?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
seccomp: Fix ioctl number for SECCOMP_IOCTL_NOTIF_ID_VALID
[linux/fpc-iii.git] / net / openvswitch / conntrack.c
blobc86e404cd65bb545872f207ce161bbb57e923c80
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
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. */
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
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
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
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
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 */
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 */
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 int ovs_ct_put_key(const struct sw_flow_key *swkey,
280 const struct sw_flow_key *output, struct sk_buff *skb)
281 {
282 if (nla_put_u32(skb, OVS_KEY_ATTR_CT_STATE, output->ct_state))
283 return -EMSGSIZE;
284
285 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
286 nla_put_u16(skb, OVS_KEY_ATTR_CT_ZONE, output->ct_zone))
287 return -EMSGSIZE;
288
289 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
290 nla_put_u32(skb, OVS_KEY_ATTR_CT_MARK, output->ct.mark))
291 return -EMSGSIZE;
292
293 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
294 nla_put(skb, OVS_KEY_ATTR_CT_LABELS, sizeof(output->ct.labels),
295 &output->ct.labels))
296 return -EMSGSIZE;
297
298 if (swkey->ct_orig_proto) {
299 if (swkey->eth.type == htons(ETH_P_IP)) {
300 struct ovs_key_ct_tuple_ipv4 orig;
301
302 memset(&orig, 0, sizeof(orig));
303 orig.ipv4_src = output->ipv4.ct_orig.src;
304 orig.ipv4_dst = output->ipv4.ct_orig.dst;
305 orig.src_port = output->ct.orig_tp.src;
306 orig.dst_port = output->ct.orig_tp.dst;
307 orig.ipv4_proto = output->ct_orig_proto;
308
309 if (nla_put(skb, OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4,
310 sizeof(orig), &orig))
311 return -EMSGSIZE;
312 } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
313 struct ovs_key_ct_tuple_ipv6 orig;
314
315 memset(&orig, 0, sizeof(orig));
316 memcpy(orig.ipv6_src, output->ipv6.ct_orig.src.s6_addr32,
317 sizeof(orig.ipv6_src));
318 memcpy(orig.ipv6_dst, output->ipv6.ct_orig.dst.s6_addr32,
319 sizeof(orig.ipv6_dst));
320 orig.src_port = output->ct.orig_tp.src;
321 orig.dst_port = output->ct.orig_tp.dst;
322 orig.ipv6_proto = output->ct_orig_proto;
323
324 if (nla_put(skb, OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6,
325 sizeof(orig), &orig))
326 return -EMSGSIZE;
327 }
328 }
329
330 return 0;
331 }
332
333 static int ovs_ct_set_mark(struct nf_conn *ct, struct sw_flow_key *key,
334 u32 ct_mark, u32 mask)
335 {
336 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
337 u32 new_mark;
338
339 new_mark = ct_mark | (ct->mark & ~(mask));
340 if (ct->mark != new_mark) {
341 ct->mark = new_mark;
342 if (nf_ct_is_confirmed(ct))
343 nf_conntrack_event_cache(IPCT_MARK, ct);
344 key->ct.mark = new_mark;
345 }
346
347 return 0;
348 #else
349 return -ENOTSUPP;
350 #endif
351 }
352
353 static struct nf_conn_labels *ovs_ct_get_conn_labels(struct nf_conn *ct)
354 {
355 struct nf_conn_labels *cl;
356
357 cl = nf_ct_labels_find(ct);
358 if (!cl) {
359 nf_ct_labels_ext_add(ct);
360 cl = nf_ct_labels_find(ct);
361 }
362
363 return cl;
364 }
365
366 /* Initialize labels for a new, yet to be committed conntrack entry. Note that
367 * since the new connection is not yet confirmed, and thus no-one else has
368 * access to it's labels, we simply write them over.
369 */
370 static int ovs_ct_init_labels(struct nf_conn *ct, struct sw_flow_key *key,
371 const struct ovs_key_ct_labels *labels,
372 const struct ovs_key_ct_labels *mask)
373 {
374 struct nf_conn_labels *cl, *master_cl;
375 bool have_mask = labels_nonzero(mask);
376
377 /* Inherit master's labels to the related connection? */
378 master_cl = ct->master ? nf_ct_labels_find(ct->master) : NULL;
379
380 if (!master_cl && !have_mask)
381 return 0; /* Nothing to do. */
382
383 cl = ovs_ct_get_conn_labels(ct);
384 if (!cl)
385 return -ENOSPC;
386
387 /* Inherit the master's labels, if any. */
388 if (master_cl)
389 *cl = *master_cl;
390
391 if (have_mask) {
392 u32 *dst = (u32 *)cl->bits;
393 int i;
394
395 for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
396 dst[i] = (dst[i] & ~mask->ct_labels_32[i]) |
397 (labels->ct_labels_32[i]
398 & mask->ct_labels_32[i]);
399 }
400
401 /* Labels are included in the IPCTNL_MSG_CT_NEW event only if the
402 * IPCT_LABEL bit is set in the event cache.
403 */
404 nf_conntrack_event_cache(IPCT_LABEL, ct);
405
406 memcpy(&key->ct.labels, cl->bits, OVS_CT_LABELS_LEN);
407
408 return 0;
409 }
410
411 static int ovs_ct_set_labels(struct nf_conn *ct, struct sw_flow_key *key,
412 const struct ovs_key_ct_labels *labels,
413 const struct ovs_key_ct_labels *mask)
414 {
415 struct nf_conn_labels *cl;
416 int err;
417
418 cl = ovs_ct_get_conn_labels(ct);
419 if (!cl)
420 return -ENOSPC;
421
422 err = nf_connlabels_replace(ct, labels->ct_labels_32,
423 mask->ct_labels_32,
424 OVS_CT_LABELS_LEN_32);
425 if (err)
426 return err;
427
428 memcpy(&key->ct.labels, cl->bits, OVS_CT_LABELS_LEN);
429
430 return 0;
431 }
432
433 /* 'skb' should already be pulled to nh_ofs. */
434 static int ovs_ct_helper(struct sk_buff *skb, u16 proto)
435 {
436 const struct nf_conntrack_helper *helper;
437 const struct nf_conn_help *help;
438 enum ip_conntrack_info ctinfo;
439 unsigned int protoff;
440 struct nf_conn *ct;
441 int err;
442
443 ct = nf_ct_get(skb, &ctinfo);
444 if (!ct || ctinfo == IP_CT_RELATED_REPLY)
445 return NF_ACCEPT;
446
447 help = nfct_help(ct);
448 if (!help)
449 return NF_ACCEPT;
450
451 helper = rcu_dereference(help->helper);
452 if (!helper)
453 return NF_ACCEPT;
454
455 switch (proto) {
456 case NFPROTO_IPV4:
457 protoff = ip_hdrlen(skb);
458 break;
459 case NFPROTO_IPV6: {
460 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
461 __be16 frag_off;
462 int ofs;
463
464 ofs = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr,
465 &frag_off);
466 if (ofs < 0 || (frag_off & htons(~0x7)) != 0) {
467 pr_debug("proto header not found\n");
468 return NF_ACCEPT;
469 }
470 protoff = ofs;
471 break;
472 }
473 default:
474 WARN_ONCE(1, "helper invoked on non-IP family!");
475 return NF_DROP;
476 }
477
478 err = helper->help(skb, protoff, ct, ctinfo);
479 if (err != NF_ACCEPT)
480 return err;
481
482 /* Adjust seqs after helper. This is needed due to some helpers (e.g.,
483 * FTP with NAT) adusting the TCP payload size when mangling IP
484 * addresses and/or port numbers in the text-based control connection.
485 */
486 if (test_bit(IPS_SEQ_ADJUST_BIT, &ct->status) &&
487 !nf_ct_seq_adjust(skb, ct, ctinfo, protoff))
488 return NF_DROP;
489 return NF_ACCEPT;
490 }
491
492 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
493 * value if 'skb' is freed.
494 */
495 static int handle_fragments(struct net *net, struct sw_flow_key *key,
496 u16 zone, struct sk_buff *skb)
497 {
498 struct ovs_skb_cb ovs_cb = *OVS_CB(skb);
499 int err;
500
501 if (key->eth.type == htons(ETH_P_IP)) {
502 enum ip_defrag_users user = IP_DEFRAG_CONNTRACK_IN + zone;
503
504 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
505 err = ip_defrag(net, skb, user);
506 if (err)
507 return err;
508
509 ovs_cb.mru = IPCB(skb)->frag_max_size;
510 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
511 } else if (key->eth.type == htons(ETH_P_IPV6)) {
512 enum ip6_defrag_users user = IP6_DEFRAG_CONNTRACK_IN + zone;
513
514 memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm));
515 err = nf_ct_frag6_gather(net, skb, user);
516 if (err) {
517 if (err != -EINPROGRESS)
518 kfree_skb(skb);
519 return err;
520 }
521
522 key->ip.proto = ipv6_hdr(skb)->nexthdr;
523 ovs_cb.mru = IP6CB(skb)->frag_max_size;
524 #endif
525 } else {
526 kfree_skb(skb);
527 return -EPFNOSUPPORT;
528 }
529
530 /* The key extracted from the fragment that completed this datagram
531 * likely didn't have an L4 header, so regenerate it.
532 */
533 ovs_flow_key_update_l3l4(skb, key);
534
535 key->ip.frag = OVS_FRAG_TYPE_NONE;
536 skb_clear_hash(skb);
537 skb->ignore_df = 1;
538 *OVS_CB(skb) = ovs_cb;
539
540 return 0;
541 }
542
543 static struct nf_conntrack_expect *
544 ovs_ct_expect_find(struct net *net, const struct nf_conntrack_zone *zone,
545 u16 proto, const struct sk_buff *skb)
546 {
547 struct nf_conntrack_tuple tuple;
548 struct nf_conntrack_expect *exp;
549
550 if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), proto, net, &tuple))
551 return NULL;
552
553 exp = __nf_ct_expect_find(net, zone, &tuple);
554 if (exp) {
555 struct nf_conntrack_tuple_hash *h;
556
557 /* Delete existing conntrack entry, if it clashes with the
558 * expectation. This can happen since conntrack ALGs do not
559 * check for clashes between (new) expectations and existing
560 * conntrack entries. nf_conntrack_in() will check the
561 * expectations only if a conntrack entry can not be found,
562 * which can lead to OVS finding the expectation (here) in the
563 * init direction, but which will not be removed by the
564 * nf_conntrack_in() call, if a matching conntrack entry is
565 * found instead. In this case all init direction packets
566 * would be reported as new related packets, while reply
567 * direction packets would be reported as un-related
568 * established packets.
569 */
570 h = nf_conntrack_find_get(net, zone, &tuple);
571 if (h) {
572 struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
573
574 nf_ct_delete(ct, 0, 0);
575 nf_conntrack_put(&ct->ct_general);
576 }
577 }
578
579 return exp;
580 }
581
582 /* This replicates logic from nf_conntrack_core.c that is not exported. */
583 static enum ip_conntrack_info
584 ovs_ct_get_info(const struct nf_conntrack_tuple_hash *h)
585 {
586 const struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
587
588 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY)
589 return IP_CT_ESTABLISHED_REPLY;
590 /* Once we've had two way comms, always ESTABLISHED. */
591 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status))
592 return IP_CT_ESTABLISHED;
593 if (test_bit(IPS_EXPECTED_BIT, &ct->status))
594 return IP_CT_RELATED;
595 return IP_CT_NEW;
596 }
597
598 /* Find an existing connection which this packet belongs to without
599 * re-attributing statistics or modifying the connection state. This allows an
600 * skb->_nfct lost due to an upcall to be recovered during actions execution.
601 *
602 * Must be called with rcu_read_lock.
603 *
604 * On success, populates skb->_nfct and returns the connection. Returns NULL
605 * if there is no existing entry.
606 */
607 static struct nf_conn *
608 ovs_ct_find_existing(struct net *net, const struct nf_conntrack_zone *zone,
609 u8 l3num, struct sk_buff *skb, bool natted)
610 {
611 struct nf_conntrack_tuple tuple;
612 struct nf_conntrack_tuple_hash *h;
613 struct nf_conn *ct;
614
615 if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), l3num,
616 net, &tuple)) {
617 pr_debug("ovs_ct_find_existing: Can't get tuple\n");
618 return NULL;
619 }
620
621 /* Must invert the tuple if skb has been transformed by NAT. */
622 if (natted) {
623 struct nf_conntrack_tuple inverse;
624
625 if (!nf_ct_invert_tuple(&inverse, &tuple)) {
626 pr_debug("ovs_ct_find_existing: Inversion failed!\n");
627 return NULL;
628 }
629 tuple = inverse;
630 }
631
632 /* look for tuple match */
633 h = nf_conntrack_find_get(net, zone, &tuple);
634 if (!h)
635 return NULL; /* Not found. */
636
637 ct = nf_ct_tuplehash_to_ctrack(h);
638
639 /* Inverted packet tuple matches the reverse direction conntrack tuple,
640 * select the other tuplehash to get the right 'ctinfo' bits for this
641 * packet.
642 */
643 if (natted)
644 h = &ct->tuplehash[!h->tuple.dst.dir];
645
646 nf_ct_set(skb, ct, ovs_ct_get_info(h));
647 return ct;
648 }
649
650 static
651 struct nf_conn *ovs_ct_executed(struct net *net,
652 const struct sw_flow_key *key,
653 const struct ovs_conntrack_info *info,
654 struct sk_buff *skb,
655 bool *ct_executed)
656 {
657 struct nf_conn *ct = NULL;
658
659 /* If no ct, check if we have evidence that an existing conntrack entry
660 * might be found for this skb. This happens when we lose a skb->_nfct
661 * due to an upcall, or if the direction is being forced. If the
662 * connection was not confirmed, it is not cached and needs to be run
663 * through conntrack again.
664 */
665 *ct_executed = (key->ct_state & OVS_CS_F_TRACKED) &&
666 !(key->ct_state & OVS_CS_F_INVALID) &&
667 (key->ct_zone == info->zone.id);
668
669 if (*ct_executed || (!key->ct_state && info->force)) {
670 ct = ovs_ct_find_existing(net, &info->zone, info->family, skb,
671 !!(key->ct_state &
672 OVS_CS_F_NAT_MASK));
673 }
674
675 return ct;
676 }
677
678 /* Determine whether skb->_nfct is equal to the result of conntrack lookup. */
679 static bool skb_nfct_cached(struct net *net,
680 const struct sw_flow_key *key,
681 const struct ovs_conntrack_info *info,
682 struct sk_buff *skb)
683 {
684 enum ip_conntrack_info ctinfo;
685 struct nf_conn *ct;
686 bool ct_executed = true;
687
688 ct = nf_ct_get(skb, &ctinfo);
689 if (!ct)
690 ct = ovs_ct_executed(net, key, info, skb, &ct_executed);
691
692 if (ct)
693 nf_ct_get(skb, &ctinfo);
694 else
695 return false;
696
697 if (!net_eq(net, read_pnet(&ct->ct_net)))
698 return false;
699 if (!nf_ct_zone_equal_any(info->ct, nf_ct_zone(ct)))
700 return false;
701 if (info->helper) {
702 struct nf_conn_help *help;
703
704 help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER);
705 if (help && rcu_access_pointer(help->helper) != info->helper)
706 return false;
707 }
708 if (info->nf_ct_timeout) {
709 struct nf_conn_timeout *timeout_ext;
710
711 timeout_ext = nf_ct_timeout_find(ct);
712 if (!timeout_ext || info->nf_ct_timeout !=
713 rcu_dereference(timeout_ext->timeout))
714 return false;
715 }
716 /* Force conntrack entry direction to the current packet? */
717 if (info->force && CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) {
718 /* Delete the conntrack entry if confirmed, else just release
719 * the reference.
720 */
721 if (nf_ct_is_confirmed(ct))
722 nf_ct_delete(ct, 0, 0);
723
724 nf_conntrack_put(&ct->ct_general);
725 nf_ct_set(skb, NULL, 0);
726 return false;
727 }
728
729 return ct_executed;
730 }
731
732 #if IS_ENABLED(CONFIG_NF_NAT)
733 /* Modelled after nf_nat_ipv[46]_fn().
734 * range is only used for new, uninitialized NAT state.
735 * Returns either NF_ACCEPT or NF_DROP.
736 */
737 static int ovs_ct_nat_execute(struct sk_buff *skb, struct nf_conn *ct,
738 enum ip_conntrack_info ctinfo,
739 const struct nf_nat_range2 *range,
740 enum nf_nat_manip_type maniptype)
741 {
742 int hooknum, nh_off, err = NF_ACCEPT;
743
744 nh_off = skb_network_offset(skb);
745 skb_pull_rcsum(skb, nh_off);
746
747 /* See HOOK2MANIP(). */
748 if (maniptype == NF_NAT_MANIP_SRC)
749 hooknum = NF_INET_LOCAL_IN; /* Source NAT */
750 else
751 hooknum = NF_INET_LOCAL_OUT; /* Destination NAT */
752
753 switch (ctinfo) {
754 case IP_CT_RELATED:
755 case IP_CT_RELATED_REPLY:
756 if (IS_ENABLED(CONFIG_NF_NAT) &&
757 skb->protocol == htons(ETH_P_IP) &&
758 ip_hdr(skb)->protocol == IPPROTO_ICMP) {
759 if (!nf_nat_icmp_reply_translation(skb, ct, ctinfo,
760 hooknum))
761 err = NF_DROP;
762 goto push;
763 } else if (IS_ENABLED(CONFIG_IPV6) &&
764 skb->protocol == htons(ETH_P_IPV6)) {
765 __be16 frag_off;
766 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
767 int hdrlen = ipv6_skip_exthdr(skb,
768 sizeof(struct ipv6hdr),
769 &nexthdr, &frag_off);
770
771 if (hdrlen >= 0 && nexthdr == IPPROTO_ICMPV6) {
772 if (!nf_nat_icmpv6_reply_translation(skb, ct,
773 ctinfo,
774 hooknum,
775 hdrlen))
776 err = NF_DROP;
777 goto push;
778 }
779 }
780 /* Non-ICMP, fall thru to initialize if needed. */
781 /* fall through */
782 case IP_CT_NEW:
783 /* Seen it before? This can happen for loopback, retrans,
784 * or local packets.
785 */
786 if (!nf_nat_initialized(ct, maniptype)) {
787 /* Initialize according to the NAT action. */
788 err = (range && range->flags & NF_NAT_RANGE_MAP_IPS)
789 /* Action is set up to establish a new
790 * mapping.
791 */
792 ? nf_nat_setup_info(ct, range, maniptype)
793 : nf_nat_alloc_null_binding(ct, hooknum);
794 if (err != NF_ACCEPT)
795 goto push;
796 }
797 break;
798
799 case IP_CT_ESTABLISHED:
800 case IP_CT_ESTABLISHED_REPLY:
801 break;
802
803 default:
804 err = NF_DROP;
805 goto push;
806 }
807
808 err = nf_nat_packet(ct, ctinfo, hooknum, skb);
809 push:
810 skb_push(skb, nh_off);
811 skb_postpush_rcsum(skb, skb->data, nh_off);
812
813 return err;
814 }
815
816 static void ovs_nat_update_key(struct sw_flow_key *key,
817 const struct sk_buff *skb,
818 enum nf_nat_manip_type maniptype)
819 {
820 if (maniptype == NF_NAT_MANIP_SRC) {
821 __be16 src;
822
823 key->ct_state |= OVS_CS_F_SRC_NAT;
824 if (key->eth.type == htons(ETH_P_IP))
825 key->ipv4.addr.src = ip_hdr(skb)->saddr;
826 else if (key->eth.type == htons(ETH_P_IPV6))
827 memcpy(&key->ipv6.addr.src, &ipv6_hdr(skb)->saddr,
828 sizeof(key->ipv6.addr.src));
829 else
830 return;
831
832 if (key->ip.proto == IPPROTO_UDP)
833 src = udp_hdr(skb)->source;
834 else if (key->ip.proto == IPPROTO_TCP)
835 src = tcp_hdr(skb)->source;
836 else if (key->ip.proto == IPPROTO_SCTP)
837 src = sctp_hdr(skb)->source;
838 else
839 return;
840
841 key->tp.src = src;
842 } else {
843 __be16 dst;
844
845 key->ct_state |= OVS_CS_F_DST_NAT;
846 if (key->eth.type == htons(ETH_P_IP))
847 key->ipv4.addr.dst = ip_hdr(skb)->daddr;
848 else if (key->eth.type == htons(ETH_P_IPV6))
849 memcpy(&key->ipv6.addr.dst, &ipv6_hdr(skb)->daddr,
850 sizeof(key->ipv6.addr.dst));
851 else
852 return;
853
854 if (key->ip.proto == IPPROTO_UDP)
855 dst = udp_hdr(skb)->dest;
856 else if (key->ip.proto == IPPROTO_TCP)
857 dst = tcp_hdr(skb)->dest;
858 else if (key->ip.proto == IPPROTO_SCTP)
859 dst = sctp_hdr(skb)->dest;
860 else
861 return;
862
863 key->tp.dst = dst;
864 }
865 }
866
867 /* Returns NF_DROP if the packet should be dropped, NF_ACCEPT otherwise. */
868 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
869 const struct ovs_conntrack_info *info,
870 struct sk_buff *skb, struct nf_conn *ct,
871 enum ip_conntrack_info ctinfo)
872 {
873 enum nf_nat_manip_type maniptype;
874 int err;
875
876 /* Add NAT extension if not confirmed yet. */
877 if (!nf_ct_is_confirmed(ct) && !nf_ct_nat_ext_add(ct))
878 return NF_ACCEPT; /* Can't NAT. */
879
880 /* Determine NAT type.
881 * Check if the NAT type can be deduced from the tracked connection.
882 * Make sure new expected connections (IP_CT_RELATED) are NATted only
883 * when committing.
884 */
885 if (info->nat & OVS_CT_NAT && ctinfo != IP_CT_NEW &&
886 ct->status & IPS_NAT_MASK &&
887 (ctinfo != IP_CT_RELATED || info->commit)) {
888 /* NAT an established or related connection like before. */
889 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY)
890 /* This is the REPLY direction for a connection
891 * for which NAT was applied in the forward
892 * direction. Do the reverse NAT.
893 */
894 maniptype = ct->status & IPS_SRC_NAT
895 ? NF_NAT_MANIP_DST : NF_NAT_MANIP_SRC;
896 else
897 maniptype = ct->status & IPS_SRC_NAT
898 ? NF_NAT_MANIP_SRC : NF_NAT_MANIP_DST;
899 } else if (info->nat & OVS_CT_SRC_NAT) {
900 maniptype = NF_NAT_MANIP_SRC;
901 } else if (info->nat & OVS_CT_DST_NAT) {
902 maniptype = NF_NAT_MANIP_DST;
903 } else {
904 return NF_ACCEPT; /* Connection is not NATed. */
905 }
906 err = ovs_ct_nat_execute(skb, ct, ctinfo, &info->range, maniptype);
907
908 if (err == NF_ACCEPT &&
909 ct->status & IPS_SRC_NAT && ct->status & IPS_DST_NAT) {
910 if (maniptype == NF_NAT_MANIP_SRC)
911 maniptype = NF_NAT_MANIP_DST;
912 else
913 maniptype = NF_NAT_MANIP_SRC;
914
915 err = ovs_ct_nat_execute(skb, ct, ctinfo, &info->range,
916 maniptype);
917 }
918
919 /* Mark NAT done if successful and update the flow key. */
920 if (err == NF_ACCEPT)
921 ovs_nat_update_key(key, skb, maniptype);
922
923 return err;
924 }
925 #else /* !CONFIG_NF_NAT */
926 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
927 const struct ovs_conntrack_info *info,
928 struct sk_buff *skb, struct nf_conn *ct,
929 enum ip_conntrack_info ctinfo)
930 {
931 return NF_ACCEPT;
932 }
933 #endif
934
935 /* Pass 'skb' through conntrack in 'net', using zone configured in 'info', if
936 * not done already. Update key with new CT state after passing the packet
937 * through conntrack.
938 * Note that if the packet is deemed invalid by conntrack, skb->_nfct will be
939 * set to NULL and 0 will be returned.
940 */
941 static int __ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
942 const struct ovs_conntrack_info *info,
943 struct sk_buff *skb)
944 {
945 /* If we are recirculating packets to match on conntrack fields and
946 * committing with a separate conntrack action, then we don't need to
947 * actually run the packet through conntrack twice unless it's for a
948 * different zone.
949 */
950 bool cached = skb_nfct_cached(net, key, info, skb);
951 enum ip_conntrack_info ctinfo;
952 struct nf_conn *ct;
953
954 if (!cached) {
955 struct nf_hook_state state = {
956 .hook = NF_INET_PRE_ROUTING,
957 .pf = info->family,
958 .net = net,
959 };
960 struct nf_conn *tmpl = info->ct;
961 int err;
962
963 /* Associate skb with specified zone. */
964 if (tmpl) {
965 if (skb_nfct(skb))
966 nf_conntrack_put(skb_nfct(skb));
967 nf_conntrack_get(&tmpl->ct_general);
968 nf_ct_set(skb, tmpl, IP_CT_NEW);
969 }
970
971 err = nf_conntrack_in(skb, &state);
972 if (err != NF_ACCEPT)
973 return -ENOENT;
974
975 /* Clear CT state NAT flags to mark that we have not yet done
976 * NAT after the nf_conntrack_in() call. We can actually clear
977 * the whole state, as it will be re-initialized below.
978 */
979 key->ct_state = 0;
980
981 /* Update the key, but keep the NAT flags. */
982 ovs_ct_update_key(skb, info, key, true, true);
983 }
984
985 ct = nf_ct_get(skb, &ctinfo);
986 if (ct) {
987 /* Packets starting a new connection must be NATted before the
988 * helper, so that the helper knows about the NAT. We enforce
989 * this by delaying both NAT and helper calls for unconfirmed
990 * connections until the committing CT action. For later
991 * packets NAT and Helper may be called in either order.
992 *
993 * NAT will be done only if the CT action has NAT, and only
994 * once per packet (per zone), as guarded by the NAT bits in
995 * the key->ct_state.
996 */
997 if (info->nat && !(key->ct_state & OVS_CS_F_NAT_MASK) &&
998 (nf_ct_is_confirmed(ct) || info->commit) &&
999 ovs_ct_nat(net, key, info, skb, ct, ctinfo) != NF_ACCEPT) {
1000 return -EINVAL;
1001 }
1002
1003 /* Userspace may decide to perform a ct lookup without a helper
1004 * specified followed by a (recirculate and) commit with one.
1005 * Therefore, for unconfirmed connections which we will commit,
1006 * we need to attach the helper here.
1007 */
1008 if (!nf_ct_is_confirmed(ct) && info->commit &&
1009 info->helper && !nfct_help(ct)) {
1010 int err = __nf_ct_try_assign_helper(ct, info->ct,
1011 GFP_ATOMIC);
1012 if (err)
1013 return err;
1014
1015 /* helper installed, add seqadj if NAT is required */
1016 if (info->nat && !nfct_seqadj(ct)) {
1017 if (!nfct_seqadj_ext_add(ct))
1018 return -EINVAL;
1019 }
1020 }
1021
1022 /* Call the helper only if:
1023 * - nf_conntrack_in() was executed above ("!cached") for a
1024 * confirmed connection, or
1025 * - When committing an unconfirmed connection.
1026 */
1027 if ((nf_ct_is_confirmed(ct) ? !cached : info->commit) &&
1028 ovs_ct_helper(skb, info->family) != NF_ACCEPT) {
1029 return -EINVAL;
1030 }
1031 }
1032
1033 return 0;
1034 }
1035
1036 /* Lookup connection and read fields into key. */
1037 static int ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
1038 const struct ovs_conntrack_info *info,
1039 struct sk_buff *skb)
1040 {
1041 struct nf_conntrack_expect *exp;
1042
1043 /* If we pass an expected packet through nf_conntrack_in() the
1044 * expectation is typically removed, but the packet could still be
1045 * lost in upcall processing. To prevent this from happening we
1046 * perform an explicit expectation lookup. Expected connections are
1047 * always new, and will be passed through conntrack only when they are
1048 * committed, as it is OK to remove the expectation at that time.
1049 */
1050 exp = ovs_ct_expect_find(net, &info->zone, info->family, skb);
1051 if (exp) {
1052 u8 state;
1053
1054 /* NOTE: New connections are NATted and Helped only when
1055 * committed, so we are not calling into NAT here.
1056 */
1057 state = OVS_CS_F_TRACKED | OVS_CS_F_NEW | OVS_CS_F_RELATED;
1058 __ovs_ct_update_key(key, state, &info->zone, exp->master);
1059 } else {
1060 struct nf_conn *ct;
1061 int err;
1062
1063 err = __ovs_ct_lookup(net, key, info, skb);
1064 if (err)
1065 return err;
1066
1067 ct = (struct nf_conn *)skb_nfct(skb);
1068 if (ct)
1069 nf_ct_deliver_cached_events(ct);
1070 }
1071
1072 return 0;
1073 }
1074
1075 static bool labels_nonzero(const struct ovs_key_ct_labels *labels)
1076 {
1077 size_t i;
1078
1079 for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
1080 if (labels->ct_labels_32[i])
1081 return true;
1082
1083 return false;
1084 }
1085
1086 #if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1087 static struct hlist_head *ct_limit_hash_bucket(
1088 const struct ovs_ct_limit_info *info, u16 zone)
1089 {
1090 return &info->limits[zone & (CT_LIMIT_HASH_BUCKETS - 1)];
1091 }
1092
1093 /* Call with ovs_mutex */
1094 static void ct_limit_set(const struct ovs_ct_limit_info *info,
1095 struct ovs_ct_limit *new_ct_limit)
1096 {
1097 struct ovs_ct_limit *ct_limit;
1098 struct hlist_head *head;
1099
1100 head = ct_limit_hash_bucket(info, new_ct_limit->zone);
1101 hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
1102 if (ct_limit->zone == new_ct_limit->zone) {
1103 hlist_replace_rcu(&ct_limit->hlist_node,
1104 &new_ct_limit->hlist_node);
1105 kfree_rcu(ct_limit, rcu);
1106 return;
1107 }
1108 }
1109
1110 hlist_add_head_rcu(&new_ct_limit->hlist_node, head);
1111 }
1112
1113 /* Call with ovs_mutex */
1114 static void ct_limit_del(const struct ovs_ct_limit_info *info, u16 zone)
1115 {
1116 struct ovs_ct_limit *ct_limit;
1117 struct hlist_head *head;
1118 struct hlist_node *n;
1119
1120 head = ct_limit_hash_bucket(info, zone);
1121 hlist_for_each_entry_safe(ct_limit, n, head, hlist_node) {
1122 if (ct_limit->zone == zone) {
1123 hlist_del_rcu(&ct_limit->hlist_node);
1124 kfree_rcu(ct_limit, rcu);
1125 return;
1126 }
1127 }
1128 }
1129
1130 /* Call with RCU read lock */
1131 static u32 ct_limit_get(const struct ovs_ct_limit_info *info, u16 zone)
1132 {
1133 struct ovs_ct_limit *ct_limit;
1134 struct hlist_head *head;
1135
1136 head = ct_limit_hash_bucket(info, zone);
1137 hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
1138 if (ct_limit->zone == zone)
1139 return ct_limit->limit;
1140 }
1141
1142 return info->default_limit;
1143 }
1144
1145 static int ovs_ct_check_limit(struct net *net,
1146 const struct ovs_conntrack_info *info,
1147 const struct nf_conntrack_tuple *tuple)
1148 {
1149 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1150 const struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
1151 u32 per_zone_limit, connections;
1152 u32 conncount_key;
1153
1154 conncount_key = info->zone.id;
1155
1156 per_zone_limit = ct_limit_get(ct_limit_info, info->zone.id);
1157 if (per_zone_limit == OVS_CT_LIMIT_UNLIMITED)
1158 return 0;
1159
1160 connections = nf_conncount_count(net, ct_limit_info->data,
1161 &conncount_key, tuple, &info->zone);
1162 if (connections > per_zone_limit)
1163 return -ENOMEM;
1164
1165 return 0;
1166 }
1167 #endif
1168
1169 /* Lookup connection and confirm if unconfirmed. */
1170 static int ovs_ct_commit(struct net *net, struct sw_flow_key *key,
1171 const struct ovs_conntrack_info *info,
1172 struct sk_buff *skb)
1173 {
1174 enum ip_conntrack_info ctinfo;
1175 struct nf_conn *ct;
1176 int err;
1177
1178 err = __ovs_ct_lookup(net, key, info, skb);
1179 if (err)
1180 return err;
1181
1182 /* The connection could be invalid, in which case this is a no-op.*/
1183 ct = nf_ct_get(skb, &ctinfo);
1184 if (!ct)
1185 return 0;
1186
1187 #if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1188 if (static_branch_unlikely(&ovs_ct_limit_enabled)) {
1189 if (!nf_ct_is_confirmed(ct)) {
1190 err = ovs_ct_check_limit(net, info,
1191 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
1192 if (err) {
1193 net_warn_ratelimited("openvswitch: zone: %u "
1194 "exceeds conntrack limit\n",
1195 info->zone.id);
1196 return err;
1197 }
1198 }
1199 }
1200 #endif
1201
1202 /* Set the conntrack event mask if given. NEW and DELETE events have
1203 * their own groups, but the NFNLGRP_CONNTRACK_UPDATE group listener
1204 * typically would receive many kinds of updates. Setting the event
1205 * mask allows those events to be filtered. The set event mask will
1206 * remain in effect for the lifetime of the connection unless changed
1207 * by a further CT action with both the commit flag and the eventmask
1208 * option. */
1209 if (info->have_eventmask) {
1210 struct nf_conntrack_ecache *cache = nf_ct_ecache_find(ct);
1211
1212 if (cache)
1213 cache->ctmask = info->eventmask;
1214 }
1215
1216 /* Apply changes before confirming the connection so that the initial
1217 * conntrack NEW netlink event carries the values given in the CT
1218 * action.
1219 */
1220 if (info->mark.mask) {
1221 err = ovs_ct_set_mark(ct, key, info->mark.value,
1222 info->mark.mask);
1223 if (err)
1224 return err;
1225 }
1226 if (!nf_ct_is_confirmed(ct)) {
1227 err = ovs_ct_init_labels(ct, key, &info->labels.value,
1228 &info->labels.mask);
1229 if (err)
1230 return err;
1231 } else if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1232 labels_nonzero(&info->labels.mask)) {
1233 err = ovs_ct_set_labels(ct, key, &info->labels.value,
1234 &info->labels.mask);
1235 if (err)
1236 return err;
1237 }
1238 /* This will take care of sending queued events even if the connection
1239 * is already confirmed.
1240 */
1241 if (nf_conntrack_confirm(skb) != NF_ACCEPT)
1242 return -EINVAL;
1243
1244 return 0;
1245 }
1246
1247 /* Trim the skb to the length specified by the IP/IPv6 header,
1248 * removing any trailing lower-layer padding. This prepares the skb
1249 * for higher-layer processing that assumes skb->len excludes padding
1250 * (such as nf_ip_checksum). The caller needs to pull the skb to the
1251 * network header, and ensure ip_hdr/ipv6_hdr points to valid data.
1252 */
1253 static int ovs_skb_network_trim(struct sk_buff *skb)
1254 {
1255 unsigned int len;
1256 int err;
1257
1258 switch (skb->protocol) {
1259 case htons(ETH_P_IP):
1260 len = ntohs(ip_hdr(skb)->tot_len);
1261 break;
1262 case htons(ETH_P_IPV6):
1263 len = sizeof(struct ipv6hdr)
1264 + ntohs(ipv6_hdr(skb)->payload_len);
1265 break;
1266 default:
1267 len = skb->len;
1268 }
1269
1270 err = pskb_trim_rcsum(skb, len);
1271 if (err)
1272 kfree_skb(skb);
1273
1274 return err;
1275 }
1276
1277 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
1278 * value if 'skb' is freed.
1279 */
1280 int ovs_ct_execute(struct net *net, struct sk_buff *skb,
1281 struct sw_flow_key *key,
1282 const struct ovs_conntrack_info *info)
1283 {
1284 int nh_ofs;
1285 int err;
1286
1287 /* The conntrack module expects to be working at L3. */
1288 nh_ofs = skb_network_offset(skb);
1289 skb_pull_rcsum(skb, nh_ofs);
1290
1291 err = ovs_skb_network_trim(skb);
1292 if (err)
1293 return err;
1294
1295 if (key->ip.frag != OVS_FRAG_TYPE_NONE) {
1296 err = handle_fragments(net, key, info->zone.id, skb);
1297 if (err)
1298 return err;
1299 }
1300
1301 if (info->commit)
1302 err = ovs_ct_commit(net, key, info, skb);
1303 else
1304 err = ovs_ct_lookup(net, key, info, skb);
1305
1306 skb_push(skb, nh_ofs);
1307 skb_postpush_rcsum(skb, skb->data, nh_ofs);
1308 if (err)
1309 kfree_skb(skb);
1310 return err;
1311 }
1312
1313 int ovs_ct_clear(struct sk_buff *skb, struct sw_flow_key *key)
1314 {
1315 if (skb_nfct(skb)) {
1316 nf_conntrack_put(skb_nfct(skb));
1317 nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
1318 ovs_ct_fill_key(skb, key);
1319 }
1320
1321 return 0;
1322 }
1323
1324 static int ovs_ct_add_helper(struct ovs_conntrack_info *info, const char *name,
1325 const struct sw_flow_key *key, bool log)
1326 {
1327 struct nf_conntrack_helper *helper;
1328 struct nf_conn_help *help;
1329 int ret = 0;
1330
1331 helper = nf_conntrack_helper_try_module_get(name, info->family,
1332 key->ip.proto);
1333 if (!helper) {
1334 OVS_NLERR(log, "Unknown helper \"%s\"", name);
1335 return -EINVAL;
1336 }
1337
1338 help = nf_ct_helper_ext_add(info->ct, GFP_KERNEL);
1339 if (!help) {
1340 nf_conntrack_helper_put(helper);
1341 return -ENOMEM;
1342 }
1343
1344 #if IS_ENABLED(CONFIG_NF_NAT)
1345 if (info->nat) {
1346 ret = nf_nat_helper_try_module_get(name, info->family,
1347 key->ip.proto);
1348 if (ret) {
1349 nf_conntrack_helper_put(helper);
1350 OVS_NLERR(log, "Failed to load \"%s\" NAT helper, error: %d",
1351 name, ret);
1352 return ret;
1353 }
1354 }
1355 #endif
1356 rcu_assign_pointer(help->helper, helper);
1357 info->helper = helper;
1358 return ret;
1359 }
1360
1361 #if IS_ENABLED(CONFIG_NF_NAT)
1362 static int parse_nat(const struct nlattr *attr,
1363 struct ovs_conntrack_info *info, bool log)
1364 {
1365 struct nlattr *a;
1366 int rem;
1367 bool have_ip_max = false;
1368 bool have_proto_max = false;
1369 bool ip_vers = (info->family == NFPROTO_IPV6);
1370
1371 nla_for_each_nested(a, attr, rem) {
1372 static const int ovs_nat_attr_lens[OVS_NAT_ATTR_MAX + 1][2] = {
1373 [OVS_NAT_ATTR_SRC] = {0, 0},
1374 [OVS_NAT_ATTR_DST] = {0, 0},
1375 [OVS_NAT_ATTR_IP_MIN] = {sizeof(struct in_addr),
1376 sizeof(struct in6_addr)},
1377 [OVS_NAT_ATTR_IP_MAX] = {sizeof(struct in_addr),
1378 sizeof(struct in6_addr)},
1379 [OVS_NAT_ATTR_PROTO_MIN] = {sizeof(u16), sizeof(u16)},
1380 [OVS_NAT_ATTR_PROTO_MAX] = {sizeof(u16), sizeof(u16)},
1381 [OVS_NAT_ATTR_PERSISTENT] = {0, 0},
1382 [OVS_NAT_ATTR_PROTO_HASH] = {0, 0},
1383 [OVS_NAT_ATTR_PROTO_RANDOM] = {0, 0},
1384 };
1385 int type = nla_type(a);
1386
1387 if (type > OVS_NAT_ATTR_MAX) {
1388 OVS_NLERR(log, "Unknown NAT attribute (type=%d, max=%d)",
1389 type, OVS_NAT_ATTR_MAX);
1390 return -EINVAL;
1391 }
1392
1393 if (nla_len(a) != ovs_nat_attr_lens[type][ip_vers]) {
1394 OVS_NLERR(log, "NAT attribute type %d has unexpected length (%d != %d)",
1395 type, nla_len(a),
1396 ovs_nat_attr_lens[type][ip_vers]);
1397 return -EINVAL;
1398 }
1399
1400 switch (type) {
1401 case OVS_NAT_ATTR_SRC:
1402 case OVS_NAT_ATTR_DST:
1403 if (info->nat) {
1404 OVS_NLERR(log, "Only one type of NAT may be specified");
1405 return -ERANGE;
1406 }
1407 info->nat |= OVS_CT_NAT;
1408 info->nat |= ((type == OVS_NAT_ATTR_SRC)
1409 ? OVS_CT_SRC_NAT : OVS_CT_DST_NAT);
1410 break;
1411
1412 case OVS_NAT_ATTR_IP_MIN:
1413 nla_memcpy(&info->range.min_addr, a,
1414 sizeof(info->range.min_addr));
1415 info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1416 break;
1417
1418 case OVS_NAT_ATTR_IP_MAX:
1419 have_ip_max = true;
1420 nla_memcpy(&info->range.max_addr, a,
1421 sizeof(info->range.max_addr));
1422 info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1423 break;
1424
1425 case OVS_NAT_ATTR_PROTO_MIN:
1426 info->range.min_proto.all = htons(nla_get_u16(a));
1427 info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1428 break;
1429
1430 case OVS_NAT_ATTR_PROTO_MAX:
1431 have_proto_max = true;
1432 info->range.max_proto.all = htons(nla_get_u16(a));
1433 info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1434 break;
1435
1436 case OVS_NAT_ATTR_PERSISTENT:
1437 info->range.flags |= NF_NAT_RANGE_PERSISTENT;
1438 break;
1439
1440 case OVS_NAT_ATTR_PROTO_HASH:
1441 info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM;
1442 break;
1443
1444 case OVS_NAT_ATTR_PROTO_RANDOM:
1445 info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM_FULLY;
1446 break;
1447
1448 default:
1449 OVS_NLERR(log, "Unknown nat attribute (%d)", type);
1450 return -EINVAL;
1451 }
1452 }
1453
1454 if (rem > 0) {
1455 OVS_NLERR(log, "NAT attribute has %d unknown bytes", rem);
1456 return -EINVAL;
1457 }
1458 if (!info->nat) {
1459 /* Do not allow flags if no type is given. */
1460 if (info->range.flags) {
1461 OVS_NLERR(log,
1462 "NAT flags may be given only when NAT range (SRC or DST) is also specified."
1463 );
1464 return -EINVAL;
1465 }
1466 info->nat = OVS_CT_NAT; /* NAT existing connections. */
1467 } else if (!info->commit) {
1468 OVS_NLERR(log,
1469 "NAT attributes may be specified only when CT COMMIT flag is also specified."
1470 );
1471 return -EINVAL;
1472 }
1473 /* Allow missing IP_MAX. */
1474 if (info->range.flags & NF_NAT_RANGE_MAP_IPS && !have_ip_max) {
1475 memcpy(&info->range.max_addr, &info->range.min_addr,
1476 sizeof(info->range.max_addr));
1477 }
1478 /* Allow missing PROTO_MAX. */
1479 if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1480 !have_proto_max) {
1481 info->range.max_proto.all = info->range.min_proto.all;
1482 }
1483 return 0;
1484 }
1485 #endif
1486
1487 static const struct ovs_ct_len_tbl ovs_ct_attr_lens[OVS_CT_ATTR_MAX + 1] = {
1488 [OVS_CT_ATTR_COMMIT] = { .minlen = 0, .maxlen = 0 },
1489 [OVS_CT_ATTR_FORCE_COMMIT] = { .minlen = 0, .maxlen = 0 },
1490 [OVS_CT_ATTR_ZONE] = { .minlen = sizeof(u16),
1491 .maxlen = sizeof(u16) },
1492 [OVS_CT_ATTR_MARK] = { .minlen = sizeof(struct md_mark),
1493 .maxlen = sizeof(struct md_mark) },
1494 [OVS_CT_ATTR_LABELS] = { .minlen = sizeof(struct md_labels),
1495 .maxlen = sizeof(struct md_labels) },
1496 [OVS_CT_ATTR_HELPER] = { .minlen = 1,
1497 .maxlen = NF_CT_HELPER_NAME_LEN },
1498 #if IS_ENABLED(CONFIG_NF_NAT)
1499 /* NAT length is checked when parsing the nested attributes. */
1500 [OVS_CT_ATTR_NAT] = { .minlen = 0, .maxlen = INT_MAX },
1501 #endif
1502 [OVS_CT_ATTR_EVENTMASK] = { .minlen = sizeof(u32),
1503 .maxlen = sizeof(u32) },
1504 [OVS_CT_ATTR_TIMEOUT] = { .minlen = 1,
1505 .maxlen = CTNL_TIMEOUT_NAME_MAX },
1506 };
1507
1508 static int parse_ct(const struct nlattr *attr, struct ovs_conntrack_info *info,
1509 const char **helper, bool log)
1510 {
1511 struct nlattr *a;
1512 int rem;
1513
1514 nla_for_each_nested(a, attr, rem) {
1515 int type = nla_type(a);
1516 int maxlen;
1517 int minlen;
1518
1519 if (type > OVS_CT_ATTR_MAX) {
1520 OVS_NLERR(log,
1521 "Unknown conntrack attr (type=%d, max=%d)",
1522 type, OVS_CT_ATTR_MAX);
1523 return -EINVAL;
1524 }
1525
1526 maxlen = ovs_ct_attr_lens[type].maxlen;
1527 minlen = ovs_ct_attr_lens[type].minlen;
1528 if (nla_len(a) < minlen || nla_len(a) > maxlen) {
1529 OVS_NLERR(log,
1530 "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
1531 type, nla_len(a), maxlen);
1532 return -EINVAL;
1533 }
1534
1535 switch (type) {
1536 case OVS_CT_ATTR_FORCE_COMMIT:
1537 info->force = true;
1538 /* fall through. */
1539 case OVS_CT_ATTR_COMMIT:
1540 info->commit = true;
1541 break;
1542 #ifdef CONFIG_NF_CONNTRACK_ZONES
1543 case OVS_CT_ATTR_ZONE:
1544 info->zone.id = nla_get_u16(a);
1545 break;
1546 #endif
1547 #ifdef CONFIG_NF_CONNTRACK_MARK
1548 case OVS_CT_ATTR_MARK: {
1549 struct md_mark *mark = nla_data(a);
1550
1551 if (!mark->mask) {
1552 OVS_NLERR(log, "ct_mark mask cannot be 0");
1553 return -EINVAL;
1554 }
1555 info->mark = *mark;
1556 break;
1557 }
1558 #endif
1559 #ifdef CONFIG_NF_CONNTRACK_LABELS
1560 case OVS_CT_ATTR_LABELS: {
1561 struct md_labels *labels = nla_data(a);
1562
1563 if (!labels_nonzero(&labels->mask)) {
1564 OVS_NLERR(log, "ct_labels mask cannot be 0");
1565 return -EINVAL;
1566 }
1567 info->labels = *labels;
1568 break;
1569 }
1570 #endif
1571 case OVS_CT_ATTR_HELPER:
1572 *helper = nla_data(a);
1573 if (!memchr(*helper, '\0', nla_len(a))) {
1574 OVS_NLERR(log, "Invalid conntrack helper");
1575 return -EINVAL;
1576 }
1577 break;
1578 #if IS_ENABLED(CONFIG_NF_NAT)
1579 case OVS_CT_ATTR_NAT: {
1580 int err = parse_nat(a, info, log);
1581
1582 if (err)
1583 return err;
1584 break;
1585 }
1586 #endif
1587 case OVS_CT_ATTR_EVENTMASK:
1588 info->have_eventmask = true;
1589 info->eventmask = nla_get_u32(a);
1590 break;
1591 #ifdef CONFIG_NF_CONNTRACK_TIMEOUT
1592 case OVS_CT_ATTR_TIMEOUT:
1593 memcpy(info->timeout, nla_data(a), nla_len(a));
1594 if (!memchr(info->timeout, '\0', nla_len(a))) {
1595 OVS_NLERR(log, "Invalid conntrack timeout");
1596 return -EINVAL;
1597 }
1598 break;
1599 #endif
1600
1601 default:
1602 OVS_NLERR(log, "Unknown conntrack attr (%d)",
1603 type);
1604 return -EINVAL;
1605 }
1606 }
1607
1608 #ifdef CONFIG_NF_CONNTRACK_MARK
1609 if (!info->commit && info->mark.mask) {
1610 OVS_NLERR(log,
1611 "Setting conntrack mark requires 'commit' flag.");
1612 return -EINVAL;
1613 }
1614 #endif
1615 #ifdef CONFIG_NF_CONNTRACK_LABELS
1616 if (!info->commit && labels_nonzero(&info->labels.mask)) {
1617 OVS_NLERR(log,
1618 "Setting conntrack labels requires 'commit' flag.");
1619 return -EINVAL;
1620 }
1621 #endif
1622 if (rem > 0) {
1623 OVS_NLERR(log, "Conntrack attr has %d unknown bytes", rem);
1624 return -EINVAL;
1625 }
1626
1627 return 0;
1628 }
1629
1630 bool ovs_ct_verify(struct net *net, enum ovs_key_attr attr)
1631 {
1632 if (attr == OVS_KEY_ATTR_CT_STATE)
1633 return true;
1634 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1635 attr == OVS_KEY_ATTR_CT_ZONE)
1636 return true;
1637 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1638 attr == OVS_KEY_ATTR_CT_MARK)
1639 return true;
1640 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1641 attr == OVS_KEY_ATTR_CT_LABELS) {
1642 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1643
1644 return ovs_net->xt_label;
1645 }
1646
1647 return false;
1648 }
1649
1650 int ovs_ct_copy_action(struct net *net, const struct nlattr *attr,
1651 const struct sw_flow_key *key,
1652 struct sw_flow_actions **sfa, bool log)
1653 {
1654 struct ovs_conntrack_info ct_info;
1655 const char *helper = NULL;
1656 u16 family;
1657 int err;
1658
1659 family = key_to_nfproto(key);
1660 if (family == NFPROTO_UNSPEC) {
1661 OVS_NLERR(log, "ct family unspecified");
1662 return -EINVAL;
1663 }
1664
1665 memset(&ct_info, 0, sizeof(ct_info));
1666 ct_info.family = family;
1667
1668 nf_ct_zone_init(&ct_info.zone, NF_CT_DEFAULT_ZONE_ID,
1669 NF_CT_DEFAULT_ZONE_DIR, 0);
1670
1671 err = parse_ct(attr, &ct_info, &helper, log);
1672 if (err)
1673 return err;
1674
1675 /* Set up template for tracking connections in specific zones. */
1676 ct_info.ct = nf_ct_tmpl_alloc(net, &ct_info.zone, GFP_KERNEL);
1677 if (!ct_info.ct) {
1678 OVS_NLERR(log, "Failed to allocate conntrack template");
1679 return -ENOMEM;
1680 }
1681
1682 if (ct_info.timeout[0]) {
1683 if (nf_ct_set_timeout(net, ct_info.ct, family, key->ip.proto,
1684 ct_info.timeout))
1685 pr_info_ratelimited("Failed to associated timeout "
1686 "policy `%s'\n", ct_info.timeout);
1687 else
1688 ct_info.nf_ct_timeout = rcu_dereference(
1689 nf_ct_timeout_find(ct_info.ct)->timeout);
1690
1691 }
1692
1693 if (helper) {
1694 err = ovs_ct_add_helper(&ct_info, helper, key, log);
1695 if (err)
1696 goto err_free_ct;
1697 }
1698
1699 err = ovs_nla_add_action(sfa, OVS_ACTION_ATTR_CT, &ct_info,
1700 sizeof(ct_info), log);
1701 if (err)
1702 goto err_free_ct;
1703
1704 __set_bit(IPS_CONFIRMED_BIT, &ct_info.ct->status);
1705 nf_conntrack_get(&ct_info.ct->ct_general);
1706 return 0;
1707 err_free_ct:
1708 __ovs_ct_free_action(&ct_info);
1709 return err;
1710 }
1711
1712 #if IS_ENABLED(CONFIG_NF_NAT)
1713 static bool ovs_ct_nat_to_attr(const struct ovs_conntrack_info *info,
1714 struct sk_buff *skb)
1715 {
1716 struct nlattr *start;
1717
1718 start = nla_nest_start_noflag(skb, OVS_CT_ATTR_NAT);
1719 if (!start)
1720 return false;
1721
1722 if (info->nat & OVS_CT_SRC_NAT) {
1723 if (nla_put_flag(skb, OVS_NAT_ATTR_SRC))
1724 return false;
1725 } else if (info->nat & OVS_CT_DST_NAT) {
1726 if (nla_put_flag(skb, OVS_NAT_ATTR_DST))
1727 return false;
1728 } else {
1729 goto out;
1730 }
1731
1732 if (info->range.flags & NF_NAT_RANGE_MAP_IPS) {
1733 if (IS_ENABLED(CONFIG_NF_NAT) &&
1734 info->family == NFPROTO_IPV4) {
1735 if (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MIN,
1736 info->range.min_addr.ip) ||
1737 (info->range.max_addr.ip
1738 != info->range.min_addr.ip &&
1739 (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MAX,
1740 info->range.max_addr.ip))))
1741 return false;
1742 } else if (IS_ENABLED(CONFIG_IPV6) &&
1743 info->family == NFPROTO_IPV6) {
1744 if (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MIN,
1745 &info->range.min_addr.in6) ||
1746 (memcmp(&info->range.max_addr.in6,
1747 &info->range.min_addr.in6,
1748 sizeof(info->range.max_addr.in6)) &&
1749 (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MAX,
1750 &info->range.max_addr.in6))))
1751 return false;
1752 } else {
1753 return false;
1754 }
1755 }
1756 if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1757 (nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MIN,
1758 ntohs(info->range.min_proto.all)) ||
1759 (info->range.max_proto.all != info->range.min_proto.all &&
1760 nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MAX,
1761 ntohs(info->range.max_proto.all)))))
1762 return false;
1763
1764 if (info->range.flags & NF_NAT_RANGE_PERSISTENT &&
1765 nla_put_flag(skb, OVS_NAT_ATTR_PERSISTENT))
1766 return false;
1767 if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM &&
1768 nla_put_flag(skb, OVS_NAT_ATTR_PROTO_HASH))
1769 return false;
1770 if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM_FULLY &&
1771 nla_put_flag(skb, OVS_NAT_ATTR_PROTO_RANDOM))
1772 return false;
1773 out:
1774 nla_nest_end(skb, start);
1775
1776 return true;
1777 }
1778 #endif
1779
1780 int ovs_ct_action_to_attr(const struct ovs_conntrack_info *ct_info,
1781 struct sk_buff *skb)
1782 {
1783 struct nlattr *start;
1784
1785 start = nla_nest_start_noflag(skb, OVS_ACTION_ATTR_CT);
1786 if (!start)
1787 return -EMSGSIZE;
1788
1789 if (ct_info->commit && nla_put_flag(skb, ct_info->force
1790 ? OVS_CT_ATTR_FORCE_COMMIT
1791 : OVS_CT_ATTR_COMMIT))
1792 return -EMSGSIZE;
1793 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1794 nla_put_u16(skb, OVS_CT_ATTR_ZONE, ct_info->zone.id))
1795 return -EMSGSIZE;
1796 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && ct_info->mark.mask &&
1797 nla_put(skb, OVS_CT_ATTR_MARK, sizeof(ct_info->mark),
1798 &ct_info->mark))
1799 return -EMSGSIZE;
1800 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1801 labels_nonzero(&ct_info->labels.mask) &&
1802 nla_put(skb, OVS_CT_ATTR_LABELS, sizeof(ct_info->labels),
1803 &ct_info->labels))
1804 return -EMSGSIZE;
1805 if (ct_info->helper) {
1806 if (nla_put_string(skb, OVS_CT_ATTR_HELPER,
1807 ct_info->helper->name))
1808 return -EMSGSIZE;
1809 }
1810 if (ct_info->have_eventmask &&
1811 nla_put_u32(skb, OVS_CT_ATTR_EVENTMASK, ct_info->eventmask))
1812 return -EMSGSIZE;
1813 if (ct_info->timeout[0]) {
1814 if (nla_put_string(skb, OVS_CT_ATTR_TIMEOUT, ct_info->timeout))
1815 return -EMSGSIZE;
1816 }
1817
1818 #if IS_ENABLED(CONFIG_NF_NAT)
1819 if (ct_info->nat && !ovs_ct_nat_to_attr(ct_info, skb))
1820 return -EMSGSIZE;
1821 #endif
1822 nla_nest_end(skb, start);
1823
1824 return 0;
1825 }
1826
1827 void ovs_ct_free_action(const struct nlattr *a)
1828 {
1829 struct ovs_conntrack_info *ct_info = nla_data(a);
1830
1831 __ovs_ct_free_action(ct_info);
1832 }
1833
1834 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info)
1835 {
1836 if (ct_info->helper) {
1837 #if IS_ENABLED(CONFIG_NF_NAT)
1838 if (ct_info->nat)
1839 nf_nat_helper_put(ct_info->helper);
1840 #endif
1841 nf_conntrack_helper_put(ct_info->helper);
1842 }
1843 if (ct_info->ct) {
1844 if (ct_info->timeout[0])
1845 nf_ct_destroy_timeout(ct_info->ct);
1846 nf_ct_tmpl_free(ct_info->ct);
1847 }
1848 }
1849
1850 #if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1851 static int ovs_ct_limit_init(struct net *net, struct ovs_net *ovs_net)
1852 {
1853 int i, err;
1854
1855 ovs_net->ct_limit_info = kmalloc(sizeof(*ovs_net->ct_limit_info),
1856 GFP_KERNEL);
1857 if (!ovs_net->ct_limit_info)
1858 return -ENOMEM;
1859
1860 ovs_net->ct_limit_info->default_limit = OVS_CT_LIMIT_DEFAULT;
1861 ovs_net->ct_limit_info->limits =
1862 kmalloc_array(CT_LIMIT_HASH_BUCKETS, sizeof(struct hlist_head),
1863 GFP_KERNEL);
1864 if (!ovs_net->ct_limit_info->limits) {
1865 kfree(ovs_net->ct_limit_info);
1866 return -ENOMEM;
1867 }
1868
1869 for (i = 0; i < CT_LIMIT_HASH_BUCKETS; i++)
1870 INIT_HLIST_HEAD(&ovs_net->ct_limit_info->limits[i]);
1871
1872 ovs_net->ct_limit_info->data =
1873 nf_conncount_init(net, NFPROTO_INET, sizeof(u32));
1874
1875 if (IS_ERR(ovs_net->ct_limit_info->data)) {
1876 err = PTR_ERR(ovs_net->ct_limit_info->data);
1877 kfree(ovs_net->ct_limit_info->limits);
1878 kfree(ovs_net->ct_limit_info);
1879 pr_err("openvswitch: failed to init nf_conncount %d\n", err);
1880 return err;
1881 }
1882 return 0;
1883 }
1884
1885 static void ovs_ct_limit_exit(struct net *net, struct ovs_net *ovs_net)
1886 {
1887 const struct ovs_ct_limit_info *info = ovs_net->ct_limit_info;
1888 int i;
1889
1890 nf_conncount_destroy(net, NFPROTO_INET, info->data);
1891 for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) {
1892 struct hlist_head *head = &info->limits[i];
1893 struct ovs_ct_limit *ct_limit;
1894
1895 hlist_for_each_entry_rcu(ct_limit, head, hlist_node,
1896 lockdep_ovsl_is_held())
1897 kfree_rcu(ct_limit, rcu);
1898 }
1899 kfree(ovs_net->ct_limit_info->limits);
1900 kfree(ovs_net->ct_limit_info);
1901 }
1902
1903 static struct sk_buff *
1904 ovs_ct_limit_cmd_reply_start(struct genl_info *info, u8 cmd,
1905 struct ovs_header **ovs_reply_header)
1906 {
1907 struct ovs_header *ovs_header = info->userhdr;
1908 struct sk_buff *skb;
1909
1910 skb = genlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1911 if (!skb)
1912 return ERR_PTR(-ENOMEM);
1913
1914 *ovs_reply_header = genlmsg_put(skb, info->snd_portid,
1915 info->snd_seq,
1916 &dp_ct_limit_genl_family, 0, cmd);
1917
1918 if (!*ovs_reply_header) {
1919 nlmsg_free(skb);
1920 return ERR_PTR(-EMSGSIZE);
1921 }
1922 (*ovs_reply_header)->dp_ifindex = ovs_header->dp_ifindex;
1923
1924 return skb;
1925 }
1926
1927 static bool check_zone_id(int zone_id, u16 *pzone)
1928 {
1929 if (zone_id >= 0 && zone_id <= 65535) {
1930 *pzone = (u16)zone_id;
1931 return true;
1932 }
1933 return false;
1934 }
1935
1936 static int ovs_ct_limit_set_zone_limit(struct nlattr *nla_zone_limit,
1937 struct ovs_ct_limit_info *info)
1938 {
1939 struct ovs_zone_limit *zone_limit;
1940 int rem;
1941 u16 zone;
1942
1943 rem = NLA_ALIGN(nla_len(nla_zone_limit));
1944 zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1945
1946 while (rem >= sizeof(*zone_limit)) {
1947 if (unlikely(zone_limit->zone_id ==
1948 OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1949 ovs_lock();
1950 info->default_limit = zone_limit->limit;
1951 ovs_unlock();
1952 } else if (unlikely(!check_zone_id(
1953 zone_limit->zone_id, &zone))) {
1954 OVS_NLERR(true, "zone id is out of range");
1955 } else {
1956 struct ovs_ct_limit *ct_limit;
1957
1958 ct_limit = kmalloc(sizeof(*ct_limit), GFP_KERNEL);
1959 if (!ct_limit)
1960 return -ENOMEM;
1961
1962 ct_limit->zone = zone;
1963 ct_limit->limit = zone_limit->limit;
1964
1965 ovs_lock();
1966 ct_limit_set(info, ct_limit);
1967 ovs_unlock();
1968 }
1969 rem -= NLA_ALIGN(sizeof(*zone_limit));
1970 zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
1971 NLA_ALIGN(sizeof(*zone_limit)));
1972 }
1973
1974 if (rem)
1975 OVS_NLERR(true, "set zone limit has %d unknown bytes", rem);
1976
1977 return 0;
1978 }
1979
1980 static int ovs_ct_limit_del_zone_limit(struct nlattr *nla_zone_limit,
1981 struct ovs_ct_limit_info *info)
1982 {
1983 struct ovs_zone_limit *zone_limit;
1984 int rem;
1985 u16 zone;
1986
1987 rem = NLA_ALIGN(nla_len(nla_zone_limit));
1988 zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1989
1990 while (rem >= sizeof(*zone_limit)) {
1991 if (unlikely(zone_limit->zone_id ==
1992 OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1993 ovs_lock();
1994 info->default_limit = OVS_CT_LIMIT_DEFAULT;
1995 ovs_unlock();
1996 } else if (unlikely(!check_zone_id(
1997 zone_limit->zone_id, &zone))) {
1998 OVS_NLERR(true, "zone id is out of range");
1999 } else {
2000 ovs_lock();
2001 ct_limit_del(info, zone);
2002 ovs_unlock();
2003 }
2004 rem -= NLA_ALIGN(sizeof(*zone_limit));
2005 zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
2006 NLA_ALIGN(sizeof(*zone_limit)));
2007 }
2008
2009 if (rem)
2010 OVS_NLERR(true, "del zone limit has %d unknown bytes", rem);
2011
2012 return 0;
2013 }
2014
2015 static int ovs_ct_limit_get_default_limit(struct ovs_ct_limit_info *info,
2016 struct sk_buff *reply)
2017 {
2018 struct ovs_zone_limit zone_limit;
2019 int err;
2020
2021 zone_limit.zone_id = OVS_ZONE_LIMIT_DEFAULT_ZONE;
2022 zone_limit.limit = info->default_limit;
2023 err = nla_put_nohdr(reply, sizeof(zone_limit), &zone_limit);
2024 if (err)
2025 return err;
2026
2027 return 0;
2028 }
2029
2030 static int __ovs_ct_limit_get_zone_limit(struct net *net,
2031 struct nf_conncount_data *data,
2032 u16 zone_id, u32 limit,
2033 struct sk_buff *reply)
2034 {
2035 struct nf_conntrack_zone ct_zone;
2036 struct ovs_zone_limit zone_limit;
2037 u32 conncount_key = zone_id;
2038
2039 zone_limit.zone_id = zone_id;
2040 zone_limit.limit = limit;
2041 nf_ct_zone_init(&ct_zone, zone_id, NF_CT_DEFAULT_ZONE_DIR, 0);
2042
2043 zone_limit.count = nf_conncount_count(net, data, &conncount_key, NULL,
2044 &ct_zone);
2045 return nla_put_nohdr(reply, sizeof(zone_limit), &zone_limit);
2046 }
2047
2048 static int ovs_ct_limit_get_zone_limit(struct net *net,
2049 struct nlattr *nla_zone_limit,
2050 struct ovs_ct_limit_info *info,
2051 struct sk_buff *reply)
2052 {
2053 struct ovs_zone_limit *zone_limit;
2054 int rem, err;
2055 u32 limit;
2056 u16 zone;
2057
2058 rem = NLA_ALIGN(nla_len(nla_zone_limit));
2059 zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
2060
2061 while (rem >= sizeof(*zone_limit)) {
2062 if (unlikely(zone_limit->zone_id ==
2063 OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
2064 err = ovs_ct_limit_get_default_limit(info, reply);
2065 if (err)
2066 return err;
2067 } else if (unlikely(!check_zone_id(zone_limit->zone_id,
2068 &zone))) {
2069 OVS_NLERR(true, "zone id is out of range");
2070 } else {
2071 rcu_read_lock();
2072 limit = ct_limit_get(info, zone);
2073 rcu_read_unlock();
2074
2075 err = __ovs_ct_limit_get_zone_limit(
2076 net, info->data, zone, limit, reply);
2077 if (err)
2078 return err;
2079 }
2080 rem -= NLA_ALIGN(sizeof(*zone_limit));
2081 zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
2082 NLA_ALIGN(sizeof(*zone_limit)));
2083 }
2084
2085 if (rem)
2086 OVS_NLERR(true, "get zone limit has %d unknown bytes", rem);
2087
2088 return 0;
2089 }
2090
2091 static int ovs_ct_limit_get_all_zone_limit(struct net *net,
2092 struct ovs_ct_limit_info *info,
2093 struct sk_buff *reply)
2094 {
2095 struct ovs_ct_limit *ct_limit;
2096 struct hlist_head *head;
2097 int i, err = 0;
2098
2099 err = ovs_ct_limit_get_default_limit(info, reply);
2100 if (err)
2101 return err;
2102
2103 rcu_read_lock();
2104 for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) {
2105 head = &info->limits[i];
2106 hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
2107 err = __ovs_ct_limit_get_zone_limit(net, info->data,
2108 ct_limit->zone, ct_limit->limit, reply);
2109 if (err)
2110 goto exit_err;
2111 }
2112 }
2113
2114 exit_err:
2115 rcu_read_unlock();
2116 return err;
2117 }
2118
2119 static int ovs_ct_limit_cmd_set(struct sk_buff *skb, struct genl_info *info)
2120 {
2121 struct nlattr **a = info->attrs;
2122 struct sk_buff *reply;
2123 struct ovs_header *ovs_reply_header;
2124 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
2125 struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2126 int err;
2127
2128 reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_SET,
2129 &ovs_reply_header);
2130 if (IS_ERR(reply))
2131 return PTR_ERR(reply);
2132
2133 if (!a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2134 err = -EINVAL;
2135 goto exit_err;
2136 }
2137
2138 err = ovs_ct_limit_set_zone_limit(a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT],
2139 ct_limit_info);
2140 if (err)
2141 goto exit_err;
2142
2143 static_branch_enable(&ovs_ct_limit_enabled);
2144
2145 genlmsg_end(reply, ovs_reply_header);
2146 return genlmsg_reply(reply, info);
2147
2148 exit_err:
2149 nlmsg_free(reply);
2150 return err;
2151 }
2152
2153 static int ovs_ct_limit_cmd_del(struct sk_buff *skb, struct genl_info *info)
2154 {
2155 struct nlattr **a = info->attrs;
2156 struct sk_buff *reply;
2157 struct ovs_header *ovs_reply_header;
2158 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
2159 struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2160 int err;
2161
2162 reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_DEL,
2163 &ovs_reply_header);
2164 if (IS_ERR(reply))
2165 return PTR_ERR(reply);
2166
2167 if (!a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2168 err = -EINVAL;
2169 goto exit_err;
2170 }
2171
2172 err = ovs_ct_limit_del_zone_limit(a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT],
2173 ct_limit_info);
2174 if (err)
2175 goto exit_err;
2176
2177 genlmsg_end(reply, ovs_reply_header);
2178 return genlmsg_reply(reply, info);
2179
2180 exit_err:
2181 nlmsg_free(reply);
2182 return err;
2183 }
2184
2185 static int ovs_ct_limit_cmd_get(struct sk_buff *skb, struct genl_info *info)
2186 {
2187 struct nlattr **a = info->attrs;
2188 struct nlattr *nla_reply;
2189 struct sk_buff *reply;
2190 struct ovs_header *ovs_reply_header;
2191 struct net *net = sock_net(skb->sk);
2192 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2193 struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2194 int err;
2195
2196 reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_GET,
2197 &ovs_reply_header);
2198 if (IS_ERR(reply))
2199 return PTR_ERR(reply);
2200
2201 nla_reply = nla_nest_start_noflag(reply, OVS_CT_LIMIT_ATTR_ZONE_LIMIT);
2202 if (!nla_reply) {
2203 err = -EMSGSIZE;
2204 goto exit_err;
2205 }
2206
2207 if (a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2208 err = ovs_ct_limit_get_zone_limit(
2209 net, a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT], ct_limit_info,
2210 reply);
2211 if (err)
2212 goto exit_err;
2213 } else {
2214 err = ovs_ct_limit_get_all_zone_limit(net, ct_limit_info,
2215 reply);
2216 if (err)
2217 goto exit_err;
2218 }
2219
2220 nla_nest_end(reply, nla_reply);
2221 genlmsg_end(reply, ovs_reply_header);
2222 return genlmsg_reply(reply, info);
2223
2224 exit_err:
2225 nlmsg_free(reply);
2226 return err;
2227 }
2228
2229 static struct genl_ops ct_limit_genl_ops[] = {
2230 { .cmd = OVS_CT_LIMIT_CMD_SET,
2231 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2232 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN
2233 * privilege. */
2234 .doit = ovs_ct_limit_cmd_set,
2235 },
2236 { .cmd = OVS_CT_LIMIT_CMD_DEL,
2237 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2238 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN
2239 * privilege. */
2240 .doit = ovs_ct_limit_cmd_del,
2241 },
2242 { .cmd = OVS_CT_LIMIT_CMD_GET,
2243 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
2244 .flags = 0, /* OK for unprivileged users. */
2245 .doit = ovs_ct_limit_cmd_get,
2246 },
2247 };
2248
2249 static const struct genl_multicast_group ovs_ct_limit_multicast_group = {
2250 .name = OVS_CT_LIMIT_MCGROUP,
2251 };
2252
2253 struct genl_family dp_ct_limit_genl_family __ro_after_init = {
2254 .hdrsize = sizeof(struct ovs_header),
2255 .name = OVS_CT_LIMIT_FAMILY,
2256 .version = OVS_CT_LIMIT_VERSION,
2257 .maxattr = OVS_CT_LIMIT_ATTR_MAX,
2258 .policy = ct_limit_policy,
2259 .netnsok = true,
2260 .parallel_ops = true,
2261 .ops = ct_limit_genl_ops,
2262 .n_ops = ARRAY_SIZE(ct_limit_genl_ops),
2263 .mcgrps = &ovs_ct_limit_multicast_group,
2264 .n_mcgrps = 1,
2265 .module = THIS_MODULE,
2266 };
2267 #endif
2268
2269 int ovs_ct_init(struct net *net)
2270 {
2271 unsigned int n_bits = sizeof(struct ovs_key_ct_labels) * BITS_PER_BYTE;
2272 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2273
2274 if (nf_connlabels_get(net, n_bits - 1)) {
2275 ovs_net->xt_label = false;
2276 OVS_NLERR(true, "Failed to set connlabel length");
2277 } else {
2278 ovs_net->xt_label = true;
2279 }
2280
2281 #if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2282 return ovs_ct_limit_init(net, ovs_net);
2283 #else
2284 return 0;
2285 #endif
2286 }
2287
2288 void ovs_ct_exit(struct net *net)
2289 {
2290 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2291
2292 #if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2293 ovs_ct_limit_exit(net, ovs_net);
2294 #endif
2295
2296 if (ovs_net->xt_label)
2297 nf_connlabels_put(net);
2298 }
Linux with added FPC-III support