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
);
154 if (!nf_ct_is_confirmed(ct
))
155 state
|= OVS_CS_F_NEW
;
157 state
|= OVS_CS_F_RELATED
;
158 zone
= nf_ct_zone(ct
);
159 } else if (post_ct
) {
160 state
= OVS_CS_F_TRACKED
| OVS_CS_F_INVALID
;
162 __ovs_ct_update_key(key
, state
, zone
, ct
);
165 void ovs_ct_fill_key(const struct sk_buff
*skb
, struct sw_flow_key
*key
)
167 ovs_ct_update_key(skb
, key
, false);
170 int ovs_ct_put_key(const struct sw_flow_key
*key
, struct sk_buff
*skb
)
172 if (nla_put_u32(skb
, OVS_KEY_ATTR_CT_STATE
, key
->ct
.state
))
175 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES
) &&
176 nla_put_u16(skb
, OVS_KEY_ATTR_CT_ZONE
, key
->ct
.zone
))
179 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK
) &&
180 nla_put_u32(skb
, OVS_KEY_ATTR_CT_MARK
, key
->ct
.mark
))
183 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS
) &&
184 nla_put(skb
, OVS_KEY_ATTR_CT_LABELS
, sizeof(key
->ct
.labels
),
191 static int ovs_ct_set_mark(struct sk_buff
*skb
, struct sw_flow_key
*key
,
192 u32 ct_mark
, u32 mask
)
194 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
195 enum ip_conntrack_info ctinfo
;
200 /* The connection could be invalid, in which case set_mark is no-op. */
201 ct
= nf_ct_get(skb
, &ctinfo
);
205 new_mark
= ct_mark
| (ct
->mark
& ~(mask
));
206 if (ct
->mark
!= new_mark
) {
208 nf_conntrack_event_cache(IPCT_MARK
, ct
);
209 key
->ct
.mark
= new_mark
;
218 static int ovs_ct_set_labels(struct sk_buff
*skb
, struct sw_flow_key
*key
,
219 const struct ovs_key_ct_labels
*labels
,
220 const struct ovs_key_ct_labels
*mask
)
222 enum ip_conntrack_info ctinfo
;
223 struct nf_conn_labels
*cl
;
227 /* The connection could be invalid, in which case set_label is no-op.*/
228 ct
= nf_ct_get(skb
, &ctinfo
);
232 cl
= nf_ct_labels_find(ct
);
234 nf_ct_labels_ext_add(ct
);
235 cl
= nf_ct_labels_find(ct
);
237 if (!cl
|| cl
->words
* sizeof(long) < OVS_CT_LABELS_LEN
)
240 err
= nf_connlabels_replace(ct
, (u32
*)labels
, (u32
*)mask
,
241 OVS_CT_LABELS_LEN
/ sizeof(u32
));
245 ovs_ct_get_labels(ct
, &key
->ct
.labels
);
249 /* 'skb' should already be pulled to nh_ofs. */
250 static int ovs_ct_helper(struct sk_buff
*skb
, u16 proto
)
252 const struct nf_conntrack_helper
*helper
;
253 const struct nf_conn_help
*help
;
254 enum ip_conntrack_info ctinfo
;
255 unsigned int protoff
;
258 ct
= nf_ct_get(skb
, &ctinfo
);
259 if (!ct
|| ctinfo
== IP_CT_RELATED_REPLY
)
262 help
= nfct_help(ct
);
266 helper
= rcu_dereference(help
->helper
);
272 protoff
= ip_hdrlen(skb
);
275 u8 nexthdr
= ipv6_hdr(skb
)->nexthdr
;
279 ofs
= ipv6_skip_exthdr(skb
, sizeof(struct ipv6hdr
), &nexthdr
,
281 if (ofs
< 0 || (frag_off
& htons(~0x7)) != 0) {
282 pr_debug("proto header not found\n");
289 WARN_ONCE(1, "helper invoked on non-IP family!");
293 return helper
->help(skb
, protoff
, ct
, ctinfo
);
296 static int handle_fragments(struct net
*net
, struct sw_flow_key
*key
,
297 u16 zone
, struct sk_buff
*skb
)
299 struct ovs_skb_cb ovs_cb
= *OVS_CB(skb
);
301 if (key
->eth
.type
== htons(ETH_P_IP
)) {
302 enum ip_defrag_users user
= IP_DEFRAG_CONNTRACK_IN
+ zone
;
305 memset(IPCB(skb
), 0, sizeof(struct inet_skb_parm
));
306 err
= ip_defrag(skb
, user
);
310 ovs_cb
.mru
= IPCB(skb
)->frag_max_size
;
311 } else if (key
->eth
.type
== htons(ETH_P_IPV6
)) {
312 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
313 enum ip6_defrag_users user
= IP6_DEFRAG_CONNTRACK_IN
+ zone
;
314 struct sk_buff
*reasm
;
316 memset(IP6CB(skb
), 0, sizeof(struct inet6_skb_parm
));
317 reasm
= nf_ct_frag6_gather(skb
, user
);
324 key
->ip
.proto
= ipv6_hdr(reasm
)->nexthdr
;
325 skb_morph(skb
, reasm
);
327 ovs_cb
.mru
= IP6CB(skb
)->frag_max_size
;
329 return -EPFNOSUPPORT
;
332 return -EPFNOSUPPORT
;
335 key
->ip
.frag
= OVS_FRAG_TYPE_NONE
;
338 *OVS_CB(skb
) = ovs_cb
;
343 static struct nf_conntrack_expect
*
344 ovs_ct_expect_find(struct net
*net
, const struct nf_conntrack_zone
*zone
,
345 u16 proto
, const struct sk_buff
*skb
)
347 struct nf_conntrack_tuple tuple
;
349 if (!nf_ct_get_tuplepr(skb
, skb_network_offset(skb
), proto
, &tuple
))
351 return __nf_ct_expect_find(net
, zone
, &tuple
);
354 /* Determine whether skb->nfct is equal to the result of conntrack lookup. */
355 static bool skb_nfct_cached(const struct net
*net
, const struct sk_buff
*skb
,
356 const struct ovs_conntrack_info
*info
)
358 enum ip_conntrack_info ctinfo
;
361 ct
= nf_ct_get(skb
, &ctinfo
);
364 if (!net_eq(net
, read_pnet(&ct
->ct_net
)))
366 if (!nf_ct_zone_equal_any(info
->ct
, nf_ct_zone(ct
)))
369 struct nf_conn_help
*help
;
371 help
= nf_ct_ext_find(ct
, NF_CT_EXT_HELPER
);
372 if (help
&& rcu_access_pointer(help
->helper
) != info
->helper
)
379 static int __ovs_ct_lookup(struct net
*net
, struct sw_flow_key
*key
,
380 const struct ovs_conntrack_info
*info
,
383 /* If we are recirculating packets to match on conntrack fields and
384 * committing with a separate conntrack action, then we don't need to
385 * actually run the packet through conntrack twice unless it's for a
388 if (!skb_nfct_cached(net
, skb
, info
)) {
389 struct nf_conn
*tmpl
= info
->ct
;
391 /* Associate skb with specified zone. */
394 nf_conntrack_put(skb
->nfct
);
395 nf_conntrack_get(&tmpl
->ct_general
);
396 skb
->nfct
= &tmpl
->ct_general
;
397 skb
->nfctinfo
= IP_CT_NEW
;
400 if (nf_conntrack_in(net
, info
->family
, NF_INET_PRE_ROUTING
,
404 if (ovs_ct_helper(skb
, info
->family
) != NF_ACCEPT
) {
405 WARN_ONCE(1, "helper rejected packet");
410 ovs_ct_update_key(skb
, key
, true);
415 /* Lookup connection and read fields into key. */
416 static int ovs_ct_lookup(struct net
*net
, struct sw_flow_key
*key
,
417 const struct ovs_conntrack_info
*info
,
420 struct nf_conntrack_expect
*exp
;
422 exp
= ovs_ct_expect_find(net
, &info
->zone
, info
->family
, skb
);
426 state
= OVS_CS_F_TRACKED
| OVS_CS_F_NEW
| OVS_CS_F_RELATED
;
427 __ovs_ct_update_key(key
, state
, &info
->zone
, exp
->master
);
431 err
= __ovs_ct_lookup(net
, key
, info
, skb
);
439 /* Lookup connection and confirm if unconfirmed. */
440 static int ovs_ct_commit(struct net
*net
, struct sw_flow_key
*key
,
441 const struct ovs_conntrack_info
*info
,
447 state
= key
->ct
.state
;
448 if (key
->ct
.zone
== info
->zone
.id
&&
449 ((state
& OVS_CS_F_TRACKED
) && !(state
& OVS_CS_F_NEW
))) {
450 /* Previous lookup has shown that this connection is already
451 * tracked and committed. Skip committing.
456 err
= __ovs_ct_lookup(net
, key
, info
, skb
);
459 if (nf_conntrack_confirm(skb
) != NF_ACCEPT
)
465 static bool labels_nonzero(const struct ovs_key_ct_labels
*labels
)
469 for (i
= 0; i
< sizeof(*labels
); i
++)
470 if (labels
->ct_labels
[i
])
476 int ovs_ct_execute(struct net
*net
, struct sk_buff
*skb
,
477 struct sw_flow_key
*key
,
478 const struct ovs_conntrack_info
*info
)
483 /* The conntrack module expects to be working at L3. */
484 nh_ofs
= skb_network_offset(skb
);
485 skb_pull(skb
, nh_ofs
);
487 if (key
->ip
.frag
!= OVS_FRAG_TYPE_NONE
) {
488 err
= handle_fragments(net
, key
, info
->zone
.id
, skb
);
494 err
= ovs_ct_commit(net
, key
, info
, skb
);
496 err
= ovs_ct_lookup(net
, key
, info
, skb
);
500 if (info
->mark
.mask
) {
501 err
= ovs_ct_set_mark(skb
, key
, info
->mark
.value
,
506 if (labels_nonzero(&info
->labels
.mask
))
507 err
= ovs_ct_set_labels(skb
, key
, &info
->labels
.value
,
510 skb_push(skb
, nh_ofs
);
514 static int ovs_ct_add_helper(struct ovs_conntrack_info
*info
, const char *name
,
515 const struct sw_flow_key
*key
, bool log
)
517 struct nf_conntrack_helper
*helper
;
518 struct nf_conn_help
*help
;
520 helper
= nf_conntrack_helper_try_module_get(name
, info
->family
,
523 OVS_NLERR(log
, "Unknown helper \"%s\"", name
);
527 help
= nf_ct_helper_ext_add(info
->ct
, helper
, GFP_KERNEL
);
529 module_put(helper
->me
);
533 rcu_assign_pointer(help
->helper
, helper
);
534 info
->helper
= helper
;
538 static const struct ovs_ct_len_tbl ovs_ct_attr_lens
[OVS_CT_ATTR_MAX
+ 1] = {
539 [OVS_CT_ATTR_COMMIT
] = { .minlen
= 0, .maxlen
= 0 },
540 [OVS_CT_ATTR_ZONE
] = { .minlen
= sizeof(u16
),
541 .maxlen
= sizeof(u16
) },
542 [OVS_CT_ATTR_MARK
] = { .minlen
= sizeof(struct md_mark
),
543 .maxlen
= sizeof(struct md_mark
) },
544 [OVS_CT_ATTR_LABELS
] = { .minlen
= sizeof(struct md_labels
),
545 .maxlen
= sizeof(struct md_labels
) },
546 [OVS_CT_ATTR_HELPER
] = { .minlen
= 1,
547 .maxlen
= NF_CT_HELPER_NAME_LEN
}
550 static int parse_ct(const struct nlattr
*attr
, struct ovs_conntrack_info
*info
,
551 const char **helper
, bool log
)
556 nla_for_each_nested(a
, attr
, rem
) {
557 int type
= nla_type(a
);
558 int maxlen
= ovs_ct_attr_lens
[type
].maxlen
;
559 int minlen
= ovs_ct_attr_lens
[type
].minlen
;
561 if (type
> OVS_CT_ATTR_MAX
) {
563 "Unknown conntrack attr (type=%d, max=%d)",
564 type
, OVS_CT_ATTR_MAX
);
567 if (nla_len(a
) < minlen
|| nla_len(a
) > maxlen
) {
569 "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
570 type
, nla_len(a
), maxlen
);
575 case OVS_CT_ATTR_COMMIT
:
578 #ifdef CONFIG_NF_CONNTRACK_ZONES
579 case OVS_CT_ATTR_ZONE
:
580 info
->zone
.id
= nla_get_u16(a
);
583 #ifdef CONFIG_NF_CONNTRACK_MARK
584 case OVS_CT_ATTR_MARK
: {
585 struct md_mark
*mark
= nla_data(a
);
588 OVS_NLERR(log
, "ct_mark mask cannot be 0");
595 #ifdef CONFIG_NF_CONNTRACK_LABELS
596 case OVS_CT_ATTR_LABELS
: {
597 struct md_labels
*labels
= nla_data(a
);
599 if (!labels_nonzero(&labels
->mask
)) {
600 OVS_NLERR(log
, "ct_labels mask cannot be 0");
603 info
->labels
= *labels
;
607 case OVS_CT_ATTR_HELPER
:
608 *helper
= nla_data(a
);
609 if (!memchr(*helper
, '\0', nla_len(a
))) {
610 OVS_NLERR(log
, "Invalid conntrack helper");
615 OVS_NLERR(log
, "Unknown conntrack attr (%d)",
622 OVS_NLERR(log
, "Conntrack attr has %d unknown bytes", rem
);
629 bool ovs_ct_verify(struct net
*net
, enum ovs_key_attr attr
)
631 if (attr
== OVS_KEY_ATTR_CT_STATE
)
633 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES
) &&
634 attr
== OVS_KEY_ATTR_CT_ZONE
)
636 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK
) &&
637 attr
== OVS_KEY_ATTR_CT_MARK
)
639 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS
) &&
640 attr
== OVS_KEY_ATTR_CT_LABELS
) {
641 struct ovs_net
*ovs_net
= net_generic(net
, ovs_net_id
);
643 return ovs_net
->xt_label
;
649 int ovs_ct_copy_action(struct net
*net
, const struct nlattr
*attr
,
650 const struct sw_flow_key
*key
,
651 struct sw_flow_actions
**sfa
, bool log
)
653 struct ovs_conntrack_info ct_info
;
654 const char *helper
= NULL
;
658 family
= key_to_nfproto(key
);
659 if (family
== NFPROTO_UNSPEC
) {
660 OVS_NLERR(log
, "ct family unspecified");
664 memset(&ct_info
, 0, sizeof(ct_info
));
665 ct_info
.family
= family
;
667 nf_ct_zone_init(&ct_info
.zone
, NF_CT_DEFAULT_ZONE_ID
,
668 NF_CT_DEFAULT_ZONE_DIR
, 0);
670 err
= parse_ct(attr
, &ct_info
, &helper
, log
);
674 /* Set up template for tracking connections in specific zones. */
675 ct_info
.ct
= nf_ct_tmpl_alloc(net
, &ct_info
.zone
, GFP_KERNEL
);
677 OVS_NLERR(log
, "Failed to allocate conntrack template");
681 err
= ovs_ct_add_helper(&ct_info
, helper
, key
, log
);
686 err
= ovs_nla_add_action(sfa
, OVS_ACTION_ATTR_CT
, &ct_info
,
687 sizeof(ct_info
), log
);
691 __set_bit(IPS_CONFIRMED_BIT
, &ct_info
.ct
->status
);
692 nf_conntrack_get(&ct_info
.ct
->ct_general
);
695 nf_conntrack_free(ct_info
.ct
);
699 int ovs_ct_action_to_attr(const struct ovs_conntrack_info
*ct_info
,
702 struct nlattr
*start
;
704 start
= nla_nest_start(skb
, OVS_ACTION_ATTR_CT
);
708 if (ct_info
->commit
&& nla_put_flag(skb
, OVS_CT_ATTR_COMMIT
))
710 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES
) &&
711 nla_put_u16(skb
, OVS_CT_ATTR_ZONE
, ct_info
->zone
.id
))
713 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK
) && ct_info
->mark
.mask
&&
714 nla_put(skb
, OVS_CT_ATTR_MARK
, sizeof(ct_info
->mark
),
717 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS
) &&
718 labels_nonzero(&ct_info
->labels
.mask
) &&
719 nla_put(skb
, OVS_CT_ATTR_LABELS
, sizeof(ct_info
->labels
),
722 if (ct_info
->helper
) {
723 if (nla_put_string(skb
, OVS_CT_ATTR_HELPER
,
724 ct_info
->helper
->name
))
728 nla_nest_end(skb
, start
);
733 void ovs_ct_free_action(const struct nlattr
*a
)
735 struct ovs_conntrack_info
*ct_info
= nla_data(a
);
738 module_put(ct_info
->helper
->me
);
740 nf_ct_put(ct_info
->ct
);
743 void ovs_ct_init(struct net
*net
)
745 unsigned int n_bits
= sizeof(struct ovs_key_ct_labels
) * BITS_PER_BYTE
;
746 struct ovs_net
*ovs_net
= net_generic(net
, ovs_net_id
);
748 if (nf_connlabels_get(net
, n_bits
)) {
749 ovs_net
->xt_label
= false;
750 OVS_NLERR(true, "Failed to set connlabel length");
752 ovs_net
->xt_label
= true;
756 void ovs_ct_exit(struct net
*net
)
758 struct ovs_net
*ovs_net
= net_generic(net
, ovs_net_id
);
760 if (ovs_net
->xt_label
)
761 nf_connlabels_put(net
);