?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
x86/speculation/mds: Fix documentation typo
[linux/fpc-iii.git] / net / openvswitch / conntrack.c
blob0171b27a2b81bd83009bf22190c4744e8afab4fe
1 /*
2 * Copyright (c) 2015 Nicira, Inc.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of version 2 of the GNU General Public
6 * License as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 */
13
14 #include <linux/module.h>
15 #include <linux/openvswitch.h>
16 #include <linux/tcp.h>
17 #include <linux/udp.h>
18 #include <linux/sctp.h>
19 #include <net/ip.h>
20 #include <net/netfilter/nf_conntrack_core.h>
21 #include <net/netfilter/nf_conntrack_helper.h>
22 #include <net/netfilter/nf_conntrack_labels.h>
23 #include <net/netfilter/nf_conntrack_seqadj.h>
24 #include <net/netfilter/nf_conntrack_zones.h>
25 #include <net/netfilter/ipv6/nf_defrag_ipv6.h>
26 #include <net/ipv6_frag.h>
27
28 #ifdef CONFIG_NF_NAT_NEEDED
29 #include <linux/netfilter/nf_nat.h>
30 #include <net/netfilter/nf_nat_core.h>
31 #include <net/netfilter/nf_nat_l3proto.h>
32 #endif
33
34 #include "datapath.h"
35 #include "conntrack.h"
36 #include "flow.h"
37 #include "flow_netlink.h"
38
39 struct ovs_ct_len_tbl {
40 int maxlen;
41 int minlen;
42 };
43
44 /* Metadata mark for masked write to conntrack mark */
45 struct md_mark {
46 u32 value;
47 u32 mask;
48 };
49
50 /* Metadata label for masked write to conntrack label. */
51 struct md_labels {
52 struct ovs_key_ct_labels value;
53 struct ovs_key_ct_labels mask;
54 };
55
56 enum ovs_ct_nat {
57 OVS_CT_NAT = 1 << 0, /* NAT for committed connections only. */
58 OVS_CT_SRC_NAT = 1 << 1, /* Source NAT for NEW connections. */
59 OVS_CT_DST_NAT = 1 << 2, /* Destination NAT for NEW connections. */
60 };
61
62 /* Conntrack action context for execution. */
63 struct ovs_conntrack_info {
64 struct nf_conntrack_helper *helper;
65 struct nf_conntrack_zone zone;
66 struct nf_conn *ct;
67 u8 commit : 1;
68 u8 nat : 3; /* enum ovs_ct_nat */
69 u8 force : 1;
70 u8 have_eventmask : 1;
71 u16 family;
72 u32 eventmask; /* Mask of 1 << IPCT_*. */
73 struct md_mark mark;
74 struct md_labels labels;
75 #ifdef CONFIG_NF_NAT_NEEDED
76 struct nf_nat_range range; /* Only present for SRC NAT and DST NAT. */
77 #endif
78 };
79
80 static bool labels_nonzero(const struct ovs_key_ct_labels *labels);
81
82 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info);
83
84 static u16 key_to_nfproto(const struct sw_flow_key *key)
85 {
86 switch (ntohs(key->eth.type)) {
87 case ETH_P_IP:
88 return NFPROTO_IPV4;
89 case ETH_P_IPV6:
90 return NFPROTO_IPV6;
91 default:
92 return NFPROTO_UNSPEC;
93 }
94 }
95
96 /* Map SKB connection state into the values used by flow definition. */
97 static u8 ovs_ct_get_state(enum ip_conntrack_info ctinfo)
98 {
99 u8 ct_state = OVS_CS_F_TRACKED;
100
101 switch (ctinfo) {
102 case IP_CT_ESTABLISHED_REPLY:
103 case IP_CT_RELATED_REPLY:
104 ct_state |= OVS_CS_F_REPLY_DIR;
105 break;
106 default:
107 break;
108 }
109
110 switch (ctinfo) {
111 case IP_CT_ESTABLISHED:
112 case IP_CT_ESTABLISHED_REPLY:
113 ct_state |= OVS_CS_F_ESTABLISHED;
114 break;
115 case IP_CT_RELATED:
116 case IP_CT_RELATED_REPLY:
117 ct_state |= OVS_CS_F_RELATED;
118 break;
119 case IP_CT_NEW:
120 ct_state |= OVS_CS_F_NEW;
121 break;
122 default:
123 break;
124 }
125
126 return ct_state;
127 }
128
129 static u32 ovs_ct_get_mark(const struct nf_conn *ct)
130 {
131 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
132 return ct ? ct->mark : 0;
133 #else
134 return 0;
135 #endif
136 }
137
138 /* Guard against conntrack labels max size shrinking below 128 bits. */
139 #if NF_CT_LABELS_MAX_SIZE < 16
140 #error NF_CT_LABELS_MAX_SIZE must be at least 16 bytes
141 #endif
142
143 static void ovs_ct_get_labels(const struct nf_conn *ct,
144 struct ovs_key_ct_labels *labels)
145 {
146 struct nf_conn_labels *cl = ct ? nf_ct_labels_find(ct) : NULL;
147
148 if (cl)
149 memcpy(labels, cl->bits, OVS_CT_LABELS_LEN);
150 else
151 memset(labels, 0, OVS_CT_LABELS_LEN);
152 }
153
154 static void __ovs_ct_update_key_orig_tp(struct sw_flow_key *key,
155 const struct nf_conntrack_tuple *orig,
156 u8 icmp_proto)
157 {
158 key->ct_orig_proto = orig->dst.protonum;
159 if (orig->dst.protonum == icmp_proto) {
160 key->ct.orig_tp.src = htons(orig->dst.u.icmp.type);
161 key->ct.orig_tp.dst = htons(orig->dst.u.icmp.code);
162 } else {
163 key->ct.orig_tp.src = orig->src.u.all;
164 key->ct.orig_tp.dst = orig->dst.u.all;
165 }
166 }
167
168 static void __ovs_ct_update_key(struct sw_flow_key *key, u8 state,
169 const struct nf_conntrack_zone *zone,
170 const struct nf_conn *ct)
171 {
172 key->ct_state = state;
173 key->ct_zone = zone->id;
174 key->ct.mark = ovs_ct_get_mark(ct);
175 ovs_ct_get_labels(ct, &key->ct.labels);
176
177 if (ct) {
178 const struct nf_conntrack_tuple *orig;
179
180 /* Use the master if we have one. */
181 if (ct->master)
182 ct = ct->master;
183 orig = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
184
185 /* IP version must match with the master connection. */
186 if (key->eth.type == htons(ETH_P_IP) &&
187 nf_ct_l3num(ct) == NFPROTO_IPV4) {
188 key->ipv4.ct_orig.src = orig->src.u3.ip;
189 key->ipv4.ct_orig.dst = orig->dst.u3.ip;
190 __ovs_ct_update_key_orig_tp(key, orig, IPPROTO_ICMP);
191 return;
192 } else if (key->eth.type == htons(ETH_P_IPV6) &&
193 !sw_flow_key_is_nd(key) &&
194 nf_ct_l3num(ct) == NFPROTO_IPV6) {
195 key->ipv6.ct_orig.src = orig->src.u3.in6;
196 key->ipv6.ct_orig.dst = orig->dst.u3.in6;
197 __ovs_ct_update_key_orig_tp(key, orig, NEXTHDR_ICMP);
198 return;
199 }
200 }
201 /* Clear 'ct_orig_proto' to mark the non-existence of conntrack
202 * original direction key fields.
203 */
204 key->ct_orig_proto = 0;
205 }
206
207 /* Update 'key' based on skb->_nfct. If 'post_ct' is true, then OVS has
208 * previously sent the packet to conntrack via the ct action. If
209 * 'keep_nat_flags' is true, the existing NAT flags retained, else they are
210 * initialized from the connection status.
211 */
212 static void ovs_ct_update_key(const struct sk_buff *skb,
213 const struct ovs_conntrack_info *info,
214 struct sw_flow_key *key, bool post_ct,
215 bool keep_nat_flags)
216 {
217 const struct nf_conntrack_zone *zone = &nf_ct_zone_dflt;
218 enum ip_conntrack_info ctinfo;
219 struct nf_conn *ct;
220 u8 state = 0;
221
222 ct = nf_ct_get(skb, &ctinfo);
223 if (ct) {
224 state = ovs_ct_get_state(ctinfo);
225 /* All unconfirmed entries are NEW connections. */
226 if (!nf_ct_is_confirmed(ct))
227 state |= OVS_CS_F_NEW;
228 /* OVS persists the related flag for the duration of the
229 * connection.
230 */
231 if (ct->master)
232 state |= OVS_CS_F_RELATED;
233 if (keep_nat_flags) {
234 state |= key->ct_state & OVS_CS_F_NAT_MASK;
235 } else {
236 if (ct->status & IPS_SRC_NAT)
237 state |= OVS_CS_F_SRC_NAT;
238 if (ct->status & IPS_DST_NAT)
239 state |= OVS_CS_F_DST_NAT;
240 }
241 zone = nf_ct_zone(ct);
242 } else if (post_ct) {
243 state = OVS_CS_F_TRACKED | OVS_CS_F_INVALID;
244 if (info)
245 zone = &info->zone;
246 }
247 __ovs_ct_update_key(key, state, zone, ct);
248 }
249
250 /* This is called to initialize CT key fields possibly coming in from the local
251 * stack.
252 */
253 void ovs_ct_fill_key(const struct sk_buff *skb, struct sw_flow_key *key)
254 {
255 ovs_ct_update_key(skb, NULL, key, false, false);
256 }
257
258 #define IN6_ADDR_INITIALIZER(ADDR) \
259 { (ADDR).s6_addr32[0], (ADDR).s6_addr32[1], \
260 (ADDR).s6_addr32[2], (ADDR).s6_addr32[3] }
261
262 int ovs_ct_put_key(const struct sw_flow_key *swkey,
263 const struct sw_flow_key *output, struct sk_buff *skb)
264 {
265 if (nla_put_u32(skb, OVS_KEY_ATTR_CT_STATE, output->ct_state))
266 return -EMSGSIZE;
267
268 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
269 nla_put_u16(skb, OVS_KEY_ATTR_CT_ZONE, output->ct_zone))
270 return -EMSGSIZE;
271
272 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
273 nla_put_u32(skb, OVS_KEY_ATTR_CT_MARK, output->ct.mark))
274 return -EMSGSIZE;
275
276 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
277 nla_put(skb, OVS_KEY_ATTR_CT_LABELS, sizeof(output->ct.labels),
278 &output->ct.labels))
279 return -EMSGSIZE;
280
281 if (swkey->ct_orig_proto) {
282 if (swkey->eth.type == htons(ETH_P_IP)) {
283 struct ovs_key_ct_tuple_ipv4 orig = {
284 output->ipv4.ct_orig.src,
285 output->ipv4.ct_orig.dst,
286 output->ct.orig_tp.src,
287 output->ct.orig_tp.dst,
288 output->ct_orig_proto,
289 };
290 if (nla_put(skb, OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4,
291 sizeof(orig), &orig))
292 return -EMSGSIZE;
293 } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
294 struct ovs_key_ct_tuple_ipv6 orig = {
295 IN6_ADDR_INITIALIZER(output->ipv6.ct_orig.src),
296 IN6_ADDR_INITIALIZER(output->ipv6.ct_orig.dst),
297 output->ct.orig_tp.src,
298 output->ct.orig_tp.dst,
299 output->ct_orig_proto,
300 };
301 if (nla_put(skb, OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6,
302 sizeof(orig), &orig))
303 return -EMSGSIZE;
304 }
305 }
306
307 return 0;
308 }
309
310 static int ovs_ct_set_mark(struct nf_conn *ct, struct sw_flow_key *key,
311 u32 ct_mark, u32 mask)
312 {
313 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
314 u32 new_mark;
315
316 new_mark = ct_mark | (ct->mark & ~(mask));
317 if (ct->mark != new_mark) {
318 ct->mark = new_mark;
319 if (nf_ct_is_confirmed(ct))
320 nf_conntrack_event_cache(IPCT_MARK, ct);
321 key->ct.mark = new_mark;
322 }
323
324 return 0;
325 #else
326 return -ENOTSUPP;
327 #endif
328 }
329
330 static struct nf_conn_labels *ovs_ct_get_conn_labels(struct nf_conn *ct)
331 {
332 struct nf_conn_labels *cl;
333
334 cl = nf_ct_labels_find(ct);
335 if (!cl) {
336 nf_ct_labels_ext_add(ct);
337 cl = nf_ct_labels_find(ct);
338 }
339
340 return cl;
341 }
342
343 /* Initialize labels for a new, yet to be committed conntrack entry. Note that
344 * since the new connection is not yet confirmed, and thus no-one else has
345 * access to it's labels, we simply write them over.
346 */
347 static int ovs_ct_init_labels(struct nf_conn *ct, struct sw_flow_key *key,
348 const struct ovs_key_ct_labels *labels,
349 const struct ovs_key_ct_labels *mask)
350 {
351 struct nf_conn_labels *cl, *master_cl;
352 bool have_mask = labels_nonzero(mask);
353
354 /* Inherit master's labels to the related connection? */
355 master_cl = ct->master ? nf_ct_labels_find(ct->master) : NULL;
356
357 if (!master_cl && !have_mask)
358 return 0; /* Nothing to do. */
359
360 cl = ovs_ct_get_conn_labels(ct);
361 if (!cl)
362 return -ENOSPC;
363
364 /* Inherit the master's labels, if any. */
365 if (master_cl)
366 *cl = *master_cl;
367
368 if (have_mask) {
369 u32 *dst = (u32 *)cl->bits;
370 int i;
371
372 for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
373 dst[i] = (dst[i] & ~mask->ct_labels_32[i]) |
374 (labels->ct_labels_32[i]
375 & mask->ct_labels_32[i]);
376 }
377
378 /* Labels are included in the IPCTNL_MSG_CT_NEW event only if the
379 * IPCT_LABEL bit is set in the event cache.
380 */
381 nf_conntrack_event_cache(IPCT_LABEL, ct);
382
383 memcpy(&key->ct.labels, cl->bits, OVS_CT_LABELS_LEN);
384
385 return 0;
386 }
387
388 static int ovs_ct_set_labels(struct nf_conn *ct, struct sw_flow_key *key,
389 const struct ovs_key_ct_labels *labels,
390 const struct ovs_key_ct_labels *mask)
391 {
392 struct nf_conn_labels *cl;
393 int err;
394
395 cl = ovs_ct_get_conn_labels(ct);
396 if (!cl)
397 return -ENOSPC;
398
399 err = nf_connlabels_replace(ct, labels->ct_labels_32,
400 mask->ct_labels_32,
401 OVS_CT_LABELS_LEN_32);
402 if (err)
403 return err;
404
405 memcpy(&key->ct.labels, cl->bits, OVS_CT_LABELS_LEN);
406
407 return 0;
408 }
409
410 /* 'skb' should already be pulled to nh_ofs. */
411 static int ovs_ct_helper(struct sk_buff *skb, u16 proto)
412 {
413 const struct nf_conntrack_helper *helper;
414 const struct nf_conn_help *help;
415 enum ip_conntrack_info ctinfo;
416 unsigned int protoff;
417 struct nf_conn *ct;
418 int err;
419
420 ct = nf_ct_get(skb, &ctinfo);
421 if (!ct || ctinfo == IP_CT_RELATED_REPLY)
422 return NF_ACCEPT;
423
424 help = nfct_help(ct);
425 if (!help)
426 return NF_ACCEPT;
427
428 helper = rcu_dereference(help->helper);
429 if (!helper)
430 return NF_ACCEPT;
431
432 switch (proto) {
433 case NFPROTO_IPV4:
434 protoff = ip_hdrlen(skb);
435 break;
436 case NFPROTO_IPV6: {
437 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
438 __be16 frag_off;
439 int ofs;
440
441 ofs = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr,
442 &frag_off);
443 if (ofs < 0 || (frag_off & htons(~0x7)) != 0) {
444 pr_debug("proto header not found\n");
445 return NF_ACCEPT;
446 }
447 protoff = ofs;
448 break;
449 }
450 default:
451 WARN_ONCE(1, "helper invoked on non-IP family!");
452 return NF_DROP;
453 }
454
455 err = helper->help(skb, protoff, ct, ctinfo);
456 if (err != NF_ACCEPT)
457 return err;
458
459 /* Adjust seqs after helper. This is needed due to some helpers (e.g.,
460 * FTP with NAT) adusting the TCP payload size when mangling IP
461 * addresses and/or port numbers in the text-based control connection.
462 */
463 if (test_bit(IPS_SEQ_ADJUST_BIT, &ct->status) &&
464 !nf_ct_seq_adjust(skb, ct, ctinfo, protoff))
465 return NF_DROP;
466 return NF_ACCEPT;
467 }
468
469 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
470 * value if 'skb' is freed.
471 */
472 static int handle_fragments(struct net *net, struct sw_flow_key *key,
473 u16 zone, struct sk_buff *skb)
474 {
475 struct ovs_skb_cb ovs_cb = *OVS_CB(skb);
476 int err;
477
478 if (key->eth.type == htons(ETH_P_IP)) {
479 enum ip_defrag_users user = IP_DEFRAG_CONNTRACK_IN + zone;
480
481 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
482 err = ip_defrag(net, skb, user);
483 if (err)
484 return err;
485
486 ovs_cb.mru = IPCB(skb)->frag_max_size;
487 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
488 } else if (key->eth.type == htons(ETH_P_IPV6)) {
489 enum ip6_defrag_users user = IP6_DEFRAG_CONNTRACK_IN + zone;
490
491 memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm));
492 err = nf_ct_frag6_gather(net, skb, user);
493 if (err) {
494 if (err != -EINPROGRESS)
495 kfree_skb(skb);
496 return err;
497 }
498
499 key->ip.proto = ipv6_hdr(skb)->nexthdr;
500 ovs_cb.mru = IP6CB(skb)->frag_max_size;
501 #endif
502 } else {
503 kfree_skb(skb);
504 return -EPFNOSUPPORT;
505 }
506
507 key->ip.frag = OVS_FRAG_TYPE_NONE;
508 skb_clear_hash(skb);
509 skb->ignore_df = 1;
510 *OVS_CB(skb) = ovs_cb;
511
512 return 0;
513 }
514
515 static struct nf_conntrack_expect *
516 ovs_ct_expect_find(struct net *net, const struct nf_conntrack_zone *zone,
517 u16 proto, const struct sk_buff *skb)
518 {
519 struct nf_conntrack_tuple tuple;
520 struct nf_conntrack_expect *exp;
521
522 if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), proto, net, &tuple))
523 return NULL;
524
525 exp = __nf_ct_expect_find(net, zone, &tuple);
526 if (exp) {
527 struct nf_conntrack_tuple_hash *h;
528
529 /* Delete existing conntrack entry, if it clashes with the
530 * expectation. This can happen since conntrack ALGs do not
531 * check for clashes between (new) expectations and existing
532 * conntrack entries. nf_conntrack_in() will check the
533 * expectations only if a conntrack entry can not be found,
534 * which can lead to OVS finding the expectation (here) in the
535 * init direction, but which will not be removed by the
536 * nf_conntrack_in() call, if a matching conntrack entry is
537 * found instead. In this case all init direction packets
538 * would be reported as new related packets, while reply
539 * direction packets would be reported as un-related
540 * established packets.
541 */
542 h = nf_conntrack_find_get(net, zone, &tuple);
543 if (h) {
544 struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
545
546 nf_ct_delete(ct, 0, 0);
547 nf_conntrack_put(&ct->ct_general);
548 }
549 }
550
551 return exp;
552 }
553
554 /* This replicates logic from nf_conntrack_core.c that is not exported. */
555 static enum ip_conntrack_info
556 ovs_ct_get_info(const struct nf_conntrack_tuple_hash *h)
557 {
558 const struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
559
560 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY)
561 return IP_CT_ESTABLISHED_REPLY;
562 /* Once we've had two way comms, always ESTABLISHED. */
563 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status))
564 return IP_CT_ESTABLISHED;
565 if (test_bit(IPS_EXPECTED_BIT, &ct->status))
566 return IP_CT_RELATED;
567 return IP_CT_NEW;
568 }
569
570 /* Find an existing connection which this packet belongs to without
571 * re-attributing statistics or modifying the connection state. This allows an
572 * skb->_nfct lost due to an upcall to be recovered during actions execution.
573 *
574 * Must be called with rcu_read_lock.
575 *
576 * On success, populates skb->_nfct and returns the connection. Returns NULL
577 * if there is no existing entry.
578 */
579 static struct nf_conn *
580 ovs_ct_find_existing(struct net *net, const struct nf_conntrack_zone *zone,
581 u8 l3num, struct sk_buff *skb, bool natted)
582 {
583 const struct nf_conntrack_l3proto *l3proto;
584 const struct nf_conntrack_l4proto *l4proto;
585 struct nf_conntrack_tuple tuple;
586 struct nf_conntrack_tuple_hash *h;
587 struct nf_conn *ct;
588 unsigned int dataoff;
589 u8 protonum;
590
591 l3proto = __nf_ct_l3proto_find(l3num);
592 if (l3proto->get_l4proto(skb, skb_network_offset(skb), &dataoff,
593 &protonum) <= 0) {
594 pr_debug("ovs_ct_find_existing: Can't get protonum\n");
595 return NULL;
596 }
597 l4proto = __nf_ct_l4proto_find(l3num, protonum);
598 if (!nf_ct_get_tuple(skb, skb_network_offset(skb), dataoff, l3num,
599 protonum, net, &tuple, l3proto, l4proto)) {
600 pr_debug("ovs_ct_find_existing: Can't get tuple\n");
601 return NULL;
602 }
603
604 /* Must invert the tuple if skb has been transformed by NAT. */
605 if (natted) {
606 struct nf_conntrack_tuple inverse;
607
608 if (!nf_ct_invert_tuple(&inverse, &tuple, l3proto, l4proto)) {
609 pr_debug("ovs_ct_find_existing: Inversion failed!\n");
610 return NULL;
611 }
612 tuple = inverse;
613 }
614
615 /* look for tuple match */
616 h = nf_conntrack_find_get(net, zone, &tuple);
617 if (!h)
618 return NULL; /* Not found. */
619
620 ct = nf_ct_tuplehash_to_ctrack(h);
621
622 /* Inverted packet tuple matches the reverse direction conntrack tuple,
623 * select the other tuplehash to get the right 'ctinfo' bits for this
624 * packet.
625 */
626 if (natted)
627 h = &ct->tuplehash[!h->tuple.dst.dir];
628
629 nf_ct_set(skb, ct, ovs_ct_get_info(h));
630 return ct;
631 }
632
633 static
634 struct nf_conn *ovs_ct_executed(struct net *net,
635 const struct sw_flow_key *key,
636 const struct ovs_conntrack_info *info,
637 struct sk_buff *skb,
638 bool *ct_executed)
639 {
640 struct nf_conn *ct = NULL;
641
642 /* If no ct, check if we have evidence that an existing conntrack entry
643 * might be found for this skb. This happens when we lose a skb->_nfct
644 * due to an upcall, or if the direction is being forced. If the
645 * connection was not confirmed, it is not cached and needs to be run
646 * through conntrack again.
647 */
648 *ct_executed = (key->ct_state & OVS_CS_F_TRACKED) &&
649 !(key->ct_state & OVS_CS_F_INVALID) &&
650 (key->ct_zone == info->zone.id);
651
652 if (*ct_executed || (!key->ct_state && info->force)) {
653 ct = ovs_ct_find_existing(net, &info->zone, info->family, skb,
654 !!(key->ct_state &
655 OVS_CS_F_NAT_MASK));
656 }
657
658 return ct;
659 }
660
661 /* Determine whether skb->_nfct is equal to the result of conntrack lookup. */
662 static bool skb_nfct_cached(struct net *net,
663 const struct sw_flow_key *key,
664 const struct ovs_conntrack_info *info,
665 struct sk_buff *skb)
666 {
667 enum ip_conntrack_info ctinfo;
668 struct nf_conn *ct;
669 bool ct_executed = true;
670
671 ct = nf_ct_get(skb, &ctinfo);
672 if (!ct)
673 ct = ovs_ct_executed(net, key, info, skb, &ct_executed);
674
675 if (ct)
676 nf_ct_get(skb, &ctinfo);
677 else
678 return false;
679
680 if (!net_eq(net, read_pnet(&ct->ct_net)))
681 return false;
682 if (!nf_ct_zone_equal_any(info->ct, nf_ct_zone(ct)))
683 return false;
684 if (info->helper) {
685 struct nf_conn_help *help;
686
687 help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER);
688 if (help && rcu_access_pointer(help->helper) != info->helper)
689 return false;
690 }
691 /* Force conntrack entry direction to the current packet? */
692 if (info->force && CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) {
693 /* Delete the conntrack entry if confirmed, else just release
694 * the reference.
695 */
696 if (nf_ct_is_confirmed(ct))
697 nf_ct_delete(ct, 0, 0);
698
699 nf_conntrack_put(&ct->ct_general);
700 nf_ct_set(skb, NULL, 0);
701 return false;
702 }
703
704 return ct_executed;
705 }
706
707 #ifdef CONFIG_NF_NAT_NEEDED
708 /* Modelled after nf_nat_ipv[46]_fn().
709 * range is only used for new, uninitialized NAT state.
710 * Returns either NF_ACCEPT or NF_DROP.
711 */
712 static int ovs_ct_nat_execute(struct sk_buff *skb, struct nf_conn *ct,
713 enum ip_conntrack_info ctinfo,
714 const struct nf_nat_range *range,
715 enum nf_nat_manip_type maniptype)
716 {
717 int hooknum, nh_off, err = NF_ACCEPT;
718
719 nh_off = skb_network_offset(skb);
720 skb_pull_rcsum(skb, nh_off);
721
722 /* See HOOK2MANIP(). */
723 if (maniptype == NF_NAT_MANIP_SRC)
724 hooknum = NF_INET_LOCAL_IN; /* Source NAT */
725 else
726 hooknum = NF_INET_LOCAL_OUT; /* Destination NAT */
727
728 switch (ctinfo) {
729 case IP_CT_RELATED:
730 case IP_CT_RELATED_REPLY:
731 if (IS_ENABLED(CONFIG_NF_NAT_IPV4) &&
732 skb->protocol == htons(ETH_P_IP) &&
733 ip_hdr(skb)->protocol == IPPROTO_ICMP) {
734 if (!nf_nat_icmp_reply_translation(skb, ct, ctinfo,
735 hooknum))
736 err = NF_DROP;
737 goto push;
738 } else if (IS_ENABLED(CONFIG_NF_NAT_IPV6) &&
739 skb->protocol == htons(ETH_P_IPV6)) {
740 __be16 frag_off;
741 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
742 int hdrlen = ipv6_skip_exthdr(skb,
743 sizeof(struct ipv6hdr),
744 &nexthdr, &frag_off);
745
746 if (hdrlen >= 0 && nexthdr == IPPROTO_ICMPV6) {
747 if (!nf_nat_icmpv6_reply_translation(skb, ct,
748 ctinfo,
749 hooknum,
750 hdrlen))
751 err = NF_DROP;
752 goto push;
753 }
754 }
755 /* Non-ICMP, fall thru to initialize if needed. */
756 case IP_CT_NEW:
757 /* Seen it before? This can happen for loopback, retrans,
758 * or local packets.
759 */
760 if (!nf_nat_initialized(ct, maniptype)) {
761 /* Initialize according to the NAT action. */
762 err = (range && range->flags & NF_NAT_RANGE_MAP_IPS)
763 /* Action is set up to establish a new
764 * mapping.
765 */
766 ? nf_nat_setup_info(ct, range, maniptype)
767 : nf_nat_alloc_null_binding(ct, hooknum);
768 if (err != NF_ACCEPT)
769 goto push;
770 }
771 break;
772
773 case IP_CT_ESTABLISHED:
774 case IP_CT_ESTABLISHED_REPLY:
775 break;
776
777 default:
778 err = NF_DROP;
779 goto push;
780 }
781
782 err = nf_nat_packet(ct, ctinfo, hooknum, skb);
783 push:
784 skb_push(skb, nh_off);
785 skb_postpush_rcsum(skb, skb->data, nh_off);
786
787 return err;
788 }
789
790 static void ovs_nat_update_key(struct sw_flow_key *key,
791 const struct sk_buff *skb,
792 enum nf_nat_manip_type maniptype)
793 {
794 if (maniptype == NF_NAT_MANIP_SRC) {
795 __be16 src;
796
797 key->ct_state |= OVS_CS_F_SRC_NAT;
798 if (key->eth.type == htons(ETH_P_IP))
799 key->ipv4.addr.src = ip_hdr(skb)->saddr;
800 else if (key->eth.type == htons(ETH_P_IPV6))
801 memcpy(&key->ipv6.addr.src, &ipv6_hdr(skb)->saddr,
802 sizeof(key->ipv6.addr.src));
803 else
804 return;
805
806 if (key->ip.proto == IPPROTO_UDP)
807 src = udp_hdr(skb)->source;
808 else if (key->ip.proto == IPPROTO_TCP)
809 src = tcp_hdr(skb)->source;
810 else if (key->ip.proto == IPPROTO_SCTP)
811 src = sctp_hdr(skb)->source;
812 else
813 return;
814
815 key->tp.src = src;
816 } else {
817 __be16 dst;
818
819 key->ct_state |= OVS_CS_F_DST_NAT;
820 if (key->eth.type == htons(ETH_P_IP))
821 key->ipv4.addr.dst = ip_hdr(skb)->daddr;
822 else if (key->eth.type == htons(ETH_P_IPV6))
823 memcpy(&key->ipv6.addr.dst, &ipv6_hdr(skb)->daddr,
824 sizeof(key->ipv6.addr.dst));
825 else
826 return;
827
828 if (key->ip.proto == IPPROTO_UDP)
829 dst = udp_hdr(skb)->dest;
830 else if (key->ip.proto == IPPROTO_TCP)
831 dst = tcp_hdr(skb)->dest;
832 else if (key->ip.proto == IPPROTO_SCTP)
833 dst = sctp_hdr(skb)->dest;
834 else
835 return;
836
837 key->tp.dst = dst;
838 }
839 }
840
841 /* Returns NF_DROP if the packet should be dropped, NF_ACCEPT otherwise. */
842 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
843 const struct ovs_conntrack_info *info,
844 struct sk_buff *skb, struct nf_conn *ct,
845 enum ip_conntrack_info ctinfo)
846 {
847 enum nf_nat_manip_type maniptype;
848 int err;
849
850 /* Add NAT extension if not confirmed yet. */
851 if (!nf_ct_is_confirmed(ct) && !nf_ct_nat_ext_add(ct))
852 return NF_ACCEPT; /* Can't NAT. */
853
854 /* Determine NAT type.
855 * Check if the NAT type can be deduced from the tracked connection.
856 * Make sure new expected connections (IP_CT_RELATED) are NATted only
857 * when committing.
858 */
859 if (info->nat & OVS_CT_NAT && ctinfo != IP_CT_NEW &&
860 ct->status & IPS_NAT_MASK &&
861 (ctinfo != IP_CT_RELATED || info->commit)) {
862 /* NAT an established or related connection like before. */
863 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY)
864 /* This is the REPLY direction for a connection
865 * for which NAT was applied in the forward
866 * direction. Do the reverse NAT.
867 */
868 maniptype = ct->status & IPS_SRC_NAT
869 ? NF_NAT_MANIP_DST : NF_NAT_MANIP_SRC;
870 else
871 maniptype = ct->status & IPS_SRC_NAT
872 ? NF_NAT_MANIP_SRC : NF_NAT_MANIP_DST;
873 } else if (info->nat & OVS_CT_SRC_NAT) {
874 maniptype = NF_NAT_MANIP_SRC;
875 } else if (info->nat & OVS_CT_DST_NAT) {
876 maniptype = NF_NAT_MANIP_DST;
877 } else {
878 return NF_ACCEPT; /* Connection is not NATed. */
879 }
880 err = ovs_ct_nat_execute(skb, ct, ctinfo, &info->range, maniptype);
881
882 /* Mark NAT done if successful and update the flow key. */
883 if (err == NF_ACCEPT)
884 ovs_nat_update_key(key, skb, maniptype);
885
886 return err;
887 }
888 #else /* !CONFIG_NF_NAT_NEEDED */
889 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
890 const struct ovs_conntrack_info *info,
891 struct sk_buff *skb, struct nf_conn *ct,
892 enum ip_conntrack_info ctinfo)
893 {
894 return NF_ACCEPT;
895 }
896 #endif
897
898 /* Pass 'skb' through conntrack in 'net', using zone configured in 'info', if
899 * not done already. Update key with new CT state after passing the packet
900 * through conntrack.
901 * Note that if the packet is deemed invalid by conntrack, skb->_nfct will be
902 * set to NULL and 0 will be returned.
903 */
904 static int __ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
905 const struct ovs_conntrack_info *info,
906 struct sk_buff *skb)
907 {
908 /* If we are recirculating packets to match on conntrack fields and
909 * committing with a separate conntrack action, then we don't need to
910 * actually run the packet through conntrack twice unless it's for a
911 * different zone.
912 */
913 bool cached = skb_nfct_cached(net, key, info, skb);
914 enum ip_conntrack_info ctinfo;
915 struct nf_conn *ct;
916
917 if (!cached) {
918 struct nf_conn *tmpl = info->ct;
919 int err;
920
921 /* Associate skb with specified zone. */
922 if (tmpl) {
923 if (skb_nfct(skb))
924 nf_conntrack_put(skb_nfct(skb));
925 nf_conntrack_get(&tmpl->ct_general);
926 nf_ct_set(skb, tmpl, IP_CT_NEW);
927 }
928
929 err = nf_conntrack_in(net, info->family,
930 NF_INET_PRE_ROUTING, skb);
931 if (err != NF_ACCEPT)
932 return -ENOENT;
933
934 /* Clear CT state NAT flags to mark that we have not yet done
935 * NAT after the nf_conntrack_in() call. We can actually clear
936 * the whole state, as it will be re-initialized below.
937 */
938 key->ct_state = 0;
939
940 /* Update the key, but keep the NAT flags. */
941 ovs_ct_update_key(skb, info, key, true, true);
942 }
943
944 ct = nf_ct_get(skb, &ctinfo);
945 if (ct) {
946 /* Packets starting a new connection must be NATted before the
947 * helper, so that the helper knows about the NAT. We enforce
948 * this by delaying both NAT and helper calls for unconfirmed
949 * connections until the committing CT action. For later
950 * packets NAT and Helper may be called in either order.
951 *
952 * NAT will be done only if the CT action has NAT, and only
953 * once per packet (per zone), as guarded by the NAT bits in
954 * the key->ct_state.
955 */
956 if (info->nat && !(key->ct_state & OVS_CS_F_NAT_MASK) &&
957 (nf_ct_is_confirmed(ct) || info->commit) &&
958 ovs_ct_nat(net, key, info, skb, ct, ctinfo) != NF_ACCEPT) {
959 return -EINVAL;
960 }
961
962 /* Userspace may decide to perform a ct lookup without a helper
963 * specified followed by a (recirculate and) commit with one.
964 * Therefore, for unconfirmed connections which we will commit,
965 * we need to attach the helper here.
966 */
967 if (!nf_ct_is_confirmed(ct) && info->commit &&
968 info->helper && !nfct_help(ct)) {
969 int err = __nf_ct_try_assign_helper(ct, info->ct,
970 GFP_ATOMIC);
971 if (err)
972 return err;
973 }
974
975 /* Call the helper only if:
976 * - nf_conntrack_in() was executed above ("!cached") for a
977 * confirmed connection, or
978 * - When committing an unconfirmed connection.
979 */
980 if ((nf_ct_is_confirmed(ct) ? !cached : info->commit) &&
981 ovs_ct_helper(skb, info->family) != NF_ACCEPT) {
982 return -EINVAL;
983 }
984 }
985
986 return 0;
987 }
988
989 /* Lookup connection and read fields into key. */
990 static int ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
991 const struct ovs_conntrack_info *info,
992 struct sk_buff *skb)
993 {
994 struct nf_conntrack_expect *exp;
995
996 /* If we pass an expected packet through nf_conntrack_in() the
997 * expectation is typically removed, but the packet could still be
998 * lost in upcall processing. To prevent this from happening we
999 * perform an explicit expectation lookup. Expected connections are
1000 * always new, and will be passed through conntrack only when they are
1001 * committed, as it is OK to remove the expectation at that time.
1002 */
1003 exp = ovs_ct_expect_find(net, &info->zone, info->family, skb);
1004 if (exp) {
1005 u8 state;
1006
1007 /* NOTE: New connections are NATted and Helped only when
1008 * committed, so we are not calling into NAT here.
1009 */
1010 state = OVS_CS_F_TRACKED | OVS_CS_F_NEW | OVS_CS_F_RELATED;
1011 __ovs_ct_update_key(key, state, &info->zone, exp->master);
1012 } else {
1013 struct nf_conn *ct;
1014 int err;
1015
1016 err = __ovs_ct_lookup(net, key, info, skb);
1017 if (err)
1018 return err;
1019
1020 ct = (struct nf_conn *)skb_nfct(skb);
1021 if (ct)
1022 nf_ct_deliver_cached_events(ct);
1023 }
1024
1025 return 0;
1026 }
1027
1028 static bool labels_nonzero(const struct ovs_key_ct_labels *labels)
1029 {
1030 size_t i;
1031
1032 for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
1033 if (labels->ct_labels_32[i])
1034 return true;
1035
1036 return false;
1037 }
1038
1039 /* Lookup connection and confirm if unconfirmed. */
1040 static int ovs_ct_commit(struct net *net, struct sw_flow_key *key,
1041 const struct ovs_conntrack_info *info,
1042 struct sk_buff *skb)
1043 {
1044 enum ip_conntrack_info ctinfo;
1045 struct nf_conn *ct;
1046 int err;
1047
1048 err = __ovs_ct_lookup(net, key, info, skb);
1049 if (err)
1050 return err;
1051
1052 /* The connection could be invalid, in which case this is a no-op.*/
1053 ct = nf_ct_get(skb, &ctinfo);
1054 if (!ct)
1055 return 0;
1056
1057 /* Set the conntrack event mask if given. NEW and DELETE events have
1058 * their own groups, but the NFNLGRP_CONNTRACK_UPDATE group listener
1059 * typically would receive many kinds of updates. Setting the event
1060 * mask allows those events to be filtered. The set event mask will
1061 * remain in effect for the lifetime of the connection unless changed
1062 * by a further CT action with both the commit flag and the eventmask
1063 * option. */
1064 if (info->have_eventmask) {
1065 struct nf_conntrack_ecache *cache = nf_ct_ecache_find(ct);
1066
1067 if (cache)
1068 cache->ctmask = info->eventmask;
1069 }
1070
1071 /* Apply changes before confirming the connection so that the initial
1072 * conntrack NEW netlink event carries the values given in the CT
1073 * action.
1074 */
1075 if (info->mark.mask) {
1076 err = ovs_ct_set_mark(ct, key, info->mark.value,
1077 info->mark.mask);
1078 if (err)
1079 return err;
1080 }
1081 if (!nf_ct_is_confirmed(ct)) {
1082 err = ovs_ct_init_labels(ct, key, &info->labels.value,
1083 &info->labels.mask);
1084 if (err)
1085 return err;
1086 } else if (labels_nonzero(&info->labels.mask)) {
1087 err = ovs_ct_set_labels(ct, key, &info->labels.value,
1088 &info->labels.mask);
1089 if (err)
1090 return err;
1091 }
1092 /* This will take care of sending queued events even if the connection
1093 * is already confirmed.
1094 */
1095 if (nf_conntrack_confirm(skb) != NF_ACCEPT)
1096 return -EINVAL;
1097
1098 return 0;
1099 }
1100
1101 /* Trim the skb to the length specified by the IP/IPv6 header,
1102 * removing any trailing lower-layer padding. This prepares the skb
1103 * for higher-layer processing that assumes skb->len excludes padding
1104 * (such as nf_ip_checksum). The caller needs to pull the skb to the
1105 * network header, and ensure ip_hdr/ipv6_hdr points to valid data.
1106 */
1107 static int ovs_skb_network_trim(struct sk_buff *skb)
1108 {
1109 unsigned int len;
1110 int err;
1111
1112 switch (skb->protocol) {
1113 case htons(ETH_P_IP):
1114 len = ntohs(ip_hdr(skb)->tot_len);
1115 break;
1116 case htons(ETH_P_IPV6):
1117 len = sizeof(struct ipv6hdr)
1118 + ntohs(ipv6_hdr(skb)->payload_len);
1119 break;
1120 default:
1121 len = skb->len;
1122 }
1123
1124 err = pskb_trim_rcsum(skb, len);
1125 if (err)
1126 kfree_skb(skb);
1127
1128 return err;
1129 }
1130
1131 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
1132 * value if 'skb' is freed.
1133 */
1134 int ovs_ct_execute(struct net *net, struct sk_buff *skb,
1135 struct sw_flow_key *key,
1136 const struct ovs_conntrack_info *info)
1137 {
1138 int nh_ofs;
1139 int err;
1140
1141 /* The conntrack module expects to be working at L3. */
1142 nh_ofs = skb_network_offset(skb);
1143 skb_pull_rcsum(skb, nh_ofs);
1144
1145 err = ovs_skb_network_trim(skb);
1146 if (err)
1147 return err;
1148
1149 if (key->ip.frag != OVS_FRAG_TYPE_NONE) {
1150 err = handle_fragments(net, key, info->zone.id, skb);
1151 if (err)
1152 return err;
1153 }
1154
1155 if (info->commit)
1156 err = ovs_ct_commit(net, key, info, skb);
1157 else
1158 err = ovs_ct_lookup(net, key, info, skb);
1159
1160 skb_push(skb, nh_ofs);
1161 skb_postpush_rcsum(skb, skb->data, nh_ofs);
1162 if (err)
1163 kfree_skb(skb);
1164 return err;
1165 }
1166
1167 static int ovs_ct_add_helper(struct ovs_conntrack_info *info, const char *name,
1168 const struct sw_flow_key *key, bool log)
1169 {
1170 struct nf_conntrack_helper *helper;
1171 struct nf_conn_help *help;
1172
1173 helper = nf_conntrack_helper_try_module_get(name, info->family,
1174 key->ip.proto);
1175 if (!helper) {
1176 OVS_NLERR(log, "Unknown helper \"%s\"", name);
1177 return -EINVAL;
1178 }
1179
1180 help = nf_ct_helper_ext_add(info->ct, helper, GFP_KERNEL);
1181 if (!help) {
1182 nf_conntrack_helper_put(helper);
1183 return -ENOMEM;
1184 }
1185
1186 rcu_assign_pointer(help->helper, helper);
1187 info->helper = helper;
1188 return 0;
1189 }
1190
1191 #ifdef CONFIG_NF_NAT_NEEDED
1192 static int parse_nat(const struct nlattr *attr,
1193 struct ovs_conntrack_info *info, bool log)
1194 {
1195 struct nlattr *a;
1196 int rem;
1197 bool have_ip_max = false;
1198 bool have_proto_max = false;
1199 bool ip_vers = (info->family == NFPROTO_IPV6);
1200
1201 nla_for_each_nested(a, attr, rem) {
1202 static const int ovs_nat_attr_lens[OVS_NAT_ATTR_MAX + 1][2] = {
1203 [OVS_NAT_ATTR_SRC] = {0, 0},
1204 [OVS_NAT_ATTR_DST] = {0, 0},
1205 [OVS_NAT_ATTR_IP_MIN] = {sizeof(struct in_addr),
1206 sizeof(struct in6_addr)},
1207 [OVS_NAT_ATTR_IP_MAX] = {sizeof(struct in_addr),
1208 sizeof(struct in6_addr)},
1209 [OVS_NAT_ATTR_PROTO_MIN] = {sizeof(u16), sizeof(u16)},
1210 [OVS_NAT_ATTR_PROTO_MAX] = {sizeof(u16), sizeof(u16)},
1211 [OVS_NAT_ATTR_PERSISTENT] = {0, 0},
1212 [OVS_NAT_ATTR_PROTO_HASH] = {0, 0},
1213 [OVS_NAT_ATTR_PROTO_RANDOM] = {0, 0},
1214 };
1215 int type = nla_type(a);
1216
1217 if (type > OVS_NAT_ATTR_MAX) {
1218 OVS_NLERR(log, "Unknown NAT attribute (type=%d, max=%d)",
1219 type, OVS_NAT_ATTR_MAX);
1220 return -EINVAL;
1221 }
1222
1223 if (nla_len(a) != ovs_nat_attr_lens[type][ip_vers]) {
1224 OVS_NLERR(log, "NAT attribute type %d has unexpected length (%d != %d)",
1225 type, nla_len(a),
1226 ovs_nat_attr_lens[type][ip_vers]);
1227 return -EINVAL;
1228 }
1229
1230 switch (type) {
1231 case OVS_NAT_ATTR_SRC:
1232 case OVS_NAT_ATTR_DST:
1233 if (info->nat) {
1234 OVS_NLERR(log, "Only one type of NAT may be specified");
1235 return -ERANGE;
1236 }
1237 info->nat |= OVS_CT_NAT;
1238 info->nat |= ((type == OVS_NAT_ATTR_SRC)
1239 ? OVS_CT_SRC_NAT : OVS_CT_DST_NAT);
1240 break;
1241
1242 case OVS_NAT_ATTR_IP_MIN:
1243 nla_memcpy(&info->range.min_addr, a,
1244 sizeof(info->range.min_addr));
1245 info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1246 break;
1247
1248 case OVS_NAT_ATTR_IP_MAX:
1249 have_ip_max = true;
1250 nla_memcpy(&info->range.max_addr, a,
1251 sizeof(info->range.max_addr));
1252 info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1253 break;
1254
1255 case OVS_NAT_ATTR_PROTO_MIN:
1256 info->range.min_proto.all = htons(nla_get_u16(a));
1257 info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1258 break;
1259
1260 case OVS_NAT_ATTR_PROTO_MAX:
1261 have_proto_max = true;
1262 info->range.max_proto.all = htons(nla_get_u16(a));
1263 info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1264 break;
1265
1266 case OVS_NAT_ATTR_PERSISTENT:
1267 info->range.flags |= NF_NAT_RANGE_PERSISTENT;
1268 break;
1269
1270 case OVS_NAT_ATTR_PROTO_HASH:
1271 info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM;
1272 break;
1273
1274 case OVS_NAT_ATTR_PROTO_RANDOM:
1275 info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM_FULLY;
1276 break;
1277
1278 default:
1279 OVS_NLERR(log, "Unknown nat attribute (%d)", type);
1280 return -EINVAL;
1281 }
1282 }
1283
1284 if (rem > 0) {
1285 OVS_NLERR(log, "NAT attribute has %d unknown bytes", rem);
1286 return -EINVAL;
1287 }
1288 if (!info->nat) {
1289 /* Do not allow flags if no type is given. */
1290 if (info->range.flags) {
1291 OVS_NLERR(log,
1292 "NAT flags may be given only when NAT range (SRC or DST) is also specified.\n"
1293 );
1294 return -EINVAL;
1295 }
1296 info->nat = OVS_CT_NAT; /* NAT existing connections. */
1297 } else if (!info->commit) {
1298 OVS_NLERR(log,
1299 "NAT attributes may be specified only when CT COMMIT flag is also specified.\n"
1300 );
1301 return -EINVAL;
1302 }
1303 /* Allow missing IP_MAX. */
1304 if (info->range.flags & NF_NAT_RANGE_MAP_IPS && !have_ip_max) {
1305 memcpy(&info->range.max_addr, &info->range.min_addr,
1306 sizeof(info->range.max_addr));
1307 }
1308 /* Allow missing PROTO_MAX. */
1309 if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1310 !have_proto_max) {
1311 info->range.max_proto.all = info->range.min_proto.all;
1312 }
1313 return 0;
1314 }
1315 #endif
1316
1317 static const struct ovs_ct_len_tbl ovs_ct_attr_lens[OVS_CT_ATTR_MAX + 1] = {
1318 [OVS_CT_ATTR_COMMIT] = { .minlen = 0, .maxlen = 0 },
1319 [OVS_CT_ATTR_FORCE_COMMIT] = { .minlen = 0, .maxlen = 0 },
1320 [OVS_CT_ATTR_ZONE] = { .minlen = sizeof(u16),
1321 .maxlen = sizeof(u16) },
1322 [OVS_CT_ATTR_MARK] = { .minlen = sizeof(struct md_mark),
1323 .maxlen = sizeof(struct md_mark) },
1324 [OVS_CT_ATTR_LABELS] = { .minlen = sizeof(struct md_labels),
1325 .maxlen = sizeof(struct md_labels) },
1326 [OVS_CT_ATTR_HELPER] = { .minlen = 1,
1327 .maxlen = NF_CT_HELPER_NAME_LEN },
1328 #ifdef CONFIG_NF_NAT_NEEDED
1329 /* NAT length is checked when parsing the nested attributes. */
1330 [OVS_CT_ATTR_NAT] = { .minlen = 0, .maxlen = INT_MAX },
1331 #endif
1332 [OVS_CT_ATTR_EVENTMASK] = { .minlen = sizeof(u32),
1333 .maxlen = sizeof(u32) },
1334 };
1335
1336 static int parse_ct(const struct nlattr *attr, struct ovs_conntrack_info *info,
1337 const char **helper, bool log)
1338 {
1339 struct nlattr *a;
1340 int rem;
1341
1342 nla_for_each_nested(a, attr, rem) {
1343 int type = nla_type(a);
1344 int maxlen;
1345 int minlen;
1346
1347 if (type > OVS_CT_ATTR_MAX) {
1348 OVS_NLERR(log,
1349 "Unknown conntrack attr (type=%d, max=%d)",
1350 type, OVS_CT_ATTR_MAX);
1351 return -EINVAL;
1352 }
1353
1354 maxlen = ovs_ct_attr_lens[type].maxlen;
1355 minlen = ovs_ct_attr_lens[type].minlen;
1356 if (nla_len(a) < minlen || nla_len(a) > maxlen) {
1357 OVS_NLERR(log,
1358 "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
1359 type, nla_len(a), maxlen);
1360 return -EINVAL;
1361 }
1362
1363 switch (type) {
1364 case OVS_CT_ATTR_FORCE_COMMIT:
1365 info->force = true;
1366 /* fall through. */
1367 case OVS_CT_ATTR_COMMIT:
1368 info->commit = true;
1369 break;
1370 #ifdef CONFIG_NF_CONNTRACK_ZONES
1371 case OVS_CT_ATTR_ZONE:
1372 info->zone.id = nla_get_u16(a);
1373 break;
1374 #endif
1375 #ifdef CONFIG_NF_CONNTRACK_MARK
1376 case OVS_CT_ATTR_MARK: {
1377 struct md_mark *mark = nla_data(a);
1378
1379 if (!mark->mask) {
1380 OVS_NLERR(log, "ct_mark mask cannot be 0");
1381 return -EINVAL;
1382 }
1383 info->mark = *mark;
1384 break;
1385 }
1386 #endif
1387 #ifdef CONFIG_NF_CONNTRACK_LABELS
1388 case OVS_CT_ATTR_LABELS: {
1389 struct md_labels *labels = nla_data(a);
1390
1391 if (!labels_nonzero(&labels->mask)) {
1392 OVS_NLERR(log, "ct_labels mask cannot be 0");
1393 return -EINVAL;
1394 }
1395 info->labels = *labels;
1396 break;
1397 }
1398 #endif
1399 case OVS_CT_ATTR_HELPER:
1400 *helper = nla_data(a);
1401 if (!memchr(*helper, '\0', nla_len(a))) {
1402 OVS_NLERR(log, "Invalid conntrack helper");
1403 return -EINVAL;
1404 }
1405 break;
1406 #ifdef CONFIG_NF_NAT_NEEDED
1407 case OVS_CT_ATTR_NAT: {
1408 int err = parse_nat(a, info, log);
1409
1410 if (err)
1411 return err;
1412 break;
1413 }
1414 #endif
1415 case OVS_CT_ATTR_EVENTMASK:
1416 info->have_eventmask = true;
1417 info->eventmask = nla_get_u32(a);
1418 break;
1419
1420 default:
1421 OVS_NLERR(log, "Unknown conntrack attr (%d)",
1422 type);
1423 return -EINVAL;
1424 }
1425 }
1426
1427 #ifdef CONFIG_NF_CONNTRACK_MARK
1428 if (!info->commit && info->mark.mask) {
1429 OVS_NLERR(log,
1430 "Setting conntrack mark requires 'commit' flag.");
1431 return -EINVAL;
1432 }
1433 #endif
1434 #ifdef CONFIG_NF_CONNTRACK_LABELS
1435 if (!info->commit && labels_nonzero(&info->labels.mask)) {
1436 OVS_NLERR(log,
1437 "Setting conntrack labels requires 'commit' flag.");
1438 return -EINVAL;
1439 }
1440 #endif
1441 if (rem > 0) {
1442 OVS_NLERR(log, "Conntrack attr has %d unknown bytes", rem);
1443 return -EINVAL;
1444 }
1445
1446 return 0;
1447 }
1448
1449 bool ovs_ct_verify(struct net *net, enum ovs_key_attr attr)
1450 {
1451 if (attr == OVS_KEY_ATTR_CT_STATE)
1452 return true;
1453 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1454 attr == OVS_KEY_ATTR_CT_ZONE)
1455 return true;
1456 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1457 attr == OVS_KEY_ATTR_CT_MARK)
1458 return true;
1459 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1460 attr == OVS_KEY_ATTR_CT_LABELS) {
1461 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1462
1463 return ovs_net->xt_label;
1464 }
1465
1466 return false;
1467 }
1468
1469 int ovs_ct_copy_action(struct net *net, const struct nlattr *attr,
1470 const struct sw_flow_key *key,
1471 struct sw_flow_actions **sfa, bool log)
1472 {
1473 struct ovs_conntrack_info ct_info;
1474 const char *helper = NULL;
1475 u16 family;
1476 int err;
1477
1478 family = key_to_nfproto(key);
1479 if (family == NFPROTO_UNSPEC) {
1480 OVS_NLERR(log, "ct family unspecified");
1481 return -EINVAL;
1482 }
1483
1484 memset(&ct_info, 0, sizeof(ct_info));
1485 ct_info.family = family;
1486
1487 nf_ct_zone_init(&ct_info.zone, NF_CT_DEFAULT_ZONE_ID,
1488 NF_CT_DEFAULT_ZONE_DIR, 0);
1489
1490 err = parse_ct(attr, &ct_info, &helper, log);
1491 if (err)
1492 return err;
1493
1494 /* Set up template for tracking connections in specific zones. */
1495 ct_info.ct = nf_ct_tmpl_alloc(net, &ct_info.zone, GFP_KERNEL);
1496 if (!ct_info.ct) {
1497 OVS_NLERR(log, "Failed to allocate conntrack template");
1498 return -ENOMEM;
1499 }
1500
1501 __set_bit(IPS_CONFIRMED_BIT, &ct_info.ct->status);
1502 nf_conntrack_get(&ct_info.ct->ct_general);
1503
1504 if (helper) {
1505 err = ovs_ct_add_helper(&ct_info, helper, key, log);
1506 if (err)
1507 goto err_free_ct;
1508 }
1509
1510 err = ovs_nla_add_action(sfa, OVS_ACTION_ATTR_CT, &ct_info,
1511 sizeof(ct_info), log);
1512 if (err)
1513 goto err_free_ct;
1514
1515 return 0;
1516 err_free_ct:
1517 __ovs_ct_free_action(&ct_info);
1518 return err;
1519 }
1520
1521 #ifdef CONFIG_NF_NAT_NEEDED
1522 static bool ovs_ct_nat_to_attr(const struct ovs_conntrack_info *info,
1523 struct sk_buff *skb)
1524 {
1525 struct nlattr *start;
1526
1527 start = nla_nest_start(skb, OVS_CT_ATTR_NAT);
1528 if (!start)
1529 return false;
1530
1531 if (info->nat & OVS_CT_SRC_NAT) {
1532 if (nla_put_flag(skb, OVS_NAT_ATTR_SRC))
1533 return false;
1534 } else if (info->nat & OVS_CT_DST_NAT) {
1535 if (nla_put_flag(skb, OVS_NAT_ATTR_DST))
1536 return false;
1537 } else {
1538 goto out;
1539 }
1540
1541 if (info->range.flags & NF_NAT_RANGE_MAP_IPS) {
1542 if (IS_ENABLED(CONFIG_NF_NAT_IPV4) &&
1543 info->family == NFPROTO_IPV4) {
1544 if (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MIN,
1545 info->range.min_addr.ip) ||
1546 (info->range.max_addr.ip
1547 != info->range.min_addr.ip &&
1548 (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MAX,
1549 info->range.max_addr.ip))))
1550 return false;
1551 } else if (IS_ENABLED(CONFIG_NF_NAT_IPV6) &&
1552 info->family == NFPROTO_IPV6) {
1553 if (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MIN,
1554 &info->range.min_addr.in6) ||
1555 (memcmp(&info->range.max_addr.in6,
1556 &info->range.min_addr.in6,
1557 sizeof(info->range.max_addr.in6)) &&
1558 (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MAX,
1559 &info->range.max_addr.in6))))
1560 return false;
1561 } else {
1562 return false;
1563 }
1564 }
1565 if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1566 (nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MIN,
1567 ntohs(info->range.min_proto.all)) ||
1568 (info->range.max_proto.all != info->range.min_proto.all &&
1569 nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MAX,
1570 ntohs(info->range.max_proto.all)))))
1571 return false;
1572
1573 if (info->range.flags & NF_NAT_RANGE_PERSISTENT &&
1574 nla_put_flag(skb, OVS_NAT_ATTR_PERSISTENT))
1575 return false;
1576 if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM &&
1577 nla_put_flag(skb, OVS_NAT_ATTR_PROTO_HASH))
1578 return false;
1579 if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM_FULLY &&
1580 nla_put_flag(skb, OVS_NAT_ATTR_PROTO_RANDOM))
1581 return false;
1582 out:
1583 nla_nest_end(skb, start);
1584
1585 return true;
1586 }
1587 #endif
1588
1589 int ovs_ct_action_to_attr(const struct ovs_conntrack_info *ct_info,
1590 struct sk_buff *skb)
1591 {
1592 struct nlattr *start;
1593
1594 start = nla_nest_start(skb, OVS_ACTION_ATTR_CT);
1595 if (!start)
1596 return -EMSGSIZE;
1597
1598 if (ct_info->commit && nla_put_flag(skb, ct_info->force
1599 ? OVS_CT_ATTR_FORCE_COMMIT
1600 : OVS_CT_ATTR_COMMIT))
1601 return -EMSGSIZE;
1602 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1603 nla_put_u16(skb, OVS_CT_ATTR_ZONE, ct_info->zone.id))
1604 return -EMSGSIZE;
1605 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && ct_info->mark.mask &&
1606 nla_put(skb, OVS_CT_ATTR_MARK, sizeof(ct_info->mark),
1607 &ct_info->mark))
1608 return -EMSGSIZE;
1609 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1610 labels_nonzero(&ct_info->labels.mask) &&
1611 nla_put(skb, OVS_CT_ATTR_LABELS, sizeof(ct_info->labels),
1612 &ct_info->labels))
1613 return -EMSGSIZE;
1614 if (ct_info->helper) {
1615 if (nla_put_string(skb, OVS_CT_ATTR_HELPER,
1616 ct_info->helper->name))
1617 return -EMSGSIZE;
1618 }
1619 if (ct_info->have_eventmask &&
1620 nla_put_u32(skb, OVS_CT_ATTR_EVENTMASK, ct_info->eventmask))
1621 return -EMSGSIZE;
1622
1623 #ifdef CONFIG_NF_NAT_NEEDED
1624 if (ct_info->nat && !ovs_ct_nat_to_attr(ct_info, skb))
1625 return -EMSGSIZE;
1626 #endif
1627 nla_nest_end(skb, start);
1628
1629 return 0;
1630 }
1631
1632 void ovs_ct_free_action(const struct nlattr *a)
1633 {
1634 struct ovs_conntrack_info *ct_info = nla_data(a);
1635
1636 __ovs_ct_free_action(ct_info);
1637 }
1638
1639 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info)
1640 {
1641 if (ct_info->helper)
1642 nf_conntrack_helper_put(ct_info->helper);
1643 if (ct_info->ct)
1644 nf_ct_tmpl_free(ct_info->ct);
1645 }
1646
1647 void ovs_ct_init(struct net *net)
1648 {
1649 unsigned int n_bits = sizeof(struct ovs_key_ct_labels) * BITS_PER_BYTE;
1650 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1651
1652 if (nf_connlabels_get(net, n_bits - 1)) {
1653 ovs_net->xt_label = false;
1654 OVS_NLERR(true, "Failed to set connlabel length");
1655 } else {
1656 ovs_net->xt_label = true;
1657 }
1658 }
1659
1660 void ovs_ct_exit(struct net *net)
1661 {
1662 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1663
1664 if (ovs_net->xt_label)
1665 nf_connlabels_put(net);
1666 }
Linux with added FPC-III support