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); | |
4f0909ee JS |
154 | if (!nf_ct_is_confirmed(ct)) |
155 | state |= OVS_CS_F_NEW; | |
7f8a436e JS |
156 | if (ct->master) |
157 | state |= OVS_CS_F_RELATED; | |
158 | zone = nf_ct_zone(ct); | |
159 | } else if (post_ct) { | |
160 | state = OVS_CS_F_TRACKED | OVS_CS_F_INVALID; | |
161 | } | |
182e3042 | 162 | __ovs_ct_update_key(key, state, zone, ct); |
7f8a436e JS |
163 | } |
164 | ||
165 | void ovs_ct_fill_key(const struct sk_buff *skb, struct sw_flow_key *key) | |
166 | { | |
167 | ovs_ct_update_key(skb, key, false); | |
168 | } | |
169 | ||
170 | int ovs_ct_put_key(const struct sw_flow_key *key, struct sk_buff *skb) | |
171 | { | |
fbccce59 | 172 | if (nla_put_u32(skb, OVS_KEY_ATTR_CT_STATE, key->ct.state)) |
7f8a436e JS |
173 | return -EMSGSIZE; |
174 | ||
175 | if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) && | |
176 | nla_put_u16(skb, OVS_KEY_ATTR_CT_ZONE, key->ct.zone)) | |
177 | return -EMSGSIZE; | |
178 | ||
182e3042 JS |
179 | if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && |
180 | nla_put_u32(skb, OVS_KEY_ATTR_CT_MARK, key->ct.mark)) | |
181 | return -EMSGSIZE; | |
182 | ||
9723e6ab | 183 | if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) && |
33db4125 JS |
184 | nla_put(skb, OVS_KEY_ATTR_CT_LABELS, sizeof(key->ct.labels), |
185 | &key->ct.labels)) | |
c2ac6673 JS |
186 | return -EMSGSIZE; |
187 | ||
182e3042 JS |
188 | return 0; |
189 | } | |
190 | ||
191 | static int ovs_ct_set_mark(struct sk_buff *skb, struct sw_flow_key *key, | |
192 | u32 ct_mark, u32 mask) | |
193 | { | |
0d5cdef8 | 194 | #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) |
182e3042 JS |
195 | enum ip_conntrack_info ctinfo; |
196 | struct nf_conn *ct; | |
197 | u32 new_mark; | |
198 | ||
182e3042 JS |
199 | |
200 | /* The connection could be invalid, in which case set_mark is no-op. */ | |
201 | ct = nf_ct_get(skb, &ctinfo); | |
202 | if (!ct) | |
203 | return 0; | |
204 | ||
205 | new_mark = ct_mark | (ct->mark & ~(mask)); | |
206 | if (ct->mark != new_mark) { | |
207 | ct->mark = new_mark; | |
208 | nf_conntrack_event_cache(IPCT_MARK, ct); | |
209 | key->ct.mark = new_mark; | |
210 | } | |
211 | ||
7f8a436e | 212 | return 0; |
0d5cdef8 JS |
213 | #else |
214 | return -ENOTSUPP; | |
215 | #endif | |
7f8a436e JS |
216 | } |
217 | ||
33db4125 JS |
218 | static int ovs_ct_set_labels(struct sk_buff *skb, struct sw_flow_key *key, |
219 | const struct ovs_key_ct_labels *labels, | |
220 | const struct ovs_key_ct_labels *mask) | |
c2ac6673 JS |
221 | { |
222 | enum ip_conntrack_info ctinfo; | |
223 | struct nf_conn_labels *cl; | |
224 | struct nf_conn *ct; | |
225 | int err; | |
226 | ||
c2ac6673 JS |
227 | /* The connection could be invalid, in which case set_label is no-op.*/ |
228 | ct = nf_ct_get(skb, &ctinfo); | |
229 | if (!ct) | |
230 | return 0; | |
231 | ||
232 | cl = nf_ct_labels_find(ct); | |
233 | if (!cl) { | |
234 | nf_ct_labels_ext_add(ct); | |
235 | cl = nf_ct_labels_find(ct); | |
236 | } | |
33db4125 | 237 | if (!cl || cl->words * sizeof(long) < OVS_CT_LABELS_LEN) |
c2ac6673 JS |
238 | return -ENOSPC; |
239 | ||
33db4125 JS |
240 | err = nf_connlabels_replace(ct, (u32 *)labels, (u32 *)mask, |
241 | OVS_CT_LABELS_LEN / sizeof(u32)); | |
c2ac6673 JS |
242 | if (err) |
243 | return err; | |
244 | ||
33db4125 | 245 | ovs_ct_get_labels(ct, &key->ct.labels); |
c2ac6673 JS |
246 | return 0; |
247 | } | |
248 | ||
cae3a262 JS |
249 | /* 'skb' should already be pulled to nh_ofs. */ |
250 | static int ovs_ct_helper(struct sk_buff *skb, u16 proto) | |
251 | { | |
252 | const struct nf_conntrack_helper *helper; | |
253 | const struct nf_conn_help *help; | |
254 | enum ip_conntrack_info ctinfo; | |
255 | unsigned int protoff; | |
256 | struct nf_conn *ct; | |
257 | ||
258 | ct = nf_ct_get(skb, &ctinfo); | |
259 | if (!ct || ctinfo == IP_CT_RELATED_REPLY) | |
260 | return NF_ACCEPT; | |
261 | ||
262 | help = nfct_help(ct); | |
263 | if (!help) | |
264 | return NF_ACCEPT; | |
265 | ||
266 | helper = rcu_dereference(help->helper); | |
267 | if (!helper) | |
268 | return NF_ACCEPT; | |
269 | ||
270 | switch (proto) { | |
271 | case NFPROTO_IPV4: | |
272 | protoff = ip_hdrlen(skb); | |
273 | break; | |
274 | case NFPROTO_IPV6: { | |
275 | u8 nexthdr = ipv6_hdr(skb)->nexthdr; | |
276 | __be16 frag_off; | |
cc570605 | 277 | int ofs; |
cae3a262 | 278 | |
cc570605 JS |
279 | ofs = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr, |
280 | &frag_off); | |
281 | if (ofs < 0 || (frag_off & htons(~0x7)) != 0) { | |
cae3a262 JS |
282 | pr_debug("proto header not found\n"); |
283 | return NF_ACCEPT; | |
284 | } | |
cc570605 | 285 | protoff = ofs; |
cae3a262 JS |
286 | break; |
287 | } | |
288 | default: | |
289 | WARN_ONCE(1, "helper invoked on non-IP family!"); | |
290 | return NF_DROP; | |
291 | } | |
292 | ||
293 | return helper->help(skb, protoff, ct, ctinfo); | |
294 | } | |
295 | ||
74c16618 JS |
296 | /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero |
297 | * value if 'skb' is freed. | |
298 | */ | |
7f8a436e JS |
299 | static int handle_fragments(struct net *net, struct sw_flow_key *key, |
300 | u16 zone, struct sk_buff *skb) | |
301 | { | |
302 | struct ovs_skb_cb ovs_cb = *OVS_CB(skb); | |
303 | ||
304 | if (key->eth.type == htons(ETH_P_IP)) { | |
305 | enum ip_defrag_users user = IP_DEFRAG_CONNTRACK_IN + zone; | |
306 | int err; | |
307 | ||
308 | memset(IPCB(skb), 0, sizeof(struct inet_skb_parm)); | |
19bcf9f2 | 309 | err = ip_defrag(net, skb, user); |
7f8a436e JS |
310 | if (err) |
311 | return err; | |
312 | ||
313 | ovs_cb.mru = IPCB(skb)->frag_max_size; | |
7f8a436e | 314 | #if IS_ENABLED(CONFIG_NF_DEFRAG_IPV6) |
74c16618 | 315 | } else if (key->eth.type == htons(ETH_P_IPV6)) { |
7f8a436e JS |
316 | enum ip6_defrag_users user = IP6_DEFRAG_CONNTRACK_IN + zone; |
317 | struct sk_buff *reasm; | |
318 | ||
319 | memset(IP6CB(skb), 0, sizeof(struct inet6_skb_parm)); | |
b7277597 | 320 | reasm = nf_ct_frag6_gather(net, skb, user); |
7f8a436e JS |
321 | if (!reasm) |
322 | return -EINPROGRESS; | |
323 | ||
74c16618 JS |
324 | if (skb == reasm) { |
325 | kfree_skb(skb); | |
7f8a436e | 326 | return -EINVAL; |
74c16618 | 327 | } |
7f8a436e | 328 | |
6f5cadee JS |
329 | /* Don't free 'skb' even though it is one of the original |
330 | * fragments, as we're going to morph it into the head. | |
331 | */ | |
332 | skb_get(skb); | |
333 | nf_ct_frag6_consume_orig(reasm); | |
7f8a436e JS |
334 | |
335 | key->ip.proto = ipv6_hdr(reasm)->nexthdr; | |
336 | skb_morph(skb, reasm); | |
6f5cadee | 337 | skb->next = reasm->next; |
7f8a436e JS |
338 | consume_skb(reasm); |
339 | ovs_cb.mru = IP6CB(skb)->frag_max_size; | |
7f8a436e JS |
340 | #endif |
341 | } else { | |
74c16618 | 342 | kfree_skb(skb); |
7f8a436e JS |
343 | return -EPFNOSUPPORT; |
344 | } | |
345 | ||
346 | key->ip.frag = OVS_FRAG_TYPE_NONE; | |
347 | skb_clear_hash(skb); | |
348 | skb->ignore_df = 1; | |
349 | *OVS_CB(skb) = ovs_cb; | |
350 | ||
351 | return 0; | |
352 | } | |
353 | ||
354 | static struct nf_conntrack_expect * | |
355 | ovs_ct_expect_find(struct net *net, const struct nf_conntrack_zone *zone, | |
356 | u16 proto, const struct sk_buff *skb) | |
357 | { | |
358 | struct nf_conntrack_tuple tuple; | |
359 | ||
a31f1adc | 360 | if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb), proto, net, &tuple)) |
7f8a436e JS |
361 | return NULL; |
362 | return __nf_ct_expect_find(net, zone, &tuple); | |
363 | } | |
364 | ||
365 | /* Determine whether skb->nfct is equal to the result of conntrack lookup. */ | |
366 | static bool skb_nfct_cached(const struct net *net, const struct sk_buff *skb, | |
367 | const struct ovs_conntrack_info *info) | |
368 | { | |
369 | enum ip_conntrack_info ctinfo; | |
370 | struct nf_conn *ct; | |
371 | ||
372 | ct = nf_ct_get(skb, &ctinfo); | |
373 | if (!ct) | |
374 | return false; | |
375 | if (!net_eq(net, read_pnet(&ct->ct_net))) | |
376 | return false; | |
377 | if (!nf_ct_zone_equal_any(info->ct, nf_ct_zone(ct))) | |
378 | return false; | |
cae3a262 JS |
379 | if (info->helper) { |
380 | struct nf_conn_help *help; | |
381 | ||
382 | help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER); | |
383 | if (help && rcu_access_pointer(help->helper) != info->helper) | |
384 | return false; | |
385 | } | |
7f8a436e JS |
386 | |
387 | return true; | |
388 | } | |
389 | ||
4f0909ee | 390 | static int __ovs_ct_lookup(struct net *net, struct sw_flow_key *key, |
7f8a436e JS |
391 | const struct ovs_conntrack_info *info, |
392 | struct sk_buff *skb) | |
393 | { | |
394 | /* If we are recirculating packets to match on conntrack fields and | |
395 | * committing with a separate conntrack action, then we don't need to | |
396 | * actually run the packet through conntrack twice unless it's for a | |
397 | * different zone. | |
398 | */ | |
399 | if (!skb_nfct_cached(net, skb, info)) { | |
400 | struct nf_conn *tmpl = info->ct; | |
401 | ||
402 | /* Associate skb with specified zone. */ | |
403 | if (tmpl) { | |
404 | if (skb->nfct) | |
405 | nf_conntrack_put(skb->nfct); | |
406 | nf_conntrack_get(&tmpl->ct_general); | |
407 | skb->nfct = &tmpl->ct_general; | |
408 | skb->nfctinfo = IP_CT_NEW; | |
409 | } | |
410 | ||
411 | if (nf_conntrack_in(net, info->family, NF_INET_PRE_ROUTING, | |
412 | skb) != NF_ACCEPT) | |
413 | return -ENOENT; | |
cae3a262 JS |
414 | |
415 | if (ovs_ct_helper(skb, info->family) != NF_ACCEPT) { | |
416 | WARN_ONCE(1, "helper rejected packet"); | |
417 | return -EINVAL; | |
418 | } | |
7f8a436e JS |
419 | } |
420 | ||
4f0909ee JS |
421 | ovs_ct_update_key(skb, key, true); |
422 | ||
7f8a436e JS |
423 | return 0; |
424 | } | |
425 | ||
426 | /* Lookup connection and read fields into key. */ | |
427 | static int ovs_ct_lookup(struct net *net, struct sw_flow_key *key, | |
428 | const struct ovs_conntrack_info *info, | |
429 | struct sk_buff *skb) | |
430 | { | |
431 | struct nf_conntrack_expect *exp; | |
432 | ||
433 | exp = ovs_ct_expect_find(net, &info->zone, info->family, skb); | |
434 | if (exp) { | |
435 | u8 state; | |
436 | ||
437 | state = OVS_CS_F_TRACKED | OVS_CS_F_NEW | OVS_CS_F_RELATED; | |
182e3042 | 438 | __ovs_ct_update_key(key, state, &info->zone, exp->master); |
7f8a436e JS |
439 | } else { |
440 | int err; | |
441 | ||
442 | err = __ovs_ct_lookup(net, key, info, skb); | |
443 | if (err) | |
444 | return err; | |
7f8a436e JS |
445 | } |
446 | ||
447 | return 0; | |
448 | } | |
449 | ||
450 | /* Lookup connection and confirm if unconfirmed. */ | |
451 | static int ovs_ct_commit(struct net *net, struct sw_flow_key *key, | |
452 | const struct ovs_conntrack_info *info, | |
453 | struct sk_buff *skb) | |
454 | { | |
455 | u8 state; | |
456 | int err; | |
457 | ||
458 | state = key->ct.state; | |
459 | if (key->ct.zone == info->zone.id && | |
460 | ((state & OVS_CS_F_TRACKED) && !(state & OVS_CS_F_NEW))) { | |
461 | /* Previous lookup has shown that this connection is already | |
462 | * tracked and committed. Skip committing. | |
463 | */ | |
464 | return 0; | |
465 | } | |
466 | ||
467 | err = __ovs_ct_lookup(net, key, info, skb); | |
468 | if (err) | |
469 | return err; | |
470 | if (nf_conntrack_confirm(skb) != NF_ACCEPT) | |
471 | return -EINVAL; | |
472 | ||
7f8a436e JS |
473 | return 0; |
474 | } | |
475 | ||
33db4125 | 476 | static bool labels_nonzero(const struct ovs_key_ct_labels *labels) |
c2ac6673 JS |
477 | { |
478 | size_t i; | |
479 | ||
33db4125 JS |
480 | for (i = 0; i < sizeof(*labels); i++) |
481 | if (labels->ct_labels[i]) | |
c2ac6673 JS |
482 | return true; |
483 | ||
484 | return false; | |
485 | } | |
486 | ||
74c16618 JS |
487 | /* Returns 0 on success, -EINPROGRESS if 'skb' is stolen, or other nonzero |
488 | * value if 'skb' is freed. | |
489 | */ | |
7f8a436e JS |
490 | int ovs_ct_execute(struct net *net, struct sk_buff *skb, |
491 | struct sw_flow_key *key, | |
492 | const struct ovs_conntrack_info *info) | |
493 | { | |
494 | int nh_ofs; | |
495 | int err; | |
496 | ||
497 | /* The conntrack module expects to be working at L3. */ | |
498 | nh_ofs = skb_network_offset(skb); | |
499 | skb_pull(skb, nh_ofs); | |
500 | ||
501 | if (key->ip.frag != OVS_FRAG_TYPE_NONE) { | |
502 | err = handle_fragments(net, key, info->zone.id, skb); | |
503 | if (err) | |
504 | return err; | |
505 | } | |
506 | ||
ab38a7b5 | 507 | if (info->commit) |
7f8a436e JS |
508 | err = ovs_ct_commit(net, key, info, skb); |
509 | else | |
510 | err = ovs_ct_lookup(net, key, info, skb); | |
182e3042 JS |
511 | if (err) |
512 | goto err; | |
7f8a436e | 513 | |
c2ac6673 | 514 | if (info->mark.mask) { |
182e3042 JS |
515 | err = ovs_ct_set_mark(skb, key, info->mark.value, |
516 | info->mark.mask); | |
c2ac6673 JS |
517 | if (err) |
518 | goto err; | |
519 | } | |
33db4125 JS |
520 | if (labels_nonzero(&info->labels.mask)) |
521 | err = ovs_ct_set_labels(skb, key, &info->labels.value, | |
522 | &info->labels.mask); | |
182e3042 | 523 | err: |
7f8a436e | 524 | skb_push(skb, nh_ofs); |
74c16618 JS |
525 | if (err) |
526 | kfree_skb(skb); | |
7f8a436e JS |
527 | return err; |
528 | } | |
529 | ||
cae3a262 JS |
530 | static int ovs_ct_add_helper(struct ovs_conntrack_info *info, const char *name, |
531 | const struct sw_flow_key *key, bool log) | |
532 | { | |
533 | struct nf_conntrack_helper *helper; | |
534 | struct nf_conn_help *help; | |
535 | ||
536 | helper = nf_conntrack_helper_try_module_get(name, info->family, | |
537 | key->ip.proto); | |
538 | if (!helper) { | |
539 | OVS_NLERR(log, "Unknown helper \"%s\"", name); | |
540 | return -EINVAL; | |
541 | } | |
542 | ||
543 | help = nf_ct_helper_ext_add(info->ct, helper, GFP_KERNEL); | |
544 | if (!help) { | |
545 | module_put(helper->me); | |
546 | return -ENOMEM; | |
547 | } | |
548 | ||
549 | rcu_assign_pointer(help->helper, helper); | |
550 | info->helper = helper; | |
551 | return 0; | |
552 | } | |
553 | ||
7f8a436e | 554 | static const struct ovs_ct_len_tbl ovs_ct_attr_lens[OVS_CT_ATTR_MAX + 1] = { |
ab38a7b5 | 555 | [OVS_CT_ATTR_COMMIT] = { .minlen = 0, .maxlen = 0 }, |
7f8a436e JS |
556 | [OVS_CT_ATTR_ZONE] = { .minlen = sizeof(u16), |
557 | .maxlen = sizeof(u16) }, | |
182e3042 JS |
558 | [OVS_CT_ATTR_MARK] = { .minlen = sizeof(struct md_mark), |
559 | .maxlen = sizeof(struct md_mark) }, | |
33db4125 JS |
560 | [OVS_CT_ATTR_LABELS] = { .minlen = sizeof(struct md_labels), |
561 | .maxlen = sizeof(struct md_labels) }, | |
cae3a262 JS |
562 | [OVS_CT_ATTR_HELPER] = { .minlen = 1, |
563 | .maxlen = NF_CT_HELPER_NAME_LEN } | |
7f8a436e JS |
564 | }; |
565 | ||
566 | static int parse_ct(const struct nlattr *attr, struct ovs_conntrack_info *info, | |
cae3a262 | 567 | const char **helper, bool log) |
7f8a436e JS |
568 | { |
569 | struct nlattr *a; | |
570 | int rem; | |
571 | ||
572 | nla_for_each_nested(a, attr, rem) { | |
573 | int type = nla_type(a); | |
574 | int maxlen = ovs_ct_attr_lens[type].maxlen; | |
575 | int minlen = ovs_ct_attr_lens[type].minlen; | |
576 | ||
577 | if (type > OVS_CT_ATTR_MAX) { | |
578 | OVS_NLERR(log, | |
579 | "Unknown conntrack attr (type=%d, max=%d)", | |
580 | type, OVS_CT_ATTR_MAX); | |
581 | return -EINVAL; | |
582 | } | |
583 | if (nla_len(a) < minlen || nla_len(a) > maxlen) { | |
584 | OVS_NLERR(log, | |
585 | "Conntrack attr type has unexpected length (type=%d, length=%d, expected=%d)", | |
586 | type, nla_len(a), maxlen); | |
587 | return -EINVAL; | |
588 | } | |
589 | ||
590 | switch (type) { | |
ab38a7b5 JS |
591 | case OVS_CT_ATTR_COMMIT: |
592 | info->commit = true; | |
7f8a436e JS |
593 | break; |
594 | #ifdef CONFIG_NF_CONNTRACK_ZONES | |
595 | case OVS_CT_ATTR_ZONE: | |
596 | info->zone.id = nla_get_u16(a); | |
597 | break; | |
182e3042 JS |
598 | #endif |
599 | #ifdef CONFIG_NF_CONNTRACK_MARK | |
600 | case OVS_CT_ATTR_MARK: { | |
601 | struct md_mark *mark = nla_data(a); | |
602 | ||
e754ec69 JS |
603 | if (!mark->mask) { |
604 | OVS_NLERR(log, "ct_mark mask cannot be 0"); | |
605 | return -EINVAL; | |
606 | } | |
182e3042 JS |
607 | info->mark = *mark; |
608 | break; | |
609 | } | |
c2ac6673 JS |
610 | #endif |
611 | #ifdef CONFIG_NF_CONNTRACK_LABELS | |
33db4125 JS |
612 | case OVS_CT_ATTR_LABELS: { |
613 | struct md_labels *labels = nla_data(a); | |
c2ac6673 | 614 | |
e754ec69 JS |
615 | if (!labels_nonzero(&labels->mask)) { |
616 | OVS_NLERR(log, "ct_labels mask cannot be 0"); | |
617 | return -EINVAL; | |
618 | } | |
33db4125 | 619 | info->labels = *labels; |
c2ac6673 JS |
620 | break; |
621 | } | |
7f8a436e | 622 | #endif |
cae3a262 JS |
623 | case OVS_CT_ATTR_HELPER: |
624 | *helper = nla_data(a); | |
625 | if (!memchr(*helper, '\0', nla_len(a))) { | |
626 | OVS_NLERR(log, "Invalid conntrack helper"); | |
627 | return -EINVAL; | |
628 | } | |
629 | break; | |
7f8a436e JS |
630 | default: |
631 | OVS_NLERR(log, "Unknown conntrack attr (%d)", | |
632 | type); | |
633 | return -EINVAL; | |
634 | } | |
635 | } | |
636 | ||
637 | if (rem > 0) { | |
638 | OVS_NLERR(log, "Conntrack attr has %d unknown bytes", rem); | |
639 | return -EINVAL; | |
640 | } | |
641 | ||
642 | return 0; | |
643 | } | |
644 | ||
c2ac6673 | 645 | bool ovs_ct_verify(struct net *net, enum ovs_key_attr attr) |
7f8a436e JS |
646 | { |
647 | if (attr == OVS_KEY_ATTR_CT_STATE) | |
648 | return true; | |
649 | if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) && | |
650 | attr == OVS_KEY_ATTR_CT_ZONE) | |
651 | return true; | |
182e3042 JS |
652 | if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && |
653 | attr == OVS_KEY_ATTR_CT_MARK) | |
654 | return true; | |
c2ac6673 | 655 | if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) && |
33db4125 | 656 | attr == OVS_KEY_ATTR_CT_LABELS) { |
c2ac6673 JS |
657 | struct ovs_net *ovs_net = net_generic(net, ovs_net_id); |
658 | ||
659 | return ovs_net->xt_label; | |
660 | } | |
7f8a436e JS |
661 | |
662 | return false; | |
663 | } | |
664 | ||
665 | int ovs_ct_copy_action(struct net *net, const struct nlattr *attr, | |
666 | const struct sw_flow_key *key, | |
667 | struct sw_flow_actions **sfa, bool log) | |
668 | { | |
669 | struct ovs_conntrack_info ct_info; | |
cae3a262 | 670 | const char *helper = NULL; |
7f8a436e JS |
671 | u16 family; |
672 | int err; | |
673 | ||
674 | family = key_to_nfproto(key); | |
675 | if (family == NFPROTO_UNSPEC) { | |
676 | OVS_NLERR(log, "ct family unspecified"); | |
677 | return -EINVAL; | |
678 | } | |
679 | ||
680 | memset(&ct_info, 0, sizeof(ct_info)); | |
681 | ct_info.family = family; | |
682 | ||
683 | nf_ct_zone_init(&ct_info.zone, NF_CT_DEFAULT_ZONE_ID, | |
684 | NF_CT_DEFAULT_ZONE_DIR, 0); | |
685 | ||
cae3a262 | 686 | err = parse_ct(attr, &ct_info, &helper, log); |
7f8a436e JS |
687 | if (err) |
688 | return err; | |
689 | ||
690 | /* Set up template for tracking connections in specific zones. */ | |
691 | ct_info.ct = nf_ct_tmpl_alloc(net, &ct_info.zone, GFP_KERNEL); | |
692 | if (!ct_info.ct) { | |
693 | OVS_NLERR(log, "Failed to allocate conntrack template"); | |
694 | return -ENOMEM; | |
695 | } | |
cae3a262 JS |
696 | if (helper) { |
697 | err = ovs_ct_add_helper(&ct_info, helper, key, log); | |
698 | if (err) | |
699 | goto err_free_ct; | |
700 | } | |
7f8a436e JS |
701 | |
702 | err = ovs_nla_add_action(sfa, OVS_ACTION_ATTR_CT, &ct_info, | |
703 | sizeof(ct_info), log); | |
704 | if (err) | |
705 | goto err_free_ct; | |
706 | ||
707 | __set_bit(IPS_CONFIRMED_BIT, &ct_info.ct->status); | |
708 | nf_conntrack_get(&ct_info.ct->ct_general); | |
709 | return 0; | |
710 | err_free_ct: | |
711 | nf_conntrack_free(ct_info.ct); | |
712 | return err; | |
713 | } | |
714 | ||
715 | int ovs_ct_action_to_attr(const struct ovs_conntrack_info *ct_info, | |
716 | struct sk_buff *skb) | |
717 | { | |
718 | struct nlattr *start; | |
719 | ||
720 | start = nla_nest_start(skb, OVS_ACTION_ATTR_CT); | |
721 | if (!start) | |
722 | return -EMSGSIZE; | |
723 | ||
ab38a7b5 | 724 | if (ct_info->commit && nla_put_flag(skb, OVS_CT_ATTR_COMMIT)) |
7f8a436e JS |
725 | return -EMSGSIZE; |
726 | if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) && | |
727 | nla_put_u16(skb, OVS_CT_ATTR_ZONE, ct_info->zone.id)) | |
728 | return -EMSGSIZE; | |
e754ec69 | 729 | if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) && ct_info->mark.mask && |
182e3042 JS |
730 | nla_put(skb, OVS_CT_ATTR_MARK, sizeof(ct_info->mark), |
731 | &ct_info->mark)) | |
732 | return -EMSGSIZE; | |
c2ac6673 | 733 | if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) && |
e754ec69 | 734 | labels_nonzero(&ct_info->labels.mask) && |
33db4125 JS |
735 | nla_put(skb, OVS_CT_ATTR_LABELS, sizeof(ct_info->labels), |
736 | &ct_info->labels)) | |
c2ac6673 | 737 | return -EMSGSIZE; |
cae3a262 JS |
738 | if (ct_info->helper) { |
739 | if (nla_put_string(skb, OVS_CT_ATTR_HELPER, | |
740 | ct_info->helper->name)) | |
741 | return -EMSGSIZE; | |
742 | } | |
7f8a436e JS |
743 | |
744 | nla_nest_end(skb, start); | |
745 | ||
746 | return 0; | |
747 | } | |
748 | ||
749 | void ovs_ct_free_action(const struct nlattr *a) | |
750 | { | |
751 | struct ovs_conntrack_info *ct_info = nla_data(a); | |
752 | ||
cae3a262 JS |
753 | if (ct_info->helper) |
754 | module_put(ct_info->helper->me); | |
7f8a436e JS |
755 | if (ct_info->ct) |
756 | nf_ct_put(ct_info->ct); | |
757 | } | |
c2ac6673 JS |
758 | |
759 | void ovs_ct_init(struct net *net) | |
760 | { | |
33db4125 | 761 | unsigned int n_bits = sizeof(struct ovs_key_ct_labels) * BITS_PER_BYTE; |
c2ac6673 JS |
762 | struct ovs_net *ovs_net = net_generic(net, ovs_net_id); |
763 | ||
764 | if (nf_connlabels_get(net, n_bits)) { | |
765 | ovs_net->xt_label = false; | |
766 | OVS_NLERR(true, "Failed to set connlabel length"); | |
767 | } else { | |
768 | ovs_net->xt_label = true; | |
769 | } | |
770 | } | |
771 | ||
772 | void ovs_ct_exit(struct net *net) | |
773 | { | |
774 | struct ovs_net *ovs_net = net_generic(net, ovs_net_id); | |
775 | ||
776 | if (ovs_net->xt_label) | |
777 | nf_connlabels_put(net); | |
778 | } |