xref: /openbmc/linux/net/sched/act_ct.c (revision 801b27e8)
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 #include <net/tc_wrapper.h>
28 
29 #include <net/netfilter/nf_flow_table.h>
30 #include <net/netfilter/nf_conntrack.h>
31 #include <net/netfilter/nf_conntrack_core.h>
32 #include <net/netfilter/nf_conntrack_zones.h>
33 #include <net/netfilter/nf_conntrack_helper.h>
34 #include <net/netfilter/nf_conntrack_acct.h>
35 #include <net/netfilter/ipv6/nf_defrag_ipv6.h>
36 #include <net/netfilter/nf_conntrack_act_ct.h>
37 #include <net/netfilter/nf_conntrack_seqadj.h>
38 #include <uapi/linux/netfilter/nf_nat.h>
39 
40 static struct workqueue_struct *act_ct_wq;
41 static struct rhashtable zones_ht;
42 static DEFINE_MUTEX(zones_mutex);
43 
44 struct tcf_ct_flow_table {
45 	struct rhash_head node; /* In zones tables */
46 
47 	struct rcu_work rwork;
48 	struct nf_flowtable nf_ft;
49 	refcount_t ref;
50 	u16 zone;
51 
52 	bool dying;
53 };
54 
55 static const struct rhashtable_params zones_params = {
56 	.head_offset = offsetof(struct tcf_ct_flow_table, node),
57 	.key_offset = offsetof(struct tcf_ct_flow_table, zone),
58 	.key_len = sizeof_field(struct tcf_ct_flow_table, zone),
59 	.automatic_shrinking = true,
60 };
61 
62 static struct flow_action_entry *
63 tcf_ct_flow_table_flow_action_get_next(struct flow_action *flow_action)
64 {
65 	int i = flow_action->num_entries++;
66 
67 	return &flow_action->entries[i];
68 }
69 
70 static void tcf_ct_add_mangle_action(struct flow_action *action,
71 				     enum flow_action_mangle_base htype,
72 				     u32 offset,
73 				     u32 mask,
74 				     u32 val)
75 {
76 	struct flow_action_entry *entry;
77 
78 	entry = tcf_ct_flow_table_flow_action_get_next(action);
79 	entry->id = FLOW_ACTION_MANGLE;
80 	entry->mangle.htype = htype;
81 	entry->mangle.mask = ~mask;
82 	entry->mangle.offset = offset;
83 	entry->mangle.val = val;
84 }
85 
86 /* The following nat helper functions check if the inverted reverse tuple
87  * (target) is different then the current dir tuple - meaning nat for ports
88  * and/or ip is needed, and add the relevant mangle actions.
89  */
90 static void
91 tcf_ct_flow_table_add_action_nat_ipv4(const struct nf_conntrack_tuple *tuple,
92 				      struct nf_conntrack_tuple target,
93 				      struct flow_action *action)
94 {
95 	if (memcmp(&target.src.u3, &tuple->src.u3, sizeof(target.src.u3)))
96 		tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_IP4,
97 					 offsetof(struct iphdr, saddr),
98 					 0xFFFFFFFF,
99 					 be32_to_cpu(target.src.u3.ip));
100 	if (memcmp(&target.dst.u3, &tuple->dst.u3, sizeof(target.dst.u3)))
101 		tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_IP4,
102 					 offsetof(struct iphdr, daddr),
103 					 0xFFFFFFFF,
104 					 be32_to_cpu(target.dst.u3.ip));
105 }
106 
107 static void
108 tcf_ct_add_ipv6_addr_mangle_action(struct flow_action *action,
109 				   union nf_inet_addr *addr,
110 				   u32 offset)
111 {
112 	int i;
113 
114 	for (i = 0; i < sizeof(struct in6_addr) / sizeof(u32); i++)
115 		tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_IP6,
116 					 i * sizeof(u32) + offset,
117 					 0xFFFFFFFF, be32_to_cpu(addr->ip6[i]));
118 }
119 
120 static void
121 tcf_ct_flow_table_add_action_nat_ipv6(const struct nf_conntrack_tuple *tuple,
122 				      struct nf_conntrack_tuple target,
123 				      struct flow_action *action)
124 {
125 	if (memcmp(&target.src.u3, &tuple->src.u3, sizeof(target.src.u3)))
126 		tcf_ct_add_ipv6_addr_mangle_action(action, &target.src.u3,
127 						   offsetof(struct ipv6hdr,
128 							    saddr));
129 	if (memcmp(&target.dst.u3, &tuple->dst.u3, sizeof(target.dst.u3)))
130 		tcf_ct_add_ipv6_addr_mangle_action(action, &target.dst.u3,
131 						   offsetof(struct ipv6hdr,
132 							    daddr));
133 }
134 
135 static void
136 tcf_ct_flow_table_add_action_nat_tcp(const struct nf_conntrack_tuple *tuple,
137 				     struct nf_conntrack_tuple target,
138 				     struct flow_action *action)
139 {
140 	__be16 target_src = target.src.u.tcp.port;
141 	__be16 target_dst = target.dst.u.tcp.port;
142 
143 	if (target_src != tuple->src.u.tcp.port)
144 		tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_TCP,
145 					 offsetof(struct tcphdr, source),
146 					 0xFFFF, be16_to_cpu(target_src));
147 	if (target_dst != tuple->dst.u.tcp.port)
148 		tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_TCP,
149 					 offsetof(struct tcphdr, dest),
150 					 0xFFFF, be16_to_cpu(target_dst));
151 }
152 
153 static void
154 tcf_ct_flow_table_add_action_nat_udp(const struct nf_conntrack_tuple *tuple,
155 				     struct nf_conntrack_tuple target,
156 				     struct flow_action *action)
157 {
158 	__be16 target_src = target.src.u.udp.port;
159 	__be16 target_dst = target.dst.u.udp.port;
160 
161 	if (target_src != tuple->src.u.udp.port)
162 		tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_UDP,
163 					 offsetof(struct udphdr, source),
164 					 0xFFFF, be16_to_cpu(target_src));
165 	if (target_dst != tuple->dst.u.udp.port)
166 		tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_UDP,
167 					 offsetof(struct udphdr, dest),
168 					 0xFFFF, be16_to_cpu(target_dst));
169 }
170 
171 static void tcf_ct_flow_table_add_action_meta(struct nf_conn *ct,
172 					      enum ip_conntrack_dir dir,
173 					      enum ip_conntrack_info ctinfo,
174 					      struct flow_action *action)
175 {
176 	struct nf_conn_labels *ct_labels;
177 	struct flow_action_entry *entry;
178 	u32 *act_ct_labels;
179 
180 	entry = tcf_ct_flow_table_flow_action_get_next(action);
181 	entry->id = FLOW_ACTION_CT_METADATA;
182 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
183 	entry->ct_metadata.mark = READ_ONCE(ct->mark);
184 #endif
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 					  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_info ctinfo;
246 	enum ip_conntrack_dir dir;
247 	int i, err;
248 
249 	switch (tdir) {
250 	case FLOW_OFFLOAD_DIR_ORIGINAL:
251 		dir = IP_CT_DIR_ORIGINAL;
252 		ctinfo = test_bit(IPS_SEEN_REPLY_BIT, &ct->status) ?
253 			IP_CT_ESTABLISHED : IP_CT_NEW;
254 		if (ctinfo == IP_CT_ESTABLISHED)
255 			set_bit(NF_FLOW_HW_ESTABLISHED, &flow->flags);
256 		break;
257 	case FLOW_OFFLOAD_DIR_REPLY:
258 		dir = IP_CT_DIR_REPLY;
259 		ctinfo = IP_CT_ESTABLISHED_REPLY;
260 		break;
261 	default:
262 		return -EOPNOTSUPP;
263 	}
264 
265 	err = tcf_ct_flow_table_add_action_nat(net, ct, dir, action);
266 	if (err)
267 		goto err_nat;
268 
269 	tcf_ct_flow_table_add_action_meta(ct, dir, ctinfo, action);
270 	return 0;
271 
272 err_nat:
273 	/* Clear filled actions */
274 	for (i = num_entries; i < action->num_entries; i++)
275 		memset(&action->entries[i], 0, sizeof(action->entries[i]));
276 	action->num_entries = num_entries;
277 
278 	return err;
279 }
280 
281 static struct nf_flowtable_type flowtable_ct = {
282 	.action		= tcf_ct_flow_table_fill_actions,
283 	.owner		= THIS_MODULE,
284 };
285 
286 static int tcf_ct_flow_table_get(struct net *net, struct tcf_ct_params *params)
287 {
288 	struct tcf_ct_flow_table *ct_ft;
289 	int err = -ENOMEM;
290 
291 	mutex_lock(&zones_mutex);
292 	ct_ft = rhashtable_lookup_fast(&zones_ht, &params->zone, zones_params);
293 	if (ct_ft && refcount_inc_not_zero(&ct_ft->ref))
294 		goto out_unlock;
295 
296 	ct_ft = kzalloc(sizeof(*ct_ft), GFP_KERNEL);
297 	if (!ct_ft)
298 		goto err_alloc;
299 	refcount_set(&ct_ft->ref, 1);
300 
301 	ct_ft->zone = params->zone;
302 	err = rhashtable_insert_fast(&zones_ht, &ct_ft->node, zones_params);
303 	if (err)
304 		goto err_insert;
305 
306 	ct_ft->nf_ft.type = &flowtable_ct;
307 	ct_ft->nf_ft.flags |= NF_FLOWTABLE_HW_OFFLOAD |
308 			      NF_FLOWTABLE_COUNTER;
309 	err = nf_flow_table_init(&ct_ft->nf_ft);
310 	if (err)
311 		goto err_init;
312 	write_pnet(&ct_ft->nf_ft.net, net);
313 
314 	__module_get(THIS_MODULE);
315 out_unlock:
316 	params->ct_ft = ct_ft;
317 	params->nf_ft = &ct_ft->nf_ft;
318 	mutex_unlock(&zones_mutex);
319 
320 	return 0;
321 
322 err_init:
323 	rhashtable_remove_fast(&zones_ht, &ct_ft->node, zones_params);
324 err_insert:
325 	kfree(ct_ft);
326 err_alloc:
327 	mutex_unlock(&zones_mutex);
328 	return err;
329 }
330 
331 static void tcf_ct_flow_table_cleanup_work(struct work_struct *work)
332 {
333 	struct flow_block_cb *block_cb, *tmp_cb;
334 	struct tcf_ct_flow_table *ct_ft;
335 	struct flow_block *block;
336 
337 	ct_ft = container_of(to_rcu_work(work), struct tcf_ct_flow_table,
338 			     rwork);
339 	nf_flow_table_free(&ct_ft->nf_ft);
340 
341 	/* Remove any remaining callbacks before cleanup */
342 	block = &ct_ft->nf_ft.flow_block;
343 	down_write(&ct_ft->nf_ft.flow_block_lock);
344 	list_for_each_entry_safe(block_cb, tmp_cb, &block->cb_list, list) {
345 		list_del(&block_cb->list);
346 		flow_block_cb_free(block_cb);
347 	}
348 	up_write(&ct_ft->nf_ft.flow_block_lock);
349 	kfree(ct_ft);
350 
351 	module_put(THIS_MODULE);
352 }
353 
354 static void tcf_ct_flow_table_put(struct tcf_ct_flow_table *ct_ft)
355 {
356 	if (refcount_dec_and_test(&ct_ft->ref)) {
357 		rhashtable_remove_fast(&zones_ht, &ct_ft->node, zones_params);
358 		INIT_RCU_WORK(&ct_ft->rwork, tcf_ct_flow_table_cleanup_work);
359 		queue_rcu_work(act_ct_wq, &ct_ft->rwork);
360 	}
361 }
362 
363 static void tcf_ct_flow_tc_ifidx(struct flow_offload *entry,
364 				 struct nf_conn_act_ct_ext *act_ct_ext, u8 dir)
365 {
366 	entry->tuplehash[dir].tuple.xmit_type = FLOW_OFFLOAD_XMIT_TC;
367 	entry->tuplehash[dir].tuple.tc.iifidx = act_ct_ext->ifindex[dir];
368 }
369 
370 static void tcf_ct_flow_table_add(struct tcf_ct_flow_table *ct_ft,
371 				  struct nf_conn *ct,
372 				  bool tcp, bool bidirectional)
373 {
374 	struct nf_conn_act_ct_ext *act_ct_ext;
375 	struct flow_offload *entry;
376 	int err;
377 
378 	if (test_and_set_bit(IPS_OFFLOAD_BIT, &ct->status))
379 		return;
380 
381 	entry = flow_offload_alloc(ct);
382 	if (!entry) {
383 		WARN_ON_ONCE(1);
384 		goto err_alloc;
385 	}
386 
387 	if (tcp) {
388 		ct->proto.tcp.seen[0].flags |= IP_CT_TCP_FLAG_BE_LIBERAL;
389 		ct->proto.tcp.seen[1].flags |= IP_CT_TCP_FLAG_BE_LIBERAL;
390 	}
391 	if (bidirectional)
392 		__set_bit(NF_FLOW_HW_BIDIRECTIONAL, &entry->flags);
393 
394 	act_ct_ext = nf_conn_act_ct_ext_find(ct);
395 	if (act_ct_ext) {
396 		tcf_ct_flow_tc_ifidx(entry, act_ct_ext, FLOW_OFFLOAD_DIR_ORIGINAL);
397 		tcf_ct_flow_tc_ifidx(entry, act_ct_ext, FLOW_OFFLOAD_DIR_REPLY);
398 	}
399 
400 	err = flow_offload_add(&ct_ft->nf_ft, entry);
401 	if (err)
402 		goto err_add;
403 
404 	return;
405 
406 err_add:
407 	flow_offload_free(entry);
408 err_alloc:
409 	clear_bit(IPS_OFFLOAD_BIT, &ct->status);
410 }
411 
412 static void tcf_ct_flow_table_process_conn(struct tcf_ct_flow_table *ct_ft,
413 					   struct nf_conn *ct,
414 					   enum ip_conntrack_info ctinfo)
415 {
416 	bool tcp = false, bidirectional = true;
417 
418 	switch (nf_ct_protonum(ct)) {
419 	case IPPROTO_TCP:
420 		if ((ctinfo != IP_CT_ESTABLISHED &&
421 		     ctinfo != IP_CT_ESTABLISHED_REPLY) ||
422 		    !test_bit(IPS_ASSURED_BIT, &ct->status) ||
423 		    ct->proto.tcp.state != TCP_CONNTRACK_ESTABLISHED)
424 			return;
425 
426 		tcp = true;
427 		break;
428 	case IPPROTO_UDP:
429 		if (!nf_ct_is_confirmed(ct))
430 			return;
431 		if (!test_bit(IPS_ASSURED_BIT, &ct->status))
432 			bidirectional = false;
433 		break;
434 #ifdef CONFIG_NF_CT_PROTO_GRE
435 	case IPPROTO_GRE: {
436 		struct nf_conntrack_tuple *tuple;
437 
438 		if ((ctinfo != IP_CT_ESTABLISHED &&
439 		     ctinfo != IP_CT_ESTABLISHED_REPLY) ||
440 		    !test_bit(IPS_ASSURED_BIT, &ct->status) ||
441 		    ct->status & IPS_NAT_MASK)
442 			return;
443 
444 		tuple = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
445 		/* No support for GRE v1 */
446 		if (tuple->src.u.gre.key || tuple->dst.u.gre.key)
447 			return;
448 		break;
449 	}
450 #endif
451 	default:
452 		return;
453 	}
454 
455 	if (nf_ct_ext_exist(ct, NF_CT_EXT_HELPER) ||
456 	    ct->status & IPS_SEQ_ADJUST)
457 		return;
458 
459 	tcf_ct_flow_table_add(ct_ft, ct, tcp, bidirectional);
460 }
461 
462 static bool
463 tcf_ct_flow_table_fill_tuple_ipv4(struct sk_buff *skb,
464 				  struct flow_offload_tuple *tuple,
465 				  struct tcphdr **tcph)
466 {
467 	struct flow_ports *ports;
468 	unsigned int thoff;
469 	struct iphdr *iph;
470 	size_t hdrsize;
471 	u8 ipproto;
472 
473 	if (!pskb_network_may_pull(skb, sizeof(*iph)))
474 		return false;
475 
476 	iph = ip_hdr(skb);
477 	thoff = iph->ihl * 4;
478 
479 	if (ip_is_fragment(iph) ||
480 	    unlikely(thoff != sizeof(struct iphdr)))
481 		return false;
482 
483 	ipproto = iph->protocol;
484 	switch (ipproto) {
485 	case IPPROTO_TCP:
486 		hdrsize = sizeof(struct tcphdr);
487 		break;
488 	case IPPROTO_UDP:
489 		hdrsize = sizeof(*ports);
490 		break;
491 #ifdef CONFIG_NF_CT_PROTO_GRE
492 	case IPPROTO_GRE:
493 		hdrsize = sizeof(struct gre_base_hdr);
494 		break;
495 #endif
496 	default:
497 		return false;
498 	}
499 
500 	if (iph->ttl <= 1)
501 		return false;
502 
503 	if (!pskb_network_may_pull(skb, thoff + hdrsize))
504 		return false;
505 
506 	switch (ipproto) {
507 	case IPPROTO_TCP:
508 		*tcph = (void *)(skb_network_header(skb) + thoff);
509 		fallthrough;
510 	case IPPROTO_UDP:
511 		ports = (struct flow_ports *)(skb_network_header(skb) + thoff);
512 		tuple->src_port = ports->source;
513 		tuple->dst_port = ports->dest;
514 		break;
515 	case IPPROTO_GRE: {
516 		struct gre_base_hdr *greh;
517 
518 		greh = (struct gre_base_hdr *)(skb_network_header(skb) + thoff);
519 		if ((greh->flags & GRE_VERSION) != GRE_VERSION_0)
520 			return false;
521 		break;
522 	}
523 	}
524 
525 	iph = ip_hdr(skb);
526 
527 	tuple->src_v4.s_addr = iph->saddr;
528 	tuple->dst_v4.s_addr = iph->daddr;
529 	tuple->l3proto = AF_INET;
530 	tuple->l4proto = ipproto;
531 
532 	return true;
533 }
534 
535 static bool
536 tcf_ct_flow_table_fill_tuple_ipv6(struct sk_buff *skb,
537 				  struct flow_offload_tuple *tuple,
538 				  struct tcphdr **tcph)
539 {
540 	struct flow_ports *ports;
541 	struct ipv6hdr *ip6h;
542 	unsigned int thoff;
543 	size_t hdrsize;
544 	u8 nexthdr;
545 
546 	if (!pskb_network_may_pull(skb, sizeof(*ip6h)))
547 		return false;
548 
549 	ip6h = ipv6_hdr(skb);
550 	thoff = sizeof(*ip6h);
551 
552 	nexthdr = ip6h->nexthdr;
553 	switch (nexthdr) {
554 	case IPPROTO_TCP:
555 		hdrsize = sizeof(struct tcphdr);
556 		break;
557 	case IPPROTO_UDP:
558 		hdrsize = sizeof(*ports);
559 		break;
560 #ifdef CONFIG_NF_CT_PROTO_GRE
561 	case IPPROTO_GRE:
562 		hdrsize = sizeof(struct gre_base_hdr);
563 		break;
564 #endif
565 	default:
566 		return false;
567 	}
568 
569 	if (ip6h->hop_limit <= 1)
570 		return false;
571 
572 	if (!pskb_network_may_pull(skb, thoff + hdrsize))
573 		return false;
574 
575 	switch (nexthdr) {
576 	case IPPROTO_TCP:
577 		*tcph = (void *)(skb_network_header(skb) + thoff);
578 		fallthrough;
579 	case IPPROTO_UDP:
580 		ports = (struct flow_ports *)(skb_network_header(skb) + thoff);
581 		tuple->src_port = ports->source;
582 		tuple->dst_port = ports->dest;
583 		break;
584 	case IPPROTO_GRE: {
585 		struct gre_base_hdr *greh;
586 
587 		greh = (struct gre_base_hdr *)(skb_network_header(skb) + thoff);
588 		if ((greh->flags & GRE_VERSION) != GRE_VERSION_0)
589 			return false;
590 		break;
591 	}
592 	}
593 
594 	ip6h = ipv6_hdr(skb);
595 
596 	tuple->src_v6 = ip6h->saddr;
597 	tuple->dst_v6 = ip6h->daddr;
598 	tuple->l3proto = AF_INET6;
599 	tuple->l4proto = nexthdr;
600 
601 	return true;
602 }
603 
604 static bool tcf_ct_flow_table_lookup(struct tcf_ct_params *p,
605 				     struct sk_buff *skb,
606 				     u8 family)
607 {
608 	struct nf_flowtable *nf_ft = &p->ct_ft->nf_ft;
609 	struct flow_offload_tuple_rhash *tuplehash;
610 	struct flow_offload_tuple tuple = {};
611 	enum ip_conntrack_info ctinfo;
612 	struct tcphdr *tcph = NULL;
613 	bool force_refresh = false;
614 	struct flow_offload *flow;
615 	struct nf_conn *ct;
616 	u8 dir;
617 
618 	switch (family) {
619 	case NFPROTO_IPV4:
620 		if (!tcf_ct_flow_table_fill_tuple_ipv4(skb, &tuple, &tcph))
621 			return false;
622 		break;
623 	case NFPROTO_IPV6:
624 		if (!tcf_ct_flow_table_fill_tuple_ipv6(skb, &tuple, &tcph))
625 			return false;
626 		break;
627 	default:
628 		return false;
629 	}
630 
631 	tuplehash = flow_offload_lookup(nf_ft, &tuple);
632 	if (!tuplehash)
633 		return false;
634 
635 	dir = tuplehash->tuple.dir;
636 	flow = container_of(tuplehash, struct flow_offload, tuplehash[dir]);
637 	ct = flow->ct;
638 
639 	if (dir == FLOW_OFFLOAD_DIR_REPLY &&
640 	    !test_bit(NF_FLOW_HW_BIDIRECTIONAL, &flow->flags)) {
641 		/* Only offload reply direction after connection became
642 		 * assured.
643 		 */
644 		if (test_bit(IPS_ASSURED_BIT, &ct->status))
645 			set_bit(NF_FLOW_HW_BIDIRECTIONAL, &flow->flags);
646 		else if (test_bit(NF_FLOW_HW_ESTABLISHED, &flow->flags))
647 			/* If flow_table flow has already been updated to the
648 			 * established state, then don't refresh.
649 			 */
650 			return false;
651 		force_refresh = true;
652 	}
653 
654 	if (tcph && (unlikely(tcph->fin || tcph->rst))) {
655 		flow_offload_teardown(flow);
656 		return false;
657 	}
658 
659 	if (dir == FLOW_OFFLOAD_DIR_ORIGINAL)
660 		ctinfo = test_bit(IPS_SEEN_REPLY_BIT, &ct->status) ?
661 			IP_CT_ESTABLISHED : IP_CT_NEW;
662 	else
663 		ctinfo = IP_CT_ESTABLISHED_REPLY;
664 
665 	flow_offload_refresh(nf_ft, flow, force_refresh);
666 	if (!test_bit(IPS_ASSURED_BIT, &ct->status)) {
667 		/* Process this flow in SW to allow promoting to ASSURED */
668 		return false;
669 	}
670 
671 	nf_conntrack_get(&ct->ct_general);
672 	nf_ct_set(skb, ct, ctinfo);
673 	if (nf_ft->flags & NF_FLOWTABLE_COUNTER)
674 		nf_ct_acct_update(ct, dir, skb->len);
675 
676 	return true;
677 }
678 
679 static int tcf_ct_flow_tables_init(void)
680 {
681 	return rhashtable_init(&zones_ht, &zones_params);
682 }
683 
684 static void tcf_ct_flow_tables_uninit(void)
685 {
686 	rhashtable_destroy(&zones_ht);
687 }
688 
689 static struct tc_action_ops act_ct_ops;
690 
691 struct tc_ct_action_net {
692 	struct tc_action_net tn; /* Must be first */
693 	bool labels;
694 };
695 
696 /* Determine whether skb->_nfct is equal to the result of conntrack lookup. */
697 static bool tcf_ct_skb_nfct_cached(struct net *net, struct sk_buff *skb,
698 				   struct tcf_ct_params *p)
699 {
700 	enum ip_conntrack_info ctinfo;
701 	struct nf_conn *ct;
702 
703 	ct = nf_ct_get(skb, &ctinfo);
704 	if (!ct)
705 		return false;
706 	if (!net_eq(net, read_pnet(&ct->ct_net)))
707 		goto drop_ct;
708 	if (nf_ct_zone(ct)->id != p->zone)
709 		goto drop_ct;
710 	if (p->helper) {
711 		struct nf_conn_help *help;
712 
713 		help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER);
714 		if (help && rcu_access_pointer(help->helper) != p->helper)
715 			goto drop_ct;
716 	}
717 
718 	/* Force conntrack entry direction. */
719 	if ((p->ct_action & TCA_CT_ACT_FORCE) &&
720 	    CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) {
721 		if (nf_ct_is_confirmed(ct))
722 			nf_ct_kill(ct);
723 
724 		goto drop_ct;
725 	}
726 
727 	return true;
728 
729 drop_ct:
730 	nf_ct_put(ct);
731 	nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
732 
733 	return false;
734 }
735 
736 static u8 tcf_ct_skb_nf_family(struct sk_buff *skb)
737 {
738 	u8 family = NFPROTO_UNSPEC;
739 
740 	switch (skb_protocol(skb, true)) {
741 	case htons(ETH_P_IP):
742 		family = NFPROTO_IPV4;
743 		break;
744 	case htons(ETH_P_IPV6):
745 		family = NFPROTO_IPV6;
746 		break;
747 	default:
748 		break;
749 	}
750 
751 	return family;
752 }
753 
754 static int tcf_ct_ipv4_is_fragment(struct sk_buff *skb, bool *frag)
755 {
756 	unsigned int len;
757 
758 	len =  skb_network_offset(skb) + sizeof(struct iphdr);
759 	if (unlikely(skb->len < len))
760 		return -EINVAL;
761 	if (unlikely(!pskb_may_pull(skb, len)))
762 		return -ENOMEM;
763 
764 	*frag = ip_is_fragment(ip_hdr(skb));
765 	return 0;
766 }
767 
768 static int tcf_ct_ipv6_is_fragment(struct sk_buff *skb, bool *frag)
769 {
770 	unsigned int flags = 0, len, payload_ofs = 0;
771 	unsigned short frag_off;
772 	int nexthdr;
773 
774 	len =  skb_network_offset(skb) + sizeof(struct ipv6hdr);
775 	if (unlikely(skb->len < len))
776 		return -EINVAL;
777 	if (unlikely(!pskb_may_pull(skb, len)))
778 		return -ENOMEM;
779 
780 	nexthdr = ipv6_find_hdr(skb, &payload_ofs, -1, &frag_off, &flags);
781 	if (unlikely(nexthdr < 0))
782 		return -EPROTO;
783 
784 	*frag = flags & IP6_FH_F_FRAG;
785 	return 0;
786 }
787 
788 static int tcf_ct_handle_fragments(struct net *net, struct sk_buff *skb,
789 				   u8 family, u16 zone, bool *defrag)
790 {
791 	enum ip_conntrack_info ctinfo;
792 	struct nf_conn *ct;
793 	int err = 0;
794 	bool frag;
795 	u8 proto;
796 	u16 mru;
797 
798 	/* Previously seen (loopback)? Ignore. */
799 	ct = nf_ct_get(skb, &ctinfo);
800 	if ((ct && !nf_ct_is_template(ct)) || ctinfo == IP_CT_UNTRACKED)
801 		return 0;
802 
803 	if (family == NFPROTO_IPV4)
804 		err = tcf_ct_ipv4_is_fragment(skb, &frag);
805 	else
806 		err = tcf_ct_ipv6_is_fragment(skb, &frag);
807 	if (err || !frag)
808 		return err;
809 
810 	skb_get(skb);
811 	err = nf_ct_handle_fragments(net, skb, zone, family, &proto, &mru);
812 	if (err)
813 		return err;
814 
815 	*defrag = true;
816 	tc_skb_cb(skb)->mru = mru;
817 
818 	return 0;
819 }
820 
821 static void tcf_ct_params_free(struct tcf_ct_params *params)
822 {
823 	if (params->helper) {
824 #if IS_ENABLED(CONFIG_NF_NAT)
825 		if (params->ct_action & TCA_CT_ACT_NAT)
826 			nf_nat_helper_put(params->helper);
827 #endif
828 		nf_conntrack_helper_put(params->helper);
829 	}
830 	if (params->ct_ft)
831 		tcf_ct_flow_table_put(params->ct_ft);
832 	if (params->tmpl)
833 		nf_ct_put(params->tmpl);
834 	kfree(params);
835 }
836 
837 static void tcf_ct_params_free_rcu(struct rcu_head *head)
838 {
839 	struct tcf_ct_params *params;
840 
841 	params = container_of(head, struct tcf_ct_params, rcu);
842 	tcf_ct_params_free(params);
843 }
844 
845 static void tcf_ct_act_set_mark(struct nf_conn *ct, u32 mark, u32 mask)
846 {
847 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
848 	u32 new_mark;
849 
850 	if (!mask)
851 		return;
852 
853 	new_mark = mark | (READ_ONCE(ct->mark) & ~(mask));
854 	if (READ_ONCE(ct->mark) != new_mark) {
855 		WRITE_ONCE(ct->mark, new_mark);
856 		if (nf_ct_is_confirmed(ct))
857 			nf_conntrack_event_cache(IPCT_MARK, ct);
858 	}
859 #endif
860 }
861 
862 static void tcf_ct_act_set_labels(struct nf_conn *ct,
863 				  u32 *labels,
864 				  u32 *labels_m)
865 {
866 #if IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS)
867 	size_t labels_sz = sizeof_field(struct tcf_ct_params, labels);
868 
869 	if (!memchr_inv(labels_m, 0, labels_sz))
870 		return;
871 
872 	nf_connlabels_replace(ct, labels, labels_m, 4);
873 #endif
874 }
875 
876 static int tcf_ct_act_nat(struct sk_buff *skb,
877 			  struct nf_conn *ct,
878 			  enum ip_conntrack_info ctinfo,
879 			  int ct_action,
880 			  struct nf_nat_range2 *range,
881 			  bool commit)
882 {
883 #if IS_ENABLED(CONFIG_NF_NAT)
884 	int err, action = 0;
885 
886 	if (!(ct_action & TCA_CT_ACT_NAT))
887 		return NF_ACCEPT;
888 	if (ct_action & TCA_CT_ACT_NAT_SRC)
889 		action |= BIT(NF_NAT_MANIP_SRC);
890 	if (ct_action & TCA_CT_ACT_NAT_DST)
891 		action |= BIT(NF_NAT_MANIP_DST);
892 
893 	err = nf_ct_nat(skb, ct, ctinfo, &action, range, commit);
894 
895 	if (action & BIT(NF_NAT_MANIP_SRC))
896 		tc_skb_cb(skb)->post_ct_snat = 1;
897 	if (action & BIT(NF_NAT_MANIP_DST))
898 		tc_skb_cb(skb)->post_ct_dnat = 1;
899 
900 	return err;
901 #else
902 	return NF_ACCEPT;
903 #endif
904 }
905 
906 TC_INDIRECT_SCOPE int tcf_ct_act(struct sk_buff *skb, const struct tc_action *a,
907 				 struct tcf_result *res)
908 {
909 	struct net *net = dev_net(skb->dev);
910 	enum ip_conntrack_info ctinfo;
911 	struct tcf_ct *c = to_ct(a);
912 	struct nf_conn *tmpl = NULL;
913 	struct nf_hook_state state;
914 	bool cached, commit, clear;
915 	int nh_ofs, err, retval;
916 	struct tcf_ct_params *p;
917 	bool add_helper = false;
918 	bool skip_add = false;
919 	bool defrag = false;
920 	struct nf_conn *ct;
921 	u8 family;
922 
923 	p = rcu_dereference_bh(c->params);
924 
925 	retval = READ_ONCE(c->tcf_action);
926 	commit = p->ct_action & TCA_CT_ACT_COMMIT;
927 	clear = p->ct_action & TCA_CT_ACT_CLEAR;
928 	tmpl = p->tmpl;
929 
930 	tcf_lastuse_update(&c->tcf_tm);
931 	tcf_action_update_bstats(&c->common, skb);
932 
933 	if (clear) {
934 		tc_skb_cb(skb)->post_ct = false;
935 		ct = nf_ct_get(skb, &ctinfo);
936 		if (ct) {
937 			nf_ct_put(ct);
938 			nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
939 		}
940 
941 		goto out_clear;
942 	}
943 
944 	family = tcf_ct_skb_nf_family(skb);
945 	if (family == NFPROTO_UNSPEC)
946 		goto drop;
947 
948 	/* The conntrack module expects to be working at L3.
949 	 * We also try to pull the IPv4/6 header to linear area
950 	 */
951 	nh_ofs = skb_network_offset(skb);
952 	skb_pull_rcsum(skb, nh_ofs);
953 	err = tcf_ct_handle_fragments(net, skb, family, p->zone, &defrag);
954 	if (err == -EINPROGRESS) {
955 		retval = TC_ACT_STOLEN;
956 		goto out_clear;
957 	}
958 	if (err)
959 		goto drop;
960 
961 	err = nf_ct_skb_network_trim(skb, family);
962 	if (err)
963 		goto drop;
964 
965 	/* If we are recirculating packets to match on ct fields and
966 	 * committing with a separate ct action, then we don't need to
967 	 * actually run the packet through conntrack twice unless it's for a
968 	 * different zone.
969 	 */
970 	cached = tcf_ct_skb_nfct_cached(net, skb, p);
971 	if (!cached) {
972 		if (tcf_ct_flow_table_lookup(p, skb, family)) {
973 			skip_add = true;
974 			goto do_nat;
975 		}
976 
977 		/* Associate skb with specified zone. */
978 		if (tmpl) {
979 			nf_conntrack_put(skb_nfct(skb));
980 			nf_conntrack_get(&tmpl->ct_general);
981 			nf_ct_set(skb, tmpl, IP_CT_NEW);
982 		}
983 
984 		state.hook = NF_INET_PRE_ROUTING;
985 		state.net = net;
986 		state.pf = family;
987 		err = nf_conntrack_in(skb, &state);
988 		if (err != NF_ACCEPT)
989 			goto out_push;
990 	}
991 
992 do_nat:
993 	ct = nf_ct_get(skb, &ctinfo);
994 	if (!ct)
995 		goto out_push;
996 	nf_ct_deliver_cached_events(ct);
997 	nf_conn_act_ct_ext_fill(skb, ct, ctinfo);
998 
999 	err = tcf_ct_act_nat(skb, ct, ctinfo, p->ct_action, &p->range, commit);
1000 	if (err != NF_ACCEPT)
1001 		goto drop;
1002 
1003 	if (!nf_ct_is_confirmed(ct) && commit && p->helper && !nfct_help(ct)) {
1004 		err = __nf_ct_try_assign_helper(ct, p->tmpl, GFP_ATOMIC);
1005 		if (err)
1006 			goto drop;
1007 		add_helper = true;
1008 		if (p->ct_action & TCA_CT_ACT_NAT && !nfct_seqadj(ct)) {
1009 			if (!nfct_seqadj_ext_add(ct))
1010 				goto drop;
1011 		}
1012 	}
1013 
1014 	if (nf_ct_is_confirmed(ct) ? ((!cached && !skip_add) || add_helper) : commit) {
1015 		if (nf_ct_helper(skb, ct, ctinfo, family) != NF_ACCEPT)
1016 			goto drop;
1017 	}
1018 
1019 	if (commit) {
1020 		tcf_ct_act_set_mark(ct, p->mark, p->mark_mask);
1021 		tcf_ct_act_set_labels(ct, p->labels, p->labels_mask);
1022 
1023 		if (!nf_ct_is_confirmed(ct))
1024 			nf_conn_act_ct_ext_add(ct);
1025 
1026 		/* This will take care of sending queued events
1027 		 * even if the connection is already confirmed.
1028 		 */
1029 		if (nf_conntrack_confirm(skb) != NF_ACCEPT)
1030 			goto drop;
1031 	}
1032 
1033 	if (!skip_add)
1034 		tcf_ct_flow_table_process_conn(p->ct_ft, ct, ctinfo);
1035 
1036 out_push:
1037 	skb_push_rcsum(skb, nh_ofs);
1038 
1039 	tc_skb_cb(skb)->post_ct = true;
1040 	tc_skb_cb(skb)->zone = p->zone;
1041 out_clear:
1042 	if (defrag)
1043 		qdisc_skb_cb(skb)->pkt_len = skb->len;
1044 	return retval;
1045 
1046 drop:
1047 	tcf_action_inc_drop_qstats(&c->common);
1048 	return TC_ACT_SHOT;
1049 }
1050 
1051 static const struct nla_policy ct_policy[TCA_CT_MAX + 1] = {
1052 	[TCA_CT_ACTION] = { .type = NLA_U16 },
1053 	[TCA_CT_PARMS] = NLA_POLICY_EXACT_LEN(sizeof(struct tc_ct)),
1054 	[TCA_CT_ZONE] = { .type = NLA_U16 },
1055 	[TCA_CT_MARK] = { .type = NLA_U32 },
1056 	[TCA_CT_MARK_MASK] = { .type = NLA_U32 },
1057 	[TCA_CT_LABELS] = { .type = NLA_BINARY,
1058 			    .len = 128 / BITS_PER_BYTE },
1059 	[TCA_CT_LABELS_MASK] = { .type = NLA_BINARY,
1060 				 .len = 128 / BITS_PER_BYTE },
1061 	[TCA_CT_NAT_IPV4_MIN] = { .type = NLA_U32 },
1062 	[TCA_CT_NAT_IPV4_MAX] = { .type = NLA_U32 },
1063 	[TCA_CT_NAT_IPV6_MIN] = NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)),
1064 	[TCA_CT_NAT_IPV6_MAX] = NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)),
1065 	[TCA_CT_NAT_PORT_MIN] = { .type = NLA_U16 },
1066 	[TCA_CT_NAT_PORT_MAX] = { .type = NLA_U16 },
1067 	[TCA_CT_HELPER_NAME] = { .type = NLA_STRING, .len = NF_CT_HELPER_NAME_LEN },
1068 	[TCA_CT_HELPER_FAMILY] = { .type = NLA_U8 },
1069 	[TCA_CT_HELPER_PROTO] = { .type = NLA_U8 },
1070 };
1071 
1072 static int tcf_ct_fill_params_nat(struct tcf_ct_params *p,
1073 				  struct tc_ct *parm,
1074 				  struct nlattr **tb,
1075 				  struct netlink_ext_ack *extack)
1076 {
1077 	struct nf_nat_range2 *range;
1078 
1079 	if (!(p->ct_action & TCA_CT_ACT_NAT))
1080 		return 0;
1081 
1082 	if (!IS_ENABLED(CONFIG_NF_NAT)) {
1083 		NL_SET_ERR_MSG_MOD(extack, "Netfilter nat isn't enabled in kernel");
1084 		return -EOPNOTSUPP;
1085 	}
1086 
1087 	if (!(p->ct_action & (TCA_CT_ACT_NAT_SRC | TCA_CT_ACT_NAT_DST)))
1088 		return 0;
1089 
1090 	if ((p->ct_action & TCA_CT_ACT_NAT_SRC) &&
1091 	    (p->ct_action & TCA_CT_ACT_NAT_DST)) {
1092 		NL_SET_ERR_MSG_MOD(extack, "dnat and snat can't be enabled at the same time");
1093 		return -EOPNOTSUPP;
1094 	}
1095 
1096 	range = &p->range;
1097 	if (tb[TCA_CT_NAT_IPV4_MIN]) {
1098 		struct nlattr *max_attr = tb[TCA_CT_NAT_IPV4_MAX];
1099 
1100 		p->ipv4_range = true;
1101 		range->flags |= NF_NAT_RANGE_MAP_IPS;
1102 		range->min_addr.ip =
1103 			nla_get_in_addr(tb[TCA_CT_NAT_IPV4_MIN]);
1104 
1105 		range->max_addr.ip = max_attr ?
1106 				     nla_get_in_addr(max_attr) :
1107 				     range->min_addr.ip;
1108 	} else if (tb[TCA_CT_NAT_IPV6_MIN]) {
1109 		struct nlattr *max_attr = tb[TCA_CT_NAT_IPV6_MAX];
1110 
1111 		p->ipv4_range = false;
1112 		range->flags |= NF_NAT_RANGE_MAP_IPS;
1113 		range->min_addr.in6 =
1114 			nla_get_in6_addr(tb[TCA_CT_NAT_IPV6_MIN]);
1115 
1116 		range->max_addr.in6 = max_attr ?
1117 				      nla_get_in6_addr(max_attr) :
1118 				      range->min_addr.in6;
1119 	}
1120 
1121 	if (tb[TCA_CT_NAT_PORT_MIN]) {
1122 		range->flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1123 		range->min_proto.all = nla_get_be16(tb[TCA_CT_NAT_PORT_MIN]);
1124 
1125 		range->max_proto.all = tb[TCA_CT_NAT_PORT_MAX] ?
1126 				       nla_get_be16(tb[TCA_CT_NAT_PORT_MAX]) :
1127 				       range->min_proto.all;
1128 	}
1129 
1130 	return 0;
1131 }
1132 
1133 static void tcf_ct_set_key_val(struct nlattr **tb,
1134 			       void *val, int val_type,
1135 			       void *mask, int mask_type,
1136 			       int len)
1137 {
1138 	if (!tb[val_type])
1139 		return;
1140 	nla_memcpy(val, tb[val_type], len);
1141 
1142 	if (!mask)
1143 		return;
1144 
1145 	if (mask_type == TCA_CT_UNSPEC || !tb[mask_type])
1146 		memset(mask, 0xff, len);
1147 	else
1148 		nla_memcpy(mask, tb[mask_type], len);
1149 }
1150 
1151 static int tcf_ct_fill_params(struct net *net,
1152 			      struct tcf_ct_params *p,
1153 			      struct tc_ct *parm,
1154 			      struct nlattr **tb,
1155 			      struct netlink_ext_ack *extack)
1156 {
1157 	struct tc_ct_action_net *tn = net_generic(net, act_ct_ops.net_id);
1158 	struct nf_conntrack_zone zone;
1159 	int err, family, proto, len;
1160 	struct nf_conn *tmpl;
1161 	char *name;
1162 
1163 	p->zone = NF_CT_DEFAULT_ZONE_ID;
1164 
1165 	tcf_ct_set_key_val(tb,
1166 			   &p->ct_action, TCA_CT_ACTION,
1167 			   NULL, TCA_CT_UNSPEC,
1168 			   sizeof(p->ct_action));
1169 
1170 	if (p->ct_action & TCA_CT_ACT_CLEAR)
1171 		return 0;
1172 
1173 	err = tcf_ct_fill_params_nat(p, parm, tb, extack);
1174 	if (err)
1175 		return err;
1176 
1177 	if (tb[TCA_CT_MARK]) {
1178 		if (!IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)) {
1179 			NL_SET_ERR_MSG_MOD(extack, "Conntrack mark isn't enabled.");
1180 			return -EOPNOTSUPP;
1181 		}
1182 		tcf_ct_set_key_val(tb,
1183 				   &p->mark, TCA_CT_MARK,
1184 				   &p->mark_mask, TCA_CT_MARK_MASK,
1185 				   sizeof(p->mark));
1186 	}
1187 
1188 	if (tb[TCA_CT_LABELS]) {
1189 		if (!IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS)) {
1190 			NL_SET_ERR_MSG_MOD(extack, "Conntrack labels isn't enabled.");
1191 			return -EOPNOTSUPP;
1192 		}
1193 
1194 		if (!tn->labels) {
1195 			NL_SET_ERR_MSG_MOD(extack, "Failed to set connlabel length");
1196 			return -EOPNOTSUPP;
1197 		}
1198 		tcf_ct_set_key_val(tb,
1199 				   p->labels, TCA_CT_LABELS,
1200 				   p->labels_mask, TCA_CT_LABELS_MASK,
1201 				   sizeof(p->labels));
1202 	}
1203 
1204 	if (tb[TCA_CT_ZONE]) {
1205 		if (!IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES)) {
1206 			NL_SET_ERR_MSG_MOD(extack, "Conntrack zones isn't enabled.");
1207 			return -EOPNOTSUPP;
1208 		}
1209 
1210 		tcf_ct_set_key_val(tb,
1211 				   &p->zone, TCA_CT_ZONE,
1212 				   NULL, TCA_CT_UNSPEC,
1213 				   sizeof(p->zone));
1214 	}
1215 
1216 	nf_ct_zone_init(&zone, p->zone, NF_CT_DEFAULT_ZONE_DIR, 0);
1217 	tmpl = nf_ct_tmpl_alloc(net, &zone, GFP_KERNEL);
1218 	if (!tmpl) {
1219 		NL_SET_ERR_MSG_MOD(extack, "Failed to allocate conntrack template");
1220 		return -ENOMEM;
1221 	}
1222 	p->tmpl = tmpl;
1223 	if (tb[TCA_CT_HELPER_NAME]) {
1224 		name = nla_data(tb[TCA_CT_HELPER_NAME]);
1225 		len = nla_len(tb[TCA_CT_HELPER_NAME]);
1226 		if (len > 16 || name[len - 1] != '\0') {
1227 			NL_SET_ERR_MSG_MOD(extack, "Failed to parse helper name.");
1228 			err = -EINVAL;
1229 			goto err;
1230 		}
1231 		family = tb[TCA_CT_HELPER_FAMILY] ? nla_get_u8(tb[TCA_CT_HELPER_FAMILY]) : AF_INET;
1232 		proto = tb[TCA_CT_HELPER_PROTO] ? nla_get_u8(tb[TCA_CT_HELPER_PROTO]) : IPPROTO_TCP;
1233 		err = nf_ct_add_helper(tmpl, name, family, proto,
1234 				       p->ct_action & TCA_CT_ACT_NAT, &p->helper);
1235 		if (err) {
1236 			NL_SET_ERR_MSG_MOD(extack, "Failed to add helper");
1237 			goto err;
1238 		}
1239 	}
1240 
1241 	if (p->ct_action & TCA_CT_ACT_COMMIT)
1242 		__set_bit(IPS_CONFIRMED_BIT, &tmpl->status);
1243 	return 0;
1244 err:
1245 	nf_ct_put(p->tmpl);
1246 	p->tmpl = NULL;
1247 	return err;
1248 }
1249 
1250 static int tcf_ct_init(struct net *net, struct nlattr *nla,
1251 		       struct nlattr *est, struct tc_action **a,
1252 		       struct tcf_proto *tp, u32 flags,
1253 		       struct netlink_ext_ack *extack)
1254 {
1255 	struct tc_action_net *tn = net_generic(net, act_ct_ops.net_id);
1256 	bool bind = flags & TCA_ACT_FLAGS_BIND;
1257 	struct tcf_ct_params *params = NULL;
1258 	struct nlattr *tb[TCA_CT_MAX + 1];
1259 	struct tcf_chain *goto_ch = NULL;
1260 	struct tc_ct *parm;
1261 	struct tcf_ct *c;
1262 	int err, res = 0;
1263 	u32 index;
1264 
1265 	if (!nla) {
1266 		NL_SET_ERR_MSG_MOD(extack, "Ct requires attributes to be passed");
1267 		return -EINVAL;
1268 	}
1269 
1270 	err = nla_parse_nested(tb, TCA_CT_MAX, nla, ct_policy, extack);
1271 	if (err < 0)
1272 		return err;
1273 
1274 	if (!tb[TCA_CT_PARMS]) {
1275 		NL_SET_ERR_MSG_MOD(extack, "Missing required ct parameters");
1276 		return -EINVAL;
1277 	}
1278 	parm = nla_data(tb[TCA_CT_PARMS]);
1279 	index = parm->index;
1280 	err = tcf_idr_check_alloc(tn, &index, a, bind);
1281 	if (err < 0)
1282 		return err;
1283 
1284 	if (!err) {
1285 		err = tcf_idr_create_from_flags(tn, index, est, a,
1286 						&act_ct_ops, bind, flags);
1287 		if (err) {
1288 			tcf_idr_cleanup(tn, index);
1289 			return err;
1290 		}
1291 		res = ACT_P_CREATED;
1292 	} else {
1293 		if (bind)
1294 			return 0;
1295 
1296 		if (!(flags & TCA_ACT_FLAGS_REPLACE)) {
1297 			tcf_idr_release(*a, bind);
1298 			return -EEXIST;
1299 		}
1300 	}
1301 	err = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack);
1302 	if (err < 0)
1303 		goto cleanup;
1304 
1305 	c = to_ct(*a);
1306 
1307 	params = kzalloc(sizeof(*params), GFP_KERNEL);
1308 	if (unlikely(!params)) {
1309 		err = -ENOMEM;
1310 		goto cleanup;
1311 	}
1312 
1313 	err = tcf_ct_fill_params(net, params, parm, tb, extack);
1314 	if (err)
1315 		goto cleanup;
1316 
1317 	err = tcf_ct_flow_table_get(net, params);
1318 	if (err)
1319 		goto cleanup;
1320 
1321 	spin_lock_bh(&c->tcf_lock);
1322 	goto_ch = tcf_action_set_ctrlact(*a, parm->action, goto_ch);
1323 	params = rcu_replace_pointer(c->params, params,
1324 				     lockdep_is_held(&c->tcf_lock));
1325 	spin_unlock_bh(&c->tcf_lock);
1326 
1327 	if (goto_ch)
1328 		tcf_chain_put_by_act(goto_ch);
1329 	if (params)
1330 		call_rcu(&params->rcu, tcf_ct_params_free_rcu);
1331 
1332 	return res;
1333 
1334 cleanup:
1335 	if (goto_ch)
1336 		tcf_chain_put_by_act(goto_ch);
1337 	if (params)
1338 		tcf_ct_params_free(params);
1339 	tcf_idr_release(*a, bind);
1340 	return err;
1341 }
1342 
1343 static void tcf_ct_cleanup(struct tc_action *a)
1344 {
1345 	struct tcf_ct_params *params;
1346 	struct tcf_ct *c = to_ct(a);
1347 
1348 	params = rcu_dereference_protected(c->params, 1);
1349 	if (params)
1350 		call_rcu(&params->rcu, tcf_ct_params_free_rcu);
1351 }
1352 
1353 static int tcf_ct_dump_key_val(struct sk_buff *skb,
1354 			       void *val, int val_type,
1355 			       void *mask, int mask_type,
1356 			       int len)
1357 {
1358 	int err;
1359 
1360 	if (mask && !memchr_inv(mask, 0, len))
1361 		return 0;
1362 
1363 	err = nla_put(skb, val_type, len, val);
1364 	if (err)
1365 		return err;
1366 
1367 	if (mask_type != TCA_CT_UNSPEC) {
1368 		err = nla_put(skb, mask_type, len, mask);
1369 		if (err)
1370 			return err;
1371 	}
1372 
1373 	return 0;
1374 }
1375 
1376 static int tcf_ct_dump_nat(struct sk_buff *skb, struct tcf_ct_params *p)
1377 {
1378 	struct nf_nat_range2 *range = &p->range;
1379 
1380 	if (!(p->ct_action & TCA_CT_ACT_NAT))
1381 		return 0;
1382 
1383 	if (!(p->ct_action & (TCA_CT_ACT_NAT_SRC | TCA_CT_ACT_NAT_DST)))
1384 		return 0;
1385 
1386 	if (range->flags & NF_NAT_RANGE_MAP_IPS) {
1387 		if (p->ipv4_range) {
1388 			if (nla_put_in_addr(skb, TCA_CT_NAT_IPV4_MIN,
1389 					    range->min_addr.ip))
1390 				return -1;
1391 			if (nla_put_in_addr(skb, TCA_CT_NAT_IPV4_MAX,
1392 					    range->max_addr.ip))
1393 				return -1;
1394 		} else {
1395 			if (nla_put_in6_addr(skb, TCA_CT_NAT_IPV6_MIN,
1396 					     &range->min_addr.in6))
1397 				return -1;
1398 			if (nla_put_in6_addr(skb, TCA_CT_NAT_IPV6_MAX,
1399 					     &range->max_addr.in6))
1400 				return -1;
1401 		}
1402 	}
1403 
1404 	if (range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) {
1405 		if (nla_put_be16(skb, TCA_CT_NAT_PORT_MIN,
1406 				 range->min_proto.all))
1407 			return -1;
1408 		if (nla_put_be16(skb, TCA_CT_NAT_PORT_MAX,
1409 				 range->max_proto.all))
1410 			return -1;
1411 	}
1412 
1413 	return 0;
1414 }
1415 
1416 static int tcf_ct_dump_helper(struct sk_buff *skb, struct nf_conntrack_helper *helper)
1417 {
1418 	if (!helper)
1419 		return 0;
1420 
1421 	if (nla_put_string(skb, TCA_CT_HELPER_NAME, helper->name) ||
1422 	    nla_put_u8(skb, TCA_CT_HELPER_FAMILY, helper->tuple.src.l3num) ||
1423 	    nla_put_u8(skb, TCA_CT_HELPER_PROTO, helper->tuple.dst.protonum))
1424 		return -1;
1425 
1426 	return 0;
1427 }
1428 
1429 static inline int tcf_ct_dump(struct sk_buff *skb, struct tc_action *a,
1430 			      int bind, int ref)
1431 {
1432 	unsigned char *b = skb_tail_pointer(skb);
1433 	struct tcf_ct *c = to_ct(a);
1434 	struct tcf_ct_params *p;
1435 
1436 	struct tc_ct opt = {
1437 		.index   = c->tcf_index,
1438 		.refcnt  = refcount_read(&c->tcf_refcnt) - ref,
1439 		.bindcnt = atomic_read(&c->tcf_bindcnt) - bind,
1440 	};
1441 	struct tcf_t t;
1442 
1443 	spin_lock_bh(&c->tcf_lock);
1444 	p = rcu_dereference_protected(c->params,
1445 				      lockdep_is_held(&c->tcf_lock));
1446 	opt.action = c->tcf_action;
1447 
1448 	if (tcf_ct_dump_key_val(skb,
1449 				&p->ct_action, TCA_CT_ACTION,
1450 				NULL, TCA_CT_UNSPEC,
1451 				sizeof(p->ct_action)))
1452 		goto nla_put_failure;
1453 
1454 	if (p->ct_action & TCA_CT_ACT_CLEAR)
1455 		goto skip_dump;
1456 
1457 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1458 	    tcf_ct_dump_key_val(skb,
1459 				&p->mark, TCA_CT_MARK,
1460 				&p->mark_mask, TCA_CT_MARK_MASK,
1461 				sizeof(p->mark)))
1462 		goto nla_put_failure;
1463 
1464 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1465 	    tcf_ct_dump_key_val(skb,
1466 				p->labels, TCA_CT_LABELS,
1467 				p->labels_mask, TCA_CT_LABELS_MASK,
1468 				sizeof(p->labels)))
1469 		goto nla_put_failure;
1470 
1471 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1472 	    tcf_ct_dump_key_val(skb,
1473 				&p->zone, TCA_CT_ZONE,
1474 				NULL, TCA_CT_UNSPEC,
1475 				sizeof(p->zone)))
1476 		goto nla_put_failure;
1477 
1478 	if (tcf_ct_dump_nat(skb, p))
1479 		goto nla_put_failure;
1480 
1481 	if (tcf_ct_dump_helper(skb, p->helper))
1482 		goto nla_put_failure;
1483 
1484 skip_dump:
1485 	if (nla_put(skb, TCA_CT_PARMS, sizeof(opt), &opt))
1486 		goto nla_put_failure;
1487 
1488 	tcf_tm_dump(&t, &c->tcf_tm);
1489 	if (nla_put_64bit(skb, TCA_CT_TM, sizeof(t), &t, TCA_CT_PAD))
1490 		goto nla_put_failure;
1491 	spin_unlock_bh(&c->tcf_lock);
1492 
1493 	return skb->len;
1494 nla_put_failure:
1495 	spin_unlock_bh(&c->tcf_lock);
1496 	nlmsg_trim(skb, b);
1497 	return -1;
1498 }
1499 
1500 static void tcf_stats_update(struct tc_action *a, u64 bytes, u64 packets,
1501 			     u64 drops, u64 lastuse, bool hw)
1502 {
1503 	struct tcf_ct *c = to_ct(a);
1504 
1505 	tcf_action_update_stats(a, bytes, packets, drops, hw);
1506 	c->tcf_tm.lastuse = max_t(u64, c->tcf_tm.lastuse, lastuse);
1507 }
1508 
1509 static int tcf_ct_offload_act_setup(struct tc_action *act, void *entry_data,
1510 				    u32 *index_inc, bool bind,
1511 				    struct netlink_ext_ack *extack)
1512 {
1513 	if (bind) {
1514 		struct flow_action_entry *entry = entry_data;
1515 
1516 		entry->id = FLOW_ACTION_CT;
1517 		entry->ct.action = tcf_ct_action(act);
1518 		entry->ct.zone = tcf_ct_zone(act);
1519 		entry->ct.flow_table = tcf_ct_ft(act);
1520 		*index_inc = 1;
1521 	} else {
1522 		struct flow_offload_action *fl_action = entry_data;
1523 
1524 		fl_action->id = FLOW_ACTION_CT;
1525 	}
1526 
1527 	return 0;
1528 }
1529 
1530 static struct tc_action_ops act_ct_ops = {
1531 	.kind		=	"ct",
1532 	.id		=	TCA_ID_CT,
1533 	.owner		=	THIS_MODULE,
1534 	.act		=	tcf_ct_act,
1535 	.dump		=	tcf_ct_dump,
1536 	.init		=	tcf_ct_init,
1537 	.cleanup	=	tcf_ct_cleanup,
1538 	.stats_update	=	tcf_stats_update,
1539 	.offload_act_setup =	tcf_ct_offload_act_setup,
1540 	.size		=	sizeof(struct tcf_ct),
1541 };
1542 
1543 static __net_init int ct_init_net(struct net *net)
1544 {
1545 	unsigned int n_bits = sizeof_field(struct tcf_ct_params, labels) * 8;
1546 	struct tc_ct_action_net *tn = net_generic(net, act_ct_ops.net_id);
1547 
1548 	if (nf_connlabels_get(net, n_bits - 1)) {
1549 		tn->labels = false;
1550 		pr_err("act_ct: Failed to set connlabels length");
1551 	} else {
1552 		tn->labels = true;
1553 	}
1554 
1555 	return tc_action_net_init(net, &tn->tn, &act_ct_ops);
1556 }
1557 
1558 static void __net_exit ct_exit_net(struct list_head *net_list)
1559 {
1560 	struct net *net;
1561 
1562 	rtnl_lock();
1563 	list_for_each_entry(net, net_list, exit_list) {
1564 		struct tc_ct_action_net *tn = net_generic(net, act_ct_ops.net_id);
1565 
1566 		if (tn->labels)
1567 			nf_connlabels_put(net);
1568 	}
1569 	rtnl_unlock();
1570 
1571 	tc_action_net_exit(net_list, act_ct_ops.net_id);
1572 }
1573 
1574 static struct pernet_operations ct_net_ops = {
1575 	.init = ct_init_net,
1576 	.exit_batch = ct_exit_net,
1577 	.id   = &act_ct_ops.net_id,
1578 	.size = sizeof(struct tc_ct_action_net),
1579 };
1580 
1581 static int __init ct_init_module(void)
1582 {
1583 	int err;
1584 
1585 	act_ct_wq = alloc_ordered_workqueue("act_ct_workqueue", 0);
1586 	if (!act_ct_wq)
1587 		return -ENOMEM;
1588 
1589 	err = tcf_ct_flow_tables_init();
1590 	if (err)
1591 		goto err_tbl_init;
1592 
1593 	err = tcf_register_action(&act_ct_ops, &ct_net_ops);
1594 	if (err)
1595 		goto err_register;
1596 
1597 	static_branch_inc(&tcf_frag_xmit_count);
1598 
1599 	return 0;
1600 
1601 err_register:
1602 	tcf_ct_flow_tables_uninit();
1603 err_tbl_init:
1604 	destroy_workqueue(act_ct_wq);
1605 	return err;
1606 }
1607 
1608 static void __exit ct_cleanup_module(void)
1609 {
1610 	static_branch_dec(&tcf_frag_xmit_count);
1611 	tcf_unregister_action(&act_ct_ops, &ct_net_ops);
1612 	tcf_ct_flow_tables_uninit();
1613 	destroy_workqueue(act_ct_wq);
1614 }
1615 
1616 module_init(ct_init_module);
1617 module_exit(ct_cleanup_module);
1618 MODULE_AUTHOR("Paul Blakey <paulb@mellanox.com>");
1619 MODULE_AUTHOR("Yossi Kuperman <yossiku@mellanox.com>");
1620 MODULE_AUTHOR("Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>");
1621 MODULE_DESCRIPTION("Connection tracking action");
1622 MODULE_LICENSE("GPL v2");
1623