xref: /linux/net/sched/act_ct.c (revision c48a7c44a1d02516309015b6134c9bb982e17008)
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 bool tcf_ct_flow_is_outdated(const struct flow_offload *flow)
282 {
283 	return test_bit(IPS_SEEN_REPLY_BIT, &flow->ct->status) &&
284 	       test_bit(IPS_HW_OFFLOAD_BIT, &flow->ct->status) &&
285 	       !test_bit(NF_FLOW_HW_PENDING, &flow->flags) &&
286 	       !test_bit(NF_FLOW_HW_ESTABLISHED, &flow->flags);
287 }
288 
289 static struct nf_flowtable_type flowtable_ct = {
290 	.gc		= tcf_ct_flow_is_outdated,
291 	.action		= tcf_ct_flow_table_fill_actions,
292 	.owner		= THIS_MODULE,
293 };
294 
295 static int tcf_ct_flow_table_get(struct net *net, struct tcf_ct_params *params)
296 {
297 	struct tcf_ct_flow_table *ct_ft;
298 	int err = -ENOMEM;
299 
300 	mutex_lock(&zones_mutex);
301 	ct_ft = rhashtable_lookup_fast(&zones_ht, &params->zone, zones_params);
302 	if (ct_ft && refcount_inc_not_zero(&ct_ft->ref))
303 		goto out_unlock;
304 
305 	ct_ft = kzalloc(sizeof(*ct_ft), GFP_KERNEL);
306 	if (!ct_ft)
307 		goto err_alloc;
308 	refcount_set(&ct_ft->ref, 1);
309 
310 	ct_ft->zone = params->zone;
311 	err = rhashtable_insert_fast(&zones_ht, &ct_ft->node, zones_params);
312 	if (err)
313 		goto err_insert;
314 
315 	ct_ft->nf_ft.type = &flowtable_ct;
316 	ct_ft->nf_ft.flags |= NF_FLOWTABLE_HW_OFFLOAD |
317 			      NF_FLOWTABLE_COUNTER;
318 	err = nf_flow_table_init(&ct_ft->nf_ft);
319 	if (err)
320 		goto err_init;
321 	write_pnet(&ct_ft->nf_ft.net, net);
322 
323 	__module_get(THIS_MODULE);
324 out_unlock:
325 	params->ct_ft = ct_ft;
326 	params->nf_ft = &ct_ft->nf_ft;
327 	mutex_unlock(&zones_mutex);
328 
329 	return 0;
330 
331 err_init:
332 	rhashtable_remove_fast(&zones_ht, &ct_ft->node, zones_params);
333 err_insert:
334 	kfree(ct_ft);
335 err_alloc:
336 	mutex_unlock(&zones_mutex);
337 	return err;
338 }
339 
340 static void tcf_ct_flow_table_cleanup_work(struct work_struct *work)
341 {
342 	struct flow_block_cb *block_cb, *tmp_cb;
343 	struct tcf_ct_flow_table *ct_ft;
344 	struct flow_block *block;
345 
346 	ct_ft = container_of(to_rcu_work(work), struct tcf_ct_flow_table,
347 			     rwork);
348 	nf_flow_table_free(&ct_ft->nf_ft);
349 
350 	/* Remove any remaining callbacks before cleanup */
351 	block = &ct_ft->nf_ft.flow_block;
352 	down_write(&ct_ft->nf_ft.flow_block_lock);
353 	list_for_each_entry_safe(block_cb, tmp_cb, &block->cb_list, list) {
354 		list_del(&block_cb->list);
355 		flow_block_cb_free(block_cb);
356 	}
357 	up_write(&ct_ft->nf_ft.flow_block_lock);
358 	kfree(ct_ft);
359 
360 	module_put(THIS_MODULE);
361 }
362 
363 static void tcf_ct_flow_table_put(struct tcf_ct_flow_table *ct_ft)
364 {
365 	if (refcount_dec_and_test(&ct_ft->ref)) {
366 		rhashtable_remove_fast(&zones_ht, &ct_ft->node, zones_params);
367 		INIT_RCU_WORK(&ct_ft->rwork, tcf_ct_flow_table_cleanup_work);
368 		queue_rcu_work(act_ct_wq, &ct_ft->rwork);
369 	}
370 }
371 
372 static void tcf_ct_flow_tc_ifidx(struct flow_offload *entry,
373 				 struct nf_conn_act_ct_ext *act_ct_ext, u8 dir)
374 {
375 	entry->tuplehash[dir].tuple.xmit_type = FLOW_OFFLOAD_XMIT_TC;
376 	entry->tuplehash[dir].tuple.tc.iifidx = act_ct_ext->ifindex[dir];
377 }
378 
379 static void tcf_ct_flow_ct_ext_ifidx_update(struct flow_offload *entry)
380 {
381 	struct nf_conn_act_ct_ext *act_ct_ext;
382 
383 	act_ct_ext = nf_conn_act_ct_ext_find(entry->ct);
384 	if (act_ct_ext) {
385 		tcf_ct_flow_tc_ifidx(entry, act_ct_ext, FLOW_OFFLOAD_DIR_ORIGINAL);
386 		tcf_ct_flow_tc_ifidx(entry, act_ct_ext, FLOW_OFFLOAD_DIR_REPLY);
387 	}
388 }
389 
390 static void tcf_ct_flow_table_add(struct tcf_ct_flow_table *ct_ft,
391 				  struct nf_conn *ct,
392 				  bool tcp, bool bidirectional)
393 {
394 	struct nf_conn_act_ct_ext *act_ct_ext;
395 	struct flow_offload *entry;
396 	int err;
397 
398 	if (test_and_set_bit(IPS_OFFLOAD_BIT, &ct->status))
399 		return;
400 
401 	entry = flow_offload_alloc(ct);
402 	if (!entry) {
403 		WARN_ON_ONCE(1);
404 		goto err_alloc;
405 	}
406 
407 	if (tcp) {
408 		ct->proto.tcp.seen[0].flags |= IP_CT_TCP_FLAG_BE_LIBERAL;
409 		ct->proto.tcp.seen[1].flags |= IP_CT_TCP_FLAG_BE_LIBERAL;
410 	}
411 	if (bidirectional)
412 		__set_bit(NF_FLOW_HW_BIDIRECTIONAL, &entry->flags);
413 
414 	act_ct_ext = nf_conn_act_ct_ext_find(ct);
415 	if (act_ct_ext) {
416 		tcf_ct_flow_tc_ifidx(entry, act_ct_ext, FLOW_OFFLOAD_DIR_ORIGINAL);
417 		tcf_ct_flow_tc_ifidx(entry, act_ct_ext, FLOW_OFFLOAD_DIR_REPLY);
418 	}
419 
420 	err = flow_offload_add(&ct_ft->nf_ft, entry);
421 	if (err)
422 		goto err_add;
423 
424 	return;
425 
426 err_add:
427 	flow_offload_free(entry);
428 err_alloc:
429 	clear_bit(IPS_OFFLOAD_BIT, &ct->status);
430 }
431 
432 static void tcf_ct_flow_table_process_conn(struct tcf_ct_flow_table *ct_ft,
433 					   struct nf_conn *ct,
434 					   enum ip_conntrack_info ctinfo)
435 {
436 	bool tcp = false, bidirectional = true;
437 
438 	switch (nf_ct_protonum(ct)) {
439 	case IPPROTO_TCP:
440 		if ((ctinfo != IP_CT_ESTABLISHED &&
441 		     ctinfo != IP_CT_ESTABLISHED_REPLY) ||
442 		    !test_bit(IPS_ASSURED_BIT, &ct->status) ||
443 		    ct->proto.tcp.state != TCP_CONNTRACK_ESTABLISHED)
444 			return;
445 
446 		tcp = true;
447 		break;
448 	case IPPROTO_UDP:
449 		if (!nf_ct_is_confirmed(ct))
450 			return;
451 		if (!test_bit(IPS_ASSURED_BIT, &ct->status))
452 			bidirectional = false;
453 		break;
454 #ifdef CONFIG_NF_CT_PROTO_GRE
455 	case IPPROTO_GRE: {
456 		struct nf_conntrack_tuple *tuple;
457 
458 		if ((ctinfo != IP_CT_ESTABLISHED &&
459 		     ctinfo != IP_CT_ESTABLISHED_REPLY) ||
460 		    !test_bit(IPS_ASSURED_BIT, &ct->status) ||
461 		    ct->status & IPS_NAT_MASK)
462 			return;
463 
464 		tuple = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
465 		/* No support for GRE v1 */
466 		if (tuple->src.u.gre.key || tuple->dst.u.gre.key)
467 			return;
468 		break;
469 	}
470 #endif
471 	default:
472 		return;
473 	}
474 
475 	if (nf_ct_ext_exist(ct, NF_CT_EXT_HELPER) ||
476 	    ct->status & IPS_SEQ_ADJUST)
477 		return;
478 
479 	tcf_ct_flow_table_add(ct_ft, ct, tcp, bidirectional);
480 }
481 
482 static bool
483 tcf_ct_flow_table_fill_tuple_ipv4(struct sk_buff *skb,
484 				  struct flow_offload_tuple *tuple,
485 				  struct tcphdr **tcph)
486 {
487 	struct flow_ports *ports;
488 	unsigned int thoff;
489 	struct iphdr *iph;
490 	size_t hdrsize;
491 	u8 ipproto;
492 
493 	if (!pskb_network_may_pull(skb, sizeof(*iph)))
494 		return false;
495 
496 	iph = ip_hdr(skb);
497 	thoff = iph->ihl * 4;
498 
499 	if (ip_is_fragment(iph) ||
500 	    unlikely(thoff != sizeof(struct iphdr)))
501 		return false;
502 
503 	ipproto = iph->protocol;
504 	switch (ipproto) {
505 	case IPPROTO_TCP:
506 		hdrsize = sizeof(struct tcphdr);
507 		break;
508 	case IPPROTO_UDP:
509 		hdrsize = sizeof(*ports);
510 		break;
511 #ifdef CONFIG_NF_CT_PROTO_GRE
512 	case IPPROTO_GRE:
513 		hdrsize = sizeof(struct gre_base_hdr);
514 		break;
515 #endif
516 	default:
517 		return false;
518 	}
519 
520 	if (iph->ttl <= 1)
521 		return false;
522 
523 	if (!pskb_network_may_pull(skb, thoff + hdrsize))
524 		return false;
525 
526 	switch (ipproto) {
527 	case IPPROTO_TCP:
528 		*tcph = (void *)(skb_network_header(skb) + thoff);
529 		fallthrough;
530 	case IPPROTO_UDP:
531 		ports = (struct flow_ports *)(skb_network_header(skb) + thoff);
532 		tuple->src_port = ports->source;
533 		tuple->dst_port = ports->dest;
534 		break;
535 	case IPPROTO_GRE: {
536 		struct gre_base_hdr *greh;
537 
538 		greh = (struct gre_base_hdr *)(skb_network_header(skb) + thoff);
539 		if ((greh->flags & GRE_VERSION) != GRE_VERSION_0)
540 			return false;
541 		break;
542 	}
543 	}
544 
545 	iph = ip_hdr(skb);
546 
547 	tuple->src_v4.s_addr = iph->saddr;
548 	tuple->dst_v4.s_addr = iph->daddr;
549 	tuple->l3proto = AF_INET;
550 	tuple->l4proto = ipproto;
551 
552 	return true;
553 }
554 
555 static bool
556 tcf_ct_flow_table_fill_tuple_ipv6(struct sk_buff *skb,
557 				  struct flow_offload_tuple *tuple,
558 				  struct tcphdr **tcph)
559 {
560 	struct flow_ports *ports;
561 	struct ipv6hdr *ip6h;
562 	unsigned int thoff;
563 	size_t hdrsize;
564 	u8 nexthdr;
565 
566 	if (!pskb_network_may_pull(skb, sizeof(*ip6h)))
567 		return false;
568 
569 	ip6h = ipv6_hdr(skb);
570 	thoff = sizeof(*ip6h);
571 
572 	nexthdr = ip6h->nexthdr;
573 	switch (nexthdr) {
574 	case IPPROTO_TCP:
575 		hdrsize = sizeof(struct tcphdr);
576 		break;
577 	case IPPROTO_UDP:
578 		hdrsize = sizeof(*ports);
579 		break;
580 #ifdef CONFIG_NF_CT_PROTO_GRE
581 	case IPPROTO_GRE:
582 		hdrsize = sizeof(struct gre_base_hdr);
583 		break;
584 #endif
585 	default:
586 		return false;
587 	}
588 
589 	if (ip6h->hop_limit <= 1)
590 		return false;
591 
592 	if (!pskb_network_may_pull(skb, thoff + hdrsize))
593 		return false;
594 
595 	switch (nexthdr) {
596 	case IPPROTO_TCP:
597 		*tcph = (void *)(skb_network_header(skb) + thoff);
598 		fallthrough;
599 	case IPPROTO_UDP:
600 		ports = (struct flow_ports *)(skb_network_header(skb) + thoff);
601 		tuple->src_port = ports->source;
602 		tuple->dst_port = ports->dest;
603 		break;
604 	case IPPROTO_GRE: {
605 		struct gre_base_hdr *greh;
606 
607 		greh = (struct gre_base_hdr *)(skb_network_header(skb) + thoff);
608 		if ((greh->flags & GRE_VERSION) != GRE_VERSION_0)
609 			return false;
610 		break;
611 	}
612 	}
613 
614 	ip6h = ipv6_hdr(skb);
615 
616 	tuple->src_v6 = ip6h->saddr;
617 	tuple->dst_v6 = ip6h->daddr;
618 	tuple->l3proto = AF_INET6;
619 	tuple->l4proto = nexthdr;
620 
621 	return true;
622 }
623 
624 static bool tcf_ct_flow_table_lookup(struct tcf_ct_params *p,
625 				     struct sk_buff *skb,
626 				     u8 family)
627 {
628 	struct nf_flowtable *nf_ft = &p->ct_ft->nf_ft;
629 	struct flow_offload_tuple_rhash *tuplehash;
630 	struct flow_offload_tuple tuple = {};
631 	enum ip_conntrack_info ctinfo;
632 	struct tcphdr *tcph = NULL;
633 	bool force_refresh = false;
634 	struct flow_offload *flow;
635 	struct nf_conn *ct;
636 	u8 dir;
637 
638 	switch (family) {
639 	case NFPROTO_IPV4:
640 		if (!tcf_ct_flow_table_fill_tuple_ipv4(skb, &tuple, &tcph))
641 			return false;
642 		break;
643 	case NFPROTO_IPV6:
644 		if (!tcf_ct_flow_table_fill_tuple_ipv6(skb, &tuple, &tcph))
645 			return false;
646 		break;
647 	default:
648 		return false;
649 	}
650 
651 	tuplehash = flow_offload_lookup(nf_ft, &tuple);
652 	if (!tuplehash)
653 		return false;
654 
655 	dir = tuplehash->tuple.dir;
656 	flow = container_of(tuplehash, struct flow_offload, tuplehash[dir]);
657 	ct = flow->ct;
658 
659 	if (dir == FLOW_OFFLOAD_DIR_REPLY &&
660 	    !test_bit(NF_FLOW_HW_BIDIRECTIONAL, &flow->flags)) {
661 		/* Only offload reply direction after connection became
662 		 * assured.
663 		 */
664 		if (test_bit(IPS_ASSURED_BIT, &ct->status))
665 			set_bit(NF_FLOW_HW_BIDIRECTIONAL, &flow->flags);
666 		else if (test_bit(NF_FLOW_HW_ESTABLISHED, &flow->flags))
667 			/* If flow_table flow has already been updated to the
668 			 * established state, then don't refresh.
669 			 */
670 			return false;
671 		force_refresh = true;
672 	}
673 
674 	if (tcph && (unlikely(tcph->fin || tcph->rst))) {
675 		flow_offload_teardown(flow);
676 		return false;
677 	}
678 
679 	if (dir == FLOW_OFFLOAD_DIR_ORIGINAL)
680 		ctinfo = test_bit(IPS_SEEN_REPLY_BIT, &ct->status) ?
681 			IP_CT_ESTABLISHED : IP_CT_NEW;
682 	else
683 		ctinfo = IP_CT_ESTABLISHED_REPLY;
684 
685 	nf_conn_act_ct_ext_fill(skb, ct, ctinfo);
686 	tcf_ct_flow_ct_ext_ifidx_update(flow);
687 	flow_offload_refresh(nf_ft, flow, force_refresh);
688 	if (!test_bit(IPS_ASSURED_BIT, &ct->status)) {
689 		/* Process this flow in SW to allow promoting to ASSURED */
690 		return false;
691 	}
692 
693 	nf_conntrack_get(&ct->ct_general);
694 	nf_ct_set(skb, ct, ctinfo);
695 	if (nf_ft->flags & NF_FLOWTABLE_COUNTER)
696 		nf_ct_acct_update(ct, dir, skb->len);
697 
698 	return true;
699 }
700 
701 static int tcf_ct_flow_tables_init(void)
702 {
703 	return rhashtable_init(&zones_ht, &zones_params);
704 }
705 
706 static void tcf_ct_flow_tables_uninit(void)
707 {
708 	rhashtable_destroy(&zones_ht);
709 }
710 
711 static struct tc_action_ops act_ct_ops;
712 
713 struct tc_ct_action_net {
714 	struct tc_action_net tn; /* Must be first */
715 };
716 
717 /* Determine whether skb->_nfct is equal to the result of conntrack lookup. */
718 static bool tcf_ct_skb_nfct_cached(struct net *net, struct sk_buff *skb,
719 				   struct tcf_ct_params *p)
720 {
721 	enum ip_conntrack_info ctinfo;
722 	struct nf_conn *ct;
723 
724 	ct = nf_ct_get(skb, &ctinfo);
725 	if (!ct)
726 		return false;
727 	if (!net_eq(net, read_pnet(&ct->ct_net)))
728 		goto drop_ct;
729 	if (nf_ct_zone(ct)->id != p->zone)
730 		goto drop_ct;
731 	if (p->helper) {
732 		struct nf_conn_help *help;
733 
734 		help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER);
735 		if (help && rcu_access_pointer(help->helper) != p->helper)
736 			goto drop_ct;
737 	}
738 
739 	/* Force conntrack entry direction. */
740 	if ((p->ct_action & TCA_CT_ACT_FORCE) &&
741 	    CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) {
742 		if (nf_ct_is_confirmed(ct))
743 			nf_ct_kill(ct);
744 
745 		goto drop_ct;
746 	}
747 
748 	return true;
749 
750 drop_ct:
751 	nf_ct_put(ct);
752 	nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
753 
754 	return false;
755 }
756 
757 static u8 tcf_ct_skb_nf_family(struct sk_buff *skb)
758 {
759 	u8 family = NFPROTO_UNSPEC;
760 
761 	switch (skb_protocol(skb, true)) {
762 	case htons(ETH_P_IP):
763 		family = NFPROTO_IPV4;
764 		break;
765 	case htons(ETH_P_IPV6):
766 		family = NFPROTO_IPV6;
767 		break;
768 	default:
769 		break;
770 	}
771 
772 	return family;
773 }
774 
775 static int tcf_ct_ipv4_is_fragment(struct sk_buff *skb, bool *frag)
776 {
777 	unsigned int len;
778 
779 	len =  skb_network_offset(skb) + sizeof(struct iphdr);
780 	if (unlikely(skb->len < len))
781 		return -EINVAL;
782 	if (unlikely(!pskb_may_pull(skb, len)))
783 		return -ENOMEM;
784 
785 	*frag = ip_is_fragment(ip_hdr(skb));
786 	return 0;
787 }
788 
789 static int tcf_ct_ipv6_is_fragment(struct sk_buff *skb, bool *frag)
790 {
791 	unsigned int flags = 0, len, payload_ofs = 0;
792 	unsigned short frag_off;
793 	int nexthdr;
794 
795 	len =  skb_network_offset(skb) + sizeof(struct ipv6hdr);
796 	if (unlikely(skb->len < len))
797 		return -EINVAL;
798 	if (unlikely(!pskb_may_pull(skb, len)))
799 		return -ENOMEM;
800 
801 	nexthdr = ipv6_find_hdr(skb, &payload_ofs, -1, &frag_off, &flags);
802 	if (unlikely(nexthdr < 0))
803 		return -EPROTO;
804 
805 	*frag = flags & IP6_FH_F_FRAG;
806 	return 0;
807 }
808 
809 static int tcf_ct_handle_fragments(struct net *net, struct sk_buff *skb,
810 				   u8 family, u16 zone, bool *defrag)
811 {
812 	enum ip_conntrack_info ctinfo;
813 	struct nf_conn *ct;
814 	int err = 0;
815 	bool frag;
816 	u8 proto;
817 	u16 mru;
818 
819 	/* Previously seen (loopback)? Ignore. */
820 	ct = nf_ct_get(skb, &ctinfo);
821 	if ((ct && !nf_ct_is_template(ct)) || ctinfo == IP_CT_UNTRACKED)
822 		return 0;
823 
824 	if (family == NFPROTO_IPV4)
825 		err = tcf_ct_ipv4_is_fragment(skb, &frag);
826 	else
827 		err = tcf_ct_ipv6_is_fragment(skb, &frag);
828 	if (err || !frag)
829 		return err;
830 
831 	skb_get(skb);
832 	err = nf_ct_handle_fragments(net, skb, zone, family, &proto, &mru);
833 	if (err)
834 		return err;
835 
836 	*defrag = true;
837 	tc_skb_cb(skb)->mru = mru;
838 
839 	return 0;
840 }
841 
842 static void tcf_ct_params_free(struct tcf_ct_params *params)
843 {
844 	if (params->helper) {
845 #if IS_ENABLED(CONFIG_NF_NAT)
846 		if (params->ct_action & TCA_CT_ACT_NAT)
847 			nf_nat_helper_put(params->helper);
848 #endif
849 		nf_conntrack_helper_put(params->helper);
850 	}
851 	if (params->ct_ft)
852 		tcf_ct_flow_table_put(params->ct_ft);
853 	if (params->tmpl) {
854 		if (params->put_labels)
855 			nf_connlabels_put(nf_ct_net(params->tmpl));
856 
857 		nf_ct_put(params->tmpl);
858 	}
859 
860 	kfree(params);
861 }
862 
863 static void tcf_ct_params_free_rcu(struct rcu_head *head)
864 {
865 	struct tcf_ct_params *params;
866 
867 	params = container_of(head, struct tcf_ct_params, rcu);
868 	tcf_ct_params_free(params);
869 }
870 
871 static void tcf_ct_act_set_mark(struct nf_conn *ct, u32 mark, u32 mask)
872 {
873 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
874 	u32 new_mark;
875 
876 	if (!mask)
877 		return;
878 
879 	new_mark = mark | (READ_ONCE(ct->mark) & ~(mask));
880 	if (READ_ONCE(ct->mark) != new_mark) {
881 		WRITE_ONCE(ct->mark, new_mark);
882 		if (nf_ct_is_confirmed(ct))
883 			nf_conntrack_event_cache(IPCT_MARK, ct);
884 	}
885 #endif
886 }
887 
888 static void tcf_ct_act_set_labels(struct nf_conn *ct,
889 				  u32 *labels,
890 				  u32 *labels_m)
891 {
892 #if IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS)
893 	size_t labels_sz = sizeof_field(struct tcf_ct_params, labels);
894 
895 	if (!memchr_inv(labels_m, 0, labels_sz))
896 		return;
897 
898 	nf_connlabels_replace(ct, labels, labels_m, 4);
899 #endif
900 }
901 
902 static int tcf_ct_act_nat(struct sk_buff *skb,
903 			  struct nf_conn *ct,
904 			  enum ip_conntrack_info ctinfo,
905 			  int ct_action,
906 			  struct nf_nat_range2 *range,
907 			  bool commit)
908 {
909 #if IS_ENABLED(CONFIG_NF_NAT)
910 	int err, action = 0;
911 
912 	if (!(ct_action & TCA_CT_ACT_NAT))
913 		return NF_ACCEPT;
914 	if (ct_action & TCA_CT_ACT_NAT_SRC)
915 		action |= BIT(NF_NAT_MANIP_SRC);
916 	if (ct_action & TCA_CT_ACT_NAT_DST)
917 		action |= BIT(NF_NAT_MANIP_DST);
918 
919 	err = nf_ct_nat(skb, ct, ctinfo, &action, range, commit);
920 
921 	if (action & BIT(NF_NAT_MANIP_SRC))
922 		tc_skb_cb(skb)->post_ct_snat = 1;
923 	if (action & BIT(NF_NAT_MANIP_DST))
924 		tc_skb_cb(skb)->post_ct_dnat = 1;
925 
926 	return err;
927 #else
928 	return NF_ACCEPT;
929 #endif
930 }
931 
932 TC_INDIRECT_SCOPE int tcf_ct_act(struct sk_buff *skb, const struct tc_action *a,
933 				 struct tcf_result *res)
934 {
935 	struct net *net = dev_net(skb->dev);
936 	enum ip_conntrack_info ctinfo;
937 	struct tcf_ct *c = to_ct(a);
938 	struct nf_conn *tmpl = NULL;
939 	struct nf_hook_state state;
940 	bool cached, commit, clear;
941 	int nh_ofs, err, retval;
942 	struct tcf_ct_params *p;
943 	bool add_helper = false;
944 	bool skip_add = false;
945 	bool defrag = false;
946 	struct nf_conn *ct;
947 	u8 family;
948 
949 	p = rcu_dereference_bh(c->params);
950 
951 	retval = READ_ONCE(c->tcf_action);
952 	commit = p->ct_action & TCA_CT_ACT_COMMIT;
953 	clear = p->ct_action & TCA_CT_ACT_CLEAR;
954 	tmpl = p->tmpl;
955 
956 	tcf_lastuse_update(&c->tcf_tm);
957 	tcf_action_update_bstats(&c->common, skb);
958 
959 	if (clear) {
960 		tc_skb_cb(skb)->post_ct = false;
961 		ct = nf_ct_get(skb, &ctinfo);
962 		if (ct) {
963 			nf_ct_put(ct);
964 			nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
965 		}
966 
967 		goto out_clear;
968 	}
969 
970 	family = tcf_ct_skb_nf_family(skb);
971 	if (family == NFPROTO_UNSPEC)
972 		goto drop;
973 
974 	/* The conntrack module expects to be working at L3.
975 	 * We also try to pull the IPv4/6 header to linear area
976 	 */
977 	nh_ofs = skb_network_offset(skb);
978 	skb_pull_rcsum(skb, nh_ofs);
979 	err = tcf_ct_handle_fragments(net, skb, family, p->zone, &defrag);
980 	if (err == -EINPROGRESS) {
981 		retval = TC_ACT_STOLEN;
982 		goto out_clear;
983 	}
984 	if (err)
985 		goto drop;
986 
987 	err = nf_ct_skb_network_trim(skb, family);
988 	if (err)
989 		goto drop;
990 
991 	/* If we are recirculating packets to match on ct fields and
992 	 * committing with a separate ct action, then we don't need to
993 	 * actually run the packet through conntrack twice unless it's for a
994 	 * different zone.
995 	 */
996 	cached = tcf_ct_skb_nfct_cached(net, skb, p);
997 	if (!cached) {
998 		if (tcf_ct_flow_table_lookup(p, skb, family)) {
999 			skip_add = true;
1000 			goto do_nat;
1001 		}
1002 
1003 		/* Associate skb with specified zone. */
1004 		if (tmpl) {
1005 			nf_conntrack_put(skb_nfct(skb));
1006 			nf_conntrack_get(&tmpl->ct_general);
1007 			nf_ct_set(skb, tmpl, IP_CT_NEW);
1008 		}
1009 
1010 		state.hook = NF_INET_PRE_ROUTING;
1011 		state.net = net;
1012 		state.pf = family;
1013 		err = nf_conntrack_in(skb, &state);
1014 		if (err != NF_ACCEPT)
1015 			goto out_push;
1016 	}
1017 
1018 do_nat:
1019 	ct = nf_ct_get(skb, &ctinfo);
1020 	if (!ct)
1021 		goto out_push;
1022 	nf_ct_deliver_cached_events(ct);
1023 	nf_conn_act_ct_ext_fill(skb, ct, ctinfo);
1024 
1025 	err = tcf_ct_act_nat(skb, ct, ctinfo, p->ct_action, &p->range, commit);
1026 	if (err != NF_ACCEPT)
1027 		goto drop;
1028 
1029 	if (!nf_ct_is_confirmed(ct) && commit && p->helper && !nfct_help(ct)) {
1030 		err = __nf_ct_try_assign_helper(ct, p->tmpl, GFP_ATOMIC);
1031 		if (err)
1032 			goto drop;
1033 		add_helper = true;
1034 		if (p->ct_action & TCA_CT_ACT_NAT && !nfct_seqadj(ct)) {
1035 			if (!nfct_seqadj_ext_add(ct))
1036 				goto drop;
1037 		}
1038 	}
1039 
1040 	if (nf_ct_is_confirmed(ct) ? ((!cached && !skip_add) || add_helper) : commit) {
1041 		if (nf_ct_helper(skb, ct, ctinfo, family) != NF_ACCEPT)
1042 			goto drop;
1043 	}
1044 
1045 	if (commit) {
1046 		tcf_ct_act_set_mark(ct, p->mark, p->mark_mask);
1047 		tcf_ct_act_set_labels(ct, p->labels, p->labels_mask);
1048 
1049 		if (!nf_ct_is_confirmed(ct))
1050 			nf_conn_act_ct_ext_add(skb, ct, ctinfo);
1051 
1052 		/* This will take care of sending queued events
1053 		 * even if the connection is already confirmed.
1054 		 */
1055 		if (nf_conntrack_confirm(skb) != NF_ACCEPT)
1056 			goto drop;
1057 	}
1058 
1059 	if (!skip_add)
1060 		tcf_ct_flow_table_process_conn(p->ct_ft, ct, ctinfo);
1061 
1062 out_push:
1063 	skb_push_rcsum(skb, nh_ofs);
1064 
1065 	tc_skb_cb(skb)->post_ct = true;
1066 	tc_skb_cb(skb)->zone = p->zone;
1067 out_clear:
1068 	if (defrag)
1069 		qdisc_skb_cb(skb)->pkt_len = skb->len;
1070 	return retval;
1071 
1072 drop:
1073 	tcf_action_inc_drop_qstats(&c->common);
1074 	return TC_ACT_SHOT;
1075 }
1076 
1077 static const struct nla_policy ct_policy[TCA_CT_MAX + 1] = {
1078 	[TCA_CT_ACTION] = { .type = NLA_U16 },
1079 	[TCA_CT_PARMS] = NLA_POLICY_EXACT_LEN(sizeof(struct tc_ct)),
1080 	[TCA_CT_ZONE] = { .type = NLA_U16 },
1081 	[TCA_CT_MARK] = { .type = NLA_U32 },
1082 	[TCA_CT_MARK_MASK] = { .type = NLA_U32 },
1083 	[TCA_CT_LABELS] = { .type = NLA_BINARY,
1084 			    .len = 128 / BITS_PER_BYTE },
1085 	[TCA_CT_LABELS_MASK] = { .type = NLA_BINARY,
1086 				 .len = 128 / BITS_PER_BYTE },
1087 	[TCA_CT_NAT_IPV4_MIN] = { .type = NLA_U32 },
1088 	[TCA_CT_NAT_IPV4_MAX] = { .type = NLA_U32 },
1089 	[TCA_CT_NAT_IPV6_MIN] = NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)),
1090 	[TCA_CT_NAT_IPV6_MAX] = NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)),
1091 	[TCA_CT_NAT_PORT_MIN] = { .type = NLA_U16 },
1092 	[TCA_CT_NAT_PORT_MAX] = { .type = NLA_U16 },
1093 	[TCA_CT_HELPER_NAME] = { .type = NLA_STRING, .len = NF_CT_HELPER_NAME_LEN },
1094 	[TCA_CT_HELPER_FAMILY] = { .type = NLA_U8 },
1095 	[TCA_CT_HELPER_PROTO] = { .type = NLA_U8 },
1096 };
1097 
1098 static int tcf_ct_fill_params_nat(struct tcf_ct_params *p,
1099 				  struct tc_ct *parm,
1100 				  struct nlattr **tb,
1101 				  struct netlink_ext_ack *extack)
1102 {
1103 	struct nf_nat_range2 *range;
1104 
1105 	if (!(p->ct_action & TCA_CT_ACT_NAT))
1106 		return 0;
1107 
1108 	if (!IS_ENABLED(CONFIG_NF_NAT)) {
1109 		NL_SET_ERR_MSG_MOD(extack, "Netfilter nat isn't enabled in kernel");
1110 		return -EOPNOTSUPP;
1111 	}
1112 
1113 	if (!(p->ct_action & (TCA_CT_ACT_NAT_SRC | TCA_CT_ACT_NAT_DST)))
1114 		return 0;
1115 
1116 	if ((p->ct_action & TCA_CT_ACT_NAT_SRC) &&
1117 	    (p->ct_action & TCA_CT_ACT_NAT_DST)) {
1118 		NL_SET_ERR_MSG_MOD(extack, "dnat and snat can't be enabled at the same time");
1119 		return -EOPNOTSUPP;
1120 	}
1121 
1122 	range = &p->range;
1123 	if (tb[TCA_CT_NAT_IPV4_MIN]) {
1124 		struct nlattr *max_attr = tb[TCA_CT_NAT_IPV4_MAX];
1125 
1126 		p->ipv4_range = true;
1127 		range->flags |= NF_NAT_RANGE_MAP_IPS;
1128 		range->min_addr.ip =
1129 			nla_get_in_addr(tb[TCA_CT_NAT_IPV4_MIN]);
1130 
1131 		range->max_addr.ip = max_attr ?
1132 				     nla_get_in_addr(max_attr) :
1133 				     range->min_addr.ip;
1134 	} else if (tb[TCA_CT_NAT_IPV6_MIN]) {
1135 		struct nlattr *max_attr = tb[TCA_CT_NAT_IPV6_MAX];
1136 
1137 		p->ipv4_range = false;
1138 		range->flags |= NF_NAT_RANGE_MAP_IPS;
1139 		range->min_addr.in6 =
1140 			nla_get_in6_addr(tb[TCA_CT_NAT_IPV6_MIN]);
1141 
1142 		range->max_addr.in6 = max_attr ?
1143 				      nla_get_in6_addr(max_attr) :
1144 				      range->min_addr.in6;
1145 	}
1146 
1147 	if (tb[TCA_CT_NAT_PORT_MIN]) {
1148 		range->flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1149 		range->min_proto.all = nla_get_be16(tb[TCA_CT_NAT_PORT_MIN]);
1150 
1151 		range->max_proto.all = tb[TCA_CT_NAT_PORT_MAX] ?
1152 				       nla_get_be16(tb[TCA_CT_NAT_PORT_MAX]) :
1153 				       range->min_proto.all;
1154 	}
1155 
1156 	return 0;
1157 }
1158 
1159 static void tcf_ct_set_key_val(struct nlattr **tb,
1160 			       void *val, int val_type,
1161 			       void *mask, int mask_type,
1162 			       int len)
1163 {
1164 	if (!tb[val_type])
1165 		return;
1166 	nla_memcpy(val, tb[val_type], len);
1167 
1168 	if (!mask)
1169 		return;
1170 
1171 	if (mask_type == TCA_CT_UNSPEC || !tb[mask_type])
1172 		memset(mask, 0xff, len);
1173 	else
1174 		nla_memcpy(mask, tb[mask_type], len);
1175 }
1176 
1177 static int tcf_ct_fill_params(struct net *net,
1178 			      struct tcf_ct_params *p,
1179 			      struct tc_ct *parm,
1180 			      struct nlattr **tb,
1181 			      struct netlink_ext_ack *extack)
1182 {
1183 	struct nf_conntrack_zone zone;
1184 	int err, family, proto, len;
1185 	bool put_labels = false;
1186 	struct nf_conn *tmpl;
1187 	char *name;
1188 
1189 	p->zone = NF_CT_DEFAULT_ZONE_ID;
1190 
1191 	tcf_ct_set_key_val(tb,
1192 			   &p->ct_action, TCA_CT_ACTION,
1193 			   NULL, TCA_CT_UNSPEC,
1194 			   sizeof(p->ct_action));
1195 
1196 	if (p->ct_action & TCA_CT_ACT_CLEAR)
1197 		return 0;
1198 
1199 	err = tcf_ct_fill_params_nat(p, parm, tb, extack);
1200 	if (err)
1201 		return err;
1202 
1203 	if (tb[TCA_CT_MARK]) {
1204 		if (!IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)) {
1205 			NL_SET_ERR_MSG_MOD(extack, "Conntrack mark isn't enabled.");
1206 			return -EOPNOTSUPP;
1207 		}
1208 		tcf_ct_set_key_val(tb,
1209 				   &p->mark, TCA_CT_MARK,
1210 				   &p->mark_mask, TCA_CT_MARK_MASK,
1211 				   sizeof(p->mark));
1212 	}
1213 
1214 	if (tb[TCA_CT_LABELS]) {
1215 		unsigned int n_bits = sizeof_field(struct tcf_ct_params, labels) * 8;
1216 
1217 		if (!IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS)) {
1218 			NL_SET_ERR_MSG_MOD(extack, "Conntrack labels isn't enabled.");
1219 			return -EOPNOTSUPP;
1220 		}
1221 
1222 		if (nf_connlabels_get(net, n_bits - 1)) {
1223 			NL_SET_ERR_MSG_MOD(extack, "Failed to set connlabel length");
1224 			return -EOPNOTSUPP;
1225 		} else {
1226 			put_labels = true;
1227 		}
1228 
1229 		tcf_ct_set_key_val(tb,
1230 				   p->labels, TCA_CT_LABELS,
1231 				   p->labels_mask, TCA_CT_LABELS_MASK,
1232 				   sizeof(p->labels));
1233 	}
1234 
1235 	if (tb[TCA_CT_ZONE]) {
1236 		if (!IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES)) {
1237 			NL_SET_ERR_MSG_MOD(extack, "Conntrack zones isn't enabled.");
1238 			return -EOPNOTSUPP;
1239 		}
1240 
1241 		tcf_ct_set_key_val(tb,
1242 				   &p->zone, TCA_CT_ZONE,
1243 				   NULL, TCA_CT_UNSPEC,
1244 				   sizeof(p->zone));
1245 	}
1246 
1247 	nf_ct_zone_init(&zone, p->zone, NF_CT_DEFAULT_ZONE_DIR, 0);
1248 	tmpl = nf_ct_tmpl_alloc(net, &zone, GFP_KERNEL);
1249 	if (!tmpl) {
1250 		NL_SET_ERR_MSG_MOD(extack, "Failed to allocate conntrack template");
1251 		return -ENOMEM;
1252 	}
1253 	p->tmpl = tmpl;
1254 	if (tb[TCA_CT_HELPER_NAME]) {
1255 		name = nla_data(tb[TCA_CT_HELPER_NAME]);
1256 		len = nla_len(tb[TCA_CT_HELPER_NAME]);
1257 		if (len > 16 || name[len - 1] != '\0') {
1258 			NL_SET_ERR_MSG_MOD(extack, "Failed to parse helper name.");
1259 			err = -EINVAL;
1260 			goto err;
1261 		}
1262 		family = tb[TCA_CT_HELPER_FAMILY] ? nla_get_u8(tb[TCA_CT_HELPER_FAMILY]) : AF_INET;
1263 		proto = tb[TCA_CT_HELPER_PROTO] ? nla_get_u8(tb[TCA_CT_HELPER_PROTO]) : IPPROTO_TCP;
1264 		err = nf_ct_add_helper(tmpl, name, family, proto,
1265 				       p->ct_action & TCA_CT_ACT_NAT, &p->helper);
1266 		if (err) {
1267 			NL_SET_ERR_MSG_MOD(extack, "Failed to add helper");
1268 			goto err;
1269 		}
1270 	}
1271 
1272 	p->put_labels = put_labels;
1273 
1274 	if (p->ct_action & TCA_CT_ACT_COMMIT)
1275 		__set_bit(IPS_CONFIRMED_BIT, &tmpl->status);
1276 	return 0;
1277 err:
1278 	if (put_labels)
1279 		nf_connlabels_put(net);
1280 
1281 	nf_ct_put(p->tmpl);
1282 	p->tmpl = NULL;
1283 	return err;
1284 }
1285 
1286 static int tcf_ct_init(struct net *net, struct nlattr *nla,
1287 		       struct nlattr *est, struct tc_action **a,
1288 		       struct tcf_proto *tp, u32 flags,
1289 		       struct netlink_ext_ack *extack)
1290 {
1291 	struct tc_action_net *tn = net_generic(net, act_ct_ops.net_id);
1292 	bool bind = flags & TCA_ACT_FLAGS_BIND;
1293 	struct tcf_ct_params *params = NULL;
1294 	struct nlattr *tb[TCA_CT_MAX + 1];
1295 	struct tcf_chain *goto_ch = NULL;
1296 	struct tc_ct *parm;
1297 	struct tcf_ct *c;
1298 	int err, res = 0;
1299 	u32 index;
1300 
1301 	if (!nla) {
1302 		NL_SET_ERR_MSG_MOD(extack, "Ct requires attributes to be passed");
1303 		return -EINVAL;
1304 	}
1305 
1306 	err = nla_parse_nested(tb, TCA_CT_MAX, nla, ct_policy, extack);
1307 	if (err < 0)
1308 		return err;
1309 
1310 	if (!tb[TCA_CT_PARMS]) {
1311 		NL_SET_ERR_MSG_MOD(extack, "Missing required ct parameters");
1312 		return -EINVAL;
1313 	}
1314 	parm = nla_data(tb[TCA_CT_PARMS]);
1315 	index = parm->index;
1316 	err = tcf_idr_check_alloc(tn, &index, a, bind);
1317 	if (err < 0)
1318 		return err;
1319 
1320 	if (!err) {
1321 		err = tcf_idr_create_from_flags(tn, index, est, a,
1322 						&act_ct_ops, bind, flags);
1323 		if (err) {
1324 			tcf_idr_cleanup(tn, index);
1325 			return err;
1326 		}
1327 		res = ACT_P_CREATED;
1328 	} else {
1329 		if (bind)
1330 			return 0;
1331 
1332 		if (!(flags & TCA_ACT_FLAGS_REPLACE)) {
1333 			tcf_idr_release(*a, bind);
1334 			return -EEXIST;
1335 		}
1336 	}
1337 	err = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack);
1338 	if (err < 0)
1339 		goto cleanup;
1340 
1341 	c = to_ct(*a);
1342 
1343 	params = kzalloc(sizeof(*params), GFP_KERNEL);
1344 	if (unlikely(!params)) {
1345 		err = -ENOMEM;
1346 		goto cleanup;
1347 	}
1348 
1349 	err = tcf_ct_fill_params(net, params, parm, tb, extack);
1350 	if (err)
1351 		goto cleanup;
1352 
1353 	err = tcf_ct_flow_table_get(net, params);
1354 	if (err)
1355 		goto cleanup;
1356 
1357 	spin_lock_bh(&c->tcf_lock);
1358 	goto_ch = tcf_action_set_ctrlact(*a, parm->action, goto_ch);
1359 	params = rcu_replace_pointer(c->params, params,
1360 				     lockdep_is_held(&c->tcf_lock));
1361 	spin_unlock_bh(&c->tcf_lock);
1362 
1363 	if (goto_ch)
1364 		tcf_chain_put_by_act(goto_ch);
1365 	if (params)
1366 		call_rcu(&params->rcu, tcf_ct_params_free_rcu);
1367 
1368 	return res;
1369 
1370 cleanup:
1371 	if (goto_ch)
1372 		tcf_chain_put_by_act(goto_ch);
1373 	if (params)
1374 		tcf_ct_params_free(params);
1375 	tcf_idr_release(*a, bind);
1376 	return err;
1377 }
1378 
1379 static void tcf_ct_cleanup(struct tc_action *a)
1380 {
1381 	struct tcf_ct_params *params;
1382 	struct tcf_ct *c = to_ct(a);
1383 
1384 	params = rcu_dereference_protected(c->params, 1);
1385 	if (params)
1386 		call_rcu(&params->rcu, tcf_ct_params_free_rcu);
1387 }
1388 
1389 static int tcf_ct_dump_key_val(struct sk_buff *skb,
1390 			       void *val, int val_type,
1391 			       void *mask, int mask_type,
1392 			       int len)
1393 {
1394 	int err;
1395 
1396 	if (mask && !memchr_inv(mask, 0, len))
1397 		return 0;
1398 
1399 	err = nla_put(skb, val_type, len, val);
1400 	if (err)
1401 		return err;
1402 
1403 	if (mask_type != TCA_CT_UNSPEC) {
1404 		err = nla_put(skb, mask_type, len, mask);
1405 		if (err)
1406 			return err;
1407 	}
1408 
1409 	return 0;
1410 }
1411 
1412 static int tcf_ct_dump_nat(struct sk_buff *skb, struct tcf_ct_params *p)
1413 {
1414 	struct nf_nat_range2 *range = &p->range;
1415 
1416 	if (!(p->ct_action & TCA_CT_ACT_NAT))
1417 		return 0;
1418 
1419 	if (!(p->ct_action & (TCA_CT_ACT_NAT_SRC | TCA_CT_ACT_NAT_DST)))
1420 		return 0;
1421 
1422 	if (range->flags & NF_NAT_RANGE_MAP_IPS) {
1423 		if (p->ipv4_range) {
1424 			if (nla_put_in_addr(skb, TCA_CT_NAT_IPV4_MIN,
1425 					    range->min_addr.ip))
1426 				return -1;
1427 			if (nla_put_in_addr(skb, TCA_CT_NAT_IPV4_MAX,
1428 					    range->max_addr.ip))
1429 				return -1;
1430 		} else {
1431 			if (nla_put_in6_addr(skb, TCA_CT_NAT_IPV6_MIN,
1432 					     &range->min_addr.in6))
1433 				return -1;
1434 			if (nla_put_in6_addr(skb, TCA_CT_NAT_IPV6_MAX,
1435 					     &range->max_addr.in6))
1436 				return -1;
1437 		}
1438 	}
1439 
1440 	if (range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) {
1441 		if (nla_put_be16(skb, TCA_CT_NAT_PORT_MIN,
1442 				 range->min_proto.all))
1443 			return -1;
1444 		if (nla_put_be16(skb, TCA_CT_NAT_PORT_MAX,
1445 				 range->max_proto.all))
1446 			return -1;
1447 	}
1448 
1449 	return 0;
1450 }
1451 
1452 static int tcf_ct_dump_helper(struct sk_buff *skb, struct nf_conntrack_helper *helper)
1453 {
1454 	if (!helper)
1455 		return 0;
1456 
1457 	if (nla_put_string(skb, TCA_CT_HELPER_NAME, helper->name) ||
1458 	    nla_put_u8(skb, TCA_CT_HELPER_FAMILY, helper->tuple.src.l3num) ||
1459 	    nla_put_u8(skb, TCA_CT_HELPER_PROTO, helper->tuple.dst.protonum))
1460 		return -1;
1461 
1462 	return 0;
1463 }
1464 
1465 static inline int tcf_ct_dump(struct sk_buff *skb, struct tc_action *a,
1466 			      int bind, int ref)
1467 {
1468 	unsigned char *b = skb_tail_pointer(skb);
1469 	struct tcf_ct *c = to_ct(a);
1470 	struct tcf_ct_params *p;
1471 
1472 	struct tc_ct opt = {
1473 		.index   = c->tcf_index,
1474 		.refcnt  = refcount_read(&c->tcf_refcnt) - ref,
1475 		.bindcnt = atomic_read(&c->tcf_bindcnt) - bind,
1476 	};
1477 	struct tcf_t t;
1478 
1479 	spin_lock_bh(&c->tcf_lock);
1480 	p = rcu_dereference_protected(c->params,
1481 				      lockdep_is_held(&c->tcf_lock));
1482 	opt.action = c->tcf_action;
1483 
1484 	if (tcf_ct_dump_key_val(skb,
1485 				&p->ct_action, TCA_CT_ACTION,
1486 				NULL, TCA_CT_UNSPEC,
1487 				sizeof(p->ct_action)))
1488 		goto nla_put_failure;
1489 
1490 	if (p->ct_action & TCA_CT_ACT_CLEAR)
1491 		goto skip_dump;
1492 
1493 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1494 	    tcf_ct_dump_key_val(skb,
1495 				&p->mark, TCA_CT_MARK,
1496 				&p->mark_mask, TCA_CT_MARK_MASK,
1497 				sizeof(p->mark)))
1498 		goto nla_put_failure;
1499 
1500 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1501 	    tcf_ct_dump_key_val(skb,
1502 				p->labels, TCA_CT_LABELS,
1503 				p->labels_mask, TCA_CT_LABELS_MASK,
1504 				sizeof(p->labels)))
1505 		goto nla_put_failure;
1506 
1507 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1508 	    tcf_ct_dump_key_val(skb,
1509 				&p->zone, TCA_CT_ZONE,
1510 				NULL, TCA_CT_UNSPEC,
1511 				sizeof(p->zone)))
1512 		goto nla_put_failure;
1513 
1514 	if (tcf_ct_dump_nat(skb, p))
1515 		goto nla_put_failure;
1516 
1517 	if (tcf_ct_dump_helper(skb, p->helper))
1518 		goto nla_put_failure;
1519 
1520 skip_dump:
1521 	if (nla_put(skb, TCA_CT_PARMS, sizeof(opt), &opt))
1522 		goto nla_put_failure;
1523 
1524 	tcf_tm_dump(&t, &c->tcf_tm);
1525 	if (nla_put_64bit(skb, TCA_CT_TM, sizeof(t), &t, TCA_CT_PAD))
1526 		goto nla_put_failure;
1527 	spin_unlock_bh(&c->tcf_lock);
1528 
1529 	return skb->len;
1530 nla_put_failure:
1531 	spin_unlock_bh(&c->tcf_lock);
1532 	nlmsg_trim(skb, b);
1533 	return -1;
1534 }
1535 
1536 static void tcf_stats_update(struct tc_action *a, u64 bytes, u64 packets,
1537 			     u64 drops, u64 lastuse, bool hw)
1538 {
1539 	struct tcf_ct *c = to_ct(a);
1540 
1541 	tcf_action_update_stats(a, bytes, packets, drops, hw);
1542 	c->tcf_tm.lastuse = max_t(u64, c->tcf_tm.lastuse, lastuse);
1543 }
1544 
1545 static int tcf_ct_offload_act_setup(struct tc_action *act, void *entry_data,
1546 				    u32 *index_inc, bool bind,
1547 				    struct netlink_ext_ack *extack)
1548 {
1549 	if (bind) {
1550 		struct flow_action_entry *entry = entry_data;
1551 
1552 		entry->id = FLOW_ACTION_CT;
1553 		entry->ct.action = tcf_ct_action(act);
1554 		entry->ct.zone = tcf_ct_zone(act);
1555 		entry->ct.flow_table = tcf_ct_ft(act);
1556 		*index_inc = 1;
1557 	} else {
1558 		struct flow_offload_action *fl_action = entry_data;
1559 
1560 		fl_action->id = FLOW_ACTION_CT;
1561 	}
1562 
1563 	return 0;
1564 }
1565 
1566 static struct tc_action_ops act_ct_ops = {
1567 	.kind		=	"ct",
1568 	.id		=	TCA_ID_CT,
1569 	.owner		=	THIS_MODULE,
1570 	.act		=	tcf_ct_act,
1571 	.dump		=	tcf_ct_dump,
1572 	.init		=	tcf_ct_init,
1573 	.cleanup	=	tcf_ct_cleanup,
1574 	.stats_update	=	tcf_stats_update,
1575 	.offload_act_setup =	tcf_ct_offload_act_setup,
1576 	.size		=	sizeof(struct tcf_ct),
1577 };
1578 
1579 static __net_init int ct_init_net(struct net *net)
1580 {
1581 	struct tc_ct_action_net *tn = net_generic(net, act_ct_ops.net_id);
1582 
1583 	return tc_action_net_init(net, &tn->tn, &act_ct_ops);
1584 }
1585 
1586 static void __net_exit ct_exit_net(struct list_head *net_list)
1587 {
1588 	tc_action_net_exit(net_list, act_ct_ops.net_id);
1589 }
1590 
1591 static struct pernet_operations ct_net_ops = {
1592 	.init = ct_init_net,
1593 	.exit_batch = ct_exit_net,
1594 	.id   = &act_ct_ops.net_id,
1595 	.size = sizeof(struct tc_ct_action_net),
1596 };
1597 
1598 static int __init ct_init_module(void)
1599 {
1600 	int err;
1601 
1602 	act_ct_wq = alloc_ordered_workqueue("act_ct_workqueue", 0);
1603 	if (!act_ct_wq)
1604 		return -ENOMEM;
1605 
1606 	err = tcf_ct_flow_tables_init();
1607 	if (err)
1608 		goto err_tbl_init;
1609 
1610 	err = tcf_register_action(&act_ct_ops, &ct_net_ops);
1611 	if (err)
1612 		goto err_register;
1613 
1614 	static_branch_inc(&tcf_frag_xmit_count);
1615 
1616 	return 0;
1617 
1618 err_register:
1619 	tcf_ct_flow_tables_uninit();
1620 err_tbl_init:
1621 	destroy_workqueue(act_ct_wq);
1622 	return err;
1623 }
1624 
1625 static void __exit ct_cleanup_module(void)
1626 {
1627 	static_branch_dec(&tcf_frag_xmit_count);
1628 	tcf_unregister_action(&act_ct_ops, &ct_net_ops);
1629 	tcf_ct_flow_tables_uninit();
1630 	destroy_workqueue(act_ct_wq);
1631 }
1632 
1633 module_init(ct_init_module);
1634 module_exit(ct_cleanup_module);
1635 MODULE_AUTHOR("Paul Blakey <paulb@mellanox.com>");
1636 MODULE_AUTHOR("Yossi Kuperman <yossiku@mellanox.com>");
1637 MODULE_AUTHOR("Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>");
1638 MODULE_DESCRIPTION("Connection tracking action");
1639 MODULE_LICENSE("GPL v2");
1640