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_label value
;
42 struct ovs_key_ct_label 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_label label
;
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_label(const struct nf_conn
*ct
,
113 struct ovs_key_ct_label
*label
)
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_LABEL_LEN
)
121 len
= OVS_CT_LABEL_LEN
;
122 else if (len
< OVS_CT_LABEL_LEN
)
123 memset(label
, 0, OVS_CT_LABEL_LEN
);
124 memcpy(label
, cl
->bits
, len
);
126 memset(label
, 0, OVS_CT_LABEL_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_label(ct
, &key
->ct
.label
);
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_u8(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_LABEL
, sizeof(key
->ct
.label
),
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_label(struct sk_buff
*skb
, struct sw_flow_key
*key
,
217 const struct ovs_key_ct_label
*label
,
218 const struct ovs_key_ct_label
*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_LABEL_LEN
)
241 err
= nf_connlabels_replace(ct
, (u32
*)label
, (u32
*)mask
,
242 OVS_CT_LABEL_LEN
/ sizeof(u32
));
246 ovs_ct_get_label(ct
, &key
->ct
.label
);
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
;
279 protoff
= ipv6_skip_exthdr(skb
, sizeof(struct ipv6hdr
),
280 &nexthdr
, &frag_off
);
281 if (protoff
< 0 || (frag_off
& htons(~0x7)) != 0) {
282 pr_debug("proto header not found\n");
288 WARN_ONCE(1, "helper invoked on non-IP family!");
292 return helper
->help(skb
, protoff
, ct
, ctinfo
);
295 static int handle_fragments(struct net
*net
, struct sw_flow_key
*key
,
296 u16 zone
, struct sk_buff
*skb
)
298 struct ovs_skb_cb ovs_cb
= *OVS_CB(skb
);
300 if (key
->eth
.type
== htons(ETH_P_IP
)) {
301 enum ip_defrag_users user
= IP_DEFRAG_CONNTRACK_IN
+ zone
;
304 memset(IPCB(skb
), 0, sizeof(struct inet_skb_parm
));
305 err
= ip_defrag(skb
, user
);
309 ovs_cb
.mru
= IPCB(skb
)->frag_max_size
;
310 } else if (key
->eth
.type
== htons(ETH_P_IPV6
)) {
311 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
312 enum ip6_defrag_users user
= IP6_DEFRAG_CONNTRACK_IN
+ zone
;
313 struct sk_buff
*reasm
;
315 memset(IP6CB(skb
), 0, sizeof(struct inet6_skb_parm
));
316 reasm
= nf_ct_frag6_gather(skb
, user
);
323 key
->ip
.proto
= ipv6_hdr(reasm
)->nexthdr
;
324 skb_morph(skb
, reasm
);
326 ovs_cb
.mru
= IP6CB(skb
)->frag_max_size
;
328 return -EPFNOSUPPORT
;
331 return -EPFNOSUPPORT
;
334 key
->ip
.frag
= OVS_FRAG_TYPE_NONE
;
337 *OVS_CB(skb
) = ovs_cb
;
342 static struct nf_conntrack_expect
*
343 ovs_ct_expect_find(struct net
*net
, const struct nf_conntrack_zone
*zone
,
344 u16 proto
, const struct sk_buff
*skb
)
346 struct nf_conntrack_tuple tuple
;
348 if (!nf_ct_get_tuplepr(skb
, skb_network_offset(skb
), proto
, &tuple
))
350 return __nf_ct_expect_find(net
, zone
, &tuple
);
353 /* Determine whether skb->nfct is equal to the result of conntrack lookup. */
354 static bool skb_nfct_cached(const struct net
*net
, const struct sk_buff
*skb
,
355 const struct ovs_conntrack_info
*info
)
357 enum ip_conntrack_info ctinfo
;
360 ct
= nf_ct_get(skb
, &ctinfo
);
363 if (!net_eq(net
, read_pnet(&ct
->ct_net
)))
365 if (!nf_ct_zone_equal_any(info
->ct
, nf_ct_zone(ct
)))
368 struct nf_conn_help
*help
;
370 help
= nf_ct_ext_find(ct
, NF_CT_EXT_HELPER
);
371 if (help
&& rcu_access_pointer(help
->helper
) != info
->helper
)
378 static int __ovs_ct_lookup(struct net
*net
, const struct sw_flow_key
*key
,
379 const struct ovs_conntrack_info
*info
,
382 /* If we are recirculating packets to match on conntrack fields and
383 * committing with a separate conntrack action, then we don't need to
384 * actually run the packet through conntrack twice unless it's for a
387 if (!skb_nfct_cached(net
, skb
, info
)) {
388 struct nf_conn
*tmpl
= info
->ct
;
390 /* Associate skb with specified zone. */
393 nf_conntrack_put(skb
->nfct
);
394 nf_conntrack_get(&tmpl
->ct_general
);
395 skb
->nfct
= &tmpl
->ct_general
;
396 skb
->nfctinfo
= IP_CT_NEW
;
399 if (nf_conntrack_in(net
, info
->family
, NF_INET_PRE_ROUTING
,
403 if (ovs_ct_helper(skb
, info
->family
) != NF_ACCEPT
) {
404 WARN_ONCE(1, "helper rejected packet");
412 /* Lookup connection and read fields into key. */
413 static int ovs_ct_lookup(struct net
*net
, struct sw_flow_key
*key
,
414 const struct ovs_conntrack_info
*info
,
417 struct nf_conntrack_expect
*exp
;
419 exp
= ovs_ct_expect_find(net
, &info
->zone
, info
->family
, skb
);
423 state
= OVS_CS_F_TRACKED
| OVS_CS_F_NEW
| OVS_CS_F_RELATED
;
424 __ovs_ct_update_key(key
, state
, &info
->zone
, exp
->master
);
428 err
= __ovs_ct_lookup(net
, key
, info
, skb
);
432 ovs_ct_update_key(skb
, key
, true);
438 /* Lookup connection and confirm if unconfirmed. */
439 static int ovs_ct_commit(struct net
*net
, struct sw_flow_key
*key
,
440 const struct ovs_conntrack_info
*info
,
446 state
= key
->ct
.state
;
447 if (key
->ct
.zone
== info
->zone
.id
&&
448 ((state
& OVS_CS_F_TRACKED
) && !(state
& OVS_CS_F_NEW
))) {
449 /* Previous lookup has shown that this connection is already
450 * tracked and committed. Skip committing.
455 err
= __ovs_ct_lookup(net
, key
, info
, skb
);
458 if (nf_conntrack_confirm(skb
) != NF_ACCEPT
)
461 ovs_ct_update_key(skb
, key
, true);
466 static bool label_nonzero(const struct ovs_key_ct_label
*label
)
470 for (i
= 0; i
< sizeof(*label
); i
++)
471 if (label
->ct_label
[i
])
477 int ovs_ct_execute(struct net
*net
, struct sk_buff
*skb
,
478 struct sw_flow_key
*key
,
479 const struct ovs_conntrack_info
*info
)
484 /* The conntrack module expects to be working at L3. */
485 nh_ofs
= skb_network_offset(skb
);
486 skb_pull(skb
, nh_ofs
);
488 if (key
->ip
.frag
!= OVS_FRAG_TYPE_NONE
) {
489 err
= handle_fragments(net
, key
, info
->zone
.id
, skb
);
494 if (info
->flags
& OVS_CT_F_COMMIT
)
495 err
= ovs_ct_commit(net
, key
, info
, skb
);
497 err
= ovs_ct_lookup(net
, key
, info
, skb
);
501 if (info
->mark
.mask
) {
502 err
= ovs_ct_set_mark(skb
, key
, info
->mark
.value
,
507 if (label_nonzero(&info
->label
.mask
))
508 err
= ovs_ct_set_label(skb
, key
, &info
->label
.value
,
511 skb_push(skb
, nh_ofs
);
515 static int ovs_ct_add_helper(struct ovs_conntrack_info
*info
, const char *name
,
516 const struct sw_flow_key
*key
, bool log
)
518 struct nf_conntrack_helper
*helper
;
519 struct nf_conn_help
*help
;
521 helper
= nf_conntrack_helper_try_module_get(name
, info
->family
,
524 OVS_NLERR(log
, "Unknown helper \"%s\"", name
);
528 help
= nf_ct_helper_ext_add(info
->ct
, helper
, GFP_KERNEL
);
530 module_put(helper
->me
);
534 rcu_assign_pointer(help
->helper
, helper
);
535 info
->helper
= helper
;
539 static const struct ovs_ct_len_tbl ovs_ct_attr_lens
[OVS_CT_ATTR_MAX
+ 1] = {
540 [OVS_CT_ATTR_FLAGS
] = { .minlen
= sizeof(u32
),
541 .maxlen
= sizeof(u32
) },
542 [OVS_CT_ATTR_ZONE
] = { .minlen
= sizeof(u16
),
543 .maxlen
= sizeof(u16
) },
544 [OVS_CT_ATTR_MARK
] = { .minlen
= sizeof(struct md_mark
),
545 .maxlen
= sizeof(struct md_mark
) },
546 [OVS_CT_ATTR_LABEL
] = { .minlen
= sizeof(struct md_label
),
547 .maxlen
= sizeof(struct md_label
) },
548 [OVS_CT_ATTR_HELPER
] = { .minlen
= 1,
549 .maxlen
= NF_CT_HELPER_NAME_LEN
}
552 static int parse_ct(const struct nlattr
*attr
, struct ovs_conntrack_info
*info
,
553 const char **helper
, bool log
)
558 nla_for_each_nested(a
, attr
, rem
) {
559 int type
= nla_type(a
);
560 int maxlen
= ovs_ct_attr_lens
[type
].maxlen
;
561 int minlen
= ovs_ct_attr_lens
[type
].minlen
;
563 if (type
> OVS_CT_ATTR_MAX
) {
565 "Unknown conntrack attr (type=%d, max=%d)",
566 type
, OVS_CT_ATTR_MAX
);
569 if (nla_len(a
) < minlen
|| nla_len(a
) > maxlen
) {
571 "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
572 type
, nla_len(a
), maxlen
);
577 case OVS_CT_ATTR_FLAGS
:
578 info
->flags
= nla_get_u32(a
);
580 #ifdef CONFIG_NF_CONNTRACK_ZONES
581 case OVS_CT_ATTR_ZONE
:
582 info
->zone
.id
= nla_get_u16(a
);
585 #ifdef CONFIG_NF_CONNTRACK_MARK
586 case OVS_CT_ATTR_MARK
: {
587 struct md_mark
*mark
= nla_data(a
);
593 #ifdef CONFIG_NF_CONNTRACK_LABELS
594 case OVS_CT_ATTR_LABEL
: {
595 struct md_label
*label
= nla_data(a
);
597 info
->label
= *label
;
601 case OVS_CT_ATTR_HELPER
:
602 *helper
= nla_data(a
);
603 if (!memchr(*helper
, '\0', nla_len(a
))) {
604 OVS_NLERR(log
, "Invalid conntrack helper");
609 OVS_NLERR(log
, "Unknown conntrack attr (%d)",
616 OVS_NLERR(log
, "Conntrack attr has %d unknown bytes", rem
);
623 bool ovs_ct_verify(struct net
*net
, enum ovs_key_attr attr
)
625 if (attr
== OVS_KEY_ATTR_CT_STATE
)
627 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES
) &&
628 attr
== OVS_KEY_ATTR_CT_ZONE
)
630 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK
) &&
631 attr
== OVS_KEY_ATTR_CT_MARK
)
633 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS
) &&
634 attr
== OVS_KEY_ATTR_CT_LABEL
) {
635 struct ovs_net
*ovs_net
= net_generic(net
, ovs_net_id
);
637 return ovs_net
->xt_label
;
643 int ovs_ct_copy_action(struct net
*net
, const struct nlattr
*attr
,
644 const struct sw_flow_key
*key
,
645 struct sw_flow_actions
**sfa
, bool log
)
647 struct ovs_conntrack_info ct_info
;
648 const char *helper
= NULL
;
652 family
= key_to_nfproto(key
);
653 if (family
== NFPROTO_UNSPEC
) {
654 OVS_NLERR(log
, "ct family unspecified");
658 memset(&ct_info
, 0, sizeof(ct_info
));
659 ct_info
.family
= family
;
661 nf_ct_zone_init(&ct_info
.zone
, NF_CT_DEFAULT_ZONE_ID
,
662 NF_CT_DEFAULT_ZONE_DIR
, 0);
664 err
= parse_ct(attr
, &ct_info
, &helper
, log
);
668 /* Set up template for tracking connections in specific zones. */
669 ct_info
.ct
= nf_ct_tmpl_alloc(net
, &ct_info
.zone
, GFP_KERNEL
);
671 OVS_NLERR(log
, "Failed to allocate conntrack template");
675 err
= ovs_ct_add_helper(&ct_info
, helper
, key
, log
);
680 err
= ovs_nla_add_action(sfa
, OVS_ACTION_ATTR_CT
, &ct_info
,
681 sizeof(ct_info
), log
);
685 __set_bit(IPS_CONFIRMED_BIT
, &ct_info
.ct
->status
);
686 nf_conntrack_get(&ct_info
.ct
->ct_general
);
689 nf_conntrack_free(ct_info
.ct
);
693 int ovs_ct_action_to_attr(const struct ovs_conntrack_info
*ct_info
,
696 struct nlattr
*start
;
698 start
= nla_nest_start(skb
, OVS_ACTION_ATTR_CT
);
702 if (nla_put_u32(skb
, OVS_CT_ATTR_FLAGS
, ct_info
->flags
))
704 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES
) &&
705 nla_put_u16(skb
, OVS_CT_ATTR_ZONE
, ct_info
->zone
.id
))
707 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK
) &&
708 nla_put(skb
, OVS_CT_ATTR_MARK
, sizeof(ct_info
->mark
),
711 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS
) &&
712 nla_put(skb
, OVS_CT_ATTR_LABEL
, sizeof(ct_info
->label
),
715 if (ct_info
->helper
) {
716 if (nla_put_string(skb
, OVS_CT_ATTR_HELPER
,
717 ct_info
->helper
->name
))
721 nla_nest_end(skb
, start
);
726 void ovs_ct_free_action(const struct nlattr
*a
)
728 struct ovs_conntrack_info
*ct_info
= nla_data(a
);
731 module_put(ct_info
->helper
->me
);
733 nf_ct_put(ct_info
->ct
);
736 void ovs_ct_init(struct net
*net
)
738 unsigned int n_bits
= sizeof(struct ovs_key_ct_label
) * BITS_PER_BYTE
;
739 struct ovs_net
*ovs_net
= net_generic(net
, ovs_net_id
);
741 if (nf_connlabels_get(net
, n_bits
)) {
742 ovs_net
->xt_label
= false;
743 OVS_NLERR(true, "Failed to set connlabel length");
745 ovs_net
->xt_label
= true;
749 void ovs_ct_exit(struct net
*net
)
751 struct ovs_net
*ovs_net
= net_generic(net
, ovs_net_id
);
753 if (ovs_net
->xt_label
)
754 nf_connlabels_put(net
);