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 void __ovs_ct_free_action(struct ovs_conntrack_info
*ct_info
);
58 static u16
key_to_nfproto(const struct sw_flow_key
*key
)
60 switch (ntohs(key
->eth
.type
)) {
66 return NFPROTO_UNSPEC
;
70 /* Map SKB connection state into the values used by flow definition. */
71 static u8
ovs_ct_get_state(enum ip_conntrack_info ctinfo
)
73 u8 ct_state
= OVS_CS_F_TRACKED
;
76 case IP_CT_ESTABLISHED_REPLY
:
77 case IP_CT_RELATED_REPLY
:
78 ct_state
|= OVS_CS_F_REPLY_DIR
;
85 case IP_CT_ESTABLISHED
:
86 case IP_CT_ESTABLISHED_REPLY
:
87 ct_state
|= OVS_CS_F_ESTABLISHED
;
90 case IP_CT_RELATED_REPLY
:
91 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 const struct ovs_conntrack_info
*info
,
145 struct sw_flow_key
*key
, bool post_ct
)
147 const struct nf_conntrack_zone
*zone
= &nf_ct_zone_dflt
;
148 enum ip_conntrack_info ctinfo
;
152 ct
= nf_ct_get(skb
, &ctinfo
);
154 state
= ovs_ct_get_state(ctinfo
);
155 /* All unconfirmed entries are NEW connections. */
156 if (!nf_ct_is_confirmed(ct
))
157 state
|= OVS_CS_F_NEW
;
158 /* OVS persists the related flag for the duration of the
162 state
|= OVS_CS_F_RELATED
;
163 zone
= nf_ct_zone(ct
);
164 } else if (post_ct
) {
165 state
= OVS_CS_F_TRACKED
| OVS_CS_F_INVALID
;
169 __ovs_ct_update_key(key
, state
, zone
, ct
);
172 /* This is called to initialize CT key fields possibly coming in from the local
175 void ovs_ct_fill_key(const struct sk_buff
*skb
, struct sw_flow_key
*key
)
177 ovs_ct_update_key(skb
, NULL
, key
, false);
180 int ovs_ct_put_key(const struct sw_flow_key
*key
, struct sk_buff
*skb
)
182 if (nla_put_u32(skb
, OVS_KEY_ATTR_CT_STATE
, key
->ct
.state
))
185 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES
) &&
186 nla_put_u16(skb
, OVS_KEY_ATTR_CT_ZONE
, key
->ct
.zone
))
189 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK
) &&
190 nla_put_u32(skb
, OVS_KEY_ATTR_CT_MARK
, key
->ct
.mark
))
193 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS
) &&
194 nla_put(skb
, OVS_KEY_ATTR_CT_LABELS
, sizeof(key
->ct
.labels
),
201 static int ovs_ct_set_mark(struct sk_buff
*skb
, struct sw_flow_key
*key
,
202 u32 ct_mark
, u32 mask
)
204 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
205 enum ip_conntrack_info ctinfo
;
209 /* The connection could be invalid, in which case set_mark is no-op. */
210 ct
= nf_ct_get(skb
, &ctinfo
);
214 new_mark
= ct_mark
| (ct
->mark
& ~(mask
));
215 if (ct
->mark
!= new_mark
) {
217 nf_conntrack_event_cache(IPCT_MARK
, ct
);
218 key
->ct
.mark
= new_mark
;
227 static int ovs_ct_set_labels(struct sk_buff
*skb
, struct sw_flow_key
*key
,
228 const struct ovs_key_ct_labels
*labels
,
229 const struct ovs_key_ct_labels
*mask
)
231 enum ip_conntrack_info ctinfo
;
232 struct nf_conn_labels
*cl
;
236 /* The connection could be invalid, in which case set_label is no-op.*/
237 ct
= nf_ct_get(skb
, &ctinfo
);
241 cl
= nf_ct_labels_find(ct
);
243 nf_ct_labels_ext_add(ct
);
244 cl
= nf_ct_labels_find(ct
);
246 if (!cl
|| cl
->words
* sizeof(long) < OVS_CT_LABELS_LEN
)
249 err
= nf_connlabels_replace(ct
, (u32
*)labels
, (u32
*)mask
,
250 OVS_CT_LABELS_LEN
/ sizeof(u32
));
254 ovs_ct_get_labels(ct
, &key
->ct
.labels
);
258 /* 'skb' should already be pulled to nh_ofs. */
259 static int ovs_ct_helper(struct sk_buff
*skb
, u16 proto
)
261 const struct nf_conntrack_helper
*helper
;
262 const struct nf_conn_help
*help
;
263 enum ip_conntrack_info ctinfo
;
264 unsigned int protoff
;
267 ct
= nf_ct_get(skb
, &ctinfo
);
268 if (!ct
|| ctinfo
== IP_CT_RELATED_REPLY
)
271 help
= nfct_help(ct
);
275 helper
= rcu_dereference(help
->helper
);
281 protoff
= ip_hdrlen(skb
);
284 u8 nexthdr
= ipv6_hdr(skb
)->nexthdr
;
288 ofs
= ipv6_skip_exthdr(skb
, sizeof(struct ipv6hdr
), &nexthdr
,
290 if (ofs
< 0 || (frag_off
& htons(~0x7)) != 0) {
291 pr_debug("proto header not found\n");
298 WARN_ONCE(1, "helper invoked on non-IP family!");
302 return helper
->help(skb
, protoff
, ct
, ctinfo
);
305 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
306 * value if 'skb' is freed.
308 static int handle_fragments(struct net
*net
, struct sw_flow_key
*key
,
309 u16 zone
, struct sk_buff
*skb
)
311 struct ovs_skb_cb ovs_cb
= *OVS_CB(skb
);
314 if (key
->eth
.type
== htons(ETH_P_IP
)) {
315 enum ip_defrag_users user
= IP_DEFRAG_CONNTRACK_IN
+ zone
;
317 memset(IPCB(skb
), 0, sizeof(struct inet_skb_parm
));
318 err
= ip_defrag(net
, skb
, user
);
322 ovs_cb
.mru
= IPCB(skb
)->frag_max_size
;
323 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
324 } else if (key
->eth
.type
== htons(ETH_P_IPV6
)) {
325 enum ip6_defrag_users user
= IP6_DEFRAG_CONNTRACK_IN
+ zone
;
327 memset(IP6CB(skb
), 0, sizeof(struct inet6_skb_parm
));
328 err
= nf_ct_frag6_gather(net
, skb
, user
);
332 key
->ip
.proto
= ipv6_hdr(skb
)->nexthdr
;
333 ovs_cb
.mru
= IP6CB(skb
)->frag_max_size
;
337 return -EPFNOSUPPORT
;
340 key
->ip
.frag
= OVS_FRAG_TYPE_NONE
;
343 *OVS_CB(skb
) = ovs_cb
;
348 static struct nf_conntrack_expect
*
349 ovs_ct_expect_find(struct net
*net
, const struct nf_conntrack_zone
*zone
,
350 u16 proto
, const struct sk_buff
*skb
)
352 struct nf_conntrack_tuple tuple
;
354 if (!nf_ct_get_tuplepr(skb
, skb_network_offset(skb
), proto
, net
, &tuple
))
356 return __nf_ct_expect_find(net
, zone
, &tuple
);
359 /* Determine whether skb->nfct is equal to the result of conntrack lookup. */
360 static bool skb_nfct_cached(const struct net
*net
, const struct sk_buff
*skb
,
361 const struct ovs_conntrack_info
*info
)
363 enum ip_conntrack_info ctinfo
;
366 ct
= nf_ct_get(skb
, &ctinfo
);
369 if (!net_eq(net
, read_pnet(&ct
->ct_net
)))
371 if (!nf_ct_zone_equal_any(info
->ct
, nf_ct_zone(ct
)))
374 struct nf_conn_help
*help
;
376 help
= nf_ct_ext_find(ct
, NF_CT_EXT_HELPER
);
377 if (help
&& rcu_access_pointer(help
->helper
) != info
->helper
)
384 /* Pass 'skb' through conntrack in 'net', using zone configured in 'info', if
385 * not done already. Update key with new CT state.
386 * Note that if the packet is deemed invalid by conntrack, skb->nfct will be
387 * set to NULL and 0 will be returned.
389 static int __ovs_ct_lookup(struct net
*net
, struct sw_flow_key
*key
,
390 const struct ovs_conntrack_info
*info
,
393 /* If we are recirculating packets to match on conntrack fields and
394 * committing with a separate conntrack action, then we don't need to
395 * actually run the packet through conntrack twice unless it's for a
398 if (!skb_nfct_cached(net
, skb
, info
)) {
399 struct nf_conn
*tmpl
= info
->ct
;
401 /* Associate skb with specified zone. */
404 nf_conntrack_put(skb
->nfct
);
405 nf_conntrack_get(&tmpl
->ct_general
);
406 skb
->nfct
= &tmpl
->ct_general
;
407 skb
->nfctinfo
= IP_CT_NEW
;
410 if (nf_conntrack_in(net
, info
->family
, NF_INET_PRE_ROUTING
,
414 if (ovs_ct_helper(skb
, info
->family
) != NF_ACCEPT
) {
415 WARN_ONCE(1, "helper rejected packet");
420 ovs_ct_update_key(skb
, info
, key
, true);
425 /* Lookup connection and read fields into key. */
426 static int ovs_ct_lookup(struct net
*net
, struct sw_flow_key
*key
,
427 const struct ovs_conntrack_info
*info
,
430 struct nf_conntrack_expect
*exp
;
432 /* If we pass an expected packet through nf_conntrack_in() the
433 * expectation is typically removed, but the packet could still be
434 * lost in upcall processing. To prevent this from happening we
435 * perform an explicit expectation lookup. Expected connections are
436 * always new, and will be passed through conntrack only when they are
437 * committed, as it is OK to remove the expectation at that time.
439 exp
= ovs_ct_expect_find(net
, &info
->zone
, info
->family
, skb
);
443 state
= OVS_CS_F_TRACKED
| OVS_CS_F_NEW
| OVS_CS_F_RELATED
;
444 __ovs_ct_update_key(key
, state
, &info
->zone
, exp
->master
);
448 err
= __ovs_ct_lookup(net
, key
, info
, skb
);
456 /* Lookup connection and confirm if unconfirmed. */
457 static int ovs_ct_commit(struct net
*net
, struct sw_flow_key
*key
,
458 const struct ovs_conntrack_info
*info
,
464 state
= key
->ct
.state
;
465 if (key
->ct
.zone
== info
->zone
.id
&&
466 ((state
& OVS_CS_F_TRACKED
) && !(state
& OVS_CS_F_NEW
))) {
467 /* Previous lookup has shown that this connection is already
468 * tracked and committed. Skip committing.
473 err
= __ovs_ct_lookup(net
, key
, info
, skb
);
476 /* This is a no-op if the connection has already been confirmed. */
477 if (nf_conntrack_confirm(skb
) != NF_ACCEPT
)
483 static bool labels_nonzero(const struct ovs_key_ct_labels
*labels
)
487 for (i
= 0; i
< sizeof(*labels
); i
++)
488 if (labels
->ct_labels
[i
])
494 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
495 * value if 'skb' is freed.
497 int ovs_ct_execute(struct net
*net
, struct sk_buff
*skb
,
498 struct sw_flow_key
*key
,
499 const struct ovs_conntrack_info
*info
)
504 /* The conntrack module expects to be working at L3. */
505 nh_ofs
= skb_network_offset(skb
);
506 skb_pull(skb
, nh_ofs
);
508 if (key
->ip
.frag
!= OVS_FRAG_TYPE_NONE
) {
509 err
= handle_fragments(net
, key
, info
->zone
.id
, skb
);
515 err
= ovs_ct_commit(net
, key
, info
, skb
);
517 err
= ovs_ct_lookup(net
, key
, info
, skb
);
521 if (info
->mark
.mask
) {
522 err
= ovs_ct_set_mark(skb
, key
, info
->mark
.value
,
527 if (labels_nonzero(&info
->labels
.mask
))
528 err
= ovs_ct_set_labels(skb
, key
, &info
->labels
.value
,
531 skb_push(skb
, nh_ofs
);
537 static int ovs_ct_add_helper(struct ovs_conntrack_info
*info
, const char *name
,
538 const struct sw_flow_key
*key
, bool log
)
540 struct nf_conntrack_helper
*helper
;
541 struct nf_conn_help
*help
;
543 helper
= nf_conntrack_helper_try_module_get(name
, info
->family
,
546 OVS_NLERR(log
, "Unknown helper \"%s\"", name
);
550 help
= nf_ct_helper_ext_add(info
->ct
, helper
, GFP_KERNEL
);
552 module_put(helper
->me
);
556 rcu_assign_pointer(help
->helper
, helper
);
557 info
->helper
= helper
;
561 static const struct ovs_ct_len_tbl ovs_ct_attr_lens
[OVS_CT_ATTR_MAX
+ 1] = {
562 [OVS_CT_ATTR_COMMIT
] = { .minlen
= 0, .maxlen
= 0 },
563 [OVS_CT_ATTR_ZONE
] = { .minlen
= sizeof(u16
),
564 .maxlen
= sizeof(u16
) },
565 [OVS_CT_ATTR_MARK
] = { .minlen
= sizeof(struct md_mark
),
566 .maxlen
= sizeof(struct md_mark
) },
567 [OVS_CT_ATTR_LABELS
] = { .minlen
= sizeof(struct md_labels
),
568 .maxlen
= sizeof(struct md_labels
) },
569 [OVS_CT_ATTR_HELPER
] = { .minlen
= 1,
570 .maxlen
= NF_CT_HELPER_NAME_LEN
}
573 static int parse_ct(const struct nlattr
*attr
, struct ovs_conntrack_info
*info
,
574 const char **helper
, bool log
)
579 nla_for_each_nested(a
, attr
, rem
) {
580 int type
= nla_type(a
);
581 int maxlen
= ovs_ct_attr_lens
[type
].maxlen
;
582 int minlen
= ovs_ct_attr_lens
[type
].minlen
;
584 if (type
> OVS_CT_ATTR_MAX
) {
586 "Unknown conntrack attr (type=%d, max=%d)",
587 type
, OVS_CT_ATTR_MAX
);
590 if (nla_len(a
) < minlen
|| nla_len(a
) > maxlen
) {
592 "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
593 type
, nla_len(a
), maxlen
);
598 case OVS_CT_ATTR_COMMIT
:
601 #ifdef CONFIG_NF_CONNTRACK_ZONES
602 case OVS_CT_ATTR_ZONE
:
603 info
->zone
.id
= nla_get_u16(a
);
606 #ifdef CONFIG_NF_CONNTRACK_MARK
607 case OVS_CT_ATTR_MARK
: {
608 struct md_mark
*mark
= nla_data(a
);
611 OVS_NLERR(log
, "ct_mark mask cannot be 0");
618 #ifdef CONFIG_NF_CONNTRACK_LABELS
619 case OVS_CT_ATTR_LABELS
: {
620 struct md_labels
*labels
= nla_data(a
);
622 if (!labels_nonzero(&labels
->mask
)) {
623 OVS_NLERR(log
, "ct_labels mask cannot be 0");
626 info
->labels
= *labels
;
630 case OVS_CT_ATTR_HELPER
:
631 *helper
= nla_data(a
);
632 if (!memchr(*helper
, '\0', nla_len(a
))) {
633 OVS_NLERR(log
, "Invalid conntrack helper");
638 OVS_NLERR(log
, "Unknown conntrack attr (%d)",
645 OVS_NLERR(log
, "Conntrack attr has %d unknown bytes", rem
);
652 bool ovs_ct_verify(struct net
*net
, enum ovs_key_attr attr
)
654 if (attr
== OVS_KEY_ATTR_CT_STATE
)
656 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES
) &&
657 attr
== OVS_KEY_ATTR_CT_ZONE
)
659 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK
) &&
660 attr
== OVS_KEY_ATTR_CT_MARK
)
662 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS
) &&
663 attr
== OVS_KEY_ATTR_CT_LABELS
) {
664 struct ovs_net
*ovs_net
= net_generic(net
, ovs_net_id
);
666 return ovs_net
->xt_label
;
672 int ovs_ct_copy_action(struct net
*net
, const struct nlattr
*attr
,
673 const struct sw_flow_key
*key
,
674 struct sw_flow_actions
**sfa
, bool log
)
676 struct ovs_conntrack_info ct_info
;
677 const char *helper
= NULL
;
681 family
= key_to_nfproto(key
);
682 if (family
== NFPROTO_UNSPEC
) {
683 OVS_NLERR(log
, "ct family unspecified");
687 memset(&ct_info
, 0, sizeof(ct_info
));
688 ct_info
.family
= family
;
690 nf_ct_zone_init(&ct_info
.zone
, NF_CT_DEFAULT_ZONE_ID
,
691 NF_CT_DEFAULT_ZONE_DIR
, 0);
693 err
= parse_ct(attr
, &ct_info
, &helper
, log
);
697 /* Set up template for tracking connections in specific zones. */
698 ct_info
.ct
= nf_ct_tmpl_alloc(net
, &ct_info
.zone
, GFP_KERNEL
);
700 OVS_NLERR(log
, "Failed to allocate conntrack template");
704 __set_bit(IPS_CONFIRMED_BIT
, &ct_info
.ct
->status
);
705 nf_conntrack_get(&ct_info
.ct
->ct_general
);
708 err
= ovs_ct_add_helper(&ct_info
, helper
, key
, log
);
713 err
= ovs_nla_add_action(sfa
, OVS_ACTION_ATTR_CT
, &ct_info
,
714 sizeof(ct_info
), log
);
720 __ovs_ct_free_action(&ct_info
);
724 int ovs_ct_action_to_attr(const struct ovs_conntrack_info
*ct_info
,
727 struct nlattr
*start
;
729 start
= nla_nest_start(skb
, OVS_ACTION_ATTR_CT
);
733 if (ct_info
->commit
&& nla_put_flag(skb
, OVS_CT_ATTR_COMMIT
))
735 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES
) &&
736 nla_put_u16(skb
, OVS_CT_ATTR_ZONE
, ct_info
->zone
.id
))
738 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK
) && ct_info
->mark
.mask
&&
739 nla_put(skb
, OVS_CT_ATTR_MARK
, sizeof(ct_info
->mark
),
742 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS
) &&
743 labels_nonzero(&ct_info
->labels
.mask
) &&
744 nla_put(skb
, OVS_CT_ATTR_LABELS
, sizeof(ct_info
->labels
),
747 if (ct_info
->helper
) {
748 if (nla_put_string(skb
, OVS_CT_ATTR_HELPER
,
749 ct_info
->helper
->name
))
753 nla_nest_end(skb
, start
);
758 void ovs_ct_free_action(const struct nlattr
*a
)
760 struct ovs_conntrack_info
*ct_info
= nla_data(a
);
762 __ovs_ct_free_action(ct_info
);
765 static void __ovs_ct_free_action(struct ovs_conntrack_info
*ct_info
)
768 module_put(ct_info
->helper
->me
);
770 nf_ct_put(ct_info
->ct
);
773 void ovs_ct_init(struct net
*net
)
775 unsigned int n_bits
= sizeof(struct ovs_key_ct_labels
) * BITS_PER_BYTE
;
776 struct ovs_net
*ovs_net
= net_generic(net
, ovs_net_id
);
778 if (nf_connlabels_get(net
, n_bits
)) {
779 ovs_net
->xt_label
= false;
780 OVS_NLERR(true, "Failed to set connlabel length");
782 ovs_net
->xt_label
= true;
786 void ovs_ct_exit(struct net
*net
)
788 struct ovs_net
*ovs_net
= net_generic(net
, ovs_net_id
);
790 if (ovs_net
->xt_label
)
791 nf_connlabels_put(net
);