1 // SPDX-License-Identifier: GPL-2.0
2 /* Connection tracking via netlink socket. Allows for user space
3 * protocol helpers and general trouble making from userspace.
4 *
5 * (C) 2001 by Jay Schulist <jschlst@samba.org>
6 * (C) 2002-2006 by Harald Welte <laforge@gnumonks.org>
7 * (C) 2003 by Patrick Mchardy <kaber@trash.net>
8 * (C) 2005-2012 by Pablo Neira Ayuso <pablo@netfilter.org>
9 *
10 * Initial connection tracking via netlink development funded and
11 * generally made possible by Network Robots, Inc. (www.networkrobots.com)
12 *
13 * Further development of this code funded by Astaro AG (http://www.astaro.com)
14 */
15
16 #include <linux/init.h>
17 #include <linux/module.h>
18 #include <linux/kernel.h>
19 #include <linux/rculist.h>
20 #include <linux/rculist_nulls.h>
21 #include <linux/types.h>
22 #include <linux/timer.h>
23 #include <linux/security.h>
24 #include <linux/skbuff.h>
25 #include <linux/errno.h>
26 #include <linux/netlink.h>
27 #include <linux/spinlock.h>
28 #include <linux/interrupt.h>
29 #include <linux/slab.h>
30 #include <linux/siphash.h>
31
32 #include <linux/netfilter.h>
33 #include <net/ipv6.h>
34 #include <net/netlink.h>
35 #include <net/sock.h>
36 #include <net/netfilter/nf_conntrack.h>
37 #include <net/netfilter/nf_conntrack_core.h>
38 #include <net/netfilter/nf_conntrack_expect.h>
39 #include <net/netfilter/nf_conntrack_helper.h>
40 #include <net/netfilter/nf_conntrack_seqadj.h>
41 #include <net/netfilter/nf_conntrack_l4proto.h>
42 #include <net/netfilter/nf_conntrack_tuple.h>
43 #include <net/netfilter/nf_conntrack_acct.h>
44 #include <net/netfilter/nf_conntrack_zones.h>
45 #include <net/netfilter/nf_conntrack_timestamp.h>
46 #include <net/netfilter/nf_conntrack_labels.h>
47 #include <net/netfilter/nf_conntrack_synproxy.h>
48 #if IS_ENABLED(CONFIG_NF_NAT)
49 #include <net/netfilter/nf_nat.h>
50 #include <net/netfilter/nf_nat_helper.h>
51 #endif
52
53 #include <linux/netfilter/nfnetlink.h>
54 #include <linux/netfilter/nfnetlink_conntrack.h>
55
56 #include "nf_internals.h"
57
58 MODULE_LICENSE("GPL");
59 MODULE_DESCRIPTION("List and change connection tracking table");
60
61 struct ctnetlink_list_dump_ctx {
62 unsigned long last_id;
63 unsigned int cpu;
64 bool done;
65 };
66
ctnetlink_dump_tuples_proto(struct sk_buff * skb,const struct nf_conntrack_tuple * tuple,const struct nf_conntrack_l4proto * l4proto)67 static int ctnetlink_dump_tuples_proto(struct sk_buff *skb,
68 const struct nf_conntrack_tuple *tuple,
69 const struct nf_conntrack_l4proto *l4proto)
70 {
71 int ret = 0;
72 struct nlattr *nest_parms;
73
74 nest_parms = nla_nest_start(skb, CTA_TUPLE_PROTO);
75 if (!nest_parms)
76 goto nla_put_failure;
77 if (nla_put_u8(skb, CTA_PROTO_NUM, tuple->dst.protonum))
78 goto nla_put_failure;
79
80 if (likely(l4proto->tuple_to_nlattr))
81 ret = l4proto->tuple_to_nlattr(skb, tuple);
82
83 nla_nest_end(skb, nest_parms);
84
85 return ret;
86
87 nla_put_failure:
88 return -1;
89 }
90
ipv4_tuple_to_nlattr(struct sk_buff * skb,const struct nf_conntrack_tuple * tuple)91 static int ipv4_tuple_to_nlattr(struct sk_buff *skb,
92 const struct nf_conntrack_tuple *tuple)
93 {
94 if (nla_put_in_addr(skb, CTA_IP_V4_SRC, tuple->src.u3.ip) ||
95 nla_put_in_addr(skb, CTA_IP_V4_DST, tuple->dst.u3.ip))
96 return -EMSGSIZE;
97 return 0;
98 }
99
ipv6_tuple_to_nlattr(struct sk_buff * skb,const struct nf_conntrack_tuple * tuple)100 static int ipv6_tuple_to_nlattr(struct sk_buff *skb,
101 const struct nf_conntrack_tuple *tuple)
102 {
103 if (nla_put_in6_addr(skb, CTA_IP_V6_SRC, &tuple->src.u3.in6) ||
104 nla_put_in6_addr(skb, CTA_IP_V6_DST, &tuple->dst.u3.in6))
105 return -EMSGSIZE;
106 return 0;
107 }
108
ctnetlink_dump_tuples_ip(struct sk_buff * skb,const struct nf_conntrack_tuple * tuple)109 static int ctnetlink_dump_tuples_ip(struct sk_buff *skb,
110 const struct nf_conntrack_tuple *tuple)
111 {
112 int ret = 0;
113 struct nlattr *nest_parms;
114
115 nest_parms = nla_nest_start(skb, CTA_TUPLE_IP);
116 if (!nest_parms)
117 goto nla_put_failure;
118
119 switch (tuple->src.l3num) {
120 case NFPROTO_IPV4:
121 ret = ipv4_tuple_to_nlattr(skb, tuple);
122 break;
123 case NFPROTO_IPV6:
124 ret = ipv6_tuple_to_nlattr(skb, tuple);
125 break;
126 }
127
128 nla_nest_end(skb, nest_parms);
129
130 return ret;
131
132 nla_put_failure:
133 return -1;
134 }
135
ctnetlink_dump_tuples(struct sk_buff * skb,const struct nf_conntrack_tuple * tuple)136 static int ctnetlink_dump_tuples(struct sk_buff *skb,
137 const struct nf_conntrack_tuple *tuple)
138 {
139 const struct nf_conntrack_l4proto *l4proto;
140 int ret;
141
142 rcu_read_lock();
143 ret = ctnetlink_dump_tuples_ip(skb, tuple);
144
145 if (ret >= 0) {
146 l4proto = nf_ct_l4proto_find(tuple->dst.protonum);
147 ret = ctnetlink_dump_tuples_proto(skb, tuple, l4proto);
148 }
149 rcu_read_unlock();
150 return ret;
151 }
152
ctnetlink_dump_zone_id(struct sk_buff * skb,int attrtype,const struct nf_conntrack_zone * zone,int dir)153 static int ctnetlink_dump_zone_id(struct sk_buff *skb, int attrtype,
154 const struct nf_conntrack_zone *zone, int dir)
155 {
156 if (zone->id == NF_CT_DEFAULT_ZONE_ID || zone->dir != dir)
157 return 0;
158 if (nla_put_be16(skb, attrtype, htons(zone->id)))
159 goto nla_put_failure;
160 return 0;
161
162 nla_put_failure:
163 return -1;
164 }
165
ctnetlink_dump_status(struct sk_buff * skb,const struct nf_conn * ct)166 static int ctnetlink_dump_status(struct sk_buff *skb, const struct nf_conn *ct)
167 {
168 if (nla_put_be32(skb, CTA_STATUS, htonl(ct->status)))
169 goto nla_put_failure;
170 return 0;
171
172 nla_put_failure:
173 return -1;
174 }
175
ctnetlink_dump_timeout(struct sk_buff * skb,const struct nf_conn * ct,bool skip_zero)176 static int ctnetlink_dump_timeout(struct sk_buff *skb, const struct nf_conn *ct,
177 bool skip_zero)
178 {
179 long timeout;
180
181 if (nf_ct_is_confirmed(ct))
182 timeout = nf_ct_expires(ct) / HZ;
183 else
184 timeout = ct->timeout / HZ;
185
186 if (skip_zero && timeout == 0)
187 return 0;
188
189 if (nla_put_be32(skb, CTA_TIMEOUT, htonl(timeout)))
190 goto nla_put_failure;
191 return 0;
192
193 nla_put_failure:
194 return -1;
195 }
196
ctnetlink_dump_protoinfo(struct sk_buff * skb,struct nf_conn * ct,bool destroy)197 static int ctnetlink_dump_protoinfo(struct sk_buff *skb, struct nf_conn *ct,
198 bool destroy)
199 {
200 const struct nf_conntrack_l4proto *l4proto;
201 struct nlattr *nest_proto;
202 int ret;
203
204 l4proto = nf_ct_l4proto_find(nf_ct_protonum(ct));
205 if (!l4proto->to_nlattr)
206 return 0;
207
208 nest_proto = nla_nest_start(skb, CTA_PROTOINFO);
209 if (!nest_proto)
210 goto nla_put_failure;
211
212 ret = l4proto->to_nlattr(skb, nest_proto, ct, destroy);
213
214 nla_nest_end(skb, nest_proto);
215
216 return ret;
217
218 nla_put_failure:
219 return -1;
220 }
221
ctnetlink_dump_helpinfo(struct sk_buff * skb,const struct nf_conn * ct)222 static int ctnetlink_dump_helpinfo(struct sk_buff *skb,
223 const struct nf_conn *ct)
224 {
225 struct nlattr *nest_helper;
226 const struct nf_conn_help *help = nfct_help(ct);
227 struct nf_conntrack_helper *helper;
228
229 if (!help)
230 return 0;
231
232 rcu_read_lock();
233 helper = rcu_dereference(help->helper);
234 if (!helper)
235 goto out;
236
237 nest_helper = nla_nest_start(skb, CTA_HELP);
238 if (!nest_helper)
239 goto nla_put_failure;
240 if (nla_put_string(skb, CTA_HELP_NAME, helper->name))
241 goto nla_put_failure;
242
243 if (rcu_access_pointer(helper->help) &&
244 helper->to_nlattr)
245 helper->to_nlattr(skb, ct);
246
247 nla_nest_end(skb, nest_helper);
248 out:
249 rcu_read_unlock();
250 return 0;
251
252 nla_put_failure:
253 rcu_read_unlock();
254 return -1;
255 }
256
257 static int
dump_counters(struct sk_buff * skb,struct nf_conn_acct * acct,enum ip_conntrack_dir dir,int type)258 dump_counters(struct sk_buff *skb, struct nf_conn_acct *acct,
259 enum ip_conntrack_dir dir, int type)
260 {
261 enum ctattr_type attr = dir ? CTA_COUNTERS_REPLY: CTA_COUNTERS_ORIG;
262 struct nf_conn_counter *counter = acct->counter;
263 struct nlattr *nest_count;
264 u64 pkts, bytes;
265
266 if (type == IPCTNL_MSG_CT_GET_CTRZERO) {
267 pkts = atomic64_xchg(&counter[dir].packets, 0);
268 bytes = atomic64_xchg(&counter[dir].bytes, 0);
269 } else {
270 pkts = atomic64_read(&counter[dir].packets);
271 bytes = atomic64_read(&counter[dir].bytes);
272 }
273
274 nest_count = nla_nest_start(skb, attr);
275 if (!nest_count)
276 goto nla_put_failure;
277
278 if (nla_put_be64(skb, CTA_COUNTERS_PACKETS, cpu_to_be64(pkts),
279 CTA_COUNTERS_PAD) ||
280 nla_put_be64(skb, CTA_COUNTERS_BYTES, cpu_to_be64(bytes),
281 CTA_COUNTERS_PAD))
282 goto nla_put_failure;
283
284 nla_nest_end(skb, nest_count);
285
286 return 0;
287
288 nla_put_failure:
289 return -1;
290 }
291
292 static int
ctnetlink_dump_acct(struct sk_buff * skb,const struct nf_conn * ct,int type)293 ctnetlink_dump_acct(struct sk_buff *skb, const struct nf_conn *ct, int type)
294 {
295 struct nf_conn_acct *acct = nf_conn_acct_find(ct);
296
297 if (!acct)
298 return 0;
299
300 if (dump_counters(skb, acct, IP_CT_DIR_ORIGINAL, type) < 0)
301 return -1;
302 if (dump_counters(skb, acct, IP_CT_DIR_REPLY, type) < 0)
303 return -1;
304
305 return 0;
306 }
307
308 static int
ctnetlink_dump_timestamp(struct sk_buff * skb,const struct nf_conn * ct)309 ctnetlink_dump_timestamp(struct sk_buff *skb, const struct nf_conn *ct)
310 {
311 struct nlattr *nest_count;
312 const struct nf_conn_tstamp *tstamp;
313
314 tstamp = nf_conn_tstamp_find(ct);
315 if (!tstamp)
316 return 0;
317
318 nest_count = nla_nest_start(skb, CTA_TIMESTAMP);
319 if (!nest_count)
320 goto nla_put_failure;
321
322 if (nla_put_be64(skb, CTA_TIMESTAMP_START, cpu_to_be64(tstamp->start),
323 CTA_TIMESTAMP_PAD) ||
324 (tstamp->stop != 0 && nla_put_be64(skb, CTA_TIMESTAMP_STOP,
325 cpu_to_be64(tstamp->stop),
326 CTA_TIMESTAMP_PAD)))
327 goto nla_put_failure;
328 nla_nest_end(skb, nest_count);
329
330 return 0;
331
332 nla_put_failure:
333 return -1;
334 }
335
336 #ifdef CONFIG_NF_CONNTRACK_MARK
ctnetlink_dump_mark(struct sk_buff * skb,const struct nf_conn * ct,bool dump)337 static int ctnetlink_dump_mark(struct sk_buff *skb, const struct nf_conn *ct,
338 bool dump)
339 {
340 u32 mark = READ_ONCE(ct->mark);
341
342 if (!mark && !dump)
343 return 0;
344
345 if (nla_put_be32(skb, CTA_MARK, htonl(mark)))
346 goto nla_put_failure;
347 return 0;
348
349 nla_put_failure:
350 return -1;
351 }
352 #else
353 #define ctnetlink_dump_mark(a, b, c) (0)
354 #endif
355
356 #ifdef CONFIG_NF_CONNTRACK_SECMARK
ctnetlink_dump_secctx(struct sk_buff * skb,const struct nf_conn * ct)357 static int ctnetlink_dump_secctx(struct sk_buff *skb, const struct nf_conn *ct)
358 {
359 struct nlattr *nest_secctx;
360 struct lsm_context ctx;
361 int ret;
362
363 ret = security_secid_to_secctx(ct->secmark, &ctx);
364 if (ret < 0)
365 return 0;
366
367 ret = -1;
368 nest_secctx = nla_nest_start(skb, CTA_SECCTX);
369 if (!nest_secctx)
370 goto nla_put_failure;
371
372 if (nla_put_string(skb, CTA_SECCTX_NAME, ctx.context))
373 goto nla_put_failure;
374 nla_nest_end(skb, nest_secctx);
375
376 ret = 0;
377 nla_put_failure:
378 security_release_secctx(&ctx);
379 return ret;
380 }
381 #else
382 #define ctnetlink_dump_secctx(a, b) (0)
383 #endif
384
385 #ifdef CONFIG_NF_CONNTRACK_EVENTS
386 static int
ctnetlink_dump_event_timestamp(struct sk_buff * skb,const struct nf_conn * ct)387 ctnetlink_dump_event_timestamp(struct sk_buff *skb, const struct nf_conn *ct)
388 {
389 #ifdef CONFIG_NF_CONNTRACK_TIMESTAMP
390 const struct nf_conntrack_ecache *e = nf_ct_ecache_find(ct);
391
392 if (e) {
393 u64 ts = local64_read(&e->timestamp);
394
395 if (ts)
396 return nla_put_be64(skb, CTA_TIMESTAMP_EVENT,
397 cpu_to_be64(ts), CTA_TIMESTAMP_PAD);
398 }
399 #endif
400 return 0;
401 }
402
ctnetlink_label_size(const struct nf_conn * ct)403 static inline int ctnetlink_label_size(const struct nf_conn *ct)
404 {
405 struct nf_conn_labels *labels = nf_ct_labels_find(ct);
406
407 if (!labels)
408 return 0;
409 return nla_total_size(sizeof(labels->bits));
410 }
411 #endif
412
413 static int
ctnetlink_dump_labels(struct sk_buff * skb,const struct nf_conn * ct)414 ctnetlink_dump_labels(struct sk_buff *skb, const struct nf_conn *ct)
415 {
416 struct nf_conn_labels *labels = nf_ct_labels_find(ct);
417 unsigned int i;
418
419 if (!labels)
420 return 0;
421
422 i = 0;
423 do {
424 if (labels->bits[i] != 0)
425 return nla_put(skb, CTA_LABELS, sizeof(labels->bits),
426 labels->bits);
427 i++;
428 } while (i < ARRAY_SIZE(labels->bits));
429
430 return 0;
431 }
432
433 #define master_tuple(ct) &(ct->master->tuplehash[IP_CT_DIR_ORIGINAL].tuple)
434
ctnetlink_dump_master(struct sk_buff * skb,const struct nf_conn * ct)435 static int ctnetlink_dump_master(struct sk_buff *skb, const struct nf_conn *ct)
436 {
437 struct nlattr *nest_parms;
438
439 if (!(ct->status & IPS_EXPECTED))
440 return 0;
441
442 nest_parms = nla_nest_start(skb, CTA_TUPLE_MASTER);
443 if (!nest_parms)
444 goto nla_put_failure;
445 if (ctnetlink_dump_tuples(skb, master_tuple(ct)) < 0)
446 goto nla_put_failure;
447 nla_nest_end(skb, nest_parms);
448
449 return 0;
450
451 nla_put_failure:
452 return -1;
453 }
454
455 static int
dump_ct_seq_adj(struct sk_buff * skb,const struct nf_ct_seqadj * seq,int type)456 dump_ct_seq_adj(struct sk_buff *skb, const struct nf_ct_seqadj *seq, int type)
457 {
458 struct nlattr *nest_parms;
459
460 nest_parms = nla_nest_start(skb, type);
461 if (!nest_parms)
462 goto nla_put_failure;
463
464 if (nla_put_be32(skb, CTA_SEQADJ_CORRECTION_POS,
465 htonl(seq->correction_pos)) ||
466 nla_put_be32(skb, CTA_SEQADJ_OFFSET_BEFORE,
467 htonl(seq->offset_before)) ||
468 nla_put_be32(skb, CTA_SEQADJ_OFFSET_AFTER,
469 htonl(seq->offset_after)))
470 goto nla_put_failure;
471
472 nla_nest_end(skb, nest_parms);
473
474 return 0;
475
476 nla_put_failure:
477 return -1;
478 }
479
ctnetlink_dump_ct_seq_adj(struct sk_buff * skb,struct nf_conn * ct)480 static int ctnetlink_dump_ct_seq_adj(struct sk_buff *skb, struct nf_conn *ct)
481 {
482 struct nf_conn_seqadj *seqadj = nfct_seqadj(ct);
483 struct nf_ct_seqadj *seq;
484
485 if (!(ct->status & IPS_SEQ_ADJUST) || !seqadj)
486 return 0;
487
488 spin_lock_bh(&ct->lock);
489 seq = &seqadj->seq[IP_CT_DIR_ORIGINAL];
490 if (dump_ct_seq_adj(skb, seq, CTA_SEQ_ADJ_ORIG) == -1)
491 goto err;
492
493 seq = &seqadj->seq[IP_CT_DIR_REPLY];
494 if (dump_ct_seq_adj(skb, seq, CTA_SEQ_ADJ_REPLY) == -1)
495 goto err;
496
497 spin_unlock_bh(&ct->lock);
498 return 0;
499 err:
500 spin_unlock_bh(&ct->lock);
501 return -1;
502 }
503
ctnetlink_dump_ct_synproxy(struct sk_buff * skb,struct nf_conn * ct)504 static int ctnetlink_dump_ct_synproxy(struct sk_buff *skb, struct nf_conn *ct)
505 {
506 struct nf_conn_synproxy *synproxy = nfct_synproxy(ct);
507 struct nlattr *nest_parms;
508
509 if (!synproxy)
510 return 0;
511
512 nest_parms = nla_nest_start(skb, CTA_SYNPROXY);
513 if (!nest_parms)
514 goto nla_put_failure;
515
516 if (nla_put_be32(skb, CTA_SYNPROXY_ISN, htonl(synproxy->isn)) ||
517 nla_put_be32(skb, CTA_SYNPROXY_ITS, htonl(synproxy->its)) ||
518 nla_put_be32(skb, CTA_SYNPROXY_TSOFF, htonl(synproxy->tsoff)))
519 goto nla_put_failure;
520
521 nla_nest_end(skb, nest_parms);
522
523 return 0;
524
525 nla_put_failure:
526 return -1;
527 }
528
ctnetlink_dump_id(struct sk_buff * skb,const struct nf_conn * ct)529 static int ctnetlink_dump_id(struct sk_buff *skb, const struct nf_conn *ct)
530 {
531 __be32 id = (__force __be32)nf_ct_get_id(ct);
532
533 if (nla_put_be32(skb, CTA_ID, id))
534 goto nla_put_failure;
535 return 0;
536
537 nla_put_failure:
538 return -1;
539 }
540
ctnetlink_dump_use(struct sk_buff * skb,const struct nf_conn * ct)541 static int ctnetlink_dump_use(struct sk_buff *skb, const struct nf_conn *ct)
542 {
543 if (nla_put_be32(skb, CTA_USE, htonl(refcount_read(&ct->ct_general.use))))
544 goto nla_put_failure;
545 return 0;
546
547 nla_put_failure:
548 return -1;
549 }
550
551 /* all these functions access ct->ext. Caller must either hold a reference
552 * on ct or prevent its deletion by holding either the bucket spinlock or
553 * pcpu dying list lock.
554 */
ctnetlink_dump_extinfo(struct sk_buff * skb,struct nf_conn * ct,u32 type)555 static int ctnetlink_dump_extinfo(struct sk_buff *skb,
556 struct nf_conn *ct, u32 type)
557 {
558 if (ctnetlink_dump_acct(skb, ct, type) < 0 ||
559 ctnetlink_dump_timestamp(skb, ct) < 0 ||
560 ctnetlink_dump_helpinfo(skb, ct) < 0 ||
561 ctnetlink_dump_labels(skb, ct) < 0 ||
562 ctnetlink_dump_ct_seq_adj(skb, ct) < 0 ||
563 ctnetlink_dump_ct_synproxy(skb, ct) < 0)
564 return -1;
565
566 return 0;
567 }
568
ctnetlink_dump_info(struct sk_buff * skb,struct nf_conn * ct)569 static int ctnetlink_dump_info(struct sk_buff *skb, struct nf_conn *ct)
570 {
571 if (ctnetlink_dump_status(skb, ct) < 0 ||
572 ctnetlink_dump_mark(skb, ct, true) < 0 ||
573 ctnetlink_dump_secctx(skb, ct) < 0 ||
574 ctnetlink_dump_id(skb, ct) < 0 ||
575 ctnetlink_dump_use(skb, ct) < 0 ||
576 ctnetlink_dump_master(skb, ct) < 0)
577 return -1;
578
579 if (!test_bit(IPS_OFFLOAD_BIT, &ct->status) &&
580 (ctnetlink_dump_timeout(skb, ct, false) < 0 ||
581 ctnetlink_dump_protoinfo(skb, ct, false) < 0))
582 return -1;
583
584 return 0;
585 }
586
587 static int
ctnetlink_fill_info(struct sk_buff * skb,u32 portid,u32 seq,u32 type,struct nf_conn * ct,bool extinfo,unsigned int flags)588 ctnetlink_fill_info(struct sk_buff *skb, u32 portid, u32 seq, u32 type,
589 struct nf_conn *ct, bool extinfo, unsigned int flags)
590 {
591 const struct nf_conntrack_zone *zone;
592 struct nlmsghdr *nlh;
593 struct nlattr *nest_parms;
594 unsigned int event;
595
596 if (portid)
597 flags |= NLM_F_MULTI;
598 event = nfnl_msg_type(NFNL_SUBSYS_CTNETLINK, IPCTNL_MSG_CT_NEW);
599 nlh = nfnl_msg_put(skb, portid, seq, event, flags, nf_ct_l3num(ct),
600 NFNETLINK_V0, 0);
601 if (!nlh)
602 goto nlmsg_failure;
603
604 zone = nf_ct_zone(ct);
605
606 nest_parms = nla_nest_start(skb, CTA_TUPLE_ORIG);
607 if (!nest_parms)
608 goto nla_put_failure;
609 if (ctnetlink_dump_tuples(skb, nf_ct_tuple(ct, IP_CT_DIR_ORIGINAL)) < 0)
610 goto nla_put_failure;
611 if (ctnetlink_dump_zone_id(skb, CTA_TUPLE_ZONE, zone,
612 NF_CT_ZONE_DIR_ORIG) < 0)
613 goto nla_put_failure;
614 nla_nest_end(skb, nest_parms);
615
616 nest_parms = nla_nest_start(skb, CTA_TUPLE_REPLY);
617 if (!nest_parms)
618 goto nla_put_failure;
619 if (ctnetlink_dump_tuples(skb, nf_ct_tuple(ct, IP_CT_DIR_REPLY)) < 0)
620 goto nla_put_failure;
621 if (ctnetlink_dump_zone_id(skb, CTA_TUPLE_ZONE, zone,
622 NF_CT_ZONE_DIR_REPL) < 0)
623 goto nla_put_failure;
624 nla_nest_end(skb, nest_parms);
625
626 if (ctnetlink_dump_zone_id(skb, CTA_ZONE, zone,
627 NF_CT_DEFAULT_ZONE_DIR) < 0)
628 goto nla_put_failure;
629
630 if (ctnetlink_dump_info(skb, ct) < 0)
631 goto nla_put_failure;
632 if (extinfo && ctnetlink_dump_extinfo(skb, ct, type) < 0)
633 goto nla_put_failure;
634
635 nlmsg_end(skb, nlh);
636 return skb->len;
637
638 nlmsg_failure:
639 nla_put_failure:
640 nlmsg_cancel(skb, nlh);
641 return -1;
642 }
643
644 static const struct nla_policy cta_ip_nla_policy[CTA_IP_MAX + 1] = {
645 [CTA_IP_V4_SRC] = { .type = NLA_U32 },
646 [CTA_IP_V4_DST] = { .type = NLA_U32 },
647 [CTA_IP_V6_SRC] = { .len = sizeof(__be32) * 4 },
648 [CTA_IP_V6_DST] = { .len = sizeof(__be32) * 4 },
649 };
650
651 #if defined(CONFIG_NETFILTER_NETLINK_GLUE_CT) || defined(CONFIG_NF_CONNTRACK_EVENTS)
ctnetlink_proto_size(const struct nf_conn * ct)652 static size_t ctnetlink_proto_size(const struct nf_conn *ct)
653 {
654 const struct nf_conntrack_l4proto *l4proto;
655 size_t len, len4 = 0;
656
657 len = nla_policy_len(cta_ip_nla_policy, CTA_IP_MAX + 1);
658 len *= 3u; /* ORIG, REPLY, MASTER */
659
660 l4proto = nf_ct_l4proto_find(nf_ct_protonum(ct));
661 len += l4proto->nlattr_size;
662 if (l4proto->nlattr_tuple_size) {
663 len4 = l4proto->nlattr_tuple_size();
664 len4 *= 3u; /* ORIG, REPLY, MASTER */
665 }
666
667 return len + len4;
668 }
669
ctnetlink_acct_size(const struct nf_conn * ct)670 static inline size_t ctnetlink_acct_size(const struct nf_conn *ct)
671 {
672 if (!nf_ct_ext_exist(ct, NF_CT_EXT_ACCT))
673 return 0;
674 return 2 * nla_total_size(0) /* CTA_COUNTERS_ORIG|REPL */
675 + 2 * nla_total_size_64bit(sizeof(uint64_t)) /* CTA_COUNTERS_PACKETS */
676 + 2 * nla_total_size_64bit(sizeof(uint64_t)) /* CTA_COUNTERS_BYTES */
677 ;
678 }
679
ctnetlink_secctx_size(const struct nf_conn * ct)680 static inline int ctnetlink_secctx_size(const struct nf_conn *ct)
681 {
682 #ifdef CONFIG_NF_CONNTRACK_SECMARK
683 int ret;
684
685 ret = security_secid_to_secctx(ct->secmark, NULL);
686 if (ret < 0)
687 return 0;
688
689 return nla_total_size(0) /* CTA_SECCTX */
690 + nla_total_size(sizeof(char) * ret); /* CTA_SECCTX_NAME */
691 #else
692 return 0;
693 #endif
694 }
695
ctnetlink_timestamp_size(const struct nf_conn * ct)696 static inline size_t ctnetlink_timestamp_size(const struct nf_conn *ct)
697 {
698 #ifdef CONFIG_NF_CONNTRACK_TIMESTAMP
699 if (!nf_ct_ext_exist(ct, NF_CT_EXT_TSTAMP))
700 return 0;
701 return nla_total_size(0) + 2 * nla_total_size_64bit(sizeof(uint64_t));
702 #else
703 return 0;
704 #endif
705 }
706 #endif
707
708 #ifdef CONFIG_NF_CONNTRACK_EVENTS
ctnetlink_nlmsg_size(const struct nf_conn * ct)709 static size_t ctnetlink_nlmsg_size(const struct nf_conn *ct)
710 {
711 return NLMSG_ALIGN(sizeof(struct nfgenmsg))
712 + 3 * nla_total_size(0) /* CTA_TUPLE_ORIG|REPL|MASTER */
713 + 3 * nla_total_size(0) /* CTA_TUPLE_IP */
714 + 3 * nla_total_size(0) /* CTA_TUPLE_PROTO */
715 + 3 * nla_total_size(sizeof(u_int8_t)) /* CTA_PROTO_NUM */
716 + nla_total_size(sizeof(u_int32_t)) /* CTA_ID */
717 + nla_total_size(sizeof(u_int32_t)) /* CTA_STATUS */
718 + ctnetlink_acct_size(ct)
719 + ctnetlink_timestamp_size(ct)
720 + nla_total_size(sizeof(u_int32_t)) /* CTA_TIMEOUT */
721 + nla_total_size(0) /* CTA_PROTOINFO */
722 + nla_total_size(0) /* CTA_HELP */
723 + nla_total_size(NF_CT_HELPER_NAME_LEN) /* CTA_HELP_NAME */
724 + ctnetlink_secctx_size(ct)
725 #if IS_ENABLED(CONFIG_NF_NAT)
726 + 2 * nla_total_size(0) /* CTA_NAT_SEQ_ADJ_ORIG|REPL */
727 + 6 * nla_total_size(sizeof(u_int32_t)) /* CTA_NAT_SEQ_OFFSET */
728 #endif
729 #ifdef CONFIG_NF_CONNTRACK_MARK
730 + nla_total_size(sizeof(u_int32_t)) /* CTA_MARK */
731 #endif
732 #ifdef CONFIG_NF_CONNTRACK_ZONES
733 + nla_total_size(sizeof(u_int16_t)) /* CTA_ZONE|CTA_TUPLE_ZONE */
734 #endif
735 + ctnetlink_proto_size(ct)
736 + ctnetlink_label_size(ct)
737 #ifdef CONFIG_NF_CONNTRACK_TIMESTAMP
738 + nla_total_size(sizeof(u64)) /* CTA_TIMESTAMP_EVENT */
739 #endif
740 ;
741 }
742
743 static int
ctnetlink_conntrack_event(unsigned int events,const struct nf_ct_event * item)744 ctnetlink_conntrack_event(unsigned int events, const struct nf_ct_event *item)
745 {
746 const struct nf_conntrack_zone *zone;
747 struct net *net;
748 struct nlmsghdr *nlh;
749 struct nlattr *nest_parms;
750 struct nf_conn *ct = item->ct;
751 struct sk_buff *skb;
752 unsigned int type;
753 unsigned int flags = 0, group;
754 int err;
755
756 if (events & (1 << IPCT_DESTROY)) {
757 type = IPCTNL_MSG_CT_DELETE;
758 group = NFNLGRP_CONNTRACK_DESTROY;
759 } else if (events & ((1 << IPCT_NEW) | (1 << IPCT_RELATED))) {
760 type = IPCTNL_MSG_CT_NEW;
761 flags = NLM_F_CREATE|NLM_F_EXCL;
762 group = NFNLGRP_CONNTRACK_NEW;
763 } else if (events) {
764 type = IPCTNL_MSG_CT_NEW;
765 group = NFNLGRP_CONNTRACK_UPDATE;
766 } else
767 return 0;
768
769 net = nf_ct_net(ct);
770 if (!item->report && !nfnetlink_has_listeners(net, group))
771 return 0;
772
773 skb = nlmsg_new(ctnetlink_nlmsg_size(ct), GFP_ATOMIC);
774 if (skb == NULL)
775 goto errout;
776
777 type = nfnl_msg_type(NFNL_SUBSYS_CTNETLINK, type);
778 nlh = nfnl_msg_put(skb, item->portid, 0, type, flags, nf_ct_l3num(ct),
779 NFNETLINK_V0, 0);
780 if (!nlh)
781 goto nlmsg_failure;
782
783 zone = nf_ct_zone(ct);
784
785 nest_parms = nla_nest_start(skb, CTA_TUPLE_ORIG);
786 if (!nest_parms)
787 goto nla_put_failure;
788 if (ctnetlink_dump_tuples(skb, nf_ct_tuple(ct, IP_CT_DIR_ORIGINAL)) < 0)
789 goto nla_put_failure;
790 if (ctnetlink_dump_zone_id(skb, CTA_TUPLE_ZONE, zone,
791 NF_CT_ZONE_DIR_ORIG) < 0)
792 goto nla_put_failure;
793 nla_nest_end(skb, nest_parms);
794
795 nest_parms = nla_nest_start(skb, CTA_TUPLE_REPLY);
796 if (!nest_parms)
797 goto nla_put_failure;
798 if (ctnetlink_dump_tuples(skb, nf_ct_tuple(ct, IP_CT_DIR_REPLY)) < 0)
799 goto nla_put_failure;
800 if (ctnetlink_dump_zone_id(skb, CTA_TUPLE_ZONE, zone,
801 NF_CT_ZONE_DIR_REPL) < 0)
802 goto nla_put_failure;
803 nla_nest_end(skb, nest_parms);
804
805 if (ctnetlink_dump_zone_id(skb, CTA_ZONE, zone,
806 NF_CT_DEFAULT_ZONE_DIR) < 0)
807 goto nla_put_failure;
808
809 if (ctnetlink_dump_id(skb, ct) < 0)
810 goto nla_put_failure;
811
812 if (ctnetlink_dump_status(skb, ct) < 0)
813 goto nla_put_failure;
814
815 if (events & (1 << IPCT_DESTROY)) {
816 if (ctnetlink_dump_timeout(skb, ct, true) < 0)
817 goto nla_put_failure;
818
819 if (ctnetlink_dump_acct(skb, ct, type) < 0 ||
820 ctnetlink_dump_timestamp(skb, ct) < 0 ||
821 ctnetlink_dump_protoinfo(skb, ct, true) < 0)
822 goto nla_put_failure;
823 } else {
824 if (ctnetlink_dump_timeout(skb, ct, false) < 0)
825 goto nla_put_failure;
826
827 if (events & (1 << IPCT_PROTOINFO) &&
828 ctnetlink_dump_protoinfo(skb, ct, false) < 0)
829 goto nla_put_failure;
830
831 if ((events & (1 << IPCT_HELPER) || nfct_help(ct))
832 && ctnetlink_dump_helpinfo(skb, ct) < 0)
833 goto nla_put_failure;
834
835 #ifdef CONFIG_NF_CONNTRACK_SECMARK
836 if ((events & (1 << IPCT_SECMARK) || ct->secmark)
837 && ctnetlink_dump_secctx(skb, ct) < 0)
838 goto nla_put_failure;
839 #endif
840 if (events & (1 << IPCT_LABEL) &&
841 ctnetlink_dump_labels(skb, ct) < 0)
842 goto nla_put_failure;
843
844 if (events & (1 << IPCT_RELATED) &&
845 ctnetlink_dump_master(skb, ct) < 0)
846 goto nla_put_failure;
847
848 if (events & (1 << IPCT_SEQADJ) &&
849 ctnetlink_dump_ct_seq_adj(skb, ct) < 0)
850 goto nla_put_failure;
851
852 if (events & (1 << IPCT_SYNPROXY) &&
853 ctnetlink_dump_ct_synproxy(skb, ct) < 0)
854 goto nla_put_failure;
855 }
856
857 #ifdef CONFIG_NF_CONNTRACK_MARK
858 if (ctnetlink_dump_mark(skb, ct, events & (1 << IPCT_MARK)))
859 goto nla_put_failure;
860 #endif
861
862 if (ctnetlink_dump_event_timestamp(skb, ct))
863 goto nla_put_failure;
864
865 nlmsg_end(skb, nlh);
866 err = nfnetlink_send(skb, net, item->portid, group, item->report,
867 GFP_ATOMIC);
868 if (err == -ENOBUFS || err == -EAGAIN)
869 return -ENOBUFS;
870
871 return 0;
872
873 nla_put_failure:
874 nlmsg_cancel(skb, nlh);
875 nlmsg_failure:
876 kfree_skb(skb);
877 errout:
878 if (nfnetlink_set_err(net, 0, group, -ENOBUFS) > 0)
879 return -ENOBUFS;
880
881 return 0;
882 }
883 #endif /* CONFIG_NF_CONNTRACK_EVENTS */
884
ctnetlink_done(struct netlink_callback * cb)885 static int ctnetlink_done(struct netlink_callback *cb)
886 {
887 kfree(cb->data);
888 return 0;
889 }
890
891 struct ctnetlink_filter_u32 {
892 u32 val;
893 u32 mask;
894 };
895
896 struct ctnetlink_filter {
897 u8 family;
898 bool zone_filter;
899
900 u_int32_t orig_flags;
901 u_int32_t reply_flags;
902
903 struct nf_conntrack_tuple orig;
904 struct nf_conntrack_tuple reply;
905 struct nf_conntrack_zone zone;
906
907 struct ctnetlink_filter_u32 mark;
908 struct ctnetlink_filter_u32 status;
909 };
910
911 static const struct nla_policy cta_filter_nla_policy[CTA_FILTER_MAX + 1] = {
912 [CTA_FILTER_ORIG_FLAGS] = NLA_POLICY_MASK(NLA_U32, CTA_FILTER_F_ALL),
913 [CTA_FILTER_REPLY_FLAGS] = NLA_POLICY_MASK(NLA_U32, CTA_FILTER_F_ALL),
914 };
915
ctnetlink_parse_filter(const struct nlattr * attr,struct ctnetlink_filter * filter)916 static int ctnetlink_parse_filter(const struct nlattr *attr,
917 struct ctnetlink_filter *filter)
918 {
919 struct nlattr *tb[CTA_FILTER_MAX + 1];
920 int ret = 0;
921
922 ret = nla_parse_nested(tb, CTA_FILTER_MAX, attr, cta_filter_nla_policy,
923 NULL);
924 if (ret)
925 return ret;
926
927 if (tb[CTA_FILTER_ORIG_FLAGS])
928 filter->orig_flags = nla_get_u32(tb[CTA_FILTER_ORIG_FLAGS]);
929
930 if (tb[CTA_FILTER_REPLY_FLAGS])
931 filter->reply_flags = nla_get_u32(tb[CTA_FILTER_REPLY_FLAGS]);
932
933 return 0;
934 }
935
936 static int ctnetlink_parse_zone(const struct nlattr *attr,
937 struct nf_conntrack_zone *zone);
938 static int ctnetlink_parse_tuple_filter(const struct nlattr * const cda[],
939 struct nf_conntrack_tuple *tuple,
940 u32 type, u_int8_t l3num,
941 struct nf_conntrack_zone *zone,
942 u_int32_t flags);
943
ctnetlink_filter_parse_mark(struct ctnetlink_filter_u32 * mark,const struct nlattr * const cda[])944 static int ctnetlink_filter_parse_mark(struct ctnetlink_filter_u32 *mark,
945 const struct nlattr * const cda[])
946 {
947 #ifdef CONFIG_NF_CONNTRACK_MARK
948 if (cda[CTA_MARK]) {
949 mark->val = ntohl(nla_get_be32(cda[CTA_MARK]));
950
951 if (cda[CTA_MARK_MASK])
952 mark->mask = ntohl(nla_get_be32(cda[CTA_MARK_MASK]));
953 else
954 mark->mask = 0xffffffff;
955 } else if (cda[CTA_MARK_MASK]) {
956 return -EINVAL;
957 }
958 #endif
959 return 0;
960 }
961
ctnetlink_filter_parse_status(struct ctnetlink_filter_u32 * status,const struct nlattr * const cda[])962 static int ctnetlink_filter_parse_status(struct ctnetlink_filter_u32 *status,
963 const struct nlattr * const cda[])
964 {
965 if (cda[CTA_STATUS]) {
966 status->val = ntohl(nla_get_be32(cda[CTA_STATUS]));
967 if (cda[CTA_STATUS_MASK])
968 status->mask = ntohl(nla_get_be32(cda[CTA_STATUS_MASK]));
969 else
970 status->mask = status->val;
971
972 /* status->val == 0? always true, else always false. */
973 if (status->mask == 0)
974 return -EINVAL;
975 } else if (cda[CTA_STATUS_MASK]) {
976 return -EINVAL;
977 }
978
979 /* CTA_STATUS is NLA_U32, if this fires UAPI needs to be extended */
980 BUILD_BUG_ON(__IPS_MAX_BIT >= 32);
981 return 0;
982 }
983
984 static struct ctnetlink_filter *
ctnetlink_alloc_filter(const struct nlattr * const cda[],u8 family)985 ctnetlink_alloc_filter(const struct nlattr * const cda[], u8 family)
986 {
987 struct ctnetlink_filter *filter;
988 int err;
989
990 #ifndef CONFIG_NF_CONNTRACK_MARK
991 if (cda[CTA_MARK] || cda[CTA_MARK_MASK])
992 return ERR_PTR(-EOPNOTSUPP);
993 #endif
994
995 filter = kzalloc_obj(*filter);
996 if (filter == NULL)
997 return ERR_PTR(-ENOMEM);
998
999 filter->family = family;
1000
1001 err = ctnetlink_filter_parse_mark(&filter->mark, cda);
1002 if (err)
1003 goto err_filter;
1004
1005 err = ctnetlink_filter_parse_status(&filter->status, cda);
1006 if (err)
1007 goto err_filter;
1008
1009 if (cda[CTA_ZONE]) {
1010 err = ctnetlink_parse_zone(cda[CTA_ZONE], &filter->zone);
1011 if (err < 0)
1012 goto err_filter;
1013 filter->zone_filter = true;
1014 }
1015
1016 if (!cda[CTA_FILTER])
1017 return filter;
1018
1019 err = ctnetlink_parse_filter(cda[CTA_FILTER], filter);
1020 if (err < 0)
1021 goto err_filter;
1022
1023 if (filter->orig_flags) {
1024 if (!cda[CTA_TUPLE_ORIG]) {
1025 err = -EINVAL;
1026 goto err_filter;
1027 }
1028
1029 err = ctnetlink_parse_tuple_filter(cda, &filter->orig,
1030 CTA_TUPLE_ORIG,
1031 filter->family,
1032 &filter->zone,
1033 filter->orig_flags);
1034 if (err < 0)
1035 goto err_filter;
1036 }
1037
1038 if (filter->reply_flags) {
1039 if (!cda[CTA_TUPLE_REPLY]) {
1040 err = -EINVAL;
1041 goto err_filter;
1042 }
1043
1044 err = ctnetlink_parse_tuple_filter(cda, &filter->reply,
1045 CTA_TUPLE_REPLY,
1046 filter->family,
1047 &filter->zone,
1048 filter->reply_flags);
1049 if (err < 0)
1050 goto err_filter;
1051 }
1052
1053 return filter;
1054
1055 err_filter:
1056 kfree(filter);
1057
1058 return ERR_PTR(err);
1059 }
1060
ctnetlink_needs_filter(u8 family,const struct nlattr * const * cda)1061 static bool ctnetlink_needs_filter(u8 family, const struct nlattr * const *cda)
1062 {
1063 return family || cda[CTA_MARK] || cda[CTA_FILTER] || cda[CTA_STATUS] || cda[CTA_ZONE];
1064 }
1065
ctnetlink_start(struct netlink_callback * cb)1066 static int ctnetlink_start(struct netlink_callback *cb)
1067 {
1068 const struct nlattr * const *cda = cb->data;
1069 struct ctnetlink_filter *filter = NULL;
1070 struct nfgenmsg *nfmsg = nlmsg_data(cb->nlh);
1071 u8 family = nfmsg->nfgen_family;
1072
1073 if (ctnetlink_needs_filter(family, cda)) {
1074 filter = ctnetlink_alloc_filter(cda, family);
1075 if (IS_ERR(filter))
1076 return PTR_ERR(filter);
1077 }
1078
1079 cb->data = filter;
1080 if (filter)
1081 cb->answer_flags = NLM_F_DUMP_FILTERED;
1082 return 0;
1083 }
1084
ctnetlink_filter_match_tuple(struct nf_conntrack_tuple * filter_tuple,struct nf_conntrack_tuple * ct_tuple,u_int32_t flags,int family)1085 static int ctnetlink_filter_match_tuple(struct nf_conntrack_tuple *filter_tuple,
1086 struct nf_conntrack_tuple *ct_tuple,
1087 u_int32_t flags, int family)
1088 {
1089 switch (family) {
1090 case NFPROTO_IPV4:
1091 if ((flags & CTA_FILTER_FLAG(CTA_IP_SRC)) &&
1092 filter_tuple->src.u3.ip != ct_tuple->src.u3.ip)
1093 return 0;
1094
1095 if ((flags & CTA_FILTER_FLAG(CTA_IP_DST)) &&
1096 filter_tuple->dst.u3.ip != ct_tuple->dst.u3.ip)
1097 return 0;
1098 break;
1099 case NFPROTO_IPV6:
1100 if ((flags & CTA_FILTER_FLAG(CTA_IP_SRC)) &&
1101 !ipv6_addr_cmp(&filter_tuple->src.u3.in6,
1102 &ct_tuple->src.u3.in6))
1103 return 0;
1104
1105 if ((flags & CTA_FILTER_FLAG(CTA_IP_DST)) &&
1106 !ipv6_addr_cmp(&filter_tuple->dst.u3.in6,
1107 &ct_tuple->dst.u3.in6))
1108 return 0;
1109 break;
1110 }
1111
1112 if ((flags & CTA_FILTER_FLAG(CTA_PROTO_NUM)) &&
1113 filter_tuple->dst.protonum != ct_tuple->dst.protonum)
1114 return 0;
1115
1116 switch (ct_tuple->dst.protonum) {
1117 case IPPROTO_TCP:
1118 case IPPROTO_UDP:
1119 if ((flags & CTA_FILTER_FLAG(CTA_PROTO_SRC_PORT)) &&
1120 filter_tuple->src.u.tcp.port != ct_tuple->src.u.tcp.port)
1121 return 0;
1122
1123 if ((flags & CTA_FILTER_FLAG(CTA_PROTO_DST_PORT)) &&
1124 filter_tuple->dst.u.tcp.port != ct_tuple->dst.u.tcp.port)
1125 return 0;
1126 break;
1127 case IPPROTO_ICMP:
1128 if ((flags & CTA_FILTER_FLAG(CTA_PROTO_ICMP_TYPE)) &&
1129 filter_tuple->dst.u.icmp.type != ct_tuple->dst.u.icmp.type)
1130 return 0;
1131 if ((flags & CTA_FILTER_FLAG(CTA_PROTO_ICMP_CODE)) &&
1132 filter_tuple->dst.u.icmp.code != ct_tuple->dst.u.icmp.code)
1133 return 0;
1134 if ((flags & CTA_FILTER_FLAG(CTA_PROTO_ICMP_ID)) &&
1135 filter_tuple->src.u.icmp.id != ct_tuple->src.u.icmp.id)
1136 return 0;
1137 break;
1138 case IPPROTO_ICMPV6:
1139 if ((flags & CTA_FILTER_FLAG(CTA_PROTO_ICMPV6_TYPE)) &&
1140 filter_tuple->dst.u.icmp.type != ct_tuple->dst.u.icmp.type)
1141 return 0;
1142 if ((flags & CTA_FILTER_FLAG(CTA_PROTO_ICMPV6_CODE)) &&
1143 filter_tuple->dst.u.icmp.code != ct_tuple->dst.u.icmp.code)
1144 return 0;
1145 if ((flags & CTA_FILTER_FLAG(CTA_PROTO_ICMPV6_ID)) &&
1146 filter_tuple->src.u.icmp.id != ct_tuple->src.u.icmp.id)
1147 return 0;
1148 break;
1149 }
1150
1151 return 1;
1152 }
1153
ctnetlink_filter_match(struct nf_conn * ct,void * data)1154 static int ctnetlink_filter_match(struct nf_conn *ct, void *data)
1155 {
1156 struct ctnetlink_filter *filter = data;
1157 struct nf_conntrack_tuple *tuple;
1158 u32 status;
1159
1160 if (filter == NULL)
1161 goto out;
1162
1163 /* Match entries of a given L3 protocol number.
1164 * If it is not specified, ie. l3proto == 0,
1165 * then match everything.
1166 */
1167 if (filter->family && nf_ct_l3num(ct) != filter->family)
1168 goto ignore_entry;
1169
1170 if (filter->zone_filter &&
1171 !nf_ct_zone_equal_any(ct, &filter->zone))
1172 goto ignore_entry;
1173
1174 if (filter->orig_flags) {
1175 tuple = nf_ct_tuple(ct, IP_CT_DIR_ORIGINAL);
1176 if (!ctnetlink_filter_match_tuple(&filter->orig, tuple,
1177 filter->orig_flags,
1178 filter->family))
1179 goto ignore_entry;
1180 }
1181
1182 if (filter->reply_flags) {
1183 tuple = nf_ct_tuple(ct, IP_CT_DIR_REPLY);
1184 if (!ctnetlink_filter_match_tuple(&filter->reply, tuple,
1185 filter->reply_flags,
1186 filter->family))
1187 goto ignore_entry;
1188 }
1189
1190 #ifdef CONFIG_NF_CONNTRACK_MARK
1191 if ((READ_ONCE(ct->mark) & filter->mark.mask) != filter->mark.val)
1192 goto ignore_entry;
1193 #endif
1194 status = (u32)READ_ONCE(ct->status);
1195 if ((status & filter->status.mask) != filter->status.val)
1196 goto ignore_entry;
1197
1198 out:
1199 return 1;
1200
1201 ignore_entry:
1202 return 0;
1203 }
1204
ctnetlink_get_id(const struct nf_conn * ct)1205 static unsigned long ctnetlink_get_id(const struct nf_conn *ct)
1206 {
1207 unsigned long id = nf_ct_get_id(ct);
1208
1209 return id ? id : 1;
1210 }
1211
1212 static int
ctnetlink_dump_table(struct sk_buff * skb,struct netlink_callback * cb)1213 ctnetlink_dump_table(struct sk_buff *skb, struct netlink_callback *cb)
1214 {
1215 unsigned int flags = cb->data ? NLM_F_DUMP_FILTERED : 0;
1216 struct net *net = sock_net(skb->sk);
1217 unsigned long last_id = cb->args[1];
1218 struct nf_conntrack_tuple_hash *h;
1219 struct hlist_nulls_node *n;
1220 struct nf_conn *nf_ct_evict[8];
1221 struct nf_conn *ct;
1222 int res, i;
1223 spinlock_t *lockp;
1224
1225 i = 0;
1226
1227 local_bh_disable();
1228 for (; cb->args[0] < nf_conntrack_htable_size; cb->args[0]++) {
1229 restart:
1230 while (i) {
1231 i--;
1232 if (nf_ct_should_gc(nf_ct_evict[i]))
1233 nf_ct_kill(nf_ct_evict[i]);
1234 nf_ct_put(nf_ct_evict[i]);
1235 }
1236
1237 lockp = &nf_conntrack_locks[cb->args[0] % CONNTRACK_LOCKS];
1238 nf_conntrack_lock(lockp);
1239 if (cb->args[0] >= nf_conntrack_htable_size) {
1240 spin_unlock(lockp);
1241 goto out;
1242 }
1243 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[cb->args[0]],
1244 hnnode) {
1245 ct = nf_ct_tuplehash_to_ctrack(h);
1246 if (nf_ct_is_expired(ct)) {
1247 /* need to defer nf_ct_kill() until lock is released */
1248 if (i < ARRAY_SIZE(nf_ct_evict) &&
1249 refcount_inc_not_zero(&ct->ct_general.use))
1250 nf_ct_evict[i++] = ct;
1251 continue;
1252 }
1253
1254 if (!net_eq(net, nf_ct_net(ct)))
1255 continue;
1256
1257 if (NF_CT_DIRECTION(h) != IP_CT_DIR_ORIGINAL)
1258 continue;
1259
1260 if (cb->args[1]) {
1261 if (ctnetlink_get_id(ct) != last_id)
1262 continue;
1263 cb->args[1] = 0;
1264 }
1265 if (!ctnetlink_filter_match(ct, cb->data))
1266 continue;
1267
1268 res =
1269 ctnetlink_fill_info(skb, NETLINK_CB(cb->skb).portid,
1270 cb->nlh->nlmsg_seq,
1271 NFNL_MSG_TYPE(cb->nlh->nlmsg_type),
1272 ct, true, flags);
1273 if (res < 0) {
1274 cb->args[1] = ctnetlink_get_id(ct);
1275 spin_unlock(lockp);
1276 goto out;
1277 }
1278 }
1279 spin_unlock(lockp);
1280 if (cb->args[1]) {
1281 cb->args[1] = 0;
1282 goto restart;
1283 }
1284 }
1285 out:
1286 local_bh_enable();
1287 if (last_id) {
1288 /* nf ct hash resize happened, now clear the leftover. */
1289 if (cb->args[1] == last_id)
1290 cb->args[1] = 0;
1291 }
1292
1293 while (i) {
1294 i--;
1295 if (nf_ct_should_gc(nf_ct_evict[i]))
1296 nf_ct_kill(nf_ct_evict[i]);
1297 nf_ct_put(nf_ct_evict[i]);
1298 }
1299
1300 return skb->len;
1301 }
1302
ipv4_nlattr_to_tuple(struct nlattr * tb[],struct nf_conntrack_tuple * t,u_int32_t flags)1303 static int ipv4_nlattr_to_tuple(struct nlattr *tb[],
1304 struct nf_conntrack_tuple *t,
1305 u_int32_t flags)
1306 {
1307 if (flags & CTA_FILTER_FLAG(CTA_IP_SRC)) {
1308 if (!tb[CTA_IP_V4_SRC])
1309 return -EINVAL;
1310
1311 t->src.u3.ip = nla_get_in_addr(tb[CTA_IP_V4_SRC]);
1312 }
1313
1314 if (flags & CTA_FILTER_FLAG(CTA_IP_DST)) {
1315 if (!tb[CTA_IP_V4_DST])
1316 return -EINVAL;
1317
1318 t->dst.u3.ip = nla_get_in_addr(tb[CTA_IP_V4_DST]);
1319 }
1320
1321 return 0;
1322 }
1323
ipv6_nlattr_to_tuple(struct nlattr * tb[],struct nf_conntrack_tuple * t,u_int32_t flags)1324 static int ipv6_nlattr_to_tuple(struct nlattr *tb[],
1325 struct nf_conntrack_tuple *t,
1326 u_int32_t flags)
1327 {
1328 if (flags & CTA_FILTER_FLAG(CTA_IP_SRC)) {
1329 if (!tb[CTA_IP_V6_SRC])
1330 return -EINVAL;
1331
1332 t->src.u3.in6 = nla_get_in6_addr(tb[CTA_IP_V6_SRC]);
1333 }
1334
1335 if (flags & CTA_FILTER_FLAG(CTA_IP_DST)) {
1336 if (!tb[CTA_IP_V6_DST])
1337 return -EINVAL;
1338
1339 t->dst.u3.in6 = nla_get_in6_addr(tb[CTA_IP_V6_DST]);
1340 }
1341
1342 return 0;
1343 }
1344
ctnetlink_parse_tuple_ip(struct nlattr * attr,struct nf_conntrack_tuple * tuple,u_int32_t flags)1345 static int ctnetlink_parse_tuple_ip(struct nlattr *attr,
1346 struct nf_conntrack_tuple *tuple,
1347 u_int32_t flags)
1348 {
1349 struct nlattr *tb[CTA_IP_MAX+1];
1350 int ret = 0;
1351
1352 ret = nla_parse_nested_deprecated(tb, CTA_IP_MAX, attr,
1353 cta_ip_nla_policy, NULL);
1354 if (ret < 0)
1355 return ret;
1356
1357 switch (tuple->src.l3num) {
1358 case NFPROTO_IPV4:
1359 ret = ipv4_nlattr_to_tuple(tb, tuple, flags);
1360 break;
1361 case NFPROTO_IPV6:
1362 ret = ipv6_nlattr_to_tuple(tb, tuple, flags);
1363 break;
1364 }
1365
1366 return ret;
1367 }
1368
1369 static const struct nla_policy proto_nla_policy[CTA_PROTO_MAX+1] = {
1370 [CTA_PROTO_NUM] = { .type = NLA_U8 },
1371 };
1372
ctnetlink_parse_tuple_proto(struct nlattr * attr,struct nf_conntrack_tuple * tuple,u_int32_t flags)1373 static int ctnetlink_parse_tuple_proto(struct nlattr *attr,
1374 struct nf_conntrack_tuple *tuple,
1375 u_int32_t flags)
1376 {
1377 const struct nf_conntrack_l4proto *l4proto;
1378 struct nlattr *tb[CTA_PROTO_MAX+1];
1379 int ret = 0;
1380
1381 ret = nla_parse_nested_deprecated(tb, CTA_PROTO_MAX, attr,
1382 proto_nla_policy, NULL);
1383 if (ret < 0)
1384 return ret;
1385
1386 if (!(flags & CTA_FILTER_FLAG(CTA_PROTO_NUM)))
1387 return 0;
1388
1389 if (!tb[CTA_PROTO_NUM])
1390 return -EINVAL;
1391
1392 tuple->dst.protonum = nla_get_u8(tb[CTA_PROTO_NUM]);
1393
1394 rcu_read_lock();
1395 l4proto = nf_ct_l4proto_find(tuple->dst.protonum);
1396
1397 if (likely(l4proto->nlattr_to_tuple)) {
1398 ret = nla_validate_nested_deprecated(attr, CTA_PROTO_MAX,
1399 l4proto->nla_policy,
1400 NULL);
1401 if (ret == 0)
1402 ret = l4proto->nlattr_to_tuple(tb, tuple, flags);
1403 }
1404
1405 rcu_read_unlock();
1406
1407 return ret;
1408 }
1409
1410 static int
ctnetlink_parse_zone(const struct nlattr * attr,struct nf_conntrack_zone * zone)1411 ctnetlink_parse_zone(const struct nlattr *attr,
1412 struct nf_conntrack_zone *zone)
1413 {
1414 nf_ct_zone_init(zone, NF_CT_DEFAULT_ZONE_ID,
1415 NF_CT_DEFAULT_ZONE_DIR, 0);
1416 #ifdef CONFIG_NF_CONNTRACK_ZONES
1417 if (attr)
1418 zone->id = ntohs(nla_get_be16(attr));
1419 #else
1420 if (attr)
1421 return -EOPNOTSUPP;
1422 #endif
1423 return 0;
1424 }
1425
1426 static int
ctnetlink_parse_tuple_zone(struct nlattr * attr,enum ctattr_type type,struct nf_conntrack_zone * zone)1427 ctnetlink_parse_tuple_zone(struct nlattr *attr, enum ctattr_type type,
1428 struct nf_conntrack_zone *zone)
1429 {
1430 int ret;
1431
1432 if (zone->id != NF_CT_DEFAULT_ZONE_ID)
1433 return -EINVAL;
1434
1435 ret = ctnetlink_parse_zone(attr, zone);
1436 if (ret < 0)
1437 return ret;
1438
1439 if (type == CTA_TUPLE_REPLY)
1440 zone->dir = NF_CT_ZONE_DIR_REPL;
1441 else
1442 zone->dir = NF_CT_ZONE_DIR_ORIG;
1443
1444 return 0;
1445 }
1446
1447 static const struct nla_policy tuple_nla_policy[CTA_TUPLE_MAX+1] = {
1448 [CTA_TUPLE_IP] = { .type = NLA_NESTED },
1449 [CTA_TUPLE_PROTO] = { .type = NLA_NESTED },
1450 [CTA_TUPLE_ZONE] = { .type = NLA_U16 },
1451 };
1452
1453 #define CTA_FILTER_F_ALL_CTA_PROTO \
1454 (CTA_FILTER_F_CTA_PROTO_SRC_PORT | \
1455 CTA_FILTER_F_CTA_PROTO_DST_PORT | \
1456 CTA_FILTER_F_CTA_PROTO_ICMP_TYPE | \
1457 CTA_FILTER_F_CTA_PROTO_ICMP_CODE | \
1458 CTA_FILTER_F_CTA_PROTO_ICMP_ID | \
1459 CTA_FILTER_F_CTA_PROTO_ICMPV6_TYPE | \
1460 CTA_FILTER_F_CTA_PROTO_ICMPV6_CODE | \
1461 CTA_FILTER_F_CTA_PROTO_ICMPV6_ID)
1462
1463 static int
ctnetlink_parse_tuple_filter(const struct nlattr * const cda[],struct nf_conntrack_tuple * tuple,u32 type,u_int8_t l3num,struct nf_conntrack_zone * zone,u_int32_t flags)1464 ctnetlink_parse_tuple_filter(const struct nlattr * const cda[],
1465 struct nf_conntrack_tuple *tuple, u32 type,
1466 u_int8_t l3num, struct nf_conntrack_zone *zone,
1467 u_int32_t flags)
1468 {
1469 struct nlattr *tb[CTA_TUPLE_MAX+1];
1470 int err;
1471
1472 memset(tuple, 0, sizeof(*tuple));
1473
1474 err = nla_parse_nested_deprecated(tb, CTA_TUPLE_MAX, cda[type],
1475 tuple_nla_policy, NULL);
1476 if (err < 0)
1477 return err;
1478
1479 if (l3num != NFPROTO_IPV4 && l3num != NFPROTO_IPV6)
1480 return -EOPNOTSUPP;
1481 tuple->src.l3num = l3num;
1482
1483 if (flags & CTA_FILTER_FLAG(CTA_IP_DST) ||
1484 flags & CTA_FILTER_FLAG(CTA_IP_SRC)) {
1485 if (!tb[CTA_TUPLE_IP])
1486 return -EINVAL;
1487
1488 err = ctnetlink_parse_tuple_ip(tb[CTA_TUPLE_IP], tuple, flags);
1489 if (err < 0)
1490 return err;
1491 }
1492
1493 if (flags & CTA_FILTER_FLAG(CTA_PROTO_NUM)) {
1494 if (!tb[CTA_TUPLE_PROTO])
1495 return -EINVAL;
1496
1497 err = ctnetlink_parse_tuple_proto(tb[CTA_TUPLE_PROTO], tuple, flags);
1498 if (err < 0)
1499 return err;
1500 } else if (flags & CTA_FILTER_FLAG(ALL_CTA_PROTO)) {
1501 /* Can't manage proto flags without a protonum */
1502 return -EINVAL;
1503 }
1504
1505 if ((flags & CTA_FILTER_FLAG(CTA_TUPLE_ZONE)) && tb[CTA_TUPLE_ZONE]) {
1506 if (!zone)
1507 return -EINVAL;
1508
1509 err = ctnetlink_parse_tuple_zone(tb[CTA_TUPLE_ZONE],
1510 type, zone);
1511 if (err < 0)
1512 return err;
1513 }
1514
1515 /* orig and expect tuples get DIR_ORIGINAL */
1516 if (type == CTA_TUPLE_REPLY)
1517 tuple->dst.dir = IP_CT_DIR_REPLY;
1518 else
1519 tuple->dst.dir = IP_CT_DIR_ORIGINAL;
1520
1521 return 0;
1522 }
1523
1524 static int
ctnetlink_parse_tuple(const struct nlattr * const cda[],struct nf_conntrack_tuple * tuple,u32 type,u_int8_t l3num,struct nf_conntrack_zone * zone)1525 ctnetlink_parse_tuple(const struct nlattr * const cda[],
1526 struct nf_conntrack_tuple *tuple, u32 type,
1527 u_int8_t l3num, struct nf_conntrack_zone *zone)
1528 {
1529 return ctnetlink_parse_tuple_filter(cda, tuple, type, l3num, zone,
1530 CTA_FILTER_FLAG(ALL));
1531 }
1532
1533 static const struct nla_policy help_nla_policy[CTA_HELP_MAX+1] = {
1534 [CTA_HELP_NAME] = { .type = NLA_NUL_STRING,
1535 .len = NF_CT_HELPER_NAME_LEN - 1 },
1536 };
1537
ctnetlink_parse_help(const struct nlattr * attr,char ** helper_name,struct nlattr ** helpinfo)1538 static int ctnetlink_parse_help(const struct nlattr *attr, char **helper_name,
1539 struct nlattr **helpinfo)
1540 {
1541 int err;
1542 struct nlattr *tb[CTA_HELP_MAX+1];
1543
1544 err = nla_parse_nested_deprecated(tb, CTA_HELP_MAX, attr,
1545 help_nla_policy, NULL);
1546 if (err < 0)
1547 return err;
1548
1549 if (!tb[CTA_HELP_NAME])
1550 return -EINVAL;
1551
1552 *helper_name = nla_data(tb[CTA_HELP_NAME]);
1553
1554 if (tb[CTA_HELP_INFO])
1555 *helpinfo = tb[CTA_HELP_INFO];
1556
1557 return 0;
1558 }
1559
1560 static const struct nla_policy ct_nla_policy[CTA_MAX+1] = {
1561 [CTA_TUPLE_ORIG] = { .type = NLA_NESTED },
1562 [CTA_TUPLE_REPLY] = { .type = NLA_NESTED },
1563 [CTA_STATUS] = { .type = NLA_U32 },
1564 [CTA_PROTOINFO] = { .type = NLA_NESTED },
1565 [CTA_HELP] = { .type = NLA_NESTED },
1566 [CTA_NAT_SRC] = { .type = NLA_NESTED },
1567 [CTA_TIMEOUT] = { .type = NLA_U32 },
1568 [CTA_MARK] = { .type = NLA_U32 },
1569 [CTA_ID] = { .type = NLA_U32 },
1570 [CTA_NAT_DST] = { .type = NLA_NESTED },
1571 [CTA_TUPLE_MASTER] = { .type = NLA_NESTED },
1572 [CTA_NAT_SEQ_ADJ_ORIG] = { .type = NLA_NESTED },
1573 [CTA_NAT_SEQ_ADJ_REPLY] = { .type = NLA_NESTED },
1574 [CTA_ZONE] = { .type = NLA_U16 },
1575 [CTA_MARK_MASK] = { .type = NLA_U32 },
1576 [CTA_LABELS] = { .type = NLA_BINARY,
1577 .len = NF_CT_LABELS_MAX_SIZE },
1578 [CTA_LABELS_MASK] = { .type = NLA_BINARY,
1579 .len = NF_CT_LABELS_MAX_SIZE },
1580 [CTA_FILTER] = { .type = NLA_NESTED },
1581 [CTA_STATUS_MASK] = { .type = NLA_U32 },
1582 [CTA_TIMESTAMP_EVENT] = { .type = NLA_REJECT },
1583 };
1584
ctnetlink_flush_iterate(struct nf_conn * ct,void * data)1585 static int ctnetlink_flush_iterate(struct nf_conn *ct, void *data)
1586 {
1587 return ctnetlink_filter_match(ct, data);
1588 }
1589
ctnetlink_flush_conntrack(struct net * net,const struct nlattr * const cda[],u32 portid,int report,u8 family)1590 static int ctnetlink_flush_conntrack(struct net *net,
1591 const struct nlattr * const cda[],
1592 u32 portid, int report, u8 family)
1593 {
1594 struct ctnetlink_filter *filter = NULL;
1595 struct nf_ct_iter_data iter = {
1596 .net = net,
1597 .portid = portid,
1598 .report = report,
1599 };
1600
1601 if (ctnetlink_needs_filter(family, cda)) {
1602 filter = ctnetlink_alloc_filter(cda, family);
1603 if (IS_ERR(filter))
1604 return PTR_ERR(filter);
1605
1606 iter.data = filter;
1607 }
1608
1609 nf_ct_iterate_cleanup_net(ctnetlink_flush_iterate, &iter);
1610 kfree(filter);
1611
1612 return 0;
1613 }
1614
ctnetlink_del_conntrack(struct sk_buff * skb,const struct nfnl_info * info,const struct nlattr * const cda[])1615 static int ctnetlink_del_conntrack(struct sk_buff *skb,
1616 const struct nfnl_info *info,
1617 const struct nlattr * const cda[])
1618 {
1619 u8 family = info->nfmsg->nfgen_family;
1620 struct nf_conntrack_tuple_hash *h;
1621 struct nf_conntrack_tuple tuple;
1622 struct nf_conntrack_zone zone;
1623 struct nf_conn *ct;
1624 int err;
1625
1626 err = ctnetlink_parse_zone(cda[CTA_ZONE], &zone);
1627 if (err < 0)
1628 return err;
1629
1630 if (cda[CTA_TUPLE_ORIG] && !cda[CTA_FILTER])
1631 err = ctnetlink_parse_tuple(cda, &tuple, CTA_TUPLE_ORIG,
1632 family, &zone);
1633 else if (cda[CTA_TUPLE_REPLY] && !cda[CTA_FILTER])
1634 err = ctnetlink_parse_tuple(cda, &tuple, CTA_TUPLE_REPLY,
1635 family, &zone);
1636 else {
1637 u8 u3 = info->nfmsg->version || cda[CTA_FILTER] ? family : AF_UNSPEC;
1638
1639 return ctnetlink_flush_conntrack(info->net, cda,
1640 NETLINK_CB(skb).portid,
1641 nlmsg_report(info->nlh), u3);
1642 }
1643
1644 if (err < 0)
1645 return err;
1646
1647 h = nf_conntrack_find_get(info->net, &zone, &tuple);
1648 if (!h)
1649 return -ENOENT;
1650
1651 ct = nf_ct_tuplehash_to_ctrack(h);
1652
1653 if (cda[CTA_ID]) {
1654 __be32 id = nla_get_be32(cda[CTA_ID]);
1655
1656 if (id != (__force __be32)nf_ct_get_id(ct)) {
1657 nf_ct_put(ct);
1658 return -ENOENT;
1659 }
1660 }
1661
1662 nf_ct_delete(ct, NETLINK_CB(skb).portid, nlmsg_report(info->nlh));
1663 nf_ct_put(ct);
1664
1665 return 0;
1666 }
1667
ctnetlink_get_conntrack(struct sk_buff * skb,const struct nfnl_info * info,const struct nlattr * const cda[])1668 static int ctnetlink_get_conntrack(struct sk_buff *skb,
1669 const struct nfnl_info *info,
1670 const struct nlattr * const cda[])
1671 {
1672 u_int8_t u3 = info->nfmsg->nfgen_family;
1673 struct nf_conntrack_tuple_hash *h;
1674 struct nf_conntrack_tuple tuple;
1675 struct nf_conntrack_zone zone;
1676 struct sk_buff *skb2;
1677 struct nf_conn *ct;
1678 int err;
1679
1680 if (info->nlh->nlmsg_flags & NLM_F_DUMP) {
1681 struct netlink_dump_control c = {
1682 .start = ctnetlink_start,
1683 .dump = ctnetlink_dump_table,
1684 .done = ctnetlink_done,
1685 .data = (void *)cda,
1686 };
1687
1688 return netlink_dump_start(info->sk, skb, info->nlh, &c);
1689 }
1690
1691 err = ctnetlink_parse_zone(cda[CTA_ZONE], &zone);
1692 if (err < 0)
1693 return err;
1694
1695 if (cda[CTA_TUPLE_ORIG])
1696 err = ctnetlink_parse_tuple(cda, &tuple, CTA_TUPLE_ORIG,
1697 u3, &zone);
1698 else if (cda[CTA_TUPLE_REPLY])
1699 err = ctnetlink_parse_tuple(cda, &tuple, CTA_TUPLE_REPLY,
1700 u3, &zone);
1701 else
1702 return -EINVAL;
1703
1704 if (err < 0)
1705 return err;
1706
1707 h = nf_conntrack_find_get(info->net, &zone, &tuple);
1708 if (!h)
1709 return -ENOENT;
1710
1711 ct = nf_ct_tuplehash_to_ctrack(h);
1712
1713 skb2 = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
1714 if (!skb2) {
1715 nf_ct_put(ct);
1716 return -ENOMEM;
1717 }
1718
1719 err = ctnetlink_fill_info(skb2, NETLINK_CB(skb).portid,
1720 info->nlh->nlmsg_seq,
1721 NFNL_MSG_TYPE(info->nlh->nlmsg_type), ct,
1722 true, 0);
1723 nf_ct_put(ct);
1724 if (err <= 0) {
1725 kfree_skb(skb2);
1726 return -ENOMEM;
1727 }
1728
1729 return nfnetlink_unicast(skb2, info->net, NETLINK_CB(skb).portid);
1730 }
1731
1732 #ifdef CONFIG_NF_CONNTRACK_EVENTS
ctnetlink_dump_one_entry(struct sk_buff * skb,struct netlink_callback * cb,struct nf_conn * ct,bool dying)1733 static int ctnetlink_dump_one_entry(struct sk_buff *skb,
1734 struct netlink_callback *cb,
1735 struct nf_conn *ct,
1736 bool dying)
1737 {
1738 struct ctnetlink_list_dump_ctx *ctx = (void *)cb->ctx;
1739 struct nfgenmsg *nfmsg = nlmsg_data(cb->nlh);
1740 u8 l3proto = nfmsg->nfgen_family;
1741 int res;
1742
1743 if (l3proto && nf_ct_l3num(ct) != l3proto)
1744 return 0;
1745
1746 if (ctx->last_id) {
1747 if (ctnetlink_get_id(ct) != ctx->last_id)
1748 return 0;
1749
1750 ctx->last_id = 0;
1751 }
1752
1753 /* We can't dump extension info for the unconfirmed
1754 * list because unconfirmed conntracks can have
1755 * ct->ext reallocated (and thus freed).
1756 *
1757 * In the dying list case ct->ext can't be free'd
1758 * until after we drop pcpu->lock.
1759 */
1760 res = ctnetlink_fill_info(skb, NETLINK_CB(cb->skb).portid,
1761 cb->nlh->nlmsg_seq,
1762 NFNL_MSG_TYPE(cb->nlh->nlmsg_type),
1763 ct, dying, 0);
1764 if (res < 0)
1765 ctx->last_id = ctnetlink_get_id(ct);
1766
1767 return res;
1768 }
1769 #endif
1770
1771 static int
ctnetlink_dump_unconfirmed(struct sk_buff * skb,struct netlink_callback * cb)1772 ctnetlink_dump_unconfirmed(struct sk_buff *skb, struct netlink_callback *cb)
1773 {
1774 return 0;
1775 }
1776
1777 static int
ctnetlink_dump_dying(struct sk_buff * skb,struct netlink_callback * cb)1778 ctnetlink_dump_dying(struct sk_buff *skb, struct netlink_callback *cb)
1779 {
1780 struct ctnetlink_list_dump_ctx *ctx = (void *)cb->ctx;
1781 #ifdef CONFIG_NF_CONNTRACK_EVENTS
1782 const struct net *net = sock_net(skb->sk);
1783 struct nf_conntrack_net_ecache *ecache_net;
1784 unsigned long last_id = ctx->last_id;
1785 struct nf_conntrack_tuple_hash *h;
1786 struct hlist_nulls_node *n;
1787 #endif
1788
1789 if (ctx->done)
1790 return 0;
1791
1792 ctx->last_id = 0;
1793
1794 #ifdef CONFIG_NF_CONNTRACK_EVENTS
1795 ecache_net = nf_conn_pernet_ecache(net);
1796 spin_lock_bh(&ecache_net->dying_lock);
1797
1798 hlist_nulls_for_each_entry(h, n, &ecache_net->dying_list, hnnode) {
1799 struct nf_conn *ct;
1800 int res;
1801
1802 ct = nf_ct_tuplehash_to_ctrack(h);
1803 if (last_id && last_id != ctnetlink_get_id(ct))
1804 continue;
1805
1806 res = ctnetlink_dump_one_entry(skb, cb, ct, true);
1807 if (res < 0) {
1808 spin_unlock_bh(&ecache_net->dying_lock);
1809 return skb->len;
1810 }
1811
1812 last_id = 0;
1813 }
1814
1815 spin_unlock_bh(&ecache_net->dying_lock);
1816 #endif
1817 ctx->done = true;
1818
1819 return skb->len;
1820 }
1821
ctnetlink_get_ct_dying(struct sk_buff * skb,const struct nfnl_info * info,const struct nlattr * const cda[])1822 static int ctnetlink_get_ct_dying(struct sk_buff *skb,
1823 const struct nfnl_info *info,
1824 const struct nlattr * const cda[])
1825 {
1826 if (info->nlh->nlmsg_flags & NLM_F_DUMP) {
1827 struct netlink_dump_control c = {
1828 .dump = ctnetlink_dump_dying,
1829 };
1830 return netlink_dump_start(info->sk, skb, info->nlh, &c);
1831 }
1832
1833 return -EOPNOTSUPP;
1834 }
1835
ctnetlink_get_ct_unconfirmed(struct sk_buff * skb,const struct nfnl_info * info,const struct nlattr * const cda[])1836 static int ctnetlink_get_ct_unconfirmed(struct sk_buff *skb,
1837 const struct nfnl_info *info,
1838 const struct nlattr * const cda[])
1839 {
1840 if (info->nlh->nlmsg_flags & NLM_F_DUMP) {
1841 struct netlink_dump_control c = {
1842 .dump = ctnetlink_dump_unconfirmed,
1843 };
1844 return netlink_dump_start(info->sk, skb, info->nlh, &c);
1845 }
1846
1847 return -EOPNOTSUPP;
1848 }
1849
1850 #if IS_ENABLED(CONFIG_NF_NAT)
1851 static int
ctnetlink_parse_nat_setup(struct nf_conn * ct,enum nf_nat_manip_type manip,const struct nlattr * attr)1852 ctnetlink_parse_nat_setup(struct nf_conn *ct,
1853 enum nf_nat_manip_type manip,
1854 const struct nlattr *attr)
1855 __must_hold(RCU)
1856 {
1857 const struct nf_nat_hook *nat_hook;
1858 int err;
1859
1860 nat_hook = rcu_dereference(nf_nat_hook);
1861 if (!nat_hook) {
1862 #ifdef CONFIG_MODULES
1863 rcu_read_unlock();
1864 nfnl_unlock(NFNL_SUBSYS_CTNETLINK);
1865 if (request_module("nf-nat") < 0) {
1866 nfnl_lock(NFNL_SUBSYS_CTNETLINK);
1867 rcu_read_lock();
1868 return -EOPNOTSUPP;
1869 }
1870 nfnl_lock(NFNL_SUBSYS_CTNETLINK);
1871 rcu_read_lock();
1872 nat_hook = rcu_dereference(nf_nat_hook);
1873 if (nat_hook)
1874 return -EAGAIN;
1875 #endif
1876 return -EOPNOTSUPP;
1877 }
1878
1879 err = nat_hook->parse_nat_setup(ct, manip, attr);
1880 if (err == -EAGAIN) {
1881 #ifdef CONFIG_MODULES
1882 rcu_read_unlock();
1883 nfnl_unlock(NFNL_SUBSYS_CTNETLINK);
1884 if (request_module("nf-nat-%u", nf_ct_l3num(ct)) < 0) {
1885 nfnl_lock(NFNL_SUBSYS_CTNETLINK);
1886 rcu_read_lock();
1887 return -EOPNOTSUPP;
1888 }
1889 nfnl_lock(NFNL_SUBSYS_CTNETLINK);
1890 rcu_read_lock();
1891 #else
1892 err = -EOPNOTSUPP;
1893 #endif
1894 }
1895 return err;
1896 }
1897 #endif
1898
1899 static int
ctnetlink_change_status(struct nf_conn * ct,const struct nlattr * const cda[])1900 ctnetlink_change_status(struct nf_conn *ct, const struct nlattr * const cda[])
1901 {
1902 return nf_ct_change_status_common(ct, ntohl(nla_get_be32(cda[CTA_STATUS])));
1903 }
1904
1905 static int
ctnetlink_setup_nat(struct nf_conn * ct,const struct nlattr * const cda[])1906 ctnetlink_setup_nat(struct nf_conn *ct, const struct nlattr * const cda[])
1907 {
1908 #if IS_ENABLED(CONFIG_NF_NAT)
1909 int ret;
1910
1911 if (!cda[CTA_NAT_DST] && !cda[CTA_NAT_SRC])
1912 return 0;
1913
1914 ret = ctnetlink_parse_nat_setup(ct, NF_NAT_MANIP_DST,
1915 cda[CTA_NAT_DST]);
1916 if (ret < 0)
1917 return ret;
1918
1919 return ctnetlink_parse_nat_setup(ct, NF_NAT_MANIP_SRC,
1920 cda[CTA_NAT_SRC]);
1921 #else
1922 if (!cda[CTA_NAT_DST] && !cda[CTA_NAT_SRC])
1923 return 0;
1924 return -EOPNOTSUPP;
1925 #endif
1926 }
1927
ctnetlink_change_helper(struct nf_conn * ct,const struct nlattr * const cda[])1928 static int ctnetlink_change_helper(struct nf_conn *ct,
1929 const struct nlattr * const cda[])
1930 {
1931 struct nf_conntrack_helper *helper;
1932 struct nf_conn_help *help = nfct_help(ct);
1933 char *helpname = NULL;
1934 struct nlattr *helpinfo = NULL;
1935 int err;
1936
1937 err = ctnetlink_parse_help(cda[CTA_HELP], &helpname, &helpinfo);
1938 if (err < 0)
1939 return err;
1940
1941 rcu_read_lock();
1942 /* don't change helper of sibling connections */
1943 if (ct->master) {
1944 /* If we try to change the helper to the same thing twice,
1945 * treat the second attempt as a no-op instead of returning
1946 * an error.
1947 */
1948 err = -EBUSY;
1949 if (help) {
1950 helper = rcu_dereference(help->helper);
1951 if (helper && !strcmp(helper->name, helpname))
1952 err = 0;
1953 }
1954 rcu_read_unlock();
1955 return err;
1956 }
1957
1958 helper = __nf_conntrack_helper_find(helpname, nf_ct_l3num(ct),
1959 nf_ct_protonum(ct));
1960 if (helper == NULL) {
1961 rcu_read_unlock();
1962 return -EOPNOTSUPP;
1963 }
1964
1965 if (help) {
1966 if (rcu_access_pointer(help->helper) == helper) {
1967 /* update private helper data if allowed. */
1968 if (rcu_access_pointer(helper->help) &&
1969 helper->from_nlattr)
1970 helper->from_nlattr(helpinfo, ct);
1971 err = 0;
1972 } else
1973 err = -EBUSY;
1974 } else {
1975 /* we cannot set a helper for an existing conntrack */
1976 err = -EOPNOTSUPP;
1977 }
1978
1979 rcu_read_unlock();
1980 return err;
1981 }
1982
ctnetlink_change_timeout(struct nf_conn * ct,const struct nlattr * const cda[])1983 static int ctnetlink_change_timeout(struct nf_conn *ct,
1984 const struct nlattr * const cda[])
1985 {
1986 return __nf_ct_change_timeout(ct, (u64)ntohl(nla_get_be32(cda[CTA_TIMEOUT])) * HZ);
1987 }
1988
1989 #if defined(CONFIG_NF_CONNTRACK_MARK)
ctnetlink_change_mark(struct nf_conn * ct,const struct nlattr * const cda[])1990 static void ctnetlink_change_mark(struct nf_conn *ct,
1991 const struct nlattr * const cda[])
1992 {
1993 u32 mark, newmark, mask = 0;
1994
1995 if (cda[CTA_MARK_MASK])
1996 mask = ~ntohl(nla_get_be32(cda[CTA_MARK_MASK]));
1997
1998 mark = ntohl(nla_get_be32(cda[CTA_MARK]));
1999 newmark = (READ_ONCE(ct->mark) & mask) ^ mark;
2000 if (newmark != READ_ONCE(ct->mark))
2001 WRITE_ONCE(ct->mark, newmark);
2002 }
2003 #endif
2004
2005 static const struct nla_policy protoinfo_policy[CTA_PROTOINFO_MAX+1] = {
2006 [CTA_PROTOINFO_TCP] = { .type = NLA_NESTED },
2007 [CTA_PROTOINFO_SCTP] = { .type = NLA_NESTED },
2008 };
2009
ctnetlink_change_protoinfo(struct nf_conn * ct,const struct nlattr * const cda[])2010 static int ctnetlink_change_protoinfo(struct nf_conn *ct,
2011 const struct nlattr * const cda[])
2012 {
2013 const struct nlattr *attr = cda[CTA_PROTOINFO];
2014 const struct nf_conntrack_l4proto *l4proto;
2015 struct nlattr *tb[CTA_PROTOINFO_MAX+1];
2016 int err = 0;
2017
2018 err = nla_parse_nested_deprecated(tb, CTA_PROTOINFO_MAX, attr,
2019 protoinfo_policy, NULL);
2020 if (err < 0)
2021 return err;
2022
2023 l4proto = nf_ct_l4proto_find(nf_ct_protonum(ct));
2024 if (l4proto->from_nlattr)
2025 err = l4proto->from_nlattr(tb, ct);
2026
2027 return err;
2028 }
2029
2030 static const struct nla_policy seqadj_policy[CTA_SEQADJ_MAX+1] = {
2031 [CTA_SEQADJ_CORRECTION_POS] = { .type = NLA_U32 },
2032 [CTA_SEQADJ_OFFSET_BEFORE] = { .type = NLA_U32 },
2033 [CTA_SEQADJ_OFFSET_AFTER] = { .type = NLA_U32 },
2034 };
2035
change_seq_adj(struct nf_ct_seqadj * seq,const struct nlattr * const attr)2036 static int change_seq_adj(struct nf_ct_seqadj *seq,
2037 const struct nlattr * const attr)
2038 {
2039 int err;
2040 struct nlattr *cda[CTA_SEQADJ_MAX+1];
2041
2042 err = nla_parse_nested_deprecated(cda, CTA_SEQADJ_MAX, attr,
2043 seqadj_policy, NULL);
2044 if (err < 0)
2045 return err;
2046
2047 if (!cda[CTA_SEQADJ_CORRECTION_POS])
2048 return -EINVAL;
2049
2050 seq->correction_pos =
2051 ntohl(nla_get_be32(cda[CTA_SEQADJ_CORRECTION_POS]));
2052
2053 if (!cda[CTA_SEQADJ_OFFSET_BEFORE])
2054 return -EINVAL;
2055
2056 seq->offset_before =
2057 ntohl(nla_get_be32(cda[CTA_SEQADJ_OFFSET_BEFORE]));
2058
2059 if (!cda[CTA_SEQADJ_OFFSET_AFTER])
2060 return -EINVAL;
2061
2062 seq->offset_after =
2063 ntohl(nla_get_be32(cda[CTA_SEQADJ_OFFSET_AFTER]));
2064
2065 return 0;
2066 }
2067
2068 static int
ctnetlink_change_seq_adj(struct nf_conn * ct,const struct nlattr * const cda[])2069 ctnetlink_change_seq_adj(struct nf_conn *ct,
2070 const struct nlattr * const cda[])
2071 {
2072 struct nf_conn_seqadj *seqadj = nfct_seqadj(ct);
2073 int ret = 0;
2074
2075 if (!seqadj)
2076 return 0;
2077
2078 spin_lock_bh(&ct->lock);
2079 if (cda[CTA_SEQ_ADJ_ORIG]) {
2080 ret = change_seq_adj(&seqadj->seq[IP_CT_DIR_ORIGINAL],
2081 cda[CTA_SEQ_ADJ_ORIG]);
2082 if (ret < 0)
2083 goto err;
2084
2085 set_bit(IPS_SEQ_ADJUST_BIT, &ct->status);
2086 }
2087
2088 if (cda[CTA_SEQ_ADJ_REPLY]) {
2089 ret = change_seq_adj(&seqadj->seq[IP_CT_DIR_REPLY],
2090 cda[CTA_SEQ_ADJ_REPLY]);
2091 if (ret < 0)
2092 goto err;
2093
2094 set_bit(IPS_SEQ_ADJUST_BIT, &ct->status);
2095 }
2096
2097 spin_unlock_bh(&ct->lock);
2098 return 0;
2099 err:
2100 spin_unlock_bh(&ct->lock);
2101 return ret;
2102 }
2103
2104 static const struct nla_policy synproxy_policy[CTA_SYNPROXY_MAX + 1] = {
2105 [CTA_SYNPROXY_ISN] = { .type = NLA_U32 },
2106 [CTA_SYNPROXY_ITS] = { .type = NLA_U32 },
2107 [CTA_SYNPROXY_TSOFF] = { .type = NLA_U32 },
2108 };
2109
ctnetlink_change_synproxy(struct nf_conn * ct,const struct nlattr * const cda[])2110 static int ctnetlink_change_synproxy(struct nf_conn *ct,
2111 const struct nlattr * const cda[])
2112 {
2113 struct nf_conn_synproxy *synproxy = nfct_synproxy(ct);
2114 struct nlattr *tb[CTA_SYNPROXY_MAX + 1];
2115 int err;
2116
2117 if (!synproxy)
2118 return 0;
2119
2120 err = nla_parse_nested_deprecated(tb, CTA_SYNPROXY_MAX,
2121 cda[CTA_SYNPROXY], synproxy_policy,
2122 NULL);
2123 if (err < 0)
2124 return err;
2125
2126 if (!tb[CTA_SYNPROXY_ISN] ||
2127 !tb[CTA_SYNPROXY_ITS] ||
2128 !tb[CTA_SYNPROXY_TSOFF])
2129 return -EINVAL;
2130
2131 synproxy->isn = ntohl(nla_get_be32(tb[CTA_SYNPROXY_ISN]));
2132 synproxy->its = ntohl(nla_get_be32(tb[CTA_SYNPROXY_ITS]));
2133 synproxy->tsoff = ntohl(nla_get_be32(tb[CTA_SYNPROXY_TSOFF]));
2134
2135 return 0;
2136 }
2137
2138 static int
ctnetlink_attach_labels(struct nf_conn * ct,const struct nlattr * const cda[])2139 ctnetlink_attach_labels(struct nf_conn *ct, const struct nlattr * const cda[])
2140 {
2141 #ifdef CONFIG_NF_CONNTRACK_LABELS
2142 size_t len = nla_len(cda[CTA_LABELS]);
2143 const void *mask = cda[CTA_LABELS_MASK];
2144
2145 if (len & (sizeof(u32)-1)) /* must be multiple of u32 */
2146 return -EINVAL;
2147
2148 if (mask) {
2149 if (nla_len(cda[CTA_LABELS_MASK]) == 0 ||
2150 nla_len(cda[CTA_LABELS_MASK]) != len)
2151 return -EINVAL;
2152 mask = nla_data(cda[CTA_LABELS_MASK]);
2153 }
2154
2155 len /= sizeof(u32);
2156
2157 return nf_connlabels_replace(ct, nla_data(cda[CTA_LABELS]), mask, len);
2158 #else
2159 return -EOPNOTSUPP;
2160 #endif
2161 }
2162
2163 static int
ctnetlink_change_conntrack(struct nf_conn * ct,const struct nlattr * const cda[])2164 ctnetlink_change_conntrack(struct nf_conn *ct,
2165 const struct nlattr * const cda[])
2166 {
2167 int err;
2168
2169 /* only allow NAT changes and master assignation for new conntracks */
2170 if (cda[CTA_NAT_SRC] || cda[CTA_NAT_DST] || cda[CTA_TUPLE_MASTER])
2171 return -EOPNOTSUPP;
2172
2173 if (cda[CTA_HELP]) {
2174 err = ctnetlink_change_helper(ct, cda);
2175 if (err < 0)
2176 return err;
2177 }
2178
2179 if (cda[CTA_TIMEOUT]) {
2180 err = ctnetlink_change_timeout(ct, cda);
2181 if (err < 0)
2182 return err;
2183 }
2184
2185 if (cda[CTA_STATUS]) {
2186 err = ctnetlink_change_status(ct, cda);
2187 if (err < 0)
2188 return err;
2189 }
2190
2191 if (cda[CTA_PROTOINFO]) {
2192 err = ctnetlink_change_protoinfo(ct, cda);
2193 if (err < 0)
2194 return err;
2195 }
2196
2197 #if defined(CONFIG_NF_CONNTRACK_MARK)
2198 if (cda[CTA_MARK])
2199 ctnetlink_change_mark(ct, cda);
2200 #endif
2201
2202 if (cda[CTA_SEQ_ADJ_ORIG] || cda[CTA_SEQ_ADJ_REPLY]) {
2203 err = ctnetlink_change_seq_adj(ct, cda);
2204 if (err < 0)
2205 return err;
2206 }
2207
2208 if (cda[CTA_SYNPROXY]) {
2209 err = ctnetlink_change_synproxy(ct, cda);
2210 if (err < 0)
2211 return err;
2212 }
2213
2214 if (cda[CTA_LABELS]) {
2215 err = ctnetlink_attach_labels(ct, cda);
2216 if (err < 0)
2217 return err;
2218 }
2219
2220 return 0;
2221 }
2222
2223 static struct nf_conn *
ctnetlink_create_conntrack(struct net * net,const struct nf_conntrack_zone * zone,const struct nlattr * const cda[],struct nf_conntrack_tuple * otuple,struct nf_conntrack_tuple * rtuple,u8 u3)2224 ctnetlink_create_conntrack(struct net *net,
2225 const struct nf_conntrack_zone *zone,
2226 const struct nlattr * const cda[],
2227 struct nf_conntrack_tuple *otuple,
2228 struct nf_conntrack_tuple *rtuple,
2229 u8 u3)
2230 {
2231 struct nf_conn *ct;
2232 int err = -EINVAL;
2233 struct nf_conntrack_helper *helper;
2234 struct nf_conn_tstamp *tstamp;
2235 u64 timeout;
2236
2237 ct = nf_conntrack_alloc(net, zone, otuple, rtuple, GFP_ATOMIC);
2238 if (IS_ERR(ct))
2239 return ERR_PTR(-ENOMEM);
2240
2241 if (!cda[CTA_TIMEOUT])
2242 goto err1;
2243
2244 rcu_read_lock();
2245 if (cda[CTA_HELP]) {
2246 char *helpname = NULL;
2247 struct nlattr *helpinfo = NULL;
2248
2249 err = ctnetlink_parse_help(cda[CTA_HELP], &helpname, &helpinfo);
2250 if (err < 0)
2251 goto err2;
2252
2253 helper = __nf_conntrack_helper_find(helpname, nf_ct_l3num(ct),
2254 nf_ct_protonum(ct));
2255 if (helper == NULL) {
2256 rcu_read_unlock();
2257 #ifdef CONFIG_MODULES
2258 if (request_module("nfct-helper-%s", helpname) < 0) {
2259 err = -EOPNOTSUPP;
2260 goto err1;
2261 }
2262
2263 rcu_read_lock();
2264 helper = __nf_conntrack_helper_find(helpname,
2265 nf_ct_l3num(ct),
2266 nf_ct_protonum(ct));
2267 if (helper) {
2268 err = -EAGAIN;
2269 goto err2;
2270 }
2271 rcu_read_unlock();
2272 #endif
2273 err = -EOPNOTSUPP;
2274 goto err1;
2275 } else {
2276 struct nf_conn_help *help;
2277
2278 help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
2279 if (help == NULL) {
2280 err = -ENOMEM;
2281 goto err2;
2282 }
2283 /* set private helper data if allowed. */
2284 if (rcu_access_pointer(helper->help) &&
2285 helper->from_nlattr)
2286 helper->from_nlattr(helpinfo, ct);
2287
2288 /* disable helper auto-assignment for this entry */
2289 ct->status |= IPS_HELPER;
2290 if (!refcount_inc_not_zero(&helper->ct_refcnt)) {
2291 err = -ENOENT;
2292 goto err2;
2293 }
2294 RCU_INIT_POINTER(help->helper, helper);
2295 }
2296 }
2297
2298 err = ctnetlink_setup_nat(ct, cda);
2299 if (err < 0)
2300 goto err2;
2301
2302 nf_ct_acct_ext_add(ct, GFP_ATOMIC);
2303 nf_ct_tstamp_ext_add(ct, GFP_ATOMIC);
2304 nf_ct_ecache_ext_add(ct, 0, 0, GFP_ATOMIC);
2305 nf_ct_labels_ext_add(ct);
2306 nfct_seqadj_ext_add(ct);
2307 nfct_synproxy_ext_add(ct);
2308
2309 /* we must add conntrack extensions before confirmation. */
2310 ct->status |= IPS_CONFIRMED;
2311
2312 timeout = (u64)ntohl(nla_get_be32(cda[CTA_TIMEOUT])) * HZ;
2313 __nf_ct_set_timeout(ct, timeout);
2314
2315 if (cda[CTA_STATUS]) {
2316 err = ctnetlink_change_status(ct, cda);
2317 if (err < 0)
2318 goto err2;
2319 }
2320
2321 if (cda[CTA_SEQ_ADJ_ORIG] || cda[CTA_SEQ_ADJ_REPLY]) {
2322 err = ctnetlink_change_seq_adj(ct, cda);
2323 if (err < 0)
2324 goto err2;
2325 }
2326
2327 memset(&ct->proto, 0, sizeof(ct->proto));
2328 if (cda[CTA_PROTOINFO]) {
2329 err = ctnetlink_change_protoinfo(ct, cda);
2330 if (err < 0)
2331 goto err2;
2332 }
2333
2334 if (cda[CTA_SYNPROXY]) {
2335 err = ctnetlink_change_synproxy(ct, cda);
2336 if (err < 0)
2337 goto err2;
2338 }
2339
2340 #if defined(CONFIG_NF_CONNTRACK_MARK)
2341 if (cda[CTA_MARK])
2342 ctnetlink_change_mark(ct, cda);
2343 #endif
2344
2345 /* setup master conntrack: this is a confirmed expectation */
2346 if (cda[CTA_TUPLE_MASTER]) {
2347 struct nf_conntrack_tuple master;
2348 struct nf_conntrack_tuple_hash *master_h;
2349 struct nf_conn *master_ct;
2350
2351 err = ctnetlink_parse_tuple(cda, &master, CTA_TUPLE_MASTER,
2352 u3, NULL);
2353 if (err < 0)
2354 goto err2;
2355
2356 master_h = nf_conntrack_find_get(net, zone, &master);
2357 if (master_h == NULL) {
2358 err = -ENOENT;
2359 goto err2;
2360 }
2361 master_ct = nf_ct_tuplehash_to_ctrack(master_h);
2362 __set_bit(IPS_EXPECTED_BIT, &ct->status);
2363 ct->master = master_ct;
2364 }
2365 tstamp = nf_conn_tstamp_find(ct);
2366 if (tstamp)
2367 tstamp->start = ktime_get_real_ns();
2368
2369 err = nf_conntrack_hash_check_insert(ct);
2370 if (err < 0)
2371 goto err3;
2372
2373 rcu_read_unlock();
2374
2375 return ct;
2376
2377 err3:
2378 if (ct->master)
2379 nf_ct_put(ct->master);
2380 err2:
2381 rcu_read_unlock();
2382 err1:
2383 nf_conntrack_free(ct);
2384 return ERR_PTR(err);
2385 }
2386
ctnetlink_new_conntrack(struct sk_buff * skb,const struct nfnl_info * info,const struct nlattr * const cda[])2387 static int ctnetlink_new_conntrack(struct sk_buff *skb,
2388 const struct nfnl_info *info,
2389 const struct nlattr * const cda[])
2390 {
2391 struct nf_conntrack_tuple otuple, rtuple;
2392 struct nf_conntrack_tuple_hash *h = NULL;
2393 u_int8_t u3 = info->nfmsg->nfgen_family;
2394 struct nf_conntrack_zone zone;
2395 struct nf_conn *ct;
2396 int err;
2397
2398 err = ctnetlink_parse_zone(cda[CTA_ZONE], &zone);
2399 if (err < 0)
2400 return err;
2401
2402 if (cda[CTA_TUPLE_ORIG]) {
2403 err = ctnetlink_parse_tuple(cda, &otuple, CTA_TUPLE_ORIG,
2404 u3, &zone);
2405 if (err < 0)
2406 return err;
2407 }
2408
2409 if (cda[CTA_TUPLE_REPLY]) {
2410 err = ctnetlink_parse_tuple(cda, &rtuple, CTA_TUPLE_REPLY,
2411 u3, &zone);
2412 if (err < 0)
2413 return err;
2414 }
2415
2416 if (cda[CTA_TUPLE_ORIG])
2417 h = nf_conntrack_find_get(info->net, &zone, &otuple);
2418 else if (cda[CTA_TUPLE_REPLY])
2419 h = nf_conntrack_find_get(info->net, &zone, &rtuple);
2420
2421 if (h == NULL) {
2422 err = -ENOENT;
2423 if (info->nlh->nlmsg_flags & NLM_F_CREATE) {
2424 enum ip_conntrack_events events;
2425
2426 if (!cda[CTA_TUPLE_ORIG] || !cda[CTA_TUPLE_REPLY])
2427 return -EINVAL;
2428 if (otuple.dst.protonum != rtuple.dst.protonum)
2429 return -EINVAL;
2430
2431 ct = ctnetlink_create_conntrack(info->net, &zone, cda,
2432 &otuple, &rtuple, u3);
2433 if (IS_ERR(ct))
2434 return PTR_ERR(ct);
2435
2436 err = 0;
2437 if (test_bit(IPS_EXPECTED_BIT, &ct->status))
2438 events = 1 << IPCT_RELATED;
2439 else
2440 events = 1 << IPCT_NEW;
2441
2442 if (cda[CTA_LABELS] &&
2443 ctnetlink_attach_labels(ct, cda) == 0)
2444 events |= (1 << IPCT_LABEL);
2445
2446 nf_conntrack_eventmask_report((1 << IPCT_REPLY) |
2447 (1 << IPCT_ASSURED) |
2448 (1 << IPCT_HELPER) |
2449 (1 << IPCT_PROTOINFO) |
2450 (1 << IPCT_SEQADJ) |
2451 (1 << IPCT_MARK) |
2452 (1 << IPCT_SYNPROXY) |
2453 events,
2454 ct, NETLINK_CB(skb).portid,
2455 nlmsg_report(info->nlh));
2456 nf_ct_put(ct);
2457 }
2458
2459 return err;
2460 }
2461 /* implicit 'else' */
2462
2463 err = -EEXIST;
2464 ct = nf_ct_tuplehash_to_ctrack(h);
2465 if (!(info->nlh->nlmsg_flags & NLM_F_EXCL)) {
2466 err = ctnetlink_change_conntrack(ct, cda);
2467 if (err == 0) {
2468 nf_conntrack_eventmask_report((1 << IPCT_REPLY) |
2469 (1 << IPCT_ASSURED) |
2470 (1 << IPCT_HELPER) |
2471 (1 << IPCT_LABEL) |
2472 (1 << IPCT_PROTOINFO) |
2473 (1 << IPCT_SEQADJ) |
2474 (1 << IPCT_MARK) |
2475 (1 << IPCT_SYNPROXY),
2476 ct, NETLINK_CB(skb).portid,
2477 nlmsg_report(info->nlh));
2478 }
2479 }
2480
2481 nf_ct_put(ct);
2482 return err;
2483 }
2484
2485 static int
ctnetlink_ct_stat_cpu_fill_info(struct sk_buff * skb,u32 portid,u32 seq,__u16 cpu,const struct ip_conntrack_stat * st)2486 ctnetlink_ct_stat_cpu_fill_info(struct sk_buff *skb, u32 portid, u32 seq,
2487 __u16 cpu, const struct ip_conntrack_stat *st)
2488 {
2489 struct nlmsghdr *nlh;
2490 unsigned int flags = portid ? NLM_F_MULTI : 0, event;
2491
2492 event = nfnl_msg_type(NFNL_SUBSYS_CTNETLINK,
2493 IPCTNL_MSG_CT_GET_STATS_CPU);
2494 nlh = nfnl_msg_put(skb, portid, seq, event, flags, AF_UNSPEC,
2495 NFNETLINK_V0, htons(cpu));
2496 if (!nlh)
2497 goto nlmsg_failure;
2498
2499 if (nla_put_be32(skb, CTA_STATS_FOUND, htonl(st->found)) ||
2500 nla_put_be32(skb, CTA_STATS_INVALID, htonl(st->invalid)) ||
2501 nla_put_be32(skb, CTA_STATS_INSERT, htonl(st->insert)) ||
2502 nla_put_be32(skb, CTA_STATS_INSERT_FAILED,
2503 htonl(st->insert_failed)) ||
2504 nla_put_be32(skb, CTA_STATS_DROP, htonl(st->drop)) ||
2505 nla_put_be32(skb, CTA_STATS_EARLY_DROP, htonl(st->early_drop)) ||
2506 nla_put_be32(skb, CTA_STATS_ERROR, htonl(st->error)) ||
2507 nla_put_be32(skb, CTA_STATS_SEARCH_RESTART,
2508 htonl(st->search_restart)) ||
2509 nla_put_be32(skb, CTA_STATS_CLASH_RESOLVE,
2510 htonl(st->clash_resolve)) ||
2511 nla_put_be32(skb, CTA_STATS_CHAIN_TOOLONG,
2512 htonl(st->chaintoolong)))
2513 goto nla_put_failure;
2514
2515 nlmsg_end(skb, nlh);
2516 return skb->len;
2517
2518 nla_put_failure:
2519 nlmsg_failure:
2520 nlmsg_cancel(skb, nlh);
2521 return -1;
2522 }
2523
2524 static int
ctnetlink_ct_stat_cpu_dump(struct sk_buff * skb,struct netlink_callback * cb)2525 ctnetlink_ct_stat_cpu_dump(struct sk_buff *skb, struct netlink_callback *cb)
2526 {
2527 int cpu;
2528 struct net *net = sock_net(skb->sk);
2529
2530 if (cb->args[0] == nr_cpu_ids)
2531 return 0;
2532
2533 for (cpu = cb->args[0]; cpu < nr_cpu_ids; cpu++) {
2534 const struct ip_conntrack_stat *st;
2535
2536 if (!cpu_possible(cpu))
2537 continue;
2538
2539 st = per_cpu_ptr(net->ct.stat, cpu);
2540 if (ctnetlink_ct_stat_cpu_fill_info(skb,
2541 NETLINK_CB(cb->skb).portid,
2542 cb->nlh->nlmsg_seq,
2543 cpu, st) < 0)
2544 break;
2545 }
2546 cb->args[0] = cpu;
2547
2548 return skb->len;
2549 }
2550
ctnetlink_stat_ct_cpu(struct sk_buff * skb,const struct nfnl_info * info,const struct nlattr * const cda[])2551 static int ctnetlink_stat_ct_cpu(struct sk_buff *skb,
2552 const struct nfnl_info *info,
2553 const struct nlattr * const cda[])
2554 {
2555 if (info->nlh->nlmsg_flags & NLM_F_DUMP) {
2556 struct netlink_dump_control c = {
2557 .dump = ctnetlink_ct_stat_cpu_dump,
2558 };
2559 return netlink_dump_start(info->sk, skb, info->nlh, &c);
2560 }
2561
2562 return 0;
2563 }
2564
2565 static int
ctnetlink_stat_ct_fill_info(struct sk_buff * skb,u32 portid,u32 seq,u32 type,struct net * net)2566 ctnetlink_stat_ct_fill_info(struct sk_buff *skb, u32 portid, u32 seq, u32 type,
2567 struct net *net)
2568 {
2569 unsigned int flags = portid ? NLM_F_MULTI : 0, event;
2570 unsigned int nr_conntracks;
2571 struct nlmsghdr *nlh;
2572
2573 event = nfnl_msg_type(NFNL_SUBSYS_CTNETLINK, IPCTNL_MSG_CT_GET_STATS);
2574 nlh = nfnl_msg_put(skb, portid, seq, event, flags, AF_UNSPEC,
2575 NFNETLINK_V0, 0);
2576 if (!nlh)
2577 goto nlmsg_failure;
2578
2579 nr_conntracks = nf_conntrack_count(net);
2580 if (nla_put_be32(skb, CTA_STATS_GLOBAL_ENTRIES, htonl(nr_conntracks)))
2581 goto nla_put_failure;
2582
2583 if (nla_put_be32(skb, CTA_STATS_GLOBAL_MAX_ENTRIES, htonl(nf_conntrack_max)))
2584 goto nla_put_failure;
2585
2586 nlmsg_end(skb, nlh);
2587 return skb->len;
2588
2589 nla_put_failure:
2590 nlmsg_failure:
2591 nlmsg_cancel(skb, nlh);
2592 return -1;
2593 }
2594
ctnetlink_stat_ct(struct sk_buff * skb,const struct nfnl_info * info,const struct nlattr * const cda[])2595 static int ctnetlink_stat_ct(struct sk_buff *skb, const struct nfnl_info *info,
2596 const struct nlattr * const cda[])
2597 {
2598 struct sk_buff *skb2;
2599 int err;
2600
2601 skb2 = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
2602 if (skb2 == NULL)
2603 return -ENOMEM;
2604
2605 err = ctnetlink_stat_ct_fill_info(skb2, NETLINK_CB(skb).portid,
2606 info->nlh->nlmsg_seq,
2607 NFNL_MSG_TYPE(info->nlh->nlmsg_type),
2608 sock_net(skb->sk));
2609 if (err <= 0) {
2610 kfree_skb(skb2);
2611 return -ENOMEM;
2612 }
2613
2614 return nfnetlink_unicast(skb2, info->net, NETLINK_CB(skb).portid);
2615 }
2616
2617 static const struct nla_policy exp_nla_policy[CTA_EXPECT_MAX+1] = {
2618 [CTA_EXPECT_MASTER] = { .type = NLA_NESTED },
2619 [CTA_EXPECT_TUPLE] = { .type = NLA_NESTED },
2620 [CTA_EXPECT_MASK] = { .type = NLA_NESTED },
2621 [CTA_EXPECT_TIMEOUT] = { .type = NLA_U32 },
2622 [CTA_EXPECT_ID] = { .type = NLA_U32 },
2623 [CTA_EXPECT_HELP_NAME] = { .type = NLA_NUL_STRING,
2624 .len = NF_CT_HELPER_NAME_LEN - 1 },
2625 [CTA_EXPECT_ZONE] = { .type = NLA_U16 },
2626 [CTA_EXPECT_FLAGS] = NLA_POLICY_MASK(NLA_BE32, NF_CT_EXPECT_MASK),
2627 [CTA_EXPECT_CLASS] = { .type = NLA_U32 },
2628 [CTA_EXPECT_NAT] = { .type = NLA_NESTED },
2629 [CTA_EXPECT_FN] = { .type = NLA_NUL_STRING },
2630 };
2631
2632 static struct nf_conntrack_expect *
2633 ctnetlink_alloc_expect(const struct nlattr *const cda[], struct nf_conn *ct,
2634 const struct nf_conntrack_helper *assign_helper,
2635 struct nf_conntrack_tuple *tuple,
2636 struct nf_conntrack_tuple *mask);
2637
2638 #ifdef CONFIG_NETFILTER_NETLINK_GLUE_CT
2639 static size_t
ctnetlink_glue_build_size(const struct nf_conn * ct)2640 ctnetlink_glue_build_size(const struct nf_conn *ct)
2641 {
2642 return 3 * nla_total_size(0) /* CTA_TUPLE_ORIG|REPL|MASTER */
2643 + 3 * nla_total_size(0) /* CTA_TUPLE_IP */
2644 + 3 * nla_total_size(0) /* CTA_TUPLE_PROTO */
2645 + 3 * nla_total_size(sizeof(u_int8_t)) /* CTA_PROTO_NUM */
2646 + nla_total_size(sizeof(u_int32_t)) /* CTA_ID */
2647 + nla_total_size(sizeof(u_int32_t)) /* CTA_STATUS */
2648 + nla_total_size(sizeof(u_int32_t)) /* CTA_TIMEOUT */
2649 + nla_total_size(0) /* CTA_PROTOINFO */
2650 + nla_total_size(0) /* CTA_HELP */
2651 + nla_total_size(NF_CT_HELPER_NAME_LEN) /* CTA_HELP_NAME */
2652 + ctnetlink_secctx_size(ct)
2653 + ctnetlink_acct_size(ct)
2654 + ctnetlink_timestamp_size(ct)
2655 #if IS_ENABLED(CONFIG_NF_NAT)
2656 + 2 * nla_total_size(0) /* CTA_NAT_SEQ_ADJ_ORIG|REPL */
2657 + 6 * nla_total_size(sizeof(u_int32_t)) /* CTA_NAT_SEQ_OFFSET */
2658 #endif
2659 #ifdef CONFIG_NF_CONNTRACK_MARK
2660 + nla_total_size(sizeof(u_int32_t)) /* CTA_MARK */
2661 #endif
2662 #ifdef CONFIG_NF_CONNTRACK_ZONES
2663 + nla_total_size(sizeof(u_int16_t)) /* CTA_ZONE|CTA_TUPLE_ZONE */
2664 #endif
2665 + ctnetlink_proto_size(ct)
2666 ;
2667 }
2668
__ctnetlink_glue_build(struct sk_buff * skb,struct nf_conn * ct)2669 static int __ctnetlink_glue_build(struct sk_buff *skb, struct nf_conn *ct)
2670 {
2671 const struct nf_conntrack_zone *zone;
2672 struct nlattr *nest_parms;
2673
2674 zone = nf_ct_zone(ct);
2675
2676 nest_parms = nla_nest_start(skb, CTA_TUPLE_ORIG);
2677 if (!nest_parms)
2678 goto nla_put_failure;
2679 if (ctnetlink_dump_tuples(skb, nf_ct_tuple(ct, IP_CT_DIR_ORIGINAL)) < 0)
2680 goto nla_put_failure;
2681 if (ctnetlink_dump_zone_id(skb, CTA_TUPLE_ZONE, zone,
2682 NF_CT_ZONE_DIR_ORIG) < 0)
2683 goto nla_put_failure;
2684 nla_nest_end(skb, nest_parms);
2685
2686 nest_parms = nla_nest_start(skb, CTA_TUPLE_REPLY);
2687 if (!nest_parms)
2688 goto nla_put_failure;
2689 if (ctnetlink_dump_tuples(skb, nf_ct_tuple(ct, IP_CT_DIR_REPLY)) < 0)
2690 goto nla_put_failure;
2691 if (ctnetlink_dump_zone_id(skb, CTA_TUPLE_ZONE, zone,
2692 NF_CT_ZONE_DIR_REPL) < 0)
2693 goto nla_put_failure;
2694 nla_nest_end(skb, nest_parms);
2695
2696 if (ctnetlink_dump_zone_id(skb, CTA_ZONE, zone,
2697 NF_CT_DEFAULT_ZONE_DIR) < 0)
2698 goto nla_put_failure;
2699
2700 if (ctnetlink_dump_id(skb, ct) < 0)
2701 goto nla_put_failure;
2702
2703 if (ctnetlink_dump_status(skb, ct) < 0)
2704 goto nla_put_failure;
2705
2706 if (ctnetlink_dump_timeout(skb, ct, false) < 0)
2707 goto nla_put_failure;
2708
2709 if (ctnetlink_dump_protoinfo(skb, ct, false) < 0)
2710 goto nla_put_failure;
2711
2712 if (ctnetlink_dump_acct(skb, ct, IPCTNL_MSG_CT_GET) < 0 ||
2713 ctnetlink_dump_timestamp(skb, ct) < 0)
2714 goto nla_put_failure;
2715
2716 if (ctnetlink_dump_helpinfo(skb, ct) < 0)
2717 goto nla_put_failure;
2718
2719 #ifdef CONFIG_NF_CONNTRACK_SECMARK
2720 if (ct->secmark && ctnetlink_dump_secctx(skb, ct) < 0)
2721 goto nla_put_failure;
2722 #endif
2723 if (ct->master && ctnetlink_dump_master(skb, ct) < 0)
2724 goto nla_put_failure;
2725
2726 if ((ct->status & IPS_SEQ_ADJUST) &&
2727 ctnetlink_dump_ct_seq_adj(skb, ct) < 0)
2728 goto nla_put_failure;
2729
2730 if (ctnetlink_dump_ct_synproxy(skb, ct) < 0)
2731 goto nla_put_failure;
2732
2733 #ifdef CONFIG_NF_CONNTRACK_MARK
2734 if (ctnetlink_dump_mark(skb, ct, true) < 0)
2735 goto nla_put_failure;
2736 #endif
2737 if (ctnetlink_dump_labels(skb, ct) < 0)
2738 goto nla_put_failure;
2739 return 0;
2740
2741 nla_put_failure:
2742 return -ENOSPC;
2743 }
2744
2745 static int
ctnetlink_glue_build(struct sk_buff * skb,struct nf_conn * ct,enum ip_conntrack_info ctinfo,u_int16_t ct_attr,u_int16_t ct_info_attr)2746 ctnetlink_glue_build(struct sk_buff *skb, struct nf_conn *ct,
2747 enum ip_conntrack_info ctinfo,
2748 u_int16_t ct_attr, u_int16_t ct_info_attr)
2749 {
2750 struct nlattr *nest_parms;
2751
2752 nest_parms = nla_nest_start(skb, ct_attr);
2753 if (!nest_parms)
2754 goto nla_put_failure;
2755
2756 if (__ctnetlink_glue_build(skb, ct) < 0)
2757 goto nla_put_failure;
2758
2759 nla_nest_end(skb, nest_parms);
2760
2761 if (nla_put_be32(skb, ct_info_attr, htonl(ctinfo)))
2762 goto nla_put_failure;
2763
2764 return 0;
2765
2766 nla_put_failure:
2767 return -ENOSPC;
2768 }
2769
2770 static int
ctnetlink_update_status(struct nf_conn * ct,const struct nlattr * const cda[])2771 ctnetlink_update_status(struct nf_conn *ct, const struct nlattr * const cda[])
2772 {
2773 unsigned int status = ntohl(nla_get_be32(cda[CTA_STATUS]));
2774 unsigned long d = ct->status ^ status;
2775
2776 if (d & IPS_SEEN_REPLY && !(status & IPS_SEEN_REPLY))
2777 /* SEEN_REPLY bit can only be set */
2778 return -EBUSY;
2779
2780 if (d & IPS_ASSURED && !(status & IPS_ASSURED))
2781 /* ASSURED bit can only be set */
2782 return -EBUSY;
2783
2784 /* This check is less strict than ctnetlink_change_status()
2785 * because callers often flip IPS_EXPECTED bits when sending
2786 * an NFQA_CT attribute to the kernel. So ignore the
2787 * unchangeable bits but do not error out. Also user programs
2788 * are allowed to clear the bits that they are allowed to change.
2789 */
2790 __nf_ct_change_status(ct, status, ~status);
2791 return 0;
2792 }
2793
2794 static int
ctnetlink_glue_parse_ct(const struct nlattr * cda[],struct nf_conn * ct)2795 ctnetlink_glue_parse_ct(const struct nlattr *cda[], struct nf_conn *ct)
2796 {
2797 int err;
2798
2799 if (cda[CTA_TIMEOUT]) {
2800 err = ctnetlink_change_timeout(ct, cda);
2801 if (err < 0)
2802 return err;
2803 }
2804 if (cda[CTA_STATUS]) {
2805 err = ctnetlink_update_status(ct, cda);
2806 if (err < 0)
2807 return err;
2808 }
2809 if (cda[CTA_HELP]) {
2810 err = ctnetlink_change_helper(ct, cda);
2811 if (err < 0)
2812 return err;
2813 }
2814 if (cda[CTA_LABELS]) {
2815 err = ctnetlink_attach_labels(ct, cda);
2816 if (err < 0)
2817 return err;
2818 }
2819 #if defined(CONFIG_NF_CONNTRACK_MARK)
2820 if (cda[CTA_MARK]) {
2821 ctnetlink_change_mark(ct, cda);
2822 }
2823 #endif
2824 return 0;
2825 }
2826
2827 static int
ctnetlink_glue_parse(const struct nlattr * attr,struct nf_conn * ct)2828 ctnetlink_glue_parse(const struct nlattr *attr, struct nf_conn *ct)
2829 {
2830 struct nlattr *cda[CTA_MAX+1];
2831 int ret;
2832
2833 ret = nla_parse_nested_deprecated(cda, CTA_MAX, attr, ct_nla_policy,
2834 NULL);
2835 if (ret < 0)
2836 return ret;
2837
2838 return ctnetlink_glue_parse_ct((const struct nlattr **)cda, ct);
2839 }
2840
ctnetlink_glue_exp_parse(const struct nlattr * const * cda,const struct nf_conn * ct,struct nf_conntrack_tuple * tuple,struct nf_conntrack_tuple * mask)2841 static int ctnetlink_glue_exp_parse(const struct nlattr * const *cda,
2842 const struct nf_conn *ct,
2843 struct nf_conntrack_tuple *tuple,
2844 struct nf_conntrack_tuple *mask)
2845 {
2846 int err;
2847
2848 err = ctnetlink_parse_tuple(cda, tuple, CTA_EXPECT_TUPLE,
2849 nf_ct_l3num(ct), NULL);
2850 if (err < 0)
2851 return err;
2852
2853 return ctnetlink_parse_tuple(cda, mask, CTA_EXPECT_MASK,
2854 nf_ct_l3num(ct), NULL);
2855 }
2856
2857 static int
ctnetlink_glue_attach_expect(const struct nlattr * attr,struct nf_conn * ct,u32 portid,u32 report)2858 ctnetlink_glue_attach_expect(const struct nlattr *attr, struct nf_conn *ct,
2859 u32 portid, u32 report)
2860 {
2861 struct nf_conntrack_helper *assign_helper = NULL;
2862 struct nlattr *cda[CTA_EXPECT_MAX+1];
2863 struct nf_conntrack_tuple tuple, mask;
2864 struct nf_conntrack_expect *exp;
2865 int err;
2866
2867 err = nla_parse_nested_deprecated(cda, CTA_EXPECT_MAX, attr,
2868 exp_nla_policy, NULL);
2869 if (err < 0)
2870 return err;
2871
2872 if (!cda[CTA_EXPECT_TUPLE] || !cda[CTA_EXPECT_MASK])
2873 return -EINVAL;
2874
2875 err = ctnetlink_glue_exp_parse((const struct nlattr * const *)cda,
2876 ct, &tuple, &mask);
2877 if (err < 0)
2878 return err;
2879
2880 if (cda[CTA_EXPECT_HELP_NAME]) {
2881 const char *helpname = nla_data(cda[CTA_EXPECT_HELP_NAME]);
2882
2883 assign_helper = __nf_conntrack_helper_find(helpname,
2884 nf_ct_l3num(ct),
2885 tuple.dst.protonum);
2886 if (!assign_helper)
2887 return -EOPNOTSUPP;
2888 }
2889
2890 exp = ctnetlink_alloc_expect((const struct nlattr * const *)cda, ct,
2891 assign_helper, &tuple, &mask);
2892 if (IS_ERR(exp))
2893 return PTR_ERR(exp);
2894
2895 err = nf_ct_expect_related_report(exp, portid, report, 0);
2896 nf_ct_expect_put(exp);
2897 return err;
2898 }
2899
ctnetlink_glue_seqadj(struct sk_buff * skb,struct nf_conn * ct,enum ip_conntrack_info ctinfo,int diff)2900 static void ctnetlink_glue_seqadj(struct sk_buff *skb, struct nf_conn *ct,
2901 enum ip_conntrack_info ctinfo, int diff)
2902 {
2903 if (!(ct->status & IPS_NAT_MASK))
2904 return;
2905
2906 nf_ct_tcp_seqadj_set(skb, ct, ctinfo, diff);
2907 }
2908
2909 static const struct nfnl_ct_hook ctnetlink_glue_hook = {
2910 .build_size = ctnetlink_glue_build_size,
2911 .build = ctnetlink_glue_build,
2912 .parse = ctnetlink_glue_parse,
2913 .attach_expect = ctnetlink_glue_attach_expect,
2914 .seq_adjust = ctnetlink_glue_seqadj,
2915 };
2916 #endif /* CONFIG_NETFILTER_NETLINK_GLUE_CT */
2917
2918 /***********************************************************************
2919 * EXPECT
2920 ***********************************************************************/
2921
ctnetlink_exp_dump_tuple(struct sk_buff * skb,const struct nf_conntrack_tuple * tuple,u32 type)2922 static int ctnetlink_exp_dump_tuple(struct sk_buff *skb,
2923 const struct nf_conntrack_tuple *tuple,
2924 u32 type)
2925 {
2926 struct nlattr *nest_parms;
2927
2928 nest_parms = nla_nest_start(skb, type);
2929 if (!nest_parms)
2930 goto nla_put_failure;
2931 if (ctnetlink_dump_tuples(skb, tuple) < 0)
2932 goto nla_put_failure;
2933 nla_nest_end(skb, nest_parms);
2934
2935 return 0;
2936
2937 nla_put_failure:
2938 return -1;
2939 }
2940
ctnetlink_exp_dump_mask(struct sk_buff * skb,const struct nf_conntrack_tuple * tuple,const struct nf_conntrack_tuple_mask * mask)2941 static int ctnetlink_exp_dump_mask(struct sk_buff *skb,
2942 const struct nf_conntrack_tuple *tuple,
2943 const struct nf_conntrack_tuple_mask *mask)
2944 {
2945 const struct nf_conntrack_l4proto *l4proto;
2946 struct nf_conntrack_tuple m;
2947 struct nlattr *nest_parms;
2948 int ret;
2949
2950 memset(&m, 0xFF, sizeof(m));
2951 memcpy(&m.src.u3, &mask->src.u3, sizeof(m.src.u3));
2952 m.src.u.all = mask->src.u.all;
2953 m.src.l3num = tuple->src.l3num;
2954 m.dst.protonum = tuple->dst.protonum;
2955
2956 nest_parms = nla_nest_start(skb, CTA_EXPECT_MASK);
2957 if (!nest_parms)
2958 goto nla_put_failure;
2959
2960 rcu_read_lock();
2961 ret = ctnetlink_dump_tuples_ip(skb, &m);
2962 if (ret >= 0) {
2963 l4proto = nf_ct_l4proto_find(tuple->dst.protonum);
2964 ret = ctnetlink_dump_tuples_proto(skb, &m, l4proto);
2965 }
2966 rcu_read_unlock();
2967
2968 if (unlikely(ret < 0))
2969 goto nla_put_failure;
2970
2971 nla_nest_end(skb, nest_parms);
2972
2973 return 0;
2974
2975 nla_put_failure:
2976 return -1;
2977 }
2978
2979 #if IS_ENABLED(CONFIG_NF_NAT)
2980 static const union nf_inet_addr any_addr;
2981 #endif
2982
nf_expect_get_id(const struct nf_conntrack_expect * exp)2983 static __be32 nf_expect_get_id(const struct nf_conntrack_expect *exp)
2984 {
2985 static siphash_aligned_key_t exp_id_seed;
2986 unsigned long a, b, c, d;
2987
2988 net_get_random_once(&exp_id_seed, sizeof(exp_id_seed));
2989
2990 a = (unsigned long)exp;
2991 b = (unsigned long)exp->helper;
2992 c = (unsigned long)exp->master;
2993 d = (unsigned long)siphash(&exp->tuple, sizeof(exp->tuple), &exp_id_seed);
2994
2995 #ifdef CONFIG_64BIT
2996 return (__force __be32)siphash_4u64((u64)a, (u64)b, (u64)c, (u64)d, &exp_id_seed);
2997 #else
2998 return (__force __be32)siphash_4u32((u32)a, (u32)b, (u32)c, (u32)d, &exp_id_seed);
2999 #endif
3000 }
3001
3002 static int
ctnetlink_exp_dump_expect(struct sk_buff * skb,const struct nf_conntrack_expect * exp)3003 ctnetlink_exp_dump_expect(struct sk_buff *skb,
3004 const struct nf_conntrack_expect *exp)
3005 {
3006 __s32 timeout = (__s32)(READ_ONCE(exp->timeout) - nfct_time_stamp) / HZ;
3007 struct nf_conntrack_helper *helper;
3008 #if IS_ENABLED(CONFIG_NF_NAT)
3009 struct nlattr *nest_parms;
3010 struct nf_conntrack_tuple nat_tuple = {};
3011 #endif
3012 struct nf_ct_helper_expectfn *expfn;
3013
3014 if (timeout < 0)
3015 timeout = 0;
3016
3017 if (ctnetlink_exp_dump_tuple(skb, &exp->tuple, CTA_EXPECT_TUPLE) < 0)
3018 goto nla_put_failure;
3019 if (ctnetlink_exp_dump_mask(skb, &exp->tuple, &exp->mask) < 0)
3020 goto nla_put_failure;
3021 if (ctnetlink_exp_dump_tuple(skb, &exp->master_tuple, CTA_EXPECT_MASTER) < 0)
3022 goto nla_put_failure;
3023
3024 #if IS_ENABLED(CONFIG_NF_NAT)
3025 if (!nf_inet_addr_cmp(&exp->saved_addr, &any_addr) ||
3026 exp->saved_proto.all) {
3027 nest_parms = nla_nest_start(skb, CTA_EXPECT_NAT);
3028 if (!nest_parms)
3029 goto nla_put_failure;
3030
3031 if (nla_put_be32(skb, CTA_EXPECT_NAT_DIR, htonl(exp->dir)))
3032 goto nla_put_failure;
3033
3034 nat_tuple.src.l3num = exp->master_tuple.src.l3num;
3035 nat_tuple.src.u3 = exp->saved_addr;
3036 nat_tuple.dst.protonum = exp->master_tuple.dst.protonum;
3037 nat_tuple.src.u = exp->saved_proto;
3038
3039 if (ctnetlink_exp_dump_tuple(skb, &nat_tuple,
3040 CTA_EXPECT_NAT_TUPLE) < 0)
3041 goto nla_put_failure;
3042 nla_nest_end(skb, nest_parms);
3043 }
3044 #endif
3045 if (nla_put_be32(skb, CTA_EXPECT_TIMEOUT, htonl(timeout)) ||
3046 nla_put_be32(skb, CTA_EXPECT_ID, nf_expect_get_id(exp)) ||
3047 nla_put_be32(skb, CTA_EXPECT_FLAGS, htonl(exp->flags & NF_CT_EXPECT_MASK)) ||
3048 nla_put_be32(skb, CTA_EXPECT_CLASS, htonl(exp->class)))
3049 goto nla_put_failure;
3050
3051 helper = rcu_dereference(exp->helper);
3052 if (helper &&
3053 nla_put_string(skb, CTA_EXPECT_HELP_NAME, helper->name))
3054 goto nla_put_failure;
3055
3056 expfn = nf_ct_helper_expectfn_find_by_symbol(exp->expectfn);
3057 if (expfn != NULL &&
3058 nla_put_string(skb, CTA_EXPECT_FN, expfn->name))
3059 goto nla_put_failure;
3060
3061 return 0;
3062
3063 nla_put_failure:
3064 return -1;
3065 }
3066
3067 static int
ctnetlink_exp_fill_info(struct sk_buff * skb,u32 portid,u32 seq,int event,const struct nf_conntrack_expect * exp)3068 ctnetlink_exp_fill_info(struct sk_buff *skb, u32 portid, u32 seq,
3069 int event, const struct nf_conntrack_expect *exp)
3070 {
3071 struct nlmsghdr *nlh;
3072 unsigned int flags = portid ? NLM_F_MULTI : 0;
3073
3074 event = nfnl_msg_type(NFNL_SUBSYS_CTNETLINK_EXP, event);
3075 nlh = nfnl_msg_put(skb, portid, seq, event, flags,
3076 exp->tuple.src.l3num, NFNETLINK_V0, 0);
3077 if (!nlh)
3078 goto nlmsg_failure;
3079
3080 if (ctnetlink_exp_dump_expect(skb, exp) < 0)
3081 goto nla_put_failure;
3082
3083 nlmsg_end(skb, nlh);
3084 return skb->len;
3085
3086 nlmsg_failure:
3087 nla_put_failure:
3088 nlmsg_cancel(skb, nlh);
3089 return -1;
3090 }
3091
3092 #ifdef CONFIG_NF_CONNTRACK_EVENTS
3093 static int
ctnetlink_expect_event(unsigned int events,const struct nf_exp_event * item)3094 ctnetlink_expect_event(unsigned int events, const struct nf_exp_event *item)
3095 {
3096 struct nf_conntrack_expect *exp = item->exp;
3097 struct net *net = nf_ct_exp_net(exp);
3098 struct nlmsghdr *nlh;
3099 struct sk_buff *skb;
3100 unsigned int type, group;
3101 int flags = 0;
3102
3103 if (events & (1 << IPEXP_DESTROY)) {
3104 type = IPCTNL_MSG_EXP_DELETE;
3105 group = NFNLGRP_CONNTRACK_EXP_DESTROY;
3106 } else if (events & (1 << IPEXP_NEW)) {
3107 type = IPCTNL_MSG_EXP_NEW;
3108 flags = NLM_F_CREATE|NLM_F_EXCL;
3109 group = NFNLGRP_CONNTRACK_EXP_NEW;
3110 } else
3111 return 0;
3112
3113 if (!item->report && !nfnetlink_has_listeners(net, group))
3114 return 0;
3115
3116 skb = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_ATOMIC);
3117 if (skb == NULL)
3118 goto errout;
3119
3120 type = nfnl_msg_type(NFNL_SUBSYS_CTNETLINK_EXP, type);
3121 nlh = nfnl_msg_put(skb, item->portid, 0, type, flags,
3122 exp->tuple.src.l3num, NFNETLINK_V0, 0);
3123 if (!nlh)
3124 goto nlmsg_failure;
3125
3126 if (ctnetlink_exp_dump_expect(skb, exp) < 0)
3127 goto nla_put_failure;
3128
3129 nlmsg_end(skb, nlh);
3130 nfnetlink_send(skb, net, item->portid, group, item->report, GFP_ATOMIC);
3131 return 0;
3132
3133 nla_put_failure:
3134 nlmsg_cancel(skb, nlh);
3135 nlmsg_failure:
3136 kfree_skb(skb);
3137 errout:
3138 nfnetlink_set_err(net, 0, 0, -ENOBUFS);
3139 return 0;
3140 }
3141 #endif
3142
ctnetlink_exp_id(const struct nf_conntrack_expect * exp)3143 static unsigned long ctnetlink_exp_id(const struct nf_conntrack_expect *exp)
3144 {
3145 unsigned long id = (unsigned long)exp;
3146
3147 id += nf_ct_get_id(exp->master);
3148 id += exp->class;
3149
3150 return id ? id : 1;
3151 }
3152
3153 static int
ctnetlink_exp_dump_table(struct sk_buff * skb,struct netlink_callback * cb)3154 ctnetlink_exp_dump_table(struct sk_buff *skb, struct netlink_callback *cb)
3155 {
3156 struct net *net = sock_net(skb->sk);
3157 struct nfgenmsg *nfmsg = nlmsg_data(cb->nlh);
3158 u_int8_t l3proto = nfmsg->nfgen_family;
3159 unsigned long last_id = cb->args[1];
3160 struct nf_conntrack_expect *exp;
3161
3162 rcu_read_lock();
3163 for (; cb->args[0] < nf_ct_expect_hsize; cb->args[0]++) {
3164 restart:
3165 hlist_for_each_entry_rcu(exp, &nf_ct_expect_hash[cb->args[0]],
3166 hnode) {
3167 if (nf_ct_exp_is_expired(exp))
3168 continue;
3169
3170 if (l3proto && exp->tuple.src.l3num != l3proto)
3171 continue;
3172
3173 if (!net_eq(nf_ct_exp_net(exp), net))
3174 continue;
3175
3176 if (cb->args[1]) {
3177 if (ctnetlink_exp_id(exp) != last_id)
3178 continue;
3179 cb->args[1] = 0;
3180 }
3181 if (ctnetlink_exp_fill_info(skb,
3182 NETLINK_CB(cb->skb).portid,
3183 cb->nlh->nlmsg_seq,
3184 IPCTNL_MSG_EXP_NEW,
3185 exp) < 0) {
3186 cb->args[1] = ctnetlink_exp_id(exp);
3187 goto out;
3188 }
3189 }
3190 if (cb->args[1]) {
3191 cb->args[1] = 0;
3192 goto restart;
3193 }
3194 }
3195 out:
3196 rcu_read_unlock();
3197 return skb->len;
3198 }
3199
3200 static int
ctnetlink_exp_ct_dump_table(struct sk_buff * skb,struct netlink_callback * cb)3201 ctnetlink_exp_ct_dump_table(struct sk_buff *skb, struct netlink_callback *cb)
3202 {
3203 struct nfgenmsg *nfmsg = nlmsg_data(cb->nlh);
3204 struct nf_conn *ct = cb->data;
3205 struct nf_conn_help *help;
3206 u_int8_t l3proto = nfmsg->nfgen_family;
3207 unsigned long last_id = cb->args[1];
3208 struct nf_conntrack_expect *exp;
3209
3210 if (cb->args[0])
3211 return 0;
3212
3213 help = nfct_help(ct);
3214 if (!help)
3215 return 0;
3216
3217 rcu_read_lock();
3218
3219 restart:
3220 hlist_for_each_entry_rcu(exp, &help->expectations, lnode) {
3221 if (l3proto && exp->tuple.src.l3num != l3proto)
3222 continue;
3223 if (cb->args[1]) {
3224 if (ctnetlink_exp_id(exp) != last_id)
3225 continue;
3226 cb->args[1] = 0;
3227 }
3228 if (ctnetlink_exp_fill_info(skb, NETLINK_CB(cb->skb).portid,
3229 cb->nlh->nlmsg_seq,
3230 IPCTNL_MSG_EXP_NEW,
3231 exp) < 0) {
3232 cb->args[1] = ctnetlink_exp_id(exp);
3233 goto out;
3234 }
3235 }
3236 if (cb->args[1]) {
3237 cb->args[1] = 0;
3238 goto restart;
3239 }
3240 cb->args[0] = 1;
3241 out:
3242 rcu_read_unlock();
3243 return skb->len;
3244 }
3245
ctnetlink_dump_exp_ct_start(struct netlink_callback * cb)3246 static int ctnetlink_dump_exp_ct_start(struct netlink_callback *cb)
3247 {
3248 struct nf_conn *ct = cb->data;
3249
3250 if (!refcount_inc_not_zero(&ct->ct_general.use))
3251 return -ENOENT;
3252 return 0;
3253 }
3254
ctnetlink_dump_exp_ct_done(struct netlink_callback * cb)3255 static int ctnetlink_dump_exp_ct_done(struct netlink_callback *cb)
3256 {
3257 struct nf_conn *ct = cb->data;
3258
3259 if (ct)
3260 nf_ct_put(ct);
3261 return 0;
3262 }
3263
ctnetlink_dump_exp_ct(struct net * net,struct sock * ctnl,struct sk_buff * skb,const struct nlmsghdr * nlh,const struct nlattr * const cda[],struct netlink_ext_ack * extack)3264 static int ctnetlink_dump_exp_ct(struct net *net, struct sock *ctnl,
3265 struct sk_buff *skb,
3266 const struct nlmsghdr *nlh,
3267 const struct nlattr * const cda[],
3268 struct netlink_ext_ack *extack)
3269 {
3270 int err;
3271 struct nfgenmsg *nfmsg = nlmsg_data(nlh);
3272 u_int8_t u3 = nfmsg->nfgen_family;
3273 struct nf_conntrack_tuple tuple;
3274 struct nf_conntrack_tuple_hash *h;
3275 struct nf_conn *ct;
3276 struct nf_conntrack_zone zone;
3277 struct netlink_dump_control c = {
3278 .dump = ctnetlink_exp_ct_dump_table,
3279 .start = ctnetlink_dump_exp_ct_start,
3280 .done = ctnetlink_dump_exp_ct_done,
3281 };
3282
3283 err = ctnetlink_parse_tuple(cda, &tuple, CTA_EXPECT_MASTER,
3284 u3, NULL);
3285 if (err < 0)
3286 return err;
3287
3288 err = ctnetlink_parse_zone(cda[CTA_EXPECT_ZONE], &zone);
3289 if (err < 0)
3290 return err;
3291
3292 h = nf_conntrack_find_get(net, &zone, &tuple);
3293 if (!h)
3294 return -ENOENT;
3295
3296 ct = nf_ct_tuplehash_to_ctrack(h);
3297 /* No expectation linked to this connection tracking. */
3298 if (!nfct_help(ct)) {
3299 nf_ct_put(ct);
3300 return 0;
3301 }
3302
3303 c.data = ct;
3304
3305 err = netlink_dump_start(ctnl, skb, nlh, &c);
3306 nf_ct_put(ct);
3307
3308 return err;
3309 }
3310
ctnetlink_get_expect(struct sk_buff * skb,const struct nfnl_info * info,const struct nlattr * const cda[])3311 static int ctnetlink_get_expect(struct sk_buff *skb,
3312 const struct nfnl_info *info,
3313 const struct nlattr * const cda[])
3314 {
3315 u_int8_t u3 = info->nfmsg->nfgen_family;
3316 struct nf_conntrack_tuple tuple;
3317 struct nf_conntrack_expect *exp;
3318 struct nf_conntrack_zone zone;
3319 struct sk_buff *skb2;
3320 int err;
3321
3322 if (info->nlh->nlmsg_flags & NLM_F_DUMP) {
3323 if (cda[CTA_EXPECT_MASTER])
3324 return ctnetlink_dump_exp_ct(info->net, info->sk, skb,
3325 info->nlh, cda,
3326 info->extack);
3327 else {
3328 struct netlink_dump_control c = {
3329 .dump = ctnetlink_exp_dump_table,
3330 };
3331 return netlink_dump_start(info->sk, skb, info->nlh, &c);
3332 }
3333 }
3334
3335 err = ctnetlink_parse_zone(cda[CTA_EXPECT_ZONE], &zone);
3336 if (err < 0)
3337 return err;
3338
3339 if (cda[CTA_EXPECT_TUPLE])
3340 err = ctnetlink_parse_tuple(cda, &tuple, CTA_EXPECT_TUPLE,
3341 u3, NULL);
3342 else if (cda[CTA_EXPECT_MASTER])
3343 err = ctnetlink_parse_tuple(cda, &tuple, CTA_EXPECT_MASTER,
3344 u3, NULL);
3345 else
3346 return -EINVAL;
3347
3348 if (err < 0)
3349 return err;
3350
3351 skb2 = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
3352 if (!skb2)
3353 return -ENOMEM;
3354
3355 spin_lock_bh(&nf_conntrack_expect_lock);
3356 exp = nf_ct_expect_find_get(info->net, &zone, &tuple);
3357 if (!exp) {
3358 spin_unlock_bh(&nf_conntrack_expect_lock);
3359 kfree_skb(skb2);
3360 return -ENOENT;
3361 }
3362
3363 if (cda[CTA_EXPECT_ID]) {
3364 __be32 id = nla_get_be32(cda[CTA_EXPECT_ID]);
3365
3366 if (id != nf_expect_get_id(exp)) {
3367 nf_ct_expect_put(exp);
3368 spin_unlock_bh(&nf_conntrack_expect_lock);
3369 kfree_skb(skb2);
3370 return -ENOENT;
3371 }
3372 }
3373
3374 rcu_read_lock();
3375 err = ctnetlink_exp_fill_info(skb2, NETLINK_CB(skb).portid,
3376 info->nlh->nlmsg_seq, IPCTNL_MSG_EXP_NEW,
3377 exp);
3378 rcu_read_unlock();
3379 nf_ct_expect_put(exp);
3380 spin_unlock_bh(&nf_conntrack_expect_lock);
3381
3382 if (err <= 0) {
3383 kfree_skb(skb2);
3384 return -ENOMEM;
3385 }
3386
3387 return nfnetlink_unicast(skb2, info->net, NETLINK_CB(skb).portid);
3388 }
3389
expect_iter_name(struct nf_conntrack_expect * exp,void * data)3390 static bool expect_iter_name(struct nf_conntrack_expect *exp, void *data)
3391 {
3392 struct nf_conntrack_helper *helper;
3393 const char *name = data;
3394
3395 helper = rcu_dereference_protected(exp->helper,
3396 lockdep_is_held(&nf_conntrack_expect_lock));
3397 if (!helper)
3398 return false;
3399
3400 return strcmp(helper->name, name) == 0;
3401 }
3402
expect_iter_all(struct nf_conntrack_expect * exp,void * data)3403 static bool expect_iter_all(struct nf_conntrack_expect *exp, void *data)
3404 {
3405 return true;
3406 }
3407
ctnetlink_del_expect(struct sk_buff * skb,const struct nfnl_info * info,const struct nlattr * const cda[])3408 static int ctnetlink_del_expect(struct sk_buff *skb,
3409 const struct nfnl_info *info,
3410 const struct nlattr * const cda[])
3411 {
3412 u_int8_t u3 = info->nfmsg->nfgen_family;
3413 struct nf_conntrack_expect *exp;
3414 struct nf_conntrack_tuple tuple;
3415 struct nf_conntrack_zone zone;
3416 int err;
3417
3418 if (cda[CTA_EXPECT_TUPLE]) {
3419 /* delete a single expect by tuple */
3420 err = ctnetlink_parse_zone(cda[CTA_EXPECT_ZONE], &zone);
3421 if (err < 0)
3422 return err;
3423
3424 err = ctnetlink_parse_tuple(cda, &tuple, CTA_EXPECT_TUPLE,
3425 u3, NULL);
3426 if (err < 0)
3427 return err;
3428
3429 spin_lock_bh(&nf_conntrack_expect_lock);
3430
3431 /* bump usage count to 2 */
3432 exp = nf_ct_expect_find_get(info->net, &zone, &tuple);
3433 if (!exp) {
3434 spin_unlock_bh(&nf_conntrack_expect_lock);
3435 return -ENOENT;
3436 }
3437
3438 if (cda[CTA_EXPECT_ID]) {
3439 __be32 id = nla_get_be32(cda[CTA_EXPECT_ID]);
3440
3441 if (id != nf_expect_get_id(exp)) {
3442 nf_ct_expect_put(exp);
3443 spin_unlock_bh(&nf_conntrack_expect_lock);
3444 return -ENOENT;
3445 }
3446 }
3447
3448 /* after list removal, usage count == 1 */
3449 nf_ct_unlink_expect_report(exp, NETLINK_CB(skb).portid,
3450 nlmsg_report(info->nlh));
3451 spin_unlock_bh(&nf_conntrack_expect_lock);
3452 /* have to put what we 'get' above.
3453 * after this line usage count == 0 */
3454 nf_ct_expect_put(exp);
3455 } else if (cda[CTA_EXPECT_HELP_NAME]) {
3456 char *name = nla_data(cda[CTA_EXPECT_HELP_NAME]);
3457
3458 nf_ct_expect_iterate_net(info->net, expect_iter_name, name,
3459 NETLINK_CB(skb).portid,
3460 nlmsg_report(info->nlh));
3461 } else {
3462 /* This basically means we have to flush everything*/
3463 nf_ct_expect_iterate_net(info->net, expect_iter_all, NULL,
3464 NETLINK_CB(skb).portid,
3465 nlmsg_report(info->nlh));
3466 }
3467
3468 return 0;
3469 }
3470 static int
ctnetlink_change_expect(struct nf_conntrack_expect * x,const struct nlattr * const cda[])3471 ctnetlink_change_expect(struct nf_conntrack_expect *x,
3472 const struct nlattr * const cda[])
3473 {
3474 if (cda[CTA_EXPECT_TIMEOUT])
3475 WRITE_ONCE(x->timeout, nfct_time_stamp +
3476 ntohl(nla_get_be32(cda[CTA_EXPECT_TIMEOUT])) * HZ);
3477
3478 return 0;
3479 }
3480
3481 #if IS_ENABLED(CONFIG_NF_NAT)
3482 static const struct nla_policy exp_nat_nla_policy[CTA_EXPECT_NAT_MAX+1] = {
3483 [CTA_EXPECT_NAT_DIR] = NLA_POLICY_MAX(NLA_BE32, IP_CT_DIR_REPLY),
3484 [CTA_EXPECT_NAT_TUPLE] = { .type = NLA_NESTED },
3485 };
3486 #endif
3487
3488 static int
ctnetlink_parse_expect_nat(const struct nlattr * attr,struct nf_conntrack_expect * exp,u_int8_t u3)3489 ctnetlink_parse_expect_nat(const struct nlattr *attr,
3490 struct nf_conntrack_expect *exp,
3491 u_int8_t u3)
3492 {
3493 #if IS_ENABLED(CONFIG_NF_NAT)
3494 struct nlattr *tb[CTA_EXPECT_NAT_MAX+1];
3495 struct nf_conntrack_tuple nat_tuple = {};
3496 int err;
3497
3498 err = nla_parse_nested_deprecated(tb, CTA_EXPECT_NAT_MAX, attr,
3499 exp_nat_nla_policy, NULL);
3500 if (err < 0)
3501 return err;
3502
3503 if (!tb[CTA_EXPECT_NAT_DIR] || !tb[CTA_EXPECT_NAT_TUPLE])
3504 return -EINVAL;
3505
3506 err = ctnetlink_parse_tuple((const struct nlattr * const *)tb,
3507 &nat_tuple, CTA_EXPECT_NAT_TUPLE,
3508 u3, NULL);
3509 if (err < 0)
3510 return err;
3511
3512 exp->saved_addr = nat_tuple.src.u3;
3513 exp->saved_proto = nat_tuple.src.u;
3514 exp->dir = ntohl(nla_get_be32(tb[CTA_EXPECT_NAT_DIR]));
3515
3516 return 0;
3517 #else
3518 return -EOPNOTSUPP;
3519 #endif
3520 }
3521
3522 static struct nf_conntrack_expect *
ctnetlink_alloc_expect(const struct nlattr * const cda[],struct nf_conn * ct,const struct nf_conntrack_helper * assign_helper,struct nf_conntrack_tuple * tuple,struct nf_conntrack_tuple * mask)3523 ctnetlink_alloc_expect(const struct nlattr * const cda[], struct nf_conn *ct,
3524 const struct nf_conntrack_helper *assign_helper,
3525 struct nf_conntrack_tuple *tuple,
3526 struct nf_conntrack_tuple *mask)
3527 {
3528 struct net *net = read_pnet(&ct->ct_net);
3529 struct nf_conntrack_helper *helper;
3530 struct nf_conntrack_ecache *ecache;
3531 struct nf_conntrack_expect *exp;
3532 struct nf_conn_help *help;
3533 u32 class = 0;
3534 int err;
3535
3536 help = nfct_help(ct);
3537 if (!help)
3538 return ERR_PTR(-EOPNOTSUPP);
3539
3540 helper = rcu_dereference(help->helper);
3541 if (!helper)
3542 return ERR_PTR(-EOPNOTSUPP);
3543
3544 if (cda[CTA_EXPECT_CLASS]) {
3545 class = ntohl(nla_get_be32(cda[CTA_EXPECT_CLASS]));
3546 if (class > helper->expect_class_max)
3547 return ERR_PTR(-EINVAL);
3548 }
3549 exp = nf_ct_expect_alloc(ct);
3550 if (!exp)
3551 return ERR_PTR(-ENOMEM);
3552
3553 if (cda[CTA_EXPECT_FLAGS]) {
3554 exp->flags = ntohl(nla_get_be32(cda[CTA_EXPECT_FLAGS]));
3555 exp->flags &= ~NF_CT_EXPECT_USERSPACE;
3556 }
3557 if (cda[CTA_EXPECT_FN]) {
3558 const char *name = nla_data(cda[CTA_EXPECT_FN]);
3559 struct nf_ct_helper_expectfn *expfn;
3560
3561 expfn = nf_ct_helper_expectfn_find_by_name(name);
3562 if (expfn == NULL) {
3563 err = -EINVAL;
3564 goto err_out;
3565 }
3566 exp->expectfn = expfn->expectfn;
3567 }
3568
3569 exp->class = class;
3570 exp->master = ct;
3571 write_pnet(&exp->net, net);
3572 #ifdef CONFIG_NF_CONNTRACK_ZONES
3573 exp->zone = ct->zone;
3574 #endif
3575 rcu_assign_pointer(exp->helper, helper);
3576 rcu_assign_pointer(exp->assign_helper, assign_helper);
3577 exp->master_tuple = ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
3578 exp->tuple = *tuple;
3579 exp->mask.src.u3 = mask->src.u3;
3580 exp->mask.src.u.all = mask->src.u.all;
3581
3582 ecache = nf_ct_ecache_find(ct);
3583 if (ecache)
3584 exp->event_mask = ecache->expmask;
3585
3586 if (cda[CTA_EXPECT_NAT]) {
3587 err = ctnetlink_parse_expect_nat(cda[CTA_EXPECT_NAT],
3588 exp, nf_ct_l3num(ct));
3589 if (err < 0)
3590 goto err_out;
3591 }
3592 return exp;
3593 err_out:
3594 nf_ct_expect_put(exp);
3595 return ERR_PTR(err);
3596 }
3597
3598 static int
ctnetlink_create_expect(struct net * net,const struct nf_conntrack_zone * zone,const struct nlattr * const cda[],u_int8_t u3,u32 portid,int report)3599 ctnetlink_create_expect(struct net *net,
3600 const struct nf_conntrack_zone *zone,
3601 const struct nlattr * const cda[],
3602 u_int8_t u3, u32 portid, int report)
3603 {
3604 struct nf_conntrack_tuple tuple, mask, master_tuple;
3605 struct nf_conntrack_tuple_hash *h = NULL;
3606 struct nf_conntrack_expect *exp;
3607 struct nf_conn *ct;
3608 int err;
3609
3610 /* caller guarantees that those three CTA_EXPECT_* exist */
3611 err = ctnetlink_parse_tuple(cda, &tuple, CTA_EXPECT_TUPLE,
3612 u3, NULL);
3613 if (err < 0)
3614 return err;
3615 err = ctnetlink_parse_tuple(cda, &mask, CTA_EXPECT_MASK,
3616 u3, NULL);
3617 if (err < 0)
3618 return err;
3619 err = ctnetlink_parse_tuple(cda, &master_tuple, CTA_EXPECT_MASTER,
3620 u3, NULL);
3621 if (err < 0)
3622 return err;
3623
3624 /* Look for master conntrack of this expectation */
3625 h = nf_conntrack_find_get(net, zone, &master_tuple);
3626 if (!h)
3627 return -ENOENT;
3628 ct = nf_ct_tuplehash_to_ctrack(h);
3629
3630 rcu_read_lock();
3631 exp = ctnetlink_alloc_expect(cda, ct, NULL, &tuple, &mask);
3632 if (IS_ERR(exp)) {
3633 err = PTR_ERR(exp);
3634 goto err_rcu;
3635 }
3636
3637 err = nf_ct_expect_related_report(exp, portid, report, 0);
3638 nf_ct_expect_put(exp);
3639 err_rcu:
3640 rcu_read_unlock();
3641 nf_ct_put(ct);
3642
3643 return err;
3644 }
3645
ctnetlink_new_expect(struct sk_buff * skb,const struct nfnl_info * info,const struct nlattr * const cda[])3646 static int ctnetlink_new_expect(struct sk_buff *skb,
3647 const struct nfnl_info *info,
3648 const struct nlattr * const cda[])
3649 {
3650 u_int8_t u3 = info->nfmsg->nfgen_family;
3651 struct nf_conntrack_tuple tuple;
3652 struct nf_conntrack_expect *exp;
3653 struct nf_conntrack_zone zone;
3654 int err;
3655
3656 if (!cda[CTA_EXPECT_TUPLE]
3657 || !cda[CTA_EXPECT_MASK]
3658 || !cda[CTA_EXPECT_MASTER])
3659 return -EINVAL;
3660
3661 err = ctnetlink_parse_zone(cda[CTA_EXPECT_ZONE], &zone);
3662 if (err < 0)
3663 return err;
3664
3665 err = ctnetlink_parse_tuple(cda, &tuple, CTA_EXPECT_TUPLE,
3666 u3, NULL);
3667 if (err < 0)
3668 return err;
3669
3670 spin_lock_bh(&nf_conntrack_expect_lock);
3671 exp = __nf_ct_expect_find(info->net, &zone, &tuple);
3672 if (!exp) {
3673 spin_unlock_bh(&nf_conntrack_expect_lock);
3674 err = -ENOENT;
3675 if (info->nlh->nlmsg_flags & NLM_F_CREATE) {
3676 err = ctnetlink_create_expect(info->net, &zone, cda, u3,
3677 NETLINK_CB(skb).portid,
3678 nlmsg_report(info->nlh));
3679 }
3680 return err;
3681 }
3682
3683 err = -EEXIST;
3684 if (!(info->nlh->nlmsg_flags & NLM_F_EXCL))
3685 err = ctnetlink_change_expect(exp, cda);
3686 spin_unlock_bh(&nf_conntrack_expect_lock);
3687
3688 return err;
3689 }
3690
3691 static int
ctnetlink_exp_stat_fill_info(struct sk_buff * skb,u32 portid,u32 seq,int cpu,const struct ip_conntrack_stat * st)3692 ctnetlink_exp_stat_fill_info(struct sk_buff *skb, u32 portid, u32 seq, int cpu,
3693 const struct ip_conntrack_stat *st)
3694 {
3695 struct nlmsghdr *nlh;
3696 unsigned int flags = portid ? NLM_F_MULTI : 0, event;
3697
3698 event = nfnl_msg_type(NFNL_SUBSYS_CTNETLINK,
3699 IPCTNL_MSG_EXP_GET_STATS_CPU);
3700 nlh = nfnl_msg_put(skb, portid, seq, event, flags, AF_UNSPEC,
3701 NFNETLINK_V0, htons(cpu));
3702 if (!nlh)
3703 goto nlmsg_failure;
3704
3705 if (nla_put_be32(skb, CTA_STATS_EXP_NEW, htonl(st->expect_new)) ||
3706 nla_put_be32(skb, CTA_STATS_EXP_CREATE, htonl(st->expect_create)) ||
3707 nla_put_be32(skb, CTA_STATS_EXP_DELETE, htonl(st->expect_delete)))
3708 goto nla_put_failure;
3709
3710 nlmsg_end(skb, nlh);
3711 return skb->len;
3712
3713 nla_put_failure:
3714 nlmsg_failure:
3715 nlmsg_cancel(skb, nlh);
3716 return -1;
3717 }
3718
3719 static int
ctnetlink_exp_stat_cpu_dump(struct sk_buff * skb,struct netlink_callback * cb)3720 ctnetlink_exp_stat_cpu_dump(struct sk_buff *skb, struct netlink_callback *cb)
3721 {
3722 int cpu;
3723 struct net *net = sock_net(skb->sk);
3724
3725 if (cb->args[0] == nr_cpu_ids)
3726 return 0;
3727
3728 for (cpu = cb->args[0]; cpu < nr_cpu_ids; cpu++) {
3729 const struct ip_conntrack_stat *st;
3730
3731 if (!cpu_possible(cpu))
3732 continue;
3733
3734 st = per_cpu_ptr(net->ct.stat, cpu);
3735 if (ctnetlink_exp_stat_fill_info(skb, NETLINK_CB(cb->skb).portid,
3736 cb->nlh->nlmsg_seq,
3737 cpu, st) < 0)
3738 break;
3739 }
3740 cb->args[0] = cpu;
3741
3742 return skb->len;
3743 }
3744
ctnetlink_stat_exp_cpu(struct sk_buff * skb,const struct nfnl_info * info,const struct nlattr * const cda[])3745 static int ctnetlink_stat_exp_cpu(struct sk_buff *skb,
3746 const struct nfnl_info *info,
3747 const struct nlattr * const cda[])
3748 {
3749 if (info->nlh->nlmsg_flags & NLM_F_DUMP) {
3750 struct netlink_dump_control c = {
3751 .dump = ctnetlink_exp_stat_cpu_dump,
3752 };
3753 return netlink_dump_start(info->sk, skb, info->nlh, &c);
3754 }
3755
3756 return 0;
3757 }
3758
3759 #ifdef CONFIG_NF_CONNTRACK_EVENTS
3760 static struct nf_ct_event_notifier ctnl_notifier = {
3761 .ct_event = ctnetlink_conntrack_event,
3762 .exp_event = ctnetlink_expect_event,
3763 };
3764 #endif
3765
3766 static const struct nfnl_callback ctnl_cb[IPCTNL_MSG_MAX] = {
3767 [IPCTNL_MSG_CT_NEW] = {
3768 .call = ctnetlink_new_conntrack,
3769 .type = NFNL_CB_MUTEX,
3770 .attr_count = CTA_MAX,
3771 .policy = ct_nla_policy
3772 },
3773 [IPCTNL_MSG_CT_GET] = {
3774 .call = ctnetlink_get_conntrack,
3775 .type = NFNL_CB_MUTEX,
3776 .attr_count = CTA_MAX,
3777 .policy = ct_nla_policy
3778 },
3779 [IPCTNL_MSG_CT_DELETE] = {
3780 .call = ctnetlink_del_conntrack,
3781 .type = NFNL_CB_MUTEX,
3782 .attr_count = CTA_MAX,
3783 .policy = ct_nla_policy
3784 },
3785 [IPCTNL_MSG_CT_GET_CTRZERO] = {
3786 .call = ctnetlink_get_conntrack,
3787 .type = NFNL_CB_MUTEX,
3788 .attr_count = CTA_MAX,
3789 .policy = ct_nla_policy
3790 },
3791 [IPCTNL_MSG_CT_GET_STATS_CPU] = {
3792 .call = ctnetlink_stat_ct_cpu,
3793 .type = NFNL_CB_MUTEX,
3794 },
3795 [IPCTNL_MSG_CT_GET_STATS] = {
3796 .call = ctnetlink_stat_ct,
3797 .type = NFNL_CB_MUTEX,
3798 },
3799 [IPCTNL_MSG_CT_GET_DYING] = {
3800 .call = ctnetlink_get_ct_dying,
3801 .type = NFNL_CB_MUTEX,
3802 },
3803 [IPCTNL_MSG_CT_GET_UNCONFIRMED] = {
3804 .call = ctnetlink_get_ct_unconfirmed,
3805 .type = NFNL_CB_MUTEX,
3806 },
3807 };
3808
3809 static const struct nfnl_callback ctnl_exp_cb[IPCTNL_MSG_EXP_MAX] = {
3810 [IPCTNL_MSG_EXP_GET] = {
3811 .call = ctnetlink_get_expect,
3812 .type = NFNL_CB_MUTEX,
3813 .attr_count = CTA_EXPECT_MAX,
3814 .policy = exp_nla_policy
3815 },
3816 [IPCTNL_MSG_EXP_NEW] = {
3817 .call = ctnetlink_new_expect,
3818 .type = NFNL_CB_MUTEX,
3819 .attr_count = CTA_EXPECT_MAX,
3820 .policy = exp_nla_policy
3821 },
3822 [IPCTNL_MSG_EXP_DELETE] = {
3823 .call = ctnetlink_del_expect,
3824 .type = NFNL_CB_MUTEX,
3825 .attr_count = CTA_EXPECT_MAX,
3826 .policy = exp_nla_policy
3827 },
3828 [IPCTNL_MSG_EXP_GET_STATS_CPU] = {
3829 .call = ctnetlink_stat_exp_cpu,
3830 .type = NFNL_CB_MUTEX,
3831 },
3832 };
3833
3834 static const struct nfnetlink_subsystem ctnl_subsys = {
3835 .name = "conntrack",
3836 .subsys_id = NFNL_SUBSYS_CTNETLINK,
3837 .cb_count = IPCTNL_MSG_MAX,
3838 .cb = ctnl_cb,
3839 };
3840
3841 static const struct nfnetlink_subsystem ctnl_exp_subsys = {
3842 .name = "conntrack_expect",
3843 .subsys_id = NFNL_SUBSYS_CTNETLINK_EXP,
3844 .cb_count = IPCTNL_MSG_EXP_MAX,
3845 .cb = ctnl_exp_cb,
3846 };
3847
3848 MODULE_ALIAS("ip_conntrack_netlink");
3849 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_CTNETLINK);
3850 MODULE_ALIAS_NFNL_SUBSYS(NFNL_SUBSYS_CTNETLINK_EXP);
3851
ctnetlink_net_init(struct net * net)3852 static int __net_init ctnetlink_net_init(struct net *net)
3853 {
3854 #ifdef CONFIG_NF_CONNTRACK_EVENTS
3855 nf_conntrack_register_notifier(net, &ctnl_notifier);
3856 #endif
3857 return 0;
3858 }
3859
ctnetlink_net_pre_exit(struct net * net)3860 static void ctnetlink_net_pre_exit(struct net *net)
3861 {
3862 #ifdef CONFIG_NF_CONNTRACK_EVENTS
3863 nf_conntrack_unregister_notifier(net);
3864 #endif
3865 }
3866
3867 static struct pernet_operations ctnetlink_net_ops = {
3868 .init = ctnetlink_net_init,
3869 .pre_exit = ctnetlink_net_pre_exit,
3870 };
3871
ctnetlink_init(void)3872 static int __init ctnetlink_init(void)
3873 {
3874 int ret;
3875
3876 NL_ASSERT_CTX_FITS(struct ctnetlink_list_dump_ctx);
3877
3878 ret = nfnetlink_subsys_register(&ctnl_subsys);
3879 if (ret < 0) {
3880 pr_err("ctnetlink_init: cannot register with nfnetlink.\n");
3881 goto err_out;
3882 }
3883
3884 ret = nfnetlink_subsys_register(&ctnl_exp_subsys);
3885 if (ret < 0) {
3886 pr_err("ctnetlink_init: cannot register exp with nfnetlink.\n");
3887 goto err_unreg_subsys;
3888 }
3889
3890 ret = register_pernet_subsys(&ctnetlink_net_ops);
3891 if (ret < 0) {
3892 pr_err("ctnetlink_init: cannot register pernet operations\n");
3893 goto err_unreg_exp_subsys;
3894 }
3895 #ifdef CONFIG_NETFILTER_NETLINK_GLUE_CT
3896 /* setup interaction between nf_queue and nf_conntrack_netlink. */
3897 RCU_INIT_POINTER(nfnl_ct_hook, &ctnetlink_glue_hook);
3898 #endif
3899 return 0;
3900
3901 err_unreg_exp_subsys:
3902 nfnetlink_subsys_unregister(&ctnl_exp_subsys);
3903 err_unreg_subsys:
3904 nfnetlink_subsys_unregister(&ctnl_subsys);
3905 err_out:
3906 return ret;
3907 }
3908
ctnetlink_exit(void)3909 static void __exit ctnetlink_exit(void)
3910 {
3911 unregister_pernet_subsys(&ctnetlink_net_ops);
3912 nfnetlink_subsys_unregister(&ctnl_exp_subsys);
3913 nfnetlink_subsys_unregister(&ctnl_subsys);
3914 #ifdef CONFIG_NETFILTER_NETLINK_GLUE_CT
3915 RCU_INIT_POINTER(nfnl_ct_hook, NULL);
3916 #endif
3917 synchronize_rcu();
3918 }
3919
3920 module_init(ctnetlink_init);
3921 module_exit(ctnetlink_exit);
3922