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 | 39 | /* Metadata label for masked write to conntrack label. */ |
33db4125 JS |
40 | struct md_labels { |
41 | struct ovs_key_ct_labels value; | |
42 | struct ovs_key_ct_labels mask; | |
c2ac6673 JS |
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; | |
ab38a7b5 | 50 | u8 commit : 1; |
7f8a436e | 51 | u16 family; |
182e3042 | 52 | struct md_mark mark; |
33db4125 | 53 | struct md_labels labels; |
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 | ||
33db4125 JS |
112 | static void ovs_ct_get_labels(const struct nf_conn *ct, |
113 | struct ovs_key_ct_labels *labels) | |
c2ac6673 JS |
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 | ||
33db4125 JS |
120 | if (len > OVS_CT_LABELS_LEN) |
121 | len = OVS_CT_LABELS_LEN; | |
122 | else if (len < OVS_CT_LABELS_LEN) | |
123 | memset(labels, 0, OVS_CT_LABELS_LEN); | |
124 | memcpy(labels, cl->bits, len); | |
c2ac6673 | 125 | } else { |
33db4125 | 126 | memset(labels, 0, OVS_CT_LABELS_LEN); |
c2ac6673 JS |
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); |
33db4125 | 137 | ovs_ct_get_labels(ct, &key->ct.labels); |
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 | { | |
fbccce59 | 170 | if (nla_put_u32(skb, OVS_KEY_ATTR_CT_STATE, key->ct.state)) |
7f8a436e JS |
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) && |
33db4125 JS |
182 | nla_put(skb, OVS_KEY_ATTR_CT_LABELS, sizeof(key->ct.labels), |
183 | &key->ct.labels)) | |
c2ac6673 JS |
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 | ||
33db4125 JS |
216 | static int ovs_ct_set_labels(struct sk_buff *skb, struct sw_flow_key *key, |
217 | const struct ovs_key_ct_labels *labels, | |
218 | const struct ovs_key_ct_labels *mask) | |
c2ac6673 JS |
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 | } | |
33db4125 | 238 | if (!cl || cl->words * sizeof(long) < OVS_CT_LABELS_LEN) |
c2ac6673 JS |
239 | return -ENOSPC; |
240 | ||
33db4125 JS |
241 | err = nf_connlabels_replace(ct, (u32 *)labels, (u32 *)mask, |
242 | OVS_CT_LABELS_LEN / sizeof(u32)); | |
c2ac6673 JS |
243 | if (err) |
244 | return err; | |
245 | ||
33db4125 | 246 | ovs_ct_get_labels(ct, &key->ct.labels); |
c2ac6673 JS |
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; | |
cc570605 | 278 | int ofs; |
cae3a262 | 279 | |
cc570605 JS |
280 | ofs = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr, |
281 | &frag_off); | |
282 | if (ofs < 0 || (frag_off & htons(~0x7)) != 0) { | |
cae3a262 JS |
283 | pr_debug("proto header not found\n"); |
284 | return NF_ACCEPT; | |
285 | } | |
cc570605 | 286 | protoff = ofs; |
cae3a262 JS |
287 | break; |
288 | } | |
289 | default: | |
290 | WARN_ONCE(1, "helper invoked on non-IP family!"); | |
291 | return NF_DROP; | |
292 | } | |
293 | ||
294 | return helper->help(skb, protoff, ct, ctinfo); | |
295 | } | |
296 | ||
7f8a436e JS |
297 | static int handle_fragments(struct net *net, struct sw_flow_key *key, |
298 | u16 zone, struct sk_buff *skb) | |
299 | { | |
300 | struct ovs_skb_cb ovs_cb = *OVS_CB(skb); | |
301 | ||
302 | if (key->eth.type == htons(ETH_P_IP)) { | |
303 | enum ip_defrag_users user = IP_DEFRAG_CONNTRACK_IN + zone; | |
304 | int err; | |
305 | ||
306 | memset(IPCB(skb), 0, sizeof(struct inet_skb_parm)); | |
307 | err = ip_defrag(skb, user); | |
308 | if (err) | |
309 | return err; | |
310 | ||
311 | ovs_cb.mru = IPCB(skb)->frag_max_size; | |
312 | } else if (key->eth.type == htons(ETH_P_IPV6)) { | |
313 | #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6) | |
314 | enum ip6_defrag_users user = IP6_DEFRAG_CONNTRACK_IN + zone; | |
315 | struct sk_buff *reasm; | |
316 | ||
317 | memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm)); | |
318 | reasm = nf_ct_frag6_gather(skb, user); | |
319 | if (!reasm) | |
320 | return -EINPROGRESS; | |
321 | ||
322 | if (skb == reasm) | |
323 | return -EINVAL; | |
324 | ||
325 | key->ip.proto = ipv6_hdr(reasm)->nexthdr; | |
326 | skb_morph(skb, reasm); | |
327 | consume_skb(reasm); | |
328 | ovs_cb.mru = IP6CB(skb)->frag_max_size; | |
329 | #else | |
330 | return -EPFNOSUPPORT; | |
331 | #endif | |
332 | } else { | |
333 | return -EPFNOSUPPORT; | |
334 | } | |
335 | ||
336 | key->ip.frag = OVS_FRAG_TYPE_NONE; | |
337 | skb_clear_hash(skb); | |
338 | skb->ignore_df = 1; | |
339 | *OVS_CB(skb) = ovs_cb; | |
340 | ||
341 | return 0; | |
342 | } | |
343 | ||
344 | static struct nf_conntrack_expect * | |
345 | ovs_ct_expect_find(struct net *net, const struct nf_conntrack_zone *zone, | |
346 | u16 proto, const struct sk_buff *skb) | |
347 | { | |
348 | struct nf_conntrack_tuple tuple; | |
349 | ||
350 | if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), proto, &tuple)) | |
351 | return NULL; | |
352 | return __nf_ct_expect_find(net, zone, &tuple); | |
353 | } | |
354 | ||
355 | /* Determine whether skb->nfct is equal to the result of conntrack lookup. */ | |
356 | static bool skb_nfct_cached(const struct net *net, const struct sk_buff *skb, | |
357 | const struct ovs_conntrack_info *info) | |
358 | { | |
359 | enum ip_conntrack_info ctinfo; | |
360 | struct nf_conn *ct; | |
361 | ||
362 | ct = nf_ct_get(skb, &ctinfo); | |
363 | if (!ct) | |
364 | return false; | |
365 | if (!net_eq(net, read_pnet(&ct->ct_net))) | |
366 | return false; | |
367 | if (!nf_ct_zone_equal_any(info->ct, nf_ct_zone(ct))) | |
368 | return false; | |
cae3a262 JS |
369 | if (info->helper) { |
370 | struct nf_conn_help *help; | |
371 | ||
372 | help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER); | |
373 | if (help && rcu_access_pointer(help->helper) != info->helper) | |
374 | return false; | |
375 | } | |
7f8a436e JS |
376 | |
377 | return true; | |
378 | } | |
379 | ||
380 | static int __ovs_ct_lookup(struct net *net, const struct sw_flow_key *key, | |
381 | const struct ovs_conntrack_info *info, | |
382 | struct sk_buff *skb) | |
383 | { | |
384 | /* If we are recirculating packets to match on conntrack fields and | |
385 | * committing with a separate conntrack action, then we don't need to | |
386 | * actually run the packet through conntrack twice unless it's for a | |
387 | * different zone. | |
388 | */ | |
389 | if (!skb_nfct_cached(net, skb, info)) { | |
390 | struct nf_conn *tmpl = info->ct; | |
391 | ||
392 | /* Associate skb with specified zone. */ | |
393 | if (tmpl) { | |
394 | if (skb->nfct) | |
395 | nf_conntrack_put(skb->nfct); | |
396 | nf_conntrack_get(&tmpl->ct_general); | |
397 | skb->nfct = &tmpl->ct_general; | |
398 | skb->nfctinfo = IP_CT_NEW; | |
399 | } | |
400 | ||
401 | if (nf_conntrack_in(net, info->family, NF_INET_PRE_ROUTING, | |
402 | skb) != NF_ACCEPT) | |
403 | return -ENOENT; | |
cae3a262 JS |
404 | |
405 | if (ovs_ct_helper(skb, info->family) != NF_ACCEPT) { | |
406 | WARN_ONCE(1, "helper rejected packet"); | |
407 | return -EINVAL; | |
408 | } | |
7f8a436e JS |
409 | } |
410 | ||
411 | return 0; | |
412 | } | |
413 | ||
414 | /* Lookup connection and read fields into key. */ | |
415 | static int ovs_ct_lookup(struct net *net, struct sw_flow_key *key, | |
416 | const struct ovs_conntrack_info *info, | |
417 | struct sk_buff *skb) | |
418 | { | |
419 | struct nf_conntrack_expect *exp; | |
420 | ||
421 | exp = ovs_ct_expect_find(net, &info->zone, info->family, skb); | |
422 | if (exp) { | |
423 | u8 state; | |
424 | ||
425 | state = OVS_CS_F_TRACKED | OVS_CS_F_NEW | OVS_CS_F_RELATED; | |
182e3042 | 426 | __ovs_ct_update_key(key, state, &info->zone, exp->master); |
7f8a436e JS |
427 | } else { |
428 | int err; | |
429 | ||
430 | err = __ovs_ct_lookup(net, key, info, skb); | |
431 | if (err) | |
432 | return err; | |
433 | ||
434 | ovs_ct_update_key(skb, key, true); | |
435 | } | |
436 | ||
437 | return 0; | |
438 | } | |
439 | ||
440 | /* Lookup connection and confirm if unconfirmed. */ | |
441 | static int ovs_ct_commit(struct net *net, struct sw_flow_key *key, | |
442 | const struct ovs_conntrack_info *info, | |
443 | struct sk_buff *skb) | |
444 | { | |
445 | u8 state; | |
446 | int err; | |
447 | ||
448 | state = key->ct.state; | |
449 | if (key->ct.zone == info->zone.id && | |
450 | ((state & OVS_CS_F_TRACKED) && !(state & OVS_CS_F_NEW))) { | |
451 | /* Previous lookup has shown that this connection is already | |
452 | * tracked and committed. Skip committing. | |
453 | */ | |
454 | return 0; | |
455 | } | |
456 | ||
457 | err = __ovs_ct_lookup(net, key, info, skb); | |
458 | if (err) | |
459 | return err; | |
460 | if (nf_conntrack_confirm(skb) != NF_ACCEPT) | |
461 | return -EINVAL; | |
462 | ||
463 | ovs_ct_update_key(skb, key, true); | |
464 | ||
465 | return 0; | |
466 | } | |
467 | ||
33db4125 | 468 | static bool labels_nonzero(const struct ovs_key_ct_labels *labels) |
c2ac6673 JS |
469 | { |
470 | size_t i; | |
471 | ||
33db4125 JS |
472 | for (i = 0; i < sizeof(*labels); i++) |
473 | if (labels->ct_labels[i]) | |
c2ac6673 JS |
474 | return true; |
475 | ||
476 | return false; | |
477 | } | |
478 | ||
7f8a436e JS |
479 | int ovs_ct_execute(struct net *net, struct sk_buff *skb, |
480 | struct sw_flow_key *key, | |
481 | const struct ovs_conntrack_info *info) | |
482 | { | |
483 | int nh_ofs; | |
484 | int err; | |
485 | ||
486 | /* The conntrack module expects to be working at L3. */ | |
487 | nh_ofs = skb_network_offset(skb); | |
488 | skb_pull(skb, nh_ofs); | |
489 | ||
490 | if (key->ip.frag != OVS_FRAG_TYPE_NONE) { | |
491 | err = handle_fragments(net, key, info->zone.id, skb); | |
492 | if (err) | |
493 | return err; | |
494 | } | |
495 | ||
ab38a7b5 | 496 | if (info->commit) |
7f8a436e JS |
497 | err = ovs_ct_commit(net, key, info, skb); |
498 | else | |
499 | err = ovs_ct_lookup(net, key, info, skb); | |
182e3042 JS |
500 | if (err) |
501 | goto err; | |
7f8a436e | 502 | |
c2ac6673 | 503 | if (info->mark.mask) { |
182e3042 JS |
504 | err = ovs_ct_set_mark(skb, key, info->mark.value, |
505 | info->mark.mask); | |
c2ac6673 JS |
506 | if (err) |
507 | goto err; | |
508 | } | |
33db4125 JS |
509 | if (labels_nonzero(&info->labels.mask)) |
510 | err = ovs_ct_set_labels(skb, key, &info->labels.value, | |
511 | &info->labels.mask); | |
182e3042 | 512 | err: |
7f8a436e JS |
513 | skb_push(skb, nh_ofs); |
514 | return err; | |
515 | } | |
516 | ||
cae3a262 JS |
517 | static int ovs_ct_add_helper(struct ovs_conntrack_info *info, const char *name, |
518 | const struct sw_flow_key *key, bool log) | |
519 | { | |
520 | struct nf_conntrack_helper *helper; | |
521 | struct nf_conn_help *help; | |
522 | ||
523 | helper = nf_conntrack_helper_try_module_get(name, info->family, | |
524 | key->ip.proto); | |
525 | if (!helper) { | |
526 | OVS_NLERR(log, "Unknown helper \"%s\"", name); | |
527 | return -EINVAL; | |
528 | } | |
529 | ||
530 | help = nf_ct_helper_ext_add(info->ct, helper, GFP_KERNEL); | |
531 | if (!help) { | |
532 | module_put(helper->me); | |
533 | return -ENOMEM; | |
534 | } | |
535 | ||
536 | rcu_assign_pointer(help->helper, helper); | |
537 | info->helper = helper; | |
538 | return 0; | |
539 | } | |
540 | ||
7f8a436e | 541 | static const struct ovs_ct_len_tbl ovs_ct_attr_lens[OVS_CT_ATTR_MAX + 1] = { |
ab38a7b5 | 542 | [OVS_CT_ATTR_COMMIT] = { .minlen = 0, .maxlen = 0 }, |
7f8a436e JS |
543 | [OVS_CT_ATTR_ZONE] = { .minlen = sizeof(u16), |
544 | .maxlen = sizeof(u16) }, | |
182e3042 JS |
545 | [OVS_CT_ATTR_MARK] = { .minlen = sizeof(struct md_mark), |
546 | .maxlen = sizeof(struct md_mark) }, | |
33db4125 JS |
547 | [OVS_CT_ATTR_LABELS] = { .minlen = sizeof(struct md_labels), |
548 | .maxlen = sizeof(struct md_labels) }, | |
cae3a262 JS |
549 | [OVS_CT_ATTR_HELPER] = { .minlen = 1, |
550 | .maxlen = NF_CT_HELPER_NAME_LEN } | |
7f8a436e JS |
551 | }; |
552 | ||
553 | static int parse_ct(const struct nlattr *attr, struct ovs_conntrack_info *info, | |
cae3a262 | 554 | const char **helper, bool log) |
7f8a436e JS |
555 | { |
556 | struct nlattr *a; | |
557 | int rem; | |
558 | ||
559 | nla_for_each_nested(a, attr, rem) { | |
560 | int type = nla_type(a); | |
561 | int maxlen = ovs_ct_attr_lens[type].maxlen; | |
562 | int minlen = ovs_ct_attr_lens[type].minlen; | |
563 | ||
564 | if (type > OVS_CT_ATTR_MAX) { | |
565 | OVS_NLERR(log, | |
566 | "Unknown conntrack attr (type=%d, max=%d)", | |
567 | type, OVS_CT_ATTR_MAX); | |
568 | return -EINVAL; | |
569 | } | |
570 | if (nla_len(a) < minlen || nla_len(a) > maxlen) { | |
571 | OVS_NLERR(log, | |
572 | "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)", | |
573 | type, nla_len(a), maxlen); | |
574 | return -EINVAL; | |
575 | } | |
576 | ||
577 | switch (type) { | |
ab38a7b5 JS |
578 | case OVS_CT_ATTR_COMMIT: |
579 | info->commit = true; | |
7f8a436e JS |
580 | break; |
581 | #ifdef CONFIG_NF_CONNTRACK_ZONES | |
582 | case OVS_CT_ATTR_ZONE: | |
583 | info->zone.id = nla_get_u16(a); | |
584 | break; | |
182e3042 JS |
585 | #endif |
586 | #ifdef CONFIG_NF_CONNTRACK_MARK | |
587 | case OVS_CT_ATTR_MARK: { | |
588 | struct md_mark *mark = nla_data(a); | |
589 | ||
590 | info->mark = *mark; | |
591 | break; | |
592 | } | |
c2ac6673 JS |
593 | #endif |
594 | #ifdef CONFIG_NF_CONNTRACK_LABELS | |
33db4125 JS |
595 | case OVS_CT_ATTR_LABELS: { |
596 | struct md_labels *labels = nla_data(a); | |
c2ac6673 | 597 | |
33db4125 | 598 | info->labels = *labels; |
c2ac6673 JS |
599 | break; |
600 | } | |
7f8a436e | 601 | #endif |
cae3a262 JS |
602 | case OVS_CT_ATTR_HELPER: |
603 | *helper = nla_data(a); | |
604 | if (!memchr(*helper, '\0', nla_len(a))) { | |
605 | OVS_NLERR(log, "Invalid conntrack helper"); | |
606 | return -EINVAL; | |
607 | } | |
608 | break; | |
7f8a436e JS |
609 | default: |
610 | OVS_NLERR(log, "Unknown conntrack attr (%d)", | |
611 | type); | |
612 | return -EINVAL; | |
613 | } | |
614 | } | |
615 | ||
616 | if (rem > 0) { | |
617 | OVS_NLERR(log, "Conntrack attr has %d unknown bytes", rem); | |
618 | return -EINVAL; | |
619 | } | |
620 | ||
621 | return 0; | |
622 | } | |
623 | ||
c2ac6673 | 624 | bool ovs_ct_verify(struct net *net, enum ovs_key_attr attr) |
7f8a436e JS |
625 | { |
626 | if (attr == OVS_KEY_ATTR_CT_STATE) | |
627 | return true; | |
628 | if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) && | |
629 | attr == OVS_KEY_ATTR_CT_ZONE) | |
630 | return true; | |
182e3042 JS |
631 | if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && |
632 | attr == OVS_KEY_ATTR_CT_MARK) | |
633 | return true; | |
c2ac6673 | 634 | if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) && |
33db4125 | 635 | attr == OVS_KEY_ATTR_CT_LABELS) { |
c2ac6673 JS |
636 | struct ovs_net *ovs_net = net_generic(net, ovs_net_id); |
637 | ||
638 | return ovs_net->xt_label; | |
639 | } | |
7f8a436e JS |
640 | |
641 | return false; | |
642 | } | |
643 | ||
644 | int ovs_ct_copy_action(struct net *net, const struct nlattr *attr, | |
645 | const struct sw_flow_key *key, | |
646 | struct sw_flow_actions **sfa, bool log) | |
647 | { | |
648 | struct ovs_conntrack_info ct_info; | |
cae3a262 | 649 | const char *helper = NULL; |
7f8a436e JS |
650 | u16 family; |
651 | int err; | |
652 | ||
653 | family = key_to_nfproto(key); | |
654 | if (family == NFPROTO_UNSPEC) { | |
655 | OVS_NLERR(log, "ct family unspecified"); | |
656 | return -EINVAL; | |
657 | } | |
658 | ||
659 | memset(&ct_info, 0, sizeof(ct_info)); | |
660 | ct_info.family = family; | |
661 | ||
662 | nf_ct_zone_init(&ct_info.zone, NF_CT_DEFAULT_ZONE_ID, | |
663 | NF_CT_DEFAULT_ZONE_DIR, 0); | |
664 | ||
cae3a262 | 665 | err = parse_ct(attr, &ct_info, &helper, log); |
7f8a436e JS |
666 | if (err) |
667 | return err; | |
668 | ||
669 | /* Set up template for tracking connections in specific zones. */ | |
670 | ct_info.ct = nf_ct_tmpl_alloc(net, &ct_info.zone, GFP_KERNEL); | |
671 | if (!ct_info.ct) { | |
672 | OVS_NLERR(log, "Failed to allocate conntrack template"); | |
673 | return -ENOMEM; | |
674 | } | |
cae3a262 JS |
675 | if (helper) { |
676 | err = ovs_ct_add_helper(&ct_info, helper, key, log); | |
677 | if (err) | |
678 | goto err_free_ct; | |
679 | } | |
7f8a436e JS |
680 | |
681 | err = ovs_nla_add_action(sfa, OVS_ACTION_ATTR_CT, &ct_info, | |
682 | sizeof(ct_info), log); | |
683 | if (err) | |
684 | goto err_free_ct; | |
685 | ||
686 | __set_bit(IPS_CONFIRMED_BIT, &ct_info.ct->status); | |
687 | nf_conntrack_get(&ct_info.ct->ct_general); | |
688 | return 0; | |
689 | err_free_ct: | |
690 | nf_conntrack_free(ct_info.ct); | |
691 | return err; | |
692 | } | |
693 | ||
694 | int ovs_ct_action_to_attr(const struct ovs_conntrack_info *ct_info, | |
695 | struct sk_buff *skb) | |
696 | { | |
697 | struct nlattr *start; | |
698 | ||
699 | start = nla_nest_start(skb, OVS_ACTION_ATTR_CT); | |
700 | if (!start) | |
701 | return -EMSGSIZE; | |
702 | ||
ab38a7b5 | 703 | if (ct_info->commit && nla_put_flag(skb, OVS_CT_ATTR_COMMIT)) |
7f8a436e JS |
704 | return -EMSGSIZE; |
705 | if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) && | |
706 | nla_put_u16(skb, OVS_CT_ATTR_ZONE, ct_info->zone.id)) | |
707 | return -EMSGSIZE; | |
182e3042 JS |
708 | if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && |
709 | nla_put(skb, OVS_CT_ATTR_MARK, sizeof(ct_info->mark), | |
710 | &ct_info->mark)) | |
711 | return -EMSGSIZE; | |
c2ac6673 | 712 | if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) && |
33db4125 JS |
713 | nla_put(skb, OVS_CT_ATTR_LABELS, sizeof(ct_info->labels), |
714 | &ct_info->labels)) | |
c2ac6673 | 715 | return -EMSGSIZE; |
cae3a262 JS |
716 | if (ct_info->helper) { |
717 | if (nla_put_string(skb, OVS_CT_ATTR_HELPER, | |
718 | ct_info->helper->name)) | |
719 | return -EMSGSIZE; | |
720 | } | |
7f8a436e JS |
721 | |
722 | nla_nest_end(skb, start); | |
723 | ||
724 | return 0; | |
725 | } | |
726 | ||
727 | void ovs_ct_free_action(const struct nlattr *a) | |
728 | { | |
729 | struct ovs_conntrack_info *ct_info = nla_data(a); | |
730 | ||
cae3a262 JS |
731 | if (ct_info->helper) |
732 | module_put(ct_info->helper->me); | |
7f8a436e JS |
733 | if (ct_info->ct) |
734 | nf_ct_put(ct_info->ct); | |
735 | } | |
c2ac6673 JS |
736 | |
737 | void ovs_ct_init(struct net *net) | |
738 | { | |
33db4125 | 739 | unsigned int n_bits = sizeof(struct ovs_key_ct_labels) * BITS_PER_BYTE; |
c2ac6673 JS |
740 | struct ovs_net *ovs_net = net_generic(net, ovs_net_id); |
741 | ||
742 | if (nf_connlabels_get(net, n_bits)) { | |
743 | ovs_net->xt_label = false; | |
744 | OVS_NLERR(true, "Failed to set connlabel length"); | |
745 | } else { | |
746 | ovs_net->xt_label = true; | |
747 | } | |
748 | } | |
749 | ||
750 | void ovs_ct_exit(struct net *net) | |
751 | { | |
752 | struct ovs_net *ovs_net = net_generic(net, ovs_net_id); | |
753 | ||
754 | if (ovs_net->xt_label) | |
755 | nf_connlabels_put(net); | |
756 | } |