case FLOW_OFFLOAD_XMIT_DIRECT:
458 ret = nf_flow_queue_xmit(net: state->net, skb, tuplehash, ETH_P_IP);
459 if (ret == NF_DROP)
460 flow_offload_teardown(flow);
461 break;
462 default:
463 WARN_ON_ONCE(1);
464 ret = NF_DROP;
465 break;
466 }
467
468 return ret;
469}
470EXPORT_SYMBOL_GPL(nf_flow_offload_ip_hook);
471
472static void nf_flow_nat_ipv6_tcp(struct sk_buff *skb, unsigned int thoff,
473 struct in6_addr *addr,
474 struct in6_addr *new_addr,
475 struct ipv6hdr *ip6h)
476{
477 struct tcphdr *tcph;
478
479 tcph = (void *)(skb_network_header(skb) + thoff);
480 inet_proto_csum_replace16(sum: &tcph->check, skb, from: addr->s6_addr32,
481 to: new_addr->s6_addr32, pseudohdr: true);
482}
483
484static void nf_flow_nat_ipv6_udp(struct sk_buff *skb, unsigned int thoff,
485 struct in6_addr *addr,
486 struct in6_addr *new_addr)
487{
488 struct udphdr *udph;
489
490 udph = (void *)(skb_network_header(skb) + thoff);
491 if (udph->check || skb->ip_summed == CHECKSUM_PARTIAL) {
492 inet_proto_csum_replace16(sum: &udph->check, skb, from: addr->s6_addr32,
493 to: new_addr->s6_addr32, pseudohdr: true);
494 if (!udph->check)
495 udph->check = CSUM_MANGLED_0;
496 }
497}
498
499static void nf_flow_nat_ipv6_l4proto(struct sk_buff *skb, struct ipv6hdr *ip6h,
500 unsigned int thoff, struct in6_addr *addr,
501 struct in6_addr *new_addr)
502{
503 switch (ip6h->nexthdr) {
504 case IPPROTO_TCP:
505 nf_flow_nat_ipv6_tcp(skb, thoff, addr, new_addr, ip6h);
506 break;
507 case IPPROTO_UDP:
508 nf_flow_nat_ipv6_udp(skb, thoff, addr, new_addr);
509 break;
510 }
511}
512
513static void nf_flow_snat_ipv6(const struct flow_offload *flow,
514 struct sk_buff *skb, struct ipv6hdr *ip6h,
515 unsigned int thoff,
516 enum flow_offload_tuple_dir dir)
517{
518 struct in6_addr addr, new_addr;
519
520 switch (dir) {
521 case FLOW_OFFLOAD_DIR_ORIGINAL:
522 addr = ip6h->saddr;
523 new_addr = flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.dst_v6;
524 ip6h->saddr = new_addr;
525 break;
526 case FLOW_OFFLOAD_DIR_REPLY:
527 addr = ip6h->daddr;
528 new_addr = flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.src_v6;
529 ip6h->daddr = new_addr;
530 break;
531 }
532
533 nf_flow_nat_ipv6_l4proto(skb, ip6h, thoff, addr: &addr, new_addr: &new_addr);
534}
535
536static void nf_flow_dnat_ipv6(const struct flow_offload *flow,
537 struct sk_buff *skb, struct ipv6hdr *ip6h,
538 unsigned int thoff,
539 enum flow_offload_tuple_dir dir)
540{
541 struct in6_addr addr, new_addr;
542
543 switch (dir) {
544 case FLOW_OFFLOAD_DIR_ORIGINAL:
545 addr = ip6h->daddr;
546 new_addr = flow->tuplehash[FLOW_OFFLOAD_DIR_REPLY].tuple.src_v6;
547 ip6h->daddr = new_addr;
548 break;
549 case FLOW_OFFLOAD_DIR_REPLY:
550 addr = ip6h->saddr;
551 new_addr = flow->tuplehash[FLOW_OFFLOAD_DIR_ORIGINAL].tuple.dst_v6;
552 ip6h->saddr = new_addr;
553 break;
554 }
555
556 nf_flow_nat_ipv6_l4proto(skb, ip6h, thoff, addr: &addr, new_addr: &new_addr);
557}
558
559static void nf_flow_nat_ipv6(const struct flow_offload *flow,
560 struct sk_buff *skb,
561 enum flow_offload_tuple_dir dir,
562 struct ipv6hdr *ip6h)
563{
564 unsigned int thoff = sizeof(*ip6h);
565
566 if (test_bit(NF_FLOW_SNAT, &flow->flags)) {
567 nf_flow_snat_port(flow, skb, thoff, protocol: ip6h->nexthdr, dir);
568 nf_flow_snat_ipv6(flow, skb, ip6h, thoff, dir);
569 }
570 if (test_bit(NF_FLOW_DNAT, &flow->flags)) {
571 nf_flow_dnat_port(flow, skb, thoff, protocol: ip6h->nexthdr, dir);
572 nf_flow_dnat_ipv6(flow, skb, ip6h, thoff, dir);
573 }
574}
575
576static int nf_flow_tuple_ipv6(struct nf_flowtable_ctx *ctx, struct sk_buff *skb,
577 struct flow_offload_tuple *tuple)
578{
579 struct flow_ports *ports;
580 struct ipv6hdr *ip6h;
581 unsigned int thoff;
582 u8 nexthdr;
583
584 thoff = sizeof(*ip6h) + ctx->offset;
585 if (!pskb_may_pull(skb, len: thoff))
586 return -1;
587
588 ip6h = (struct ipv6hdr *)(skb_network_header(skb) + ctx->offset);
589
590 nexthdr = ip6h->nexthdr;
591 switch (nexthdr) {
592 case IPPROTO_TCP:
593 ctx->hdrsize = sizeof(struct tcphdr);
594 break;
595 case IPPROTO_UDP:
596 ctx->hdrsize = sizeof(struct udphdr);
597 break;
598#ifdef CONFIG_NF_CT_PROTO_GRE
599 case IPPROTO_GRE:
600 ctx->hdrsize = sizeof(struct gre_base_hdr);
601 break;
602#endif
603 default:
604 return -1;
605 }
606
607 if (ip6h->hop_limit <= 1)
608 return -1;
609
610 if (!pskb_may_pull(skb, len: thoff + ctx->hdrsize))
611 return -1;
612
613 switch (nexthdr) {
614 case IPPROTO_TCP:
615 case IPPROTO_UDP:
616 ports = (struct flow_ports *)(skb_network_header(skb) + thoff);
617 tuple->src_port = ports->source;
618 tuple->dst_port = ports->dest;
619 break;
620 case IPPROTO_GRE: {
621 struct gre_base_hdr *greh;
622
623 greh = (struct gre_base_hdr *)(skb_network_header(skb) + thoff);
624 if ((greh->flags & GRE_VERSION) != GRE_VERSION_0)
625 return -1;
626 break;
627 }
628 }
629
630 ip6h = (struct ipv6hdr *)(skb_network_header(skb) + ctx->offset);
631
632 tuple->src_v6 = ip6h->saddr;
633 tuple->dst_v6 = ip6h->daddr;
634 tuple->l3proto = AF_INET6;
635 tuple->l4proto = nexthdr;
636 tuple->iifidx = ctx->in->ifindex;
637 nf_flow_tuple_encap(skb, tuple);
638
639 return 0;
640}
641
642static int nf_flow_offload_ipv6_forward(struct nf_flowtable_ctx *ctx,
643 struct nf_flowtable *flow_table,
644 struct flow_offload_tuple_rhash *tuplehash,
645 struct sk_buff *skb)
646{
647 enum flow_offload_tuple_dir dir;
648 struct flow_offload *flow;
649 unsigned int thoff, mtu;
650 struct ipv6hdr *ip6h;
651
652 dir = tuplehash->tuple.dir;
653 flow = container_of(tuplehash, struct flow_offload, tuplehash[dir]);
654
655 mtu = flow->tuplehash[dir].tuple.mtu + ctx->offset;
656 if (unlikely(nf_flow_exceeds_mtu(skb, mtu)))
657 return 0;
658
659 ip6h = (struct ipv6hdr *)(skb_network_header(skb) + ctx->offset);
660 thoff = sizeof(*ip6h) + ctx->offset;
661 if (nf_flow_state_check(flow, proto: ip6h->nexthdr, skb, thoff))
662 return 0;
663
664 if (!nf_flow_dst_check(tuple: &tuplehash->tuple)) {
665 flow_offload_teardown(flow);
666 return 0;
667 }
668
669 if (skb_try_make_writable(skb, write_len: thoff + ctx->hdrsize))
670 return -1;
671
672 flow_offload_refresh(flow_table, flow, force: false);
673
674 nf_flow_encap_pop(skb, tuplehash);
675
676 ip6h = ipv6_hdr(skb);
677 nf_flow_nat_ipv6(flow, skb, dir, ip6h);
678
679 ip6h->hop_limit--;
680 skb_clear_tstamp(skb);
681
682 if (flow_table->flags & NF_FLOWTABLE_COUNTER)
683 nf_ct_acct_update(ct: flow->ct, dir: tuplehash->tuple.dir, bytes: skb->len);
684
685 return 1;
686}
687
688static struct flow_offload_tuple_rhash *
689nf_flow_offload_ipv6_lookup(struct nf_flowtable_ctx *ctx,
690 struct nf_flowtable *flow_table,
691 struct sk_buff *skb)
692{
693 struct flow_offload_tuple tuple = {};
694
695 if (skb->protocol != htons(ETH_P_IPV6) &&
696 !nf_flow_skb_encap_protocol(skb, htons(ETH_P_IPV6), offset: &ctx->offset))
697 return NULL;
698
699 if (nf_flow_tuple_ipv6(ctx, skb, tuple: &tuple) < 0)
700 return NULL;
701
702 return flow_offload_lookup(flow_table, tuple: &tuple);
703}
704
705unsigned int
706nf_flow_offload_ipv6_hook(void *priv, struct sk_buff *skb,
707 const struct nf_hook_state *state)
708{
709 struct flow_offload_tuple_rhash *tuplehash;
710 struct nf_flowtable *flow_table = priv;
711 enum flow_offload_tuple_dir dir;
712 struct nf_flowtable_ctx ctx = {
713 .in = state->in,
714 };
715 const struct in6_addr *nexthop;
716 struct flow_offload *flow;
717 struct net_device *outdev;
718 struct rt6_info *rt;
719 int ret;
720
721 tuplehash = nf_flow_offload_ipv6_lookup(ctx: &ctx, flow_table, skb);
722 if (tuplehash == NULL)
723 return NF_ACCEPT;
724
725 ret = nf_flow_offload_ipv6_forward(ctx: &ctx, flow_table, tuplehash, skb);
726 if (ret < 0)
727 return NF_DROP;
728 else if (ret == 0)
729 return NF_ACCEPT;
730
731 if (unlikely(tuplehash->tuple.xmit_type == FLOW_OFFLOAD_XMIT_XFRM)) {
732 rt = (struct rt6_info *)tuplehash->tuple.dst_cache;
733 memset(skb->cb, 0, sizeof(struct inet6_skb_parm));
734 IP6CB(skb)->iif = skb->dev->ifindex;
735 IP6CB(skb)->flags = IP6SKB_FORWARDED;
736 return nf_flow_xmit_xfrm(skb, state, dst: &rt->dst);
737 }
738
739 dir = tuplehash->tuple.dir;
740 flow = container_of(tuplehash, struct flow_offload, tuplehash[dir]);
741
742 switch (tuplehash->tuple.xmit_type) {
743 case FLOW_OFFLOAD_XMIT_NEIGH:
744 rt = (struct rt6_info *)tuplehash->tuple.dst_cache;
745 outdev = rt->dst.dev;
746 skb->dev = outdev;
747 nexthop = rt6_nexthop(rt, daddr: &flow->tuplehash[!dir].tuple.src_v6);
748 skb_dst_set_noref(skb, dst: &rt->dst);
749 neigh_xmit(fam: NEIGH_ND_TABLE, outdev, nexthop, skb);
750 ret = NF_STOLEN;
751 break;
752 case FLOW_OFFLOAD_XMIT_DIRECT:
753 ret = nf_flow_queue_xmit(net: state->net, skb, tuplehash, ETH_P_IPV6);
754 if (ret == NF_DROP)
755 flow_offload_teardown(flow);
756 break;
757 default:
758 WARN_ON_ONCE(1);
759 ret = NF_DROP;
760 break;
761 }
762
763 return ret;
764}
765EXPORT_SYMBOL_GPL(nf_flow_offload_ipv6_hook);
766

source code of linux/net/netfilter/nf_flow_table_ip.c