]> git.proxmox.com Git - mirror_ovs.git/blob - datapath/conntrack.c
projects / mirror_ovs.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
datapath: Pass nf_hook_state to nf_conntrack_in()
[mirror_ovs.git] / datapath / conntrack.c
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/kconfig.h>
15 #include <linux/version.h>
16
17 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
18
19 #include <linux/module.h>
20 #include <linux/openvswitch.h>
21 #include <linux/tcp.h>
22 #include <linux/udp.h>
23 #include <linux/sctp.h>
24 #include <linux/static_key.h>
25 #include <net/ip.h>
26 #include <net/genetlink.h>
27 #include <net/netfilter/nf_conntrack_core.h>
28 #include <net/netfilter/nf_conntrack_count.h>
29 #include <net/netfilter/nf_conntrack_helper.h>
30 #include <net/netfilter/nf_conntrack_labels.h>
31 #include <net/netfilter/nf_conntrack_seqadj.h>
32 #include <net/netfilter/nf_conntrack_zones.h>
33 #include <net/netfilter/ipv6/nf_defrag_ipv6.h>
34
35 #ifdef CONFIG_NF_NAT_NEEDED
36 #include <linux/netfilter/nf_nat.h>
37 #include <net/netfilter/nf_nat_core.h>
38 #include <net/netfilter/nf_nat_l3proto.h>
39 #endif
40
41 #include "datapath.h"
42 #include "conntrack.h"
43 #include "flow.h"
44 #include "flow_netlink.h"
45 #include "gso.h"
46
47 #ifndef HAVE_NF_NAT_RANGE2
48 #define nf_nat_range2 nf_nat_range
49 #endif
50
51 struct ovs_ct_len_tbl {
52 int maxlen;
53 int minlen;
54 };
55
56 /* Metadata mark for masked write to conntrack mark */
57 struct md_mark {
58 u32 value;
59 u32 mask;
60 };
61
62 /* Metadata label for masked write to conntrack label. */
63 struct md_labels {
64 struct ovs_key_ct_labels value;
65 struct ovs_key_ct_labels mask;
66 };
67
68 enum ovs_ct_nat {
69 OVS_CT_NAT = 1 << 0, /* NAT for committed connections only. */
70 OVS_CT_SRC_NAT = 1 << 1, /* Source NAT for NEW connections. */
71 OVS_CT_DST_NAT = 1 << 2, /* Destination NAT for NEW connections. */
72 };
73
74 /* Conntrack action context for execution. */
75 struct ovs_conntrack_info {
76 struct nf_conntrack_helper *helper;
77 struct nf_conntrack_zone zone;
78 struct nf_conn *ct;
79 u8 commit : 1;
80 u8 nat : 3; /* enum ovs_ct_nat */
81 u8 random_fully_compat : 1; /* bool */
82 u8 force : 1;
83 u8 have_eventmask : 1;
84 u16 family;
85 u32 eventmask; /* Mask of 1 << IPCT_*. */
86 struct md_mark mark;
87 struct md_labels labels;
88 #ifdef CONFIG_NF_NAT_NEEDED
89 struct nf_nat_range2 range; /* Only present for SRC NAT and DST NAT. */
90 #endif
91 };
92
93 #if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
94 #define OVS_CT_LIMIT_UNLIMITED 0
95 #define OVS_CT_LIMIT_DEFAULT OVS_CT_LIMIT_UNLIMITED
96 #define CT_LIMIT_HASH_BUCKETS 512
97 static DEFINE_STATIC_KEY_FALSE(ovs_ct_limit_enabled);
98
99 struct ovs_ct_limit {
100 /* Elements in ovs_ct_limit_info->limits hash table */
101 struct hlist_node hlist_node;
102 struct rcu_head rcu;
103 u16 zone;
104 u32 limit;
105 };
106
107 struct ovs_ct_limit_info {
108 u32 default_limit;
109 struct hlist_head *limits;
110 struct nf_conncount_data *data;
111 };
112
113 static const struct nla_policy ct_limit_policy[OVS_CT_LIMIT_ATTR_MAX + 1] = {
114 [OVS_CT_LIMIT_ATTR_ZONE_LIMIT] = { .type = NLA_NESTED, },
115 };
116 #endif
117
118 static bool labels_nonzero(const struct ovs_key_ct_labels *labels);
119
120 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info);
121
122 static u16 key_to_nfproto(const struct sw_flow_key *key)
123 {
124 switch (ntohs(key->eth.type)) {
125 case ETH_P_IP:
126 return NFPROTO_IPV4;
127 case ETH_P_IPV6:
128 return NFPROTO_IPV6;
129 default:
130 return NFPROTO_UNSPEC;
131 }
132 }
133
134 /* Map SKB connection state into the values used by flow definition. */
135 static u8 ovs_ct_get_state(enum ip_conntrack_info ctinfo)
136 {
137 u8 ct_state = OVS_CS_F_TRACKED;
138
139 switch (ctinfo) {
140 case IP_CT_ESTABLISHED_REPLY:
141 case IP_CT_RELATED_REPLY:
142 ct_state |= OVS_CS_F_REPLY_DIR;
143 break;
144 default:
145 break;
146 }
147
148 switch (ctinfo) {
149 case IP_CT_ESTABLISHED:
150 case IP_CT_ESTABLISHED_REPLY:
151 ct_state |= OVS_CS_F_ESTABLISHED;
152 break;
153 case IP_CT_RELATED:
154 case IP_CT_RELATED_REPLY:
155 ct_state |= OVS_CS_F_RELATED;
156 break;
157 case IP_CT_NEW:
158 ct_state |= OVS_CS_F_NEW;
159 break;
160 default:
161 break;
162 }
163
164 return ct_state;
165 }
166
167 static u32 ovs_ct_get_mark(const struct nf_conn *ct)
168 {
169 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
170 return ct ? ct->mark : 0;
171 #else
172 return 0;
173 #endif
174 }
175
176 /* Guard against conntrack labels max size shrinking below 128 bits. */
177 #if NF_CT_LABELS_MAX_SIZE < 16
178 #error NF_CT_LABELS_MAX_SIZE must be at least 16 bytes
179 #endif
180
181 static void ovs_ct_get_labels(const struct nf_conn *ct,
182 struct ovs_key_ct_labels *labels)
183 {
184 struct nf_conn_labels *cl = ct ? nf_ct_labels_find(ct) : NULL;
185
186 if (cl)
187 memcpy(labels, cl->bits, OVS_CT_LABELS_LEN);
188 else
189 memset(labels, 0, OVS_CT_LABELS_LEN);
190 }
191
192 static void __ovs_ct_update_key_orig_tp(struct sw_flow_key *key,
193 const struct nf_conntrack_tuple *orig,
194 u8 icmp_proto)
195 {
196 key->ct_orig_proto = orig->dst.protonum;
197 if (orig->dst.protonum == icmp_proto) {
198 key->ct.orig_tp.src = htons(orig->dst.u.icmp.type);
199 key->ct.orig_tp.dst = htons(orig->dst.u.icmp.code);
200 } else {
201 key->ct.orig_tp.src = orig->src.u.all;
202 key->ct.orig_tp.dst = orig->dst.u.all;
203 }
204 }
205
206 static void __ovs_ct_update_key(struct sw_flow_key *key, u8 state,
207 const struct nf_conntrack_zone *zone,
208 const struct nf_conn *ct)
209 {
210 key->ct_state = state;
211 key->ct_zone = zone->id;
212 key->ct.mark = ovs_ct_get_mark(ct);
213 ovs_ct_get_labels(ct, &key->ct.labels);
214
215 if (ct) {
216 const struct nf_conntrack_tuple *orig;
217
218 /* Use the master if we have one. */
219 if (ct->master)
220 ct = ct->master;
221 orig = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
222
223 /* IP version must match with the master connection. */
224 if (key->eth.type == htons(ETH_P_IP) &&
225 nf_ct_l3num(ct) == NFPROTO_IPV4) {
226 key->ipv4.ct_orig.src = orig->src.u3.ip;
227 key->ipv4.ct_orig.dst = orig->dst.u3.ip;
228 __ovs_ct_update_key_orig_tp(key, orig, IPPROTO_ICMP);
229 return;
230 } else if (key->eth.type == htons(ETH_P_IPV6) &&
231 !sw_flow_key_is_nd(key) &&
232 nf_ct_l3num(ct) == NFPROTO_IPV6) {
233 key->ipv6.ct_orig.src = orig->src.u3.in6;
234 key->ipv6.ct_orig.dst = orig->dst.u3.in6;
235 __ovs_ct_update_key_orig_tp(key, orig, NEXTHDR_ICMP);
236 return;
237 }
238 }
239 /* Clear 'ct_orig_proto' to mark the non-existence of conntrack
240 * original direction key fields.
241 */
242 key->ct_orig_proto = 0;
243 }
244
245 /* Update 'key' based on skb->_nfct. If 'post_ct' is true, then OVS has
246 * previously sent the packet to conntrack via the ct action. If
247 * 'keep_nat_flags' is true, the existing NAT flags retained, else they are
248 * initialized from the connection status.
249 */
250 static void ovs_ct_update_key(const struct sk_buff *skb,
251 const struct ovs_conntrack_info *info,
252 struct sw_flow_key *key, bool post_ct,
253 bool keep_nat_flags)
254 {
255 const struct nf_conntrack_zone *zone = &nf_ct_zone_dflt;
256 enum ip_conntrack_info ctinfo;
257 struct nf_conn *ct;
258 u8 state = 0;
259
260 ct = nf_ct_get(skb, &ctinfo);
261 if (ct) {
262 state = ovs_ct_get_state(ctinfo);
263 /* All unconfirmed entries are NEW connections. */
264 if (!nf_ct_is_confirmed(ct))
265 state |= OVS_CS_F_NEW;
266 /* OVS persists the related flag for the duration of the
267 * connection.
268 */
269 if (ct->master)
270 state |= OVS_CS_F_RELATED;
271 if (keep_nat_flags) {
272 state |= key->ct_state & OVS_CS_F_NAT_MASK;
273 } else {
274 if (ct->status & IPS_SRC_NAT)
275 state |= OVS_CS_F_SRC_NAT;
276 if (ct->status & IPS_DST_NAT)
277 state |= OVS_CS_F_DST_NAT;
278 }
279 zone = nf_ct_zone(ct);
280 } else if (post_ct) {
281 state = OVS_CS_F_TRACKED | OVS_CS_F_INVALID;
282 if (info)
283 zone = &info->zone;
284 }
285 __ovs_ct_update_key(key, state, zone, ct);
286 }
287
288 /* This is called to initialize CT key fields possibly coming in from the local
289 * stack.
290 */
291 void ovs_ct_fill_key(const struct sk_buff *skb, struct sw_flow_key *key)
292 {
293 ovs_ct_update_key(skb, NULL, key, false, false);
294 }
295
296 #define IN6_ADDR_INITIALIZER(ADDR) \
297 { (ADDR).s6_addr32[0], (ADDR).s6_addr32[1], \
298 (ADDR).s6_addr32[2], (ADDR).s6_addr32[3] }
299
300 int ovs_ct_put_key(const struct sw_flow_key *swkey,
301 const struct sw_flow_key *output, struct sk_buff *skb)
302 {
303 if (nla_put_u32(skb, OVS_KEY_ATTR_CT_STATE, output->ct_state))
304 return -EMSGSIZE;
305
306 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
307 nla_put_u16(skb, OVS_KEY_ATTR_CT_ZONE, output->ct_zone))
308 return -EMSGSIZE;
309
310 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
311 nla_put_u32(skb, OVS_KEY_ATTR_CT_MARK, output->ct.mark))
312 return -EMSGSIZE;
313
314 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
315 nla_put(skb, OVS_KEY_ATTR_CT_LABELS, sizeof(output->ct.labels),
316 &output->ct.labels))
317 return -EMSGSIZE;
318
319 if (swkey->ct_orig_proto) {
320 if (swkey->eth.type == htons(ETH_P_IP)) {
321 struct ovs_key_ct_tuple_ipv4 orig = {
322 output->ipv4.ct_orig.src,
323 output->ipv4.ct_orig.dst,
324 output->ct.orig_tp.src,
325 output->ct.orig_tp.dst,
326 output->ct_orig_proto,
327 };
328 if (nla_put(skb, OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV4,
329 sizeof(orig), &orig))
330 return -EMSGSIZE;
331 } else if (swkey->eth.type == htons(ETH_P_IPV6)) {
332 struct ovs_key_ct_tuple_ipv6 orig = {
333 IN6_ADDR_INITIALIZER(output->ipv6.ct_orig.src),
334 IN6_ADDR_INITIALIZER(output->ipv6.ct_orig.dst),
335 output->ct.orig_tp.src,
336 output->ct.orig_tp.dst,
337 output->ct_orig_proto,
338 };
339 if (nla_put(skb, OVS_KEY_ATTR_CT_ORIG_TUPLE_IPV6,
340 sizeof(orig), &orig))
341 return -EMSGSIZE;
342 }
343 }
344
345 return 0;
346 }
347
348 static int ovs_ct_set_mark(struct nf_conn *ct, struct sw_flow_key *key,
349 u32 ct_mark, u32 mask)
350 {
351 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
352 u32 new_mark;
353
354 new_mark = ct_mark | (ct->mark & ~(mask));
355 if (ct->mark != new_mark) {
356 ct->mark = new_mark;
357 if (nf_ct_is_confirmed(ct))
358 nf_conntrack_event_cache(IPCT_MARK, ct);
359 key->ct.mark = new_mark;
360 }
361
362 return 0;
363 #else
364 return -ENOTSUPP;
365 #endif
366 }
367
368 static struct nf_conn_labels *ovs_ct_get_conn_labels(struct nf_conn *ct)
369 {
370 struct nf_conn_labels *cl;
371
372 cl = nf_ct_labels_find(ct);
373 if (!cl) {
374 nf_ct_labels_ext_add(ct);
375 cl = nf_ct_labels_find(ct);
376 }
377
378 return cl;
379 }
380
381 /* Initialize labels for a new, yet to be committed conntrack entry. Note that
382 * since the new connection is not yet confirmed, and thus no-one else has
383 * access to it's labels, we simply write them over.
384 */
385 static int ovs_ct_init_labels(struct nf_conn *ct, struct sw_flow_key *key,
386 const struct ovs_key_ct_labels *labels,
387 const struct ovs_key_ct_labels *mask)
388 {
389 struct nf_conn_labels *cl, *master_cl;
390 bool have_mask = labels_nonzero(mask);
391
392 /* Inherit master's labels to the related connection? */
393 master_cl = ct->master ? nf_ct_labels_find(ct->master) : NULL;
394
395 if (!master_cl && !have_mask)
396 return 0; /* Nothing to do. */
397
398 cl = ovs_ct_get_conn_labels(ct);
399 if (!cl)
400 return -ENOSPC;
401
402 /* Inherit the master's labels, if any. Must use memcpy for backport
403 * as struct assignment only copies the length field in older
404 * kernels.
405 */
406 if (master_cl)
407 memcpy(cl->bits, master_cl->bits, OVS_CT_LABELS_LEN);
408
409 if (have_mask) {
410 u32 *dst = (u32 *)cl->bits;
411 int i;
412
413 for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
414 dst[i] = (dst[i] & ~mask->ct_labels_32[i]) |
415 (labels->ct_labels_32[i]
416 & mask->ct_labels_32[i]);
417 }
418
419 /* Labels are included in the IPCTNL_MSG_CT_NEW event only if the
420 * IPCT_LABEL bit is set in the event cache.
421 */
422 nf_conntrack_event_cache(IPCT_LABEL, ct);
423
424 memcpy(&key->ct.labels, cl->bits, OVS_CT_LABELS_LEN);
425
426 return 0;
427 }
428
429 static int ovs_ct_set_labels(struct nf_conn *ct, struct sw_flow_key *key,
430 const struct ovs_key_ct_labels *labels,
431 const struct ovs_key_ct_labels *mask)
432 {
433 struct nf_conn_labels *cl;
434 int err;
435
436 cl = ovs_ct_get_conn_labels(ct);
437 if (!cl)
438 return -ENOSPC;
439
440 err = nf_connlabels_replace(ct, labels->ct_labels_32,
441 mask->ct_labels_32,
442 OVS_CT_LABELS_LEN_32);
443 if (err)
444 return err;
445
446 memcpy(&key->ct.labels, cl->bits, OVS_CT_LABELS_LEN);
447
448 return 0;
449 }
450
451 /* 'skb' should already be pulled to nh_ofs. */
452 static int ovs_ct_helper(struct sk_buff *skb, u16 proto)
453 {
454 const struct nf_conntrack_helper *helper;
455 const struct nf_conn_help *help;
456 enum ip_conntrack_info ctinfo;
457 unsigned int protoff;
458 struct nf_conn *ct;
459 u8 nexthdr;
460 int err;
461
462 #if LINUX_VERSION_CODE < KERNEL_VERSION(4,6,0)
463 bool dst_set = false;
464 struct rtable rt = { .rt_flags = 0 };
465 #endif
466
467 ct = nf_ct_get(skb, &ctinfo);
468 if (!ct || ctinfo == IP_CT_RELATED_REPLY)
469 return NF_ACCEPT;
470
471 help = nfct_help(ct);
472 if (!help)
473 return NF_ACCEPT;
474
475 helper = rcu_dereference(help->helper);
476 if (!helper)
477 return NF_ACCEPT;
478
479 switch (proto) {
480 case NFPROTO_IPV4:
481 protoff = ip_hdrlen(skb);
482 break;
483 case NFPROTO_IPV6: {
484 __be16 frag_off;
485 int ofs;
486
487 nexthdr = ipv6_hdr(skb)->nexthdr;
488 ofs = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr,
489 &frag_off);
490 if (ofs < 0 || (frag_off & htons(~0x7)) != 0) {
491 pr_debug("proto header not found\n");
492 return NF_ACCEPT;
493 }
494 protoff = ofs;
495 break;
496 }
497 default:
498 WARN_ONCE(1, "helper invoked on non-IP family!");
499 return NF_DROP;
500 }
501
502 #if LINUX_VERSION_CODE < KERNEL_VERSION(4,6,0)
503 /* Linux 4.5 and older depend on skb_dst being set when recalculating
504 * checksums after NAT helper has mangled TCP or UDP packet payload.
505 * skb_dst is cast to a rtable struct and the flags examined.
506 * Forcing these flags to have RTCF_LOCAL not set ensures checksum mod
507 * is carried out in the same way as kernel versions > 4.5
508 */
509 if (ct->status & IPS_NAT_MASK && skb->ip_summed != CHECKSUM_PARTIAL
510 && !skb_dst(skb)) {
511 dst_set = true;
512 skb_dst_set(skb, &rt.dst);
513 }
514 #endif
515 err = helper->help(skb, protoff, ct, ctinfo);
516 if (err != NF_ACCEPT)
517 return err;
518
519 #if LINUX_VERSION_CODE < KERNEL_VERSION(4,6,0)
520 if (dst_set)
521 skb_dst_set(skb, NULL);
522 #endif
523
524 /* Adjust seqs after helper. This is needed due to some helpers (e.g.,
525 * FTP with NAT) adusting the TCP payload size when mangling IP
526 * addresses and/or port numbers in the text-based control connection.
527 */
528 if (test_bit(IPS_SEQ_ADJUST_BIT, &ct->status) &&
529 !nf_ct_seq_adjust(skb, ct, ctinfo, protoff))
530 return NF_DROP;
531 return NF_ACCEPT;
532 }
533
534 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
535 * value if 'skb' is freed.
536 */
537 static int handle_fragments(struct net *net, struct sw_flow_key *key,
538 u16 zone, struct sk_buff *skb)
539 {
540 struct ovs_gso_cb ovs_cb = *OVS_GSO_CB(skb);
541 int err;
542
543 if (key->eth.type == htons(ETH_P_IP)) {
544 enum ip_defrag_users user = IP_DEFRAG_CONNTRACK_IN + zone;
545
546 memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
547 err = ip_defrag(net, skb, user);
548 if (err)
549 return err;
550
551 ovs_cb.dp_cb.mru = IPCB(skb)->frag_max_size;
552 #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6)
553 } else if (key->eth.type == htons(ETH_P_IPV6)) {
554 enum ip6_defrag_users user = IP6_DEFRAG_CONNTRACK_IN + zone;
555
556 memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm));
557 err = nf_ct_frag6_gather(net, skb, user);
558 if (err) {
559 if (err != -EINPROGRESS)
560 kfree_skb(skb);
561 return err;
562 }
563
564 key->ip.proto = ipv6_hdr(skb)->nexthdr;
565 ovs_cb.dp_cb.mru = IP6CB(skb)->frag_max_size;
566 #endif /* IP frag support */
567 } else {
568 kfree_skb(skb);
569 return -EPFNOSUPPORT;
570 }
571
572 key->ip.frag = OVS_FRAG_TYPE_NONE;
573 skb_clear_hash(skb);
574 skb->ignore_df = 1;
575 *OVS_GSO_CB(skb) = ovs_cb;
576
577 return 0;
578 }
579
580 static struct nf_conntrack_expect *
581 ovs_ct_expect_find(struct net *net, const struct nf_conntrack_zone *zone,
582 u16 proto, const struct sk_buff *skb)
583 {
584 struct nf_conntrack_tuple tuple;
585 struct nf_conntrack_expect *exp;
586
587 if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), proto, net, &tuple))
588 return NULL;
589
590 exp = __nf_ct_expect_find(net, zone, &tuple);
591 if (exp) {
592 struct nf_conntrack_tuple_hash *h;
593
594 /* Delete existing conntrack entry, if it clashes with the
595 * expectation. This can happen since conntrack ALGs do not
596 * check for clashes between (new) expectations and existing
597 * conntrack entries. nf_conntrack_in() will check the
598 * expectations only if a conntrack entry can not be found,
599 * which can lead to OVS finding the expectation (here) in the
600 * init direction, but which will not be removed by the
601 * nf_conntrack_in() call, if a matching conntrack entry is
602 * found instead. In this case all init direction packets
603 * would be reported as new related packets, while reply
604 * direction packets would be reported as un-related
605 * established packets.
606 */
607 h = nf_conntrack_find_get(net, zone, &tuple);
608 if (h) {
609 struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
610
611 nf_ct_delete(ct, 0, 0);
612 nf_conntrack_put(&ct->ct_general);
613 }
614 }
615
616 return exp;
617 }
618
619 /* This replicates logic from nf_conntrack_core.c that is not exported. */
620 static enum ip_conntrack_info
621 ovs_ct_get_info(const struct nf_conntrack_tuple_hash *h)
622 {
623 const struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
624
625 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY)
626 return IP_CT_ESTABLISHED_REPLY;
627 /* Once we've had two way comms, always ESTABLISHED. */
628 if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status))
629 return IP_CT_ESTABLISHED;
630 if (test_bit(IPS_EXPECTED_BIT, &ct->status))
631 return IP_CT_RELATED;
632 return IP_CT_NEW;
633 }
634
635 /* Find an existing connection which this packet belongs to without
636 * re-attributing statistics or modifying the connection state. This allows an
637 * skb->_nfct lost due to an upcall to be recovered during actions execution.
638 *
639 * Must be called with rcu_read_lock.
640 *
641 * On success, populates skb->_nfct and returns the connection. Returns NULL
642 * if there is no existing entry.
643 */
644 static struct nf_conn *
645 ovs_ct_find_existing(struct net *net, const struct nf_conntrack_zone *zone,
646 u8 l3num, struct sk_buff *skb, bool natted)
647 {
648 const struct nf_conntrack_l3proto *l3proto;
649 const struct nf_conntrack_l4proto *l4proto;
650 struct nf_conntrack_tuple tuple;
651 struct nf_conntrack_tuple_hash *h;
652 struct nf_conn *ct;
653 unsigned int dataoff;
654 u8 protonum;
655
656 l3proto = __nf_ct_l3proto_find(l3num);
657 if (l3proto->get_l4proto(skb, skb_network_offset(skb), &dataoff,
658 &protonum) <= 0) {
659 pr_debug("ovs_ct_find_existing: Can't get protonum\n");
660 return NULL;
661 }
662 l4proto = __nf_ct_l4proto_find(l3num, protonum);
663 if (!nf_ct_get_tuple(skb, skb_network_offset(skb), dataoff, l3num,
664 protonum, net, &tuple, l3proto, l4proto)) {
665 pr_debug("ovs_ct_find_existing: Can't get tuple\n");
666 return NULL;
667 }
668
669 /* Must invert the tuple if skb has been transformed by NAT. */
670 if (natted) {
671 struct nf_conntrack_tuple inverse;
672
673 if (!nf_ct_invert_tuple(&inverse, &tuple, l3proto, l4proto)) {
674 pr_debug("ovs_ct_find_existing: Inversion failed!\n");
675 return NULL;
676 }
677 tuple = inverse;
678 }
679
680 /* look for tuple match */
681 h = nf_conntrack_find_get(net, zone, &tuple);
682 if (!h)
683 return NULL; /* Not found. */
684
685 ct = nf_ct_tuplehash_to_ctrack(h);
686
687 /* Inverted packet tuple matches the reverse direction conntrack tuple,
688 * select the other tuplehash to get the right 'ctinfo' bits for this
689 * packet.
690 */
691 if (natted)
692 h = &ct->tuplehash[!h->tuple.dst.dir];
693
694 nf_ct_set(skb, ct, ovs_ct_get_info(h));
695 return ct;
696 }
697
698 static
699 struct nf_conn *ovs_ct_executed(struct net *net,
700 const struct sw_flow_key *key,
701 const struct ovs_conntrack_info *info,
702 struct sk_buff *skb,
703 bool *ct_executed)
704 {
705 struct nf_conn *ct = NULL;
706
707 /* If no ct, check if we have evidence that an existing conntrack entry
708 * might be found for this skb. This happens when we lose a skb->_nfct
709 * due to an upcall, or if the direction is being forced. If the
710 * connection was not confirmed, it is not cached and needs to be run
711 * through conntrack again.
712 */
713 *ct_executed = (key->ct_state & OVS_CS_F_TRACKED) &&
714 !(key->ct_state & OVS_CS_F_INVALID) &&
715 (key->ct_zone == info->zone.id);
716
717 if (*ct_executed || (!key->ct_state && info->force)) {
718 ct = ovs_ct_find_existing(net, &info->zone, info->family, skb,
719 !!(key->ct_state &
720 OVS_CS_F_NAT_MASK));
721 }
722
723 return ct;
724 }
725
726 /* Determine whether skb->_nfct is equal to the result of conntrack lookup. */
727 static bool skb_nfct_cached(struct net *net,
728 const struct sw_flow_key *key,
729 const struct ovs_conntrack_info *info,
730 struct sk_buff *skb)
731 {
732 enum ip_conntrack_info ctinfo;
733 struct nf_conn *ct;
734 bool ct_executed = true;
735
736 ct = nf_ct_get(skb, &ctinfo);
737 if (!ct)
738 ct = ovs_ct_executed(net, key, info, skb, &ct_executed);
739
740 if (ct)
741 nf_ct_get(skb, &ctinfo);
742 else
743 return false;
744
745 if (!net_eq(net, read_pnet(&ct->ct_net)))
746 return false;
747 if (!nf_ct_zone_equal_any(info->ct, nf_ct_zone(ct)))
748 return false;
749 if (info->helper) {
750 struct nf_conn_help *help;
751
752 help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER);
753 if (help && rcu_access_pointer(help->helper) != info->helper)
754 return false;
755 }
756 /* Force conntrack entry direction to the current packet? */
757 if (info->force && CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) {
758 /* Delete the conntrack entry if confirmed, else just release
759 * the reference.
760 */
761 if (nf_ct_is_confirmed(ct))
762 nf_ct_delete(ct, 0, 0);
763
764 nf_conntrack_put(&ct->ct_general);
765 nf_ct_set(skb, NULL, 0);
766 return false;
767 }
768
769 return ct_executed;
770 }
771
772 #ifdef CONFIG_NF_NAT_NEEDED
773 /* Modelled after nf_nat_ipv[46]_fn().
774 * range is only used for new, uninitialized NAT state.
775 * Returns either NF_ACCEPT or NF_DROP.
776 */
777 static int ovs_ct_nat_execute(struct sk_buff *skb, struct nf_conn *ct,
778 enum ip_conntrack_info ctinfo,
779 const struct nf_nat_range2 *range,
780 enum nf_nat_manip_type maniptype)
781 {
782 int hooknum, nh_off, err = NF_ACCEPT;
783
784 nh_off = skb_network_offset(skb);
785 skb_pull_rcsum(skb, nh_off);
786
787 /* See HOOK2MANIP(). */
788 if (maniptype == NF_NAT_MANIP_SRC)
789 hooknum = NF_INET_LOCAL_IN; /* Source NAT */
790 else
791 hooknum = NF_INET_LOCAL_OUT; /* Destination NAT */
792
793 switch (ctinfo) {
794 case IP_CT_RELATED:
795 case IP_CT_RELATED_REPLY:
796 if (IS_ENABLED(CONFIG_NF_NAT_IPV4) &&
797 skb->protocol == htons(ETH_P_IP) &&
798 ip_hdr(skb)->protocol == IPPROTO_ICMP) {
799 if (!nf_nat_icmp_reply_translation(skb, ct, ctinfo,
800 hooknum))
801 err = NF_DROP;
802 goto push;
803 } else if (IS_ENABLED(CONFIG_NF_NAT_IPV6) &&
804 skb->protocol == htons(ETH_P_IPV6)) {
805 __be16 frag_off;
806 u8 nexthdr = ipv6_hdr(skb)->nexthdr;
807 int hdrlen = ipv6_skip_exthdr(skb,
808 sizeof(struct ipv6hdr),
809 &nexthdr, &frag_off);
810
811 if (hdrlen >= 0 && nexthdr == IPPROTO_ICMPV6) {
812 if (!nf_nat_icmpv6_reply_translation(skb, ct,
813 ctinfo,
814 hooknum,
815 hdrlen))
816 err = NF_DROP;
817 goto push;
818 }
819 }
820 /* Non-ICMP, fall thru to initialize if needed. */
821 case IP_CT_NEW:
822 /* Seen it before? This can happen for loopback, retrans,
823 * or local packets.
824 */
825 if (!nf_nat_initialized(ct, maniptype)) {
826 /* Initialize according to the NAT action. */
827 err = (range && range->flags & NF_NAT_RANGE_MAP_IPS)
828 /* Action is set up to establish a new
829 * mapping.
830 */
831 ? nf_nat_setup_info(ct, range, maniptype)
832 : nf_nat_alloc_null_binding(ct, hooknum);
833 if (err != NF_ACCEPT)
834 goto push;
835 }
836 break;
837
838 case IP_CT_ESTABLISHED:
839 case IP_CT_ESTABLISHED_REPLY:
840 break;
841
842 default:
843 err = NF_DROP;
844 goto push;
845 }
846
847 err = nf_nat_packet(ct, ctinfo, hooknum, skb);
848 push:
849 skb_push(skb, nh_off);
850 skb_postpush_rcsum(skb, skb->data, nh_off);
851
852 return err;
853 }
854
855 static void ovs_nat_update_key(struct sw_flow_key *key,
856 const struct sk_buff *skb,
857 enum nf_nat_manip_type maniptype)
858 {
859 if (maniptype == NF_NAT_MANIP_SRC) {
860 __be16 src;
861
862 key->ct_state |= OVS_CS_F_SRC_NAT;
863 if (key->eth.type == htons(ETH_P_IP))
864 key->ipv4.addr.src = ip_hdr(skb)->saddr;
865 else if (key->eth.type == htons(ETH_P_IPV6))
866 memcpy(&key->ipv6.addr.src, &ipv6_hdr(skb)->saddr,
867 sizeof(key->ipv6.addr.src));
868 else
869 return;
870
871 if (key->ip.proto == IPPROTO_UDP)
872 src = udp_hdr(skb)->source;
873 else if (key->ip.proto == IPPROTO_TCP)
874 src = tcp_hdr(skb)->source;
875 else if (key->ip.proto == IPPROTO_SCTP)
876 src = sctp_hdr(skb)->source;
877 else
878 return;
879
880 key->tp.src = src;
881 } else {
882 __be16 dst;
883
884 key->ct_state |= OVS_CS_F_DST_NAT;
885 if (key->eth.type == htons(ETH_P_IP))
886 key->ipv4.addr.dst = ip_hdr(skb)->daddr;
887 else if (key->eth.type == htons(ETH_P_IPV6))
888 memcpy(&key->ipv6.addr.dst, &ipv6_hdr(skb)->daddr,
889 sizeof(key->ipv6.addr.dst));
890 else
891 return;
892
893 if (key->ip.proto == IPPROTO_UDP)
894 dst = udp_hdr(skb)->dest;
895 else if (key->ip.proto == IPPROTO_TCP)
896 dst = tcp_hdr(skb)->dest;
897 else if (key->ip.proto == IPPROTO_SCTP)
898 dst = sctp_hdr(skb)->dest;
899 else
900 return;
901
902 key->tp.dst = dst;
903 }
904 }
905
906 /* Returns NF_DROP if the packet should be dropped, NF_ACCEPT otherwise. */
907 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
908 const struct ovs_conntrack_info *info,
909 struct sk_buff *skb, struct nf_conn *ct,
910 enum ip_conntrack_info ctinfo)
911 {
912 enum nf_nat_manip_type maniptype;
913 int err;
914
915 #ifdef HAVE_NF_CT_IS_UNTRACKED
916 if (nf_ct_is_untracked(ct)) {
917 /* A NAT action may only be performed on tracked packets. */
918 return NF_ACCEPT;
919 }
920 #endif /* HAVE_NF_CT_IS_UNTRACKED */
921
922 /* Add NAT extension if not confirmed yet. */
923 if (!nf_ct_is_confirmed(ct) && !nf_ct_nat_ext_add(ct))
924 return NF_ACCEPT; /* Can't NAT. */
925
926 /* Determine NAT type.
927 * Check if the NAT type can be deduced from the tracked connection.
928 * Make sure new expected connections (IP_CT_RELATED) are NATted only
929 * when committing.
930 */
931 if (info->nat & OVS_CT_NAT && ctinfo != IP_CT_NEW &&
932 ct->status & IPS_NAT_MASK &&
933 (ctinfo != IP_CT_RELATED || info->commit)) {
934 /* NAT an established or related connection like before. */
935 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY)
936 /* This is the REPLY direction for a connection
937 * for which NAT was applied in the forward
938 * direction. Do the reverse NAT.
939 */
940 maniptype = ct->status & IPS_SRC_NAT
941 ? NF_NAT_MANIP_DST : NF_NAT_MANIP_SRC;
942 else
943 maniptype = ct->status & IPS_SRC_NAT
944 ? NF_NAT_MANIP_SRC : NF_NAT_MANIP_DST;
945 } else if (info->nat & OVS_CT_SRC_NAT) {
946 maniptype = NF_NAT_MANIP_SRC;
947 } else if (info->nat & OVS_CT_DST_NAT) {
948 maniptype = NF_NAT_MANIP_DST;
949 } else {
950 return NF_ACCEPT; /* Connection is not NATed. */
951 }
952 err = ovs_ct_nat_execute(skb, ct, ctinfo, &info->range, maniptype);
953
954 /* Mark NAT done if successful and update the flow key. */
955 if (err == NF_ACCEPT)
956 ovs_nat_update_key(key, skb, maniptype);
957
958 return err;
959 }
960 #else /* !CONFIG_NF_NAT_NEEDED */
961 static int ovs_ct_nat(struct net *net, struct sw_flow_key *key,
962 const struct ovs_conntrack_info *info,
963 struct sk_buff *skb, struct nf_conn *ct,
964 enum ip_conntrack_info ctinfo)
965 {
966 return NF_ACCEPT;
967 }
968 #endif
969
970 /* Pass 'skb' through conntrack in 'net', using zone configured in 'info', if
971 * not done already. Update key with new CT state after passing the packet
972 * through conntrack.
973 * Note that if the packet is deemed invalid by conntrack, skb->_nfct will be
974 * set to NULL and 0 will be returned.
975 */
976 static int __ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
977 const struct ovs_conntrack_info *info,
978 struct sk_buff *skb)
979 {
980 /* If we are recirculating packets to match on conntrack fields and
981 * committing with a separate conntrack action, then we don't need to
982 * actually run the packet through conntrack twice unless it's for a
983 * different zone.
984 */
985 bool cached = skb_nfct_cached(net, key, info, skb);
986 enum ip_conntrack_info ctinfo;
987 struct nf_conn *ct;
988
989 if (!cached) {
990 struct nf_hook_state state = {
991 .hook = NF_INET_PRE_ROUTING,
992 .pf = info->family,
993 .net = net,
994 };
995 struct nf_conn *tmpl = info->ct;
996 int err;
997
998 /* Associate skb with specified zone. */
999 if (tmpl) {
1000 if (skb_nfct(skb))
1001 nf_conntrack_put(skb_nfct(skb));
1002 nf_conntrack_get(&tmpl->ct_general);
1003 nf_ct_set(skb, tmpl, IP_CT_NEW);
1004 }
1005
1006 err = nf_conntrack_in(skb, &state);
1007 if (err != NF_ACCEPT)
1008 return -ENOENT;
1009
1010 /* Clear CT state NAT flags to mark that we have not yet done
1011 * NAT after the nf_conntrack_in() call. We can actually clear
1012 * the whole state, as it will be re-initialized below.
1013 */
1014 key->ct_state = 0;
1015
1016 /* Update the key, but keep the NAT flags. */
1017 ovs_ct_update_key(skb, info, key, true, true);
1018 }
1019
1020 ct = nf_ct_get(skb, &ctinfo);
1021 if (ct) {
1022 /* Packets starting a new connection must be NATted before the
1023 * helper, so that the helper knows about the NAT. We enforce
1024 * this by delaying both NAT and helper calls for unconfirmed
1025 * connections until the committing CT action. For later
1026 * packets NAT and Helper may be called in either order.
1027 *
1028 * NAT will be done only if the CT action has NAT, and only
1029 * once per packet (per zone), as guarded by the NAT bits in
1030 * the key->ct_state.
1031 */
1032 if (info->nat && !(key->ct_state & OVS_CS_F_NAT_MASK) &&
1033 (nf_ct_is_confirmed(ct) || info->commit) &&
1034 ovs_ct_nat(net, key, info, skb, ct, ctinfo) != NF_ACCEPT) {
1035 return -EINVAL;
1036 }
1037
1038 /* Userspace may decide to perform a ct lookup without a helper
1039 * specified followed by a (recirculate and) commit with one.
1040 * Therefore, for unconfirmed connections which we will commit,
1041 * we need to attach the helper here.
1042 */
1043 if (!nf_ct_is_confirmed(ct) && info->commit &&
1044 info->helper && !nfct_help(ct)) {
1045 int err = __nf_ct_try_assign_helper(ct, info->ct,
1046 GFP_ATOMIC);
1047 if (err)
1048 return err;
1049 }
1050
1051 /* Call the helper only if:
1052 * - nf_conntrack_in() was executed above ("!cached") for a
1053 * confirmed connection, or
1054 * - When committing an unconfirmed connection.
1055 */
1056 if ((nf_ct_is_confirmed(ct) ? !cached : info->commit) &&
1057 ovs_ct_helper(skb, info->family) != NF_ACCEPT) {
1058 return -EINVAL;
1059 }
1060 }
1061
1062 return 0;
1063 }
1064
1065 /* Lookup connection and read fields into key. */
1066 static int ovs_ct_lookup(struct net *net, struct sw_flow_key *key,
1067 const struct ovs_conntrack_info *info,
1068 struct sk_buff *skb)
1069 {
1070 struct nf_conntrack_expect *exp;
1071
1072 /* If we pass an expected packet through nf_conntrack_in() the
1073 * expectation is typically removed, but the packet could still be
1074 * lost in upcall processing. To prevent this from happening we
1075 * perform an explicit expectation lookup. Expected connections are
1076 * always new, and will be passed through conntrack only when they are
1077 * committed, as it is OK to remove the expectation at that time.
1078 */
1079 exp = ovs_ct_expect_find(net, &info->zone, info->family, skb);
1080 if (exp) {
1081 u8 state;
1082
1083 /* NOTE: New connections are NATted and Helped only when
1084 * committed, so we are not calling into NAT here.
1085 */
1086 state = OVS_CS_F_TRACKED | OVS_CS_F_NEW | OVS_CS_F_RELATED;
1087 __ovs_ct_update_key(key, state, &info->zone, exp->master);
1088 } else {
1089 struct nf_conn *ct;
1090 int err;
1091
1092 err = __ovs_ct_lookup(net, key, info, skb);
1093 if (err)
1094 return err;
1095
1096 ct = (struct nf_conn *)skb_nfct(skb);
1097 if (ct)
1098 nf_ct_deliver_cached_events(ct);
1099 }
1100
1101 return 0;
1102 }
1103
1104 static bool labels_nonzero(const struct ovs_key_ct_labels *labels)
1105 {
1106 size_t i;
1107
1108 for (i = 0; i < OVS_CT_LABELS_LEN_32; i++)
1109 if (labels->ct_labels_32[i])
1110 return true;
1111
1112 return false;
1113 }
1114
1115 #if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1116 static struct hlist_head *ct_limit_hash_bucket(
1117 const struct ovs_ct_limit_info *info, u16 zone)
1118 {
1119 return &info->limits[zone & (CT_LIMIT_HASH_BUCKETS - 1)];
1120 }
1121
1122 /* Call with ovs_mutex */
1123 static void ct_limit_set(const struct ovs_ct_limit_info *info,
1124 struct ovs_ct_limit *new_ct_limit)
1125 {
1126 struct ovs_ct_limit *ct_limit;
1127 struct hlist_head *head;
1128
1129 head = ct_limit_hash_bucket(info, new_ct_limit->zone);
1130 hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
1131 if (ct_limit->zone == new_ct_limit->zone) {
1132 hlist_replace_rcu(&ct_limit->hlist_node,
1133 &new_ct_limit->hlist_node);
1134 kfree_rcu(ct_limit, rcu);
1135 return;
1136 }
1137 }
1138
1139 hlist_add_head_rcu(&new_ct_limit->hlist_node, head);
1140 }
1141
1142 /* Call with ovs_mutex */
1143 static void ct_limit_del(const struct ovs_ct_limit_info *info, u16 zone)
1144 {
1145 struct ovs_ct_limit *ct_limit;
1146 struct hlist_head *head;
1147 struct hlist_node *n;
1148
1149 head = ct_limit_hash_bucket(info, zone);
1150 hlist_for_each_entry_safe(ct_limit, n, head, hlist_node) {
1151 if (ct_limit->zone == zone) {
1152 hlist_del_rcu(&ct_limit->hlist_node);
1153 kfree_rcu(ct_limit, rcu);
1154 return;
1155 }
1156 }
1157 }
1158
1159 /* Call with RCU read lock */
1160 static u32 ct_limit_get(const struct ovs_ct_limit_info *info, u16 zone)
1161 {
1162 struct ovs_ct_limit *ct_limit;
1163 struct hlist_head *head;
1164
1165 head = ct_limit_hash_bucket(info, zone);
1166 hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
1167 if (ct_limit->zone == zone)
1168 return ct_limit->limit;
1169 }
1170
1171 return info->default_limit;
1172 }
1173
1174 static int ovs_ct_check_limit(struct net *net,
1175 const struct ovs_conntrack_info *info,
1176 const struct nf_conntrack_tuple *tuple)
1177 {
1178 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1179 const struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
1180 u32 per_zone_limit, connections;
1181 u32 conncount_key;
1182
1183 conncount_key = info->zone.id;
1184
1185 per_zone_limit = ct_limit_get(ct_limit_info, info->zone.id);
1186 if (per_zone_limit == OVS_CT_LIMIT_UNLIMITED)
1187 return 0;
1188
1189 connections = nf_conncount_count(net, ct_limit_info->data,
1190 &conncount_key, tuple, &info->zone);
1191 if (connections > per_zone_limit)
1192 return -ENOMEM;
1193
1194 return 0;
1195 }
1196 #endif
1197
1198 /* Lookup connection and confirm if unconfirmed. */
1199 static int ovs_ct_commit(struct net *net, struct sw_flow_key *key,
1200 const struct ovs_conntrack_info *info,
1201 struct sk_buff *skb)
1202 {
1203 enum ip_conntrack_info ctinfo;
1204 struct nf_conn *ct;
1205 int err;
1206
1207 err = __ovs_ct_lookup(net, key, info, skb);
1208 if (err)
1209 return err;
1210
1211 /* The connection could be invalid, in which case this is a no-op.*/
1212 ct = nf_ct_get(skb, &ctinfo);
1213 if (!ct)
1214 return 0;
1215
1216 #if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1217 if (static_branch_unlikely(&ovs_ct_limit_enabled)) {
1218 if (!nf_ct_is_confirmed(ct)) {
1219 err = ovs_ct_check_limit(net, info,
1220 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
1221 if (err) {
1222 net_warn_ratelimited("openvswitch: zone: %u "
1223 "exceeds conntrack limit\n",
1224 info->zone.id);
1225 return err;
1226 }
1227 }
1228 }
1229 #endif
1230
1231 /* Set the conntrack event mask if given. NEW and DELETE events have
1232 * their own groups, but the NFNLGRP_CONNTRACK_UPDATE group listener
1233 * typically would receive many kinds of updates. Setting the event
1234 * mask allows those events to be filtered. The set event mask will
1235 * remain in effect for the lifetime of the connection unless changed
1236 * by a further CT action with both the commit flag and the eventmask
1237 * option. */
1238 if (info->have_eventmask) {
1239 struct nf_conntrack_ecache *cache = nf_ct_ecache_find(ct);
1240
1241 if (cache)
1242 cache->ctmask = info->eventmask;
1243 }
1244
1245 /* Apply changes before confirming the connection so that the initial
1246 * conntrack NEW netlink event carries the values given in the CT
1247 * action.
1248 */
1249 if (info->mark.mask) {
1250 err = ovs_ct_set_mark(ct, key, info->mark.value,
1251 info->mark.mask);
1252 if (err)
1253 return err;
1254 }
1255 if (!nf_ct_is_confirmed(ct)) {
1256 err = ovs_ct_init_labels(ct, key, &info->labels.value,
1257 &info->labels.mask);
1258 if (err)
1259 return err;
1260 } else if (labels_nonzero(&info->labels.mask)) {
1261 err = ovs_ct_set_labels(ct, key, &info->labels.value,
1262 &info->labels.mask);
1263 if (err)
1264 return err;
1265 }
1266 /* This will take care of sending queued events even if the connection
1267 * is already confirmed.
1268 */
1269 if (nf_conntrack_confirm(skb) != NF_ACCEPT)
1270 return -EINVAL;
1271
1272 return 0;
1273 }
1274
1275 /* Trim the skb to the length specified by the IP/IPv6 header,
1276 * removing any trailing lower-layer padding. This prepares the skb
1277 * for higher-layer processing that assumes skb->len excludes padding
1278 * (such as nf_ip_checksum). The caller needs to pull the skb to the
1279 * network header, and ensure ip_hdr/ipv6_hdr points to valid data.
1280 */
1281 static int ovs_skb_network_trim(struct sk_buff *skb)
1282 {
1283 unsigned int len;
1284 int err;
1285
1286 switch (skb->protocol) {
1287 case htons(ETH_P_IP):
1288 len = ntohs(ip_hdr(skb)->tot_len);
1289 break;
1290 case htons(ETH_P_IPV6):
1291 len = sizeof(struct ipv6hdr)
1292 + ntohs(ipv6_hdr(skb)->payload_len);
1293 break;
1294 default:
1295 len = skb->len;
1296 }
1297
1298 err = pskb_trim_rcsum(skb, len);
1299 if (err)
1300 kfree_skb(skb);
1301
1302 return err;
1303 }
1304
1305 /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero
1306 * value if 'skb' is freed.
1307 */
1308 int ovs_ct_execute(struct net *net, struct sk_buff *skb,
1309 struct sw_flow_key *key,
1310 const struct ovs_conntrack_info *info)
1311 {
1312 int nh_ofs;
1313 int err;
1314
1315 /* The conntrack module expects to be working at L3. */
1316 nh_ofs = skb_network_offset(skb);
1317 skb_pull_rcsum(skb, nh_ofs);
1318
1319 err = ovs_skb_network_trim(skb);
1320 if (err)
1321 return err;
1322
1323 if (key->ip.frag != OVS_FRAG_TYPE_NONE) {
1324 err = handle_fragments(net, key, info->zone.id, skb);
1325 if (err)
1326 return err;
1327 }
1328
1329 if (info->commit)
1330 err = ovs_ct_commit(net, key, info, skb);
1331 else
1332 err = ovs_ct_lookup(net, key, info, skb);
1333
1334 skb_push(skb, nh_ofs);
1335 skb_postpush_rcsum(skb, skb->data, nh_ofs);
1336 if (err)
1337 kfree_skb(skb);
1338 return err;
1339 }
1340
1341 int ovs_ct_clear(struct sk_buff *skb, struct sw_flow_key *key)
1342 {
1343 if (skb_nfct(skb)) {
1344 nf_conntrack_put(skb_nfct(skb));
1345 #ifdef HAVE_IP_CT_UNTRACKED
1346 nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
1347 #else
1348 nf_ct_set(skb, NULL, 0);
1349 #endif
1350 ovs_ct_fill_key(skb, key);
1351 }
1352
1353 return 0;
1354 }
1355
1356 static int ovs_ct_add_helper(struct ovs_conntrack_info *info, const char *name,
1357 const struct sw_flow_key *key, bool log)
1358 {
1359 struct nf_conntrack_helper *helper;
1360 struct nf_conn_help *help;
1361
1362 helper = nf_conntrack_helper_try_module_get(name, info->family,
1363 key->ip.proto);
1364 if (!helper) {
1365 OVS_NLERR(log, "Unknown helper \"%s\"", name);
1366 return -EINVAL;
1367 }
1368
1369 help = nf_ct_helper_ext_add(info->ct, helper, GFP_KERNEL);
1370 if (!help) {
1371 nf_conntrack_helper_put(helper);
1372 return -ENOMEM;
1373 }
1374
1375 rcu_assign_pointer(help->helper, helper);
1376 info->helper = helper;
1377
1378 if (info->nat)
1379 request_module("ip_nat_%s", name);
1380
1381 return 0;
1382 }
1383
1384 #ifdef CONFIG_NF_NAT_NEEDED
1385 static int parse_nat(const struct nlattr *attr,
1386 struct ovs_conntrack_info *info, bool log)
1387 {
1388 struct nlattr *a;
1389 int rem;
1390 bool have_ip_max = false;
1391 bool have_proto_max = false;
1392 bool ip_vers = (info->family == NFPROTO_IPV6);
1393
1394 nla_for_each_nested(a, attr, rem) {
1395 static const int ovs_nat_attr_lens[OVS_NAT_ATTR_MAX + 1][2] = {
1396 [OVS_NAT_ATTR_SRC] = {0, 0},
1397 [OVS_NAT_ATTR_DST] = {0, 0},
1398 [OVS_NAT_ATTR_IP_MIN] = {sizeof(struct in_addr),
1399 sizeof(struct in6_addr)},
1400 [OVS_NAT_ATTR_IP_MAX] = {sizeof(struct in_addr),
1401 sizeof(struct in6_addr)},
1402 [OVS_NAT_ATTR_PROTO_MIN] = {sizeof(u16), sizeof(u16)},
1403 [OVS_NAT_ATTR_PROTO_MAX] = {sizeof(u16), sizeof(u16)},
1404 [OVS_NAT_ATTR_PERSISTENT] = {0, 0},
1405 [OVS_NAT_ATTR_PROTO_HASH] = {0, 0},
1406 [OVS_NAT_ATTR_PROTO_RANDOM] = {0, 0},
1407 };
1408 int type = nla_type(a);
1409
1410 if (type > OVS_NAT_ATTR_MAX) {
1411 OVS_NLERR(log, "Unknown NAT attribute (type=%d, max=%d)",
1412 type, OVS_NAT_ATTR_MAX);
1413 return -EINVAL;
1414 }
1415
1416 if (nla_len(a) != ovs_nat_attr_lens[type][ip_vers]) {
1417 OVS_NLERR(log, "NAT attribute type %d has unexpected length (%d != %d)",
1418 type, nla_len(a),
1419 ovs_nat_attr_lens[type][ip_vers]);
1420 return -EINVAL;
1421 }
1422
1423 switch (type) {
1424 case OVS_NAT_ATTR_SRC:
1425 case OVS_NAT_ATTR_DST:
1426 if (info->nat) {
1427 OVS_NLERR(log, "Only one type of NAT may be specified");
1428 return -ERANGE;
1429 }
1430 info->nat |= OVS_CT_NAT;
1431 info->nat |= ((type == OVS_NAT_ATTR_SRC)
1432 ? OVS_CT_SRC_NAT : OVS_CT_DST_NAT);
1433 break;
1434
1435 case OVS_NAT_ATTR_IP_MIN:
1436 nla_memcpy(&info->range.min_addr, a,
1437 sizeof(info->range.min_addr));
1438 info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1439 break;
1440
1441 case OVS_NAT_ATTR_IP_MAX:
1442 have_ip_max = true;
1443 nla_memcpy(&info->range.max_addr, a,
1444 sizeof(info->range.max_addr));
1445 info->range.flags |= NF_NAT_RANGE_MAP_IPS;
1446 break;
1447
1448 case OVS_NAT_ATTR_PROTO_MIN:
1449 info->range.min_proto.all = htons(nla_get_u16(a));
1450 info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1451 break;
1452
1453 case OVS_NAT_ATTR_PROTO_MAX:
1454 have_proto_max = true;
1455 info->range.max_proto.all = htons(nla_get_u16(a));
1456 info->range.flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1457 break;
1458
1459 case OVS_NAT_ATTR_PERSISTENT:
1460 info->range.flags |= NF_NAT_RANGE_PERSISTENT;
1461 break;
1462
1463 case OVS_NAT_ATTR_PROTO_HASH:
1464 info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM;
1465 break;
1466
1467 case OVS_NAT_ATTR_PROTO_RANDOM:
1468 #ifdef NF_NAT_RANGE_PROTO_RANDOM_FULLY
1469 info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM_FULLY;
1470 #else
1471 info->range.flags |= NF_NAT_RANGE_PROTO_RANDOM;
1472 info->random_fully_compat = true;
1473 #endif
1474 break;
1475
1476 default:
1477 OVS_NLERR(log, "Unknown nat attribute (%d)", type);
1478 return -EINVAL;
1479 }
1480 }
1481
1482 if (rem > 0) {
1483 OVS_NLERR(log, "NAT attribute has %d unknown bytes", rem);
1484 return -EINVAL;
1485 }
1486 if (!info->nat) {
1487 /* Do not allow flags if no type is given. */
1488 if (info->range.flags) {
1489 OVS_NLERR(log,
1490 "NAT flags may be given only when NAT range (SRC or DST) is also specified."
1491 );
1492 return -EINVAL;
1493 }
1494 info->nat = OVS_CT_NAT; /* NAT existing connections. */
1495 } else if (!info->commit) {
1496 OVS_NLERR(log,
1497 "NAT attributes may be specified only when CT COMMIT flag is also specified."
1498 );
1499 return -EINVAL;
1500 }
1501 /* Allow missing IP_MAX. */
1502 if (info->range.flags & NF_NAT_RANGE_MAP_IPS && !have_ip_max) {
1503 memcpy(&info->range.max_addr, &info->range.min_addr,
1504 sizeof(info->range.max_addr));
1505 }
1506 /* Allow missing PROTO_MAX. */
1507 if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1508 !have_proto_max) {
1509 info->range.max_proto.all = info->range.min_proto.all;
1510 }
1511 return 0;
1512 }
1513 #endif
1514
1515 static const struct ovs_ct_len_tbl ovs_ct_attr_lens[OVS_CT_ATTR_MAX + 1] = {
1516 [OVS_CT_ATTR_COMMIT] = { .minlen = 0, .maxlen = 0 },
1517 [OVS_CT_ATTR_FORCE_COMMIT] = { .minlen = 0, .maxlen = 0 },
1518 [OVS_CT_ATTR_ZONE] = { .minlen = sizeof(u16),
1519 .maxlen = sizeof(u16) },
1520 [OVS_CT_ATTR_MARK] = { .minlen = sizeof(struct md_mark),
1521 .maxlen = sizeof(struct md_mark) },
1522 [OVS_CT_ATTR_LABELS] = { .minlen = sizeof(struct md_labels),
1523 .maxlen = sizeof(struct md_labels) },
1524 [OVS_CT_ATTR_HELPER] = { .minlen = 1,
1525 .maxlen = NF_CT_HELPER_NAME_LEN },
1526 #ifdef CONFIG_NF_NAT_NEEDED
1527 /* NAT length is checked when parsing the nested attributes. */
1528 [OVS_CT_ATTR_NAT] = { .minlen = 0, .maxlen = INT_MAX },
1529 #endif
1530 [OVS_CT_ATTR_EVENTMASK] = { .minlen = sizeof(u32),
1531 .maxlen = sizeof(u32) },
1532 };
1533
1534 static int parse_ct(const struct nlattr *attr, struct ovs_conntrack_info *info,
1535 const char **helper, bool log)
1536 {
1537 struct nlattr *a;
1538 int rem;
1539
1540 nla_for_each_nested(a, attr, rem) {
1541 int type = nla_type(a);
1542 int maxlen;
1543 int minlen;
1544
1545 if (type > OVS_CT_ATTR_MAX) {
1546 OVS_NLERR(log,
1547 "Unknown conntrack attr (type=%d, max=%d)",
1548 type, OVS_CT_ATTR_MAX);
1549 return -EINVAL;
1550 }
1551
1552 maxlen = ovs_ct_attr_lens[type].maxlen;
1553 minlen = ovs_ct_attr_lens[type].minlen;
1554 if (nla_len(a) < minlen || nla_len(a) > maxlen) {
1555 OVS_NLERR(log,
1556 "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)",
1557 type, nla_len(a), maxlen);
1558 return -EINVAL;
1559 }
1560
1561 switch (type) {
1562 case OVS_CT_ATTR_FORCE_COMMIT:
1563 info->force = true;
1564 /* fall through. */
1565 case OVS_CT_ATTR_COMMIT:
1566 info->commit = true;
1567 break;
1568 #ifdef CONFIG_NF_CONNTRACK_ZONES
1569 case OVS_CT_ATTR_ZONE:
1570 info->zone.id = nla_get_u16(a);
1571 break;
1572 #endif
1573 #ifdef CONFIG_NF_CONNTRACK_MARK
1574 case OVS_CT_ATTR_MARK: {
1575 struct md_mark *mark = nla_data(a);
1576
1577 if (!mark->mask) {
1578 OVS_NLERR(log, "ct_mark mask cannot be 0");
1579 return -EINVAL;
1580 }
1581 info->mark = *mark;
1582 break;
1583 }
1584 #endif
1585 #ifdef CONFIG_NF_CONNTRACK_LABELS
1586 case OVS_CT_ATTR_LABELS: {
1587 struct md_labels *labels = nla_data(a);
1588
1589 if (!labels_nonzero(&labels->mask)) {
1590 OVS_NLERR(log, "ct_labels mask cannot be 0");
1591 return -EINVAL;
1592 }
1593 info->labels = *labels;
1594 break;
1595 }
1596 #endif
1597 case OVS_CT_ATTR_HELPER:
1598 *helper = nla_data(a);
1599 if (!memchr(*helper, '\0', nla_len(a))) {
1600 OVS_NLERR(log, "Invalid conntrack helper");
1601 return -EINVAL;
1602 }
1603 break;
1604 #ifdef CONFIG_NF_NAT_NEEDED
1605 case OVS_CT_ATTR_NAT: {
1606 int err = parse_nat(a, info, log);
1607
1608 if (err)
1609 return err;
1610 break;
1611 }
1612 #endif
1613 case OVS_CT_ATTR_EVENTMASK:
1614 info->have_eventmask = true;
1615 info->eventmask = nla_get_u32(a);
1616 break;
1617
1618 default:
1619 OVS_NLERR(log, "Unknown conntrack attr (%d)",
1620 type);
1621 return -EINVAL;
1622 }
1623 }
1624
1625 #ifdef CONFIG_NF_CONNTRACK_MARK
1626 if (!info->commit && info->mark.mask) {
1627 OVS_NLERR(log,
1628 "Setting conntrack mark requires 'commit' flag.");
1629 return -EINVAL;
1630 }
1631 #endif
1632 #ifdef CONFIG_NF_CONNTRACK_LABELS
1633 if (!info->commit && labels_nonzero(&info->labels.mask)) {
1634 OVS_NLERR(log,
1635 "Setting conntrack labels requires 'commit' flag.");
1636 return -EINVAL;
1637 }
1638 #endif
1639 if (rem > 0) {
1640 OVS_NLERR(log, "Conntrack attr has %d unknown bytes", rem);
1641 return -EINVAL;
1642 }
1643
1644 return 0;
1645 }
1646
1647 bool ovs_ct_verify(struct net *net, enum ovs_key_attr attr)
1648 {
1649 if (attr == OVS_KEY_ATTR_CT_STATE)
1650 return true;
1651 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1652 attr == OVS_KEY_ATTR_CT_ZONE)
1653 return true;
1654 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1655 attr == OVS_KEY_ATTR_CT_MARK)
1656 return true;
1657 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1658 attr == OVS_KEY_ATTR_CT_LABELS) {
1659 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
1660
1661 return ovs_net->xt_label;
1662 }
1663
1664 return false;
1665 }
1666
1667 int ovs_ct_copy_action(struct net *net, const struct nlattr *attr,
1668 const struct sw_flow_key *key,
1669 struct sw_flow_actions **sfa, bool log)
1670 {
1671 struct ovs_conntrack_info ct_info;
1672 const char *helper = NULL;
1673 u16 family;
1674 int err;
1675
1676 family = key_to_nfproto(key);
1677 if (family == NFPROTO_UNSPEC) {
1678 OVS_NLERR(log, "ct family unspecified");
1679 return -EINVAL;
1680 }
1681
1682 memset(&ct_info, 0, sizeof(ct_info));
1683 ct_info.family = family;
1684
1685 nf_ct_zone_init(&ct_info.zone, NF_CT_DEFAULT_ZONE_ID,
1686 NF_CT_DEFAULT_ZONE_DIR, 0);
1687
1688 err = parse_ct(attr, &ct_info, &helper, log);
1689 if (err)
1690 return err;
1691
1692 /* Set up template for tracking connections in specific zones. */
1693 ct_info.ct = nf_ct_tmpl_alloc(net, &ct_info.zone, GFP_KERNEL);
1694 if (!ct_info.ct) {
1695 OVS_NLERR(log, "Failed to allocate conntrack template");
1696 return -ENOMEM;
1697 }
1698 if (helper) {
1699 err = ovs_ct_add_helper(&ct_info, helper, key, log);
1700 if (err)
1701 goto err_free_ct;
1702 }
1703
1704 err = ovs_nla_add_action(sfa, OVS_ACTION_ATTR_CT, &ct_info,
1705 sizeof(ct_info), log);
1706 if (err)
1707 goto err_free_ct;
1708
1709 __set_bit(IPS_CONFIRMED_BIT, &ct_info.ct->status);
1710 nf_conntrack_get(&ct_info.ct->ct_general);
1711 return 0;
1712 err_free_ct:
1713 __ovs_ct_free_action(&ct_info);
1714 return err;
1715 }
1716
1717 #ifdef CONFIG_NF_NAT_NEEDED
1718 static bool ovs_ct_nat_to_attr(const struct ovs_conntrack_info *info,
1719 struct sk_buff *skb)
1720 {
1721 struct nlattr *start;
1722
1723 start = nla_nest_start(skb, OVS_CT_ATTR_NAT);
1724 if (!start)
1725 return false;
1726
1727 if (info->nat & OVS_CT_SRC_NAT) {
1728 if (nla_put_flag(skb, OVS_NAT_ATTR_SRC))
1729 return false;
1730 } else if (info->nat & OVS_CT_DST_NAT) {
1731 if (nla_put_flag(skb, OVS_NAT_ATTR_DST))
1732 return false;
1733 } else {
1734 goto out;
1735 }
1736
1737 if (info->range.flags & NF_NAT_RANGE_MAP_IPS) {
1738 if (IS_ENABLED(CONFIG_NF_NAT_IPV4) &&
1739 info->family == NFPROTO_IPV4) {
1740 if (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MIN,
1741 info->range.min_addr.ip) ||
1742 (info->range.max_addr.ip
1743 != info->range.min_addr.ip &&
1744 (nla_put_in_addr(skb, OVS_NAT_ATTR_IP_MAX,
1745 info->range.max_addr.ip))))
1746 return false;
1747 } else if (IS_ENABLED(CONFIG_NF_NAT_IPV6) &&
1748 info->family == NFPROTO_IPV6) {
1749 if (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MIN,
1750 &info->range.min_addr.in6) ||
1751 (memcmp(&info->range.max_addr.in6,
1752 &info->range.min_addr.in6,
1753 sizeof(info->range.max_addr.in6)) &&
1754 (nla_put_in6_addr(skb, OVS_NAT_ATTR_IP_MAX,
1755 &info->range.max_addr.in6))))
1756 return false;
1757 } else {
1758 return false;
1759 }
1760 }
1761 if (info->range.flags & NF_NAT_RANGE_PROTO_SPECIFIED &&
1762 (nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MIN,
1763 ntohs(info->range.min_proto.all)) ||
1764 (info->range.max_proto.all != info->range.min_proto.all &&
1765 nla_put_u16(skb, OVS_NAT_ATTR_PROTO_MAX,
1766 ntohs(info->range.max_proto.all)))))
1767 return false;
1768
1769 if (info->range.flags & NF_NAT_RANGE_PERSISTENT &&
1770 nla_put_flag(skb, OVS_NAT_ATTR_PERSISTENT))
1771 return false;
1772 if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM &&
1773 nla_put_flag(skb, info->random_fully_compat
1774 ? OVS_NAT_ATTR_PROTO_RANDOM
1775 : OVS_NAT_ATTR_PROTO_HASH))
1776 return false;
1777 #ifdef NF_NAT_RANGE_PROTO_RANDOM_FULLY
1778 if (info->range.flags & NF_NAT_RANGE_PROTO_RANDOM_FULLY &&
1779 nla_put_flag(skb, OVS_NAT_ATTR_PROTO_RANDOM))
1780 return false;
1781 #endif
1782 out:
1783 nla_nest_end(skb, start);
1784
1785 return true;
1786 }
1787 #endif
1788
1789 int ovs_ct_action_to_attr(const struct ovs_conntrack_info *ct_info,
1790 struct sk_buff *skb)
1791 {
1792 struct nlattr *start;
1793
1794 start = nla_nest_start(skb, OVS_ACTION_ATTR_CT);
1795 if (!start)
1796 return -EMSGSIZE;
1797
1798 if (ct_info->commit && nla_put_flag(skb, ct_info->force
1799 ? OVS_CT_ATTR_FORCE_COMMIT
1800 : OVS_CT_ATTR_COMMIT))
1801 return -EMSGSIZE;
1802 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1803 nla_put_u16(skb, OVS_CT_ATTR_ZONE, ct_info->zone.id))
1804 return -EMSGSIZE;
1805 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && ct_info->mark.mask &&
1806 nla_put(skb, OVS_CT_ATTR_MARK, sizeof(ct_info->mark),
1807 &ct_info->mark))
1808 return -EMSGSIZE;
1809 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1810 labels_nonzero(&ct_info->labels.mask) &&
1811 nla_put(skb, OVS_CT_ATTR_LABELS, sizeof(ct_info->labels),
1812 &ct_info->labels))
1813 return -EMSGSIZE;
1814 if (ct_info->helper) {
1815 if (nla_put_string(skb, OVS_CT_ATTR_HELPER,
1816 ct_info->helper->name))
1817 return -EMSGSIZE;
1818 }
1819 if (ct_info->have_eventmask &&
1820 nla_put_u32(skb, OVS_CT_ATTR_EVENTMASK, ct_info->eventmask))
1821 return -EMSGSIZE;
1822
1823 #ifdef CONFIG_NF_NAT_NEEDED
1824 if (ct_info->nat && !ovs_ct_nat_to_attr(ct_info, skb))
1825 return -EMSGSIZE;
1826 #endif
1827 nla_nest_end(skb, start);
1828
1829 return 0;
1830 }
1831
1832 void ovs_ct_free_action(const struct nlattr *a)
1833 {
1834 struct ovs_conntrack_info *ct_info = nla_data(a);
1835
1836 __ovs_ct_free_action(ct_info);
1837 }
1838
1839 static void __ovs_ct_free_action(struct ovs_conntrack_info *ct_info)
1840 {
1841 if (ct_info->helper)
1842 nf_conntrack_helper_put(ct_info->helper);
1843 if (ct_info->ct)
1844 nf_ct_tmpl_free(ct_info->ct);
1845 }
1846
1847 #if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
1848 static int ovs_ct_limit_init(struct net *net, struct ovs_net *ovs_net)
1849 {
1850 int i, err;
1851
1852 ovs_net->ct_limit_info = kmalloc(sizeof(*ovs_net->ct_limit_info),
1853 GFP_KERNEL);
1854 if (!ovs_net->ct_limit_info)
1855 return -ENOMEM;
1856
1857 ovs_net->ct_limit_info->default_limit = OVS_CT_LIMIT_DEFAULT;
1858 ovs_net->ct_limit_info->limits =
1859 kmalloc_array(CT_LIMIT_HASH_BUCKETS, sizeof(struct hlist_head),
1860 GFP_KERNEL);
1861 if (!ovs_net->ct_limit_info->limits) {
1862 kfree(ovs_net->ct_limit_info);
1863 return -ENOMEM;
1864 }
1865
1866 for (i = 0; i < CT_LIMIT_HASH_BUCKETS; i++)
1867 INIT_HLIST_HEAD(&ovs_net->ct_limit_info->limits[i]);
1868
1869 ovs_net->ct_limit_info->data =
1870 nf_conncount_init(net, NFPROTO_INET, sizeof(u32));
1871
1872 if (IS_ERR(ovs_net->ct_limit_info->data)) {
1873 err = PTR_ERR(ovs_net->ct_limit_info->data);
1874 kfree(ovs_net->ct_limit_info->limits);
1875 kfree(ovs_net->ct_limit_info);
1876 pr_err("openvswitch: failed to init nf_conncount %d\n", err);
1877 return err;
1878 }
1879 return 0;
1880 }
1881
1882 static void ovs_ct_limit_exit(struct net *net, struct ovs_net *ovs_net)
1883 {
1884 const struct ovs_ct_limit_info *info = ovs_net->ct_limit_info;
1885 int i;
1886
1887 nf_conncount_destroy(net, NFPROTO_INET, info->data);
1888 for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) {
1889 struct hlist_head *head = &info->limits[i];
1890 struct ovs_ct_limit *ct_limit;
1891
1892 hlist_for_each_entry_rcu(ct_limit, head, hlist_node)
1893 kfree_rcu(ct_limit, rcu);
1894 }
1895 kfree(ovs_net->ct_limit_info->limits);
1896 kfree(ovs_net->ct_limit_info);
1897 }
1898
1899 static struct sk_buff *
1900 ovs_ct_limit_cmd_reply_start(struct genl_info *info, u8 cmd,
1901 struct ovs_header **ovs_reply_header)
1902 {
1903 struct ovs_header *ovs_header = info->userhdr;
1904 struct sk_buff *skb;
1905
1906 skb = genlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1907 if (!skb)
1908 return ERR_PTR(-ENOMEM);
1909
1910 *ovs_reply_header = genlmsg_put(skb, info->snd_portid,
1911 info->snd_seq,
1912 &dp_ct_limit_genl_family, 0, cmd);
1913
1914 if (!*ovs_reply_header) {
1915 nlmsg_free(skb);
1916 return ERR_PTR(-EMSGSIZE);
1917 }
1918 (*ovs_reply_header)->dp_ifindex = ovs_header->dp_ifindex;
1919
1920 return skb;
1921 }
1922
1923 static bool check_zone_id(int zone_id, u16 *pzone)
1924 {
1925 if (zone_id >= 0 && zone_id <= 65535) {
1926 *pzone = (u16)zone_id;
1927 return true;
1928 }
1929 return false;
1930 }
1931
1932 static int ovs_ct_limit_set_zone_limit(struct nlattr *nla_zone_limit,
1933 struct ovs_ct_limit_info *info)
1934 {
1935 struct ovs_zone_limit *zone_limit;
1936 int rem;
1937 u16 zone;
1938
1939 rem = NLA_ALIGN(nla_len(nla_zone_limit));
1940 zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1941
1942 while (rem >= sizeof(*zone_limit)) {
1943 if (unlikely(zone_limit->zone_id ==
1944 OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1945 ovs_lock();
1946 info->default_limit = zone_limit->limit;
1947 ovs_unlock();
1948 } else if (unlikely(!check_zone_id(
1949 zone_limit->zone_id, &zone))) {
1950 OVS_NLERR(true, "zone id is out of range");
1951 } else {
1952 struct ovs_ct_limit *ct_limit;
1953
1954 ct_limit = kmalloc(sizeof(*ct_limit), GFP_KERNEL);
1955 if (!ct_limit)
1956 return -ENOMEM;
1957
1958 ct_limit->zone = zone;
1959 ct_limit->limit = zone_limit->limit;
1960
1961 ovs_lock();
1962 ct_limit_set(info, ct_limit);
1963 ovs_unlock();
1964 }
1965 rem -= NLA_ALIGN(sizeof(*zone_limit));
1966 zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
1967 NLA_ALIGN(sizeof(*zone_limit)));
1968 }
1969
1970 if (rem)
1971 OVS_NLERR(true, "set zone limit has %d unknown bytes", rem);
1972
1973 return 0;
1974 }
1975
1976 static int ovs_ct_limit_del_zone_limit(struct nlattr *nla_zone_limit,
1977 struct ovs_ct_limit_info *info)
1978 {
1979 struct ovs_zone_limit *zone_limit;
1980 int rem;
1981 u16 zone;
1982
1983 rem = NLA_ALIGN(nla_len(nla_zone_limit));
1984 zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
1985
1986 while (rem >= sizeof(*zone_limit)) {
1987 if (unlikely(zone_limit->zone_id ==
1988 OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
1989 ovs_lock();
1990 info->default_limit = OVS_CT_LIMIT_DEFAULT;
1991 ovs_unlock();
1992 } else if (unlikely(!check_zone_id(
1993 zone_limit->zone_id, &zone))) {
1994 OVS_NLERR(true, "zone id is out of range");
1995 } else {
1996 ovs_lock();
1997 ct_limit_del(info, zone);
1998 ovs_unlock();
1999 }
2000 rem -= NLA_ALIGN(sizeof(*zone_limit));
2001 zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
2002 NLA_ALIGN(sizeof(*zone_limit)));
2003 }
2004
2005 if (rem)
2006 OVS_NLERR(true, "del zone limit has %d unknown bytes", rem);
2007
2008 return 0;
2009 }
2010
2011 static int ovs_ct_limit_get_default_limit(struct ovs_ct_limit_info *info,
2012 struct sk_buff *reply)
2013 {
2014 struct ovs_zone_limit zone_limit;
2015 int err;
2016
2017 zone_limit.zone_id = OVS_ZONE_LIMIT_DEFAULT_ZONE;
2018 zone_limit.limit = info->default_limit;
2019 err = nla_put_nohdr(reply, sizeof(zone_limit), &zone_limit);
2020 if (err)
2021 return err;
2022
2023 return 0;
2024 }
2025
2026 static int __ovs_ct_limit_get_zone_limit(struct net *net,
2027 struct nf_conncount_data *data,
2028 u16 zone_id, u32 limit,
2029 struct sk_buff *reply)
2030 {
2031 struct nf_conntrack_zone ct_zone;
2032 struct ovs_zone_limit zone_limit;
2033 u32 conncount_key = zone_id;
2034
2035 zone_limit.zone_id = zone_id;
2036 zone_limit.limit = limit;
2037 nf_ct_zone_init(&ct_zone, zone_id, NF_CT_DEFAULT_ZONE_DIR, 0);
2038
2039 zone_limit.count = nf_conncount_count(net, data, &conncount_key, NULL,
2040 &ct_zone);
2041 return nla_put_nohdr(reply, sizeof(zone_limit), &zone_limit);
2042 }
2043
2044 static int ovs_ct_limit_get_zone_limit(struct net *net,
2045 struct nlattr *nla_zone_limit,
2046 struct ovs_ct_limit_info *info,
2047 struct sk_buff *reply)
2048 {
2049 struct ovs_zone_limit *zone_limit;
2050 int rem, err;
2051 u32 limit;
2052 u16 zone;
2053
2054 rem = NLA_ALIGN(nla_len(nla_zone_limit));
2055 zone_limit = (struct ovs_zone_limit *)nla_data(nla_zone_limit);
2056
2057 while (rem >= sizeof(*zone_limit)) {
2058 if (unlikely(zone_limit->zone_id ==
2059 OVS_ZONE_LIMIT_DEFAULT_ZONE)) {
2060 err = ovs_ct_limit_get_default_limit(info, reply);
2061 if (err)
2062 return err;
2063 } else if (unlikely(!check_zone_id(zone_limit->zone_id,
2064 &zone))) {
2065 OVS_NLERR(true, "zone id is out of range");
2066 } else {
2067 rcu_read_lock();
2068 limit = ct_limit_get(info, zone);
2069 rcu_read_unlock();
2070
2071 err = __ovs_ct_limit_get_zone_limit(
2072 net, info->data, zone, limit, reply);
2073 if (err)
2074 return err;
2075 }
2076 rem -= NLA_ALIGN(sizeof(*zone_limit));
2077 zone_limit = (struct ovs_zone_limit *)((u8 *)zone_limit +
2078 NLA_ALIGN(sizeof(*zone_limit)));
2079 }
2080
2081 if (rem)
2082 OVS_NLERR(true, "get zone limit has %d unknown bytes", rem);
2083
2084 return 0;
2085 }
2086
2087 static int ovs_ct_limit_get_all_zone_limit(struct net *net,
2088 struct ovs_ct_limit_info *info,
2089 struct sk_buff *reply)
2090 {
2091 struct ovs_ct_limit *ct_limit;
2092 struct hlist_head *head;
2093 int i, err = 0;
2094
2095 err = ovs_ct_limit_get_default_limit(info, reply);
2096 if (err)
2097 return err;
2098
2099 rcu_read_lock();
2100 for (i = 0; i < CT_LIMIT_HASH_BUCKETS; ++i) {
2101 head = &info->limits[i];
2102 hlist_for_each_entry_rcu(ct_limit, head, hlist_node) {
2103 err = __ovs_ct_limit_get_zone_limit(net, info->data,
2104 ct_limit->zone, ct_limit->limit, reply);
2105 if (err)
2106 goto exit_err;
2107 }
2108 }
2109
2110 exit_err:
2111 rcu_read_unlock();
2112 return err;
2113 }
2114
2115 static int ovs_ct_limit_cmd_set(struct sk_buff *skb, struct genl_info *info)
2116 {
2117 struct nlattr **a = info->attrs;
2118 struct sk_buff *reply;
2119 struct ovs_header *ovs_reply_header;
2120 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
2121 struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2122 int err;
2123
2124 reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_SET,
2125 &ovs_reply_header);
2126 if (IS_ERR(reply))
2127 return PTR_ERR(reply);
2128
2129 if (!a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2130 err = -EINVAL;
2131 goto exit_err;
2132 }
2133
2134 err = ovs_ct_limit_set_zone_limit(a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT],
2135 ct_limit_info);
2136 if (err)
2137 goto exit_err;
2138
2139 static_branch_enable(&ovs_ct_limit_enabled);
2140
2141 genlmsg_end(reply, ovs_reply_header);
2142 return genlmsg_reply(reply, info);
2143
2144 exit_err:
2145 nlmsg_free(reply);
2146 return err;
2147 }
2148
2149 static int ovs_ct_limit_cmd_del(struct sk_buff *skb, struct genl_info *info)
2150 {
2151 struct nlattr **a = info->attrs;
2152 struct sk_buff *reply;
2153 struct ovs_header *ovs_reply_header;
2154 struct ovs_net *ovs_net = net_generic(sock_net(skb->sk), ovs_net_id);
2155 struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2156 int err;
2157
2158 reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_DEL,
2159 &ovs_reply_header);
2160 if (IS_ERR(reply))
2161 return PTR_ERR(reply);
2162
2163 if (!a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2164 err = -EINVAL;
2165 goto exit_err;
2166 }
2167
2168 err = ovs_ct_limit_del_zone_limit(a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT],
2169 ct_limit_info);
2170 if (err)
2171 goto exit_err;
2172
2173 genlmsg_end(reply, ovs_reply_header);
2174 return genlmsg_reply(reply, info);
2175
2176 exit_err:
2177 nlmsg_free(reply);
2178 return err;
2179 }
2180
2181 static int ovs_ct_limit_cmd_get(struct sk_buff *skb, struct genl_info *info)
2182 {
2183 struct nlattr **a = info->attrs;
2184 struct nlattr *nla_reply;
2185 struct sk_buff *reply;
2186 struct ovs_header *ovs_reply_header;
2187 struct net *net = sock_net(skb->sk);
2188 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2189 struct ovs_ct_limit_info *ct_limit_info = ovs_net->ct_limit_info;
2190 int err;
2191
2192 reply = ovs_ct_limit_cmd_reply_start(info, OVS_CT_LIMIT_CMD_GET,
2193 &ovs_reply_header);
2194 if (IS_ERR(reply))
2195 return PTR_ERR(reply);
2196
2197 nla_reply = nla_nest_start(reply, OVS_CT_LIMIT_ATTR_ZONE_LIMIT);
2198
2199 if (a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT]) {
2200 err = ovs_ct_limit_get_zone_limit(
2201 net, a[OVS_CT_LIMIT_ATTR_ZONE_LIMIT], ct_limit_info,
2202 reply);
2203 if (err)
2204 goto exit_err;
2205 } else {
2206 err = ovs_ct_limit_get_all_zone_limit(net, ct_limit_info,
2207 reply);
2208 if (err)
2209 goto exit_err;
2210 }
2211
2212 nla_nest_end(reply, nla_reply);
2213 genlmsg_end(reply, ovs_reply_header);
2214 return genlmsg_reply(reply, info);
2215
2216 exit_err:
2217 nlmsg_free(reply);
2218 return err;
2219 }
2220
2221 static struct genl_ops ct_limit_genl_ops[] = {
2222 { .cmd = OVS_CT_LIMIT_CMD_SET,
2223 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN
2224 * privilege. */
2225 .policy = ct_limit_policy,
2226 .doit = ovs_ct_limit_cmd_set,
2227 },
2228 { .cmd = OVS_CT_LIMIT_CMD_DEL,
2229 .flags = GENL_ADMIN_PERM, /* Requires CAP_NET_ADMIN
2230 * privilege. */
2231 .policy = ct_limit_policy,
2232 .doit = ovs_ct_limit_cmd_del,
2233 },
2234 { .cmd = OVS_CT_LIMIT_CMD_GET,
2235 .flags = 0, /* OK for unprivileged users. */
2236 .policy = ct_limit_policy,
2237 .doit = ovs_ct_limit_cmd_get,
2238 },
2239 };
2240
2241 static const struct genl_multicast_group ovs_ct_limit_multicast_group = {
2242 .name = OVS_CT_LIMIT_MCGROUP,
2243 };
2244
2245 struct genl_family dp_ct_limit_genl_family __ro_after_init = {
2246 .hdrsize = sizeof(struct ovs_header),
2247 .name = OVS_CT_LIMIT_FAMILY,
2248 .version = OVS_CT_LIMIT_VERSION,
2249 .maxattr = OVS_CT_LIMIT_ATTR_MAX,
2250 .netnsok = true,
2251 .parallel_ops = true,
2252 .ops = ct_limit_genl_ops,
2253 .n_ops = ARRAY_SIZE(ct_limit_genl_ops),
2254 .mcgrps = &ovs_ct_limit_multicast_group,
2255 .n_mcgrps = 1,
2256 .module = THIS_MODULE,
2257 };
2258 #endif
2259
2260 int ovs_ct_init(struct net *net)
2261 {
2262 unsigned int n_bits = sizeof(struct ovs_key_ct_labels) * BITS_PER_BYTE;
2263 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2264
2265 if (nf_connlabels_get(net, n_bits - 1)) {
2266 ovs_net->xt_label = false;
2267 OVS_NLERR(true, "Failed to set connlabel length");
2268 } else {
2269 ovs_net->xt_label = true;
2270 }
2271
2272 #if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2273 return ovs_ct_limit_init(net, ovs_net);
2274 #else
2275 return 0;
2276 #endif
2277 }
2278
2279 void ovs_ct_exit(struct net *net)
2280 {
2281 struct ovs_net *ovs_net = net_generic(net, ovs_net_id);
2282
2283 #if IS_ENABLED(CONFIG_NETFILTER_CONNCOUNT)
2284 ovs_ct_limit_exit(net, ovs_net);
2285 #endif
2286
2287 if (ovs_net->xt_label)
2288 nf_connlabels_put(net);
2289 }
2290
2291 #endif /* CONFIG_NF_CONNTRACK */
openvswitch mirror