xref: /linux/net/sched/act_ct.c (revision 666ed8bfd1de3b091cf32ca03b651757dd86cfff)
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/ipv6/nf_defrag_ipv6.h>
34 #include <uapi/linux/netfilter/nf_nat.h>
35 
36 static struct workqueue_struct *act_ct_wq;
37 static struct rhashtable zones_ht;
38 static DEFINE_MUTEX(zones_mutex);
39 
40 struct tcf_ct_flow_table {
41 	struct rhash_head node; /* In zones tables */
42 
43 	struct rcu_work rwork;
44 	struct nf_flowtable nf_ft;
45 	refcount_t ref;
46 	u16 zone;
47 
48 	bool dying;
49 };
50 
51 static const struct rhashtable_params zones_params = {
52 	.head_offset = offsetof(struct tcf_ct_flow_table, node),
53 	.key_offset = offsetof(struct tcf_ct_flow_table, zone),
54 	.key_len = sizeof_field(struct tcf_ct_flow_table, zone),
55 	.automatic_shrinking = true,
56 };
57 
58 static struct flow_action_entry *
59 tcf_ct_flow_table_flow_action_get_next(struct flow_action *flow_action)
60 {
61 	int i = flow_action->num_entries++;
62 
63 	return &flow_action->entries[i];
64 }
65 
66 static void tcf_ct_add_mangle_action(struct flow_action *action,
67 				     enum flow_action_mangle_base htype,
68 				     u32 offset,
69 				     u32 mask,
70 				     u32 val)
71 {
72 	struct flow_action_entry *entry;
73 
74 	entry = tcf_ct_flow_table_flow_action_get_next(action);
75 	entry->id = FLOW_ACTION_MANGLE;
76 	entry->mangle.htype = htype;
77 	entry->mangle.mask = ~mask;
78 	entry->mangle.offset = offset;
79 	entry->mangle.val = val;
80 }
81 
82 /* The following nat helper functions check if the inverted reverse tuple
83  * (target) is different then the current dir tuple - meaning nat for ports
84  * and/or ip is needed, and add the relevant mangle actions.
85  */
86 static void
87 tcf_ct_flow_table_add_action_nat_ipv4(const struct nf_conntrack_tuple *tuple,
88 				      struct nf_conntrack_tuple target,
89 				      struct flow_action *action)
90 {
91 	if (memcmp(&target.src.u3, &tuple->src.u3, sizeof(target.src.u3)))
92 		tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_IP4,
93 					 offsetof(struct iphdr, saddr),
94 					 0xFFFFFFFF,
95 					 be32_to_cpu(target.src.u3.ip));
96 	if (memcmp(&target.dst.u3, &tuple->dst.u3, sizeof(target.dst.u3)))
97 		tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_IP4,
98 					 offsetof(struct iphdr, daddr),
99 					 0xFFFFFFFF,
100 					 be32_to_cpu(target.dst.u3.ip));
101 }
102 
103 static void
104 tcf_ct_add_ipv6_addr_mangle_action(struct flow_action *action,
105 				   union nf_inet_addr *addr,
106 				   u32 offset)
107 {
108 	int i;
109 
110 	for (i = 0; i < sizeof(struct in6_addr) / sizeof(u32); i++)
111 		tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_IP6,
112 					 i * sizeof(u32) + offset,
113 					 0xFFFFFFFF, be32_to_cpu(addr->ip6[i]));
114 }
115 
116 static void
117 tcf_ct_flow_table_add_action_nat_ipv6(const struct nf_conntrack_tuple *tuple,
118 				      struct nf_conntrack_tuple target,
119 				      struct flow_action *action)
120 {
121 	if (memcmp(&target.src.u3, &tuple->src.u3, sizeof(target.src.u3)))
122 		tcf_ct_add_ipv6_addr_mangle_action(action, &target.src.u3,
123 						   offsetof(struct ipv6hdr,
124 							    saddr));
125 	if (memcmp(&target.dst.u3, &tuple->dst.u3, sizeof(target.dst.u3)))
126 		tcf_ct_add_ipv6_addr_mangle_action(action, &target.dst.u3,
127 						   offsetof(struct ipv6hdr,
128 							    daddr));
129 }
130 
131 static void
132 tcf_ct_flow_table_add_action_nat_tcp(const struct nf_conntrack_tuple *tuple,
133 				     struct nf_conntrack_tuple target,
134 				     struct flow_action *action)
135 {
136 	__be16 target_src = target.src.u.tcp.port;
137 	__be16 target_dst = target.dst.u.tcp.port;
138 
139 	if (target_src != tuple->src.u.tcp.port)
140 		tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_TCP,
141 					 offsetof(struct tcphdr, source),
142 					 0xFFFF, be16_to_cpu(target_src));
143 	if (target_dst != tuple->dst.u.tcp.port)
144 		tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_TCP,
145 					 offsetof(struct tcphdr, dest),
146 					 0xFFFF, be16_to_cpu(target_dst));
147 }
148 
149 static void
150 tcf_ct_flow_table_add_action_nat_udp(const struct nf_conntrack_tuple *tuple,
151 				     struct nf_conntrack_tuple target,
152 				     struct flow_action *action)
153 {
154 	__be16 target_src = target.src.u.udp.port;
155 	__be16 target_dst = target.dst.u.udp.port;
156 
157 	if (target_src != tuple->src.u.udp.port)
158 		tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_TCP,
159 					 offsetof(struct udphdr, source),
160 					 0xFFFF, be16_to_cpu(target_src));
161 	if (target_dst != tuple->dst.u.udp.port)
162 		tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_TCP,
163 					 offsetof(struct udphdr, dest),
164 					 0xFFFF, be16_to_cpu(target_dst));
165 }
166 
167 static void tcf_ct_flow_table_add_action_meta(struct nf_conn *ct,
168 					      enum ip_conntrack_dir dir,
169 					      struct flow_action *action)
170 {
171 	struct nf_conn_labels *ct_labels;
172 	struct flow_action_entry *entry;
173 	enum ip_conntrack_info ctinfo;
174 	u32 *act_ct_labels;
175 
176 	entry = tcf_ct_flow_table_flow_action_get_next(action);
177 	entry->id = FLOW_ACTION_CT_METADATA;
178 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
179 	entry->ct_metadata.mark = ct->mark;
180 #endif
181 	ctinfo = dir == IP_CT_DIR_ORIGINAL ? IP_CT_ESTABLISHED :
182 					     IP_CT_ESTABLISHED_REPLY;
183 	/* aligns with the CT reference on the SKB nf_ct_set */
184 	entry->ct_metadata.cookie = (unsigned long)ct | ctinfo;
185 
186 	act_ct_labels = entry->ct_metadata.labels;
187 	ct_labels = nf_ct_labels_find(ct);
188 	if (ct_labels)
189 		memcpy(act_ct_labels, ct_labels->bits, NF_CT_LABELS_MAX_SIZE);
190 	else
191 		memset(act_ct_labels, 0, NF_CT_LABELS_MAX_SIZE);
192 }
193 
194 static int tcf_ct_flow_table_add_action_nat(struct net *net,
195 					    struct nf_conn *ct,
196 					    enum ip_conntrack_dir dir,
197 					    struct flow_action *action)
198 {
199 	const struct nf_conntrack_tuple *tuple = &ct->tuplehash[dir].tuple;
200 	struct nf_conntrack_tuple target;
201 
202 	nf_ct_invert_tuple(&target, &ct->tuplehash[!dir].tuple);
203 
204 	switch (tuple->src.l3num) {
205 	case NFPROTO_IPV4:
206 		tcf_ct_flow_table_add_action_nat_ipv4(tuple, target,
207 						      action);
208 		break;
209 	case NFPROTO_IPV6:
210 		tcf_ct_flow_table_add_action_nat_ipv6(tuple, target,
211 						      action);
212 		break;
213 	default:
214 		return -EOPNOTSUPP;
215 	}
216 
217 	switch (nf_ct_protonum(ct)) {
218 	case IPPROTO_TCP:
219 		tcf_ct_flow_table_add_action_nat_tcp(tuple, target, action);
220 		break;
221 	case IPPROTO_UDP:
222 		tcf_ct_flow_table_add_action_nat_udp(tuple, target, action);
223 		break;
224 	default:
225 		return -EOPNOTSUPP;
226 	}
227 
228 	return 0;
229 }
230 
231 static int tcf_ct_flow_table_fill_actions(struct net *net,
232 					  const struct flow_offload *flow,
233 					  enum flow_offload_tuple_dir tdir,
234 					  struct nf_flow_rule *flow_rule)
235 {
236 	struct flow_action *action = &flow_rule->rule->action;
237 	int num_entries = action->num_entries;
238 	struct nf_conn *ct = flow->ct;
239 	enum ip_conntrack_dir dir;
240 	int i, err;
241 
242 	switch (tdir) {
243 	case FLOW_OFFLOAD_DIR_ORIGINAL:
244 		dir = IP_CT_DIR_ORIGINAL;
245 		break;
246 	case FLOW_OFFLOAD_DIR_REPLY:
247 		dir = IP_CT_DIR_REPLY;
248 		break;
249 	default:
250 		return -EOPNOTSUPP;
251 	}
252 
253 	err = tcf_ct_flow_table_add_action_nat(net, ct, dir, action);
254 	if (err)
255 		goto err_nat;
256 
257 	tcf_ct_flow_table_add_action_meta(ct, dir, action);
258 	return 0;
259 
260 err_nat:
261 	/* Clear filled actions */
262 	for (i = num_entries; i < action->num_entries; i++)
263 		memset(&action->entries[i], 0, sizeof(action->entries[i]));
264 	action->num_entries = num_entries;
265 
266 	return err;
267 }
268 
269 static struct nf_flowtable_type flowtable_ct = {
270 	.action		= tcf_ct_flow_table_fill_actions,
271 	.owner		= THIS_MODULE,
272 };
273 
274 static int tcf_ct_flow_table_get(struct tcf_ct_params *params)
275 {
276 	struct tcf_ct_flow_table *ct_ft;
277 	int err = -ENOMEM;
278 
279 	mutex_lock(&zones_mutex);
280 	ct_ft = rhashtable_lookup_fast(&zones_ht, &params->zone, zones_params);
281 	if (ct_ft && refcount_inc_not_zero(&ct_ft->ref))
282 		goto out_unlock;
283 
284 	ct_ft = kzalloc(sizeof(*ct_ft), GFP_KERNEL);
285 	if (!ct_ft)
286 		goto err_alloc;
287 	refcount_set(&ct_ft->ref, 1);
288 
289 	ct_ft->zone = params->zone;
290 	err = rhashtable_insert_fast(&zones_ht, &ct_ft->node, zones_params);
291 	if (err)
292 		goto err_insert;
293 
294 	ct_ft->nf_ft.type = &flowtable_ct;
295 	ct_ft->nf_ft.flags |= NF_FLOWTABLE_HW_OFFLOAD;
296 	err = nf_flow_table_init(&ct_ft->nf_ft);
297 	if (err)
298 		goto err_init;
299 
300 	__module_get(THIS_MODULE);
301 out_unlock:
302 	params->ct_ft = ct_ft;
303 	params->nf_ft = &ct_ft->nf_ft;
304 	mutex_unlock(&zones_mutex);
305 
306 	return 0;
307 
308 err_init:
309 	rhashtable_remove_fast(&zones_ht, &ct_ft->node, zones_params);
310 err_insert:
311 	kfree(ct_ft);
312 err_alloc:
313 	mutex_unlock(&zones_mutex);
314 	return err;
315 }
316 
317 static void tcf_ct_flow_table_cleanup_work(struct work_struct *work)
318 {
319 	struct tcf_ct_flow_table *ct_ft;
320 
321 	ct_ft = container_of(to_rcu_work(work), struct tcf_ct_flow_table,
322 			     rwork);
323 	nf_flow_table_free(&ct_ft->nf_ft);
324 	kfree(ct_ft);
325 
326 	module_put(THIS_MODULE);
327 }
328 
329 static void tcf_ct_flow_table_put(struct tcf_ct_params *params)
330 {
331 	struct tcf_ct_flow_table *ct_ft = params->ct_ft;
332 
333 	if (refcount_dec_and_test(&params->ct_ft->ref)) {
334 		rhashtable_remove_fast(&zones_ht, &ct_ft->node, zones_params);
335 		INIT_RCU_WORK(&ct_ft->rwork, tcf_ct_flow_table_cleanup_work);
336 		queue_rcu_work(act_ct_wq, &ct_ft->rwork);
337 	}
338 }
339 
340 static void tcf_ct_flow_table_add(struct tcf_ct_flow_table *ct_ft,
341 				  struct nf_conn *ct,
342 				  bool tcp)
343 {
344 	struct flow_offload *entry;
345 	int err;
346 
347 	if (test_and_set_bit(IPS_OFFLOAD_BIT, &ct->status))
348 		return;
349 
350 	entry = flow_offload_alloc(ct);
351 	if (!entry) {
352 		WARN_ON_ONCE(1);
353 		goto err_alloc;
354 	}
355 
356 	if (tcp) {
357 		ct->proto.tcp.seen[0].flags |= IP_CT_TCP_FLAG_BE_LIBERAL;
358 		ct->proto.tcp.seen[1].flags |= IP_CT_TCP_FLAG_BE_LIBERAL;
359 	}
360 
361 	err = flow_offload_add(&ct_ft->nf_ft, entry);
362 	if (err)
363 		goto err_add;
364 
365 	return;
366 
367 err_add:
368 	flow_offload_free(entry);
369 err_alloc:
370 	clear_bit(IPS_OFFLOAD_BIT, &ct->status);
371 }
372 
373 static void tcf_ct_flow_table_process_conn(struct tcf_ct_flow_table *ct_ft,
374 					   struct nf_conn *ct,
375 					   enum ip_conntrack_info ctinfo)
376 {
377 	bool tcp = false;
378 
379 	if (ctinfo != IP_CT_ESTABLISHED && ctinfo != IP_CT_ESTABLISHED_REPLY)
380 		return;
381 
382 	switch (nf_ct_protonum(ct)) {
383 	case IPPROTO_TCP:
384 		tcp = true;
385 		if (ct->proto.tcp.state != TCP_CONNTRACK_ESTABLISHED)
386 			return;
387 		break;
388 	case IPPROTO_UDP:
389 		break;
390 	default:
391 		return;
392 	}
393 
394 	if (nf_ct_ext_exist(ct, NF_CT_EXT_HELPER) ||
395 	    ct->status & IPS_SEQ_ADJUST)
396 		return;
397 
398 	tcf_ct_flow_table_add(ct_ft, ct, tcp);
399 }
400 
401 static bool
402 tcf_ct_flow_table_fill_tuple_ipv4(struct sk_buff *skb,
403 				  struct flow_offload_tuple *tuple,
404 				  struct tcphdr **tcph)
405 {
406 	struct flow_ports *ports;
407 	unsigned int thoff;
408 	struct iphdr *iph;
409 
410 	if (!pskb_network_may_pull(skb, sizeof(*iph)))
411 		return false;
412 
413 	iph = ip_hdr(skb);
414 	thoff = iph->ihl * 4;
415 
416 	if (ip_is_fragment(iph) ||
417 	    unlikely(thoff != sizeof(struct iphdr)))
418 		return false;
419 
420 	if (iph->protocol != IPPROTO_TCP &&
421 	    iph->protocol != IPPROTO_UDP)
422 		return false;
423 
424 	if (iph->ttl <= 1)
425 		return false;
426 
427 	if (!pskb_network_may_pull(skb, iph->protocol == IPPROTO_TCP ?
428 					thoff + sizeof(struct tcphdr) :
429 					thoff + sizeof(*ports)))
430 		return false;
431 
432 	iph = ip_hdr(skb);
433 	if (iph->protocol == IPPROTO_TCP)
434 		*tcph = (void *)(skb_network_header(skb) + thoff);
435 
436 	ports = (struct flow_ports *)(skb_network_header(skb) + thoff);
437 	tuple->src_v4.s_addr = iph->saddr;
438 	tuple->dst_v4.s_addr = iph->daddr;
439 	tuple->src_port = ports->source;
440 	tuple->dst_port = ports->dest;
441 	tuple->l3proto = AF_INET;
442 	tuple->l4proto = iph->protocol;
443 
444 	return true;
445 }
446 
447 static bool
448 tcf_ct_flow_table_fill_tuple_ipv6(struct sk_buff *skb,
449 				  struct flow_offload_tuple *tuple,
450 				  struct tcphdr **tcph)
451 {
452 	struct flow_ports *ports;
453 	struct ipv6hdr *ip6h;
454 	unsigned int thoff;
455 
456 	if (!pskb_network_may_pull(skb, sizeof(*ip6h)))
457 		return false;
458 
459 	ip6h = ipv6_hdr(skb);
460 
461 	if (ip6h->nexthdr != IPPROTO_TCP &&
462 	    ip6h->nexthdr != IPPROTO_UDP)
463 		return false;
464 
465 	if (ip6h->hop_limit <= 1)
466 		return false;
467 
468 	thoff = sizeof(*ip6h);
469 	if (!pskb_network_may_pull(skb, ip6h->nexthdr == IPPROTO_TCP ?
470 					thoff + sizeof(struct tcphdr) :
471 					thoff + sizeof(*ports)))
472 		return false;
473 
474 	ip6h = ipv6_hdr(skb);
475 	if (ip6h->nexthdr == IPPROTO_TCP)
476 		*tcph = (void *)(skb_network_header(skb) + thoff);
477 
478 	ports = (struct flow_ports *)(skb_network_header(skb) + thoff);
479 	tuple->src_v6 = ip6h->saddr;
480 	tuple->dst_v6 = ip6h->daddr;
481 	tuple->src_port = ports->source;
482 	tuple->dst_port = ports->dest;
483 	tuple->l3proto = AF_INET6;
484 	tuple->l4proto = ip6h->nexthdr;
485 
486 	return true;
487 }
488 
489 static bool tcf_ct_flow_table_lookup(struct tcf_ct_params *p,
490 				     struct sk_buff *skb,
491 				     u8 family)
492 {
493 	struct nf_flowtable *nf_ft = &p->ct_ft->nf_ft;
494 	struct flow_offload_tuple_rhash *tuplehash;
495 	struct flow_offload_tuple tuple = {};
496 	enum ip_conntrack_info ctinfo;
497 	struct tcphdr *tcph = NULL;
498 	struct flow_offload *flow;
499 	struct nf_conn *ct;
500 	u8 dir;
501 
502 	/* Previously seen or loopback */
503 	ct = nf_ct_get(skb, &ctinfo);
504 	if ((ct && !nf_ct_is_template(ct)) || ctinfo == IP_CT_UNTRACKED)
505 		return false;
506 
507 	switch (family) {
508 	case NFPROTO_IPV4:
509 		if (!tcf_ct_flow_table_fill_tuple_ipv4(skb, &tuple, &tcph))
510 			return false;
511 		break;
512 	case NFPROTO_IPV6:
513 		if (!tcf_ct_flow_table_fill_tuple_ipv6(skb, &tuple, &tcph))
514 			return false;
515 		break;
516 	default:
517 		return false;
518 	}
519 
520 	tuplehash = flow_offload_lookup(nf_ft, &tuple);
521 	if (!tuplehash)
522 		return false;
523 
524 	dir = tuplehash->tuple.dir;
525 	flow = container_of(tuplehash, struct flow_offload, tuplehash[dir]);
526 	ct = flow->ct;
527 
528 	if (tcph && (unlikely(tcph->fin || tcph->rst))) {
529 		flow_offload_teardown(flow);
530 		return false;
531 	}
532 
533 	ctinfo = dir == FLOW_OFFLOAD_DIR_ORIGINAL ? IP_CT_ESTABLISHED :
534 						    IP_CT_ESTABLISHED_REPLY;
535 
536 	flow_offload_refresh(nf_ft, flow);
537 	nf_conntrack_get(&ct->ct_general);
538 	nf_ct_set(skb, ct, ctinfo);
539 
540 	return true;
541 }
542 
543 static int tcf_ct_flow_tables_init(void)
544 {
545 	return rhashtable_init(&zones_ht, &zones_params);
546 }
547 
548 static void tcf_ct_flow_tables_uninit(void)
549 {
550 	rhashtable_destroy(&zones_ht);
551 }
552 
553 static struct tc_action_ops act_ct_ops;
554 static unsigned int ct_net_id;
555 
556 struct tc_ct_action_net {
557 	struct tc_action_net tn; /* Must be first */
558 	bool labels;
559 };
560 
561 /* Determine whether skb->_nfct is equal to the result of conntrack lookup. */
562 static bool tcf_ct_skb_nfct_cached(struct net *net, struct sk_buff *skb,
563 				   u16 zone_id, bool force)
564 {
565 	enum ip_conntrack_info ctinfo;
566 	struct nf_conn *ct;
567 
568 	ct = nf_ct_get(skb, &ctinfo);
569 	if (!ct)
570 		return false;
571 	if (!net_eq(net, read_pnet(&ct->ct_net)))
572 		return false;
573 	if (nf_ct_zone(ct)->id != zone_id)
574 		return false;
575 
576 	/* Force conntrack entry direction. */
577 	if (force && CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) {
578 		if (nf_ct_is_confirmed(ct))
579 			nf_ct_kill(ct);
580 
581 		nf_conntrack_put(&ct->ct_general);
582 		nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
583 
584 		return false;
585 	}
586 
587 	return true;
588 }
589 
590 /* Trim the skb to the length specified by the IP/IPv6 header,
591  * removing any trailing lower-layer padding. This prepares the skb
592  * for higher-layer processing that assumes skb->len excludes padding
593  * (such as nf_ip_checksum). The caller needs to pull the skb to the
594  * network header, and ensure ip_hdr/ipv6_hdr points to valid data.
595  */
596 static int tcf_ct_skb_network_trim(struct sk_buff *skb, int family)
597 {
598 	unsigned int len;
599 	int err;
600 
601 	switch (family) {
602 	case NFPROTO_IPV4:
603 		len = ntohs(ip_hdr(skb)->tot_len);
604 		break;
605 	case NFPROTO_IPV6:
606 		len = sizeof(struct ipv6hdr)
607 			+ ntohs(ipv6_hdr(skb)->payload_len);
608 		break;
609 	default:
610 		len = skb->len;
611 	}
612 
613 	err = pskb_trim_rcsum(skb, len);
614 
615 	return err;
616 }
617 
618 static u8 tcf_ct_skb_nf_family(struct sk_buff *skb)
619 {
620 	u8 family = NFPROTO_UNSPEC;
621 
622 	switch (skb->protocol) {
623 	case htons(ETH_P_IP):
624 		family = NFPROTO_IPV4;
625 		break;
626 	case htons(ETH_P_IPV6):
627 		family = NFPROTO_IPV6;
628 		break;
629 	default:
630 		break;
631 	}
632 
633 	return family;
634 }
635 
636 static int tcf_ct_ipv4_is_fragment(struct sk_buff *skb, bool *frag)
637 {
638 	unsigned int len;
639 
640 	len =  skb_network_offset(skb) + sizeof(struct iphdr);
641 	if (unlikely(skb->len < len))
642 		return -EINVAL;
643 	if (unlikely(!pskb_may_pull(skb, len)))
644 		return -ENOMEM;
645 
646 	*frag = ip_is_fragment(ip_hdr(skb));
647 	return 0;
648 }
649 
650 static int tcf_ct_ipv6_is_fragment(struct sk_buff *skb, bool *frag)
651 {
652 	unsigned int flags = 0, len, payload_ofs = 0;
653 	unsigned short frag_off;
654 	int nexthdr;
655 
656 	len =  skb_network_offset(skb) + sizeof(struct ipv6hdr);
657 	if (unlikely(skb->len < len))
658 		return -EINVAL;
659 	if (unlikely(!pskb_may_pull(skb, len)))
660 		return -ENOMEM;
661 
662 	nexthdr = ipv6_find_hdr(skb, &payload_ofs, -1, &frag_off, &flags);
663 	if (unlikely(nexthdr < 0))
664 		return -EPROTO;
665 
666 	*frag = flags & IP6_FH_F_FRAG;
667 	return 0;
668 }
669 
670 static int tcf_ct_handle_fragments(struct net *net, struct sk_buff *skb,
671 				   u8 family, u16 zone)
672 {
673 	enum ip_conntrack_info ctinfo;
674 	struct nf_conn *ct;
675 	int err = 0;
676 	bool frag;
677 
678 	/* Previously seen (loopback)? Ignore. */
679 	ct = nf_ct_get(skb, &ctinfo);
680 	if ((ct && !nf_ct_is_template(ct)) || ctinfo == IP_CT_UNTRACKED)
681 		return 0;
682 
683 	if (family == NFPROTO_IPV4)
684 		err = tcf_ct_ipv4_is_fragment(skb, &frag);
685 	else
686 		err = tcf_ct_ipv6_is_fragment(skb, &frag);
687 	if (err || !frag)
688 		return err;
689 
690 	skb_get(skb);
691 
692 	if (family == NFPROTO_IPV4) {
693 		enum ip_defrag_users user = IP_DEFRAG_CONNTRACK_IN + zone;
694 
695 		memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
696 		local_bh_disable();
697 		err = ip_defrag(net, skb, user);
698 		local_bh_enable();
699 		if (err && err != -EINPROGRESS)
700 			goto out_free;
701 	} else { /* NFPROTO_IPV6 */
702 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
703 		enum ip6_defrag_users user = IP6_DEFRAG_CONNTRACK_IN + zone;
704 
705 		memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm));
706 		err = nf_ct_frag6_gather(net, skb, user);
707 		if (err && err != -EINPROGRESS)
708 			goto out_free;
709 #else
710 		err = -EOPNOTSUPP;
711 		goto out_free;
712 #endif
713 	}
714 
715 	skb_clear_hash(skb);
716 	skb->ignore_df = 1;
717 	return err;
718 
719 out_free:
720 	kfree_skb(skb);
721 	return err;
722 }
723 
724 static void tcf_ct_params_free(struct rcu_head *head)
725 {
726 	struct tcf_ct_params *params = container_of(head,
727 						    struct tcf_ct_params, rcu);
728 
729 	tcf_ct_flow_table_put(params);
730 
731 	if (params->tmpl)
732 		nf_conntrack_put(&params->tmpl->ct_general);
733 	kfree(params);
734 }
735 
736 #if IS_ENABLED(CONFIG_NF_NAT)
737 /* Modelled after nf_nat_ipv[46]_fn().
738  * range is only used for new, uninitialized NAT state.
739  * Returns either NF_ACCEPT or NF_DROP.
740  */
741 static int ct_nat_execute(struct sk_buff *skb, struct nf_conn *ct,
742 			  enum ip_conntrack_info ctinfo,
743 			  const struct nf_nat_range2 *range,
744 			  enum nf_nat_manip_type maniptype)
745 {
746 	int hooknum, err = NF_ACCEPT;
747 
748 	/* See HOOK2MANIP(). */
749 	if (maniptype == NF_NAT_MANIP_SRC)
750 		hooknum = NF_INET_LOCAL_IN; /* Source NAT */
751 	else
752 		hooknum = NF_INET_LOCAL_OUT; /* Destination NAT */
753 
754 	switch (ctinfo) {
755 	case IP_CT_RELATED:
756 	case IP_CT_RELATED_REPLY:
757 		if (skb->protocol == htons(ETH_P_IP) &&
758 		    ip_hdr(skb)->protocol == IPPROTO_ICMP) {
759 			if (!nf_nat_icmp_reply_translation(skb, ct, ctinfo,
760 							   hooknum))
761 				err = NF_DROP;
762 			goto out;
763 		} else if (IS_ENABLED(CONFIG_IPV6) &&
764 			   skb->protocol == htons(ETH_P_IPV6)) {
765 			__be16 frag_off;
766 			u8 nexthdr = ipv6_hdr(skb)->nexthdr;
767 			int hdrlen = ipv6_skip_exthdr(skb,
768 						      sizeof(struct ipv6hdr),
769 						      &nexthdr, &frag_off);
770 
771 			if (hdrlen >= 0 && nexthdr == IPPROTO_ICMPV6) {
772 				if (!nf_nat_icmpv6_reply_translation(skb, ct,
773 								     ctinfo,
774 								     hooknum,
775 								     hdrlen))
776 					err = NF_DROP;
777 				goto out;
778 			}
779 		}
780 		/* Non-ICMP, fall thru to initialize if needed. */
781 		/* fall through */
782 	case IP_CT_NEW:
783 		/* Seen it before?  This can happen for loopback, retrans,
784 		 * or local packets.
785 		 */
786 		if (!nf_nat_initialized(ct, maniptype)) {
787 			/* Initialize according to the NAT action. */
788 			err = (range && range->flags & NF_NAT_RANGE_MAP_IPS)
789 				/* Action is set up to establish a new
790 				 * mapping.
791 				 */
792 				? nf_nat_setup_info(ct, range, maniptype)
793 				: nf_nat_alloc_null_binding(ct, hooknum);
794 			if (err != NF_ACCEPT)
795 				goto out;
796 		}
797 		break;
798 
799 	case IP_CT_ESTABLISHED:
800 	case IP_CT_ESTABLISHED_REPLY:
801 		break;
802 
803 	default:
804 		err = NF_DROP;
805 		goto out;
806 	}
807 
808 	err = nf_nat_packet(ct, ctinfo, hooknum, skb);
809 out:
810 	return err;
811 }
812 #endif /* CONFIG_NF_NAT */
813 
814 static void tcf_ct_act_set_mark(struct nf_conn *ct, u32 mark, u32 mask)
815 {
816 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
817 	u32 new_mark;
818 
819 	if (!mask)
820 		return;
821 
822 	new_mark = mark | (ct->mark & ~(mask));
823 	if (ct->mark != new_mark) {
824 		ct->mark = new_mark;
825 		if (nf_ct_is_confirmed(ct))
826 			nf_conntrack_event_cache(IPCT_MARK, ct);
827 	}
828 #endif
829 }
830 
831 static void tcf_ct_act_set_labels(struct nf_conn *ct,
832 				  u32 *labels,
833 				  u32 *labels_m)
834 {
835 #if IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS)
836 	size_t labels_sz = sizeof_field(struct tcf_ct_params, labels);
837 
838 	if (!memchr_inv(labels_m, 0, labels_sz))
839 		return;
840 
841 	nf_connlabels_replace(ct, labels, labels_m, 4);
842 #endif
843 }
844 
845 static int tcf_ct_act_nat(struct sk_buff *skb,
846 			  struct nf_conn *ct,
847 			  enum ip_conntrack_info ctinfo,
848 			  int ct_action,
849 			  struct nf_nat_range2 *range,
850 			  bool commit)
851 {
852 #if IS_ENABLED(CONFIG_NF_NAT)
853 	int err;
854 	enum nf_nat_manip_type maniptype;
855 
856 	if (!(ct_action & TCA_CT_ACT_NAT))
857 		return NF_ACCEPT;
858 
859 	/* Add NAT extension if not confirmed yet. */
860 	if (!nf_ct_is_confirmed(ct) && !nf_ct_nat_ext_add(ct))
861 		return NF_DROP;   /* Can't NAT. */
862 
863 	if (ctinfo != IP_CT_NEW && (ct->status & IPS_NAT_MASK) &&
864 	    (ctinfo != IP_CT_RELATED || commit)) {
865 		/* NAT an established or related connection like before. */
866 		if (CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY)
867 			/* This is the REPLY direction for a connection
868 			 * for which NAT was applied in the forward
869 			 * direction.  Do the reverse NAT.
870 			 */
871 			maniptype = ct->status & IPS_SRC_NAT
872 				? NF_NAT_MANIP_DST : NF_NAT_MANIP_SRC;
873 		else
874 			maniptype = ct->status & IPS_SRC_NAT
875 				? NF_NAT_MANIP_SRC : NF_NAT_MANIP_DST;
876 	} else if (ct_action & TCA_CT_ACT_NAT_SRC) {
877 		maniptype = NF_NAT_MANIP_SRC;
878 	} else if (ct_action & TCA_CT_ACT_NAT_DST) {
879 		maniptype = NF_NAT_MANIP_DST;
880 	} else {
881 		return NF_ACCEPT;
882 	}
883 
884 	err = ct_nat_execute(skb, ct, ctinfo, range, maniptype);
885 	if (err == NF_ACCEPT &&
886 	    ct->status & IPS_SRC_NAT && ct->status & IPS_DST_NAT) {
887 		if (maniptype == NF_NAT_MANIP_SRC)
888 			maniptype = NF_NAT_MANIP_DST;
889 		else
890 			maniptype = NF_NAT_MANIP_SRC;
891 
892 		err = ct_nat_execute(skb, ct, ctinfo, range, maniptype);
893 	}
894 	return err;
895 #else
896 	return NF_ACCEPT;
897 #endif
898 }
899 
900 static int tcf_ct_act(struct sk_buff *skb, const struct tc_action *a,
901 		      struct tcf_result *res)
902 {
903 	struct net *net = dev_net(skb->dev);
904 	bool cached, commit, clear, force;
905 	enum ip_conntrack_info ctinfo;
906 	struct tcf_ct *c = to_ct(a);
907 	struct nf_conn *tmpl = NULL;
908 	struct nf_hook_state state;
909 	int nh_ofs, err, retval;
910 	struct tcf_ct_params *p;
911 	bool skip_add = false;
912 	struct nf_conn *ct;
913 	u8 family;
914 
915 	p = rcu_dereference_bh(c->params);
916 
917 	retval = READ_ONCE(c->tcf_action);
918 	commit = p->ct_action & TCA_CT_ACT_COMMIT;
919 	clear = p->ct_action & TCA_CT_ACT_CLEAR;
920 	force = p->ct_action & TCA_CT_ACT_FORCE;
921 	tmpl = p->tmpl;
922 
923 	if (clear) {
924 		ct = nf_ct_get(skb, &ctinfo);
925 		if (ct) {
926 			nf_conntrack_put(&ct->ct_general);
927 			nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
928 		}
929 
930 		goto out;
931 	}
932 
933 	family = tcf_ct_skb_nf_family(skb);
934 	if (family == NFPROTO_UNSPEC)
935 		goto drop;
936 
937 	/* The conntrack module expects to be working at L3.
938 	 * We also try to pull the IPv4/6 header to linear area
939 	 */
940 	nh_ofs = skb_network_offset(skb);
941 	skb_pull_rcsum(skb, nh_ofs);
942 	err = tcf_ct_handle_fragments(net, skb, family, p->zone);
943 	if (err == -EINPROGRESS) {
944 		retval = TC_ACT_STOLEN;
945 		goto out;
946 	}
947 	if (err)
948 		goto drop;
949 
950 	err = tcf_ct_skb_network_trim(skb, family);
951 	if (err)
952 		goto drop;
953 
954 	/* If we are recirculating packets to match on ct fields and
955 	 * committing with a separate ct action, then we don't need to
956 	 * actually run the packet through conntrack twice unless it's for a
957 	 * different zone.
958 	 */
959 	cached = tcf_ct_skb_nfct_cached(net, skb, p->zone, force);
960 	if (!cached) {
961 		if (!commit && tcf_ct_flow_table_lookup(p, skb, family)) {
962 			skip_add = true;
963 			goto do_nat;
964 		}
965 
966 		/* Associate skb with specified zone. */
967 		if (tmpl) {
968 			ct = nf_ct_get(skb, &ctinfo);
969 			if (skb_nfct(skb))
970 				nf_conntrack_put(skb_nfct(skb));
971 			nf_conntrack_get(&tmpl->ct_general);
972 			nf_ct_set(skb, tmpl, IP_CT_NEW);
973 		}
974 
975 		state.hook = NF_INET_PRE_ROUTING;
976 		state.net = net;
977 		state.pf = family;
978 		err = nf_conntrack_in(skb, &state);
979 		if (err != NF_ACCEPT)
980 			goto out_push;
981 	}
982 
983 do_nat:
984 	ct = nf_ct_get(skb, &ctinfo);
985 	if (!ct)
986 		goto out_push;
987 	nf_ct_deliver_cached_events(ct);
988 
989 	err = tcf_ct_act_nat(skb, ct, ctinfo, p->ct_action, &p->range, commit);
990 	if (err != NF_ACCEPT)
991 		goto drop;
992 
993 	if (commit) {
994 		tcf_ct_act_set_mark(ct, p->mark, p->mark_mask);
995 		tcf_ct_act_set_labels(ct, p->labels, p->labels_mask);
996 
997 		/* This will take care of sending queued events
998 		 * even if the connection is already confirmed.
999 		 */
1000 		nf_conntrack_confirm(skb);
1001 	} else if (!skip_add) {
1002 		tcf_ct_flow_table_process_conn(p->ct_ft, ct, ctinfo);
1003 	}
1004 
1005 out_push:
1006 	skb_push_rcsum(skb, nh_ofs);
1007 
1008 out:
1009 	tcf_action_update_bstats(&c->common, skb);
1010 	return retval;
1011 
1012 drop:
1013 	tcf_action_inc_drop_qstats(&c->common);
1014 	return TC_ACT_SHOT;
1015 }
1016 
1017 static const struct nla_policy ct_policy[TCA_CT_MAX + 1] = {
1018 	[TCA_CT_ACTION] = { .type = NLA_U16 },
1019 	[TCA_CT_PARMS] = { .type = NLA_EXACT_LEN, .len = sizeof(struct tc_ct) },
1020 	[TCA_CT_ZONE] = { .type = NLA_U16 },
1021 	[TCA_CT_MARK] = { .type = NLA_U32 },
1022 	[TCA_CT_MARK_MASK] = { .type = NLA_U32 },
1023 	[TCA_CT_LABELS] = { .type = NLA_BINARY,
1024 			    .len = 128 / BITS_PER_BYTE },
1025 	[TCA_CT_LABELS_MASK] = { .type = NLA_BINARY,
1026 				 .len = 128 / BITS_PER_BYTE },
1027 	[TCA_CT_NAT_IPV4_MIN] = { .type = NLA_U32 },
1028 	[TCA_CT_NAT_IPV4_MAX] = { .type = NLA_U32 },
1029 	[TCA_CT_NAT_IPV6_MIN] = { .type = NLA_EXACT_LEN,
1030 				  .len = sizeof(struct in6_addr) },
1031 	[TCA_CT_NAT_IPV6_MAX] = { .type = NLA_EXACT_LEN,
1032 				   .len = sizeof(struct in6_addr) },
1033 	[TCA_CT_NAT_PORT_MIN] = { .type = NLA_U16 },
1034 	[TCA_CT_NAT_PORT_MAX] = { .type = NLA_U16 },
1035 };
1036 
1037 static int tcf_ct_fill_params_nat(struct tcf_ct_params *p,
1038 				  struct tc_ct *parm,
1039 				  struct nlattr **tb,
1040 				  struct netlink_ext_ack *extack)
1041 {
1042 	struct nf_nat_range2 *range;
1043 
1044 	if (!(p->ct_action & TCA_CT_ACT_NAT))
1045 		return 0;
1046 
1047 	if (!IS_ENABLED(CONFIG_NF_NAT)) {
1048 		NL_SET_ERR_MSG_MOD(extack, "Netfilter nat isn't enabled in kernel");
1049 		return -EOPNOTSUPP;
1050 	}
1051 
1052 	if (!(p->ct_action & (TCA_CT_ACT_NAT_SRC | TCA_CT_ACT_NAT_DST)))
1053 		return 0;
1054 
1055 	if ((p->ct_action & TCA_CT_ACT_NAT_SRC) &&
1056 	    (p->ct_action & TCA_CT_ACT_NAT_DST)) {
1057 		NL_SET_ERR_MSG_MOD(extack, "dnat and snat can't be enabled at the same time");
1058 		return -EOPNOTSUPP;
1059 	}
1060 
1061 	range = &p->range;
1062 	if (tb[TCA_CT_NAT_IPV4_MIN]) {
1063 		struct nlattr *max_attr = tb[TCA_CT_NAT_IPV4_MAX];
1064 
1065 		p->ipv4_range = true;
1066 		range->flags |= NF_NAT_RANGE_MAP_IPS;
1067 		range->min_addr.ip =
1068 			nla_get_in_addr(tb[TCA_CT_NAT_IPV4_MIN]);
1069 
1070 		range->max_addr.ip = max_attr ?
1071 				     nla_get_in_addr(max_attr) :
1072 				     range->min_addr.ip;
1073 	} else if (tb[TCA_CT_NAT_IPV6_MIN]) {
1074 		struct nlattr *max_attr = tb[TCA_CT_NAT_IPV6_MAX];
1075 
1076 		p->ipv4_range = false;
1077 		range->flags |= NF_NAT_RANGE_MAP_IPS;
1078 		range->min_addr.in6 =
1079 			nla_get_in6_addr(tb[TCA_CT_NAT_IPV6_MIN]);
1080 
1081 		range->max_addr.in6 = max_attr ?
1082 				      nla_get_in6_addr(max_attr) :
1083 				      range->min_addr.in6;
1084 	}
1085 
1086 	if (tb[TCA_CT_NAT_PORT_MIN]) {
1087 		range->flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1088 		range->min_proto.all = nla_get_be16(tb[TCA_CT_NAT_PORT_MIN]);
1089 
1090 		range->max_proto.all = tb[TCA_CT_NAT_PORT_MAX] ?
1091 				       nla_get_be16(tb[TCA_CT_NAT_PORT_MAX]) :
1092 				       range->min_proto.all;
1093 	}
1094 
1095 	return 0;
1096 }
1097 
1098 static void tcf_ct_set_key_val(struct nlattr **tb,
1099 			       void *val, int val_type,
1100 			       void *mask, int mask_type,
1101 			       int len)
1102 {
1103 	if (!tb[val_type])
1104 		return;
1105 	nla_memcpy(val, tb[val_type], len);
1106 
1107 	if (!mask)
1108 		return;
1109 
1110 	if (mask_type == TCA_CT_UNSPEC || !tb[mask_type])
1111 		memset(mask, 0xff, len);
1112 	else
1113 		nla_memcpy(mask, tb[mask_type], len);
1114 }
1115 
1116 static int tcf_ct_fill_params(struct net *net,
1117 			      struct tcf_ct_params *p,
1118 			      struct tc_ct *parm,
1119 			      struct nlattr **tb,
1120 			      struct netlink_ext_ack *extack)
1121 {
1122 	struct tc_ct_action_net *tn = net_generic(net, ct_net_id);
1123 	struct nf_conntrack_zone zone;
1124 	struct nf_conn *tmpl;
1125 	int err;
1126 
1127 	p->zone = NF_CT_DEFAULT_ZONE_ID;
1128 
1129 	tcf_ct_set_key_val(tb,
1130 			   &p->ct_action, TCA_CT_ACTION,
1131 			   NULL, TCA_CT_UNSPEC,
1132 			   sizeof(p->ct_action));
1133 
1134 	if (p->ct_action & TCA_CT_ACT_CLEAR)
1135 		return 0;
1136 
1137 	err = tcf_ct_fill_params_nat(p, parm, tb, extack);
1138 	if (err)
1139 		return err;
1140 
1141 	if (tb[TCA_CT_MARK]) {
1142 		if (!IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)) {
1143 			NL_SET_ERR_MSG_MOD(extack, "Conntrack mark isn't enabled.");
1144 			return -EOPNOTSUPP;
1145 		}
1146 		tcf_ct_set_key_val(tb,
1147 				   &p->mark, TCA_CT_MARK,
1148 				   &p->mark_mask, TCA_CT_MARK_MASK,
1149 				   sizeof(p->mark));
1150 	}
1151 
1152 	if (tb[TCA_CT_LABELS]) {
1153 		if (!IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS)) {
1154 			NL_SET_ERR_MSG_MOD(extack, "Conntrack labels isn't enabled.");
1155 			return -EOPNOTSUPP;
1156 		}
1157 
1158 		if (!tn->labels) {
1159 			NL_SET_ERR_MSG_MOD(extack, "Failed to set connlabel length");
1160 			return -EOPNOTSUPP;
1161 		}
1162 		tcf_ct_set_key_val(tb,
1163 				   p->labels, TCA_CT_LABELS,
1164 				   p->labels_mask, TCA_CT_LABELS_MASK,
1165 				   sizeof(p->labels));
1166 	}
1167 
1168 	if (tb[TCA_CT_ZONE]) {
1169 		if (!IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES)) {
1170 			NL_SET_ERR_MSG_MOD(extack, "Conntrack zones isn't enabled.");
1171 			return -EOPNOTSUPP;
1172 		}
1173 
1174 		tcf_ct_set_key_val(tb,
1175 				   &p->zone, TCA_CT_ZONE,
1176 				   NULL, TCA_CT_UNSPEC,
1177 				   sizeof(p->zone));
1178 	}
1179 
1180 	if (p->zone == NF_CT_DEFAULT_ZONE_ID)
1181 		return 0;
1182 
1183 	nf_ct_zone_init(&zone, p->zone, NF_CT_DEFAULT_ZONE_DIR, 0);
1184 	tmpl = nf_ct_tmpl_alloc(net, &zone, GFP_KERNEL);
1185 	if (!tmpl) {
1186 		NL_SET_ERR_MSG_MOD(extack, "Failed to allocate conntrack template");
1187 		return -ENOMEM;
1188 	}
1189 	__set_bit(IPS_CONFIRMED_BIT, &tmpl->status);
1190 	nf_conntrack_get(&tmpl->ct_general);
1191 	p->tmpl = tmpl;
1192 
1193 	return 0;
1194 }
1195 
1196 static int tcf_ct_init(struct net *net, struct nlattr *nla,
1197 		       struct nlattr *est, struct tc_action **a,
1198 		       int replace, int bind, bool rtnl_held,
1199 		       struct tcf_proto *tp, u32 flags,
1200 		       struct netlink_ext_ack *extack)
1201 {
1202 	struct tc_action_net *tn = net_generic(net, ct_net_id);
1203 	struct tcf_ct_params *params = NULL;
1204 	struct nlattr *tb[TCA_CT_MAX + 1];
1205 	struct tcf_chain *goto_ch = NULL;
1206 	struct tc_ct *parm;
1207 	struct tcf_ct *c;
1208 	int err, res = 0;
1209 	u32 index;
1210 
1211 	if (!nla) {
1212 		NL_SET_ERR_MSG_MOD(extack, "Ct requires attributes to be passed");
1213 		return -EINVAL;
1214 	}
1215 
1216 	err = nla_parse_nested(tb, TCA_CT_MAX, nla, ct_policy, extack);
1217 	if (err < 0)
1218 		return err;
1219 
1220 	if (!tb[TCA_CT_PARMS]) {
1221 		NL_SET_ERR_MSG_MOD(extack, "Missing required ct parameters");
1222 		return -EINVAL;
1223 	}
1224 	parm = nla_data(tb[TCA_CT_PARMS]);
1225 	index = parm->index;
1226 	err = tcf_idr_check_alloc(tn, &index, a, bind);
1227 	if (err < 0)
1228 		return err;
1229 
1230 	if (!err) {
1231 		err = tcf_idr_create_from_flags(tn, index, est, a,
1232 						&act_ct_ops, bind, flags);
1233 		if (err) {
1234 			tcf_idr_cleanup(tn, index);
1235 			return err;
1236 		}
1237 		res = ACT_P_CREATED;
1238 	} else {
1239 		if (bind)
1240 			return 0;
1241 
1242 		if (!replace) {
1243 			tcf_idr_release(*a, bind);
1244 			return -EEXIST;
1245 		}
1246 	}
1247 	err = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack);
1248 	if (err < 0)
1249 		goto cleanup;
1250 
1251 	c = to_ct(*a);
1252 
1253 	params = kzalloc(sizeof(*params), GFP_KERNEL);
1254 	if (unlikely(!params)) {
1255 		err = -ENOMEM;
1256 		goto cleanup;
1257 	}
1258 
1259 	err = tcf_ct_fill_params(net, params, parm, tb, extack);
1260 	if (err)
1261 		goto cleanup;
1262 
1263 	err = tcf_ct_flow_table_get(params);
1264 	if (err)
1265 		goto cleanup;
1266 
1267 	spin_lock_bh(&c->tcf_lock);
1268 	goto_ch = tcf_action_set_ctrlact(*a, parm->action, goto_ch);
1269 	params = rcu_replace_pointer(c->params, params,
1270 				     lockdep_is_held(&c->tcf_lock));
1271 	spin_unlock_bh(&c->tcf_lock);
1272 
1273 	if (goto_ch)
1274 		tcf_chain_put_by_act(goto_ch);
1275 	if (params)
1276 		call_rcu(&params->rcu, tcf_ct_params_free);
1277 	if (res == ACT_P_CREATED)
1278 		tcf_idr_insert(tn, *a);
1279 
1280 	return res;
1281 
1282 cleanup:
1283 	if (goto_ch)
1284 		tcf_chain_put_by_act(goto_ch);
1285 	kfree(params);
1286 	tcf_idr_release(*a, bind);
1287 	return err;
1288 }
1289 
1290 static void tcf_ct_cleanup(struct tc_action *a)
1291 {
1292 	struct tcf_ct_params *params;
1293 	struct tcf_ct *c = to_ct(a);
1294 
1295 	params = rcu_dereference_protected(c->params, 1);
1296 	if (params)
1297 		call_rcu(&params->rcu, tcf_ct_params_free);
1298 }
1299 
1300 static int tcf_ct_dump_key_val(struct sk_buff *skb,
1301 			       void *val, int val_type,
1302 			       void *mask, int mask_type,
1303 			       int len)
1304 {
1305 	int err;
1306 
1307 	if (mask && !memchr_inv(mask, 0, len))
1308 		return 0;
1309 
1310 	err = nla_put(skb, val_type, len, val);
1311 	if (err)
1312 		return err;
1313 
1314 	if (mask_type != TCA_CT_UNSPEC) {
1315 		err = nla_put(skb, mask_type, len, mask);
1316 		if (err)
1317 			return err;
1318 	}
1319 
1320 	return 0;
1321 }
1322 
1323 static int tcf_ct_dump_nat(struct sk_buff *skb, struct tcf_ct_params *p)
1324 {
1325 	struct nf_nat_range2 *range = &p->range;
1326 
1327 	if (!(p->ct_action & TCA_CT_ACT_NAT))
1328 		return 0;
1329 
1330 	if (!(p->ct_action & (TCA_CT_ACT_NAT_SRC | TCA_CT_ACT_NAT_DST)))
1331 		return 0;
1332 
1333 	if (range->flags & NF_NAT_RANGE_MAP_IPS) {
1334 		if (p->ipv4_range) {
1335 			if (nla_put_in_addr(skb, TCA_CT_NAT_IPV4_MIN,
1336 					    range->min_addr.ip))
1337 				return -1;
1338 			if (nla_put_in_addr(skb, TCA_CT_NAT_IPV4_MAX,
1339 					    range->max_addr.ip))
1340 				return -1;
1341 		} else {
1342 			if (nla_put_in6_addr(skb, TCA_CT_NAT_IPV6_MIN,
1343 					     &range->min_addr.in6))
1344 				return -1;
1345 			if (nla_put_in6_addr(skb, TCA_CT_NAT_IPV6_MAX,
1346 					     &range->max_addr.in6))
1347 				return -1;
1348 		}
1349 	}
1350 
1351 	if (range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) {
1352 		if (nla_put_be16(skb, TCA_CT_NAT_PORT_MIN,
1353 				 range->min_proto.all))
1354 			return -1;
1355 		if (nla_put_be16(skb, TCA_CT_NAT_PORT_MAX,
1356 				 range->max_proto.all))
1357 			return -1;
1358 	}
1359 
1360 	return 0;
1361 }
1362 
1363 static inline int tcf_ct_dump(struct sk_buff *skb, struct tc_action *a,
1364 			      int bind, int ref)
1365 {
1366 	unsigned char *b = skb_tail_pointer(skb);
1367 	struct tcf_ct *c = to_ct(a);
1368 	struct tcf_ct_params *p;
1369 
1370 	struct tc_ct opt = {
1371 		.index   = c->tcf_index,
1372 		.refcnt  = refcount_read(&c->tcf_refcnt) - ref,
1373 		.bindcnt = atomic_read(&c->tcf_bindcnt) - bind,
1374 	};
1375 	struct tcf_t t;
1376 
1377 	spin_lock_bh(&c->tcf_lock);
1378 	p = rcu_dereference_protected(c->params,
1379 				      lockdep_is_held(&c->tcf_lock));
1380 	opt.action = c->tcf_action;
1381 
1382 	if (tcf_ct_dump_key_val(skb,
1383 				&p->ct_action, TCA_CT_ACTION,
1384 				NULL, TCA_CT_UNSPEC,
1385 				sizeof(p->ct_action)))
1386 		goto nla_put_failure;
1387 
1388 	if (p->ct_action & TCA_CT_ACT_CLEAR)
1389 		goto skip_dump;
1390 
1391 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1392 	    tcf_ct_dump_key_val(skb,
1393 				&p->mark, TCA_CT_MARK,
1394 				&p->mark_mask, TCA_CT_MARK_MASK,
1395 				sizeof(p->mark)))
1396 		goto nla_put_failure;
1397 
1398 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1399 	    tcf_ct_dump_key_val(skb,
1400 				p->labels, TCA_CT_LABELS,
1401 				p->labels_mask, TCA_CT_LABELS_MASK,
1402 				sizeof(p->labels)))
1403 		goto nla_put_failure;
1404 
1405 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1406 	    tcf_ct_dump_key_val(skb,
1407 				&p->zone, TCA_CT_ZONE,
1408 				NULL, TCA_CT_UNSPEC,
1409 				sizeof(p->zone)))
1410 		goto nla_put_failure;
1411 
1412 	if (tcf_ct_dump_nat(skb, p))
1413 		goto nla_put_failure;
1414 
1415 skip_dump:
1416 	if (nla_put(skb, TCA_CT_PARMS, sizeof(opt), &opt))
1417 		goto nla_put_failure;
1418 
1419 	tcf_tm_dump(&t, &c->tcf_tm);
1420 	if (nla_put_64bit(skb, TCA_CT_TM, sizeof(t), &t, TCA_CT_PAD))
1421 		goto nla_put_failure;
1422 	spin_unlock_bh(&c->tcf_lock);
1423 
1424 	return skb->len;
1425 nla_put_failure:
1426 	spin_unlock_bh(&c->tcf_lock);
1427 	nlmsg_trim(skb, b);
1428 	return -1;
1429 }
1430 
1431 static int tcf_ct_walker(struct net *net, struct sk_buff *skb,
1432 			 struct netlink_callback *cb, int type,
1433 			 const struct tc_action_ops *ops,
1434 			 struct netlink_ext_ack *extack)
1435 {
1436 	struct tc_action_net *tn = net_generic(net, ct_net_id);
1437 
1438 	return tcf_generic_walker(tn, skb, cb, type, ops, extack);
1439 }
1440 
1441 static int tcf_ct_search(struct net *net, struct tc_action **a, u32 index)
1442 {
1443 	struct tc_action_net *tn = net_generic(net, ct_net_id);
1444 
1445 	return tcf_idr_search(tn, a, index);
1446 }
1447 
1448 static void tcf_stats_update(struct tc_action *a, u64 bytes, u32 packets,
1449 			     u64 lastuse, bool hw)
1450 {
1451 	struct tcf_ct *c = to_ct(a);
1452 
1453 	tcf_action_update_stats(a, bytes, packets, false, hw);
1454 	c->tcf_tm.lastuse = max_t(u64, c->tcf_tm.lastuse, lastuse);
1455 }
1456 
1457 static struct tc_action_ops act_ct_ops = {
1458 	.kind		=	"ct",
1459 	.id		=	TCA_ID_CT,
1460 	.owner		=	THIS_MODULE,
1461 	.act		=	tcf_ct_act,
1462 	.dump		=	tcf_ct_dump,
1463 	.init		=	tcf_ct_init,
1464 	.cleanup	=	tcf_ct_cleanup,
1465 	.walk		=	tcf_ct_walker,
1466 	.lookup		=	tcf_ct_search,
1467 	.stats_update	=	tcf_stats_update,
1468 	.size		=	sizeof(struct tcf_ct),
1469 };
1470 
1471 static __net_init int ct_init_net(struct net *net)
1472 {
1473 	unsigned int n_bits = sizeof_field(struct tcf_ct_params, labels) * 8;
1474 	struct tc_ct_action_net *tn = net_generic(net, ct_net_id);
1475 
1476 	if (nf_connlabels_get(net, n_bits - 1)) {
1477 		tn->labels = false;
1478 		pr_err("act_ct: Failed to set connlabels length");
1479 	} else {
1480 		tn->labels = true;
1481 	}
1482 
1483 	return tc_action_net_init(net, &tn->tn, &act_ct_ops);
1484 }
1485 
1486 static void __net_exit ct_exit_net(struct list_head *net_list)
1487 {
1488 	struct net *net;
1489 
1490 	rtnl_lock();
1491 	list_for_each_entry(net, net_list, exit_list) {
1492 		struct tc_ct_action_net *tn = net_generic(net, ct_net_id);
1493 
1494 		if (tn->labels)
1495 			nf_connlabels_put(net);
1496 	}
1497 	rtnl_unlock();
1498 
1499 	tc_action_net_exit(net_list, ct_net_id);
1500 }
1501 
1502 static struct pernet_operations ct_net_ops = {
1503 	.init = ct_init_net,
1504 	.exit_batch = ct_exit_net,
1505 	.id   = &ct_net_id,
1506 	.size = sizeof(struct tc_ct_action_net),
1507 };
1508 
1509 static int __init ct_init_module(void)
1510 {
1511 	int err;
1512 
1513 	act_ct_wq = alloc_ordered_workqueue("act_ct_workqueue", 0);
1514 	if (!act_ct_wq)
1515 		return -ENOMEM;
1516 
1517 	err = tcf_ct_flow_tables_init();
1518 	if (err)
1519 		goto err_tbl_init;
1520 
1521 	err = tcf_register_action(&act_ct_ops, &ct_net_ops);
1522 	if (err)
1523 		goto err_register;
1524 
1525 	return 0;
1526 
1527 err_tbl_init:
1528 	destroy_workqueue(act_ct_wq);
1529 err_register:
1530 	tcf_ct_flow_tables_uninit();
1531 	return err;
1532 }
1533 
1534 static void __exit ct_cleanup_module(void)
1535 {
1536 	tcf_unregister_action(&act_ct_ops, &ct_net_ops);
1537 	tcf_ct_flow_tables_uninit();
1538 	destroy_workqueue(act_ct_wq);
1539 }
1540 
1541 void tcf_ct_flow_table_restore_skb(struct sk_buff *skb, unsigned long cookie)
1542 {
1543 	enum ip_conntrack_info ctinfo = cookie & NFCT_INFOMASK;
1544 	struct nf_conn *ct;
1545 
1546 	ct = (struct nf_conn *)(cookie & NFCT_PTRMASK);
1547 	nf_conntrack_get(&ct->ct_general);
1548 	nf_ct_set(skb, ct, ctinfo);
1549 }
1550 EXPORT_SYMBOL_GPL(tcf_ct_flow_table_restore_skb);
1551 
1552 module_init(ct_init_module);
1553 module_exit(ct_cleanup_module);
1554 MODULE_AUTHOR("Paul Blakey <paulb@mellanox.com>");
1555 MODULE_AUTHOR("Yossi Kuperman <yossiku@mellanox.com>");
1556 MODULE_AUTHOR("Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>");
1557 MODULE_DESCRIPTION("Connection tracking action");
1558 MODULE_LICENSE("GPL v2");
1559 
1560