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 skb->nfctinfo, and returns the
429 * connection. Returns NULL 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)
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 /* look for tuple match */
457 h = nf_conntrack_find_get(net, zone, &tuple);
459 return NULL; /* Not found. */
461 ct = nf_ct_tuplehash_to_ctrack(h);
463 skb->nfct = &ct->ct_general;
464 skb->nfctinfo = ovs_ct_get_info(h);
468 /* Determine whether skb->nfct is equal to the result of conntrack lookup. */
469 static bool skb_nfct_cached(struct net *net,
470 const struct sw_flow_key *key,
471 const struct ovs_conntrack_info *info,
474 enum ip_conntrack_info ctinfo;
477 ct = nf_ct_get(skb, &ctinfo);
478 /* If no ct, check if we have evidence that an existing conntrack entry
479 * might be found for this skb. This happens when we lose a skb->nfct
480 * due to an upcall. If the connection was not confirmed, it is not
481 * cached and needs to be run through conntrack again.
483 if (!ct && key->ct.state & OVS_CS_F_TRACKED &&
484 !(key->ct.state & OVS_CS_F_INVALID) &&
485 key->ct.zone == info->zone.id)
486 ct = ovs_ct_find_existing(net, &info->zone, info->family, skb);
489 if (!net_eq(net, read_pnet(&ct->ct_net)))
491 if (!nf_ct_zone_equal_any(info->ct, nf_ct_zone(ct)))
494 struct nf_conn_help *help;
496 help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER);
497 if (help && rcu_access_pointer(help->helper) != info->helper)
504 #ifdef CONFIG_NF_NAT_NEEDED
505 /* Modelled after nf_nat_ipv[46]_fn().
506 * range is only used for new, uninitialized NAT state.
507 * Returns either NF_ACCEPT or NF_DROP.
509 static int ovs_ct_nat_execute(struct sk_buff *skb, struct nf_conn *ct,
510 enum ip_conntrack_info ctinfo,
511 const struct nf_nat_range *range,
512 enum nf_nat_manip_type maniptype)
514 int hooknum, nh_off, err = NF_ACCEPT;
516 nh_off = skb_network_offset(skb);
517 skb_pull_rcsum(skb, nh_off);
519 /* See HOOK2MANIP(). */
520 if (maniptype == NF_NAT_MANIP_SRC)
521 hooknum = NF_INET_LOCAL_IN; /* Source NAT */
523 hooknum = NF_INET_LOCAL_OUT; /* Destination NAT */
527 case IP_CT_RELATED_REPLY:
528 if (IS_ENABLED(CONFIG_NF_NAT_IPV4) &&
529 skb->protocol == htons(ETH_P_IP) &&
530 ip_hdr(skb)->protocol == IPPROTO_ICMP) {
531 if (!nf_nat_icmp_reply_translation(skb, ct, ctinfo,
535 } else if (IS_ENABLED(CONFIG_NF_NAT_IPV6) &&
536 skb->protocol == htons(ETH_P_IPV6)) {
538 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
539 int hdrlen = ipv6_skip_exthdr(skb,
540 sizeof(struct ipv6hdr),
541 &nexthdr, &frag_off);
543 if (hdrlen >= 0 && nexthdr == IPPROTO_ICMPV6) {
544 if (!nf_nat_icmpv6_reply_translation(skb, ct,
552 /* Non-ICMP, fall thru to initialize if needed. */
554 /* Seen it before? This can happen for loopback, retrans,
557 if (!nf_nat_initialized(ct, maniptype)) {
558 /* Initialize according to the NAT action. */
559 err = (range && range->flags & NF_NAT_RANGE_MAP_IPS)
560 /* Action is set up to establish a new
563 ? nf_nat_setup_info(ct, range, maniptype)
564 : nf_nat_alloc_null_binding(ct, hooknum);
565 if (err != NF_ACCEPT)
570 case IP_CT_ESTABLISHED:
571 case IP_CT_ESTABLISHED_REPLY:
579 err = nf_nat_packet(ct, ctinfo, hooknum, skb);
581 skb_push(skb, nh_off);
582 skb_postpush_rcsum(skb, skb->data, nh_off);
587 static void ovs_nat_update_key(struct sw_flow_key *key,
588 const struct sk_buff *skb,
589 enum nf_nat_manip_type maniptype)
591 if (maniptype == NF_NAT_MANIP_SRC) {
594 key->ct.state |= OVS_CS_F_SRC_NAT;
595 if (key->eth.type == htons(ETH_P_IP))
596 key->ipv4.addr.src = ip_hdr(skb)->saddr;
597 else if (key->eth.type == htons(ETH_P_IPV6))
598 memcpy(&key->ipv6.addr.src, &ipv6_hdr(skb)->saddr,
599 sizeof(key->ipv6.addr.src));
603 if (key->ip.proto == IPPROTO_UDP)
604 src = udp_hdr(skb)->source;
605 else if (key->ip.proto == IPPROTO_TCP)
606 src = tcp_hdr(skb)->source;
607 else if (key->ip.proto == IPPROTO_SCTP)
608 src = sctp_hdr(skb)->source;
616 key->ct.state |= OVS_CS_F_DST_NAT;
617 if (key->eth.type == htons(ETH_P_IP))
618 key->ipv4.addr.dst = ip_hdr(skb)->daddr;
619 else if (key->eth.type == htons(ETH_P_IPV6))
620 memcpy(&key->ipv6.addr.dst, &ipv6_hdr(skb)->daddr,
621 sizeof(key->ipv6.addr.dst));
625 if (key->ip.proto == IPPROTO_UDP)
626 dst = udp_hdr(skb)->dest;
627 else if (key->ip.proto == IPPROTO_TCP)
628 dst = tcp_hdr(skb)->dest;
629 else if (key->ip.proto == IPPROTO_SCTP)
630 dst = sctp_hdr(skb)->dest;
638 /* Returns NF_DROP if the packet should be dropped, NF_ACCEPT otherwise. */
639 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
640 const struct ovs_conntrack_info *info,
641 struct sk_buff *skb, struct nf_conn *ct,
642 enum ip_conntrack_info ctinfo)
644 enum nf_nat_manip_type maniptype;
647 if (nf_ct_is_untracked(ct)) {
648 /* A NAT action may only be performed on tracked packets. */
652 /* Add NAT extension if not confirmed yet. */
653 if (!nf_ct_is_confirmed(ct) && !nf_ct_nat_ext_add(ct))
654 return NF_ACCEPT; /* Can't NAT. */
656 /* Determine NAT type.
657 * Check if the NAT type can be deduced from the tracked connection.
658 * Make sure new expected connections (IP_CT_RELATED) are NATted only
661 if (info->nat & OVS_CT_NAT && ctinfo != IP_CT_NEW &&
662 ct->status & IPS_NAT_MASK &&
663 (ctinfo != IP_CT_RELATED || info->commit)) {
664 /* NAT an established or related connection like before. */
665 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY)
666 /* This is the REPLY direction for a connection
667 * for which NAT was applied in the forward
668 * direction. Do the reverse NAT.
670 maniptype = ct->status & IPS_SRC_NAT
671 ? NF_NAT_MANIP_DST : NF_NAT_MANIP_SRC;
673 maniptype = ct->status & IPS_SRC_NAT
674 ? NF_NAT_MANIP_SRC : NF_NAT_MANIP_DST;
675 } else if (info->nat & OVS_CT_SRC_NAT) {
676 maniptype = NF_NAT_MANIP_SRC;
677 } else if (info->nat & OVS_CT_DST_NAT) {
678 maniptype = NF_NAT_MANIP_DST;
680 return NF_ACCEPT; /* Connection is not NATed. */
682 err = ovs_ct_nat_execute(skb, ct, ctinfo, &info->range, maniptype);
684 /* Mark NAT done if successful and update the flow key. */
685 if (err == NF_ACCEPT)
686 ovs_nat_update_key(key, skb, maniptype);
690 #else /* !CONFIG_NF_NAT_NEEDED */
691 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
692 const struct ovs_conntrack_info *info,
693 struct sk_buff *skb, struct nf_conn *ct,
694 enum ip_conntrack_info ctinfo)
700 /* Pass 'skb' through conntrack in 'net', using zone configured in 'info', if
701 * not done already. Update key with new CT state after passing the packet
703 * Note that if the packet is deemed invalid by conntrack, skb->nfct will be
704 * set to NULL and 0 will be returned.
706 static int __ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
707 const struct ovs_conntrack_info *info,
710 /* If we are recirculating packets to match on conntrack fields and
711 * committing with a separate conntrack action, then we don't need to
712 * actually run the packet through conntrack twice unless it's for a
715 bool cached = skb_nfct_cached(net, key, info, skb);
716 enum ip_conntrack_info ctinfo;
720 struct nf_conn *tmpl = info->ct;
723 /* Associate skb with specified zone. */
726 nf_conntrack_put(skb->nfct);
727 nf_conntrack_get(&tmpl->ct_general);
728 skb->nfct = &tmpl->ct_general;
729 skb->nfctinfo = IP_CT_NEW;
732 err = nf_conntrack_in(net, info->family,
733 NF_INET_PRE_ROUTING, skb);
734 if (err != NF_ACCEPT)
737 /* Clear CT state NAT flags to mark that we have not yet done
738 * NAT after the nf_conntrack_in() call. We can actually clear
739 * the whole state, as it will be re-initialized below.
743 /* Update the key, but keep the NAT flags. */
744 ovs_ct_update_key(skb, info, key, true, true);
747 ct = nf_ct_get(skb, &ctinfo);
749 /* Packets starting a new connection must be NATted before the
750 * helper, so that the helper knows about the NAT. We enforce
751 * this by delaying both NAT and helper calls for unconfirmed
752 * connections until the committing CT action. For later
753 * packets NAT and Helper may be called in either order.
755 * NAT will be done only if the CT action has NAT, and only
756 * once per packet (per zone), as guarded by the NAT bits in
759 if (info->nat && !(key->ct.state & OVS_CS_F_NAT_MASK) &&
760 (nf_ct_is_confirmed(ct) || info->commit) &&
761 ovs_ct_nat(net, key, info, skb, ct, ctinfo) != NF_ACCEPT) {
765 /* Userspace may decide to perform a ct lookup without a helper
766 * specified followed by a (recirculate and) commit with one.
767 * Therefore, for unconfirmed connections which we will commit,
768 * we need to attach the helper here.
770 if (!nf_ct_is_confirmed(ct) && info->commit &&
771 info->helper && !nfct_help(ct)) {
772 int err = __nf_ct_try_assign_helper(ct, info->ct,
778 /* Call the helper only if:
779 * - nf_conntrack_in() was executed above ("!cached") for a
780 * confirmed connection, or
781 * - When committing an unconfirmed connection.
783 if ((nf_ct_is_confirmed(ct) ? !cached : info->commit) &&
784 ovs_ct_helper(skb, info->family) != NF_ACCEPT) {
792 /* Lookup connection and read fields into key. */
793 static int ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
794 const struct ovs_conntrack_info *info,
797 struct nf_conntrack_expect *exp;
799 /* If we pass an expected packet through nf_conntrack_in() the
800 * expectation is typically removed, but the packet could still be
801 * lost in upcall processing. To prevent this from happening we
802 * perform an explicit expectation lookup. Expected connections are
803 * always new, and will be passed through conntrack only when they are
804 * committed, as it is OK to remove the expectation at that time.
806 exp = ovs_ct_expect_find(net, &info->zone, info->family, skb);
810 /* NOTE: New connections are NATted and Helped only when
811 * committed, so we are not calling into NAT here.
813 state = OVS_CS_F_TRACKED | OVS_CS_F_NEW | OVS_CS_F_RELATED;
814 __ovs_ct_update_key(key, state, &info->zone, exp->master);
819 err = __ovs_ct_lookup(net, key, info, skb);
823 ct = (struct nf_conn *)skb->nfct;
825 nf_ct_deliver_cached_events(ct);
831 static bool labels_nonzero(const struct ovs_key_ct_labels *labels)
835 for (i = 0; i < sizeof(*labels); i++)
836 if (labels->ct_labels[i])
842 /* Lookup connection and confirm if unconfirmed. */
843 static int ovs_ct_commit(struct net *net, struct sw_flow_key *key,
844 const struct ovs_conntrack_info *info,
849 err = __ovs_ct_lookup(net, key, info, skb);
853 /* Apply changes before confirming the connection so that the initial
854 * conntrack NEW netlink event carries the values given in the CT
857 if (info->mark.mask) {
858 err = ovs_ct_set_mark(skb, key, info->mark.value,
863 if (labels_nonzero(&info->labels.mask)) {
864 err = ovs_ct_set_labels(skb, key, &info->labels.value,
869 /* This will take care of sending queued events even if the connection
870 * is already confirmed.
872 if (nf_conntrack_confirm(skb) != NF_ACCEPT)
878 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
879 * value if 'skb' is freed.
881 int ovs_ct_execute(struct net *net, struct sk_buff *skb,
882 struct sw_flow_key *key,
883 const struct ovs_conntrack_info *info)
888 /* The conntrack module expects to be working at L3. */
889 nh_ofs = skb_network_offset(skb);
890 skb_pull_rcsum(skb, nh_ofs);
892 if (key->ip.frag != OVS_FRAG_TYPE_NONE) {
893 err = handle_fragments(net, key, info->zone.id, skb);
899 err = ovs_ct_commit(net, key, info, skb);
901 err = ovs_ct_lookup(net, key, info, skb);
903 skb_push(skb, nh_ofs);
904 skb_postpush_rcsum(skb, skb->data, nh_ofs);
910 static int ovs_ct_add_helper(struct ovs_conntrack_info *info, const char *name,
911 const struct sw_flow_key *key, bool log)
913 struct nf_conntrack_helper *helper;
914 struct nf_conn_help *help;
916 helper = nf_conntrack_helper_try_module_get(name, info->family,
919 OVS_NLERR(log, "Unknown helper \"%s\"", name);
923 help = nf_ct_helper_ext_add(info->ct, helper, GFP_KERNEL);
925 module_put(helper->me);
929 rcu_assign_pointer(help->helper, helper);
930 info->helper = helper;
934 #ifdef CONFIG_NF_NAT_NEEDED
935 static int parse_nat(const struct nlattr *attr,
936 struct ovs_conntrack_info *info, bool log)
940 bool have_ip_max = false;
941 bool have_proto_max = false;
942 bool ip_vers = (info->family == NFPROTO_IPV6);
944 nla_for_each_nested(a, attr, rem) {
945 static const int ovs_nat_attr_lens[OVS_NAT_ATTR_MAX + 1][2] = {
946 [OVS_NAT_ATTR_SRC] = {0, 0},
947 [OVS_NAT_ATTR_DST] = {0, 0},
948 [OVS_NAT_ATTR_IP_MIN] = {sizeof(struct in_addr),
949 sizeof(struct in6_addr)},
950 [OVS_NAT_ATTR_IP_MAX] = {sizeof(struct in_addr),
951 sizeof(struct in6_addr)},
952 [OVS_NAT_ATTR_PROTO_MIN] = {sizeof(u16), sizeof(u16)},
953 [OVS_NAT_ATTR_PROTO_MAX] = {sizeof(u16), sizeof(u16)},
954 [OVS_NAT_ATTR_PERSISTENT] = {0, 0},
955 [OVS_NAT_ATTR_PROTO_HASH] = {0, 0},
956 [OVS_NAT_ATTR_PROTO_RANDOM] = {0, 0},
958 int type = nla_type(a);
960 if (type > OVS_NAT_ATTR_MAX) {
962 "Unknown NAT attribute (type=%d, max=%d).\n",
963 type, OVS_NAT_ATTR_MAX);
967 if (nla_len(a) != ovs_nat_attr_lens[type][ip_vers]) {
969 "NAT attribute type %d has unexpected length (%d != %d).\n",
971 ovs_nat_attr_lens[type][ip_vers]);
976 case OVS_NAT_ATTR_SRC:
977 case OVS_NAT_ATTR_DST:
980 "Only one type of NAT may be specified.\n"
984 info->nat |= OVS_CT_NAT;
985 info->nat |= ((type == OVS_NAT_ATTR_SRC)
986 ? OVS_CT_SRC_NAT : OVS_CT_DST_NAT);
989 case OVS_NAT_ATTR_IP_MIN:
990 nla_memcpy(&info->range.min_addr, a,
991 sizeof(info->range.min_addr));
992 info->range.flags |= NF_NAT_RANGE_MAP_IPS;
995 case OVS_NAT_ATTR_IP_MAX:
997 nla_memcpy(&info->range.max_addr, a,
998 sizeof(info->range.max_addr));
999 info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1002 case OVS_NAT_ATTR_PROTO_MIN:
1003 info->range.min_proto.all = htons(nla_get_u16(a));
1004 info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1007 case OVS_NAT_ATTR_PROTO_MAX:
1008 have_proto_max = true;
1009 info->range.max_proto.all = htons(nla_get_u16(a));
1010 info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1013 case OVS_NAT_ATTR_PERSISTENT:
1014 info->range.flags |= NF_NAT_RANGE_PERSISTENT;
1017 case OVS_NAT_ATTR_PROTO_HASH:
1018 info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM;
1021 case OVS_NAT_ATTR_PROTO_RANDOM:
1022 info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM_FULLY;
1026 OVS_NLERR(log, "Unknown nat attribute (%d).\n", type);
1032 OVS_NLERR(log, "NAT attribute has %d unknown bytes.\n", rem);
1036 /* Do not allow flags if no type is given. */
1037 if (info->range.flags) {
1039 "NAT flags may be given only when NAT range (SRC or DST) is also specified.\n"
1043 info->nat = OVS_CT_NAT; /* NAT existing connections. */
1044 } else if (!info->commit) {
1046 "NAT attributes may be specified only when CT COMMIT flag is also specified.\n"
1050 /* Allow missing IP_MAX. */
1051 if (info->range.flags & NF_NAT_RANGE_MAP_IPS && !have_ip_max) {
1052 memcpy(&info->range.max_addr, &info->range.min_addr,
1053 sizeof(info->range.max_addr));
1055 /* Allow missing PROTO_MAX. */
1056 if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1058 info->range.max_proto.all = info->range.min_proto.all;
1064 static const struct ovs_ct_len_tbl ovs_ct_attr_lens[OVS_CT_ATTR_MAX + 1] = {
1065 [OVS_CT_ATTR_COMMIT] = { .minlen = 0, .maxlen = 0 },
1066 [OVS_CT_ATTR_ZONE] = { .minlen = sizeof(u16),
1067 .maxlen = sizeof(u16) },
1068 [OVS_CT_ATTR_MARK] = { .minlen = sizeof(struct md_mark),
1069 .maxlen = sizeof(struct md_mark) },
1070 [OVS_CT_ATTR_LABELS] = { .minlen = sizeof(struct md_labels),
1071 .maxlen = sizeof(struct md_labels) },
1072 [OVS_CT_ATTR_HELPER] = { .minlen = 1,
1073 .maxlen = NF_CT_HELPER_NAME_LEN },
1074 #ifdef CONFIG_NF_NAT_NEEDED
1075 /* NAT length is checked when parsing the nested attributes. */
1076 [OVS_CT_ATTR_NAT] = { .minlen = 0, .maxlen = INT_MAX },
1080 static int parse_ct(const struct nlattr *attr, struct ovs_conntrack_info *info,
1081 const char **helper, bool log)
1086 nla_for_each_nested(a, attr, rem) {
1087 int type = nla_type(a);
1088 int maxlen = ovs_ct_attr_lens[type].maxlen;
1089 int minlen = ovs_ct_attr_lens[type].minlen;
1091 if (type > OVS_CT_ATTR_MAX) {
1093 "Unknown conntrack attr (type=%d, max=%d)",
1094 type, OVS_CT_ATTR_MAX);
1097 if (nla_len(a) < minlen || nla_len(a) > maxlen) {
1099 "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
1100 type, nla_len(a), maxlen);
1105 case OVS_CT_ATTR_COMMIT:
1106 info->commit = true;
1108 #ifdef CONFIG_NF_CONNTRACK_ZONES
1109 case OVS_CT_ATTR_ZONE:
1110 info->zone.id = nla_get_u16(a);
1113 #ifdef CONFIG_NF_CONNTRACK_MARK
1114 case OVS_CT_ATTR_MARK: {
1115 struct md_mark *mark = nla_data(a);
1118 OVS_NLERR(log, "ct_mark mask cannot be 0");
1125 #ifdef CONFIG_NF_CONNTRACK_LABELS
1126 case OVS_CT_ATTR_LABELS: {
1127 struct md_labels *labels = nla_data(a);
1129 if (!labels_nonzero(&labels->mask)) {
1130 OVS_NLERR(log, "ct_labels mask cannot be 0");
1133 info->labels = *labels;
1137 case OVS_CT_ATTR_HELPER:
1138 *helper = nla_data(a);
1139 if (!memchr(*helper, '\0', nla_len(a))) {
1140 OVS_NLERR(log, "Invalid conntrack helper");
1144 #ifdef CONFIG_NF_NAT_NEEDED
1145 case OVS_CT_ATTR_NAT: {
1146 int err = parse_nat(a, info, log);
1154 OVS_NLERR(log, "Unknown conntrack attr (%d)",
1160 #ifdef CONFIG_NF_CONNTRACK_MARK
1161 if (!info->commit && info->mark.mask) {
1163 "Setting conntrack mark requires 'commit' flag.");
1167 #ifdef CONFIG_NF_CONNTRACK_LABELS
1168 if (!info->commit && labels_nonzero(&info->labels.mask)) {
1170 "Setting conntrack labels requires 'commit' flag.");
1175 OVS_NLERR(log, "Conntrack attr has %d unknown bytes", rem);
1182 bool ovs_ct_verify(struct net *net, enum ovs_key_attr attr)
1184 if (attr == OVS_KEY_ATTR_CT_STATE)
1186 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1187 attr == OVS_KEY_ATTR_CT_ZONE)
1189 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1190 attr == OVS_KEY_ATTR_CT_MARK)
1192 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1193 attr == OVS_KEY_ATTR_CT_LABELS) {
1194 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1196 return ovs_net->xt_label;
1202 int ovs_ct_copy_action(struct net *net, const struct nlattr *attr,
1203 const struct sw_flow_key *key,
1204 struct sw_flow_actions **sfa, bool log)
1206 struct ovs_conntrack_info ct_info;
1207 const char *helper = NULL;
1211 family = key_to_nfproto(key);
1212 if (family == NFPROTO_UNSPEC) {
1213 OVS_NLERR(log, "ct family unspecified");
1217 memset(&ct_info, 0, sizeof(ct_info));
1218 ct_info.family = family;
1220 nf_ct_zone_init(&ct_info.zone, NF_CT_DEFAULT_ZONE_ID,
1221 NF_CT_DEFAULT_ZONE_DIR, 0);
1223 err = parse_ct(attr, &ct_info, &helper, log);
1227 /* Set up template for tracking connections in specific zones. */
1228 ct_info.ct = nf_ct_tmpl_alloc(net, &ct_info.zone, GFP_KERNEL);
1230 OVS_NLERR(log, "Failed to allocate conntrack template");
1234 __set_bit(IPS_CONFIRMED_BIT, &ct_info.ct->status);
1235 nf_conntrack_get(&ct_info.ct->ct_general);
1238 err = ovs_ct_add_helper(&ct_info, helper, key, log);
1243 err = ovs_nla_add_action(sfa, OVS_ACTION_ATTR_CT, &ct_info,
1244 sizeof(ct_info), log);
1250 __ovs_ct_free_action(&ct_info);
1254 #ifdef CONFIG_NF_NAT_NEEDED
1255 static bool ovs_ct_nat_to_attr(const struct ovs_conntrack_info *info,
1256 struct sk_buff *skb)
1258 struct nlattr *start;
1260 start = nla_nest_start(skb, OVS_CT_ATTR_NAT);
1264 if (info->nat & OVS_CT_SRC_NAT) {
1265 if (nla_put_flag(skb, OVS_NAT_ATTR_SRC))
1267 } else if (info->nat & OVS_CT_DST_NAT) {
1268 if (nla_put_flag(skb, OVS_NAT_ATTR_DST))
1274 if (info->range.flags & NF_NAT_RANGE_MAP_IPS) {
1275 if (IS_ENABLED(CONFIG_NF_NAT_IPV4) &&
1276 info->family == NFPROTO_IPV4) {
1277 if (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MIN,
1278 info->range.min_addr.ip) ||
1279 (info->range.max_addr.ip
1280 != info->range.min_addr.ip &&
1281 (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MAX,
1282 info->range.max_addr.ip))))
1284 } else if (IS_ENABLED(CONFIG_NF_NAT_IPV6) &&
1285 info->family == NFPROTO_IPV6) {
1286 if (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MIN,
1287 &info->range.min_addr.in6) ||
1288 (memcmp(&info->range.max_addr.in6,
1289 &info->range.min_addr.in6,
1290 sizeof(info->range.max_addr.in6)) &&
1291 (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MAX,
1292 &info->range.max_addr.in6))))
1298 if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1299 (nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MIN,
1300 ntohs(info->range.min_proto.all)) ||
1301 (info->range.max_proto.all != info->range.min_proto.all &&
1302 nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MAX,
1303 ntohs(info->range.max_proto.all)))))
1306 if (info->range.flags & NF_NAT_RANGE_PERSISTENT &&
1307 nla_put_flag(skb, OVS_NAT_ATTR_PERSISTENT))
1309 if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM &&
1310 nla_put_flag(skb, OVS_NAT_ATTR_PROTO_HASH))
1312 if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM_FULLY &&
1313 nla_put_flag(skb, OVS_NAT_ATTR_PROTO_RANDOM))
1316 nla_nest_end(skb, start);
1322 int ovs_ct_action_to_attr(const struct ovs_conntrack_info *ct_info,
1323 struct sk_buff *skb)
1325 struct nlattr *start;
1327 start = nla_nest_start(skb, OVS_ACTION_ATTR_CT);
1331 if (ct_info->commit && nla_put_flag(skb, OVS_CT_ATTR_COMMIT))
1333 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1334 nla_put_u16(skb, OVS_CT_ATTR_ZONE, ct_info->zone.id))
1336 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && ct_info->mark.mask &&
1337 nla_put(skb, OVS_CT_ATTR_MARK, sizeof(ct_info->mark),
1340 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1341 labels_nonzero(&ct_info->labels.mask) &&
1342 nla_put(skb, OVS_CT_ATTR_LABELS, sizeof(ct_info->labels),
1345 if (ct_info->helper) {
1346 if (nla_put_string(skb, OVS_CT_ATTR_HELPER,
1347 ct_info->helper->name))
1350 #ifdef CONFIG_NF_NAT_NEEDED
1351 if (ct_info->nat && !ovs_ct_nat_to_attr(ct_info, skb))
1354 nla_nest_end(skb, start);
1359 void ovs_ct_free_action(const struct nlattr *a)
1361 struct ovs_conntrack_info *ct_info = nla_data(a);
1363 __ovs_ct_free_action(ct_info);
1366 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info)
1368 if (ct_info->helper)
1369 module_put(ct_info->helper->me);
1371 nf_ct_tmpl_free(ct_info->ct);
1374 void ovs_ct_init(struct net *net)
1376 unsigned int n_bits = sizeof(struct ovs_key_ct_labels) * BITS_PER_BYTE;
1377 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1379 if (nf_connlabels_get(net, n_bits - 1)) {
1380 ovs_net->xt_label = false;
1381 OVS_NLERR(true, "Failed to set connlabel length");
1383 ovs_net->xt_label = true;
1387 void ovs_ct_exit(struct net *net)
1389 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1391 if (ovs_net->xt_label)
1392 nf_connlabels_put(net);