1 // SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB
2 /* -
3 * net/sched/act_ct.c Connection Tracking action
4 *
5 * Authors: Paul Blakey <paulb@mellanox.com>
6 * Yossi Kuperman <yossiku@mellanox.com>
7 * Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
8 */
9
10 #include <linux/module.h>
11 #include <linux/init.h>
12 #include <linux/kernel.h>
13 #include <linux/skbuff.h>
14 #include <linux/rtnetlink.h>
15 #include <linux/pkt_cls.h>
16 #include <linux/if_tunnel.h>
17 #include <linux/ip.h>
18 #include <linux/ipv6.h>
19 #include <linux/rhashtable.h>
20 #include <net/gre.h>
21 #include <net/netlink.h>
22 #include <net/pkt_sched.h>
23 #include <net/pkt_cls.h>
24 #include <net/act_api.h>
25 #include <net/ip.h>
26 #include <net/ipv6_frag.h>
27 #include <uapi/linux/tc_act/tc_ct.h>
28 #include <net/tc_act/tc_ct.h>
29 #include <net/tc_wrapper.h>
30
31 #include <net/netfilter/nf_flow_table.h>
32 #include <net/netfilter/nf_conntrack.h>
33 #include <net/netfilter/nf_conntrack_core.h>
34 #include <net/netfilter/nf_conntrack_zones.h>
35 #include <net/netfilter/nf_conntrack_helper.h>
36 #include <net/netfilter/nf_conntrack_acct.h>
37 #include <net/netfilter/ipv6/nf_defrag_ipv6.h>
38 #include <net/netfilter/nf_conntrack_act_ct.h>
39 #include <net/netfilter/nf_conntrack_seqadj.h>
40 #include <uapi/linux/netfilter/nf_nat.h>
41
42 static struct workqueue_struct *act_ct_wq;
43 static struct rhashtable zones_ht;
44 static DEFINE_MUTEX(zones_mutex);
45
46 struct zones_ht_key {
47 struct net *net;
48 u16 zone;
49 };
50
51 struct tcf_ct_flow_table {
52 struct rhash_head node; /* In zones tables */
53
54 struct rcu_work rwork;
55 struct nf_flowtable nf_ft;
56 refcount_t ref;
57 struct zones_ht_key key;
58
59 bool dying;
60 };
61
62 static const struct rhashtable_params zones_params = {
63 .head_offset = offsetof(struct tcf_ct_flow_table, node),
64 .key_offset = offsetof(struct tcf_ct_flow_table, key),
65 .key_len = offsetofend(struct zones_ht_key, zone),
66 .automatic_shrinking = true,
67 };
68
69 static struct flow_action_entry *
tcf_ct_flow_table_flow_action_get_next(struct flow_action * flow_action)70 tcf_ct_flow_table_flow_action_get_next(struct flow_action *flow_action)
71 {
72 int i = flow_action->num_entries++;
73
74 return &flow_action->entries[i];
75 }
76
tcf_ct_add_mangle_action(struct flow_action * action,enum flow_action_mangle_base htype,u32 offset,u32 mask,u32 val)77 static void tcf_ct_add_mangle_action(struct flow_action *action,
78 enum flow_action_mangle_base htype,
79 u32 offset,
80 u32 mask,
81 u32 val)
82 {
83 struct flow_action_entry *entry;
84
85 entry = tcf_ct_flow_table_flow_action_get_next(action);
86 entry->id = FLOW_ACTION_MANGLE;
87 entry->mangle.htype = htype;
88 entry->mangle.mask = ~mask;
89 entry->mangle.offset = offset;
90 entry->mangle.val = val;
91 }
92
93 /* The following nat helper functions check if the inverted reverse tuple
94 * (target) is different then the current dir tuple - meaning nat for ports
95 * and/or ip is needed, and add the relevant mangle actions.
96 */
97 static void
tcf_ct_flow_table_add_action_nat_ipv4(const struct nf_conntrack_tuple * tuple,struct nf_conntrack_tuple target,struct flow_action * action)98 tcf_ct_flow_table_add_action_nat_ipv4(const struct nf_conntrack_tuple *tuple,
99 struct nf_conntrack_tuple target,
100 struct flow_action *action)
101 {
102 if (memcmp(&target.src.u3, &tuple->src.u3, sizeof(target.src.u3)))
103 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_IP4,
104 offsetof(struct iphdr, saddr),
105 0xFFFFFFFF,
106 be32_to_cpu(target.src.u3.ip));
107 if (memcmp(&target.dst.u3, &tuple->dst.u3, sizeof(target.dst.u3)))
108 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_IP4,
109 offsetof(struct iphdr, daddr),
110 0xFFFFFFFF,
111 be32_to_cpu(target.dst.u3.ip));
112 }
113
114 static void
tcf_ct_add_ipv6_addr_mangle_action(struct flow_action * action,union nf_inet_addr * addr,u32 offset)115 tcf_ct_add_ipv6_addr_mangle_action(struct flow_action *action,
116 union nf_inet_addr *addr,
117 u32 offset)
118 {
119 int i;
120
121 for (i = 0; i < sizeof(struct in6_addr) / sizeof(u32); i++)
122 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_IP6,
123 i * sizeof(u32) + offset,
124 0xFFFFFFFF, be32_to_cpu(addr->ip6[i]));
125 }
126
127 static void
tcf_ct_flow_table_add_action_nat_ipv6(const struct nf_conntrack_tuple * tuple,struct nf_conntrack_tuple target,struct flow_action * action)128 tcf_ct_flow_table_add_action_nat_ipv6(const struct nf_conntrack_tuple *tuple,
129 struct nf_conntrack_tuple target,
130 struct flow_action *action)
131 {
132 if (memcmp(&target.src.u3, &tuple->src.u3, sizeof(target.src.u3)))
133 tcf_ct_add_ipv6_addr_mangle_action(action, &target.src.u3,
134 offsetof(struct ipv6hdr,
135 saddr));
136 if (memcmp(&target.dst.u3, &tuple->dst.u3, sizeof(target.dst.u3)))
137 tcf_ct_add_ipv6_addr_mangle_action(action, &target.dst.u3,
138 offsetof(struct ipv6hdr,
139 daddr));
140 }
141
142 static void
tcf_ct_flow_table_add_action_nat_tcp(const struct nf_conntrack_tuple * tuple,struct nf_conntrack_tuple target,struct flow_action * action)143 tcf_ct_flow_table_add_action_nat_tcp(const struct nf_conntrack_tuple *tuple,
144 struct nf_conntrack_tuple target,
145 struct flow_action *action)
146 {
147 __be16 target_src = target.src.u.tcp.port;
148 __be16 target_dst = target.dst.u.tcp.port;
149
150 if (target_src != tuple->src.u.tcp.port)
151 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_TCP,
152 offsetof(struct tcphdr, source),
153 0xFFFF, be16_to_cpu(target_src));
154 if (target_dst != tuple->dst.u.tcp.port)
155 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_TCP,
156 offsetof(struct tcphdr, dest),
157 0xFFFF, be16_to_cpu(target_dst));
158 }
159
160 static void
tcf_ct_flow_table_add_action_nat_udp(const struct nf_conntrack_tuple * tuple,struct nf_conntrack_tuple target,struct flow_action * action)161 tcf_ct_flow_table_add_action_nat_udp(const struct nf_conntrack_tuple *tuple,
162 struct nf_conntrack_tuple target,
163 struct flow_action *action)
164 {
165 __be16 target_src = target.src.u.udp.port;
166 __be16 target_dst = target.dst.u.udp.port;
167
168 if (target_src != tuple->src.u.udp.port)
169 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_UDP,
170 offsetof(struct udphdr, source),
171 0xFFFF, be16_to_cpu(target_src));
172 if (target_dst != tuple->dst.u.udp.port)
173 tcf_ct_add_mangle_action(action, FLOW_ACT_MANGLE_HDR_TYPE_UDP,
174 offsetof(struct udphdr, dest),
175 0xFFFF, be16_to_cpu(target_dst));
176 }
177
tcf_ct_flow_table_add_action_meta(struct nf_conn * ct,enum ip_conntrack_dir dir,enum ip_conntrack_info ctinfo,struct flow_action * action)178 static void tcf_ct_flow_table_add_action_meta(struct nf_conn *ct,
179 enum ip_conntrack_dir dir,
180 enum ip_conntrack_info ctinfo,
181 struct flow_action *action)
182 {
183 struct nf_conn_labels *ct_labels;
184 struct flow_action_entry *entry;
185 u32 *act_ct_labels;
186
187 entry = tcf_ct_flow_table_flow_action_get_next(action);
188 entry->id = FLOW_ACTION_CT_METADATA;
189 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
190 entry->ct_metadata.mark = READ_ONCE(ct->mark);
191 #endif
192 /* aligns with the CT reference on the SKB nf_ct_set */
193 entry->ct_metadata.cookie = (unsigned long)ct | ctinfo;
194 entry->ct_metadata.orig_dir = dir == IP_CT_DIR_ORIGINAL;
195
196 act_ct_labels = entry->ct_metadata.labels;
197 ct_labels = nf_ct_labels_find(ct);
198 if (ct_labels)
199 memcpy(act_ct_labels, ct_labels->bits, NF_CT_LABELS_MAX_SIZE);
200 else
201 memset(act_ct_labels, 0, NF_CT_LABELS_MAX_SIZE);
202 }
203
tcf_ct_flow_table_add_action_nat(struct net * net,struct nf_conn * ct,enum ip_conntrack_dir dir,struct flow_action * action)204 static int tcf_ct_flow_table_add_action_nat(struct net *net,
205 struct nf_conn *ct,
206 enum ip_conntrack_dir dir,
207 struct flow_action *action)
208 {
209 const struct nf_conntrack_tuple *tuple = &ct->tuplehash[dir].tuple;
210 struct nf_conntrack_tuple target;
211
212 if (!(ct->status & IPS_NAT_MASK))
213 return 0;
214
215 nf_ct_invert_tuple(&target, &ct->tuplehash[!dir].tuple);
216
217 switch (tuple->src.l3num) {
218 case NFPROTO_IPV4:
219 tcf_ct_flow_table_add_action_nat_ipv4(tuple, target,
220 action);
221 break;
222 case NFPROTO_IPV6:
223 tcf_ct_flow_table_add_action_nat_ipv6(tuple, target,
224 action);
225 break;
226 default:
227 return -EOPNOTSUPP;
228 }
229
230 switch (nf_ct_protonum(ct)) {
231 case IPPROTO_TCP:
232 tcf_ct_flow_table_add_action_nat_tcp(tuple, target, action);
233 break;
234 case IPPROTO_UDP:
235 tcf_ct_flow_table_add_action_nat_udp(tuple, target, action);
236 break;
237 default:
238 return -EOPNOTSUPP;
239 }
240
241 return 0;
242 }
243
tcf_ct_flow_table_fill_actions(struct net * net,struct flow_offload * flow,enum flow_offload_tuple_dir tdir,struct nf_flow_rule * flow_rule)244 static int tcf_ct_flow_table_fill_actions(struct net *net,
245 struct flow_offload *flow,
246 enum flow_offload_tuple_dir tdir,
247 struct nf_flow_rule *flow_rule)
248 {
249 struct flow_action *action = &flow_rule->rule->action;
250 int num_entries = action->num_entries;
251 struct nf_conn *ct = flow->ct;
252 enum ip_conntrack_info ctinfo;
253 enum ip_conntrack_dir dir;
254 int i, err;
255
256 switch (tdir) {
257 case FLOW_OFFLOAD_DIR_ORIGINAL:
258 dir = IP_CT_DIR_ORIGINAL;
259 ctinfo = test_bit(IPS_SEEN_REPLY_BIT, &ct->status) ?
260 IP_CT_ESTABLISHED : IP_CT_NEW;
261 if (ctinfo == IP_CT_ESTABLISHED)
262 set_bit(NF_FLOW_HW_ESTABLISHED, &flow->flags);
263 break;
264 case FLOW_OFFLOAD_DIR_REPLY:
265 dir = IP_CT_DIR_REPLY;
266 ctinfo = IP_CT_ESTABLISHED_REPLY;
267 break;
268 default:
269 return -EOPNOTSUPP;
270 }
271
272 err = tcf_ct_flow_table_add_action_nat(net, ct, dir, action);
273 if (err)
274 goto err_nat;
275
276 tcf_ct_flow_table_add_action_meta(ct, dir, ctinfo, action);
277 return 0;
278
279 err_nat:
280 /* Clear filled actions */
281 for (i = num_entries; i < action->num_entries; i++)
282 memset(&action->entries[i], 0, sizeof(action->entries[i]));
283 action->num_entries = num_entries;
284
285 return err;
286 }
287
tcf_ct_flow_is_outdated(const struct flow_offload * flow)288 static bool tcf_ct_flow_is_outdated(const struct flow_offload *flow)
289 {
290 return test_bit(IPS_SEEN_REPLY_BIT, &flow->ct->status) &&
291 test_bit(IPS_HW_OFFLOAD_BIT, &flow->ct->status) &&
292 !test_bit(NF_FLOW_HW_PENDING, &flow->flags) &&
293 !test_bit(NF_FLOW_HW_ESTABLISHED, &flow->flags);
294 }
295
296 static void tcf_ct_flow_table_get_ref(struct tcf_ct_flow_table *ct_ft);
297
tcf_ct_nf_get(struct nf_flowtable * ft)298 static void tcf_ct_nf_get(struct nf_flowtable *ft)
299 {
300 struct tcf_ct_flow_table *ct_ft =
301 container_of(ft, struct tcf_ct_flow_table, nf_ft);
302
303 tcf_ct_flow_table_get_ref(ct_ft);
304 }
305
306 static void tcf_ct_flow_table_put(struct tcf_ct_flow_table *ct_ft);
307
tcf_ct_nf_put(struct nf_flowtable * ft)308 static void tcf_ct_nf_put(struct nf_flowtable *ft)
309 {
310 struct tcf_ct_flow_table *ct_ft =
311 container_of(ft, struct tcf_ct_flow_table, nf_ft);
312
313 tcf_ct_flow_table_put(ct_ft);
314 }
315
316 static struct nf_flowtable_type flowtable_ct = {
317 .gc = tcf_ct_flow_is_outdated,
318 .action = tcf_ct_flow_table_fill_actions,
319 .get = tcf_ct_nf_get,
320 .put = tcf_ct_nf_put,
321 .owner = THIS_MODULE,
322 };
323
tcf_ct_flow_table_get(struct net * net,struct tcf_ct_params * params)324 static int tcf_ct_flow_table_get(struct net *net, struct tcf_ct_params *params)
325 {
326 struct zones_ht_key key = { .net = net, .zone = params->zone };
327 struct tcf_ct_flow_table *ct_ft;
328 int err = -ENOMEM;
329
330 mutex_lock(&zones_mutex);
331 ct_ft = rhashtable_lookup_fast(&zones_ht, &key, zones_params);
332 if (ct_ft && refcount_inc_not_zero(&ct_ft->ref))
333 goto out_unlock;
334
335 ct_ft = kzalloc_obj(*ct_ft);
336 if (!ct_ft)
337 goto err_alloc;
338 refcount_set(&ct_ft->ref, 1);
339
340 ct_ft->key = key;
341 err = rhashtable_insert_fast(&zones_ht, &ct_ft->node, zones_params);
342 if (err)
343 goto err_insert;
344
345 ct_ft->nf_ft.type = &flowtable_ct;
346 ct_ft->nf_ft.flags |= NF_FLOWTABLE_HW_OFFLOAD |
347 NF_FLOWTABLE_COUNTER;
348 err = nf_flow_table_init(&ct_ft->nf_ft);
349 if (err)
350 goto err_init;
351 write_pnet(&ct_ft->nf_ft.net, net);
352
353 __module_get(THIS_MODULE);
354 out_unlock:
355 params->ct_ft = ct_ft;
356 params->nf_ft = &ct_ft->nf_ft;
357 mutex_unlock(&zones_mutex);
358
359 return 0;
360
361 err_init:
362 rhashtable_remove_fast(&zones_ht, &ct_ft->node, zones_params);
363 err_insert:
364 kfree(ct_ft);
365 err_alloc:
366 mutex_unlock(&zones_mutex);
367 return err;
368 }
369
tcf_ct_flow_table_get_ref(struct tcf_ct_flow_table * ct_ft)370 static void tcf_ct_flow_table_get_ref(struct tcf_ct_flow_table *ct_ft)
371 {
372 refcount_inc(&ct_ft->ref);
373 }
374
tcf_ct_flow_table_cleanup_work(struct work_struct * work)375 static void tcf_ct_flow_table_cleanup_work(struct work_struct *work)
376 {
377 struct tcf_ct_flow_table *ct_ft;
378 struct flow_block *block;
379
380 ct_ft = container_of(to_rcu_work(work), struct tcf_ct_flow_table,
381 rwork);
382 nf_flow_table_free(&ct_ft->nf_ft);
383
384 block = &ct_ft->nf_ft.flow_block;
385 down_write(&ct_ft->nf_ft.flow_block_lock);
386 WARN_ON(!list_empty(&block->cb_list));
387 up_write(&ct_ft->nf_ft.flow_block_lock);
388 kfree(ct_ft);
389
390 module_put(THIS_MODULE);
391 }
392
tcf_ct_flow_table_put(struct tcf_ct_flow_table * ct_ft)393 static void tcf_ct_flow_table_put(struct tcf_ct_flow_table *ct_ft)
394 {
395 if (refcount_dec_and_test(&ct_ft->ref)) {
396 rhashtable_remove_fast(&zones_ht, &ct_ft->node, zones_params);
397 INIT_RCU_WORK(&ct_ft->rwork, tcf_ct_flow_table_cleanup_work);
398 queue_rcu_work(act_ct_wq, &ct_ft->rwork);
399 }
400 }
401
tcf_ct_flow_tc_ifidx(struct flow_offload * entry,struct nf_conn_act_ct_ext * act_ct_ext,u8 dir)402 static void tcf_ct_flow_tc_ifidx(struct flow_offload *entry,
403 struct nf_conn_act_ct_ext *act_ct_ext, u8 dir)
404 {
405 entry->tuplehash[dir].tuple.xmit_type = FLOW_OFFLOAD_XMIT_TC;
406 entry->tuplehash[dir].tuple.tc.iifidx = act_ct_ext->ifindex[dir];
407 }
408
tcf_ct_flow_ct_ext_ifidx_update(struct flow_offload * entry)409 static void tcf_ct_flow_ct_ext_ifidx_update(struct flow_offload *entry)
410 {
411 struct nf_conn_act_ct_ext *act_ct_ext;
412
413 act_ct_ext = nf_conn_act_ct_ext_find(entry->ct);
414 if (act_ct_ext) {
415 tcf_ct_flow_tc_ifidx(entry, act_ct_ext, FLOW_OFFLOAD_DIR_ORIGINAL);
416 tcf_ct_flow_tc_ifidx(entry, act_ct_ext, FLOW_OFFLOAD_DIR_REPLY);
417 }
418 }
419
tcf_ct_flow_table_add(struct tcf_ct_flow_table * ct_ft,struct nf_conn * ct,bool tcp,bool bidirectional)420 static void tcf_ct_flow_table_add(struct tcf_ct_flow_table *ct_ft,
421 struct nf_conn *ct,
422 bool tcp, bool bidirectional)
423 {
424 struct nf_conn_act_ct_ext *act_ct_ext;
425 struct flow_offload *entry;
426 int err;
427
428 if (test_and_set_bit(IPS_OFFLOAD_BIT, &ct->status))
429 return;
430
431 entry = flow_offload_alloc(ct);
432 if (!entry) {
433 WARN_ON_ONCE(1);
434 goto err_alloc;
435 }
436
437 if (tcp) {
438 ct->proto.tcp.seen[0].flags |= IP_CT_TCP_FLAG_BE_LIBERAL;
439 ct->proto.tcp.seen[1].flags |= IP_CT_TCP_FLAG_BE_LIBERAL;
440 }
441 if (bidirectional)
442 __set_bit(NF_FLOW_HW_BIDIRECTIONAL, &entry->flags);
443
444 act_ct_ext = nf_conn_act_ct_ext_find(ct);
445 if (act_ct_ext) {
446 tcf_ct_flow_tc_ifidx(entry, act_ct_ext, FLOW_OFFLOAD_DIR_ORIGINAL);
447 tcf_ct_flow_tc_ifidx(entry, act_ct_ext, FLOW_OFFLOAD_DIR_REPLY);
448 }
449
450 err = flow_offload_add(&ct_ft->nf_ft, entry);
451 if (err)
452 goto err_add;
453
454 return;
455
456 err_add:
457 flow_offload_free(entry);
458 err_alloc:
459 clear_bit(IPS_OFFLOAD_BIT, &ct->status);
460 }
461
tcf_ct_flow_table_process_conn(struct tcf_ct_flow_table * ct_ft,struct nf_conn * ct,enum ip_conntrack_info ctinfo)462 static void tcf_ct_flow_table_process_conn(struct tcf_ct_flow_table *ct_ft,
463 struct nf_conn *ct,
464 enum ip_conntrack_info ctinfo)
465 {
466 bool tcp = false, bidirectional = true;
467
468 switch (nf_ct_protonum(ct)) {
469 case IPPROTO_TCP:
470 if ((ctinfo != IP_CT_ESTABLISHED &&
471 ctinfo != IP_CT_ESTABLISHED_REPLY) ||
472 !test_bit(IPS_ASSURED_BIT, &ct->status) ||
473 ct->proto.tcp.state != TCP_CONNTRACK_ESTABLISHED)
474 return;
475
476 tcp = true;
477 break;
478 case IPPROTO_UDP:
479 if (!nf_ct_is_confirmed(ct))
480 return;
481 if (!test_bit(IPS_ASSURED_BIT, &ct->status))
482 bidirectional = false;
483 break;
484 #ifdef CONFIG_NF_CT_PROTO_GRE
485 case IPPROTO_GRE: {
486 struct nf_conntrack_tuple *tuple;
487
488 if ((ctinfo != IP_CT_ESTABLISHED &&
489 ctinfo != IP_CT_ESTABLISHED_REPLY) ||
490 !test_bit(IPS_ASSURED_BIT, &ct->status) ||
491 ct->status & IPS_NAT_MASK)
492 return;
493
494 tuple = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
495 /* No support for GRE v1 */
496 if (tuple->src.u.gre.key || tuple->dst.u.gre.key)
497 return;
498 break;
499 }
500 #endif
501 default:
502 return;
503 }
504
505 if (nf_ct_ext_exist(ct, NF_CT_EXT_HELPER) ||
506 ct->status & IPS_SEQ_ADJUST)
507 return;
508
509 tcf_ct_flow_table_add(ct_ft, ct, tcp, bidirectional);
510 }
511
512 static bool
tcf_ct_flow_table_fill_tuple_ipv4(struct sk_buff * skb,struct flow_offload_tuple * tuple,struct tcphdr ** tcph)513 tcf_ct_flow_table_fill_tuple_ipv4(struct sk_buff *skb,
514 struct flow_offload_tuple *tuple,
515 struct tcphdr **tcph)
516 {
517 struct flow_ports *ports;
518 unsigned int thoff;
519 struct iphdr *iph;
520 size_t hdrsize;
521 u8 ipproto;
522
523 if (!pskb_network_may_pull(skb, sizeof(*iph)))
524 return false;
525
526 iph = ip_hdr(skb);
527 thoff = iph->ihl * 4;
528
529 if (ip_is_fragment(iph) ||
530 unlikely(thoff != sizeof(struct iphdr)))
531 return false;
532
533 ipproto = iph->protocol;
534 switch (ipproto) {
535 case IPPROTO_TCP:
536 hdrsize = sizeof(struct tcphdr);
537 break;
538 case IPPROTO_UDP:
539 hdrsize = sizeof(*ports);
540 break;
541 #ifdef CONFIG_NF_CT_PROTO_GRE
542 case IPPROTO_GRE:
543 hdrsize = sizeof(struct gre_base_hdr);
544 break;
545 #endif
546 default:
547 return false;
548 }
549
550 if (iph->ttl <= 1)
551 return false;
552
553 if (!pskb_network_may_pull(skb, thoff + hdrsize))
554 return false;
555
556 switch (ipproto) {
557 case IPPROTO_TCP:
558 *tcph = (void *)(skb_network_header(skb) + thoff);
559 fallthrough;
560 case IPPROTO_UDP:
561 ports = (struct flow_ports *)(skb_network_header(skb) + thoff);
562 tuple->src_port = ports->source;
563 tuple->dst_port = ports->dest;
564 break;
565 case IPPROTO_GRE: {
566 struct gre_base_hdr *greh;
567
568 greh = (struct gre_base_hdr *)(skb_network_header(skb) + thoff);
569 if ((greh->flags & GRE_VERSION) != GRE_VERSION_0)
570 return false;
571 break;
572 }
573 }
574
575 iph = ip_hdr(skb);
576
577 tuple->src_v4.s_addr = iph->saddr;
578 tuple->dst_v4.s_addr = iph->daddr;
579 tuple->l3proto = AF_INET;
580 tuple->l4proto = ipproto;
581
582 return true;
583 }
584
585 static bool
tcf_ct_flow_table_fill_tuple_ipv6(struct sk_buff * skb,struct flow_offload_tuple * tuple,struct tcphdr ** tcph)586 tcf_ct_flow_table_fill_tuple_ipv6(struct sk_buff *skb,
587 struct flow_offload_tuple *tuple,
588 struct tcphdr **tcph)
589 {
590 struct flow_ports *ports;
591 struct ipv6hdr *ip6h;
592 unsigned int thoff;
593 size_t hdrsize;
594 u8 nexthdr;
595
596 if (!pskb_network_may_pull(skb, sizeof(*ip6h)))
597 return false;
598
599 ip6h = ipv6_hdr(skb);
600 thoff = sizeof(*ip6h);
601
602 nexthdr = ip6h->nexthdr;
603 switch (nexthdr) {
604 case IPPROTO_TCP:
605 hdrsize = sizeof(struct tcphdr);
606 break;
607 case IPPROTO_UDP:
608 hdrsize = sizeof(*ports);
609 break;
610 #ifdef CONFIG_NF_CT_PROTO_GRE
611 case IPPROTO_GRE:
612 hdrsize = sizeof(struct gre_base_hdr);
613 break;
614 #endif
615 default:
616 return false;
617 }
618
619 if (ip6h->hop_limit <= 1)
620 return false;
621
622 if (!pskb_network_may_pull(skb, thoff + hdrsize))
623 return false;
624
625 switch (nexthdr) {
626 case IPPROTO_TCP:
627 *tcph = (void *)(skb_network_header(skb) + thoff);
628 fallthrough;
629 case IPPROTO_UDP:
630 ports = (struct flow_ports *)(skb_network_header(skb) + thoff);
631 tuple->src_port = ports->source;
632 tuple->dst_port = ports->dest;
633 break;
634 case IPPROTO_GRE: {
635 struct gre_base_hdr *greh;
636
637 greh = (struct gre_base_hdr *)(skb_network_header(skb) + thoff);
638 if ((greh->flags & GRE_VERSION) != GRE_VERSION_0)
639 return false;
640 break;
641 }
642 }
643
644 ip6h = ipv6_hdr(skb);
645
646 tuple->src_v6 = ip6h->saddr;
647 tuple->dst_v6 = ip6h->daddr;
648 tuple->l3proto = AF_INET6;
649 tuple->l4proto = nexthdr;
650
651 return true;
652 }
653
tcf_ct_flow_table_lookup(struct tcf_ct_params * p,struct sk_buff * skb,u8 family)654 static bool tcf_ct_flow_table_lookup(struct tcf_ct_params *p,
655 struct sk_buff *skb,
656 u8 family)
657 {
658 struct nf_flowtable *nf_ft = &p->ct_ft->nf_ft;
659 struct flow_offload_tuple_rhash *tuplehash;
660 struct flow_offload_tuple tuple = {};
661 enum ip_conntrack_info ctinfo;
662 struct tcphdr *tcph = NULL;
663 bool force_refresh = false;
664 struct flow_offload *flow;
665 struct nf_conn *ct;
666 u8 dir;
667
668 switch (family) {
669 case NFPROTO_IPV4:
670 if (!tcf_ct_flow_table_fill_tuple_ipv4(skb, &tuple, &tcph))
671 return false;
672 break;
673 case NFPROTO_IPV6:
674 if (!tcf_ct_flow_table_fill_tuple_ipv6(skb, &tuple, &tcph))
675 return false;
676 break;
677 default:
678 return false;
679 }
680
681 tuplehash = flow_offload_lookup(nf_ft, &tuple);
682 if (!tuplehash)
683 return false;
684
685 dir = tuplehash->tuple.dir;
686 flow = container_of(tuplehash, struct flow_offload, tuplehash[dir]);
687 ct = flow->ct;
688
689 if (dir == FLOW_OFFLOAD_DIR_REPLY &&
690 !test_bit(NF_FLOW_HW_BIDIRECTIONAL, &flow->flags)) {
691 /* Only offload reply direction after connection became
692 * assured.
693 */
694 if (test_bit(IPS_ASSURED_BIT, &ct->status))
695 set_bit(NF_FLOW_HW_BIDIRECTIONAL, &flow->flags);
696 else if (test_bit(NF_FLOW_HW_ESTABLISHED, &flow->flags))
697 /* If flow_table flow has already been updated to the
698 * established state, then don't refresh.
699 */
700 return false;
701 force_refresh = true;
702 }
703
704 if (tcph && (unlikely(tcph->fin || tcph->rst))) {
705 flow_offload_teardown(flow);
706 return false;
707 }
708
709 if (dir == FLOW_OFFLOAD_DIR_ORIGINAL)
710 ctinfo = test_bit(IPS_SEEN_REPLY_BIT, &ct->status) ?
711 IP_CT_ESTABLISHED : IP_CT_NEW;
712 else
713 ctinfo = IP_CT_ESTABLISHED_REPLY;
714
715 nf_conn_act_ct_ext_fill(skb, ct, ctinfo);
716 tcf_ct_flow_ct_ext_ifidx_update(flow);
717 flow_offload_refresh(nf_ft, flow, force_refresh);
718 if (!test_bit(IPS_ASSURED_BIT, &ct->status)) {
719 /* Process this flow in SW to allow promoting to ASSURED */
720 return false;
721 }
722
723 nf_conntrack_get(&ct->ct_general);
724 nf_ct_set(skb, ct, ctinfo);
725 if (nf_ft->flags & NF_FLOWTABLE_COUNTER)
726 nf_ct_acct_update(ct, dir, skb->len);
727
728 return true;
729 }
730
tcf_ct_flow_tables_init(void)731 static int tcf_ct_flow_tables_init(void)
732 {
733 return rhashtable_init(&zones_ht, &zones_params);
734 }
735
tcf_ct_flow_tables_uninit(void)736 static void tcf_ct_flow_tables_uninit(void)
737 {
738 rhashtable_destroy(&zones_ht);
739 }
740
741 static struct tc_action_ops act_ct_ops;
742
743 struct tc_ct_action_net {
744 struct tc_action_net tn; /* Must be first */
745 };
746
747 /* Determine whether skb->_nfct is equal to the result of conntrack lookup. */
tcf_ct_skb_nfct_cached(struct net * net,struct sk_buff * skb,struct tcf_ct_params * p)748 static bool tcf_ct_skb_nfct_cached(struct net *net, struct sk_buff *skb,
749 struct tcf_ct_params *p)
750 {
751 enum ip_conntrack_info ctinfo;
752 struct nf_conn *ct;
753
754 ct = nf_ct_get(skb, &ctinfo);
755 if (!ct)
756 return false;
757 if (!net_eq(net, read_pnet(&ct->ct_net)))
758 goto drop_ct;
759 if (nf_ct_zone(ct)->id != p->zone)
760 goto drop_ct;
761 if (p->helper) {
762 struct nf_conn_help *help;
763
764 help = nf_ct_ext_find(ct, NF_CT_EXT_HELPER);
765 if (help && rcu_access_pointer(help->helper) != p->helper)
766 goto drop_ct;
767 }
768
769 /* Force conntrack entry direction. */
770 if ((p->ct_action & TCA_CT_ACT_FORCE) &&
771 CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) {
772 if (nf_ct_is_confirmed(ct))
773 nf_ct_kill(ct);
774
775 goto drop_ct;
776 }
777
778 return true;
779
780 drop_ct:
781 nf_ct_put(ct);
782 nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
783
784 return false;
785 }
786
tcf_ct_skb_nf_family(struct sk_buff * skb)787 static u8 tcf_ct_skb_nf_family(struct sk_buff *skb)
788 {
789 u8 family = NFPROTO_UNSPEC;
790
791 switch (skb_protocol(skb, true)) {
792 case htons(ETH_P_IP):
793 family = NFPROTO_IPV4;
794 break;
795 case htons(ETH_P_IPV6):
796 family = NFPROTO_IPV6;
797 break;
798 default:
799 break;
800 }
801
802 return family;
803 }
804
tcf_ct_ipv4_is_fragment(struct sk_buff * skb,bool * frag)805 static int tcf_ct_ipv4_is_fragment(struct sk_buff *skb, bool *frag)
806 {
807 unsigned int len;
808
809 len = skb_network_offset(skb) + sizeof(struct iphdr);
810 if (unlikely(skb->len < len))
811 return -EINVAL;
812 if (unlikely(!pskb_may_pull(skb, len)))
813 return -ENOMEM;
814
815 *frag = ip_is_fragment(ip_hdr(skb));
816 return 0;
817 }
818
tcf_ct_ipv6_is_fragment(struct sk_buff * skb,bool * frag)819 static int tcf_ct_ipv6_is_fragment(struct sk_buff *skb, bool *frag)
820 {
821 unsigned int flags = 0, len, payload_ofs = 0;
822 unsigned short frag_off;
823 int nexthdr;
824
825 len = skb_network_offset(skb) + sizeof(struct ipv6hdr);
826 if (unlikely(skb->len < len))
827 return -EINVAL;
828 if (unlikely(!pskb_may_pull(skb, len)))
829 return -ENOMEM;
830
831 nexthdr = ipv6_find_hdr(skb, &payload_ofs, -1, &frag_off, &flags);
832 if (unlikely(nexthdr < 0))
833 return -EPROTO;
834
835 *frag = flags & IP6_FH_F_FRAG;
836 return 0;
837 }
838
tcf_ct_handle_fragments(struct net * net,struct sk_buff * skb,u8 family,u16 zone,bool * defrag)839 static int tcf_ct_handle_fragments(struct net *net, struct sk_buff *skb,
840 u8 family, u16 zone, bool *defrag)
841 {
842 enum ip_conntrack_info ctinfo;
843 struct nf_conn *ct;
844 int err = 0;
845 bool frag;
846 u8 proto;
847 u16 mru;
848
849 /* Previously seen (loopback)? Ignore. */
850 ct = nf_ct_get(skb, &ctinfo);
851 if ((ct && !nf_ct_is_template(ct)) || ctinfo == IP_CT_UNTRACKED)
852 return 0;
853
854 if (family == NFPROTO_IPV4)
855 err = tcf_ct_ipv4_is_fragment(skb, &frag);
856 else
857 err = tcf_ct_ipv6_is_fragment(skb, &frag);
858 if (err || !frag)
859 return err;
860
861 err = nf_ct_handle_fragments(net, skb, zone, family, &proto, &mru);
862 if (err)
863 return err;
864
865 *defrag = true;
866 tc_skb_cb(skb)->mru = mru;
867
868 return 0;
869 }
870
tcf_ct_params_free(struct tcf_ct_params * params)871 static void tcf_ct_params_free(struct tcf_ct_params *params)
872 {
873 if (params->helper) {
874 #if IS_ENABLED(CONFIG_NF_NAT)
875 if (params->ct_action & TCA_CT_ACT_NAT)
876 nf_nat_helper_put(params->helper);
877 #endif
878 nf_conntrack_helper_put(params->helper);
879 }
880 if (params->ct_ft)
881 tcf_ct_flow_table_put(params->ct_ft);
882 if (params->tmpl) {
883 if (params->put_labels)
884 nf_connlabels_put(nf_ct_net(params->tmpl));
885
886 nf_ct_put(params->tmpl);
887 }
888
889 kfree(params);
890 }
891
tcf_ct_params_free_rcu(struct rcu_head * head)892 static void tcf_ct_params_free_rcu(struct rcu_head *head)
893 {
894 struct tcf_ct_params *params;
895
896 params = container_of(head, struct tcf_ct_params, rcu);
897 tcf_ct_params_free(params);
898 }
899
tcf_ct_act_set_mark(struct nf_conn * ct,u32 mark,u32 mask)900 static void tcf_ct_act_set_mark(struct nf_conn *ct, u32 mark, u32 mask)
901 {
902 #if IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)
903 u32 new_mark;
904
905 if (!mask)
906 return;
907
908 new_mark = mark | (READ_ONCE(ct->mark) & ~(mask));
909 if (READ_ONCE(ct->mark) != new_mark) {
910 WRITE_ONCE(ct->mark, new_mark);
911 if (nf_ct_is_confirmed(ct))
912 nf_conntrack_event_cache(IPCT_MARK, ct);
913 }
914 #endif
915 }
916
tcf_ct_act_set_labels(struct nf_conn * ct,u32 * labels,u32 * labels_m)917 static void tcf_ct_act_set_labels(struct nf_conn *ct,
918 u32 *labels,
919 u32 *labels_m)
920 {
921 #if IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS)
922 size_t labels_sz = sizeof_field(struct tcf_ct_params, labels);
923
924 if (!memchr_inv(labels_m, 0, labels_sz))
925 return;
926
927 nf_connlabels_replace(ct, labels, labels_m, 4);
928 #endif
929 }
930
tcf_ct_act_nat(struct sk_buff * skb,struct nf_conn * ct,enum ip_conntrack_info ctinfo,int ct_action,struct nf_nat_range2 * range,bool commit)931 static int tcf_ct_act_nat(struct sk_buff *skb,
932 struct nf_conn *ct,
933 enum ip_conntrack_info ctinfo,
934 int ct_action,
935 struct nf_nat_range2 *range,
936 bool commit)
937 {
938 #if IS_ENABLED(CONFIG_NF_NAT)
939 int err, action = 0;
940
941 if (!(ct_action & TCA_CT_ACT_NAT))
942 return NF_ACCEPT;
943 if (ct_action & TCA_CT_ACT_NAT_SRC)
944 action |= BIT(NF_NAT_MANIP_SRC);
945 if (ct_action & TCA_CT_ACT_NAT_DST)
946 action |= BIT(NF_NAT_MANIP_DST);
947
948 err = nf_ct_nat(skb, ct, ctinfo, &action, range, commit);
949 if (err != NF_ACCEPT)
950 return err & NF_VERDICT_MASK;
951
952 if (action & BIT(NF_NAT_MANIP_SRC))
953 qdisc_skb_cb(skb)->post_ct_snat = 1;
954 if (action & BIT(NF_NAT_MANIP_DST))
955 qdisc_skb_cb(skb)->post_ct_dnat = 1;
956
957 return err;
958 #else
959 return NF_ACCEPT;
960 #endif
961 }
962
tcf_ct_act(struct sk_buff * skb,const struct tc_action * a,struct tcf_result * res)963 TC_INDIRECT_SCOPE int tcf_ct_act(struct sk_buff *skb, const struct tc_action *a,
964 struct tcf_result *res)
965 {
966 struct net *net = dev_net(skb->dev);
967 enum ip_conntrack_info ctinfo;
968 struct tcf_ct *c = to_ct(a);
969 struct nf_conn *tmpl = NULL;
970 struct nf_hook_state state;
971 bool cached, commit, clear;
972 int nh_ofs, err, retval;
973 struct tcf_ct_params *p;
974 bool add_helper = false;
975 bool skip_add = false;
976 bool defrag = false;
977 struct nf_conn *ct;
978 u8 family;
979
980 p = rcu_dereference_bh(c->params);
981
982 retval = p->action;
983 commit = p->ct_action & TCA_CT_ACT_COMMIT;
984 clear = p->ct_action & TCA_CT_ACT_CLEAR;
985 tmpl = p->tmpl;
986
987 tcf_lastuse_update(&c->tcf_tm);
988 tcf_action_update_bstats(&c->common, skb);
989
990 if (clear) {
991 qdisc_skb_cb(skb)->post_ct = false;
992 ct = nf_ct_get(skb, &ctinfo);
993 if (ct) {
994 nf_ct_put(ct);
995 nf_ct_set(skb, NULL, IP_CT_UNTRACKED);
996 }
997
998 goto out_clear;
999 }
1000
1001 family = tcf_ct_skb_nf_family(skb);
1002 if (family == NFPROTO_UNSPEC)
1003 goto drop;
1004
1005 /* The conntrack module expects to be working at L3.
1006 * We also try to pull the IPv4/6 header to linear area
1007 */
1008 nh_ofs = skb_network_offset(skb);
1009 skb_pull_rcsum(skb, nh_ofs);
1010 err = tcf_ct_handle_fragments(net, skb, family, p->zone, &defrag);
1011 if (err)
1012 goto out_frag;
1013
1014 err = nf_ct_skb_network_trim(skb, family);
1015 if (err)
1016 goto drop;
1017
1018 /* If we are recirculating packets to match on ct fields and
1019 * committing with a separate ct action, then we don't need to
1020 * actually run the packet through conntrack twice unless it's for a
1021 * different zone.
1022 */
1023 cached = tcf_ct_skb_nfct_cached(net, skb, p);
1024 if (!cached) {
1025 if (tcf_ct_flow_table_lookup(p, skb, family)) {
1026 skip_add = true;
1027 goto do_nat;
1028 }
1029
1030 /* Associate skb with specified zone. */
1031 if (tmpl) {
1032 nf_conntrack_put(skb_nfct(skb));
1033 nf_conntrack_get(&tmpl->ct_general);
1034 nf_ct_set(skb, tmpl, IP_CT_NEW);
1035 }
1036
1037 state.hook = NF_INET_PRE_ROUTING;
1038 state.net = net;
1039 state.pf = family;
1040 err = nf_conntrack_in(skb, &state);
1041 if (err != NF_ACCEPT)
1042 goto nf_error;
1043 }
1044
1045 do_nat:
1046 ct = nf_ct_get(skb, &ctinfo);
1047 if (!ct)
1048 goto out_push;
1049 nf_ct_deliver_cached_events(ct);
1050 nf_conn_act_ct_ext_fill(skb, ct, ctinfo);
1051
1052 err = tcf_ct_act_nat(skb, ct, ctinfo, p->ct_action, &p->range, commit);
1053 if (err != NF_ACCEPT)
1054 goto nf_error;
1055
1056 if (!nf_ct_is_confirmed(ct) && commit && p->helper && !nfct_help(ct)) {
1057 err = __nf_ct_try_assign_helper(ct, p->tmpl, GFP_ATOMIC);
1058 if (err)
1059 goto drop;
1060 add_helper = true;
1061 if (p->ct_action & TCA_CT_ACT_NAT && !nfct_seqadj(ct)) {
1062 if (!nfct_seqadj_ext_add(ct))
1063 goto drop;
1064 }
1065 }
1066
1067 if (nf_ct_is_confirmed(ct) ? ((!cached && !skip_add) || add_helper) : commit) {
1068 err = nf_ct_helper(skb, ct, ctinfo, family);
1069 if (err != NF_ACCEPT)
1070 goto nf_error;
1071 }
1072
1073 if (commit) {
1074 tcf_ct_act_set_mark(ct, p->mark, p->mark_mask);
1075 tcf_ct_act_set_labels(ct, p->labels, p->labels_mask);
1076
1077 if (!nf_ct_is_confirmed(ct))
1078 nf_conn_act_ct_ext_add(skb, ct, ctinfo);
1079
1080 /* This will take care of sending queued events
1081 * even if the connection is already confirmed.
1082 */
1083 err = nf_conntrack_confirm(skb);
1084 if (err != NF_ACCEPT)
1085 goto nf_error;
1086
1087 /* The ct may be dropped if a clash has been resolved,
1088 * so it's necessary to retrieve it from skb again to
1089 * prevent UAF.
1090 */
1091 ct = nf_ct_get(skb, &ctinfo);
1092 if (!ct)
1093 skip_add = true;
1094 }
1095
1096 if (!skip_add)
1097 tcf_ct_flow_table_process_conn(p->ct_ft, ct, ctinfo);
1098
1099 out_push:
1100 skb_push_rcsum(skb, nh_ofs);
1101
1102 qdisc_skb_cb(skb)->post_ct = true;
1103 tc_skb_cb(skb)->zone = p->zone;
1104 out_clear:
1105 if (defrag)
1106 qdisc_skb_cb(skb)->pkt_len = skb->len;
1107 return retval;
1108
1109 out_frag:
1110 if (err != -EINPROGRESS)
1111 tcf_action_inc_drop_qstats(&c->common);
1112 return TC_ACT_CONSUMED;
1113
1114 drop:
1115 tcf_action_inc_drop_qstats(&c->common);
1116 return TC_ACT_SHOT;
1117
1118 nf_error:
1119 /* some verdicts store extra data in upper bits, such
1120 * as errno or queue number.
1121 */
1122 switch (err & NF_VERDICT_MASK) {
1123 case NF_DROP:
1124 goto drop;
1125 case NF_STOLEN:
1126 tcf_action_inc_drop_qstats(&c->common);
1127 return TC_ACT_CONSUMED;
1128 default:
1129 DEBUG_NET_WARN_ON_ONCE(1);
1130 goto drop;
1131 }
1132 }
1133
1134 static const struct nla_policy ct_policy[TCA_CT_MAX + 1] = {
1135 [TCA_CT_ACTION] = { .type = NLA_U16 },
1136 [TCA_CT_PARMS] = NLA_POLICY_EXACT_LEN(sizeof(struct tc_ct)),
1137 [TCA_CT_ZONE] = { .type = NLA_U16 },
1138 [TCA_CT_MARK] = { .type = NLA_U32 },
1139 [TCA_CT_MARK_MASK] = { .type = NLA_U32 },
1140 [TCA_CT_LABELS] = { .type = NLA_BINARY,
1141 .len = 128 / BITS_PER_BYTE },
1142 [TCA_CT_LABELS_MASK] = { .type = NLA_BINARY,
1143 .len = 128 / BITS_PER_BYTE },
1144 [TCA_CT_NAT_IPV4_MIN] = { .type = NLA_U32 },
1145 [TCA_CT_NAT_IPV4_MAX] = { .type = NLA_U32 },
1146 [TCA_CT_NAT_IPV6_MIN] = NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)),
1147 [TCA_CT_NAT_IPV6_MAX] = NLA_POLICY_EXACT_LEN(sizeof(struct in6_addr)),
1148 [TCA_CT_NAT_PORT_MIN] = { .type = NLA_U16 },
1149 [TCA_CT_NAT_PORT_MAX] = { .type = NLA_U16 },
1150 [TCA_CT_HELPER_NAME] = { .type = NLA_STRING, .len = NF_CT_HELPER_NAME_LEN },
1151 [TCA_CT_HELPER_FAMILY] = { .type = NLA_U8 },
1152 [TCA_CT_HELPER_PROTO] = { .type = NLA_U8 },
1153 };
1154
tcf_ct_fill_params_nat(struct tcf_ct_params * p,struct tc_ct * parm,struct nlattr ** tb,struct netlink_ext_ack * extack)1155 static int tcf_ct_fill_params_nat(struct tcf_ct_params *p,
1156 struct tc_ct *parm,
1157 struct nlattr **tb,
1158 struct netlink_ext_ack *extack)
1159 {
1160 struct nf_nat_range2 *range;
1161
1162 if (!(p->ct_action & TCA_CT_ACT_NAT))
1163 return 0;
1164
1165 if (!IS_ENABLED(CONFIG_NF_NAT)) {
1166 NL_SET_ERR_MSG_MOD(extack, "Netfilter nat isn't enabled in kernel");
1167 return -EOPNOTSUPP;
1168 }
1169
1170 if (!(p->ct_action & (TCA_CT_ACT_NAT_SRC | TCA_CT_ACT_NAT_DST)))
1171 return 0;
1172
1173 if ((p->ct_action & TCA_CT_ACT_NAT_SRC) &&
1174 (p->ct_action & TCA_CT_ACT_NAT_DST)) {
1175 NL_SET_ERR_MSG_MOD(extack, "dnat and snat can't be enabled at the same time");
1176 return -EOPNOTSUPP;
1177 }
1178
1179 range = &p->range;
1180 if (tb[TCA_CT_NAT_IPV4_MIN]) {
1181 struct nlattr *max_attr = tb[TCA_CT_NAT_IPV4_MAX];
1182
1183 p->ipv4_range = true;
1184 range->flags |= NF_NAT_RANGE_MAP_IPS;
1185 range->min_addr.ip =
1186 nla_get_in_addr(tb[TCA_CT_NAT_IPV4_MIN]);
1187
1188 range->max_addr.ip =
1189 nla_get_in_addr_default(max_attr, range->min_addr.ip);
1190 } else if (tb[TCA_CT_NAT_IPV6_MIN]) {
1191 struct nlattr *max_attr = tb[TCA_CT_NAT_IPV6_MAX];
1192
1193 p->ipv4_range = false;
1194 range->flags |= NF_NAT_RANGE_MAP_IPS;
1195 range->min_addr.in6 =
1196 nla_get_in6_addr(tb[TCA_CT_NAT_IPV6_MIN]);
1197
1198 range->max_addr.in6 = max_attr ?
1199 nla_get_in6_addr(max_attr) :
1200 range->min_addr.in6;
1201 }
1202
1203 if (tb[TCA_CT_NAT_PORT_MIN]) {
1204 range->flags |= NF_NAT_RANGE_PROTO_SPECIFIED;
1205 range->min_proto.all = nla_get_be16(tb[TCA_CT_NAT_PORT_MIN]);
1206
1207 range->max_proto.all = tb[TCA_CT_NAT_PORT_MAX] ?
1208 nla_get_be16(tb[TCA_CT_NAT_PORT_MAX]) :
1209 range->min_proto.all;
1210 }
1211
1212 return 0;
1213 }
1214
tcf_ct_set_key_val(struct nlattr ** tb,void * val,int val_type,void * mask,int mask_type,int len)1215 static void tcf_ct_set_key_val(struct nlattr **tb,
1216 void *val, int val_type,
1217 void *mask, int mask_type,
1218 int len)
1219 {
1220 if (!tb[val_type])
1221 return;
1222 nla_memcpy(val, tb[val_type], len);
1223
1224 if (!mask)
1225 return;
1226
1227 if (mask_type == TCA_CT_UNSPEC || !tb[mask_type])
1228 memset(mask, 0xff, len);
1229 else
1230 nla_memcpy(mask, tb[mask_type], len);
1231 }
1232
tcf_ct_fill_params(struct net * net,struct tcf_ct_params * p,struct tc_ct * parm,struct nlattr ** tb,struct netlink_ext_ack * extack)1233 static int tcf_ct_fill_params(struct net *net,
1234 struct tcf_ct_params *p,
1235 struct tc_ct *parm,
1236 struct nlattr **tb,
1237 struct netlink_ext_ack *extack)
1238 {
1239 struct nf_conntrack_zone zone;
1240 int err, family, proto, len;
1241 bool put_labels = false;
1242 struct nf_conn *tmpl;
1243 char *name;
1244
1245 p->zone = NF_CT_DEFAULT_ZONE_ID;
1246
1247 tcf_ct_set_key_val(tb,
1248 &p->ct_action, TCA_CT_ACTION,
1249 NULL, TCA_CT_UNSPEC,
1250 sizeof(p->ct_action));
1251
1252 if (p->ct_action & TCA_CT_ACT_CLEAR)
1253 return 0;
1254
1255 err = tcf_ct_fill_params_nat(p, parm, tb, extack);
1256 if (err)
1257 return err;
1258
1259 if (tb[TCA_CT_MARK]) {
1260 if (!IS_ENABLED(CONFIG_NF_CONNTRACK_MARK)) {
1261 NL_SET_ERR_MSG_MOD(extack, "Conntrack mark isn't enabled.");
1262 return -EOPNOTSUPP;
1263 }
1264 tcf_ct_set_key_val(tb,
1265 &p->mark, TCA_CT_MARK,
1266 &p->mark_mask, TCA_CT_MARK_MASK,
1267 sizeof(p->mark));
1268 }
1269
1270 if (tb[TCA_CT_LABELS]) {
1271 unsigned int n_bits = sizeof_field(struct tcf_ct_params, labels) * 8;
1272
1273 if (!IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS)) {
1274 NL_SET_ERR_MSG_MOD(extack, "Conntrack labels isn't enabled.");
1275 return -EOPNOTSUPP;
1276 }
1277
1278 if (nf_connlabels_get(net, n_bits - 1)) {
1279 NL_SET_ERR_MSG_MOD(extack, "Failed to set connlabel length");
1280 return -EOPNOTSUPP;
1281 } else {
1282 put_labels = true;
1283 }
1284
1285 tcf_ct_set_key_val(tb,
1286 p->labels, TCA_CT_LABELS,
1287 p->labels_mask, TCA_CT_LABELS_MASK,
1288 sizeof(p->labels));
1289 }
1290
1291 if (tb[TCA_CT_ZONE]) {
1292 if (!IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES)) {
1293 NL_SET_ERR_MSG_MOD(extack, "Conntrack zones isn't enabled.");
1294 return -EOPNOTSUPP;
1295 }
1296
1297 tcf_ct_set_key_val(tb,
1298 &p->zone, TCA_CT_ZONE,
1299 NULL, TCA_CT_UNSPEC,
1300 sizeof(p->zone));
1301 }
1302
1303 nf_ct_zone_init(&zone, p->zone, NF_CT_DEFAULT_ZONE_DIR, 0);
1304 tmpl = nf_ct_tmpl_alloc(net, &zone, GFP_KERNEL);
1305 if (!tmpl) {
1306 NL_SET_ERR_MSG_MOD(extack, "Failed to allocate conntrack template");
1307 return -ENOMEM;
1308 }
1309 p->tmpl = tmpl;
1310 if (tb[TCA_CT_HELPER_NAME]) {
1311 name = nla_data(tb[TCA_CT_HELPER_NAME]);
1312 len = nla_len(tb[TCA_CT_HELPER_NAME]);
1313 if (len > 16 || name[len - 1] != '\0') {
1314 NL_SET_ERR_MSG_MOD(extack, "Failed to parse helper name.");
1315 err = -EINVAL;
1316 goto err;
1317 }
1318 family = nla_get_u8_default(tb[TCA_CT_HELPER_FAMILY], AF_INET);
1319 proto = nla_get_u8_default(tb[TCA_CT_HELPER_PROTO],
1320 IPPROTO_TCP);
1321 err = nf_ct_add_helper(tmpl, name, family, proto,
1322 p->ct_action & TCA_CT_ACT_NAT, &p->helper);
1323 if (err) {
1324 NL_SET_ERR_MSG_MOD(extack, "Failed to add helper");
1325 goto err;
1326 }
1327 }
1328
1329 p->put_labels = put_labels;
1330
1331 if (p->ct_action & TCA_CT_ACT_COMMIT)
1332 __set_bit(IPS_CONFIRMED_BIT, &tmpl->status);
1333 return 0;
1334 err:
1335 if (put_labels)
1336 nf_connlabels_put(net);
1337
1338 nf_ct_put(p->tmpl);
1339 p->tmpl = NULL;
1340 return err;
1341 }
1342
tcf_ct_init(struct net * net,struct nlattr * nla,struct nlattr * est,struct tc_action ** a,struct tcf_proto * tp,u32 flags,struct netlink_ext_ack * extack)1343 static int tcf_ct_init(struct net *net, struct nlattr *nla,
1344 struct nlattr *est, struct tc_action **a,
1345 struct tcf_proto *tp, u32 flags,
1346 struct netlink_ext_ack *extack)
1347 {
1348 struct tc_action_net *tn = net_generic(net, act_ct_ops.net_id);
1349 bool bind = flags & TCA_ACT_FLAGS_BIND;
1350 struct tcf_ct_params *params = NULL;
1351 struct nlattr *tb[TCA_CT_MAX + 1];
1352 struct tcf_chain *goto_ch = NULL;
1353 struct tc_ct *parm;
1354 struct tcf_ct *c;
1355 int err, res = 0;
1356 u32 index;
1357
1358 if (!nla) {
1359 NL_SET_ERR_MSG_MOD(extack, "Ct requires attributes to be passed");
1360 return -EINVAL;
1361 }
1362
1363 err = nla_parse_nested(tb, TCA_CT_MAX, nla, ct_policy, extack);
1364 if (err < 0)
1365 return err;
1366
1367 if (!tb[TCA_CT_PARMS]) {
1368 NL_SET_ERR_MSG_MOD(extack, "Missing required ct parameters");
1369 return -EINVAL;
1370 }
1371 parm = nla_data(tb[TCA_CT_PARMS]);
1372 index = parm->index;
1373 err = tcf_idr_check_alloc(tn, &index, a, bind);
1374 if (err < 0)
1375 return err;
1376
1377 if (!err) {
1378 err = tcf_idr_create_from_flags(tn, index, est, a,
1379 &act_ct_ops, bind, flags);
1380 if (err) {
1381 tcf_idr_cleanup(tn, index);
1382 return err;
1383 }
1384 res = ACT_P_CREATED;
1385 } else {
1386 if (bind)
1387 return ACT_P_BOUND;
1388
1389 if (!(flags & TCA_ACT_FLAGS_REPLACE)) {
1390 tcf_idr_release(*a, bind);
1391 return -EEXIST;
1392 }
1393 }
1394 err = tcf_action_check_ctrlact(parm->action, tp, &goto_ch, extack);
1395 if (err < 0)
1396 goto cleanup;
1397
1398 c = to_ct(*a);
1399
1400 params = kzalloc_obj(*params);
1401 if (unlikely(!params)) {
1402 err = -ENOMEM;
1403 goto cleanup;
1404 }
1405
1406 err = tcf_ct_fill_params(net, params, parm, tb, extack);
1407 if (err)
1408 goto cleanup;
1409
1410 err = tcf_ct_flow_table_get(net, params);
1411 if (err)
1412 goto cleanup;
1413
1414 params->action = parm->action;
1415 spin_lock_bh(&c->tcf_lock);
1416 goto_ch = tcf_action_set_ctrlact(*a, parm->action, goto_ch);
1417 params = rcu_replace_pointer(c->params, params,
1418 lockdep_is_held(&c->tcf_lock));
1419 spin_unlock_bh(&c->tcf_lock);
1420
1421 if (goto_ch)
1422 tcf_chain_put_by_act(goto_ch);
1423 if (params)
1424 call_rcu(¶ms->rcu, tcf_ct_params_free_rcu);
1425
1426 return res;
1427
1428 cleanup:
1429 if (goto_ch)
1430 tcf_chain_put_by_act(goto_ch);
1431 if (params)
1432 tcf_ct_params_free(params);
1433 tcf_idr_release(*a, bind);
1434 return err;
1435 }
1436
tcf_ct_cleanup(struct tc_action * a)1437 static void tcf_ct_cleanup(struct tc_action *a)
1438 {
1439 struct tcf_ct_params *params;
1440 struct tcf_ct *c = to_ct(a);
1441
1442 params = rcu_dereference_protected(c->params, 1);
1443 if (params)
1444 call_rcu(¶ms->rcu, tcf_ct_params_free_rcu);
1445 }
1446
tcf_ct_dump_key_val(struct sk_buff * skb,const void * val,int val_type,const void * mask,int mask_type,int len)1447 static int tcf_ct_dump_key_val(struct sk_buff *skb,
1448 const void *val, int val_type,
1449 const void *mask, int mask_type,
1450 int len)
1451 {
1452 int err;
1453
1454 if (mask && !memchr_inv(mask, 0, len))
1455 return 0;
1456
1457 err = nla_put(skb, val_type, len, val);
1458 if (err)
1459 return err;
1460
1461 if (mask_type != TCA_CT_UNSPEC) {
1462 err = nla_put(skb, mask_type, len, mask);
1463 if (err)
1464 return err;
1465 }
1466
1467 return 0;
1468 }
1469
tcf_ct_dump_nat(struct sk_buff * skb,const struct tcf_ct_params * p)1470 static int tcf_ct_dump_nat(struct sk_buff *skb, const struct tcf_ct_params *p)
1471 {
1472 const struct nf_nat_range2 *range = &p->range;
1473
1474 if (!(p->ct_action & TCA_CT_ACT_NAT))
1475 return 0;
1476
1477 if (!(p->ct_action & (TCA_CT_ACT_NAT_SRC | TCA_CT_ACT_NAT_DST)))
1478 return 0;
1479
1480 if (range->flags & NF_NAT_RANGE_MAP_IPS) {
1481 if (p->ipv4_range) {
1482 if (nla_put_in_addr(skb, TCA_CT_NAT_IPV4_MIN,
1483 range->min_addr.ip))
1484 return -1;
1485 if (nla_put_in_addr(skb, TCA_CT_NAT_IPV4_MAX,
1486 range->max_addr.ip))
1487 return -1;
1488 } else {
1489 if (nla_put_in6_addr(skb, TCA_CT_NAT_IPV6_MIN,
1490 &range->min_addr.in6))
1491 return -1;
1492 if (nla_put_in6_addr(skb, TCA_CT_NAT_IPV6_MAX,
1493 &range->max_addr.in6))
1494 return -1;
1495 }
1496 }
1497
1498 if (range->flags & NF_NAT_RANGE_PROTO_SPECIFIED) {
1499 if (nla_put_be16(skb, TCA_CT_NAT_PORT_MIN,
1500 range->min_proto.all))
1501 return -1;
1502 if (nla_put_be16(skb, TCA_CT_NAT_PORT_MAX,
1503 range->max_proto.all))
1504 return -1;
1505 }
1506
1507 return 0;
1508 }
1509
tcf_ct_dump_helper(struct sk_buff * skb,const struct nf_conntrack_helper * helper)1510 static int tcf_ct_dump_helper(struct sk_buff *skb,
1511 const struct nf_conntrack_helper *helper)
1512 {
1513 if (!helper)
1514 return 0;
1515
1516 if (nla_put_string(skb, TCA_CT_HELPER_NAME, helper->name) ||
1517 nla_put_u8(skb, TCA_CT_HELPER_FAMILY, helper->tuple.src.l3num) ||
1518 nla_put_u8(skb, TCA_CT_HELPER_PROTO, helper->tuple.dst.protonum))
1519 return -1;
1520
1521 return 0;
1522 }
1523
tcf_ct_dump(struct sk_buff * skb,struct tc_action * a,int bind,int ref)1524 static inline int tcf_ct_dump(struct sk_buff *skb, struct tc_action *a,
1525 int bind, int ref)
1526 {
1527 unsigned char *b = skb_tail_pointer(skb);
1528 const struct tcf_ct *c = to_ct(a);
1529 const struct tcf_ct_params *p;
1530 struct tc_ct opt = {
1531 .index = c->tcf_index,
1532 .refcnt = refcount_read(&c->tcf_refcnt) - ref,
1533 .bindcnt = atomic_read(&c->tcf_bindcnt) - bind,
1534 };
1535 struct tcf_t t;
1536
1537 rcu_read_lock();
1538 p = rcu_dereference(c->params);
1539 opt.action = p->action;
1540
1541 if (tcf_ct_dump_key_val(skb,
1542 &p->ct_action, TCA_CT_ACTION,
1543 NULL, TCA_CT_UNSPEC,
1544 sizeof(p->ct_action)))
1545 goto nla_put_failure;
1546
1547 if (p->ct_action & TCA_CT_ACT_CLEAR)
1548 goto skip_dump;
1549
1550 if (IS_ENABLED(CONFIG_NF_CONNTRACK_MARK) &&
1551 tcf_ct_dump_key_val(skb,
1552 &p->mark, TCA_CT_MARK,
1553 &p->mark_mask, TCA_CT_MARK_MASK,
1554 sizeof(p->mark)))
1555 goto nla_put_failure;
1556
1557 if (IS_ENABLED(CONFIG_NF_CONNTRACK_LABELS) &&
1558 tcf_ct_dump_key_val(skb,
1559 p->labels, TCA_CT_LABELS,
1560 p->labels_mask, TCA_CT_LABELS_MASK,
1561 sizeof(p->labels)))
1562 goto nla_put_failure;
1563
1564 if (IS_ENABLED(CONFIG_NF_CONNTRACK_ZONES) &&
1565 tcf_ct_dump_key_val(skb,
1566 &p->zone, TCA_CT_ZONE,
1567 NULL, TCA_CT_UNSPEC,
1568 sizeof(p->zone)))
1569 goto nla_put_failure;
1570
1571 if (tcf_ct_dump_nat(skb, p))
1572 goto nla_put_failure;
1573
1574 if (tcf_ct_dump_helper(skb, p->helper))
1575 goto nla_put_failure;
1576
1577 skip_dump:
1578 if (nla_put(skb, TCA_CT_PARMS, sizeof(opt), &opt))
1579 goto nla_put_failure;
1580
1581 tcf_tm_dump(&t, &c->tcf_tm);
1582 if (nla_put_64bit(skb, TCA_CT_TM, sizeof(t), &t, TCA_CT_PAD))
1583 goto nla_put_failure;
1584 rcu_read_unlock();
1585
1586 return skb->len;
1587 nla_put_failure:
1588 rcu_read_unlock();
1589 nlmsg_trim(skb, b);
1590 return -1;
1591 }
1592
tcf_stats_update(struct tc_action * a,u64 bytes,u64 packets,u64 drops,u64 lastuse,bool hw)1593 static void tcf_stats_update(struct tc_action *a, u64 bytes, u64 packets,
1594 u64 drops, u64 lastuse, bool hw)
1595 {
1596 struct tcf_ct *c = to_ct(a);
1597
1598 tcf_action_update_stats(a, bytes, packets, drops, hw);
1599 c->tcf_tm.lastuse = max_t(u64, c->tcf_tm.lastuse, lastuse);
1600 }
1601
tcf_ct_offload_act_setup(struct tc_action * act,void * entry_data,u32 * index_inc,bool bind,struct netlink_ext_ack * extack)1602 static int tcf_ct_offload_act_setup(struct tc_action *act, void *entry_data,
1603 u32 *index_inc, bool bind,
1604 struct netlink_ext_ack *extack)
1605 {
1606 if (bind) {
1607 struct flow_action_entry *entry = entry_data;
1608
1609 if (tcf_ct_helper(act))
1610 return -EOPNOTSUPP;
1611
1612 entry->id = FLOW_ACTION_CT;
1613 entry->ct.action = tcf_ct_action(act);
1614 entry->ct.zone = tcf_ct_zone(act);
1615 entry->ct.flow_table = tcf_ct_ft(act);
1616 *index_inc = 1;
1617 } else {
1618 struct flow_offload_action *fl_action = entry_data;
1619
1620 fl_action->id = FLOW_ACTION_CT;
1621 }
1622
1623 return 0;
1624 }
1625
1626 static struct tc_action_ops act_ct_ops = {
1627 .kind = "ct",
1628 .id = TCA_ID_CT,
1629 .owner = THIS_MODULE,
1630 .act = tcf_ct_act,
1631 .dump = tcf_ct_dump,
1632 .init = tcf_ct_init,
1633 .cleanup = tcf_ct_cleanup,
1634 .stats_update = tcf_stats_update,
1635 .offload_act_setup = tcf_ct_offload_act_setup,
1636 .size = sizeof(struct tcf_ct),
1637 };
1638 MODULE_ALIAS_NET_ACT("ct");
1639
ct_init_net(struct net * net)1640 static __net_init int ct_init_net(struct net *net)
1641 {
1642 struct tc_ct_action_net *tn = net_generic(net, act_ct_ops.net_id);
1643
1644 return tc_action_net_init(net, &tn->tn, &act_ct_ops);
1645 }
1646
ct_exit_net(struct list_head * net_list)1647 static void __net_exit ct_exit_net(struct list_head *net_list)
1648 {
1649 tc_action_net_exit(net_list, act_ct_ops.net_id);
1650 }
1651
1652 static struct pernet_operations ct_net_ops = {
1653 .init = ct_init_net,
1654 .exit_batch = ct_exit_net,
1655 .id = &act_ct_ops.net_id,
1656 .size = sizeof(struct tc_ct_action_net),
1657 };
1658
ct_init_module(void)1659 static int __init ct_init_module(void)
1660 {
1661 int err;
1662
1663 act_ct_wq = alloc_ordered_workqueue("act_ct_workqueue", 0);
1664 if (!act_ct_wq)
1665 return -ENOMEM;
1666
1667 err = tcf_ct_flow_tables_init();
1668 if (err)
1669 goto err_tbl_init;
1670
1671 err = tcf_register_action(&act_ct_ops, &ct_net_ops);
1672 if (err)
1673 goto err_register;
1674
1675 static_branch_inc(&tcf_frag_xmit_count);
1676
1677 return 0;
1678
1679 err_register:
1680 tcf_ct_flow_tables_uninit();
1681 err_tbl_init:
1682 destroy_workqueue(act_ct_wq);
1683 return err;
1684 }
1685
ct_cleanup_module(void)1686 static void __exit ct_cleanup_module(void)
1687 {
1688 static_branch_dec(&tcf_frag_xmit_count);
1689 tcf_unregister_action(&act_ct_ops, &ct_net_ops);
1690 tcf_ct_flow_tables_uninit();
1691 destroy_workqueue(act_ct_wq);
1692 }
1693
1694 module_init(ct_init_module);
1695 module_exit(ct_cleanup_module);
1696 MODULE_AUTHOR("Paul Blakey <paulb@mellanox.com>");
1697 MODULE_AUTHOR("Yossi Kuperman <yossiku@mellanox.com>");
1698 MODULE_AUTHOR("Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>");
1699 MODULE_DESCRIPTION("Connection tracking action");
1700 MODULE_LICENSE("GPL v2");
1701