2 * Copyright (c) 2015 Nicira, Inc.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of version 2 of the GNU General Public
6 * License as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
14 #include <linux/module.h>
15 #include <linux/openvswitch.h>
16 #include <linux/tcp.h>
17 #include <linux/udp.h>
18 #include <linux/sctp.h>
20 #include <net/netfilter/nf_conntrack_core.h>
21 #include <net/netfilter/nf_conntrack_helper.h>
22 #include <net/netfilter/nf_conntrack_labels.h>
23 #include <net/netfilter/nf_conntrack_seqadj.h>
24 #include <net/netfilter/nf_conntrack_zones.h>
25 #include <net/netfilter/ipv6/nf_defrag_ipv6.h>
27 #ifdef CONFIG_NF_NAT_NEEDED
28 #include <linux/netfilter/nf_nat.h>
29 #include <net/netfilter/nf_nat_core.h>
30 #include <net/netfilter/nf_nat_l3proto.h>
34 #include "conntrack.h"
36 #include "flow_netlink.h"
38 struct ovs_ct_len_tbl
{
43 /* Metadata mark for masked write to conntrack mark */
49 /* Metadata label for masked write to conntrack label. */
51 struct ovs_key_ct_labels value
;
52 struct ovs_key_ct_labels mask
;
56 OVS_CT_NAT
= 1 << 0, /* NAT for committed connections only. */
57 OVS_CT_SRC_NAT
= 1 << 1, /* Source NAT for NEW connections. */
58 OVS_CT_DST_NAT
= 1 << 2, /* Destination NAT for NEW connections. */
61 /* Conntrack action context for execution. */
62 struct ovs_conntrack_info
{
63 struct nf_conntrack_helper
*helper
;
64 struct nf_conntrack_zone zone
;
67 u8 nat
: 3; /* enum ovs_ct_nat */
70 struct md_labels labels
;
71 #ifdef CONFIG_NF_NAT_NEEDED
72 struct nf_nat_range range
; /* Only present for SRC NAT and DST NAT. */
76 static void __ovs_ct_free_action(struct ovs_conntrack_info
*ct_info
);
78 static u16
key_to_nfproto(const struct sw_flow_key
*key
)
80 switch (ntohs(key
->eth
.type
)) {
86 return NFPROTO_UNSPEC
;
90 /* Map SKB connection state into the values used by flow definition. */
91 static u8
ovs_ct_get_state(enum ip_conntrack_info ctinfo
)
93 u8 ct_state
= OVS_CS_F_TRACKED
;
96 case IP_CT_ESTABLISHED_REPLY
:
97 case IP_CT_RELATED_REPLY
:
98 ct_state
|= OVS_CS_F_REPLY_DIR
;
105 case IP_CT_ESTABLISHED
:
106 case IP_CT_ESTABLISHED_REPLY
:
107 ct_state
|= OVS_CS_F_ESTABLISHED
;
110 case IP_CT_RELATED_REPLY
:
111 ct_state
|= OVS_CS_F_RELATED
;
114 ct_state
|= OVS_CS_F_NEW
;
123 static u32
ovs_ct_get_mark(const struct nf_conn
*ct
)
125 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
126 return ct
? ct
->mark
: 0;
132 static void ovs_ct_get_labels(const struct nf_conn
*ct
,
133 struct ovs_key_ct_labels
*labels
)
135 struct nf_conn_labels
*cl
= ct
? nf_ct_labels_find(ct
) : NULL
;
138 size_t len
= cl
->words
* sizeof(long);
140 if (len
> OVS_CT_LABELS_LEN
)
141 len
= OVS_CT_LABELS_LEN
;
142 else if (len
< OVS_CT_LABELS_LEN
)
143 memset(labels
, 0, OVS_CT_LABELS_LEN
);
144 memcpy(labels
, cl
->bits
, len
);
146 memset(labels
, 0, OVS_CT_LABELS_LEN
);
150 static void __ovs_ct_update_key(struct sw_flow_key
*key
, u8 state
,
151 const struct nf_conntrack_zone
*zone
,
152 const struct nf_conn
*ct
)
154 key
->ct
.state
= state
;
155 key
->ct
.zone
= zone
->id
;
156 key
->ct
.mark
= ovs_ct_get_mark(ct
);
157 ovs_ct_get_labels(ct
, &key
->ct
.labels
);
160 /* Update 'key' based on skb->nfct. If 'post_ct' is true, then OVS has
161 * previously sent the packet to conntrack via the ct action. If
162 * 'keep_nat_flags' is true, the existing NAT flags retained, else they are
163 * initialized from the connection status.
165 static void ovs_ct_update_key(const struct sk_buff
*skb
,
166 const struct ovs_conntrack_info
*info
,
167 struct sw_flow_key
*key
, bool post_ct
,
170 const struct nf_conntrack_zone
*zone
= &nf_ct_zone_dflt
;
171 enum ip_conntrack_info ctinfo
;
175 ct
= nf_ct_get(skb
, &ctinfo
);
177 state
= ovs_ct_get_state(ctinfo
);
178 /* All unconfirmed entries are NEW connections. */
179 if (!nf_ct_is_confirmed(ct
))
180 state
|= OVS_CS_F_NEW
;
181 /* OVS persists the related flag for the duration of the
185 state
|= OVS_CS_F_RELATED
;
186 if (keep_nat_flags
) {
187 state
|= key
->ct
.state
& OVS_CS_F_NAT_MASK
;
189 if (ct
->status
& IPS_SRC_NAT
)
190 state
|= OVS_CS_F_SRC_NAT
;
191 if (ct
->status
& IPS_DST_NAT
)
192 state
|= OVS_CS_F_DST_NAT
;
194 zone
= nf_ct_zone(ct
);
195 } else if (post_ct
) {
196 state
= OVS_CS_F_TRACKED
| OVS_CS_F_INVALID
;
200 __ovs_ct_update_key(key
, state
, zone
, ct
);
203 /* This is called to initialize CT key fields possibly coming in from the local
206 void ovs_ct_fill_key(const struct sk_buff
*skb
, struct sw_flow_key
*key
)
208 ovs_ct_update_key(skb
, NULL
, key
, false, false);
211 int ovs_ct_put_key(const struct sw_flow_key
*key
, struct sk_buff
*skb
)
213 if (nla_put_u32(skb
, OVS_KEY_ATTR_CT_STATE
, key
->ct
.state
))
216 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES
) &&
217 nla_put_u16(skb
, OVS_KEY_ATTR_CT_ZONE
, key
->ct
.zone
))
220 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK
) &&
221 nla_put_u32(skb
, OVS_KEY_ATTR_CT_MARK
, key
->ct
.mark
))
224 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS
) &&
225 nla_put(skb
, OVS_KEY_ATTR_CT_LABELS
, sizeof(key
->ct
.labels
),
232 static int ovs_ct_set_mark(struct sk_buff
*skb
, struct sw_flow_key
*key
,
233 u32 ct_mark
, u32 mask
)
235 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
236 enum ip_conntrack_info ctinfo
;
240 /* The connection could be invalid, in which case set_mark is no-op. */
241 ct
= nf_ct_get(skb
, &ctinfo
);
245 new_mark
= ct_mark
| (ct
->mark
& ~(mask
));
246 if (ct
->mark
!= new_mark
) {
248 nf_conntrack_event_cache(IPCT_MARK
, ct
);
249 key
->ct
.mark
= new_mark
;
258 static int ovs_ct_set_labels(struct sk_buff
*skb
, struct sw_flow_key
*key
,
259 const struct ovs_key_ct_labels
*labels
,
260 const struct ovs_key_ct_labels
*mask
)
262 enum ip_conntrack_info ctinfo
;
263 struct nf_conn_labels
*cl
;
267 /* The connection could be invalid, in which case set_label is no-op.*/
268 ct
= nf_ct_get(skb
, &ctinfo
);
272 cl
= nf_ct_labels_find(ct
);
274 nf_ct_labels_ext_add(ct
);
275 cl
= nf_ct_labels_find(ct
);
277 if (!cl
|| cl
->words
* sizeof(long) < OVS_CT_LABELS_LEN
)
280 err
= nf_connlabels_replace(ct
, (u32
*)labels
, (u32
*)mask
,
281 OVS_CT_LABELS_LEN
/ sizeof(u32
));
285 ovs_ct_get_labels(ct
, &key
->ct
.labels
);
289 /* 'skb' should already be pulled to nh_ofs. */
290 static int ovs_ct_helper(struct sk_buff
*skb
, u16 proto
)
292 const struct nf_conntrack_helper
*helper
;
293 const struct nf_conn_help
*help
;
294 enum ip_conntrack_info ctinfo
;
295 unsigned int protoff
;
299 ct
= nf_ct_get(skb
, &ctinfo
);
300 if (!ct
|| ctinfo
== IP_CT_RELATED_REPLY
)
303 help
= nfct_help(ct
);
307 helper
= rcu_dereference(help
->helper
);
313 protoff
= ip_hdrlen(skb
);
316 u8 nexthdr
= ipv6_hdr(skb
)->nexthdr
;
320 ofs
= ipv6_skip_exthdr(skb
, sizeof(struct ipv6hdr
), &nexthdr
,
322 if (ofs
< 0 || (frag_off
& htons(~0x7)) != 0) {
323 pr_debug("proto header not found\n");
330 WARN_ONCE(1, "helper invoked on non-IP family!");
334 err
= helper
->help(skb
, protoff
, ct
, ctinfo
);
335 if (err
!= NF_ACCEPT
)
338 /* Adjust seqs after helper. This is needed due to some helpers (e.g.,
339 * FTP with NAT) adusting the TCP payload size when mangling IP
340 * addresses and/or port numbers in the text-based control connection.
342 if (test_bit(IPS_SEQ_ADJUST_BIT
, &ct
->status
) &&
343 !nf_ct_seq_adjust(skb
, ct
, ctinfo
, protoff
))
348 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
349 * value if 'skb' is freed.
351 static int handle_fragments(struct net
*net
, struct sw_flow_key
*key
,
352 u16 zone
, struct sk_buff
*skb
)
354 struct ovs_skb_cb ovs_cb
= *OVS_CB(skb
);
357 if (key
->eth
.type
== htons(ETH_P_IP
)) {
358 enum ip_defrag_users user
= IP_DEFRAG_CONNTRACK_IN
+ zone
;
360 memset(IPCB(skb
), 0, sizeof(struct inet_skb_parm
));
361 err
= ip_defrag(net
, skb
, user
);
365 ovs_cb
.mru
= IPCB(skb
)->frag_max_size
;
366 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
367 } else if (key
->eth
.type
== htons(ETH_P_IPV6
)) {
368 enum ip6_defrag_users user
= IP6_DEFRAG_CONNTRACK_IN
+ zone
;
370 memset(IP6CB(skb
), 0, sizeof(struct inet6_skb_parm
));
371 err
= nf_ct_frag6_gather(net
, skb
, user
);
375 key
->ip
.proto
= ipv6_hdr(skb
)->nexthdr
;
376 ovs_cb
.mru
= IP6CB(skb
)->frag_max_size
;
380 return -EPFNOSUPPORT
;
383 key
->ip
.frag
= OVS_FRAG_TYPE_NONE
;
386 *OVS_CB(skb
) = ovs_cb
;
391 static struct nf_conntrack_expect
*
392 ovs_ct_expect_find(struct net
*net
, const struct nf_conntrack_zone
*zone
,
393 u16 proto
, const struct sk_buff
*skb
)
395 struct nf_conntrack_tuple tuple
;
397 if (!nf_ct_get_tuplepr(skb
, skb_network_offset(skb
), proto
, net
, &tuple
))
399 return __nf_ct_expect_find(net
, zone
, &tuple
);
402 /* This replicates logic from nf_conntrack_core.c that is not exported. */
403 static enum ip_conntrack_info
404 ovs_ct_get_info(const struct nf_conntrack_tuple_hash
*h
)
406 const struct nf_conn
*ct
= nf_ct_tuplehash_to_ctrack(h
);
408 if (NF_CT_DIRECTION(h
) == IP_CT_DIR_REPLY
)
409 return IP_CT_ESTABLISHED_REPLY
;
410 /* Once we've had two way comms, always ESTABLISHED. */
411 if (test_bit(IPS_SEEN_REPLY_BIT
, &ct
->status
))
412 return IP_CT_ESTABLISHED
;
413 if (test_bit(IPS_EXPECTED_BIT
, &ct
->status
))
414 return IP_CT_RELATED
;
418 /* Find an existing connection which this packet belongs to without
419 * re-attributing statistics or modifying the connection state. This allows an
420 * skb->nfct lost due to an upcall to be recovered during actions execution.
422 * Must be called with rcu_read_lock.
424 * On success, populates skb->nfct and skb->nfctinfo, and returns the
425 * connection. Returns NULL if there is no existing entry.
427 static struct nf_conn
*
428 ovs_ct_find_existing(struct net
*net
, const struct nf_conntrack_zone
*zone
,
429 u8 l3num
, struct sk_buff
*skb
)
431 struct nf_conntrack_l3proto
*l3proto
;
432 struct nf_conntrack_l4proto
*l4proto
;
433 struct nf_conntrack_tuple tuple
;
434 struct nf_conntrack_tuple_hash
*h
;
435 enum ip_conntrack_info ctinfo
;
437 unsigned int dataoff
;
440 l3proto
= __nf_ct_l3proto_find(l3num
);
442 pr_debug("ovs_ct_find_existing: Can't get l3proto\n");
445 if (l3proto
->get_l4proto(skb
, skb_network_offset(skb
), &dataoff
,
447 pr_debug("ovs_ct_find_existing: Can't get protonum\n");
450 l4proto
= __nf_ct_l4proto_find(l3num
, protonum
);
452 pr_debug("ovs_ct_find_existing: Can't get l4proto\n");
455 if (!nf_ct_get_tuple(skb
, skb_network_offset(skb
), dataoff
, l3num
,
456 protonum
, net
, &tuple
, l3proto
, l4proto
)) {
457 pr_debug("ovs_ct_find_existing: Can't get tuple\n");
461 /* look for tuple match */
462 h
= nf_conntrack_find_get(net
, zone
, &tuple
);
464 return NULL
; /* Not found. */
466 ct
= nf_ct_tuplehash_to_ctrack(h
);
468 ctinfo
= ovs_ct_get_info(h
);
469 if (ctinfo
== IP_CT_NEW
) {
470 /* This should not happen. */
471 WARN_ONCE(1, "ovs_ct_find_existing: new packet for %p\n", ct
);
473 skb
->nfct
= &ct
->ct_general
;
474 skb
->nfctinfo
= ctinfo
;
478 /* Determine whether skb->nfct is equal to the result of conntrack lookup. */
479 static bool skb_nfct_cached(struct net
*net
,
480 const struct sw_flow_key
*key
,
481 const struct ovs_conntrack_info
*info
,
484 enum ip_conntrack_info ctinfo
;
487 ct
= nf_ct_get(skb
, &ctinfo
);
488 /* If no ct, check if we have evidence that an existing conntrack entry
489 * might be found for this skb. This happens when we lose a skb->nfct
490 * due to an upcall. If the connection was not confirmed, it is not
491 * cached and needs to be run through conntrack again.
493 if (!ct
&& key
->ct
.state
& OVS_CS_F_TRACKED
&&
494 !(key
->ct
.state
& OVS_CS_F_INVALID
) &&
495 key
->ct
.zone
== info
->zone
.id
)
496 ct
= ovs_ct_find_existing(net
, &info
->zone
, info
->family
, skb
);
499 if (!net_eq(net
, read_pnet(&ct
->ct_net
)))
501 if (!nf_ct_zone_equal_any(info
->ct
, nf_ct_zone(ct
)))
504 struct nf_conn_help
*help
;
506 help
= nf_ct_ext_find(ct
, NF_CT_EXT_HELPER
);
507 if (help
&& rcu_access_pointer(help
->helper
) != info
->helper
)
514 #ifdef CONFIG_NF_NAT_NEEDED
515 /* Modelled after nf_nat_ipv[46]_fn().
516 * range is only used for new, uninitialized NAT state.
517 * Returns either NF_ACCEPT or NF_DROP.
519 static int ovs_ct_nat_execute(struct sk_buff
*skb
, struct nf_conn
*ct
,
520 enum ip_conntrack_info ctinfo
,
521 const struct nf_nat_range
*range
,
522 enum nf_nat_manip_type maniptype
)
524 int hooknum
, nh_off
, err
= NF_ACCEPT
;
526 nh_off
= skb_network_offset(skb
);
527 skb_pull(skb
, nh_off
);
529 /* See HOOK2MANIP(). */
530 if (maniptype
== NF_NAT_MANIP_SRC
)
531 hooknum
= NF_INET_LOCAL_IN
; /* Source NAT */
533 hooknum
= NF_INET_LOCAL_OUT
; /* Destination NAT */
537 case IP_CT_RELATED_REPLY
:
538 if (skb
->protocol
== htons(ETH_P_IP
) &&
539 ip_hdr(skb
)->protocol
== IPPROTO_ICMP
) {
540 if (!nf_nat_icmp_reply_translation(skb
, ct
, ctinfo
,
544 #if IS_ENABLED(CONFIG_NF_NAT_IPV6)
545 } else if (skb
->protocol
== htons(ETH_P_IPV6
)) {
547 u8 nexthdr
= ipv6_hdr(skb
)->nexthdr
;
548 int hdrlen
= ipv6_skip_exthdr(skb
,
549 sizeof(struct ipv6hdr
),
550 &nexthdr
, &frag_off
);
552 if (hdrlen
>= 0 && nexthdr
== IPPROTO_ICMPV6
) {
553 if (!nf_nat_icmpv6_reply_translation(skb
, ct
,
562 /* Non-ICMP, fall thru to initialize if needed. */
564 /* Seen it before? This can happen for loopback, retrans,
567 if (!nf_nat_initialized(ct
, maniptype
)) {
568 /* Initialize according to the NAT action. */
569 err
= (range
&& range
->flags
& NF_NAT_RANGE_MAP_IPS
)
570 /* Action is set up to establish a new
573 ? nf_nat_setup_info(ct
, range
, maniptype
)
574 : nf_nat_alloc_null_binding(ct
, hooknum
);
575 if (err
!= NF_ACCEPT
)
580 case IP_CT_ESTABLISHED
:
581 case IP_CT_ESTABLISHED_REPLY
:
589 err
= nf_nat_packet(ct
, ctinfo
, hooknum
, skb
);
591 skb_push(skb
, nh_off
);
596 static void ovs_nat_update_key(struct sw_flow_key
*key
,
597 const struct sk_buff
*skb
,
598 enum nf_nat_manip_type maniptype
)
600 if (maniptype
== NF_NAT_MANIP_SRC
) {
603 key
->ct
.state
|= OVS_CS_F_SRC_NAT
;
604 if (key
->eth
.type
== htons(ETH_P_IP
))
605 key
->ipv4
.addr
.src
= ip_hdr(skb
)->saddr
;
606 else if (key
->eth
.type
== htons(ETH_P_IPV6
))
607 memcpy(&key
->ipv6
.addr
.src
, &ipv6_hdr(skb
)->saddr
,
608 sizeof(key
->ipv6
.addr
.src
));
612 if (key
->ip
.proto
== IPPROTO_UDP
)
613 src
= udp_hdr(skb
)->source
;
614 else if (key
->ip
.proto
== IPPROTO_TCP
)
615 src
= tcp_hdr(skb
)->source
;
616 else if (key
->ip
.proto
== IPPROTO_SCTP
)
617 src
= sctp_hdr(skb
)->source
;
625 key
->ct
.state
|= OVS_CS_F_DST_NAT
;
626 if (key
->eth
.type
== htons(ETH_P_IP
))
627 key
->ipv4
.addr
.dst
= ip_hdr(skb
)->daddr
;
628 else if (key
->eth
.type
== htons(ETH_P_IPV6
))
629 memcpy(&key
->ipv6
.addr
.dst
, &ipv6_hdr(skb
)->daddr
,
630 sizeof(key
->ipv6
.addr
.dst
));
634 if (key
->ip
.proto
== IPPROTO_UDP
)
635 dst
= udp_hdr(skb
)->dest
;
636 else if (key
->ip
.proto
== IPPROTO_TCP
)
637 dst
= tcp_hdr(skb
)->dest
;
638 else if (key
->ip
.proto
== IPPROTO_SCTP
)
639 dst
= sctp_hdr(skb
)->dest
;
647 /* Returns NF_DROP if the packet should be dropped, NF_ACCEPT otherwise. */
648 static int ovs_ct_nat(struct net
*net
, struct sw_flow_key
*key
,
649 const struct ovs_conntrack_info
*info
,
650 struct sk_buff
*skb
, struct nf_conn
*ct
,
651 enum ip_conntrack_info ctinfo
)
653 enum nf_nat_manip_type maniptype
;
656 if (nf_ct_is_untracked(ct
)) {
657 /* A NAT action may only be performed on tracked packets. */
661 /* Add NAT extension if not confirmed yet. */
662 if (!nf_ct_is_confirmed(ct
) && !nf_ct_nat_ext_add(ct
))
663 return NF_ACCEPT
; /* Can't NAT. */
665 /* Determine NAT type.
666 * Check if the NAT type can be deduced from the tracked connection.
667 * Make sure expected traffic is NATted only when committing.
669 if (info
->nat
& OVS_CT_NAT
&& ctinfo
!= IP_CT_NEW
&&
670 ct
->status
& IPS_NAT_MASK
&&
671 (!(ct
->status
& IPS_EXPECTED_BIT
) || info
->commit
)) {
672 /* NAT an established or related connection like before. */
673 if (CTINFO2DIR(ctinfo
) == IP_CT_DIR_REPLY
)
674 /* This is the REPLY direction for a connection
675 * for which NAT was applied in the forward
676 * direction. Do the reverse NAT.
678 maniptype
= ct
->status
& IPS_SRC_NAT
679 ? NF_NAT_MANIP_DST
: NF_NAT_MANIP_SRC
;
681 maniptype
= ct
->status
& IPS_SRC_NAT
682 ? NF_NAT_MANIP_SRC
: NF_NAT_MANIP_DST
;
683 } else if (info
->nat
& OVS_CT_SRC_NAT
) {
684 maniptype
= NF_NAT_MANIP_SRC
;
685 } else if (info
->nat
& OVS_CT_DST_NAT
) {
686 maniptype
= NF_NAT_MANIP_DST
;
688 return NF_ACCEPT
; /* Connection is not NATed. */
690 err
= ovs_ct_nat_execute(skb
, ct
, ctinfo
, &info
->range
, maniptype
);
692 /* Mark NAT done if successful and update the flow key. */
693 if (err
== NF_ACCEPT
)
694 ovs_nat_update_key(key
, skb
, maniptype
);
698 #else /* !CONFIG_NF_NAT_NEEDED */
699 static int ovs_ct_nat(struct net
*net
, struct sw_flow_key
*key
,
700 const struct ovs_conntrack_info
*info
,
701 struct sk_buff
*skb
, struct nf_conn
*ct
,
702 enum ip_conntrack_info ctinfo
)
708 /* Pass 'skb' through conntrack in 'net', using zone configured in 'info', if
709 * not done already. Update key with new CT state after passing the packet
711 * Note that if the packet is deemed invalid by conntrack, skb->nfct will be
712 * set to NULL and 0 will be returned.
714 static int __ovs_ct_lookup(struct net
*net
, struct sw_flow_key
*key
,
715 const struct ovs_conntrack_info
*info
,
718 /* If we are recirculating packets to match on conntrack fields and
719 * committing with a separate conntrack action, then we don't need to
720 * actually run the packet through conntrack twice unless it's for a
723 bool cached
= skb_nfct_cached(net
, key
, info
, skb
);
724 enum ip_conntrack_info ctinfo
;
728 struct nf_conn
*tmpl
= info
->ct
;
731 /* Associate skb with specified zone. */
734 nf_conntrack_put(skb
->nfct
);
735 nf_conntrack_get(&tmpl
->ct_general
);
736 skb
->nfct
= &tmpl
->ct_general
;
737 skb
->nfctinfo
= IP_CT_NEW
;
740 /* Repeat if requested, see nf_iterate(). */
742 err
= nf_conntrack_in(net
, info
->family
,
743 NF_INET_PRE_ROUTING
, skb
);
744 } while (err
== NF_REPEAT
);
746 if (err
!= NF_ACCEPT
)
749 /* Clear CT state NAT flags to mark that we have not yet done
750 * NAT after the nf_conntrack_in() call. We can actually clear
751 * the whole state, as it will be re-initialized below.
755 /* Update the key, but keep the NAT flags. */
756 ovs_ct_update_key(skb
, info
, key
, true, true);
759 ct
= nf_ct_get(skb
, &ctinfo
);
761 /* Packets starting a new connection must be NATted before the
762 * helper, so that the helper knows about the NAT. We enforce
763 * this by delaying both NAT and helper calls for unconfirmed
764 * connections until the committing CT action. For later
765 * packets NAT and Helper may be called in either order.
767 * NAT will be done only if the CT action has NAT, and only
768 * once per packet (per zone), as guarded by the NAT bits in
771 if (info
->nat
&& !(key
->ct
.state
& OVS_CS_F_NAT_MASK
) &&
772 (nf_ct_is_confirmed(ct
) || info
->commit
) &&
773 ovs_ct_nat(net
, key
, info
, skb
, ct
, ctinfo
) != NF_ACCEPT
) {
777 /* Call the helper only if:
778 * - nf_conntrack_in() was executed above ("!cached") for a
779 * confirmed connection, or
780 * - When committing an unconfirmed connection.
782 if ((nf_ct_is_confirmed(ct
) ? !cached
: info
->commit
) &&
783 ovs_ct_helper(skb
, info
->family
) != NF_ACCEPT
) {
791 /* Lookup connection and read fields into key. */
792 static int ovs_ct_lookup(struct net
*net
, struct sw_flow_key
*key
,
793 const struct ovs_conntrack_info
*info
,
796 struct nf_conntrack_expect
*exp
;
798 /* If we pass an expected packet through nf_conntrack_in() the
799 * expectation is typically removed, but the packet could still be
800 * lost in upcall processing. To prevent this from happening we
801 * perform an explicit expectation lookup. Expected connections are
802 * always new, and will be passed through conntrack only when they are
803 * committed, as it is OK to remove the expectation at that time.
805 exp
= ovs_ct_expect_find(net
, &info
->zone
, info
->family
, skb
);
809 /* NOTE: New connections are NATted and Helped only when
810 * committed, so we are not calling into NAT here.
812 state
= OVS_CS_F_TRACKED
| OVS_CS_F_NEW
| OVS_CS_F_RELATED
;
813 __ovs_ct_update_key(key
, state
, &info
->zone
, exp
->master
);
815 return __ovs_ct_lookup(net
, key
, info
, skb
);
820 /* Lookup connection and confirm if unconfirmed. */
821 static int ovs_ct_commit(struct net
*net
, struct sw_flow_key
*key
,
822 const struct ovs_conntrack_info
*info
,
827 err
= __ovs_ct_lookup(net
, key
, info
, skb
);
830 /* This is a no-op if the connection has already been confirmed. */
831 if (nf_conntrack_confirm(skb
) != NF_ACCEPT
)
837 static bool labels_nonzero(const struct ovs_key_ct_labels
*labels
)
841 for (i
= 0; i
< sizeof(*labels
); i
++)
842 if (labels
->ct_labels
[i
])
848 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
849 * value if 'skb' is freed.
851 int ovs_ct_execute(struct net
*net
, struct sk_buff
*skb
,
852 struct sw_flow_key
*key
,
853 const struct ovs_conntrack_info
*info
)
858 /* The conntrack module expects to be working at L3. */
859 nh_ofs
= skb_network_offset(skb
);
860 skb_pull(skb
, nh_ofs
);
862 if (key
->ip
.frag
!= OVS_FRAG_TYPE_NONE
) {
863 err
= handle_fragments(net
, key
, info
->zone
.id
, skb
);
869 err
= ovs_ct_commit(net
, key
, info
, skb
);
871 err
= ovs_ct_lookup(net
, key
, info
, skb
);
875 if (info
->mark
.mask
) {
876 err
= ovs_ct_set_mark(skb
, key
, info
->mark
.value
,
881 if (labels_nonzero(&info
->labels
.mask
))
882 err
= ovs_ct_set_labels(skb
, key
, &info
->labels
.value
,
885 skb_push(skb
, nh_ofs
);
891 static int ovs_ct_add_helper(struct ovs_conntrack_info
*info
, const char *name
,
892 const struct sw_flow_key
*key
, bool log
)
894 struct nf_conntrack_helper
*helper
;
895 struct nf_conn_help
*help
;
897 helper
= nf_conntrack_helper_try_module_get(name
, info
->family
,
900 OVS_NLERR(log
, "Unknown helper \"%s\"", name
);
904 help
= nf_ct_helper_ext_add(info
->ct
, helper
, GFP_KERNEL
);
906 module_put(helper
->me
);
910 rcu_assign_pointer(help
->helper
, helper
);
911 info
->helper
= helper
;
915 #ifdef CONFIG_NF_NAT_NEEDED
916 static int parse_nat(const struct nlattr
*attr
,
917 struct ovs_conntrack_info
*info
, bool log
)
921 bool have_ip_max
= false;
922 bool have_proto_max
= false;
923 bool ip_vers
= (info
->family
== NFPROTO_IPV6
);
925 nla_for_each_nested(a
, attr
, rem
) {
926 static const int ovs_nat_attr_lens
[OVS_NAT_ATTR_MAX
+ 1][2] = {
927 [OVS_NAT_ATTR_SRC
] = {0, 0},
928 [OVS_NAT_ATTR_DST
] = {0, 0},
929 [OVS_NAT_ATTR_IP_MIN
] = {sizeof(struct in_addr
),
930 sizeof(struct in6_addr
)},
931 [OVS_NAT_ATTR_IP_MAX
] = {sizeof(struct in_addr
),
932 sizeof(struct in6_addr
)},
933 [OVS_NAT_ATTR_PROTO_MIN
] = {sizeof(u16
), sizeof(u16
)},
934 [OVS_NAT_ATTR_PROTO_MAX
] = {sizeof(u16
), sizeof(u16
)},
935 [OVS_NAT_ATTR_PERSISTENT
] = {0, 0},
936 [OVS_NAT_ATTR_PROTO_HASH
] = {0, 0},
937 [OVS_NAT_ATTR_PROTO_RANDOM
] = {0, 0},
939 int type
= nla_type(a
);
941 if (type
> OVS_NAT_ATTR_MAX
) {
943 "Unknown NAT attribute (type=%d, max=%d).\n",
944 type
, OVS_NAT_ATTR_MAX
);
948 if (nla_len(a
) != ovs_nat_attr_lens
[type
][ip_vers
]) {
950 "NAT attribute type %d has unexpected length (%d != %d).\n",
952 ovs_nat_attr_lens
[type
][ip_vers
]);
957 case OVS_NAT_ATTR_SRC
:
958 case OVS_NAT_ATTR_DST
:
961 "Only one type of NAT may be specified.\n"
965 info
->nat
|= OVS_CT_NAT
;
966 info
->nat
|= ((type
== OVS_NAT_ATTR_SRC
)
967 ? OVS_CT_SRC_NAT
: OVS_CT_DST_NAT
);
970 case OVS_NAT_ATTR_IP_MIN
:
971 nla_memcpy(&info
->range
.min_addr
, a
, nla_len(a
));
972 info
->range
.flags
|= NF_NAT_RANGE_MAP_IPS
;
975 case OVS_NAT_ATTR_IP_MAX
:
977 nla_memcpy(&info
->range
.max_addr
, a
,
978 sizeof(info
->range
.max_addr
));
979 info
->range
.flags
|= NF_NAT_RANGE_MAP_IPS
;
982 case OVS_NAT_ATTR_PROTO_MIN
:
983 info
->range
.min_proto
.all
= htons(nla_get_u16(a
));
984 info
->range
.flags
|= NF_NAT_RANGE_PROTO_SPECIFIED
;
987 case OVS_NAT_ATTR_PROTO_MAX
:
988 have_proto_max
= true;
989 info
->range
.max_proto
.all
= htons(nla_get_u16(a
));
990 info
->range
.flags
|= NF_NAT_RANGE_PROTO_SPECIFIED
;
993 case OVS_NAT_ATTR_PERSISTENT
:
994 info
->range
.flags
|= NF_NAT_RANGE_PERSISTENT
;
997 case OVS_NAT_ATTR_PROTO_HASH
:
998 info
->range
.flags
|= NF_NAT_RANGE_PROTO_RANDOM
;
1001 case OVS_NAT_ATTR_PROTO_RANDOM
:
1002 info
->range
.flags
|= NF_NAT_RANGE_PROTO_RANDOM_FULLY
;
1006 OVS_NLERR(log
, "Unknown nat attribute (%d).\n", type
);
1012 OVS_NLERR(log
, "NAT attribute has %d unknown bytes.\n", rem
);
1016 /* Do not allow flags if no type is given. */
1017 if (info
->range
.flags
) {
1019 "NAT flags may be given only when NAT range (SRC or DST) is also specified.\n"
1023 info
->nat
= OVS_CT_NAT
; /* NAT existing connections. */
1024 } else if (!info
->commit
) {
1026 "NAT attributes may be specified only when CT COMMIT flag is also specified.\n"
1030 /* Allow missing IP_MAX. */
1031 if (info
->range
.flags
& NF_NAT_RANGE_MAP_IPS
&& !have_ip_max
) {
1032 memcpy(&info
->range
.max_addr
, &info
->range
.min_addr
,
1033 sizeof(info
->range
.max_addr
));
1035 /* Allow missing PROTO_MAX. */
1036 if (info
->range
.flags
& NF_NAT_RANGE_PROTO_SPECIFIED
&&
1038 info
->range
.max_proto
.all
= info
->range
.min_proto
.all
;
1044 static const struct ovs_ct_len_tbl ovs_ct_attr_lens
[OVS_CT_ATTR_MAX
+ 1] = {
1045 [OVS_CT_ATTR_COMMIT
] = { .minlen
= 0, .maxlen
= 0 },
1046 [OVS_CT_ATTR_ZONE
] = { .minlen
= sizeof(u16
),
1047 .maxlen
= sizeof(u16
) },
1048 [OVS_CT_ATTR_MARK
] = { .minlen
= sizeof(struct md_mark
),
1049 .maxlen
= sizeof(struct md_mark
) },
1050 [OVS_CT_ATTR_LABELS
] = { .minlen
= sizeof(struct md_labels
),
1051 .maxlen
= sizeof(struct md_labels
) },
1052 [OVS_CT_ATTR_HELPER
] = { .minlen
= 1,
1053 .maxlen
= NF_CT_HELPER_NAME_LEN
},
1054 #ifdef CONFIG_NF_NAT_NEEDED
1055 /* NAT length is checked when parsing the nested attributes. */
1056 [OVS_CT_ATTR_NAT
] = { .minlen
= 0, .maxlen
= INT_MAX
},
1060 static int parse_ct(const struct nlattr
*attr
, struct ovs_conntrack_info
*info
,
1061 const char **helper
, bool log
)
1066 nla_for_each_nested(a
, attr
, rem
) {
1067 int type
= nla_type(a
);
1068 int maxlen
= ovs_ct_attr_lens
[type
].maxlen
;
1069 int minlen
= ovs_ct_attr_lens
[type
].minlen
;
1071 if (type
> OVS_CT_ATTR_MAX
) {
1073 "Unknown conntrack attr (type=%d, max=%d)",
1074 type
, OVS_CT_ATTR_MAX
);
1077 if (nla_len(a
) < minlen
|| nla_len(a
) > maxlen
) {
1079 "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
1080 type
, nla_len(a
), maxlen
);
1085 case OVS_CT_ATTR_COMMIT
:
1086 info
->commit
= true;
1088 #ifdef CONFIG_NF_CONNTRACK_ZONES
1089 case OVS_CT_ATTR_ZONE
:
1090 info
->zone
.id
= nla_get_u16(a
);
1093 #ifdef CONFIG_NF_CONNTRACK_MARK
1094 case OVS_CT_ATTR_MARK
: {
1095 struct md_mark
*mark
= nla_data(a
);
1098 OVS_NLERR(log
, "ct_mark mask cannot be 0");
1105 #ifdef CONFIG_NF_CONNTRACK_LABELS
1106 case OVS_CT_ATTR_LABELS
: {
1107 struct md_labels
*labels
= nla_data(a
);
1109 if (!labels_nonzero(&labels
->mask
)) {
1110 OVS_NLERR(log
, "ct_labels mask cannot be 0");
1113 info
->labels
= *labels
;
1117 case OVS_CT_ATTR_HELPER
:
1118 *helper
= nla_data(a
);
1119 if (!memchr(*helper
, '\0', nla_len(a
))) {
1120 OVS_NLERR(log
, "Invalid conntrack helper");
1124 #ifdef CONFIG_NF_NAT_NEEDED
1125 case OVS_CT_ATTR_NAT
: {
1126 int err
= parse_nat(a
, info
, log
);
1134 OVS_NLERR(log
, "Unknown conntrack attr (%d)",
1141 OVS_NLERR(log
, "Conntrack attr has %d unknown bytes", rem
);
1148 bool ovs_ct_verify(struct net
*net
, enum ovs_key_attr attr
)
1150 if (attr
== OVS_KEY_ATTR_CT_STATE
)
1152 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES
) &&
1153 attr
== OVS_KEY_ATTR_CT_ZONE
)
1155 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK
) &&
1156 attr
== OVS_KEY_ATTR_CT_MARK
)
1158 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS
) &&
1159 attr
== OVS_KEY_ATTR_CT_LABELS
) {
1160 struct ovs_net
*ovs_net
= net_generic(net
, ovs_net_id
);
1162 return ovs_net
->xt_label
;
1168 int ovs_ct_copy_action(struct net
*net
, const struct nlattr
*attr
,
1169 const struct sw_flow_key
*key
,
1170 struct sw_flow_actions
**sfa
, bool log
)
1172 struct ovs_conntrack_info ct_info
;
1173 const char *helper
= NULL
;
1177 family
= key_to_nfproto(key
);
1178 if (family
== NFPROTO_UNSPEC
) {
1179 OVS_NLERR(log
, "ct family unspecified");
1183 memset(&ct_info
, 0, sizeof(ct_info
));
1184 ct_info
.family
= family
;
1186 nf_ct_zone_init(&ct_info
.zone
, NF_CT_DEFAULT_ZONE_ID
,
1187 NF_CT_DEFAULT_ZONE_DIR
, 0);
1189 err
= parse_ct(attr
, &ct_info
, &helper
, log
);
1193 /* Set up template for tracking connections in specific zones. */
1194 ct_info
.ct
= nf_ct_tmpl_alloc(net
, &ct_info
.zone
, GFP_KERNEL
);
1196 OVS_NLERR(log
, "Failed to allocate conntrack template");
1200 __set_bit(IPS_CONFIRMED_BIT
, &ct_info
.ct
->status
);
1201 nf_conntrack_get(&ct_info
.ct
->ct_general
);
1204 err
= ovs_ct_add_helper(&ct_info
, helper
, key
, log
);
1209 err
= ovs_nla_add_action(sfa
, OVS_ACTION_ATTR_CT
, &ct_info
,
1210 sizeof(ct_info
), log
);
1216 __ovs_ct_free_action(&ct_info
);
1220 #ifdef CONFIG_NF_NAT_NEEDED
1221 static bool ovs_ct_nat_to_attr(const struct ovs_conntrack_info
*info
,
1222 struct sk_buff
*skb
)
1224 struct nlattr
*start
;
1226 start
= nla_nest_start(skb
, OVS_CT_ATTR_NAT
);
1230 if (info
->nat
& OVS_CT_SRC_NAT
) {
1231 if (nla_put_flag(skb
, OVS_NAT_ATTR_SRC
))
1233 } else if (info
->nat
& OVS_CT_DST_NAT
) {
1234 if (nla_put_flag(skb
, OVS_NAT_ATTR_DST
))
1240 if (info
->range
.flags
& NF_NAT_RANGE_MAP_IPS
) {
1241 if (info
->family
== NFPROTO_IPV4
) {
1242 if (nla_put_in_addr(skb
, OVS_NAT_ATTR_IP_MIN
,
1243 info
->range
.min_addr
.ip
) ||
1244 (info
->range
.max_addr
.ip
1245 != info
->range
.min_addr
.ip
&&
1246 (nla_put_in_addr(skb
, OVS_NAT_ATTR_IP_MAX
,
1247 info
->range
.max_addr
.ip
))))
1249 #if IS_ENABLED(CONFIG_NF_NAT_IPV6)
1250 } else if (info
->family
== NFPROTO_IPV6
) {
1251 if (nla_put_in6_addr(skb
, OVS_NAT_ATTR_IP_MIN
,
1252 &info
->range
.min_addr
.in6
) ||
1253 (memcmp(&info
->range
.max_addr
.in6
,
1254 &info
->range
.min_addr
.in6
,
1255 sizeof(info
->range
.max_addr
.in6
)) &&
1256 (nla_put_in6_addr(skb
, OVS_NAT_ATTR_IP_MAX
,
1257 &info
->range
.max_addr
.in6
))))
1264 if (info
->range
.flags
& NF_NAT_RANGE_PROTO_SPECIFIED
&&
1265 (nla_put_u16(skb
, OVS_NAT_ATTR_PROTO_MIN
,
1266 ntohs(info
->range
.min_proto
.all
)) ||
1267 (info
->range
.max_proto
.all
!= info
->range
.min_proto
.all
&&
1268 nla_put_u16(skb
, OVS_NAT_ATTR_PROTO_MAX
,
1269 ntohs(info
->range
.max_proto
.all
)))))
1272 if (info
->range
.flags
& NF_NAT_RANGE_PERSISTENT
&&
1273 nla_put_flag(skb
, OVS_NAT_ATTR_PERSISTENT
))
1275 if (info
->range
.flags
& NF_NAT_RANGE_PROTO_RANDOM
&&
1276 nla_put_flag(skb
, OVS_NAT_ATTR_PROTO_HASH
))
1278 if (info
->range
.flags
& NF_NAT_RANGE_PROTO_RANDOM_FULLY
&&
1279 nla_put_flag(skb
, OVS_NAT_ATTR_PROTO_RANDOM
))
1282 nla_nest_end(skb
, start
);
1288 int ovs_ct_action_to_attr(const struct ovs_conntrack_info
*ct_info
,
1289 struct sk_buff
*skb
)
1291 struct nlattr
*start
;
1293 start
= nla_nest_start(skb
, OVS_ACTION_ATTR_CT
);
1297 if (ct_info
->commit
&& nla_put_flag(skb
, OVS_CT_ATTR_COMMIT
))
1299 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES
) &&
1300 nla_put_u16(skb
, OVS_CT_ATTR_ZONE
, ct_info
->zone
.id
))
1302 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK
) && ct_info
->mark
.mask
&&
1303 nla_put(skb
, OVS_CT_ATTR_MARK
, sizeof(ct_info
->mark
),
1306 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS
) &&
1307 labels_nonzero(&ct_info
->labels
.mask
) &&
1308 nla_put(skb
, OVS_CT_ATTR_LABELS
, sizeof(ct_info
->labels
),
1311 if (ct_info
->helper
) {
1312 if (nla_put_string(skb
, OVS_CT_ATTR_HELPER
,
1313 ct_info
->helper
->name
))
1316 #ifdef CONFIG_NF_NAT_NEEDED
1317 if (ct_info
->nat
&& !ovs_ct_nat_to_attr(ct_info
, skb
))
1320 nla_nest_end(skb
, start
);
1325 void ovs_ct_free_action(const struct nlattr
*a
)
1327 struct ovs_conntrack_info
*ct_info
= nla_data(a
);
1329 __ovs_ct_free_action(ct_info
);
1332 static void __ovs_ct_free_action(struct ovs_conntrack_info
*ct_info
)
1334 if (ct_info
->helper
)
1335 module_put(ct_info
->helper
->me
);
1337 nf_ct_put(ct_info
->ct
);
1340 void ovs_ct_init(struct net
*net
)
1342 unsigned int n_bits
= sizeof(struct ovs_key_ct_labels
) * BITS_PER_BYTE
;
1343 struct ovs_net
*ovs_net
= net_generic(net
, ovs_net_id
);
1345 if (nf_connlabels_get(net
, n_bits
)) {
1346 ovs_net
->xt_label
= false;
1347 OVS_NLERR(true, "Failed to set connlabel length");
1349 ovs_net
->xt_label
= true;
1353 void ovs_ct_exit(struct net
*net
)
1355 struct ovs_net
*ovs_net
= net_generic(net
, ovs_net_id
);
1357 if (ovs_net
->xt_label
)
1358 nf_connlabels_put(net
);