projects/linux·blob master · refs

2080 lines · 53.7 KiB

1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright (c) 2015 Nicira, Inc.
4 */
5
6#include <linux/module.h>
7#include <linux/openvswitch.h>
8#include <linux/tcp.h>
9#include <linux/udp.h>
10#include <linux/sctp.h>
11#include <linux/static_key.h>
12#include <linux/string_helpers.h>
13#include <net/ip.h>
14#include <net/genetlink.h>
15#include <net/netfilter/nf_conntrack_core.h>
16#include <net/netfilter/nf_conntrack_count.h>
17#include <net/netfilter/nf_conntrack_helper.h>
18#include <net/netfilter/nf_conntrack_labels.h>
19#include <net/netfilter/nf_conntrack_seqadj.h>
20#include <net/netfilter/nf_conntrack_timeout.h>
21#include <net/netfilter/nf_conntrack_zones.h>
22#include <net/netfilter/ipv6/nf_defrag_ipv6.h>
23#include <net/ipv6_frag.h>
24
25#if IS_ENABLED(CONFIG_NF_NAT)
26#include <net/netfilter/nf_nat.h>
27#endif
28
29#include <net/netfilter/nf_conntrack_act_ct.h>
30
31#include "datapath.h"
32#include "drop.h"
33#include "conntrack.h"
34#include "flow.h"
35#include "flow_netlink.h"
36
37struct ovs_ct_len_tbl {
38 int maxlen;
39 int minlen;
40};
41
42/* Metadata mark for masked write to conntrack mark */
43struct md_mark {
44 u32 value;
45 u32 mask;
46};
47
48/* Metadata label for masked write to conntrack label. */
49struct md_labels {
50 struct ovs_key_ct_labels value;
51 struct ovs_key_ct_labels mask;
52};
53
54enum ovs_ct_nat {
55 OVS_CT_NAT = 1 << 0, /* NAT for committed connections only. */
56 OVS_CT_SRC_NAT = 1 << 1, /* Source NAT for NEW connections. */
57 OVS_CT_DST_NAT = 1 << 2, /* Destination NAT for NEW connections. */
58};
59
60/* Conntrack action context for execution. */
61struct ovs_conntrack_info {
62 struct nf_conntrack_helper *helper;
63 struct nf_conntrack_zone zone;
64 struct nf_conn *ct;
65 u8 commit : 1;
66 u8 nat : 3; /* enum ovs_ct_nat */
67 u8 force : 1;
68 u8 have_eventmask : 1;
69 u16 family;
70 u32 eventmask; /* Mask of 1 << IPCT_*. */
71 struct md_mark mark;
72 struct md_labels labels;
73 char timeout[CTNL_TIMEOUT_NAME_MAX];
74 struct nf_ct_timeout *nf_ct_timeout;
75#if IS_ENABLED(CONFIG_NF_NAT)
76 struct nf_nat_range2 range; /* Only present for SRC NAT and DST NAT. */
77#endif
78};
79
80#if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
81#define OVS_CT_LIMIT_UNLIMITED 0
82#define OVS_CT_LIMIT_DEFAULT OVS_CT_LIMIT_UNLIMITED
83#define CT_LIMIT_HASH_BUCKETS 512
84static DEFINE_STATIC_KEY_FALSE(ovs_ct_limit_enabled);
85
86struct ovs_ct_limit {
87 /* Elements in ovs_ct_limit_info->limits hash table */
88 struct hlist_node hlist_node;
89 struct rcu_head rcu;
90 u16 zone;
91 u32 limit;
92};
93
94struct ovs_ct_limit_info {
95 u32 default_limit;
96 struct hlist_head *limits;
97 struct nf_conncount_data *data;
98};
99
100static const struct nla_policy ct_limit_policy[OVS_CT_LIMIT_ATTR_MAX + 1] = {
101 [OVS_CT_LIMIT_ATTR_ZONE_LIMIT] = { .type = NLA_NESTED, },
102};
103#endif
104
105static bool labels_nonzero(const struct ovs_key_ct_labels *labels);
106
107static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info);
108
109static u16 key_to_nfproto(const struct sw_flow_key *key)
110{
111 switch (ntohs(key->eth.type)) {
112 case ETH_P_IP:
113 return NFPROTO_IPV4;
114 case ETH_P_IPV6:
115 return NFPROTO_IPV6;
116 default:
117 return NFPROTO_UNSPEC;
118 }
119}
120
121/* Map SKB connection state into the values used by flow definition. */
122static u8 ovs_ct_get_state(enum ip_conntrack_info ctinfo)
123{
124 u8 ct_state = OVS_CS_F_TRACKED;
125
126 switch (ctinfo) {
127 case IP_CT_ESTABLISHED_REPLY:
128 case IP_CT_RELATED_REPLY:
129 ct_state |= OVS_CS_F_REPLY_DIR;
130 break;
131 default:
132 break;
133 }
134
135 switch (ctinfo) {
136 case IP_CT_ESTABLISHED:
137 case IP_CT_ESTABLISHED_REPLY:
138 ct_state |= OVS_CS_F_ESTABLISHED;
139 break;
140 case IP_CT_RELATED:
141 case IP_CT_RELATED_REPLY:
142 ct_state |= OVS_CS_F_RELATED;
143 break;
144 case IP_CT_NEW:
145 ct_state |= OVS_CS_F_NEW;
146 break;
147 default:
148 break;
149 }
150
151 return ct_state;
152}
153
154static u32 ovs_ct_get_mark(const struct nf_conn *ct)
155{
156#if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
157 return ct ? READ_ONCE(ct->mark) : 0;
158#else
159 return 0;
160#endif
161}
162
163/* Guard against conntrack labels max size shrinking below 128 bits. */
164#if NF_CT_LABELS_MAX_SIZE < 16
165#error NF_CT_LABELS_MAX_SIZE must be at least 16 bytes
166#endif
167
168static void ovs_ct_get_labels(const struct nf_conn *ct,
169 struct ovs_key_ct_labels *labels)
170{
171 struct nf_conn_labels *cl = NULL;
172
173 if (ct) {
174 if (ct->master && !nf_ct_is_confirmed(ct))
175 ct = ct->master;
176 cl = nf_ct_labels_find(ct);
177 }
178 if (cl)
179 memcpy(labels, cl->bits, OVS_CT_LABELS_LEN);
180 else
181 memset(labels, 0, OVS_CT_LABELS_LEN);
182}
183
184static void __ovs_ct_update_key_orig_tp(struct sw_flow_key *key,
185 const struct nf_conntrack_tuple *orig,
186 u8 icmp_proto)
187{
188 key->ct_orig_proto = orig->dst.protonum;
189 if (orig->dst.protonum == icmp_proto) {
190 key->ct.orig_tp.src = htons(orig->dst.u.icmp.type);
191 key->ct.orig_tp.dst = htons(orig->dst.u.icmp.code);
192 } else {
193 key->ct.orig_tp.src = orig->src.u.all;
194 key->ct.orig_tp.dst = orig->dst.u.all;
195 }
196}
197
198static void __ovs_ct_update_key(struct sw_flow_key *key, u8 state,
199 const struct nf_conntrack_zone *zone,
200 const struct nf_conn *ct)
201{
202 key->ct_state = state;
203 key->ct_zone = zone->id;
204 key->ct.mark = ovs_ct_get_mark(ct);
205 ovs_ct_get_labels(ct, &key->ct.labels);
206
207 if (ct) {
208 const struct nf_conntrack_tuple *orig;
209
210 /* Use the master if we have one. */
211 if (ct->master)
212 ct = ct->master;
213 orig = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
214
215 /* IP version must match with the master connection. */
216 if (key->eth.type == htons(ETH_P_IP) &&
217 nf_ct_l3num(ct) == NFPROTO_IPV4) {
218 key->ipv4.ct_orig.src = orig->src.u3.ip;
219 key->ipv4.ct_orig.dst = orig->dst.u3.ip;
220 __ovs_ct_update_key_orig_tp(key, orig, IPPROTO_ICMP);
221 return;
222 } else if (key->eth.type == htons(ETH_P_IPV6) &&
223 !sw_flow_key_is_nd(key) &&
224 nf_ct_l3num(ct) == NFPROTO_IPV6) {
225 key->ipv6.ct_orig.src = orig->src.u3.in6;
226 key->ipv6.ct_orig.dst = orig->dst.u3.in6;
227 __ovs_ct_update_key_orig_tp(key, orig, NEXTHDR_ICMP);
228 return;
229 }
230 }
231 /* Clear 'ct_orig_proto' to mark the non-existence of conntrack
232 * original direction key fields.
233 */
234 key->ct_orig_proto = 0;
235}
236
237/* Update 'key' based on skb->_nfct. If 'post_ct' is true, then OVS has
238 * previously sent the packet to conntrack via the ct action. If
239 * 'keep_nat_flags' is true, the existing NAT flags retained, else they are
240 * initialized from the connection status.
241 */
242static void ovs_ct_update_key(const struct sk_buff *skb,
243 const struct ovs_conntrack_info *info,
244 struct sw_flow_key *key, bool post_ct,
245 bool keep_nat_flags)
246{
247 const struct nf_conntrack_zone *zone = &nf_ct_zone_dflt;
248 enum ip_conntrack_info ctinfo;
249 struct nf_conn *ct;
250 u8 state = 0;
251
252 ct = nf_ct_get(skb, &ctinfo);
253 if (ct) {
254 state = ovs_ct_get_state(ctinfo);
255 /* All unconfirmed entries are NEW connections. */
256 if (!nf_ct_is_confirmed(ct))
257 state |= OVS_CS_F_NEW;
258 /* OVS persists the related flag for the duration of the
259 * connection.
260 */
261 if (ct->master)
262 state |= OVS_CS_F_RELATED;
263 if (keep_nat_flags) {
264 state |= key->ct_state & OVS_CS_F_NAT_MASK;
265 } else {
266 if (ct->status & IPS_SRC_NAT)
267 state |= OVS_CS_F_SRC_NAT;
268 if (ct->status & IPS_DST_NAT)
269 state |= OVS_CS_F_DST_NAT;
270 }
271 zone = nf_ct_zone(ct);
272 } else if (post_ct) {
273 state = OVS_CS_F_TRACKED | OVS_CS_F_INVALID;
274 if (info)
275 zone = &info->zone;
276 }
277 __ovs_ct_update_key(key, state, zone, ct);
278}
279
280/* This is called to initialize CT key fields possibly coming in from the local
281 * stack.
282 */
283void ovs_ct_fill_key(const struct sk_buff *skb,
284 struct sw_flow_key *key,
285 bool post_ct)
286{
287 ovs_ct_update_key(skb, NULL, key, post_ct, false);
288}
289
290int ovs_ct_put_key(const struct sw_flow_key *swkey,
291 const struct sw_flow_key *output, struct sk_buff *skb)
292{
293 if (nla_put_u32(skb, OVS_KEY_ATTR_CT_STATE, output->ct_state))
294 return -EMSGSIZE;
295
296 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
297 nla_put_u16(skb, OVS_KEY_ATTR_CT_ZONE, output->ct_zone))
298 return -EMSGSIZE;
299
300 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
301 nla_put_u32(skb, OVS_KEY_ATTR_CT_MARK, output->ct.mark))
302 return -EMSGSIZE;
303
304 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
305 nla_put(skb, OVS_KEY_ATTR_CT_LABELS, sizeof(output->ct.labels),
306 &output->ct.labels))
307 return -EMSGSIZE;
308
309 if (swkey->ct_orig_proto) {
310 if (swkey->eth.type == htons(ETH_P_IP)) {
311 struct ovs_key_ct_tuple_ipv4 orig;
312
313 memset(&orig, 0, sizeof(orig));
314 orig.ipv4_src = output->ipv4.ct_orig.src;
315 orig.ipv4_dst = output->ipv4.ct_orig.dst;
316 orig.src_port = output->ct.orig_tp.src;
317 orig.dst_port = output->ct.orig_tp.dst;
318 orig.ipv4_proto = output->ct_orig_proto;
319
320 if (nla_put(skb, OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4,
321 sizeof(orig), &orig))
322 return -EMSGSIZE;
323 } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
324 struct ovs_key_ct_tuple_ipv6 orig;
325
326 memset(&orig, 0, sizeof(orig));
327 memcpy(orig.ipv6_src, output->ipv6.ct_orig.src.s6_addr32,
328 sizeof(orig.ipv6_src));
329 memcpy(orig.ipv6_dst, output->ipv6.ct_orig.dst.s6_addr32,
330 sizeof(orig.ipv6_dst));
331 orig.src_port = output->ct.orig_tp.src;
332 orig.dst_port = output->ct.orig_tp.dst;
333 orig.ipv6_proto = output->ct_orig_proto;
334
335 if (nla_put(skb, OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6,
336 sizeof(orig), &orig))
337 return -EMSGSIZE;
338 }
339 }
340
341 return 0;
342}
343
344static int ovs_ct_set_mark(struct nf_conn *ct, struct sw_flow_key *key,
345 u32 ct_mark, u32 mask)
346{
347#if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
348 u32 new_mark;
349
350 new_mark = ct_mark | (READ_ONCE(ct->mark) & ~(mask));
351 if (READ_ONCE(ct->mark) != new_mark) {
352 WRITE_ONCE(ct->mark, new_mark);
353 if (nf_ct_is_confirmed(ct))
354 nf_conntrack_event_cache(IPCT_MARK, ct);
355 key->ct.mark = new_mark;
356 }
357
358 return 0;
359#else
360 return -ENOTSUPP;
361#endif
362}
363
364static struct nf_conn_labels *ovs_ct_get_conn_labels(struct nf_conn *ct)
365{
366 struct nf_conn_labels *cl;
367
368 cl = nf_ct_labels_find(ct);
369 if (!cl && !nf_ct_is_confirmed(ct)) {
370 nf_ct_labels_ext_add(ct);
371 cl = nf_ct_labels_find(ct);
372 }
373
374 return cl;
375}
376
377/* Initialize labels for a new, yet to be committed conntrack entry. Note that
378 * since the new connection is not yet confirmed, and thus no-one else has
379 * access to it's labels, we simply write them over.
380 */
381static int ovs_ct_init_labels(struct nf_conn *ct, struct sw_flow_key *key,
382 const struct ovs_key_ct_labels *labels,
383 const struct ovs_key_ct_labels *mask)
384{
385 struct nf_conn_labels *cl, *master_cl;
386 bool have_mask = labels_nonzero(mask);
387
388 /* Inherit master's labels to the related connection? */
389 master_cl = ct->master ? nf_ct_labels_find(ct->master) : NULL;
390
391 if (!master_cl && !have_mask)
392 return 0; /* Nothing to do. */
393
394 cl = ovs_ct_get_conn_labels(ct);
395 if (!cl)
396 return -ENOSPC;
397
398 /* Inherit the master's labels, if any. */
399 if (master_cl)
400 *cl = *master_cl;
401
402 if (have_mask) {
403 u32 *dst = (u32 *)cl->bits;
404 int i;
405
406 for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
407 dst[i] = (dst[i] & ~mask->ct_labels_32[i]) |
408 (labels->ct_labels_32[i]
409 & mask->ct_labels_32[i]);
410 }
411
412 /* Labels are included in the IPCTNL_MSG_CT_NEW event only if the
413 * IPCT_LABEL bit is set in the event cache.
414 */
415 nf_conntrack_event_cache(IPCT_LABEL, ct);
416
417 memcpy(&key->ct.labels, cl->bits, OVS_CT_LABELS_LEN);
418
419 return 0;
420}
421
422static int ovs_ct_set_labels(struct nf_conn *ct, struct sw_flow_key *key,
423 const struct ovs_key_ct_labels *labels,
424 const struct ovs_key_ct_labels *mask)
425{
426 struct nf_conn_labels *cl;
427 int err;
428
429 cl = ovs_ct_get_conn_labels(ct);
430 if (!cl)
431 return -ENOSPC;
432
433 err = nf_connlabels_replace(ct, labels->ct_labels_32,
434 mask->ct_labels_32,
435 OVS_CT_LABELS_LEN_32);
436 if (err)
437 return err;
438
439 memcpy(&key->ct.labels, cl->bits, OVS_CT_LABELS_LEN);
440
441 return 0;
442}
443
444static int ovs_ct_handle_fragments(struct net *net, struct sw_flow_key *key,
445 u16 zone, int family, struct sk_buff *skb)
446{
447 struct ovs_skb_cb ovs_cb = *OVS_CB(skb);
448 int err;
449
450 err = nf_ct_handle_fragments(net, skb, zone, family, &key->ip.proto, &ovs_cb.mru);
451 if (err)
452 return err;
453
454 /* The key extracted from the fragment that completed this datagram
455 * likely didn't have an L4 header, so regenerate it.
456 */
457 ovs_flow_key_update_l3l4(skb, key);
458 key->ip.frag = OVS_FRAG_TYPE_NONE;
459 *OVS_CB(skb) = ovs_cb;
460
461 return 0;
462}
463
464/* This replicates logic from nf_conntrack_core.c that is not exported. */
465static enum ip_conntrack_info
466ovs_ct_get_info(const struct nf_conntrack_tuple_hash *h)
467{
468 const struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
469
470 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY)
471 return IP_CT_ESTABLISHED_REPLY;
472 /* Once we've had two way comms, always ESTABLISHED. */
473 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status))
474 return IP_CT_ESTABLISHED;
475 if (test_bit(IPS_EXPECTED_BIT, &ct->status))
476 return IP_CT_RELATED;
477 return IP_CT_NEW;
478}
479
480/* Find an existing connection which this packet belongs to without
481 * re-attributing statistics or modifying the connection state. This allows an
482 * skb->_nfct lost due to an upcall to be recovered during actions execution.
483 *
484 * Must be called with rcu_read_lock.
485 *
486 * On success, populates skb->_nfct and returns the connection. Returns NULL
487 * if there is no existing entry.
488 */
489static struct nf_conn *
490ovs_ct_find_existing(struct net *net, const struct nf_conntrack_zone *zone,
491 u8 l3num, struct sk_buff *skb, bool natted)
492{
493 struct nf_conntrack_tuple tuple;
494 struct nf_conntrack_tuple_hash *h;
495 struct nf_conn *ct;
496
497 if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), l3num,
498 net, &tuple)) {
499 pr_debug("ovs_ct_find_existing: Can't get tuple\n");
500 return NULL;
501 }
502
503 /* Must invert the tuple if skb has been transformed by NAT. */
504 if (natted) {
505 struct nf_conntrack_tuple inverse;
506
507 if (!nf_ct_invert_tuple(&inverse, &tuple)) {
508 pr_debug("ovs_ct_find_existing: Inversion failed!\n");
509 return NULL;
510 }
511 tuple = inverse;
512 }
513
514 /* look for tuple match */
515 h = nf_conntrack_find_get(net, zone, &tuple);
516 if (!h)
517 return NULL; /* Not found. */
518
519 ct = nf_ct_tuplehash_to_ctrack(h);
520
521 /* Inverted packet tuple matches the reverse direction conntrack tuple,
522 * select the other tuplehash to get the right 'ctinfo' bits for this
523 * packet.
524 */
525 if (natted)
526 h = &ct->tuplehash[!h->tuple.dst.dir];
527
528 nf_ct_set(skb, ct, ovs_ct_get_info(h));
529 return ct;
530}
531
532static
533struct nf_conn *ovs_ct_executed(struct net *net,
534 const struct sw_flow_key *key,
535 const struct ovs_conntrack_info *info,
536 struct sk_buff *skb,
537 bool *ct_executed)
538{
539 struct nf_conn *ct = NULL;
540
541 /* If no ct, check if we have evidence that an existing conntrack entry
542 * might be found for this skb. This happens when we lose a skb->_nfct
543 * due to an upcall, or if the direction is being forced. If the
544 * connection was not confirmed, it is not cached and needs to be run
545 * through conntrack again.
546 */
547 *ct_executed = (key->ct_state & OVS_CS_F_TRACKED) &&
548 !(key->ct_state & OVS_CS_F_INVALID) &&
549 (key->ct_zone == info->zone.id);
550
551 if (*ct_executed || (!key->ct_state && info->force)) {
552 ct = ovs_ct_find_existing(net, &info->zone, info->family, skb,
553 !!(key->ct_state &
554 OVS_CS_F_NAT_MASK));
555 }
556
557 return ct;
558}
559
560/* Determine whether skb->_nfct is equal to the result of conntrack lookup. */
561static bool skb_nfct_cached(struct net *net,
562 const struct sw_flow_key *key,
563 const struct ovs_conntrack_info *info,
564 struct sk_buff *skb)
565{
566 enum ip_conntrack_info ctinfo;
567 struct nf_conn *ct;
568 bool ct_executed = true;
569
570 ct = nf_ct_get(skb, &ctinfo);
571 if (!ct)
572 ct = ovs_ct_executed(net, key, info, skb, &ct_executed);
573
574 if (ct)
575 nf_ct_get(skb, &ctinfo);
576 else
577 return false;
578
579 if (!net_eq(net, read_pnet(&ct->ct_net)))
580 return false;
581 if (!nf_ct_zone_equal_any(info->ct, nf_ct_zone(ct)))
582 return false;
583 if (info->helper) {
584 struct nf_conn_help *help;
585
586 help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER);
587 if (help && rcu_access_pointer(help->helper) != info->helper)
588 return false;
589 }
590 if (info->nf_ct_timeout) {
591 struct nf_conn_timeout *timeout_ext;
592
593 timeout_ext = nf_ct_timeout_find(ct);
594 if (!timeout_ext || info->nf_ct_timeout !=
595 rcu_dereference(timeout_ext->timeout))
596 return false;
597 }
598 /* Force conntrack entry direction to the current packet? */
599 if (info->force && CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) {
600 /* Delete the conntrack entry if confirmed, else just release
601 * the reference.
602 */
603 if (nf_ct_is_confirmed(ct))
604 nf_ct_delete(ct, 0, 0);
605
606 nf_reset_ct(skb);
607 nf_ct_set(skb, NULL, 0);
608 return false;
609 }
610
611 return ct_executed;
612}
613
614#if IS_ENABLED(CONFIG_NF_NAT)
615static void ovs_nat_update_key(struct sw_flow_key *key,
616 const struct sk_buff *skb,
617 enum nf_nat_manip_type maniptype)
618{
619 if (maniptype == NF_NAT_MANIP_SRC) {
620 __be16 src;
621
622 key->ct_state |= OVS_CS_F_SRC_NAT;
623 if (key->eth.type == htons(ETH_P_IP))
624 key->ipv4.addr.src = ip_hdr(skb)->saddr;
625 else if (key->eth.type == htons(ETH_P_IPV6))
626 memcpy(&key->ipv6.addr.src, &ipv6_hdr(skb)->saddr,
627 sizeof(key->ipv6.addr.src));
628 else
629 return;
630
631 if (key->ip.proto == IPPROTO_UDP)
632 src = udp_hdr(skb)->source;
633 else if (key->ip.proto == IPPROTO_TCP)
634 src = tcp_hdr(skb)->source;
635 else if (key->ip.proto == IPPROTO_SCTP)
636 src = sctp_hdr(skb)->source;
637 else
638 return;
639
640 key->tp.src = src;
641 } else {
642 __be16 dst;
643
644 key->ct_state |= OVS_CS_F_DST_NAT;
645 if (key->eth.type == htons(ETH_P_IP))
646 key->ipv4.addr.dst = ip_hdr(skb)->daddr;
647 else if (key->eth.type == htons(ETH_P_IPV6))
648 memcpy(&key->ipv6.addr.dst, &ipv6_hdr(skb)->daddr,
649 sizeof(key->ipv6.addr.dst));
650 else
651 return;
652
653 if (key->ip.proto == IPPROTO_UDP)
654 dst = udp_hdr(skb)->dest;
655 else if (key->ip.proto == IPPROTO_TCP)
656 dst = tcp_hdr(skb)->dest;
657 else if (key->ip.proto == IPPROTO_SCTP)
658 dst = sctp_hdr(skb)->dest;
659 else
660 return;
661
662 key->tp.dst = dst;
663 }
664}
665
666/* Returns NF_DROP if the packet should be dropped, NF_ACCEPT otherwise. */
667static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
668 const struct ovs_conntrack_info *info,
669 struct sk_buff *skb, struct nf_conn *ct,
670 enum ip_conntrack_info ctinfo)
671{
672 int err, action = 0;
673
674 if (!(info->nat & OVS_CT_NAT))
675 return NF_ACCEPT;
676 if (info->nat & OVS_CT_SRC_NAT)
677 action |= BIT(NF_NAT_MANIP_SRC);
678 if (info->nat & OVS_CT_DST_NAT)
679 action |= BIT(NF_NAT_MANIP_DST);
680
681 err = nf_ct_nat(skb, ct, ctinfo, &action, &info->range, info->commit);
682 if (err != NF_ACCEPT)
683 return err;
684
685 if (action & BIT(NF_NAT_MANIP_SRC))
686 ovs_nat_update_key(key, skb, NF_NAT_MANIP_SRC);
687 if (action & BIT(NF_NAT_MANIP_DST))
688 ovs_nat_update_key(key, skb, NF_NAT_MANIP_DST);
689
690 return err;
691}
692#else /* !CONFIG_NF_NAT */
693static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
694 const struct ovs_conntrack_info *info,
695 struct sk_buff *skb, struct nf_conn *ct,
696 enum ip_conntrack_info ctinfo)
697{
698 return NF_ACCEPT;
699}
700#endif
701
702static int verdict_to_errno(unsigned int verdict)
703{
704 switch (verdict & NF_VERDICT_MASK) {
705 case NF_ACCEPT:
706 return 0;
707 case NF_DROP:
708 return -EINVAL;
709 case NF_STOLEN:
710 return -EINPROGRESS;
711 default:
712 break;
713 }
714
715 return -EINVAL;
716}
717
718/* Pass 'skb' through conntrack in 'net', using zone configured in 'info', if
719 * not done already. Update key with new CT state after passing the packet
720 * through conntrack.
721 * Note that if the packet is deemed invalid by conntrack, skb->_nfct will be
722 * set to NULL and 0 will be returned.
723 */
724static int __ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
725 const struct ovs_conntrack_info *info,
726 struct sk_buff *skb)
727{
728 /* If we are recirculating packets to match on conntrack fields and
729 * committing with a separate conntrack action, then we don't need to
730 * actually run the packet through conntrack twice unless it's for a
731 * different zone.
732 */
733 bool cached = skb_nfct_cached(net, key, info, skb);
734 enum ip_conntrack_info ctinfo;
735 struct nf_conn *ct;
736
737 /* If the ct entry is not confirmed and shared with some other skb,
738 * e.g., a cloned one, we can't just modify it with the commit as we
739 * must not modify the extension set. Reset.
740 */
741 if (cached && info->commit) {
742 ct = nf_ct_get(skb, &ctinfo);
743 if (ct && !nf_ct_is_confirmed(ct) && nf_ct_shared(ct)) {
744 nf_reset_ct(skb);
745 cached = false;
746 }
747 }
748
749 if (!cached) {
750 struct nf_hook_state state = {
751 .hook = NF_INET_PRE_ROUTING,
752 .pf = info->family,
753 .net = net,
754 };
755 struct nf_conn *tmpl = info->ct;
756 int err;
757
758 /* Associate skb with specified zone. */
759 if (tmpl) {
760 nf_reset_ct(skb);
761 nf_conntrack_get(&tmpl->ct_general);
762 nf_ct_set(skb, tmpl, IP_CT_NEW);
763 }
764
765 err = nf_conntrack_in(skb, &state);
766 if (err != NF_ACCEPT)
767 return verdict_to_errno(err);
768
769 /* Clear CT state NAT flags to mark that we have not yet done
770 * NAT after the nf_conntrack_in() call. We can actually clear
771 * the whole state, as it will be re-initialized below.
772 */
773 key->ct_state = 0;
774
775 /* Update the key, but keep the NAT flags. */
776 ovs_ct_update_key(skb, info, key, true, true);
777 }
778
779 ct = nf_ct_get(skb, &ctinfo);
780 if (ct) {
781 /* Packets starting a new connection must be NATted before the
782 * helper, so that the helper knows about the NAT. We enforce
783 * this by delaying both NAT and helper calls for unconfirmed
784 * connections until the committing CT action. For later
785 * packets NAT and Helper may be called in either order.
786 *
787 * NAT will be done only if the CT action has NAT, and only
788 * once per packet (per zone), as guarded by the NAT bits in
789 * the key->ct_state.
790 */
791 if (info->nat && !(key->ct_state & OVS_CS_F_NAT_MASK) &&
792 (nf_ct_is_confirmed(ct) || info->commit)) {
793 int err = ovs_ct_nat(net, key, info, skb, ct, ctinfo);
794
795 err = verdict_to_errno(err);
796 if (err)
797 return err;
798 }
799
800 /* Userspace may decide to perform a ct lookup without a helper
801 * specified followed by a (recirculate and) commit with one,
802 * or attach a helper in a later commit. Therefore, for
803 * connections which we will commit, we may need to attach
804 * the helper here.
805 */
806 if (!nf_ct_is_confirmed(ct) && info->commit &&
807 info->helper && !nfct_help(ct)) {
808 int err = __nf_ct_try_assign_helper(ct, info->ct,
809 GFP_ATOMIC);
810 if (err)
811 return err;
812
813 /* helper installed, add seqadj if NAT is required */
814 if (info->nat && !nfct_seqadj(ct)) {
815 if (!nfct_seqadj_ext_add(ct))
816 return -EINVAL;
817 }
818 }
819
820 /* Call the helper only if nf_conntrack_in() was executed
821 * above ("!cached").
822 *
823 * For unconfirmed connections it will be called later during
824 * commit as we need to have all the other extensions allocated
825 * before the call.
826 */
827 if (nf_ct_is_confirmed(ct) && !cached) {
828 int err = nf_ct_helper(skb, ct, ctinfo, info->family);
829
830 err = verdict_to_errno(err);
831 if (err)
832 return err;
833 }
834
835 if (nf_ct_protonum(ct) == IPPROTO_TCP &&
836 nf_ct_is_confirmed(ct) && nf_conntrack_tcp_established(ct)) {
837 /* Be liberal for tcp packets so that out-of-window
838 * packets are not marked invalid.
839 */
840 nf_ct_set_tcp_be_liberal(ct);
841 }
842
843 nf_conn_act_ct_ext_fill(skb, ct, ctinfo);
844 }
845
846 return 0;
847}
848
849/* Lookup connection and read fields into key. */
850static int ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
851 const struct ovs_conntrack_info *info,
852 struct sk_buff *skb)
853{
854 struct nf_conn *ct;
855 int err;
856
857 err = __ovs_ct_lookup(net, key, info, skb);
858 if (err)
859 return err;
860
861 ct = (struct nf_conn *)skb_nfct(skb);
862 if (ct)
863 nf_ct_deliver_cached_events(ct);
864
865 return 0;
866}
867
868static bool labels_nonzero(const struct ovs_key_ct_labels *labels)
869{
870 size_t i;
871
872 for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
873 if (labels->ct_labels_32[i])
874 return true;
875
876 return false;
877}
878
879#if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
880static struct hlist_head *ct_limit_hash_bucket(
881 const struct ovs_ct_limit_info *info, u16 zone)
882{
883 return &info->limits[zone & (CT_LIMIT_HASH_BUCKETS - 1)];
884}
885
886/* Call with ovs_mutex */
887static void ct_limit_set(const struct ovs_ct_limit_info *info,
888 struct ovs_ct_limit *new_ct_limit)
889{
890 struct ovs_ct_limit *ct_limit;
891 struct hlist_head *head;
892
893 head = ct_limit_hash_bucket(info, new_ct_limit->zone);
894 hlist_for_each_entry_rcu(ct_limit, head, hlist_node,
895 lockdep_ovsl_is_held()) {
896 if (ct_limit->zone == new_ct_limit->zone) {
897 hlist_replace_rcu(&ct_limit->hlist_node,
898 &new_ct_limit->hlist_node);
899 kfree_rcu(ct_limit, rcu);
900 return;
901 }
902 }
903
904 hlist_add_head_rcu(&new_ct_limit->hlist_node, head);
905}
906
907/* Call with ovs_mutex */
908static void ct_limit_del(const struct ovs_ct_limit_info *info, u16 zone)
909{
910 struct ovs_ct_limit *ct_limit;
911 struct hlist_head *head;
912 struct hlist_node *n;
913
914 head = ct_limit_hash_bucket(info, zone);
915 hlist_for_each_entry_safe(ct_limit, n, head, hlist_node) {
916 if (ct_limit->zone == zone) {
917 hlist_del_rcu(&ct_limit->hlist_node);
918 kfree_rcu(ct_limit, rcu);
919 return;
920 }
921 }
922}
923
924/* Call with RCU read lock */
925static u32 ct_limit_get(const struct ovs_ct_limit_info *info, u16 zone)
926{
927 struct ovs_ct_limit *ct_limit;
928 struct hlist_head *head;
929
930 head = ct_limit_hash_bucket(info, zone);
931 hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
932 if (ct_limit->zone == zone)
933 return ct_limit->limit;
934 }
935
936 return info->default_limit;
937}
938
939static int ovs_ct_check_limit(struct net *net,
940 const struct sk_buff *skb,
941 const struct ovs_conntrack_info *info)
942{
943 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
944 const struct ovs_ct_limit_info *ct_limit_info;
945 u32 per_zone_limit, connections;
946 u32 conncount_key;
947
948 ct_limit_info = rcu_dereference(ovs_net->ct_limit_info);
949 if (!ct_limit_info)
950 return 0;
951
952 conncount_key = info->zone.id;
953
954 per_zone_limit = ct_limit_get(ct_limit_info, info->zone.id);
955 if (per_zone_limit == OVS_CT_LIMIT_UNLIMITED)
956 return 0;
957
958 connections = nf_conncount_count_skb(net, skb, info->family,
959 ct_limit_info->data,
960 &conncount_key);
961 if (connections > per_zone_limit)
962 return -ENOMEM;
963
964 return 0;
965}
966#endif
967
968/* Lookup connection and confirm if unconfirmed. */
969static int ovs_ct_commit(struct net *net, struct sw_flow_key *key,
970 const struct ovs_conntrack_info *info,
971 struct sk_buff *skb)
972{
973 enum ip_conntrack_info ctinfo;
974 struct nf_conn *ct;
975 int err;
976
977 err = __ovs_ct_lookup(net, key, info, skb);
978 if (err)
979 return err;
980
981 /* The connection could be invalid, in which case this is a no-op.*/
982 ct = nf_ct_get(skb, &ctinfo);
983 if (!ct)
984 return 0;
985
986#if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
987 if (static_branch_unlikely(&ovs_ct_limit_enabled)) {
988 if (!nf_ct_is_confirmed(ct)) {
989 err = ovs_ct_check_limit(net, skb, info);
990 if (err) {
991 net_warn_ratelimited("openvswitch: zone: %u "
992 "exceeds conntrack limit\n",
993 info->zone.id);
994 return err;
995 }
996 }
997 }
998#endif
999
1000 /* Set the conntrack event mask if given. NEW and DELETE events have
1001 * their own groups, but the NFNLGRP_CONNTRACK_UPDATE group listener
1002 * typically would receive many kinds of updates. Setting the event
1003 * mask allows those events to be filtered. The set event mask will
1004 * remain in effect for the lifetime of the connection unless changed
1005 * by a further CT action with both the commit flag and the eventmask
1006 * option. */
1007 if (info->have_eventmask) {
1008 struct nf_conntrack_ecache *cache = nf_ct_ecache_find(ct);
1009
1010 if (cache)
1011 cache->ctmask = info->eventmask;
1012 }
1013
1014 /* Apply changes before confirming the connection so that the initial
1015 * conntrack NEW netlink event carries the values given in the CT
1016 * action.
1017 */
1018 if (info->mark.mask) {
1019 err = ovs_ct_set_mark(ct, key, info->mark.value,
1020 info->mark.mask);
1021 if (err)
1022 return err;
1023 }
1024 if (!nf_ct_is_confirmed(ct)) {
1025 err = ovs_ct_init_labels(ct, key, &info->labels.value,
1026 &info->labels.mask);
1027 if (err)
1028 return err;
1029
1030 nf_conn_act_ct_ext_add(skb, ct, ctinfo);
1031
1032 /* Call the helpers now. We couldn't do this before as
1033 * all the extensions must be allocated before the call.
1034 */
1035 err = nf_ct_helper(skb, ct, ctinfo, info->family);
1036 err = verdict_to_errno(err);
1037 if (err)
1038 return err;
1039 } else if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1040 labels_nonzero(&info->labels.mask)) {
1041 err = ovs_ct_set_labels(ct, key, &info->labels.value,
1042 &info->labels.mask);
1043 if (err)
1044 return err;
1045 }
1046 /* This will take care of sending queued events even if the connection
1047 * is already confirmed.
1048 */
1049 err = nf_conntrack_confirm(skb);
1050
1051 return verdict_to_errno(err);
1052}
1053
1054/* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
1055 * value if 'skb' is freed.
1056 */
1057int ovs_ct_execute(struct net *net, struct sk_buff *skb,
1058 struct sw_flow_key *key,
1059 const struct ovs_conntrack_info *info)
1060{
1061 int nh_ofs;
1062 int err;
1063
1064 /* The conntrack module expects to be working at L3. */
1065 nh_ofs = skb_network_offset(skb);
1066 skb_pull_rcsum(skb, nh_ofs);
1067
1068 err = nf_ct_skb_network_trim(skb, info->family);
1069 if (err) {
1070 kfree_skb(skb);
1071 return err;
1072 }
1073
1074 if (key->ip.frag != OVS_FRAG_TYPE_NONE) {
1075 err = ovs_ct_handle_fragments(net, key, info->zone.id,
1076 info->family, skb);
1077 if (err)
1078 return err;
1079 }
1080
1081 if (info->commit)
1082 err = ovs_ct_commit(net, key, info, skb);
1083 else
1084 err = ovs_ct_lookup(net, key, info, skb);
1085
1086 /* conntrack core returned NF_STOLEN */
1087 if (err == -EINPROGRESS)
1088 return err;
1089
1090 skb_push_rcsum(skb, nh_ofs);
1091 if (err)
1092 ovs_kfree_skb_reason(skb, OVS_DROP_CONNTRACK);
1093 return err;
1094}
1095
1096int ovs_ct_clear(struct sk_buff *skb, struct sw_flow_key *key)
1097{
1098 nf_reset_ct(skb);
1099 nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
1100
1101 if (key)
1102 ovs_ct_fill_key(skb, key, false);
1103
1104 return 0;
1105}
1106
1107#if IS_ENABLED(CONFIG_NF_NAT)
1108static int parse_nat(const struct nlattr *attr,
1109 struct ovs_conntrack_info *info, bool log)
1110{
1111 struct nlattr *a;
1112 int rem;
1113 bool have_ip_max = false;
1114 bool have_proto_max = false;
1115 bool ip_vers = (info->family == NFPROTO_IPV6);
1116
1117 nla_for_each_nested(a, attr, rem) {
1118 static const int ovs_nat_attr_lens[OVS_NAT_ATTR_MAX + 1][2] = {
1119 [OVS_NAT_ATTR_SRC] = {0, 0},
1120 [OVS_NAT_ATTR_DST] = {0, 0},
1121 [OVS_NAT_ATTR_IP_MIN] = {sizeof(struct in_addr),
1122 sizeof(struct in6_addr)},
1123 [OVS_NAT_ATTR_IP_MAX] = {sizeof(struct in_addr),
1124 sizeof(struct in6_addr)},
1125 [OVS_NAT_ATTR_PROTO_MIN] = {sizeof(u16), sizeof(u16)},
1126 [OVS_NAT_ATTR_PROTO_MAX] = {sizeof(u16), sizeof(u16)},
1127 [OVS_NAT_ATTR_PERSISTENT] = {0, 0},
1128 [OVS_NAT_ATTR_PROTO_HASH] = {0, 0},
1129 [OVS_NAT_ATTR_PROTO_RANDOM] = {0, 0},
1130 };
1131 int type = nla_type(a);
1132
1133 if (type > OVS_NAT_ATTR_MAX) {
1134 OVS_NLERR(log, "Unknown NAT attribute (type=%d, max=%d)",
1135 type, OVS_NAT_ATTR_MAX);
1136 return -EINVAL;
1137 }
1138
1139 if (nla_len(a) != ovs_nat_attr_lens[type][ip_vers]) {
1140 OVS_NLERR(log, "NAT attribute type %d has unexpected length (%d != %d)",
1141 type, nla_len(a),
1142 ovs_nat_attr_lens[type][ip_vers]);
1143 return -EINVAL;
1144 }
1145
1146 switch (type) {
1147 case OVS_NAT_ATTR_SRC:
1148 case OVS_NAT_ATTR_DST:
1149 if (info->nat) {
1150 OVS_NLERR(log, "Only one type of NAT may be specified");
1151 return -ERANGE;
1152 }
1153 info->nat |= OVS_CT_NAT;
1154 info->nat |= ((type == OVS_NAT_ATTR_SRC)
1155 ? OVS_CT_SRC_NAT : OVS_CT_DST_NAT);
1156 break;
1157
1158 case OVS_NAT_ATTR_IP_MIN:
1159 nla_memcpy(&info->range.min_addr, a,
1160 sizeof(info->range.min_addr));
1161 info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1162 break;
1163
1164 case OVS_NAT_ATTR_IP_MAX:
1165 have_ip_max = true;
1166 nla_memcpy(&info->range.max_addr, a,
1167 sizeof(info->range.max_addr));
1168 info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1169 break;
1170
1171 case OVS_NAT_ATTR_PROTO_MIN:
1172 info->range.min_proto.all = htons(nla_get_u16(a));
1173 info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1174 break;
1175
1176 case OVS_NAT_ATTR_PROTO_MAX:
1177 have_proto_max = true;
1178 info->range.max_proto.all = htons(nla_get_u16(a));
1179 info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1180 break;
1181
1182 case OVS_NAT_ATTR_PERSISTENT:
1183 info->range.flags |= NF_NAT_RANGE_PERSISTENT;
1184 break;
1185
1186 case OVS_NAT_ATTR_PROTO_HASH:
1187 info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM;
1188 break;
1189
1190 case OVS_NAT_ATTR_PROTO_RANDOM:
1191 info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM_FULLY;
1192 break;
1193
1194 default:
1195 OVS_NLERR(log, "Unknown nat attribute (%d)", type);
1196 return -EINVAL;
1197 }
1198 }
1199
1200 if (rem > 0) {
1201 OVS_NLERR(log, "NAT attribute has %d unknown bytes", rem);
1202 return -EINVAL;
1203 }
1204 if (!info->nat) {
1205 /* Do not allow flags if no type is given. */
1206 if (info->range.flags) {
1207 OVS_NLERR(log,
1208 "NAT flags may be given only when NAT range (SRC or DST) is also specified."
1209 );
1210 return -EINVAL;
1211 }
1212 info->nat = OVS_CT_NAT; /* NAT existing connections. */
1213 } else if (!info->commit) {
1214 OVS_NLERR(log,
1215 "NAT attributes may be specified only when CT COMMIT flag is also specified."
1216 );
1217 return -EINVAL;
1218 }
1219 /* Allow missing IP_MAX. */
1220 if (info->range.flags & NF_NAT_RANGE_MAP_IPS && !have_ip_max) {
1221 memcpy(&info->range.max_addr, &info->range.min_addr,
1222 sizeof(info->range.max_addr));
1223 }
1224 /* Allow missing PROTO_MAX. */
1225 if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1226 !have_proto_max) {
1227 info->range.max_proto.all = info->range.min_proto.all;
1228 }
1229 return 0;
1230}
1231#endif
1232
1233static const struct ovs_ct_len_tbl ovs_ct_attr_lens[OVS_CT_ATTR_MAX + 1] = {
1234 [OVS_CT_ATTR_COMMIT] = { .minlen = 0, .maxlen = 0 },
1235 [OVS_CT_ATTR_FORCE_COMMIT] = { .minlen = 0, .maxlen = 0 },
1236 [OVS_CT_ATTR_ZONE] = { .minlen = sizeof(u16),
1237 .maxlen = sizeof(u16) },
1238 [OVS_CT_ATTR_MARK] = { .minlen = sizeof(struct md_mark),
1239 .maxlen = sizeof(struct md_mark) },
1240 [OVS_CT_ATTR_LABELS] = { .minlen = sizeof(struct md_labels),
1241 .maxlen = sizeof(struct md_labels) },
1242 [OVS_CT_ATTR_HELPER] = { .minlen = 1,
1243 .maxlen = NF_CT_HELPER_NAME_LEN },
1244#if IS_ENABLED(CONFIG_NF_NAT)
1245 /* NAT length is checked when parsing the nested attributes. */
1246 [OVS_CT_ATTR_NAT] = { .minlen = 0, .maxlen = INT_MAX },
1247#endif
1248 [OVS_CT_ATTR_EVENTMASK] = { .minlen = sizeof(u32),
1249 .maxlen = sizeof(u32) },
1250 [OVS_CT_ATTR_TIMEOUT] = { .minlen = 1,
1251 .maxlen = CTNL_TIMEOUT_NAME_MAX },
1252};
1253
1254static int parse_ct(const struct nlattr *attr, struct ovs_conntrack_info *info,
1255 const char **helper, bool log)
1256{
1257 struct nlattr *a;
1258 int rem;
1259
1260 nla_for_each_nested(a, attr, rem) {
1261 int type = nla_type(a);
1262 int maxlen;
1263 int minlen;
1264
1265 if (type > OVS_CT_ATTR_MAX) {
1266 OVS_NLERR(log,
1267 "Unknown conntrack attr (type=%d, max=%d)",
1268 type, OVS_CT_ATTR_MAX);
1269 return -EINVAL;
1270 }
1271
1272 maxlen = ovs_ct_attr_lens[type].maxlen;
1273 minlen = ovs_ct_attr_lens[type].minlen;
1274 if (nla_len(a) < minlen || nla_len(a) > maxlen) {
1275 OVS_NLERR(log,
1276 "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
1277 type, nla_len(a), maxlen);
1278 return -EINVAL;
1279 }
1280
1281 switch (type) {
1282 case OVS_CT_ATTR_FORCE_COMMIT:
1283 info->force = true;
1284 fallthrough;
1285 case OVS_CT_ATTR_COMMIT:
1286 info->commit = true;
1287 break;
1288#ifdef CONFIG_NF_CONNTRACK_ZONES
1289 case OVS_CT_ATTR_ZONE:
1290 info->zone.id = nla_get_u16(a);
1291 break;
1292#endif
1293#ifdef CONFIG_NF_CONNTRACK_MARK
1294 case OVS_CT_ATTR_MARK: {
1295 struct md_mark *mark = nla_data(a);
1296
1297 if (!mark->mask) {
1298 OVS_NLERR(log, "ct_mark mask cannot be 0");
1299 return -EINVAL;
1300 }
1301 info->mark = *mark;
1302 break;
1303 }
1304#endif
1305#ifdef CONFIG_NF_CONNTRACK_LABELS
1306 case OVS_CT_ATTR_LABELS: {
1307 struct md_labels *labels = nla_data(a);
1308
1309 if (!labels_nonzero(&labels->mask)) {
1310 OVS_NLERR(log, "ct_labels mask cannot be 0");
1311 return -EINVAL;
1312 }
1313 info->labels = *labels;
1314 break;
1315 }
1316#endif
1317 case OVS_CT_ATTR_HELPER:
1318 *helper = nla_data(a);
1319 if (!string_is_terminated(*helper, nla_len(a))) {
1320 OVS_NLERR(log, "Invalid conntrack helper");
1321 return -EINVAL;
1322 }
1323 break;
1324#if IS_ENABLED(CONFIG_NF_NAT)
1325 case OVS_CT_ATTR_NAT: {
1326 int err = parse_nat(a, info, log);
1327
1328 if (err)
1329 return err;
1330 break;
1331 }
1332#endif
1333 case OVS_CT_ATTR_EVENTMASK:
1334 info->have_eventmask = true;
1335 info->eventmask = nla_get_u32(a);
1336 break;
1337#ifdef CONFIG_NF_CONNTRACK_TIMEOUT
1338 case OVS_CT_ATTR_TIMEOUT:
1339 memcpy(info->timeout, nla_data(a), nla_len(a));
1340 if (!string_is_terminated(info->timeout, nla_len(a))) {
1341 OVS_NLERR(log, "Invalid conntrack timeout");
1342 return -EINVAL;
1343 }
1344 break;
1345#endif
1346
1347 default:
1348 OVS_NLERR(log, "Unknown conntrack attr (%d)",
1349 type);
1350 return -EINVAL;
1351 }
1352 }
1353
1354#ifdef CONFIG_NF_CONNTRACK_MARK
1355 if (!info->commit && info->mark.mask) {
1356 OVS_NLERR(log,
1357 "Setting conntrack mark requires 'commit' flag.");
1358 return -EINVAL;
1359 }
1360#endif
1361#ifdef CONFIG_NF_CONNTRACK_LABELS
1362 if (!info->commit && labels_nonzero(&info->labels.mask)) {
1363 OVS_NLERR(log,
1364 "Setting conntrack labels requires 'commit' flag.");
1365 return -EINVAL;
1366 }
1367#endif
1368 if (rem > 0) {
1369 OVS_NLERR(log, "Conntrack attr has %d unknown bytes", rem);
1370 return -EINVAL;
1371 }
1372
1373 return 0;
1374}
1375
1376bool ovs_ct_verify(struct net *net, enum ovs_key_attr attr)
1377{
1378 if (attr == OVS_KEY_ATTR_CT_STATE)
1379 return true;
1380 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1381 attr == OVS_KEY_ATTR_CT_ZONE)
1382 return true;
1383 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1384 attr == OVS_KEY_ATTR_CT_MARK)
1385 return true;
1386 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1387 attr == OVS_KEY_ATTR_CT_LABELS) {
1388 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1389
1390 return ovs_net->xt_label;
1391 }
1392
1393 return false;
1394}
1395
1396int ovs_ct_copy_action(struct net *net, const struct nlattr *attr,
1397 const struct sw_flow_key *key,
1398 struct sw_flow_actions **sfa, bool log)
1399{
1400 struct ovs_conntrack_info ct_info;
1401 const char *helper = NULL;
1402 u16 family;
1403 int err;
1404
1405 family = key_to_nfproto(key);
1406 if (family == NFPROTO_UNSPEC) {
1407 OVS_NLERR(log, "ct family unspecified");
1408 return -EINVAL;
1409 }
1410
1411 memset(&ct_info, 0, sizeof(ct_info));
1412 ct_info.family = family;
1413
1414 nf_ct_zone_init(&ct_info.zone, NF_CT_DEFAULT_ZONE_ID,
1415 NF_CT_DEFAULT_ZONE_DIR, 0);
1416
1417 err = parse_ct(attr, &ct_info, &helper, log);
1418 if (err)
1419 return err;
1420
1421 /* Set up template for tracking connections in specific zones. */
1422 ct_info.ct = nf_ct_tmpl_alloc(net, &ct_info.zone, GFP_KERNEL);
1423 if (!ct_info.ct) {
1424 OVS_NLERR(log, "Failed to allocate conntrack template");
1425 return -ENOMEM;
1426 }
1427
1428 if (ct_info.timeout[0]) {
1429 if (nf_ct_set_timeout(net, ct_info.ct, family, key->ip.proto,
1430 ct_info.timeout))
1431 OVS_NLERR(log,
1432 "Failed to associated timeout policy '%s'",
1433 ct_info.timeout);
1434 else
1435 ct_info.nf_ct_timeout = rcu_dereference(
1436 nf_ct_timeout_find(ct_info.ct)->timeout);
1437
1438 }
1439
1440 if (helper) {
1441 err = nf_ct_add_helper(ct_info.ct, helper, ct_info.family,
1442 key->ip.proto, ct_info.nat, &ct_info.helper);
1443 if (err) {
1444 OVS_NLERR(log, "Failed to add %s helper %d", helper, err);
1445 goto err_free_ct;
1446 }
1447 }
1448
1449 err = ovs_nla_add_action(sfa, OVS_ACTION_ATTR_CT, &ct_info,
1450 sizeof(ct_info), log);
1451 if (err)
1452 goto err_free_ct;
1453
1454 if (ct_info.commit)
1455 __set_bit(IPS_CONFIRMED_BIT, &ct_info.ct->status);
1456 return 0;
1457err_free_ct:
1458 __ovs_ct_free_action(&ct_info);
1459 return err;
1460}
1461
1462#if IS_ENABLED(CONFIG_NF_NAT)
1463static bool ovs_ct_nat_to_attr(const struct ovs_conntrack_info *info,
1464 struct sk_buff *skb)
1465{
1466 struct nlattr *start;
1467
1468 start = nla_nest_start_noflag(skb, OVS_CT_ATTR_NAT);
1469 if (!start)
1470 return false;
1471
1472 if (info->nat & OVS_CT_SRC_NAT) {
1473 if (nla_put_flag(skb, OVS_NAT_ATTR_SRC))
1474 return false;
1475 } else if (info->nat & OVS_CT_DST_NAT) {
1476 if (nla_put_flag(skb, OVS_NAT_ATTR_DST))
1477 return false;
1478 } else {
1479 goto out;
1480 }
1481
1482 if (info->range.flags & NF_NAT_RANGE_MAP_IPS) {
1483 if (IS_ENABLED(CONFIG_NF_NAT) &&
1484 info->family == NFPROTO_IPV4) {
1485 if (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MIN,
1486 info->range.min_addr.ip) ||
1487 (info->range.max_addr.ip
1488 != info->range.min_addr.ip &&
1489 (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MAX,
1490 info->range.max_addr.ip))))
1491 return false;
1492 } else if (IS_ENABLED(CONFIG_IPV6) &&
1493 info->family == NFPROTO_IPV6) {
1494 if (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MIN,
1495 &info->range.min_addr.in6) ||
1496 (memcmp(&info->range.max_addr.in6,
1497 &info->range.min_addr.in6,
1498 sizeof(info->range.max_addr.in6)) &&
1499 (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MAX,
1500 &info->range.max_addr.in6))))
1501 return false;
1502 } else {
1503 return false;
1504 }
1505 }
1506 if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1507 (nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MIN,
1508 ntohs(info->range.min_proto.all)) ||
1509 (info->range.max_proto.all != info->range.min_proto.all &&
1510 nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MAX,
1511 ntohs(info->range.max_proto.all)))))
1512 return false;
1513
1514 if (info->range.flags & NF_NAT_RANGE_PERSISTENT &&
1515 nla_put_flag(skb, OVS_NAT_ATTR_PERSISTENT))
1516 return false;
1517 if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM &&
1518 nla_put_flag(skb, OVS_NAT_ATTR_PROTO_HASH))
1519 return false;
1520 if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM_FULLY &&
1521 nla_put_flag(skb, OVS_NAT_ATTR_PROTO_RANDOM))
1522 return false;
1523out:
1524 nla_nest_end(skb, start);
1525
1526 return true;
1527}
1528#endif
1529
1530int ovs_ct_action_to_attr(const struct ovs_conntrack_info *ct_info,
1531 struct sk_buff *skb)
1532{
1533 struct nlattr *start;
1534
1535 start = nla_nest_start_noflag(skb, OVS_ACTION_ATTR_CT);
1536 if (!start)
1537 return -EMSGSIZE;
1538
1539 if (ct_info->commit && nla_put_flag(skb, ct_info->force
1540 ? OVS_CT_ATTR_FORCE_COMMIT
1541 : OVS_CT_ATTR_COMMIT))
1542 return -EMSGSIZE;
1543 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1544 nla_put_u16(skb, OVS_CT_ATTR_ZONE, ct_info->zone.id))
1545 return -EMSGSIZE;
1546 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && ct_info->mark.mask &&
1547 nla_put(skb, OVS_CT_ATTR_MARK, sizeof(ct_info->mark),
1548 &ct_info->mark))
1549 return -EMSGSIZE;
1550 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1551 labels_nonzero(&ct_info->labels.mask) &&
1552 nla_put(skb, OVS_CT_ATTR_LABELS, sizeof(ct_info->labels),
1553 &ct_info->labels))
1554 return -EMSGSIZE;
1555 if (ct_info->helper) {
1556 if (nla_put_string(skb, OVS_CT_ATTR_HELPER,
1557 ct_info->helper->name))
1558 return -EMSGSIZE;
1559 }
1560 if (ct_info->have_eventmask &&
1561 nla_put_u32(skb, OVS_CT_ATTR_EVENTMASK, ct_info->eventmask))
1562 return -EMSGSIZE;
1563 if (ct_info->timeout[0]) {
1564 if (nla_put_string(skb, OVS_CT_ATTR_TIMEOUT, ct_info->timeout))
1565 return -EMSGSIZE;
1566 }
1567
1568#if IS_ENABLED(CONFIG_NF_NAT)
1569 if (ct_info->nat && !ovs_ct_nat_to_attr(ct_info, skb))
1570 return -EMSGSIZE;
1571#endif
1572 nla_nest_end(skb, start);
1573
1574 return 0;
1575}
1576
1577void ovs_ct_free_action(const struct nlattr *a)
1578{
1579 struct ovs_conntrack_info *ct_info = nla_data(a);
1580
1581 __ovs_ct_free_action(ct_info);
1582}
1583
1584static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info)
1585{
1586 if (ct_info->helper) {
1587#if IS_ENABLED(CONFIG_NF_NAT)
1588 if (ct_info->nat)
1589 nf_nat_helper_put(ct_info->helper);
1590#endif
1591 nf_conntrack_helper_put(ct_info->helper);
1592 }
1593 if (ct_info->ct) {
1594 if (ct_info->timeout[0])
1595 nf_ct_destroy_timeout(ct_info->ct);
1596 nf_ct_tmpl_free(ct_info->ct);
1597 }
1598}
1599
1600#if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1601static int ovs_ct_limit_init(struct net *net, struct ovs_net *ovs_net)
1602{
1603 struct ovs_ct_limit_info *info;
1604 int i, err;
1605
1606 info = kmalloc_obj(*info);
1607 if (!info)
1608 return -ENOMEM;
1609
1610 info->default_limit = OVS_CT_LIMIT_DEFAULT;
1611 info->limits =
1612 kmalloc_objs(struct hlist_head, CT_LIMIT_HASH_BUCKETS);
1613 if (!info->limits) {
1614 kfree(info);
1615 return -ENOMEM;
1616 }
1617
1618 for (i = 0; i < CT_LIMIT_HASH_BUCKETS; i++)
1619 INIT_HLIST_HEAD(&info->limits[i]);
1620
1621 info->data = nf_conncount_init(net, sizeof(u32));
1622
1623 if (IS_ERR(info->data)) {
1624 err = PTR_ERR(info->data);
1625 kfree(info->limits);
1626 kfree(info);
1627 pr_err("openvswitch: failed to init nf_conncount %d\n", err);
1628 return err;
1629 }
1630 rcu_assign_pointer(ovs_net->ct_limit_info, info);
1631 return 0;
1632}
1633
1634static void *ovs_ct_limit_exit_start(struct ovs_net *ovs_net)
1635{
1636 return rcu_replace_pointer(ovs_net->ct_limit_info, NULL,
1637 lockdep_ovsl_is_held());
1638}
1639
1640/* The CT limit state must be detached by ovs_ct_limit_exit_start() and an
1641 * RCU grace period must elapse before this function runs. The pernet core
1642 * guarantees the grace period between the .pre_exit and .exit callbacks.
1643 */
1644static void ovs_ct_limit_exit_finish(struct net *net, void *data)
1645{
1646 const struct ovs_ct_limit_info *info = data;
1647 int i;
1648
1649 if (!info)
1650 return;
1651
1652 nf_conncount_destroy(net, info->data);
1653 for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) {
1654 struct hlist_head *head = &info->limits[i];
1655 struct ovs_ct_limit *ct_limit;
1656 struct hlist_node *next;
1657
1658 hlist_for_each_entry_safe(ct_limit, next, head, hlist_node)
1659 kfree(ct_limit);
1660 }
1661 kfree(info->limits);
1662 kfree(info);
1663}
1664
1665static struct sk_buff *
1666ovs_ct_limit_cmd_reply_start(struct genl_info *info, u8 cmd,
1667 struct ovs_header **ovs_reply_header)
1668{
1669 struct ovs_header *ovs_header = genl_info_userhdr(info);
1670 struct sk_buff *skb;
1671
1672 skb = genlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1673 if (!skb)
1674 return ERR_PTR(-ENOMEM);
1675
1676 *ovs_reply_header = genlmsg_put(skb, info->snd_portid,
1677 info->snd_seq,
1678 &dp_ct_limit_genl_family, 0, cmd);
1679
1680 if (!*ovs_reply_header) {
1681 nlmsg_free(skb);
1682 return ERR_PTR(-EMSGSIZE);
1683 }
1684 (*ovs_reply_header)->dp_ifindex = ovs_header->dp_ifindex;
1685
1686 return skb;
1687}
1688
1689static bool check_zone_id(int zone_id, u16 *pzone)
1690{
1691 if (zone_id >= 0 && zone_id <= 65535) {
1692 *pzone = (u16)zone_id;
1693 return true;
1694 }
1695 return false;
1696}
1697
1698static int ovs_ct_limit_set_zone_limit(struct ovs_net *ovs_net,
1699 struct nlattr *nla_zone_limit)
1700{
1701 struct ovs_zone_limit *zone_limit;
1702 struct ovs_ct_limit_info *info;
1703 u16 zone;
1704 int rem;
1705
1706 rem = NLA_ALIGN(nla_len(nla_zone_limit));
1707 zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1708
1709 while (rem >= sizeof(*zone_limit)) {
1710 if (unlikely(zone_limit->zone_id ==
1711 OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1712 ovs_lock();
1713 info = ovsl_dereference(ovs_net->ct_limit_info);
1714 info->default_limit = zone_limit->limit;
1715 ovs_unlock();
1716 } else if (unlikely(!check_zone_id(
1717 zone_limit->zone_id, &zone))) {
1718 OVS_NLERR(true, "zone id is out of range");
1719 } else {
1720 struct ovs_ct_limit *ct_limit;
1721
1722 ct_limit = kmalloc_obj(*ct_limit, GFP_KERNEL_ACCOUNT);
1723 if (!ct_limit)
1724 return -ENOMEM;
1725
1726 ct_limit->zone = zone;
1727 ct_limit->limit = zone_limit->limit;
1728
1729 ovs_lock();
1730 info = ovsl_dereference(ovs_net->ct_limit_info);
1731 ct_limit_set(info, ct_limit);
1732 ovs_unlock();
1733 }
1734 rem -= NLA_ALIGN(sizeof(*zone_limit));
1735 zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
1736 NLA_ALIGN(sizeof(*zone_limit)));
1737 }
1738
1739 if (rem)
1740 OVS_NLERR(true, "set zone limit has %d unknown bytes", rem);
1741
1742 return 0;
1743}
1744
1745static int ovs_ct_limit_del_zone_limit(struct ovs_net *ovs_net,
1746 struct nlattr *nla_zone_limit)
1747{
1748 struct ovs_zone_limit *zone_limit;
1749 struct ovs_ct_limit_info *info;
1750 u16 zone;
1751 int rem;
1752
1753 rem = NLA_ALIGN(nla_len(nla_zone_limit));
1754 zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1755
1756 while (rem >= sizeof(*zone_limit)) {
1757 if (unlikely(zone_limit->zone_id ==
1758 OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1759 ovs_lock();
1760 info = ovsl_dereference(ovs_net->ct_limit_info);
1761 info->default_limit = OVS_CT_LIMIT_DEFAULT;
1762 ovs_unlock();
1763 } else if (unlikely(!check_zone_id(
1764 zone_limit->zone_id, &zone))) {
1765 OVS_NLERR(true, "zone id is out of range");
1766 } else {
1767 ovs_lock();
1768 info = ovsl_dereference(ovs_net->ct_limit_info);
1769 ct_limit_del(info, zone);
1770 ovs_unlock();
1771 }
1772 rem -= NLA_ALIGN(sizeof(*zone_limit));
1773 zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
1774 NLA_ALIGN(sizeof(*zone_limit)));
1775 }
1776
1777 if (rem)
1778 OVS_NLERR(true, "del zone limit has %d unknown bytes", rem);
1779
1780 return 0;
1781}
1782
1783static int ovs_ct_limit_get_default_limit(struct ovs_ct_limit_info *info,
1784 struct sk_buff *reply)
1785{
1786 struct ovs_zone_limit zone_limit = {
1787 .zone_id = OVS_ZONE_LIMIT_DEFAULT_ZONE,
1788 .limit = info->default_limit,
1789 };
1790
1791 return nla_put_nohdr(reply, sizeof(zone_limit), &zone_limit);
1792}
1793
1794static int __ovs_ct_limit_get_zone_limit(struct net *net,
1795 struct nf_conncount_data *data,
1796 u16 zone_id, u32 limit,
1797 struct sk_buff *reply)
1798{
1799 struct nf_conntrack_zone ct_zone;
1800 struct ovs_zone_limit zone_limit;
1801 u32 conncount_key = zone_id;
1802
1803 zone_limit.zone_id = zone_id;
1804 zone_limit.limit = limit;
1805 nf_ct_zone_init(&ct_zone, zone_id, NF_CT_DEFAULT_ZONE_DIR, 0);
1806
1807 zone_limit.count = nf_conncount_count_skb(net, NULL, 0, data,
1808 &conncount_key);
1809 return nla_put_nohdr(reply, sizeof(zone_limit), &zone_limit);
1810}
1811
1812/* Called with RCU read lock held. */
1813static int ovs_ct_limit_get_zone_limit(struct net *net,
1814 struct nlattr *nla_zone_limit,
1815 struct ovs_ct_limit_info *info,
1816 struct sk_buff *reply)
1817{
1818 struct ovs_zone_limit *zone_limit;
1819 int rem, err;
1820 u32 limit;
1821 u16 zone;
1822
1823 rem = NLA_ALIGN(nla_len(nla_zone_limit));
1824 zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1825
1826 while (rem >= sizeof(*zone_limit)) {
1827 if (unlikely(zone_limit->zone_id ==
1828 OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1829 err = ovs_ct_limit_get_default_limit(info, reply);
1830 if (err)
1831 return err;
1832 } else if (unlikely(!check_zone_id(zone_limit->zone_id,
1833 &zone))) {
1834 OVS_NLERR(true, "zone id is out of range");
1835 } else {
1836 limit = ct_limit_get(info, zone);
1837
1838 err = __ovs_ct_limit_get_zone_limit(
1839 net, info->data, zone, limit, reply);
1840 if (err)
1841 return err;
1842 }
1843 rem -= NLA_ALIGN(sizeof(*zone_limit));
1844 zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
1845 NLA_ALIGN(sizeof(*zone_limit)));
1846 }
1847
1848 if (rem)
1849 OVS_NLERR(true, "get zone limit has %d unknown bytes", rem);
1850
1851 return 0;
1852}
1853
1854/* Called with RCU read lock held. */
1855static int ovs_ct_limit_get_all_zone_limit(struct net *net,
1856 struct ovs_ct_limit_info *info,
1857 struct sk_buff *reply)
1858{
1859 struct ovs_ct_limit *ct_limit;
1860 struct hlist_head *head;
1861 int i, err = 0;
1862
1863 err = ovs_ct_limit_get_default_limit(info, reply);
1864 if (err)
1865 return err;
1866
1867 for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) {
1868 head = &info->limits[i];
1869 hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
1870 err = __ovs_ct_limit_get_zone_limit(net, info->data,
1871 ct_limit->zone, ct_limit->limit, reply);
1872 if (err)
1873 return err;
1874 }
1875 }
1876
1877 return err;
1878}
1879
1880static int ovs_ct_limit_cmd_set(struct sk_buff *skb, struct genl_info *info)
1881{
1882 struct nlattr **a = info->attrs;
1883 struct sk_buff *reply;
1884 struct ovs_header *ovs_reply_header;
1885 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1886 int err;
1887
1888 reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_SET,
1889 &ovs_reply_header);
1890 if (IS_ERR(reply))
1891 return PTR_ERR(reply);
1892
1893 if (!a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
1894 err = -EINVAL;
1895 goto exit_err;
1896 }
1897
1898 err = ovs_ct_limit_set_zone_limit(ovs_net,
1899 a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]);
1900 if (err)
1901 goto exit_err;
1902
1903 static_branch_enable(&ovs_ct_limit_enabled);
1904
1905 genlmsg_end(reply, ovs_reply_header);
1906 return genlmsg_reply(reply, info);
1907
1908exit_err:
1909 nlmsg_free(reply);
1910 return err;
1911}
1912
1913static int ovs_ct_limit_cmd_del(struct sk_buff *skb, struct genl_info *info)
1914{
1915 struct nlattr **a = info->attrs;
1916 struct sk_buff *reply;
1917 struct ovs_header *ovs_reply_header;
1918 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
1919 int err;
1920
1921 reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_DEL,
1922 &ovs_reply_header);
1923 if (IS_ERR(reply))
1924 return PTR_ERR(reply);
1925
1926 if (!a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
1927 err = -EINVAL;
1928 goto exit_err;
1929 }
1930
1931 err = ovs_ct_limit_del_zone_limit(ovs_net,
1932 a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]);
1933 if (err)
1934 goto exit_err;
1935
1936 genlmsg_end(reply, ovs_reply_header);
1937 return genlmsg_reply(reply, info);
1938
1939exit_err:
1940 nlmsg_free(reply);
1941 return err;
1942}
1943
1944static int ovs_ct_limit_cmd_get(struct sk_buff *skb, struct genl_info *info)
1945{
1946 struct nlattr **a = info->attrs;
1947 struct nlattr *nla_reply;
1948 struct sk_buff *reply;
1949 struct ovs_header *ovs_reply_header;
1950 struct net *net = sock_net(skb->sk);
1951 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1952 struct ovs_ct_limit_info *ct_limit_info;
1953 int err;
1954
1955 reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_GET,
1956 &ovs_reply_header);
1957 if (IS_ERR(reply))
1958 return PTR_ERR(reply);
1959
1960 nla_reply = nla_nest_start_noflag(reply, OVS_CT_LIMIT_ATTR_ZONE_LIMIT);
1961 if (!nla_reply) {
1962 err = -EMSGSIZE;
1963 goto exit_err;
1964 }
1965
1966 rcu_read_lock();
1967 ct_limit_info = rcu_dereference(ovs_net->ct_limit_info);
1968 if (a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
1969 err = ovs_ct_limit_get_zone_limit(
1970 net, a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT], ct_limit_info,
1971 reply);
1972 } else {
1973 err = ovs_ct_limit_get_all_zone_limit(net, ct_limit_info,
1974 reply);
1975 }
1976 rcu_read_unlock();
1977 if (err)
1978 goto exit_err;
1979
1980 nla_nest_end(reply, nla_reply);
1981 genlmsg_end(reply, ovs_reply_header);
1982 return genlmsg_reply(reply, info);
1983
1984exit_err:
1985 nlmsg_free(reply);
1986 return err;
1987}
1988
1989static const struct genl_small_ops ct_limit_genl_ops[] = {
1990 { .cmd = OVS_CT_LIMIT_CMD_SET,
1991 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1992 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN
1993 * privilege.
1994 */
1995 .doit = ovs_ct_limit_cmd_set,
1996 },
1997 { .cmd = OVS_CT_LIMIT_CMD_DEL,
1998 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
1999 .flags = GENL_UNS_ADMIN_PERM, /* Requires CAP_NET_ADMIN
2000 * privilege.
2001 */
2002 .doit = ovs_ct_limit_cmd_del,
2003 },
2004 { .cmd = OVS_CT_LIMIT_CMD_GET,
2005 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2006 .flags = 0, /* OK for unprivileged users. */
2007 .doit = ovs_ct_limit_cmd_get,
2008 },
2009};
2010
2011static const struct genl_multicast_group ovs_ct_limit_multicast_group = {
2012 .name = OVS_CT_LIMIT_MCGROUP,
2013};
2014
2015struct genl_family dp_ct_limit_genl_family __ro_after_init = {
2016 .hdrsize = sizeof(struct ovs_header),
2017 .name = OVS_CT_LIMIT_FAMILY,
2018 .version = OVS_CT_LIMIT_VERSION,
2019 .maxattr = OVS_CT_LIMIT_ATTR_MAX,
2020 .policy = ct_limit_policy,
2021 .netnsok = true,
2022 .parallel_ops = true,
2023 .small_ops = ct_limit_genl_ops,
2024 .n_small_ops = ARRAY_SIZE(ct_limit_genl_ops),
2025 .resv_start_op = OVS_CT_LIMIT_CMD_GET + 1,
2026 .mcgrps = &ovs_ct_limit_multicast_group,
2027 .n_mcgrps = 1,
2028 .module = THIS_MODULE,
2029};
2030#endif
2031
2032int ovs_ct_init(struct net *net)
2033{
2034 unsigned int n_bits = sizeof(struct ovs_key_ct_labels) * BITS_PER_BYTE;
2035 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2036 int err = 0;
2037
2038 if (nf_connlabels_get(net, n_bits - 1)) {
2039 ovs_net->xt_label = false;
2040 OVS_NLERR(true, "Failed to set connlabel length");
2041 } else {
2042 ovs_net->xt_label = true;
2043 }
2044
2045#if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2046 err = ovs_ct_limit_init(net, ovs_net);
2047 if (err && ovs_net->xt_label)
2048 nf_connlabels_put(net);
2049#endif
2050 return err;
2051}
2052
2053/* Must be called with ovs_mutex held. Detaches the RCU-protected
2054 * ct_limit_info and stores it in ovs_net->ct_limit_exit_data for
2055 * ovs_ct_exit_finish() to complete the teardown after an RCU grace period.
2056 */
2057void ovs_ct_exit_start(struct net *net __maybe_unused)
2058{
2059#if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2060 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2061
2062 ovs_net->ct_limit_exit_data = ovs_ct_limit_exit_start(ovs_net);
2063#endif
2064}
2065
2066/* Completes the CT limit teardown. The pernet core guarantees an RCU
2067 * grace period between detaching the state in ovs_ct_exit_start() and
2068 * this call, so no RCU readers remain.
2069 */
2070void ovs_ct_exit_finish(struct net *net)
2071{
2072 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2073
2074#if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2075 ovs_ct_limit_exit_finish(net, ovs_net->ct_limit_exit_data);
2076#endif
2077
2078 if (ovs_net->xt_label)
2079 nf_connlabels_put(net);
2080}