1 // SPDX-License-Identifier: GPL-2.0-only
2 /* Connection state tracking for netfilter. This is separated from,
3 but required by, the NAT layer; it can also be used by an iptables
4 extension. */
5
6 /* (C) 1999-2001 Paul `Rusty' Russell
7 * (C) 2002-2006 Netfilter Core Team <coreteam@netfilter.org>
8 * (C) 2003,2004 USAGI/WIDE Project <http://www.linux-ipv6.org>
9 * (C) 2005-2012 Patrick McHardy <kaber@trash.net>
10 */
11
12 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
13
14 #include <linux/types.h>
15 #include <linux/netfilter.h>
16 #include <linux/module.h>
17 #include <linux/sched.h>
18 #include <linux/skbuff.h>
19 #include <linux/proc_fs.h>
20 #include <linux/vmalloc.h>
21 #include <linux/stddef.h>
22 #include <linux/slab.h>
23 #include <linux/random.h>
24 #include <linux/siphash.h>
25 #include <linux/err.h>
26 #include <linux/percpu.h>
27 #include <linux/moduleparam.h>
28 #include <linux/notifier.h>
29 #include <linux/kernel.h>
30 #include <linux/netdevice.h>
31 #include <linux/socket.h>
32 #include <linux/mm.h>
33 #include <linux/nsproxy.h>
34 #include <linux/rculist_nulls.h>
35
36 #include <net/netfilter/nf_conntrack.h>
37 #include <net/netfilter/nf_conntrack_bpf.h>
38 #include <net/netfilter/nf_conntrack_l4proto.h>
39 #include <net/netfilter/nf_conntrack_expect.h>
40 #include <net/netfilter/nf_conntrack_helper.h>
41 #include <net/netfilter/nf_conntrack_core.h>
42 #include <net/netfilter/nf_conntrack_extend.h>
43 #include <net/netfilter/nf_conntrack_acct.h>
44 #include <net/netfilter/nf_conntrack_ecache.h>
45 #include <net/netfilter/nf_conntrack_zones.h>
46 #include <net/netfilter/nf_conntrack_timestamp.h>
47 #include <net/netfilter/nf_conntrack_timeout.h>
48 #include <net/netfilter/nf_conntrack_labels.h>
49 #include <net/netfilter/nf_conntrack_synproxy.h>
50 #include <net/netfilter/nf_nat.h>
51 #include <net/netfilter/nf_nat_helper.h>
52 #include <net/netns/hash.h>
53 #include <net/ip.h>
54
55 #include "nf_internals.h"
56
57 __cacheline_aligned_in_smp spinlock_t nf_conntrack_locks[CONNTRACK_LOCKS];
58 EXPORT_SYMBOL_GPL(nf_conntrack_locks);
59
60 __cacheline_aligned_in_smp DEFINE_SPINLOCK(nf_conntrack_expect_lock);
61 EXPORT_SYMBOL_GPL(nf_conntrack_expect_lock);
62
63 struct hlist_nulls_head *nf_conntrack_hash __read_mostly;
64 EXPORT_SYMBOL_GPL(nf_conntrack_hash);
65
66 struct conntrack_gc_work {
67 struct delayed_work dwork;
68 u32 next_bucket;
69 u32 avg_timeout;
70 u32 count;
71 u32 start_time;
72 bool exiting;
73 bool early_drop;
74 };
75
76 static __read_mostly struct kmem_cache *nf_conntrack_cachep;
77 static DEFINE_SPINLOCK(nf_conntrack_locks_all_lock);
78 static __read_mostly bool nf_conntrack_locks_all;
79
80 /* serialize hash resizes and nf_ct_iterate_cleanup */
81 static DEFINE_MUTEX(nf_conntrack_mutex);
82
83 #define GC_SCAN_INTERVAL_MAX (60ul * HZ)
84 #define GC_SCAN_INTERVAL_MIN (1ul * HZ)
85
86 /* clamp timeouts to this value (TCP unacked) */
87 #define GC_SCAN_INTERVAL_CLAMP (300ul * HZ)
88
89 /* Initial bias pretending we have 100 entries at the upper bound so we don't
90 * wakeup often just because we have three entries with a 1s timeout while still
91 * allowing non-idle machines to wakeup more often when needed.
92 */
93 #define GC_SCAN_INITIAL_COUNT 100
94 #define GC_SCAN_INTERVAL_INIT GC_SCAN_INTERVAL_MAX
95
96 #define GC_SCAN_MAX_DURATION msecs_to_jiffies(10)
97 #define GC_SCAN_EXPIRED_MAX (64000u / HZ)
98
99 #define MIN_CHAINLEN 50u
100 #define MAX_CHAINLEN (80u - MIN_CHAINLEN)
101
102 static struct conntrack_gc_work conntrack_gc_work;
103
nf_conntrack_lock(spinlock_t * lock)104 void nf_conntrack_lock(spinlock_t *lock) __acquires(lock)
105 {
106 /* 1) Acquire the lock */
107 spin_lock(lock);
108
109 /* 2) read nf_conntrack_locks_all, with ACQUIRE semantics
110 * It pairs with the smp_store_release() in nf_conntrack_all_unlock()
111 */
112 if (likely(smp_load_acquire(&nf_conntrack_locks_all) == false))
113 return;
114
115 /* fast path failed, unlock */
116 spin_unlock(lock);
117
118 /* Slow path 1) get global lock */
119 spin_lock(&nf_conntrack_locks_all_lock);
120
121 /* Slow path 2) get the lock we want */
122 spin_lock(lock);
123
124 /* Slow path 3) release the global lock */
125 spin_unlock(&nf_conntrack_locks_all_lock);
126 }
127 EXPORT_SYMBOL_GPL(nf_conntrack_lock);
128
nf_conntrack_double_unlock(unsigned int h1,unsigned int h2)129 static void nf_conntrack_double_unlock(unsigned int h1, unsigned int h2)
130 {
131 h1 %= CONNTRACK_LOCKS;
132 h2 %= CONNTRACK_LOCKS;
133 spin_unlock(&nf_conntrack_locks[h1]);
134 if (h1 != h2)
135 spin_unlock(&nf_conntrack_locks[h2]);
136 }
137
138 /* return true if we need to recompute hashes (in case hash table was resized) */
nf_conntrack_double_lock(struct net * net,unsigned int h1,unsigned int h2,unsigned int sequence)139 static bool nf_conntrack_double_lock(struct net *net, unsigned int h1,
140 unsigned int h2, unsigned int sequence)
141 {
142 h1 %= CONNTRACK_LOCKS;
143 h2 %= CONNTRACK_LOCKS;
144 if (h1 <= h2) {
145 nf_conntrack_lock(&nf_conntrack_locks[h1]);
146 if (h1 != h2)
147 spin_lock_nested(&nf_conntrack_locks[h2],
148 SINGLE_DEPTH_NESTING);
149 } else {
150 nf_conntrack_lock(&nf_conntrack_locks[h2]);
151 spin_lock_nested(&nf_conntrack_locks[h1],
152 SINGLE_DEPTH_NESTING);
153 }
154 if (read_seqcount_retry(&nf_conntrack_generation, sequence)) {
155 nf_conntrack_double_unlock(h1, h2);
156 return true;
157 }
158 return false;
159 }
160
nf_conntrack_all_lock(void)161 static void nf_conntrack_all_lock(void)
162 __acquires(&nf_conntrack_locks_all_lock)
163 {
164 int i;
165
166 spin_lock(&nf_conntrack_locks_all_lock);
167
168 /* For nf_contrack_locks_all, only the latest time when another
169 * CPU will see an update is controlled, by the "release" of the
170 * spin_lock below.
171 * The earliest time is not controlled, an thus KCSAN could detect
172 * a race when nf_conntract_lock() reads the variable.
173 * WRITE_ONCE() is used to ensure the compiler will not
174 * optimize the write.
175 */
176 WRITE_ONCE(nf_conntrack_locks_all, true);
177
178 for (i = 0; i < CONNTRACK_LOCKS; i++) {
179 spin_lock(&nf_conntrack_locks[i]);
180
181 /* This spin_unlock provides the "release" to ensure that
182 * nf_conntrack_locks_all==true is visible to everyone that
183 * acquired spin_lock(&nf_conntrack_locks[]).
184 */
185 spin_unlock(&nf_conntrack_locks[i]);
186 }
187 }
188
nf_conntrack_all_unlock(void)189 static void nf_conntrack_all_unlock(void)
190 __releases(&nf_conntrack_locks_all_lock)
191 {
192 /* All prior stores must be complete before we clear
193 * 'nf_conntrack_locks_all'. Otherwise nf_conntrack_lock()
194 * might observe the false value but not the entire
195 * critical section.
196 * It pairs with the smp_load_acquire() in nf_conntrack_lock()
197 */
198 smp_store_release(&nf_conntrack_locks_all, false);
199 spin_unlock(&nf_conntrack_locks_all_lock);
200 }
201
202 unsigned int nf_conntrack_htable_size __read_mostly;
203 EXPORT_SYMBOL_GPL(nf_conntrack_htable_size);
204
205 unsigned int nf_conntrack_max __read_mostly;
206 EXPORT_SYMBOL_GPL(nf_conntrack_max);
207 seqcount_spinlock_t nf_conntrack_generation __read_mostly;
208 static siphash_aligned_key_t nf_conntrack_hash_rnd;
209
hash_conntrack_raw(const struct nf_conntrack_tuple * tuple,unsigned int zoneid,const struct net * net)210 static u32 hash_conntrack_raw(const struct nf_conntrack_tuple *tuple,
211 unsigned int zoneid,
212 const struct net *net)
213 {
214 siphash_key_t key;
215
216 get_random_once(&nf_conntrack_hash_rnd, sizeof(nf_conntrack_hash_rnd));
217
218 key = nf_conntrack_hash_rnd;
219
220 key.key[0] ^= zoneid;
221 key.key[1] ^= net_hash_mix(net);
222
223 return siphash((void *)tuple,
224 offsetofend(struct nf_conntrack_tuple, dst.__nfct_hash_offsetend),
225 &key);
226 }
227
scale_hash(u32 hash)228 static u32 scale_hash(u32 hash)
229 {
230 return reciprocal_scale(hash, nf_conntrack_htable_size);
231 }
232
__hash_conntrack(const struct net * net,const struct nf_conntrack_tuple * tuple,unsigned int zoneid,unsigned int size)233 static u32 __hash_conntrack(const struct net *net,
234 const struct nf_conntrack_tuple *tuple,
235 unsigned int zoneid,
236 unsigned int size)
237 {
238 return reciprocal_scale(hash_conntrack_raw(tuple, zoneid, net), size);
239 }
240
hash_conntrack(const struct net * net,const struct nf_conntrack_tuple * tuple,unsigned int zoneid)241 static u32 hash_conntrack(const struct net *net,
242 const struct nf_conntrack_tuple *tuple,
243 unsigned int zoneid)
244 {
245 return scale_hash(hash_conntrack_raw(tuple, zoneid, net));
246 }
247
nf_ct_get_tuple_ports(const struct sk_buff * skb,unsigned int dataoff,struct nf_conntrack_tuple * tuple)248 static bool nf_ct_get_tuple_ports(const struct sk_buff *skb,
249 unsigned int dataoff,
250 struct nf_conntrack_tuple *tuple)
251 { struct {
252 __be16 sport;
253 __be16 dport;
254 } _inet_hdr, *inet_hdr;
255
256 /* Actually only need first 4 bytes to get ports. */
257 inet_hdr = skb_header_pointer(skb, dataoff, sizeof(_inet_hdr), &_inet_hdr);
258 if (!inet_hdr)
259 return false;
260
261 tuple->src.u.udp.port = inet_hdr->sport;
262 tuple->dst.u.udp.port = inet_hdr->dport;
263 return true;
264 }
265
266 static bool
nf_ct_get_tuple(const struct sk_buff * skb,unsigned int nhoff,unsigned int dataoff,u_int16_t l3num,u_int8_t protonum,struct net * net,struct nf_conntrack_tuple * tuple)267 nf_ct_get_tuple(const struct sk_buff *skb,
268 unsigned int nhoff,
269 unsigned int dataoff,
270 u_int16_t l3num,
271 u_int8_t protonum,
272 struct net *net,
273 struct nf_conntrack_tuple *tuple)
274 {
275 unsigned int size;
276 const __be32 *ap;
277 __be32 _addrs[8];
278
279 memset(tuple, 0, sizeof(*tuple));
280
281 tuple->src.l3num = l3num;
282 switch (l3num) {
283 case NFPROTO_IPV4:
284 nhoff += offsetof(struct iphdr, saddr);
285 size = 2 * sizeof(__be32);
286 break;
287 case NFPROTO_IPV6:
288 nhoff += offsetof(struct ipv6hdr, saddr);
289 size = sizeof(_addrs);
290 break;
291 default:
292 return true;
293 }
294
295 ap = skb_header_pointer(skb, nhoff, size, _addrs);
296 if (!ap)
297 return false;
298
299 switch (l3num) {
300 case NFPROTO_IPV4:
301 tuple->src.u3.ip = ap[0];
302 tuple->dst.u3.ip = ap[1];
303 break;
304 case NFPROTO_IPV6:
305 memcpy(tuple->src.u3.ip6, ap, sizeof(tuple->src.u3.ip6));
306 memcpy(tuple->dst.u3.ip6, ap + 4, sizeof(tuple->dst.u3.ip6));
307 break;
308 }
309
310 tuple->dst.protonum = protonum;
311 tuple->dst.dir = IP_CT_DIR_ORIGINAL;
312
313 switch (protonum) {
314 #if IS_ENABLED(CONFIG_IPV6)
315 case IPPROTO_ICMPV6:
316 return icmpv6_pkt_to_tuple(skb, dataoff, net, tuple);
317 #endif
318 case IPPROTO_ICMP:
319 return icmp_pkt_to_tuple(skb, dataoff, net, tuple);
320 #ifdef CONFIG_NF_CT_PROTO_GRE
321 case IPPROTO_GRE:
322 return gre_pkt_to_tuple(skb, dataoff, net, tuple);
323 #endif
324 case IPPROTO_TCP:
325 case IPPROTO_UDP:
326 #ifdef CONFIG_NF_CT_PROTO_UDPLITE
327 case IPPROTO_UDPLITE:
328 #endif
329 #ifdef CONFIG_NF_CT_PROTO_SCTP
330 case IPPROTO_SCTP:
331 #endif
332 #ifdef CONFIG_NF_CT_PROTO_DCCP
333 case IPPROTO_DCCP:
334 #endif
335 /* fallthrough */
336 return nf_ct_get_tuple_ports(skb, dataoff, tuple);
337 default:
338 break;
339 }
340
341 return true;
342 }
343
ipv4_get_l4proto(const struct sk_buff * skb,unsigned int nhoff,u_int8_t * protonum)344 static int ipv4_get_l4proto(const struct sk_buff *skb, unsigned int nhoff,
345 u_int8_t *protonum)
346 {
347 int dataoff = -1;
348 const struct iphdr *iph;
349 struct iphdr _iph;
350
351 iph = skb_header_pointer(skb, nhoff, sizeof(_iph), &_iph);
352 if (!iph)
353 return -1;
354
355 /* Conntrack defragments packets, we might still see fragments
356 * inside ICMP packets though.
357 */
358 if (iph->frag_off & htons(IP_OFFSET))
359 return -1;
360
361 dataoff = nhoff + (iph->ihl << 2);
362 *protonum = iph->protocol;
363
364 /* Check bogus IP headers */
365 if (dataoff > skb->len) {
366 pr_debug("bogus IPv4 packet: nhoff %u, ihl %u, skblen %u\n",
367 nhoff, iph->ihl << 2, skb->len);
368 return -1;
369 }
370 return dataoff;
371 }
372
373 #if IS_ENABLED(CONFIG_IPV6)
ipv6_get_l4proto(const struct sk_buff * skb,unsigned int nhoff,u8 * protonum)374 static int ipv6_get_l4proto(const struct sk_buff *skb, unsigned int nhoff,
375 u8 *protonum)
376 {
377 int protoff = -1;
378 unsigned int extoff = nhoff + sizeof(struct ipv6hdr);
379 __be16 frag_off;
380 u8 nexthdr;
381
382 if (skb_copy_bits(skb, nhoff + offsetof(struct ipv6hdr, nexthdr),
383 &nexthdr, sizeof(nexthdr)) != 0) {
384 pr_debug("can't get nexthdr\n");
385 return -1;
386 }
387 protoff = ipv6_skip_exthdr(skb, extoff, &nexthdr, &frag_off);
388 /*
389 * (protoff == skb->len) means the packet has not data, just
390 * IPv6 and possibly extensions headers, but it is tracked anyway
391 */
392 if (protoff < 0 || (frag_off & htons(~0x7)) != 0) {
393 pr_debug("can't find proto in pkt\n");
394 return -1;
395 }
396
397 *protonum = nexthdr;
398 return protoff;
399 }
400 #endif
401
get_l4proto(const struct sk_buff * skb,unsigned int nhoff,u8 pf,u8 * l4num)402 static int get_l4proto(const struct sk_buff *skb,
403 unsigned int nhoff, u8 pf, u8 *l4num)
404 {
405 switch (pf) {
406 case NFPROTO_IPV4:
407 return ipv4_get_l4proto(skb, nhoff, l4num);
408 #if IS_ENABLED(CONFIG_IPV6)
409 case NFPROTO_IPV6:
410 return ipv6_get_l4proto(skb, nhoff, l4num);
411 #endif
412 default:
413 *l4num = 0;
414 break;
415 }
416 return -1;
417 }
418
nf_ct_get_tuplepr(const struct sk_buff * skb,unsigned int nhoff,u_int16_t l3num,struct net * net,struct nf_conntrack_tuple * tuple)419 bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff,
420 u_int16_t l3num,
421 struct net *net, struct nf_conntrack_tuple *tuple)
422 {
423 u8 protonum;
424 int protoff;
425
426 protoff = get_l4proto(skb, nhoff, l3num, &protonum);
427 if (protoff <= 0)
428 return false;
429
430 return nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, net, tuple);
431 }
432 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
433
434 bool
nf_ct_invert_tuple(struct nf_conntrack_tuple * inverse,const struct nf_conntrack_tuple * orig)435 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
436 const struct nf_conntrack_tuple *orig)
437 {
438 memset(inverse, 0, sizeof(*inverse));
439
440 inverse->src.l3num = orig->src.l3num;
441
442 switch (orig->src.l3num) {
443 case NFPROTO_IPV4:
444 inverse->src.u3.ip = orig->dst.u3.ip;
445 inverse->dst.u3.ip = orig->src.u3.ip;
446 break;
447 case NFPROTO_IPV6:
448 inverse->src.u3.in6 = orig->dst.u3.in6;
449 inverse->dst.u3.in6 = orig->src.u3.in6;
450 break;
451 default:
452 break;
453 }
454
455 inverse->dst.dir = !orig->dst.dir;
456
457 inverse->dst.protonum = orig->dst.protonum;
458
459 switch (orig->dst.protonum) {
460 case IPPROTO_ICMP:
461 return nf_conntrack_invert_icmp_tuple(inverse, orig);
462 #if IS_ENABLED(CONFIG_IPV6)
463 case IPPROTO_ICMPV6:
464 return nf_conntrack_invert_icmpv6_tuple(inverse, orig);
465 #endif
466 }
467
468 inverse->src.u.all = orig->dst.u.all;
469 inverse->dst.u.all = orig->src.u.all;
470 return true;
471 }
472 EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
473
474 /* Generate a almost-unique pseudo-id for a given conntrack.
475 *
476 * intentionally doesn't re-use any of the seeds used for hash
477 * table location, we assume id gets exposed to userspace.
478 *
479 * Following nf_conn items do not change throughout lifetime
480 * of the nf_conn:
481 *
482 * 1. nf_conn address
483 * 2. nf_conn->master address (normally NULL)
484 * 3. the associated net namespace
485 * 4. the original direction tuple
486 */
nf_ct_get_id(const struct nf_conn * ct)487 u32 nf_ct_get_id(const struct nf_conn *ct)
488 {
489 static siphash_aligned_key_t ct_id_seed;
490 unsigned long a, b, c, d;
491
492 net_get_random_once(&ct_id_seed, sizeof(ct_id_seed));
493
494 a = (unsigned long)ct;
495 b = (unsigned long)ct->master;
496 c = (unsigned long)nf_ct_net(ct);
497 d = (unsigned long)siphash(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
498 sizeof(ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple),
499 &ct_id_seed);
500 #ifdef CONFIG_64BIT
501 return siphash_4u64((u64)a, (u64)b, (u64)c, (u64)d, &ct_id_seed);
502 #else
503 return siphash_4u32((u32)a, (u32)b, (u32)c, (u32)d, &ct_id_seed);
504 #endif
505 }
506 EXPORT_SYMBOL_GPL(nf_ct_get_id);
507
508 static void
clean_from_lists(struct nf_conn * ct)509 clean_from_lists(struct nf_conn *ct)
510 {
511 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
512 hlist_nulls_del_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode);
513
514 /* Destroy all pending expectations */
515 nf_ct_remove_expectations(ct);
516 }
517
518 #define NFCT_ALIGN(len) (((len) + NFCT_INFOMASK) & ~NFCT_INFOMASK)
519
520 /* Released via nf_ct_destroy() */
nf_ct_tmpl_alloc(struct net * net,const struct nf_conntrack_zone * zone,gfp_t flags)521 struct nf_conn *nf_ct_tmpl_alloc(struct net *net,
522 const struct nf_conntrack_zone *zone,
523 gfp_t flags)
524 {
525 struct nf_conn *tmpl, *p;
526
527 if (ARCH_KMALLOC_MINALIGN <= NFCT_INFOMASK) {
528 tmpl = kzalloc(sizeof(*tmpl) + NFCT_INFOMASK, flags);
529 if (!tmpl)
530 return NULL;
531
532 p = tmpl;
533 tmpl = (struct nf_conn *)NFCT_ALIGN((unsigned long)p);
534 if (tmpl != p) {
535 tmpl = (struct nf_conn *)NFCT_ALIGN((unsigned long)p);
536 tmpl->proto.tmpl_padto = (char *)tmpl - (char *)p;
537 }
538 } else {
539 tmpl = kzalloc(sizeof(*tmpl), flags);
540 if (!tmpl)
541 return NULL;
542 }
543
544 tmpl->status = IPS_TEMPLATE;
545 write_pnet(&tmpl->ct_net, net);
546 nf_ct_zone_add(tmpl, zone);
547 refcount_set(&tmpl->ct_general.use, 1);
548
549 return tmpl;
550 }
551 EXPORT_SYMBOL_GPL(nf_ct_tmpl_alloc);
552
nf_ct_tmpl_free(struct nf_conn * tmpl)553 void nf_ct_tmpl_free(struct nf_conn *tmpl)
554 {
555 kfree(tmpl->ext);
556
557 if (ARCH_KMALLOC_MINALIGN <= NFCT_INFOMASK)
558 kfree((char *)tmpl - tmpl->proto.tmpl_padto);
559 else
560 kfree(tmpl);
561 }
562 EXPORT_SYMBOL_GPL(nf_ct_tmpl_free);
563
destroy_gre_conntrack(struct nf_conn * ct)564 static void destroy_gre_conntrack(struct nf_conn *ct)
565 {
566 #ifdef CONFIG_NF_CT_PROTO_GRE
567 struct nf_conn *master = ct->master;
568
569 if (master)
570 nf_ct_gre_keymap_destroy(master);
571 #endif
572 }
573
nf_ct_destroy(struct nf_conntrack * nfct)574 void nf_ct_destroy(struct nf_conntrack *nfct)
575 {
576 struct nf_conn *ct = (struct nf_conn *)nfct;
577
578 WARN_ON(refcount_read(&nfct->use) != 0);
579
580 if (unlikely(nf_ct_is_template(ct))) {
581 nf_ct_tmpl_free(ct);
582 return;
583 }
584
585 if (unlikely(nf_ct_protonum(ct) == IPPROTO_GRE))
586 destroy_gre_conntrack(ct);
587
588 /* Expectations will have been removed in clean_from_lists,
589 * except TFTP can create an expectation on the first packet,
590 * before connection is in the list, so we need to clean here,
591 * too.
592 */
593 nf_ct_remove_expectations(ct);
594
595 if (ct->master)
596 nf_ct_put(ct->master);
597
598 nf_conntrack_free(ct);
599 }
600 EXPORT_SYMBOL(nf_ct_destroy);
601
__nf_ct_delete_from_lists(struct nf_conn * ct)602 static void __nf_ct_delete_from_lists(struct nf_conn *ct)
603 {
604 struct net *net = nf_ct_net(ct);
605 unsigned int hash, reply_hash;
606 unsigned int sequence;
607
608 do {
609 sequence = read_seqcount_begin(&nf_conntrack_generation);
610 hash = hash_conntrack(net,
611 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
612 nf_ct_zone_id(nf_ct_zone(ct), IP_CT_DIR_ORIGINAL));
613 reply_hash = hash_conntrack(net,
614 &ct->tuplehash[IP_CT_DIR_REPLY].tuple,
615 nf_ct_zone_id(nf_ct_zone(ct), IP_CT_DIR_REPLY));
616 } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
617
618 clean_from_lists(ct);
619 nf_conntrack_double_unlock(hash, reply_hash);
620 }
621
nf_ct_delete_from_lists(struct nf_conn * ct)622 static void nf_ct_delete_from_lists(struct nf_conn *ct)
623 {
624 nf_ct_helper_destroy(ct);
625 local_bh_disable();
626
627 __nf_ct_delete_from_lists(ct);
628
629 local_bh_enable();
630 }
631
nf_ct_add_to_ecache_list(struct nf_conn * ct)632 static void nf_ct_add_to_ecache_list(struct nf_conn *ct)
633 {
634 #ifdef CONFIG_NF_CONNTRACK_EVENTS
635 struct nf_conntrack_net *cnet = nf_ct_pernet(nf_ct_net(ct));
636
637 spin_lock(&cnet->ecache.dying_lock);
638 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
639 &cnet->ecache.dying_list);
640 spin_unlock(&cnet->ecache.dying_lock);
641 #endif
642 }
643
nf_ct_delete(struct nf_conn * ct,u32 portid,int report)644 bool nf_ct_delete(struct nf_conn *ct, u32 portid, int report)
645 {
646 struct nf_conn_tstamp *tstamp;
647 struct net *net;
648
649 if (test_and_set_bit(IPS_DYING_BIT, &ct->status))
650 return false;
651
652 tstamp = nf_conn_tstamp_find(ct);
653 if (tstamp) {
654 s32 timeout = READ_ONCE(ct->timeout) - nfct_time_stamp;
655
656 tstamp->stop = ktime_get_real_ns();
657 if (timeout < 0)
658 tstamp->stop -= jiffies_to_nsecs(-timeout);
659 }
660
661 if (nf_conntrack_event_report(IPCT_DESTROY, ct,
662 portid, report) < 0) {
663 /* destroy event was not delivered. nf_ct_put will
664 * be done by event cache worker on redelivery.
665 */
666 nf_ct_helper_destroy(ct);
667 local_bh_disable();
668 __nf_ct_delete_from_lists(ct);
669 nf_ct_add_to_ecache_list(ct);
670 local_bh_enable();
671
672 nf_conntrack_ecache_work(nf_ct_net(ct), NFCT_ECACHE_DESTROY_FAIL);
673 return false;
674 }
675
676 net = nf_ct_net(ct);
677 if (nf_conntrack_ecache_dwork_pending(net))
678 nf_conntrack_ecache_work(net, NFCT_ECACHE_DESTROY_SENT);
679 nf_ct_delete_from_lists(ct);
680 nf_ct_put(ct);
681 return true;
682 }
683 EXPORT_SYMBOL_GPL(nf_ct_delete);
684
685 static inline bool
nf_ct_key_equal(struct nf_conntrack_tuple_hash * h,const struct nf_conntrack_tuple * tuple,const struct nf_conntrack_zone * zone,const struct net * net)686 nf_ct_key_equal(struct nf_conntrack_tuple_hash *h,
687 const struct nf_conntrack_tuple *tuple,
688 const struct nf_conntrack_zone *zone,
689 const struct net *net)
690 {
691 struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
692
693 /* A conntrack can be recreated with the equal tuple,
694 * so we need to check that the conntrack is confirmed
695 */
696 return nf_ct_tuple_equal(tuple, &h->tuple) &&
697 nf_ct_zone_equal(ct, zone, NF_CT_DIRECTION(h)) &&
698 nf_ct_is_confirmed(ct) &&
699 net_eq(net, nf_ct_net(ct));
700 }
701
702 static inline bool
nf_ct_match(const struct nf_conn * ct1,const struct nf_conn * ct2)703 nf_ct_match(const struct nf_conn *ct1, const struct nf_conn *ct2)
704 {
705 return nf_ct_tuple_equal(&ct1->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
706 &ct2->tuplehash[IP_CT_DIR_ORIGINAL].tuple) &&
707 nf_ct_tuple_equal(&ct1->tuplehash[IP_CT_DIR_REPLY].tuple,
708 &ct2->tuplehash[IP_CT_DIR_REPLY].tuple) &&
709 nf_ct_zone_equal(ct1, nf_ct_zone(ct2), IP_CT_DIR_ORIGINAL) &&
710 nf_ct_zone_equal(ct1, nf_ct_zone(ct2), IP_CT_DIR_REPLY) &&
711 net_eq(nf_ct_net(ct1), nf_ct_net(ct2));
712 }
713
714 /* caller must hold rcu readlock and none of the nf_conntrack_locks */
nf_ct_gc_expired(struct nf_conn * ct)715 static void nf_ct_gc_expired(struct nf_conn *ct)
716 {
717 if (!refcount_inc_not_zero(&ct->ct_general.use))
718 return;
719
720 /* load ->status after refcount increase */
721 smp_acquire__after_ctrl_dep();
722
723 if (nf_ct_should_gc(ct))
724 nf_ct_kill(ct);
725
726 nf_ct_put(ct);
727 }
728
729 /*
730 * Warning :
731 * - Caller must take a reference on returned object
732 * and recheck nf_ct_tuple_equal(tuple, &h->tuple)
733 */
734 static struct nf_conntrack_tuple_hash *
____nf_conntrack_find(struct net * net,const struct nf_conntrack_zone * zone,const struct nf_conntrack_tuple * tuple,u32 hash)735 ____nf_conntrack_find(struct net *net, const struct nf_conntrack_zone *zone,
736 const struct nf_conntrack_tuple *tuple, u32 hash)
737 {
738 struct nf_conntrack_tuple_hash *h;
739 struct hlist_nulls_head *ct_hash;
740 struct hlist_nulls_node *n;
741 unsigned int bucket, hsize;
742
743 begin:
744 nf_conntrack_get_ht(&ct_hash, &hsize);
745 bucket = reciprocal_scale(hash, hsize);
746
747 hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[bucket], hnnode) {
748 struct nf_conn *ct;
749
750 ct = nf_ct_tuplehash_to_ctrack(h);
751 if (nf_ct_is_expired(ct)) {
752 nf_ct_gc_expired(ct);
753 continue;
754 }
755
756 if (nf_ct_key_equal(h, tuple, zone, net))
757 return h;
758 }
759 /*
760 * if the nulls value we got at the end of this lookup is
761 * not the expected one, we must restart lookup.
762 * We probably met an item that was moved to another chain.
763 */
764 if (get_nulls_value(n) != bucket) {
765 NF_CT_STAT_INC_ATOMIC(net, search_restart);
766 goto begin;
767 }
768
769 return NULL;
770 }
771
772 /* Find a connection corresponding to a tuple. */
773 static struct nf_conntrack_tuple_hash *
__nf_conntrack_find_get(struct net * net,const struct nf_conntrack_zone * zone,const struct nf_conntrack_tuple * tuple,u32 hash)774 __nf_conntrack_find_get(struct net *net, const struct nf_conntrack_zone *zone,
775 const struct nf_conntrack_tuple *tuple, u32 hash)
776 {
777 struct nf_conntrack_tuple_hash *h;
778 struct nf_conn *ct;
779
780 h = ____nf_conntrack_find(net, zone, tuple, hash);
781 if (h) {
782 /* We have a candidate that matches the tuple we're interested
783 * in, try to obtain a reference and re-check tuple
784 */
785 ct = nf_ct_tuplehash_to_ctrack(h);
786 if (likely(refcount_inc_not_zero(&ct->ct_general.use))) {
787 /* re-check key after refcount */
788 smp_acquire__after_ctrl_dep();
789
790 if (likely(nf_ct_key_equal(h, tuple, zone, net)))
791 return h;
792
793 /* TYPESAFE_BY_RCU recycled the candidate */
794 nf_ct_put(ct);
795 }
796
797 h = NULL;
798 }
799
800 return h;
801 }
802
803 struct nf_conntrack_tuple_hash *
nf_conntrack_find_get(struct net * net,const struct nf_conntrack_zone * zone,const struct nf_conntrack_tuple * tuple)804 nf_conntrack_find_get(struct net *net, const struct nf_conntrack_zone *zone,
805 const struct nf_conntrack_tuple *tuple)
806 {
807 unsigned int rid, zone_id = nf_ct_zone_id(zone, IP_CT_DIR_ORIGINAL);
808 struct nf_conntrack_tuple_hash *thash;
809
810 rcu_read_lock();
811
812 thash = __nf_conntrack_find_get(net, zone, tuple,
813 hash_conntrack_raw(tuple, zone_id, net));
814
815 if (thash)
816 goto out_unlock;
817
818 rid = nf_ct_zone_id(zone, IP_CT_DIR_REPLY);
819 if (rid != zone_id)
820 thash = __nf_conntrack_find_get(net, zone, tuple,
821 hash_conntrack_raw(tuple, rid, net));
822
823 out_unlock:
824 rcu_read_unlock();
825 return thash;
826 }
827 EXPORT_SYMBOL_GPL(nf_conntrack_find_get);
828
__nf_conntrack_hash_insert(struct nf_conn * ct,unsigned int hash,unsigned int reply_hash)829 static void __nf_conntrack_hash_insert(struct nf_conn *ct,
830 unsigned int hash,
831 unsigned int reply_hash)
832 {
833 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode,
834 &nf_conntrack_hash[hash]);
835 hlist_nulls_add_head_rcu(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode,
836 &nf_conntrack_hash[reply_hash]);
837 }
838
nf_ct_ext_valid_pre(const struct nf_ct_ext * ext)839 static bool nf_ct_ext_valid_pre(const struct nf_ct_ext *ext)
840 {
841 /* if ext->gen_id is not equal to nf_conntrack_ext_genid, some extensions
842 * may contain stale pointers to e.g. helper that has been removed.
843 *
844 * The helper can't clear this because the nf_conn object isn't in
845 * any hash and synchronize_rcu() isn't enough because associated skb
846 * might sit in a queue.
847 */
848 return !ext || ext->gen_id == atomic_read(&nf_conntrack_ext_genid);
849 }
850
nf_ct_ext_valid_post(struct nf_ct_ext * ext)851 static bool nf_ct_ext_valid_post(struct nf_ct_ext *ext)
852 {
853 if (!ext)
854 return true;
855
856 if (ext->gen_id != atomic_read(&nf_conntrack_ext_genid))
857 return false;
858
859 /* inserted into conntrack table, nf_ct_iterate_cleanup()
860 * will find it. Disable nf_ct_ext_find() id check.
861 */
862 WRITE_ONCE(ext->gen_id, 0);
863 return true;
864 }
865
866 int
nf_conntrack_hash_check_insert(struct nf_conn * ct)867 nf_conntrack_hash_check_insert(struct nf_conn *ct)
868 {
869 const struct nf_conntrack_zone *zone;
870 struct net *net = nf_ct_net(ct);
871 unsigned int hash, reply_hash;
872 struct nf_conntrack_tuple_hash *h;
873 struct hlist_nulls_node *n;
874 unsigned int max_chainlen;
875 unsigned int chainlen = 0;
876 unsigned int sequence;
877 int err = -EEXIST;
878
879 zone = nf_ct_zone(ct);
880
881 if (!nf_ct_ext_valid_pre(ct->ext))
882 return -EAGAIN;
883
884 local_bh_disable();
885 do {
886 sequence = read_seqcount_begin(&nf_conntrack_generation);
887 hash = hash_conntrack(net,
888 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
889 nf_ct_zone_id(nf_ct_zone(ct), IP_CT_DIR_ORIGINAL));
890 reply_hash = hash_conntrack(net,
891 &ct->tuplehash[IP_CT_DIR_REPLY].tuple,
892 nf_ct_zone_id(nf_ct_zone(ct), IP_CT_DIR_REPLY));
893 } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
894
895 max_chainlen = MIN_CHAINLEN + get_random_u32_below(MAX_CHAINLEN);
896
897 /* See if there's one in the list already, including reverse */
898 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[hash], hnnode) {
899 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
900 zone, net))
901 goto out;
902
903 if (chainlen++ > max_chainlen)
904 goto chaintoolong;
905 }
906
907 chainlen = 0;
908
909 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[reply_hash], hnnode) {
910 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_REPLY].tuple,
911 zone, net))
912 goto out;
913 if (chainlen++ > max_chainlen)
914 goto chaintoolong;
915 }
916
917 /* If genid has changed, we can't insert anymore because ct
918 * extensions could have stale pointers and nf_ct_iterate_destroy
919 * might have completed its table scan already.
920 *
921 * Increment of the ext genid right after this check is fine:
922 * nf_ct_iterate_destroy blocks until locks are released.
923 */
924 if (!nf_ct_ext_valid_post(ct->ext)) {
925 err = -EAGAIN;
926 goto out;
927 }
928
929 smp_wmb();
930 /* The caller holds a reference to this object */
931 refcount_set(&ct->ct_general.use, 2);
932 __nf_conntrack_hash_insert(ct, hash, reply_hash);
933 nf_conntrack_double_unlock(hash, reply_hash);
934 NF_CT_STAT_INC(net, insert);
935 local_bh_enable();
936
937 return 0;
938 chaintoolong:
939 NF_CT_STAT_INC(net, chaintoolong);
940 err = -ENOSPC;
941 out:
942 nf_conntrack_double_unlock(hash, reply_hash);
943 local_bh_enable();
944 return err;
945 }
946 EXPORT_SYMBOL_GPL(nf_conntrack_hash_check_insert);
947
nf_ct_acct_add(struct nf_conn * ct,u32 dir,unsigned int packets,unsigned int bytes)948 void nf_ct_acct_add(struct nf_conn *ct, u32 dir, unsigned int packets,
949 unsigned int bytes)
950 {
951 struct nf_conn_acct *acct;
952
953 acct = nf_conn_acct_find(ct);
954 if (acct) {
955 struct nf_conn_counter *counter = acct->counter;
956
957 atomic64_add(packets, &counter[dir].packets);
958 atomic64_add(bytes, &counter[dir].bytes);
959 }
960 }
961 EXPORT_SYMBOL_GPL(nf_ct_acct_add);
962
nf_ct_acct_merge(struct nf_conn * ct,enum ip_conntrack_info ctinfo,const struct nf_conn * loser_ct)963 static void nf_ct_acct_merge(struct nf_conn *ct, enum ip_conntrack_info ctinfo,
964 const struct nf_conn *loser_ct)
965 {
966 struct nf_conn_acct *acct;
967
968 acct = nf_conn_acct_find(loser_ct);
969 if (acct) {
970 struct nf_conn_counter *counter = acct->counter;
971 unsigned int bytes;
972
973 /* u32 should be fine since we must have seen one packet. */
974 bytes = atomic64_read(&counter[CTINFO2DIR(ctinfo)].bytes);
975 nf_ct_acct_update(ct, CTINFO2DIR(ctinfo), bytes);
976 }
977 }
978
__nf_conntrack_insert_prepare(struct nf_conn * ct)979 static void __nf_conntrack_insert_prepare(struct nf_conn *ct)
980 {
981 struct nf_conn_tstamp *tstamp;
982
983 refcount_inc(&ct->ct_general.use);
984
985 /* set conntrack timestamp, if enabled. */
986 tstamp = nf_conn_tstamp_find(ct);
987 if (tstamp)
988 tstamp->start = ktime_get_real_ns();
989 }
990
991 /* caller must hold locks to prevent concurrent changes */
__nf_ct_resolve_clash(struct sk_buff * skb,struct nf_conntrack_tuple_hash * h)992 static int __nf_ct_resolve_clash(struct sk_buff *skb,
993 struct nf_conntrack_tuple_hash *h)
994 {
995 /* This is the conntrack entry already in hashes that won race. */
996 struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
997 enum ip_conntrack_info ctinfo;
998 struct nf_conn *loser_ct;
999
1000 loser_ct = nf_ct_get(skb, &ctinfo);
1001
1002 if (nf_ct_is_dying(ct))
1003 return NF_DROP;
1004
1005 if (((ct->status & IPS_NAT_DONE_MASK) == 0) ||
1006 nf_ct_match(ct, loser_ct)) {
1007 struct net *net = nf_ct_net(ct);
1008
1009 nf_conntrack_get(&ct->ct_general);
1010
1011 nf_ct_acct_merge(ct, ctinfo, loser_ct);
1012 nf_ct_put(loser_ct);
1013 nf_ct_set(skb, ct, ctinfo);
1014
1015 NF_CT_STAT_INC(net, clash_resolve);
1016 return NF_ACCEPT;
1017 }
1018
1019 return NF_DROP;
1020 }
1021
1022 /**
1023 * nf_ct_resolve_clash_harder - attempt to insert clashing conntrack entry
1024 *
1025 * @skb: skb that causes the collision
1026 * @repl_idx: hash slot for reply direction
1027 *
1028 * Called when origin or reply direction had a clash.
1029 * The skb can be handled without packet drop provided the reply direction
1030 * is unique or there the existing entry has the identical tuple in both
1031 * directions.
1032 *
1033 * Caller must hold conntrack table locks to prevent concurrent updates.
1034 *
1035 * Returns NF_DROP if the clash could not be handled.
1036 */
nf_ct_resolve_clash_harder(struct sk_buff * skb,u32 repl_idx)1037 static int nf_ct_resolve_clash_harder(struct sk_buff *skb, u32 repl_idx)
1038 {
1039 struct nf_conn *loser_ct = (struct nf_conn *)skb_nfct(skb);
1040 const struct nf_conntrack_zone *zone;
1041 struct nf_conntrack_tuple_hash *h;
1042 struct hlist_nulls_node *n;
1043 struct net *net;
1044
1045 zone = nf_ct_zone(loser_ct);
1046 net = nf_ct_net(loser_ct);
1047
1048 /* Reply direction must never result in a clash, unless both origin
1049 * and reply tuples are identical.
1050 */
1051 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[repl_idx], hnnode) {
1052 if (nf_ct_key_equal(h,
1053 &loser_ct->tuplehash[IP_CT_DIR_REPLY].tuple,
1054 zone, net))
1055 return __nf_ct_resolve_clash(skb, h);
1056 }
1057
1058 /* We want the clashing entry to go away real soon: 1 second timeout. */
1059 WRITE_ONCE(loser_ct->timeout, nfct_time_stamp + HZ);
1060
1061 /* IPS_NAT_CLASH removes the entry automatically on the first
1062 * reply. Also prevents UDP tracker from moving the entry to
1063 * ASSURED state, i.e. the entry can always be evicted under
1064 * pressure.
1065 */
1066 loser_ct->status |= IPS_FIXED_TIMEOUT | IPS_NAT_CLASH;
1067
1068 __nf_conntrack_insert_prepare(loser_ct);
1069
1070 /* fake add for ORIGINAL dir: we want lookups to only find the entry
1071 * already in the table. This also hides the clashing entry from
1072 * ctnetlink iteration, i.e. conntrack -L won't show them.
1073 */
1074 hlist_nulls_add_fake(&loser_ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode);
1075
1076 hlist_nulls_add_head_rcu(&loser_ct->tuplehash[IP_CT_DIR_REPLY].hnnode,
1077 &nf_conntrack_hash[repl_idx]);
1078
1079 NF_CT_STAT_INC(net, clash_resolve);
1080 return NF_ACCEPT;
1081 }
1082
1083 /**
1084 * nf_ct_resolve_clash - attempt to handle clash without packet drop
1085 *
1086 * @skb: skb that causes the clash
1087 * @h: tuplehash of the clashing entry already in table
1088 * @reply_hash: hash slot for reply direction
1089 *
1090 * A conntrack entry can be inserted to the connection tracking table
1091 * if there is no existing entry with an identical tuple.
1092 *
1093 * If there is one, @skb (and the associated, unconfirmed conntrack) has
1094 * to be dropped. In case @skb is retransmitted, next conntrack lookup
1095 * will find the already-existing entry.
1096 *
1097 * The major problem with such packet drop is the extra delay added by
1098 * the packet loss -- it will take some time for a retransmit to occur
1099 * (or the sender to time out when waiting for a reply).
1100 *
1101 * This function attempts to handle the situation without packet drop.
1102 *
1103 * If @skb has no NAT transformation or if the colliding entries are
1104 * exactly the same, only the to-be-confirmed conntrack entry is discarded
1105 * and @skb is associated with the conntrack entry already in the table.
1106 *
1107 * Failing that, the new, unconfirmed conntrack is still added to the table
1108 * provided that the collision only occurs in the ORIGINAL direction.
1109 * The new entry will be added only in the non-clashing REPLY direction,
1110 * so packets in the ORIGINAL direction will continue to match the existing
1111 * entry. The new entry will also have a fixed timeout so it expires --
1112 * due to the collision, it will only see reply traffic.
1113 *
1114 * Returns NF_DROP if the clash could not be resolved.
1115 */
1116 static __cold noinline int
nf_ct_resolve_clash(struct sk_buff * skb,struct nf_conntrack_tuple_hash * h,u32 reply_hash)1117 nf_ct_resolve_clash(struct sk_buff *skb, struct nf_conntrack_tuple_hash *h,
1118 u32 reply_hash)
1119 {
1120 /* This is the conntrack entry already in hashes that won race. */
1121 struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
1122 const struct nf_conntrack_l4proto *l4proto;
1123 enum ip_conntrack_info ctinfo;
1124 struct nf_conn *loser_ct;
1125 struct net *net;
1126 int ret;
1127
1128 loser_ct = nf_ct_get(skb, &ctinfo);
1129 net = nf_ct_net(loser_ct);
1130
1131 l4proto = nf_ct_l4proto_find(nf_ct_protonum(ct));
1132 if (!l4proto->allow_clash)
1133 goto drop;
1134
1135 ret = __nf_ct_resolve_clash(skb, h);
1136 if (ret == NF_ACCEPT)
1137 return ret;
1138
1139 ret = nf_ct_resolve_clash_harder(skb, reply_hash);
1140 if (ret == NF_ACCEPT)
1141 return ret;
1142
1143 drop:
1144 NF_CT_STAT_INC(net, drop);
1145 NF_CT_STAT_INC(net, insert_failed);
1146 return NF_DROP;
1147 }
1148
1149 /* Confirm a connection given skb; places it in hash table */
1150 int
__nf_conntrack_confirm(struct sk_buff * skb)1151 __nf_conntrack_confirm(struct sk_buff *skb)
1152 {
1153 unsigned int chainlen = 0, sequence, max_chainlen;
1154 const struct nf_conntrack_zone *zone;
1155 unsigned int hash, reply_hash;
1156 struct nf_conntrack_tuple_hash *h;
1157 struct nf_conn *ct;
1158 struct nf_conn_help *help;
1159 struct hlist_nulls_node *n;
1160 enum ip_conntrack_info ctinfo;
1161 struct net *net;
1162 int ret = NF_DROP;
1163
1164 ct = nf_ct_get(skb, &ctinfo);
1165 net = nf_ct_net(ct);
1166
1167 /* ipt_REJECT uses nf_conntrack_attach to attach related
1168 ICMP/TCP RST packets in other direction. Actual packet
1169 which created connection will be IP_CT_NEW or for an
1170 expected connection, IP_CT_RELATED. */
1171 if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
1172 return NF_ACCEPT;
1173
1174 zone = nf_ct_zone(ct);
1175 local_bh_disable();
1176
1177 do {
1178 sequence = read_seqcount_begin(&nf_conntrack_generation);
1179 /* reuse the hash saved before */
1180 hash = *(unsigned long *)&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev;
1181 hash = scale_hash(hash);
1182 reply_hash = hash_conntrack(net,
1183 &ct->tuplehash[IP_CT_DIR_REPLY].tuple,
1184 nf_ct_zone_id(nf_ct_zone(ct), IP_CT_DIR_REPLY));
1185 } while (nf_conntrack_double_lock(net, hash, reply_hash, sequence));
1186
1187 /* We're not in hash table, and we refuse to set up related
1188 * connections for unconfirmed conns. But packet copies and
1189 * REJECT will give spurious warnings here.
1190 */
1191
1192 /* Another skb with the same unconfirmed conntrack may
1193 * win the race. This may happen for bridge(br_flood)
1194 * or broadcast/multicast packets do skb_clone with
1195 * unconfirmed conntrack.
1196 */
1197 if (unlikely(nf_ct_is_confirmed(ct))) {
1198 WARN_ON_ONCE(1);
1199 nf_conntrack_double_unlock(hash, reply_hash);
1200 local_bh_enable();
1201 return NF_DROP;
1202 }
1203
1204 if (!nf_ct_ext_valid_pre(ct->ext)) {
1205 NF_CT_STAT_INC(net, insert_failed);
1206 goto dying;
1207 }
1208
1209 /* We have to check the DYING flag after unlink to prevent
1210 * a race against nf_ct_get_next_corpse() possibly called from
1211 * user context, else we insert an already 'dead' hash, blocking
1212 * further use of that particular connection -JM.
1213 */
1214 ct->status |= IPS_CONFIRMED;
1215
1216 if (unlikely(nf_ct_is_dying(ct))) {
1217 NF_CT_STAT_INC(net, insert_failed);
1218 goto dying;
1219 }
1220
1221 max_chainlen = MIN_CHAINLEN + get_random_u32_below(MAX_CHAINLEN);
1222 /* See if there's one in the list already, including reverse:
1223 NAT could have grabbed it without realizing, since we're
1224 not in the hash. If there is, we lost race. */
1225 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[hash], hnnode) {
1226 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
1227 zone, net))
1228 goto out;
1229 if (chainlen++ > max_chainlen)
1230 goto chaintoolong;
1231 }
1232
1233 chainlen = 0;
1234 hlist_nulls_for_each_entry(h, n, &nf_conntrack_hash[reply_hash], hnnode) {
1235 if (nf_ct_key_equal(h, &ct->tuplehash[IP_CT_DIR_REPLY].tuple,
1236 zone, net))
1237 goto out;
1238 if (chainlen++ > max_chainlen) {
1239 chaintoolong:
1240 NF_CT_STAT_INC(net, chaintoolong);
1241 NF_CT_STAT_INC(net, insert_failed);
1242 ret = NF_DROP;
1243 goto dying;
1244 }
1245 }
1246
1247 /* Timer relative to confirmation time, not original
1248 setting time, otherwise we'd get timer wrap in
1249 weird delay cases. */
1250 ct->timeout += nfct_time_stamp;
1251
1252 __nf_conntrack_insert_prepare(ct);
1253
1254 /* Since the lookup is lockless, hash insertion must be done after
1255 * starting the timer and setting the CONFIRMED bit. The RCU barriers
1256 * guarantee that no other CPU can find the conntrack before the above
1257 * stores are visible.
1258 */
1259 __nf_conntrack_hash_insert(ct, hash, reply_hash);
1260 nf_conntrack_double_unlock(hash, reply_hash);
1261 local_bh_enable();
1262
1263 /* ext area is still valid (rcu read lock is held,
1264 * but will go out of scope soon, we need to remove
1265 * this conntrack again.
1266 */
1267 if (!nf_ct_ext_valid_post(ct->ext)) {
1268 nf_ct_kill(ct);
1269 NF_CT_STAT_INC_ATOMIC(net, drop);
1270 return NF_DROP;
1271 }
1272
1273 help = nfct_help(ct);
1274 if (help && help->helper)
1275 nf_conntrack_event_cache(IPCT_HELPER, ct);
1276
1277 nf_conntrack_event_cache(master_ct(ct) ?
1278 IPCT_RELATED : IPCT_NEW, ct);
1279 return NF_ACCEPT;
1280
1281 out:
1282 ret = nf_ct_resolve_clash(skb, h, reply_hash);
1283 dying:
1284 nf_conntrack_double_unlock(hash, reply_hash);
1285 local_bh_enable();
1286 return ret;
1287 }
1288 EXPORT_SYMBOL_GPL(__nf_conntrack_confirm);
1289
1290 /* Returns true if a connection corresponds to the tuple (required
1291 for NAT). */
1292 int
nf_conntrack_tuple_taken(const struct nf_conntrack_tuple * tuple,const struct nf_conn * ignored_conntrack)1293 nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
1294 const struct nf_conn *ignored_conntrack)
1295 {
1296 struct net *net = nf_ct_net(ignored_conntrack);
1297 const struct nf_conntrack_zone *zone;
1298 struct nf_conntrack_tuple_hash *h;
1299 struct hlist_nulls_head *ct_hash;
1300 unsigned int hash, hsize;
1301 struct hlist_nulls_node *n;
1302 struct nf_conn *ct;
1303
1304 zone = nf_ct_zone(ignored_conntrack);
1305
1306 rcu_read_lock();
1307 begin:
1308 nf_conntrack_get_ht(&ct_hash, &hsize);
1309 hash = __hash_conntrack(net, tuple, nf_ct_zone_id(zone, IP_CT_DIR_REPLY), hsize);
1310
1311 hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[hash], hnnode) {
1312 ct = nf_ct_tuplehash_to_ctrack(h);
1313
1314 if (ct == ignored_conntrack)
1315 continue;
1316
1317 if (nf_ct_is_expired(ct)) {
1318 nf_ct_gc_expired(ct);
1319 continue;
1320 }
1321
1322 if (nf_ct_key_equal(h, tuple, zone, net)) {
1323 /* Tuple is taken already, so caller will need to find
1324 * a new source port to use.
1325 *
1326 * Only exception:
1327 * If the *original tuples* are identical, then both
1328 * conntracks refer to the same flow.
1329 * This is a rare situation, it can occur e.g. when
1330 * more than one UDP packet is sent from same socket
1331 * in different threads.
1332 *
1333 * Let nf_ct_resolve_clash() deal with this later.
1334 */
1335 if (nf_ct_tuple_equal(&ignored_conntrack->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
1336 &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple) &&
1337 nf_ct_zone_equal(ct, zone, IP_CT_DIR_ORIGINAL))
1338 continue;
1339
1340 NF_CT_STAT_INC_ATOMIC(net, found);
1341 rcu_read_unlock();
1342 return 1;
1343 }
1344 }
1345
1346 if (get_nulls_value(n) != hash) {
1347 NF_CT_STAT_INC_ATOMIC(net, search_restart);
1348 goto begin;
1349 }
1350
1351 rcu_read_unlock();
1352
1353 return 0;
1354 }
1355 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
1356
1357 #define NF_CT_EVICTION_RANGE 8
1358
1359 /* There's a small race here where we may free a just-assured
1360 connection. Too bad: we're in trouble anyway. */
early_drop_list(struct net * net,struct hlist_nulls_head * head)1361 static unsigned int early_drop_list(struct net *net,
1362 struct hlist_nulls_head *head)
1363 {
1364 struct nf_conntrack_tuple_hash *h;
1365 struct hlist_nulls_node *n;
1366 unsigned int drops = 0;
1367 struct nf_conn *tmp;
1368
1369 hlist_nulls_for_each_entry_rcu(h, n, head, hnnode) {
1370 tmp = nf_ct_tuplehash_to_ctrack(h);
1371
1372 if (nf_ct_is_expired(tmp)) {
1373 nf_ct_gc_expired(tmp);
1374 continue;
1375 }
1376
1377 if (test_bit(IPS_ASSURED_BIT, &tmp->status) ||
1378 !net_eq(nf_ct_net(tmp), net) ||
1379 nf_ct_is_dying(tmp))
1380 continue;
1381
1382 if (!refcount_inc_not_zero(&tmp->ct_general.use))
1383 continue;
1384
1385 /* load ->ct_net and ->status after refcount increase */
1386 smp_acquire__after_ctrl_dep();
1387
1388 /* kill only if still in same netns -- might have moved due to
1389 * SLAB_TYPESAFE_BY_RCU rules.
1390 *
1391 * We steal the timer reference. If that fails timer has
1392 * already fired or someone else deleted it. Just drop ref
1393 * and move to next entry.
1394 */
1395 if (net_eq(nf_ct_net(tmp), net) &&
1396 nf_ct_is_confirmed(tmp) &&
1397 nf_ct_delete(tmp, 0, 0))
1398 drops++;
1399
1400 nf_ct_put(tmp);
1401 }
1402
1403 return drops;
1404 }
1405
early_drop(struct net * net,unsigned int hash)1406 static noinline int early_drop(struct net *net, unsigned int hash)
1407 {
1408 unsigned int i, bucket;
1409
1410 for (i = 0; i < NF_CT_EVICTION_RANGE; i++) {
1411 struct hlist_nulls_head *ct_hash;
1412 unsigned int hsize, drops;
1413
1414 rcu_read_lock();
1415 nf_conntrack_get_ht(&ct_hash, &hsize);
1416 if (!i)
1417 bucket = reciprocal_scale(hash, hsize);
1418 else
1419 bucket = (bucket + 1) % hsize;
1420
1421 drops = early_drop_list(net, &ct_hash[bucket]);
1422 rcu_read_unlock();
1423
1424 if (drops) {
1425 NF_CT_STAT_ADD_ATOMIC(net, early_drop, drops);
1426 return true;
1427 }
1428 }
1429
1430 return false;
1431 }
1432
gc_worker_skip_ct(const struct nf_conn * ct)1433 static bool gc_worker_skip_ct(const struct nf_conn *ct)
1434 {
1435 return !nf_ct_is_confirmed(ct) || nf_ct_is_dying(ct);
1436 }
1437
gc_worker_can_early_drop(const struct nf_conn * ct)1438 static bool gc_worker_can_early_drop(const struct nf_conn *ct)
1439 {
1440 const struct nf_conntrack_l4proto *l4proto;
1441 u8 protonum = nf_ct_protonum(ct);
1442
1443 if (!test_bit(IPS_ASSURED_BIT, &ct->status))
1444 return true;
1445
1446 l4proto = nf_ct_l4proto_find(protonum);
1447 if (l4proto->can_early_drop && l4proto->can_early_drop(ct))
1448 return true;
1449
1450 return false;
1451 }
1452
gc_worker(struct work_struct * work)1453 static void gc_worker(struct work_struct *work)
1454 {
1455 unsigned int i, hashsz, nf_conntrack_max95 = 0;
1456 u32 end_time, start_time = nfct_time_stamp;
1457 struct conntrack_gc_work *gc_work;
1458 unsigned int expired_count = 0;
1459 unsigned long next_run;
1460 s32 delta_time;
1461 long count;
1462
1463 gc_work = container_of(work, struct conntrack_gc_work, dwork.work);
1464
1465 i = gc_work->next_bucket;
1466 if (gc_work->early_drop)
1467 nf_conntrack_max95 = nf_conntrack_max / 100u * 95u;
1468
1469 if (i == 0) {
1470 gc_work->avg_timeout = GC_SCAN_INTERVAL_INIT;
1471 gc_work->count = GC_SCAN_INITIAL_COUNT;
1472 gc_work->start_time = start_time;
1473 }
1474
1475 next_run = gc_work->avg_timeout;
1476 count = gc_work->count;
1477
1478 end_time = start_time + GC_SCAN_MAX_DURATION;
1479
1480 do {
1481 struct nf_conntrack_tuple_hash *h;
1482 struct hlist_nulls_head *ct_hash;
1483 struct hlist_nulls_node *n;
1484 struct nf_conn *tmp;
1485
1486 rcu_read_lock();
1487
1488 nf_conntrack_get_ht(&ct_hash, &hashsz);
1489 if (i >= hashsz) {
1490 rcu_read_unlock();
1491 break;
1492 }
1493
1494 hlist_nulls_for_each_entry_rcu(h, n, &ct_hash[i], hnnode) {
1495 struct nf_conntrack_net *cnet;
1496 struct net *net;
1497 long expires;
1498
1499 tmp = nf_ct_tuplehash_to_ctrack(h);
1500
1501 if (test_bit(IPS_OFFLOAD_BIT, &tmp->status)) {
1502 nf_ct_offload_timeout(tmp);
1503 if (!nf_conntrack_max95)
1504 continue;
1505 }
1506
1507 if (expired_count > GC_SCAN_EXPIRED_MAX) {
1508 rcu_read_unlock();
1509
1510 gc_work->next_bucket = i;
1511 gc_work->avg_timeout = next_run;
1512 gc_work->count = count;
1513
1514 delta_time = nfct_time_stamp - gc_work->start_time;
1515
1516 /* re-sched immediately if total cycle time is exceeded */
1517 next_run = delta_time < (s32)GC_SCAN_INTERVAL_MAX;
1518 goto early_exit;
1519 }
1520
1521 if (nf_ct_is_expired(tmp)) {
1522 nf_ct_gc_expired(tmp);
1523 expired_count++;
1524 continue;
1525 }
1526
1527 expires = clamp(nf_ct_expires(tmp), GC_SCAN_INTERVAL_MIN, GC_SCAN_INTERVAL_CLAMP);
1528 expires = (expires - (long)next_run) / ++count;
1529 next_run += expires;
1530
1531 if (nf_conntrack_max95 == 0 || gc_worker_skip_ct(tmp))
1532 continue;
1533
1534 net = nf_ct_net(tmp);
1535 cnet = nf_ct_pernet(net);
1536 if (atomic_read(&cnet->count) < nf_conntrack_max95)
1537 continue;
1538
1539 /* need to take reference to avoid possible races */
1540 if (!refcount_inc_not_zero(&tmp->ct_general.use))
1541 continue;
1542
1543 /* load ->status after refcount increase */
1544 smp_acquire__after_ctrl_dep();
1545
1546 if (gc_worker_skip_ct(tmp)) {
1547 nf_ct_put(tmp);
1548 continue;
1549 }
1550
1551 if (gc_worker_can_early_drop(tmp)) {
1552 nf_ct_kill(tmp);
1553 expired_count++;
1554 }
1555
1556 nf_ct_put(tmp);
1557 }
1558
1559 /* could check get_nulls_value() here and restart if ct
1560 * was moved to another chain. But given gc is best-effort
1561 * we will just continue with next hash slot.
1562 */
1563 rcu_read_unlock();
1564 cond_resched();
1565 i++;
1566
1567 delta_time = nfct_time_stamp - end_time;
1568 if (delta_time > 0 && i < hashsz) {
1569 gc_work->avg_timeout = next_run;
1570 gc_work->count = count;
1571 gc_work->next_bucket = i;
1572 next_run = 0;
1573 goto early_exit;
1574 }
1575 } while (i < hashsz);
1576
1577 gc_work->next_bucket = 0;
1578
1579 next_run = clamp(next_run, GC_SCAN_INTERVAL_MIN, GC_SCAN_INTERVAL_MAX);
1580
1581 delta_time = max_t(s32, nfct_time_stamp - gc_work->start_time, 1);
1582 if (next_run > (unsigned long)delta_time)
1583 next_run -= delta_time;
1584 else
1585 next_run = 1;
1586
1587 early_exit:
1588 if (gc_work->exiting)
1589 return;
1590
1591 if (next_run)
1592 gc_work->early_drop = false;
1593
1594 queue_delayed_work(system_power_efficient_wq, &gc_work->dwork, next_run);
1595 }
1596
conntrack_gc_work_init(struct conntrack_gc_work * gc_work)1597 static void conntrack_gc_work_init(struct conntrack_gc_work *gc_work)
1598 {
1599 INIT_DELAYED_WORK(&gc_work->dwork, gc_worker);
1600 gc_work->exiting = false;
1601 }
1602
1603 static struct nf_conn *
__nf_conntrack_alloc(struct net * net,const struct nf_conntrack_zone * zone,const struct nf_conntrack_tuple * orig,const struct nf_conntrack_tuple * repl,gfp_t gfp,u32 hash)1604 __nf_conntrack_alloc(struct net *net,
1605 const struct nf_conntrack_zone *zone,
1606 const struct nf_conntrack_tuple *orig,
1607 const struct nf_conntrack_tuple *repl,
1608 gfp_t gfp, u32 hash)
1609 {
1610 struct nf_conntrack_net *cnet = nf_ct_pernet(net);
1611 unsigned int ct_count;
1612 struct nf_conn *ct;
1613
1614 /* We don't want any race condition at early drop stage */
1615 ct_count = atomic_inc_return(&cnet->count);
1616
1617 if (nf_conntrack_max && unlikely(ct_count > nf_conntrack_max)) {
1618 if (!early_drop(net, hash)) {
1619 if (!conntrack_gc_work.early_drop)
1620 conntrack_gc_work.early_drop = true;
1621 atomic_dec(&cnet->count);
1622 net_warn_ratelimited("nf_conntrack: table full, dropping packet\n");
1623 return ERR_PTR(-ENOMEM);
1624 }
1625 }
1626
1627 /*
1628 * Do not use kmem_cache_zalloc(), as this cache uses
1629 * SLAB_TYPESAFE_BY_RCU.
1630 */
1631 ct = kmem_cache_alloc(nf_conntrack_cachep, gfp);
1632 if (ct == NULL)
1633 goto out;
1634
1635 spin_lock_init(&ct->lock);
1636 ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
1637 ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL;
1638 ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
1639 /* save hash for reusing when confirming */
1640 *(unsigned long *)(&ct->tuplehash[IP_CT_DIR_REPLY].hnnode.pprev) = hash;
1641 ct->status = 0;
1642 WRITE_ONCE(ct->timeout, 0);
1643 write_pnet(&ct->ct_net, net);
1644 memset_after(ct, 0, __nfct_init_offset);
1645
1646 nf_ct_zone_add(ct, zone);
1647
1648 /* Because we use RCU lookups, we set ct_general.use to zero before
1649 * this is inserted in any list.
1650 */
1651 refcount_set(&ct->ct_general.use, 0);
1652 return ct;
1653 out:
1654 atomic_dec(&cnet->count);
1655 return ERR_PTR(-ENOMEM);
1656 }
1657
nf_conntrack_alloc(struct net * net,const struct nf_conntrack_zone * zone,const struct nf_conntrack_tuple * orig,const struct nf_conntrack_tuple * repl,gfp_t gfp)1658 struct nf_conn *nf_conntrack_alloc(struct net *net,
1659 const struct nf_conntrack_zone *zone,
1660 const struct nf_conntrack_tuple *orig,
1661 const struct nf_conntrack_tuple *repl,
1662 gfp_t gfp)
1663 {
1664 return __nf_conntrack_alloc(net, zone, orig, repl, gfp, 0);
1665 }
1666 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
1667
nf_conntrack_free(struct nf_conn * ct)1668 void nf_conntrack_free(struct nf_conn *ct)
1669 {
1670 struct net *net = nf_ct_net(ct);
1671 struct nf_conntrack_net *cnet;
1672
1673 /* A freed object has refcnt == 0, that's
1674 * the golden rule for SLAB_TYPESAFE_BY_RCU
1675 */
1676 WARN_ON(refcount_read(&ct->ct_general.use) != 0);
1677
1678 if (ct->status & IPS_SRC_NAT_DONE) {
1679 const struct nf_nat_hook *nat_hook;
1680
1681 rcu_read_lock();
1682 nat_hook = rcu_dereference(nf_nat_hook);
1683 if (nat_hook)
1684 nat_hook->remove_nat_bysrc(ct);
1685 rcu_read_unlock();
1686 }
1687
1688 kfree(ct->ext);
1689 kmem_cache_free(nf_conntrack_cachep, ct);
1690 cnet = nf_ct_pernet(net);
1691
1692 smp_mb__before_atomic();
1693 atomic_dec(&cnet->count);
1694 }
1695 EXPORT_SYMBOL_GPL(nf_conntrack_free);
1696
1697
1698 /* Allocate a new conntrack: we return -ENOMEM if classification
1699 failed due to stress. Otherwise it really is unclassifiable. */
1700 static noinline struct nf_conntrack_tuple_hash *
init_conntrack(struct net * net,struct nf_conn * tmpl,const struct nf_conntrack_tuple * tuple,struct sk_buff * skb,unsigned int dataoff,u32 hash)1701 init_conntrack(struct net *net, struct nf_conn *tmpl,
1702 const struct nf_conntrack_tuple *tuple,
1703 struct sk_buff *skb,
1704 unsigned int dataoff, u32 hash)
1705 {
1706 struct nf_conn *ct;
1707 struct nf_conn_help *help;
1708 struct nf_conntrack_tuple repl_tuple;
1709 #ifdef CONFIG_NF_CONNTRACK_EVENTS
1710 struct nf_conntrack_ecache *ecache;
1711 #endif
1712 struct nf_conntrack_expect *exp = NULL;
1713 const struct nf_conntrack_zone *zone;
1714 struct nf_conn_timeout *timeout_ext;
1715 struct nf_conntrack_zone tmp;
1716 struct nf_conntrack_net *cnet;
1717
1718 if (!nf_ct_invert_tuple(&repl_tuple, tuple))
1719 return NULL;
1720
1721 zone = nf_ct_zone_tmpl(tmpl, skb, &tmp);
1722 ct = __nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC,
1723 hash);
1724 if (IS_ERR(ct))
1725 return (struct nf_conntrack_tuple_hash *)ct;
1726
1727 if (!nf_ct_add_synproxy(ct, tmpl)) {
1728 nf_conntrack_free(ct);
1729 return ERR_PTR(-ENOMEM);
1730 }
1731
1732 timeout_ext = tmpl ? nf_ct_timeout_find(tmpl) : NULL;
1733
1734 if (timeout_ext)
1735 nf_ct_timeout_ext_add(ct, rcu_dereference(timeout_ext->timeout),
1736 GFP_ATOMIC);
1737
1738 nf_ct_acct_ext_add(ct, GFP_ATOMIC);
1739 nf_ct_tstamp_ext_add(ct, GFP_ATOMIC);
1740 nf_ct_labels_ext_add(ct);
1741
1742 #ifdef CONFIG_NF_CONNTRACK_EVENTS
1743 ecache = tmpl ? nf_ct_ecache_find(tmpl) : NULL;
1744
1745 if ((ecache || net->ct.sysctl_events) &&
1746 !nf_ct_ecache_ext_add(ct, ecache ? ecache->ctmask : 0,
1747 ecache ? ecache->expmask : 0,
1748 GFP_ATOMIC)) {
1749 nf_conntrack_free(ct);
1750 return ERR_PTR(-ENOMEM);
1751 }
1752 #endif
1753
1754 cnet = nf_ct_pernet(net);
1755 if (cnet->expect_count) {
1756 spin_lock_bh(&nf_conntrack_expect_lock);
1757 exp = nf_ct_find_expectation(net, zone, tuple, !tmpl || nf_ct_is_confirmed(tmpl));
1758 if (exp) {
1759 /* Welcome, Mr. Bond. We've been expecting you... */
1760 __set_bit(IPS_EXPECTED_BIT, &ct->status);
1761 /* exp->master safe, refcnt bumped in nf_ct_find_expectation */
1762 ct->master = exp->master;
1763 if (exp->helper) {
1764 help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
1765 if (help)
1766 rcu_assign_pointer(help->helper, exp->helper);
1767 }
1768
1769 #ifdef CONFIG_NF_CONNTRACK_MARK
1770 ct->mark = READ_ONCE(exp->master->mark);
1771 #endif
1772 #ifdef CONFIG_NF_CONNTRACK_SECMARK
1773 ct->secmark = exp->master->secmark;
1774 #endif
1775 NF_CT_STAT_INC(net, expect_new);
1776 }
1777 spin_unlock_bh(&nf_conntrack_expect_lock);
1778 }
1779 if (!exp && tmpl)
1780 __nf_ct_try_assign_helper(ct, tmpl, GFP_ATOMIC);
1781
1782 /* Other CPU might have obtained a pointer to this object before it was
1783 * released. Because refcount is 0, refcount_inc_not_zero() will fail.
1784 *
1785 * After refcount_set(1) it will succeed; ensure that zeroing of
1786 * ct->status and the correct ct->net pointer are visible; else other
1787 * core might observe CONFIRMED bit which means the entry is valid and
1788 * in the hash table, but its not (anymore).
1789 */
1790 smp_wmb();
1791
1792 /* Now it is going to be associated with an sk_buff, set refcount to 1. */
1793 refcount_set(&ct->ct_general.use, 1);
1794
1795 if (exp) {
1796 if (exp->expectfn)
1797 exp->expectfn(ct, exp);
1798 nf_ct_expect_put(exp);
1799 }
1800
1801 return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
1802 }
1803
1804 /* On success, returns 0, sets skb->_nfct | ctinfo */
1805 static int
resolve_normal_ct(struct nf_conn * tmpl,struct sk_buff * skb,unsigned int dataoff,u_int8_t protonum,const struct nf_hook_state * state)1806 resolve_normal_ct(struct nf_conn *tmpl,
1807 struct sk_buff *skb,
1808 unsigned int dataoff,
1809 u_int8_t protonum,
1810 const struct nf_hook_state *state)
1811 {
1812 const struct nf_conntrack_zone *zone;
1813 struct nf_conntrack_tuple tuple;
1814 struct nf_conntrack_tuple_hash *h;
1815 enum ip_conntrack_info ctinfo;
1816 struct nf_conntrack_zone tmp;
1817 u32 hash, zone_id, rid;
1818 struct nf_conn *ct;
1819
1820 if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
1821 dataoff, state->pf, protonum, state->net,
1822 &tuple))
1823 return 0;
1824
1825 /* look for tuple match */
1826 zone = nf_ct_zone_tmpl(tmpl, skb, &tmp);
1827
1828 zone_id = nf_ct_zone_id(zone, IP_CT_DIR_ORIGINAL);
1829 hash = hash_conntrack_raw(&tuple, zone_id, state->net);
1830 h = __nf_conntrack_find_get(state->net, zone, &tuple, hash);
1831
1832 if (!h) {
1833 rid = nf_ct_zone_id(zone, IP_CT_DIR_REPLY);
1834 if (zone_id != rid) {
1835 u32 tmp = hash_conntrack_raw(&tuple, rid, state->net);
1836
1837 h = __nf_conntrack_find_get(state->net, zone, &tuple, tmp);
1838 }
1839 }
1840
1841 if (!h) {
1842 h = init_conntrack(state->net, tmpl, &tuple,
1843 skb, dataoff, hash);
1844 if (!h)
1845 return 0;
1846 if (IS_ERR(h))
1847 return PTR_ERR(h);
1848 }
1849 ct = nf_ct_tuplehash_to_ctrack(h);
1850
1851 /* It exists; we have (non-exclusive) reference. */
1852 if (NF_CT_DIRECTION(h) == IP_CT_DIR_REPLY) {
1853 ctinfo = IP_CT_ESTABLISHED_REPLY;
1854 } else {
1855 unsigned long status = READ_ONCE(ct->status);
1856
1857 /* Once we've had two way comms, always ESTABLISHED. */
1858 if (likely(status & IPS_SEEN_REPLY))
1859 ctinfo = IP_CT_ESTABLISHED;
1860 else if (status & IPS_EXPECTED)
1861 ctinfo = IP_CT_RELATED;
1862 else
1863 ctinfo = IP_CT_NEW;
1864 }
1865 nf_ct_set(skb, ct, ctinfo);
1866 return 0;
1867 }
1868
1869 /*
1870 * icmp packets need special treatment to handle error messages that are
1871 * related to a connection.
1872 *
1873 * Callers need to check if skb has a conntrack assigned when this
1874 * helper returns; in such case skb belongs to an already known connection.
1875 */
1876 static unsigned int __cold
nf_conntrack_handle_icmp(struct nf_conn * tmpl,struct sk_buff * skb,unsigned int dataoff,u8 protonum,const struct nf_hook_state * state)1877 nf_conntrack_handle_icmp(struct nf_conn *tmpl,
1878 struct sk_buff *skb,
1879 unsigned int dataoff,
1880 u8 protonum,
1881 const struct nf_hook_state *state)
1882 {
1883 int ret;
1884
1885 if (state->pf == NFPROTO_IPV4 && protonum == IPPROTO_ICMP)
1886 ret = nf_conntrack_icmpv4_error(tmpl, skb, dataoff, state);
1887 #if IS_ENABLED(CONFIG_IPV6)
1888 else if (state->pf == NFPROTO_IPV6 && protonum == IPPROTO_ICMPV6)
1889 ret = nf_conntrack_icmpv6_error(tmpl, skb, dataoff, state);
1890 #endif
1891 else
1892 return NF_ACCEPT;
1893
1894 if (ret <= 0)
1895 NF_CT_STAT_INC_ATOMIC(state->net, error);
1896
1897 return ret;
1898 }
1899
generic_packet(struct nf_conn * ct,struct sk_buff * skb,enum ip_conntrack_info ctinfo)1900 static int generic_packet(struct nf_conn *ct, struct sk_buff *skb,
1901 enum ip_conntrack_info ctinfo)
1902 {
1903 const unsigned int *timeout = nf_ct_timeout_lookup(ct);
1904
1905 if (!timeout)
1906 timeout = &nf_generic_pernet(nf_ct_net(ct))->timeout;
1907
1908 nf_ct_refresh_acct(ct, ctinfo, skb, *timeout);
1909 return NF_ACCEPT;
1910 }
1911
1912 /* Returns verdict for packet, or -1 for invalid. */
nf_conntrack_handle_packet(struct nf_conn * ct,struct sk_buff * skb,unsigned int dataoff,enum ip_conntrack_info ctinfo,const struct nf_hook_state * state)1913 static int nf_conntrack_handle_packet(struct nf_conn *ct,
1914 struct sk_buff *skb,
1915 unsigned int dataoff,
1916 enum ip_conntrack_info ctinfo,
1917 const struct nf_hook_state *state)
1918 {
1919 switch (nf_ct_protonum(ct)) {
1920 case IPPROTO_TCP:
1921 return nf_conntrack_tcp_packet(ct, skb, dataoff,
1922 ctinfo, state);
1923 case IPPROTO_UDP:
1924 return nf_conntrack_udp_packet(ct, skb, dataoff,
1925 ctinfo, state);
1926 case IPPROTO_ICMP:
1927 return nf_conntrack_icmp_packet(ct, skb, ctinfo, state);
1928 #if IS_ENABLED(CONFIG_IPV6)
1929 case IPPROTO_ICMPV6:
1930 return nf_conntrack_icmpv6_packet(ct, skb, ctinfo, state);
1931 #endif
1932 #ifdef CONFIG_NF_CT_PROTO_UDPLITE
1933 case IPPROTO_UDPLITE:
1934 return nf_conntrack_udplite_packet(ct, skb, dataoff,
1935 ctinfo, state);
1936 #endif
1937 #ifdef CONFIG_NF_CT_PROTO_SCTP
1938 case IPPROTO_SCTP:
1939 return nf_conntrack_sctp_packet(ct, skb, dataoff,
1940 ctinfo, state);
1941 #endif
1942 #ifdef CONFIG_NF_CT_PROTO_DCCP
1943 case IPPROTO_DCCP:
1944 return nf_conntrack_dccp_packet(ct, skb, dataoff,
1945 ctinfo, state);
1946 #endif
1947 #ifdef CONFIG_NF_CT_PROTO_GRE
1948 case IPPROTO_GRE:
1949 return nf_conntrack_gre_packet(ct, skb, dataoff,
1950 ctinfo, state);
1951 #endif
1952 }
1953
1954 return generic_packet(ct, skb, ctinfo);
1955 }
1956
1957 unsigned int
nf_conntrack_in(struct sk_buff * skb,const struct nf_hook_state * state)1958 nf_conntrack_in(struct sk_buff *skb, const struct nf_hook_state *state)
1959 {
1960 enum ip_conntrack_info ctinfo;
1961 struct nf_conn *ct, *tmpl;
1962 u_int8_t protonum;
1963 int dataoff, ret;
1964
1965 tmpl = nf_ct_get(skb, &ctinfo);
1966 if (tmpl || ctinfo == IP_CT_UNTRACKED) {
1967 /* Previously seen (loopback or untracked)? Ignore. */
1968 if ((tmpl && !nf_ct_is_template(tmpl)) ||
1969 ctinfo == IP_CT_UNTRACKED)
1970 return NF_ACCEPT;
1971 skb->_nfct = 0;
1972 }
1973
1974 /* rcu_read_lock()ed by nf_hook_thresh */
1975 dataoff = get_l4proto(skb, skb_network_offset(skb), state->pf, &protonum);
1976 if (dataoff <= 0) {
1977 NF_CT_STAT_INC_ATOMIC(state->net, invalid);
1978 ret = NF_ACCEPT;
1979 goto out;
1980 }
1981
1982 if (protonum == IPPROTO_ICMP || protonum == IPPROTO_ICMPV6) {
1983 ret = nf_conntrack_handle_icmp(tmpl, skb, dataoff,
1984 protonum, state);
1985 if (ret <= 0) {
1986 ret = -ret;
1987 goto out;
1988 }
1989 /* ICMP[v6] protocol trackers may assign one conntrack. */
1990 if (skb->_nfct)
1991 goto out;
1992 }
1993 repeat:
1994 ret = resolve_normal_ct(tmpl, skb, dataoff,
1995 protonum, state);
1996 if (ret < 0) {
1997 /* Too stressed to deal. */
1998 NF_CT_STAT_INC_ATOMIC(state->net, drop);
1999 ret = NF_DROP;
2000 goto out;
2001 }
2002
2003 ct = nf_ct_get(skb, &ctinfo);
2004 if (!ct) {
2005 /* Not valid part of a connection */
2006 NF_CT_STAT_INC_ATOMIC(state->net, invalid);
2007 ret = NF_ACCEPT;
2008 goto out;
2009 }
2010
2011 ret = nf_conntrack_handle_packet(ct, skb, dataoff, ctinfo, state);
2012 if (ret <= 0) {
2013 /* Invalid: inverse of the return code tells
2014 * the netfilter core what to do */
2015 nf_ct_put(ct);
2016 skb->_nfct = 0;
2017 /* Special case: TCP tracker reports an attempt to reopen a
2018 * closed/aborted connection. We have to go back and create a
2019 * fresh conntrack.
2020 */
2021 if (ret == -NF_REPEAT)
2022 goto repeat;
2023
2024 NF_CT_STAT_INC_ATOMIC(state->net, invalid);
2025 if (ret == NF_DROP)
2026 NF_CT_STAT_INC_ATOMIC(state->net, drop);
2027
2028 ret = -ret;
2029 goto out;
2030 }
2031
2032 if (ctinfo == IP_CT_ESTABLISHED_REPLY &&
2033 !test_and_set_bit(IPS_SEEN_REPLY_BIT, &ct->status))
2034 nf_conntrack_event_cache(IPCT_REPLY, ct);
2035 out:
2036 if (tmpl)
2037 nf_ct_put(tmpl);
2038
2039 return ret;
2040 }
2041 EXPORT_SYMBOL_GPL(nf_conntrack_in);
2042
2043 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
__nf_ct_refresh_acct(struct nf_conn * ct,enum ip_conntrack_info ctinfo,const struct sk_buff * skb,u32 extra_jiffies,bool do_acct)2044 void __nf_ct_refresh_acct(struct nf_conn *ct,
2045 enum ip_conntrack_info ctinfo,
2046 const struct sk_buff *skb,
2047 u32 extra_jiffies,
2048 bool do_acct)
2049 {
2050 /* Only update if this is not a fixed timeout */
2051 if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
2052 goto acct;
2053
2054 /* If not in hash table, timer will not be active yet */
2055 if (nf_ct_is_confirmed(ct))
2056 extra_jiffies += nfct_time_stamp;
2057
2058 if (READ_ONCE(ct->timeout) != extra_jiffies)
2059 WRITE_ONCE(ct->timeout, extra_jiffies);
2060 acct:
2061 if (do_acct)
2062 nf_ct_acct_update(ct, CTINFO2DIR(ctinfo), skb->len);
2063 }
2064 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
2065
nf_ct_kill_acct(struct nf_conn * ct,enum ip_conntrack_info ctinfo,const struct sk_buff * skb)2066 bool nf_ct_kill_acct(struct nf_conn *ct,
2067 enum ip_conntrack_info ctinfo,
2068 const struct sk_buff *skb)
2069 {
2070 nf_ct_acct_update(ct, CTINFO2DIR(ctinfo), skb->len);
2071
2072 return nf_ct_delete(ct, 0, 0);
2073 }
2074 EXPORT_SYMBOL_GPL(nf_ct_kill_acct);
2075
2076 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
2077
2078 #include <linux/netfilter/nfnetlink.h>
2079 #include <linux/netfilter/nfnetlink_conntrack.h>
2080 #include <linux/mutex.h>
2081
2082 /* Generic function for tcp/udp/sctp/dccp and alike. */
nf_ct_port_tuple_to_nlattr(struct sk_buff * skb,const struct nf_conntrack_tuple * tuple)2083 int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
2084 const struct nf_conntrack_tuple *tuple)
2085 {
2086 if (nla_put_be16(skb, CTA_PROTO_SRC_PORT, tuple->src.u.tcp.port) ||
2087 nla_put_be16(skb, CTA_PROTO_DST_PORT, tuple->dst.u.tcp.port))
2088 goto nla_put_failure;
2089 return 0;
2090
2091 nla_put_failure:
2092 return -1;
2093 }
2094 EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
2095
2096 const struct nla_policy nf_ct_port_nla_policy[CTA_PROTO_MAX+1] = {
2097 [CTA_PROTO_SRC_PORT] = { .type = NLA_U16 },
2098 [CTA_PROTO_DST_PORT] = { .type = NLA_U16 },
2099 };
2100 EXPORT_SYMBOL_GPL(nf_ct_port_nla_policy);
2101
nf_ct_port_nlattr_to_tuple(struct nlattr * tb[],struct nf_conntrack_tuple * t,u_int32_t flags)2102 int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
2103 struct nf_conntrack_tuple *t,
2104 u_int32_t flags)
2105 {
2106 if (flags & CTA_FILTER_FLAG(CTA_PROTO_SRC_PORT)) {
2107 if (!tb[CTA_PROTO_SRC_PORT])
2108 return -EINVAL;
2109
2110 t->src.u.tcp.port = nla_get_be16(tb[CTA_PROTO_SRC_PORT]);
2111 }
2112
2113 if (flags & CTA_FILTER_FLAG(CTA_PROTO_DST_PORT)) {
2114 if (!tb[CTA_PROTO_DST_PORT])
2115 return -EINVAL;
2116
2117 t->dst.u.tcp.port = nla_get_be16(tb[CTA_PROTO_DST_PORT]);
2118 }
2119
2120 return 0;
2121 }
2122 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
2123
nf_ct_port_nlattr_tuple_size(void)2124 unsigned int nf_ct_port_nlattr_tuple_size(void)
2125 {
2126 static unsigned int size __read_mostly;
2127
2128 if (!size)
2129 size = nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
2130
2131 return size;
2132 }
2133 EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_tuple_size);
2134 #endif
2135
2136 /* Used by ipt_REJECT and ip6t_REJECT. */
nf_conntrack_attach(struct sk_buff * nskb,const struct sk_buff * skb)2137 static void nf_conntrack_attach(struct sk_buff *nskb, const struct sk_buff *skb)
2138 {
2139 struct nf_conn *ct;
2140 enum ip_conntrack_info ctinfo;
2141
2142 /* This ICMP is in reverse direction to the packet which caused it */
2143 ct = nf_ct_get(skb, &ctinfo);
2144 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL)
2145 ctinfo = IP_CT_RELATED_REPLY;
2146 else
2147 ctinfo = IP_CT_RELATED;
2148
2149 /* Attach to new skbuff, and increment count */
2150 nf_ct_set(nskb, ct, ctinfo);
2151 nf_conntrack_get(skb_nfct(nskb));
2152 }
2153
__nf_conntrack_update(struct net * net,struct sk_buff * skb,struct nf_conn * ct,enum ip_conntrack_info ctinfo)2154 static int __nf_conntrack_update(struct net *net, struct sk_buff *skb,
2155 struct nf_conn *ct,
2156 enum ip_conntrack_info ctinfo)
2157 {
2158 const struct nf_nat_hook *nat_hook;
2159 struct nf_conntrack_tuple_hash *h;
2160 struct nf_conntrack_tuple tuple;
2161 unsigned int status;
2162 int dataoff;
2163 u16 l3num;
2164 u8 l4num;
2165
2166 l3num = nf_ct_l3num(ct);
2167
2168 dataoff = get_l4proto(skb, skb_network_offset(skb), l3num, &l4num);
2169 if (dataoff <= 0)
2170 return NF_DROP;
2171
2172 if (!nf_ct_get_tuple(skb, skb_network_offset(skb), dataoff, l3num,
2173 l4num, net, &tuple))
2174 return NF_DROP;
2175
2176 if (ct->status & IPS_SRC_NAT) {
2177 memcpy(tuple.src.u3.all,
2178 ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.u3.all,
2179 sizeof(tuple.src.u3.all));
2180 tuple.src.u.all =
2181 ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.src.u.all;
2182 }
2183
2184 if (ct->status & IPS_DST_NAT) {
2185 memcpy(tuple.dst.u3.all,
2186 ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.dst.u3.all,
2187 sizeof(tuple.dst.u3.all));
2188 tuple.dst.u.all =
2189 ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple.dst.u.all;
2190 }
2191
2192 h = nf_conntrack_find_get(net, nf_ct_zone(ct), &tuple);
2193 if (!h)
2194 return NF_ACCEPT;
2195
2196 /* Store status bits of the conntrack that is clashing to re-do NAT
2197 * mangling according to what it has been done already to this packet.
2198 */
2199 status = ct->status;
2200
2201 nf_ct_put(ct);
2202 ct = nf_ct_tuplehash_to_ctrack(h);
2203 nf_ct_set(skb, ct, ctinfo);
2204
2205 nat_hook = rcu_dereference(nf_nat_hook);
2206 if (!nat_hook)
2207 return NF_ACCEPT;
2208
2209 if (status & IPS_SRC_NAT) {
2210 unsigned int verdict = nat_hook->manip_pkt(skb, ct,
2211 NF_NAT_MANIP_SRC,
2212 IP_CT_DIR_ORIGINAL);
2213 if (verdict != NF_ACCEPT)
2214 return verdict;
2215 }
2216
2217 if (status & IPS_DST_NAT) {
2218 unsigned int verdict = nat_hook->manip_pkt(skb, ct,
2219 NF_NAT_MANIP_DST,
2220 IP_CT_DIR_ORIGINAL);
2221 if (verdict != NF_ACCEPT)
2222 return verdict;
2223 }
2224
2225 return NF_ACCEPT;
2226 }
2227
2228 /* This packet is coming from userspace via nf_queue, complete the packet
2229 * processing after the helper invocation in nf_confirm().
2230 */
nf_confirm_cthelper(struct sk_buff * skb,struct nf_conn * ct,enum ip_conntrack_info ctinfo)2231 static int nf_confirm_cthelper(struct sk_buff *skb, struct nf_conn *ct,
2232 enum ip_conntrack_info ctinfo)
2233 {
2234 const struct nf_conntrack_helper *helper;
2235 const struct nf_conn_help *help;
2236 int protoff;
2237
2238 help = nfct_help(ct);
2239 if (!help)
2240 return NF_ACCEPT;
2241
2242 helper = rcu_dereference(help->helper);
2243 if (!helper)
2244 return NF_ACCEPT;
2245
2246 if (!(helper->flags & NF_CT_HELPER_F_USERSPACE))
2247 return NF_ACCEPT;
2248
2249 switch (nf_ct_l3num(ct)) {
2250 case NFPROTO_IPV4:
2251 protoff = skb_network_offset(skb) + ip_hdrlen(skb);
2252 break;
2253 #if IS_ENABLED(CONFIG_IPV6)
2254 case NFPROTO_IPV6: {
2255 __be16 frag_off;
2256 u8 pnum;
2257
2258 pnum = ipv6_hdr(skb)->nexthdr;
2259 protoff = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &pnum,
2260 &frag_off);
2261 if (protoff < 0 || (frag_off & htons(~0x7)) != 0)
2262 return NF_ACCEPT;
2263 break;
2264 }
2265 #endif
2266 default:
2267 return NF_ACCEPT;
2268 }
2269
2270 if (test_bit(IPS_SEQ_ADJUST_BIT, &ct->status) &&
2271 !nf_is_loopback_packet(skb)) {
2272 if (!nf_ct_seq_adjust(skb, ct, ctinfo, protoff)) {
2273 NF_CT_STAT_INC_ATOMIC(nf_ct_net(ct), drop);
2274 return NF_DROP;
2275 }
2276 }
2277
2278 /* We've seen it coming out the other side: confirm it */
2279 return nf_conntrack_confirm(skb);
2280 }
2281
nf_conntrack_update(struct net * net,struct sk_buff * skb)2282 static int nf_conntrack_update(struct net *net, struct sk_buff *skb)
2283 {
2284 enum ip_conntrack_info ctinfo;
2285 struct nf_conn *ct;
2286
2287 ct = nf_ct_get(skb, &ctinfo);
2288 if (!ct)
2289 return NF_ACCEPT;
2290
2291 if (!nf_ct_is_confirmed(ct)) {
2292 int ret = __nf_conntrack_update(net, skb, ct, ctinfo);
2293
2294 if (ret != NF_ACCEPT)
2295 return ret;
2296
2297 ct = nf_ct_get(skb, &ctinfo);
2298 if (!ct)
2299 return NF_ACCEPT;
2300 }
2301
2302 return nf_confirm_cthelper(skb, ct, ctinfo);
2303 }
2304
nf_conntrack_get_tuple_skb(struct nf_conntrack_tuple * dst_tuple,const struct sk_buff * skb)2305 static bool nf_conntrack_get_tuple_skb(struct nf_conntrack_tuple *dst_tuple,
2306 const struct sk_buff *skb)
2307 {
2308 const struct nf_conntrack_tuple *src_tuple;
2309 const struct nf_conntrack_tuple_hash *hash;
2310 struct nf_conntrack_tuple srctuple;
2311 enum ip_conntrack_info ctinfo;
2312 struct nf_conn *ct;
2313
2314 ct = nf_ct_get(skb, &ctinfo);
2315 if (ct) {
2316 src_tuple = nf_ct_tuple(ct, CTINFO2DIR(ctinfo));
2317 memcpy(dst_tuple, src_tuple, sizeof(*dst_tuple));
2318 return true;
2319 }
2320
2321 if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb),
2322 NFPROTO_IPV4, dev_net(skb->dev),
2323 &srctuple))
2324 return false;
2325
2326 hash = nf_conntrack_find_get(dev_net(skb->dev),
2327 &nf_ct_zone_dflt,
2328 &srctuple);
2329 if (!hash)
2330 return false;
2331
2332 ct = nf_ct_tuplehash_to_ctrack(hash);
2333 src_tuple = nf_ct_tuple(ct, !hash->tuple.dst.dir);
2334 memcpy(dst_tuple, src_tuple, sizeof(*dst_tuple));
2335 nf_ct_put(ct);
2336
2337 return true;
2338 }
2339
2340 /* Bring out ya dead! */
2341 static struct nf_conn *
get_next_corpse(int (* iter)(struct nf_conn * i,void * data),const struct nf_ct_iter_data * iter_data,unsigned int * bucket)2342 get_next_corpse(int (*iter)(struct nf_conn *i, void *data),
2343 const struct nf_ct_iter_data *iter_data, unsigned int *bucket)
2344 {
2345 struct nf_conntrack_tuple_hash *h;
2346 struct nf_conn *ct;
2347 struct hlist_nulls_node *n;
2348 spinlock_t *lockp;
2349
2350 for (; *bucket < nf_conntrack_htable_size; (*bucket)++) {
2351 struct hlist_nulls_head *hslot = &nf_conntrack_hash[*bucket];
2352
2353 if (hlist_nulls_empty(hslot))
2354 continue;
2355
2356 lockp = &nf_conntrack_locks[*bucket % CONNTRACK_LOCKS];
2357 local_bh_disable();
2358 nf_conntrack_lock(lockp);
2359 hlist_nulls_for_each_entry(h, n, hslot, hnnode) {
2360 if (NF_CT_DIRECTION(h) != IP_CT_DIR_REPLY)
2361 continue;
2362 /* All nf_conn objects are added to hash table twice, one
2363 * for original direction tuple, once for the reply tuple.
2364 *
2365 * Exception: In the IPS_NAT_CLASH case, only the reply
2366 * tuple is added (the original tuple already existed for
2367 * a different object).
2368 *
2369 * We only need to call the iterator once for each
2370 * conntrack, so we just use the 'reply' direction
2371 * tuple while iterating.
2372 */
2373 ct = nf_ct_tuplehash_to_ctrack(h);
2374
2375 if (iter_data->net &&
2376 !net_eq(iter_data->net, nf_ct_net(ct)))
2377 continue;
2378
2379 if (iter(ct, iter_data->data))
2380 goto found;
2381 }
2382 spin_unlock(lockp);
2383 local_bh_enable();
2384 cond_resched();
2385 }
2386
2387 return NULL;
2388 found:
2389 refcount_inc(&ct->ct_general.use);
2390 spin_unlock(lockp);
2391 local_bh_enable();
2392 return ct;
2393 }
2394
nf_ct_iterate_cleanup(int (* iter)(struct nf_conn * i,void * data),const struct nf_ct_iter_data * iter_data)2395 static void nf_ct_iterate_cleanup(int (*iter)(struct nf_conn *i, void *data),
2396 const struct nf_ct_iter_data *iter_data)
2397 {
2398 unsigned int bucket = 0;
2399 struct nf_conn *ct;
2400
2401 might_sleep();
2402
2403 mutex_lock(&nf_conntrack_mutex);
2404 while ((ct = get_next_corpse(iter, iter_data, &bucket)) != NULL) {
2405 /* Time to push up daises... */
2406
2407 nf_ct_delete(ct, iter_data->portid, iter_data->report);
2408 nf_ct_put(ct);
2409 cond_resched();
2410 }
2411 mutex_unlock(&nf_conntrack_mutex);
2412 }
2413
nf_ct_iterate_cleanup_net(int (* iter)(struct nf_conn * i,void * data),const struct nf_ct_iter_data * iter_data)2414 void nf_ct_iterate_cleanup_net(int (*iter)(struct nf_conn *i, void *data),
2415 const struct nf_ct_iter_data *iter_data)
2416 {
2417 struct net *net = iter_data->net;
2418 struct nf_conntrack_net *cnet = nf_ct_pernet(net);
2419
2420 might_sleep();
2421
2422 if (atomic_read(&cnet->count) == 0)
2423 return;
2424
2425 nf_ct_iterate_cleanup(iter, iter_data);
2426 }
2427 EXPORT_SYMBOL_GPL(nf_ct_iterate_cleanup_net);
2428
2429 /**
2430 * nf_ct_iterate_destroy - destroy unconfirmed conntracks and iterate table
2431 * @iter: callback to invoke for each conntrack
2432 * @data: data to pass to @iter
2433 *
2434 * Like nf_ct_iterate_cleanup, but first marks conntracks on the
2435 * unconfirmed list as dying (so they will not be inserted into
2436 * main table).
2437 *
2438 * Can only be called in module exit path.
2439 */
2440 void
nf_ct_iterate_destroy(int (* iter)(struct nf_conn * i,void * data),void * data)2441 nf_ct_iterate_destroy(int (*iter)(struct nf_conn *i, void *data), void *data)
2442 {
2443 struct nf_ct_iter_data iter_data = {};
2444 struct net *net;
2445
2446 down_read(&net_rwsem);
2447 for_each_net(net) {
2448 struct nf_conntrack_net *cnet = nf_ct_pernet(net);
2449
2450 if (atomic_read(&cnet->count) == 0)
2451 continue;
2452 nf_queue_nf_hook_drop(net);
2453 }
2454 up_read(&net_rwsem);
2455
2456 /* Need to wait for netns cleanup worker to finish, if its
2457 * running -- it might have deleted a net namespace from
2458 * the global list, so hook drop above might not have
2459 * affected all namespaces.
2460 */
2461 net_ns_barrier();
2462
2463 /* a skb w. unconfirmed conntrack could have been reinjected just
2464 * before we called nf_queue_nf_hook_drop().
2465 *
2466 * This makes sure its inserted into conntrack table.
2467 */
2468 synchronize_net();
2469
2470 nf_ct_ext_bump_genid();
2471 iter_data.data = data;
2472 nf_ct_iterate_cleanup(iter, &iter_data);
2473
2474 /* Another cpu might be in a rcu read section with
2475 * rcu protected pointer cleared in iter callback
2476 * or hidden via nf_ct_ext_bump_genid() above.
2477 *
2478 * Wait until those are done.
2479 */
2480 synchronize_rcu();
2481 }
2482 EXPORT_SYMBOL_GPL(nf_ct_iterate_destroy);
2483
kill_all(struct nf_conn * i,void * data)2484 static int kill_all(struct nf_conn *i, void *data)
2485 {
2486 return 1;
2487 }
2488
nf_conntrack_cleanup_start(void)2489 void nf_conntrack_cleanup_start(void)
2490 {
2491 cleanup_nf_conntrack_bpf();
2492 conntrack_gc_work.exiting = true;
2493 }
2494
nf_conntrack_cleanup_end(void)2495 void nf_conntrack_cleanup_end(void)
2496 {
2497 RCU_INIT_POINTER(nf_ct_hook, NULL);
2498 cancel_delayed_work_sync(&conntrack_gc_work.dwork);
2499 kvfree(nf_conntrack_hash);
2500
2501 nf_conntrack_proto_fini();
2502 nf_conntrack_helper_fini();
2503 nf_conntrack_expect_fini();
2504
2505 kmem_cache_destroy(nf_conntrack_cachep);
2506 }
2507
2508 /*
2509 * Mishearing the voices in his head, our hero wonders how he's
2510 * supposed to kill the mall.
2511 */
nf_conntrack_cleanup_net(struct net * net)2512 void nf_conntrack_cleanup_net(struct net *net)
2513 {
2514 LIST_HEAD(single);
2515
2516 list_add(&net->exit_list, &single);
2517 nf_conntrack_cleanup_net_list(&single);
2518 }
2519
nf_conntrack_cleanup_net_list(struct list_head * net_exit_list)2520 void nf_conntrack_cleanup_net_list(struct list_head *net_exit_list)
2521 {
2522 struct nf_ct_iter_data iter_data = {};
2523 struct net *net;
2524 int busy;
2525
2526 /*
2527 * This makes sure all current packets have passed through
2528 * netfilter framework. Roll on, two-stage module
2529 * delete...
2530 */
2531 synchronize_rcu_expedited();
2532 i_see_dead_people:
2533 busy = 0;
2534 list_for_each_entry(net, net_exit_list, exit_list) {
2535 struct nf_conntrack_net *cnet = nf_ct_pernet(net);
2536
2537 iter_data.net = net;
2538 nf_ct_iterate_cleanup_net(kill_all, &iter_data);
2539 if (atomic_read(&cnet->count) != 0)
2540 busy = 1;
2541 }
2542 if (busy) {
2543 schedule();
2544 goto i_see_dead_people;
2545 }
2546
2547 list_for_each_entry(net, net_exit_list, exit_list) {
2548 nf_conntrack_ecache_pernet_fini(net);
2549 nf_conntrack_expect_pernet_fini(net);
2550 free_percpu(net->ct.stat);
2551 }
2552 }
2553
nf_ct_alloc_hashtable(unsigned int * sizep,int nulls)2554 void *nf_ct_alloc_hashtable(unsigned int *sizep, int nulls)
2555 {
2556 struct hlist_nulls_head *hash;
2557 unsigned int nr_slots, i;
2558
2559 if (*sizep > (UINT_MAX / sizeof(struct hlist_nulls_head)))
2560 return NULL;
2561
2562 BUILD_BUG_ON(sizeof(struct hlist_nulls_head) != sizeof(struct hlist_head));
2563 nr_slots = *sizep = roundup(*sizep, PAGE_SIZE / sizeof(struct hlist_nulls_head));
2564
2565 hash = kvcalloc(nr_slots, sizeof(struct hlist_nulls_head), GFP_KERNEL);
2566
2567 if (hash && nulls)
2568 for (i = 0; i < nr_slots; i++)
2569 INIT_HLIST_NULLS_HEAD(&hash[i], i);
2570
2571 return hash;
2572 }
2573 EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
2574
nf_conntrack_hash_resize(unsigned int hashsize)2575 int nf_conntrack_hash_resize(unsigned int hashsize)
2576 {
2577 int i, bucket;
2578 unsigned int old_size;
2579 struct hlist_nulls_head *hash, *old_hash;
2580 struct nf_conntrack_tuple_hash *h;
2581 struct nf_conn *ct;
2582
2583 if (!hashsize)
2584 return -EINVAL;
2585
2586 hash = nf_ct_alloc_hashtable(&hashsize, 1);
2587 if (!hash)
2588 return -ENOMEM;
2589
2590 mutex_lock(&nf_conntrack_mutex);
2591 old_size = nf_conntrack_htable_size;
2592 if (old_size == hashsize) {
2593 mutex_unlock(&nf_conntrack_mutex);
2594 kvfree(hash);
2595 return 0;
2596 }
2597
2598 local_bh_disable();
2599 nf_conntrack_all_lock();
2600 write_seqcount_begin(&nf_conntrack_generation);
2601
2602 /* Lookups in the old hash might happen in parallel, which means we
2603 * might get false negatives during connection lookup. New connections
2604 * created because of a false negative won't make it into the hash
2605 * though since that required taking the locks.
2606 */
2607
2608 for (i = 0; i < nf_conntrack_htable_size; i++) {
2609 while (!hlist_nulls_empty(&nf_conntrack_hash[i])) {
2610 unsigned int zone_id;
2611
2612 h = hlist_nulls_entry(nf_conntrack_hash[i].first,
2613 struct nf_conntrack_tuple_hash, hnnode);
2614 ct = nf_ct_tuplehash_to_ctrack(h);
2615 hlist_nulls_del_rcu(&h->hnnode);
2616
2617 zone_id = nf_ct_zone_id(nf_ct_zone(ct), NF_CT_DIRECTION(h));
2618 bucket = __hash_conntrack(nf_ct_net(ct),
2619 &h->tuple, zone_id, hashsize);
2620 hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
2621 }
2622 }
2623 old_hash = nf_conntrack_hash;
2624
2625 nf_conntrack_hash = hash;
2626 nf_conntrack_htable_size = hashsize;
2627
2628 write_seqcount_end(&nf_conntrack_generation);
2629 nf_conntrack_all_unlock();
2630 local_bh_enable();
2631
2632 mutex_unlock(&nf_conntrack_mutex);
2633
2634 synchronize_net();
2635 kvfree(old_hash);
2636 return 0;
2637 }
2638
nf_conntrack_set_hashsize(const char * val,const struct kernel_param * kp)2639 int nf_conntrack_set_hashsize(const char *val, const struct kernel_param *kp)
2640 {
2641 unsigned int hashsize;
2642 int rc;
2643
2644 if (current->nsproxy->net_ns != &init_net)
2645 return -EOPNOTSUPP;
2646
2647 /* On boot, we can set this without any fancy locking. */
2648 if (!nf_conntrack_hash)
2649 return param_set_uint(val, kp);
2650
2651 rc = kstrtouint(val, 0, &hashsize);
2652 if (rc)
2653 return rc;
2654
2655 return nf_conntrack_hash_resize(hashsize);
2656 }
2657
nf_conntrack_init_start(void)2658 int nf_conntrack_init_start(void)
2659 {
2660 unsigned long nr_pages = totalram_pages();
2661 int max_factor = 8;
2662 int ret = -ENOMEM;
2663 int i;
2664
2665 seqcount_spinlock_init(&nf_conntrack_generation,
2666 &nf_conntrack_locks_all_lock);
2667
2668 for (i = 0; i < CONNTRACK_LOCKS; i++)
2669 spin_lock_init(&nf_conntrack_locks[i]);
2670
2671 if (!nf_conntrack_htable_size) {
2672 nf_conntrack_htable_size
2673 = (((nr_pages << PAGE_SHIFT) / 16384)
2674 / sizeof(struct hlist_head));
2675 if (BITS_PER_LONG >= 64 &&
2676 nr_pages > (4 * (1024 * 1024 * 1024 / PAGE_SIZE)))
2677 nf_conntrack_htable_size = 262144;
2678 else if (nr_pages > (1024 * 1024 * 1024 / PAGE_SIZE))
2679 nf_conntrack_htable_size = 65536;
2680
2681 if (nf_conntrack_htable_size < 1024)
2682 nf_conntrack_htable_size = 1024;
2683 /* Use a max. factor of one by default to keep the average
2684 * hash chain length at 2 entries. Each entry has to be added
2685 * twice (once for original direction, once for reply).
2686 * When a table size is given we use the old value of 8 to
2687 * avoid implicit reduction of the max entries setting.
2688 */
2689 max_factor = 1;
2690 }
2691
2692 nf_conntrack_hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size, 1);
2693 if (!nf_conntrack_hash)
2694 return -ENOMEM;
2695
2696 nf_conntrack_max = max_factor * nf_conntrack_htable_size;
2697
2698 nf_conntrack_cachep = kmem_cache_create("nf_conntrack",
2699 sizeof(struct nf_conn),
2700 NFCT_INFOMASK + 1,
2701 SLAB_TYPESAFE_BY_RCU | SLAB_HWCACHE_ALIGN, NULL);
2702 if (!nf_conntrack_cachep)
2703 goto err_cachep;
2704
2705 ret = nf_conntrack_expect_init();
2706 if (ret < 0)
2707 goto err_expect;
2708
2709 ret = nf_conntrack_helper_init();
2710 if (ret < 0)
2711 goto err_helper;
2712
2713 ret = nf_conntrack_proto_init();
2714 if (ret < 0)
2715 goto err_proto;
2716
2717 conntrack_gc_work_init(&conntrack_gc_work);
2718 queue_delayed_work(system_power_efficient_wq, &conntrack_gc_work.dwork, HZ);
2719
2720 ret = register_nf_conntrack_bpf();
2721 if (ret < 0)
2722 goto err_kfunc;
2723
2724 return 0;
2725
2726 err_kfunc:
2727 cancel_delayed_work_sync(&conntrack_gc_work.dwork);
2728 nf_conntrack_proto_fini();
2729 err_proto:
2730 nf_conntrack_helper_fini();
2731 err_helper:
2732 nf_conntrack_expect_fini();
2733 err_expect:
2734 kmem_cache_destroy(nf_conntrack_cachep);
2735 err_cachep:
2736 kvfree(nf_conntrack_hash);
2737 return ret;
2738 }
2739
nf_conntrack_set_closing(struct nf_conntrack * nfct)2740 static void nf_conntrack_set_closing(struct nf_conntrack *nfct)
2741 {
2742 struct nf_conn *ct = nf_ct_to_nf_conn(nfct);
2743
2744 switch (nf_ct_protonum(ct)) {
2745 case IPPROTO_TCP:
2746 nf_conntrack_tcp_set_closing(ct);
2747 break;
2748 }
2749 }
2750
2751 static const struct nf_ct_hook nf_conntrack_hook = {
2752 .update = nf_conntrack_update,
2753 .destroy = nf_ct_destroy,
2754 .get_tuple_skb = nf_conntrack_get_tuple_skb,
2755 .attach = nf_conntrack_attach,
2756 .set_closing = nf_conntrack_set_closing,
2757 .confirm = __nf_conntrack_confirm,
2758 };
2759
nf_conntrack_init_end(void)2760 void nf_conntrack_init_end(void)
2761 {
2762 RCU_INIT_POINTER(nf_ct_hook, &nf_conntrack_hook);
2763 }
2764
2765 /*
2766 * We need to use special "null" values, not used in hash table
2767 */
2768 #define UNCONFIRMED_NULLS_VAL ((1<<30)+0)
2769
nf_conntrack_init_net(struct net * net)2770 int nf_conntrack_init_net(struct net *net)
2771 {
2772 struct nf_conntrack_net *cnet = nf_ct_pernet(net);
2773 int ret = -ENOMEM;
2774
2775 BUILD_BUG_ON(IP_CT_UNTRACKED == IP_CT_NUMBER);
2776 BUILD_BUG_ON_NOT_POWER_OF_2(CONNTRACK_LOCKS);
2777 atomic_set(&cnet->count, 0);
2778
2779 net->ct.stat = alloc_percpu(struct ip_conntrack_stat);
2780 if (!net->ct.stat)
2781 return ret;
2782
2783 ret = nf_conntrack_expect_pernet_init(net);
2784 if (ret < 0)
2785 goto err_expect;
2786
2787 nf_conntrack_acct_pernet_init(net);
2788 nf_conntrack_tstamp_pernet_init(net);
2789 nf_conntrack_ecache_pernet_init(net);
2790 nf_conntrack_proto_pernet_init(net);
2791
2792 return 0;
2793
2794 err_expect:
2795 free_percpu(net->ct.stat);
2796 return ret;
2797 }
2798
2799 /* ctnetlink code shared by both ctnetlink and nf_conntrack_bpf */
2800
__nf_ct_change_timeout(struct nf_conn * ct,u64 timeout)2801 int __nf_ct_change_timeout(struct nf_conn *ct, u64 timeout)
2802 {
2803 if (test_bit(IPS_FIXED_TIMEOUT_BIT, &ct->status))
2804 return -EPERM;
2805
2806 __nf_ct_set_timeout(ct, timeout);
2807
2808 if (test_bit(IPS_DYING_BIT, &ct->status))
2809 return -ETIME;
2810
2811 return 0;
2812 }
2813 EXPORT_SYMBOL_GPL(__nf_ct_change_timeout);
2814
__nf_ct_change_status(struct nf_conn * ct,unsigned long on,unsigned long off)2815 void __nf_ct_change_status(struct nf_conn *ct, unsigned long on, unsigned long off)
2816 {
2817 unsigned int bit;
2818
2819 /* Ignore these unchangable bits */
2820 on &= ~IPS_UNCHANGEABLE_MASK;
2821 off &= ~IPS_UNCHANGEABLE_MASK;
2822
2823 for (bit = 0; bit < __IPS_MAX_BIT; bit++) {
2824 if (on & (1 << bit))
2825 set_bit(bit, &ct->status);
2826 else if (off & (1 << bit))
2827 clear_bit(bit, &ct->status);
2828 }
2829 }
2830 EXPORT_SYMBOL_GPL(__nf_ct_change_status);
2831
nf_ct_change_status_common(struct nf_conn * ct,unsigned int status)2832 int nf_ct_change_status_common(struct nf_conn *ct, unsigned int status)
2833 {
2834 unsigned long d;
2835
2836 d = ct->status ^ status;
2837
2838 if (d & (IPS_EXPECTED|IPS_CONFIRMED|IPS_DYING))
2839 /* unchangeable */
2840 return -EBUSY;
2841
2842 if (d & IPS_SEEN_REPLY && !(status & IPS_SEEN_REPLY))
2843 /* SEEN_REPLY bit can only be set */
2844 return -EBUSY;
2845
2846 if (d & IPS_ASSURED && !(status & IPS_ASSURED))
2847 /* ASSURED bit can only be set */
2848 return -EBUSY;
2849
2850 __nf_ct_change_status(ct, status, 0);
2851 return 0;
2852 }
2853 EXPORT_SYMBOL_GPL(nf_ct_change_status_common);
2854