xref: /openbmc/linux/net/sched/act_ct.c (revision 6562c9ac)
1 // SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB
2 /* -
3  * net/sched/act_ct.c  Connection Tracking action
4  *
5  * Authors:   Paul Blakey <paulb@mellanox.com>
6  *            Yossi Kuperman <yossiku@mellanox.com>
7  *            Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
8  */
9 
10 #include <linux/module.h>
11 #include <linux/init.h>
12 #include <linux/kernel.h>
13 #include <linux/skbuff.h>
14 #include <linux/rtnetlink.h>
15 #include <linux/pkt_cls.h>
16 #include <linux/ip.h>
17 #include <linux/ipv6.h>
18 #include <linux/rhashtable.h>
19 #include <net/netlink.h>
20 #include <net/pkt_sched.h>
21 #include <net/pkt_cls.h>
22 #include <net/act_api.h>
23 #include <net/ip.h>
24 #include <net/ipv6_frag.h>
25 #include <uapi/linux/tc_act/tc_ct.h>
26 #include <net/tc_act/tc_ct.h>
27 
28 #include <net/netfilter/nf_flow_table.h>
29 #include <net/netfilter/nf_conntrack.h>
30 #include <net/netfilter/nf_conntrack_core.h>
31 #include <net/netfilter/nf_conntrack_zones.h>
32 #include <net/netfilter/nf_conntrack_helper.h>
33 #include <net/netfilter/nf_conntrack_acct.h>
34 #include <net/netfilter/ipv6/nf_defrag_ipv6.h>
35 #include <net/netfilter/nf_conntrack_act_ct.h>
36 #include <uapi/linux/netfilter/nf_nat.h>
37 
38 static struct workqueue_struct *act_ct_wq;
39 static struct rhashtable zones_ht;
40 static DEFINE_MUTEX(zones_mutex);
41 
42 struct tcf_ct_flow_table {
43 	struct rhash_head node; /* In zones tables */
44 
45 	struct rcu_work rwork;
46 	struct nf_flowtable nf_ft;
47 	refcount_t ref;
48 	u16 zone;
49 
50 	bool dying;
51 };
52 
53 static const struct rhashtable_params zones_params = {
54 	.head_offset = offsetof(struct tcf_ct_flow_table, node),
55 	.key_offset = offsetof(struct tcf_ct_flow_table, zone),
56 	.key_len = sizeof_field(struct tcf_ct_flow_table, zone),
57 	.automatic_shrinking = true,
58 };
59 
60 static struct flow_action_entry *
61 tcf_ct_flow_table_flow_action_get_next(struct flow_action *flow_action)
62 {
63 	int i = flow_action->num_entries++;
64 
65 	return &flow_action->entries[i];
66 }
67 
68 static void tcf_ct_add_mangle_action(struct flow_action *action,
69 				     enum flow_action_mangle_base htype,
70 				     u32 offset,
71 				     u32 mask,
72 				     u32 val)
73 {
74 	struct flow_action_entry *entry;
75 
76 	entry = tcf_ct_flow_table_flow_action_get_next(action);
77 	entry->id = FLOW_ACTION_MANGLE;
78 	entry->mangle.htype = htype;
79 	entry->mangle.mask = ~mask;
80 	entry->mangle.offset = offset;
81 	entry->mangle.val = val;
82 }
83 
84 /* The following nat helper functions check if the inverted reverse tuple
85  * (target) is different then the current dir tuple - meaning nat for ports
86  * and/or ip is needed, and add the relevant mangle actions.
87  */
88 static void
89 tcf_ct_flow_table_add_action_nat_ipv4(const struct nf_conntrack_tuple *tuple,
90 				      struct nf_conntrack_tuple target,
91 				      struct flow_action *action)
92 {
93 	if (memcmp(&target.src.u3, &tuple->src.u3, sizeof(target.src.u3)))
94 		tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_IP4,
95 					 offsetof(struct iphdr, saddr),
96 					 0xFFFFFFFF,
97 					 be32_to_cpu(target.src.u3.ip));
98 	if (memcmp(&target.dst.u3, &tuple->dst.u3, sizeof(target.dst.u3)))
99 		tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_IP4,
100 					 offsetof(struct iphdr, daddr),
101 					 0xFFFFFFFF,
102 					 be32_to_cpu(target.dst.u3.ip));
103 }
104 
105 static void
106 tcf_ct_add_ipv6_addr_mangle_action(struct flow_action *action,
107 				   union nf_inet_addr *addr,
108 				   u32 offset)
109 {
110 	int i;
111 
112 	for (i = 0; i < sizeof(struct in6_addr) / sizeof(u32); i++)
113 		tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_IP6,
114 					 i * sizeof(u32) + offset,
115 					 0xFFFFFFFF, be32_to_cpu(addr->ip6[i]));
116 }
117 
118 static void
119 tcf_ct_flow_table_add_action_nat_ipv6(const struct nf_conntrack_tuple *tuple,
120 				      struct nf_conntrack_tuple target,
121 				      struct flow_action *action)
122 {
123 	if (memcmp(&target.src.u3, &tuple->src.u3, sizeof(target.src.u3)))
124 		tcf_ct_add_ipv6_addr_mangle_action(action, &target.src.u3,
125 						   offsetof(struct ipv6hdr,
126 							    saddr));
127 	if (memcmp(&target.dst.u3, &tuple->dst.u3, sizeof(target.dst.u3)))
128 		tcf_ct_add_ipv6_addr_mangle_action(action, &target.dst.u3,
129 						   offsetof(struct ipv6hdr,
130 							    daddr));
131 }
132 
133 static void
134 tcf_ct_flow_table_add_action_nat_tcp(const struct nf_conntrack_tuple *tuple,
135 				     struct nf_conntrack_tuple target,
136 				     struct flow_action *action)
137 {
138 	__be16 target_src = target.src.u.tcp.port;
139 	__be16 target_dst = target.dst.u.tcp.port;
140 
141 	if (target_src != tuple->src.u.tcp.port)
142 		tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_TCP,
143 					 offsetof(struct tcphdr, source),
144 					 0xFFFF, be16_to_cpu(target_src));
145 	if (target_dst != tuple->dst.u.tcp.port)
146 		tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_TCP,
147 					 offsetof(struct tcphdr, dest),
148 					 0xFFFF, be16_to_cpu(target_dst));
149 }
150 
151 static void
152 tcf_ct_flow_table_add_action_nat_udp(const struct nf_conntrack_tuple *tuple,
153 				     struct nf_conntrack_tuple target,
154 				     struct flow_action *action)
155 {
156 	__be16 target_src = target.src.u.udp.port;
157 	__be16 target_dst = target.dst.u.udp.port;
158 
159 	if (target_src != tuple->src.u.udp.port)
160 		tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_UDP,
161 					 offsetof(struct udphdr, source),
162 					 0xFFFF, be16_to_cpu(target_src));
163 	if (target_dst != tuple->dst.u.udp.port)
164 		tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_UDP,
165 					 offsetof(struct udphdr, dest),
166 					 0xFFFF, be16_to_cpu(target_dst));
167 }
168 
169 static void tcf_ct_flow_table_add_action_meta(struct nf_conn *ct,
170 					      enum ip_conntrack_dir dir,
171 					      struct flow_action *action)
172 {
173 	struct nf_conn_labels *ct_labels;
174 	struct flow_action_entry *entry;
175 	enum ip_conntrack_info ctinfo;
176 	u32 *act_ct_labels;
177 
178 	entry = tcf_ct_flow_table_flow_action_get_next(action);
179 	entry->id = FLOW_ACTION_CT_METADATA;
180 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
181 	entry->ct_metadata.mark = ct->mark;
182 #endif
183 	ctinfo = dir == IP_CT_DIR_ORIGINAL ? IP_CT_ESTABLISHED :
184 					     IP_CT_ESTABLISHED_REPLY;
185 	/* aligns with the CT reference on the SKB nf_ct_set */
186 	entry->ct_metadata.cookie = (unsigned long)ct | ctinfo;
187 	entry->ct_metadata.orig_dir = dir == IP_CT_DIR_ORIGINAL;
188 
189 	act_ct_labels = entry->ct_metadata.labels;
190 	ct_labels = nf_ct_labels_find(ct);
191 	if (ct_labels)
192 		memcpy(act_ct_labels, ct_labels->bits, NF_CT_LABELS_MAX_SIZE);
193 	else
194 		memset(act_ct_labels, 0, NF_CT_LABELS_MAX_SIZE);
195 }
196 
197 static int tcf_ct_flow_table_add_action_nat(struct net *net,
198 					    struct nf_conn *ct,
199 					    enum ip_conntrack_dir dir,
200 					    struct flow_action *action)
201 {
202 	const struct nf_conntrack_tuple *tuple = &ct->tuplehash[dir].tuple;
203 	struct nf_conntrack_tuple target;
204 
205 	if (!(ct->status & IPS_NAT_MASK))
206 		return 0;
207 
208 	nf_ct_invert_tuple(&target, &ct->tuplehash[!dir].tuple);
209 
210 	switch (tuple->src.l3num) {
211 	case NFPROTO_IPV4:
212 		tcf_ct_flow_table_add_action_nat_ipv4(tuple, target,
213 						      action);
214 		break;
215 	case NFPROTO_IPV6:
216 		tcf_ct_flow_table_add_action_nat_ipv6(tuple, target,
217 						      action);
218 		break;
219 	default:
220 		return -EOPNOTSUPP;
221 	}
222 
223 	switch (nf_ct_protonum(ct)) {
224 	case IPPROTO_TCP:
225 		tcf_ct_flow_table_add_action_nat_tcp(tuple, target, action);
226 		break;
227 	case IPPROTO_UDP:
228 		tcf_ct_flow_table_add_action_nat_udp(tuple, target, action);
229 		break;
230 	default:
231 		return -EOPNOTSUPP;
232 	}
233 
234 	return 0;
235 }
236 
237 static int tcf_ct_flow_table_fill_actions(struct net *net,
238 					  const struct flow_offload *flow,
239 					  enum flow_offload_tuple_dir tdir,
240 					  struct nf_flow_rule *flow_rule)
241 {
242 	struct flow_action *action = &flow_rule->rule->action;
243 	int num_entries = action->num_entries;
244 	struct nf_conn *ct = flow->ct;
245 	enum ip_conntrack_dir dir;
246 	int i, err;
247 
248 	switch (tdir) {
249 	case FLOW_OFFLOAD_DIR_ORIGINAL:
250 		dir = IP_CT_DIR_ORIGINAL;
251 		break;
252 	case FLOW_OFFLOAD_DIR_REPLY:
253 		dir = IP_CT_DIR_REPLY;
254 		break;
255 	default:
256 		return -EOPNOTSUPP;
257 	}
258 
259 	err = tcf_ct_flow_table_add_action_nat(net, ct, dir, action);
260 	if (err)
261 		goto err_nat;
262 
263 	tcf_ct_flow_table_add_action_meta(ct, dir, action);
264 	return 0;
265 
266 err_nat:
267 	/* Clear filled actions */
268 	for (i = num_entries; i < action->num_entries; i++)
269 		memset(&action->entries[i], 0, sizeof(action->entries[i]));
270 	action->num_entries = num_entries;
271 
272 	return err;
273 }
274 
275 static struct nf_flowtable_type flowtable_ct = {
276 	.action		= tcf_ct_flow_table_fill_actions,
277 	.owner		= THIS_MODULE,
278 };
279 
280 static int tcf_ct_flow_table_get(struct net *net, struct tcf_ct_params *params)
281 {
282 	struct tcf_ct_flow_table *ct_ft;
283 	int err = -ENOMEM;
284 
285 	mutex_lock(&zones_mutex);
286 	ct_ft = rhashtable_lookup_fast(&zones_ht, &params->zone, zones_params);
287 	if (ct_ft && refcount_inc_not_zero(&ct_ft->ref))
288 		goto out_unlock;
289 
290 	ct_ft = kzalloc(sizeof(*ct_ft), GFP_KERNEL);
291 	if (!ct_ft)
292 		goto err_alloc;
293 	refcount_set(&ct_ft->ref, 1);
294 
295 	ct_ft->zone = params->zone;
296 	err = rhashtable_insert_fast(&zones_ht, &ct_ft->node, zones_params);
297 	if (err)
298 		goto err_insert;
299 
300 	ct_ft->nf_ft.type = &flowtable_ct;
301 	ct_ft->nf_ft.flags |= NF_FLOWTABLE_HW_OFFLOAD |
302 			      NF_FLOWTABLE_COUNTER;
303 	err = nf_flow_table_init(&ct_ft->nf_ft);
304 	if (err)
305 		goto err_init;
306 	write_pnet(&ct_ft->nf_ft.net, net);
307 
308 	__module_get(THIS_MODULE);
309 out_unlock:
310 	params->ct_ft = ct_ft;
311 	params->nf_ft = &ct_ft->nf_ft;
312 	mutex_unlock(&zones_mutex);
313 
314 	return 0;
315 
316 err_init:
317 	rhashtable_remove_fast(&zones_ht, &ct_ft->node, zones_params);
318 err_insert:
319 	kfree(ct_ft);
320 err_alloc:
321 	mutex_unlock(&zones_mutex);
322 	return err;
323 }
324 
325 static void tcf_ct_flow_table_cleanup_work(struct work_struct *work)
326 {
327 	struct flow_block_cb *block_cb, *tmp_cb;
328 	struct tcf_ct_flow_table *ct_ft;
329 	struct flow_block *block;
330 
331 	ct_ft = container_of(to_rcu_work(work), struct tcf_ct_flow_table,
332 			     rwork);
333 	nf_flow_table_free(&ct_ft->nf_ft);
334 
335 	/* Remove any remaining callbacks before cleanup */
336 	block = &ct_ft->nf_ft.flow_block;
337 	down_write(&ct_ft->nf_ft.flow_block_lock);
338 	list_for_each_entry_safe(block_cb, tmp_cb, &block->cb_list, list) {
339 		list_del(&block_cb->list);
340 		flow_block_cb_free(block_cb);
341 	}
342 	up_write(&ct_ft->nf_ft.flow_block_lock);
343 	kfree(ct_ft);
344 
345 	module_put(THIS_MODULE);
346 }
347 
348 static void tcf_ct_flow_table_put(struct tcf_ct_params *params)
349 {
350 	struct tcf_ct_flow_table *ct_ft = params->ct_ft;
351 
352 	if (refcount_dec_and_test(&params->ct_ft->ref)) {
353 		rhashtable_remove_fast(&zones_ht, &ct_ft->node, zones_params);
354 		INIT_RCU_WORK(&ct_ft->rwork, tcf_ct_flow_table_cleanup_work);
355 		queue_rcu_work(act_ct_wq, &ct_ft->rwork);
356 	}
357 }
358 
359 static void tcf_ct_flow_tc_ifidx(struct flow_offload *entry,
360 				 struct nf_conn_act_ct_ext *act_ct_ext, u8 dir)
361 {
362 	entry->tuplehash[dir].tuple.xmit_type = FLOW_OFFLOAD_XMIT_TC;
363 	entry->tuplehash[dir].tuple.tc.iifidx = act_ct_ext->ifindex[dir];
364 }
365 
366 static void tcf_ct_flow_table_add(struct tcf_ct_flow_table *ct_ft,
367 				  struct nf_conn *ct,
368 				  bool tcp)
369 {
370 	struct nf_conn_act_ct_ext *act_ct_ext;
371 	struct flow_offload *entry;
372 	int err;
373 
374 	if (test_and_set_bit(IPS_OFFLOAD_BIT, &ct->status))
375 		return;
376 
377 	entry = flow_offload_alloc(ct);
378 	if (!entry) {
379 		WARN_ON_ONCE(1);
380 		goto err_alloc;
381 	}
382 
383 	if (tcp) {
384 		ct->proto.tcp.seen[0].flags |= IP_CT_TCP_FLAG_BE_LIBERAL;
385 		ct->proto.tcp.seen[1].flags |= IP_CT_TCP_FLAG_BE_LIBERAL;
386 	}
387 
388 	act_ct_ext = nf_conn_act_ct_ext_find(ct);
389 	if (act_ct_ext) {
390 		tcf_ct_flow_tc_ifidx(entry, act_ct_ext, FLOW_OFFLOAD_DIR_ORIGINAL);
391 		tcf_ct_flow_tc_ifidx(entry, act_ct_ext, FLOW_OFFLOAD_DIR_REPLY);
392 	}
393 
394 	err = flow_offload_add(&ct_ft->nf_ft, entry);
395 	if (err)
396 		goto err_add;
397 
398 	return;
399 
400 err_add:
401 	flow_offload_free(entry);
402 err_alloc:
403 	clear_bit(IPS_OFFLOAD_BIT, &ct->status);
404 }
405 
406 static void tcf_ct_flow_table_process_conn(struct tcf_ct_flow_table *ct_ft,
407 					   struct nf_conn *ct,
408 					   enum ip_conntrack_info ctinfo)
409 {
410 	bool tcp = false;
411 
412 	if ((ctinfo != IP_CT_ESTABLISHED && ctinfo != IP_CT_ESTABLISHED_REPLY) ||
413 	    !test_bit(IPS_ASSURED_BIT, &ct->status))
414 		return;
415 
416 	switch (nf_ct_protonum(ct)) {
417 	case IPPROTO_TCP:
418 		tcp = true;
419 		if (ct->proto.tcp.state != TCP_CONNTRACK_ESTABLISHED)
420 			return;
421 		break;
422 	case IPPROTO_UDP:
423 		break;
424 #ifdef CONFIG_NF_CT_PROTO_GRE
425 	case IPPROTO_GRE: {
426 		struct nf_conntrack_tuple *tuple;
427 
428 		if (ct->status & IPS_NAT_MASK)
429 			return;
430 		tuple = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
431 		/* No support for GRE v1 */
432 		if (tuple->src.u.gre.key || tuple->dst.u.gre.key)
433 			return;
434 		break;
435 	}
436 #endif
437 	default:
438 		return;
439 	}
440 
441 	if (nf_ct_ext_exist(ct, NF_CT_EXT_HELPER) ||
442 	    ct->status & IPS_SEQ_ADJUST)
443 		return;
444 
445 	tcf_ct_flow_table_add(ct_ft, ct, tcp);
446 }
447 
448 static bool
449 tcf_ct_flow_table_fill_tuple_ipv4(struct sk_buff *skb,
450 				  struct flow_offload_tuple *tuple,
451 				  struct tcphdr **tcph)
452 {
453 	struct flow_ports *ports;
454 	unsigned int thoff;
455 	struct iphdr *iph;
456 	size_t hdrsize;
457 	u8 ipproto;
458 
459 	if (!pskb_network_may_pull(skb, sizeof(*iph)))
460 		return false;
461 
462 	iph = ip_hdr(skb);
463 	thoff = iph->ihl * 4;
464 
465 	if (ip_is_fragment(iph) ||
466 	    unlikely(thoff != sizeof(struct iphdr)))
467 		return false;
468 
469 	ipproto = iph->protocol;
470 	switch (ipproto) {
471 	case IPPROTO_TCP:
472 		hdrsize = sizeof(struct tcphdr);
473 		break;
474 	case IPPROTO_UDP:
475 		hdrsize = sizeof(*ports);
476 		break;
477 #ifdef CONFIG_NF_CT_PROTO_GRE
478 	case IPPROTO_GRE:
479 		hdrsize = sizeof(struct gre_base_hdr);
480 		break;
481 #endif
482 	default:
483 		return false;
484 	}
485 
486 	if (iph->ttl <= 1)
487 		return false;
488 
489 	if (!pskb_network_may_pull(skb, thoff + hdrsize))
490 		return false;
491 
492 	switch (ipproto) {
493 	case IPPROTO_TCP:
494 		*tcph = (void *)(skb_network_header(skb) + thoff);
495 		fallthrough;
496 	case IPPROTO_UDP:
497 		ports = (struct flow_ports *)(skb_network_header(skb) + thoff);
498 		tuple->src_port = ports->source;
499 		tuple->dst_port = ports->dest;
500 		break;
501 	case IPPROTO_GRE: {
502 		struct gre_base_hdr *greh;
503 
504 		greh = (struct gre_base_hdr *)(skb_network_header(skb) + thoff);
505 		if ((greh->flags & GRE_VERSION) != GRE_VERSION_0)
506 			return false;
507 		break;
508 	}
509 	}
510 
511 	iph = ip_hdr(skb);
512 
513 	tuple->src_v4.s_addr = iph->saddr;
514 	tuple->dst_v4.s_addr = iph->daddr;
515 	tuple->l3proto = AF_INET;
516 	tuple->l4proto = ipproto;
517 
518 	return true;
519 }
520 
521 static bool
522 tcf_ct_flow_table_fill_tuple_ipv6(struct sk_buff *skb,
523 				  struct flow_offload_tuple *tuple,
524 				  struct tcphdr **tcph)
525 {
526 	struct flow_ports *ports;
527 	struct ipv6hdr *ip6h;
528 	unsigned int thoff;
529 	size_t hdrsize;
530 	u8 nexthdr;
531 
532 	if (!pskb_network_may_pull(skb, sizeof(*ip6h)))
533 		return false;
534 
535 	ip6h = ipv6_hdr(skb);
536 	thoff = sizeof(*ip6h);
537 
538 	nexthdr = ip6h->nexthdr;
539 	switch (nexthdr) {
540 	case IPPROTO_TCP:
541 		hdrsize = sizeof(struct tcphdr);
542 		break;
543 	case IPPROTO_UDP:
544 		hdrsize = sizeof(*ports);
545 		break;
546 #ifdef CONFIG_NF_CT_PROTO_GRE
547 	case IPPROTO_GRE:
548 		hdrsize = sizeof(struct gre_base_hdr);
549 		break;
550 #endif
551 	default:
552 		return false;
553 	}
554 
555 	if (ip6h->hop_limit <= 1)
556 		return false;
557 
558 	if (!pskb_network_may_pull(skb, thoff + hdrsize))
559 		return false;
560 
561 	switch (nexthdr) {
562 	case IPPROTO_TCP:
563 		*tcph = (void *)(skb_network_header(skb) + thoff);
564 		fallthrough;
565 	case IPPROTO_UDP:
566 		ports = (struct flow_ports *)(skb_network_header(skb) + thoff);
567 		tuple->src_port = ports->source;
568 		tuple->dst_port = ports->dest;
569 		break;
570 	case IPPROTO_GRE: {
571 		struct gre_base_hdr *greh;
572 
573 		greh = (struct gre_base_hdr *)(skb_network_header(skb) + thoff);
574 		if ((greh->flags & GRE_VERSION) != GRE_VERSION_0)
575 			return false;
576 		break;
577 	}
578 	}
579 
580 	ip6h = ipv6_hdr(skb);
581 
582 	tuple->src_v6 = ip6h->saddr;
583 	tuple->dst_v6 = ip6h->daddr;
584 	tuple->l3proto = AF_INET6;
585 	tuple->l4proto = nexthdr;
586 
587 	return true;
588 }
589 
590 static bool tcf_ct_flow_table_lookup(struct tcf_ct_params *p,
591 				     struct sk_buff *skb,
592 				     u8 family)
593 {
594 	struct nf_flowtable *nf_ft = &p->ct_ft->nf_ft;
595 	struct flow_offload_tuple_rhash *tuplehash;
596 	struct flow_offload_tuple tuple = {};
597 	enum ip_conntrack_info ctinfo;
598 	struct tcphdr *tcph = NULL;
599 	struct flow_offload *flow;
600 	struct nf_conn *ct;
601 	u8 dir;
602 
603 	switch (family) {
604 	case NFPROTO_IPV4:
605 		if (!tcf_ct_flow_table_fill_tuple_ipv4(skb, &tuple, &tcph))
606 			return false;
607 		break;
608 	case NFPROTO_IPV6:
609 		if (!tcf_ct_flow_table_fill_tuple_ipv6(skb, &tuple, &tcph))
610 			return false;
611 		break;
612 	default:
613 		return false;
614 	}
615 
616 	tuplehash = flow_offload_lookup(nf_ft, &tuple);
617 	if (!tuplehash)
618 		return false;
619 
620 	dir = tuplehash->tuple.dir;
621 	flow = container_of(tuplehash, struct flow_offload, tuplehash[dir]);
622 	ct = flow->ct;
623 
624 	if (tcph && (unlikely(tcph->fin || tcph->rst))) {
625 		flow_offload_teardown(flow);
626 		return false;
627 	}
628 
629 	ctinfo = dir == FLOW_OFFLOAD_DIR_ORIGINAL ? IP_CT_ESTABLISHED :
630 						    IP_CT_ESTABLISHED_REPLY;
631 
632 	flow_offload_refresh(nf_ft, flow);
633 	nf_conntrack_get(&ct->ct_general);
634 	nf_ct_set(skb, ct, ctinfo);
635 	if (nf_ft->flags & NF_FLOWTABLE_COUNTER)
636 		nf_ct_acct_update(ct, dir, skb->len);
637 
638 	return true;
639 }
640 
641 static int tcf_ct_flow_tables_init(void)
642 {
643 	return rhashtable_init(&zones_ht, &zones_params);
644 }
645 
646 static void tcf_ct_flow_tables_uninit(void)
647 {
648 	rhashtable_destroy(&zones_ht);
649 }
650 
651 static struct tc_action_ops act_ct_ops;
652 static unsigned int ct_net_id;
653 
654 struct tc_ct_action_net {
655 	struct tc_action_net tn; /* Must be first */
656 	bool labels;
657 };
658 
659 /* Determine whether skb->_nfct is equal to the result of conntrack lookup. */
660 static bool tcf_ct_skb_nfct_cached(struct net *net, struct sk_buff *skb,
661 				   u16 zone_id, bool force)
662 {
663 	enum ip_conntrack_info ctinfo;
664 	struct nf_conn *ct;
665 
666 	ct = nf_ct_get(skb, &ctinfo);
667 	if (!ct)
668 		return false;
669 	if (!net_eq(net, read_pnet(&ct->ct_net)))
670 		goto drop_ct;
671 	if (nf_ct_zone(ct)->id != zone_id)
672 		goto drop_ct;
673 
674 	/* Force conntrack entry direction. */
675 	if (force && CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) {
676 		if (nf_ct_is_confirmed(ct))
677 			nf_ct_kill(ct);
678 
679 		goto drop_ct;
680 	}
681 
682 	return true;
683 
684 drop_ct:
685 	nf_ct_put(ct);
686 	nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
687 
688 	return false;
689 }
690 
691 /* Trim the skb to the length specified by the IP/IPv6 header,
692  * removing any trailing lower-layer padding. This prepares the skb
693  * for higher-layer processing that assumes skb->len excludes padding
694  * (such as nf_ip_checksum). The caller needs to pull the skb to the
695  * network header, and ensure ip_hdr/ipv6_hdr points to valid data.
696  */
697 static int tcf_ct_skb_network_trim(struct sk_buff *skb, int family)
698 {
699 	unsigned int len;
700 	int err;
701 
702 	switch (family) {
703 	case NFPROTO_IPV4:
704 		len = ntohs(ip_hdr(skb)->tot_len);
705 		break;
706 	case NFPROTO_IPV6:
707 		len = sizeof(struct ipv6hdr)
708 			+ ntohs(ipv6_hdr(skb)->payload_len);
709 		break;
710 	default:
711 		len = skb->len;
712 	}
713 
714 	err = pskb_trim_rcsum(skb, len);
715 
716 	return err;
717 }
718 
719 static u8 tcf_ct_skb_nf_family(struct sk_buff *skb)
720 {
721 	u8 family = NFPROTO_UNSPEC;
722 
723 	switch (skb_protocol(skb, true)) {
724 	case htons(ETH_P_IP):
725 		family = NFPROTO_IPV4;
726 		break;
727 	case htons(ETH_P_IPV6):
728 		family = NFPROTO_IPV6;
729 		break;
730 	default:
731 		break;
732 	}
733 
734 	return family;
735 }
736 
737 static int tcf_ct_ipv4_is_fragment(struct sk_buff *skb, bool *frag)
738 {
739 	unsigned int len;
740 
741 	len =  skb_network_offset(skb) + sizeof(struct iphdr);
742 	if (unlikely(skb->len < len))
743 		return -EINVAL;
744 	if (unlikely(!pskb_may_pull(skb, len)))
745 		return -ENOMEM;
746 
747 	*frag = ip_is_fragment(ip_hdr(skb));
748 	return 0;
749 }
750 
751 static int tcf_ct_ipv6_is_fragment(struct sk_buff *skb, bool *frag)
752 {
753 	unsigned int flags = 0, len, payload_ofs = 0;
754 	unsigned short frag_off;
755 	int nexthdr;
756 
757 	len =  skb_network_offset(skb) + sizeof(struct ipv6hdr);
758 	if (unlikely(skb->len < len))
759 		return -EINVAL;
760 	if (unlikely(!pskb_may_pull(skb, len)))
761 		return -ENOMEM;
762 
763 	nexthdr = ipv6_find_hdr(skb, &payload_ofs, -1, &frag_off, &flags);
764 	if (unlikely(nexthdr < 0))
765 		return -EPROTO;
766 
767 	*frag = flags & IP6_FH_F_FRAG;
768 	return 0;
769 }
770 
771 static int tcf_ct_handle_fragments(struct net *net, struct sk_buff *skb,
772 				   u8 family, u16 zone, bool *defrag)
773 {
774 	enum ip_conntrack_info ctinfo;
775 	struct nf_conn *ct;
776 	int err = 0;
777 	bool frag;
778 	u16 mru;
779 
780 	/* Previously seen (loopback)? Ignore. */
781 	ct = nf_ct_get(skb, &ctinfo);
782 	if ((ct && !nf_ct_is_template(ct)) || ctinfo == IP_CT_UNTRACKED)
783 		return 0;
784 
785 	if (family == NFPROTO_IPV4)
786 		err = tcf_ct_ipv4_is_fragment(skb, &frag);
787 	else
788 		err = tcf_ct_ipv6_is_fragment(skb, &frag);
789 	if (err || !frag)
790 		return err;
791 
792 	skb_get(skb);
793 	mru = tc_skb_cb(skb)->mru;
794 
795 	if (family == NFPROTO_IPV4) {
796 		enum ip_defrag_users user = IP_DEFRAG_CONNTRACK_IN + zone;
797 
798 		memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
799 		local_bh_disable();
800 		err = ip_defrag(net, skb, user);
801 		local_bh_enable();
802 		if (err && err != -EINPROGRESS)
803 			return err;
804 
805 		if (!err) {
806 			*defrag = true;
807 			mru = IPCB(skb)->frag_max_size;
808 		}
809 	} else { /* NFPROTO_IPV6 */
810 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
811 		enum ip6_defrag_users user = IP6_DEFRAG_CONNTRACK_IN + zone;
812 
813 		memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm));
814 		err = nf_ct_frag6_gather(net, skb, user);
815 		if (err && err != -EINPROGRESS)
816 			goto out_free;
817 
818 		if (!err) {
819 			*defrag = true;
820 			mru = IP6CB(skb)->frag_max_size;
821 		}
822 #else
823 		err = -EOPNOTSUPP;
824 		goto out_free;
825 #endif
826 	}
827 
828 	if (err != -EINPROGRESS)
829 		tc_skb_cb(skb)->mru = mru;
830 	skb_clear_hash(skb);
831 	skb->ignore_df = 1;
832 	return err;
833 
834 out_free:
835 	kfree_skb(skb);
836 	return err;
837 }
838 
839 static void tcf_ct_params_free(struct rcu_head *head)
840 {
841 	struct tcf_ct_params *params = container_of(head,
842 						    struct tcf_ct_params, rcu);
843 
844 	tcf_ct_flow_table_put(params);
845 
846 	if (params->tmpl)
847 		nf_ct_put(params->tmpl);
848 	kfree(params);
849 }
850 
851 #if IS_ENABLED(CONFIG_NF_NAT)
852 /* Modelled after nf_nat_ipv[46]_fn().
853  * range is only used for new, uninitialized NAT state.
854  * Returns either NF_ACCEPT or NF_DROP.
855  */
856 static int ct_nat_execute(struct sk_buff *skb, struct nf_conn *ct,
857 			  enum ip_conntrack_info ctinfo,
858 			  const struct nf_nat_range2 *range,
859 			  enum nf_nat_manip_type maniptype)
860 {
861 	__be16 proto = skb_protocol(skb, true);
862 	int hooknum, err = NF_ACCEPT;
863 
864 	/* See HOOK2MANIP(). */
865 	if (maniptype == NF_NAT_MANIP_SRC)
866 		hooknum = NF_INET_LOCAL_IN; /* Source NAT */
867 	else
868 		hooknum = NF_INET_LOCAL_OUT; /* Destination NAT */
869 
870 	switch (ctinfo) {
871 	case IP_CT_RELATED:
872 	case IP_CT_RELATED_REPLY:
873 		if (proto == htons(ETH_P_IP) &&
874 		    ip_hdr(skb)->protocol == IPPROTO_ICMP) {
875 			if (!nf_nat_icmp_reply_translation(skb, ct, ctinfo,
876 							   hooknum))
877 				err = NF_DROP;
878 			goto out;
879 		} else if (IS_ENABLED(CONFIG_IPV6) && proto == htons(ETH_P_IPV6)) {
880 			__be16 frag_off;
881 			u8 nexthdr = ipv6_hdr(skb)->nexthdr;
882 			int hdrlen = ipv6_skip_exthdr(skb,
883 						      sizeof(struct ipv6hdr),
884 						      &nexthdr, &frag_off);
885 
886 			if (hdrlen >= 0 && nexthdr == IPPROTO_ICMPV6) {
887 				if (!nf_nat_icmpv6_reply_translation(skb, ct,
888 								     ctinfo,
889 								     hooknum,
890 								     hdrlen))
891 					err = NF_DROP;
892 				goto out;
893 			}
894 		}
895 		/* Non-ICMP, fall thru to initialize if needed. */
896 		fallthrough;
897 	case IP_CT_NEW:
898 		/* Seen it before?  This can happen for loopback, retrans,
899 		 * or local packets.
900 		 */
901 		if (!nf_nat_initialized(ct, maniptype)) {
902 			/* Initialize according to the NAT action. */
903 			err = (range && range->flags & NF_NAT_RANGE_MAP_IPS)
904 				/* Action is set up to establish a new
905 				 * mapping.
906 				 */
907 				? nf_nat_setup_info(ct, range, maniptype)
908 				: nf_nat_alloc_null_binding(ct, hooknum);
909 			if (err != NF_ACCEPT)
910 				goto out;
911 		}
912 		break;
913 
914 	case IP_CT_ESTABLISHED:
915 	case IP_CT_ESTABLISHED_REPLY:
916 		break;
917 
918 	default:
919 		err = NF_DROP;
920 		goto out;
921 	}
922 
923 	err = nf_nat_packet(ct, ctinfo, hooknum, skb);
924 	if (err == NF_ACCEPT) {
925 		if (maniptype == NF_NAT_MANIP_SRC)
926 			tc_skb_cb(skb)->post_ct_snat = 1;
927 		if (maniptype == NF_NAT_MANIP_DST)
928 			tc_skb_cb(skb)->post_ct_dnat = 1;
929 	}
930 out:
931 	return err;
932 }
933 #endif /* CONFIG_NF_NAT */
934 
935 static void tcf_ct_act_set_mark(struct nf_conn *ct, u32 mark, u32 mask)
936 {
937 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
938 	u32 new_mark;
939 
940 	if (!mask)
941 		return;
942 
943 	new_mark = mark | (ct->mark & ~(mask));
944 	if (ct->mark != new_mark) {
945 		ct->mark = new_mark;
946 		if (nf_ct_is_confirmed(ct))
947 			nf_conntrack_event_cache(IPCT_MARK, ct);
948 	}
949 #endif
950 }
951 
952 static void tcf_ct_act_set_labels(struct nf_conn *ct,
953 				  u32 *labels,
954 				  u32 *labels_m)
955 {
956 #if IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS)
957 	size_t labels_sz = sizeof_field(struct tcf_ct_params, labels);
958 
959 	if (!memchr_inv(labels_m, 0, labels_sz))
960 		return;
961 
962 	nf_connlabels_replace(ct, labels, labels_m, 4);
963 #endif
964 }
965 
966 static int tcf_ct_act_nat(struct sk_buff *skb,
967 			  struct nf_conn *ct,
968 			  enum ip_conntrack_info ctinfo,
969 			  int ct_action,
970 			  struct nf_nat_range2 *range,
971 			  bool commit)
972 {
973 #if IS_ENABLED(CONFIG_NF_NAT)
974 	int err;
975 	enum nf_nat_manip_type maniptype;
976 
977 	if (!(ct_action & TCA_CT_ACT_NAT))
978 		return NF_ACCEPT;
979 
980 	/* Add NAT extension if not confirmed yet. */
981 	if (!nf_ct_is_confirmed(ct) && !nf_ct_nat_ext_add(ct))
982 		return NF_DROP;   /* Can't NAT. */
983 
984 	if (ctinfo != IP_CT_NEW && (ct->status & IPS_NAT_MASK) &&
985 	    (ctinfo != IP_CT_RELATED || commit)) {
986 		/* NAT an established or related connection like before. */
987 		if (CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY)
988 			/* This is the REPLY direction for a connection
989 			 * for which NAT was applied in the forward
990 			 * direction.  Do the reverse NAT.
991 			 */
992 			maniptype = ct->status & IPS_SRC_NAT
993 				? NF_NAT_MANIP_DST : NF_NAT_MANIP_SRC;
994 		else
995 			maniptype = ct->status & IPS_SRC_NAT
996 				? NF_NAT_MANIP_SRC : NF_NAT_MANIP_DST;
997 	} else if (ct_action & TCA_CT_ACT_NAT_SRC) {
998 		maniptype = NF_NAT_MANIP_SRC;
999 	} else if (ct_action & TCA_CT_ACT_NAT_DST) {
1000 		maniptype = NF_NAT_MANIP_DST;
1001 	} else {
1002 		return NF_ACCEPT;
1003 	}
1004 
1005 	err = ct_nat_execute(skb, ct, ctinfo, range, maniptype);
1006 	if (err == NF_ACCEPT && ct->status & IPS_DST_NAT) {
1007 		if (ct->status & IPS_SRC_NAT) {
1008 			if (maniptype == NF_NAT_MANIP_SRC)
1009 				maniptype = NF_NAT_MANIP_DST;
1010 			else
1011 				maniptype = NF_NAT_MANIP_SRC;
1012 
1013 			err = ct_nat_execute(skb, ct, ctinfo, range,
1014 					     maniptype);
1015 		} else if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) {
1016 			err = ct_nat_execute(skb, ct, ctinfo, NULL,
1017 					     NF_NAT_MANIP_SRC);
1018 		}
1019 	}
1020 	return err;
1021 #else
1022 	return NF_ACCEPT;
1023 #endif
1024 }
1025 
1026 static int tcf_ct_act(struct sk_buff *skb, const struct tc_action *a,
1027 		      struct tcf_result *res)
1028 {
1029 	struct net *net = dev_net(skb->dev);
1030 	bool cached, commit, clear, force;
1031 	enum ip_conntrack_info ctinfo;
1032 	struct tcf_ct *c = to_ct(a);
1033 	struct nf_conn *tmpl = NULL;
1034 	struct nf_hook_state state;
1035 	int nh_ofs, err, retval;
1036 	struct tcf_ct_params *p;
1037 	bool skip_add = false;
1038 	bool defrag = false;
1039 	struct nf_conn *ct;
1040 	u8 family;
1041 
1042 	p = rcu_dereference_bh(c->params);
1043 
1044 	retval = READ_ONCE(c->tcf_action);
1045 	commit = p->ct_action & TCA_CT_ACT_COMMIT;
1046 	clear = p->ct_action & TCA_CT_ACT_CLEAR;
1047 	force = p->ct_action & TCA_CT_ACT_FORCE;
1048 	tmpl = p->tmpl;
1049 
1050 	tcf_lastuse_update(&c->tcf_tm);
1051 	tcf_action_update_bstats(&c->common, skb);
1052 
1053 	if (clear) {
1054 		tc_skb_cb(skb)->post_ct = false;
1055 		ct = nf_ct_get(skb, &ctinfo);
1056 		if (ct) {
1057 			nf_ct_put(ct);
1058 			nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
1059 		}
1060 
1061 		goto out_clear;
1062 	}
1063 
1064 	family = tcf_ct_skb_nf_family(skb);
1065 	if (family == NFPROTO_UNSPEC)
1066 		goto drop;
1067 
1068 	/* The conntrack module expects to be working at L3.
1069 	 * We also try to pull the IPv4/6 header to linear area
1070 	 */
1071 	nh_ofs = skb_network_offset(skb);
1072 	skb_pull_rcsum(skb, nh_ofs);
1073 	err = tcf_ct_handle_fragments(net, skb, family, p->zone, &defrag);
1074 	if (err == -EINPROGRESS) {
1075 		retval = TC_ACT_STOLEN;
1076 		goto out_clear;
1077 	}
1078 	if (err)
1079 		goto drop;
1080 
1081 	err = tcf_ct_skb_network_trim(skb, family);
1082 	if (err)
1083 		goto drop;
1084 
1085 	/* If we are recirculating packets to match on ct fields and
1086 	 * committing with a separate ct action, then we don't need to
1087 	 * actually run the packet through conntrack twice unless it's for a
1088 	 * different zone.
1089 	 */
1090 	cached = tcf_ct_skb_nfct_cached(net, skb, p->zone, force);
1091 	if (!cached) {
1092 		if (tcf_ct_flow_table_lookup(p, skb, family)) {
1093 			skip_add = true;
1094 			goto do_nat;
1095 		}
1096 
1097 		/* Associate skb with specified zone. */
1098 		if (tmpl) {
1099 			nf_conntrack_put(skb_nfct(skb));
1100 			nf_conntrack_get(&tmpl->ct_general);
1101 			nf_ct_set(skb, tmpl, IP_CT_NEW);
1102 		}
1103 
1104 		state.hook = NF_INET_PRE_ROUTING;
1105 		state.net = net;
1106 		state.pf = family;
1107 		err = nf_conntrack_in(skb, &state);
1108 		if (err != NF_ACCEPT)
1109 			goto out_push;
1110 	}
1111 
1112 do_nat:
1113 	ct = nf_ct_get(skb, &ctinfo);
1114 	if (!ct)
1115 		goto out_push;
1116 	nf_ct_deliver_cached_events(ct);
1117 	nf_conn_act_ct_ext_fill(skb, ct, ctinfo);
1118 
1119 	err = tcf_ct_act_nat(skb, ct, ctinfo, p->ct_action, &p->range, commit);
1120 	if (err != NF_ACCEPT)
1121 		goto drop;
1122 
1123 	if (commit) {
1124 		tcf_ct_act_set_mark(ct, p->mark, p->mark_mask);
1125 		tcf_ct_act_set_labels(ct, p->labels, p->labels_mask);
1126 
1127 		if (!nf_ct_is_confirmed(ct))
1128 			nf_conn_act_ct_ext_add(ct);
1129 
1130 		/* This will take care of sending queued events
1131 		 * even if the connection is already confirmed.
1132 		 */
1133 		if (nf_conntrack_confirm(skb) != NF_ACCEPT)
1134 			goto drop;
1135 	}
1136 
1137 	if (!skip_add)
1138 		tcf_ct_flow_table_process_conn(p->ct_ft, ct, ctinfo);
1139 
1140 out_push:
1141 	skb_push_rcsum(skb, nh_ofs);
1142 
1143 	tc_skb_cb(skb)->post_ct = true;
1144 	tc_skb_cb(skb)->zone = p->zone;
1145 out_clear:
1146 	if (defrag)
1147 		qdisc_skb_cb(skb)->pkt_len = skb->len;
1148 	return retval;
1149 
1150 drop:
1151 	tcf_action_inc_drop_qstats(&c->common);
1152 	return TC_ACT_SHOT;
1153 }
1154 
1155 static const struct nla_policy ct_policy[TCA_CT_MAX + 1] = {
1156 	[TCA_CT_ACTION] = { .type = NLA_U16 },
1157 	[TCA_CT_PARMS] = NLA_POLICY_EXACT_LEN(sizeof(struct tc_ct)),
1158 	[TCA_CT_ZONE] = { .type = NLA_U16 },
1159 	[TCA_CT_MARK] = { .type = NLA_U32 },
1160 	[TCA_CT_MARK_MASK] = { .type = NLA_U32 },
1161 	[TCA_CT_LABELS] = { .type = NLA_BINARY,
1162 			    .len = 128 / BITS_PER_BYTE },
1163 	[TCA_CT_LABELS_MASK] = { .type = NLA_BINARY,
1164 				 .len = 128 / BITS_PER_BYTE },
1165 	[TCA_CT_NAT_IPV4_MIN] = { .type = NLA_U32 },
1166 	[TCA_CT_NAT_IPV4_MAX] = { .type = NLA_U32 },
1167 	[TCA_CT_NAT_IPV6_MIN] = NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)),
1168 	[TCA_CT_NAT_IPV6_MAX] = NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)),
1169 	[TCA_CT_NAT_PORT_MIN] = { .type = NLA_U16 },
1170 	[TCA_CT_NAT_PORT_MAX] = { .type = NLA_U16 },
1171 };
1172 
1173 static int tcf_ct_fill_params_nat(struct tcf_ct_params *p,
1174 				  struct tc_ct *parm,
1175 				  struct nlattr **tb,
1176 				  struct netlink_ext_ack *extack)
1177 {
1178 	struct nf_nat_range2 *range;
1179 
1180 	if (!(p->ct_action & TCA_CT_ACT_NAT))
1181 		return 0;
1182 
1183 	if (!IS_ENABLED(CONFIG_NF_NAT)) {
1184 		NL_SET_ERR_MSG_MOD(extack, "Netfilter nat isn't enabled in kernel");
1185 		return -EOPNOTSUPP;
1186 	}
1187 
1188 	if (!(p->ct_action & (TCA_CT_ACT_NAT_SRC | TCA_CT_ACT_NAT_DST)))
1189 		return 0;
1190 
1191 	if ((p->ct_action & TCA_CT_ACT_NAT_SRC) &&
1192 	    (p->ct_action & TCA_CT_ACT_NAT_DST)) {
1193 		NL_SET_ERR_MSG_MOD(extack, "dnat and snat can't be enabled at the same time");
1194 		return -EOPNOTSUPP;
1195 	}
1196 
1197 	range = &p->range;
1198 	if (tb[TCA_CT_NAT_IPV4_MIN]) {
1199 		struct nlattr *max_attr = tb[TCA_CT_NAT_IPV4_MAX];
1200 
1201 		p->ipv4_range = true;
1202 		range->flags |= NF_NAT_RANGE_MAP_IPS;
1203 		range->min_addr.ip =
1204 			nla_get_in_addr(tb[TCA_CT_NAT_IPV4_MIN]);
1205 
1206 		range->max_addr.ip = max_attr ?
1207 				     nla_get_in_addr(max_attr) :
1208 				     range->min_addr.ip;
1209 	} else if (tb[TCA_CT_NAT_IPV6_MIN]) {
1210 		struct nlattr *max_attr = tb[TCA_CT_NAT_IPV6_MAX];
1211 
1212 		p->ipv4_range = false;
1213 		range->flags |= NF_NAT_RANGE_MAP_IPS;
1214 		range->min_addr.in6 =
1215 			nla_get_in6_addr(tb[TCA_CT_NAT_IPV6_MIN]);
1216 
1217 		range->max_addr.in6 = max_attr ?
1218 				      nla_get_in6_addr(max_attr) :
1219 				      range->min_addr.in6;
1220 	}
1221 
1222 	if (tb[TCA_CT_NAT_PORT_MIN]) {
1223 		range->flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1224 		range->min_proto.all = nla_get_be16(tb[TCA_CT_NAT_PORT_MIN]);
1225 
1226 		range->max_proto.all = tb[TCA_CT_NAT_PORT_MAX] ?
1227 				       nla_get_be16(tb[TCA_CT_NAT_PORT_MAX]) :
1228 				       range->min_proto.all;
1229 	}
1230 
1231 	return 0;
1232 }
1233 
1234 static void tcf_ct_set_key_val(struct nlattr **tb,
1235 			       void *val, int val_type,
1236 			       void *mask, int mask_type,
1237 			       int len)
1238 {
1239 	if (!tb[val_type])
1240 		return;
1241 	nla_memcpy(val, tb[val_type], len);
1242 
1243 	if (!mask)
1244 		return;
1245 
1246 	if (mask_type == TCA_CT_UNSPEC || !tb[mask_type])
1247 		memset(mask, 0xff, len);
1248 	else
1249 		nla_memcpy(mask, tb[mask_type], len);
1250 }
1251 
1252 static int tcf_ct_fill_params(struct net *net,
1253 			      struct tcf_ct_params *p,
1254 			      struct tc_ct *parm,
1255 			      struct nlattr **tb,
1256 			      struct netlink_ext_ack *extack)
1257 {
1258 	struct tc_ct_action_net *tn = net_generic(net, ct_net_id);
1259 	struct nf_conntrack_zone zone;
1260 	struct nf_conn *tmpl;
1261 	int err;
1262 
1263 	p->zone = NF_CT_DEFAULT_ZONE_ID;
1264 
1265 	tcf_ct_set_key_val(tb,
1266 			   &p->ct_action, TCA_CT_ACTION,
1267 			   NULL, TCA_CT_UNSPEC,
1268 			   sizeof(p->ct_action));
1269 
1270 	if (p->ct_action & TCA_CT_ACT_CLEAR)
1271 		return 0;
1272 
1273 	err = tcf_ct_fill_params_nat(p, parm, tb, extack);
1274 	if (err)
1275 		return err;
1276 
1277 	if (tb[TCA_CT_MARK]) {
1278 		if (!IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)) {
1279 			NL_SET_ERR_MSG_MOD(extack, "Conntrack mark isn't enabled.");
1280 			return -EOPNOTSUPP;
1281 		}
1282 		tcf_ct_set_key_val(tb,
1283 				   &p->mark, TCA_CT_MARK,
1284 				   &p->mark_mask, TCA_CT_MARK_MASK,
1285 				   sizeof(p->mark));
1286 	}
1287 
1288 	if (tb[TCA_CT_LABELS]) {
1289 		if (!IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS)) {
1290 			NL_SET_ERR_MSG_MOD(extack, "Conntrack labels isn't enabled.");
1291 			return -EOPNOTSUPP;
1292 		}
1293 
1294 		if (!tn->labels) {
1295 			NL_SET_ERR_MSG_MOD(extack, "Failed to set connlabel length");
1296 			return -EOPNOTSUPP;
1297 		}
1298 		tcf_ct_set_key_val(tb,
1299 				   p->labels, TCA_CT_LABELS,
1300 				   p->labels_mask, TCA_CT_LABELS_MASK,
1301 				   sizeof(p->labels));
1302 	}
1303 
1304 	if (tb[TCA_CT_ZONE]) {
1305 		if (!IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES)) {
1306 			NL_SET_ERR_MSG_MOD(extack, "Conntrack zones isn't enabled.");
1307 			return -EOPNOTSUPP;
1308 		}
1309 
1310 		tcf_ct_set_key_val(tb,
1311 				   &p->zone, TCA_CT_ZONE,
1312 				   NULL, TCA_CT_UNSPEC,
1313 				   sizeof(p->zone));
1314 	}
1315 
1316 	nf_ct_zone_init(&zone, p->zone, NF_CT_DEFAULT_ZONE_DIR, 0);
1317 	tmpl = nf_ct_tmpl_alloc(net, &zone, GFP_KERNEL);
1318 	if (!tmpl) {
1319 		NL_SET_ERR_MSG_MOD(extack, "Failed to allocate conntrack template");
1320 		return -ENOMEM;
1321 	}
1322 	__set_bit(IPS_CONFIRMED_BIT, &tmpl->status);
1323 	p->tmpl = tmpl;
1324 
1325 	return 0;
1326 }
1327 
1328 static int tcf_ct_init(struct net *net, struct nlattr *nla,
1329 		       struct nlattr *est, struct tc_action **a,
1330 		       struct tcf_proto *tp, u32 flags,
1331 		       struct netlink_ext_ack *extack)
1332 {
1333 	struct tc_action_net *tn = net_generic(net, ct_net_id);
1334 	bool bind = flags & TCA_ACT_FLAGS_BIND;
1335 	struct tcf_ct_params *params = NULL;
1336 	struct nlattr *tb[TCA_CT_MAX + 1];
1337 	struct tcf_chain *goto_ch = NULL;
1338 	struct tc_ct *parm;
1339 	struct tcf_ct *c;
1340 	int err, res = 0;
1341 	u32 index;
1342 
1343 	if (!nla) {
1344 		NL_SET_ERR_MSG_MOD(extack, "Ct requires attributes to be passed");
1345 		return -EINVAL;
1346 	}
1347 
1348 	err = nla_parse_nested(tb, TCA_CT_MAX, nla, ct_policy, extack);
1349 	if (err < 0)
1350 		return err;
1351 
1352 	if (!tb[TCA_CT_PARMS]) {
1353 		NL_SET_ERR_MSG_MOD(extack, "Missing required ct parameters");
1354 		return -EINVAL;
1355 	}
1356 	parm = nla_data(tb[TCA_CT_PARMS]);
1357 	index = parm->index;
1358 	err = tcf_idr_check_alloc(tn, &index, a, bind);
1359 	if (err < 0)
1360 		return err;
1361 
1362 	if (!err) {
1363 		err = tcf_idr_create_from_flags(tn, index, est, a,
1364 						&act_ct_ops, bind, flags);
1365 		if (err) {
1366 			tcf_idr_cleanup(tn, index);
1367 			return err;
1368 		}
1369 		res = ACT_P_CREATED;
1370 	} else {
1371 		if (bind)
1372 			return 0;
1373 
1374 		if (!(flags & TCA_ACT_FLAGS_REPLACE)) {
1375 			tcf_idr_release(*a, bind);
1376 			return -EEXIST;
1377 		}
1378 	}
1379 	err = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack);
1380 	if (err < 0)
1381 		goto cleanup;
1382 
1383 	c = to_ct(*a);
1384 
1385 	params = kzalloc(sizeof(*params), GFP_KERNEL);
1386 	if (unlikely(!params)) {
1387 		err = -ENOMEM;
1388 		goto cleanup;
1389 	}
1390 
1391 	err = tcf_ct_fill_params(net, params, parm, tb, extack);
1392 	if (err)
1393 		goto cleanup;
1394 
1395 	err = tcf_ct_flow_table_get(net, params);
1396 	if (err)
1397 		goto cleanup_params;
1398 
1399 	spin_lock_bh(&c->tcf_lock);
1400 	goto_ch = tcf_action_set_ctrlact(*a, parm->action, goto_ch);
1401 	params = rcu_replace_pointer(c->params, params,
1402 				     lockdep_is_held(&c->tcf_lock));
1403 	spin_unlock_bh(&c->tcf_lock);
1404 
1405 	if (goto_ch)
1406 		tcf_chain_put_by_act(goto_ch);
1407 	if (params)
1408 		call_rcu(&params->rcu, tcf_ct_params_free);
1409 
1410 	return res;
1411 
1412 cleanup_params:
1413 	if (params->tmpl)
1414 		nf_ct_put(params->tmpl);
1415 cleanup:
1416 	if (goto_ch)
1417 		tcf_chain_put_by_act(goto_ch);
1418 	kfree(params);
1419 	tcf_idr_release(*a, bind);
1420 	return err;
1421 }
1422 
1423 static void tcf_ct_cleanup(struct tc_action *a)
1424 {
1425 	struct tcf_ct_params *params;
1426 	struct tcf_ct *c = to_ct(a);
1427 
1428 	params = rcu_dereference_protected(c->params, 1);
1429 	if (params)
1430 		call_rcu(&params->rcu, tcf_ct_params_free);
1431 }
1432 
1433 static int tcf_ct_dump_key_val(struct sk_buff *skb,
1434 			       void *val, int val_type,
1435 			       void *mask, int mask_type,
1436 			       int len)
1437 {
1438 	int err;
1439 
1440 	if (mask && !memchr_inv(mask, 0, len))
1441 		return 0;
1442 
1443 	err = nla_put(skb, val_type, len, val);
1444 	if (err)
1445 		return err;
1446 
1447 	if (mask_type != TCA_CT_UNSPEC) {
1448 		err = nla_put(skb, mask_type, len, mask);
1449 		if (err)
1450 			return err;
1451 	}
1452 
1453 	return 0;
1454 }
1455 
1456 static int tcf_ct_dump_nat(struct sk_buff *skb, struct tcf_ct_params *p)
1457 {
1458 	struct nf_nat_range2 *range = &p->range;
1459 
1460 	if (!(p->ct_action & TCA_CT_ACT_NAT))
1461 		return 0;
1462 
1463 	if (!(p->ct_action & (TCA_CT_ACT_NAT_SRC | TCA_CT_ACT_NAT_DST)))
1464 		return 0;
1465 
1466 	if (range->flags & NF_NAT_RANGE_MAP_IPS) {
1467 		if (p->ipv4_range) {
1468 			if (nla_put_in_addr(skb, TCA_CT_NAT_IPV4_MIN,
1469 					    range->min_addr.ip))
1470 				return -1;
1471 			if (nla_put_in_addr(skb, TCA_CT_NAT_IPV4_MAX,
1472 					    range->max_addr.ip))
1473 				return -1;
1474 		} else {
1475 			if (nla_put_in6_addr(skb, TCA_CT_NAT_IPV6_MIN,
1476 					     &range->min_addr.in6))
1477 				return -1;
1478 			if (nla_put_in6_addr(skb, TCA_CT_NAT_IPV6_MAX,
1479 					     &range->max_addr.in6))
1480 				return -1;
1481 		}
1482 	}
1483 
1484 	if (range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) {
1485 		if (nla_put_be16(skb, TCA_CT_NAT_PORT_MIN,
1486 				 range->min_proto.all))
1487 			return -1;
1488 		if (nla_put_be16(skb, TCA_CT_NAT_PORT_MAX,
1489 				 range->max_proto.all))
1490 			return -1;
1491 	}
1492 
1493 	return 0;
1494 }
1495 
1496 static inline int tcf_ct_dump(struct sk_buff *skb, struct tc_action *a,
1497 			      int bind, int ref)
1498 {
1499 	unsigned char *b = skb_tail_pointer(skb);
1500 	struct tcf_ct *c = to_ct(a);
1501 	struct tcf_ct_params *p;
1502 
1503 	struct tc_ct opt = {
1504 		.index   = c->tcf_index,
1505 		.refcnt  = refcount_read(&c->tcf_refcnt) - ref,
1506 		.bindcnt = atomic_read(&c->tcf_bindcnt) - bind,
1507 	};
1508 	struct tcf_t t;
1509 
1510 	spin_lock_bh(&c->tcf_lock);
1511 	p = rcu_dereference_protected(c->params,
1512 				      lockdep_is_held(&c->tcf_lock));
1513 	opt.action = c->tcf_action;
1514 
1515 	if (tcf_ct_dump_key_val(skb,
1516 				&p->ct_action, TCA_CT_ACTION,
1517 				NULL, TCA_CT_UNSPEC,
1518 				sizeof(p->ct_action)))
1519 		goto nla_put_failure;
1520 
1521 	if (p->ct_action & TCA_CT_ACT_CLEAR)
1522 		goto skip_dump;
1523 
1524 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1525 	    tcf_ct_dump_key_val(skb,
1526 				&p->mark, TCA_CT_MARK,
1527 				&p->mark_mask, TCA_CT_MARK_MASK,
1528 				sizeof(p->mark)))
1529 		goto nla_put_failure;
1530 
1531 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1532 	    tcf_ct_dump_key_val(skb,
1533 				p->labels, TCA_CT_LABELS,
1534 				p->labels_mask, TCA_CT_LABELS_MASK,
1535 				sizeof(p->labels)))
1536 		goto nla_put_failure;
1537 
1538 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1539 	    tcf_ct_dump_key_val(skb,
1540 				&p->zone, TCA_CT_ZONE,
1541 				NULL, TCA_CT_UNSPEC,
1542 				sizeof(p->zone)))
1543 		goto nla_put_failure;
1544 
1545 	if (tcf_ct_dump_nat(skb, p))
1546 		goto nla_put_failure;
1547 
1548 skip_dump:
1549 	if (nla_put(skb, TCA_CT_PARMS, sizeof(opt), &opt))
1550 		goto nla_put_failure;
1551 
1552 	tcf_tm_dump(&t, &c->tcf_tm);
1553 	if (nla_put_64bit(skb, TCA_CT_TM, sizeof(t), &t, TCA_CT_PAD))
1554 		goto nla_put_failure;
1555 	spin_unlock_bh(&c->tcf_lock);
1556 
1557 	return skb->len;
1558 nla_put_failure:
1559 	spin_unlock_bh(&c->tcf_lock);
1560 	nlmsg_trim(skb, b);
1561 	return -1;
1562 }
1563 
1564 static int tcf_ct_walker(struct net *net, struct sk_buff *skb,
1565 			 struct netlink_callback *cb, int type,
1566 			 const struct tc_action_ops *ops,
1567 			 struct netlink_ext_ack *extack)
1568 {
1569 	struct tc_action_net *tn = net_generic(net, ct_net_id);
1570 
1571 	return tcf_generic_walker(tn, skb, cb, type, ops, extack);
1572 }
1573 
1574 static int tcf_ct_search(struct net *net, struct tc_action **a, u32 index)
1575 {
1576 	struct tc_action_net *tn = net_generic(net, ct_net_id);
1577 
1578 	return tcf_idr_search(tn, a, index);
1579 }
1580 
1581 static void tcf_stats_update(struct tc_action *a, u64 bytes, u64 packets,
1582 			     u64 drops, u64 lastuse, bool hw)
1583 {
1584 	struct tcf_ct *c = to_ct(a);
1585 
1586 	tcf_action_update_stats(a, bytes, packets, drops, hw);
1587 	c->tcf_tm.lastuse = max_t(u64, c->tcf_tm.lastuse, lastuse);
1588 }
1589 
1590 static int tcf_ct_offload_act_setup(struct tc_action *act, void *entry_data,
1591 				    u32 *index_inc, bool bind,
1592 				    struct netlink_ext_ack *extack)
1593 {
1594 	if (bind) {
1595 		struct flow_action_entry *entry = entry_data;
1596 
1597 		entry->id = FLOW_ACTION_CT;
1598 		entry->ct.action = tcf_ct_action(act);
1599 		entry->ct.zone = tcf_ct_zone(act);
1600 		entry->ct.flow_table = tcf_ct_ft(act);
1601 		*index_inc = 1;
1602 	} else {
1603 		struct flow_offload_action *fl_action = entry_data;
1604 
1605 		fl_action->id = FLOW_ACTION_CT;
1606 	}
1607 
1608 	return 0;
1609 }
1610 
1611 static struct tc_action_ops act_ct_ops = {
1612 	.kind		=	"ct",
1613 	.id		=	TCA_ID_CT,
1614 	.owner		=	THIS_MODULE,
1615 	.act		=	tcf_ct_act,
1616 	.dump		=	tcf_ct_dump,
1617 	.init		=	tcf_ct_init,
1618 	.cleanup	=	tcf_ct_cleanup,
1619 	.walk		=	tcf_ct_walker,
1620 	.lookup		=	tcf_ct_search,
1621 	.stats_update	=	tcf_stats_update,
1622 	.offload_act_setup =	tcf_ct_offload_act_setup,
1623 	.size		=	sizeof(struct tcf_ct),
1624 };
1625 
1626 static __net_init int ct_init_net(struct net *net)
1627 {
1628 	unsigned int n_bits = sizeof_field(struct tcf_ct_params, labels) * 8;
1629 	struct tc_ct_action_net *tn = net_generic(net, ct_net_id);
1630 
1631 	if (nf_connlabels_get(net, n_bits - 1)) {
1632 		tn->labels = false;
1633 		pr_err("act_ct: Failed to set connlabels length");
1634 	} else {
1635 		tn->labels = true;
1636 	}
1637 
1638 	return tc_action_net_init(net, &tn->tn, &act_ct_ops);
1639 }
1640 
1641 static void __net_exit ct_exit_net(struct list_head *net_list)
1642 {
1643 	struct net *net;
1644 
1645 	rtnl_lock();
1646 	list_for_each_entry(net, net_list, exit_list) {
1647 		struct tc_ct_action_net *tn = net_generic(net, ct_net_id);
1648 
1649 		if (tn->labels)
1650 			nf_connlabels_put(net);
1651 	}
1652 	rtnl_unlock();
1653 
1654 	tc_action_net_exit(net_list, ct_net_id);
1655 }
1656 
1657 static struct pernet_operations ct_net_ops = {
1658 	.init = ct_init_net,
1659 	.exit_batch = ct_exit_net,
1660 	.id   = &ct_net_id,
1661 	.size = sizeof(struct tc_ct_action_net),
1662 };
1663 
1664 static int __init ct_init_module(void)
1665 {
1666 	int err;
1667 
1668 	act_ct_wq = alloc_ordered_workqueue("act_ct_workqueue", 0);
1669 	if (!act_ct_wq)
1670 		return -ENOMEM;
1671 
1672 	err = tcf_ct_flow_tables_init();
1673 	if (err)
1674 		goto err_tbl_init;
1675 
1676 	err = tcf_register_action(&act_ct_ops, &ct_net_ops);
1677 	if (err)
1678 		goto err_register;
1679 
1680 	static_branch_inc(&tcf_frag_xmit_count);
1681 
1682 	return 0;
1683 
1684 err_register:
1685 	tcf_ct_flow_tables_uninit();
1686 err_tbl_init:
1687 	destroy_workqueue(act_ct_wq);
1688 	return err;
1689 }
1690 
1691 static void __exit ct_cleanup_module(void)
1692 {
1693 	static_branch_dec(&tcf_frag_xmit_count);
1694 	tcf_unregister_action(&act_ct_ops, &ct_net_ops);
1695 	tcf_ct_flow_tables_uninit();
1696 	destroy_workqueue(act_ct_wq);
1697 }
1698 
1699 module_init(ct_init_module);
1700 module_exit(ct_cleanup_module);
1701 MODULE_AUTHOR("Paul Blakey <paulb@mellanox.com>");
1702 MODULE_AUTHOR("Yossi Kuperman <yossiku@mellanox.com>");
1703 MODULE_AUTHOR("Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>");
1704 MODULE_DESCRIPTION("Connection tracking action");
1705 MODULE_LICENSE("GPL v2");
1706