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(flow_action: 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(p: &target.src.u3, q: &tuple->src.u3, size: sizeof(target.src.u3)))
96	tcf_ct_add_mangle_action(action, htype: FLOW_ACT_MANGLE_HDR_TYPE_IP4,
97	offsetof(struct iphdr, saddr),
98	mask: 0xFFFFFFFF,
99	be32_to_cpu(target.src.u3.ip));
100	if (memcmp(p: &target.dst.u3, q: &tuple->dst.u3, size: sizeof(target.dst.u3)))
101	tcf_ct_add_mangle_action(action, htype: FLOW_ACT_MANGLE_HDR_TYPE_IP4,
102	offsetof(struct iphdr, daddr),
103	mask: 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, htype: FLOW_ACT_MANGLE_HDR_TYPE_IP6,
116	offset: i * sizeof(u32) + offset,
117	mask: 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(p: &target.src.u3, q: &tuple->src.u3, size: sizeof(target.src.u3)))
126	tcf_ct_add_ipv6_addr_mangle_action(action, addr: &target.src.u3,
127	offsetof(struct ipv6hdr,
128	saddr));
129	if (memcmp(p: &target.dst.u3, q: &tuple->dst.u3, size: sizeof(target.dst.u3)))
130	tcf_ct_add_ipv6_addr_mangle_action(action, addr: &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, htype: FLOW_ACT_MANGLE_HDR_TYPE_TCP,
145	offsetof(struct tcphdr, source),
146	mask: 0xFFFF, be16_to_cpu(target_src));
147	if (target_dst != tuple->dst.u.tcp.port)
148	tcf_ct_add_mangle_action(action, htype: FLOW_ACT_MANGLE_HDR_TYPE_TCP,
149	offsetof(struct tcphdr, dest),
150	mask: 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, htype: FLOW_ACT_MANGLE_HDR_TYPE_UDP,
163	offsetof(struct udphdr, source),
164	mask: 0xFFFF, be16_to_cpu(target_src));
165	if (target_dst != tuple->dst.u.udp.port)
166	tcf_ct_add_mangle_action(action, htype: FLOW_ACT_MANGLE_HDR_TYPE_UDP,
167	offsetof(struct udphdr, dest),
168	mask: 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(flow_action: 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(inverse: &target, orig: &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(nr: NF_FLOW_HW_ESTABLISHED, addr: &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 void tcf_ct_flow_table_get_ref(struct tcf_ct_flow_table *ct_ft);
290
291	static void tcf_ct_nf_get(struct nf_flowtable *ft)
292	{
293	struct tcf_ct_flow_table *ct_ft =
294	container_of(ft, struct tcf_ct_flow_table, nf_ft);
295
296	tcf_ct_flow_table_get_ref(ct_ft);
297	}
298
299	static void tcf_ct_flow_table_put(struct tcf_ct_flow_table *ct_ft);
300
301	static void tcf_ct_nf_put(struct nf_flowtable *ft)
302	{
303	struct tcf_ct_flow_table *ct_ft =
304	container_of(ft, struct tcf_ct_flow_table, nf_ft);
305
306	tcf_ct_flow_table_put(ct_ft);
307	}
308
309	static struct nf_flowtable_type flowtable_ct = {
310	.gc = tcf_ct_flow_is_outdated,
311	.action = tcf_ct_flow_table_fill_actions,
312	.get = tcf_ct_nf_get,
313	.put = tcf_ct_nf_put,
314	.owner = THIS_MODULE,
315	};
316
317	static int tcf_ct_flow_table_get(struct net *net, struct tcf_ct_params *params)
318	{
319	struct tcf_ct_flow_table *ct_ft;
320	int err = -ENOMEM;
321
322	mutex_lock(&zones_mutex);
323	ct_ft = rhashtable_lookup_fast(ht: &zones_ht, key: &params->zone, params: zones_params);
324	if (ct_ft && refcount_inc_not_zero(r: &ct_ft->ref))
325	goto out_unlock;
326
327	ct_ft = kzalloc(size: sizeof(*ct_ft), GFP_KERNEL);
328	if (!ct_ft)
329	goto err_alloc;
330	refcount_set(r: &ct_ft->ref, n: 1);
331
332	ct_ft->zone = params->zone;
333	err = rhashtable_insert_fast(ht: &zones_ht, obj: &ct_ft->node, params: zones_params);
334	if (err)
335	goto err_insert;
336
337	ct_ft->nf_ft.type = &flowtable_ct;
338	ct_ft->nf_ft.flags |= NF_FLOWTABLE_HW_OFFLOAD |
339	NF_FLOWTABLE_COUNTER;
340	err = nf_flow_table_init(flow_table: &ct_ft->nf_ft);
341	if (err)
342	goto err_init;
343	write_pnet(pnet: &ct_ft->nf_ft.net, net);
344
345	__module_get(THIS_MODULE);
346	out_unlock:
347	params->ct_ft = ct_ft;
348	params->nf_ft = &ct_ft->nf_ft;
349	mutex_unlock(lock: &zones_mutex);
350
351	return 0;
352
353	err_init:
354	rhashtable_remove_fast(ht: &zones_ht, obj: &ct_ft->node, params: zones_params);
355	err_insert:
356	kfree(objp: ct_ft);
357	err_alloc:
358	mutex_unlock(lock: &zones_mutex);
359	return err;
360	}
361
362	static void tcf_ct_flow_table_get_ref(struct tcf_ct_flow_table *ct_ft)
363	{
364	refcount_inc(r: &ct_ft->ref);
365	}
366
367	static void tcf_ct_flow_table_cleanup_work(struct work_struct *work)
368	{
369	struct tcf_ct_flow_table *ct_ft;
370	struct flow_block *block;
371
372	ct_ft = container_of(to_rcu_work(work), struct tcf_ct_flow_table,
373	rwork);
374	nf_flow_table_free(flow_table: &ct_ft->nf_ft);
375
376	block = &ct_ft->nf_ft.flow_block;
377	down_write(sem: &ct_ft->nf_ft.flow_block_lock);
378	WARN_ON(!list_empty(&block->cb_list));
379	up_write(sem: &ct_ft->nf_ft.flow_block_lock);
380	kfree(objp: ct_ft);
381
382	module_put(THIS_MODULE);
383	}
384
385	static void tcf_ct_flow_table_put(struct tcf_ct_flow_table *ct_ft)
386	{
387	if (refcount_dec_and_test(r: &ct_ft->ref)) {
388	rhashtable_remove_fast(ht: &zones_ht, obj: &ct_ft->node, params: zones_params);
389	INIT_RCU_WORK(&ct_ft->rwork, tcf_ct_flow_table_cleanup_work);
390	queue_rcu_work(wq: act_ct_wq, rwork: &ct_ft->rwork);
391	}
392	}
393
394	static void tcf_ct_flow_tc_ifidx(struct flow_offload *entry,
395	struct nf_conn_act_ct_ext *act_ct_ext, u8 dir)
396	{
397	entry->tuplehash[dir].tuple.xmit_type = FLOW_OFFLOAD_XMIT_TC;
398	entry->tuplehash[dir].tuple.tc.iifidx = act_ct_ext->ifindex[dir];
399	}
400
401	static void tcf_ct_flow_ct_ext_ifidx_update(struct flow_offload *entry)
402	{
403	struct nf_conn_act_ct_ext *act_ct_ext;
404
405	act_ct_ext = nf_conn_act_ct_ext_find(ct: entry->ct);
406	if (act_ct_ext) {
407	tcf_ct_flow_tc_ifidx(entry, act_ct_ext, dir: FLOW_OFFLOAD_DIR_ORIGINAL);
408	tcf_ct_flow_tc_ifidx(entry, act_ct_ext, dir: FLOW_OFFLOAD_DIR_REPLY);
409	}
410	}
411
412	static void tcf_ct_flow_table_add(struct tcf_ct_flow_table *ct_ft,
413	struct nf_conn *ct,
414	bool tcp, bool bidirectional)
415	{
416	struct nf_conn_act_ct_ext *act_ct_ext;
417	struct flow_offload *entry;
418	int err;
419
420	if (test_and_set_bit(nr: IPS_OFFLOAD_BIT, addr: &ct->status))
421	return;
422
423	entry = flow_offload_alloc(ct);
424	if (!entry) {
425	WARN_ON_ONCE(1);
426	goto err_alloc;
427	}
428
429	if (tcp) {
430	ct->proto.tcp.seen[0].flags |= IP_CT_TCP_FLAG_BE_LIBERAL;
431	ct->proto.tcp.seen[1].flags |= IP_CT_TCP_FLAG_BE_LIBERAL;
432	}
433	if (bidirectional)
434	__set_bit(NF_FLOW_HW_BIDIRECTIONAL, &entry->flags);
435
436	act_ct_ext = nf_conn_act_ct_ext_find(ct);
437	if (act_ct_ext) {
438	tcf_ct_flow_tc_ifidx(entry, act_ct_ext, dir: FLOW_OFFLOAD_DIR_ORIGINAL);
439	tcf_ct_flow_tc_ifidx(entry, act_ct_ext, dir: FLOW_OFFLOAD_DIR_REPLY);
440	}
441
442	err = flow_offload_add(flow_table: &ct_ft->nf_ft, flow: entry);
443	if (err)
444	goto err_add;
445
446	return;
447
448	err_add:
449	flow_offload_free(flow: entry);
450	err_alloc:
451	clear_bit(nr: IPS_OFFLOAD_BIT, addr: &ct->status);
452	}
453
454	static void tcf_ct_flow_table_process_conn(struct tcf_ct_flow_table *ct_ft,
455	struct nf_conn *ct,
456	enum ip_conntrack_info ctinfo)
457	{
458	bool tcp = false, bidirectional = true;
459
460	switch (nf_ct_protonum(ct)) {
461	case IPPROTO_TCP:
462	if ((ctinfo != IP_CT_ESTABLISHED &&
463	ctinfo != IP_CT_ESTABLISHED_REPLY) ||
464	!test_bit(IPS_ASSURED_BIT, &ct->status) ||
465	ct->proto.tcp.state != TCP_CONNTRACK_ESTABLISHED)
466	return;
467
468	tcp = true;
469	break;
470	case IPPROTO_UDP:
471	if (!nf_ct_is_confirmed(ct))
472	return;
473	if (!test_bit(IPS_ASSURED_BIT, &ct->status))
474	bidirectional = false;
475	break;
476	#ifdef CONFIG_NF_CT_PROTO_GRE
477	case IPPROTO_GRE: {
478	struct nf_conntrack_tuple *tuple;
479
480	if ((ctinfo != IP_CT_ESTABLISHED &&
481	ctinfo != IP_CT_ESTABLISHED_REPLY) ||
482	!test_bit(IPS_ASSURED_BIT, &ct->status) ||
483	ct->status & IPS_NAT_MASK)
484	return;
485
486	tuple = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
487	/* No support for GRE v1 */
488	if (tuple->src.u.gre.key || tuple->dst.u.gre.key)
489	return;
490	break;
491	}
492	#endif
493	default:
494	return;
495	}
496
497	if (nf_ct_ext_exist(ct, id: NF_CT_EXT_HELPER) ||
498	ct->status & IPS_SEQ_ADJUST)
499	return;
500
501	tcf_ct_flow_table_add(ct_ft, ct, tcp, bidirectional);
502	}
503
504	static bool
505	tcf_ct_flow_table_fill_tuple_ipv4(struct sk_buff *skb,
506	struct flow_offload_tuple *tuple,
507	struct tcphdr **tcph)
508	{
509	struct flow_ports *ports;
510	unsigned int thoff;
511	struct iphdr *iph;
512	size_t hdrsize;
513	u8 ipproto;
514
515	if (!pskb_network_may_pull(skb, len: sizeof(*iph)))
516	return false;
517
518	iph = ip_hdr(skb);
519	thoff = iph->ihl * 4;
520
521	if (ip_is_fragment(iph) ||
522	unlikely(thoff != sizeof(struct iphdr)))
523	return false;
524
525	ipproto = iph->protocol;
526	switch (ipproto) {
527	case IPPROTO_TCP:
528	hdrsize = sizeof(struct tcphdr);
529	break;
530	case IPPROTO_UDP:
531	hdrsize = sizeof(*ports);
532	break;
533	#ifdef CONFIG_NF_CT_PROTO_GRE
534	case IPPROTO_GRE:
535	hdrsize = sizeof(struct gre_base_hdr);
536	break;
537	#endif
538	default:
539	return false;
540	}
541
542	if (iph->ttl <= 1)
543	return false;
544
545	if (!pskb_network_may_pull(skb, len: thoff + hdrsize))
546	return false;
547
548	switch (ipproto) {
549	case IPPROTO_TCP:
550	*tcph = (void *)(skb_network_header(skb) + thoff);
551	fallthrough;
552	case IPPROTO_UDP:
553	ports = (struct flow_ports *)(skb_network_header(skb) + thoff);
554	tuple->src_port = ports->source;
555	tuple->dst_port = ports->dest;
556	break;
557	case IPPROTO_GRE: {
558	struct gre_base_hdr *greh;
559
560	greh = (struct gre_base_hdr *)(skb_network_header(skb) + thoff);
561	if ((greh->flags & GRE_VERSION) != GRE_VERSION_0)
562	return false;
563	break;
564	}
565	}
566
567	iph = ip_hdr(skb);
568
569	tuple->src_v4.s_addr = iph->saddr;
570	tuple->dst_v4.s_addr = iph->daddr;
571	tuple->l3proto = AF_INET;
572	tuple->l4proto = ipproto;
573
574	return true;
575	}
576
577	static bool
578	tcf_ct_flow_table_fill_tuple_ipv6(struct sk_buff *skb,
579	struct flow_offload_tuple *tuple,
580	struct tcphdr **tcph)
581	{
582	struct flow_ports *ports;
583	struct ipv6hdr *ip6h;
584	unsigned int thoff;
585	size_t hdrsize;
586	u8 nexthdr;
587
588	if (!pskb_network_may_pull(skb, len: sizeof(*ip6h)))
589	return false;
590
591	ip6h = ipv6_hdr(skb);
592	thoff = sizeof(*ip6h);
593
594	nexthdr = ip6h->nexthdr;
595	switch (nexthdr) {
596	case IPPROTO_TCP:
597	hdrsize = sizeof(struct tcphdr);
598	break;
599	case IPPROTO_UDP:
600	hdrsize = sizeof(*ports);
601	break;
602	#ifdef CONFIG_NF_CT_PROTO_GRE
603	case IPPROTO_GRE:
604	hdrsize = sizeof(struct gre_base_hdr);
605	break;
606	#endif
607	default:
608	return false;
609	}
610
611	if (ip6h->hop_limit <= 1)
612	return false;
613
614	if (!pskb_network_may_pull(skb, len: thoff + hdrsize))
615	return false;
616
617	switch (nexthdr) {
618	case IPPROTO_TCP:
619	*tcph = (void *)(skb_network_header(skb) + thoff);
620	fallthrough;
621	case IPPROTO_UDP:
622	ports = (struct flow_ports *)(skb_network_header(skb) + thoff);
623	tuple->src_port = ports->source;
624	tuple->dst_port = ports->dest;
625	break;
626	case IPPROTO_GRE: {
627	struct gre_base_hdr *greh;
628
629	greh = (struct gre_base_hdr *)(skb_network_header(skb) + thoff);
630	if ((greh->flags & GRE_VERSION) != GRE_VERSION_0)
631	return false;
632	break;
633	}
634	}
635
636	ip6h = ipv6_hdr(skb);
637
638	tuple->src_v6 = ip6h->saddr;
639	tuple->dst_v6 = ip6h->daddr;
640	tuple->l3proto = AF_INET6;
641	tuple->l4proto = nexthdr;
642
643	return true;
644	}
645
646	static bool tcf_ct_flow_table_lookup(struct tcf_ct_params *p,
647	struct sk_buff *skb,
648	u8 family)
649	{
650	struct nf_flowtable *nf_ft = &p->ct_ft->nf_ft;
651	struct flow_offload_tuple_rhash *tuplehash;
652	struct flow_offload_tuple tuple = {};
653	enum ip_conntrack_info ctinfo;
654	struct tcphdr *tcph = NULL;
655	bool force_refresh = false;
656	struct flow_offload *flow;
657	struct nf_conn *ct;
658	u8 dir;
659
660	switch (family) {
661	case NFPROTO_IPV4:
662	if (!tcf_ct_flow_table_fill_tuple_ipv4(skb, tuple: &tuple, tcph: &tcph))
663	return false;
664	break;
665	case NFPROTO_IPV6:
666	if (!tcf_ct_flow_table_fill_tuple_ipv6(skb, tuple: &tuple, tcph: &tcph))
667	return false;
668	break;
669	default:
670	return false;
671	}
672
673	tuplehash = flow_offload_lookup(flow_table: nf_ft, tuple: &tuple);
674	if (!tuplehash)
675	return false;
676
677	dir = tuplehash->tuple.dir;
678	flow = container_of(tuplehash, struct flow_offload, tuplehash[dir]);
679	ct = flow->ct;
680
681	if (dir == FLOW_OFFLOAD_DIR_REPLY &&
682	!test_bit(NF_FLOW_HW_BIDIRECTIONAL, &flow->flags)) {
683	/* Only offload reply direction after connection became
684	* assured.
685	*/
686	if (test_bit(IPS_ASSURED_BIT, &ct->status))
687	set_bit(nr: NF_FLOW_HW_BIDIRECTIONAL, addr: &flow->flags);
688	else if (test_bit(NF_FLOW_HW_ESTABLISHED, &flow->flags))
689	/* If flow_table flow has already been updated to the
690	* established state, then don't refresh.
691	*/
692	return false;
693	force_refresh = true;
694	}
695
696	if (tcph && (unlikely(tcph->fin || tcph->rst))) {
697	flow_offload_teardown(flow);
698	return false;
699	}
700
701	if (dir == FLOW_OFFLOAD_DIR_ORIGINAL)
702	ctinfo = test_bit(IPS_SEEN_REPLY_BIT, &ct->status) ?
703	IP_CT_ESTABLISHED : IP_CT_NEW;
704	else
705	ctinfo = IP_CT_ESTABLISHED_REPLY;
706
707	nf_conn_act_ct_ext_fill(skb, ct, ctinfo);
708	tcf_ct_flow_ct_ext_ifidx_update(entry: flow);
709	flow_offload_refresh(flow_table: nf_ft, flow, force: force_refresh);
710	if (!test_bit(IPS_ASSURED_BIT, &ct->status)) {
711	/* Process this flow in SW to allow promoting to ASSURED */
712	return false;
713	}
714
715	nf_conntrack_get(nfct: &ct->ct_general);
716	nf_ct_set(skb, ct, info: ctinfo);
717	if (nf_ft->flags & NF_FLOWTABLE_COUNTER)
718	nf_ct_acct_update(ct, dir, bytes: skb->len);
719
720	return true;
721	}
722
723	static int tcf_ct_flow_tables_init(void)
724	{
725	return rhashtable_init(ht: &zones_ht, params: &zones_params);
726	}
727
728	static void tcf_ct_flow_tables_uninit(void)
729	{
730	rhashtable_destroy(ht: &zones_ht);
731	}
732
733	static struct tc_action_ops act_ct_ops;
734
735	struct tc_ct_action_net {
736	struct tc_action_net tn; /* Must be first */
737	};
738
739	/* Determine whether skb->_nfct is equal to the result of conntrack lookup. */
740	static bool tcf_ct_skb_nfct_cached(struct net *net, struct sk_buff *skb,
741	struct tcf_ct_params *p)
742	{
743	enum ip_conntrack_info ctinfo;
744	struct nf_conn *ct;
745
746	ct = nf_ct_get(skb, ctinfo: &ctinfo);
747	if (!ct)
748	return false;
749	if (!net_eq(net1: net, net2: read_pnet(pnet: &ct->ct_net)))
750	goto drop_ct;
751	if (nf_ct_zone(ct)->id != p->zone)
752	goto drop_ct;
753	if (p->helper) {
754	struct nf_conn_help *help;
755
756	help = nf_ct_ext_find(ct, id: NF_CT_EXT_HELPER);
757	if (help && rcu_access_pointer(help->helper) != p->helper)
758	goto drop_ct;
759	}
760
761	/* Force conntrack entry direction. */
762	if ((p->ct_action & TCA_CT_ACT_FORCE) &&
763	CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) {
764	if (nf_ct_is_confirmed(ct))
765	nf_ct_kill(ct);
766
767	goto drop_ct;
768	}
769
770	return true;
771
772	drop_ct:
773	nf_ct_put(ct);
774	nf_ct_set(skb, NULL, info: IP_CT_UNTRACKED);
775
776	return false;
777	}
778
779	static u8 tcf_ct_skb_nf_family(struct sk_buff *skb)
780	{
781	u8 family = NFPROTO_UNSPEC;
782
783	switch (skb_protocol(skb, skip_vlan: true)) {
784	case htons(ETH_P_IP):
785	family = NFPROTO_IPV4;
786	break;
787	case htons(ETH_P_IPV6):
788	family = NFPROTO_IPV6;
789	break;
790	default:
791	break;
792	}
793
794	return family;
795	}
796
797	static int tcf_ct_ipv4_is_fragment(struct sk_buff *skb, bool *frag)
798	{
799	unsigned int len;
800
801	len = skb_network_offset(skb) + sizeof(struct iphdr);
802	if (unlikely(skb->len < len))
803	return -EINVAL;
804	if (unlikely(!pskb_may_pull(skb, len)))
805	return -ENOMEM;
806
807	*frag = ip_is_fragment(iph: ip_hdr(skb));
808	return 0;
809	}
810
811	static int tcf_ct_ipv6_is_fragment(struct sk_buff *skb, bool *frag)
812	{
813	unsigned int flags = 0, len, payload_ofs = 0;
814	unsigned short frag_off;
815	int nexthdr;
816
817	len = skb_network_offset(skb) + sizeof(struct ipv6hdr);
818	if (unlikely(skb->len < len))
819	return -EINVAL;
820	if (unlikely(!pskb_may_pull(skb, len)))
821	return -ENOMEM;
822
823	nexthdr = ipv6_find_hdr(skb, offset: &payload_ofs, target: -1, fragoff: &frag_off, fragflg: &flags);
824	if (unlikely(nexthdr < 0))
825	return -EPROTO;
826
827	*frag = flags & IP6_FH_F_FRAG;
828	return 0;
829	}
830
831	static int tcf_ct_handle_fragments(struct net *net, struct sk_buff *skb,
832	u8 family, u16 zone, bool *defrag)
833	{
834	enum ip_conntrack_info ctinfo;
835	struct nf_conn *ct;
836	int err = 0;
837	bool frag;
838	u8 proto;
839	u16 mru;
840
841	/* Previously seen (loopback)? Ignore. */
842	ct = nf_ct_get(skb, ctinfo: &ctinfo);
843	if ((ct && !nf_ct_is_template(ct)) || ctinfo == IP_CT_UNTRACKED)
844	return 0;
845
846	if (family == NFPROTO_IPV4)
847	err = tcf_ct_ipv4_is_fragment(skb, frag: &frag);
848	else
849	err = tcf_ct_ipv6_is_fragment(skb, frag: &frag);
850	if (err || !frag)
851	return err;
852
853	err = nf_ct_handle_fragments(net, skb, zone, family, proto: &proto, mru: &mru);
854	if (err)
855	return err;
856
857	*defrag = true;
858	tc_skb_cb(skb)->mru = mru;
859
860	return 0;
861	}
862
863	static void tcf_ct_params_free(struct tcf_ct_params *params)
864	{
865	if (params->helper) {
866	#if IS_ENABLED(CONFIG_NF_NAT)
867	if (params->ct_action & TCA_CT_ACT_NAT)
868	nf_nat_helper_put(helper: params->helper);
869	#endif
870	nf_conntrack_helper_put(helper: params->helper);
871	}
872	if (params->ct_ft)
873	tcf_ct_flow_table_put(ct_ft: params->ct_ft);
874	if (params->tmpl) {
875	if (params->put_labels)
876	nf_connlabels_put(net: nf_ct_net(ct: params->tmpl));
877
878	nf_ct_put(ct: params->tmpl);
879	}
880
881	kfree(objp: params);
882	}
883
884	static void tcf_ct_params_free_rcu(struct rcu_head *head)
885	{
886	struct tcf_ct_params *params;
887
888	params = container_of(head, struct tcf_ct_params, rcu);
889	tcf_ct_params_free(params);
890	}
891
892	static void tcf_ct_act_set_mark(struct nf_conn *ct, u32 mark, u32 mask)
893	{
894	#if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
895	u32 new_mark;
896
897	if (!mask)
898	return;
899
900	new_mark = mark | (READ_ONCE(ct->mark) & ~(mask));
901	if (READ_ONCE(ct->mark) != new_mark) {
902	WRITE_ONCE(ct->mark, new_mark);
903	if (nf_ct_is_confirmed(ct))
904	nf_conntrack_event_cache(event: IPCT_MARK, ct);
905	}
906	#endif
907	}
908
909	static void tcf_ct_act_set_labels(struct nf_conn *ct,
910	u32 *labels,
911	u32 *labels_m)
912	{
913	#if IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS)
914	size_t labels_sz = sizeof_field(struct tcf_ct_params, labels);
915
916	if (!memchr_inv(p: labels_m, c: 0, size: labels_sz))
917	return;
918
919	nf_connlabels_replace(ct, data: labels, mask: labels_m, words: 4);
920	#endif
921	}
922
923	static int tcf_ct_act_nat(struct sk_buff *skb,
924	struct nf_conn *ct,
925	enum ip_conntrack_info ctinfo,
926	int ct_action,
927	struct nf_nat_range2 *range,
928	bool commit)
929	{
930	#if IS_ENABLED(CONFIG_NF_NAT)
931	int err, action = 0;
932
933	if (!(ct_action & TCA_CT_ACT_NAT))
934	return NF_ACCEPT;
935	if (ct_action & TCA_CT_ACT_NAT_SRC)
936	action |= BIT(NF_NAT_MANIP_SRC);
937	if (ct_action & TCA_CT_ACT_NAT_DST)
938	action |= BIT(NF_NAT_MANIP_DST);
939
940	err = nf_ct_nat(skb, ct, ctinfo, action: &action, range, commit);
941
942	if (action & BIT(NF_NAT_MANIP_SRC))
943	tc_skb_cb(skb)->post_ct_snat = 1;
944	if (action & BIT(NF_NAT_MANIP_DST))
945	tc_skb_cb(skb)->post_ct_dnat = 1;
946
947	return err;
948	#else
949	return NF_ACCEPT;
950	#endif
951	}
952
953	TC_INDIRECT_SCOPE int tcf_ct_act(struct sk_buff *skb, const struct tc_action *a,
954	struct tcf_result *res)
955	{
956	struct net *net = dev_net(dev: skb->dev);
957	enum ip_conntrack_info ctinfo;
958	struct tcf_ct *c = to_ct(a);
959	struct nf_conn *tmpl = NULL;
960	struct nf_hook_state state;
961	bool cached, commit, clear;
962	int nh_ofs, err, retval;
963	struct tcf_ct_params *p;
964	bool add_helper = false;
965	bool skip_add = false;
966	bool defrag = false;
967	struct nf_conn *ct;
968	u8 family;
969
970	p = rcu_dereference_bh(c->params);
971
972	retval = READ_ONCE(c->tcf_action);
973	commit = p->ct_action & TCA_CT_ACT_COMMIT;
974	clear = p->ct_action & TCA_CT_ACT_CLEAR;
975	tmpl = p->tmpl;
976
977	tcf_lastuse_update(tm: &c->tcf_tm);
978	tcf_action_update_bstats(a: &c->common, skb);
979
980	if (clear) {
981	tc_skb_cb(skb)->post_ct = false;
982	ct = nf_ct_get(skb, ctinfo: &ctinfo);
983	if (ct) {
984	nf_ct_put(ct);
985	nf_ct_set(skb, NULL, info: IP_CT_UNTRACKED);
986	}
987
988	goto out_clear;
989	}
990
991	family = tcf_ct_skb_nf_family(skb);
992	if (family == NFPROTO_UNSPEC)
993	goto drop;
994
995	/* The conntrack module expects to be working at L3.
996	* We also try to pull the IPv4/6 header to linear area
997	*/
998	nh_ofs = skb_network_offset(skb);
999	skb_pull_rcsum(skb, len: nh_ofs);
1000	err = tcf_ct_handle_fragments(net, skb, family, zone: p->zone, defrag: &defrag);
1001	if (err)
1002	goto out_frag;
1003
1004	err = nf_ct_skb_network_trim(skb, family);
1005	if (err)
1006	goto drop;
1007
1008	/* If we are recirculating packets to match on ct fields and
1009	* committing with a separate ct action, then we don't need to
1010	* actually run the packet through conntrack twice unless it's for a
1011	* different zone.
1012	*/
1013	cached = tcf_ct_skb_nfct_cached(net, skb, p);
1014	if (!cached) {
1015	if (tcf_ct_flow_table_lookup(p, skb, family)) {
1016	skip_add = true;
1017	goto do_nat;
1018	}
1019
1020	/* Associate skb with specified zone. */
1021	if (tmpl) {
1022	nf_conntrack_put(nfct: skb_nfct(skb));
1023	nf_conntrack_get(nfct: &tmpl->ct_general);
1024	nf_ct_set(skb, ct: tmpl, info: IP_CT_NEW);
1025	}
1026
1027	state.hook = NF_INET_PRE_ROUTING;
1028	state.net = net;
1029	state.pf = family;
1030	err = nf_conntrack_in(skb, state: &state);
1031	if (err != NF_ACCEPT)
1032	goto out_push;
1033	}
1034
1035	do_nat:
1036	ct = nf_ct_get(skb, ctinfo: &ctinfo);
1037	if (!ct)
1038	goto out_push;
1039	nf_ct_deliver_cached_events(ct);
1040	nf_conn_act_ct_ext_fill(skb, ct, ctinfo);
1041
1042	err = tcf_ct_act_nat(skb, ct, ctinfo, ct_action: p->ct_action, range: &p->range, commit);
1043	if (err != NF_ACCEPT)
1044	goto drop;
1045
1046	if (!nf_ct_is_confirmed(ct) && commit && p->helper && !nfct_help(ct)) {
1047	err = __nf_ct_try_assign_helper(ct, tmpl: p->tmpl, GFP_ATOMIC);
1048	if (err)
1049	goto drop;
1050	add_helper = true;
1051	if (p->ct_action & TCA_CT_ACT_NAT && !nfct_seqadj(ct)) {
1052	if (!nfct_seqadj_ext_add(ct))
1053	goto drop;
1054	}
1055	}
1056
1057	if (nf_ct_is_confirmed(ct) ? ((!cached && !skip_add) || add_helper) : commit) {
1058	if (nf_ct_helper(skb, ct, ctinfo, proto: family) != NF_ACCEPT)
1059	goto drop;
1060	}
1061
1062	if (commit) {
1063	tcf_ct_act_set_mark(ct, mark: p->mark, mask: p->mark_mask);
1064	tcf_ct_act_set_labels(ct, labels: p->labels, labels_m: p->labels_mask);
1065
1066	if (!nf_ct_is_confirmed(ct))
1067	nf_conn_act_ct_ext_add(skb, ct, ctinfo);
1068
1069	/* This will take care of sending queued events
1070	* even if the connection is already confirmed.
1071	*/
1072	if (nf_conntrack_confirm(skb) != NF_ACCEPT)
1073	goto drop;
1074	}
1075
1076	if (!skip_add)
1077	tcf_ct_flow_table_process_conn(ct_ft: p->ct_ft, ct, ctinfo);
1078
1079	out_push:
1080	skb_push_rcsum(skb, len: nh_ofs);
1081
1082	tc_skb_cb(skb)->post_ct = true;
1083	tc_skb_cb(skb)->zone = p->zone;
1084	out_clear:
1085	if (defrag)
1086	qdisc_skb_cb(skb)->pkt_len = skb->len;
1087	return retval;
1088
1089	out_frag:
1090	if (err != -EINPROGRESS)
1091	tcf_action_inc_drop_qstats(a: &c->common);
1092	return TC_ACT_CONSUMED;
1093
1094	drop:
1095	tcf_action_inc_drop_qstats(a: &c->common);
1096	return TC_ACT_SHOT;
1097	}
1098
1099	static const struct nla_policy ct_policy[TCA_CT_MAX + 1] = {
1100	[TCA_CT_ACTION] = { .type = NLA_U16 },
1101	[TCA_CT_PARMS] = NLA_POLICY_EXACT_LEN(sizeof(struct tc_ct)),
1102	[TCA_CT_ZONE] = { .type = NLA_U16 },
1103	[TCA_CT_MARK] = { .type = NLA_U32 },
1104	[TCA_CT_MARK_MASK] = { .type = NLA_U32 },
1105	[TCA_CT_LABELS] = { .type = NLA_BINARY,
1106	.len = 128 / BITS_PER_BYTE },
1107	[TCA_CT_LABELS_MASK] = { .type = NLA_BINARY,
1108	.len = 128 / BITS_PER_BYTE },
1109	[TCA_CT_NAT_IPV4_MIN] = { .type = NLA_U32 },
1110	[TCA_CT_NAT_IPV4_MAX] = { .type = NLA_U32 },
1111	[TCA_CT_NAT_IPV6_MIN] = NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)),
1112	[TCA_CT_NAT_IPV6_MAX] = NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)),
1113	[TCA_CT_NAT_PORT_MIN] = { .type = NLA_U16 },
1114	[TCA_CT_NAT_PORT_MAX] = { .type = NLA_U16 },
1115	[TCA_CT_HELPER_NAME] = { .type = NLA_STRING, .len = NF_CT_HELPER_NAME_LEN },
1116	[TCA_CT_HELPER_FAMILY] = { .type = NLA_U8 },
1117	[TCA_CT_HELPER_PROTO] = { .type = NLA_U8 },
1118	};
1119
1120	static int tcf_ct_fill_params_nat(struct tcf_ct_params *p,
1121	struct tc_ct *parm,
1122	struct nlattr **tb,
1123	struct netlink_ext_ack *extack)
1124	{
1125	struct nf_nat_range2 *range;
1126
1127	if (!(p->ct_action & TCA_CT_ACT_NAT))
1128	return 0;
1129
1130	if (!IS_ENABLED(CONFIG_NF_NAT)) {
1131	NL_SET_ERR_MSG_MOD(extack, "Netfilter nat isn't enabled in kernel");
1132	return -EOPNOTSUPP;
1133	}
1134
1135	if (!(p->ct_action & (TCA_CT_ACT_NAT_SRC | TCA_CT_ACT_NAT_DST)))
1136	return 0;
1137
1138	if ((p->ct_action & TCA_CT_ACT_NAT_SRC) &&
1139	(p->ct_action & TCA_CT_ACT_NAT_DST)) {
1140	NL_SET_ERR_MSG_MOD(extack, "dnat and snat can't be enabled at the same time");
1141	return -EOPNOTSUPP;
1142	}
1143
1144	range = &p->range;
1145	if (tb[TCA_CT_NAT_IPV4_MIN]) {
1146	struct nlattr *max_attr = tb[TCA_CT_NAT_IPV4_MAX];
1147
1148	p->ipv4_range = true;
1149	range->flags |= NF_NAT_RANGE_MAP_IPS;
1150	range->min_addr.ip =
1151	nla_get_in_addr(nla: tb[TCA_CT_NAT_IPV4_MIN]);
1152
1153	range->max_addr.ip = max_attr ?
1154	nla_get_in_addr(nla: max_attr) :
1155	range->min_addr.ip;
1156	} else if (tb[TCA_CT_NAT_IPV6_MIN]) {
1157	struct nlattr *max_attr = tb[TCA_CT_NAT_IPV6_MAX];
1158
1159	p->ipv4_range = false;
1160	range->flags |= NF_NAT_RANGE_MAP_IPS;
1161	range->min_addr.in6 =
1162	nla_get_in6_addr(nla: tb[TCA_CT_NAT_IPV6_MIN]);
1163
1164	range->max_addr.in6 = max_attr ?
1165	nla_get_in6_addr(nla: max_attr) :
1166	range->min_addr.in6;
1167	}
1168
1169	if (tb[TCA_CT_NAT_PORT_MIN]) {
1170	range->flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1171	range->min_proto.all = nla_get_be16(nla: tb[TCA_CT_NAT_PORT_MIN]);
1172
1173	range->max_proto.all = tb[TCA_CT_NAT_PORT_MAX] ?
1174	nla_get_be16(nla: tb[TCA_CT_NAT_PORT_MAX]) :
1175	range->min_proto.all;
1176	}
1177
1178	return 0;
1179	}
1180
1181	static void tcf_ct_set_key_val(struct nlattr **tb,
1182	void *val, int val_type,
1183	void *mask, int mask_type,
1184	int len)
1185	{
1186	if (!tb[val_type])
1187	return;
1188	nla_memcpy(dest: val, src: tb[val_type], count: len);
1189
1190	if (!mask)
1191	return;
1192
1193	if (mask_type == TCA_CT_UNSPEC || !tb[mask_type])
1194	memset(mask, 0xff, len);
1195	else
1196	nla_memcpy(dest: mask, src: tb[mask_type], count: len);
1197	}
1198
1199	static int tcf_ct_fill_params(struct net *net,
1200	struct tcf_ct_params *p,
1201	struct tc_ct *parm,
1202	struct nlattr **tb,
1203	struct netlink_ext_ack *extack)
1204	{
1205	struct nf_conntrack_zone zone;
1206	int err, family, proto, len;
1207	bool put_labels = false;
1208	struct nf_conn *tmpl;
1209	char *name;
1210
1211	p->zone = NF_CT_DEFAULT_ZONE_ID;
1212
1213	tcf_ct_set_key_val(tb,
1214	val: &p->ct_action, val_type: TCA_CT_ACTION,
1215	NULL, mask_type: TCA_CT_UNSPEC,
1216	len: sizeof(p->ct_action));
1217
1218	if (p->ct_action & TCA_CT_ACT_CLEAR)
1219	return 0;
1220
1221	err = tcf_ct_fill_params_nat(p, parm, tb, extack);
1222	if (err)
1223	return err;
1224
1225	if (tb[TCA_CT_MARK]) {
1226	if (!IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)) {
1227	NL_SET_ERR_MSG_MOD(extack, "Conntrack mark isn't enabled.");
1228	return -EOPNOTSUPP;
1229	}
1230	tcf_ct_set_key_val(tb,
1231	val: &p->mark, val_type: TCA_CT_MARK,
1232	mask: &p->mark_mask, mask_type: TCA_CT_MARK_MASK,
1233	len: sizeof(p->mark));
1234	}
1235
1236	if (tb[TCA_CT_LABELS]) {
1237	unsigned int n_bits = sizeof_field(struct tcf_ct_params, labels) * 8;
1238
1239	if (!IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS)) {
1240	NL_SET_ERR_MSG_MOD(extack, "Conntrack labels isn't enabled.");
1241	return -EOPNOTSUPP;
1242	}
1243
1244	if (nf_connlabels_get(net, bit: n_bits - 1)) {
1245	NL_SET_ERR_MSG_MOD(extack, "Failed to set connlabel length");
1246	return -EOPNOTSUPP;
1247	} else {
1248	put_labels = true;
1249	}
1250
1251	tcf_ct_set_key_val(tb,
1252	val: p->labels, val_type: TCA_CT_LABELS,
1253	mask: p->labels_mask, mask_type: TCA_CT_LABELS_MASK,
1254	len: sizeof(p->labels));
1255	}
1256
1257	if (tb[TCA_CT_ZONE]) {
1258	if (!IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES)) {
1259	NL_SET_ERR_MSG_MOD(extack, "Conntrack zones isn't enabled.");
1260	return -EOPNOTSUPP;
1261	}
1262
1263	tcf_ct_set_key_val(tb,
1264	val: &p->zone, val_type: TCA_CT_ZONE,
1265	NULL, mask_type: TCA_CT_UNSPEC,
1266	len: sizeof(p->zone));
1267	}
1268
1269	nf_ct_zone_init(zone: &zone, id: p->zone, NF_CT_DEFAULT_ZONE_DIR, flags: 0);
1270	tmpl = nf_ct_tmpl_alloc(net, zone: &zone, GFP_KERNEL);
1271	if (!tmpl) {
1272	NL_SET_ERR_MSG_MOD(extack, "Failed to allocate conntrack template");
1273	return -ENOMEM;
1274	}
1275	p->tmpl = tmpl;
1276	if (tb[TCA_CT_HELPER_NAME]) {
1277	name = nla_data(nla: tb[TCA_CT_HELPER_NAME]);
1278	len = nla_len(nla: tb[TCA_CT_HELPER_NAME]);
1279	if (len > 16 || name[len - 1] != '\0') {
1280	NL_SET_ERR_MSG_MOD(extack, "Failed to parse helper name.");
1281	err = -EINVAL;
1282	goto err;
1283	}
1284	family = tb[TCA_CT_HELPER_FAMILY] ? nla_get_u8(nla: tb[TCA_CT_HELPER_FAMILY]) : AF_INET;
1285	proto = tb[TCA_CT_HELPER_PROTO] ? nla_get_u8(nla: tb[TCA_CT_HELPER_PROTO]) : IPPROTO_TCP;
1286	err = nf_ct_add_helper(ct: tmpl, name, family, proto,
1287	nat: p->ct_action & TCA_CT_ACT_NAT, hp: &p->helper);
1288	if (err) {
1289	NL_SET_ERR_MSG_MOD(extack, "Failed to add helper");
1290	goto err;
1291	}
1292	}
1293
1294	p->put_labels = put_labels;
1295
1296	if (p->ct_action & TCA_CT_ACT_COMMIT)
1297	__set_bit(IPS_CONFIRMED_BIT, &tmpl->status);
1298	return 0;
1299	err:
1300	if (put_labels)
1301	nf_connlabels_put(net);
1302
1303	nf_ct_put(ct: p->tmpl);
1304	p->tmpl = NULL;
1305	return err;
1306	}
1307
1308	static int tcf_ct_init(struct net *net, struct nlattr *nla,
1309	struct nlattr *est, struct tc_action **a,
1310	struct tcf_proto *tp, u32 flags,
1311	struct netlink_ext_ack *extack)
1312	{
1313	struct tc_action_net *tn = net_generic(net, id: act_ct_ops.net_id);
1314	bool bind = flags & TCA_ACT_FLAGS_BIND;
1315	struct tcf_ct_params *params = NULL;
1316	struct nlattr *tb[TCA_CT_MAX + 1];
1317	struct tcf_chain *goto_ch = NULL;
1318	struct tc_ct *parm;
1319	struct tcf_ct *c;
1320	int err, res = 0;
1321	u32 index;
1322
1323	if (!nla) {
1324	NL_SET_ERR_MSG_MOD(extack, "Ct requires attributes to be passed");
1325	return -EINVAL;
1326	}
1327
1328	err = nla_parse_nested(tb, TCA_CT_MAX, nla, policy: ct_policy, extack);
1329	if (err < 0)
1330	return err;
1331
1332	if (!tb[TCA_CT_PARMS]) {
1333	NL_SET_ERR_MSG_MOD(extack, "Missing required ct parameters");
1334	return -EINVAL;
1335	}
1336	parm = nla_data(nla: tb[TCA_CT_PARMS]);
1337	index = parm->index;
1338	err = tcf_idr_check_alloc(tn, index: &index, a, bind);
1339	if (err < 0)
1340	return err;
1341
1342	if (!err) {
1343	err = tcf_idr_create_from_flags(tn, index, est, a,
1344	ops: &act_ct_ops, bind, flags);
1345	if (err) {
1346	tcf_idr_cleanup(tn, index);
1347	return err;
1348	}
1349	res = ACT_P_CREATED;
1350	} else {
1351	if (bind)
1352	return ACT_P_BOUND;
1353
1354	if (!(flags & TCA_ACT_FLAGS_REPLACE)) {
1355	tcf_idr_release(a: *a, bind);
1356	return -EEXIST;
1357	}
1358	}
1359	err = tcf_action_check_ctrlact(action: parm->action, tp, handle: &goto_ch, newchain: extack);
1360	if (err < 0)
1361	goto cleanup;
1362
1363	c = to_ct(*a);
1364
1365	params = kzalloc(size: sizeof(*params), GFP_KERNEL);
1366	if (unlikely(!params)) {
1367	err = -ENOMEM;
1368	goto cleanup;
1369	}
1370
1371	err = tcf_ct_fill_params(net, p: params, parm, tb, extack);
1372	if (err)
1373	goto cleanup;
1374
1375	err = tcf_ct_flow_table_get(net, params);
1376	if (err)
1377	goto cleanup;
1378
1379	spin_lock_bh(lock: &c->tcf_lock);
1380	goto_ch = tcf_action_set_ctrlact(a: *a, action: parm->action, newchain: goto_ch);
1381	params = rcu_replace_pointer(c->params, params,
1382	lockdep_is_held(&c->tcf_lock));
1383	spin_unlock_bh(lock: &c->tcf_lock);
1384
1385	if (goto_ch)
1386	tcf_chain_put_by_act(chain: goto_ch);
1387	if (params)
1388	call_rcu(head: &params->rcu, func: tcf_ct_params_free_rcu);
1389
1390	return res;
1391
1392	cleanup:
1393	if (goto_ch)
1394	tcf_chain_put_by_act(chain: goto_ch);
1395	if (params)
1396	tcf_ct_params_free(params);
1397	tcf_idr_release(a: *a, bind);
1398	return err;
1399	}
1400
1401	static void tcf_ct_cleanup(struct tc_action *a)
1402	{
1403	struct tcf_ct_params *params;
1404	struct tcf_ct *c = to_ct(a);
1405
1406	params = rcu_dereference_protected(c->params, 1);
1407	if (params)
1408	call_rcu(head: &params->rcu, func: tcf_ct_params_free_rcu);
1409	}
1410
1411	static int tcf_ct_dump_key_val(struct sk_buff *skb,
1412	void *val, int val_type,
1413	void *mask, int mask_type,
1414	int len)
1415	{
1416	int err;
1417
1418	if (mask && !memchr_inv(p: mask, c: 0, size: len))
1419	return 0;
1420
1421	err = nla_put(skb, attrtype: val_type, attrlen: len, data: val);
1422	if (err)
1423	return err;
1424
1425	if (mask_type != TCA_CT_UNSPEC) {
1426	err = nla_put(skb, attrtype: mask_type, attrlen: len, data: mask);
1427	if (err)
1428	return err;
1429	}
1430
1431	return 0;
1432	}
1433
1434	static int tcf_ct_dump_nat(struct sk_buff *skb, struct tcf_ct_params *p)
1435	{
1436	struct nf_nat_range2 *range = &p->range;
1437
1438	if (!(p->ct_action & TCA_CT_ACT_NAT))
1439	return 0;
1440
1441	if (!(p->ct_action & (TCA_CT_ACT_NAT_SRC | TCA_CT_ACT_NAT_DST)))
1442	return 0;
1443
1444	if (range->flags & NF_NAT_RANGE_MAP_IPS) {
1445	if (p->ipv4_range) {
1446	if (nla_put_in_addr(skb, attrtype: TCA_CT_NAT_IPV4_MIN,
1447	addr: range->min_addr.ip))
1448	return -1;
1449	if (nla_put_in_addr(skb, attrtype: TCA_CT_NAT_IPV4_MAX,
1450	addr: range->max_addr.ip))
1451	return -1;
1452	} else {
1453	if (nla_put_in6_addr(skb, attrtype: TCA_CT_NAT_IPV6_MIN,
1454	addr: &range->min_addr.in6))
1455	return -1;
1456	if (nla_put_in6_addr(skb, attrtype: TCA_CT_NAT_IPV6_MAX,
1457	addr: &range->max_addr.in6))
1458	return -1;
1459	}
1460	}
1461
1462	if (range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) {
1463	if (nla_put_be16(skb, attrtype: TCA_CT_NAT_PORT_MIN,
1464	value: range->min_proto.all))
1465	return -1;
1466	if (nla_put_be16(skb, attrtype: TCA_CT_NAT_PORT_MAX,
1467	value: range->max_proto.all))
1468	return -1;
1469	}
1470
1471	return 0;
1472	}
1473
1474	static int tcf_ct_dump_helper(struct sk_buff *skb, struct nf_conntrack_helper *helper)
1475	{
1476	if (!helper)
1477	return 0;
1478
1479	if (nla_put_string(skb, attrtype: TCA_CT_HELPER_NAME, str: helper->name) ||
1480	nla_put_u8(skb, attrtype: TCA_CT_HELPER_FAMILY, value: helper->tuple.src.l3num) ||
1481	nla_put_u8(skb, attrtype: TCA_CT_HELPER_PROTO, value: helper->tuple.dst.protonum))
1482	return -1;
1483
1484	return 0;
1485	}
1486
1487	static inline int tcf_ct_dump(struct sk_buff *skb, struct tc_action *a,
1488	int bind, int ref)
1489	{
1490	unsigned char *b = skb_tail_pointer(skb);
1491	struct tcf_ct *c = to_ct(a);
1492	struct tcf_ct_params *p;
1493
1494	struct tc_ct opt = {
1495	.index = c->tcf_index,
1496	.refcnt = refcount_read(r: &c->tcf_refcnt) - ref,
1497	.bindcnt = atomic_read(v: &c->tcf_bindcnt) - bind,
1498	};
1499	struct tcf_t t;
1500
1501	spin_lock_bh(lock: &c->tcf_lock);
1502	p = rcu_dereference_protected(c->params,
1503	lockdep_is_held(&c->tcf_lock));
1504	opt.action = c->tcf_action;
1505
1506	if (tcf_ct_dump_key_val(skb,
1507	val: &p->ct_action, val_type: TCA_CT_ACTION,
1508	NULL, mask_type: TCA_CT_UNSPEC,
1509	len: sizeof(p->ct_action)))
1510	goto nla_put_failure;
1511
1512	if (p->ct_action & TCA_CT_ACT_CLEAR)
1513	goto skip_dump;
1514
1515	if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1516	tcf_ct_dump_key_val(skb,
1517	val: &p->mark, val_type: TCA_CT_MARK,
1518	mask: &p->mark_mask, mask_type: TCA_CT_MARK_MASK,
1519	len: sizeof(p->mark)))
1520	goto nla_put_failure;
1521
1522	if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1523	tcf_ct_dump_key_val(skb,
1524	val: p->labels, val_type: TCA_CT_LABELS,
1525	mask: p->labels_mask, mask_type: TCA_CT_LABELS_MASK,
1526	len: sizeof(p->labels)))
1527	goto nla_put_failure;
1528
1529	if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1530	tcf_ct_dump_key_val(skb,
1531	val: &p->zone, val_type: TCA_CT_ZONE,
1532	NULL, mask_type: TCA_CT_UNSPEC,
1533	len: sizeof(p->zone)))
1534	goto nla_put_failure;
1535
1536	if (tcf_ct_dump_nat(skb, p))
1537	goto nla_put_failure;
1538
1539	if (tcf_ct_dump_helper(skb, helper: p->helper))
1540	goto nla_put_failure;
1541
1542	skip_dump:
1543	if (nla_put(skb, attrtype: TCA_CT_PARMS, attrlen: sizeof(opt), data: &opt))
1544	goto nla_put_failure;
1545
1546	tcf_tm_dump(dtm: &t, stm: &c->tcf_tm);
1547	if (nla_put_64bit(skb, attrtype: TCA_CT_TM, attrlen: sizeof(t), data: &t, padattr: TCA_CT_PAD))
1548	goto nla_put_failure;
1549	spin_unlock_bh(lock: &c->tcf_lock);
1550
1551	return skb->len;
1552	nla_put_failure:
1553	spin_unlock_bh(lock: &c->tcf_lock);
1554	nlmsg_trim(skb, mark: b);
1555	return -1;
1556	}
1557
1558	static void tcf_stats_update(struct tc_action *a, u64 bytes, u64 packets,
1559	u64 drops, u64 lastuse, bool hw)
1560	{
1561	struct tcf_ct *c = to_ct(a);
1562
1563	tcf_action_update_stats(a, bytes, packets, drops, hw);
1564	c->tcf_tm.lastuse = max_t(u64, c->tcf_tm.lastuse, lastuse);
1565	}
1566
1567	static int tcf_ct_offload_act_setup(struct tc_action *act, void *entry_data,
1568	u32 *index_inc, bool bind,
1569	struct netlink_ext_ack *extack)
1570	{
1571	if (bind) {
1572	struct flow_action_entry *entry = entry_data;
1573
1574	if (tcf_ct_helper(a: act))
1575	return -EOPNOTSUPP;
1576
1577	entry->id = FLOW_ACTION_CT;
1578	entry->ct.action = tcf_ct_action(a: act);
1579	entry->ct.zone = tcf_ct_zone(a: act);
1580	entry->ct.flow_table = tcf_ct_ft(a: act);
1581	*index_inc = 1;
1582	} else {
1583	struct flow_offload_action *fl_action = entry_data;
1584
1585	fl_action->id = FLOW_ACTION_CT;
1586	}
1587
1588	return 0;
1589	}
1590
1591	static struct tc_action_ops act_ct_ops = {
1592	.kind = "ct",
1593	.id = TCA_ID_CT,
1594	.owner = THIS_MODULE,
1595	.act = tcf_ct_act,
1596	.dump = tcf_ct_dump,
1597	.init = tcf_ct_init,
1598	.cleanup = tcf_ct_cleanup,
1599	.stats_update = tcf_stats_update,
1600	.offload_act_setup = tcf_ct_offload_act_setup,
1601	.size = sizeof(struct tcf_ct),
1602	};
1603	MODULE_ALIAS_NET_ACT("ct");
1604
1605	static __net_init int ct_init_net(struct net *net)
1606	{
1607	struct tc_ct_action_net *tn = net_generic(net, id: act_ct_ops.net_id);
1608
1609	return tc_action_net_init(net, tn: &tn->tn, ops: &act_ct_ops);
1610	}
1611
1612	static void __net_exit ct_exit_net(struct list_head *net_list)
1613	{
1614	tc_action_net_exit(net_list, id: act_ct_ops.net_id);
1615	}
1616
1617	static struct pernet_operations ct_net_ops = {
1618	.init = ct_init_net,
1619	.exit_batch = ct_exit_net,
1620	.id = &act_ct_ops.net_id,
1621	.size = sizeof(struct tc_ct_action_net),
1622	};
1623
1624	static int __init ct_init_module(void)
1625	{
1626	int err;
1627
1628	act_ct_wq = alloc_ordered_workqueue("act_ct_workqueue", 0);
1629	if (!act_ct_wq)
1630	return -ENOMEM;
1631
1632	err = tcf_ct_flow_tables_init();
1633	if (err)
1634	goto err_tbl_init;
1635
1636	err = tcf_register_action(a: &act_ct_ops, ops: &ct_net_ops);
1637	if (err)
1638	goto err_register;
1639
1640	static_branch_inc(&tcf_frag_xmit_count);
1641
1642	return 0;
1643
1644	err_register:
1645	tcf_ct_flow_tables_uninit();
1646	err_tbl_init:
1647	destroy_workqueue(wq: act_ct_wq);
1648	return err;
1649	}
1650
1651	static void __exit ct_cleanup_module(void)
1652	{
1653	static_branch_dec(&tcf_frag_xmit_count);
1654	tcf_unregister_action(a: &act_ct_ops, ops: &ct_net_ops);
1655	tcf_ct_flow_tables_uninit();
1656	destroy_workqueue(wq: act_ct_wq);
1657	}
1658
1659	module_init(ct_init_module);
1660	module_exit(ct_cleanup_module);
1661	MODULE_AUTHOR("Paul Blakey <paulb@mellanox.com>");
1662	MODULE_AUTHOR("Yossi Kuperman <yossiku@mellanox.com>");
1663	MODULE_AUTHOR("Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>");
1664	MODULE_DESCRIPTION("Connection tracking action");
1665	MODULE_LICENSE("GPL v2");
1666