Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
6 *
7 * Generic INET transport hashtables
8 *
9 * Authors: Lotsa people, from code originally in tcp
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#if IS_ENABLED(CONFIG_IPV6)
24#include <net/inet6_hashtables.h>
25#endif
26#include <net/secure_seq.h>
27#include <net/hotdata.h>
28#include <net/ip.h>
29#include <net/tcp.h>
30#include <net/sock_reuseport.h>
31
32u32 inet_ehashfn(const struct net *net, const __be32 laddr,
33 const __u16 lport, const __be32 faddr,
34 const __be16 fport)
35{
36 net_get_random_once(&inet_ehash_secret, sizeof(inet_ehash_secret));
37
38 return __inet_ehashfn(laddr, lport, faddr, fport,
39 inet_ehash_secret + net_hash_mix(net));
40}
41EXPORT_SYMBOL_GPL(inet_ehashfn);
42
43/* This function handles inet_sock, but also timewait and request sockets
44 * for IPv4/IPv6.
45 */
46static u32 sk_ehashfn(const struct sock *sk)
47{
48#if IS_ENABLED(CONFIG_IPV6)
49 if (sk->sk_family == AF_INET6 &&
50 !ipv6_addr_v4mapped(&sk->sk_v6_daddr))
51 return inet6_ehashfn(sock_net(sk),
52 &sk->sk_v6_rcv_saddr, sk->sk_num,
53 &sk->sk_v6_daddr, sk->sk_dport);
54#endif
55 return inet_ehashfn(sock_net(sk),
56 sk->sk_rcv_saddr, sk->sk_num,
57 sk->sk_daddr, sk->sk_dport);
58}
59
60/*
61 * Allocate and initialize a new local port bind bucket.
62 * The bindhash mutex for snum's hash chain must be held here.
63 */
64struct inet_bind_bucket *inet_bind_bucket_create(struct kmem_cache *cachep,
65 struct net *net,
66 struct inet_bind_hashbucket *head,
67 const unsigned short snum,
68 int l3mdev)
69{
70 struct inet_bind_bucket *tb = kmem_cache_alloc(cachep, GFP_ATOMIC);
71
72 if (tb) {
73 write_pnet(&tb->ib_net, net);
74 tb->l3mdev = l3mdev;
75 tb->port = snum;
76 tb->fastreuse = 0;
77 tb->fastreuseport = 0;
78 INIT_HLIST_HEAD(&tb->bhash2);
79 hlist_add_head(&tb->node, &head->chain);
80 }
81 return tb;
82}
83
84/*
85 * Caller must hold hashbucket lock for this tb with local BH disabled
86 */
87void inet_bind_bucket_destroy(struct kmem_cache *cachep, struct inet_bind_bucket *tb)
88{
89 if (hlist_empty(&tb->bhash2)) {
90 __hlist_del(&tb->node);
91 kmem_cache_free(cachep, tb);
92 }
93}
94
95bool inet_bind_bucket_match(const struct inet_bind_bucket *tb, const struct net *net,
96 unsigned short port, int l3mdev)
97{
98 return net_eq(ib_net(tb), net) && tb->port == port &&
99 tb->l3mdev == l3mdev;
100}
101
102static void inet_bind2_bucket_init(struct inet_bind2_bucket *tb2,
103 struct net *net,
104 struct inet_bind_hashbucket *head,
105 struct inet_bind_bucket *tb,
106 const struct sock *sk)
107{
108 write_pnet(&tb2->ib_net, net);
109 tb2->l3mdev = tb->l3mdev;
110 tb2->port = tb->port;
111#if IS_ENABLED(CONFIG_IPV6)
112 BUILD_BUG_ON(USHRT_MAX < (IPV6_ADDR_ANY | IPV6_ADDR_MAPPED));
113 if (sk->sk_family == AF_INET6) {
114 tb2->addr_type = ipv6_addr_type(&sk->sk_v6_rcv_saddr);
115 tb2->v6_rcv_saddr = sk->sk_v6_rcv_saddr;
116 } else {
117 tb2->addr_type = IPV6_ADDR_MAPPED;
118 ipv6_addr_set_v4mapped(sk->sk_rcv_saddr, &tb2->v6_rcv_saddr);
119 }
120#else
121 tb2->rcv_saddr = sk->sk_rcv_saddr;
122#endif
123 INIT_HLIST_HEAD(&tb2->owners);
124 hlist_add_head(&tb2->node, &head->chain);
125 hlist_add_head(&tb2->bhash_node, &tb->bhash2);
126}
127
128struct inet_bind2_bucket *inet_bind2_bucket_create(struct kmem_cache *cachep,
129 struct net *net,
130 struct inet_bind_hashbucket *head,
131 struct inet_bind_bucket *tb,
132 const struct sock *sk)
133{
134 struct inet_bind2_bucket *tb2 = kmem_cache_alloc(cachep, GFP_ATOMIC);
135
136 if (tb2)
137 inet_bind2_bucket_init(tb2, net, head, tb, sk);
138
139 return tb2;
140}
141
142/* Caller must hold hashbucket lock for this tb with local BH disabled */
143void inet_bind2_bucket_destroy(struct kmem_cache *cachep, struct inet_bind2_bucket *tb)
144{
145 if (hlist_empty(&tb->owners)) {
146 __hlist_del(&tb->node);
147 __hlist_del(&tb->bhash_node);
148 kmem_cache_free(cachep, tb);
149 }
150}
151
152static bool inet_bind2_bucket_addr_match(const struct inet_bind2_bucket *tb2,
153 const struct sock *sk)
154{
155#if IS_ENABLED(CONFIG_IPV6)
156 if (sk->sk_family == AF_INET6)
157 return ipv6_addr_equal(&tb2->v6_rcv_saddr, &sk->sk_v6_rcv_saddr);
158
159 if (tb2->addr_type != IPV6_ADDR_MAPPED)
160 return false;
161#endif
162 return tb2->rcv_saddr == sk->sk_rcv_saddr;
163}
164
165void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
166 struct inet_bind2_bucket *tb2, unsigned short port)
167{
168 inet_sk(sk)->inet_num = port;
169 inet_csk(sk)->icsk_bind_hash = tb;
170 inet_csk(sk)->icsk_bind2_hash = tb2;
171 sk_add_bind_node(sk, &tb2->owners);
172}
173
174/*
175 * Get rid of any references to a local port held by the given sock.
176 */
177static void __inet_put_port(struct sock *sk)
178{
179 struct inet_hashinfo *hashinfo = tcp_or_dccp_get_hashinfo(sk);
180 struct inet_bind_hashbucket *head, *head2;
181 struct net *net = sock_net(sk);
182 struct inet_bind_bucket *tb;
183 int bhash;
184
185 bhash = inet_bhashfn(net, inet_sk(sk)->inet_num, hashinfo->bhash_size);
186 head = &hashinfo->bhash[bhash];
187 head2 = inet_bhashfn_portaddr(hashinfo, sk, net, inet_sk(sk)->inet_num);
188
189 spin_lock(&head->lock);
190 tb = inet_csk(sk)->icsk_bind_hash;
191 inet_csk(sk)->icsk_bind_hash = NULL;
192 inet_sk(sk)->inet_num = 0;
193
194 spin_lock(&head2->lock);
195 if (inet_csk(sk)->icsk_bind2_hash) {
196 struct inet_bind2_bucket *tb2 = inet_csk(sk)->icsk_bind2_hash;
197
198 __sk_del_bind_node(sk);
199 inet_csk(sk)->icsk_bind2_hash = NULL;
200 inet_bind2_bucket_destroy(hashinfo->bind2_bucket_cachep, tb2);
201 }
202 spin_unlock(&head2->lock);
203
204 inet_bind_bucket_destroy(hashinfo->bind_bucket_cachep, tb);
205 spin_unlock(&head->lock);
206}
207
208void inet_put_port(struct sock *sk)
209{
210 local_bh_disable();
211 __inet_put_port(sk);
212 local_bh_enable();
213}
214EXPORT_SYMBOL(inet_put_port);
215
216int __inet_inherit_port(const struct sock *sk, struct sock *child)
217{
218 struct inet_hashinfo *table = tcp_or_dccp_get_hashinfo(sk);
219 unsigned short port = inet_sk(child)->inet_num;
220 struct inet_bind_hashbucket *head, *head2;
221 bool created_inet_bind_bucket = false;
222 struct net *net = sock_net(sk);
223 bool update_fastreuse = false;
224 struct inet_bind2_bucket *tb2;
225 struct inet_bind_bucket *tb;
226 int bhash, l3mdev;
227
228 bhash = inet_bhashfn(net, port, table->bhash_size);
229 head = &table->bhash[bhash];
230 head2 = inet_bhashfn_portaddr(table, child, net, port);
231
232 spin_lock(&head->lock);
233 spin_lock(&head2->lock);
234 tb = inet_csk(sk)->icsk_bind_hash;
235 tb2 = inet_csk(sk)->icsk_bind2_hash;
236 if (unlikely(!tb || !tb2)) {
237 spin_unlock(&head2->lock);
238 spin_unlock(&head->lock);
239 return -ENOENT;
240 }
241 if (tb->port != port) {
242 l3mdev = inet_sk_bound_l3mdev(sk);
243
244 /* NOTE: using tproxy and redirecting skbs to a proxy
245 * on a different listener port breaks the assumption
246 * that the listener socket's icsk_bind_hash is the same
247 * as that of the child socket. We have to look up or
248 * create a new bind bucket for the child here. */
249 inet_bind_bucket_for_each(tb, &head->chain) {
250 if (inet_bind_bucket_match(tb, net, port, l3mdev))
251 break;
252 }
253 if (!tb) {
254 tb = inet_bind_bucket_create(table->bind_bucket_cachep,
255 net, head, port, l3mdev);
256 if (!tb) {
257 spin_unlock(&head2->lock);
258 spin_unlock(&head->lock);
259 return -ENOMEM;
260 }
261 created_inet_bind_bucket = true;
262 }
263 update_fastreuse = true;
264
265 goto bhash2_find;
266 } else if (!inet_bind2_bucket_addr_match(tb2, child)) {
267 l3mdev = inet_sk_bound_l3mdev(sk);
268
269bhash2_find:
270 tb2 = inet_bind2_bucket_find(head2, net, port, l3mdev, child);
271 if (!tb2) {
272 tb2 = inet_bind2_bucket_create(table->bind2_bucket_cachep,
273 net, head2, tb, child);
274 if (!tb2)
275 goto error;
276 }
277 }
278 if (update_fastreuse)
279 inet_csk_update_fastreuse(tb, child);
280 inet_bind_hash(child, tb, tb2, port);
281 spin_unlock(&head2->lock);
282 spin_unlock(&head->lock);
283
284 return 0;
285
286error:
287 if (created_inet_bind_bucket)
288 inet_bind_bucket_destroy(table->bind_bucket_cachep, tb);
289 spin_unlock(&head2->lock);
290 spin_unlock(&head->lock);
291 return -ENOMEM;
292}
293EXPORT_SYMBOL_GPL(__inet_inherit_port);
294
295static struct inet_listen_hashbucket *
296inet_lhash2_bucket_sk(struct inet_hashinfo *h, struct sock *sk)
297{
298 u32 hash;
299
300#if IS_ENABLED(CONFIG_IPV6)
301 if (sk->sk_family == AF_INET6)
302 hash = ipv6_portaddr_hash(sock_net(sk),
303 &sk->sk_v6_rcv_saddr,
304 inet_sk(sk)->inet_num);
305 else
306#endif
307 hash = ipv4_portaddr_hash(sock_net(sk),
308 inet_sk(sk)->inet_rcv_saddr,
309 inet_sk(sk)->inet_num);
310 return inet_lhash2_bucket(h, hash);
311}
312
313static inline int compute_score(struct sock *sk, const struct net *net,
314 const unsigned short hnum, const __be32 daddr,
315 const int dif, const int sdif)
316{
317 int score = -1;
318
319 if (net_eq(sock_net(sk), net) && sk->sk_num == hnum &&
320 !ipv6_only_sock(sk)) {
321 if (sk->sk_rcv_saddr != daddr)
322 return -1;
323
324 if (!inet_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif))
325 return -1;
326 score = sk->sk_bound_dev_if ? 2 : 1;
327
328 if (sk->sk_family == PF_INET)
329 score++;
330 if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
331 score++;
332 }
333 return score;
334}
335
336/**
337 * inet_lookup_reuseport() - execute reuseport logic on AF_INET socket if necessary.
338 * @net: network namespace.
339 * @sk: AF_INET socket, must be in TCP_LISTEN state for TCP or TCP_CLOSE for UDP.
340 * @skb: context for a potential SK_REUSEPORT program.
341 * @doff: header offset.
342 * @saddr: source address.
343 * @sport: source port.
344 * @daddr: destination address.
345 * @hnum: destination port in host byte order.
346 * @ehashfn: hash function used to generate the fallback hash.
347 *
348 * Return: NULL if sk doesn't have SO_REUSEPORT set, otherwise a pointer to
349 * the selected sock or an error.
350 */
351struct sock *inet_lookup_reuseport(const struct net *net, struct sock *sk,
352 struct sk_buff *skb, int doff,
353 __be32 saddr, __be16 sport,
354 __be32 daddr, unsigned short hnum,
355 inet_ehashfn_t *ehashfn)
356{
357 struct sock *reuse_sk = NULL;
358 u32 phash;
359
360 if (sk->sk_reuseport) {
361 phash = INDIRECT_CALL_2(ehashfn, udp_ehashfn, inet_ehashfn,
362 net, daddr, hnum, saddr, sport);
363 reuse_sk = reuseport_select_sock(sk, phash, skb, doff);
364 }
365 return reuse_sk;
366}
367EXPORT_SYMBOL_GPL(inet_lookup_reuseport);
368
369/*
370 * Here are some nice properties to exploit here. The BSD API
371 * does not allow a listening sock to specify the remote port nor the
372 * remote address for the connection. So always assume those are both
373 * wildcarded during the search since they can never be otherwise.
374 */
375
376/* called with rcu_read_lock() : No refcount taken on the socket */
377static struct sock *inet_lhash2_lookup(const struct net *net,
378 struct inet_listen_hashbucket *ilb2,
379 struct sk_buff *skb, int doff,
380 const __be32 saddr, __be16 sport,
381 const __be32 daddr, const unsigned short hnum,
382 const int dif, const int sdif)
383{
384 struct sock *sk, *result = NULL;
385 struct hlist_nulls_node *node;
386 int score, hiscore = 0;
387
388 sk_nulls_for_each_rcu(sk, node, &ilb2->nulls_head) {
389 score = compute_score(sk, net, hnum, daddr, dif, sdif);
390 if (score > hiscore) {
391 result = inet_lookup_reuseport(net, sk, skb, doff,
392 saddr, sport, daddr, hnum, inet_ehashfn);
393 if (result)
394 return result;
395
396 result = sk;
397 hiscore = score;
398 }
399 }
400
401 return result;
402}
403
404struct sock *inet_lookup_run_sk_lookup(const struct net *net,
405 int protocol,
406 struct sk_buff *skb, int doff,
407 __be32 saddr, __be16 sport,
408 __be32 daddr, u16 hnum, const int dif,
409 inet_ehashfn_t *ehashfn)
410{
411 struct sock *sk, *reuse_sk;
412 bool no_reuseport;
413
414 no_reuseport = bpf_sk_lookup_run_v4(net, protocol, saddr, sport,
415 daddr, hnum, dif, &sk);
416 if (no_reuseport || IS_ERR_OR_NULL(sk))
417 return sk;
418
419 reuse_sk = inet_lookup_reuseport(net, sk, skb, doff, saddr, sport, daddr, hnum,
420 ehashfn);
421 if (reuse_sk)
422 sk = reuse_sk;
423 return sk;
424}
425
426struct sock *__inet_lookup_listener(const struct net *net,
427 struct inet_hashinfo *hashinfo,
428 struct sk_buff *skb, int doff,
429 const __be32 saddr, __be16 sport,
430 const __be32 daddr, const unsigned short hnum,
431 const int dif, const int sdif)
432{
433 struct inet_listen_hashbucket *ilb2;
434 struct sock *result = NULL;
435 unsigned int hash2;
436
437 /* Lookup redirect from BPF */
438 if (static_branch_unlikely(&bpf_sk_lookup_enabled) &&
439 hashinfo == net->ipv4.tcp_death_row.hashinfo) {
440 result = inet_lookup_run_sk_lookup(net, IPPROTO_TCP, skb, doff,
441 saddr, sport, daddr, hnum, dif,
442 inet_ehashfn);
443 if (result)
444 goto done;
445 }
446
447 hash2 = ipv4_portaddr_hash(net, daddr, hnum);
448 ilb2 = inet_lhash2_bucket(hashinfo, hash2);
449
450 result = inet_lhash2_lookup(net, ilb2, skb, doff,
451 saddr, sport, daddr, hnum,
452 dif, sdif);
453 if (result)
454 goto done;
455
456 /* Lookup lhash2 with INADDR_ANY */
457 hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
458 ilb2 = inet_lhash2_bucket(hashinfo, hash2);
459
460 result = inet_lhash2_lookup(net, ilb2, skb, doff,
461 saddr, sport, htonl(INADDR_ANY), hnum,
462 dif, sdif);
463done:
464 if (IS_ERR(result))
465 return NULL;
466 return result;
467}
468EXPORT_SYMBOL_GPL(__inet_lookup_listener);
469
470/* All sockets share common refcount, but have different destructors */
471void sock_gen_put(struct sock *sk)
472{
473 if (!refcount_dec_and_test(&sk->sk_refcnt))
474 return;
475
476 if (sk->sk_state == TCP_TIME_WAIT)
477 inet_twsk_free(inet_twsk(sk));
478 else if (sk->sk_state == TCP_NEW_SYN_RECV)
479 reqsk_free(inet_reqsk(sk));
480 else
481 sk_free(sk);
482}
483EXPORT_SYMBOL_GPL(sock_gen_put);
484
485void sock_edemux(struct sk_buff *skb)
486{
487 sock_gen_put(skb->sk);
488}
489EXPORT_SYMBOL(sock_edemux);
490
491struct sock *__inet_lookup_established(const struct net *net,
492 struct inet_hashinfo *hashinfo,
493 const __be32 saddr, const __be16 sport,
494 const __be32 daddr, const u16 hnum,
495 const int dif, const int sdif)
496{
497 INET_ADDR_COOKIE(acookie, saddr, daddr);
498 const __portpair ports = INET_COMBINED_PORTS(sport, hnum);
499 struct sock *sk;
500 const struct hlist_nulls_node *node;
501 /* Optimize here for direct hit, only listening connections can
502 * have wildcards anyways.
503 */
504 unsigned int hash = inet_ehashfn(net, daddr, hnum, saddr, sport);
505 unsigned int slot = hash & hashinfo->ehash_mask;
506 struct inet_ehash_bucket *head = &hashinfo->ehash[slot];
507
508begin:
509 sk_nulls_for_each_rcu(sk, node, &head->chain) {
510 if (sk->sk_hash != hash)
511 continue;
512 if (likely(inet_match(net, sk, acookie, ports, dif, sdif))) {
513 if (unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
514 goto out;
515 if (unlikely(!inet_match(net, sk, acookie,
516 ports, dif, sdif))) {
517 sock_gen_put(sk);
518 goto begin;
519 }
520 goto found;
521 }
522 }
523 /*
524 * if the nulls value we got at the end of this lookup is
525 * not the expected one, we must restart lookup.
526 * We probably met an item that was moved to another chain.
527 */
528 if (get_nulls_value(node) != slot)
529 goto begin;
530out:
531 sk = NULL;
532found:
533 return sk;
534}
535EXPORT_SYMBOL_GPL(__inet_lookup_established);
536
537/* called with local bh disabled */
538static int __inet_check_established(struct inet_timewait_death_row *death_row,
539 struct sock *sk, __u16 lport,
540 struct inet_timewait_sock **twp)
541{
542 struct inet_hashinfo *hinfo = death_row->hashinfo;
543 struct inet_sock *inet = inet_sk(sk);
544 __be32 daddr = inet->inet_rcv_saddr;
545 __be32 saddr = inet->inet_daddr;
546 int dif = sk->sk_bound_dev_if;
547 struct net *net = sock_net(sk);
548 int sdif = l3mdev_master_ifindex_by_index(net, dif);
549 INET_ADDR_COOKIE(acookie, saddr, daddr);
550 const __portpair ports = INET_COMBINED_PORTS(inet->inet_dport, lport);
551 unsigned int hash = inet_ehashfn(net, daddr, lport,
552 saddr, inet->inet_dport);
553 struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash);
554 spinlock_t *lock = inet_ehash_lockp(hinfo, hash);
555 struct sock *sk2;
556 const struct hlist_nulls_node *node;
557 struct inet_timewait_sock *tw = NULL;
558
559 spin_lock(lock);
560
561 sk_nulls_for_each(sk2, node, &head->chain) {
562 if (sk2->sk_hash != hash)
563 continue;
564
565 if (likely(inet_match(net, sk2, acookie, ports, dif, sdif))) {
566 if (sk2->sk_state == TCP_TIME_WAIT) {
567 tw = inet_twsk(sk2);
568 if (sk->sk_protocol == IPPROTO_TCP &&
569 tcp_twsk_unique(sk, sk2, twp))
570 break;
571 }
572 goto not_unique;
573 }
574 }
575
576 /* Must record num and sport now. Otherwise we will see
577 * in hash table socket with a funny identity.
578 */
579 inet->inet_num = lport;
580 inet->inet_sport = htons(lport);
581 sk->sk_hash = hash;
582 WARN_ON(!sk_unhashed(sk));
583 __sk_nulls_add_node_rcu(sk, &head->chain);
584 if (tw) {
585 sk_nulls_del_node_init_rcu((struct sock *)tw);
586 __NET_INC_STATS(net, LINUX_MIB_TIMEWAITRECYCLED);
587 }
588 spin_unlock(lock);
589 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
590
591 if (twp) {
592 *twp = tw;
593 } else if (tw) {
594 /* Silly. Should hash-dance instead... */
595 inet_twsk_deschedule_put(tw);
596 }
597 return 0;
598
599not_unique:
600 spin_unlock(lock);
601 return -EADDRNOTAVAIL;
602}
603
604static u64 inet_sk_port_offset(const struct sock *sk)
605{
606 const struct inet_sock *inet = inet_sk(sk);
607
608 return secure_ipv4_port_ephemeral(inet->inet_rcv_saddr,
609 inet->inet_daddr,
610 inet->inet_dport);
611}
612
613/* Searches for an exsiting socket in the ehash bucket list.
614 * Returns true if found, false otherwise.
615 */
616static bool inet_ehash_lookup_by_sk(struct sock *sk,
617 struct hlist_nulls_head *list)
618{
619 const __portpair ports = INET_COMBINED_PORTS(sk->sk_dport, sk->sk_num);
620 const int sdif = sk->sk_bound_dev_if;
621 const int dif = sk->sk_bound_dev_if;
622 const struct hlist_nulls_node *node;
623 struct net *net = sock_net(sk);
624 struct sock *esk;
625
626 INET_ADDR_COOKIE(acookie, sk->sk_daddr, sk->sk_rcv_saddr);
627
628 sk_nulls_for_each_rcu(esk, node, list) {
629 if (esk->sk_hash != sk->sk_hash)
630 continue;
631 if (sk->sk_family == AF_INET) {
632 if (unlikely(inet_match(net, esk, acookie,
633 ports, dif, sdif))) {
634 return true;
635 }
636 }
637#if IS_ENABLED(CONFIG_IPV6)
638 else if (sk->sk_family == AF_INET6) {
639 if (unlikely(inet6_match(net, esk,
640 &sk->sk_v6_daddr,
641 &sk->sk_v6_rcv_saddr,
642 ports, dif, sdif))) {
643 return true;
644 }
645 }
646#endif
647 }
648 return false;
649}
650
651/* Insert a socket into ehash, and eventually remove another one
652 * (The another one can be a SYN_RECV or TIMEWAIT)
653 * If an existing socket already exists, socket sk is not inserted,
654 * and sets found_dup_sk parameter to true.
655 */
656bool inet_ehash_insert(struct sock *sk, struct sock *osk, bool *found_dup_sk)
657{
658 struct inet_hashinfo *hashinfo = tcp_or_dccp_get_hashinfo(sk);
659 struct inet_ehash_bucket *head;
660 struct hlist_nulls_head *list;
661 spinlock_t *lock;
662 bool ret = true;
663
664 WARN_ON_ONCE(!sk_unhashed(sk));
665
666 sk->sk_hash = sk_ehashfn(sk);
667 head = inet_ehash_bucket(hashinfo, sk->sk_hash);
668 list = &head->chain;
669 lock = inet_ehash_lockp(hashinfo, sk->sk_hash);
670
671 spin_lock(lock);
672 if (osk) {
673 WARN_ON_ONCE(sk->sk_hash != osk->sk_hash);
674 ret = sk_nulls_del_node_init_rcu(osk);
675 } else if (found_dup_sk) {
676 *found_dup_sk = inet_ehash_lookup_by_sk(sk, list);
677 if (*found_dup_sk)
678 ret = false;
679 }
680
681 if (ret)
682 __sk_nulls_add_node_rcu(sk, list);
683
684 spin_unlock(lock);
685
686 return ret;
687}
688
689bool inet_ehash_nolisten(struct sock *sk, struct sock *osk, bool *found_dup_sk)
690{
691 bool ok = inet_ehash_insert(sk, osk, found_dup_sk);
692
693 if (ok) {
694 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
695 } else {
696 this_cpu_inc(*sk->sk_prot->orphan_count);
697 inet_sk_set_state(sk, TCP_CLOSE);
698 sock_set_flag(sk, SOCK_DEAD);
699 inet_csk_destroy_sock(sk);
700 }
701 return ok;
702}
703EXPORT_SYMBOL_GPL(inet_ehash_nolisten);
704
705static int inet_reuseport_add_sock(struct sock *sk,
706 struct inet_listen_hashbucket *ilb)
707{
708 struct inet_bind_bucket *tb = inet_csk(sk)->icsk_bind_hash;
709 const struct hlist_nulls_node *node;
710 struct sock *sk2;
711 kuid_t uid = sock_i_uid(sk);
712
713 sk_nulls_for_each_rcu(sk2, node, &ilb->nulls_head) {
714 if (sk2 != sk &&
715 sk2->sk_family == sk->sk_family &&
716 ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
717 sk2->sk_bound_dev_if == sk->sk_bound_dev_if &&
718 inet_csk(sk2)->icsk_bind_hash == tb &&
719 sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
720 inet_rcv_saddr_equal(sk, sk2, false))
721 return reuseport_add_sock(sk, sk2,
722 inet_rcv_saddr_any(sk));
723 }
724
725 return reuseport_alloc(sk, inet_rcv_saddr_any(sk));
726}
727
728int __inet_hash(struct sock *sk, struct sock *osk)
729{
730 struct inet_hashinfo *hashinfo = tcp_or_dccp_get_hashinfo(sk);
731 struct inet_listen_hashbucket *ilb2;
732 int err = 0;
733
734 if (sk->sk_state != TCP_LISTEN) {
735 local_bh_disable();
736 inet_ehash_nolisten(sk, osk, NULL);
737 local_bh_enable();
738 return 0;
739 }
740 WARN_ON(!sk_unhashed(sk));
741 ilb2 = inet_lhash2_bucket_sk(hashinfo, sk);
742
743 spin_lock(&ilb2->lock);
744 if (sk->sk_reuseport) {
745 err = inet_reuseport_add_sock(sk, ilb2);
746 if (err)
747 goto unlock;
748 }
749 sock_set_flag(sk, SOCK_RCU_FREE);
750 if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
751 sk->sk_family == AF_INET6)
752 __sk_nulls_add_node_tail_rcu(sk, &ilb2->nulls_head);
753 else
754 __sk_nulls_add_node_rcu(sk, &ilb2->nulls_head);
755 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
756unlock:
757 spin_unlock(&ilb2->lock);
758
759 return err;
760}
761EXPORT_SYMBOL(__inet_hash);
762
763int inet_hash(struct sock *sk)
764{
765 int err = 0;
766
767 if (sk->sk_state != TCP_CLOSE)
768 err = __inet_hash(sk, NULL);
769
770 return err;
771}
772EXPORT_SYMBOL_GPL(inet_hash);
773
774void inet_unhash(struct sock *sk)
775{
776 struct inet_hashinfo *hashinfo = tcp_or_dccp_get_hashinfo(sk);
777
778 if (sk_unhashed(sk))
779 return;
780
781 if (sk->sk_state == TCP_LISTEN) {
782 struct inet_listen_hashbucket *ilb2;
783
784 ilb2 = inet_lhash2_bucket_sk(hashinfo, sk);
785 /* Don't disable bottom halves while acquiring the lock to
786 * avoid circular locking dependency on PREEMPT_RT.
787 */
788 spin_lock(&ilb2->lock);
789 if (sk_unhashed(sk)) {
790 spin_unlock(&ilb2->lock);
791 return;
792 }
793
794 if (rcu_access_pointer(sk->sk_reuseport_cb))
795 reuseport_stop_listen_sock(sk);
796
797 __sk_nulls_del_node_init_rcu(sk);
798 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
799 spin_unlock(&ilb2->lock);
800 } else {
801 spinlock_t *lock = inet_ehash_lockp(hashinfo, sk->sk_hash);
802
803 spin_lock_bh(lock);
804 if (sk_unhashed(sk)) {
805 spin_unlock_bh(lock);
806 return;
807 }
808 __sk_nulls_del_node_init_rcu(sk);
809 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
810 spin_unlock_bh(lock);
811 }
812}
813EXPORT_SYMBOL_GPL(inet_unhash);
814
815static bool inet_bind2_bucket_match(const struct inet_bind2_bucket *tb,
816 const struct net *net, unsigned short port,
817 int l3mdev, const struct sock *sk)
818{
819 if (!net_eq(ib2_net(tb), net) || tb->port != port ||
820 tb->l3mdev != l3mdev)
821 return false;
822
823 return inet_bind2_bucket_addr_match(tb, sk);
824}
825
826bool inet_bind2_bucket_match_addr_any(const struct inet_bind2_bucket *tb, const struct net *net,
827 unsigned short port, int l3mdev, const struct sock *sk)
828{
829 if (!net_eq(ib2_net(tb), net) || tb->port != port ||
830 tb->l3mdev != l3mdev)
831 return false;
832
833#if IS_ENABLED(CONFIG_IPV6)
834 if (tb->addr_type == IPV6_ADDR_ANY)
835 return true;
836
837 if (tb->addr_type != IPV6_ADDR_MAPPED)
838 return false;
839
840 if (sk->sk_family == AF_INET6 &&
841 !ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr))
842 return false;
843#endif
844 return tb->rcv_saddr == 0;
845}
846
847/* The socket's bhash2 hashbucket spinlock must be held when this is called */
848struct inet_bind2_bucket *
849inet_bind2_bucket_find(const struct inet_bind_hashbucket *head, const struct net *net,
850 unsigned short port, int l3mdev, const struct sock *sk)
851{
852 struct inet_bind2_bucket *bhash2 = NULL;
853
854 inet_bind_bucket_for_each(bhash2, &head->chain)
855 if (inet_bind2_bucket_match(bhash2, net, port, l3mdev, sk))
856 break;
857
858 return bhash2;
859}
860
861struct inet_bind_hashbucket *
862inet_bhash2_addr_any_hashbucket(const struct sock *sk, const struct net *net, int port)
863{
864 struct inet_hashinfo *hinfo = tcp_or_dccp_get_hashinfo(sk);
865 u32 hash;
866
867#if IS_ENABLED(CONFIG_IPV6)
868 if (sk->sk_family == AF_INET6)
869 hash = ipv6_portaddr_hash(net, &in6addr_any, port);
870 else
871#endif
872 hash = ipv4_portaddr_hash(net, 0, port);
873
874 return &hinfo->bhash2[hash & (hinfo->bhash_size - 1)];
875}
876
877static void inet_update_saddr(struct sock *sk, void *saddr, int family)
878{
879 if (family == AF_INET) {
880 inet_sk(sk)->inet_saddr = *(__be32 *)saddr;
881 sk_rcv_saddr_set(sk, inet_sk(sk)->inet_saddr);
882 }
883#if IS_ENABLED(CONFIG_IPV6)
884 else {
885 sk->sk_v6_rcv_saddr = *(struct in6_addr *)saddr;
886 }
887#endif
888}
889
890static int __inet_bhash2_update_saddr(struct sock *sk, void *saddr, int family, bool reset)
891{
892 struct inet_hashinfo *hinfo = tcp_or_dccp_get_hashinfo(sk);
893 struct inet_bind_hashbucket *head, *head2;
894 struct inet_bind2_bucket *tb2, *new_tb2;
895 int l3mdev = inet_sk_bound_l3mdev(sk);
896 int port = inet_sk(sk)->inet_num;
897 struct net *net = sock_net(sk);
898 int bhash;
899
900 if (!inet_csk(sk)->icsk_bind2_hash) {
901 /* Not bind()ed before. */
902 if (reset)
903 inet_reset_saddr(sk);
904 else
905 inet_update_saddr(sk, saddr, family);
906
907 return 0;
908 }
909
910 /* Allocate a bind2 bucket ahead of time to avoid permanently putting
911 * the bhash2 table in an inconsistent state if a new tb2 bucket
912 * allocation fails.
913 */
914 new_tb2 = kmem_cache_alloc(hinfo->bind2_bucket_cachep, GFP_ATOMIC);
915 if (!new_tb2) {
916 if (reset) {
917 /* The (INADDR_ANY, port) bucket might have already
918 * been freed, then we cannot fixup icsk_bind2_hash,
919 * so we give up and unlink sk from bhash/bhash2 not
920 * to leave inconsistency in bhash2.
921 */
922 inet_put_port(sk);
923 inet_reset_saddr(sk);
924 }
925
926 return -ENOMEM;
927 }
928
929 bhash = inet_bhashfn(net, port, hinfo->bhash_size);
930 head = &hinfo->bhash[bhash];
931 head2 = inet_bhashfn_portaddr(hinfo, sk, net, port);
932
933 /* If we change saddr locklessly, another thread
934 * iterating over bhash might see corrupted address.
935 */
936 spin_lock_bh(&head->lock);
937
938 spin_lock(&head2->lock);
939 __sk_del_bind_node(sk);
940 inet_bind2_bucket_destroy(hinfo->bind2_bucket_cachep, inet_csk(sk)->icsk_bind2_hash);
941 spin_unlock(&head2->lock);
942
943 if (reset)
944 inet_reset_saddr(sk);
945 else
946 inet_update_saddr(sk, saddr, family);
947
948 head2 = inet_bhashfn_portaddr(hinfo, sk, net, port);
949
950 spin_lock(&head2->lock);
951 tb2 = inet_bind2_bucket_find(head2, net, port, l3mdev, sk);
952 if (!tb2) {
953 tb2 = new_tb2;
954 inet_bind2_bucket_init(tb2, net, head2, inet_csk(sk)->icsk_bind_hash, sk);
955 }
956 inet_csk(sk)->icsk_bind2_hash = tb2;
957 sk_add_bind_node(sk, &tb2->owners);
958 spin_unlock(&head2->lock);
959
960 spin_unlock_bh(&head->lock);
961
962 if (tb2 != new_tb2)
963 kmem_cache_free(hinfo->bind2_bucket_cachep, new_tb2);
964
965 return 0;
966}
967
968int inet_bhash2_update_saddr(struct sock *sk, void *saddr, int family)
969{
970 return __inet_bhash2_update_saddr(sk, saddr, family, false);
971}
972EXPORT_SYMBOL_GPL(inet_bhash2_update_saddr);
973
974void inet_bhash2_reset_saddr(struct sock *sk)
975{
976 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
977 __inet_bhash2_update_saddr(sk, NULL, 0, true);
978}
979EXPORT_SYMBOL_GPL(inet_bhash2_reset_saddr);
980
981/* RFC 6056 3.3.4. Algorithm 4: Double-Hash Port Selection Algorithm
982 * Note that we use 32bit integers (vs RFC 'short integers')
983 * because 2^16 is not a multiple of num_ephemeral and this
984 * property might be used by clever attacker.
985 *
986 * RFC claims using TABLE_LENGTH=10 buckets gives an improvement, though
987 * attacks were since demonstrated, thus we use 65536 by default instead
988 * to really give more isolation and privacy, at the expense of 256kB
989 * of kernel memory.
990 */
991#define INET_TABLE_PERTURB_SIZE (1 << CONFIG_INET_TABLE_PERTURB_ORDER)
992static u32 *table_perturb;
993
994int __inet_hash_connect(struct inet_timewait_death_row *death_row,
995 struct sock *sk, u64 port_offset,
996 int (*check_established)(struct inet_timewait_death_row *,
997 struct sock *, __u16, struct inet_timewait_sock **))
998{
999 struct inet_hashinfo *hinfo = death_row->hashinfo;
1000 struct inet_bind_hashbucket *head, *head2;
1001 struct inet_timewait_sock *tw = NULL;
1002 int port = inet_sk(sk)->inet_num;
1003 struct net *net = sock_net(sk);
1004 struct inet_bind2_bucket *tb2;
1005 struct inet_bind_bucket *tb;
1006 bool tb_created = false;
1007 u32 remaining, offset;
1008 int ret, i, low, high;
1009 bool local_ports;
1010 int step, l3mdev;
1011 u32 index;
1012
1013 if (port) {
1014 local_bh_disable();
1015 ret = check_established(death_row, sk, port, NULL);
1016 local_bh_enable();
1017 return ret;
1018 }
1019
1020 l3mdev = inet_sk_bound_l3mdev(sk);
1021
1022 local_ports = inet_sk_get_local_port_range(sk, &low, &high);
1023 step = local_ports ? 1 : 2;
1024
1025 high++; /* [32768, 60999] -> [32768, 61000[ */
1026 remaining = high - low;
1027 if (!local_ports && remaining > 1)
1028 remaining &= ~1U;
1029
1030 get_random_sleepable_once(table_perturb,
1031 INET_TABLE_PERTURB_SIZE * sizeof(*table_perturb));
1032 index = port_offset & (INET_TABLE_PERTURB_SIZE - 1);
1033
1034 offset = READ_ONCE(table_perturb[index]) + (port_offset >> 32);
1035 offset %= remaining;
1036
1037 /* In first pass we try ports of @low parity.
1038 * inet_csk_get_port() does the opposite choice.
1039 */
1040 if (!local_ports)
1041 offset &= ~1U;
1042other_parity_scan:
1043 port = low + offset;
1044 for (i = 0; i < remaining; i += step, port += step) {
1045 if (unlikely(port >= high))
1046 port -= remaining;
1047 if (inet_is_local_reserved_port(net, port))
1048 continue;
1049 head = &hinfo->bhash[inet_bhashfn(net, port,
1050 hinfo->bhash_size)];
1051 spin_lock_bh(&head->lock);
1052
1053 /* Does not bother with rcv_saddr checks, because
1054 * the established check is already unique enough.
1055 */
1056 inet_bind_bucket_for_each(tb, &head->chain) {
1057 if (inet_bind_bucket_match(tb, net, port, l3mdev)) {
1058 if (tb->fastreuse >= 0 ||
1059 tb->fastreuseport >= 0)
1060 goto next_port;
1061 WARN_ON(hlist_empty(&tb->bhash2));
1062 if (!check_established(death_row, sk,
1063 port, &tw))
1064 goto ok;
1065 goto next_port;
1066 }
1067 }
1068
1069 tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep,
1070 net, head, port, l3mdev);
1071 if (!tb) {
1072 spin_unlock_bh(&head->lock);
1073 return -ENOMEM;
1074 }
1075 tb_created = true;
1076 tb->fastreuse = -1;
1077 tb->fastreuseport = -1;
1078 goto ok;
1079next_port:
1080 spin_unlock_bh(&head->lock);
1081 cond_resched();
1082 }
1083
1084 if (!local_ports) {
1085 offset++;
1086 if ((offset & 1) && remaining > 1)
1087 goto other_parity_scan;
1088 }
1089 return -EADDRNOTAVAIL;
1090
1091ok:
1092 /* Find the corresponding tb2 bucket since we need to
1093 * add the socket to the bhash2 table as well
1094 */
1095 head2 = inet_bhashfn_portaddr(hinfo, sk, net, port);
1096 spin_lock(&head2->lock);
1097
1098 tb2 = inet_bind2_bucket_find(head2, net, port, l3mdev, sk);
1099 if (!tb2) {
1100 tb2 = inet_bind2_bucket_create(hinfo->bind2_bucket_cachep, net,
1101 head2, tb, sk);
1102 if (!tb2)
1103 goto error;
1104 }
1105
1106 /* Here we want to add a little bit of randomness to the next source
1107 * port that will be chosen. We use a max() with a random here so that
1108 * on low contention the randomness is maximal and on high contention
1109 * it may be inexistent.
1110 */
1111 i = max_t(int, i, get_random_u32_below(8) * step);
1112 WRITE_ONCE(table_perturb[index], READ_ONCE(table_perturb[index]) + i + step);
1113
1114 /* Head lock still held and bh's disabled */
1115 inet_bind_hash(sk, tb, tb2, port);
1116
1117 if (sk_unhashed(sk)) {
1118 inet_sk(sk)->inet_sport = htons(port);
1119 inet_ehash_nolisten(sk, (struct sock *)tw, NULL);
1120 }
1121 if (tw)
1122 inet_twsk_bind_unhash(tw, hinfo);
1123
1124 spin_unlock(&head2->lock);
1125 spin_unlock(&head->lock);
1126
1127 if (tw)
1128 inet_twsk_deschedule_put(tw);
1129 local_bh_enable();
1130 return 0;
1131
1132error:
1133 if (sk_hashed(sk)) {
1134 spinlock_t *lock = inet_ehash_lockp(hinfo, sk->sk_hash);
1135
1136 sock_prot_inuse_add(net, sk->sk_prot, -1);
1137
1138 spin_lock(lock);
1139 __sk_nulls_del_node_init_rcu(sk);
1140 spin_unlock(lock);
1141
1142 sk->sk_hash = 0;
1143 inet_sk(sk)->inet_sport = 0;
1144 inet_sk(sk)->inet_num = 0;
1145
1146 if (tw)
1147 inet_twsk_bind_unhash(tw, hinfo);
1148 }
1149
1150 spin_unlock(&head2->lock);
1151 if (tb_created)
1152 inet_bind_bucket_destroy(hinfo->bind_bucket_cachep, tb);
1153 spin_unlock(&head->lock);
1154
1155 if (tw)
1156 inet_twsk_deschedule_put(tw);
1157
1158 local_bh_enable();
1159
1160 return -ENOMEM;
1161}
1162
1163/*
1164 * Bind a port for a connect operation and hash it.
1165 */
1166int inet_hash_connect(struct inet_timewait_death_row *death_row,
1167 struct sock *sk)
1168{
1169 u64 port_offset = 0;
1170
1171 if (!inet_sk(sk)->inet_num)
1172 port_offset = inet_sk_port_offset(sk);
1173 return __inet_hash_connect(death_row, sk, port_offset,
1174 __inet_check_established);
1175}
1176EXPORT_SYMBOL_GPL(inet_hash_connect);
1177
1178static void init_hashinfo_lhash2(struct inet_hashinfo *h)
1179{
1180 int i;
1181
1182 for (i = 0; i <= h->lhash2_mask; i++) {
1183 spin_lock_init(&h->lhash2[i].lock);
1184 INIT_HLIST_NULLS_HEAD(&h->lhash2[i].nulls_head,
1185 i + LISTENING_NULLS_BASE);
1186 }
1187}
1188
1189void __init inet_hashinfo2_init(struct inet_hashinfo *h, const char *name,
1190 unsigned long numentries, int scale,
1191 unsigned long low_limit,
1192 unsigned long high_limit)
1193{
1194 h->lhash2 = alloc_large_system_hash(name,
1195 sizeof(*h->lhash2),
1196 numentries,
1197 scale,
1198 0,
1199 NULL,
1200 &h->lhash2_mask,
1201 low_limit,
1202 high_limit);
1203 init_hashinfo_lhash2(h);
1204
1205 /* this one is used for source ports of outgoing connections */
1206 table_perturb = alloc_large_system_hash("Table-perturb",
1207 sizeof(*table_perturb),
1208 INET_TABLE_PERTURB_SIZE,
1209 0, 0, NULL, NULL,
1210 INET_TABLE_PERTURB_SIZE,
1211 INET_TABLE_PERTURB_SIZE);
1212}
1213
1214int inet_hashinfo2_init_mod(struct inet_hashinfo *h)
1215{
1216 h->lhash2 = kmalloc_array(INET_LHTABLE_SIZE, sizeof(*h->lhash2), GFP_KERNEL);
1217 if (!h->lhash2)
1218 return -ENOMEM;
1219
1220 h->lhash2_mask = INET_LHTABLE_SIZE - 1;
1221 /* INET_LHTABLE_SIZE must be a power of 2 */
1222 BUG_ON(INET_LHTABLE_SIZE & h->lhash2_mask);
1223
1224 init_hashinfo_lhash2(h);
1225 return 0;
1226}
1227EXPORT_SYMBOL_GPL(inet_hashinfo2_init_mod);
1228
1229int inet_ehash_locks_alloc(struct inet_hashinfo *hashinfo)
1230{
1231 unsigned int locksz = sizeof(spinlock_t);
1232 unsigned int i, nblocks = 1;
1233
1234 if (locksz != 0) {
1235 /* allocate 2 cache lines or at least one spinlock per cpu */
1236 nblocks = max(2U * L1_CACHE_BYTES / locksz, 1U);
1237 nblocks = roundup_pow_of_two(nblocks * num_possible_cpus());
1238
1239 /* no more locks than number of hash buckets */
1240 nblocks = min(nblocks, hashinfo->ehash_mask + 1);
1241
1242 hashinfo->ehash_locks = kvmalloc_array(nblocks, locksz, GFP_KERNEL);
1243 if (!hashinfo->ehash_locks)
1244 return -ENOMEM;
1245
1246 for (i = 0; i < nblocks; i++)
1247 spin_lock_init(&hashinfo->ehash_locks[i]);
1248 }
1249 hashinfo->ehash_locks_mask = nblocks - 1;
1250 return 0;
1251}
1252EXPORT_SYMBOL_GPL(inet_ehash_locks_alloc);
1253
1254struct inet_hashinfo *inet_pernet_hashinfo_alloc(struct inet_hashinfo *hashinfo,
1255 unsigned int ehash_entries)
1256{
1257 struct inet_hashinfo *new_hashinfo;
1258 int i;
1259
1260 new_hashinfo = kmemdup(hashinfo, sizeof(*hashinfo), GFP_KERNEL);
1261 if (!new_hashinfo)
1262 goto err;
1263
1264 new_hashinfo->ehash = vmalloc_huge(ehash_entries * sizeof(struct inet_ehash_bucket),
1265 GFP_KERNEL_ACCOUNT);
1266 if (!new_hashinfo->ehash)
1267 goto free_hashinfo;
1268
1269 new_hashinfo->ehash_mask = ehash_entries - 1;
1270
1271 if (inet_ehash_locks_alloc(new_hashinfo))
1272 goto free_ehash;
1273
1274 for (i = 0; i < ehash_entries; i++)
1275 INIT_HLIST_NULLS_HEAD(&new_hashinfo->ehash[i].chain, i);
1276
1277 new_hashinfo->pernet = true;
1278
1279 return new_hashinfo;
1280
1281free_ehash:
1282 vfree(new_hashinfo->ehash);
1283free_hashinfo:
1284 kfree(new_hashinfo);
1285err:
1286 return NULL;
1287}
1288EXPORT_SYMBOL_GPL(inet_pernet_hashinfo_alloc);
1289
1290void inet_pernet_hashinfo_free(struct inet_hashinfo *hashinfo)
1291{
1292 if (!hashinfo->pernet)
1293 return;
1294
1295 inet_ehash_locks_free(hashinfo);
1296 vfree(hashinfo->ehash);
1297 kfree(hashinfo);
1298}
1299EXPORT_SYMBOL_GPL(inet_pernet_hashinfo_free);
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
6 *
7 * Generic INET transport hashtables
8 *
9 * Authors: Lotsa people, from code originally in tcp
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/* This function handles inet_sock, but also timewait and request sockets
41 * for IPv4/IPv6.
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 * Allocate and initialize a new local port bind bucket.
59 * The bindhash mutex for snum's hash chain must be held here.
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 * Caller must hold hashbucket lock for this tb with local BH disabled
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 * Get rid of any references to a local port held by the given sock.
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 /* NOTE: using tproxy and redirecting skbs to a proxy
148 * on a different listener port breaks the assumption
149 * that the listener socket's icsk_bind_hash is the same
150 * as that of the child socket. We have to look up or
151 * create a new bind bucket for the child here. */
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 * Here are some nice properties to exploit here. The BSD API
267 * does not allow a listening sock to specify the remote port nor the
268 * remote address for the connection. So always assume those are both
269 * wildcarded during the search since they can never be otherwise.
270 */
271
272/* called with rcu_read_lock() : No refcount taken on the socket */
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; /* only TCP is supported */
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 /* Lookup redirect from BPF */
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 /* Lookup lhash2 with INADDR_ANY */
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/* All sockets share common refcount, but have different destructors */
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 /* Optimize here for direct hit, only listening connections can
400 * have wildcards anyways.
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 * if the nulls value we got at the end of this lookup is
425 * not the expected one, we must restart lookup.
426 * We probably met an item that was moved to another chain.
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/* called with local bh disabled */
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 /* Must record num and sport now. Otherwise we will see
477 * in hash table socket with a funny identity.
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 /* Silly. Should hash-dance instead... */
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/* insert a socket into ehash, and eventually remove another one
514 * (The another one can be a SYN_RECV or TIMEWAIT
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 /* No definite answer... Walk to established hash table */
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++; /* [32768, 60999] -> [32768, 61000[ */
699 remaining = high - low;
700 if (likely(remaining > 1))
701 remaining &= ~1U;
702
703 offset = (hint + port_offset) % remaining;
704 /* In first pass we try ports of @low parity.
705 * inet_csk_get_port() does the opposite choice.
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 /* Does not bother with rcv_saddr checks, because
720 * the established check is already unique enough.
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 /* Head lock still held and bh's disabled */
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 * Bind a port for a connect operation and hash it.
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 /* INET_LHTABLE_SIZE must be a power of 2 */
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 /* allocate 2 cache lines or at least one spinlock per cpu */
854 nblocks = max(2U * L1_CACHE_BYTES / locksz, 1U);
855 nblocks = roundup_pow_of_two(nblocks * num_possible_cpus());
856
857 /* no more locks than number of hash buckets */
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);