Commit | Line | Data |
---|---|---|
7f8a436e JS |
1 | /* |
2 | * Copyright (c) 2015 Nicira, Inc. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of version 2 of the GNU General Public | |
6 | * License as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, but | |
9 | * WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | */ | |
13 | ||
14 | #include <linux/module.h> | |
15 | #include <linux/openvswitch.h> | |
16 | #include <net/ip.h> | |
17 | #include <net/netfilter/nf_conntrack_core.h> | |
cae3a262 | 18 | #include <net/netfilter/nf_conntrack_helper.h> |
c2ac6673 | 19 | #include <net/netfilter/nf_conntrack_labels.h> |
7f8a436e JS |
20 | #include <net/netfilter/nf_conntrack_zones.h> |
21 | #include <net/netfilter/ipv6/nf_defrag_ipv6.h> | |
22 | ||
23 | #include "datapath.h" | |
24 | #include "conntrack.h" | |
25 | #include "flow.h" | |
26 | #include "flow_netlink.h" | |
27 | ||
28 | struct ovs_ct_len_tbl { | |
29 | size_t maxlen; | |
30 | size_t minlen; | |
31 | }; | |
32 | ||
182e3042 JS |
33 | /* Metadata mark for masked write to conntrack mark */ |
34 | struct md_mark { | |
35 | u32 value; | |
36 | u32 mask; | |
37 | }; | |
38 | ||
c2ac6673 JS |
39 | /* Metadata label for masked write to conntrack label. */ |
40 | struct md_label { | |
41 | struct ovs_key_ct_label value; | |
42 | struct ovs_key_ct_label mask; | |
43 | }; | |
44 | ||
7f8a436e JS |
45 | /* Conntrack action context for execution. */ |
46 | struct ovs_conntrack_info { | |
cae3a262 | 47 | struct nf_conntrack_helper *helper; |
7f8a436e JS |
48 | struct nf_conntrack_zone zone; |
49 | struct nf_conn *ct; | |
50 | u32 flags; | |
51 | u16 family; | |
182e3042 | 52 | struct md_mark mark; |
c2ac6673 | 53 | struct md_label label; |
7f8a436e JS |
54 | }; |
55 | ||
56 | static u16 key_to_nfproto(const struct sw_flow_key *key) | |
57 | { | |
58 | switch (ntohs(key->eth.type)) { | |
59 | case ETH_P_IP: | |
60 | return NFPROTO_IPV4; | |
61 | case ETH_P_IPV6: | |
62 | return NFPROTO_IPV6; | |
63 | default: | |
64 | return NFPROTO_UNSPEC; | |
65 | } | |
66 | } | |
67 | ||
68 | /* Map SKB connection state into the values used by flow definition. */ | |
69 | static u8 ovs_ct_get_state(enum ip_conntrack_info ctinfo) | |
70 | { | |
71 | u8 ct_state = OVS_CS_F_TRACKED; | |
72 | ||
73 | switch (ctinfo) { | |
74 | case IP_CT_ESTABLISHED_REPLY: | |
75 | case IP_CT_RELATED_REPLY: | |
76 | case IP_CT_NEW_REPLY: | |
77 | ct_state |= OVS_CS_F_REPLY_DIR; | |
78 | break; | |
79 | default: | |
80 | break; | |
81 | } | |
82 | ||
83 | switch (ctinfo) { | |
84 | case IP_CT_ESTABLISHED: | |
85 | case IP_CT_ESTABLISHED_REPLY: | |
86 | ct_state |= OVS_CS_F_ESTABLISHED; | |
87 | break; | |
88 | case IP_CT_RELATED: | |
89 | case IP_CT_RELATED_REPLY: | |
90 | ct_state |= OVS_CS_F_RELATED; | |
91 | break; | |
92 | case IP_CT_NEW: | |
93 | case IP_CT_NEW_REPLY: | |
94 | ct_state |= OVS_CS_F_NEW; | |
95 | break; | |
96 | default: | |
97 | break; | |
98 | } | |
99 | ||
100 | return ct_state; | |
101 | } | |
102 | ||
0d5cdef8 JS |
103 | static u32 ovs_ct_get_mark(const struct nf_conn *ct) |
104 | { | |
105 | #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) | |
106 | return ct ? ct->mark : 0; | |
107 | #else | |
108 | return 0; | |
109 | #endif | |
110 | } | |
111 | ||
c2ac6673 JS |
112 | static void ovs_ct_get_label(const struct nf_conn *ct, |
113 | struct ovs_key_ct_label *label) | |
114 | { | |
115 | struct nf_conn_labels *cl = ct ? nf_ct_labels_find(ct) : NULL; | |
116 | ||
117 | if (cl) { | |
118 | size_t len = cl->words * sizeof(long); | |
119 | ||
120 | if (len > OVS_CT_LABEL_LEN) | |
121 | len = OVS_CT_LABEL_LEN; | |
122 | else if (len < OVS_CT_LABEL_LEN) | |
123 | memset(label, 0, OVS_CT_LABEL_LEN); | |
124 | memcpy(label, cl->bits, len); | |
125 | } else { | |
126 | memset(label, 0, OVS_CT_LABEL_LEN); | |
127 | } | |
128 | } | |
129 | ||
7f8a436e | 130 | static void __ovs_ct_update_key(struct sw_flow_key *key, u8 state, |
182e3042 JS |
131 | const struct nf_conntrack_zone *zone, |
132 | const struct nf_conn *ct) | |
7f8a436e JS |
133 | { |
134 | key->ct.state = state; | |
135 | key->ct.zone = zone->id; | |
0d5cdef8 | 136 | key->ct.mark = ovs_ct_get_mark(ct); |
c2ac6673 | 137 | ovs_ct_get_label(ct, &key->ct.label); |
7f8a436e JS |
138 | } |
139 | ||
140 | /* Update 'key' based on skb->nfct. If 'post_ct' is true, then OVS has | |
141 | * previously sent the packet to conntrack via the ct action. | |
142 | */ | |
143 | static void ovs_ct_update_key(const struct sk_buff *skb, | |
144 | struct sw_flow_key *key, bool post_ct) | |
145 | { | |
146 | const struct nf_conntrack_zone *zone = &nf_ct_zone_dflt; | |
147 | enum ip_conntrack_info ctinfo; | |
148 | struct nf_conn *ct; | |
149 | u8 state = 0; | |
150 | ||
151 | ct = nf_ct_get(skb, &ctinfo); | |
152 | if (ct) { | |
153 | state = ovs_ct_get_state(ctinfo); | |
154 | if (ct->master) | |
155 | state |= OVS_CS_F_RELATED; | |
156 | zone = nf_ct_zone(ct); | |
157 | } else if (post_ct) { | |
158 | state = OVS_CS_F_TRACKED | OVS_CS_F_INVALID; | |
159 | } | |
182e3042 | 160 | __ovs_ct_update_key(key, state, zone, ct); |
7f8a436e JS |
161 | } |
162 | ||
163 | void ovs_ct_fill_key(const struct sk_buff *skb, struct sw_flow_key *key) | |
164 | { | |
165 | ovs_ct_update_key(skb, key, false); | |
166 | } | |
167 | ||
168 | int ovs_ct_put_key(const struct sw_flow_key *key, struct sk_buff *skb) | |
169 | { | |
170 | if (nla_put_u8(skb, OVS_KEY_ATTR_CT_STATE, key->ct.state)) | |
171 | return -EMSGSIZE; | |
172 | ||
173 | if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) && | |
174 | nla_put_u16(skb, OVS_KEY_ATTR_CT_ZONE, key->ct.zone)) | |
175 | return -EMSGSIZE; | |
176 | ||
182e3042 JS |
177 | if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && |
178 | nla_put_u32(skb, OVS_KEY_ATTR_CT_MARK, key->ct.mark)) | |
179 | return -EMSGSIZE; | |
180 | ||
9723e6ab | 181 | if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) && |
c2ac6673 JS |
182 | nla_put(skb, OVS_KEY_ATTR_CT_LABEL, sizeof(key->ct.label), |
183 | &key->ct.label)) | |
184 | return -EMSGSIZE; | |
185 | ||
182e3042 JS |
186 | return 0; |
187 | } | |
188 | ||
189 | static int ovs_ct_set_mark(struct sk_buff *skb, struct sw_flow_key *key, | |
190 | u32 ct_mark, u32 mask) | |
191 | { | |
0d5cdef8 | 192 | #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) |
182e3042 JS |
193 | enum ip_conntrack_info ctinfo; |
194 | struct nf_conn *ct; | |
195 | u32 new_mark; | |
196 | ||
182e3042 JS |
197 | |
198 | /* The connection could be invalid, in which case set_mark is no-op. */ | |
199 | ct = nf_ct_get(skb, &ctinfo); | |
200 | if (!ct) | |
201 | return 0; | |
202 | ||
203 | new_mark = ct_mark | (ct->mark & ~(mask)); | |
204 | if (ct->mark != new_mark) { | |
205 | ct->mark = new_mark; | |
206 | nf_conntrack_event_cache(IPCT_MARK, ct); | |
207 | key->ct.mark = new_mark; | |
208 | } | |
209 | ||
7f8a436e | 210 | return 0; |
0d5cdef8 JS |
211 | #else |
212 | return -ENOTSUPP; | |
213 | #endif | |
7f8a436e JS |
214 | } |
215 | ||
c2ac6673 JS |
216 | static int ovs_ct_set_label(struct sk_buff *skb, struct sw_flow_key *key, |
217 | const struct ovs_key_ct_label *label, | |
218 | const struct ovs_key_ct_label *mask) | |
219 | { | |
220 | enum ip_conntrack_info ctinfo; | |
221 | struct nf_conn_labels *cl; | |
222 | struct nf_conn *ct; | |
223 | int err; | |
224 | ||
225 | if (!IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS)) | |
226 | return -ENOTSUPP; | |
227 | ||
228 | /* The connection could be invalid, in which case set_label is no-op.*/ | |
229 | ct = nf_ct_get(skb, &ctinfo); | |
230 | if (!ct) | |
231 | return 0; | |
232 | ||
233 | cl = nf_ct_labels_find(ct); | |
234 | if (!cl) { | |
235 | nf_ct_labels_ext_add(ct); | |
236 | cl = nf_ct_labels_find(ct); | |
237 | } | |
238 | if (!cl || cl->words * sizeof(long) < OVS_CT_LABEL_LEN) | |
239 | return -ENOSPC; | |
240 | ||
241 | err = nf_connlabels_replace(ct, (u32 *)label, (u32 *)mask, | |
242 | OVS_CT_LABEL_LEN / sizeof(u32)); | |
243 | if (err) | |
244 | return err; | |
245 | ||
246 | ovs_ct_get_label(ct, &key->ct.label); | |
247 | return 0; | |
248 | } | |
249 | ||
cae3a262 JS |
250 | /* 'skb' should already be pulled to nh_ofs. */ |
251 | static int ovs_ct_helper(struct sk_buff *skb, u16 proto) | |
252 | { | |
253 | const struct nf_conntrack_helper *helper; | |
254 | const struct nf_conn_help *help; | |
255 | enum ip_conntrack_info ctinfo; | |
256 | unsigned int protoff; | |
257 | struct nf_conn *ct; | |
258 | ||
259 | ct = nf_ct_get(skb, &ctinfo); | |
260 | if (!ct || ctinfo == IP_CT_RELATED_REPLY) | |
261 | return NF_ACCEPT; | |
262 | ||
263 | help = nfct_help(ct); | |
264 | if (!help) | |
265 | return NF_ACCEPT; | |
266 | ||
267 | helper = rcu_dereference(help->helper); | |
268 | if (!helper) | |
269 | return NF_ACCEPT; | |
270 | ||
271 | switch (proto) { | |
272 | case NFPROTO_IPV4: | |
273 | protoff = ip_hdrlen(skb); | |
274 | break; | |
275 | case NFPROTO_IPV6: { | |
276 | u8 nexthdr = ipv6_hdr(skb)->nexthdr; | |
277 | __be16 frag_off; | |
278 | ||
279 | protoff = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), | |
280 | &nexthdr, &frag_off); | |
281 | if (protoff < 0 || (frag_off & htons(~0x7)) != 0) { | |
282 | pr_debug("proto header not found\n"); | |
283 | return NF_ACCEPT; | |
284 | } | |
285 | break; | |
286 | } | |
287 | default: | |
288 | WARN_ONCE(1, "helper invoked on non-IP family!"); | |
289 | return NF_DROP; | |
290 | } | |
291 | ||
292 | return helper->help(skb, protoff, ct, ctinfo); | |
293 | } | |
294 | ||
7f8a436e JS |
295 | static int handle_fragments(struct net *net, struct sw_flow_key *key, |
296 | u16 zone, struct sk_buff *skb) | |
297 | { | |
298 | struct ovs_skb_cb ovs_cb = *OVS_CB(skb); | |
299 | ||
300 | if (key->eth.type == htons(ETH_P_IP)) { | |
301 | enum ip_defrag_users user = IP_DEFRAG_CONNTRACK_IN + zone; | |
302 | int err; | |
303 | ||
304 | memset(IPCB(skb), 0, sizeof(struct inet_skb_parm)); | |
305 | err = ip_defrag(skb, user); | |
306 | if (err) | |
307 | return err; | |
308 | ||
309 | ovs_cb.mru = IPCB(skb)->frag_max_size; | |
310 | } else if (key->eth.type == htons(ETH_P_IPV6)) { | |
311 | #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6) | |
312 | enum ip6_defrag_users user = IP6_DEFRAG_CONNTRACK_IN + zone; | |
313 | struct sk_buff *reasm; | |
314 | ||
315 | memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm)); | |
316 | reasm = nf_ct_frag6_gather(skb, user); | |
317 | if (!reasm) | |
318 | return -EINPROGRESS; | |
319 | ||
320 | if (skb == reasm) | |
321 | return -EINVAL; | |
322 | ||
323 | key->ip.proto = ipv6_hdr(reasm)->nexthdr; | |
324 | skb_morph(skb, reasm); | |
325 | consume_skb(reasm); | |
326 | ovs_cb.mru = IP6CB(skb)->frag_max_size; | |
327 | #else | |
328 | return -EPFNOSUPPORT; | |
329 | #endif | |
330 | } else { | |
331 | return -EPFNOSUPPORT; | |
332 | } | |
333 | ||
334 | key->ip.frag = OVS_FRAG_TYPE_NONE; | |
335 | skb_clear_hash(skb); | |
336 | skb->ignore_df = 1; | |
337 | *OVS_CB(skb) = ovs_cb; | |
338 | ||
339 | return 0; | |
340 | } | |
341 | ||
342 | static struct nf_conntrack_expect * | |
343 | ovs_ct_expect_find(struct net *net, const struct nf_conntrack_zone *zone, | |
344 | u16 proto, const struct sk_buff *skb) | |
345 | { | |
346 | struct nf_conntrack_tuple tuple; | |
347 | ||
348 | if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), proto, &tuple)) | |
349 | return NULL; | |
350 | return __nf_ct_expect_find(net, zone, &tuple); | |
351 | } | |
352 | ||
353 | /* Determine whether skb->nfct is equal to the result of conntrack lookup. */ | |
354 | static bool skb_nfct_cached(const struct net *net, const struct sk_buff *skb, | |
355 | const struct ovs_conntrack_info *info) | |
356 | { | |
357 | enum ip_conntrack_info ctinfo; | |
358 | struct nf_conn *ct; | |
359 | ||
360 | ct = nf_ct_get(skb, &ctinfo); | |
361 | if (!ct) | |
362 | return false; | |
363 | if (!net_eq(net, read_pnet(&ct->ct_net))) | |
364 | return false; | |
365 | if (!nf_ct_zone_equal_any(info->ct, nf_ct_zone(ct))) | |
366 | return false; | |
cae3a262 JS |
367 | if (info->helper) { |
368 | struct nf_conn_help *help; | |
369 | ||
370 | help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER); | |
371 | if (help && rcu_access_pointer(help->helper) != info->helper) | |
372 | return false; | |
373 | } | |
7f8a436e JS |
374 | |
375 | return true; | |
376 | } | |
377 | ||
378 | static int __ovs_ct_lookup(struct net *net, const struct sw_flow_key *key, | |
379 | const struct ovs_conntrack_info *info, | |
380 | struct sk_buff *skb) | |
381 | { | |
382 | /* If we are recirculating packets to match on conntrack fields and | |
383 | * committing with a separate conntrack action, then we don't need to | |
384 | * actually run the packet through conntrack twice unless it's for a | |
385 | * different zone. | |
386 | */ | |
387 | if (!skb_nfct_cached(net, skb, info)) { | |
388 | struct nf_conn *tmpl = info->ct; | |
389 | ||
390 | /* Associate skb with specified zone. */ | |
391 | if (tmpl) { | |
392 | if (skb->nfct) | |
393 | nf_conntrack_put(skb->nfct); | |
394 | nf_conntrack_get(&tmpl->ct_general); | |
395 | skb->nfct = &tmpl->ct_general; | |
396 | skb->nfctinfo = IP_CT_NEW; | |
397 | } | |
398 | ||
399 | if (nf_conntrack_in(net, info->family, NF_INET_PRE_ROUTING, | |
400 | skb) != NF_ACCEPT) | |
401 | return -ENOENT; | |
cae3a262 JS |
402 | |
403 | if (ovs_ct_helper(skb, info->family) != NF_ACCEPT) { | |
404 | WARN_ONCE(1, "helper rejected packet"); | |
405 | return -EINVAL; | |
406 | } | |
7f8a436e JS |
407 | } |
408 | ||
409 | return 0; | |
410 | } | |
411 | ||
412 | /* Lookup connection and read fields into key. */ | |
413 | static int ovs_ct_lookup(struct net *net, struct sw_flow_key *key, | |
414 | const struct ovs_conntrack_info *info, | |
415 | struct sk_buff *skb) | |
416 | { | |
417 | struct nf_conntrack_expect *exp; | |
418 | ||
419 | exp = ovs_ct_expect_find(net, &info->zone, info->family, skb); | |
420 | if (exp) { | |
421 | u8 state; | |
422 | ||
423 | state = OVS_CS_F_TRACKED | OVS_CS_F_NEW | OVS_CS_F_RELATED; | |
182e3042 | 424 | __ovs_ct_update_key(key, state, &info->zone, exp->master); |
7f8a436e JS |
425 | } else { |
426 | int err; | |
427 | ||
428 | err = __ovs_ct_lookup(net, key, info, skb); | |
429 | if (err) | |
430 | return err; | |
431 | ||
432 | ovs_ct_update_key(skb, key, true); | |
433 | } | |
434 | ||
435 | return 0; | |
436 | } | |
437 | ||
438 | /* Lookup connection and confirm if unconfirmed. */ | |
439 | static int ovs_ct_commit(struct net *net, struct sw_flow_key *key, | |
440 | const struct ovs_conntrack_info *info, | |
441 | struct sk_buff *skb) | |
442 | { | |
443 | u8 state; | |
444 | int err; | |
445 | ||
446 | state = key->ct.state; | |
447 | if (key->ct.zone == info->zone.id && | |
448 | ((state & OVS_CS_F_TRACKED) && !(state & OVS_CS_F_NEW))) { | |
449 | /* Previous lookup has shown that this connection is already | |
450 | * tracked and committed. Skip committing. | |
451 | */ | |
452 | return 0; | |
453 | } | |
454 | ||
455 | err = __ovs_ct_lookup(net, key, info, skb); | |
456 | if (err) | |
457 | return err; | |
458 | if (nf_conntrack_confirm(skb) != NF_ACCEPT) | |
459 | return -EINVAL; | |
460 | ||
461 | ovs_ct_update_key(skb, key, true); | |
462 | ||
463 | return 0; | |
464 | } | |
465 | ||
c2ac6673 JS |
466 | static bool label_nonzero(const struct ovs_key_ct_label *label) |
467 | { | |
468 | size_t i; | |
469 | ||
470 | for (i = 0; i < sizeof(*label); i++) | |
471 | if (label->ct_label[i]) | |
472 | return true; | |
473 | ||
474 | return false; | |
475 | } | |
476 | ||
7f8a436e JS |
477 | int ovs_ct_execute(struct net *net, struct sk_buff *skb, |
478 | struct sw_flow_key *key, | |
479 | const struct ovs_conntrack_info *info) | |
480 | { | |
481 | int nh_ofs; | |
482 | int err; | |
483 | ||
484 | /* The conntrack module expects to be working at L3. */ | |
485 | nh_ofs = skb_network_offset(skb); | |
486 | skb_pull(skb, nh_ofs); | |
487 | ||
488 | if (key->ip.frag != OVS_FRAG_TYPE_NONE) { | |
489 | err = handle_fragments(net, key, info->zone.id, skb); | |
490 | if (err) | |
491 | return err; | |
492 | } | |
493 | ||
494 | if (info->flags & OVS_CT_F_COMMIT) | |
495 | err = ovs_ct_commit(net, key, info, skb); | |
496 | else | |
497 | err = ovs_ct_lookup(net, key, info, skb); | |
182e3042 JS |
498 | if (err) |
499 | goto err; | |
7f8a436e | 500 | |
c2ac6673 | 501 | if (info->mark.mask) { |
182e3042 JS |
502 | err = ovs_ct_set_mark(skb, key, info->mark.value, |
503 | info->mark.mask); | |
c2ac6673 JS |
504 | if (err) |
505 | goto err; | |
506 | } | |
507 | if (label_nonzero(&info->label.mask)) | |
508 | err = ovs_ct_set_label(skb, key, &info->label.value, | |
509 | &info->label.mask); | |
182e3042 | 510 | err: |
7f8a436e JS |
511 | skb_push(skb, nh_ofs); |
512 | return err; | |
513 | } | |
514 | ||
cae3a262 JS |
515 | static int ovs_ct_add_helper(struct ovs_conntrack_info *info, const char *name, |
516 | const struct sw_flow_key *key, bool log) | |
517 | { | |
518 | struct nf_conntrack_helper *helper; | |
519 | struct nf_conn_help *help; | |
520 | ||
521 | helper = nf_conntrack_helper_try_module_get(name, info->family, | |
522 | key->ip.proto); | |
523 | if (!helper) { | |
524 | OVS_NLERR(log, "Unknown helper \"%s\"", name); | |
525 | return -EINVAL; | |
526 | } | |
527 | ||
528 | help = nf_ct_helper_ext_add(info->ct, helper, GFP_KERNEL); | |
529 | if (!help) { | |
530 | module_put(helper->me); | |
531 | return -ENOMEM; | |
532 | } | |
533 | ||
534 | rcu_assign_pointer(help->helper, helper); | |
535 | info->helper = helper; | |
536 | return 0; | |
537 | } | |
538 | ||
7f8a436e JS |
539 | static const struct ovs_ct_len_tbl ovs_ct_attr_lens[OVS_CT_ATTR_MAX + 1] = { |
540 | [OVS_CT_ATTR_FLAGS] = { .minlen = sizeof(u32), | |
541 | .maxlen = sizeof(u32) }, | |
542 | [OVS_CT_ATTR_ZONE] = { .minlen = sizeof(u16), | |
543 | .maxlen = sizeof(u16) }, | |
182e3042 JS |
544 | [OVS_CT_ATTR_MARK] = { .minlen = sizeof(struct md_mark), |
545 | .maxlen = sizeof(struct md_mark) }, | |
c2ac6673 JS |
546 | [OVS_CT_ATTR_LABEL] = { .minlen = sizeof(struct md_label), |
547 | .maxlen = sizeof(struct md_label) }, | |
cae3a262 JS |
548 | [OVS_CT_ATTR_HELPER] = { .minlen = 1, |
549 | .maxlen = NF_CT_HELPER_NAME_LEN } | |
7f8a436e JS |
550 | }; |
551 | ||
552 | static int parse_ct(const struct nlattr *attr, struct ovs_conntrack_info *info, | |
cae3a262 | 553 | const char **helper, bool log) |
7f8a436e JS |
554 | { |
555 | struct nlattr *a; | |
556 | int rem; | |
557 | ||
558 | nla_for_each_nested(a, attr, rem) { | |
559 | int type = nla_type(a); | |
560 | int maxlen = ovs_ct_attr_lens[type].maxlen; | |
561 | int minlen = ovs_ct_attr_lens[type].minlen; | |
562 | ||
563 | if (type > OVS_CT_ATTR_MAX) { | |
564 | OVS_NLERR(log, | |
565 | "Unknown conntrack attr (type=%d, max=%d)", | |
566 | type, OVS_CT_ATTR_MAX); | |
567 | return -EINVAL; | |
568 | } | |
569 | if (nla_len(a) < minlen || nla_len(a) > maxlen) { | |
570 | OVS_NLERR(log, | |
571 | "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)", | |
572 | type, nla_len(a), maxlen); | |
573 | return -EINVAL; | |
574 | } | |
575 | ||
576 | switch (type) { | |
577 | case OVS_CT_ATTR_FLAGS: | |
578 | info->flags = nla_get_u32(a); | |
579 | break; | |
580 | #ifdef CONFIG_NF_CONNTRACK_ZONES | |
581 | case OVS_CT_ATTR_ZONE: | |
582 | info->zone.id = nla_get_u16(a); | |
583 | break; | |
182e3042 JS |
584 | #endif |
585 | #ifdef CONFIG_NF_CONNTRACK_MARK | |
586 | case OVS_CT_ATTR_MARK: { | |
587 | struct md_mark *mark = nla_data(a); | |
588 | ||
589 | info->mark = *mark; | |
590 | break; | |
591 | } | |
c2ac6673 JS |
592 | #endif |
593 | #ifdef CONFIG_NF_CONNTRACK_LABELS | |
594 | case OVS_CT_ATTR_LABEL: { | |
595 | struct md_label *label = nla_data(a); | |
596 | ||
597 | info->label = *label; | |
598 | break; | |
599 | } | |
7f8a436e | 600 | #endif |
cae3a262 JS |
601 | case OVS_CT_ATTR_HELPER: |
602 | *helper = nla_data(a); | |
603 | if (!memchr(*helper, '\0', nla_len(a))) { | |
604 | OVS_NLERR(log, "Invalid conntrack helper"); | |
605 | return -EINVAL; | |
606 | } | |
607 | break; | |
7f8a436e JS |
608 | default: |
609 | OVS_NLERR(log, "Unknown conntrack attr (%d)", | |
610 | type); | |
611 | return -EINVAL; | |
612 | } | |
613 | } | |
614 | ||
615 | if (rem > 0) { | |
616 | OVS_NLERR(log, "Conntrack attr has %d unknown bytes", rem); | |
617 | return -EINVAL; | |
618 | } | |
619 | ||
620 | return 0; | |
621 | } | |
622 | ||
c2ac6673 | 623 | bool ovs_ct_verify(struct net *net, enum ovs_key_attr attr) |
7f8a436e JS |
624 | { |
625 | if (attr == OVS_KEY_ATTR_CT_STATE) | |
626 | return true; | |
627 | if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) && | |
628 | attr == OVS_KEY_ATTR_CT_ZONE) | |
629 | return true; | |
182e3042 JS |
630 | if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && |
631 | attr == OVS_KEY_ATTR_CT_MARK) | |
632 | return true; | |
c2ac6673 JS |
633 | if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) && |
634 | attr == OVS_KEY_ATTR_CT_LABEL) { | |
635 | struct ovs_net *ovs_net = net_generic(net, ovs_net_id); | |
636 | ||
637 | return ovs_net->xt_label; | |
638 | } | |
7f8a436e JS |
639 | |
640 | return false; | |
641 | } | |
642 | ||
643 | int ovs_ct_copy_action(struct net *net, const struct nlattr *attr, | |
644 | const struct sw_flow_key *key, | |
645 | struct sw_flow_actions **sfa, bool log) | |
646 | { | |
647 | struct ovs_conntrack_info ct_info; | |
cae3a262 | 648 | const char *helper = NULL; |
7f8a436e JS |
649 | u16 family; |
650 | int err; | |
651 | ||
652 | family = key_to_nfproto(key); | |
653 | if (family == NFPROTO_UNSPEC) { | |
654 | OVS_NLERR(log, "ct family unspecified"); | |
655 | return -EINVAL; | |
656 | } | |
657 | ||
658 | memset(&ct_info, 0, sizeof(ct_info)); | |
659 | ct_info.family = family; | |
660 | ||
661 | nf_ct_zone_init(&ct_info.zone, NF_CT_DEFAULT_ZONE_ID, | |
662 | NF_CT_DEFAULT_ZONE_DIR, 0); | |
663 | ||
cae3a262 | 664 | err = parse_ct(attr, &ct_info, &helper, log); |
7f8a436e JS |
665 | if (err) |
666 | return err; | |
667 | ||
668 | /* Set up template for tracking connections in specific zones. */ | |
669 | ct_info.ct = nf_ct_tmpl_alloc(net, &ct_info.zone, GFP_KERNEL); | |
670 | if (!ct_info.ct) { | |
671 | OVS_NLERR(log, "Failed to allocate conntrack template"); | |
672 | return -ENOMEM; | |
673 | } | |
cae3a262 JS |
674 | if (helper) { |
675 | err = ovs_ct_add_helper(&ct_info, helper, key, log); | |
676 | if (err) | |
677 | goto err_free_ct; | |
678 | } | |
7f8a436e JS |
679 | |
680 | err = ovs_nla_add_action(sfa, OVS_ACTION_ATTR_CT, &ct_info, | |
681 | sizeof(ct_info), log); | |
682 | if (err) | |
683 | goto err_free_ct; | |
684 | ||
685 | __set_bit(IPS_CONFIRMED_BIT, &ct_info.ct->status); | |
686 | nf_conntrack_get(&ct_info.ct->ct_general); | |
687 | return 0; | |
688 | err_free_ct: | |
689 | nf_conntrack_free(ct_info.ct); | |
690 | return err; | |
691 | } | |
692 | ||
693 | int ovs_ct_action_to_attr(const struct ovs_conntrack_info *ct_info, | |
694 | struct sk_buff *skb) | |
695 | { | |
696 | struct nlattr *start; | |
697 | ||
698 | start = nla_nest_start(skb, OVS_ACTION_ATTR_CT); | |
699 | if (!start) | |
700 | return -EMSGSIZE; | |
701 | ||
702 | if (nla_put_u32(skb, OVS_CT_ATTR_FLAGS, ct_info->flags)) | |
703 | return -EMSGSIZE; | |
704 | if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) && | |
705 | nla_put_u16(skb, OVS_CT_ATTR_ZONE, ct_info->zone.id)) | |
706 | return -EMSGSIZE; | |
182e3042 JS |
707 | if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && |
708 | nla_put(skb, OVS_CT_ATTR_MARK, sizeof(ct_info->mark), | |
709 | &ct_info->mark)) | |
710 | return -EMSGSIZE; | |
c2ac6673 JS |
711 | if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) && |
712 | nla_put(skb, OVS_CT_ATTR_LABEL, sizeof(ct_info->label), | |
713 | &ct_info->label)) | |
714 | return -EMSGSIZE; | |
cae3a262 JS |
715 | if (ct_info->helper) { |
716 | if (nla_put_string(skb, OVS_CT_ATTR_HELPER, | |
717 | ct_info->helper->name)) | |
718 | return -EMSGSIZE; | |
719 | } | |
7f8a436e JS |
720 | |
721 | nla_nest_end(skb, start); | |
722 | ||
723 | return 0; | |
724 | } | |
725 | ||
726 | void ovs_ct_free_action(const struct nlattr *a) | |
727 | { | |
728 | struct ovs_conntrack_info *ct_info = nla_data(a); | |
729 | ||
cae3a262 JS |
730 | if (ct_info->helper) |
731 | module_put(ct_info->helper->me); | |
7f8a436e JS |
732 | if (ct_info->ct) |
733 | nf_ct_put(ct_info->ct); | |
734 | } | |
c2ac6673 JS |
735 | |
736 | void ovs_ct_init(struct net *net) | |
737 | { | |
738 | unsigned int n_bits = sizeof(struct ovs_key_ct_label) * BITS_PER_BYTE; | |
739 | struct ovs_net *ovs_net = net_generic(net, ovs_net_id); | |
740 | ||
741 | if (nf_connlabels_get(net, n_bits)) { | |
742 | ovs_net->xt_label = false; | |
743 | OVS_NLERR(true, "Failed to set connlabel length"); | |
744 | } else { | |
745 | ovs_net->xt_label = true; | |
746 | } | |
747 | } | |
748 | ||
749 | void ovs_ct_exit(struct net *net) | |
750 | { | |
751 | struct ovs_net *ovs_net = net_generic(net, ovs_net_id); | |
752 | ||
753 | if (ovs_net->xt_label) | |
754 | nf_connlabels_put(net); | |
755 | } |