2 * Copyright (c) 2015 Nicira, Inc.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of version 2 of the GNU General Public
6 * License as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
14 #include <linux/kconfig.h>
15 #include <linux/version.h>
17 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
19 #include <linux/module.h>
20 #include <linux/openvswitch.h>
21 #include <linux/tcp.h>
22 #include <linux/udp.h>
23 #include <linux/sctp.h>
25 #include <net/netfilter/nf_conntrack_core.h>
26 #include <net/netfilter/nf_conntrack_helper.h>
27 #include <net/netfilter/nf_conntrack_labels.h>
28 #include <net/netfilter/nf_conntrack_seqadj.h>
29 #include <net/netfilter/nf_conntrack_zones.h>
30 #include <net/netfilter/ipv6/nf_defrag_ipv6.h>
32 #ifdef CONFIG_NF_NAT_NEEDED
33 #include <linux/netfilter/nf_nat.h>
34 #include <net/netfilter/nf_nat_core.h>
35 #include <net/netfilter/nf_nat_l3proto.h>
39 #include "conntrack.h"
41 #include "flow_netlink.h"
44 struct ovs_ct_len_tbl {
49 /* Metadata mark for masked write to conntrack mark */
55 /* Metadata label for masked write to conntrack label. */
57 struct ovs_key_ct_labels value;
58 struct ovs_key_ct_labels mask;
62 OVS_CT_NAT = 1 << 0, /* NAT for committed connections only. */
63 OVS_CT_SRC_NAT = 1 << 1, /* Source NAT for NEW connections. */
64 OVS_CT_DST_NAT = 1 << 2, /* Destination NAT for NEW connections. */
67 /* Conntrack action context for execution. */
68 struct ovs_conntrack_info {
69 struct nf_conntrack_helper *helper;
70 struct nf_conntrack_zone zone;
73 u8 nat : 3; /* enum ovs_ct_nat */
74 u8 random_fully_compat : 1; /* bool */
77 struct md_labels labels;
78 #ifdef CONFIG_NF_NAT_NEEDED
79 struct nf_nat_range range; /* Only present for SRC NAT and DST NAT. */
83 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info);
85 static u16 key_to_nfproto(const struct sw_flow_key *key)
87 switch (ntohs(key->eth.type)) {
93 return NFPROTO_UNSPEC;
97 /* Map SKB connection state into the values used by flow definition. */
98 static u8 ovs_ct_get_state(enum ip_conntrack_info ctinfo)
100 u8 ct_state = OVS_CS_F_TRACKED;
103 case IP_CT_ESTABLISHED_REPLY:
104 case IP_CT_RELATED_REPLY:
105 ct_state |= OVS_CS_F_REPLY_DIR;
112 case IP_CT_ESTABLISHED:
113 case IP_CT_ESTABLISHED_REPLY:
114 ct_state |= OVS_CS_F_ESTABLISHED;
117 case IP_CT_RELATED_REPLY:
118 ct_state |= OVS_CS_F_RELATED;
121 ct_state |= OVS_CS_F_NEW;
130 static u32 ovs_ct_get_mark(const struct nf_conn *ct)
132 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
133 return ct ? ct->mark : 0;
139 static void ovs_ct_get_labels(const struct nf_conn *ct,
140 struct ovs_key_ct_labels *labels)
142 struct nf_conn_labels *cl = ct ? nf_ct_labels_find(ct) : NULL;
145 size_t len = cl->words * sizeof(long);
147 if (len > OVS_CT_LABELS_LEN)
148 len = OVS_CT_LABELS_LEN;
149 else if (len < OVS_CT_LABELS_LEN)
150 memset(labels, 0, OVS_CT_LABELS_LEN);
151 memcpy(labels, cl->bits, len);
153 memset(labels, 0, OVS_CT_LABELS_LEN);
157 static void __ovs_ct_update_key(struct sw_flow_key *key, u8 state,
158 const struct nf_conntrack_zone *zone,
159 const struct nf_conn *ct)
161 key->ct.state = state;
162 key->ct.zone = zone->id;
163 key->ct.mark = ovs_ct_get_mark(ct);
164 ovs_ct_get_labels(ct, &key->ct.labels);
167 /* Update 'key' based on skb->nfct. If 'post_ct' is true, then OVS has
168 * previously sent the packet to conntrack via the ct action. If
169 * 'keep_nat_flags' is true, the existing NAT flags retained, else they are
170 * initialized from the connection status.
172 static void ovs_ct_update_key(const struct sk_buff *skb,
173 const struct ovs_conntrack_info *info,
174 struct sw_flow_key *key, bool post_ct,
177 const struct nf_conntrack_zone *zone = &nf_ct_zone_dflt;
178 enum ip_conntrack_info ctinfo;
182 ct = nf_ct_get(skb, &ctinfo);
184 state = ovs_ct_get_state(ctinfo);
185 /* All unconfirmed entries are NEW connections. */
186 if (!nf_ct_is_confirmed(ct))
187 state |= OVS_CS_F_NEW;
188 /* OVS persists the related flag for the duration of the
192 state |= OVS_CS_F_RELATED;
193 if (keep_nat_flags) {
194 state |= key->ct.state & OVS_CS_F_NAT_MASK;
196 if (ct->status & IPS_SRC_NAT)
197 state |= OVS_CS_F_SRC_NAT;
198 if (ct->status & IPS_DST_NAT)
199 state |= OVS_CS_F_DST_NAT;
201 zone = nf_ct_zone(ct);
202 } else if (post_ct) {
203 state = OVS_CS_F_TRACKED | OVS_CS_F_INVALID;
207 __ovs_ct_update_key(key, state, zone, ct);
210 /* This is called to initialize CT key fields possibly coming in from the local
213 void ovs_ct_fill_key(const struct sk_buff *skb, struct sw_flow_key *key)
215 ovs_ct_update_key(skb, NULL, key, false, false);
218 int ovs_ct_put_key(const struct sw_flow_key *key, struct sk_buff *skb)
220 if (nla_put_u32(skb, OVS_KEY_ATTR_CT_STATE, key->ct.state))
223 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
224 nla_put_u16(skb, OVS_KEY_ATTR_CT_ZONE, key->ct.zone))
227 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
228 nla_put_u32(skb, OVS_KEY_ATTR_CT_MARK, key->ct.mark))
231 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
232 nla_put(skb, OVS_KEY_ATTR_CT_LABELS, sizeof(key->ct.labels),
239 static int ovs_ct_set_mark(struct sk_buff *skb, struct sw_flow_key *key,
240 u32 ct_mark, u32 mask)
242 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
243 enum ip_conntrack_info ctinfo;
247 /* The connection could be invalid, in which case set_mark is no-op. */
248 ct = nf_ct_get(skb, &ctinfo);
252 new_mark = ct_mark | (ct->mark & ~(mask));
253 if (ct->mark != new_mark) {
255 nf_conntrack_event_cache(IPCT_MARK, ct);
256 key->ct.mark = new_mark;
265 static int ovs_ct_set_labels(struct sk_buff *skb, struct sw_flow_key *key,
266 const struct ovs_key_ct_labels *labels,
267 const struct ovs_key_ct_labels *mask)
269 enum ip_conntrack_info ctinfo;
270 struct nf_conn_labels *cl;
274 /* The connection could be invalid, in which case set_label is no-op.*/
275 ct = nf_ct_get(skb, &ctinfo);
279 cl = nf_ct_labels_find(ct);
281 nf_ct_labels_ext_add(ct);
282 cl = nf_ct_labels_find(ct);
284 if (!cl || cl->words * sizeof(long) < OVS_CT_LABELS_LEN)
287 err = nf_connlabels_replace(ct, (u32 *)labels, (u32 *)mask,
288 OVS_CT_LABELS_LEN / sizeof(u32));
292 ovs_ct_get_labels(ct, &key->ct.labels);
296 /* 'skb' should already be pulled to nh_ofs. */
297 static int ovs_ct_helper(struct sk_buff *skb, u16 proto)
299 const struct nf_conntrack_helper *helper;
300 const struct nf_conn_help *help;
301 enum ip_conntrack_info ctinfo;
302 unsigned int protoff;
307 ct = nf_ct_get(skb, &ctinfo);
308 if (!ct || ctinfo == IP_CT_RELATED_REPLY)
311 help = nfct_help(ct);
315 helper = rcu_dereference(help->helper);
321 protoff = ip_hdrlen(skb);
327 nexthdr = ipv6_hdr(skb)->nexthdr;
328 ofs = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr,
330 if (ofs < 0 || (frag_off & htons(~0x7)) != 0) {
331 pr_debug("proto header not found\n");
338 WARN_ONCE(1, "helper invoked on non-IP family!");
342 #if LINUX_VERSION_CODE < KERNEL_VERSION(4,6,0)
343 /* Linux 4.5 and older depend on skb_dst being set when recalculating
344 * checksums after NAT helper has mangled TCP or UDP packet payload.
345 * This dependency is avoided when skb is CHECKSUM_PARTIAL or when UDP
348 * The dependency is not triggered when the main NAT code updates
349 * checksums after translating the IP header (address, port), so this
350 * fix only needs to be executed on packets that are both being NATted
351 * and that have a helper assigned.
353 if (ct->status & IPS_NAT_MASK && skb->ip_summed != CHECKSUM_PARTIAL) {
354 u8 ipproto = (proto == NFPROTO_IPV4)
355 ? ip_hdr(skb)->protocol : nexthdr;
360 offset = offsetof(struct tcphdr, check);
363 /* Skip if no csum. */
364 if (udp_hdr(skb)->check)
365 offset = offsetof(struct udphdr, check);
369 if (unlikely(!pskb_may_pull(skb, protoff + offset + 2)))
372 skb->csum_start = skb_headroom(skb) + protoff;
373 skb->csum_offset = offset;
374 skb->ip_summed = CHECKSUM_PARTIAL;
378 err = helper->help(skb, protoff, ct, ctinfo);
379 if (err != NF_ACCEPT)
382 /* Adjust seqs after helper. This is needed due to some helpers (e.g.,
383 * FTP with NAT) adusting the TCP payload size when mangling IP
384 * addresses and/or port numbers in the text-based control connection.
386 if (test_bit(IPS_SEQ_ADJUST_BIT, &ct->status) &&
387 !nf_ct_seq_adjust(skb, ct, ctinfo, protoff))
392 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
393 * value if 'skb' is freed.
395 static int handle_fragments(struct net *net, struct sw_flow_key *key,
396 u16 zone, struct sk_buff *skb)
398 struct ovs_gso_cb ovs_cb = *OVS_GSO_CB(skb);
402 OVS_NLERR(true, "%s: skb has no dev; dropping", __func__);
406 if (key->eth.type == htons(ETH_P_IP)) {
407 enum ip_defrag_users user = IP_DEFRAG_CONNTRACK_IN + zone;
409 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
410 err = ip_defrag(net, skb, user);
414 ovs_cb.dp_cb.mru = IPCB(skb)->frag_max_size;
415 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
416 } else if (key->eth.type == htons(ETH_P_IPV6)) {
417 enum ip6_defrag_users user = IP6_DEFRAG_CONNTRACK_IN + zone;
420 memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm));
421 err = nf_ct_frag6_gather(net, skb, user);
425 key->ip.proto = ipv6_hdr(skb)->nexthdr;
426 ovs_cb.dp_cb.mru = IP6CB(skb)->frag_max_size;
427 #endif /* IP frag support */
430 return -EPFNOSUPPORT;
433 key->ip.frag = OVS_FRAG_TYPE_NONE;
436 *OVS_GSO_CB(skb) = ovs_cb;
441 static struct nf_conntrack_expect *
442 ovs_ct_expect_find(struct net *net, const struct nf_conntrack_zone *zone,
443 u16 proto, const struct sk_buff *skb)
445 struct nf_conntrack_tuple tuple;
447 if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), proto, net, &tuple))
449 return __nf_ct_expect_find(net, zone, &tuple);
452 /* This replicates logic from nf_conntrack_core.c that is not exported. */
453 static enum ip_conntrack_info
454 ovs_ct_get_info(const struct nf_conntrack_tuple_hash *h)
456 const struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
458 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY)
459 return IP_CT_ESTABLISHED_REPLY;
460 /* Once we've had two way comms, always ESTABLISHED. */
461 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status))
462 return IP_CT_ESTABLISHED;
463 if (test_bit(IPS_EXPECTED_BIT, &ct->status))
464 return IP_CT_RELATED;
468 /* Find an existing connection which this packet belongs to without
469 * re-attributing statistics or modifying the connection state. This allows an
470 * skb->nfct lost due to an upcall to be recovered during actions execution.
472 * Must be called with rcu_read_lock.
474 * On success, populates skb->nfct and skb->nfctinfo, and returns the
475 * connection. Returns NULL if there is no existing entry.
477 static struct nf_conn *
478 ovs_ct_find_existing(struct net *net, const struct nf_conntrack_zone *zone,
479 u8 l3num, struct sk_buff *skb)
481 struct nf_conntrack_l3proto *l3proto;
482 struct nf_conntrack_l4proto *l4proto;
483 struct nf_conntrack_tuple tuple;
484 struct nf_conntrack_tuple_hash *h;
485 enum ip_conntrack_info ctinfo;
487 unsigned int dataoff;
490 l3proto = __nf_ct_l3proto_find(l3num);
491 if (l3proto->get_l4proto(skb, skb_network_offset(skb), &dataoff,
493 pr_debug("ovs_ct_find_existing: Can't get protonum\n");
496 l4proto = __nf_ct_l4proto_find(l3num, protonum);
497 if (!nf_ct_get_tuple(skb, skb_network_offset(skb), dataoff, l3num,
498 protonum, net, &tuple, l3proto, l4proto)) {
499 pr_debug("ovs_ct_find_existing: Can't get tuple\n");
503 /* look for tuple match */
504 h = nf_conntrack_find_get(net, zone, &tuple);
506 return NULL; /* Not found. */
508 ct = nf_ct_tuplehash_to_ctrack(h);
510 ctinfo = ovs_ct_get_info(h);
511 if (ctinfo == IP_CT_NEW) {
512 /* This should not happen. */
513 WARN_ONCE(1, "ovs_ct_find_existing: new packet for %p\n", ct);
515 skb->nfct = &ct->ct_general;
516 skb->nfctinfo = ctinfo;
520 /* Determine whether skb->nfct is equal to the result of conntrack lookup. */
521 static bool skb_nfct_cached(struct net *net,
522 const struct sw_flow_key *key,
523 const struct ovs_conntrack_info *info,
526 enum ip_conntrack_info ctinfo;
529 ct = nf_ct_get(skb, &ctinfo);
530 /* If no ct, check if we have evidence that an existing conntrack entry
531 * might be found for this skb. This happens when we lose a skb->nfct
532 * due to an upcall. If the connection was not confirmed, it is not
533 * cached and needs to be run through conntrack again.
535 if (!ct && key->ct.state & OVS_CS_F_TRACKED &&
536 !(key->ct.state & OVS_CS_F_INVALID) &&
537 key->ct.zone == info->zone.id)
538 ct = ovs_ct_find_existing(net, &info->zone, info->family, skb);
541 if (!net_eq(net, read_pnet(&ct->ct_net)))
543 if (!nf_ct_zone_equal_any(info->ct, nf_ct_zone(ct)))
546 struct nf_conn_help *help;
548 help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER);
549 if (help && rcu_access_pointer(help->helper) != info->helper)
556 #ifdef CONFIG_NF_NAT_NEEDED
557 /* Modelled after nf_nat_ipv[46]_fn().
558 * range is only used for new, uninitialized NAT state.
559 * Returns either NF_ACCEPT or NF_DROP.
561 static int ovs_ct_nat_execute(struct sk_buff *skb, struct nf_conn *ct,
562 enum ip_conntrack_info ctinfo,
563 const struct nf_nat_range *range,
564 enum nf_nat_manip_type maniptype)
566 int hooknum, nh_off, err = NF_ACCEPT;
568 nh_off = skb_network_offset(skb);
569 skb_pull(skb, nh_off);
571 /* See HOOK2MANIP(). */
572 if (maniptype == NF_NAT_MANIP_SRC)
573 hooknum = NF_INET_LOCAL_IN; /* Source NAT */
575 hooknum = NF_INET_LOCAL_OUT; /* Destination NAT */
579 case IP_CT_RELATED_REPLY:
580 if (IS_ENABLED(CONFIG_NF_NAT_IPV4) &&
581 skb->protocol == htons(ETH_P_IP) &&
582 ip_hdr(skb)->protocol == IPPROTO_ICMP) {
583 if (!nf_nat_icmp_reply_translation(skb, ct, ctinfo,
587 } else if (IS_ENABLED(CONFIG_NF_NAT_IPV6) &&
588 skb->protocol == htons(ETH_P_IPV6)) {
590 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
591 int hdrlen = ipv6_skip_exthdr(skb,
592 sizeof(struct ipv6hdr),
593 &nexthdr, &frag_off);
595 if (hdrlen >= 0 && nexthdr == IPPROTO_ICMPV6) {
596 if (!nf_nat_icmpv6_reply_translation(skb, ct,
604 /* Non-ICMP, fall thru to initialize if needed. */
606 /* Seen it before? This can happen for loopback, retrans,
609 if (!nf_nat_initialized(ct, maniptype)) {
610 /* Initialize according to the NAT action. */
611 err = (range && range->flags & NF_NAT_RANGE_MAP_IPS)
612 /* Action is set up to establish a new
615 ? nf_nat_setup_info(ct, range, maniptype)
616 : nf_nat_alloc_null_binding(ct, hooknum);
617 if (err != NF_ACCEPT)
622 case IP_CT_ESTABLISHED:
623 case IP_CT_ESTABLISHED_REPLY:
631 err = nf_nat_packet(ct, ctinfo, hooknum, skb);
633 skb_push(skb, nh_off);
638 static void ovs_nat_update_key(struct sw_flow_key *key,
639 const struct sk_buff *skb,
640 enum nf_nat_manip_type maniptype)
642 if (maniptype == NF_NAT_MANIP_SRC) {
645 key->ct.state |= OVS_CS_F_SRC_NAT;
646 if (key->eth.type == htons(ETH_P_IP))
647 key->ipv4.addr.src = ip_hdr(skb)->saddr;
648 else if (key->eth.type == htons(ETH_P_IPV6))
649 memcpy(&key->ipv6.addr.src, &ipv6_hdr(skb)->saddr,
650 sizeof(key->ipv6.addr.src));
654 if (key->ip.proto == IPPROTO_UDP)
655 src = udp_hdr(skb)->source;
656 else if (key->ip.proto == IPPROTO_TCP)
657 src = tcp_hdr(skb)->source;
658 else if (key->ip.proto == IPPROTO_SCTP)
659 src = sctp_hdr(skb)->source;
667 key->ct.state |= OVS_CS_F_DST_NAT;
668 if (key->eth.type == htons(ETH_P_IP))
669 key->ipv4.addr.dst = ip_hdr(skb)->daddr;
670 else if (key->eth.type == htons(ETH_P_IPV6))
671 memcpy(&key->ipv6.addr.dst, &ipv6_hdr(skb)->daddr,
672 sizeof(key->ipv6.addr.dst));
676 if (key->ip.proto == IPPROTO_UDP)
677 dst = udp_hdr(skb)->dest;
678 else if (key->ip.proto == IPPROTO_TCP)
679 dst = tcp_hdr(skb)->dest;
680 else if (key->ip.proto == IPPROTO_SCTP)
681 dst = sctp_hdr(skb)->dest;
689 /* Returns NF_DROP if the packet should be dropped, NF_ACCEPT otherwise. */
690 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
691 const struct ovs_conntrack_info *info,
692 struct sk_buff *skb, struct nf_conn *ct,
693 enum ip_conntrack_info ctinfo)
695 enum nf_nat_manip_type maniptype;
698 if (nf_ct_is_untracked(ct)) {
699 /* A NAT action may only be performed on tracked packets. */
703 /* Add NAT extension if not confirmed yet. */
704 if (!nf_ct_is_confirmed(ct) && !nf_ct_nat_ext_add(ct))
705 return NF_ACCEPT; /* Can't NAT. */
707 /* Determine NAT type.
708 * Check if the NAT type can be deduced from the tracked connection.
709 * Make sure new expected connections (IP_CT_RELATED) are NATted only
712 if (info->nat & OVS_CT_NAT && ctinfo != IP_CT_NEW &&
713 ct->status & IPS_NAT_MASK &&
714 (ctinfo != IP_CT_RELATED || info->commit)) {
715 /* NAT an established or related connection like before. */
716 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY)
717 /* This is the REPLY direction for a connection
718 * for which NAT was applied in the forward
719 * direction. Do the reverse NAT.
721 maniptype = ct->status & IPS_SRC_NAT
722 ? NF_NAT_MANIP_DST : NF_NAT_MANIP_SRC;
724 maniptype = ct->status & IPS_SRC_NAT
725 ? NF_NAT_MANIP_SRC : NF_NAT_MANIP_DST;
726 } else if (info->nat & OVS_CT_SRC_NAT) {
727 maniptype = NF_NAT_MANIP_SRC;
728 } else if (info->nat & OVS_CT_DST_NAT) {
729 maniptype = NF_NAT_MANIP_DST;
731 return NF_ACCEPT; /* Connection is not NATed. */
733 err = ovs_ct_nat_execute(skb, ct, ctinfo, &info->range, maniptype);
735 /* Mark NAT done if successful and update the flow key. */
736 if (err == NF_ACCEPT)
737 ovs_nat_update_key(key, skb, maniptype);
741 #else /* !CONFIG_NF_NAT_NEEDED */
742 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
743 const struct ovs_conntrack_info *info,
744 struct sk_buff *skb, struct nf_conn *ct,
745 enum ip_conntrack_info ctinfo)
751 /* Pass 'skb' through conntrack in 'net', using zone configured in 'info', if
752 * not done already. Update key with new CT state after passing the packet
754 * Note that if the packet is deemed invalid by conntrack, skb->nfct will be
755 * set to NULL and 0 will be returned.
757 static int __ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
758 const struct ovs_conntrack_info *info,
761 /* If we are recirculating packets to match on conntrack fields and
762 * committing with a separate conntrack action, then we don't need to
763 * actually run the packet through conntrack twice unless it's for a
766 bool cached = skb_nfct_cached(net, key, info, skb);
767 enum ip_conntrack_info ctinfo;
771 struct nf_conn *tmpl = info->ct;
774 /* Associate skb with specified zone. */
777 nf_conntrack_put(skb->nfct);
778 nf_conntrack_get(&tmpl->ct_general);
779 skb->nfct = &tmpl->ct_general;
780 skb->nfctinfo = IP_CT_NEW;
783 /* Repeat if requested, see nf_iterate(). */
785 err = nf_conntrack_in(net, info->family,
786 NF_INET_FORWARD, skb);
787 } while (err == NF_REPEAT);
789 if (err != NF_ACCEPT)
792 /* Clear CT state NAT flags to mark that we have not yet done
793 * NAT after the nf_conntrack_in() call. We can actually clear
794 * the whole state, as it will be re-initialized below.
798 /* Update the key, but keep the NAT flags. */
799 ovs_ct_update_key(skb, info, key, true, true);
802 ct = nf_ct_get(skb, &ctinfo);
804 /* Packets starting a new connection must be NATted before the
805 * helper, so that the helper knows about the NAT. We enforce
806 * this by delaying both NAT and helper calls for unconfirmed
807 * connections until the committing CT action. For later
808 * packets NAT and Helper may be called in either order.
810 * NAT will be done only if the CT action has NAT, and only
811 * once per packet (per zone), as guarded by the NAT bits in
814 if (info->nat && !(key->ct.state & OVS_CS_F_NAT_MASK) &&
815 (nf_ct_is_confirmed(ct) || info->commit) &&
816 ovs_ct_nat(net, key, info, skb, ct, ctinfo) != NF_ACCEPT) {
820 /* Userspace may decide to perform a ct lookup without a helper
821 * specified followed by a (recirculate and) commit with one.
822 * Therefore, for unconfirmed connections which we will commit,
823 * we need to attach the helper here.
825 if (!nf_ct_is_confirmed(ct) && info->commit &&
826 info->helper && !nfct_help(ct)) {
827 int err = __nf_ct_try_assign_helper(ct, info->ct,
833 /* Call the helper only if:
834 * - nf_conntrack_in() was executed above ("!cached") for a
835 * confirmed connection, or
836 * - When committing an unconfirmed connection.
838 if ((nf_ct_is_confirmed(ct) ? !cached : info->commit) &&
839 ovs_ct_helper(skb, info->family) != NF_ACCEPT) {
847 /* Lookup connection and read fields into key. */
848 static int ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
849 const struct ovs_conntrack_info *info,
852 struct nf_conntrack_expect *exp;
854 /* If we pass an expected packet through nf_conntrack_in() the
855 * expectation is typically removed, but the packet could still be
856 * lost in upcall processing. To prevent this from happening we
857 * perform an explicit expectation lookup. Expected connections are
858 * always new, and will be passed through conntrack only when they are
859 * committed, as it is OK to remove the expectation at that time.
861 exp = ovs_ct_expect_find(net, &info->zone, info->family, skb);
865 /* NOTE: New connections are NATted and Helped only when
866 * committed, so we are not calling into NAT here.
868 state = OVS_CS_F_TRACKED | OVS_CS_F_NEW | OVS_CS_F_RELATED;
869 __ovs_ct_update_key(key, state, &info->zone, exp->master);
871 return __ovs_ct_lookup(net, key, info, skb);
876 /* Lookup connection and confirm if unconfirmed. */
877 static int ovs_ct_commit(struct net *net, struct sw_flow_key *key,
878 const struct ovs_conntrack_info *info,
883 err = __ovs_ct_lookup(net, key, info, skb);
886 /* This is a no-op if the connection has already been confirmed. */
887 if (nf_conntrack_confirm(skb) != NF_ACCEPT)
893 static bool labels_nonzero(const struct ovs_key_ct_labels *labels)
897 for (i = 0; i < sizeof(*labels); i++)
898 if (labels->ct_labels[i])
904 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
905 * value if 'skb' is freed.
907 int ovs_ct_execute(struct net *net, struct sk_buff *skb,
908 struct sw_flow_key *key,
909 const struct ovs_conntrack_info *info)
914 /* The conntrack module expects to be working at L3. */
915 nh_ofs = skb_network_offset(skb);
916 skb_pull(skb, nh_ofs);
918 if (key->ip.frag != OVS_FRAG_TYPE_NONE) {
919 err = handle_fragments(net, key, info->zone.id, skb);
925 err = ovs_ct_commit(net, key, info, skb);
927 err = ovs_ct_lookup(net, key, info, skb);
931 if (info->mark.mask) {
932 err = ovs_ct_set_mark(skb, key, info->mark.value,
937 if (labels_nonzero(&info->labels.mask))
938 err = ovs_ct_set_labels(skb, key, &info->labels.value,
941 skb_push(skb, nh_ofs);
947 static int ovs_ct_add_helper(struct ovs_conntrack_info *info, const char *name,
948 const struct sw_flow_key *key, bool log)
950 struct nf_conntrack_helper *helper;
951 struct nf_conn_help *help;
953 helper = nf_conntrack_helper_try_module_get(name, info->family,
956 OVS_NLERR(log, "Unknown helper \"%s\"", name);
960 help = nf_ct_helper_ext_add(info->ct, helper, GFP_KERNEL);
962 module_put(helper->me);
966 rcu_assign_pointer(help->helper, helper);
967 info->helper = helper;
971 #ifdef CONFIG_NF_NAT_NEEDED
972 static int parse_nat(const struct nlattr *attr,
973 struct ovs_conntrack_info *info, bool log)
977 bool have_ip_max = false;
978 bool have_proto_max = false;
979 bool ip_vers = (info->family == NFPROTO_IPV6);
981 nla_for_each_nested(a, attr, rem) {
982 static const int ovs_nat_attr_lens[OVS_NAT_ATTR_MAX + 1][2] = {
983 [OVS_NAT_ATTR_SRC] = {0, 0},
984 [OVS_NAT_ATTR_DST] = {0, 0},
985 [OVS_NAT_ATTR_IP_MIN] = {sizeof(struct in_addr),
986 sizeof(struct in6_addr)},
987 [OVS_NAT_ATTR_IP_MAX] = {sizeof(struct in_addr),
988 sizeof(struct in6_addr)},
989 [OVS_NAT_ATTR_PROTO_MIN] = {sizeof(u16), sizeof(u16)},
990 [OVS_NAT_ATTR_PROTO_MAX] = {sizeof(u16), sizeof(u16)},
991 [OVS_NAT_ATTR_PERSISTENT] = {0, 0},
992 [OVS_NAT_ATTR_PROTO_HASH] = {0, 0},
993 [OVS_NAT_ATTR_PROTO_RANDOM] = {0, 0},
995 int type = nla_type(a);
997 if (type > OVS_NAT_ATTR_MAX) {
999 "Unknown NAT attribute (type=%d, max=%d).\n",
1000 type, OVS_NAT_ATTR_MAX);
1004 if (nla_len(a) != ovs_nat_attr_lens[type][ip_vers]) {
1006 "NAT attribute type %d has unexpected length (%d != %d).\n",
1008 ovs_nat_attr_lens[type][ip_vers]);
1013 case OVS_NAT_ATTR_SRC:
1014 case OVS_NAT_ATTR_DST:
1017 "Only one type of NAT may be specified.\n"
1021 info->nat |= OVS_CT_NAT;
1022 info->nat |= ((type == OVS_NAT_ATTR_SRC)
1023 ? OVS_CT_SRC_NAT : OVS_CT_DST_NAT);
1026 case OVS_NAT_ATTR_IP_MIN:
1027 nla_memcpy(&info->range.min_addr, a,
1028 sizeof(info->range.min_addr));
1029 info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1032 case OVS_NAT_ATTR_IP_MAX:
1034 nla_memcpy(&info->range.max_addr, a,
1035 sizeof(info->range.max_addr));
1036 info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1039 case OVS_NAT_ATTR_PROTO_MIN:
1040 info->range.min_proto.all = htons(nla_get_u16(a));
1041 info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1044 case OVS_NAT_ATTR_PROTO_MAX:
1045 have_proto_max = true;
1046 info->range.max_proto.all = htons(nla_get_u16(a));
1047 info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1050 case OVS_NAT_ATTR_PERSISTENT:
1051 info->range.flags |= NF_NAT_RANGE_PERSISTENT;
1054 case OVS_NAT_ATTR_PROTO_HASH:
1055 info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM;
1058 case OVS_NAT_ATTR_PROTO_RANDOM:
1059 #ifdef NF_NAT_RANGE_PROTO_RANDOM_FULLY
1060 info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM_FULLY;
1062 info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM;
1063 info->random_fully_compat = true;
1068 OVS_NLERR(log, "Unknown nat attribute (%d).\n", type);
1074 OVS_NLERR(log, "NAT attribute has %d unknown bytes.\n", rem);
1078 /* Do not allow flags if no type is given. */
1079 if (info->range.flags) {
1081 "NAT flags may be given only when NAT range (SRC or DST) is also specified.\n"
1085 info->nat = OVS_CT_NAT; /* NAT existing connections. */
1086 } else if (!info->commit) {
1088 "NAT attributes may be specified only when CT COMMIT flag is also specified.\n"
1092 /* Allow missing IP_MAX. */
1093 if (info->range.flags & NF_NAT_RANGE_MAP_IPS && !have_ip_max) {
1094 memcpy(&info->range.max_addr, &info->range.min_addr,
1095 sizeof(info->range.max_addr));
1097 /* Allow missing PROTO_MAX. */
1098 if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1100 info->range.max_proto.all = info->range.min_proto.all;
1106 static const struct ovs_ct_len_tbl ovs_ct_attr_lens[OVS_CT_ATTR_MAX + 1] = {
1107 [OVS_CT_ATTR_COMMIT] = { .minlen = 0, .maxlen = 0 },
1108 [OVS_CT_ATTR_ZONE] = { .minlen = sizeof(u16),
1109 .maxlen = sizeof(u16) },
1110 [OVS_CT_ATTR_MARK] = { .minlen = sizeof(struct md_mark),
1111 .maxlen = sizeof(struct md_mark) },
1112 [OVS_CT_ATTR_LABELS] = { .minlen = sizeof(struct md_labels),
1113 .maxlen = sizeof(struct md_labels) },
1114 [OVS_CT_ATTR_HELPER] = { .minlen = 1,
1115 .maxlen = NF_CT_HELPER_NAME_LEN },
1116 #ifdef CONFIG_NF_NAT_NEEDED
1117 /* NAT length is checked when parsing the nested attributes. */
1118 [OVS_CT_ATTR_NAT] = { .minlen = 0, .maxlen = INT_MAX },
1122 static int parse_ct(const struct nlattr *attr, struct ovs_conntrack_info *info,
1123 const char **helper, bool log)
1128 nla_for_each_nested(a, attr, rem) {
1129 int type = nla_type(a);
1130 int maxlen = ovs_ct_attr_lens[type].maxlen;
1131 int minlen = ovs_ct_attr_lens[type].minlen;
1133 if (type > OVS_CT_ATTR_MAX) {
1135 "Unknown conntrack attr (type=%d, max=%d)",
1136 type, OVS_CT_ATTR_MAX);
1139 if (nla_len(a) < minlen || nla_len(a) > maxlen) {
1141 "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
1142 type, nla_len(a), maxlen);
1147 case OVS_CT_ATTR_COMMIT:
1148 info->commit = true;
1150 #ifdef CONFIG_NF_CONNTRACK_ZONES
1151 case OVS_CT_ATTR_ZONE:
1152 info->zone.id = nla_get_u16(a);
1155 #ifdef CONFIG_NF_CONNTRACK_MARK
1156 case OVS_CT_ATTR_MARK: {
1157 struct md_mark *mark = nla_data(a);
1160 OVS_NLERR(log, "ct_mark mask cannot be 0");
1167 #ifdef CONFIG_NF_CONNTRACK_LABELS
1168 case OVS_CT_ATTR_LABELS: {
1169 struct md_labels *labels = nla_data(a);
1171 if (!labels_nonzero(&labels->mask)) {
1172 OVS_NLERR(log, "ct_labels mask cannot be 0");
1175 info->labels = *labels;
1179 case OVS_CT_ATTR_HELPER:
1180 *helper = nla_data(a);
1181 if (!memchr(*helper, '\0', nla_len(a))) {
1182 OVS_NLERR(log, "Invalid conntrack helper");
1186 #ifdef CONFIG_NF_NAT_NEEDED
1187 case OVS_CT_ATTR_NAT: {
1188 int err = parse_nat(a, info, log);
1196 OVS_NLERR(log, "Unknown conntrack attr (%d)",
1203 OVS_NLERR(log, "Conntrack attr has %d unknown bytes", rem);
1210 bool ovs_ct_verify(struct net *net, enum ovs_key_attr attr)
1212 if (attr == OVS_KEY_ATTR_CT_STATE)
1214 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1215 attr == OVS_KEY_ATTR_CT_ZONE)
1217 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1218 attr == OVS_KEY_ATTR_CT_MARK)
1220 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1221 attr == OVS_KEY_ATTR_CT_LABELS) {
1222 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1224 return ovs_net->xt_label;
1230 int ovs_ct_copy_action(struct net *net, const struct nlattr *attr,
1231 const struct sw_flow_key *key,
1232 struct sw_flow_actions **sfa, bool log)
1234 struct ovs_conntrack_info ct_info;
1235 const char *helper = NULL;
1239 family = key_to_nfproto(key);
1240 if (family == NFPROTO_UNSPEC) {
1241 OVS_NLERR(log, "ct family unspecified");
1245 memset(&ct_info, 0, sizeof(ct_info));
1246 ct_info.family = family;
1248 nf_ct_zone_init(&ct_info.zone, NF_CT_DEFAULT_ZONE_ID,
1249 NF_CT_DEFAULT_ZONE_DIR, 0);
1251 err = parse_ct(attr, &ct_info, &helper, log);
1255 /* Set up template for tracking connections in specific zones. */
1256 ct_info.ct = nf_ct_tmpl_alloc(net, &ct_info.zone, GFP_KERNEL);
1258 OVS_NLERR(log, "Failed to allocate conntrack template");
1262 __set_bit(IPS_CONFIRMED_BIT, &ct_info.ct->status);
1263 nf_conntrack_get(&ct_info.ct->ct_general);
1266 err = ovs_ct_add_helper(&ct_info, helper, key, log);
1271 err = ovs_nla_add_action(sfa, OVS_ACTION_ATTR_CT, &ct_info,
1272 sizeof(ct_info), log);
1278 __ovs_ct_free_action(&ct_info);
1282 #ifdef CONFIG_NF_NAT_NEEDED
1283 static bool ovs_ct_nat_to_attr(const struct ovs_conntrack_info *info,
1284 struct sk_buff *skb)
1286 struct nlattr *start;
1288 start = nla_nest_start(skb, OVS_CT_ATTR_NAT);
1292 if (info->nat & OVS_CT_SRC_NAT) {
1293 if (nla_put_flag(skb, OVS_NAT_ATTR_SRC))
1295 } else if (info->nat & OVS_CT_DST_NAT) {
1296 if (nla_put_flag(skb, OVS_NAT_ATTR_DST))
1302 if (info->range.flags & NF_NAT_RANGE_MAP_IPS) {
1303 if (IS_ENABLED(CONFIG_NF_NAT_IPV4) &&
1304 info->family == NFPROTO_IPV4) {
1305 if (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MIN,
1306 info->range.min_addr.ip) ||
1307 (info->range.max_addr.ip
1308 != info->range.min_addr.ip &&
1309 (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MAX,
1310 info->range.max_addr.ip))))
1312 } else if (IS_ENABLED(CONFIG_NF_NAT_IPV6) &&
1313 info->family == NFPROTO_IPV6) {
1314 if (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MIN,
1315 &info->range.min_addr.in6) ||
1316 (memcmp(&info->range.max_addr.in6,
1317 &info->range.min_addr.in6,
1318 sizeof(info->range.max_addr.in6)) &&
1319 (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MAX,
1320 &info->range.max_addr.in6))))
1326 if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1327 (nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MIN,
1328 ntohs(info->range.min_proto.all)) ||
1329 (info->range.max_proto.all != info->range.min_proto.all &&
1330 nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MAX,
1331 ntohs(info->range.max_proto.all)))))
1334 if (info->range.flags & NF_NAT_RANGE_PERSISTENT &&
1335 nla_put_flag(skb, OVS_NAT_ATTR_PERSISTENT))
1337 if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM &&
1338 nla_put_flag(skb, info->random_fully_compat
1339 ? OVS_NAT_ATTR_PROTO_RANDOM
1340 : OVS_NAT_ATTR_PROTO_HASH))
1342 #ifdef NF_NAT_RANGE_PROTO_RANDOM_FULLY
1343 if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM_FULLY &&
1344 nla_put_flag(skb, OVS_NAT_ATTR_PROTO_RANDOM))
1348 nla_nest_end(skb, start);
1354 int ovs_ct_action_to_attr(const struct ovs_conntrack_info *ct_info,
1355 struct sk_buff *skb)
1357 struct nlattr *start;
1359 start = nla_nest_start(skb, OVS_ACTION_ATTR_CT);
1363 if (ct_info->commit && nla_put_flag(skb, OVS_CT_ATTR_COMMIT))
1365 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1366 nla_put_u16(skb, OVS_CT_ATTR_ZONE, ct_info->zone.id))
1368 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && ct_info->mark.mask &&
1369 nla_put(skb, OVS_CT_ATTR_MARK, sizeof(ct_info->mark),
1372 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1373 labels_nonzero(&ct_info->labels.mask) &&
1374 nla_put(skb, OVS_CT_ATTR_LABELS, sizeof(ct_info->labels),
1377 if (ct_info->helper) {
1378 if (nla_put_string(skb, OVS_CT_ATTR_HELPER,
1379 ct_info->helper->name))
1382 #ifdef CONFIG_NF_NAT_NEEDED
1383 if (ct_info->nat && !ovs_ct_nat_to_attr(ct_info, skb))
1386 nla_nest_end(skb, start);
1391 void ovs_ct_free_action(const struct nlattr *a)
1393 struct ovs_conntrack_info *ct_info = nla_data(a);
1395 __ovs_ct_free_action(ct_info);
1398 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info)
1400 if (ct_info->helper)
1401 module_put(ct_info->helper->me);
1403 nf_ct_tmpl_free(ct_info->ct);
1406 void ovs_ct_init(struct net *net)
1408 unsigned int n_bits = sizeof(struct ovs_key_ct_labels) * BITS_PER_BYTE;
1409 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1411 if (nf_connlabels_get(net, n_bits - 1)) {
1412 ovs_net->xt_label = false;
1413 OVS_NLERR(true, "Failed to set connlabel length");
1415 ovs_net->xt_label = true;
1419 void ovs_ct_exit(struct net *net)
1421 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1423 if (ovs_net->xt_label)
1424 nf_connlabels_put(net);
1427 #endif /* CONFIG_NF_CONNTRACK */