xref: /openbmc/linux/net/sched/act_ct.c (revision 4e95bc26)
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 <net/netlink.h>
19 #include <net/pkt_sched.h>
20 #include <net/pkt_cls.h>
21 #include <net/act_api.h>
22 #include <net/ip.h>
23 #include <net/ipv6_frag.h>
24 #include <uapi/linux/tc_act/tc_ct.h>
25 #include <net/tc_act/tc_ct.h>
26 
27 #include <linux/netfilter/nf_nat.h>
28 #include <net/netfilter/nf_conntrack.h>
29 #include <net/netfilter/nf_conntrack_core.h>
30 #include <net/netfilter/nf_conntrack_zones.h>
31 #include <net/netfilter/nf_conntrack_helper.h>
32 #include <net/netfilter/ipv6/nf_defrag_ipv6.h>
33 
34 static struct tc_action_ops act_ct_ops;
35 static unsigned int ct_net_id;
36 
37 struct tc_ct_action_net {
38 	struct tc_action_net tn; /* Must be first */
39 	bool labels;
40 };
41 
42 /* Determine whether skb->_nfct is equal to the result of conntrack lookup. */
43 static bool tcf_ct_skb_nfct_cached(struct net *net, struct sk_buff *skb,
44 				   u16 zone_id, bool force)
45 {
46 	enum ip_conntrack_info ctinfo;
47 	struct nf_conn *ct;
48 
49 	ct = nf_ct_get(skb, &ctinfo);
50 	if (!ct)
51 		return false;
52 	if (!net_eq(net, read_pnet(&ct->ct_net)))
53 		return false;
54 	if (nf_ct_zone(ct)->id != zone_id)
55 		return false;
56 
57 	/* Force conntrack entry direction. */
58 	if (force && CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) {
59 		if (nf_ct_is_confirmed(ct))
60 			nf_ct_kill(ct);
61 
62 		nf_conntrack_put(&ct->ct_general);
63 		nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
64 
65 		return false;
66 	}
67 
68 	return true;
69 }
70 
71 /* Trim the skb to the length specified by the IP/IPv6 header,
72  * removing any trailing lower-layer padding. This prepares the skb
73  * for higher-layer processing that assumes skb->len excludes padding
74  * (such as nf_ip_checksum). The caller needs to pull the skb to the
75  * network header, and ensure ip_hdr/ipv6_hdr points to valid data.
76  */
77 static int tcf_ct_skb_network_trim(struct sk_buff *skb, int family)
78 {
79 	unsigned int len;
80 	int err;
81 
82 	switch (family) {
83 	case NFPROTO_IPV4:
84 		len = ntohs(ip_hdr(skb)->tot_len);
85 		break;
86 	case NFPROTO_IPV6:
87 		len = sizeof(struct ipv6hdr)
88 			+ ntohs(ipv6_hdr(skb)->payload_len);
89 		break;
90 	default:
91 		len = skb->len;
92 	}
93 
94 	err = pskb_trim_rcsum(skb, len);
95 
96 	return err;
97 }
98 
99 static u8 tcf_ct_skb_nf_family(struct sk_buff *skb)
100 {
101 	u8 family = NFPROTO_UNSPEC;
102 
103 	switch (skb->protocol) {
104 	case htons(ETH_P_IP):
105 		family = NFPROTO_IPV4;
106 		break;
107 	case htons(ETH_P_IPV6):
108 		family = NFPROTO_IPV6;
109 		break;
110 	default:
111 		break;
112 	}
113 
114 	return family;
115 }
116 
117 static int tcf_ct_ipv4_is_fragment(struct sk_buff *skb, bool *frag)
118 {
119 	unsigned int len;
120 
121 	len =  skb_network_offset(skb) + sizeof(struct iphdr);
122 	if (unlikely(skb->len < len))
123 		return -EINVAL;
124 	if (unlikely(!pskb_may_pull(skb, len)))
125 		return -ENOMEM;
126 
127 	*frag = ip_is_fragment(ip_hdr(skb));
128 	return 0;
129 }
130 
131 static int tcf_ct_ipv6_is_fragment(struct sk_buff *skb, bool *frag)
132 {
133 	unsigned int flags = 0, len, payload_ofs = 0;
134 	unsigned short frag_off;
135 	int nexthdr;
136 
137 	len =  skb_network_offset(skb) + sizeof(struct ipv6hdr);
138 	if (unlikely(skb->len < len))
139 		return -EINVAL;
140 	if (unlikely(!pskb_may_pull(skb, len)))
141 		return -ENOMEM;
142 
143 	nexthdr = ipv6_find_hdr(skb, &payload_ofs, -1, &frag_off, &flags);
144 	if (unlikely(nexthdr < 0))
145 		return -EPROTO;
146 
147 	*frag = flags & IP6_FH_F_FRAG;
148 	return 0;
149 }
150 
151 static int tcf_ct_handle_fragments(struct net *net, struct sk_buff *skb,
152 				   u8 family, u16 zone)
153 {
154 	enum ip_conntrack_info ctinfo;
155 	struct nf_conn *ct;
156 	int err = 0;
157 	bool frag;
158 
159 	/* Previously seen (loopback)? Ignore. */
160 	ct = nf_ct_get(skb, &ctinfo);
161 	if ((ct && !nf_ct_is_template(ct)) || ctinfo == IP_CT_UNTRACKED)
162 		return 0;
163 
164 	if (family == NFPROTO_IPV4)
165 		err = tcf_ct_ipv4_is_fragment(skb, &frag);
166 	else
167 		err = tcf_ct_ipv6_is_fragment(skb, &frag);
168 	if (err || !frag)
169 		return err;
170 
171 	skb_get(skb);
172 
173 	if (family == NFPROTO_IPV4) {
174 		enum ip_defrag_users user = IP_DEFRAG_CONNTRACK_IN + zone;
175 
176 		memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
177 		local_bh_disable();
178 		err = ip_defrag(net, skb, user);
179 		local_bh_enable();
180 		if (err && err != -EINPROGRESS)
181 			goto out_free;
182 	} else { /* NFPROTO_IPV6 */
183 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
184 		enum ip6_defrag_users user = IP6_DEFRAG_CONNTRACK_IN + zone;
185 
186 		memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm));
187 		err = nf_ct_frag6_gather(net, skb, user);
188 		if (err && err != -EINPROGRESS)
189 			goto out_free;
190 #else
191 		err = -EOPNOTSUPP;
192 		goto out_free;
193 #endif
194 	}
195 
196 	skb_clear_hash(skb);
197 	skb->ignore_df = 1;
198 	return err;
199 
200 out_free:
201 	kfree_skb(skb);
202 	return err;
203 }
204 
205 static void tcf_ct_params_free(struct rcu_head *head)
206 {
207 	struct tcf_ct_params *params = container_of(head,
208 						    struct tcf_ct_params, rcu);
209 
210 	if (params->tmpl)
211 		nf_conntrack_put(&params->tmpl->ct_general);
212 	kfree(params);
213 }
214 
215 #if IS_ENABLED(CONFIG_NF_NAT)
216 /* Modelled after nf_nat_ipv[46]_fn().
217  * range is only used for new, uninitialized NAT state.
218  * Returns either NF_ACCEPT or NF_DROP.
219  */
220 static int ct_nat_execute(struct sk_buff *skb, struct nf_conn *ct,
221 			  enum ip_conntrack_info ctinfo,
222 			  const struct nf_nat_range2 *range,
223 			  enum nf_nat_manip_type maniptype)
224 {
225 	int hooknum, err = NF_ACCEPT;
226 
227 	/* See HOOK2MANIP(). */
228 	if (maniptype == NF_NAT_MANIP_SRC)
229 		hooknum = NF_INET_LOCAL_IN; /* Source NAT */
230 	else
231 		hooknum = NF_INET_LOCAL_OUT; /* Destination NAT */
232 
233 	switch (ctinfo) {
234 	case IP_CT_RELATED:
235 	case IP_CT_RELATED_REPLY:
236 		if (skb->protocol == htons(ETH_P_IP) &&
237 		    ip_hdr(skb)->protocol == IPPROTO_ICMP) {
238 			if (!nf_nat_icmp_reply_translation(skb, ct, ctinfo,
239 							   hooknum))
240 				err = NF_DROP;
241 			goto out;
242 		} else if (IS_ENABLED(CONFIG_IPV6) &&
243 			   skb->protocol == htons(ETH_P_IPV6)) {
244 			__be16 frag_off;
245 			u8 nexthdr = ipv6_hdr(skb)->nexthdr;
246 			int hdrlen = ipv6_skip_exthdr(skb,
247 						      sizeof(struct ipv6hdr),
248 						      &nexthdr, &frag_off);
249 
250 			if (hdrlen >= 0 && nexthdr == IPPROTO_ICMPV6) {
251 				if (!nf_nat_icmpv6_reply_translation(skb, ct,
252 								     ctinfo,
253 								     hooknum,
254 								     hdrlen))
255 					err = NF_DROP;
256 				goto out;
257 			}
258 		}
259 		/* Non-ICMP, fall thru to initialize if needed. */
260 		/* fall through */
261 	case IP_CT_NEW:
262 		/* Seen it before?  This can happen for loopback, retrans,
263 		 * or local packets.
264 		 */
265 		if (!nf_nat_initialized(ct, maniptype)) {
266 			/* Initialize according to the NAT action. */
267 			err = (range && range->flags & NF_NAT_RANGE_MAP_IPS)
268 				/* Action is set up to establish a new
269 				 * mapping.
270 				 */
271 				? nf_nat_setup_info(ct, range, maniptype)
272 				: nf_nat_alloc_null_binding(ct, hooknum);
273 			if (err != NF_ACCEPT)
274 				goto out;
275 		}
276 		break;
277 
278 	case IP_CT_ESTABLISHED:
279 	case IP_CT_ESTABLISHED_REPLY:
280 		break;
281 
282 	default:
283 		err = NF_DROP;
284 		goto out;
285 	}
286 
287 	err = nf_nat_packet(ct, ctinfo, hooknum, skb);
288 out:
289 	return err;
290 }
291 #endif /* CONFIG_NF_NAT */
292 
293 static void tcf_ct_act_set_mark(struct nf_conn *ct, u32 mark, u32 mask)
294 {
295 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
296 	u32 new_mark;
297 
298 	if (!mask)
299 		return;
300 
301 	new_mark = mark | (ct->mark & ~(mask));
302 	if (ct->mark != new_mark) {
303 		ct->mark = new_mark;
304 		if (nf_ct_is_confirmed(ct))
305 			nf_conntrack_event_cache(IPCT_MARK, ct);
306 	}
307 #endif
308 }
309 
310 static void tcf_ct_act_set_labels(struct nf_conn *ct,
311 				  u32 *labels,
312 				  u32 *labels_m)
313 {
314 #if IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS)
315 	size_t labels_sz = FIELD_SIZEOF(struct tcf_ct_params, labels);
316 
317 	if (!memchr_inv(labels_m, 0, labels_sz))
318 		return;
319 
320 	nf_connlabels_replace(ct, labels, labels_m, 4);
321 #endif
322 }
323 
324 static int tcf_ct_act_nat(struct sk_buff *skb,
325 			  struct nf_conn *ct,
326 			  enum ip_conntrack_info ctinfo,
327 			  int ct_action,
328 			  struct nf_nat_range2 *range,
329 			  bool commit)
330 {
331 #if IS_ENABLED(CONFIG_NF_NAT)
332 	enum nf_nat_manip_type maniptype;
333 
334 	if (!(ct_action & TCA_CT_ACT_NAT))
335 		return NF_ACCEPT;
336 
337 	/* Add NAT extension if not confirmed yet. */
338 	if (!nf_ct_is_confirmed(ct) && !nf_ct_nat_ext_add(ct))
339 		return NF_DROP;   /* Can't NAT. */
340 
341 	if (ctinfo != IP_CT_NEW && (ct->status & IPS_NAT_MASK) &&
342 	    (ctinfo != IP_CT_RELATED || commit)) {
343 		/* NAT an established or related connection like before. */
344 		if (CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY)
345 			/* This is the REPLY direction for a connection
346 			 * for which NAT was applied in the forward
347 			 * direction.  Do the reverse NAT.
348 			 */
349 			maniptype = ct->status & IPS_SRC_NAT
350 				? NF_NAT_MANIP_DST : NF_NAT_MANIP_SRC;
351 		else
352 			maniptype = ct->status & IPS_SRC_NAT
353 				? NF_NAT_MANIP_SRC : NF_NAT_MANIP_DST;
354 	} else if (ct_action & TCA_CT_ACT_NAT_SRC) {
355 		maniptype = NF_NAT_MANIP_SRC;
356 	} else if (ct_action & TCA_CT_ACT_NAT_DST) {
357 		maniptype = NF_NAT_MANIP_DST;
358 	} else {
359 		return NF_ACCEPT;
360 	}
361 
362 	return ct_nat_execute(skb, ct, ctinfo, range, maniptype);
363 #else
364 	return NF_ACCEPT;
365 #endif
366 }
367 
368 static int tcf_ct_act(struct sk_buff *skb, const struct tc_action *a,
369 		      struct tcf_result *res)
370 {
371 	struct net *net = dev_net(skb->dev);
372 	bool cached, commit, clear, force;
373 	enum ip_conntrack_info ctinfo;
374 	struct tcf_ct *c = to_ct(a);
375 	struct nf_conn *tmpl = NULL;
376 	struct nf_hook_state state;
377 	int nh_ofs, err, retval;
378 	struct tcf_ct_params *p;
379 	struct nf_conn *ct;
380 	u8 family;
381 
382 	p = rcu_dereference_bh(c->params);
383 
384 	retval = READ_ONCE(c->tcf_action);
385 	commit = p->ct_action & TCA_CT_ACT_COMMIT;
386 	clear = p->ct_action & TCA_CT_ACT_CLEAR;
387 	force = p->ct_action & TCA_CT_ACT_FORCE;
388 	tmpl = p->tmpl;
389 
390 	if (clear) {
391 		ct = nf_ct_get(skb, &ctinfo);
392 		if (ct) {
393 			nf_conntrack_put(&ct->ct_general);
394 			nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
395 		}
396 
397 		goto out;
398 	}
399 
400 	family = tcf_ct_skb_nf_family(skb);
401 	if (family == NFPROTO_UNSPEC)
402 		goto drop;
403 
404 	/* The conntrack module expects to be working at L3.
405 	 * We also try to pull the IPv4/6 header to linear area
406 	 */
407 	nh_ofs = skb_network_offset(skb);
408 	skb_pull_rcsum(skb, nh_ofs);
409 	err = tcf_ct_handle_fragments(net, skb, family, p->zone);
410 	if (err == -EINPROGRESS) {
411 		retval = TC_ACT_STOLEN;
412 		goto out;
413 	}
414 	if (err)
415 		goto drop;
416 
417 	err = tcf_ct_skb_network_trim(skb, family);
418 	if (err)
419 		goto drop;
420 
421 	/* If we are recirculating packets to match on ct fields and
422 	 * committing with a separate ct action, then we don't need to
423 	 * actually run the packet through conntrack twice unless it's for a
424 	 * different zone.
425 	 */
426 	cached = tcf_ct_skb_nfct_cached(net, skb, p->zone, force);
427 	if (!cached) {
428 		/* Associate skb with specified zone. */
429 		if (tmpl) {
430 			ct = nf_ct_get(skb, &ctinfo);
431 			if (skb_nfct(skb))
432 				nf_conntrack_put(skb_nfct(skb));
433 			nf_conntrack_get(&tmpl->ct_general);
434 			nf_ct_set(skb, tmpl, IP_CT_NEW);
435 		}
436 
437 		state.hook = NF_INET_PRE_ROUTING;
438 		state.net = net;
439 		state.pf = family;
440 		err = nf_conntrack_in(skb, &state);
441 		if (err != NF_ACCEPT)
442 			goto out_push;
443 	}
444 
445 	ct = nf_ct_get(skb, &ctinfo);
446 	if (!ct)
447 		goto out_push;
448 	nf_ct_deliver_cached_events(ct);
449 
450 	err = tcf_ct_act_nat(skb, ct, ctinfo, p->ct_action, &p->range, commit);
451 	if (err != NF_ACCEPT)
452 		goto drop;
453 
454 	if (commit) {
455 		tcf_ct_act_set_mark(ct, p->mark, p->mark_mask);
456 		tcf_ct_act_set_labels(ct, p->labels, p->labels_mask);
457 
458 		/* This will take care of sending queued events
459 		 * even if the connection is already confirmed.
460 		 */
461 		nf_conntrack_confirm(skb);
462 	}
463 
464 out_push:
465 	skb_push_rcsum(skb, nh_ofs);
466 
467 out:
468 	bstats_cpu_update(this_cpu_ptr(a->cpu_bstats), skb);
469 	return retval;
470 
471 drop:
472 	qstats_drop_inc(this_cpu_ptr(a->cpu_qstats));
473 	return TC_ACT_SHOT;
474 }
475 
476 static const struct nla_policy ct_policy[TCA_CT_MAX + 1] = {
477 	[TCA_CT_UNSPEC] = { .strict_start_type = TCA_CT_UNSPEC + 1 },
478 	[TCA_CT_ACTION] = { .type = NLA_U16 },
479 	[TCA_CT_PARMS] = { .type = NLA_EXACT_LEN, .len = sizeof(struct tc_ct) },
480 	[TCA_CT_ZONE] = { .type = NLA_U16 },
481 	[TCA_CT_MARK] = { .type = NLA_U32 },
482 	[TCA_CT_MARK_MASK] = { .type = NLA_U32 },
483 	[TCA_CT_LABELS] = { .type = NLA_BINARY,
484 			    .len = 128 / BITS_PER_BYTE },
485 	[TCA_CT_LABELS_MASK] = { .type = NLA_BINARY,
486 				 .len = 128 / BITS_PER_BYTE },
487 	[TCA_CT_NAT_IPV4_MIN] = { .type = NLA_U32 },
488 	[TCA_CT_NAT_IPV4_MAX] = { .type = NLA_U32 },
489 	[TCA_CT_NAT_IPV6_MIN] = { .type = NLA_EXACT_LEN,
490 				  .len = sizeof(struct in6_addr) },
491 	[TCA_CT_NAT_IPV6_MAX] = { .type = NLA_EXACT_LEN,
492 				   .len = sizeof(struct in6_addr) },
493 	[TCA_CT_NAT_PORT_MIN] = { .type = NLA_U16 },
494 	[TCA_CT_NAT_PORT_MAX] = { .type = NLA_U16 },
495 };
496 
497 static int tcf_ct_fill_params_nat(struct tcf_ct_params *p,
498 				  struct tc_ct *parm,
499 				  struct nlattr **tb,
500 				  struct netlink_ext_ack *extack)
501 {
502 	struct nf_nat_range2 *range;
503 
504 	if (!(p->ct_action & TCA_CT_ACT_NAT))
505 		return 0;
506 
507 	if (!IS_ENABLED(CONFIG_NF_NAT)) {
508 		NL_SET_ERR_MSG_MOD(extack, "Netfilter nat isn't enabled in kernel");
509 		return -EOPNOTSUPP;
510 	}
511 
512 	if (!(p->ct_action & (TCA_CT_ACT_NAT_SRC | TCA_CT_ACT_NAT_DST)))
513 		return 0;
514 
515 	if ((p->ct_action & TCA_CT_ACT_NAT_SRC) &&
516 	    (p->ct_action & TCA_CT_ACT_NAT_DST)) {
517 		NL_SET_ERR_MSG_MOD(extack, "dnat and snat can't be enabled at the same time");
518 		return -EOPNOTSUPP;
519 	}
520 
521 	range = &p->range;
522 	if (tb[TCA_CT_NAT_IPV4_MIN]) {
523 		struct nlattr *max_attr = tb[TCA_CT_NAT_IPV4_MAX];
524 
525 		p->ipv4_range = true;
526 		range->flags |= NF_NAT_RANGE_MAP_IPS;
527 		range->min_addr.ip =
528 			nla_get_in_addr(tb[TCA_CT_NAT_IPV4_MIN]);
529 
530 		range->max_addr.ip = max_attr ?
531 				     nla_get_in_addr(max_attr) :
532 				     range->min_addr.ip;
533 	} else if (tb[TCA_CT_NAT_IPV6_MIN]) {
534 		struct nlattr *max_attr = tb[TCA_CT_NAT_IPV6_MAX];
535 
536 		p->ipv4_range = false;
537 		range->flags |= NF_NAT_RANGE_MAP_IPS;
538 		range->min_addr.in6 =
539 			nla_get_in6_addr(tb[TCA_CT_NAT_IPV6_MIN]);
540 
541 		range->max_addr.in6 = max_attr ?
542 				      nla_get_in6_addr(max_attr) :
543 				      range->min_addr.in6;
544 	}
545 
546 	if (tb[TCA_CT_NAT_PORT_MIN]) {
547 		range->flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
548 		range->min_proto.all = nla_get_be16(tb[TCA_CT_NAT_PORT_MIN]);
549 
550 		range->max_proto.all = tb[TCA_CT_NAT_PORT_MAX] ?
551 				       nla_get_be16(tb[TCA_CT_NAT_PORT_MAX]) :
552 				       range->min_proto.all;
553 	}
554 
555 	return 0;
556 }
557 
558 static void tcf_ct_set_key_val(struct nlattr **tb,
559 			       void *val, int val_type,
560 			       void *mask, int mask_type,
561 			       int len)
562 {
563 	if (!tb[val_type])
564 		return;
565 	nla_memcpy(val, tb[val_type], len);
566 
567 	if (!mask)
568 		return;
569 
570 	if (mask_type == TCA_CT_UNSPEC || !tb[mask_type])
571 		memset(mask, 0xff, len);
572 	else
573 		nla_memcpy(mask, tb[mask_type], len);
574 }
575 
576 static int tcf_ct_fill_params(struct net *net,
577 			      struct tcf_ct_params *p,
578 			      struct tc_ct *parm,
579 			      struct nlattr **tb,
580 			      struct netlink_ext_ack *extack)
581 {
582 	struct tc_ct_action_net *tn = net_generic(net, ct_net_id);
583 	struct nf_conntrack_zone zone;
584 	struct nf_conn *tmpl;
585 	int err;
586 
587 	p->zone = NF_CT_DEFAULT_ZONE_ID;
588 
589 	tcf_ct_set_key_val(tb,
590 			   &p->ct_action, TCA_CT_ACTION,
591 			   NULL, TCA_CT_UNSPEC,
592 			   sizeof(p->ct_action));
593 
594 	if (p->ct_action & TCA_CT_ACT_CLEAR)
595 		return 0;
596 
597 	err = tcf_ct_fill_params_nat(p, parm, tb, extack);
598 	if (err)
599 		return err;
600 
601 	if (tb[TCA_CT_MARK]) {
602 		if (!IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)) {
603 			NL_SET_ERR_MSG_MOD(extack, "Conntrack mark isn't enabled.");
604 			return -EOPNOTSUPP;
605 		}
606 		tcf_ct_set_key_val(tb,
607 				   &p->mark, TCA_CT_MARK,
608 				   &p->mark_mask, TCA_CT_MARK_MASK,
609 				   sizeof(p->mark));
610 	}
611 
612 	if (tb[TCA_CT_LABELS]) {
613 		if (!IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS)) {
614 			NL_SET_ERR_MSG_MOD(extack, "Conntrack labels isn't enabled.");
615 			return -EOPNOTSUPP;
616 		}
617 
618 		if (!tn->labels) {
619 			NL_SET_ERR_MSG_MOD(extack, "Failed to set connlabel length");
620 			return -EOPNOTSUPP;
621 		}
622 		tcf_ct_set_key_val(tb,
623 				   p->labels, TCA_CT_LABELS,
624 				   p->labels_mask, TCA_CT_LABELS_MASK,
625 				   sizeof(p->labels));
626 	}
627 
628 	if (tb[TCA_CT_ZONE]) {
629 		if (!IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES)) {
630 			NL_SET_ERR_MSG_MOD(extack, "Conntrack zones isn't enabled.");
631 			return -EOPNOTSUPP;
632 		}
633 
634 		tcf_ct_set_key_val(tb,
635 				   &p->zone, TCA_CT_ZONE,
636 				   NULL, TCA_CT_UNSPEC,
637 				   sizeof(p->zone));
638 	}
639 
640 	if (p->zone == NF_CT_DEFAULT_ZONE_ID)
641 		return 0;
642 
643 	nf_ct_zone_init(&zone, p->zone, NF_CT_DEFAULT_ZONE_DIR, 0);
644 	tmpl = nf_ct_tmpl_alloc(net, &zone, GFP_KERNEL);
645 	if (!tmpl) {
646 		NL_SET_ERR_MSG_MOD(extack, "Failed to allocate conntrack template");
647 		return -ENOMEM;
648 	}
649 	__set_bit(IPS_CONFIRMED_BIT, &tmpl->status);
650 	nf_conntrack_get(&tmpl->ct_general);
651 	p->tmpl = tmpl;
652 
653 	return 0;
654 }
655 
656 static int tcf_ct_init(struct net *net, struct nlattr *nla,
657 		       struct nlattr *est, struct tc_action **a,
658 		       int replace, int bind, bool rtnl_held,
659 		       struct tcf_proto *tp,
660 		       struct netlink_ext_ack *extack)
661 {
662 	struct tc_action_net *tn = net_generic(net, ct_net_id);
663 	struct tcf_ct_params *params = NULL;
664 	struct nlattr *tb[TCA_CT_MAX + 1];
665 	struct tcf_chain *goto_ch = NULL;
666 	struct tc_ct *parm;
667 	struct tcf_ct *c;
668 	int err, res = 0;
669 
670 	if (!nla) {
671 		NL_SET_ERR_MSG_MOD(extack, "Ct requires attributes to be passed");
672 		return -EINVAL;
673 	}
674 
675 	err = nla_parse_nested(tb, TCA_CT_MAX, nla, ct_policy, extack);
676 	if (err < 0)
677 		return err;
678 
679 	if (!tb[TCA_CT_PARMS]) {
680 		NL_SET_ERR_MSG_MOD(extack, "Missing required ct parameters");
681 		return -EINVAL;
682 	}
683 	parm = nla_data(tb[TCA_CT_PARMS]);
684 
685 	err = tcf_idr_check_alloc(tn, &parm->index, a, bind);
686 	if (err < 0)
687 		return err;
688 
689 	if (!err) {
690 		err = tcf_idr_create(tn, parm->index, est, a,
691 				     &act_ct_ops, bind, true);
692 		if (err) {
693 			tcf_idr_cleanup(tn, parm->index);
694 			return err;
695 		}
696 		res = ACT_P_CREATED;
697 	} else {
698 		if (bind)
699 			return 0;
700 
701 		if (!replace) {
702 			tcf_idr_release(*a, bind);
703 			return -EEXIST;
704 		}
705 	}
706 	err = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack);
707 	if (err < 0)
708 		goto cleanup;
709 
710 	c = to_ct(*a);
711 
712 	params = kzalloc(sizeof(*params), GFP_KERNEL);
713 	if (unlikely(!params)) {
714 		err = -ENOMEM;
715 		goto cleanup;
716 	}
717 
718 	err = tcf_ct_fill_params(net, params, parm, tb, extack);
719 	if (err)
720 		goto cleanup;
721 
722 	spin_lock_bh(&c->tcf_lock);
723 	goto_ch = tcf_action_set_ctrlact(*a, parm->action, goto_ch);
724 	rcu_swap_protected(c->params, params, lockdep_is_held(&c->tcf_lock));
725 	spin_unlock_bh(&c->tcf_lock);
726 
727 	if (goto_ch)
728 		tcf_chain_put_by_act(goto_ch);
729 	if (params)
730 		kfree_rcu(params, rcu);
731 	if (res == ACT_P_CREATED)
732 		tcf_idr_insert(tn, *a);
733 
734 	return res;
735 
736 cleanup:
737 	if (goto_ch)
738 		tcf_chain_put_by_act(goto_ch);
739 	kfree(params);
740 	tcf_idr_release(*a, bind);
741 	return err;
742 }
743 
744 static void tcf_ct_cleanup(struct tc_action *a)
745 {
746 	struct tcf_ct_params *params;
747 	struct tcf_ct *c = to_ct(a);
748 
749 	params = rcu_dereference_protected(c->params, 1);
750 	if (params)
751 		call_rcu(&params->rcu, tcf_ct_params_free);
752 }
753 
754 static int tcf_ct_dump_key_val(struct sk_buff *skb,
755 			       void *val, int val_type,
756 			       void *mask, int mask_type,
757 			       int len)
758 {
759 	int err;
760 
761 	if (mask && !memchr_inv(mask, 0, len))
762 		return 0;
763 
764 	err = nla_put(skb, val_type, len, val);
765 	if (err)
766 		return err;
767 
768 	if (mask_type != TCA_CT_UNSPEC) {
769 		err = nla_put(skb, mask_type, len, mask);
770 		if (err)
771 			return err;
772 	}
773 
774 	return 0;
775 }
776 
777 static int tcf_ct_dump_nat(struct sk_buff *skb, struct tcf_ct_params *p)
778 {
779 	struct nf_nat_range2 *range = &p->range;
780 
781 	if (!(p->ct_action & TCA_CT_ACT_NAT))
782 		return 0;
783 
784 	if (!(p->ct_action & (TCA_CT_ACT_NAT_SRC | TCA_CT_ACT_NAT_DST)))
785 		return 0;
786 
787 	if (range->flags & NF_NAT_RANGE_MAP_IPS) {
788 		if (p->ipv4_range) {
789 			if (nla_put_in_addr(skb, TCA_CT_NAT_IPV4_MIN,
790 					    range->min_addr.ip))
791 				return -1;
792 			if (nla_put_in_addr(skb, TCA_CT_NAT_IPV4_MAX,
793 					    range->max_addr.ip))
794 				return -1;
795 		} else {
796 			if (nla_put_in6_addr(skb, TCA_CT_NAT_IPV6_MIN,
797 					     &range->min_addr.in6))
798 				return -1;
799 			if (nla_put_in6_addr(skb, TCA_CT_NAT_IPV6_MAX,
800 					     &range->max_addr.in6))
801 				return -1;
802 		}
803 	}
804 
805 	if (range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) {
806 		if (nla_put_be16(skb, TCA_CT_NAT_PORT_MIN,
807 				 range->min_proto.all))
808 			return -1;
809 		if (nla_put_be16(skb, TCA_CT_NAT_PORT_MAX,
810 				 range->max_proto.all))
811 			return -1;
812 	}
813 
814 	return 0;
815 }
816 
817 static inline int tcf_ct_dump(struct sk_buff *skb, struct tc_action *a,
818 			      int bind, int ref)
819 {
820 	unsigned char *b = skb_tail_pointer(skb);
821 	struct tcf_ct *c = to_ct(a);
822 	struct tcf_ct_params *p;
823 
824 	struct tc_ct opt = {
825 		.index   = c->tcf_index,
826 		.refcnt  = refcount_read(&c->tcf_refcnt) - ref,
827 		.bindcnt = atomic_read(&c->tcf_bindcnt) - bind,
828 	};
829 	struct tcf_t t;
830 
831 	spin_lock_bh(&c->tcf_lock);
832 	p = rcu_dereference_protected(c->params,
833 				      lockdep_is_held(&c->tcf_lock));
834 	opt.action = c->tcf_action;
835 
836 	if (tcf_ct_dump_key_val(skb,
837 				&p->ct_action, TCA_CT_ACTION,
838 				NULL, TCA_CT_UNSPEC,
839 				sizeof(p->ct_action)))
840 		goto nla_put_failure;
841 
842 	if (p->ct_action & TCA_CT_ACT_CLEAR)
843 		goto skip_dump;
844 
845 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
846 	    tcf_ct_dump_key_val(skb,
847 				&p->mark, TCA_CT_MARK,
848 				&p->mark_mask, TCA_CT_MARK_MASK,
849 				sizeof(p->mark)))
850 		goto nla_put_failure;
851 
852 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
853 	    tcf_ct_dump_key_val(skb,
854 				p->labels, TCA_CT_LABELS,
855 				p->labels_mask, TCA_CT_LABELS_MASK,
856 				sizeof(p->labels)))
857 		goto nla_put_failure;
858 
859 	if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
860 	    tcf_ct_dump_key_val(skb,
861 				&p->zone, TCA_CT_ZONE,
862 				NULL, TCA_CT_UNSPEC,
863 				sizeof(p->zone)))
864 		goto nla_put_failure;
865 
866 	if (tcf_ct_dump_nat(skb, p))
867 		goto nla_put_failure;
868 
869 skip_dump:
870 	if (nla_put(skb, TCA_CT_PARMS, sizeof(opt), &opt))
871 		goto nla_put_failure;
872 
873 	tcf_tm_dump(&t, &c->tcf_tm);
874 	if (nla_put_64bit(skb, TCA_CT_TM, sizeof(t), &t, TCA_CT_PAD))
875 		goto nla_put_failure;
876 	spin_unlock_bh(&c->tcf_lock);
877 
878 	return skb->len;
879 nla_put_failure:
880 	spin_unlock_bh(&c->tcf_lock);
881 	nlmsg_trim(skb, b);
882 	return -1;
883 }
884 
885 static int tcf_ct_walker(struct net *net, struct sk_buff *skb,
886 			 struct netlink_callback *cb, int type,
887 			 const struct tc_action_ops *ops,
888 			 struct netlink_ext_ack *extack)
889 {
890 	struct tc_action_net *tn = net_generic(net, ct_net_id);
891 
892 	return tcf_generic_walker(tn, skb, cb, type, ops, extack);
893 }
894 
895 static int tcf_ct_search(struct net *net, struct tc_action **a, u32 index)
896 {
897 	struct tc_action_net *tn = net_generic(net, ct_net_id);
898 
899 	return tcf_idr_search(tn, a, index);
900 }
901 
902 static void tcf_stats_update(struct tc_action *a, u64 bytes, u32 packets,
903 			     u64 lastuse, bool hw)
904 {
905 	struct tcf_ct *c = to_ct(a);
906 
907 	_bstats_cpu_update(this_cpu_ptr(a->cpu_bstats), bytes, packets);
908 
909 	if (hw)
910 		_bstats_cpu_update(this_cpu_ptr(a->cpu_bstats_hw),
911 				   bytes, packets);
912 	c->tcf_tm.lastuse = max_t(u64, c->tcf_tm.lastuse, lastuse);
913 }
914 
915 static struct tc_action_ops act_ct_ops = {
916 	.kind		=	"ct",
917 	.id		=	TCA_ID_CT,
918 	.owner		=	THIS_MODULE,
919 	.act		=	tcf_ct_act,
920 	.dump		=	tcf_ct_dump,
921 	.init		=	tcf_ct_init,
922 	.cleanup	=	tcf_ct_cleanup,
923 	.walk		=	tcf_ct_walker,
924 	.lookup		=	tcf_ct_search,
925 	.stats_update	=	tcf_stats_update,
926 	.size		=	sizeof(struct tcf_ct),
927 };
928 
929 static __net_init int ct_init_net(struct net *net)
930 {
931 	unsigned int n_bits = FIELD_SIZEOF(struct tcf_ct_params, labels) * 8;
932 	struct tc_ct_action_net *tn = net_generic(net, ct_net_id);
933 
934 	if (nf_connlabels_get(net, n_bits - 1)) {
935 		tn->labels = false;
936 		pr_err("act_ct: Failed to set connlabels length");
937 	} else {
938 		tn->labels = true;
939 	}
940 
941 	return tc_action_net_init(&tn->tn, &act_ct_ops);
942 }
943 
944 static void __net_exit ct_exit_net(struct list_head *net_list)
945 {
946 	struct net *net;
947 
948 	rtnl_lock();
949 	list_for_each_entry(net, net_list, exit_list) {
950 		struct tc_ct_action_net *tn = net_generic(net, ct_net_id);
951 
952 		if (tn->labels)
953 			nf_connlabels_put(net);
954 	}
955 	rtnl_unlock();
956 
957 	tc_action_net_exit(net_list, ct_net_id);
958 }
959 
960 static struct pernet_operations ct_net_ops = {
961 	.init = ct_init_net,
962 	.exit_batch = ct_exit_net,
963 	.id   = &ct_net_id,
964 	.size = sizeof(struct tc_ct_action_net),
965 };
966 
967 static int __init ct_init_module(void)
968 {
969 	return tcf_register_action(&act_ct_ops, &ct_net_ops);
970 }
971 
972 static void __exit ct_cleanup_module(void)
973 {
974 	tcf_unregister_action(&act_ct_ops, &ct_net_ops);
975 }
976 
977 module_init(ct_init_module);
978 module_exit(ct_cleanup_module);
979 MODULE_AUTHOR("Paul Blakey <paulb@mellanox.com>");
980 MODULE_AUTHOR("Yossi Kuperman <yossiku@mellanox.com>");
981 MODULE_AUTHOR("Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>");
982 MODULE_DESCRIPTION("Connection tracking action");
983 MODULE_LICENSE("GPL v2");
984 
985