1
2
3
4
5
6
7
8
9
10
11
12#include <linux/module.h>
13#include <linux/random.h>
14#include <linux/sched.h>
15#include <linux/slab.h>
16#include <linux/wait.h>
17#include <linux/vmalloc.h>
18#include <linux/memblock.h>
19
20#include <net/addrconf.h>
21#include <net/inet_connection_sock.h>
22#include <net/inet_hashtables.h>
23#include <net/secure_seq.h>
24#include <net/ip.h>
25#include <net/tcp.h>
26#include <net/sock_reuseport.h>
27
28static u32 inet_ehashfn(const struct net *net, const __be32 laddr,
29 const __u16 lport, const __be32 faddr,
30 const __be16 fport)
31{
32 static u32 inet_ehash_secret __read_mostly;
33
34 net_get_random_once(&inet_ehash_secret, sizeof(inet_ehash_secret));
35
36 return __inet_ehashfn(laddr, lport, faddr, fport,
37 inet_ehash_secret + net_hash_mix(net));
38}
39
40
41
42
43static u32 sk_ehashfn(const struct sock *sk)
44{
45#if IS_ENABLED(CONFIG_IPV6)
46 if (sk->sk_family == AF_INET6 &&
47 !ipv6_addr_v4mapped(&sk->sk_v6_daddr))
48 return inet6_ehashfn(sock_net(sk),
49 &sk->sk_v6_rcv_saddr, sk->sk_num,
50 &sk->sk_v6_daddr, sk->sk_dport);
51#endif
52 return inet_ehashfn(sock_net(sk),
53 sk->sk_rcv_saddr, sk->sk_num,
54 sk->sk_daddr, sk->sk_dport);
55}
56
57
58
59
60
61struct inet_bind_bucket *inet_bind_bucket_create(struct kmem_cache *cachep,
62 struct net *net,
63 struct inet_bind_hashbucket *head,
64 const unsigned short snum,
65 int l3mdev)
66{
67 struct inet_bind_bucket *tb = kmem_cache_alloc(cachep, GFP_ATOMIC);
68
69 if (tb) {
70 write_pnet(&tb->ib_net, net);
71 tb->l3mdev = l3mdev;
72 tb->port = snum;
73 tb->fastreuse = 0;
74 tb->fastreuseport = 0;
75 INIT_HLIST_HEAD(&tb->owners);
76 hlist_add_head(&tb->node, &head->chain);
77 }
78 return tb;
79}
80
81
82
83
84void inet_bind_bucket_destroy(struct kmem_cache *cachep, struct inet_bind_bucket *tb)
85{
86 if (hlist_empty(&tb->owners)) {
87 __hlist_del(&tb->node);
88 kmem_cache_free(cachep, tb);
89 }
90}
91
92void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
93 const unsigned short snum)
94{
95 inet_sk(sk)->inet_num = snum;
96 sk_add_bind_node(sk, &tb->owners);
97 inet_csk(sk)->icsk_bind_hash = tb;
98}
99
100
101
102
103static void __inet_put_port(struct sock *sk)
104{
105 struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
106 const int bhash = inet_bhashfn(sock_net(sk), inet_sk(sk)->inet_num,
107 hashinfo->bhash_size);
108 struct inet_bind_hashbucket *head = &hashinfo->bhash[bhash];
109 struct inet_bind_bucket *tb;
110
111 spin_lock(&head->lock);
112 tb = inet_csk(sk)->icsk_bind_hash;
113 __sk_del_bind_node(sk);
114 inet_csk(sk)->icsk_bind_hash = NULL;
115 inet_sk(sk)->inet_num = 0;
116 inet_bind_bucket_destroy(hashinfo->bind_bucket_cachep, tb);
117 spin_unlock(&head->lock);
118}
119
120void inet_put_port(struct sock *sk)
121{
122 local_bh_disable();
123 __inet_put_port(sk);
124 local_bh_enable();
125}
126EXPORT_SYMBOL(inet_put_port);
127
128int __inet_inherit_port(const struct sock *sk, struct sock *child)
129{
130 struct inet_hashinfo *table = sk->sk_prot->h.hashinfo;
131 unsigned short port = inet_sk(child)->inet_num;
132 const int bhash = inet_bhashfn(sock_net(sk), port,
133 table->bhash_size);
134 struct inet_bind_hashbucket *head = &table->bhash[bhash];
135 struct inet_bind_bucket *tb;
136 int l3mdev;
137
138 spin_lock(&head->lock);
139 tb = inet_csk(sk)->icsk_bind_hash;
140 if (unlikely(!tb)) {
141 spin_unlock(&head->lock);
142 return -ENOENT;
143 }
144 if (tb->port != port) {
145 l3mdev = inet_sk_bound_l3mdev(sk);
146
147
148
149
150
151
152 inet_bind_bucket_for_each(tb, &head->chain) {
153 if (net_eq(ib_net(tb), sock_net(sk)) &&
154 tb->l3mdev == l3mdev && tb->port == port)
155 break;
156 }
157 if (!tb) {
158 tb = inet_bind_bucket_create(table->bind_bucket_cachep,
159 sock_net(sk), head, port,
160 l3mdev);
161 if (!tb) {
162 spin_unlock(&head->lock);
163 return -ENOMEM;
164 }
165 }
166 inet_csk_update_fastreuse(tb, child);
167 }
168 inet_bind_hash(child, tb, port);
169 spin_unlock(&head->lock);
170
171 return 0;
172}
173EXPORT_SYMBOL_GPL(__inet_inherit_port);
174
175static struct inet_listen_hashbucket *
176inet_lhash2_bucket_sk(struct inet_hashinfo *h, struct sock *sk)
177{
178 u32 hash;
179
180#if IS_ENABLED(CONFIG_IPV6)
181 if (sk->sk_family == AF_INET6)
182 hash = ipv6_portaddr_hash(sock_net(sk),
183 &sk->sk_v6_rcv_saddr,
184 inet_sk(sk)->inet_num);
185 else
186#endif
187 hash = ipv4_portaddr_hash(sock_net(sk),
188 inet_sk(sk)->inet_rcv_saddr,
189 inet_sk(sk)->inet_num);
190 return inet_lhash2_bucket(h, hash);
191}
192
193static void inet_hash2(struct inet_hashinfo *h, struct sock *sk)
194{
195 struct inet_listen_hashbucket *ilb2;
196
197 if (!h->lhash2)
198 return;
199
200 ilb2 = inet_lhash2_bucket_sk(h, sk);
201
202 spin_lock(&ilb2->lock);
203 if (sk->sk_reuseport && sk->sk_family == AF_INET6)
204 hlist_add_tail_rcu(&inet_csk(sk)->icsk_listen_portaddr_node,
205 &ilb2->head);
206 else
207 hlist_add_head_rcu(&inet_csk(sk)->icsk_listen_portaddr_node,
208 &ilb2->head);
209 ilb2->count++;
210 spin_unlock(&ilb2->lock);
211}
212
213static void inet_unhash2(struct inet_hashinfo *h, struct sock *sk)
214{
215 struct inet_listen_hashbucket *ilb2;
216
217 if (!h->lhash2 ||
218 WARN_ON_ONCE(hlist_unhashed(&inet_csk(sk)->icsk_listen_portaddr_node)))
219 return;
220
221 ilb2 = inet_lhash2_bucket_sk(h, sk);
222
223 spin_lock(&ilb2->lock);
224 hlist_del_init_rcu(&inet_csk(sk)->icsk_listen_portaddr_node);
225 ilb2->count--;
226 spin_unlock(&ilb2->lock);
227}
228
229static inline int compute_score(struct sock *sk, struct net *net,
230 const unsigned short hnum, const __be32 daddr,
231 const int dif, const int sdif, bool exact_dif)
232{
233 int score = -1;
234
235 if (net_eq(sock_net(sk), net) && sk->sk_num == hnum &&
236 !ipv6_only_sock(sk)) {
237 if (sk->sk_rcv_saddr != daddr)
238 return -1;
239
240 if (!inet_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif))
241 return -1;
242
243 score = sk->sk_family == PF_INET ? 2 : 1;
244 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
245 score++;
246 }
247 return score;
248}
249
250static inline struct sock *lookup_reuseport(struct net *net, struct sock *sk,
251 struct sk_buff *skb, int doff,
252 __be32 saddr, __be16 sport,
253 __be32 daddr, unsigned short hnum)
254{
255 struct sock *reuse_sk = NULL;
256 u32 phash;
257
258 if (sk->sk_reuseport) {
259 phash = inet_ehashfn(net, daddr, hnum, saddr, sport);
260 reuse_sk = reuseport_select_sock(sk, phash, skb, doff);
261 }
262 return reuse_sk;
263}
264
265
266
267
268
269
270
271
272
273static struct sock *inet_lhash2_lookup(struct net *net,
274 struct inet_listen_hashbucket *ilb2,
275 struct sk_buff *skb, int doff,
276 const __be32 saddr, __be16 sport,
277 const __be32 daddr, const unsigned short hnum,
278 const int dif, const int sdif)
279{
280 bool exact_dif = inet_exact_dif_match(net, skb);
281 struct inet_connection_sock *icsk;
282 struct sock *sk, *result = NULL;
283 int score, hiscore = 0;
284
285 inet_lhash2_for_each_icsk_rcu(icsk, &ilb2->head) {
286 sk = (struct sock *)icsk;
287 score = compute_score(sk, net, hnum, daddr,
288 dif, sdif, exact_dif);
289 if (score > hiscore) {
290 result = lookup_reuseport(net, sk, skb, doff,
291 saddr, sport, daddr, hnum);
292 if (result)
293 return result;
294
295 result = sk;
296 hiscore = score;
297 }
298 }
299
300 return result;
301}
302
303static inline struct sock *inet_lookup_run_bpf(struct net *net,
304 struct inet_hashinfo *hashinfo,
305 struct sk_buff *skb, int doff,
306 __be32 saddr, __be16 sport,
307 __be32 daddr, u16 hnum)
308{
309 struct sock *sk, *reuse_sk;
310 bool no_reuseport;
311
312 if (hashinfo != &tcp_hashinfo)
313 return NULL;
314
315 no_reuseport = bpf_sk_lookup_run_v4(net, IPPROTO_TCP,
316 saddr, sport, daddr, hnum, &sk);
317 if (no_reuseport || IS_ERR_OR_NULL(sk))
318 return sk;
319
320 reuse_sk = lookup_reuseport(net, sk, skb, doff, saddr, sport, daddr, hnum);
321 if (reuse_sk)
322 sk = reuse_sk;
323 return sk;
324}
325
326struct sock *__inet_lookup_listener(struct net *net,
327 struct inet_hashinfo *hashinfo,
328 struct sk_buff *skb, int doff,
329 const __be32 saddr, __be16 sport,
330 const __be32 daddr, const unsigned short hnum,
331 const int dif, const int sdif)
332{
333 struct inet_listen_hashbucket *ilb2;
334 struct sock *result = NULL;
335 unsigned int hash2;
336
337
338 if (static_branch_unlikely(&bpf_sk_lookup_enabled)) {
339 result = inet_lookup_run_bpf(net, hashinfo, skb, doff,
340 saddr, sport, daddr, hnum);
341 if (result)
342 goto done;
343 }
344
345 hash2 = ipv4_portaddr_hash(net, daddr, hnum);
346 ilb2 = inet_lhash2_bucket(hashinfo, hash2);
347
348 result = inet_lhash2_lookup(net, ilb2, skb, doff,
349 saddr, sport, daddr, hnum,
350 dif, sdif);
351 if (result)
352 goto done;
353
354
355 hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
356 ilb2 = inet_lhash2_bucket(hashinfo, hash2);
357
358 result = inet_lhash2_lookup(net, ilb2, skb, doff,
359 saddr, sport, htonl(INADDR_ANY), hnum,
360 dif, sdif);
361done:
362 if (IS_ERR(result))
363 return NULL;
364 return result;
365}
366EXPORT_SYMBOL_GPL(__inet_lookup_listener);
367
368
369void sock_gen_put(struct sock *sk)
370{
371 if (!refcount_dec_and_test(&sk->sk_refcnt))
372 return;
373
374 if (sk->sk_state == TCP_TIME_WAIT)
375 inet_twsk_free(inet_twsk(sk));
376 else if (sk->sk_state == TCP_NEW_SYN_RECV)
377 reqsk_free(inet_reqsk(sk));
378 else
379 sk_free(sk);
380}
381EXPORT_SYMBOL_GPL(sock_gen_put);
382
383void sock_edemux(struct sk_buff *skb)
384{
385 sock_gen_put(skb->sk);
386}
387EXPORT_SYMBOL(sock_edemux);
388
389struct sock *__inet_lookup_established(struct net *net,
390 struct inet_hashinfo *hashinfo,
391 const __be32 saddr, const __be16 sport,
392 const __be32 daddr, const u16 hnum,
393 const int dif, const int sdif)
394{
395 INET_ADDR_COOKIE(acookie, saddr, daddr);
396 const __portpair ports = INET_COMBINED_PORTS(sport, hnum);
397 struct sock *sk;
398 const struct hlist_nulls_node *node;
399
400
401
402 unsigned int hash = inet_ehashfn(net, daddr, hnum, saddr, sport);
403 unsigned int slot = hash & hashinfo->ehash_mask;
404 struct inet_ehash_bucket *head = &hashinfo->ehash[slot];
405
406begin:
407 sk_nulls_for_each_rcu(sk, node, &head->chain) {
408 if (sk->sk_hash != hash)
409 continue;
410 if (likely(INET_MATCH(sk, net, acookie,
411 saddr, daddr, ports, dif, sdif))) {
412 if (unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
413 goto out;
414 if (unlikely(!INET_MATCH(sk, net, acookie,
415 saddr, daddr, ports,
416 dif, sdif))) {
417 sock_gen_put(sk);
418 goto begin;
419 }
420 goto found;
421 }
422 }
423
424
425
426
427
428 if (get_nulls_value(node) != slot)
429 goto begin;
430out:
431 sk = NULL;
432found:
433 return sk;
434}
435EXPORT_SYMBOL_GPL(__inet_lookup_established);
436
437
438static int __inet_check_established(struct inet_timewait_death_row *death_row,
439 struct sock *sk, __u16 lport,
440 struct inet_timewait_sock **twp)
441{
442 struct inet_hashinfo *hinfo = death_row->hashinfo;
443 struct inet_sock *inet = inet_sk(sk);
444 __be32 daddr = inet->inet_rcv_saddr;
445 __be32 saddr = inet->inet_daddr;
446 int dif = sk->sk_bound_dev_if;
447 struct net *net = sock_net(sk);
448 int sdif = l3mdev_master_ifindex_by_index(net, dif);
449 INET_ADDR_COOKIE(acookie, saddr, daddr);
450 const __portpair ports = INET_COMBINED_PORTS(inet->inet_dport, lport);
451 unsigned int hash = inet_ehashfn(net, daddr, lport,
452 saddr, inet->inet_dport);
453 struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash);
454 spinlock_t *lock = inet_ehash_lockp(hinfo, hash);
455 struct sock *sk2;
456 const struct hlist_nulls_node *node;
457 struct inet_timewait_sock *tw = NULL;
458
459 spin_lock(lock);
460
461 sk_nulls_for_each(sk2, node, &head->chain) {
462 if (sk2->sk_hash != hash)
463 continue;
464
465 if (likely(INET_MATCH(sk2, net, acookie,
466 saddr, daddr, ports, dif, sdif))) {
467 if (sk2->sk_state == TCP_TIME_WAIT) {
468 tw = inet_twsk(sk2);
469 if (twsk_unique(sk, sk2, twp))
470 break;
471 }
472 goto not_unique;
473 }
474 }
475
476
477
478
479 inet->inet_num = lport;
480 inet->inet_sport = htons(lport);
481 sk->sk_hash = hash;
482 WARN_ON(!sk_unhashed(sk));
483 __sk_nulls_add_node_rcu(sk, &head->chain);
484 if (tw) {
485 sk_nulls_del_node_init_rcu((struct sock *)tw);
486 __NET_INC_STATS(net, LINUX_MIB_TIMEWAITRECYCLED);
487 }
488 spin_unlock(lock);
489 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
490
491 if (twp) {
492 *twp = tw;
493 } else if (tw) {
494
495 inet_twsk_deschedule_put(tw);
496 }
497 return 0;
498
499not_unique:
500 spin_unlock(lock);
501 return -EADDRNOTAVAIL;
502}
503
504static u32 inet_sk_port_offset(const struct sock *sk)
505{
506 const struct inet_sock *inet = inet_sk(sk);
507
508 return secure_ipv4_port_ephemeral(inet->inet_rcv_saddr,
509 inet->inet_daddr,
510 inet->inet_dport);
511}
512
513
514
515
516bool inet_ehash_insert(struct sock *sk, struct sock *osk)
517{
518 struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
519 struct hlist_nulls_head *list;
520 struct inet_ehash_bucket *head;
521 spinlock_t *lock;
522 bool ret = true;
523
524 WARN_ON_ONCE(!sk_unhashed(sk));
525
526 sk->sk_hash = sk_ehashfn(sk);
527 head = inet_ehash_bucket(hashinfo, sk->sk_hash);
528 list = &head->chain;
529 lock = inet_ehash_lockp(hashinfo, sk->sk_hash);
530
531 spin_lock(lock);
532 if (osk) {
533 WARN_ON_ONCE(sk->sk_hash != osk->sk_hash);
534 ret = sk_nulls_del_node_init_rcu(osk);
535 }
536 if (ret)
537 __sk_nulls_add_node_rcu(sk, list);
538 spin_unlock(lock);
539 return ret;
540}
541
542bool inet_ehash_nolisten(struct sock *sk, struct sock *osk)
543{
544 bool ok = inet_ehash_insert(sk, osk);
545
546 if (ok) {
547 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
548 } else {
549 percpu_counter_inc(sk->sk_prot->orphan_count);
550 inet_sk_set_state(sk, TCP_CLOSE);
551 sock_set_flag(sk, SOCK_DEAD);
552 inet_csk_destroy_sock(sk);
553 }
554 return ok;
555}
556EXPORT_SYMBOL_GPL(inet_ehash_nolisten);
557
558static int inet_reuseport_add_sock(struct sock *sk,
559 struct inet_listen_hashbucket *ilb)
560{
561 struct inet_bind_bucket *tb = inet_csk(sk)->icsk_bind_hash;
562 const struct hlist_nulls_node *node;
563 struct sock *sk2;
564 kuid_t uid = sock_i_uid(sk);
565
566 sk_nulls_for_each_rcu(sk2, node, &ilb->nulls_head) {
567 if (sk2 != sk &&
568 sk2->sk_family == sk->sk_family &&
569 ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
570 sk2->sk_bound_dev_if == sk->sk_bound_dev_if &&
571 inet_csk(sk2)->icsk_bind_hash == tb &&
572 sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
573 inet_rcv_saddr_equal(sk, sk2, false))
574 return reuseport_add_sock(sk, sk2,
575 inet_rcv_saddr_any(sk));
576 }
577
578 return reuseport_alloc(sk, inet_rcv_saddr_any(sk));
579}
580
581int __inet_hash(struct sock *sk, struct sock *osk)
582{
583 struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
584 struct inet_listen_hashbucket *ilb;
585 int err = 0;
586
587 if (sk->sk_state != TCP_LISTEN) {
588 inet_ehash_nolisten(sk, osk);
589 return 0;
590 }
591 WARN_ON(!sk_unhashed(sk));
592 ilb = &hashinfo->listening_hash[inet_sk_listen_hashfn(sk)];
593
594 spin_lock(&ilb->lock);
595 if (sk->sk_reuseport) {
596 err = inet_reuseport_add_sock(sk, ilb);
597 if (err)
598 goto unlock;
599 }
600 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
601 sk->sk_family == AF_INET6)
602 __sk_nulls_add_node_tail_rcu(sk, &ilb->nulls_head);
603 else
604 __sk_nulls_add_node_rcu(sk, &ilb->nulls_head);
605 inet_hash2(hashinfo, sk);
606 ilb->count++;
607 sock_set_flag(sk, SOCK_RCU_FREE);
608 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
609unlock:
610 spin_unlock(&ilb->lock);
611
612 return err;
613}
614EXPORT_SYMBOL(__inet_hash);
615
616int inet_hash(struct sock *sk)
617{
618 int err = 0;
619
620 if (sk->sk_state != TCP_CLOSE) {
621 local_bh_disable();
622 err = __inet_hash(sk, NULL);
623 local_bh_enable();
624 }
625
626 return err;
627}
628EXPORT_SYMBOL_GPL(inet_hash);
629
630void inet_unhash(struct sock *sk)
631{
632 struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
633 struct inet_listen_hashbucket *ilb = NULL;
634 spinlock_t *lock;
635
636 if (sk_unhashed(sk))
637 return;
638
639 if (sk->sk_state == TCP_LISTEN) {
640 ilb = &hashinfo->listening_hash[inet_sk_listen_hashfn(sk)];
641 lock = &ilb->lock;
642 } else {
643 lock = inet_ehash_lockp(hashinfo, sk->sk_hash);
644 }
645 spin_lock_bh(lock);
646 if (sk_unhashed(sk))
647 goto unlock;
648
649 if (rcu_access_pointer(sk->sk_reuseport_cb))
650 reuseport_detach_sock(sk);
651 if (ilb) {
652 inet_unhash2(hashinfo, sk);
653 ilb->count--;
654 }
655 __sk_nulls_del_node_init_rcu(sk);
656 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
657unlock:
658 spin_unlock_bh(lock);
659}
660EXPORT_SYMBOL_GPL(inet_unhash);
661
662int __inet_hash_connect(struct inet_timewait_death_row *death_row,
663 struct sock *sk, u32 port_offset,
664 int (*check_established)(struct inet_timewait_death_row *,
665 struct sock *, __u16, struct inet_timewait_sock **))
666{
667 struct inet_hashinfo *hinfo = death_row->hashinfo;
668 struct inet_timewait_sock *tw = NULL;
669 struct inet_bind_hashbucket *head;
670 int port = inet_sk(sk)->inet_num;
671 struct net *net = sock_net(sk);
672 struct inet_bind_bucket *tb;
673 u32 remaining, offset;
674 int ret, i, low, high;
675 static u32 hint;
676 int l3mdev;
677
678 if (port) {
679 head = &hinfo->bhash[inet_bhashfn(net, port,
680 hinfo->bhash_size)];
681 tb = inet_csk(sk)->icsk_bind_hash;
682 spin_lock_bh(&head->lock);
683 if (sk_head(&tb->owners) == sk && !sk->sk_bind_node.next) {
684 inet_ehash_nolisten(sk, NULL);
685 spin_unlock_bh(&head->lock);
686 return 0;
687 }
688 spin_unlock(&head->lock);
689
690 ret = check_established(death_row, sk, port, NULL);
691 local_bh_enable();
692 return ret;
693 }
694
695 l3mdev = inet_sk_bound_l3mdev(sk);
696
697 inet_get_local_port_range(net, &low, &high);
698 high++;
699 remaining = high - low;
700 if (likely(remaining > 1))
701 remaining &= ~1U;
702
703 offset = (hint + port_offset) % remaining;
704
705
706
707 offset &= ~1U;
708other_parity_scan:
709 port = low + offset;
710 for (i = 0; i < remaining; i += 2, port += 2) {
711 if (unlikely(port >= high))
712 port -= remaining;
713 if (inet_is_local_reserved_port(net, port))
714 continue;
715 head = &hinfo->bhash[inet_bhashfn(net, port,
716 hinfo->bhash_size)];
717 spin_lock_bh(&head->lock);
718
719
720
721
722 inet_bind_bucket_for_each(tb, &head->chain) {
723 if (net_eq(ib_net(tb), net) && tb->l3mdev == l3mdev &&
724 tb->port == port) {
725 if (tb->fastreuse >= 0 ||
726 tb->fastreuseport >= 0)
727 goto next_port;
728 WARN_ON(hlist_empty(&tb->owners));
729 if (!check_established(death_row, sk,
730 port, &tw))
731 goto ok;
732 goto next_port;
733 }
734 }
735
736 tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep,
737 net, head, port, l3mdev);
738 if (!tb) {
739 spin_unlock_bh(&head->lock);
740 return -ENOMEM;
741 }
742 tb->fastreuse = -1;
743 tb->fastreuseport = -1;
744 goto ok;
745next_port:
746 spin_unlock_bh(&head->lock);
747 cond_resched();
748 }
749
750 offset++;
751 if ((offset & 1) && remaining > 1)
752 goto other_parity_scan;
753
754 return -EADDRNOTAVAIL;
755
756ok:
757 hint += i + 2;
758
759
760 inet_bind_hash(sk, tb, port);
761 if (sk_unhashed(sk)) {
762 inet_sk(sk)->inet_sport = htons(port);
763 inet_ehash_nolisten(sk, (struct sock *)tw);
764 }
765 if (tw)
766 inet_twsk_bind_unhash(tw, hinfo);
767 spin_unlock(&head->lock);
768 if (tw)
769 inet_twsk_deschedule_put(tw);
770 local_bh_enable();
771 return 0;
772}
773
774
775
776
777int inet_hash_connect(struct inet_timewait_death_row *death_row,
778 struct sock *sk)
779{
780 u32 port_offset = 0;
781
782 if (!inet_sk(sk)->inet_num)
783 port_offset = inet_sk_port_offset(sk);
784 return __inet_hash_connect(death_row, sk, port_offset,
785 __inet_check_established);
786}
787EXPORT_SYMBOL_GPL(inet_hash_connect);
788
789void inet_hashinfo_init(struct inet_hashinfo *h)
790{
791 int i;
792
793 for (i = 0; i < INET_LHTABLE_SIZE; i++) {
794 spin_lock_init(&h->listening_hash[i].lock);
795 INIT_HLIST_NULLS_HEAD(&h->listening_hash[i].nulls_head,
796 i + LISTENING_NULLS_BASE);
797 h->listening_hash[i].count = 0;
798 }
799
800 h->lhash2 = NULL;
801}
802EXPORT_SYMBOL_GPL(inet_hashinfo_init);
803
804static void init_hashinfo_lhash2(struct inet_hashinfo *h)
805{
806 int i;
807
808 for (i = 0; i <= h->lhash2_mask; i++) {
809 spin_lock_init(&h->lhash2[i].lock);
810 INIT_HLIST_HEAD(&h->lhash2[i].head);
811 h->lhash2[i].count = 0;
812 }
813}
814
815void __init inet_hashinfo2_init(struct inet_hashinfo *h, const char *name,
816 unsigned long numentries, int scale,
817 unsigned long low_limit,
818 unsigned long high_limit)
819{
820 h->lhash2 = alloc_large_system_hash(name,
821 sizeof(*h->lhash2),
822 numentries,
823 scale,
824 0,
825 NULL,
826 &h->lhash2_mask,
827 low_limit,
828 high_limit);
829 init_hashinfo_lhash2(h);
830}
831
832int inet_hashinfo2_init_mod(struct inet_hashinfo *h)
833{
834 h->lhash2 = kmalloc_array(INET_LHTABLE_SIZE, sizeof(*h->lhash2), GFP_KERNEL);
835 if (!h->lhash2)
836 return -ENOMEM;
837
838 h->lhash2_mask = INET_LHTABLE_SIZE - 1;
839
840 BUG_ON(INET_LHTABLE_SIZE & h->lhash2_mask);
841
842 init_hashinfo_lhash2(h);
843 return 0;
844}
845EXPORT_SYMBOL_GPL(inet_hashinfo2_init_mod);
846
847int inet_ehash_locks_alloc(struct inet_hashinfo *hashinfo)
848{
849 unsigned int locksz = sizeof(spinlock_t);
850 unsigned int i, nblocks = 1;
851
852 if (locksz != 0) {
853
854 nblocks = max(2U * L1_CACHE_BYTES / locksz, 1U);
855 nblocks = roundup_pow_of_two(nblocks * num_possible_cpus());
856
857
858 nblocks = min(nblocks, hashinfo->ehash_mask + 1);
859
860 hashinfo->ehash_locks = kvmalloc_array(nblocks, locksz, GFP_KERNEL);
861 if (!hashinfo->ehash_locks)
862 return -ENOMEM;
863
864 for (i = 0; i < nblocks; i++)
865 spin_lock_init(&hashinfo->ehash_locks[i]);
866 }
867 hashinfo->ehash_locks_mask = nblocks - 1;
868 return 0;
869}
870EXPORT_SYMBOL_GPL(inet_ehash_locks_alloc);
871