1 // SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB
3 * net/sched/act_ct.c Connection Tracking action
5 * Authors: Paul Blakey <paulb@mellanox.com>
6 * Yossi Kuperman <yossiku@mellanox.com>
7 * Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
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>
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>
24 #include <net/ipv6_frag.h>
25 #include <uapi/linux/tc_act/tc_ct.h>
26 #include <net/tc_act/tc_ct.h>
28 #include <net/netfilter/nf_flow_table.h>
29 #include <net/netfilter/nf_conntrack.h>
30 #include <net/netfilter/nf_conntrack_core.h>
31 #include <net/netfilter/nf_conntrack_zones.h>
32 #include <net/netfilter/nf_conntrack_helper.h>
33 #include <net/netfilter/nf_conntrack_acct.h>
34 #include <net/netfilter/ipv6/nf_defrag_ipv6.h>
35 #include <uapi/linux/netfilter/nf_nat.h>
37 static struct workqueue_struct *act_ct_wq;
38 static struct rhashtable zones_ht;
39 static DEFINE_MUTEX(zones_mutex);
41 struct tcf_ct_flow_table {
42 struct rhash_head node; /* In zones tables */
44 struct rcu_work rwork;
45 struct nf_flowtable nf_ft;
52 static const struct rhashtable_params zones_params = {
53 .head_offset = offsetof(struct tcf_ct_flow_table, node),
54 .key_offset = offsetof(struct tcf_ct_flow_table, zone),
55 .key_len = sizeof_field(struct tcf_ct_flow_table, zone),
56 .automatic_shrinking = true,
59 static struct flow_action_entry *
60 tcf_ct_flow_table_flow_action_get_next(struct flow_action *flow_action)
62 int i = flow_action->num_entries++;
64 return &flow_action->entries[i];
67 static void tcf_ct_add_mangle_action(struct flow_action *action,
68 enum flow_action_mangle_base htype,
73 struct flow_action_entry *entry;
75 entry = tcf_ct_flow_table_flow_action_get_next(action);
76 entry->id = FLOW_ACTION_MANGLE;
77 entry->mangle.htype = htype;
78 entry->mangle.mask = ~mask;
79 entry->mangle.offset = offset;
80 entry->mangle.val = val;
83 /* The following nat helper functions check if the inverted reverse tuple
84 * (target) is different then the current dir tuple - meaning nat for ports
85 * and/or ip is needed, and add the relevant mangle actions.
88 tcf_ct_flow_table_add_action_nat_ipv4(const struct nf_conntrack_tuple *tuple,
89 struct nf_conntrack_tuple target,
90 struct flow_action *action)
92 if (memcmp(&target.src.u3, &tuple->src.u3, sizeof(target.src.u3)))
93 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_IP4,
94 offsetof(struct iphdr, saddr),
96 be32_to_cpu(target.src.u3.ip));
97 if (memcmp(&target.dst.u3, &tuple->dst.u3, sizeof(target.dst.u3)))
98 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_IP4,
99 offsetof(struct iphdr, daddr),
101 be32_to_cpu(target.dst.u3.ip));
105 tcf_ct_add_ipv6_addr_mangle_action(struct flow_action *action,
106 union nf_inet_addr *addr,
111 for (i = 0; i < sizeof(struct in6_addr) / sizeof(u32); i++)
112 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_IP6,
113 i * sizeof(u32) + offset,
114 0xFFFFFFFF, be32_to_cpu(addr->ip6[i]));
118 tcf_ct_flow_table_add_action_nat_ipv6(const struct nf_conntrack_tuple *tuple,
119 struct nf_conntrack_tuple target,
120 struct flow_action *action)
122 if (memcmp(&target.src.u3, &tuple->src.u3, sizeof(target.src.u3)))
123 tcf_ct_add_ipv6_addr_mangle_action(action, &target.src.u3,
124 offsetof(struct ipv6hdr,
126 if (memcmp(&target.dst.u3, &tuple->dst.u3, sizeof(target.dst.u3)))
127 tcf_ct_add_ipv6_addr_mangle_action(action, &target.dst.u3,
128 offsetof(struct ipv6hdr,
133 tcf_ct_flow_table_add_action_nat_tcp(const struct nf_conntrack_tuple *tuple,
134 struct nf_conntrack_tuple target,
135 struct flow_action *action)
137 __be16 target_src = target.src.u.tcp.port;
138 __be16 target_dst = target.dst.u.tcp.port;
140 if (target_src != tuple->src.u.tcp.port)
141 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_TCP,
142 offsetof(struct tcphdr, source),
143 0xFFFF, be16_to_cpu(target_src));
144 if (target_dst != tuple->dst.u.tcp.port)
145 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_TCP,
146 offsetof(struct tcphdr, dest),
147 0xFFFF, be16_to_cpu(target_dst));
151 tcf_ct_flow_table_add_action_nat_udp(const struct nf_conntrack_tuple *tuple,
152 struct nf_conntrack_tuple target,
153 struct flow_action *action)
155 __be16 target_src = target.src.u.udp.port;
156 __be16 target_dst = target.dst.u.udp.port;
158 if (target_src != tuple->src.u.udp.port)
159 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_TCP,
160 offsetof(struct udphdr, source),
161 0xFFFF, be16_to_cpu(target_src));
162 if (target_dst != tuple->dst.u.udp.port)
163 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_TCP,
164 offsetof(struct udphdr, dest),
165 0xFFFF, be16_to_cpu(target_dst));
168 static void tcf_ct_flow_table_add_action_meta(struct nf_conn *ct,
169 enum ip_conntrack_dir dir,
170 struct flow_action *action)
172 struct nf_conn_labels *ct_labels;
173 struct flow_action_entry *entry;
174 enum ip_conntrack_info ctinfo;
177 entry = tcf_ct_flow_table_flow_action_get_next(action);
178 entry->id = FLOW_ACTION_CT_METADATA;
179 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
180 entry->ct_metadata.mark = ct->mark;
182 ctinfo = dir == IP_CT_DIR_ORIGINAL ? IP_CT_ESTABLISHED :
183 IP_CT_ESTABLISHED_REPLY;
184 /* aligns with the CT reference on the SKB nf_ct_set */
185 entry->ct_metadata.cookie = (unsigned long)ct | ctinfo;
187 act_ct_labels = entry->ct_metadata.labels;
188 ct_labels = nf_ct_labels_find(ct);
190 memcpy(act_ct_labels, ct_labels->bits, NF_CT_LABELS_MAX_SIZE);
192 memset(act_ct_labels, 0, NF_CT_LABELS_MAX_SIZE);
195 static int tcf_ct_flow_table_add_action_nat(struct net *net,
197 enum ip_conntrack_dir dir,
198 struct flow_action *action)
200 const struct nf_conntrack_tuple *tuple = &ct->tuplehash[dir].tuple;
201 struct nf_conntrack_tuple target;
203 if (!(ct->status & IPS_NAT_MASK))
206 nf_ct_invert_tuple(&target, &ct->tuplehash[!dir].tuple);
208 switch (tuple->src.l3num) {
210 tcf_ct_flow_table_add_action_nat_ipv4(tuple, target,
214 tcf_ct_flow_table_add_action_nat_ipv6(tuple, target,
221 switch (nf_ct_protonum(ct)) {
223 tcf_ct_flow_table_add_action_nat_tcp(tuple, target, action);
226 tcf_ct_flow_table_add_action_nat_udp(tuple, target, action);
235 static int tcf_ct_flow_table_fill_actions(struct net *net,
236 const struct flow_offload *flow,
237 enum flow_offload_tuple_dir tdir,
238 struct nf_flow_rule *flow_rule)
240 struct flow_action *action = &flow_rule->rule->action;
241 int num_entries = action->num_entries;
242 struct nf_conn *ct = flow->ct;
243 enum ip_conntrack_dir dir;
247 case FLOW_OFFLOAD_DIR_ORIGINAL:
248 dir = IP_CT_DIR_ORIGINAL;
250 case FLOW_OFFLOAD_DIR_REPLY:
251 dir = IP_CT_DIR_REPLY;
257 err = tcf_ct_flow_table_add_action_nat(net, ct, dir, action);
261 tcf_ct_flow_table_add_action_meta(ct, dir, action);
265 /* Clear filled actions */
266 for (i = num_entries; i < action->num_entries; i++)
267 memset(&action->entries[i], 0, sizeof(action->entries[i]));
268 action->num_entries = num_entries;
273 static struct nf_flowtable_type flowtable_ct = {
274 .action = tcf_ct_flow_table_fill_actions,
275 .owner = THIS_MODULE,
278 static int tcf_ct_flow_table_get(struct tcf_ct_params *params)
280 struct tcf_ct_flow_table *ct_ft;
283 mutex_lock(&zones_mutex);
284 ct_ft = rhashtable_lookup_fast(&zones_ht, ¶ms->zone, zones_params);
285 if (ct_ft && refcount_inc_not_zero(&ct_ft->ref))
288 ct_ft = kzalloc(sizeof(*ct_ft), GFP_KERNEL);
291 refcount_set(&ct_ft->ref, 1);
293 ct_ft->zone = params->zone;
294 err = rhashtable_insert_fast(&zones_ht, &ct_ft->node, zones_params);
298 ct_ft->nf_ft.type = &flowtable_ct;
299 ct_ft->nf_ft.flags |= NF_FLOWTABLE_HW_OFFLOAD;
300 err = nf_flow_table_init(&ct_ft->nf_ft);
304 __module_get(THIS_MODULE);
306 params->ct_ft = ct_ft;
307 params->nf_ft = &ct_ft->nf_ft;
308 mutex_unlock(&zones_mutex);
313 rhashtable_remove_fast(&zones_ht, &ct_ft->node, zones_params);
317 mutex_unlock(&zones_mutex);
321 static void tcf_ct_flow_table_cleanup_work(struct work_struct *work)
323 struct tcf_ct_flow_table *ct_ft;
325 ct_ft = container_of(to_rcu_work(work), struct tcf_ct_flow_table,
327 nf_flow_table_free(&ct_ft->nf_ft);
330 module_put(THIS_MODULE);
333 static void tcf_ct_flow_table_put(struct tcf_ct_params *params)
335 struct tcf_ct_flow_table *ct_ft = params->ct_ft;
337 if (refcount_dec_and_test(¶ms->ct_ft->ref)) {
338 rhashtable_remove_fast(&zones_ht, &ct_ft->node, zones_params);
339 INIT_RCU_WORK(&ct_ft->rwork, tcf_ct_flow_table_cleanup_work);
340 queue_rcu_work(act_ct_wq, &ct_ft->rwork);
344 static void tcf_ct_flow_table_add(struct tcf_ct_flow_table *ct_ft,
348 struct flow_offload *entry;
351 if (test_and_set_bit(IPS_OFFLOAD_BIT, &ct->status))
354 entry = flow_offload_alloc(ct);
361 ct->proto.tcp.seen[0].flags |= IP_CT_TCP_FLAG_BE_LIBERAL;
362 ct->proto.tcp.seen[1].flags |= IP_CT_TCP_FLAG_BE_LIBERAL;
365 err = flow_offload_add(&ct_ft->nf_ft, entry);
372 flow_offload_free(entry);
374 clear_bit(IPS_OFFLOAD_BIT, &ct->status);
377 static void tcf_ct_flow_table_process_conn(struct tcf_ct_flow_table *ct_ft,
379 enum ip_conntrack_info ctinfo)
383 if (ctinfo != IP_CT_ESTABLISHED && ctinfo != IP_CT_ESTABLISHED_REPLY)
386 switch (nf_ct_protonum(ct)) {
389 if (ct->proto.tcp.state != TCP_CONNTRACK_ESTABLISHED)
398 if (nf_ct_ext_exist(ct, NF_CT_EXT_HELPER) ||
399 ct->status & IPS_SEQ_ADJUST)
402 tcf_ct_flow_table_add(ct_ft, ct, tcp);
406 tcf_ct_flow_table_fill_tuple_ipv4(struct sk_buff *skb,
407 struct flow_offload_tuple *tuple,
408 struct tcphdr **tcph)
410 struct flow_ports *ports;
414 if (!pskb_network_may_pull(skb, sizeof(*iph)))
418 thoff = iph->ihl * 4;
420 if (ip_is_fragment(iph) ||
421 unlikely(thoff != sizeof(struct iphdr)))
424 if (iph->protocol != IPPROTO_TCP &&
425 iph->protocol != IPPROTO_UDP)
431 if (!pskb_network_may_pull(skb, iph->protocol == IPPROTO_TCP ?
432 thoff + sizeof(struct tcphdr) :
433 thoff + sizeof(*ports)))
437 if (iph->protocol == IPPROTO_TCP)
438 *tcph = (void *)(skb_network_header(skb) + thoff);
440 ports = (struct flow_ports *)(skb_network_header(skb) + thoff);
441 tuple->src_v4.s_addr = iph->saddr;
442 tuple->dst_v4.s_addr = iph->daddr;
443 tuple->src_port = ports->source;
444 tuple->dst_port = ports->dest;
445 tuple->l3proto = AF_INET;
446 tuple->l4proto = iph->protocol;
452 tcf_ct_flow_table_fill_tuple_ipv6(struct sk_buff *skb,
453 struct flow_offload_tuple *tuple,
454 struct tcphdr **tcph)
456 struct flow_ports *ports;
457 struct ipv6hdr *ip6h;
460 if (!pskb_network_may_pull(skb, sizeof(*ip6h)))
463 ip6h = ipv6_hdr(skb);
465 if (ip6h->nexthdr != IPPROTO_TCP &&
466 ip6h->nexthdr != IPPROTO_UDP)
469 if (ip6h->hop_limit <= 1)
472 thoff = sizeof(*ip6h);
473 if (!pskb_network_may_pull(skb, ip6h->nexthdr == IPPROTO_TCP ?
474 thoff + sizeof(struct tcphdr) :
475 thoff + sizeof(*ports)))
478 ip6h = ipv6_hdr(skb);
479 if (ip6h->nexthdr == IPPROTO_TCP)
480 *tcph = (void *)(skb_network_header(skb) + thoff);
482 ports = (struct flow_ports *)(skb_network_header(skb) + thoff);
483 tuple->src_v6 = ip6h->saddr;
484 tuple->dst_v6 = ip6h->daddr;
485 tuple->src_port = ports->source;
486 tuple->dst_port = ports->dest;
487 tuple->l3proto = AF_INET6;
488 tuple->l4proto = ip6h->nexthdr;
493 static bool tcf_ct_flow_table_lookup(struct tcf_ct_params *p,
497 struct nf_flowtable *nf_ft = &p->ct_ft->nf_ft;
498 struct flow_offload_tuple_rhash *tuplehash;
499 struct flow_offload_tuple tuple = {};
500 enum ip_conntrack_info ctinfo;
501 struct tcphdr *tcph = NULL;
502 struct flow_offload *flow;
506 /* Previously seen or loopback */
507 ct = nf_ct_get(skb, &ctinfo);
508 if ((ct && !nf_ct_is_template(ct)) || ctinfo == IP_CT_UNTRACKED)
513 if (!tcf_ct_flow_table_fill_tuple_ipv4(skb, &tuple, &tcph))
517 if (!tcf_ct_flow_table_fill_tuple_ipv6(skb, &tuple, &tcph))
524 tuplehash = flow_offload_lookup(nf_ft, &tuple);
528 dir = tuplehash->tuple.dir;
529 flow = container_of(tuplehash, struct flow_offload, tuplehash[dir]);
532 if (tcph && (unlikely(tcph->fin || tcph->rst))) {
533 flow_offload_teardown(flow);
537 ctinfo = dir == FLOW_OFFLOAD_DIR_ORIGINAL ? IP_CT_ESTABLISHED :
538 IP_CT_ESTABLISHED_REPLY;
540 flow_offload_refresh(nf_ft, flow);
541 nf_conntrack_get(&ct->ct_general);
542 nf_ct_set(skb, ct, ctinfo);
543 nf_ct_acct_update(ct, dir, skb->len);
548 static int tcf_ct_flow_tables_init(void)
550 return rhashtable_init(&zones_ht, &zones_params);
553 static void tcf_ct_flow_tables_uninit(void)
555 rhashtable_destroy(&zones_ht);
558 static struct tc_action_ops act_ct_ops;
559 static unsigned int ct_net_id;
561 struct tc_ct_action_net {
562 struct tc_action_net tn; /* Must be first */
566 /* Determine whether skb->_nfct is equal to the result of conntrack lookup. */
567 static bool tcf_ct_skb_nfct_cached(struct net *net, struct sk_buff *skb,
568 u16 zone_id, bool force)
570 enum ip_conntrack_info ctinfo;
573 ct = nf_ct_get(skb, &ctinfo);
576 if (!net_eq(net, read_pnet(&ct->ct_net)))
578 if (nf_ct_zone(ct)->id != zone_id)
581 /* Force conntrack entry direction. */
582 if (force && CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) {
583 if (nf_ct_is_confirmed(ct))
586 nf_conntrack_put(&ct->ct_general);
587 nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
595 /* Trim the skb to the length specified by the IP/IPv6 header,
596 * removing any trailing lower-layer padding. This prepares the skb
597 * for higher-layer processing that assumes skb->len excludes padding
598 * (such as nf_ip_checksum). The caller needs to pull the skb to the
599 * network header, and ensure ip_hdr/ipv6_hdr points to valid data.
601 static int tcf_ct_skb_network_trim(struct sk_buff *skb, int family)
608 len = ntohs(ip_hdr(skb)->tot_len);
611 len = sizeof(struct ipv6hdr)
612 + ntohs(ipv6_hdr(skb)->payload_len);
618 err = pskb_trim_rcsum(skb, len);
623 static u8 tcf_ct_skb_nf_family(struct sk_buff *skb)
625 u8 family = NFPROTO_UNSPEC;
627 switch (skb_protocol(skb, true)) {
628 case htons(ETH_P_IP):
629 family = NFPROTO_IPV4;
631 case htons(ETH_P_IPV6):
632 family = NFPROTO_IPV6;
641 static int tcf_ct_ipv4_is_fragment(struct sk_buff *skb, bool *frag)
645 len = skb_network_offset(skb) + sizeof(struct iphdr);
646 if (unlikely(skb->len < len))
648 if (unlikely(!pskb_may_pull(skb, len)))
651 *frag = ip_is_fragment(ip_hdr(skb));
655 static int tcf_ct_ipv6_is_fragment(struct sk_buff *skb, bool *frag)
657 unsigned int flags = 0, len, payload_ofs = 0;
658 unsigned short frag_off;
661 len = skb_network_offset(skb) + sizeof(struct ipv6hdr);
662 if (unlikely(skb->len < len))
664 if (unlikely(!pskb_may_pull(skb, len)))
667 nexthdr = ipv6_find_hdr(skb, &payload_ofs, -1, &frag_off, &flags);
668 if (unlikely(nexthdr < 0))
671 *frag = flags & IP6_FH_F_FRAG;
675 static int tcf_ct_handle_fragments(struct net *net, struct sk_buff *skb,
676 u8 family, u16 zone, bool *defrag)
678 enum ip_conntrack_info ctinfo;
679 struct qdisc_skb_cb cb;
684 /* Previously seen (loopback)? Ignore. */
685 ct = nf_ct_get(skb, &ctinfo);
686 if ((ct && !nf_ct_is_template(ct)) || ctinfo == IP_CT_UNTRACKED)
689 if (family == NFPROTO_IPV4)
690 err = tcf_ct_ipv4_is_fragment(skb, &frag);
692 err = tcf_ct_ipv6_is_fragment(skb, &frag);
697 cb = *qdisc_skb_cb(skb);
699 if (family == NFPROTO_IPV4) {
700 enum ip_defrag_users user = IP_DEFRAG_CONNTRACK_IN + zone;
702 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
704 err = ip_defrag(net, skb, user);
706 if (err && err != -EINPROGRESS)
711 cb.mru = IPCB(skb)->frag_max_size;
713 } else { /* NFPROTO_IPV6 */
714 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
715 enum ip6_defrag_users user = IP6_DEFRAG_CONNTRACK_IN + zone;
717 memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm));
718 err = nf_ct_frag6_gather(net, skb, user);
719 if (err && err != -EINPROGRESS)
724 cb.mru = IP6CB(skb)->frag_max_size;
732 *qdisc_skb_cb(skb) = cb;
742 static void tcf_ct_params_free(struct rcu_head *head)
744 struct tcf_ct_params *params = container_of(head,
745 struct tcf_ct_params, rcu);
747 tcf_ct_flow_table_put(params);
750 nf_conntrack_put(¶ms->tmpl->ct_general);
754 #if IS_ENABLED(CONFIG_NF_NAT)
755 /* Modelled after nf_nat_ipv[46]_fn().
756 * range is only used for new, uninitialized NAT state.
757 * Returns either NF_ACCEPT or NF_DROP.
759 static int ct_nat_execute(struct sk_buff *skb, struct nf_conn *ct,
760 enum ip_conntrack_info ctinfo,
761 const struct nf_nat_range2 *range,
762 enum nf_nat_manip_type maniptype)
764 __be16 proto = skb_protocol(skb, true);
765 int hooknum, err = NF_ACCEPT;
767 /* See HOOK2MANIP(). */
768 if (maniptype == NF_NAT_MANIP_SRC)
769 hooknum = NF_INET_LOCAL_IN; /* Source NAT */
771 hooknum = NF_INET_LOCAL_OUT; /* Destination NAT */
775 case IP_CT_RELATED_REPLY:
776 if (proto == htons(ETH_P_IP) &&
777 ip_hdr(skb)->protocol == IPPROTO_ICMP) {
778 if (!nf_nat_icmp_reply_translation(skb, ct, ctinfo,
782 } else if (IS_ENABLED(CONFIG_IPV6) && proto == htons(ETH_P_IPV6)) {
784 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
785 int hdrlen = ipv6_skip_exthdr(skb,
786 sizeof(struct ipv6hdr),
787 &nexthdr, &frag_off);
789 if (hdrlen >= 0 && nexthdr == IPPROTO_ICMPV6) {
790 if (!nf_nat_icmpv6_reply_translation(skb, ct,
798 /* Non-ICMP, fall thru to initialize if needed. */
801 /* Seen it before? This can happen for loopback, retrans,
804 if (!nf_nat_initialized(ct, maniptype)) {
805 /* Initialize according to the NAT action. */
806 err = (range && range->flags & NF_NAT_RANGE_MAP_IPS)
807 /* Action is set up to establish a new
810 ? nf_nat_setup_info(ct, range, maniptype)
811 : nf_nat_alloc_null_binding(ct, hooknum);
812 if (err != NF_ACCEPT)
817 case IP_CT_ESTABLISHED:
818 case IP_CT_ESTABLISHED_REPLY:
826 err = nf_nat_packet(ct, ctinfo, hooknum, skb);
830 #endif /* CONFIG_NF_NAT */
832 static void tcf_ct_act_set_mark(struct nf_conn *ct, u32 mark, u32 mask)
834 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
840 new_mark = mark | (ct->mark & ~(mask));
841 if (ct->mark != new_mark) {
843 if (nf_ct_is_confirmed(ct))
844 nf_conntrack_event_cache(IPCT_MARK, ct);
849 static void tcf_ct_act_set_labels(struct nf_conn *ct,
853 #if IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS)
854 size_t labels_sz = sizeof_field(struct tcf_ct_params, labels);
856 if (!memchr_inv(labels_m, 0, labels_sz))
859 nf_connlabels_replace(ct, labels, labels_m, 4);
863 static int tcf_ct_act_nat(struct sk_buff *skb,
865 enum ip_conntrack_info ctinfo,
867 struct nf_nat_range2 *range,
870 #if IS_ENABLED(CONFIG_NF_NAT)
872 enum nf_nat_manip_type maniptype;
874 if (!(ct_action & TCA_CT_ACT_NAT))
877 /* Add NAT extension if not confirmed yet. */
878 if (!nf_ct_is_confirmed(ct) && !nf_ct_nat_ext_add(ct))
879 return NF_DROP; /* Can't NAT. */
881 if (ctinfo != IP_CT_NEW && (ct->status & IPS_NAT_MASK) &&
882 (ctinfo != IP_CT_RELATED || commit)) {
883 /* NAT an established or related connection like before. */
884 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY)
885 /* This is the REPLY direction for a connection
886 * for which NAT was applied in the forward
887 * direction. Do the reverse NAT.
889 maniptype = ct->status & IPS_SRC_NAT
890 ? NF_NAT_MANIP_DST : NF_NAT_MANIP_SRC;
892 maniptype = ct->status & IPS_SRC_NAT
893 ? NF_NAT_MANIP_SRC : NF_NAT_MANIP_DST;
894 } else if (ct_action & TCA_CT_ACT_NAT_SRC) {
895 maniptype = NF_NAT_MANIP_SRC;
896 } else if (ct_action & TCA_CT_ACT_NAT_DST) {
897 maniptype = NF_NAT_MANIP_DST;
902 err = ct_nat_execute(skb, ct, ctinfo, range, maniptype);
903 if (err == NF_ACCEPT &&
904 ct->status & IPS_SRC_NAT && ct->status & IPS_DST_NAT) {
905 if (maniptype == NF_NAT_MANIP_SRC)
906 maniptype = NF_NAT_MANIP_DST;
908 maniptype = NF_NAT_MANIP_SRC;
910 err = ct_nat_execute(skb, ct, ctinfo, range, maniptype);
918 static int tcf_ct_act(struct sk_buff *skb, const struct tc_action *a,
919 struct tcf_result *res)
921 struct net *net = dev_net(skb->dev);
922 bool cached, commit, clear, force;
923 enum ip_conntrack_info ctinfo;
924 struct tcf_ct *c = to_ct(a);
925 struct nf_conn *tmpl = NULL;
926 struct nf_hook_state state;
927 int nh_ofs, err, retval;
928 struct tcf_ct_params *p;
929 bool skip_add = false;
934 p = rcu_dereference_bh(c->params);
936 retval = READ_ONCE(c->tcf_action);
937 commit = p->ct_action & TCA_CT_ACT_COMMIT;
938 clear = p->ct_action & TCA_CT_ACT_CLEAR;
939 force = p->ct_action & TCA_CT_ACT_FORCE;
942 tcf_lastuse_update(&c->tcf_tm);
945 ct = nf_ct_get(skb, &ctinfo);
947 nf_conntrack_put(&ct->ct_general);
948 nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
954 family = tcf_ct_skb_nf_family(skb);
955 if (family == NFPROTO_UNSPEC)
958 /* The conntrack module expects to be working at L3.
959 * We also try to pull the IPv4/6 header to linear area
961 nh_ofs = skb_network_offset(skb);
962 skb_pull_rcsum(skb, nh_ofs);
963 err = tcf_ct_handle_fragments(net, skb, family, p->zone, &defrag);
964 if (err == -EINPROGRESS) {
965 retval = TC_ACT_STOLEN;
971 err = tcf_ct_skb_network_trim(skb, family);
975 /* If we are recirculating packets to match on ct fields and
976 * committing with a separate ct action, then we don't need to
977 * actually run the packet through conntrack twice unless it's for a
980 cached = tcf_ct_skb_nfct_cached(net, skb, p->zone, force);
982 if (!commit && tcf_ct_flow_table_lookup(p, skb, family)) {
987 /* Associate skb with specified zone. */
989 ct = nf_ct_get(skb, &ctinfo);
991 nf_conntrack_put(skb_nfct(skb));
992 nf_conntrack_get(&tmpl->ct_general);
993 nf_ct_set(skb, tmpl, IP_CT_NEW);
996 state.hook = NF_INET_PRE_ROUTING;
999 err = nf_conntrack_in(skb, &state);
1000 if (err != NF_ACCEPT)
1005 ct = nf_ct_get(skb, &ctinfo);
1008 nf_ct_deliver_cached_events(ct);
1010 err = tcf_ct_act_nat(skb, ct, ctinfo, p->ct_action, &p->range, commit);
1011 if (err != NF_ACCEPT)
1015 tcf_ct_act_set_mark(ct, p->mark, p->mark_mask);
1016 tcf_ct_act_set_labels(ct, p->labels, p->labels_mask);
1018 /* This will take care of sending queued events
1019 * even if the connection is already confirmed.
1021 nf_conntrack_confirm(skb);
1022 } else if (!skip_add) {
1023 tcf_ct_flow_table_process_conn(p->ct_ft, ct, ctinfo);
1027 skb_push_rcsum(skb, nh_ofs);
1030 tcf_action_update_bstats(&c->common, skb);
1032 qdisc_skb_cb(skb)->pkt_len = skb->len;
1036 tcf_action_inc_drop_qstats(&c->common);
1040 static const struct nla_policy ct_policy[TCA_CT_MAX + 1] = {
1041 [TCA_CT_ACTION] = { .type = NLA_U16 },
1042 [TCA_CT_PARMS] = { .type = NLA_EXACT_LEN, .len = sizeof(struct tc_ct) },
1043 [TCA_CT_ZONE] = { .type = NLA_U16 },
1044 [TCA_CT_MARK] = { .type = NLA_U32 },
1045 [TCA_CT_MARK_MASK] = { .type = NLA_U32 },
1046 [TCA_CT_LABELS] = { .type = NLA_BINARY,
1047 .len = 128 / BITS_PER_BYTE },
1048 [TCA_CT_LABELS_MASK] = { .type = NLA_BINARY,
1049 .len = 128 / BITS_PER_BYTE },
1050 [TCA_CT_NAT_IPV4_MIN] = { .type = NLA_U32 },
1051 [TCA_CT_NAT_IPV4_MAX] = { .type = NLA_U32 },
1052 [TCA_CT_NAT_IPV6_MIN] = { .type = NLA_EXACT_LEN,
1053 .len = sizeof(struct in6_addr) },
1054 [TCA_CT_NAT_IPV6_MAX] = { .type = NLA_EXACT_LEN,
1055 .len = sizeof(struct in6_addr) },
1056 [TCA_CT_NAT_PORT_MIN] = { .type = NLA_U16 },
1057 [TCA_CT_NAT_PORT_MAX] = { .type = NLA_U16 },
1060 static int tcf_ct_fill_params_nat(struct tcf_ct_params *p,
1063 struct netlink_ext_ack *extack)
1065 struct nf_nat_range2 *range;
1067 if (!(p->ct_action & TCA_CT_ACT_NAT))
1070 if (!IS_ENABLED(CONFIG_NF_NAT)) {
1071 NL_SET_ERR_MSG_MOD(extack, "Netfilter nat isn't enabled in kernel");
1075 if (!(p->ct_action & (TCA_CT_ACT_NAT_SRC | TCA_CT_ACT_NAT_DST)))
1078 if ((p->ct_action & TCA_CT_ACT_NAT_SRC) &&
1079 (p->ct_action & TCA_CT_ACT_NAT_DST)) {
1080 NL_SET_ERR_MSG_MOD(extack, "dnat and snat can't be enabled at the same time");
1085 if (tb[TCA_CT_NAT_IPV4_MIN]) {
1086 struct nlattr *max_attr = tb[TCA_CT_NAT_IPV4_MAX];
1088 p->ipv4_range = true;
1089 range->flags |= NF_NAT_RANGE_MAP_IPS;
1090 range->min_addr.ip =
1091 nla_get_in_addr(tb[TCA_CT_NAT_IPV4_MIN]);
1093 range->max_addr.ip = max_attr ?
1094 nla_get_in_addr(max_attr) :
1096 } else if (tb[TCA_CT_NAT_IPV6_MIN]) {
1097 struct nlattr *max_attr = tb[TCA_CT_NAT_IPV6_MAX];
1099 p->ipv4_range = false;
1100 range->flags |= NF_NAT_RANGE_MAP_IPS;
1101 range->min_addr.in6 =
1102 nla_get_in6_addr(tb[TCA_CT_NAT_IPV6_MIN]);
1104 range->max_addr.in6 = max_attr ?
1105 nla_get_in6_addr(max_attr) :
1106 range->min_addr.in6;
1109 if (tb[TCA_CT_NAT_PORT_MIN]) {
1110 range->flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1111 range->min_proto.all = nla_get_be16(tb[TCA_CT_NAT_PORT_MIN]);
1113 range->max_proto.all = tb[TCA_CT_NAT_PORT_MAX] ?
1114 nla_get_be16(tb[TCA_CT_NAT_PORT_MAX]) :
1115 range->min_proto.all;
1121 static void tcf_ct_set_key_val(struct nlattr **tb,
1122 void *val, int val_type,
1123 void *mask, int mask_type,
1128 nla_memcpy(val, tb[val_type], len);
1133 if (mask_type == TCA_CT_UNSPEC || !tb[mask_type])
1134 memset(mask, 0xff, len);
1136 nla_memcpy(mask, tb[mask_type], len);
1139 static int tcf_ct_fill_params(struct net *net,
1140 struct tcf_ct_params *p,
1143 struct netlink_ext_ack *extack)
1145 struct tc_ct_action_net *tn = net_generic(net, ct_net_id);
1146 struct nf_conntrack_zone zone;
1147 struct nf_conn *tmpl;
1150 p->zone = NF_CT_DEFAULT_ZONE_ID;
1152 tcf_ct_set_key_val(tb,
1153 &p->ct_action, TCA_CT_ACTION,
1154 NULL, TCA_CT_UNSPEC,
1155 sizeof(p->ct_action));
1157 if (p->ct_action & TCA_CT_ACT_CLEAR)
1160 err = tcf_ct_fill_params_nat(p, parm, tb, extack);
1164 if (tb[TCA_CT_MARK]) {
1165 if (!IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)) {
1166 NL_SET_ERR_MSG_MOD(extack, "Conntrack mark isn't enabled.");
1169 tcf_ct_set_key_val(tb,
1170 &p->mark, TCA_CT_MARK,
1171 &p->mark_mask, TCA_CT_MARK_MASK,
1175 if (tb[TCA_CT_LABELS]) {
1176 if (!IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS)) {
1177 NL_SET_ERR_MSG_MOD(extack, "Conntrack labels isn't enabled.");
1182 NL_SET_ERR_MSG_MOD(extack, "Failed to set connlabel length");
1185 tcf_ct_set_key_val(tb,
1186 p->labels, TCA_CT_LABELS,
1187 p->labels_mask, TCA_CT_LABELS_MASK,
1191 if (tb[TCA_CT_ZONE]) {
1192 if (!IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES)) {
1193 NL_SET_ERR_MSG_MOD(extack, "Conntrack zones isn't enabled.");
1197 tcf_ct_set_key_val(tb,
1198 &p->zone, TCA_CT_ZONE,
1199 NULL, TCA_CT_UNSPEC,
1203 if (p->zone == NF_CT_DEFAULT_ZONE_ID)
1206 nf_ct_zone_init(&zone, p->zone, NF_CT_DEFAULT_ZONE_DIR, 0);
1207 tmpl = nf_ct_tmpl_alloc(net, &zone, GFP_KERNEL);
1209 NL_SET_ERR_MSG_MOD(extack, "Failed to allocate conntrack template");
1212 __set_bit(IPS_CONFIRMED_BIT, &tmpl->status);
1213 nf_conntrack_get(&tmpl->ct_general);
1219 static int tcf_ct_init(struct net *net, struct nlattr *nla,
1220 struct nlattr *est, struct tc_action **a,
1221 int replace, int bind, bool rtnl_held,
1222 struct tcf_proto *tp, u32 flags,
1223 struct netlink_ext_ack *extack)
1225 struct tc_action_net *tn = net_generic(net, ct_net_id);
1226 struct tcf_ct_params *params = NULL;
1227 struct nlattr *tb[TCA_CT_MAX + 1];
1228 struct tcf_chain *goto_ch = NULL;
1235 NL_SET_ERR_MSG_MOD(extack, "Ct requires attributes to be passed");
1239 err = nla_parse_nested(tb, TCA_CT_MAX, nla, ct_policy, extack);
1243 if (!tb[TCA_CT_PARMS]) {
1244 NL_SET_ERR_MSG_MOD(extack, "Missing required ct parameters");
1247 parm = nla_data(tb[TCA_CT_PARMS]);
1248 index = parm->index;
1249 err = tcf_idr_check_alloc(tn, &index, a, bind);
1254 err = tcf_idr_create_from_flags(tn, index, est, a,
1255 &act_ct_ops, bind, flags);
1257 tcf_idr_cleanup(tn, index);
1260 res = ACT_P_CREATED;
1266 tcf_idr_release(*a, bind);
1270 err = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack);
1276 params = kzalloc(sizeof(*params), GFP_KERNEL);
1277 if (unlikely(!params)) {
1282 err = tcf_ct_fill_params(net, params, parm, tb, extack);
1286 err = tcf_ct_flow_table_get(params);
1290 spin_lock_bh(&c->tcf_lock);
1291 goto_ch = tcf_action_set_ctrlact(*a, parm->action, goto_ch);
1292 params = rcu_replace_pointer(c->params, params,
1293 lockdep_is_held(&c->tcf_lock));
1294 spin_unlock_bh(&c->tcf_lock);
1297 tcf_chain_put_by_act(goto_ch);
1299 call_rcu(¶ms->rcu, tcf_ct_params_free);
1300 if (res == ACT_P_CREATED)
1301 tcf_idr_insert(tn, *a);
1307 tcf_chain_put_by_act(goto_ch);
1309 tcf_idr_release(*a, bind);
1313 static void tcf_ct_cleanup(struct tc_action *a)
1315 struct tcf_ct_params *params;
1316 struct tcf_ct *c = to_ct(a);
1318 params = rcu_dereference_protected(c->params, 1);
1320 call_rcu(¶ms->rcu, tcf_ct_params_free);
1323 static int tcf_ct_dump_key_val(struct sk_buff *skb,
1324 void *val, int val_type,
1325 void *mask, int mask_type,
1330 if (mask && !memchr_inv(mask, 0, len))
1333 err = nla_put(skb, val_type, len, val);
1337 if (mask_type != TCA_CT_UNSPEC) {
1338 err = nla_put(skb, mask_type, len, mask);
1346 static int tcf_ct_dump_nat(struct sk_buff *skb, struct tcf_ct_params *p)
1348 struct nf_nat_range2 *range = &p->range;
1350 if (!(p->ct_action & TCA_CT_ACT_NAT))
1353 if (!(p->ct_action & (TCA_CT_ACT_NAT_SRC | TCA_CT_ACT_NAT_DST)))
1356 if (range->flags & NF_NAT_RANGE_MAP_IPS) {
1357 if (p->ipv4_range) {
1358 if (nla_put_in_addr(skb, TCA_CT_NAT_IPV4_MIN,
1359 range->min_addr.ip))
1361 if (nla_put_in_addr(skb, TCA_CT_NAT_IPV4_MAX,
1362 range->max_addr.ip))
1365 if (nla_put_in6_addr(skb, TCA_CT_NAT_IPV6_MIN,
1366 &range->min_addr.in6))
1368 if (nla_put_in6_addr(skb, TCA_CT_NAT_IPV6_MAX,
1369 &range->max_addr.in6))
1374 if (range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) {
1375 if (nla_put_be16(skb, TCA_CT_NAT_PORT_MIN,
1376 range->min_proto.all))
1378 if (nla_put_be16(skb, TCA_CT_NAT_PORT_MAX,
1379 range->max_proto.all))
1386 static inline int tcf_ct_dump(struct sk_buff *skb, struct tc_action *a,
1389 unsigned char *b = skb_tail_pointer(skb);
1390 struct tcf_ct *c = to_ct(a);
1391 struct tcf_ct_params *p;
1393 struct tc_ct opt = {
1394 .index = c->tcf_index,
1395 .refcnt = refcount_read(&c->tcf_refcnt) - ref,
1396 .bindcnt = atomic_read(&c->tcf_bindcnt) - bind,
1400 spin_lock_bh(&c->tcf_lock);
1401 p = rcu_dereference_protected(c->params,
1402 lockdep_is_held(&c->tcf_lock));
1403 opt.action = c->tcf_action;
1405 if (tcf_ct_dump_key_val(skb,
1406 &p->ct_action, TCA_CT_ACTION,
1407 NULL, TCA_CT_UNSPEC,
1408 sizeof(p->ct_action)))
1409 goto nla_put_failure;
1411 if (p->ct_action & TCA_CT_ACT_CLEAR)
1414 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1415 tcf_ct_dump_key_val(skb,
1416 &p->mark, TCA_CT_MARK,
1417 &p->mark_mask, TCA_CT_MARK_MASK,
1419 goto nla_put_failure;
1421 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1422 tcf_ct_dump_key_val(skb,
1423 p->labels, TCA_CT_LABELS,
1424 p->labels_mask, TCA_CT_LABELS_MASK,
1426 goto nla_put_failure;
1428 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1429 tcf_ct_dump_key_val(skb,
1430 &p->zone, TCA_CT_ZONE,
1431 NULL, TCA_CT_UNSPEC,
1433 goto nla_put_failure;
1435 if (tcf_ct_dump_nat(skb, p))
1436 goto nla_put_failure;
1439 if (nla_put(skb, TCA_CT_PARMS, sizeof(opt), &opt))
1440 goto nla_put_failure;
1442 tcf_tm_dump(&t, &c->tcf_tm);
1443 if (nla_put_64bit(skb, TCA_CT_TM, sizeof(t), &t, TCA_CT_PAD))
1444 goto nla_put_failure;
1445 spin_unlock_bh(&c->tcf_lock);
1449 spin_unlock_bh(&c->tcf_lock);
1454 static int tcf_ct_walker(struct net *net, struct sk_buff *skb,
1455 struct netlink_callback *cb, int type,
1456 const struct tc_action_ops *ops,
1457 struct netlink_ext_ack *extack)
1459 struct tc_action_net *tn = net_generic(net, ct_net_id);
1461 return tcf_generic_walker(tn, skb, cb, type, ops, extack);
1464 static int tcf_ct_search(struct net *net, struct tc_action **a, u32 index)
1466 struct tc_action_net *tn = net_generic(net, ct_net_id);
1468 return tcf_idr_search(tn, a, index);
1471 static void tcf_stats_update(struct tc_action *a, u64 bytes, u64 packets,
1472 u64 drops, u64 lastuse, bool hw)
1474 struct tcf_ct *c = to_ct(a);
1476 tcf_action_update_stats(a, bytes, packets, drops, hw);
1477 c->tcf_tm.lastuse = max_t(u64, c->tcf_tm.lastuse, lastuse);
1480 static struct tc_action_ops act_ct_ops = {
1483 .owner = THIS_MODULE,
1485 .dump = tcf_ct_dump,
1486 .init = tcf_ct_init,
1487 .cleanup = tcf_ct_cleanup,
1488 .walk = tcf_ct_walker,
1489 .lookup = tcf_ct_search,
1490 .stats_update = tcf_stats_update,
1491 .size = sizeof(struct tcf_ct),
1494 static __net_init int ct_init_net(struct net *net)
1496 unsigned int n_bits = sizeof_field(struct tcf_ct_params, labels) * 8;
1497 struct tc_ct_action_net *tn = net_generic(net, ct_net_id);
1499 if (nf_connlabels_get(net, n_bits - 1)) {
1501 pr_err("act_ct: Failed to set connlabels length");
1506 return tc_action_net_init(net, &tn->tn, &act_ct_ops);
1509 static void __net_exit ct_exit_net(struct list_head *net_list)
1514 list_for_each_entry(net, net_list, exit_list) {
1515 struct tc_ct_action_net *tn = net_generic(net, ct_net_id);
1518 nf_connlabels_put(net);
1522 tc_action_net_exit(net_list, ct_net_id);
1525 static struct pernet_operations ct_net_ops = {
1526 .init = ct_init_net,
1527 .exit_batch = ct_exit_net,
1529 .size = sizeof(struct tc_ct_action_net),
1532 static int __init ct_init_module(void)
1536 act_ct_wq = alloc_ordered_workqueue("act_ct_workqueue", 0);
1540 err = tcf_ct_flow_tables_init();
1544 err = tcf_register_action(&act_ct_ops, &ct_net_ops);
1551 tcf_ct_flow_tables_uninit();
1553 destroy_workqueue(act_ct_wq);
1557 static void __exit ct_cleanup_module(void)
1559 tcf_unregister_action(&act_ct_ops, &ct_net_ops);
1560 tcf_ct_flow_tables_uninit();
1561 destroy_workqueue(act_ct_wq);
1564 module_init(ct_init_module);
1565 module_exit(ct_cleanup_module);
1566 MODULE_AUTHOR("Paul Blakey <paulb@mellanox.com>");
1567 MODULE_AUTHOR("Yossi Kuperman <yossiku@mellanox.com>");
1568 MODULE_AUTHOR("Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>");
1569 MODULE_DESCRIPTION("Connection tracking action");
1570 MODULE_LICENSE("GPL v2");