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 if (!nf_ct_is_confirmed(ct))
156 state |= OVS_CS_F_NEW;
158 state |= OVS_CS_F_RELATED;
159 zone = nf_ct_zone(ct);
160 } else if (post_ct) {
161 state = OVS_CS_F_TRACKED | OVS_CS_F_INVALID;
165 __ovs_ct_update_key(key, state, zone, ct);
168 void ovs_ct_fill_key(const struct sk_buff *skb, struct sw_flow_key *key)
170 ovs_ct_update_key(skb, NULL, key, false);
173 int ovs_ct_put_key(const struct sw_flow_key *key, struct sk_buff *skb)
175 if (nla_put_u32(skb, OVS_KEY_ATTR_CT_STATE, key->ct.state))
178 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
179 nla_put_u16(skb, OVS_KEY_ATTR_CT_ZONE, key->ct.zone))
182 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
183 nla_put_u32(skb, OVS_KEY_ATTR_CT_MARK, key->ct.mark))
186 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
187 nla_put(skb, OVS_KEY_ATTR_CT_LABELS, sizeof(key->ct.labels),
194 static int ovs_ct_set_mark(struct sk_buff *skb, struct sw_flow_key *key,
195 u32 ct_mark, u32 mask)
197 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
198 enum ip_conntrack_info ctinfo;
203 /* The connection could be invalid, in which case set_mark is no-op. */
204 ct = nf_ct_get(skb, &ctinfo);
208 new_mark = ct_mark | (ct->mark & ~(mask));
209 if (ct->mark != new_mark) {
211 nf_conntrack_event_cache(IPCT_MARK, ct);
212 key->ct.mark = new_mark;
221 static int ovs_ct_set_labels(struct sk_buff *skb, struct sw_flow_key *key,
222 const struct ovs_key_ct_labels *labels,
223 const struct ovs_key_ct_labels *mask)
225 enum ip_conntrack_info ctinfo;
226 struct nf_conn_labels *cl;
230 /* The connection could be invalid, in which case set_label is no-op.*/
231 ct = nf_ct_get(skb, &ctinfo);
235 cl = nf_ct_labels_find(ct);
237 nf_ct_labels_ext_add(ct);
238 cl = nf_ct_labels_find(ct);
240 if (!cl || cl->words * sizeof(long) < OVS_CT_LABELS_LEN)
243 err = nf_connlabels_replace(ct, (u32 *)labels, (u32 *)mask,
244 OVS_CT_LABELS_LEN / sizeof(u32));
248 ovs_ct_get_labels(ct, &key->ct.labels);
252 /* 'skb' should already be pulled to nh_ofs. */
253 static int ovs_ct_helper(struct sk_buff *skb, u16 proto)
255 const struct nf_conntrack_helper *helper;
256 const struct nf_conn_help *help;
257 enum ip_conntrack_info ctinfo;
258 unsigned int protoff;
261 ct = nf_ct_get(skb, &ctinfo);
262 if (!ct || ctinfo == IP_CT_RELATED_REPLY)
265 help = nfct_help(ct);
269 helper = rcu_dereference(help->helper);
275 protoff = ip_hdrlen(skb);
278 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
282 ofs = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr,
284 if (ofs < 0 || (frag_off & htons(~0x7)) != 0) {
285 pr_debug("proto header not found\n");
292 WARN_ONCE(1, "helper invoked on non-IP family!");
296 return helper->help(skb, protoff, ct, ctinfo);
299 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
300 * value if 'skb' is freed.
302 static int handle_fragments(struct net *net, struct sw_flow_key *key,
303 u16 zone, struct sk_buff *skb)
305 struct ovs_skb_cb ovs_cb = *OVS_CB(skb);
308 if (key->eth.type == htons(ETH_P_IP)) {
309 enum ip_defrag_users user = IP_DEFRAG_CONNTRACK_IN + zone;
311 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
312 err = ip_defrag(net, skb, user);
316 ovs_cb.mru = IPCB(skb)->frag_max_size;
317 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
318 } else if (key->eth.type == htons(ETH_P_IPV6)) {
319 enum ip6_defrag_users user = IP6_DEFRAG_CONNTRACK_IN + zone;
321 memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm));
322 err = nf_ct_frag6_gather(net, skb, user);
326 key->ip.proto = ipv6_hdr(skb)->nexthdr;
327 ovs_cb.mru = IP6CB(skb)->frag_max_size;
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, net, &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, 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");
409 ovs_ct_update_key(skb, info, key, true);
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);
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)
464 static bool labels_nonzero(const struct ovs_key_ct_labels *labels)
468 for (i = 0; i < sizeof(*labels); i++)
469 if (labels->ct_labels[i])
475 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
476 * value if 'skb' is freed.
478 int ovs_ct_execute(struct net *net, struct sk_buff *skb,
479 struct sw_flow_key *key,
480 const struct ovs_conntrack_info *info)
485 /* The conntrack module expects to be working at L3. */
486 nh_ofs = skb_network_offset(skb);
487 skb_pull(skb, nh_ofs);
489 if (key->ip.frag != OVS_FRAG_TYPE_NONE) {
490 err = handle_fragments(net, key, info->zone.id, skb);
496 err = ovs_ct_commit(net, key, info, skb);
498 err = ovs_ct_lookup(net, key, info, skb);
502 if (info->mark.mask) {
503 err = ovs_ct_set_mark(skb, key, info->mark.value,
508 if (labels_nonzero(&info->labels.mask))
509 err = ovs_ct_set_labels(skb, key, &info->labels.value,
512 skb_push(skb, nh_ofs);
518 static int ovs_ct_add_helper(struct ovs_conntrack_info *info, const char *name,
519 const struct sw_flow_key *key, bool log)
521 struct nf_conntrack_helper *helper;
522 struct nf_conn_help *help;
524 helper = nf_conntrack_helper_try_module_get(name, info->family,
527 OVS_NLERR(log, "Unknown helper \"%s\"", name);
531 help = nf_ct_helper_ext_add(info->ct, helper, GFP_KERNEL);
533 module_put(helper->me);
537 rcu_assign_pointer(help->helper, helper);
538 info->helper = helper;
542 static const struct ovs_ct_len_tbl ovs_ct_attr_lens[OVS_CT_ATTR_MAX + 1] = {
543 [OVS_CT_ATTR_COMMIT] = { .minlen = 0, .maxlen = 0 },
544 [OVS_CT_ATTR_ZONE] = { .minlen = sizeof(u16),
545 .maxlen = sizeof(u16) },
546 [OVS_CT_ATTR_MARK] = { .minlen = sizeof(struct md_mark),
547 .maxlen = sizeof(struct md_mark) },
548 [OVS_CT_ATTR_LABELS] = { .minlen = sizeof(struct md_labels),
549 .maxlen = sizeof(struct md_labels) },
550 [OVS_CT_ATTR_HELPER] = { .minlen = 1,
551 .maxlen = NF_CT_HELPER_NAME_LEN }
554 static int parse_ct(const struct nlattr *attr, struct ovs_conntrack_info *info,
555 const char **helper, bool log)
560 nla_for_each_nested(a, attr, rem) {
561 int type = nla_type(a);
562 int maxlen = ovs_ct_attr_lens[type].maxlen;
563 int minlen = ovs_ct_attr_lens[type].minlen;
565 if (type > OVS_CT_ATTR_MAX) {
567 "Unknown conntrack attr (type=%d, max=%d)",
568 type, OVS_CT_ATTR_MAX);
571 if (nla_len(a) < minlen || nla_len(a) > maxlen) {
573 "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
574 type, nla_len(a), maxlen);
579 case OVS_CT_ATTR_COMMIT:
582 #ifdef CONFIG_NF_CONNTRACK_ZONES
583 case OVS_CT_ATTR_ZONE:
584 info->zone.id = nla_get_u16(a);
587 #ifdef CONFIG_NF_CONNTRACK_MARK
588 case OVS_CT_ATTR_MARK: {
589 struct md_mark *mark = nla_data(a);
592 OVS_NLERR(log, "ct_mark mask cannot be 0");
599 #ifdef CONFIG_NF_CONNTRACK_LABELS
600 case OVS_CT_ATTR_LABELS: {
601 struct md_labels *labels = nla_data(a);
603 if (!labels_nonzero(&labels->mask)) {
604 OVS_NLERR(log, "ct_labels mask cannot be 0");
607 info->labels = *labels;
611 case OVS_CT_ATTR_HELPER:
612 *helper = nla_data(a);
613 if (!memchr(*helper, '\0', nla_len(a))) {
614 OVS_NLERR(log, "Invalid conntrack helper");
619 OVS_NLERR(log, "Unknown conntrack attr (%d)",
626 OVS_NLERR(log, "Conntrack attr has %d unknown bytes", rem);
633 bool ovs_ct_verify(struct net *net, enum ovs_key_attr attr)
635 if (attr == OVS_KEY_ATTR_CT_STATE)
637 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
638 attr == OVS_KEY_ATTR_CT_ZONE)
640 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
641 attr == OVS_KEY_ATTR_CT_MARK)
643 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
644 attr == OVS_KEY_ATTR_CT_LABELS) {
645 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
647 return ovs_net->xt_label;
653 int ovs_ct_copy_action(struct net *net, const struct nlattr *attr,
654 const struct sw_flow_key *key,
655 struct sw_flow_actions **sfa, bool log)
657 struct ovs_conntrack_info ct_info;
658 const char *helper = NULL;
662 family = key_to_nfproto(key);
663 if (family == NFPROTO_UNSPEC) {
664 OVS_NLERR(log, "ct family unspecified");
668 memset(&ct_info, 0, sizeof(ct_info));
669 ct_info.family = family;
671 nf_ct_zone_init(&ct_info.zone, NF_CT_DEFAULT_ZONE_ID,
672 NF_CT_DEFAULT_ZONE_DIR, 0);
674 err = parse_ct(attr, &ct_info, &helper, log);
678 /* Set up template for tracking connections in specific zones. */
679 ct_info.ct = nf_ct_tmpl_alloc(net, &ct_info.zone, GFP_KERNEL);
681 OVS_NLERR(log, "Failed to allocate conntrack template");
685 __set_bit(IPS_CONFIRMED_BIT, &ct_info.ct->status);
686 nf_conntrack_get(&ct_info.ct->ct_general);
689 err = ovs_ct_add_helper(&ct_info, helper, key, log);
694 err = ovs_nla_add_action(sfa, OVS_ACTION_ATTR_CT, &ct_info,
695 sizeof(ct_info), log);
701 __ovs_ct_free_action(&ct_info);
705 int ovs_ct_action_to_attr(const struct ovs_conntrack_info *ct_info,
708 struct nlattr *start;
710 start = nla_nest_start(skb, OVS_ACTION_ATTR_CT);
714 if (ct_info->commit && nla_put_flag(skb, OVS_CT_ATTR_COMMIT))
716 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
717 nla_put_u16(skb, OVS_CT_ATTR_ZONE, ct_info->zone.id))
719 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && ct_info->mark.mask &&
720 nla_put(skb, OVS_CT_ATTR_MARK, sizeof(ct_info->mark),
723 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
724 labels_nonzero(&ct_info->labels.mask) &&
725 nla_put(skb, OVS_CT_ATTR_LABELS, sizeof(ct_info->labels),
728 if (ct_info->helper) {
729 if (nla_put_string(skb, OVS_CT_ATTR_HELPER,
730 ct_info->helper->name))
734 nla_nest_end(skb, start);
739 void ovs_ct_free_action(const struct nlattr *a)
741 struct ovs_conntrack_info *ct_info = nla_data(a);
743 __ovs_ct_free_action(ct_info);
746 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info)
749 module_put(ct_info->helper->me);
751 nf_ct_put(ct_info->ct);
754 void ovs_ct_init(struct net *net)
756 unsigned int n_bits = sizeof(struct ovs_key_ct_labels) * BITS_PER_BYTE;
757 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
759 if (nf_connlabels_get(net, n_bits)) {
760 ovs_net->xt_label = false;
761 OVS_NLERR(true, "Failed to set connlabel length");
763 ovs_net->xt_label = true;
767 void ovs_ct_exit(struct net *net)
769 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
771 if (ovs_net->xt_label)
772 nf_connlabels_put(net);