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/module.h>
15 #include <linux/openvswitch.h>
16 #include <linux/tcp.h>
17 #include <linux/udp.h>
18 #include <linux/sctp.h>
20 #include <net/netfilter/nf_conntrack_core.h>
21 #include <net/netfilter/nf_conntrack_helper.h>
22 #include <net/netfilter/nf_conntrack_labels.h>
23 #include <net/netfilter/nf_conntrack_seqadj.h>
24 #include <net/netfilter/nf_conntrack_zones.h>
25 #include <net/netfilter/ipv6/nf_defrag_ipv6.h>
27 #ifdef CONFIG_NF_NAT_NEEDED
28 #include <linux/netfilter/nf_nat.h>
29 #include <net/netfilter/nf_nat_core.h>
30 #include <net/netfilter/nf_nat_l3proto.h>
34 #include "conntrack.h"
36 #include "flow_netlink.h"
38 struct ovs_ct_len_tbl {
43 /* Metadata mark for masked write to conntrack mark */
49 /* Metadata label for masked write to conntrack label. */
51 struct ovs_key_ct_labels value;
52 struct ovs_key_ct_labels mask;
56 OVS_CT_NAT = 1 << 0, /* NAT for committed connections only. */
57 OVS_CT_SRC_NAT = 1 << 1, /* Source NAT for NEW connections. */
58 OVS_CT_DST_NAT = 1 << 2, /* Destination NAT for NEW connections. */
61 /* Conntrack action context for execution. */
62 struct ovs_conntrack_info {
63 struct nf_conntrack_helper *helper;
64 struct nf_conntrack_zone zone;
67 u8 nat : 3; /* enum ovs_ct_nat */
70 struct md_labels labels;
71 #ifdef CONFIG_NF_NAT_NEEDED
72 struct nf_nat_range range; /* Only present for SRC NAT and DST NAT. */
76 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info);
78 static u16 key_to_nfproto(const struct sw_flow_key *key)
80 switch (ntohs(key->eth.type)) {
86 return NFPROTO_UNSPEC;
90 /* Map SKB connection state into the values used by flow definition. */
91 static u8 ovs_ct_get_state(enum ip_conntrack_info ctinfo)
93 u8 ct_state = OVS_CS_F_TRACKED;
96 case IP_CT_ESTABLISHED_REPLY:
97 case IP_CT_RELATED_REPLY:
98 ct_state |= OVS_CS_F_REPLY_DIR;
105 case IP_CT_ESTABLISHED:
106 case IP_CT_ESTABLISHED_REPLY:
107 ct_state |= OVS_CS_F_ESTABLISHED;
110 case IP_CT_RELATED_REPLY:
111 ct_state |= OVS_CS_F_RELATED;
114 ct_state |= OVS_CS_F_NEW;
123 static u32 ovs_ct_get_mark(const struct nf_conn *ct)
125 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
126 return ct ? ct->mark : 0;
132 static void ovs_ct_get_labels(const struct nf_conn *ct,
133 struct ovs_key_ct_labels *labels)
135 struct nf_conn_labels *cl = ct ? nf_ct_labels_find(ct) : NULL;
138 size_t len = sizeof(cl->bits);
140 if (len > OVS_CT_LABELS_LEN)
141 len = OVS_CT_LABELS_LEN;
142 else if (len < OVS_CT_LABELS_LEN)
143 memset(labels, 0, OVS_CT_LABELS_LEN);
144 memcpy(labels, cl->bits, len);
146 memset(labels, 0, OVS_CT_LABELS_LEN);
150 static void __ovs_ct_update_key(struct sw_flow_key *key, u8 state,
151 const struct nf_conntrack_zone *zone,
152 const struct nf_conn *ct)
154 key->ct.state = state;
155 key->ct.zone = zone->id;
156 key->ct.mark = ovs_ct_get_mark(ct);
157 ovs_ct_get_labels(ct, &key->ct.labels);
160 /* Update 'key' based on skb->_nfct. If 'post_ct' is true, then OVS has
161 * previously sent the packet to conntrack via the ct action. If
162 * 'keep_nat_flags' is true, the existing NAT flags retained, else they are
163 * initialized from the connection status.
165 static void ovs_ct_update_key(const struct sk_buff *skb,
166 const struct ovs_conntrack_info *info,
167 struct sw_flow_key *key, bool post_ct,
170 const struct nf_conntrack_zone *zone = &nf_ct_zone_dflt;
171 enum ip_conntrack_info ctinfo;
175 ct = nf_ct_get(skb, &ctinfo);
177 state = ovs_ct_get_state(ctinfo);
178 /* All unconfirmed entries are NEW connections. */
179 if (!nf_ct_is_confirmed(ct))
180 state |= OVS_CS_F_NEW;
181 /* OVS persists the related flag for the duration of the
185 state |= OVS_CS_F_RELATED;
186 if (keep_nat_flags) {
187 state |= key->ct.state & OVS_CS_F_NAT_MASK;
189 if (ct->status & IPS_SRC_NAT)
190 state |= OVS_CS_F_SRC_NAT;
191 if (ct->status & IPS_DST_NAT)
192 state |= OVS_CS_F_DST_NAT;
194 zone = nf_ct_zone(ct);
195 } else if (post_ct) {
196 state = OVS_CS_F_TRACKED | OVS_CS_F_INVALID;
200 __ovs_ct_update_key(key, state, zone, ct);
203 /* This is called to initialize CT key fields possibly coming in from the local
206 void ovs_ct_fill_key(const struct sk_buff *skb, struct sw_flow_key *key)
208 ovs_ct_update_key(skb, NULL, key, false, false);
211 int ovs_ct_put_key(const struct sw_flow_key *key, struct sk_buff *skb)
213 if (nla_put_u32(skb, OVS_KEY_ATTR_CT_STATE, key->ct.state))
216 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
217 nla_put_u16(skb, OVS_KEY_ATTR_CT_ZONE, key->ct.zone))
220 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
221 nla_put_u32(skb, OVS_KEY_ATTR_CT_MARK, key->ct.mark))
224 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
225 nla_put(skb, OVS_KEY_ATTR_CT_LABELS, sizeof(key->ct.labels),
232 static int ovs_ct_set_mark(struct sk_buff *skb, struct sw_flow_key *key,
233 u32 ct_mark, u32 mask)
235 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
236 enum ip_conntrack_info ctinfo;
240 /* The connection could be invalid, in which case set_mark is no-op. */
241 ct = nf_ct_get(skb, &ctinfo);
245 new_mark = ct_mark | (ct->mark & ~(mask));
246 if (ct->mark != new_mark) {
248 nf_conntrack_event_cache(IPCT_MARK, ct);
249 key->ct.mark = new_mark;
258 static int ovs_ct_set_labels(struct sk_buff *skb, struct sw_flow_key *key,
259 const struct ovs_key_ct_labels *labels,
260 const struct ovs_key_ct_labels *mask)
262 enum ip_conntrack_info ctinfo;
263 struct nf_conn_labels *cl;
267 /* The connection could be invalid, in which case set_label is no-op.*/
268 ct = nf_ct_get(skb, &ctinfo);
272 cl = nf_ct_labels_find(ct);
274 nf_ct_labels_ext_add(ct);
275 cl = nf_ct_labels_find(ct);
277 if (!cl || sizeof(cl->bits) < OVS_CT_LABELS_LEN)
280 err = nf_connlabels_replace(ct, (u32 *)labels, (u32 *)mask,
281 OVS_CT_LABELS_LEN / sizeof(u32));
285 ovs_ct_get_labels(ct, &key->ct.labels);
289 /* 'skb' should already be pulled to nh_ofs. */
290 static int ovs_ct_helper(struct sk_buff *skb, u16 proto)
292 const struct nf_conntrack_helper *helper;
293 const struct nf_conn_help *help;
294 enum ip_conntrack_info ctinfo;
295 unsigned int protoff;
299 ct = nf_ct_get(skb, &ctinfo);
300 if (!ct || ctinfo == IP_CT_RELATED_REPLY)
303 help = nfct_help(ct);
307 helper = rcu_dereference(help->helper);
313 protoff = ip_hdrlen(skb);
316 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
320 ofs = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr,
322 if (ofs < 0 || (frag_off & htons(~0x7)) != 0) {
323 pr_debug("proto header not found\n");
330 WARN_ONCE(1, "helper invoked on non-IP family!");
334 err = helper->help(skb, protoff, ct, ctinfo);
335 if (err != NF_ACCEPT)
338 /* Adjust seqs after helper. This is needed due to some helpers (e.g.,
339 * FTP with NAT) adusting the TCP payload size when mangling IP
340 * addresses and/or port numbers in the text-based control connection.
342 if (test_bit(IPS_SEQ_ADJUST_BIT, &ct->status) &&
343 !nf_ct_seq_adjust(skb, ct, ctinfo, protoff))
348 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
349 * value if 'skb' is freed.
351 static int handle_fragments(struct net *net, struct sw_flow_key *key,
352 u16 zone, struct sk_buff *skb)
354 struct ovs_skb_cb ovs_cb = *OVS_CB(skb);
357 if (key->eth.type == htons(ETH_P_IP)) {
358 enum ip_defrag_users user = IP_DEFRAG_CONNTRACK_IN + zone;
360 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
361 err = ip_defrag(net, skb, user);
365 ovs_cb.mru = IPCB(skb)->frag_max_size;
366 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
367 } else if (key->eth.type == htons(ETH_P_IPV6)) {
368 enum ip6_defrag_users user = IP6_DEFRAG_CONNTRACK_IN + zone;
371 memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm));
372 err = nf_ct_frag6_gather(net, skb, user);
374 if (err != -EINPROGRESS)
379 key->ip.proto = ipv6_hdr(skb)->nexthdr;
380 ovs_cb.mru = IP6CB(skb)->frag_max_size;
384 return -EPFNOSUPPORT;
387 key->ip.frag = OVS_FRAG_TYPE_NONE;
390 *OVS_CB(skb) = ovs_cb;
395 static struct nf_conntrack_expect *
396 ovs_ct_expect_find(struct net *net, const struct nf_conntrack_zone *zone,
397 u16 proto, const struct sk_buff *skb)
399 struct nf_conntrack_tuple tuple;
401 if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), proto, net, &tuple))
403 return __nf_ct_expect_find(net, zone, &tuple);
406 /* This replicates logic from nf_conntrack_core.c that is not exported. */
407 static enum ip_conntrack_info
408 ovs_ct_get_info(const struct nf_conntrack_tuple_hash *h)
410 const struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
412 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY)
413 return IP_CT_ESTABLISHED_REPLY;
414 /* Once we've had two way comms, always ESTABLISHED. */
415 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status))
416 return IP_CT_ESTABLISHED;
417 if (test_bit(IPS_EXPECTED_BIT, &ct->status))
418 return IP_CT_RELATED;
422 /* Find an existing connection which this packet belongs to without
423 * re-attributing statistics or modifying the connection state. This allows an
424 * skb->_nfct lost due to an upcall to be recovered during actions execution.
426 * Must be called with rcu_read_lock.
428 * On success, populates skb->_nfct and returns the connection. Returns NULL
429 * if there is no existing entry.
431 static struct nf_conn *
432 ovs_ct_find_existing(struct net *net, const struct nf_conntrack_zone *zone,
433 u8 l3num, struct sk_buff *skb, bool natted)
435 struct nf_conntrack_l3proto *l3proto;
436 struct nf_conntrack_l4proto *l4proto;
437 struct nf_conntrack_tuple tuple;
438 struct nf_conntrack_tuple_hash *h;
440 unsigned int dataoff;
443 l3proto = __nf_ct_l3proto_find(l3num);
444 if (l3proto->get_l4proto(skb, skb_network_offset(skb), &dataoff,
446 pr_debug("ovs_ct_find_existing: Can't get protonum\n");
449 l4proto = __nf_ct_l4proto_find(l3num, protonum);
450 if (!nf_ct_get_tuple(skb, skb_network_offset(skb), dataoff, l3num,
451 protonum, net, &tuple, l3proto, l4proto)) {
452 pr_debug("ovs_ct_find_existing: Can't get tuple\n");
456 /* Must invert the tuple if skb has been transformed by NAT. */
458 struct nf_conntrack_tuple inverse;
460 if (!nf_ct_invert_tuple(&inverse, &tuple, l3proto, l4proto)) {
461 pr_debug("ovs_ct_find_existing: Inversion failed!\n");
467 /* look for tuple match */
468 h = nf_conntrack_find_get(net, zone, &tuple);
470 return NULL; /* Not found. */
472 ct = nf_ct_tuplehash_to_ctrack(h);
474 /* Inverted packet tuple matches the reverse direction conntrack tuple,
475 * select the other tuplehash to get the right 'ctinfo' bits for this
479 h = &ct->tuplehash[!h->tuple.dst.dir];
481 nf_ct_set(skb, ct, ovs_ct_get_info(h));
485 /* Determine whether skb->_nfct is equal to the result of conntrack lookup. */
486 static bool skb_nfct_cached(struct net *net,
487 const struct sw_flow_key *key,
488 const struct ovs_conntrack_info *info,
491 enum ip_conntrack_info ctinfo;
494 ct = nf_ct_get(skb, &ctinfo);
495 /* If no ct, check if we have evidence that an existing conntrack entry
496 * might be found for this skb. This happens when we lose a skb->_nfct
497 * due to an upcall. If the connection was not confirmed, it is not
498 * cached and needs to be run through conntrack again.
500 if (!ct && key->ct.state & OVS_CS_F_TRACKED &&
501 !(key->ct.state & OVS_CS_F_INVALID) &&
502 key->ct.zone == info->zone.id)
503 ct = ovs_ct_find_existing(net, &info->zone, info->family, skb,
505 & OVS_CS_F_NAT_MASK));
508 if (!net_eq(net, read_pnet(&ct->ct_net)))
510 if (!nf_ct_zone_equal_any(info->ct, nf_ct_zone(ct)))
513 struct nf_conn_help *help;
515 help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER);
516 if (help && rcu_access_pointer(help->helper) != info->helper)
523 #ifdef CONFIG_NF_NAT_NEEDED
524 /* Modelled after nf_nat_ipv[46]_fn().
525 * range is only used for new, uninitialized NAT state.
526 * Returns either NF_ACCEPT or NF_DROP.
528 static int ovs_ct_nat_execute(struct sk_buff *skb, struct nf_conn *ct,
529 enum ip_conntrack_info ctinfo,
530 const struct nf_nat_range *range,
531 enum nf_nat_manip_type maniptype)
533 int hooknum, nh_off, err = NF_ACCEPT;
535 nh_off = skb_network_offset(skb);
536 skb_pull_rcsum(skb, nh_off);
538 /* See HOOK2MANIP(). */
539 if (maniptype == NF_NAT_MANIP_SRC)
540 hooknum = NF_INET_LOCAL_IN; /* Source NAT */
542 hooknum = NF_INET_LOCAL_OUT; /* Destination NAT */
546 case IP_CT_RELATED_REPLY:
547 if (IS_ENABLED(CONFIG_NF_NAT_IPV4) &&
548 skb->protocol == htons(ETH_P_IP) &&
549 ip_hdr(skb)->protocol == IPPROTO_ICMP) {
550 if (!nf_nat_icmp_reply_translation(skb, ct, ctinfo,
554 } else if (IS_ENABLED(CONFIG_NF_NAT_IPV6) &&
555 skb->protocol == htons(ETH_P_IPV6)) {
557 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
558 int hdrlen = ipv6_skip_exthdr(skb,
559 sizeof(struct ipv6hdr),
560 &nexthdr, &frag_off);
562 if (hdrlen >= 0 && nexthdr == IPPROTO_ICMPV6) {
563 if (!nf_nat_icmpv6_reply_translation(skb, ct,
571 /* Non-ICMP, fall thru to initialize if needed. */
573 /* Seen it before? This can happen for loopback, retrans,
576 if (!nf_nat_initialized(ct, maniptype)) {
577 /* Initialize according to the NAT action. */
578 err = (range && range->flags & NF_NAT_RANGE_MAP_IPS)
579 /* Action is set up to establish a new
582 ? nf_nat_setup_info(ct, range, maniptype)
583 : nf_nat_alloc_null_binding(ct, hooknum);
584 if (err != NF_ACCEPT)
589 case IP_CT_ESTABLISHED:
590 case IP_CT_ESTABLISHED_REPLY:
598 err = nf_nat_packet(ct, ctinfo, hooknum, skb);
600 skb_push(skb, nh_off);
601 skb_postpush_rcsum(skb, skb->data, nh_off);
606 static void ovs_nat_update_key(struct sw_flow_key *key,
607 const struct sk_buff *skb,
608 enum nf_nat_manip_type maniptype)
610 if (maniptype == NF_NAT_MANIP_SRC) {
613 key->ct.state |= OVS_CS_F_SRC_NAT;
614 if (key->eth.type == htons(ETH_P_IP))
615 key->ipv4.addr.src = ip_hdr(skb)->saddr;
616 else if (key->eth.type == htons(ETH_P_IPV6))
617 memcpy(&key->ipv6.addr.src, &ipv6_hdr(skb)->saddr,
618 sizeof(key->ipv6.addr.src));
622 if (key->ip.proto == IPPROTO_UDP)
623 src = udp_hdr(skb)->source;
624 else if (key->ip.proto == IPPROTO_TCP)
625 src = tcp_hdr(skb)->source;
626 else if (key->ip.proto == IPPROTO_SCTP)
627 src = sctp_hdr(skb)->source;
635 key->ct.state |= OVS_CS_F_DST_NAT;
636 if (key->eth.type == htons(ETH_P_IP))
637 key->ipv4.addr.dst = ip_hdr(skb)->daddr;
638 else if (key->eth.type == htons(ETH_P_IPV6))
639 memcpy(&key->ipv6.addr.dst, &ipv6_hdr(skb)->daddr,
640 sizeof(key->ipv6.addr.dst));
644 if (key->ip.proto == IPPROTO_UDP)
645 dst = udp_hdr(skb)->dest;
646 else if (key->ip.proto == IPPROTO_TCP)
647 dst = tcp_hdr(skb)->dest;
648 else if (key->ip.proto == IPPROTO_SCTP)
649 dst = sctp_hdr(skb)->dest;
657 /* Returns NF_DROP if the packet should be dropped, NF_ACCEPT otherwise. */
658 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
659 const struct ovs_conntrack_info *info,
660 struct sk_buff *skb, struct nf_conn *ct,
661 enum ip_conntrack_info ctinfo)
663 enum nf_nat_manip_type maniptype;
666 if (nf_ct_is_untracked(ct)) {
667 /* A NAT action may only be performed on tracked packets. */
671 /* Add NAT extension if not confirmed yet. */
672 if (!nf_ct_is_confirmed(ct) && !nf_ct_nat_ext_add(ct))
673 return NF_ACCEPT; /* Can't NAT. */
675 /* Determine NAT type.
676 * Check if the NAT type can be deduced from the tracked connection.
677 * Make sure new expected connections (IP_CT_RELATED) are NATted only
680 if (info->nat & OVS_CT_NAT && ctinfo != IP_CT_NEW &&
681 ct->status & IPS_NAT_MASK &&
682 (ctinfo != IP_CT_RELATED || info->commit)) {
683 /* NAT an established or related connection like before. */
684 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY)
685 /* This is the REPLY direction for a connection
686 * for which NAT was applied in the forward
687 * direction. Do the reverse NAT.
689 maniptype = ct->status & IPS_SRC_NAT
690 ? NF_NAT_MANIP_DST : NF_NAT_MANIP_SRC;
692 maniptype = ct->status & IPS_SRC_NAT
693 ? NF_NAT_MANIP_SRC : NF_NAT_MANIP_DST;
694 } else if (info->nat & OVS_CT_SRC_NAT) {
695 maniptype = NF_NAT_MANIP_SRC;
696 } else if (info->nat & OVS_CT_DST_NAT) {
697 maniptype = NF_NAT_MANIP_DST;
699 return NF_ACCEPT; /* Connection is not NATed. */
701 err = ovs_ct_nat_execute(skb, ct, ctinfo, &info->range, maniptype);
703 /* Mark NAT done if successful and update the flow key. */
704 if (err == NF_ACCEPT)
705 ovs_nat_update_key(key, skb, maniptype);
709 #else /* !CONFIG_NF_NAT_NEEDED */
710 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
711 const struct ovs_conntrack_info *info,
712 struct sk_buff *skb, struct nf_conn *ct,
713 enum ip_conntrack_info ctinfo)
719 /* Pass 'skb' through conntrack in 'net', using zone configured in 'info', if
720 * not done already. Update key with new CT state after passing the packet
722 * Note that if the packet is deemed invalid by conntrack, skb->_nfct will be
723 * set to NULL and 0 will be returned.
725 static int __ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
726 const struct ovs_conntrack_info *info,
729 /* If we are recirculating packets to match on conntrack fields and
730 * committing with a separate conntrack action, then we don't need to
731 * actually run the packet through conntrack twice unless it's for a
734 bool cached = skb_nfct_cached(net, key, info, skb);
735 enum ip_conntrack_info ctinfo;
739 struct nf_conn *tmpl = info->ct;
742 /* Associate skb with specified zone. */
745 nf_conntrack_put(skb_nfct(skb));
746 nf_conntrack_get(&tmpl->ct_general);
747 nf_ct_set(skb, tmpl, IP_CT_NEW);
750 err = nf_conntrack_in(net, info->family,
751 NF_INET_PRE_ROUTING, skb);
752 if (err != NF_ACCEPT)
755 /* Clear CT state NAT flags to mark that we have not yet done
756 * NAT after the nf_conntrack_in() call. We can actually clear
757 * the whole state, as it will be re-initialized below.
761 /* Update the key, but keep the NAT flags. */
762 ovs_ct_update_key(skb, info, key, true, true);
765 ct = nf_ct_get(skb, &ctinfo);
767 /* Packets starting a new connection must be NATted before the
768 * helper, so that the helper knows about the NAT. We enforce
769 * this by delaying both NAT and helper calls for unconfirmed
770 * connections until the committing CT action. For later
771 * packets NAT and Helper may be called in either order.
773 * NAT will be done only if the CT action has NAT, and only
774 * once per packet (per zone), as guarded by the NAT bits in
777 if (info->nat && !(key->ct.state & OVS_CS_F_NAT_MASK) &&
778 (nf_ct_is_confirmed(ct) || info->commit) &&
779 ovs_ct_nat(net, key, info, skb, ct, ctinfo) != NF_ACCEPT) {
783 /* Userspace may decide to perform a ct lookup without a helper
784 * specified followed by a (recirculate and) commit with one.
785 * Therefore, for unconfirmed connections which we will commit,
786 * we need to attach the helper here.
788 if (!nf_ct_is_confirmed(ct) && info->commit &&
789 info->helper && !nfct_help(ct)) {
790 int err = __nf_ct_try_assign_helper(ct, info->ct,
796 /* Call the helper only if:
797 * - nf_conntrack_in() was executed above ("!cached") for a
798 * confirmed connection, or
799 * - When committing an unconfirmed connection.
801 if ((nf_ct_is_confirmed(ct) ? !cached : info->commit) &&
802 ovs_ct_helper(skb, info->family) != NF_ACCEPT) {
810 /* Lookup connection and read fields into key. */
811 static int ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
812 const struct ovs_conntrack_info *info,
815 struct nf_conntrack_expect *exp;
817 /* If we pass an expected packet through nf_conntrack_in() the
818 * expectation is typically removed, but the packet could still be
819 * lost in upcall processing. To prevent this from happening we
820 * perform an explicit expectation lookup. Expected connections are
821 * always new, and will be passed through conntrack only when they are
822 * committed, as it is OK to remove the expectation at that time.
824 exp = ovs_ct_expect_find(net, &info->zone, info->family, skb);
828 /* NOTE: New connections are NATted and Helped only when
829 * committed, so we are not calling into NAT here.
831 state = OVS_CS_F_TRACKED | OVS_CS_F_NEW | OVS_CS_F_RELATED;
832 __ovs_ct_update_key(key, state, &info->zone, exp->master);
837 err = __ovs_ct_lookup(net, key, info, skb);
841 ct = (struct nf_conn *)skb_nfct(skb);
843 nf_ct_deliver_cached_events(ct);
849 static bool labels_nonzero(const struct ovs_key_ct_labels *labels)
853 for (i = 0; i < sizeof(*labels); i++)
854 if (labels->ct_labels[i])
860 /* Lookup connection and confirm if unconfirmed. */
861 static int ovs_ct_commit(struct net *net, struct sw_flow_key *key,
862 const struct ovs_conntrack_info *info,
867 err = __ovs_ct_lookup(net, key, info, skb);
871 /* Apply changes before confirming the connection so that the initial
872 * conntrack NEW netlink event carries the values given in the CT
875 if (info->mark.mask) {
876 err = ovs_ct_set_mark(skb, key, info->mark.value,
881 if (labels_nonzero(&info->labels.mask)) {
882 err = ovs_ct_set_labels(skb, key, &info->labels.value,
887 /* This will take care of sending queued events even if the connection
888 * is already confirmed.
890 if (nf_conntrack_confirm(skb) != NF_ACCEPT)
896 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
897 * value if 'skb' is freed.
899 int ovs_ct_execute(struct net *net, struct sk_buff *skb,
900 struct sw_flow_key *key,
901 const struct ovs_conntrack_info *info)
906 /* The conntrack module expects to be working at L3. */
907 nh_ofs = skb_network_offset(skb);
908 skb_pull_rcsum(skb, nh_ofs);
910 if (key->ip.frag != OVS_FRAG_TYPE_NONE) {
911 err = handle_fragments(net, key, info->zone.id, skb);
917 err = ovs_ct_commit(net, key, info, skb);
919 err = ovs_ct_lookup(net, key, info, skb);
921 skb_push(skb, nh_ofs);
922 skb_postpush_rcsum(skb, skb->data, nh_ofs);
928 static int ovs_ct_add_helper(struct ovs_conntrack_info *info, const char *name,
929 const struct sw_flow_key *key, bool log)
931 struct nf_conntrack_helper *helper;
932 struct nf_conn_help *help;
934 helper = nf_conntrack_helper_try_module_get(name, info->family,
937 OVS_NLERR(log, "Unknown helper \"%s\"", name);
941 help = nf_ct_helper_ext_add(info->ct, helper, GFP_KERNEL);
943 module_put(helper->me);
947 rcu_assign_pointer(help->helper, helper);
948 info->helper = helper;
952 #ifdef CONFIG_NF_NAT_NEEDED
953 static int parse_nat(const struct nlattr *attr,
954 struct ovs_conntrack_info *info, bool log)
958 bool have_ip_max = false;
959 bool have_proto_max = false;
960 bool ip_vers = (info->family == NFPROTO_IPV6);
962 nla_for_each_nested(a, attr, rem) {
963 static const int ovs_nat_attr_lens[OVS_NAT_ATTR_MAX + 1][2] = {
964 [OVS_NAT_ATTR_SRC] = {0, 0},
965 [OVS_NAT_ATTR_DST] = {0, 0},
966 [OVS_NAT_ATTR_IP_MIN] = {sizeof(struct in_addr),
967 sizeof(struct in6_addr)},
968 [OVS_NAT_ATTR_IP_MAX] = {sizeof(struct in_addr),
969 sizeof(struct in6_addr)},
970 [OVS_NAT_ATTR_PROTO_MIN] = {sizeof(u16), sizeof(u16)},
971 [OVS_NAT_ATTR_PROTO_MAX] = {sizeof(u16), sizeof(u16)},
972 [OVS_NAT_ATTR_PERSISTENT] = {0, 0},
973 [OVS_NAT_ATTR_PROTO_HASH] = {0, 0},
974 [OVS_NAT_ATTR_PROTO_RANDOM] = {0, 0},
976 int type = nla_type(a);
978 if (type > OVS_NAT_ATTR_MAX) {
980 "Unknown NAT attribute (type=%d, max=%d).\n",
981 type, OVS_NAT_ATTR_MAX);
985 if (nla_len(a) != ovs_nat_attr_lens[type][ip_vers]) {
987 "NAT attribute type %d has unexpected length (%d != %d).\n",
989 ovs_nat_attr_lens[type][ip_vers]);
994 case OVS_NAT_ATTR_SRC:
995 case OVS_NAT_ATTR_DST:
998 "Only one type of NAT may be specified.\n"
1002 info->nat |= OVS_CT_NAT;
1003 info->nat |= ((type == OVS_NAT_ATTR_SRC)
1004 ? OVS_CT_SRC_NAT : OVS_CT_DST_NAT);
1007 case OVS_NAT_ATTR_IP_MIN:
1008 nla_memcpy(&info->range.min_addr, a,
1009 sizeof(info->range.min_addr));
1010 info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1013 case OVS_NAT_ATTR_IP_MAX:
1015 nla_memcpy(&info->range.max_addr, a,
1016 sizeof(info->range.max_addr));
1017 info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1020 case OVS_NAT_ATTR_PROTO_MIN:
1021 info->range.min_proto.all = htons(nla_get_u16(a));
1022 info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1025 case OVS_NAT_ATTR_PROTO_MAX:
1026 have_proto_max = true;
1027 info->range.max_proto.all = htons(nla_get_u16(a));
1028 info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1031 case OVS_NAT_ATTR_PERSISTENT:
1032 info->range.flags |= NF_NAT_RANGE_PERSISTENT;
1035 case OVS_NAT_ATTR_PROTO_HASH:
1036 info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM;
1039 case OVS_NAT_ATTR_PROTO_RANDOM:
1040 info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM_FULLY;
1044 OVS_NLERR(log, "Unknown nat attribute (%d).\n", type);
1050 OVS_NLERR(log, "NAT attribute has %d unknown bytes.\n", rem);
1054 /* Do not allow flags if no type is given. */
1055 if (info->range.flags) {
1057 "NAT flags may be given only when NAT range (SRC or DST) is also specified.\n"
1061 info->nat = OVS_CT_NAT; /* NAT existing connections. */
1062 } else if (!info->commit) {
1064 "NAT attributes may be specified only when CT COMMIT flag is also specified.\n"
1068 /* Allow missing IP_MAX. */
1069 if (info->range.flags & NF_NAT_RANGE_MAP_IPS && !have_ip_max) {
1070 memcpy(&info->range.max_addr, &info->range.min_addr,
1071 sizeof(info->range.max_addr));
1073 /* Allow missing PROTO_MAX. */
1074 if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1076 info->range.max_proto.all = info->range.min_proto.all;
1082 static const struct ovs_ct_len_tbl ovs_ct_attr_lens[OVS_CT_ATTR_MAX + 1] = {
1083 [OVS_CT_ATTR_COMMIT] = { .minlen = 0, .maxlen = 0 },
1084 [OVS_CT_ATTR_ZONE] = { .minlen = sizeof(u16),
1085 .maxlen = sizeof(u16) },
1086 [OVS_CT_ATTR_MARK] = { .minlen = sizeof(struct md_mark),
1087 .maxlen = sizeof(struct md_mark) },
1088 [OVS_CT_ATTR_LABELS] = { .minlen = sizeof(struct md_labels),
1089 .maxlen = sizeof(struct md_labels) },
1090 [OVS_CT_ATTR_HELPER] = { .minlen = 1,
1091 .maxlen = NF_CT_HELPER_NAME_LEN },
1092 #ifdef CONFIG_NF_NAT_NEEDED
1093 /* NAT length is checked when parsing the nested attributes. */
1094 [OVS_CT_ATTR_NAT] = { .minlen = 0, .maxlen = INT_MAX },
1098 static int parse_ct(const struct nlattr *attr, struct ovs_conntrack_info *info,
1099 const char **helper, bool log)
1104 nla_for_each_nested(a, attr, rem) {
1105 int type = nla_type(a);
1106 int maxlen = ovs_ct_attr_lens[type].maxlen;
1107 int minlen = ovs_ct_attr_lens[type].minlen;
1109 if (type > OVS_CT_ATTR_MAX) {
1111 "Unknown conntrack attr (type=%d, max=%d)",
1112 type, OVS_CT_ATTR_MAX);
1115 if (nla_len(a) < minlen || nla_len(a) > maxlen) {
1117 "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
1118 type, nla_len(a), maxlen);
1123 case OVS_CT_ATTR_COMMIT:
1124 info->commit = true;
1126 #ifdef CONFIG_NF_CONNTRACK_ZONES
1127 case OVS_CT_ATTR_ZONE:
1128 info->zone.id = nla_get_u16(a);
1131 #ifdef CONFIG_NF_CONNTRACK_MARK
1132 case OVS_CT_ATTR_MARK: {
1133 struct md_mark *mark = nla_data(a);
1136 OVS_NLERR(log, "ct_mark mask cannot be 0");
1143 #ifdef CONFIG_NF_CONNTRACK_LABELS
1144 case OVS_CT_ATTR_LABELS: {
1145 struct md_labels *labels = nla_data(a);
1147 if (!labels_nonzero(&labels->mask)) {
1148 OVS_NLERR(log, "ct_labels mask cannot be 0");
1151 info->labels = *labels;
1155 case OVS_CT_ATTR_HELPER:
1156 *helper = nla_data(a);
1157 if (!memchr(*helper, '\0', nla_len(a))) {
1158 OVS_NLERR(log, "Invalid conntrack helper");
1162 #ifdef CONFIG_NF_NAT_NEEDED
1163 case OVS_CT_ATTR_NAT: {
1164 int err = parse_nat(a, info, log);
1172 OVS_NLERR(log, "Unknown conntrack attr (%d)",
1178 #ifdef CONFIG_NF_CONNTRACK_MARK
1179 if (!info->commit && info->mark.mask) {
1181 "Setting conntrack mark requires 'commit' flag.");
1185 #ifdef CONFIG_NF_CONNTRACK_LABELS
1186 if (!info->commit && labels_nonzero(&info->labels.mask)) {
1188 "Setting conntrack labels requires 'commit' flag.");
1193 OVS_NLERR(log, "Conntrack attr has %d unknown bytes", rem);
1200 bool ovs_ct_verify(struct net *net, enum ovs_key_attr attr)
1202 if (attr == OVS_KEY_ATTR_CT_STATE)
1204 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1205 attr == OVS_KEY_ATTR_CT_ZONE)
1207 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1208 attr == OVS_KEY_ATTR_CT_MARK)
1210 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1211 attr == OVS_KEY_ATTR_CT_LABELS) {
1212 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1214 return ovs_net->xt_label;
1220 int ovs_ct_copy_action(struct net *net, const struct nlattr *attr,
1221 const struct sw_flow_key *key,
1222 struct sw_flow_actions **sfa, bool log)
1224 struct ovs_conntrack_info ct_info;
1225 const char *helper = NULL;
1229 family = key_to_nfproto(key);
1230 if (family == NFPROTO_UNSPEC) {
1231 OVS_NLERR(log, "ct family unspecified");
1235 memset(&ct_info, 0, sizeof(ct_info));
1236 ct_info.family = family;
1238 nf_ct_zone_init(&ct_info.zone, NF_CT_DEFAULT_ZONE_ID,
1239 NF_CT_DEFAULT_ZONE_DIR, 0);
1241 err = parse_ct(attr, &ct_info, &helper, log);
1245 /* Set up template for tracking connections in specific zones. */
1246 ct_info.ct = nf_ct_tmpl_alloc(net, &ct_info.zone, GFP_KERNEL);
1248 OVS_NLERR(log, "Failed to allocate conntrack template");
1252 __set_bit(IPS_CONFIRMED_BIT, &ct_info.ct->status);
1253 nf_conntrack_get(&ct_info.ct->ct_general);
1256 err = ovs_ct_add_helper(&ct_info, helper, key, log);
1261 err = ovs_nla_add_action(sfa, OVS_ACTION_ATTR_CT, &ct_info,
1262 sizeof(ct_info), log);
1268 __ovs_ct_free_action(&ct_info);
1272 #ifdef CONFIG_NF_NAT_NEEDED
1273 static bool ovs_ct_nat_to_attr(const struct ovs_conntrack_info *info,
1274 struct sk_buff *skb)
1276 struct nlattr *start;
1278 start = nla_nest_start(skb, OVS_CT_ATTR_NAT);
1282 if (info->nat & OVS_CT_SRC_NAT) {
1283 if (nla_put_flag(skb, OVS_NAT_ATTR_SRC))
1285 } else if (info->nat & OVS_CT_DST_NAT) {
1286 if (nla_put_flag(skb, OVS_NAT_ATTR_DST))
1292 if (info->range.flags & NF_NAT_RANGE_MAP_IPS) {
1293 if (IS_ENABLED(CONFIG_NF_NAT_IPV4) &&
1294 info->family == NFPROTO_IPV4) {
1295 if (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MIN,
1296 info->range.min_addr.ip) ||
1297 (info->range.max_addr.ip
1298 != info->range.min_addr.ip &&
1299 (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MAX,
1300 info->range.max_addr.ip))))
1302 } else if (IS_ENABLED(CONFIG_NF_NAT_IPV6) &&
1303 info->family == NFPROTO_IPV6) {
1304 if (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MIN,
1305 &info->range.min_addr.in6) ||
1306 (memcmp(&info->range.max_addr.in6,
1307 &info->range.min_addr.in6,
1308 sizeof(info->range.max_addr.in6)) &&
1309 (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MAX,
1310 &info->range.max_addr.in6))))
1316 if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1317 (nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MIN,
1318 ntohs(info->range.min_proto.all)) ||
1319 (info->range.max_proto.all != info->range.min_proto.all &&
1320 nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MAX,
1321 ntohs(info->range.max_proto.all)))))
1324 if (info->range.flags & NF_NAT_RANGE_PERSISTENT &&
1325 nla_put_flag(skb, OVS_NAT_ATTR_PERSISTENT))
1327 if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM &&
1328 nla_put_flag(skb, OVS_NAT_ATTR_PROTO_HASH))
1330 if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM_FULLY &&
1331 nla_put_flag(skb, OVS_NAT_ATTR_PROTO_RANDOM))
1334 nla_nest_end(skb, start);
1340 int ovs_ct_action_to_attr(const struct ovs_conntrack_info *ct_info,
1341 struct sk_buff *skb)
1343 struct nlattr *start;
1345 start = nla_nest_start(skb, OVS_ACTION_ATTR_CT);
1349 if (ct_info->commit && nla_put_flag(skb, OVS_CT_ATTR_COMMIT))
1351 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1352 nla_put_u16(skb, OVS_CT_ATTR_ZONE, ct_info->zone.id))
1354 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && ct_info->mark.mask &&
1355 nla_put(skb, OVS_CT_ATTR_MARK, sizeof(ct_info->mark),
1358 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1359 labels_nonzero(&ct_info->labels.mask) &&
1360 nla_put(skb, OVS_CT_ATTR_LABELS, sizeof(ct_info->labels),
1363 if (ct_info->helper) {
1364 if (nla_put_string(skb, OVS_CT_ATTR_HELPER,
1365 ct_info->helper->name))
1368 #ifdef CONFIG_NF_NAT_NEEDED
1369 if (ct_info->nat && !ovs_ct_nat_to_attr(ct_info, skb))
1372 nla_nest_end(skb, start);
1377 void ovs_ct_free_action(const struct nlattr *a)
1379 struct ovs_conntrack_info *ct_info = nla_data(a);
1381 __ovs_ct_free_action(ct_info);
1384 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info)
1386 if (ct_info->helper)
1387 module_put(ct_info->helper->me);
1389 nf_ct_tmpl_free(ct_info->ct);
1392 void ovs_ct_init(struct net *net)
1394 unsigned int n_bits = sizeof(struct ovs_key_ct_labels) * BITS_PER_BYTE;
1395 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1397 if (nf_connlabels_get(net, n_bits - 1)) {
1398 ovs_net->xt_label = false;
1399 OVS_NLERR(true, "Failed to set connlabel length");
1401 ovs_net->xt_label = true;
1405 void ovs_ct_exit(struct net *net)
1407 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1409 if (ovs_net->xt_label)
1410 nf_connlabels_put(net);