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 ? 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, struct sw_flow_key *key)
276 ovs_ct_update_key(skb, NULL, key, false, false);
279 int ovs_ct_put_key(const struct sw_flow_key *swkey,
280 const struct sw_flow_key *output, struct sk_buff *skb)
282 if (nla_put_u32(skb, OVS_KEY_ATTR_CT_STATE, output->ct_state))
285 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
286 nla_put_u16(skb, OVS_KEY_ATTR_CT_ZONE, output->ct_zone))
289 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
290 nla_put_u32(skb, OVS_KEY_ATTR_CT_MARK, output->ct.mark))
293 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
294 nla_put(skb, OVS_KEY_ATTR_CT_LABELS, sizeof(output->ct.labels),
298 if (swkey->ct_orig_proto) {
299 if (swkey->eth.type == htons(ETH_P_IP)) {
300 struct ovs_key_ct_tuple_ipv4 orig;
302 memset(&orig, 0, sizeof(orig));
303 orig.ipv4_src = output->ipv4.ct_orig.src;
304 orig.ipv4_dst = output->ipv4.ct_orig.dst;
305 orig.src_port = output->ct.orig_tp.src;
306 orig.dst_port = output->ct.orig_tp.dst;
307 orig.ipv4_proto = output->ct_orig_proto;
309 if (nla_put(skb, OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4,
310 sizeof(orig), &orig))
312 } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
313 struct ovs_key_ct_tuple_ipv6 orig;
315 memset(&orig, 0, sizeof(orig));
316 memcpy(orig.ipv6_src, output->ipv6.ct_orig.src.s6_addr32,
317 sizeof(orig.ipv6_src));
318 memcpy(orig.ipv6_dst, output->ipv6.ct_orig.dst.s6_addr32,
319 sizeof(orig.ipv6_dst));
320 orig.src_port = output->ct.orig_tp.src;
321 orig.dst_port = output->ct.orig_tp.dst;
322 orig.ipv6_proto = output->ct_orig_proto;
324 if (nla_put(skb, OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6,
325 sizeof(orig), &orig))
333 static int ovs_ct_set_mark(struct nf_conn *ct, struct sw_flow_key *key,
334 u32 ct_mark, u32 mask)
336 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
339 new_mark = ct_mark | (ct->mark & ~(mask));
340 if (ct->mark != new_mark) {
342 if (nf_ct_is_confirmed(ct))
343 nf_conntrack_event_cache(IPCT_MARK, ct);
344 key->ct.mark = new_mark;
353 static struct nf_conn_labels *ovs_ct_get_conn_labels(struct nf_conn *ct)
355 struct nf_conn_labels *cl;
357 cl = nf_ct_labels_find(ct);
359 nf_ct_labels_ext_add(ct);
360 cl = nf_ct_labels_find(ct);
366 /* Initialize labels for a new, yet to be committed conntrack entry. Note that
367 * since the new connection is not yet confirmed, and thus no-one else has
368 * access to it's labels, we simply write them over.
370 static int ovs_ct_init_labels(struct nf_conn *ct, struct sw_flow_key *key,
371 const struct ovs_key_ct_labels *labels,
372 const struct ovs_key_ct_labels *mask)
374 struct nf_conn_labels *cl, *master_cl;
375 bool have_mask = labels_nonzero(mask);
377 /* Inherit master's labels to the related connection? */
378 master_cl = ct->master ? nf_ct_labels_find(ct->master) : NULL;
380 if (!master_cl && !have_mask)
381 return 0; /* Nothing to do. */
383 cl = ovs_ct_get_conn_labels(ct);
387 /* Inherit the master's labels, if any. */
392 u32 *dst = (u32 *)cl->bits;
395 for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
396 dst[i] = (dst[i] & ~mask->ct_labels_32[i]) |
397 (labels->ct_labels_32[i]
398 & mask->ct_labels_32[i]);
401 /* Labels are included in the IPCTNL_MSG_CT_NEW event only if the
402 * IPCT_LABEL bit is set in the event cache.
404 nf_conntrack_event_cache(IPCT_LABEL, ct);
406 memcpy(&key->ct.labels, cl->bits, OVS_CT_LABELS_LEN);
411 static int ovs_ct_set_labels(struct nf_conn *ct, struct sw_flow_key *key,
412 const struct ovs_key_ct_labels *labels,
413 const struct ovs_key_ct_labels *mask)
415 struct nf_conn_labels *cl;
418 cl = ovs_ct_get_conn_labels(ct);
422 err = nf_connlabels_replace(ct, labels->ct_labels_32,
424 OVS_CT_LABELS_LEN_32);
428 memcpy(&key->ct.labels, cl->bits, OVS_CT_LABELS_LEN);
433 /* 'skb' should already be pulled to nh_ofs. */
434 static int ovs_ct_helper(struct sk_buff *skb, u16 proto)
436 const struct nf_conntrack_helper *helper;
437 const struct nf_conn_help *help;
438 enum ip_conntrack_info ctinfo;
439 unsigned int protoff;
443 ct = nf_ct_get(skb, &ctinfo);
444 if (!ct || ctinfo == IP_CT_RELATED_REPLY)
447 help = nfct_help(ct);
451 helper = rcu_dereference(help->helper);
457 protoff = ip_hdrlen(skb);
460 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
464 ofs = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr,
466 if (ofs < 0 || (frag_off & htons(~0x7)) != 0) {
467 pr_debug("proto header not found\n");
474 WARN_ONCE(1, "helper invoked on non-IP family!");
478 err = helper->help(skb, protoff, ct, ctinfo);
479 if (err != NF_ACCEPT)
482 /* Adjust seqs after helper. This is needed due to some helpers (e.g.,
483 * FTP with NAT) adusting the TCP payload size when mangling IP
484 * addresses and/or port numbers in the text-based control connection.
486 if (test_bit(IPS_SEQ_ADJUST_BIT, &ct->status) &&
487 !nf_ct_seq_adjust(skb, ct, ctinfo, protoff))
492 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
493 * value if 'skb' is freed.
495 static int handle_fragments(struct net *net, struct sw_flow_key *key,
496 u16 zone, struct sk_buff *skb)
498 struct ovs_skb_cb ovs_cb = *OVS_CB(skb);
501 if (key->eth.type == htons(ETH_P_IP)) {
502 enum ip_defrag_users user = IP_DEFRAG_CONNTRACK_IN + zone;
504 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
505 err = ip_defrag(net, skb, user);
509 ovs_cb.mru = IPCB(skb)->frag_max_size;
510 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
511 } else if (key->eth.type == htons(ETH_P_IPV6)) {
512 enum ip6_defrag_users user = IP6_DEFRAG_CONNTRACK_IN + zone;
514 memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm));
515 err = nf_ct_frag6_gather(net, skb, user);
517 if (err != -EINPROGRESS)
522 key->ip.proto = ipv6_hdr(skb)->nexthdr;
523 ovs_cb.mru = IP6CB(skb)->frag_max_size;
527 return -EPFNOSUPPORT;
530 /* The key extracted from the fragment that completed this datagram
531 * likely didn't have an L4 header, so regenerate it.
533 ovs_flow_key_update_l3l4(skb, key);
535 key->ip.frag = OVS_FRAG_TYPE_NONE;
538 *OVS_CB(skb) = ovs_cb;
543 static struct nf_conntrack_expect *
544 ovs_ct_expect_find(struct net *net, const struct nf_conntrack_zone *zone,
545 u16 proto, const struct sk_buff *skb)
547 struct nf_conntrack_tuple tuple;
548 struct nf_conntrack_expect *exp;
550 if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), proto, net, &tuple))
553 exp = __nf_ct_expect_find(net, zone, &tuple);
555 struct nf_conntrack_tuple_hash *h;
557 /* Delete existing conntrack entry, if it clashes with the
558 * expectation. This can happen since conntrack ALGs do not
559 * check for clashes between (new) expectations and existing
560 * conntrack entries. nf_conntrack_in() will check the
561 * expectations only if a conntrack entry can not be found,
562 * which can lead to OVS finding the expectation (here) in the
563 * init direction, but which will not be removed by the
564 * nf_conntrack_in() call, if a matching conntrack entry is
565 * found instead. In this case all init direction packets
566 * would be reported as new related packets, while reply
567 * direction packets would be reported as un-related
568 * established packets.
570 h = nf_conntrack_find_get(net, zone, &tuple);
572 struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
574 nf_ct_delete(ct, 0, 0);
575 nf_conntrack_put(&ct->ct_general);
582 /* This replicates logic from nf_conntrack_core.c that is not exported. */
583 static enum ip_conntrack_info
584 ovs_ct_get_info(const struct nf_conntrack_tuple_hash *h)
586 const struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
588 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY)
589 return IP_CT_ESTABLISHED_REPLY;
590 /* Once we've had two way comms, always ESTABLISHED. */
591 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status))
592 return IP_CT_ESTABLISHED;
593 if (test_bit(IPS_EXPECTED_BIT, &ct->status))
594 return IP_CT_RELATED;
598 /* Find an existing connection which this packet belongs to without
599 * re-attributing statistics or modifying the connection state. This allows an
600 * skb->_nfct lost due to an upcall to be recovered during actions execution.
602 * Must be called with rcu_read_lock.
604 * On success, populates skb->_nfct and returns the connection. Returns NULL
605 * if there is no existing entry.
607 static struct nf_conn *
608 ovs_ct_find_existing(struct net *net, const struct nf_conntrack_zone *zone,
609 u8 l3num, struct sk_buff *skb, bool natted)
611 struct nf_conntrack_tuple tuple;
612 struct nf_conntrack_tuple_hash *h;
615 if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), l3num,
617 pr_debug("ovs_ct_find_existing: Can't get tuple\n");
621 /* Must invert the tuple if skb has been transformed by NAT. */
623 struct nf_conntrack_tuple inverse;
625 if (!nf_ct_invert_tuple(&inverse, &tuple)) {
626 pr_debug("ovs_ct_find_existing: Inversion failed!\n");
632 /* look for tuple match */
633 h = nf_conntrack_find_get(net, zone, &tuple);
635 return NULL; /* Not found. */
637 ct = nf_ct_tuplehash_to_ctrack(h);
639 /* Inverted packet tuple matches the reverse direction conntrack tuple,
640 * select the other tuplehash to get the right 'ctinfo' bits for this
644 h = &ct->tuplehash[!h->tuple.dst.dir];
646 nf_ct_set(skb, ct, ovs_ct_get_info(h));
651 struct nf_conn *ovs_ct_executed(struct net *net,
652 const struct sw_flow_key *key,
653 const struct ovs_conntrack_info *info,
657 struct nf_conn *ct = NULL;
659 /* If no ct, check if we have evidence that an existing conntrack entry
660 * might be found for this skb. This happens when we lose a skb->_nfct
661 * due to an upcall, or if the direction is being forced. If the
662 * connection was not confirmed, it is not cached and needs to be run
663 * through conntrack again.
665 *ct_executed = (key->ct_state & OVS_CS_F_TRACKED) &&
666 !(key->ct_state & OVS_CS_F_INVALID) &&
667 (key->ct_zone == info->zone.id);
669 if (*ct_executed || (!key->ct_state && info->force)) {
670 ct = ovs_ct_find_existing(net, &info->zone, info->family, skb,
678 /* Determine whether skb->_nfct is equal to the result of conntrack lookup. */
679 static bool skb_nfct_cached(struct net *net,
680 const struct sw_flow_key *key,
681 const struct ovs_conntrack_info *info,
684 enum ip_conntrack_info ctinfo;
686 bool ct_executed = true;
688 ct = nf_ct_get(skb, &ctinfo);
690 ct = ovs_ct_executed(net, key, info, skb, &ct_executed);
693 nf_ct_get(skb, &ctinfo);
697 if (!net_eq(net, read_pnet(&ct->ct_net)))
699 if (!nf_ct_zone_equal_any(info->ct, nf_ct_zone(ct)))
702 struct nf_conn_help *help;
704 help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER);
705 if (help && rcu_access_pointer(help->helper) != info->helper)
708 if (info->nf_ct_timeout) {
709 struct nf_conn_timeout *timeout_ext;
711 timeout_ext = nf_ct_timeout_find(ct);
712 if (!timeout_ext || info->nf_ct_timeout !=
713 rcu_dereference(timeout_ext->timeout))
716 /* Force conntrack entry direction to the current packet? */
717 if (info->force && CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) {
718 /* Delete the conntrack entry if confirmed, else just release
721 if (nf_ct_is_confirmed(ct))
722 nf_ct_delete(ct, 0, 0);
724 nf_conntrack_put(&ct->ct_general);
725 nf_ct_set(skb, NULL, 0);
732 #if IS_ENABLED(CONFIG_NF_NAT)
733 /* Modelled after nf_nat_ipv[46]_fn().
734 * range is only used for new, uninitialized NAT state.
735 * Returns either NF_ACCEPT or NF_DROP.
737 static int ovs_ct_nat_execute(struct sk_buff *skb, struct nf_conn *ct,
738 enum ip_conntrack_info ctinfo,
739 const struct nf_nat_range2 *range,
740 enum nf_nat_manip_type maniptype)
742 int hooknum, nh_off, err = NF_ACCEPT;
744 nh_off = skb_network_offset(skb);
745 skb_pull_rcsum(skb, nh_off);
747 /* See HOOK2MANIP(). */
748 if (maniptype == NF_NAT_MANIP_SRC)
749 hooknum = NF_INET_LOCAL_IN; /* Source NAT */
751 hooknum = NF_INET_LOCAL_OUT; /* Destination NAT */
755 case IP_CT_RELATED_REPLY:
756 if (IS_ENABLED(CONFIG_NF_NAT) &&
757 skb->protocol == htons(ETH_P_IP) &&
758 ip_hdr(skb)->protocol == IPPROTO_ICMP) {
759 if (!nf_nat_icmp_reply_translation(skb, ct, ctinfo,
763 } else if (IS_ENABLED(CONFIG_IPV6) &&
764 skb->protocol == htons(ETH_P_IPV6)) {
766 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
767 int hdrlen = ipv6_skip_exthdr(skb,
768 sizeof(struct ipv6hdr),
769 &nexthdr, &frag_off);
771 if (hdrlen >= 0 && nexthdr == IPPROTO_ICMPV6) {
772 if (!nf_nat_icmpv6_reply_translation(skb, ct,
780 /* Non-ICMP, fall thru to initialize if needed. */
783 /* Seen it before? This can happen for loopback, retrans,
786 if (!nf_nat_initialized(ct, maniptype)) {
787 /* Initialize according to the NAT action. */
788 err = (range && range->flags & NF_NAT_RANGE_MAP_IPS)
789 /* Action is set up to establish a new
792 ? nf_nat_setup_info(ct, range, maniptype)
793 : nf_nat_alloc_null_binding(ct, hooknum);
794 if (err != NF_ACCEPT)
799 case IP_CT_ESTABLISHED:
800 case IP_CT_ESTABLISHED_REPLY:
808 err = nf_nat_packet(ct, ctinfo, hooknum, skb);
810 skb_push(skb, nh_off);
811 skb_postpush_rcsum(skb, skb->data, nh_off);
816 static void ovs_nat_update_key(struct sw_flow_key *key,
817 const struct sk_buff *skb,
818 enum nf_nat_manip_type maniptype)
820 if (maniptype == NF_NAT_MANIP_SRC) {
823 key->ct_state |= OVS_CS_F_SRC_NAT;
824 if (key->eth.type == htons(ETH_P_IP))
825 key->ipv4.addr.src = ip_hdr(skb)->saddr;
826 else if (key->eth.type == htons(ETH_P_IPV6))
827 memcpy(&key->ipv6.addr.src, &ipv6_hdr(skb)->saddr,
828 sizeof(key->ipv6.addr.src));
832 if (key->ip.proto == IPPROTO_UDP)
833 src = udp_hdr(skb)->source;
834 else if (key->ip.proto == IPPROTO_TCP)
835 src = tcp_hdr(skb)->source;
836 else if (key->ip.proto == IPPROTO_SCTP)
837 src = sctp_hdr(skb)->source;
845 key->ct_state |= OVS_CS_F_DST_NAT;
846 if (key->eth.type == htons(ETH_P_IP))
847 key->ipv4.addr.dst = ip_hdr(skb)->daddr;
848 else if (key->eth.type == htons(ETH_P_IPV6))
849 memcpy(&key->ipv6.addr.dst, &ipv6_hdr(skb)->daddr,
850 sizeof(key->ipv6.addr.dst));
854 if (key->ip.proto == IPPROTO_UDP)
855 dst = udp_hdr(skb)->dest;
856 else if (key->ip.proto == IPPROTO_TCP)
857 dst = tcp_hdr(skb)->dest;
858 else if (key->ip.proto == IPPROTO_SCTP)
859 dst = sctp_hdr(skb)->dest;
867 /* Returns NF_DROP if the packet should be dropped, NF_ACCEPT otherwise. */
868 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
869 const struct ovs_conntrack_info *info,
870 struct sk_buff *skb, struct nf_conn *ct,
871 enum ip_conntrack_info ctinfo)
873 enum nf_nat_manip_type maniptype;
876 /* Add NAT extension if not confirmed yet. */
877 if (!nf_ct_is_confirmed(ct) && !nf_ct_nat_ext_add(ct))
878 return NF_ACCEPT; /* Can't NAT. */
880 /* Determine NAT type.
881 * Check if the NAT type can be deduced from the tracked connection.
882 * Make sure new expected connections (IP_CT_RELATED) are NATted only
885 if (info->nat & OVS_CT_NAT && ctinfo != IP_CT_NEW &&
886 ct->status & IPS_NAT_MASK &&
887 (ctinfo != IP_CT_RELATED || info->commit)) {
888 /* NAT an established or related connection like before. */
889 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY)
890 /* This is the REPLY direction for a connection
891 * for which NAT was applied in the forward
892 * direction. Do the reverse NAT.
894 maniptype = ct->status & IPS_SRC_NAT
895 ? NF_NAT_MANIP_DST : NF_NAT_MANIP_SRC;
897 maniptype = ct->status & IPS_SRC_NAT
898 ? NF_NAT_MANIP_SRC : NF_NAT_MANIP_DST;
899 } else if (info->nat & OVS_CT_SRC_NAT) {
900 maniptype = NF_NAT_MANIP_SRC;
901 } else if (info->nat & OVS_CT_DST_NAT) {
902 maniptype = NF_NAT_MANIP_DST;
904 return NF_ACCEPT; /* Connection is not NATed. */
906 err = ovs_ct_nat_execute(skb, ct, ctinfo, &info->range, maniptype);
908 if (err == NF_ACCEPT && ct->status & IPS_DST_NAT) {
909 if (ct->status & IPS_SRC_NAT) {
910 if (maniptype == NF_NAT_MANIP_SRC)
911 maniptype = NF_NAT_MANIP_DST;
913 maniptype = NF_NAT_MANIP_SRC;
915 err = ovs_ct_nat_execute(skb, ct, ctinfo, &info->range,
917 } else if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) {
918 err = ovs_ct_nat_execute(skb, ct, ctinfo, NULL,
923 /* Mark NAT done if successful and update the flow key. */
924 if (err == NF_ACCEPT)
925 ovs_nat_update_key(key, skb, maniptype);
929 #else /* !CONFIG_NF_NAT */
930 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
931 const struct ovs_conntrack_info *info,
932 struct sk_buff *skb, struct nf_conn *ct,
933 enum ip_conntrack_info ctinfo)
939 /* Pass 'skb' through conntrack in 'net', using zone configured in 'info', if
940 * not done already. Update key with new CT state after passing the packet
942 * Note that if the packet is deemed invalid by conntrack, skb->_nfct will be
943 * set to NULL and 0 will be returned.
945 static int __ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
946 const struct ovs_conntrack_info *info,
949 /* If we are recirculating packets to match on conntrack fields and
950 * committing with a separate conntrack action, then we don't need to
951 * actually run the packet through conntrack twice unless it's for a
954 bool cached = skb_nfct_cached(net, key, info, skb);
955 enum ip_conntrack_info ctinfo;
959 struct nf_hook_state state = {
960 .hook = NF_INET_PRE_ROUTING,
964 struct nf_conn *tmpl = info->ct;
967 /* Associate skb with specified zone. */
970 nf_conntrack_put(skb_nfct(skb));
971 nf_conntrack_get(&tmpl->ct_general);
972 nf_ct_set(skb, tmpl, IP_CT_NEW);
975 err = nf_conntrack_in(skb, &state);
976 if (err != NF_ACCEPT)
979 /* Clear CT state NAT flags to mark that we have not yet done
980 * NAT after the nf_conntrack_in() call. We can actually clear
981 * the whole state, as it will be re-initialized below.
985 /* Update the key, but keep the NAT flags. */
986 ovs_ct_update_key(skb, info, key, true, true);
989 ct = nf_ct_get(skb, &ctinfo);
991 bool add_helper = false;
993 /* Packets starting a new connection must be NATted before the
994 * helper, so that the helper knows about the NAT. We enforce
995 * this by delaying both NAT and helper calls for unconfirmed
996 * connections until the committing CT action. For later
997 * packets NAT and Helper may be called in either order.
999 * NAT will be done only if the CT action has NAT, and only
1000 * once per packet (per zone), as guarded by the NAT bits in
1001 * the key->ct_state.
1003 if (info->nat && !(key->ct_state & OVS_CS_F_NAT_MASK) &&
1004 (nf_ct_is_confirmed(ct) || info->commit) &&
1005 ovs_ct_nat(net, key, info, skb, ct, ctinfo) != NF_ACCEPT) {
1009 /* Userspace may decide to perform a ct lookup without a helper
1010 * specified followed by a (recirculate and) commit with one,
1011 * or attach a helper in a later commit. Therefore, for
1012 * connections which we will commit, we may need to attach
1015 if (info->commit && info->helper && !nfct_help(ct)) {
1016 int err = __nf_ct_try_assign_helper(ct, info->ct,
1022 /* helper installed, add seqadj if NAT is required */
1023 if (info->nat && !nfct_seqadj(ct)) {
1024 if (!nfct_seqadj_ext_add(ct))
1029 /* Call the helper only if:
1030 * - nf_conntrack_in() was executed above ("!cached") or a
1031 * helper was just attached ("add_helper") for a confirmed
1033 * - When committing an unconfirmed connection.
1035 if ((nf_ct_is_confirmed(ct) ? !cached || add_helper :
1037 ovs_ct_helper(skb, info->family) != NF_ACCEPT) {
1045 /* Lookup connection and read fields into key. */
1046 static int ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
1047 const struct ovs_conntrack_info *info,
1048 struct sk_buff *skb)
1050 struct nf_conntrack_expect *exp;
1052 /* If we pass an expected packet through nf_conntrack_in() the
1053 * expectation is typically removed, but the packet could still be
1054 * lost in upcall processing. To prevent this from happening we
1055 * perform an explicit expectation lookup. Expected connections are
1056 * always new, and will be passed through conntrack only when they are
1057 * committed, as it is OK to remove the expectation at that time.
1059 exp = ovs_ct_expect_find(net, &info->zone, info->family, skb);
1063 /* NOTE: New connections are NATted and Helped only when
1064 * committed, so we are not calling into NAT here.
1066 state = OVS_CS_F_TRACKED | OVS_CS_F_NEW | OVS_CS_F_RELATED;
1067 __ovs_ct_update_key(key, state, &info->zone, exp->master);
1072 err = __ovs_ct_lookup(net, key, info, skb);
1076 ct = (struct nf_conn *)skb_nfct(skb);
1078 nf_ct_deliver_cached_events(ct);
1084 static bool labels_nonzero(const struct ovs_key_ct_labels *labels)
1088 for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
1089 if (labels->ct_labels_32[i])
1095 #if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1096 static struct hlist_head *ct_limit_hash_bucket(
1097 const struct ovs_ct_limit_info *info, u16 zone)
1099 return &info->limits[zone & (CT_LIMIT_HASH_BUCKETS - 1)];
1102 /* Call with ovs_mutex */
1103 static void ct_limit_set(const struct ovs_ct_limit_info *info,
1104 struct ovs_ct_limit *new_ct_limit)
1106 struct ovs_ct_limit *ct_limit;
1107 struct hlist_head *head;
1109 head = ct_limit_hash_bucket(info, new_ct_limit->zone);
1110 hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
1111 if (ct_limit->zone == new_ct_limit->zone) {
1112 hlist_replace_rcu(&ct_limit->hlist_node,
1113 &new_ct_limit->hlist_node);
1114 kfree_rcu(ct_limit, rcu);
1119 hlist_add_head_rcu(&new_ct_limit->hlist_node, head);
1122 /* Call with ovs_mutex */
1123 static void ct_limit_del(const struct ovs_ct_limit_info *info, u16 zone)
1125 struct ovs_ct_limit *ct_limit;
1126 struct hlist_head *head;
1127 struct hlist_node *n;
1129 head = ct_limit_hash_bucket(info, zone);
1130 hlist_for_each_entry_safe(ct_limit, n, head, hlist_node) {
1131 if (ct_limit->zone == zone) {
1132 hlist_del_rcu(&ct_limit->hlist_node);
1133 kfree_rcu(ct_limit, rcu);
1139 /* Call with RCU read lock */
1140 static u32 ct_limit_get(const struct ovs_ct_limit_info *info, u16 zone)
1142 struct ovs_ct_limit *ct_limit;
1143 struct hlist_head *head;
1145 head = ct_limit_hash_bucket(info, zone);
1146 hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
1147 if (ct_limit->zone == zone)
1148 return ct_limit->limit;
1151 return info->default_limit;
1154 static int ovs_ct_check_limit(struct net *net,
1155 const struct ovs_conntrack_info *info,
1156 const struct nf_conntrack_tuple *tuple)
1158 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1159 const struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
1160 u32 per_zone_limit, connections;
1163 conncount_key = info->zone.id;
1165 per_zone_limit = ct_limit_get(ct_limit_info, info->zone.id);
1166 if (per_zone_limit == OVS_CT_LIMIT_UNLIMITED)
1169 connections = nf_conncount_count(net, ct_limit_info->data,
1170 &conncount_key, tuple, &info->zone);
1171 if (connections > per_zone_limit)
1178 /* Lookup connection and confirm if unconfirmed. */
1179 static int ovs_ct_commit(struct net *net, struct sw_flow_key *key,
1180 const struct ovs_conntrack_info *info,
1181 struct sk_buff *skb)
1183 enum ip_conntrack_info ctinfo;
1187 err = __ovs_ct_lookup(net, key, info, skb);
1191 /* The connection could be invalid, in which case this is a no-op.*/
1192 ct = nf_ct_get(skb, &ctinfo);
1196 #if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1197 if (static_branch_unlikely(&ovs_ct_limit_enabled)) {
1198 if (!nf_ct_is_confirmed(ct)) {
1199 err = ovs_ct_check_limit(net, info,
1200 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
1202 net_warn_ratelimited("openvswitch: zone: %u "
1203 "exceeds conntrack limit\n",
1211 /* Set the conntrack event mask if given. NEW and DELETE events have
1212 * their own groups, but the NFNLGRP_CONNTRACK_UPDATE group listener
1213 * typically would receive many kinds of updates. Setting the event
1214 * mask allows those events to be filtered. The set event mask will
1215 * remain in effect for the lifetime of the connection unless changed
1216 * by a further CT action with both the commit flag and the eventmask
1218 if (info->have_eventmask) {
1219 struct nf_conntrack_ecache *cache = nf_ct_ecache_find(ct);
1222 cache->ctmask = info->eventmask;
1225 /* Apply changes before confirming the connection so that the initial
1226 * conntrack NEW netlink event carries the values given in the CT
1229 if (info->mark.mask) {
1230 err = ovs_ct_set_mark(ct, key, info->mark.value,
1235 if (!nf_ct_is_confirmed(ct)) {
1236 err = ovs_ct_init_labels(ct, key, &info->labels.value,
1237 &info->labels.mask);
1240 } else if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1241 labels_nonzero(&info->labels.mask)) {
1242 err = ovs_ct_set_labels(ct, key, &info->labels.value,
1243 &info->labels.mask);
1247 /* This will take care of sending queued events even if the connection
1248 * is already confirmed.
1250 if (nf_conntrack_confirm(skb) != NF_ACCEPT)
1256 /* Trim the skb to the length specified by the IP/IPv6 header,
1257 * removing any trailing lower-layer padding. This prepares the skb
1258 * for higher-layer processing that assumes skb->len excludes padding
1259 * (such as nf_ip_checksum). The caller needs to pull the skb to the
1260 * network header, and ensure ip_hdr/ipv6_hdr points to valid data.
1262 static int ovs_skb_network_trim(struct sk_buff *skb)
1267 switch (skb->protocol) {
1268 case htons(ETH_P_IP):
1269 len = ntohs(ip_hdr(skb)->tot_len);
1271 case htons(ETH_P_IPV6):
1272 len = sizeof(struct ipv6hdr)
1273 + ntohs(ipv6_hdr(skb)->payload_len);
1279 err = pskb_trim_rcsum(skb, len);
1286 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
1287 * value if 'skb' is freed.
1289 int ovs_ct_execute(struct net *net, struct sk_buff *skb,
1290 struct sw_flow_key *key,
1291 const struct ovs_conntrack_info *info)
1296 /* The conntrack module expects to be working at L3. */
1297 nh_ofs = skb_network_offset(skb);
1298 skb_pull_rcsum(skb, nh_ofs);
1300 err = ovs_skb_network_trim(skb);
1304 if (key->ip.frag != OVS_FRAG_TYPE_NONE) {
1305 err = handle_fragments(net, key, info->zone.id, skb);
1311 err = ovs_ct_commit(net, key, info, skb);
1313 err = ovs_ct_lookup(net, key, info, skb);
1315 skb_push(skb, nh_ofs);
1316 skb_postpush_rcsum(skb, skb->data, nh_ofs);
1322 int ovs_ct_clear(struct sk_buff *skb, struct sw_flow_key *key)
1324 if (skb_nfct(skb)) {
1325 nf_conntrack_put(skb_nfct(skb));
1326 nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
1327 ovs_ct_fill_key(skb, key);
1333 static int ovs_ct_add_helper(struct ovs_conntrack_info *info, const char *name,
1334 const struct sw_flow_key *key, bool log)
1336 struct nf_conntrack_helper *helper;
1337 struct nf_conn_help *help;
1340 helper = nf_conntrack_helper_try_module_get(name, info->family,
1343 OVS_NLERR(log, "Unknown helper \"%s\"", name);
1347 help = nf_ct_helper_ext_add(info->ct, GFP_KERNEL);
1349 nf_conntrack_helper_put(helper);
1353 #if IS_ENABLED(CONFIG_NF_NAT)
1355 ret = nf_nat_helper_try_module_get(name, info->family,
1358 nf_conntrack_helper_put(helper);
1359 OVS_NLERR(log, "Failed to load \"%s\" NAT helper, error: %d",
1365 rcu_assign_pointer(help->helper, helper);
1366 info->helper = helper;
1370 #if IS_ENABLED(CONFIG_NF_NAT)
1371 static int parse_nat(const struct nlattr *attr,
1372 struct ovs_conntrack_info *info, bool log)
1376 bool have_ip_max = false;
1377 bool have_proto_max = false;
1378 bool ip_vers = (info->family == NFPROTO_IPV6);
1380 nla_for_each_nested(a, attr, rem) {
1381 static const int ovs_nat_attr_lens[OVS_NAT_ATTR_MAX + 1][2] = {
1382 [OVS_NAT_ATTR_SRC] = {0, 0},
1383 [OVS_NAT_ATTR_DST] = {0, 0},
1384 [OVS_NAT_ATTR_IP_MIN] = {sizeof(struct in_addr),
1385 sizeof(struct in6_addr)},
1386 [OVS_NAT_ATTR_IP_MAX] = {sizeof(struct in_addr),
1387 sizeof(struct in6_addr)},
1388 [OVS_NAT_ATTR_PROTO_MIN] = {sizeof(u16), sizeof(u16)},
1389 [OVS_NAT_ATTR_PROTO_MAX] = {sizeof(u16), sizeof(u16)},
1390 [OVS_NAT_ATTR_PERSISTENT] = {0, 0},
1391 [OVS_NAT_ATTR_PROTO_HASH] = {0, 0},
1392 [OVS_NAT_ATTR_PROTO_RANDOM] = {0, 0},
1394 int type = nla_type(a);
1396 if (type > OVS_NAT_ATTR_MAX) {
1397 OVS_NLERR(log, "Unknown NAT attribute (type=%d, max=%d)",
1398 type, OVS_NAT_ATTR_MAX);
1402 if (nla_len(a) != ovs_nat_attr_lens[type][ip_vers]) {
1403 OVS_NLERR(log, "NAT attribute type %d has unexpected length (%d != %d)",
1405 ovs_nat_attr_lens[type][ip_vers]);
1410 case OVS_NAT_ATTR_SRC:
1411 case OVS_NAT_ATTR_DST:
1413 OVS_NLERR(log, "Only one type of NAT may be specified");
1416 info->nat |= OVS_CT_NAT;
1417 info->nat |= ((type == OVS_NAT_ATTR_SRC)
1418 ? OVS_CT_SRC_NAT : OVS_CT_DST_NAT);
1421 case OVS_NAT_ATTR_IP_MIN:
1422 nla_memcpy(&info->range.min_addr, a,
1423 sizeof(info->range.min_addr));
1424 info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1427 case OVS_NAT_ATTR_IP_MAX:
1429 nla_memcpy(&info->range.max_addr, a,
1430 sizeof(info->range.max_addr));
1431 info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1434 case OVS_NAT_ATTR_PROTO_MIN:
1435 info->range.min_proto.all = htons(nla_get_u16(a));
1436 info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1439 case OVS_NAT_ATTR_PROTO_MAX:
1440 have_proto_max = true;
1441 info->range.max_proto.all = htons(nla_get_u16(a));
1442 info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1445 case OVS_NAT_ATTR_PERSISTENT:
1446 info->range.flags |= NF_NAT_RANGE_PERSISTENT;
1449 case OVS_NAT_ATTR_PROTO_HASH:
1450 info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM;
1453 case OVS_NAT_ATTR_PROTO_RANDOM:
1454 info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM_FULLY;
1458 OVS_NLERR(log, "Unknown nat attribute (%d)", type);
1464 OVS_NLERR(log, "NAT attribute has %d unknown bytes", rem);
1468 /* Do not allow flags if no type is given. */
1469 if (info->range.flags) {
1471 "NAT flags may be given only when NAT range (SRC or DST) is also specified."
1475 info->nat = OVS_CT_NAT; /* NAT existing connections. */
1476 } else if (!info->commit) {
1478 "NAT attributes may be specified only when CT COMMIT flag is also specified."
1482 /* Allow missing IP_MAX. */
1483 if (info->range.flags & NF_NAT_RANGE_MAP_IPS && !have_ip_max) {
1484 memcpy(&info->range.max_addr, &info->range.min_addr,
1485 sizeof(info->range.max_addr));
1487 /* Allow missing PROTO_MAX. */
1488 if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1490 info->range.max_proto.all = info->range.min_proto.all;
1496 static const struct ovs_ct_len_tbl ovs_ct_attr_lens[OVS_CT_ATTR_MAX + 1] = {
1497 [OVS_CT_ATTR_COMMIT] = { .minlen = 0, .maxlen = 0 },
1498 [OVS_CT_ATTR_FORCE_COMMIT] = { .minlen = 0, .maxlen = 0 },
1499 [OVS_CT_ATTR_ZONE] = { .minlen = sizeof(u16),
1500 .maxlen = sizeof(u16) },
1501 [OVS_CT_ATTR_MARK] = { .minlen = sizeof(struct md_mark),
1502 .maxlen = sizeof(struct md_mark) },
1503 [OVS_CT_ATTR_LABELS] = { .minlen = sizeof(struct md_labels),
1504 .maxlen = sizeof(struct md_labels) },
1505 [OVS_CT_ATTR_HELPER] = { .minlen = 1,
1506 .maxlen = NF_CT_HELPER_NAME_LEN },
1507 #if IS_ENABLED(CONFIG_NF_NAT)
1508 /* NAT length is checked when parsing the nested attributes. */
1509 [OVS_CT_ATTR_NAT] = { .minlen = 0, .maxlen = INT_MAX },
1511 [OVS_CT_ATTR_EVENTMASK] = { .minlen = sizeof(u32),
1512 .maxlen = sizeof(u32) },
1513 [OVS_CT_ATTR_TIMEOUT] = { .minlen = 1,
1514 .maxlen = CTNL_TIMEOUT_NAME_MAX },
1517 static int parse_ct(const struct nlattr *attr, struct ovs_conntrack_info *info,
1518 const char **helper, bool log)
1523 nla_for_each_nested(a, attr, rem) {
1524 int type = nla_type(a);
1528 if (type > OVS_CT_ATTR_MAX) {
1530 "Unknown conntrack attr (type=%d, max=%d)",
1531 type, OVS_CT_ATTR_MAX);
1535 maxlen = ovs_ct_attr_lens[type].maxlen;
1536 minlen = ovs_ct_attr_lens[type].minlen;
1537 if (nla_len(a) < minlen || nla_len(a) > maxlen) {
1539 "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
1540 type, nla_len(a), maxlen);
1545 case OVS_CT_ATTR_FORCE_COMMIT:
1548 case OVS_CT_ATTR_COMMIT:
1549 info->commit = true;
1551 #ifdef CONFIG_NF_CONNTRACK_ZONES
1552 case OVS_CT_ATTR_ZONE:
1553 info->zone.id = nla_get_u16(a);
1556 #ifdef CONFIG_NF_CONNTRACK_MARK
1557 case OVS_CT_ATTR_MARK: {
1558 struct md_mark *mark = nla_data(a);
1561 OVS_NLERR(log, "ct_mark mask cannot be 0");
1568 #ifdef CONFIG_NF_CONNTRACK_LABELS
1569 case OVS_CT_ATTR_LABELS: {
1570 struct md_labels *labels = nla_data(a);
1572 if (!labels_nonzero(&labels->mask)) {
1573 OVS_NLERR(log, "ct_labels mask cannot be 0");
1576 info->labels = *labels;
1580 case OVS_CT_ATTR_HELPER:
1581 *helper = nla_data(a);
1582 if (!memchr(*helper, '\0', nla_len(a))) {
1583 OVS_NLERR(log, "Invalid conntrack helper");
1587 #if IS_ENABLED(CONFIG_NF_NAT)
1588 case OVS_CT_ATTR_NAT: {
1589 int err = parse_nat(a, info, log);
1596 case OVS_CT_ATTR_EVENTMASK:
1597 info->have_eventmask = true;
1598 info->eventmask = nla_get_u32(a);
1600 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT
1601 case OVS_CT_ATTR_TIMEOUT:
1602 memcpy(info->timeout, nla_data(a), nla_len(a));
1603 if (!memchr(info->timeout, '\0', nla_len(a))) {
1604 OVS_NLERR(log, "Invalid conntrack timeout");
1611 OVS_NLERR(log, "Unknown conntrack attr (%d)",
1617 #ifdef CONFIG_NF_CONNTRACK_MARK
1618 if (!info->commit && info->mark.mask) {
1620 "Setting conntrack mark requires 'commit' flag.");
1624 #ifdef CONFIG_NF_CONNTRACK_LABELS
1625 if (!info->commit && labels_nonzero(&info->labels.mask)) {
1627 "Setting conntrack labels requires 'commit' flag.");
1632 OVS_NLERR(log, "Conntrack attr has %d unknown bytes", rem);
1639 bool ovs_ct_verify(struct net *net, enum ovs_key_attr attr)
1641 if (attr == OVS_KEY_ATTR_CT_STATE)
1643 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1644 attr == OVS_KEY_ATTR_CT_ZONE)
1646 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1647 attr == OVS_KEY_ATTR_CT_MARK)
1649 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1650 attr == OVS_KEY_ATTR_CT_LABELS) {
1651 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1653 return ovs_net->xt_label;
1659 int ovs_ct_copy_action(struct net *net, const struct nlattr *attr,
1660 const struct sw_flow_key *key,
1661 struct sw_flow_actions **sfa, bool log)
1663 struct ovs_conntrack_info ct_info;
1664 const char *helper = NULL;
1668 family = key_to_nfproto(key);
1669 if (family == NFPROTO_UNSPEC) {
1670 OVS_NLERR(log, "ct family unspecified");
1674 memset(&ct_info, 0, sizeof(ct_info));
1675 ct_info.family = family;
1677 nf_ct_zone_init(&ct_info.zone, NF_CT_DEFAULT_ZONE_ID,
1678 NF_CT_DEFAULT_ZONE_DIR, 0);
1680 err = parse_ct(attr, &ct_info, &helper, log);
1684 /* Set up template for tracking connections in specific zones. */
1685 ct_info.ct = nf_ct_tmpl_alloc(net, &ct_info.zone, GFP_KERNEL);
1687 OVS_NLERR(log, "Failed to allocate conntrack template");
1691 if (ct_info.timeout[0]) {
1692 if (nf_ct_set_timeout(net, ct_info.ct, family, key->ip.proto,
1694 pr_info_ratelimited("Failed to associated timeout "
1695 "policy `%s'\n", ct_info.timeout);
1697 ct_info.nf_ct_timeout = rcu_dereference(
1698 nf_ct_timeout_find(ct_info.ct)->timeout);
1703 err = ovs_ct_add_helper(&ct_info, helper, key, log);
1708 err = ovs_nla_add_action(sfa, OVS_ACTION_ATTR_CT, &ct_info,
1709 sizeof(ct_info), log);
1713 __set_bit(IPS_CONFIRMED_BIT, &ct_info.ct->status);
1714 nf_conntrack_get(&ct_info.ct->ct_general);
1717 __ovs_ct_free_action(&ct_info);
1721 #if IS_ENABLED(CONFIG_NF_NAT)
1722 static bool ovs_ct_nat_to_attr(const struct ovs_conntrack_info *info,
1723 struct sk_buff *skb)
1725 struct nlattr *start;
1727 start = nla_nest_start_noflag(skb, OVS_CT_ATTR_NAT);
1731 if (info->nat & OVS_CT_SRC_NAT) {
1732 if (nla_put_flag(skb, OVS_NAT_ATTR_SRC))
1734 } else if (info->nat & OVS_CT_DST_NAT) {
1735 if (nla_put_flag(skb, OVS_NAT_ATTR_DST))
1741 if (info->range.flags & NF_NAT_RANGE_MAP_IPS) {
1742 if (IS_ENABLED(CONFIG_NF_NAT) &&
1743 info->family == NFPROTO_IPV4) {
1744 if (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MIN,
1745 info->range.min_addr.ip) ||
1746 (info->range.max_addr.ip
1747 != info->range.min_addr.ip &&
1748 (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MAX,
1749 info->range.max_addr.ip))))
1751 } else if (IS_ENABLED(CONFIG_IPV6) &&
1752 info->family == NFPROTO_IPV6) {
1753 if (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MIN,
1754 &info->range.min_addr.in6) ||
1755 (memcmp(&info->range.max_addr.in6,
1756 &info->range.min_addr.in6,
1757 sizeof(info->range.max_addr.in6)) &&
1758 (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MAX,
1759 &info->range.max_addr.in6))))
1765 if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1766 (nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MIN,
1767 ntohs(info->range.min_proto.all)) ||
1768 (info->range.max_proto.all != info->range.min_proto.all &&
1769 nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MAX,
1770 ntohs(info->range.max_proto.all)))))
1773 if (info->range.flags & NF_NAT_RANGE_PERSISTENT &&
1774 nla_put_flag(skb, OVS_NAT_ATTR_PERSISTENT))
1776 if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM &&
1777 nla_put_flag(skb, OVS_NAT_ATTR_PROTO_HASH))
1779 if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM_FULLY &&
1780 nla_put_flag(skb, OVS_NAT_ATTR_PROTO_RANDOM))
1783 nla_nest_end(skb, start);
1789 int ovs_ct_action_to_attr(const struct ovs_conntrack_info *ct_info,
1790 struct sk_buff *skb)
1792 struct nlattr *start;
1794 start = nla_nest_start_noflag(skb, OVS_ACTION_ATTR_CT);
1798 if (ct_info->commit && nla_put_flag(skb, ct_info->force
1799 ? OVS_CT_ATTR_FORCE_COMMIT
1800 : OVS_CT_ATTR_COMMIT))
1802 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1803 nla_put_u16(skb, OVS_CT_ATTR_ZONE, ct_info->zone.id))
1805 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && ct_info->mark.mask &&
1806 nla_put(skb, OVS_CT_ATTR_MARK, sizeof(ct_info->mark),
1809 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1810 labels_nonzero(&ct_info->labels.mask) &&
1811 nla_put(skb, OVS_CT_ATTR_LABELS, sizeof(ct_info->labels),
1814 if (ct_info->helper) {
1815 if (nla_put_string(skb, OVS_CT_ATTR_HELPER,
1816 ct_info->helper->name))
1819 if (ct_info->have_eventmask &&
1820 nla_put_u32(skb, OVS_CT_ATTR_EVENTMASK, ct_info->eventmask))
1822 if (ct_info->timeout[0]) {
1823 if (nla_put_string(skb, OVS_CT_ATTR_TIMEOUT, ct_info->timeout))
1827 #if IS_ENABLED(CONFIG_NF_NAT)
1828 if (ct_info->nat && !ovs_ct_nat_to_attr(ct_info, skb))
1831 nla_nest_end(skb, start);
1836 void ovs_ct_free_action(const struct nlattr *a)
1838 struct ovs_conntrack_info *ct_info = nla_data(a);
1840 __ovs_ct_free_action(ct_info);
1843 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info)
1845 if (ct_info->helper) {
1846 #if IS_ENABLED(CONFIG_NF_NAT)
1848 nf_nat_helper_put(ct_info->helper);
1850 nf_conntrack_helper_put(ct_info->helper);
1853 if (ct_info->timeout[0])
1854 nf_ct_destroy_timeout(ct_info->ct);
1855 nf_ct_tmpl_free(ct_info->ct);
1859 #if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1860 static int ovs_ct_limit_init(struct net *net, struct ovs_net *ovs_net)
1864 ovs_net->ct_limit_info = kmalloc(sizeof(*ovs_net->ct_limit_info),
1866 if (!ovs_net->ct_limit_info)
1869 ovs_net->ct_limit_info->default_limit = OVS_CT_LIMIT_DEFAULT;
1870 ovs_net->ct_limit_info->limits =
1871 kmalloc_array(CT_LIMIT_HASH_BUCKETS, sizeof(struct hlist_head),
1873 if (!ovs_net->ct_limit_info->limits) {
1874 kfree(ovs_net->ct_limit_info);
1878 for (i = 0; i < CT_LIMIT_HASH_BUCKETS; i++)
1879 INIT_HLIST_HEAD(&ovs_net->ct_limit_info->limits[i]);
1881 ovs_net->ct_limit_info->data =
1882 nf_conncount_init(net, NFPROTO_INET, sizeof(u32));
1884 if (IS_ERR(ovs_net->ct_limit_info->data)) {
1885 err = PTR_ERR(ovs_net->ct_limit_info->data);
1886 kfree(ovs_net->ct_limit_info->limits);
1887 kfree(ovs_net->ct_limit_info);
1888 pr_err("openvswitch: failed to init nf_conncount %d\n", err);
1894 static void ovs_ct_limit_exit(struct net *net, struct ovs_net *ovs_net)
1896 const struct ovs_ct_limit_info *info = ovs_net->ct_limit_info;
1899 nf_conncount_destroy(net, NFPROTO_INET, info->data);
1900 for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) {
1901 struct hlist_head *head = &info->limits[i];
1902 struct ovs_ct_limit *ct_limit;
1904 hlist_for_each_entry_rcu(ct_limit, head, hlist_node,
1905 lockdep_ovsl_is_held())
1906 kfree_rcu(ct_limit, rcu);
1908 kfree(ovs_net->ct_limit_info->limits);
1909 kfree(ovs_net->ct_limit_info);
1912 static struct sk_buff *
1913 ovs_ct_limit_cmd_reply_start(struct genl_info *info, u8 cmd,
1914 struct ovs_header **ovs_reply_header)
1916 struct ovs_header *ovs_header = info->userhdr;
1917 struct sk_buff *skb;
1919 skb = genlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1921 return ERR_PTR(-ENOMEM);
1923 *ovs_reply_header = genlmsg_put(skb, info->snd_portid,
1925 &dp_ct_limit_genl_family, 0, cmd);
1927 if (!*ovs_reply_header) {
1929 return ERR_PTR(-EMSGSIZE);
1931 (*ovs_reply_header)->dp_ifindex = ovs_header->dp_ifindex;
1936 static bool check_zone_id(int zone_id, u16 *pzone)
1938 if (zone_id >= 0 && zone_id <= 65535) {
1939 *pzone = (u16)zone_id;
1945 static int ovs_ct_limit_set_zone_limit(struct nlattr *nla_zone_limit,
1946 struct ovs_ct_limit_info *info)
1948 struct ovs_zone_limit *zone_limit;
1952 rem = NLA_ALIGN(nla_len(nla_zone_limit));
1953 zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1955 while (rem >= sizeof(*zone_limit)) {
1956 if (unlikely(zone_limit->zone_id ==
1957 OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1959 info->default_limit = zone_limit->limit;
1961 } else if (unlikely(!check_zone_id(
1962 zone_limit->zone_id, &zone))) {
1963 OVS_NLERR(true, "zone id is out of range");
1965 struct ovs_ct_limit *ct_limit;
1967 ct_limit = kmalloc(sizeof(*ct_limit), GFP_KERNEL);
1971 ct_limit->zone = zone;
1972 ct_limit->limit = zone_limit->limit;
1975 ct_limit_set(info, ct_limit);
1978 rem -= NLA_ALIGN(sizeof(*zone_limit));
1979 zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
1980 NLA_ALIGN(sizeof(*zone_limit)));
1984 OVS_NLERR(true, "set zone limit has %d unknown bytes", rem);
1989 static int ovs_ct_limit_del_zone_limit(struct nlattr *nla_zone_limit,
1990 struct ovs_ct_limit_info *info)
1992 struct ovs_zone_limit *zone_limit;
1996 rem = NLA_ALIGN(nla_len(nla_zone_limit));
1997 zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1999 while (rem >= sizeof(*zone_limit)) {
2000 if (unlikely(zone_limit->zone_id ==
2001 OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
2003 info->default_limit = OVS_CT_LIMIT_DEFAULT;
2005 } else if (unlikely(!check_zone_id(
2006 zone_limit->zone_id, &zone))) {
2007 OVS_NLERR(true, "zone id is out of range");
2010 ct_limit_del(info, zone);
2013 rem -= NLA_ALIGN(sizeof(*zone_limit));
2014 zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
2015 NLA_ALIGN(sizeof(*zone_limit)));
2019 OVS_NLERR(true, "del zone limit has %d unknown bytes", rem);
2024 static int ovs_ct_limit_get_default_limit(struct ovs_ct_limit_info *info,
2025 struct sk_buff *reply)
2027 struct ovs_zone_limit zone_limit;
2030 zone_limit.zone_id = OVS_ZONE_LIMIT_DEFAULT_ZONE;
2031 zone_limit.limit = info->default_limit;
2032 err = nla_put_nohdr(reply, sizeof(zone_limit), &zone_limit);
2039 static int __ovs_ct_limit_get_zone_limit(struct net *net,
2040 struct nf_conncount_data *data,
2041 u16 zone_id, u32 limit,
2042 struct sk_buff *reply)
2044 struct nf_conntrack_zone ct_zone;
2045 struct ovs_zone_limit zone_limit;
2046 u32 conncount_key = zone_id;
2048 zone_limit.zone_id = zone_id;
2049 zone_limit.limit = limit;
2050 nf_ct_zone_init(&ct_zone, zone_id, NF_CT_DEFAULT_ZONE_DIR, 0);
2052 zone_limit.count = nf_conncount_count(net, data, &conncount_key, NULL,
2054 return nla_put_nohdr(reply, sizeof(zone_limit), &zone_limit);
2057 static int ovs_ct_limit_get_zone_limit(struct net *net,
2058 struct nlattr *nla_zone_limit,
2059 struct ovs_ct_limit_info *info,
2060 struct sk_buff *reply)
2062 struct ovs_zone_limit *zone_limit;
2067 rem = NLA_ALIGN(nla_len(nla_zone_limit));
2068 zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
2070 while (rem >= sizeof(*zone_limit)) {
2071 if (unlikely(zone_limit->zone_id ==
2072 OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
2073 err = ovs_ct_limit_get_default_limit(info, reply);
2076 } else if (unlikely(!check_zone_id(zone_limit->zone_id,
2078 OVS_NLERR(true, "zone id is out of range");
2081 limit = ct_limit_get(info, zone);
2084 err = __ovs_ct_limit_get_zone_limit(
2085 net, info->data, zone, limit, reply);
2089 rem -= NLA_ALIGN(sizeof(*zone_limit));
2090 zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
2091 NLA_ALIGN(sizeof(*zone_limit)));
2095 OVS_NLERR(true, "get zone limit has %d unknown bytes", rem);
2100 static int ovs_ct_limit_get_all_zone_limit(struct net *net,
2101 struct ovs_ct_limit_info *info,
2102 struct sk_buff *reply)
2104 struct ovs_ct_limit *ct_limit;
2105 struct hlist_head *head;
2108 err = ovs_ct_limit_get_default_limit(info, reply);
2113 for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) {
2114 head = &info->limits[i];
2115 hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
2116 err = __ovs_ct_limit_get_zone_limit(net, info->data,
2117 ct_limit->zone, ct_limit->limit, reply);
2128 static int ovs_ct_limit_cmd_set(struct sk_buff *skb, struct genl_info *info)
2130 struct nlattr **a = info->attrs;
2131 struct sk_buff *reply;
2132 struct ovs_header *ovs_reply_header;
2133 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
2134 struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2137 reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_SET,
2140 return PTR_ERR(reply);
2142 if (!a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2147 err = ovs_ct_limit_set_zone_limit(a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT],
2152 static_branch_enable(&ovs_ct_limit_enabled);
2154 genlmsg_end(reply, ovs_reply_header);
2155 return genlmsg_reply(reply, info);
2162 static int ovs_ct_limit_cmd_del(struct sk_buff *skb, struct genl_info *info)
2164 struct nlattr **a = info->attrs;
2165 struct sk_buff *reply;
2166 struct ovs_header *ovs_reply_header;
2167 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
2168 struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2171 reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_DEL,
2174 return PTR_ERR(reply);
2176 if (!a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2181 err = ovs_ct_limit_del_zone_limit(a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT],
2186 genlmsg_end(reply, ovs_reply_header);
2187 return genlmsg_reply(reply, info);
2194 static int ovs_ct_limit_cmd_get(struct sk_buff *skb, struct genl_info *info)
2196 struct nlattr **a = info->attrs;
2197 struct nlattr *nla_reply;
2198 struct sk_buff *reply;
2199 struct ovs_header *ovs_reply_header;
2200 struct net *net = sock_net(skb->sk);
2201 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2202 struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2205 reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_GET,
2208 return PTR_ERR(reply);
2210 nla_reply = nla_nest_start_noflag(reply, OVS_CT_LIMIT_ATTR_ZONE_LIMIT);
2216 if (a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2217 err = ovs_ct_limit_get_zone_limit(
2218 net, a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT], ct_limit_info,
2223 err = ovs_ct_limit_get_all_zone_limit(net, ct_limit_info,
2229 nla_nest_end(reply, nla_reply);
2230 genlmsg_end(reply, ovs_reply_header);
2231 return genlmsg_reply(reply, info);
2238 static struct genl_ops ct_limit_genl_ops[] = {
2239 { .cmd = OVS_CT_LIMIT_CMD_SET,
2240 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2241 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN
2243 .doit = ovs_ct_limit_cmd_set,
2245 { .cmd = OVS_CT_LIMIT_CMD_DEL,
2246 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2247 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN
2249 .doit = ovs_ct_limit_cmd_del,
2251 { .cmd = OVS_CT_LIMIT_CMD_GET,
2252 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2253 .flags = 0, /* OK for unprivileged users. */
2254 .doit = ovs_ct_limit_cmd_get,
2258 static const struct genl_multicast_group ovs_ct_limit_multicast_group = {
2259 .name = OVS_CT_LIMIT_MCGROUP,
2262 struct genl_family dp_ct_limit_genl_family __ro_after_init = {
2263 .hdrsize = sizeof(struct ovs_header),
2264 .name = OVS_CT_LIMIT_FAMILY,
2265 .version = OVS_CT_LIMIT_VERSION,
2266 .maxattr = OVS_CT_LIMIT_ATTR_MAX,
2267 .policy = ct_limit_policy,
2269 .parallel_ops = true,
2270 .ops = ct_limit_genl_ops,
2271 .n_ops = ARRAY_SIZE(ct_limit_genl_ops),
2272 .mcgrps = &ovs_ct_limit_multicast_group,
2274 .module = THIS_MODULE,
2278 int ovs_ct_init(struct net *net)
2280 unsigned int n_bits = sizeof(struct ovs_key_ct_labels) * BITS_PER_BYTE;
2281 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2283 if (nf_connlabels_get(net, n_bits - 1)) {
2284 ovs_net->xt_label = false;
2285 OVS_NLERR(true, "Failed to set connlabel length");
2287 ovs_net->xt_label = true;
2290 #if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2291 return ovs_ct_limit_init(net, ovs_net);
2297 void ovs_ct_exit(struct net *net)
2299 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2301 #if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2302 ovs_ct_limit_exit(net, ovs_net);
2305 if (ovs_net->xt_label)
2306 nf_connlabels_put(net);