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 <net/ip.h>
13 #include <net/genetlink.h>
14 #include <net/netfilter/nf_conntrack_core.h>
15 #include <net/netfilter/nf_conntrack_count.h>
16 #include <net/netfilter/nf_conntrack_helper.h>
17 #include <net/netfilter/nf_conntrack_labels.h>
18 #include <net/netfilter/nf_conntrack_seqadj.h>
19 #include <net/netfilter/nf_conntrack_timeout.h>
20 #include <net/netfilter/nf_conntrack_zones.h>
21 #include <net/netfilter/ipv6/nf_defrag_ipv6.h>
22 #include <net/ipv6_frag.h>
23
24 #if IS_ENABLED(CONFIG_NF_NAT)
25 #include <net/netfilter/nf_nat.h>
26 #endif
27
28 #include "datapath.h"
29 #include "conntrack.h"
30 #include "flow.h"
31 #include "flow_netlink.h"
32
33 struct ovs_ct_len_tbl {
34 int maxlen;
35 int minlen;
36 };
37
38 /* Metadata mark for masked write to conntrack mark */
39 struct md_mark {
40 u32 value;
41 u32 mask;
42 };
43
44 /* Metadata label for masked write to conntrack label. */
45 struct md_labels {
46 struct ovs_key_ct_labels value;
47 struct ovs_key_ct_labels mask;
48 };
49
50 enum ovs_ct_nat {
51 OVS_CT_NAT = 1 << 0, /* NAT for committed connections only. */
52 OVS_CT_SRC_NAT = 1 << 1, /* Source NAT for NEW connections. */
53 OVS_CT_DST_NAT = 1 << 2, /* Destination NAT for NEW connections. */
54 };
55
56 /* Conntrack action context for execution. */
57 struct ovs_conntrack_info {
58 struct nf_conntrack_helper *helper;
59 struct nf_conntrack_zone zone;
60 struct nf_conn *ct;
61 u8 commit : 1;
62 u8 nat : 3; /* enum ovs_ct_nat */
63 u8 force : 1;
64 u8 have_eventmask : 1;
65 u16 family;
66 u32 eventmask; /* Mask of 1 << IPCT_*. */
67 struct md_mark mark;
68 struct md_labels labels;
69 char timeout[CTNL_TIMEOUT_NAME_MAX];
70 struct nf_ct_timeout *nf_ct_timeout;
71 #if IS_ENABLED(CONFIG_NF_NAT)
72 struct nf_nat_range2 range; /* Only present for SRC NAT and DST NAT. */
73 #endif
74 };
75
76 #if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
77 #define OVS_CT_LIMIT_UNLIMITED 0
78 #define OVS_CT_LIMIT_DEFAULT OVS_CT_LIMIT_UNLIMITED
79 #define CT_LIMIT_HASH_BUCKETS 512
80 static DEFINE_STATIC_KEY_FALSE(ovs_ct_limit_enabled);
81
82 struct ovs_ct_limit {
83 /* Elements in ovs_ct_limit_info->limits hash table */
84 struct hlist_node hlist_node;
85 struct rcu_head rcu;
86 u16 zone;
87 u32 limit;
88 };
89
90 struct ovs_ct_limit_info {
91 u32 default_limit;
92 struct hlist_head *limits;
93 struct nf_conncount_data *data;
94 };
95
96 static const struct nla_policy ct_limit_policy[OVS_CT_LIMIT_ATTR_MAX + 1] = {
97 [OVS_CT_LIMIT_ATTR_ZONE_LIMIT] = { .type = NLA_NESTED, },
98 };
99 #endif
100
101 static bool labels_nonzero(const struct ovs_key_ct_labels *labels);
102
103 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info);
104
key_to_nfproto(const struct sw_flow_key * key)105 static u16 key_to_nfproto(const struct sw_flow_key *key)
106 {
107 switch (ntohs(key->eth.type)) {
108 case ETH_P_IP:
109 return NFPROTO_IPV4;
110 case ETH_P_IPV6:
111 return NFPROTO_IPV6;
112 default:
113 return NFPROTO_UNSPEC;
114 }
115 }
116
117 /* Map SKB connection state into the values used by flow definition. */
ovs_ct_get_state(enum ip_conntrack_info ctinfo)118 static u8 ovs_ct_get_state(enum ip_conntrack_info ctinfo)
119 {
120 u8 ct_state = OVS_CS_F_TRACKED;
121
122 switch (ctinfo) {
123 case IP_CT_ESTABLISHED_REPLY:
124 case IP_CT_RELATED_REPLY:
125 ct_state |= OVS_CS_F_REPLY_DIR;
126 break;
127 default:
128 break;
129 }
130
131 switch (ctinfo) {
132 case IP_CT_ESTABLISHED:
133 case IP_CT_ESTABLISHED_REPLY:
134 ct_state |= OVS_CS_F_ESTABLISHED;
135 break;
136 case IP_CT_RELATED:
137 case IP_CT_RELATED_REPLY:
138 ct_state |= OVS_CS_F_RELATED;
139 break;
140 case IP_CT_NEW:
141 ct_state |= OVS_CS_F_NEW;
142 break;
143 default:
144 break;
145 }
146
147 return ct_state;
148 }
149
ovs_ct_get_mark(const struct nf_conn * ct)150 static u32 ovs_ct_get_mark(const struct nf_conn *ct)
151 {
152 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
153 return ct ? ct->mark : 0;
154 #else
155 return 0;
156 #endif
157 }
158
159 /* Guard against conntrack labels max size shrinking below 128 bits. */
160 #if NF_CT_LABELS_MAX_SIZE < 16
161 #error NF_CT_LABELS_MAX_SIZE must be at least 16 bytes
162 #endif
163
ovs_ct_get_labels(const struct nf_conn * ct,struct ovs_key_ct_labels * labels)164 static void ovs_ct_get_labels(const struct nf_conn *ct,
165 struct ovs_key_ct_labels *labels)
166 {
167 struct nf_conn_labels *cl = ct ? nf_ct_labels_find(ct) : NULL;
168
169 if (cl)
170 memcpy(labels, cl->bits, OVS_CT_LABELS_LEN);
171 else
172 memset(labels, 0, OVS_CT_LABELS_LEN);
173 }
174
__ovs_ct_update_key_orig_tp(struct sw_flow_key * key,const struct nf_conntrack_tuple * orig,u8 icmp_proto)175 static void __ovs_ct_update_key_orig_tp(struct sw_flow_key *key,
176 const struct nf_conntrack_tuple *orig,
177 u8 icmp_proto)
178 {
179 key->ct_orig_proto = orig->dst.protonum;
180 if (orig->dst.protonum == icmp_proto) {
181 key->ct.orig_tp.src = htons(orig->dst.u.icmp.type);
182 key->ct.orig_tp.dst = htons(orig->dst.u.icmp.code);
183 } else {
184 key->ct.orig_tp.src = orig->src.u.all;
185 key->ct.orig_tp.dst = orig->dst.u.all;
186 }
187 }
188
__ovs_ct_update_key(struct sw_flow_key * key,u8 state,const struct nf_conntrack_zone * zone,const struct nf_conn * ct)189 static void __ovs_ct_update_key(struct sw_flow_key *key, u8 state,
190 const struct nf_conntrack_zone *zone,
191 const struct nf_conn *ct)
192 {
193 key->ct_state = state;
194 key->ct_zone = zone->id;
195 key->ct.mark = ovs_ct_get_mark(ct);
196 ovs_ct_get_labels(ct, &key->ct.labels);
197
198 if (ct) {
199 const struct nf_conntrack_tuple *orig;
200
201 /* Use the master if we have one. */
202 if (ct->master)
203 ct = ct->master;
204 orig = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
205
206 /* IP version must match with the master connection. */
207 if (key->eth.type == htons(ETH_P_IP) &&
208 nf_ct_l3num(ct) == NFPROTO_IPV4) {
209 key->ipv4.ct_orig.src = orig->src.u3.ip;
210 key->ipv4.ct_orig.dst = orig->dst.u3.ip;
211 __ovs_ct_update_key_orig_tp(key, orig, IPPROTO_ICMP);
212 return;
213 } else if (key->eth.type == htons(ETH_P_IPV6) &&
214 !sw_flow_key_is_nd(key) &&
215 nf_ct_l3num(ct) == NFPROTO_IPV6) {
216 key->ipv6.ct_orig.src = orig->src.u3.in6;
217 key->ipv6.ct_orig.dst = orig->dst.u3.in6;
218 __ovs_ct_update_key_orig_tp(key, orig, NEXTHDR_ICMP);
219 return;
220 }
221 }
222 /* Clear 'ct_orig_proto' to mark the non-existence of conntrack
223 * original direction key fields.
224 */
225 key->ct_orig_proto = 0;
226 }
227
228 /* Update 'key' based on skb->_nfct. If 'post_ct' is true, then OVS has
229 * previously sent the packet to conntrack via the ct action. If
230 * 'keep_nat_flags' is true, the existing NAT flags retained, else they are
231 * initialized from the connection status.
232 */
ovs_ct_update_key(const struct sk_buff * skb,const struct ovs_conntrack_info * info,struct sw_flow_key * key,bool post_ct,bool keep_nat_flags)233 static void ovs_ct_update_key(const struct sk_buff *skb,
234 const struct ovs_conntrack_info *info,
235 struct sw_flow_key *key, bool post_ct,
236 bool keep_nat_flags)
237 {
238 const struct nf_conntrack_zone *zone = &nf_ct_zone_dflt;
239 enum ip_conntrack_info ctinfo;
240 struct nf_conn *ct;
241 u8 state = 0;
242
243 ct = nf_ct_get(skb, &ctinfo);
244 if (ct) {
245 state = ovs_ct_get_state(ctinfo);
246 /* All unconfirmed entries are NEW connections. */
247 if (!nf_ct_is_confirmed(ct))
248 state |= OVS_CS_F_NEW;
249 /* OVS persists the related flag for the duration of the
250 * connection.
251 */
252 if (ct->master)
253 state |= OVS_CS_F_RELATED;
254 if (keep_nat_flags) {
255 state |= key->ct_state & OVS_CS_F_NAT_MASK;
256 } else {
257 if (ct->status & IPS_SRC_NAT)
258 state |= OVS_CS_F_SRC_NAT;
259 if (ct->status & IPS_DST_NAT)
260 state |= OVS_CS_F_DST_NAT;
261 }
262 zone = nf_ct_zone(ct);
263 } else if (post_ct) {
264 state = OVS_CS_F_TRACKED | OVS_CS_F_INVALID;
265 if (info)
266 zone = &info->zone;
267 }
268 __ovs_ct_update_key(key, state, zone, ct);
269 }
270
271 /* This is called to initialize CT key fields possibly coming in from the local
272 * stack.
273 */
ovs_ct_fill_key(const struct sk_buff * skb,struct sw_flow_key * key)274 void ovs_ct_fill_key(const struct sk_buff *skb, struct sw_flow_key *key)
275 {
276 ovs_ct_update_key(skb, NULL, key, false, false);
277 }
278
279 #define IN6_ADDR_INITIALIZER(ADDR) \
280 { (ADDR).s6_addr32[0], (ADDR).s6_addr32[1], \
281 (ADDR).s6_addr32[2], (ADDR).s6_addr32[3] }
282
ovs_ct_put_key(const struct sw_flow_key * swkey,const struct sw_flow_key * output,struct sk_buff * skb)283 int ovs_ct_put_key(const struct sw_flow_key *swkey,
284 const struct sw_flow_key *output, struct sk_buff *skb)
285 {
286 if (nla_put_u32(skb, OVS_KEY_ATTR_CT_STATE, output->ct_state))
287 return -EMSGSIZE;
288
289 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
290 nla_put_u16(skb, OVS_KEY_ATTR_CT_ZONE, output->ct_zone))
291 return -EMSGSIZE;
292
293 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
294 nla_put_u32(skb, OVS_KEY_ATTR_CT_MARK, output->ct.mark))
295 return -EMSGSIZE;
296
297 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
298 nla_put(skb, OVS_KEY_ATTR_CT_LABELS, sizeof(output->ct.labels),
299 &output->ct.labels))
300 return -EMSGSIZE;
301
302 if (swkey->ct_orig_proto) {
303 if (swkey->eth.type == htons(ETH_P_IP)) {
304 struct ovs_key_ct_tuple_ipv4 orig = {
305 output->ipv4.ct_orig.src,
306 output->ipv4.ct_orig.dst,
307 output->ct.orig_tp.src,
308 output->ct.orig_tp.dst,
309 output->ct_orig_proto,
310 };
311 if (nla_put(skb, OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4,
312 sizeof(orig), &orig))
313 return -EMSGSIZE;
314 } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
315 struct ovs_key_ct_tuple_ipv6 orig = {
316 IN6_ADDR_INITIALIZER(output->ipv6.ct_orig.src),
317 IN6_ADDR_INITIALIZER(output->ipv6.ct_orig.dst),
318 output->ct.orig_tp.src,
319 output->ct.orig_tp.dst,
320 output->ct_orig_proto,
321 };
322 if (nla_put(skb, OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6,
323 sizeof(orig), &orig))
324 return -EMSGSIZE;
325 }
326 }
327
328 return 0;
329 }
330
ovs_ct_set_mark(struct nf_conn * ct,struct sw_flow_key * key,u32 ct_mark,u32 mask)331 static int ovs_ct_set_mark(struct nf_conn *ct, struct sw_flow_key *key,
332 u32 ct_mark, u32 mask)
333 {
334 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
335 u32 new_mark;
336
337 new_mark = ct_mark | (ct->mark & ~(mask));
338 if (ct->mark != new_mark) {
339 ct->mark = new_mark;
340 if (nf_ct_is_confirmed(ct))
341 nf_conntrack_event_cache(IPCT_MARK, ct);
342 key->ct.mark = new_mark;
343 }
344
345 return 0;
346 #else
347 return -ENOTSUPP;
348 #endif
349 }
350
ovs_ct_get_conn_labels(struct nf_conn * ct)351 static struct nf_conn_labels *ovs_ct_get_conn_labels(struct nf_conn *ct)
352 {
353 struct nf_conn_labels *cl;
354
355 cl = nf_ct_labels_find(ct);
356 if (!cl) {
357 nf_ct_labels_ext_add(ct);
358 cl = nf_ct_labels_find(ct);
359 }
360
361 return cl;
362 }
363
364 /* Initialize labels for a new, yet to be committed conntrack entry. Note that
365 * since the new connection is not yet confirmed, and thus no-one else has
366 * access to it's labels, we simply write them over.
367 */
ovs_ct_init_labels(struct nf_conn * ct,struct sw_flow_key * key,const struct ovs_key_ct_labels * labels,const struct ovs_key_ct_labels * mask)368 static int ovs_ct_init_labels(struct nf_conn *ct, struct sw_flow_key *key,
369 const struct ovs_key_ct_labels *labels,
370 const struct ovs_key_ct_labels *mask)
371 {
372 struct nf_conn_labels *cl, *master_cl;
373 bool have_mask = labels_nonzero(mask);
374
375 /* Inherit master's labels to the related connection? */
376 master_cl = ct->master ? nf_ct_labels_find(ct->master) : NULL;
377
378 if (!master_cl && !have_mask)
379 return 0; /* Nothing to do. */
380
381 cl = ovs_ct_get_conn_labels(ct);
382 if (!cl)
383 return -ENOSPC;
384
385 /* Inherit the master's labels, if any. */
386 if (master_cl)
387 *cl = *master_cl;
388
389 if (have_mask) {
390 u32 *dst = (u32 *)cl->bits;
391 int i;
392
393 for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
394 dst[i] = (dst[i] & ~mask->ct_labels_32[i]) |
395 (labels->ct_labels_32[i]
396 & mask->ct_labels_32[i]);
397 }
398
399 /* Labels are included in the IPCTNL_MSG_CT_NEW event only if the
400 * IPCT_LABEL bit is set in the event cache.
401 */
402 nf_conntrack_event_cache(IPCT_LABEL, ct);
403
404 memcpy(&key->ct.labels, cl->bits, OVS_CT_LABELS_LEN);
405
406 return 0;
407 }
408
ovs_ct_set_labels(struct nf_conn * ct,struct sw_flow_key * key,const struct ovs_key_ct_labels * labels,const struct ovs_key_ct_labels * mask)409 static int ovs_ct_set_labels(struct nf_conn *ct, struct sw_flow_key *key,
410 const struct ovs_key_ct_labels *labels,
411 const struct ovs_key_ct_labels *mask)
412 {
413 struct nf_conn_labels *cl;
414 int err;
415
416 cl = ovs_ct_get_conn_labels(ct);
417 if (!cl)
418 return -ENOSPC;
419
420 err = nf_connlabels_replace(ct, labels->ct_labels_32,
421 mask->ct_labels_32,
422 OVS_CT_LABELS_LEN_32);
423 if (err)
424 return err;
425
426 memcpy(&key->ct.labels, cl->bits, OVS_CT_LABELS_LEN);
427
428 return 0;
429 }
430
431 /* 'skb' should already be pulled to nh_ofs. */
ovs_ct_helper(struct sk_buff * skb,u16 proto)432 static int ovs_ct_helper(struct sk_buff *skb, u16 proto)
433 {
434 const struct nf_conntrack_helper *helper;
435 const struct nf_conn_help *help;
436 enum ip_conntrack_info ctinfo;
437 unsigned int protoff;
438 struct nf_conn *ct;
439 int err;
440
441 ct = nf_ct_get(skb, &ctinfo);
442 if (!ct || ctinfo == IP_CT_RELATED_REPLY)
443 return NF_ACCEPT;
444
445 help = nfct_help(ct);
446 if (!help)
447 return NF_ACCEPT;
448
449 helper = rcu_dereference(help->helper);
450 if (!helper)
451 return NF_ACCEPT;
452
453 switch (proto) {
454 case NFPROTO_IPV4:
455 protoff = ip_hdrlen(skb);
456 break;
457 case NFPROTO_IPV6: {
458 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
459 __be16 frag_off;
460 int ofs;
461
462 ofs = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr,
463 &frag_off);
464 if (ofs < 0 || (frag_off & htons(~0x7)) != 0) {
465 pr_debug("proto header not found\n");
466 return NF_ACCEPT;
467 }
468 protoff = ofs;
469 break;
470 }
471 default:
472 WARN_ONCE(1, "helper invoked on non-IP family!");
473 return NF_DROP;
474 }
475
476 err = helper->help(skb, protoff, ct, ctinfo);
477 if (err != NF_ACCEPT)
478 return err;
479
480 /* Adjust seqs after helper. This is needed due to some helpers (e.g.,
481 * FTP with NAT) adusting the TCP payload size when mangling IP
482 * addresses and/or port numbers in the text-based control connection.
483 */
484 if (test_bit(IPS_SEQ_ADJUST_BIT, &ct->status) &&
485 !nf_ct_seq_adjust(skb, ct, ctinfo, protoff))
486 return NF_DROP;
487 return NF_ACCEPT;
488 }
489
490 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
491 * value if 'skb' is freed.
492 */
handle_fragments(struct net * net,struct sw_flow_key * key,u16 zone,struct sk_buff * skb)493 static int handle_fragments(struct net *net, struct sw_flow_key *key,
494 u16 zone, struct sk_buff *skb)
495 {
496 struct ovs_skb_cb ovs_cb = *OVS_CB(skb);
497 int err;
498
499 if (key->eth.type == htons(ETH_P_IP)) {
500 enum ip_defrag_users user = IP_DEFRAG_CONNTRACK_IN + zone;
501
502 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
503 err = ip_defrag(net, skb, user);
504 if (err)
505 return err;
506
507 ovs_cb.mru = IPCB(skb)->frag_max_size;
508 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
509 } else if (key->eth.type == htons(ETH_P_IPV6)) {
510 enum ip6_defrag_users user = IP6_DEFRAG_CONNTRACK_IN + zone;
511
512 memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm));
513 err = nf_ct_frag6_gather(net, skb, user);
514 if (err) {
515 if (err != -EINPROGRESS)
516 kfree_skb(skb);
517 return err;
518 }
519
520 key->ip.proto = ipv6_hdr(skb)->nexthdr;
521 ovs_cb.mru = IP6CB(skb)->frag_max_size;
522 #endif
523 } else {
524 kfree_skb(skb);
525 return -EPFNOSUPPORT;
526 }
527
528 /* The key extracted from the fragment that completed this datagram
529 * likely didn't have an L4 header, so regenerate it.
530 */
531 ovs_flow_key_update_l3l4(skb, key);
532
533 key->ip.frag = OVS_FRAG_TYPE_NONE;
534 skb_clear_hash(skb);
535 skb->ignore_df = 1;
536 *OVS_CB(skb) = ovs_cb;
537
538 return 0;
539 }
540
541 static struct nf_conntrack_expect *
ovs_ct_expect_find(struct net * net,const struct nf_conntrack_zone * zone,u16 proto,const struct sk_buff * skb)542 ovs_ct_expect_find(struct net *net, const struct nf_conntrack_zone *zone,
543 u16 proto, const struct sk_buff *skb)
544 {
545 struct nf_conntrack_tuple tuple;
546 struct nf_conntrack_expect *exp;
547
548 if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), proto, net, &tuple))
549 return NULL;
550
551 exp = __nf_ct_expect_find(net, zone, &tuple);
552 if (exp) {
553 struct nf_conntrack_tuple_hash *h;
554
555 /* Delete existing conntrack entry, if it clashes with the
556 * expectation. This can happen since conntrack ALGs do not
557 * check for clashes between (new) expectations and existing
558 * conntrack entries. nf_conntrack_in() will check the
559 * expectations only if a conntrack entry can not be found,
560 * which can lead to OVS finding the expectation (here) in the
561 * init direction, but which will not be removed by the
562 * nf_conntrack_in() call, if a matching conntrack entry is
563 * found instead. In this case all init direction packets
564 * would be reported as new related packets, while reply
565 * direction packets would be reported as un-related
566 * established packets.
567 */
568 h = nf_conntrack_find_get(net, zone, &tuple);
569 if (h) {
570 struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
571
572 nf_ct_delete(ct, 0, 0);
573 nf_conntrack_put(&ct->ct_general);
574 }
575 }
576
577 return exp;
578 }
579
580 /* This replicates logic from nf_conntrack_core.c that is not exported. */
581 static enum ip_conntrack_info
ovs_ct_get_info(const struct nf_conntrack_tuple_hash * h)582 ovs_ct_get_info(const struct nf_conntrack_tuple_hash *h)
583 {
584 const struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
585
586 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY)
587 return IP_CT_ESTABLISHED_REPLY;
588 /* Once we've had two way comms, always ESTABLISHED. */
589 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status))
590 return IP_CT_ESTABLISHED;
591 if (test_bit(IPS_EXPECTED_BIT, &ct->status))
592 return IP_CT_RELATED;
593 return IP_CT_NEW;
594 }
595
596 /* Find an existing connection which this packet belongs to without
597 * re-attributing statistics or modifying the connection state. This allows an
598 * skb->_nfct lost due to an upcall to be recovered during actions execution.
599 *
600 * Must be called with rcu_read_lock.
601 *
602 * On success, populates skb->_nfct and returns the connection. Returns NULL
603 * if there is no existing entry.
604 */
605 static struct nf_conn *
ovs_ct_find_existing(struct net * net,const struct nf_conntrack_zone * zone,u8 l3num,struct sk_buff * skb,bool natted)606 ovs_ct_find_existing(struct net *net, const struct nf_conntrack_zone *zone,
607 u8 l3num, struct sk_buff *skb, bool natted)
608 {
609 struct nf_conntrack_tuple tuple;
610 struct nf_conntrack_tuple_hash *h;
611 struct nf_conn *ct;
612
613 if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), l3num,
614 net, &tuple)) {
615 pr_debug("ovs_ct_find_existing: Can't get tuple\n");
616 return NULL;
617 }
618
619 /* Must invert the tuple if skb has been transformed by NAT. */
620 if (natted) {
621 struct nf_conntrack_tuple inverse;
622
623 if (!nf_ct_invert_tuple(&inverse, &tuple)) {
624 pr_debug("ovs_ct_find_existing: Inversion failed!\n");
625 return NULL;
626 }
627 tuple = inverse;
628 }
629
630 /* look for tuple match */
631 h = nf_conntrack_find_get(net, zone, &tuple);
632 if (!h)
633 return NULL; /* Not found. */
634
635 ct = nf_ct_tuplehash_to_ctrack(h);
636
637 /* Inverted packet tuple matches the reverse direction conntrack tuple,
638 * select the other tuplehash to get the right 'ctinfo' bits for this
639 * packet.
640 */
641 if (natted)
642 h = &ct->tuplehash[!h->tuple.dst.dir];
643
644 nf_ct_set(skb, ct, ovs_ct_get_info(h));
645 return ct;
646 }
647
648 static
ovs_ct_executed(struct net * net,const struct sw_flow_key * key,const struct ovs_conntrack_info * info,struct sk_buff * skb,bool * ct_executed)649 struct nf_conn *ovs_ct_executed(struct net *net,
650 const struct sw_flow_key *key,
651 const struct ovs_conntrack_info *info,
652 struct sk_buff *skb,
653 bool *ct_executed)
654 {
655 struct nf_conn *ct = NULL;
656
657 /* If no ct, check if we have evidence that an existing conntrack entry
658 * might be found for this skb. This happens when we lose a skb->_nfct
659 * due to an upcall, or if the direction is being forced. If the
660 * connection was not confirmed, it is not cached and needs to be run
661 * through conntrack again.
662 */
663 *ct_executed = (key->ct_state & OVS_CS_F_TRACKED) &&
664 !(key->ct_state & OVS_CS_F_INVALID) &&
665 (key->ct_zone == info->zone.id);
666
667 if (*ct_executed || (!key->ct_state && info->force)) {
668 ct = ovs_ct_find_existing(net, &info->zone, info->family, skb,
669 !!(key->ct_state &
670 OVS_CS_F_NAT_MASK));
671 }
672
673 return ct;
674 }
675
676 /* Determine whether skb->_nfct is equal to the result of conntrack lookup. */
skb_nfct_cached(struct net * net,const struct sw_flow_key * key,const struct ovs_conntrack_info * info,struct sk_buff * skb)677 static bool skb_nfct_cached(struct net *net,
678 const struct sw_flow_key *key,
679 const struct ovs_conntrack_info *info,
680 struct sk_buff *skb)
681 {
682 enum ip_conntrack_info ctinfo;
683 struct nf_conn *ct;
684 bool ct_executed = true;
685
686 ct = nf_ct_get(skb, &ctinfo);
687 if (!ct)
688 ct = ovs_ct_executed(net, key, info, skb, &ct_executed);
689
690 if (ct)
691 nf_ct_get(skb, &ctinfo);
692 else
693 return false;
694
695 if (!net_eq(net, read_pnet(&ct->ct_net)))
696 return false;
697 if (!nf_ct_zone_equal_any(info->ct, nf_ct_zone(ct)))
698 return false;
699 if (info->helper) {
700 struct nf_conn_help *help;
701
702 help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER);
703 if (help && rcu_access_pointer(help->helper) != info->helper)
704 return false;
705 }
706 if (info->nf_ct_timeout) {
707 struct nf_conn_timeout *timeout_ext;
708
709 timeout_ext = nf_ct_timeout_find(ct);
710 if (!timeout_ext || info->nf_ct_timeout !=
711 rcu_dereference(timeout_ext->timeout))
712 return false;
713 }
714 /* Force conntrack entry direction to the current packet? */
715 if (info->force && CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) {
716 /* Delete the conntrack entry if confirmed, else just release
717 * the reference.
718 */
719 if (nf_ct_is_confirmed(ct))
720 nf_ct_delete(ct, 0, 0);
721
722 nf_conntrack_put(&ct->ct_general);
723 nf_ct_set(skb, NULL, 0);
724 return false;
725 }
726
727 return ct_executed;
728 }
729
730 #if IS_ENABLED(CONFIG_NF_NAT)
731 /* Modelled after nf_nat_ipv[46]_fn().
732 * range is only used for new, uninitialized NAT state.
733 * Returns either NF_ACCEPT or NF_DROP.
734 */
ovs_ct_nat_execute(struct sk_buff * skb,struct nf_conn * ct,enum ip_conntrack_info ctinfo,const struct nf_nat_range2 * range,enum nf_nat_manip_type maniptype)735 static int ovs_ct_nat_execute(struct sk_buff *skb, struct nf_conn *ct,
736 enum ip_conntrack_info ctinfo,
737 const struct nf_nat_range2 *range,
738 enum nf_nat_manip_type maniptype)
739 {
740 int hooknum, nh_off, err = NF_ACCEPT;
741
742 nh_off = skb_network_offset(skb);
743 skb_pull_rcsum(skb, nh_off);
744
745 /* See HOOK2MANIP(). */
746 if (maniptype == NF_NAT_MANIP_SRC)
747 hooknum = NF_INET_LOCAL_IN; /* Source NAT */
748 else
749 hooknum = NF_INET_LOCAL_OUT; /* Destination NAT */
750
751 switch (ctinfo) {
752 case IP_CT_RELATED:
753 case IP_CT_RELATED_REPLY:
754 if (IS_ENABLED(CONFIG_NF_NAT) &&
755 skb->protocol == htons(ETH_P_IP) &&
756 ip_hdr(skb)->protocol == IPPROTO_ICMP) {
757 if (!nf_nat_icmp_reply_translation(skb, ct, ctinfo,
758 hooknum))
759 err = NF_DROP;
760 goto push;
761 } else if (IS_ENABLED(CONFIG_IPV6) &&
762 skb->protocol == htons(ETH_P_IPV6)) {
763 __be16 frag_off;
764 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
765 int hdrlen = ipv6_skip_exthdr(skb,
766 sizeof(struct ipv6hdr),
767 &nexthdr, &frag_off);
768
769 if (hdrlen >= 0 && nexthdr == IPPROTO_ICMPV6) {
770 if (!nf_nat_icmpv6_reply_translation(skb, ct,
771 ctinfo,
772 hooknum,
773 hdrlen))
774 err = NF_DROP;
775 goto push;
776 }
777 }
778 /* Non-ICMP, fall thru to initialize if needed. */
779 /* fall through */
780 case IP_CT_NEW:
781 /* Seen it before? This can happen for loopback, retrans,
782 * or local packets.
783 */
784 if (!nf_nat_initialized(ct, maniptype)) {
785 /* Initialize according to the NAT action. */
786 err = (range && range->flags & NF_NAT_RANGE_MAP_IPS)
787 /* Action is set up to establish a new
788 * mapping.
789 */
790 ? nf_nat_setup_info(ct, range, maniptype)
791 : nf_nat_alloc_null_binding(ct, hooknum);
792 if (err != NF_ACCEPT)
793 goto push;
794 }
795 break;
796
797 case IP_CT_ESTABLISHED:
798 case IP_CT_ESTABLISHED_REPLY:
799 break;
800
801 default:
802 err = NF_DROP;
803 goto push;
804 }
805
806 err = nf_nat_packet(ct, ctinfo, hooknum, skb);
807 push:
808 skb_push(skb, nh_off);
809 skb_postpush_rcsum(skb, skb->data, nh_off);
810
811 return err;
812 }
813
ovs_nat_update_key(struct sw_flow_key * key,const struct sk_buff * skb,enum nf_nat_manip_type maniptype)814 static void ovs_nat_update_key(struct sw_flow_key *key,
815 const struct sk_buff *skb,
816 enum nf_nat_manip_type maniptype)
817 {
818 if (maniptype == NF_NAT_MANIP_SRC) {
819 __be16 src;
820
821 key->ct_state |= OVS_CS_F_SRC_NAT;
822 if (key->eth.type == htons(ETH_P_IP))
823 key->ipv4.addr.src = ip_hdr(skb)->saddr;
824 else if (key->eth.type == htons(ETH_P_IPV6))
825 memcpy(&key->ipv6.addr.src, &ipv6_hdr(skb)->saddr,
826 sizeof(key->ipv6.addr.src));
827 else
828 return;
829
830 if (key->ip.proto == IPPROTO_UDP)
831 src = udp_hdr(skb)->source;
832 else if (key->ip.proto == IPPROTO_TCP)
833 src = tcp_hdr(skb)->source;
834 else if (key->ip.proto == IPPROTO_SCTP)
835 src = sctp_hdr(skb)->source;
836 else
837 return;
838
839 key->tp.src = src;
840 } else {
841 __be16 dst;
842
843 key->ct_state |= OVS_CS_F_DST_NAT;
844 if (key->eth.type == htons(ETH_P_IP))
845 key->ipv4.addr.dst = ip_hdr(skb)->daddr;
846 else if (key->eth.type == htons(ETH_P_IPV6))
847 memcpy(&key->ipv6.addr.dst, &ipv6_hdr(skb)->daddr,
848 sizeof(key->ipv6.addr.dst));
849 else
850 return;
851
852 if (key->ip.proto == IPPROTO_UDP)
853 dst = udp_hdr(skb)->dest;
854 else if (key->ip.proto == IPPROTO_TCP)
855 dst = tcp_hdr(skb)->dest;
856 else if (key->ip.proto == IPPROTO_SCTP)
857 dst = sctp_hdr(skb)->dest;
858 else
859 return;
860
861 key->tp.dst = dst;
862 }
863 }
864
865 /* Returns NF_DROP if the packet should be dropped, NF_ACCEPT otherwise. */
ovs_ct_nat(struct net * net,struct sw_flow_key * key,const struct ovs_conntrack_info * info,struct sk_buff * skb,struct nf_conn * ct,enum ip_conntrack_info ctinfo)866 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
867 const struct ovs_conntrack_info *info,
868 struct sk_buff *skb, struct nf_conn *ct,
869 enum ip_conntrack_info ctinfo)
870 {
871 enum nf_nat_manip_type maniptype;
872 int err;
873
874 /* Add NAT extension if not confirmed yet. */
875 if (!nf_ct_is_confirmed(ct) && !nf_ct_nat_ext_add(ct))
876 return NF_ACCEPT; /* Can't NAT. */
877
878 /* Determine NAT type.
879 * Check if the NAT type can be deduced from the tracked connection.
880 * Make sure new expected connections (IP_CT_RELATED) are NATted only
881 * when committing.
882 */
883 if (info->nat & OVS_CT_NAT && ctinfo != IP_CT_NEW &&
884 ct->status & IPS_NAT_MASK &&
885 (ctinfo != IP_CT_RELATED || info->commit)) {
886 /* NAT an established or related connection like before. */
887 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY)
888 /* This is the REPLY direction for a connection
889 * for which NAT was applied in the forward
890 * direction. Do the reverse NAT.
891 */
892 maniptype = ct->status & IPS_SRC_NAT
893 ? NF_NAT_MANIP_DST : NF_NAT_MANIP_SRC;
894 else
895 maniptype = ct->status & IPS_SRC_NAT
896 ? NF_NAT_MANIP_SRC : NF_NAT_MANIP_DST;
897 } else if (info->nat & OVS_CT_SRC_NAT) {
898 maniptype = NF_NAT_MANIP_SRC;
899 } else if (info->nat & OVS_CT_DST_NAT) {
900 maniptype = NF_NAT_MANIP_DST;
901 } else {
902 return NF_ACCEPT; /* Connection is not NATed. */
903 }
904 err = ovs_ct_nat_execute(skb, ct, ctinfo, &info->range, maniptype);
905
906 if (err == NF_ACCEPT &&
907 ct->status & IPS_SRC_NAT && ct->status & IPS_DST_NAT) {
908 if (maniptype == NF_NAT_MANIP_SRC)
909 maniptype = NF_NAT_MANIP_DST;
910 else
911 maniptype = NF_NAT_MANIP_SRC;
912
913 err = ovs_ct_nat_execute(skb, ct, ctinfo, &info->range,
914 maniptype);
915 }
916
917 /* Mark NAT done if successful and update the flow key. */
918 if (err == NF_ACCEPT)
919 ovs_nat_update_key(key, skb, maniptype);
920
921 return err;
922 }
923 #else /* !CONFIG_NF_NAT */
ovs_ct_nat(struct net * net,struct sw_flow_key * key,const struct ovs_conntrack_info * info,struct sk_buff * skb,struct nf_conn * ct,enum ip_conntrack_info ctinfo)924 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
925 const struct ovs_conntrack_info *info,
926 struct sk_buff *skb, struct nf_conn *ct,
927 enum ip_conntrack_info ctinfo)
928 {
929 return NF_ACCEPT;
930 }
931 #endif
932
933 /* Pass 'skb' through conntrack in 'net', using zone configured in 'info', if
934 * not done already. Update key with new CT state after passing the packet
935 * through conntrack.
936 * Note that if the packet is deemed invalid by conntrack, skb->_nfct will be
937 * set to NULL and 0 will be returned.
938 */
__ovs_ct_lookup(struct net * net,struct sw_flow_key * key,const struct ovs_conntrack_info * info,struct sk_buff * skb)939 static int __ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
940 const struct ovs_conntrack_info *info,
941 struct sk_buff *skb)
942 {
943 /* If we are recirculating packets to match on conntrack fields and
944 * committing with a separate conntrack action, then we don't need to
945 * actually run the packet through conntrack twice unless it's for a
946 * different zone.
947 */
948 bool cached = skb_nfct_cached(net, key, info, skb);
949 enum ip_conntrack_info ctinfo;
950 struct nf_conn *ct;
951
952 if (!cached) {
953 struct nf_hook_state state = {
954 .hook = NF_INET_PRE_ROUTING,
955 .pf = info->family,
956 .net = net,
957 };
958 struct nf_conn *tmpl = info->ct;
959 int err;
960
961 /* Associate skb with specified zone. */
962 if (tmpl) {
963 if (skb_nfct(skb))
964 nf_conntrack_put(skb_nfct(skb));
965 nf_conntrack_get(&tmpl->ct_general);
966 nf_ct_set(skb, tmpl, IP_CT_NEW);
967 }
968
969 err = nf_conntrack_in(skb, &state);
970 if (err != NF_ACCEPT)
971 return -ENOENT;
972
973 /* Clear CT state NAT flags to mark that we have not yet done
974 * NAT after the nf_conntrack_in() call. We can actually clear
975 * the whole state, as it will be re-initialized below.
976 */
977 key->ct_state = 0;
978
979 /* Update the key, but keep the NAT flags. */
980 ovs_ct_update_key(skb, info, key, true, true);
981 }
982
983 ct = nf_ct_get(skb, &ctinfo);
984 if (ct) {
985 /* Packets starting a new connection must be NATted before the
986 * helper, so that the helper knows about the NAT. We enforce
987 * this by delaying both NAT and helper calls for unconfirmed
988 * connections until the committing CT action. For later
989 * packets NAT and Helper may be called in either order.
990 *
991 * NAT will be done only if the CT action has NAT, and only
992 * once per packet (per zone), as guarded by the NAT bits in
993 * the key->ct_state.
994 */
995 if (info->nat && !(key->ct_state & OVS_CS_F_NAT_MASK) &&
996 (nf_ct_is_confirmed(ct) || info->commit) &&
997 ovs_ct_nat(net, key, info, skb, ct, ctinfo) != NF_ACCEPT) {
998 return -EINVAL;
999 }
1000
1001 /* Userspace may decide to perform a ct lookup without a helper
1002 * specified followed by a (recirculate and) commit with one.
1003 * Therefore, for unconfirmed connections which we will commit,
1004 * we need to attach the helper here.
1005 */
1006 if (!nf_ct_is_confirmed(ct) && info->commit &&
1007 info->helper && !nfct_help(ct)) {
1008 int err = __nf_ct_try_assign_helper(ct, info->ct,
1009 GFP_ATOMIC);
1010 if (err)
1011 return err;
1012
1013 /* helper installed, add seqadj if NAT is required */
1014 if (info->nat && !nfct_seqadj(ct)) {
1015 if (!nfct_seqadj_ext_add(ct))
1016 return -EINVAL;
1017 }
1018 }
1019
1020 /* Call the helper only if:
1021 * - nf_conntrack_in() was executed above ("!cached") for a
1022 * confirmed connection, or
1023 * - When committing an unconfirmed connection.
1024 */
1025 if ((nf_ct_is_confirmed(ct) ? !cached : info->commit) &&
1026 ovs_ct_helper(skb, info->family) != NF_ACCEPT) {
1027 return -EINVAL;
1028 }
1029 }
1030
1031 return 0;
1032 }
1033
1034 /* Lookup connection and read fields into key. */
ovs_ct_lookup(struct net * net,struct sw_flow_key * key,const struct ovs_conntrack_info * info,struct sk_buff * skb)1035 static int ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
1036 const struct ovs_conntrack_info *info,
1037 struct sk_buff *skb)
1038 {
1039 struct nf_conntrack_expect *exp;
1040
1041 /* If we pass an expected packet through nf_conntrack_in() the
1042 * expectation is typically removed, but the packet could still be
1043 * lost in upcall processing. To prevent this from happening we
1044 * perform an explicit expectation lookup. Expected connections are
1045 * always new, and will be passed through conntrack only when they are
1046 * committed, as it is OK to remove the expectation at that time.
1047 */
1048 exp = ovs_ct_expect_find(net, &info->zone, info->family, skb);
1049 if (exp) {
1050 u8 state;
1051
1052 /* NOTE: New connections are NATted and Helped only when
1053 * committed, so we are not calling into NAT here.
1054 */
1055 state = OVS_CS_F_TRACKED | OVS_CS_F_NEW | OVS_CS_F_RELATED;
1056 __ovs_ct_update_key(key, state, &info->zone, exp->master);
1057 } else {
1058 struct nf_conn *ct;
1059 int err;
1060
1061 err = __ovs_ct_lookup(net, key, info, skb);
1062 if (err)
1063 return err;
1064
1065 ct = (struct nf_conn *)skb_nfct(skb);
1066 if (ct)
1067 nf_ct_deliver_cached_events(ct);
1068 }
1069
1070 return 0;
1071 }
1072
labels_nonzero(const struct ovs_key_ct_labels * labels)1073 static bool labels_nonzero(const struct ovs_key_ct_labels *labels)
1074 {
1075 size_t i;
1076
1077 for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
1078 if (labels->ct_labels_32[i])
1079 return true;
1080
1081 return false;
1082 }
1083
1084 #if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
ct_limit_hash_bucket(const struct ovs_ct_limit_info * info,u16 zone)1085 static struct hlist_head *ct_limit_hash_bucket(
1086 const struct ovs_ct_limit_info *info, u16 zone)
1087 {
1088 return &info->limits[zone & (CT_LIMIT_HASH_BUCKETS - 1)];
1089 }
1090
1091 /* Call with ovs_mutex */
ct_limit_set(const struct ovs_ct_limit_info * info,struct ovs_ct_limit * new_ct_limit)1092 static void ct_limit_set(const struct ovs_ct_limit_info *info,
1093 struct ovs_ct_limit *new_ct_limit)
1094 {
1095 struct ovs_ct_limit *ct_limit;
1096 struct hlist_head *head;
1097
1098 head = ct_limit_hash_bucket(info, new_ct_limit->zone);
1099 hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
1100 if (ct_limit->zone == new_ct_limit->zone) {
1101 hlist_replace_rcu(&ct_limit->hlist_node,
1102 &new_ct_limit->hlist_node);
1103 kfree_rcu(ct_limit, rcu);
1104 return;
1105 }
1106 }
1107
1108 hlist_add_head_rcu(&new_ct_limit->hlist_node, head);
1109 }
1110
1111 /* Call with ovs_mutex */
ct_limit_del(const struct ovs_ct_limit_info * info,u16 zone)1112 static void ct_limit_del(const struct ovs_ct_limit_info *info, u16 zone)
1113 {
1114 struct ovs_ct_limit *ct_limit;
1115 struct hlist_head *head;
1116 struct hlist_node *n;
1117
1118 head = ct_limit_hash_bucket(info, zone);
1119 hlist_for_each_entry_safe(ct_limit, n, head, hlist_node) {
1120 if (ct_limit->zone == zone) {
1121 hlist_del_rcu(&ct_limit->hlist_node);
1122 kfree_rcu(ct_limit, rcu);
1123 return;
1124 }
1125 }
1126 }
1127
1128 /* Call with RCU read lock */
ct_limit_get(const struct ovs_ct_limit_info * info,u16 zone)1129 static u32 ct_limit_get(const struct ovs_ct_limit_info *info, u16 zone)
1130 {
1131 struct ovs_ct_limit *ct_limit;
1132 struct hlist_head *head;
1133
1134 head = ct_limit_hash_bucket(info, zone);
1135 hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
1136 if (ct_limit->zone == zone)
1137 return ct_limit->limit;
1138 }
1139
1140 return info->default_limit;
1141 }
1142
ovs_ct_check_limit(struct net * net,const struct ovs_conntrack_info * info,const struct nf_conntrack_tuple * tuple)1143 static int ovs_ct_check_limit(struct net *net,
1144 const struct ovs_conntrack_info *info,
1145 const struct nf_conntrack_tuple *tuple)
1146 {
1147 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1148 const struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
1149 u32 per_zone_limit, connections;
1150 u32 conncount_key;
1151
1152 conncount_key = info->zone.id;
1153
1154 per_zone_limit = ct_limit_get(ct_limit_info, info->zone.id);
1155 if (per_zone_limit == OVS_CT_LIMIT_UNLIMITED)
1156 return 0;
1157
1158 connections = nf_conncount_count(net, ct_limit_info->data,
1159 &conncount_key, tuple, &info->zone);
1160 if (connections > per_zone_limit)
1161 return -ENOMEM;
1162
1163 return 0;
1164 }
1165 #endif
1166
1167 /* Lookup connection and confirm if unconfirmed. */
ovs_ct_commit(struct net * net,struct sw_flow_key * key,const struct ovs_conntrack_info * info,struct sk_buff * skb)1168 static int ovs_ct_commit(struct net *net, struct sw_flow_key *key,
1169 const struct ovs_conntrack_info *info,
1170 struct sk_buff *skb)
1171 {
1172 enum ip_conntrack_info ctinfo;
1173 struct nf_conn *ct;
1174 int err;
1175
1176 err = __ovs_ct_lookup(net, key, info, skb);
1177 if (err)
1178 return err;
1179
1180 /* The connection could be invalid, in which case this is a no-op.*/
1181 ct = nf_ct_get(skb, &ctinfo);
1182 if (!ct)
1183 return 0;
1184
1185 #if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1186 if (static_branch_unlikely(&ovs_ct_limit_enabled)) {
1187 if (!nf_ct_is_confirmed(ct)) {
1188 err = ovs_ct_check_limit(net, info,
1189 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
1190 if (err) {
1191 net_warn_ratelimited("openvswitch: zone: %u "
1192 "exceeds conntrack limit\n",
1193 info->zone.id);
1194 return err;
1195 }
1196 }
1197 }
1198 #endif
1199
1200 /* Set the conntrack event mask if given. NEW and DELETE events have
1201 * their own groups, but the NFNLGRP_CONNTRACK_UPDATE group listener
1202 * typically would receive many kinds of updates. Setting the event
1203 * mask allows those events to be filtered. The set event mask will
1204 * remain in effect for the lifetime of the connection unless changed
1205 * by a further CT action with both the commit flag and the eventmask
1206 * option. */
1207 if (info->have_eventmask) {
1208 struct nf_conntrack_ecache *cache = nf_ct_ecache_find(ct);
1209
1210 if (cache)
1211 cache->ctmask = info->eventmask;
1212 }
1213
1214 /* Apply changes before confirming the connection so that the initial
1215 * conntrack NEW netlink event carries the values given in the CT
1216 * action.
1217 */
1218 if (info->mark.mask) {
1219 err = ovs_ct_set_mark(ct, key, info->mark.value,
1220 info->mark.mask);
1221 if (err)
1222 return err;
1223 }
1224 if (!nf_ct_is_confirmed(ct)) {
1225 err = ovs_ct_init_labels(ct, key, &info->labels.value,
1226 &info->labels.mask);
1227 if (err)
1228 return err;
1229 } else if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1230 labels_nonzero(&info->labels.mask)) {
1231 err = ovs_ct_set_labels(ct, key, &info->labels.value,
1232 &info->labels.mask);
1233 if (err)
1234 return err;
1235 }
1236 /* This will take care of sending queued events even if the connection
1237 * is already confirmed.
1238 */
1239 if (nf_conntrack_confirm(skb) != NF_ACCEPT)
1240 return -EINVAL;
1241
1242 return 0;
1243 }
1244
1245 /* Trim the skb to the length specified by the IP/IPv6 header,
1246 * removing any trailing lower-layer padding. This prepares the skb
1247 * for higher-layer processing that assumes skb->len excludes padding
1248 * (such as nf_ip_checksum). The caller needs to pull the skb to the
1249 * network header, and ensure ip_hdr/ipv6_hdr points to valid data.
1250 */
ovs_skb_network_trim(struct sk_buff * skb)1251 static int ovs_skb_network_trim(struct sk_buff *skb)
1252 {
1253 unsigned int len;
1254 int err;
1255
1256 switch (skb->protocol) {
1257 case htons(ETH_P_IP):
1258 len = ntohs(ip_hdr(skb)->tot_len);
1259 break;
1260 case htons(ETH_P_IPV6):
1261 len = sizeof(struct ipv6hdr)
1262 + ntohs(ipv6_hdr(skb)->payload_len);
1263 break;
1264 default:
1265 len = skb->len;
1266 }
1267
1268 err = pskb_trim_rcsum(skb, len);
1269 if (err)
1270 kfree_skb(skb);
1271
1272 return err;
1273 }
1274
1275 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
1276 * value if 'skb' is freed.
1277 */
ovs_ct_execute(struct net * net,struct sk_buff * skb,struct sw_flow_key * key,const struct ovs_conntrack_info * info)1278 int ovs_ct_execute(struct net *net, struct sk_buff *skb,
1279 struct sw_flow_key *key,
1280 const struct ovs_conntrack_info *info)
1281 {
1282 int nh_ofs;
1283 int err;
1284
1285 /* The conntrack module expects to be working at L3. */
1286 nh_ofs = skb_network_offset(skb);
1287 skb_pull_rcsum(skb, nh_ofs);
1288
1289 err = ovs_skb_network_trim(skb);
1290 if (err)
1291 return err;
1292
1293 if (key->ip.frag != OVS_FRAG_TYPE_NONE) {
1294 err = handle_fragments(net, key, info->zone.id, skb);
1295 if (err)
1296 return err;
1297 }
1298
1299 if (info->commit)
1300 err = ovs_ct_commit(net, key, info, skb);
1301 else
1302 err = ovs_ct_lookup(net, key, info, skb);
1303
1304 skb_push(skb, nh_ofs);
1305 skb_postpush_rcsum(skb, skb->data, nh_ofs);
1306 if (err)
1307 kfree_skb(skb);
1308 return err;
1309 }
1310
ovs_ct_clear(struct sk_buff * skb,struct sw_flow_key * key)1311 int ovs_ct_clear(struct sk_buff *skb, struct sw_flow_key *key)
1312 {
1313 if (skb_nfct(skb)) {
1314 nf_conntrack_put(skb_nfct(skb));
1315 nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
1316 ovs_ct_fill_key(skb, key);
1317 }
1318
1319 return 0;
1320 }
1321
ovs_ct_add_helper(struct ovs_conntrack_info * info,const char * name,const struct sw_flow_key * key,bool log)1322 static int ovs_ct_add_helper(struct ovs_conntrack_info *info, const char *name,
1323 const struct sw_flow_key *key, bool log)
1324 {
1325 struct nf_conntrack_helper *helper;
1326 struct nf_conn_help *help;
1327 int ret = 0;
1328
1329 helper = nf_conntrack_helper_try_module_get(name, info->family,
1330 key->ip.proto);
1331 if (!helper) {
1332 OVS_NLERR(log, "Unknown helper \"%s\"", name);
1333 return -EINVAL;
1334 }
1335
1336 help = nf_ct_helper_ext_add(info->ct, GFP_KERNEL);
1337 if (!help) {
1338 nf_conntrack_helper_put(helper);
1339 return -ENOMEM;
1340 }
1341
1342 #if IS_ENABLED(CONFIG_NF_NAT)
1343 if (info->nat) {
1344 ret = nf_nat_helper_try_module_get(name, info->family,
1345 key->ip.proto);
1346 if (ret) {
1347 nf_conntrack_helper_put(helper);
1348 OVS_NLERR(log, "Failed to load \"%s\" NAT helper, error: %d",
1349 name, ret);
1350 return ret;
1351 }
1352 }
1353 #endif
1354 rcu_assign_pointer(help->helper, helper);
1355 info->helper = helper;
1356 return ret;
1357 }
1358
1359 #if IS_ENABLED(CONFIG_NF_NAT)
parse_nat(const struct nlattr * attr,struct ovs_conntrack_info * info,bool log)1360 static int parse_nat(const struct nlattr *attr,
1361 struct ovs_conntrack_info *info, bool log)
1362 {
1363 struct nlattr *a;
1364 int rem;
1365 bool have_ip_max = false;
1366 bool have_proto_max = false;
1367 bool ip_vers = (info->family == NFPROTO_IPV6);
1368
1369 nla_for_each_nested(a, attr, rem) {
1370 static const int ovs_nat_attr_lens[OVS_NAT_ATTR_MAX + 1][2] = {
1371 [OVS_NAT_ATTR_SRC] = {0, 0},
1372 [OVS_NAT_ATTR_DST] = {0, 0},
1373 [OVS_NAT_ATTR_IP_MIN] = {sizeof(struct in_addr),
1374 sizeof(struct in6_addr)},
1375 [OVS_NAT_ATTR_IP_MAX] = {sizeof(struct in_addr),
1376 sizeof(struct in6_addr)},
1377 [OVS_NAT_ATTR_PROTO_MIN] = {sizeof(u16), sizeof(u16)},
1378 [OVS_NAT_ATTR_PROTO_MAX] = {sizeof(u16), sizeof(u16)},
1379 [OVS_NAT_ATTR_PERSISTENT] = {0, 0},
1380 [OVS_NAT_ATTR_PROTO_HASH] = {0, 0},
1381 [OVS_NAT_ATTR_PROTO_RANDOM] = {0, 0},
1382 };
1383 int type = nla_type(a);
1384
1385 if (type > OVS_NAT_ATTR_MAX) {
1386 OVS_NLERR(log, "Unknown NAT attribute (type=%d, max=%d)",
1387 type, OVS_NAT_ATTR_MAX);
1388 return -EINVAL;
1389 }
1390
1391 if (nla_len(a) != ovs_nat_attr_lens[type][ip_vers]) {
1392 OVS_NLERR(log, "NAT attribute type %d has unexpected length (%d != %d)",
1393 type, nla_len(a),
1394 ovs_nat_attr_lens[type][ip_vers]);
1395 return -EINVAL;
1396 }
1397
1398 switch (type) {
1399 case OVS_NAT_ATTR_SRC:
1400 case OVS_NAT_ATTR_DST:
1401 if (info->nat) {
1402 OVS_NLERR(log, "Only one type of NAT may be specified");
1403 return -ERANGE;
1404 }
1405 info->nat |= OVS_CT_NAT;
1406 info->nat |= ((type == OVS_NAT_ATTR_SRC)
1407 ? OVS_CT_SRC_NAT : OVS_CT_DST_NAT);
1408 break;
1409
1410 case OVS_NAT_ATTR_IP_MIN:
1411 nla_memcpy(&info->range.min_addr, a,
1412 sizeof(info->range.min_addr));
1413 info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1414 break;
1415
1416 case OVS_NAT_ATTR_IP_MAX:
1417 have_ip_max = true;
1418 nla_memcpy(&info->range.max_addr, a,
1419 sizeof(info->range.max_addr));
1420 info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1421 break;
1422
1423 case OVS_NAT_ATTR_PROTO_MIN:
1424 info->range.min_proto.all = htons(nla_get_u16(a));
1425 info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1426 break;
1427
1428 case OVS_NAT_ATTR_PROTO_MAX:
1429 have_proto_max = true;
1430 info->range.max_proto.all = htons(nla_get_u16(a));
1431 info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1432 break;
1433
1434 case OVS_NAT_ATTR_PERSISTENT:
1435 info->range.flags |= NF_NAT_RANGE_PERSISTENT;
1436 break;
1437
1438 case OVS_NAT_ATTR_PROTO_HASH:
1439 info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM;
1440 break;
1441
1442 case OVS_NAT_ATTR_PROTO_RANDOM:
1443 info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM_FULLY;
1444 break;
1445
1446 default:
1447 OVS_NLERR(log, "Unknown nat attribute (%d)", type);
1448 return -EINVAL;
1449 }
1450 }
1451
1452 if (rem > 0) {
1453 OVS_NLERR(log, "NAT attribute has %d unknown bytes", rem);
1454 return -EINVAL;
1455 }
1456 if (!info->nat) {
1457 /* Do not allow flags if no type is given. */
1458 if (info->range.flags) {
1459 OVS_NLERR(log,
1460 "NAT flags may be given only when NAT range (SRC or DST) is also specified."
1461 );
1462 return -EINVAL;
1463 }
1464 info->nat = OVS_CT_NAT; /* NAT existing connections. */
1465 } else if (!info->commit) {
1466 OVS_NLERR(log,
1467 "NAT attributes may be specified only when CT COMMIT flag is also specified."
1468 );
1469 return -EINVAL;
1470 }
1471 /* Allow missing IP_MAX. */
1472 if (info->range.flags & NF_NAT_RANGE_MAP_IPS && !have_ip_max) {
1473 memcpy(&info->range.max_addr, &info->range.min_addr,
1474 sizeof(info->range.max_addr));
1475 }
1476 /* Allow missing PROTO_MAX. */
1477 if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1478 !have_proto_max) {
1479 info->range.max_proto.all = info->range.min_proto.all;
1480 }
1481 return 0;
1482 }
1483 #endif
1484
1485 static const struct ovs_ct_len_tbl ovs_ct_attr_lens[OVS_CT_ATTR_MAX + 1] = {
1486 [OVS_CT_ATTR_COMMIT] = { .minlen = 0, .maxlen = 0 },
1487 [OVS_CT_ATTR_FORCE_COMMIT] = { .minlen = 0, .maxlen = 0 },
1488 [OVS_CT_ATTR_ZONE] = { .minlen = sizeof(u16),
1489 .maxlen = sizeof(u16) },
1490 [OVS_CT_ATTR_MARK] = { .minlen = sizeof(struct md_mark),
1491 .maxlen = sizeof(struct md_mark) },
1492 [OVS_CT_ATTR_LABELS] = { .minlen = sizeof(struct md_labels),
1493 .maxlen = sizeof(struct md_labels) },
1494 [OVS_CT_ATTR_HELPER] = { .minlen = 1,
1495 .maxlen = NF_CT_HELPER_NAME_LEN },
1496 #if IS_ENABLED(CONFIG_NF_NAT)
1497 /* NAT length is checked when parsing the nested attributes. */
1498 [OVS_CT_ATTR_NAT] = { .minlen = 0, .maxlen = INT_MAX },
1499 #endif
1500 [OVS_CT_ATTR_EVENTMASK] = { .minlen = sizeof(u32),
1501 .maxlen = sizeof(u32) },
1502 [OVS_CT_ATTR_TIMEOUT] = { .minlen = 1,
1503 .maxlen = CTNL_TIMEOUT_NAME_MAX },
1504 };
1505
parse_ct(const struct nlattr * attr,struct ovs_conntrack_info * info,const char ** helper,bool log)1506 static int parse_ct(const struct nlattr *attr, struct ovs_conntrack_info *info,
1507 const char **helper, bool log)
1508 {
1509 struct nlattr *a;
1510 int rem;
1511
1512 nla_for_each_nested(a, attr, rem) {
1513 int type = nla_type(a);
1514 int maxlen;
1515 int minlen;
1516
1517 if (type > OVS_CT_ATTR_MAX) {
1518 OVS_NLERR(log,
1519 "Unknown conntrack attr (type=%d, max=%d)",
1520 type, OVS_CT_ATTR_MAX);
1521 return -EINVAL;
1522 }
1523
1524 maxlen = ovs_ct_attr_lens[type].maxlen;
1525 minlen = ovs_ct_attr_lens[type].minlen;
1526 if (nla_len(a) < minlen || nla_len(a) > maxlen) {
1527 OVS_NLERR(log,
1528 "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
1529 type, nla_len(a), maxlen);
1530 return -EINVAL;
1531 }
1532
1533 switch (type) {
1534 case OVS_CT_ATTR_FORCE_COMMIT:
1535 info->force = true;
1536 /* fall through. */
1537 case OVS_CT_ATTR_COMMIT:
1538 info->commit = true;
1539 break;
1540 #ifdef CONFIG_NF_CONNTRACK_ZONES
1541 case OVS_CT_ATTR_ZONE:
1542 info->zone.id = nla_get_u16(a);
1543 break;
1544 #endif
1545 #ifdef CONFIG_NF_CONNTRACK_MARK
1546 case OVS_CT_ATTR_MARK: {
1547 struct md_mark *mark = nla_data(a);
1548
1549 if (!mark->mask) {
1550 OVS_NLERR(log, "ct_mark mask cannot be 0");
1551 return -EINVAL;
1552 }
1553 info->mark = *mark;
1554 break;
1555 }
1556 #endif
1557 #ifdef CONFIG_NF_CONNTRACK_LABELS
1558 case OVS_CT_ATTR_LABELS: {
1559 struct md_labels *labels = nla_data(a);
1560
1561 if (!labels_nonzero(&labels->mask)) {
1562 OVS_NLERR(log, "ct_labels mask cannot be 0");
1563 return -EINVAL;
1564 }
1565 info->labels = *labels;
1566 break;
1567 }
1568 #endif
1569 case OVS_CT_ATTR_HELPER:
1570 *helper = nla_data(a);
1571 if (!memchr(*helper, '\0', nla_len(a))) {
1572 OVS_NLERR(log, "Invalid conntrack helper");
1573 return -EINVAL;
1574 }
1575 break;
1576 #if IS_ENABLED(CONFIG_NF_NAT)
1577 case OVS_CT_ATTR_NAT: {
1578 int err = parse_nat(a, info, log);
1579
1580 if (err)
1581 return err;
1582 break;
1583 }
1584 #endif
1585 case OVS_CT_ATTR_EVENTMASK:
1586 info->have_eventmask = true;
1587 info->eventmask = nla_get_u32(a);
1588 break;
1589 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT
1590 case OVS_CT_ATTR_TIMEOUT:
1591 memcpy(info->timeout, nla_data(a), nla_len(a));
1592 if (!memchr(info->timeout, '\0', nla_len(a))) {
1593 OVS_NLERR(log, "Invalid conntrack timeout");
1594 return -EINVAL;
1595 }
1596 break;
1597 #endif
1598
1599 default:
1600 OVS_NLERR(log, "Unknown conntrack attr (%d)",
1601 type);
1602 return -EINVAL;
1603 }
1604 }
1605
1606 #ifdef CONFIG_NF_CONNTRACK_MARK
1607 if (!info->commit && info->mark.mask) {
1608 OVS_NLERR(log,
1609 "Setting conntrack mark requires 'commit' flag.");
1610 return -EINVAL;
1611 }
1612 #endif
1613 #ifdef CONFIG_NF_CONNTRACK_LABELS
1614 if (!info->commit && labels_nonzero(&info->labels.mask)) {
1615 OVS_NLERR(log,
1616 "Setting conntrack labels requires 'commit' flag.");
1617 return -EINVAL;
1618 }
1619 #endif
1620 if (rem > 0) {
1621 OVS_NLERR(log, "Conntrack attr has %d unknown bytes", rem);
1622 return -EINVAL;
1623 }
1624
1625 return 0;
1626 }
1627
ovs_ct_verify(struct net * net,enum ovs_key_attr attr)1628 bool ovs_ct_verify(struct net *net, enum ovs_key_attr attr)
1629 {
1630 if (attr == OVS_KEY_ATTR_CT_STATE)
1631 return true;
1632 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1633 attr == OVS_KEY_ATTR_CT_ZONE)
1634 return true;
1635 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1636 attr == OVS_KEY_ATTR_CT_MARK)
1637 return true;
1638 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1639 attr == OVS_KEY_ATTR_CT_LABELS) {
1640 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1641
1642 return ovs_net->xt_label;
1643 }
1644
1645 return false;
1646 }
1647
ovs_ct_copy_action(struct net * net,const struct nlattr * attr,const struct sw_flow_key * key,struct sw_flow_actions ** sfa,bool log)1648 int ovs_ct_copy_action(struct net *net, const struct nlattr *attr,
1649 const struct sw_flow_key *key,
1650 struct sw_flow_actions **sfa, bool log)
1651 {
1652 struct ovs_conntrack_info ct_info;
1653 const char *helper = NULL;
1654 u16 family;
1655 int err;
1656
1657 family = key_to_nfproto(key);
1658 if (family == NFPROTO_UNSPEC) {
1659 OVS_NLERR(log, "ct family unspecified");
1660 return -EINVAL;
1661 }
1662
1663 memset(&ct_info, 0, sizeof(ct_info));
1664 ct_info.family = family;
1665
1666 nf_ct_zone_init(&ct_info.zone, NF_CT_DEFAULT_ZONE_ID,
1667 NF_CT_DEFAULT_ZONE_DIR, 0);
1668
1669 err = parse_ct(attr, &ct_info, &helper, log);
1670 if (err)
1671 return err;
1672
1673 /* Set up template for tracking connections in specific zones. */
1674 ct_info.ct = nf_ct_tmpl_alloc(net, &ct_info.zone, GFP_KERNEL);
1675 if (!ct_info.ct) {
1676 OVS_NLERR(log, "Failed to allocate conntrack template");
1677 return -ENOMEM;
1678 }
1679
1680 if (ct_info.timeout[0]) {
1681 if (nf_ct_set_timeout(net, ct_info.ct, family, key->ip.proto,
1682 ct_info.timeout))
1683 pr_info_ratelimited("Failed to associated timeout "
1684 "policy `%s'\n", ct_info.timeout);
1685 else
1686 ct_info.nf_ct_timeout = rcu_dereference(
1687 nf_ct_timeout_find(ct_info.ct)->timeout);
1688
1689 }
1690
1691 if (helper) {
1692 err = ovs_ct_add_helper(&ct_info, helper, key, log);
1693 if (err)
1694 goto err_free_ct;
1695 }
1696
1697 err = ovs_nla_add_action(sfa, OVS_ACTION_ATTR_CT, &ct_info,
1698 sizeof(ct_info), log);
1699 if (err)
1700 goto err_free_ct;
1701
1702 __set_bit(IPS_CONFIRMED_BIT, &ct_info.ct->status);
1703 nf_conntrack_get(&ct_info.ct->ct_general);
1704 return 0;
1705 err_free_ct:
1706 __ovs_ct_free_action(&ct_info);
1707 return err;
1708 }
1709
1710 #if IS_ENABLED(CONFIG_NF_NAT)
ovs_ct_nat_to_attr(const struct ovs_conntrack_info * info,struct sk_buff * skb)1711 static bool ovs_ct_nat_to_attr(const struct ovs_conntrack_info *info,
1712 struct sk_buff *skb)
1713 {
1714 struct nlattr *start;
1715
1716 start = nla_nest_start_noflag(skb, OVS_CT_ATTR_NAT);
1717 if (!start)
1718 return false;
1719
1720 if (info->nat & OVS_CT_SRC_NAT) {
1721 if (nla_put_flag(skb, OVS_NAT_ATTR_SRC))
1722 return false;
1723 } else if (info->nat & OVS_CT_DST_NAT) {
1724 if (nla_put_flag(skb, OVS_NAT_ATTR_DST))
1725 return false;
1726 } else {
1727 goto out;
1728 }
1729
1730 if (info->range.flags & NF_NAT_RANGE_MAP_IPS) {
1731 if (IS_ENABLED(CONFIG_NF_NAT) &&
1732 info->family == NFPROTO_IPV4) {
1733 if (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MIN,
1734 info->range.min_addr.ip) ||
1735 (info->range.max_addr.ip
1736 != info->range.min_addr.ip &&
1737 (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MAX,
1738 info->range.max_addr.ip))))
1739 return false;
1740 } else if (IS_ENABLED(CONFIG_IPV6) &&
1741 info->family == NFPROTO_IPV6) {
1742 if (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MIN,
1743 &info->range.min_addr.in6) ||
1744 (memcmp(&info->range.max_addr.in6,
1745 &info->range.min_addr.in6,
1746 sizeof(info->range.max_addr.in6)) &&
1747 (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MAX,
1748 &info->range.max_addr.in6))))
1749 return false;
1750 } else {
1751 return false;
1752 }
1753 }
1754 if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1755 (nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MIN,
1756 ntohs(info->range.min_proto.all)) ||
1757 (info->range.max_proto.all != info->range.min_proto.all &&
1758 nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MAX,
1759 ntohs(info->range.max_proto.all)))))
1760 return false;
1761
1762 if (info->range.flags & NF_NAT_RANGE_PERSISTENT &&
1763 nla_put_flag(skb, OVS_NAT_ATTR_PERSISTENT))
1764 return false;
1765 if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM &&
1766 nla_put_flag(skb, OVS_NAT_ATTR_PROTO_HASH))
1767 return false;
1768 if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM_FULLY &&
1769 nla_put_flag(skb, OVS_NAT_ATTR_PROTO_RANDOM))
1770 return false;
1771 out:
1772 nla_nest_end(skb, start);
1773
1774 return true;
1775 }
1776 #endif
1777
ovs_ct_action_to_attr(const struct ovs_conntrack_info * ct_info,struct sk_buff * skb)1778 int ovs_ct_action_to_attr(const struct ovs_conntrack_info *ct_info,
1779 struct sk_buff *skb)
1780 {
1781 struct nlattr *start;
1782
1783 start = nla_nest_start_noflag(skb, OVS_ACTION_ATTR_CT);
1784 if (!start)
1785 return -EMSGSIZE;
1786
1787 if (ct_info->commit && nla_put_flag(skb, ct_info->force
1788 ? OVS_CT_ATTR_FORCE_COMMIT
1789 : OVS_CT_ATTR_COMMIT))
1790 return -EMSGSIZE;
1791 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1792 nla_put_u16(skb, OVS_CT_ATTR_ZONE, ct_info->zone.id))
1793 return -EMSGSIZE;
1794 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && ct_info->mark.mask &&
1795 nla_put(skb, OVS_CT_ATTR_MARK, sizeof(ct_info->mark),
1796 &ct_info->mark))
1797 return -EMSGSIZE;
1798 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1799 labels_nonzero(&ct_info->labels.mask) &&
1800 nla_put(skb, OVS_CT_ATTR_LABELS, sizeof(ct_info->labels),
1801 &ct_info->labels))
1802 return -EMSGSIZE;
1803 if (ct_info->helper) {
1804 if (nla_put_string(skb, OVS_CT_ATTR_HELPER,
1805 ct_info->helper->name))
1806 return -EMSGSIZE;
1807 }
1808 if (ct_info->have_eventmask &&
1809 nla_put_u32(skb, OVS_CT_ATTR_EVENTMASK, ct_info->eventmask))
1810 return -EMSGSIZE;
1811 if (ct_info->timeout[0]) {
1812 if (nla_put_string(skb, OVS_CT_ATTR_TIMEOUT, ct_info->timeout))
1813 return -EMSGSIZE;
1814 }
1815
1816 #if IS_ENABLED(CONFIG_NF_NAT)
1817 if (ct_info->nat && !ovs_ct_nat_to_attr(ct_info, skb))
1818 return -EMSGSIZE;
1819 #endif
1820 nla_nest_end(skb, start);
1821
1822 return 0;
1823 }
1824
ovs_ct_free_action(const struct nlattr * a)1825 void ovs_ct_free_action(const struct nlattr *a)
1826 {
1827 struct ovs_conntrack_info *ct_info = nla_data(a);
1828
1829 __ovs_ct_free_action(ct_info);
1830 }
1831
__ovs_ct_free_action(struct ovs_conntrack_info * ct_info)1832 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info)
1833 {
1834 if (ct_info->helper) {
1835 #if IS_ENABLED(CONFIG_NF_NAT)
1836 if (ct_info->nat)
1837 nf_nat_helper_put(ct_info->helper);
1838 #endif
1839 nf_conntrack_helper_put(ct_info->helper);
1840 }
1841 if (ct_info->ct) {
1842 if (ct_info->timeout[0])
1843 nf_ct_destroy_timeout(ct_info->ct);
1844 nf_ct_tmpl_free(ct_info->ct);
1845 }
1846 }
1847
1848 #if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
ovs_ct_limit_init(struct net * net,struct ovs_net * ovs_net)1849 static int ovs_ct_limit_init(struct net *net, struct ovs_net *ovs_net)
1850 {
1851 int i, err;
1852
1853 ovs_net->ct_limit_info = kmalloc(sizeof(*ovs_net->ct_limit_info),
1854 GFP_KERNEL);
1855 if (!ovs_net->ct_limit_info)
1856 return -ENOMEM;
1857
1858 ovs_net->ct_limit_info->default_limit = OVS_CT_LIMIT_DEFAULT;
1859 ovs_net->ct_limit_info->limits =
1860 kmalloc_array(CT_LIMIT_HASH_BUCKETS, sizeof(struct hlist_head),
1861 GFP_KERNEL);
1862 if (!ovs_net->ct_limit_info->limits) {
1863 kfree(ovs_net->ct_limit_info);
1864 return -ENOMEM;
1865 }
1866
1867 for (i = 0; i < CT_LIMIT_HASH_BUCKETS; i++)
1868 INIT_HLIST_HEAD(&ovs_net->ct_limit_info->limits[i]);
1869
1870 ovs_net->ct_limit_info->data =
1871 nf_conncount_init(net, NFPROTO_INET, sizeof(u32));
1872
1873 if (IS_ERR(ovs_net->ct_limit_info->data)) {
1874 err = PTR_ERR(ovs_net->ct_limit_info->data);
1875 kfree(ovs_net->ct_limit_info->limits);
1876 kfree(ovs_net->ct_limit_info);
1877 pr_err("openvswitch: failed to init nf_conncount %d\n", err);
1878 return err;
1879 }
1880 return 0;
1881 }
1882
ovs_ct_limit_exit(struct net * net,struct ovs_net * ovs_net)1883 static void ovs_ct_limit_exit(struct net *net, struct ovs_net *ovs_net)
1884 {
1885 const struct ovs_ct_limit_info *info = ovs_net->ct_limit_info;
1886 int i;
1887
1888 nf_conncount_destroy(net, NFPROTO_INET, info->data);
1889 for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) {
1890 struct hlist_head *head = &info->limits[i];
1891 struct ovs_ct_limit *ct_limit;
1892
1893 hlist_for_each_entry_rcu(ct_limit, head, hlist_node)
1894 kfree_rcu(ct_limit, rcu);
1895 }
1896 kfree(ovs_net->ct_limit_info->limits);
1897 kfree(ovs_net->ct_limit_info);
1898 }
1899
1900 static struct sk_buff *
ovs_ct_limit_cmd_reply_start(struct genl_info * info,u8 cmd,struct ovs_header ** ovs_reply_header)1901 ovs_ct_limit_cmd_reply_start(struct genl_info *info, u8 cmd,
1902 struct ovs_header **ovs_reply_header)
1903 {
1904 struct ovs_header *ovs_header = info->userhdr;
1905 struct sk_buff *skb;
1906
1907 skb = genlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1908 if (!skb)
1909 return ERR_PTR(-ENOMEM);
1910
1911 *ovs_reply_header = genlmsg_put(skb, info->snd_portid,
1912 info->snd_seq,
1913 &dp_ct_limit_genl_family, 0, cmd);
1914
1915 if (!*ovs_reply_header) {
1916 nlmsg_free(skb);
1917 return ERR_PTR(-EMSGSIZE);
1918 }
1919 (*ovs_reply_header)->dp_ifindex = ovs_header->dp_ifindex;
1920
1921 return skb;
1922 }
1923
check_zone_id(int zone_id,u16 * pzone)1924 static bool check_zone_id(int zone_id, u16 *pzone)
1925 {
1926 if (zone_id >= 0 && zone_id <= 65535) {
1927 *pzone = (u16)zone_id;
1928 return true;
1929 }
1930 return false;
1931 }
1932
ovs_ct_limit_set_zone_limit(struct nlattr * nla_zone_limit,struct ovs_ct_limit_info * info)1933 static int ovs_ct_limit_set_zone_limit(struct nlattr *nla_zone_limit,
1934 struct ovs_ct_limit_info *info)
1935 {
1936 struct ovs_zone_limit *zone_limit;
1937 int rem;
1938 u16 zone;
1939
1940 rem = NLA_ALIGN(nla_len(nla_zone_limit));
1941 zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1942
1943 while (rem >= sizeof(*zone_limit)) {
1944 if (unlikely(zone_limit->zone_id ==
1945 OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1946 ovs_lock();
1947 info->default_limit = zone_limit->limit;
1948 ovs_unlock();
1949 } else if (unlikely(!check_zone_id(
1950 zone_limit->zone_id, &zone))) {
1951 OVS_NLERR(true, "zone id is out of range");
1952 } else {
1953 struct ovs_ct_limit *ct_limit;
1954
1955 ct_limit = kmalloc(sizeof(*ct_limit), GFP_KERNEL);
1956 if (!ct_limit)
1957 return -ENOMEM;
1958
1959 ct_limit->zone = zone;
1960 ct_limit->limit = zone_limit->limit;
1961
1962 ovs_lock();
1963 ct_limit_set(info, ct_limit);
1964 ovs_unlock();
1965 }
1966 rem -= NLA_ALIGN(sizeof(*zone_limit));
1967 zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
1968 NLA_ALIGN(sizeof(*zone_limit)));
1969 }
1970
1971 if (rem)
1972 OVS_NLERR(true, "set zone limit has %d unknown bytes", rem);
1973
1974 return 0;
1975 }
1976
ovs_ct_limit_del_zone_limit(struct nlattr * nla_zone_limit,struct ovs_ct_limit_info * info)1977 static int ovs_ct_limit_del_zone_limit(struct nlattr *nla_zone_limit,
1978 struct ovs_ct_limit_info *info)
1979 {
1980 struct ovs_zone_limit *zone_limit;
1981 int rem;
1982 u16 zone;
1983
1984 rem = NLA_ALIGN(nla_len(nla_zone_limit));
1985 zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1986
1987 while (rem >= sizeof(*zone_limit)) {
1988 if (unlikely(zone_limit->zone_id ==
1989 OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1990 ovs_lock();
1991 info->default_limit = OVS_CT_LIMIT_DEFAULT;
1992 ovs_unlock();
1993 } else if (unlikely(!check_zone_id(
1994 zone_limit->zone_id, &zone))) {
1995 OVS_NLERR(true, "zone id is out of range");
1996 } else {
1997 ovs_lock();
1998 ct_limit_del(info, zone);
1999 ovs_unlock();
2000 }
2001 rem -= NLA_ALIGN(sizeof(*zone_limit));
2002 zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
2003 NLA_ALIGN(sizeof(*zone_limit)));
2004 }
2005
2006 if (rem)
2007 OVS_NLERR(true, "del zone limit has %d unknown bytes", rem);
2008
2009 return 0;
2010 }
2011
ovs_ct_limit_get_default_limit(struct ovs_ct_limit_info * info,struct sk_buff * reply)2012 static int ovs_ct_limit_get_default_limit(struct ovs_ct_limit_info *info,
2013 struct sk_buff *reply)
2014 {
2015 struct ovs_zone_limit zone_limit;
2016 int err;
2017
2018 zone_limit.zone_id = OVS_ZONE_LIMIT_DEFAULT_ZONE;
2019 zone_limit.limit = info->default_limit;
2020 err = nla_put_nohdr(reply, sizeof(zone_limit), &zone_limit);
2021 if (err)
2022 return err;
2023
2024 return 0;
2025 }
2026
__ovs_ct_limit_get_zone_limit(struct net * net,struct nf_conncount_data * data,u16 zone_id,u32 limit,struct sk_buff * reply)2027 static int __ovs_ct_limit_get_zone_limit(struct net *net,
2028 struct nf_conncount_data *data,
2029 u16 zone_id, u32 limit,
2030 struct sk_buff *reply)
2031 {
2032 struct nf_conntrack_zone ct_zone;
2033 struct ovs_zone_limit zone_limit;
2034 u32 conncount_key = zone_id;
2035
2036 zone_limit.zone_id = zone_id;
2037 zone_limit.limit = limit;
2038 nf_ct_zone_init(&ct_zone, zone_id, NF_CT_DEFAULT_ZONE_DIR, 0);
2039
2040 zone_limit.count = nf_conncount_count(net, data, &conncount_key, NULL,
2041 &ct_zone);
2042 return nla_put_nohdr(reply, sizeof(zone_limit), &zone_limit);
2043 }
2044
ovs_ct_limit_get_zone_limit(struct net * net,struct nlattr * nla_zone_limit,struct ovs_ct_limit_info * info,struct sk_buff * reply)2045 static int ovs_ct_limit_get_zone_limit(struct net *net,
2046 struct nlattr *nla_zone_limit,
2047 struct ovs_ct_limit_info *info,
2048 struct sk_buff *reply)
2049 {
2050 struct ovs_zone_limit *zone_limit;
2051 int rem, err;
2052 u32 limit;
2053 u16 zone;
2054
2055 rem = NLA_ALIGN(nla_len(nla_zone_limit));
2056 zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
2057
2058 while (rem >= sizeof(*zone_limit)) {
2059 if (unlikely(zone_limit->zone_id ==
2060 OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
2061 err = ovs_ct_limit_get_default_limit(info, reply);
2062 if (err)
2063 return err;
2064 } else if (unlikely(!check_zone_id(zone_limit->zone_id,
2065 &zone))) {
2066 OVS_NLERR(true, "zone id is out of range");
2067 } else {
2068 rcu_read_lock();
2069 limit = ct_limit_get(info, zone);
2070 rcu_read_unlock();
2071
2072 err = __ovs_ct_limit_get_zone_limit(
2073 net, info->data, zone, limit, reply);
2074 if (err)
2075 return err;
2076 }
2077 rem -= NLA_ALIGN(sizeof(*zone_limit));
2078 zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
2079 NLA_ALIGN(sizeof(*zone_limit)));
2080 }
2081
2082 if (rem)
2083 OVS_NLERR(true, "get zone limit has %d unknown bytes", rem);
2084
2085 return 0;
2086 }
2087
ovs_ct_limit_get_all_zone_limit(struct net * net,struct ovs_ct_limit_info * info,struct sk_buff * reply)2088 static int ovs_ct_limit_get_all_zone_limit(struct net *net,
2089 struct ovs_ct_limit_info *info,
2090 struct sk_buff *reply)
2091 {
2092 struct ovs_ct_limit *ct_limit;
2093 struct hlist_head *head;
2094 int i, err = 0;
2095
2096 err = ovs_ct_limit_get_default_limit(info, reply);
2097 if (err)
2098 return err;
2099
2100 rcu_read_lock();
2101 for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) {
2102 head = &info->limits[i];
2103 hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
2104 err = __ovs_ct_limit_get_zone_limit(net, info->data,
2105 ct_limit->zone, ct_limit->limit, reply);
2106 if (err)
2107 goto exit_err;
2108 }
2109 }
2110
2111 exit_err:
2112 rcu_read_unlock();
2113 return err;
2114 }
2115
ovs_ct_limit_cmd_set(struct sk_buff * skb,struct genl_info * info)2116 static int ovs_ct_limit_cmd_set(struct sk_buff *skb, struct genl_info *info)
2117 {
2118 struct nlattr **a = info->attrs;
2119 struct sk_buff *reply;
2120 struct ovs_header *ovs_reply_header;
2121 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
2122 struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2123 int err;
2124
2125 reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_SET,
2126 &ovs_reply_header);
2127 if (IS_ERR(reply))
2128 return PTR_ERR(reply);
2129
2130 if (!a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2131 err = -EINVAL;
2132 goto exit_err;
2133 }
2134
2135 err = ovs_ct_limit_set_zone_limit(a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT],
2136 ct_limit_info);
2137 if (err)
2138 goto exit_err;
2139
2140 static_branch_enable(&ovs_ct_limit_enabled);
2141
2142 genlmsg_end(reply, ovs_reply_header);
2143 return genlmsg_reply(reply, info);
2144
2145 exit_err:
2146 nlmsg_free(reply);
2147 return err;
2148 }
2149
ovs_ct_limit_cmd_del(struct sk_buff * skb,struct genl_info * info)2150 static int ovs_ct_limit_cmd_del(struct sk_buff *skb, struct genl_info *info)
2151 {
2152 struct nlattr **a = info->attrs;
2153 struct sk_buff *reply;
2154 struct ovs_header *ovs_reply_header;
2155 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
2156 struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2157 int err;
2158
2159 reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_DEL,
2160 &ovs_reply_header);
2161 if (IS_ERR(reply))
2162 return PTR_ERR(reply);
2163
2164 if (!a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2165 err = -EINVAL;
2166 goto exit_err;
2167 }
2168
2169 err = ovs_ct_limit_del_zone_limit(a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT],
2170 ct_limit_info);
2171 if (err)
2172 goto exit_err;
2173
2174 genlmsg_end(reply, ovs_reply_header);
2175 return genlmsg_reply(reply, info);
2176
2177 exit_err:
2178 nlmsg_free(reply);
2179 return err;
2180 }
2181
ovs_ct_limit_cmd_get(struct sk_buff * skb,struct genl_info * info)2182 static int ovs_ct_limit_cmd_get(struct sk_buff *skb, struct genl_info *info)
2183 {
2184 struct nlattr **a = info->attrs;
2185 struct nlattr *nla_reply;
2186 struct sk_buff *reply;
2187 struct ovs_header *ovs_reply_header;
2188 struct net *net = sock_net(skb->sk);
2189 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2190 struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2191 int err;
2192
2193 reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_GET,
2194 &ovs_reply_header);
2195 if (IS_ERR(reply))
2196 return PTR_ERR(reply);
2197
2198 nla_reply = nla_nest_start_noflag(reply, OVS_CT_LIMIT_ATTR_ZONE_LIMIT);
2199 if (!nla_reply) {
2200 err = -EMSGSIZE;
2201 goto exit_err;
2202 }
2203
2204 if (a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2205 err = ovs_ct_limit_get_zone_limit(
2206 net, a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT], ct_limit_info,
2207 reply);
2208 if (err)
2209 goto exit_err;
2210 } else {
2211 err = ovs_ct_limit_get_all_zone_limit(net, ct_limit_info,
2212 reply);
2213 if (err)
2214 goto exit_err;
2215 }
2216
2217 nla_nest_end(reply, nla_reply);
2218 genlmsg_end(reply, ovs_reply_header);
2219 return genlmsg_reply(reply, info);
2220
2221 exit_err:
2222 nlmsg_free(reply);
2223 return err;
2224 }
2225
2226 static struct genl_ops ct_limit_genl_ops[] = {
2227 { .cmd = OVS_CT_LIMIT_CMD_SET,
2228 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2229 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN
2230 * privilege. */
2231 .doit = ovs_ct_limit_cmd_set,
2232 },
2233 { .cmd = OVS_CT_LIMIT_CMD_DEL,
2234 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2235 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN
2236 * privilege. */
2237 .doit = ovs_ct_limit_cmd_del,
2238 },
2239 { .cmd = OVS_CT_LIMIT_CMD_GET,
2240 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2241 .flags = 0, /* OK for unprivileged users. */
2242 .doit = ovs_ct_limit_cmd_get,
2243 },
2244 };
2245
2246 static const struct genl_multicast_group ovs_ct_limit_multicast_group = {
2247 .name = OVS_CT_LIMIT_MCGROUP,
2248 };
2249
2250 struct genl_family dp_ct_limit_genl_family __ro_after_init = {
2251 .hdrsize = sizeof(struct ovs_header),
2252 .name = OVS_CT_LIMIT_FAMILY,
2253 .version = OVS_CT_LIMIT_VERSION,
2254 .maxattr = OVS_CT_LIMIT_ATTR_MAX,
2255 .policy = ct_limit_policy,
2256 .netnsok = true,
2257 .parallel_ops = true,
2258 .ops = ct_limit_genl_ops,
2259 .n_ops = ARRAY_SIZE(ct_limit_genl_ops),
2260 .mcgrps = &ovs_ct_limit_multicast_group,
2261 .n_mcgrps = 1,
2262 .module = THIS_MODULE,
2263 };
2264 #endif
2265
ovs_ct_init(struct net * net)2266 int ovs_ct_init(struct net *net)
2267 {
2268 unsigned int n_bits = sizeof(struct ovs_key_ct_labels) * BITS_PER_BYTE;
2269 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2270
2271 if (nf_connlabels_get(net, n_bits - 1)) {
2272 ovs_net->xt_label = false;
2273 OVS_NLERR(true, "Failed to set connlabel length");
2274 } else {
2275 ovs_net->xt_label = true;
2276 }
2277
2278 #if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2279 return ovs_ct_limit_init(net, ovs_net);
2280 #else
2281 return 0;
2282 #endif
2283 }
2284
ovs_ct_exit(struct net * net)2285 void ovs_ct_exit(struct net *net)
2286 {
2287 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2288
2289 #if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2290 ovs_ct_limit_exit(net, ovs_net);
2291 #endif
2292
2293 if (ovs_net->xt_label)
2294 nf_connlabels_put(net);
2295 }
2296