1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (c) 2015 Nicira, Inc.
6 #include <linux/module.h>
7 #include <linux/openvswitch.h>
10 #include <linux/sctp.h>
11 #include <linux/static_key.h>
13 #include <net/genetlink.h>
14 #include <net/netfilter/nf_conntrack_core.h>
15 #include <net/netfilter/nf_conntrack_count.h>
16 #include <net/netfilter/nf_conntrack_helper.h>
17 #include <net/netfilter/nf_conntrack_labels.h>
18 #include <net/netfilter/nf_conntrack_seqadj.h>
19 #include <net/netfilter/nf_conntrack_timeout.h>
20 #include <net/netfilter/nf_conntrack_zones.h>
21 #include <net/netfilter/ipv6/nf_defrag_ipv6.h>
22 #include <net/ipv6_frag.h>
24 #if IS_ENABLED(CONFIG_NF_NAT)
25 #include <net/netfilter/nf_nat.h>
29 #include "conntrack.h"
31 #include "flow_netlink.h"
33 struct ovs_ct_len_tbl {
38 /* Metadata mark for masked write to conntrack mark */
44 /* Metadata label for masked write to conntrack label. */
46 struct ovs_key_ct_labels value;
47 struct ovs_key_ct_labels mask;
51 OVS_CT_NAT = 1 << 0, /* NAT for committed connections only. */
52 OVS_CT_SRC_NAT = 1 << 1, /* Source NAT for NEW connections. */
53 OVS_CT_DST_NAT = 1 << 2, /* Destination NAT for NEW connections. */
56 /* Conntrack action context for execution. */
57 struct ovs_conntrack_info {
58 struct nf_conntrack_helper *helper;
59 struct nf_conntrack_zone zone;
62 u8 nat : 3; /* enum ovs_ct_nat */
64 u8 have_eventmask : 1;
66 u32 eventmask; /* Mask of 1 << IPCT_*. */
68 struct md_labels labels;
69 char timeout[CTNL_TIMEOUT_NAME_MAX];
70 struct nf_ct_timeout *nf_ct_timeout;
71 #if IS_ENABLED(CONFIG_NF_NAT)
72 struct nf_nat_range2 range; /* Only present for SRC NAT and DST NAT. */
76 #if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
77 #define OVS_CT_LIMIT_UNLIMITED 0
78 #define OVS_CT_LIMIT_DEFAULT OVS_CT_LIMIT_UNLIMITED
79 #define CT_LIMIT_HASH_BUCKETS 512
80 static DEFINE_STATIC_KEY_FALSE(ovs_ct_limit_enabled);
83 /* Elements in ovs_ct_limit_info->limits hash table */
84 struct hlist_node hlist_node;
90 struct ovs_ct_limit_info {
92 struct hlist_head *limits;
93 struct nf_conncount_data *data;
96 static const struct nla_policy ct_limit_policy[OVS_CT_LIMIT_ATTR_MAX + 1] = {
97 [OVS_CT_LIMIT_ATTR_ZONE_LIMIT] = { .type = NLA_NESTED, },
101 static bool labels_nonzero(const struct ovs_key_ct_labels *labels);
103 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info);
105 static u16 key_to_nfproto(const struct sw_flow_key *key)
107 switch (ntohs(key->eth.type)) {
113 return NFPROTO_UNSPEC;
117 /* Map SKB connection state into the values used by flow definition. */
118 static u8 ovs_ct_get_state(enum ip_conntrack_info ctinfo)
120 u8 ct_state = OVS_CS_F_TRACKED;
123 case IP_CT_ESTABLISHED_REPLY:
124 case IP_CT_RELATED_REPLY:
125 ct_state |= OVS_CS_F_REPLY_DIR;
132 case IP_CT_ESTABLISHED:
133 case IP_CT_ESTABLISHED_REPLY:
134 ct_state |= OVS_CS_F_ESTABLISHED;
137 case IP_CT_RELATED_REPLY:
138 ct_state |= OVS_CS_F_RELATED;
141 ct_state |= OVS_CS_F_NEW;
150 static u32 ovs_ct_get_mark(const struct nf_conn *ct)
152 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
153 return ct ? READ_ONCE(ct->mark) : 0;
159 /* Guard against conntrack labels max size shrinking below 128 bits. */
160 #if NF_CT_LABELS_MAX_SIZE < 16
161 #error NF_CT_LABELS_MAX_SIZE must be at least 16 bytes
164 static void ovs_ct_get_labels(const struct nf_conn *ct,
165 struct ovs_key_ct_labels *labels)
167 struct nf_conn_labels *cl = ct ? nf_ct_labels_find(ct) : NULL;
170 memcpy(labels, cl->bits, OVS_CT_LABELS_LEN);
172 memset(labels, 0, OVS_CT_LABELS_LEN);
175 static void __ovs_ct_update_key_orig_tp(struct sw_flow_key *key,
176 const struct nf_conntrack_tuple *orig,
179 key->ct_orig_proto = orig->dst.protonum;
180 if (orig->dst.protonum == icmp_proto) {
181 key->ct.orig_tp.src = htons(orig->dst.u.icmp.type);
182 key->ct.orig_tp.dst = htons(orig->dst.u.icmp.code);
184 key->ct.orig_tp.src = orig->src.u.all;
185 key->ct.orig_tp.dst = orig->dst.u.all;
189 static void __ovs_ct_update_key(struct sw_flow_key *key, u8 state,
190 const struct nf_conntrack_zone *zone,
191 const struct nf_conn *ct)
193 key->ct_state = state;
194 key->ct_zone = zone->id;
195 key->ct.mark = ovs_ct_get_mark(ct);
196 ovs_ct_get_labels(ct, &key->ct.labels);
199 const struct nf_conntrack_tuple *orig;
201 /* Use the master if we have one. */
204 orig = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
206 /* IP version must match with the master connection. */
207 if (key->eth.type == htons(ETH_P_IP) &&
208 nf_ct_l3num(ct) == NFPROTO_IPV4) {
209 key->ipv4.ct_orig.src = orig->src.u3.ip;
210 key->ipv4.ct_orig.dst = orig->dst.u3.ip;
211 __ovs_ct_update_key_orig_tp(key, orig, IPPROTO_ICMP);
213 } else if (key->eth.type == htons(ETH_P_IPV6) &&
214 !sw_flow_key_is_nd(key) &&
215 nf_ct_l3num(ct) == NFPROTO_IPV6) {
216 key->ipv6.ct_orig.src = orig->src.u3.in6;
217 key->ipv6.ct_orig.dst = orig->dst.u3.in6;
218 __ovs_ct_update_key_orig_tp(key, orig, NEXTHDR_ICMP);
222 /* Clear 'ct_orig_proto' to mark the non-existence of conntrack
223 * original direction key fields.
225 key->ct_orig_proto = 0;
228 /* Update 'key' based on skb->_nfct. If 'post_ct' is true, then OVS has
229 * previously sent the packet to conntrack via the ct action. If
230 * 'keep_nat_flags' is true, the existing NAT flags retained, else they are
231 * initialized from the connection status.
233 static void ovs_ct_update_key(const struct sk_buff *skb,
234 const struct ovs_conntrack_info *info,
235 struct sw_flow_key *key, bool post_ct,
238 const struct nf_conntrack_zone *zone = &nf_ct_zone_dflt;
239 enum ip_conntrack_info ctinfo;
243 ct = nf_ct_get(skb, &ctinfo);
245 state = ovs_ct_get_state(ctinfo);
246 /* All unconfirmed entries are NEW connections. */
247 if (!nf_ct_is_confirmed(ct))
248 state |= OVS_CS_F_NEW;
249 /* OVS persists the related flag for the duration of the
253 state |= OVS_CS_F_RELATED;
254 if (keep_nat_flags) {
255 state |= key->ct_state & OVS_CS_F_NAT_MASK;
257 if (ct->status & IPS_SRC_NAT)
258 state |= OVS_CS_F_SRC_NAT;
259 if (ct->status & IPS_DST_NAT)
260 state |= OVS_CS_F_DST_NAT;
262 zone = nf_ct_zone(ct);
263 } else if (post_ct) {
264 state = OVS_CS_F_TRACKED | OVS_CS_F_INVALID;
268 __ovs_ct_update_key(key, state, zone, ct);
271 /* This is called to initialize CT key fields possibly coming in from the local
274 void ovs_ct_fill_key(const struct sk_buff *skb,
275 struct sw_flow_key *key,
278 ovs_ct_update_key(skb, NULL, key, post_ct, false);
281 int ovs_ct_put_key(const struct sw_flow_key *swkey,
282 const struct sw_flow_key *output, struct sk_buff *skb)
284 if (nla_put_u32(skb, OVS_KEY_ATTR_CT_STATE, output->ct_state))
287 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
288 nla_put_u16(skb, OVS_KEY_ATTR_CT_ZONE, output->ct_zone))
291 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
292 nla_put_u32(skb, OVS_KEY_ATTR_CT_MARK, output->ct.mark))
295 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
296 nla_put(skb, OVS_KEY_ATTR_CT_LABELS, sizeof(output->ct.labels),
300 if (swkey->ct_orig_proto) {
301 if (swkey->eth.type == htons(ETH_P_IP)) {
302 struct ovs_key_ct_tuple_ipv4 orig;
304 memset(&orig, 0, sizeof(orig));
305 orig.ipv4_src = output->ipv4.ct_orig.src;
306 orig.ipv4_dst = output->ipv4.ct_orig.dst;
307 orig.src_port = output->ct.orig_tp.src;
308 orig.dst_port = output->ct.orig_tp.dst;
309 orig.ipv4_proto = output->ct_orig_proto;
311 if (nla_put(skb, OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4,
312 sizeof(orig), &orig))
314 } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
315 struct ovs_key_ct_tuple_ipv6 orig;
317 memset(&orig, 0, sizeof(orig));
318 memcpy(orig.ipv6_src, output->ipv6.ct_orig.src.s6_addr32,
319 sizeof(orig.ipv6_src));
320 memcpy(orig.ipv6_dst, output->ipv6.ct_orig.dst.s6_addr32,
321 sizeof(orig.ipv6_dst));
322 orig.src_port = output->ct.orig_tp.src;
323 orig.dst_port = output->ct.orig_tp.dst;
324 orig.ipv6_proto = output->ct_orig_proto;
326 if (nla_put(skb, OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6,
327 sizeof(orig), &orig))
335 static int ovs_ct_set_mark(struct nf_conn *ct, struct sw_flow_key *key,
336 u32 ct_mark, u32 mask)
338 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
341 new_mark = ct_mark | (READ_ONCE(ct->mark) & ~(mask));
342 if (READ_ONCE(ct->mark) != new_mark) {
343 WRITE_ONCE(ct->mark, new_mark);
344 if (nf_ct_is_confirmed(ct))
345 nf_conntrack_event_cache(IPCT_MARK, ct);
346 key->ct.mark = new_mark;
355 static struct nf_conn_labels *ovs_ct_get_conn_labels(struct nf_conn *ct)
357 struct nf_conn_labels *cl;
359 cl = nf_ct_labels_find(ct);
361 nf_ct_labels_ext_add(ct);
362 cl = nf_ct_labels_find(ct);
368 /* Initialize labels for a new, yet to be committed conntrack entry. Note that
369 * since the new connection is not yet confirmed, and thus no-one else has
370 * access to it's labels, we simply write them over.
372 static int ovs_ct_init_labels(struct nf_conn *ct, struct sw_flow_key *key,
373 const struct ovs_key_ct_labels *labels,
374 const struct ovs_key_ct_labels *mask)
376 struct nf_conn_labels *cl, *master_cl;
377 bool have_mask = labels_nonzero(mask);
379 /* Inherit master's labels to the related connection? */
380 master_cl = ct->master ? nf_ct_labels_find(ct->master) : NULL;
382 if (!master_cl && !have_mask)
383 return 0; /* Nothing to do. */
385 cl = ovs_ct_get_conn_labels(ct);
389 /* Inherit the master's labels, if any. */
394 u32 *dst = (u32 *)cl->bits;
397 for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
398 dst[i] = (dst[i] & ~mask->ct_labels_32[i]) |
399 (labels->ct_labels_32[i]
400 & mask->ct_labels_32[i]);
403 /* Labels are included in the IPCTNL_MSG_CT_NEW event only if the
404 * IPCT_LABEL bit is set in the event cache.
406 nf_conntrack_event_cache(IPCT_LABEL, ct);
408 memcpy(&key->ct.labels, cl->bits, OVS_CT_LABELS_LEN);
413 static int ovs_ct_set_labels(struct nf_conn *ct, struct sw_flow_key *key,
414 const struct ovs_key_ct_labels *labels,
415 const struct ovs_key_ct_labels *mask)
417 struct nf_conn_labels *cl;
420 cl = ovs_ct_get_conn_labels(ct);
424 err = nf_connlabels_replace(ct, labels->ct_labels_32,
426 OVS_CT_LABELS_LEN_32);
430 memcpy(&key->ct.labels, cl->bits, OVS_CT_LABELS_LEN);
435 /* 'skb' should already be pulled to nh_ofs. */
436 static int ovs_ct_helper(struct sk_buff *skb, u16 proto)
438 const struct nf_conntrack_helper *helper;
439 const struct nf_conn_help *help;
440 enum ip_conntrack_info ctinfo;
441 unsigned int protoff;
445 ct = nf_ct_get(skb, &ctinfo);
446 if (!ct || ctinfo == IP_CT_RELATED_REPLY)
449 help = nfct_help(ct);
453 helper = rcu_dereference(help->helper);
459 protoff = ip_hdrlen(skb);
462 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
466 ofs = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr,
468 if (ofs < 0 || (frag_off & htons(~0x7)) != 0) {
469 pr_debug("proto header not found\n");
476 WARN_ONCE(1, "helper invoked on non-IP family!");
480 err = helper->help(skb, protoff, ct, ctinfo);
481 if (err != NF_ACCEPT)
484 /* Adjust seqs after helper. This is needed due to some helpers (e.g.,
485 * FTP with NAT) adusting the TCP payload size when mangling IP
486 * addresses and/or port numbers in the text-based control connection.
488 if (test_bit(IPS_SEQ_ADJUST_BIT, &ct->status) &&
489 !nf_ct_seq_adjust(skb, ct, ctinfo, protoff))
494 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
495 * value if 'skb' is freed.
497 static int handle_fragments(struct net *net, struct sw_flow_key *key,
498 u16 zone, struct sk_buff *skb)
500 struct ovs_skb_cb ovs_cb = *OVS_CB(skb);
503 if (key->eth.type == htons(ETH_P_IP)) {
504 enum ip_defrag_users user = IP_DEFRAG_CONNTRACK_IN + zone;
506 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
507 err = ip_defrag(net, skb, user);
511 ovs_cb.mru = IPCB(skb)->frag_max_size;
512 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
513 } else if (key->eth.type == htons(ETH_P_IPV6)) {
514 enum ip6_defrag_users user = IP6_DEFRAG_CONNTRACK_IN + zone;
516 memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm));
517 err = nf_ct_frag6_gather(net, skb, user);
519 if (err != -EINPROGRESS)
524 key->ip.proto = ipv6_hdr(skb)->nexthdr;
525 ovs_cb.mru = IP6CB(skb)->frag_max_size;
529 return -EPFNOSUPPORT;
532 /* The key extracted from the fragment that completed this datagram
533 * likely didn't have an L4 header, so regenerate it.
535 ovs_flow_key_update_l3l4(skb, key);
537 key->ip.frag = OVS_FRAG_TYPE_NONE;
540 *OVS_CB(skb) = ovs_cb;
545 static struct nf_conntrack_expect *
546 ovs_ct_expect_find(struct net *net, const struct nf_conntrack_zone *zone,
547 u16 proto, const struct sk_buff *skb)
549 struct nf_conntrack_tuple tuple;
550 struct nf_conntrack_expect *exp;
552 if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), proto, net, &tuple))
555 exp = __nf_ct_expect_find(net, zone, &tuple);
557 struct nf_conntrack_tuple_hash *h;
559 /* Delete existing conntrack entry, if it clashes with the
560 * expectation. This can happen since conntrack ALGs do not
561 * check for clashes between (new) expectations and existing
562 * conntrack entries. nf_conntrack_in() will check the
563 * expectations only if a conntrack entry can not be found,
564 * which can lead to OVS finding the expectation (here) in the
565 * init direction, but which will not be removed by the
566 * nf_conntrack_in() call, if a matching conntrack entry is
567 * found instead. In this case all init direction packets
568 * would be reported as new related packets, while reply
569 * direction packets would be reported as un-related
570 * established packets.
572 h = nf_conntrack_find_get(net, zone, &tuple);
574 struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
576 nf_ct_delete(ct, 0, 0);
584 /* This replicates logic from nf_conntrack_core.c that is not exported. */
585 static enum ip_conntrack_info
586 ovs_ct_get_info(const struct nf_conntrack_tuple_hash *h)
588 const struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
590 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY)
591 return IP_CT_ESTABLISHED_REPLY;
592 /* Once we've had two way comms, always ESTABLISHED. */
593 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status))
594 return IP_CT_ESTABLISHED;
595 if (test_bit(IPS_EXPECTED_BIT, &ct->status))
596 return IP_CT_RELATED;
600 /* Find an existing connection which this packet belongs to without
601 * re-attributing statistics or modifying the connection state. This allows an
602 * skb->_nfct lost due to an upcall to be recovered during actions execution.
604 * Must be called with rcu_read_lock.
606 * On success, populates skb->_nfct and returns the connection. Returns NULL
607 * if there is no existing entry.
609 static struct nf_conn *
610 ovs_ct_find_existing(struct net *net, const struct nf_conntrack_zone *zone,
611 u8 l3num, struct sk_buff *skb, bool natted)
613 struct nf_conntrack_tuple tuple;
614 struct nf_conntrack_tuple_hash *h;
617 if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), l3num,
619 pr_debug("ovs_ct_find_existing: Can't get tuple\n");
623 /* Must invert the tuple if skb has been transformed by NAT. */
625 struct nf_conntrack_tuple inverse;
627 if (!nf_ct_invert_tuple(&inverse, &tuple)) {
628 pr_debug("ovs_ct_find_existing: Inversion failed!\n");
634 /* look for tuple match */
635 h = nf_conntrack_find_get(net, zone, &tuple);
637 return NULL; /* Not found. */
639 ct = nf_ct_tuplehash_to_ctrack(h);
641 /* Inverted packet tuple matches the reverse direction conntrack tuple,
642 * select the other tuplehash to get the right 'ctinfo' bits for this
646 h = &ct->tuplehash[!h->tuple.dst.dir];
648 nf_ct_set(skb, ct, ovs_ct_get_info(h));
653 struct nf_conn *ovs_ct_executed(struct net *net,
654 const struct sw_flow_key *key,
655 const struct ovs_conntrack_info *info,
659 struct nf_conn *ct = NULL;
661 /* If no ct, check if we have evidence that an existing conntrack entry
662 * might be found for this skb. This happens when we lose a skb->_nfct
663 * due to an upcall, or if the direction is being forced. If the
664 * connection was not confirmed, it is not cached and needs to be run
665 * through conntrack again.
667 *ct_executed = (key->ct_state & OVS_CS_F_TRACKED) &&
668 !(key->ct_state & OVS_CS_F_INVALID) &&
669 (key->ct_zone == info->zone.id);
671 if (*ct_executed || (!key->ct_state && info->force)) {
672 ct = ovs_ct_find_existing(net, &info->zone, info->family, skb,
680 /* Determine whether skb->_nfct is equal to the result of conntrack lookup. */
681 static bool skb_nfct_cached(struct net *net,
682 const struct sw_flow_key *key,
683 const struct ovs_conntrack_info *info,
686 enum ip_conntrack_info ctinfo;
688 bool ct_executed = true;
690 ct = nf_ct_get(skb, &ctinfo);
692 ct = ovs_ct_executed(net, key, info, skb, &ct_executed);
695 nf_ct_get(skb, &ctinfo);
699 if (!net_eq(net, read_pnet(&ct->ct_net)))
701 if (!nf_ct_zone_equal_any(info->ct, nf_ct_zone(ct)))
704 struct nf_conn_help *help;
706 help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER);
707 if (help && rcu_access_pointer(help->helper) != info->helper)
710 if (info->nf_ct_timeout) {
711 struct nf_conn_timeout *timeout_ext;
713 timeout_ext = nf_ct_timeout_find(ct);
714 if (!timeout_ext || info->nf_ct_timeout !=
715 rcu_dereference(timeout_ext->timeout))
718 /* Force conntrack entry direction to the current packet? */
719 if (info->force && CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) {
720 /* Delete the conntrack entry if confirmed, else just release
723 if (nf_ct_is_confirmed(ct))
724 nf_ct_delete(ct, 0, 0);
727 nf_ct_set(skb, NULL, 0);
734 #if IS_ENABLED(CONFIG_NF_NAT)
735 static void ovs_nat_update_key(struct sw_flow_key *key,
736 const struct sk_buff *skb,
737 enum nf_nat_manip_type maniptype)
739 if (maniptype == NF_NAT_MANIP_SRC) {
742 key->ct_state |= OVS_CS_F_SRC_NAT;
743 if (key->eth.type == htons(ETH_P_IP))
744 key->ipv4.addr.src = ip_hdr(skb)->saddr;
745 else if (key->eth.type == htons(ETH_P_IPV6))
746 memcpy(&key->ipv6.addr.src, &ipv6_hdr(skb)->saddr,
747 sizeof(key->ipv6.addr.src));
751 if (key->ip.proto == IPPROTO_UDP)
752 src = udp_hdr(skb)->source;
753 else if (key->ip.proto == IPPROTO_TCP)
754 src = tcp_hdr(skb)->source;
755 else if (key->ip.proto == IPPROTO_SCTP)
756 src = sctp_hdr(skb)->source;
764 key->ct_state |= OVS_CS_F_DST_NAT;
765 if (key->eth.type == htons(ETH_P_IP))
766 key->ipv4.addr.dst = ip_hdr(skb)->daddr;
767 else if (key->eth.type == htons(ETH_P_IPV6))
768 memcpy(&key->ipv6.addr.dst, &ipv6_hdr(skb)->daddr,
769 sizeof(key->ipv6.addr.dst));
773 if (key->ip.proto == IPPROTO_UDP)
774 dst = udp_hdr(skb)->dest;
775 else if (key->ip.proto == IPPROTO_TCP)
776 dst = tcp_hdr(skb)->dest;
777 else if (key->ip.proto == IPPROTO_SCTP)
778 dst = sctp_hdr(skb)->dest;
786 /* Modelled after nf_nat_ipv[46]_fn().
787 * range is only used for new, uninitialized NAT state.
788 * Returns either NF_ACCEPT or NF_DROP.
790 static int ovs_ct_nat_execute(struct sk_buff *skb, struct nf_conn *ct,
791 enum ip_conntrack_info ctinfo,
792 const struct nf_nat_range2 *range,
793 enum nf_nat_manip_type maniptype, struct sw_flow_key *key)
795 int hooknum, nh_off, err = NF_ACCEPT;
797 nh_off = skb_network_offset(skb);
798 skb_pull_rcsum(skb, nh_off);
800 /* See HOOK2MANIP(). */
801 if (maniptype == NF_NAT_MANIP_SRC)
802 hooknum = NF_INET_LOCAL_IN; /* Source NAT */
804 hooknum = NF_INET_LOCAL_OUT; /* Destination NAT */
808 case IP_CT_RELATED_REPLY:
809 if (IS_ENABLED(CONFIG_NF_NAT) &&
810 skb->protocol == htons(ETH_P_IP) &&
811 ip_hdr(skb)->protocol == IPPROTO_ICMP) {
812 if (!nf_nat_icmp_reply_translation(skb, ct, ctinfo,
816 } else if (IS_ENABLED(CONFIG_IPV6) &&
817 skb->protocol == htons(ETH_P_IPV6)) {
819 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
820 int hdrlen = ipv6_skip_exthdr(skb,
821 sizeof(struct ipv6hdr),
822 &nexthdr, &frag_off);
824 if (hdrlen >= 0 && nexthdr == IPPROTO_ICMPV6) {
825 if (!nf_nat_icmpv6_reply_translation(skb, ct,
833 /* Non-ICMP, fall thru to initialize if needed. */
836 /* Seen it before? This can happen for loopback, retrans,
839 if (!nf_nat_initialized(ct, maniptype)) {
840 /* Initialize according to the NAT action. */
841 err = (range && range->flags & NF_NAT_RANGE_MAP_IPS)
842 /* Action is set up to establish a new
845 ? nf_nat_setup_info(ct, range, maniptype)
846 : nf_nat_alloc_null_binding(ct, hooknum);
847 if (err != NF_ACCEPT)
852 case IP_CT_ESTABLISHED:
853 case IP_CT_ESTABLISHED_REPLY:
861 err = nf_nat_packet(ct, ctinfo, hooknum, skb);
863 skb_push_rcsum(skb, nh_off);
865 /* Update the flow key if NAT successful. */
866 if (err == NF_ACCEPT)
867 ovs_nat_update_key(key, skb, maniptype);
872 /* Returns NF_DROP if the packet should be dropped, NF_ACCEPT otherwise. */
873 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
874 const struct ovs_conntrack_info *info,
875 struct sk_buff *skb, struct nf_conn *ct,
876 enum ip_conntrack_info ctinfo)
878 enum nf_nat_manip_type maniptype;
881 /* Add NAT extension if not confirmed yet. */
882 if (!nf_ct_is_confirmed(ct) && !nf_ct_nat_ext_add(ct))
883 return NF_ACCEPT; /* Can't NAT. */
885 /* Determine NAT type.
886 * Check if the NAT type can be deduced from the tracked connection.
887 * Make sure new expected connections (IP_CT_RELATED) are NATted only
890 if (info->nat & OVS_CT_NAT && ctinfo != IP_CT_NEW &&
891 ct->status & IPS_NAT_MASK &&
892 (ctinfo != IP_CT_RELATED || info->commit)) {
893 /* NAT an established or related connection like before. */
894 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY)
895 /* This is the REPLY direction for a connection
896 * for which NAT was applied in the forward
897 * direction. Do the reverse NAT.
899 maniptype = ct->status & IPS_SRC_NAT
900 ? NF_NAT_MANIP_DST : NF_NAT_MANIP_SRC;
902 maniptype = ct->status & IPS_SRC_NAT
903 ? NF_NAT_MANIP_SRC : NF_NAT_MANIP_DST;
904 } else if (info->nat & OVS_CT_SRC_NAT) {
905 maniptype = NF_NAT_MANIP_SRC;
906 } else if (info->nat & OVS_CT_DST_NAT) {
907 maniptype = NF_NAT_MANIP_DST;
909 return NF_ACCEPT; /* Connection is not NATed. */
911 err = ovs_ct_nat_execute(skb, ct, ctinfo, &info->range, maniptype, key);
913 if (err == NF_ACCEPT && ct->status & IPS_DST_NAT) {
914 if (ct->status & IPS_SRC_NAT) {
915 if (maniptype == NF_NAT_MANIP_SRC)
916 maniptype = NF_NAT_MANIP_DST;
918 maniptype = NF_NAT_MANIP_SRC;
920 err = ovs_ct_nat_execute(skb, ct, ctinfo, &info->range,
922 } else if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) {
923 err = ovs_ct_nat_execute(skb, ct, ctinfo, NULL,
924 NF_NAT_MANIP_SRC, key);
930 #else /* !CONFIG_NF_NAT */
931 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
932 const struct ovs_conntrack_info *info,
933 struct sk_buff *skb, struct nf_conn *ct,
934 enum ip_conntrack_info ctinfo)
940 /* Pass 'skb' through conntrack in 'net', using zone configured in 'info', if
941 * not done already. Update key with new CT state after passing the packet
943 * Note that if the packet is deemed invalid by conntrack, skb->_nfct will be
944 * set to NULL and 0 will be returned.
946 static int __ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
947 const struct ovs_conntrack_info *info,
950 /* If we are recirculating packets to match on conntrack fields and
951 * committing with a separate conntrack action, then we don't need to
952 * actually run the packet through conntrack twice unless it's for a
955 bool cached = skb_nfct_cached(net, key, info, skb);
956 enum ip_conntrack_info ctinfo;
960 struct nf_hook_state state = {
961 .hook = NF_INET_PRE_ROUTING,
965 struct nf_conn *tmpl = info->ct;
968 /* Associate skb with specified zone. */
970 ct = nf_ct_get(skb, &ctinfo);
972 nf_conntrack_get(&tmpl->ct_general);
973 nf_ct_set(skb, tmpl, IP_CT_NEW);
976 err = nf_conntrack_in(skb, &state);
977 if (err != NF_ACCEPT)
980 /* Clear CT state NAT flags to mark that we have not yet done
981 * NAT after the nf_conntrack_in() call. We can actually clear
982 * the whole state, as it will be re-initialized below.
986 /* Update the key, but keep the NAT flags. */
987 ovs_ct_update_key(skb, info, key, true, true);
990 ct = nf_ct_get(skb, &ctinfo);
992 bool add_helper = false;
994 /* Packets starting a new connection must be NATted before the
995 * helper, so that the helper knows about the NAT. We enforce
996 * this by delaying both NAT and helper calls for unconfirmed
997 * connections until the committing CT action. For later
998 * packets NAT and Helper may be called in either order.
1000 * NAT will be done only if the CT action has NAT, and only
1001 * once per packet (per zone), as guarded by the NAT bits in
1002 * the key->ct_state.
1004 if (info->nat && !(key->ct_state & OVS_CS_F_NAT_MASK) &&
1005 (nf_ct_is_confirmed(ct) || info->commit) &&
1006 ovs_ct_nat(net, key, info, skb, ct, ctinfo) != NF_ACCEPT) {
1010 /* Userspace may decide to perform a ct lookup without a helper
1011 * specified followed by a (recirculate and) commit with one,
1012 * or attach a helper in a later commit. Therefore, for
1013 * connections which we will commit, we may need to attach
1016 if (info->commit && info->helper && !nfct_help(ct)) {
1017 int err = __nf_ct_try_assign_helper(ct, info->ct,
1023 /* helper installed, add seqadj if NAT is required */
1024 if (info->nat && !nfct_seqadj(ct)) {
1025 if (!nfct_seqadj_ext_add(ct))
1030 /* Call the helper only if:
1031 * - nf_conntrack_in() was executed above ("!cached") or a
1032 * helper was just attached ("add_helper") for a confirmed
1034 * - When committing an unconfirmed connection.
1036 if ((nf_ct_is_confirmed(ct) ? !cached || add_helper :
1038 ovs_ct_helper(skb, info->family) != NF_ACCEPT) {
1042 if (nf_ct_protonum(ct) == IPPROTO_TCP &&
1043 nf_ct_is_confirmed(ct) && nf_conntrack_tcp_established(ct)) {
1044 /* Be liberal for tcp packets so that out-of-window
1045 * packets are not marked invalid.
1047 nf_ct_set_tcp_be_liberal(ct);
1054 /* Lookup connection and read fields into key. */
1055 static int ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
1056 const struct ovs_conntrack_info *info,
1057 struct sk_buff *skb)
1059 struct nf_conntrack_expect *exp;
1061 /* If we pass an expected packet through nf_conntrack_in() the
1062 * expectation is typically removed, but the packet could still be
1063 * lost in upcall processing. To prevent this from happening we
1064 * perform an explicit expectation lookup. Expected connections are
1065 * always new, and will be passed through conntrack only when they are
1066 * committed, as it is OK to remove the expectation at that time.
1068 exp = ovs_ct_expect_find(net, &info->zone, info->family, skb);
1072 /* NOTE: New connections are NATted and Helped only when
1073 * committed, so we are not calling into NAT here.
1075 state = OVS_CS_F_TRACKED | OVS_CS_F_NEW | OVS_CS_F_RELATED;
1076 __ovs_ct_update_key(key, state, &info->zone, exp->master);
1081 err = __ovs_ct_lookup(net, key, info, skb);
1085 ct = (struct nf_conn *)skb_nfct(skb);
1087 nf_ct_deliver_cached_events(ct);
1093 static bool labels_nonzero(const struct ovs_key_ct_labels *labels)
1097 for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
1098 if (labels->ct_labels_32[i])
1104 #if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1105 static struct hlist_head *ct_limit_hash_bucket(
1106 const struct ovs_ct_limit_info *info, u16 zone)
1108 return &info->limits[zone & (CT_LIMIT_HASH_BUCKETS - 1)];
1111 /* Call with ovs_mutex */
1112 static void ct_limit_set(const struct ovs_ct_limit_info *info,
1113 struct ovs_ct_limit *new_ct_limit)
1115 struct ovs_ct_limit *ct_limit;
1116 struct hlist_head *head;
1118 head = ct_limit_hash_bucket(info, new_ct_limit->zone);
1119 hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
1120 if (ct_limit->zone == new_ct_limit->zone) {
1121 hlist_replace_rcu(&ct_limit->hlist_node,
1122 &new_ct_limit->hlist_node);
1123 kfree_rcu(ct_limit, rcu);
1128 hlist_add_head_rcu(&new_ct_limit->hlist_node, head);
1131 /* Call with ovs_mutex */
1132 static void ct_limit_del(const struct ovs_ct_limit_info *info, u16 zone)
1134 struct ovs_ct_limit *ct_limit;
1135 struct hlist_head *head;
1136 struct hlist_node *n;
1138 head = ct_limit_hash_bucket(info, zone);
1139 hlist_for_each_entry_safe(ct_limit, n, head, hlist_node) {
1140 if (ct_limit->zone == zone) {
1141 hlist_del_rcu(&ct_limit->hlist_node);
1142 kfree_rcu(ct_limit, rcu);
1148 /* Call with RCU read lock */
1149 static u32 ct_limit_get(const struct ovs_ct_limit_info *info, u16 zone)
1151 struct ovs_ct_limit *ct_limit;
1152 struct hlist_head *head;
1154 head = ct_limit_hash_bucket(info, zone);
1155 hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
1156 if (ct_limit->zone == zone)
1157 return ct_limit->limit;
1160 return info->default_limit;
1163 static int ovs_ct_check_limit(struct net *net,
1164 const struct ovs_conntrack_info *info,
1165 const struct nf_conntrack_tuple *tuple)
1167 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1168 const struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
1169 u32 per_zone_limit, connections;
1172 conncount_key = info->zone.id;
1174 per_zone_limit = ct_limit_get(ct_limit_info, info->zone.id);
1175 if (per_zone_limit == OVS_CT_LIMIT_UNLIMITED)
1178 connections = nf_conncount_count(net, ct_limit_info->data,
1179 &conncount_key, tuple, &info->zone);
1180 if (connections > per_zone_limit)
1187 /* Lookup connection and confirm if unconfirmed. */
1188 static int ovs_ct_commit(struct net *net, struct sw_flow_key *key,
1189 const struct ovs_conntrack_info *info,
1190 struct sk_buff *skb)
1192 enum ip_conntrack_info ctinfo;
1196 err = __ovs_ct_lookup(net, key, info, skb);
1200 /* The connection could be invalid, in which case this is a no-op.*/
1201 ct = nf_ct_get(skb, &ctinfo);
1205 #if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1206 if (static_branch_unlikely(&ovs_ct_limit_enabled)) {
1207 if (!nf_ct_is_confirmed(ct)) {
1208 err = ovs_ct_check_limit(net, info,
1209 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
1211 net_warn_ratelimited("openvswitch: zone: %u "
1212 "exceeds conntrack limit\n",
1220 /* Set the conntrack event mask if given. NEW and DELETE events have
1221 * their own groups, but the NFNLGRP_CONNTRACK_UPDATE group listener
1222 * typically would receive many kinds of updates. Setting the event
1223 * mask allows those events to be filtered. The set event mask will
1224 * remain in effect for the lifetime of the connection unless changed
1225 * by a further CT action with both the commit flag and the eventmask
1227 if (info->have_eventmask) {
1228 struct nf_conntrack_ecache *cache = nf_ct_ecache_find(ct);
1231 cache->ctmask = info->eventmask;
1234 /* Apply changes before confirming the connection so that the initial
1235 * conntrack NEW netlink event carries the values given in the CT
1238 if (info->mark.mask) {
1239 err = ovs_ct_set_mark(ct, key, info->mark.value,
1244 if (!nf_ct_is_confirmed(ct)) {
1245 err = ovs_ct_init_labels(ct, key, &info->labels.value,
1246 &info->labels.mask);
1249 } else if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1250 labels_nonzero(&info->labels.mask)) {
1251 err = ovs_ct_set_labels(ct, key, &info->labels.value,
1252 &info->labels.mask);
1256 /* This will take care of sending queued events even if the connection
1257 * is already confirmed.
1259 if (nf_conntrack_confirm(skb) != NF_ACCEPT)
1265 /* Trim the skb to the length specified by the IP/IPv6 header,
1266 * removing any trailing lower-layer padding. This prepares the skb
1267 * for higher-layer processing that assumes skb->len excludes padding
1268 * (such as nf_ip_checksum). The caller needs to pull the skb to the
1269 * network header, and ensure ip_hdr/ipv6_hdr points to valid data.
1271 static int ovs_skb_network_trim(struct sk_buff *skb)
1276 switch (skb->protocol) {
1277 case htons(ETH_P_IP):
1278 len = ntohs(ip_hdr(skb)->tot_len);
1280 case htons(ETH_P_IPV6):
1281 len = sizeof(struct ipv6hdr)
1282 + ntohs(ipv6_hdr(skb)->payload_len);
1288 err = pskb_trim_rcsum(skb, len);
1295 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
1296 * value if 'skb' is freed.
1298 int ovs_ct_execute(struct net *net, struct sk_buff *skb,
1299 struct sw_flow_key *key,
1300 const struct ovs_conntrack_info *info)
1305 /* The conntrack module expects to be working at L3. */
1306 nh_ofs = skb_network_offset(skb);
1307 skb_pull_rcsum(skb, nh_ofs);
1309 err = ovs_skb_network_trim(skb);
1313 if (key->ip.frag != OVS_FRAG_TYPE_NONE) {
1314 err = handle_fragments(net, key, info->zone.id, skb);
1320 err = ovs_ct_commit(net, key, info, skb);
1322 err = ovs_ct_lookup(net, key, info, skb);
1324 skb_push_rcsum(skb, nh_ofs);
1330 int ovs_ct_clear(struct sk_buff *skb, struct sw_flow_key *key)
1332 enum ip_conntrack_info ctinfo;
1335 ct = nf_ct_get(skb, &ctinfo);
1338 nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
1341 ovs_ct_fill_key(skb, key, false);
1346 static int ovs_ct_add_helper(struct ovs_conntrack_info *info, const char *name,
1347 const struct sw_flow_key *key, bool log)
1349 struct nf_conntrack_helper *helper;
1350 struct nf_conn_help *help;
1353 helper = nf_conntrack_helper_try_module_get(name, info->family,
1356 OVS_NLERR(log, "Unknown helper \"%s\"", name);
1360 help = nf_ct_helper_ext_add(info->ct, GFP_KERNEL);
1362 nf_conntrack_helper_put(helper);
1366 #if IS_ENABLED(CONFIG_NF_NAT)
1368 ret = nf_nat_helper_try_module_get(name, info->family,
1371 nf_conntrack_helper_put(helper);
1372 OVS_NLERR(log, "Failed to load \"%s\" NAT helper, error: %d",
1378 rcu_assign_pointer(help->helper, helper);
1379 info->helper = helper;
1383 #if IS_ENABLED(CONFIG_NF_NAT)
1384 static int parse_nat(const struct nlattr *attr,
1385 struct ovs_conntrack_info *info, bool log)
1389 bool have_ip_max = false;
1390 bool have_proto_max = false;
1391 bool ip_vers = (info->family == NFPROTO_IPV6);
1393 nla_for_each_nested(a, attr, rem) {
1394 static const int ovs_nat_attr_lens[OVS_NAT_ATTR_MAX + 1][2] = {
1395 [OVS_NAT_ATTR_SRC] = {0, 0},
1396 [OVS_NAT_ATTR_DST] = {0, 0},
1397 [OVS_NAT_ATTR_IP_MIN] = {sizeof(struct in_addr),
1398 sizeof(struct in6_addr)},
1399 [OVS_NAT_ATTR_IP_MAX] = {sizeof(struct in_addr),
1400 sizeof(struct in6_addr)},
1401 [OVS_NAT_ATTR_PROTO_MIN] = {sizeof(u16), sizeof(u16)},
1402 [OVS_NAT_ATTR_PROTO_MAX] = {sizeof(u16), sizeof(u16)},
1403 [OVS_NAT_ATTR_PERSISTENT] = {0, 0},
1404 [OVS_NAT_ATTR_PROTO_HASH] = {0, 0},
1405 [OVS_NAT_ATTR_PROTO_RANDOM] = {0, 0},
1407 int type = nla_type(a);
1409 if (type > OVS_NAT_ATTR_MAX) {
1410 OVS_NLERR(log, "Unknown NAT attribute (type=%d, max=%d)",
1411 type, OVS_NAT_ATTR_MAX);
1415 if (nla_len(a) != ovs_nat_attr_lens[type][ip_vers]) {
1416 OVS_NLERR(log, "NAT attribute type %d has unexpected length (%d != %d)",
1418 ovs_nat_attr_lens[type][ip_vers]);
1423 case OVS_NAT_ATTR_SRC:
1424 case OVS_NAT_ATTR_DST:
1426 OVS_NLERR(log, "Only one type of NAT may be specified");
1429 info->nat |= OVS_CT_NAT;
1430 info->nat |= ((type == OVS_NAT_ATTR_SRC)
1431 ? OVS_CT_SRC_NAT : OVS_CT_DST_NAT);
1434 case OVS_NAT_ATTR_IP_MIN:
1435 nla_memcpy(&info->range.min_addr, a,
1436 sizeof(info->range.min_addr));
1437 info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1440 case OVS_NAT_ATTR_IP_MAX:
1442 nla_memcpy(&info->range.max_addr, a,
1443 sizeof(info->range.max_addr));
1444 info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1447 case OVS_NAT_ATTR_PROTO_MIN:
1448 info->range.min_proto.all = htons(nla_get_u16(a));
1449 info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1452 case OVS_NAT_ATTR_PROTO_MAX:
1453 have_proto_max = true;
1454 info->range.max_proto.all = htons(nla_get_u16(a));
1455 info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1458 case OVS_NAT_ATTR_PERSISTENT:
1459 info->range.flags |= NF_NAT_RANGE_PERSISTENT;
1462 case OVS_NAT_ATTR_PROTO_HASH:
1463 info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM;
1466 case OVS_NAT_ATTR_PROTO_RANDOM:
1467 info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM_FULLY;
1471 OVS_NLERR(log, "Unknown nat attribute (%d)", type);
1477 OVS_NLERR(log, "NAT attribute has %d unknown bytes", rem);
1481 /* Do not allow flags if no type is given. */
1482 if (info->range.flags) {
1484 "NAT flags may be given only when NAT range (SRC or DST) is also specified."
1488 info->nat = OVS_CT_NAT; /* NAT existing connections. */
1489 } else if (!info->commit) {
1491 "NAT attributes may be specified only when CT COMMIT flag is also specified."
1495 /* Allow missing IP_MAX. */
1496 if (info->range.flags & NF_NAT_RANGE_MAP_IPS && !have_ip_max) {
1497 memcpy(&info->range.max_addr, &info->range.min_addr,
1498 sizeof(info->range.max_addr));
1500 /* Allow missing PROTO_MAX. */
1501 if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1503 info->range.max_proto.all = info->range.min_proto.all;
1509 static const struct ovs_ct_len_tbl ovs_ct_attr_lens[OVS_CT_ATTR_MAX + 1] = {
1510 [OVS_CT_ATTR_COMMIT] = { .minlen = 0, .maxlen = 0 },
1511 [OVS_CT_ATTR_FORCE_COMMIT] = { .minlen = 0, .maxlen = 0 },
1512 [OVS_CT_ATTR_ZONE] = { .minlen = sizeof(u16),
1513 .maxlen = sizeof(u16) },
1514 [OVS_CT_ATTR_MARK] = { .minlen = sizeof(struct md_mark),
1515 .maxlen = sizeof(struct md_mark) },
1516 [OVS_CT_ATTR_LABELS] = { .minlen = sizeof(struct md_labels),
1517 .maxlen = sizeof(struct md_labels) },
1518 [OVS_CT_ATTR_HELPER] = { .minlen = 1,
1519 .maxlen = NF_CT_HELPER_NAME_LEN },
1520 #if IS_ENABLED(CONFIG_NF_NAT)
1521 /* NAT length is checked when parsing the nested attributes. */
1522 [OVS_CT_ATTR_NAT] = { .minlen = 0, .maxlen = INT_MAX },
1524 [OVS_CT_ATTR_EVENTMASK] = { .minlen = sizeof(u32),
1525 .maxlen = sizeof(u32) },
1526 [OVS_CT_ATTR_TIMEOUT] = { .minlen = 1,
1527 .maxlen = CTNL_TIMEOUT_NAME_MAX },
1530 static int parse_ct(const struct nlattr *attr, struct ovs_conntrack_info *info,
1531 const char **helper, bool log)
1536 nla_for_each_nested(a, attr, rem) {
1537 int type = nla_type(a);
1541 if (type > OVS_CT_ATTR_MAX) {
1543 "Unknown conntrack attr (type=%d, max=%d)",
1544 type, OVS_CT_ATTR_MAX);
1548 maxlen = ovs_ct_attr_lens[type].maxlen;
1549 minlen = ovs_ct_attr_lens[type].minlen;
1550 if (nla_len(a) < minlen || nla_len(a) > maxlen) {
1552 "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
1553 type, nla_len(a), maxlen);
1558 case OVS_CT_ATTR_FORCE_COMMIT:
1561 case OVS_CT_ATTR_COMMIT:
1562 info->commit = true;
1564 #ifdef CONFIG_NF_CONNTRACK_ZONES
1565 case OVS_CT_ATTR_ZONE:
1566 info->zone.id = nla_get_u16(a);
1569 #ifdef CONFIG_NF_CONNTRACK_MARK
1570 case OVS_CT_ATTR_MARK: {
1571 struct md_mark *mark = nla_data(a);
1574 OVS_NLERR(log, "ct_mark mask cannot be 0");
1581 #ifdef CONFIG_NF_CONNTRACK_LABELS
1582 case OVS_CT_ATTR_LABELS: {
1583 struct md_labels *labels = nla_data(a);
1585 if (!labels_nonzero(&labels->mask)) {
1586 OVS_NLERR(log, "ct_labels mask cannot be 0");
1589 info->labels = *labels;
1593 case OVS_CT_ATTR_HELPER:
1594 *helper = nla_data(a);
1595 if (!memchr(*helper, '\0', nla_len(a))) {
1596 OVS_NLERR(log, "Invalid conntrack helper");
1600 #if IS_ENABLED(CONFIG_NF_NAT)
1601 case OVS_CT_ATTR_NAT: {
1602 int err = parse_nat(a, info, log);
1609 case OVS_CT_ATTR_EVENTMASK:
1610 info->have_eventmask = true;
1611 info->eventmask = nla_get_u32(a);
1613 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT
1614 case OVS_CT_ATTR_TIMEOUT:
1615 memcpy(info->timeout, nla_data(a), nla_len(a));
1616 if (!memchr(info->timeout, '\0', nla_len(a))) {
1617 OVS_NLERR(log, "Invalid conntrack timeout");
1624 OVS_NLERR(log, "Unknown conntrack attr (%d)",
1630 #ifdef CONFIG_NF_CONNTRACK_MARK
1631 if (!info->commit && info->mark.mask) {
1633 "Setting conntrack mark requires 'commit' flag.");
1637 #ifdef CONFIG_NF_CONNTRACK_LABELS
1638 if (!info->commit && labels_nonzero(&info->labels.mask)) {
1640 "Setting conntrack labels requires 'commit' flag.");
1645 OVS_NLERR(log, "Conntrack attr has %d unknown bytes", rem);
1652 bool ovs_ct_verify(struct net *net, enum ovs_key_attr attr)
1654 if (attr == OVS_KEY_ATTR_CT_STATE)
1656 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1657 attr == OVS_KEY_ATTR_CT_ZONE)
1659 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1660 attr == OVS_KEY_ATTR_CT_MARK)
1662 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1663 attr == OVS_KEY_ATTR_CT_LABELS) {
1664 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1666 return ovs_net->xt_label;
1672 int ovs_ct_copy_action(struct net *net, const struct nlattr *attr,
1673 const struct sw_flow_key *key,
1674 struct sw_flow_actions **sfa, bool log)
1676 struct ovs_conntrack_info ct_info;
1677 const char *helper = NULL;
1681 family = key_to_nfproto(key);
1682 if (family == NFPROTO_UNSPEC) {
1683 OVS_NLERR(log, "ct family unspecified");
1687 memset(&ct_info, 0, sizeof(ct_info));
1688 ct_info.family = family;
1690 nf_ct_zone_init(&ct_info.zone, NF_CT_DEFAULT_ZONE_ID,
1691 NF_CT_DEFAULT_ZONE_DIR, 0);
1693 err = parse_ct(attr, &ct_info, &helper, log);
1697 /* Set up template for tracking connections in specific zones. */
1698 ct_info.ct = nf_ct_tmpl_alloc(net, &ct_info.zone, GFP_KERNEL);
1700 OVS_NLERR(log, "Failed to allocate conntrack template");
1704 if (ct_info.timeout[0]) {
1705 if (nf_ct_set_timeout(net, ct_info.ct, family, key->ip.proto,
1707 pr_info_ratelimited("Failed to associated timeout "
1708 "policy `%s'\n", ct_info.timeout);
1710 ct_info.nf_ct_timeout = rcu_dereference(
1711 nf_ct_timeout_find(ct_info.ct)->timeout);
1716 err = ovs_ct_add_helper(&ct_info, helper, key, log);
1721 err = ovs_nla_add_action(sfa, OVS_ACTION_ATTR_CT, &ct_info,
1722 sizeof(ct_info), log);
1726 __set_bit(IPS_CONFIRMED_BIT, &ct_info.ct->status);
1729 __ovs_ct_free_action(&ct_info);
1733 #if IS_ENABLED(CONFIG_NF_NAT)
1734 static bool ovs_ct_nat_to_attr(const struct ovs_conntrack_info *info,
1735 struct sk_buff *skb)
1737 struct nlattr *start;
1739 start = nla_nest_start_noflag(skb, OVS_CT_ATTR_NAT);
1743 if (info->nat & OVS_CT_SRC_NAT) {
1744 if (nla_put_flag(skb, OVS_NAT_ATTR_SRC))
1746 } else if (info->nat & OVS_CT_DST_NAT) {
1747 if (nla_put_flag(skb, OVS_NAT_ATTR_DST))
1753 if (info->range.flags & NF_NAT_RANGE_MAP_IPS) {
1754 if (IS_ENABLED(CONFIG_NF_NAT) &&
1755 info->family == NFPROTO_IPV4) {
1756 if (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MIN,
1757 info->range.min_addr.ip) ||
1758 (info->range.max_addr.ip
1759 != info->range.min_addr.ip &&
1760 (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MAX,
1761 info->range.max_addr.ip))))
1763 } else if (IS_ENABLED(CONFIG_IPV6) &&
1764 info->family == NFPROTO_IPV6) {
1765 if (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MIN,
1766 &info->range.min_addr.in6) ||
1767 (memcmp(&info->range.max_addr.in6,
1768 &info->range.min_addr.in6,
1769 sizeof(info->range.max_addr.in6)) &&
1770 (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MAX,
1771 &info->range.max_addr.in6))))
1777 if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1778 (nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MIN,
1779 ntohs(info->range.min_proto.all)) ||
1780 (info->range.max_proto.all != info->range.min_proto.all &&
1781 nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MAX,
1782 ntohs(info->range.max_proto.all)))))
1785 if (info->range.flags & NF_NAT_RANGE_PERSISTENT &&
1786 nla_put_flag(skb, OVS_NAT_ATTR_PERSISTENT))
1788 if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM &&
1789 nla_put_flag(skb, OVS_NAT_ATTR_PROTO_HASH))
1791 if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM_FULLY &&
1792 nla_put_flag(skb, OVS_NAT_ATTR_PROTO_RANDOM))
1795 nla_nest_end(skb, start);
1801 int ovs_ct_action_to_attr(const struct ovs_conntrack_info *ct_info,
1802 struct sk_buff *skb)
1804 struct nlattr *start;
1806 start = nla_nest_start_noflag(skb, OVS_ACTION_ATTR_CT);
1810 if (ct_info->commit && nla_put_flag(skb, ct_info->force
1811 ? OVS_CT_ATTR_FORCE_COMMIT
1812 : OVS_CT_ATTR_COMMIT))
1814 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1815 nla_put_u16(skb, OVS_CT_ATTR_ZONE, ct_info->zone.id))
1817 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && ct_info->mark.mask &&
1818 nla_put(skb, OVS_CT_ATTR_MARK, sizeof(ct_info->mark),
1821 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1822 labels_nonzero(&ct_info->labels.mask) &&
1823 nla_put(skb, OVS_CT_ATTR_LABELS, sizeof(ct_info->labels),
1826 if (ct_info->helper) {
1827 if (nla_put_string(skb, OVS_CT_ATTR_HELPER,
1828 ct_info->helper->name))
1831 if (ct_info->have_eventmask &&
1832 nla_put_u32(skb, OVS_CT_ATTR_EVENTMASK, ct_info->eventmask))
1834 if (ct_info->timeout[0]) {
1835 if (nla_put_string(skb, OVS_CT_ATTR_TIMEOUT, ct_info->timeout))
1839 #if IS_ENABLED(CONFIG_NF_NAT)
1840 if (ct_info->nat && !ovs_ct_nat_to_attr(ct_info, skb))
1843 nla_nest_end(skb, start);
1848 void ovs_ct_free_action(const struct nlattr *a)
1850 struct ovs_conntrack_info *ct_info = nla_data(a);
1852 __ovs_ct_free_action(ct_info);
1855 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info)
1857 if (ct_info->helper) {
1858 #if IS_ENABLED(CONFIG_NF_NAT)
1860 nf_nat_helper_put(ct_info->helper);
1862 nf_conntrack_helper_put(ct_info->helper);
1865 if (ct_info->timeout[0])
1866 nf_ct_destroy_timeout(ct_info->ct);
1867 nf_ct_tmpl_free(ct_info->ct);
1871 #if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1872 static int ovs_ct_limit_init(struct net *net, struct ovs_net *ovs_net)
1876 ovs_net->ct_limit_info = kmalloc(sizeof(*ovs_net->ct_limit_info),
1878 if (!ovs_net->ct_limit_info)
1881 ovs_net->ct_limit_info->default_limit = OVS_CT_LIMIT_DEFAULT;
1882 ovs_net->ct_limit_info->limits =
1883 kmalloc_array(CT_LIMIT_HASH_BUCKETS, sizeof(struct hlist_head),
1885 if (!ovs_net->ct_limit_info->limits) {
1886 kfree(ovs_net->ct_limit_info);
1890 for (i = 0; i < CT_LIMIT_HASH_BUCKETS; i++)
1891 INIT_HLIST_HEAD(&ovs_net->ct_limit_info->limits[i]);
1893 ovs_net->ct_limit_info->data =
1894 nf_conncount_init(net, NFPROTO_INET, sizeof(u32));
1896 if (IS_ERR(ovs_net->ct_limit_info->data)) {
1897 err = PTR_ERR(ovs_net->ct_limit_info->data);
1898 kfree(ovs_net->ct_limit_info->limits);
1899 kfree(ovs_net->ct_limit_info);
1900 pr_err("openvswitch: failed to init nf_conncount %d\n", err);
1906 static void ovs_ct_limit_exit(struct net *net, struct ovs_net *ovs_net)
1908 const struct ovs_ct_limit_info *info = ovs_net->ct_limit_info;
1911 nf_conncount_destroy(net, NFPROTO_INET, info->data);
1912 for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) {
1913 struct hlist_head *head = &info->limits[i];
1914 struct ovs_ct_limit *ct_limit;
1916 hlist_for_each_entry_rcu(ct_limit, head, hlist_node,
1917 lockdep_ovsl_is_held())
1918 kfree_rcu(ct_limit, rcu);
1920 kfree(info->limits);
1924 static struct sk_buff *
1925 ovs_ct_limit_cmd_reply_start(struct genl_info *info, u8 cmd,
1926 struct ovs_header **ovs_reply_header)
1928 struct ovs_header *ovs_header = info->userhdr;
1929 struct sk_buff *skb;
1931 skb = genlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1933 return ERR_PTR(-ENOMEM);
1935 *ovs_reply_header = genlmsg_put(skb, info->snd_portid,
1937 &dp_ct_limit_genl_family, 0, cmd);
1939 if (!*ovs_reply_header) {
1941 return ERR_PTR(-EMSGSIZE);
1943 (*ovs_reply_header)->dp_ifindex = ovs_header->dp_ifindex;
1948 static bool check_zone_id(int zone_id, u16 *pzone)
1950 if (zone_id >= 0 && zone_id <= 65535) {
1951 *pzone = (u16)zone_id;
1957 static int ovs_ct_limit_set_zone_limit(struct nlattr *nla_zone_limit,
1958 struct ovs_ct_limit_info *info)
1960 struct ovs_zone_limit *zone_limit;
1964 rem = NLA_ALIGN(nla_len(nla_zone_limit));
1965 zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1967 while (rem >= sizeof(*zone_limit)) {
1968 if (unlikely(zone_limit->zone_id ==
1969 OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1971 info->default_limit = zone_limit->limit;
1973 } else if (unlikely(!check_zone_id(
1974 zone_limit->zone_id, &zone))) {
1975 OVS_NLERR(true, "zone id is out of range");
1977 struct ovs_ct_limit *ct_limit;
1979 ct_limit = kmalloc(sizeof(*ct_limit), GFP_KERNEL);
1983 ct_limit->zone = zone;
1984 ct_limit->limit = zone_limit->limit;
1987 ct_limit_set(info, ct_limit);
1990 rem -= NLA_ALIGN(sizeof(*zone_limit));
1991 zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
1992 NLA_ALIGN(sizeof(*zone_limit)));
1996 OVS_NLERR(true, "set zone limit has %d unknown bytes", rem);
2001 static int ovs_ct_limit_del_zone_limit(struct nlattr *nla_zone_limit,
2002 struct ovs_ct_limit_info *info)
2004 struct ovs_zone_limit *zone_limit;
2008 rem = NLA_ALIGN(nla_len(nla_zone_limit));
2009 zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
2011 while (rem >= sizeof(*zone_limit)) {
2012 if (unlikely(zone_limit->zone_id ==
2013 OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
2015 info->default_limit = OVS_CT_LIMIT_DEFAULT;
2017 } else if (unlikely(!check_zone_id(
2018 zone_limit->zone_id, &zone))) {
2019 OVS_NLERR(true, "zone id is out of range");
2022 ct_limit_del(info, zone);
2025 rem -= NLA_ALIGN(sizeof(*zone_limit));
2026 zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
2027 NLA_ALIGN(sizeof(*zone_limit)));
2031 OVS_NLERR(true, "del zone limit has %d unknown bytes", rem);
2036 static int ovs_ct_limit_get_default_limit(struct ovs_ct_limit_info *info,
2037 struct sk_buff *reply)
2039 struct ovs_zone_limit zone_limit = {
2040 .zone_id = OVS_ZONE_LIMIT_DEFAULT_ZONE,
2041 .limit = info->default_limit,
2044 return nla_put_nohdr(reply, sizeof(zone_limit), &zone_limit);
2047 static int __ovs_ct_limit_get_zone_limit(struct net *net,
2048 struct nf_conncount_data *data,
2049 u16 zone_id, u32 limit,
2050 struct sk_buff *reply)
2052 struct nf_conntrack_zone ct_zone;
2053 struct ovs_zone_limit zone_limit;
2054 u32 conncount_key = zone_id;
2056 zone_limit.zone_id = zone_id;
2057 zone_limit.limit = limit;
2058 nf_ct_zone_init(&ct_zone, zone_id, NF_CT_DEFAULT_ZONE_DIR, 0);
2060 zone_limit.count = nf_conncount_count(net, data, &conncount_key, NULL,
2062 return nla_put_nohdr(reply, sizeof(zone_limit), &zone_limit);
2065 static int ovs_ct_limit_get_zone_limit(struct net *net,
2066 struct nlattr *nla_zone_limit,
2067 struct ovs_ct_limit_info *info,
2068 struct sk_buff *reply)
2070 struct ovs_zone_limit *zone_limit;
2075 rem = NLA_ALIGN(nla_len(nla_zone_limit));
2076 zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
2078 while (rem >= sizeof(*zone_limit)) {
2079 if (unlikely(zone_limit->zone_id ==
2080 OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
2081 err = ovs_ct_limit_get_default_limit(info, reply);
2084 } else if (unlikely(!check_zone_id(zone_limit->zone_id,
2086 OVS_NLERR(true, "zone id is out of range");
2089 limit = ct_limit_get(info, zone);
2092 err = __ovs_ct_limit_get_zone_limit(
2093 net, info->data, zone, limit, reply);
2097 rem -= NLA_ALIGN(sizeof(*zone_limit));
2098 zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
2099 NLA_ALIGN(sizeof(*zone_limit)));
2103 OVS_NLERR(true, "get zone limit has %d unknown bytes", rem);
2108 static int ovs_ct_limit_get_all_zone_limit(struct net *net,
2109 struct ovs_ct_limit_info *info,
2110 struct sk_buff *reply)
2112 struct ovs_ct_limit *ct_limit;
2113 struct hlist_head *head;
2116 err = ovs_ct_limit_get_default_limit(info, reply);
2121 for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) {
2122 head = &info->limits[i];
2123 hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
2124 err = __ovs_ct_limit_get_zone_limit(net, info->data,
2125 ct_limit->zone, ct_limit->limit, reply);
2136 static int ovs_ct_limit_cmd_set(struct sk_buff *skb, struct genl_info *info)
2138 struct nlattr **a = info->attrs;
2139 struct sk_buff *reply;
2140 struct ovs_header *ovs_reply_header;
2141 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
2142 struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2145 reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_SET,
2148 return PTR_ERR(reply);
2150 if (!a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2155 err = ovs_ct_limit_set_zone_limit(a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT],
2160 static_branch_enable(&ovs_ct_limit_enabled);
2162 genlmsg_end(reply, ovs_reply_header);
2163 return genlmsg_reply(reply, info);
2170 static int ovs_ct_limit_cmd_del(struct sk_buff *skb, struct genl_info *info)
2172 struct nlattr **a = info->attrs;
2173 struct sk_buff *reply;
2174 struct ovs_header *ovs_reply_header;
2175 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
2176 struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2179 reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_DEL,
2182 return PTR_ERR(reply);
2184 if (!a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2189 err = ovs_ct_limit_del_zone_limit(a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT],
2194 genlmsg_end(reply, ovs_reply_header);
2195 return genlmsg_reply(reply, info);
2202 static int ovs_ct_limit_cmd_get(struct sk_buff *skb, struct genl_info *info)
2204 struct nlattr **a = info->attrs;
2205 struct nlattr *nla_reply;
2206 struct sk_buff *reply;
2207 struct ovs_header *ovs_reply_header;
2208 struct net *net = sock_net(skb->sk);
2209 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2210 struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2213 reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_GET,
2216 return PTR_ERR(reply);
2218 nla_reply = nla_nest_start_noflag(reply, OVS_CT_LIMIT_ATTR_ZONE_LIMIT);
2224 if (a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2225 err = ovs_ct_limit_get_zone_limit(
2226 net, a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT], ct_limit_info,
2231 err = ovs_ct_limit_get_all_zone_limit(net, ct_limit_info,
2237 nla_nest_end(reply, nla_reply);
2238 genlmsg_end(reply, ovs_reply_header);
2239 return genlmsg_reply(reply, info);
2246 static const struct genl_small_ops ct_limit_genl_ops[] = {
2247 { .cmd = OVS_CT_LIMIT_CMD_SET,
2248 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2249 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN
2251 .doit = ovs_ct_limit_cmd_set,
2253 { .cmd = OVS_CT_LIMIT_CMD_DEL,
2254 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2255 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN
2257 .doit = ovs_ct_limit_cmd_del,
2259 { .cmd = OVS_CT_LIMIT_CMD_GET,
2260 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2261 .flags = 0, /* OK for unprivileged users. */
2262 .doit = ovs_ct_limit_cmd_get,
2266 static const struct genl_multicast_group ovs_ct_limit_multicast_group = {
2267 .name = OVS_CT_LIMIT_MCGROUP,
2270 struct genl_family dp_ct_limit_genl_family __ro_after_init = {
2271 .hdrsize = sizeof(struct ovs_header),
2272 .name = OVS_CT_LIMIT_FAMILY,
2273 .version = OVS_CT_LIMIT_VERSION,
2274 .maxattr = OVS_CT_LIMIT_ATTR_MAX,
2275 .policy = ct_limit_policy,
2277 .parallel_ops = true,
2278 .small_ops = ct_limit_genl_ops,
2279 .n_small_ops = ARRAY_SIZE(ct_limit_genl_ops),
2280 .mcgrps = &ovs_ct_limit_multicast_group,
2282 .module = THIS_MODULE,
2286 int ovs_ct_init(struct net *net)
2288 unsigned int n_bits = sizeof(struct ovs_key_ct_labels) * BITS_PER_BYTE;
2289 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2291 if (nf_connlabels_get(net, n_bits - 1)) {
2292 ovs_net->xt_label = false;
2293 OVS_NLERR(true, "Failed to set connlabel length");
2295 ovs_net->xt_label = true;
2298 #if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2299 return ovs_ct_limit_init(net, ovs_net);
2305 void ovs_ct_exit(struct net *net)
2307 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2309 #if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2310 ovs_ct_limit_exit(net, ovs_net);
2313 if (ovs_net->xt_label)
2314 nf_connlabels_put(net);