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>
17 #include <net/netfilter/nf_conntrack_core.h>
18 #include <net/netfilter/nf_conntrack_helper.h>
19 #include <net/netfilter/nf_conntrack_labels.h>
20 #include <net/netfilter/nf_conntrack_zones.h>
21 #include <net/netfilter/ipv6/nf_defrag_ipv6.h>
24 #include "conntrack.h"
26 #include "flow_netlink.h"
28 struct ovs_ct_len_tbl
{
33 /* Metadata mark for masked write to conntrack mark */
39 /* Metadata label for masked write to conntrack label. */
41 struct ovs_key_ct_labels value
;
42 struct ovs_key_ct_labels mask
;
45 /* Conntrack action context for execution. */
46 struct ovs_conntrack_info
{
47 struct nf_conntrack_helper
*helper
;
48 struct nf_conntrack_zone zone
;
53 struct md_labels labels
;
56 static u16
key_to_nfproto(const struct sw_flow_key
*key
)
58 switch (ntohs(key
->eth
.type
)) {
64 return NFPROTO_UNSPEC
;
68 /* Map SKB connection state into the values used by flow definition. */
69 static u8
ovs_ct_get_state(enum ip_conntrack_info ctinfo
)
71 u8 ct_state
= OVS_CS_F_TRACKED
;
74 case IP_CT_ESTABLISHED_REPLY
:
75 case IP_CT_RELATED_REPLY
:
77 ct_state
|= OVS_CS_F_REPLY_DIR
;
84 case IP_CT_ESTABLISHED
:
85 case IP_CT_ESTABLISHED_REPLY
:
86 ct_state
|= OVS_CS_F_ESTABLISHED
;
89 case IP_CT_RELATED_REPLY
:
90 ct_state
|= OVS_CS_F_RELATED
;
94 ct_state
|= OVS_CS_F_NEW
;
103 static u32
ovs_ct_get_mark(const struct nf_conn
*ct
)
105 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
106 return ct
? ct
->mark
: 0;
112 static void ovs_ct_get_labels(const struct nf_conn
*ct
,
113 struct ovs_key_ct_labels
*labels
)
115 struct nf_conn_labels
*cl
= ct
? nf_ct_labels_find(ct
) : NULL
;
118 size_t len
= cl
->words
* sizeof(long);
120 if (len
> OVS_CT_LABELS_LEN
)
121 len
= OVS_CT_LABELS_LEN
;
122 else if (len
< OVS_CT_LABELS_LEN
)
123 memset(labels
, 0, OVS_CT_LABELS_LEN
);
124 memcpy(labels
, cl
->bits
, len
);
126 memset(labels
, 0, OVS_CT_LABELS_LEN
);
130 static void __ovs_ct_update_key(struct sw_flow_key
*key
, u8 state
,
131 const struct nf_conntrack_zone
*zone
,
132 const struct nf_conn
*ct
)
134 key
->ct
.state
= state
;
135 key
->ct
.zone
= zone
->id
;
136 key
->ct
.mark
= ovs_ct_get_mark(ct
);
137 ovs_ct_get_labels(ct
, &key
->ct
.labels
);
140 /* Update 'key' based on skb->nfct. If 'post_ct' is true, then OVS has
141 * previously sent the packet to conntrack via the ct action.
143 static void ovs_ct_update_key(const struct sk_buff
*skb
,
144 struct sw_flow_key
*key
, bool post_ct
)
146 const struct nf_conntrack_zone
*zone
= &nf_ct_zone_dflt
;
147 enum ip_conntrack_info ctinfo
;
151 ct
= nf_ct_get(skb
, &ctinfo
);
153 state
= ovs_ct_get_state(ctinfo
);
155 state
|= OVS_CS_F_RELATED
;
156 zone
= nf_ct_zone(ct
);
157 } else if (post_ct
) {
158 state
= OVS_CS_F_TRACKED
| OVS_CS_F_INVALID
;
160 __ovs_ct_update_key(key
, state
, zone
, ct
);
163 void ovs_ct_fill_key(const struct sk_buff
*skb
, struct sw_flow_key
*key
)
165 ovs_ct_update_key(skb
, key
, false);
168 int ovs_ct_put_key(const struct sw_flow_key
*key
, struct sk_buff
*skb
)
170 if (nla_put_u32(skb
, OVS_KEY_ATTR_CT_STATE
, key
->ct
.state
))
173 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES
) &&
174 nla_put_u16(skb
, OVS_KEY_ATTR_CT_ZONE
, key
->ct
.zone
))
177 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK
) &&
178 nla_put_u32(skb
, OVS_KEY_ATTR_CT_MARK
, key
->ct
.mark
))
181 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS
) &&
182 nla_put(skb
, OVS_KEY_ATTR_CT_LABELS
, sizeof(key
->ct
.labels
),
189 static int ovs_ct_set_mark(struct sk_buff
*skb
, struct sw_flow_key
*key
,
190 u32 ct_mark
, u32 mask
)
192 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
193 enum ip_conntrack_info ctinfo
;
198 /* The connection could be invalid, in which case set_mark is no-op. */
199 ct
= nf_ct_get(skb
, &ctinfo
);
203 new_mark
= ct_mark
| (ct
->mark
& ~(mask
));
204 if (ct
->mark
!= new_mark
) {
206 nf_conntrack_event_cache(IPCT_MARK
, ct
);
207 key
->ct
.mark
= new_mark
;
216 static int ovs_ct_set_labels(struct sk_buff
*skb
, struct sw_flow_key
*key
,
217 const struct ovs_key_ct_labels
*labels
,
218 const struct ovs_key_ct_labels
*mask
)
220 enum ip_conntrack_info ctinfo
;
221 struct nf_conn_labels
*cl
;
225 if (!IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS
))
228 /* The connection could be invalid, in which case set_label is no-op.*/
229 ct
= nf_ct_get(skb
, &ctinfo
);
233 cl
= nf_ct_labels_find(ct
);
235 nf_ct_labels_ext_add(ct
);
236 cl
= nf_ct_labels_find(ct
);
238 if (!cl
|| cl
->words
* sizeof(long) < OVS_CT_LABELS_LEN
)
241 err
= nf_connlabels_replace(ct
, (u32
*)labels
, (u32
*)mask
,
242 OVS_CT_LABELS_LEN
/ sizeof(u32
));
246 ovs_ct_get_labels(ct
, &key
->ct
.labels
);
250 /* 'skb' should already be pulled to nh_ofs. */
251 static int ovs_ct_helper(struct sk_buff
*skb
, u16 proto
)
253 const struct nf_conntrack_helper
*helper
;
254 const struct nf_conn_help
*help
;
255 enum ip_conntrack_info ctinfo
;
256 unsigned int protoff
;
259 ct
= nf_ct_get(skb
, &ctinfo
);
260 if (!ct
|| ctinfo
== IP_CT_RELATED_REPLY
)
263 help
= nfct_help(ct
);
267 helper
= rcu_dereference(help
->helper
);
273 protoff
= ip_hdrlen(skb
);
276 u8 nexthdr
= ipv6_hdr(skb
)->nexthdr
;
280 ofs
= ipv6_skip_exthdr(skb
, sizeof(struct ipv6hdr
), &nexthdr
,
282 if (ofs
< 0 || (frag_off
& htons(~0x7)) != 0) {
283 pr_debug("proto header not found\n");
290 WARN_ONCE(1, "helper invoked on non-IP family!");
294 return helper
->help(skb
, protoff
, ct
, ctinfo
);
297 static int handle_fragments(struct net
*net
, struct sw_flow_key
*key
,
298 u16 zone
, struct sk_buff
*skb
)
300 struct ovs_skb_cb ovs_cb
= *OVS_CB(skb
);
302 if (key
->eth
.type
== htons(ETH_P_IP
)) {
303 enum ip_defrag_users user
= IP_DEFRAG_CONNTRACK_IN
+ zone
;
306 memset(IPCB(skb
), 0, sizeof(struct inet_skb_parm
));
307 err
= ip_defrag(skb
, user
);
311 ovs_cb
.mru
= IPCB(skb
)->frag_max_size
;
312 } else if (key
->eth
.type
== htons(ETH_P_IPV6
)) {
313 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
314 enum ip6_defrag_users user
= IP6_DEFRAG_CONNTRACK_IN
+ zone
;
315 struct sk_buff
*reasm
;
317 memset(IP6CB(skb
), 0, sizeof(struct inet6_skb_parm
));
318 reasm
= nf_ct_frag6_gather(skb
, user
);
325 key
->ip
.proto
= ipv6_hdr(reasm
)->nexthdr
;
326 skb_morph(skb
, reasm
);
328 ovs_cb
.mru
= IP6CB(skb
)->frag_max_size
;
330 return -EPFNOSUPPORT
;
333 return -EPFNOSUPPORT
;
336 key
->ip
.frag
= OVS_FRAG_TYPE_NONE
;
339 *OVS_CB(skb
) = ovs_cb
;
344 static struct nf_conntrack_expect
*
345 ovs_ct_expect_find(struct net
*net
, const struct nf_conntrack_zone
*zone
,
346 u16 proto
, const struct sk_buff
*skb
)
348 struct nf_conntrack_tuple tuple
;
350 if (!nf_ct_get_tuplepr(skb
, skb_network_offset(skb
), proto
, &tuple
))
352 return __nf_ct_expect_find(net
, zone
, &tuple
);
355 /* Determine whether skb->nfct is equal to the result of conntrack lookup. */
356 static bool skb_nfct_cached(const struct net
*net
, const struct sk_buff
*skb
,
357 const struct ovs_conntrack_info
*info
)
359 enum ip_conntrack_info ctinfo
;
362 ct
= nf_ct_get(skb
, &ctinfo
);
365 if (!net_eq(net
, read_pnet(&ct
->ct_net
)))
367 if (!nf_ct_zone_equal_any(info
->ct
, nf_ct_zone(ct
)))
370 struct nf_conn_help
*help
;
372 help
= nf_ct_ext_find(ct
, NF_CT_EXT_HELPER
);
373 if (help
&& rcu_access_pointer(help
->helper
) != info
->helper
)
380 static int __ovs_ct_lookup(struct net
*net
, const struct sw_flow_key
*key
,
381 const struct ovs_conntrack_info
*info
,
384 /* If we are recirculating packets to match on conntrack fields and
385 * committing with a separate conntrack action, then we don't need to
386 * actually run the packet through conntrack twice unless it's for a
389 if (!skb_nfct_cached(net
, skb
, info
)) {
390 struct nf_conn
*tmpl
= info
->ct
;
392 /* Associate skb with specified zone. */
395 nf_conntrack_put(skb
->nfct
);
396 nf_conntrack_get(&tmpl
->ct_general
);
397 skb
->nfct
= &tmpl
->ct_general
;
398 skb
->nfctinfo
= IP_CT_NEW
;
401 if (nf_conntrack_in(net
, info
->family
, NF_INET_PRE_ROUTING
,
405 if (ovs_ct_helper(skb
, info
->family
) != NF_ACCEPT
) {
406 WARN_ONCE(1, "helper rejected packet");
414 /* Lookup connection and read fields into key. */
415 static int ovs_ct_lookup(struct net
*net
, struct sw_flow_key
*key
,
416 const struct ovs_conntrack_info
*info
,
419 struct nf_conntrack_expect
*exp
;
421 exp
= ovs_ct_expect_find(net
, &info
->zone
, info
->family
, skb
);
425 state
= OVS_CS_F_TRACKED
| OVS_CS_F_NEW
| OVS_CS_F_RELATED
;
426 __ovs_ct_update_key(key
, state
, &info
->zone
, exp
->master
);
430 err
= __ovs_ct_lookup(net
, key
, info
, skb
);
434 ovs_ct_update_key(skb
, key
, true);
440 /* Lookup connection and confirm if unconfirmed. */
441 static int ovs_ct_commit(struct net
*net
, struct sw_flow_key
*key
,
442 const struct ovs_conntrack_info
*info
,
448 state
= key
->ct
.state
;
449 if (key
->ct
.zone
== info
->zone
.id
&&
450 ((state
& OVS_CS_F_TRACKED
) && !(state
& OVS_CS_F_NEW
))) {
451 /* Previous lookup has shown that this connection is already
452 * tracked and committed. Skip committing.
457 err
= __ovs_ct_lookup(net
, key
, info
, skb
);
460 if (nf_conntrack_confirm(skb
) != NF_ACCEPT
)
463 ovs_ct_update_key(skb
, key
, true);
468 static bool labels_nonzero(const struct ovs_key_ct_labels
*labels
)
472 for (i
= 0; i
< sizeof(*labels
); i
++)
473 if (labels
->ct_labels
[i
])
479 int ovs_ct_execute(struct net
*net
, struct sk_buff
*skb
,
480 struct sw_flow_key
*key
,
481 const struct ovs_conntrack_info
*info
)
486 /* The conntrack module expects to be working at L3. */
487 nh_ofs
= skb_network_offset(skb
);
488 skb_pull(skb
, nh_ofs
);
490 if (key
->ip
.frag
!= OVS_FRAG_TYPE_NONE
) {
491 err
= handle_fragments(net
, key
, info
->zone
.id
, skb
);
497 err
= ovs_ct_commit(net
, key
, info
, skb
);
499 err
= ovs_ct_lookup(net
, key
, info
, skb
);
503 if (info
->mark
.mask
) {
504 err
= ovs_ct_set_mark(skb
, key
, info
->mark
.value
,
509 if (labels_nonzero(&info
->labels
.mask
))
510 err
= ovs_ct_set_labels(skb
, key
, &info
->labels
.value
,
513 skb_push(skb
, nh_ofs
);
517 static int ovs_ct_add_helper(struct ovs_conntrack_info
*info
, const char *name
,
518 const struct sw_flow_key
*key
, bool log
)
520 struct nf_conntrack_helper
*helper
;
521 struct nf_conn_help
*help
;
523 helper
= nf_conntrack_helper_try_module_get(name
, info
->family
,
526 OVS_NLERR(log
, "Unknown helper \"%s\"", name
);
530 help
= nf_ct_helper_ext_add(info
->ct
, helper
, GFP_KERNEL
);
532 module_put(helper
->me
);
536 rcu_assign_pointer(help
->helper
, helper
);
537 info
->helper
= helper
;
541 static const struct ovs_ct_len_tbl ovs_ct_attr_lens
[OVS_CT_ATTR_MAX
+ 1] = {
542 [OVS_CT_ATTR_COMMIT
] = { .minlen
= 0, .maxlen
= 0 },
543 [OVS_CT_ATTR_ZONE
] = { .minlen
= sizeof(u16
),
544 .maxlen
= sizeof(u16
) },
545 [OVS_CT_ATTR_MARK
] = { .minlen
= sizeof(struct md_mark
),
546 .maxlen
= sizeof(struct md_mark
) },
547 [OVS_CT_ATTR_LABELS
] = { .minlen
= sizeof(struct md_labels
),
548 .maxlen
= sizeof(struct md_labels
) },
549 [OVS_CT_ATTR_HELPER
] = { .minlen
= 1,
550 .maxlen
= NF_CT_HELPER_NAME_LEN
}
553 static int parse_ct(const struct nlattr
*attr
, struct ovs_conntrack_info
*info
,
554 const char **helper
, bool log
)
559 nla_for_each_nested(a
, attr
, rem
) {
560 int type
= nla_type(a
);
561 int maxlen
= ovs_ct_attr_lens
[type
].maxlen
;
562 int minlen
= ovs_ct_attr_lens
[type
].minlen
;
564 if (type
> OVS_CT_ATTR_MAX
) {
566 "Unknown conntrack attr (type=%d, max=%d)",
567 type
, OVS_CT_ATTR_MAX
);
570 if (nla_len(a
) < minlen
|| nla_len(a
) > maxlen
) {
572 "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
573 type
, nla_len(a
), maxlen
);
578 case OVS_CT_ATTR_COMMIT
:
581 #ifdef CONFIG_NF_CONNTRACK_ZONES
582 case OVS_CT_ATTR_ZONE
:
583 info
->zone
.id
= nla_get_u16(a
);
586 #ifdef CONFIG_NF_CONNTRACK_MARK
587 case OVS_CT_ATTR_MARK
: {
588 struct md_mark
*mark
= nla_data(a
);
594 #ifdef CONFIG_NF_CONNTRACK_LABELS
595 case OVS_CT_ATTR_LABELS
: {
596 struct md_labels
*labels
= nla_data(a
);
598 info
->labels
= *labels
;
602 case OVS_CT_ATTR_HELPER
:
603 *helper
= nla_data(a
);
604 if (!memchr(*helper
, '\0', nla_len(a
))) {
605 OVS_NLERR(log
, "Invalid conntrack helper");
610 OVS_NLERR(log
, "Unknown conntrack attr (%d)",
617 OVS_NLERR(log
, "Conntrack attr has %d unknown bytes", rem
);
624 bool ovs_ct_verify(struct net
*net
, enum ovs_key_attr attr
)
626 if (attr
== OVS_KEY_ATTR_CT_STATE
)
628 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES
) &&
629 attr
== OVS_KEY_ATTR_CT_ZONE
)
631 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK
) &&
632 attr
== OVS_KEY_ATTR_CT_MARK
)
634 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS
) &&
635 attr
== OVS_KEY_ATTR_CT_LABELS
) {
636 struct ovs_net
*ovs_net
= net_generic(net
, ovs_net_id
);
638 return ovs_net
->xt_label
;
644 int ovs_ct_copy_action(struct net
*net
, const struct nlattr
*attr
,
645 const struct sw_flow_key
*key
,
646 struct sw_flow_actions
**sfa
, bool log
)
648 struct ovs_conntrack_info ct_info
;
649 const char *helper
= NULL
;
653 family
= key_to_nfproto(key
);
654 if (family
== NFPROTO_UNSPEC
) {
655 OVS_NLERR(log
, "ct family unspecified");
659 memset(&ct_info
, 0, sizeof(ct_info
));
660 ct_info
.family
= family
;
662 nf_ct_zone_init(&ct_info
.zone
, NF_CT_DEFAULT_ZONE_ID
,
663 NF_CT_DEFAULT_ZONE_DIR
, 0);
665 err
= parse_ct(attr
, &ct_info
, &helper
, log
);
669 /* Set up template for tracking connections in specific zones. */
670 ct_info
.ct
= nf_ct_tmpl_alloc(net
, &ct_info
.zone
, GFP_KERNEL
);
672 OVS_NLERR(log
, "Failed to allocate conntrack template");
676 err
= ovs_ct_add_helper(&ct_info
, helper
, key
, log
);
681 err
= ovs_nla_add_action(sfa
, OVS_ACTION_ATTR_CT
, &ct_info
,
682 sizeof(ct_info
), log
);
686 __set_bit(IPS_CONFIRMED_BIT
, &ct_info
.ct
->status
);
687 nf_conntrack_get(&ct_info
.ct
->ct_general
);
690 nf_conntrack_free(ct_info
.ct
);
694 int ovs_ct_action_to_attr(const struct ovs_conntrack_info
*ct_info
,
697 struct nlattr
*start
;
699 start
= nla_nest_start(skb
, OVS_ACTION_ATTR_CT
);
703 if (ct_info
->commit
&& nla_put_flag(skb
, OVS_CT_ATTR_COMMIT
))
705 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES
) &&
706 nla_put_u16(skb
, OVS_CT_ATTR_ZONE
, ct_info
->zone
.id
))
708 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK
) &&
709 nla_put(skb
, OVS_CT_ATTR_MARK
, sizeof(ct_info
->mark
),
712 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS
) &&
713 nla_put(skb
, OVS_CT_ATTR_LABELS
, sizeof(ct_info
->labels
),
716 if (ct_info
->helper
) {
717 if (nla_put_string(skb
, OVS_CT_ATTR_HELPER
,
718 ct_info
->helper
->name
))
722 nla_nest_end(skb
, start
);
727 void ovs_ct_free_action(const struct nlattr
*a
)
729 struct ovs_conntrack_info
*ct_info
= nla_data(a
);
732 module_put(ct_info
->helper
->me
);
734 nf_ct_put(ct_info
->ct
);
737 void ovs_ct_init(struct net
*net
)
739 unsigned int n_bits
= sizeof(struct ovs_key_ct_labels
) * BITS_PER_BYTE
;
740 struct ovs_net
*ovs_net
= net_generic(net
, ovs_net_id
);
742 if (nf_connlabels_get(net
, n_bits
)) {
743 ovs_net
->xt_label
= false;
744 OVS_NLERR(true, "Failed to set connlabel length");
746 ovs_net
->xt_label
= true;
750 void ovs_ct_exit(struct net
*net
)
752 struct ovs_net
*ovs_net
= net_generic(net
, ovs_net_id
);
754 if (ovs_net
->xt_label
)
755 nf_connlabels_put(net
);