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/ip.h>
28 #include <net/tcp.h>
29 #include <net/sock_reuseport.h>
30 
inet_ehashfn(const struct net * net,const __be32 laddr,const __u16 lport,const __be32 faddr,const __be16 fport)31 static u32 inet_ehashfn(const struct net *net, const __be32 laddr,
32 			const __u16 lport, const __be32 faddr,
33 			const __be16 fport)
34 {
35 	static u32 inet_ehash_secret __read_mostly;
36 
37 	net_get_random_once(&inet_ehash_secret, sizeof(inet_ehash_secret));
38 
39 	return __inet_ehashfn(laddr, lport, faddr, fport,
40 			      inet_ehash_secret + net_hash_mix(net));
41 }
42 
43 /* This function handles inet_sock, but also timewait and request sockets
44  * for IPv4/IPv6.
45  */
sk_ehashfn(const struct sock * sk)46 static 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  */
inet_bind_bucket_create(struct kmem_cache * cachep,struct net * net,struct inet_bind_hashbucket * head,const unsigned short snum,int l3mdev)64 struct 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->owners);
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  */
inet_bind_bucket_destroy(struct kmem_cache * cachep,struct inet_bind_bucket * tb)87 void inet_bind_bucket_destroy(struct kmem_cache *cachep, struct inet_bind_bucket *tb)
88 {
89 	if (hlist_empty(&tb->owners)) {
90 		__hlist_del(&tb->node);
91 		kmem_cache_free(cachep, tb);
92 	}
93 }
94 
inet_bind_hash(struct sock * sk,struct inet_bind_bucket * tb,const unsigned short snum)95 void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
96 		    const unsigned short snum)
97 {
98 	inet_sk(sk)->inet_num = snum;
99 	sk_add_bind_node(sk, &tb->owners);
100 	inet_csk(sk)->icsk_bind_hash = tb;
101 }
102 
103 /*
104  * Get rid of any references to a local port held by the given sock.
105  */
__inet_put_port(struct sock * sk)106 static void __inet_put_port(struct sock *sk)
107 {
108 	struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
109 	const int bhash = inet_bhashfn(sock_net(sk), inet_sk(sk)->inet_num,
110 			hashinfo->bhash_size);
111 	struct inet_bind_hashbucket *head = &hashinfo->bhash[bhash];
112 	struct inet_bind_bucket *tb;
113 
114 	spin_lock(&head->lock);
115 	tb = inet_csk(sk)->icsk_bind_hash;
116 	__sk_del_bind_node(sk);
117 	inet_csk(sk)->icsk_bind_hash = NULL;
118 	inet_sk(sk)->inet_num = 0;
119 	inet_bind_bucket_destroy(hashinfo->bind_bucket_cachep, tb);
120 	spin_unlock(&head->lock);
121 }
122 
inet_put_port(struct sock * sk)123 void inet_put_port(struct sock *sk)
124 {
125 	local_bh_disable();
126 	__inet_put_port(sk);
127 	local_bh_enable();
128 }
129 EXPORT_SYMBOL(inet_put_port);
130 
__inet_inherit_port(const struct sock * sk,struct sock * child)131 int __inet_inherit_port(const struct sock *sk, struct sock *child)
132 {
133 	struct inet_hashinfo *table = sk->sk_prot->h.hashinfo;
134 	unsigned short port = inet_sk(child)->inet_num;
135 	const int bhash = inet_bhashfn(sock_net(sk), port,
136 			table->bhash_size);
137 	struct inet_bind_hashbucket *head = &table->bhash[bhash];
138 	struct inet_bind_bucket *tb;
139 	int l3mdev;
140 
141 	spin_lock(&head->lock);
142 	tb = inet_csk(sk)->icsk_bind_hash;
143 	if (unlikely(!tb)) {
144 		spin_unlock(&head->lock);
145 		return -ENOENT;
146 	}
147 	if (tb->port != port) {
148 		l3mdev = inet_sk_bound_l3mdev(sk);
149 
150 		/* NOTE: using tproxy and redirecting skbs to a proxy
151 		 * on a different listener port breaks the assumption
152 		 * that the listener socket's icsk_bind_hash is the same
153 		 * as that of the child socket. We have to look up or
154 		 * create a new bind bucket for the child here. */
155 		inet_bind_bucket_for_each(tb, &head->chain) {
156 			if (net_eq(ib_net(tb), sock_net(sk)) &&
157 			    tb->l3mdev == l3mdev && tb->port == port)
158 				break;
159 		}
160 		if (!tb) {
161 			tb = inet_bind_bucket_create(table->bind_bucket_cachep,
162 						     sock_net(sk), head, port,
163 						     l3mdev);
164 			if (!tb) {
165 				spin_unlock(&head->lock);
166 				return -ENOMEM;
167 			}
168 		}
169 		inet_csk_update_fastreuse(tb, child);
170 	}
171 	inet_bind_hash(child, tb, port);
172 	spin_unlock(&head->lock);
173 
174 	return 0;
175 }
176 EXPORT_SYMBOL_GPL(__inet_inherit_port);
177 
178 static struct inet_listen_hashbucket *
inet_lhash2_bucket_sk(struct inet_hashinfo * h,struct sock * sk)179 inet_lhash2_bucket_sk(struct inet_hashinfo *h, struct sock *sk)
180 {
181 	u32 hash;
182 
183 #if IS_ENABLED(CONFIG_IPV6)
184 	if (sk->sk_family == AF_INET6)
185 		hash = ipv6_portaddr_hash(sock_net(sk),
186 					  &sk->sk_v6_rcv_saddr,
187 					  inet_sk(sk)->inet_num);
188 	else
189 #endif
190 		hash = ipv4_portaddr_hash(sock_net(sk),
191 					  inet_sk(sk)->inet_rcv_saddr,
192 					  inet_sk(sk)->inet_num);
193 	return inet_lhash2_bucket(h, hash);
194 }
195 
inet_hash2(struct inet_hashinfo * h,struct sock * sk)196 static void inet_hash2(struct inet_hashinfo *h, struct sock *sk)
197 {
198 	struct inet_listen_hashbucket *ilb2;
199 
200 	if (!h->lhash2)
201 		return;
202 
203 	ilb2 = inet_lhash2_bucket_sk(h, sk);
204 
205 	spin_lock(&ilb2->lock);
206 	if (sk->sk_reuseport && sk->sk_family == AF_INET6)
207 		hlist_add_tail_rcu(&inet_csk(sk)->icsk_listen_portaddr_node,
208 				   &ilb2->head);
209 	else
210 		hlist_add_head_rcu(&inet_csk(sk)->icsk_listen_portaddr_node,
211 				   &ilb2->head);
212 	ilb2->count++;
213 	spin_unlock(&ilb2->lock);
214 }
215 
inet_unhash2(struct inet_hashinfo * h,struct sock * sk)216 static void inet_unhash2(struct inet_hashinfo *h, struct sock *sk)
217 {
218 	struct inet_listen_hashbucket *ilb2;
219 
220 	if (!h->lhash2 ||
221 	    WARN_ON_ONCE(hlist_unhashed(&inet_csk(sk)->icsk_listen_portaddr_node)))
222 		return;
223 
224 	ilb2 = inet_lhash2_bucket_sk(h, sk);
225 
226 	spin_lock(&ilb2->lock);
227 	hlist_del_init_rcu(&inet_csk(sk)->icsk_listen_portaddr_node);
228 	ilb2->count--;
229 	spin_unlock(&ilb2->lock);
230 }
231 
compute_score(struct sock * sk,struct net * net,const unsigned short hnum,const __be32 daddr,const int dif,const int sdif)232 static inline int compute_score(struct sock *sk, struct net *net,
233 				const unsigned short hnum, const __be32 daddr,
234 				const int dif, const int sdif)
235 {
236 	int score = -1;
237 
238 	if (net_eq(sock_net(sk), net) && sk->sk_num == hnum &&
239 			!ipv6_only_sock(sk)) {
240 		if (sk->sk_rcv_saddr != daddr)
241 			return -1;
242 
243 		if (!inet_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif))
244 			return -1;
245 		score =  sk->sk_bound_dev_if ? 2 : 1;
246 
247 		if (sk->sk_family == PF_INET)
248 			score++;
249 		if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
250 			score++;
251 	}
252 	return score;
253 }
254 
lookup_reuseport(struct net * net,struct sock * sk,struct sk_buff * skb,int doff,__be32 saddr,__be16 sport,__be32 daddr,unsigned short hnum)255 static inline struct sock *lookup_reuseport(struct net *net, struct sock *sk,
256 					    struct sk_buff *skb, int doff,
257 					    __be32 saddr, __be16 sport,
258 					    __be32 daddr, unsigned short hnum)
259 {
260 	struct sock *reuse_sk = NULL;
261 	u32 phash;
262 
263 	if (sk->sk_reuseport) {
264 		phash = inet_ehashfn(net, daddr, hnum, saddr, sport);
265 		reuse_sk = reuseport_select_sock(sk, phash, skb, doff);
266 	}
267 	return reuse_sk;
268 }
269 
270 /*
271  * Here are some nice properties to exploit here. The BSD API
272  * does not allow a listening sock to specify the remote port nor the
273  * remote address for the connection. So always assume those are both
274  * wildcarded during the search since they can never be otherwise.
275  */
276 
277 /* called with rcu_read_lock() : No refcount taken on the socket */
inet_lhash2_lookup(struct net * net,struct inet_listen_hashbucket * ilb2,struct sk_buff * skb,int doff,const __be32 saddr,__be16 sport,const __be32 daddr,const unsigned short hnum,const int dif,const int sdif)278 static struct sock *inet_lhash2_lookup(struct net *net,
279 				struct inet_listen_hashbucket *ilb2,
280 				struct sk_buff *skb, int doff,
281 				const __be32 saddr, __be16 sport,
282 				const __be32 daddr, const unsigned short hnum,
283 				const int dif, const int sdif)
284 {
285 	struct inet_connection_sock *icsk;
286 	struct sock *sk, *result = NULL;
287 	int score, hiscore = 0;
288 
289 	inet_lhash2_for_each_icsk_rcu(icsk, &ilb2->head) {
290 		sk = (struct sock *)icsk;
291 		score = compute_score(sk, net, hnum, daddr, dif, sdif);
292 		if (score > hiscore) {
293 			result = lookup_reuseport(net, sk, skb, doff,
294 						  saddr, sport, daddr, hnum);
295 			if (result)
296 				return result;
297 
298 			result = sk;
299 			hiscore = score;
300 		}
301 	}
302 
303 	return result;
304 }
305 
inet_lookup_run_bpf(struct net * net,struct inet_hashinfo * hashinfo,struct sk_buff * skb,int doff,__be32 saddr,__be16 sport,__be32 daddr,u16 hnum)306 static inline struct sock *inet_lookup_run_bpf(struct net *net,
307 					       struct inet_hashinfo *hashinfo,
308 					       struct sk_buff *skb, int doff,
309 					       __be32 saddr, __be16 sport,
310 					       __be32 daddr, u16 hnum)
311 {
312 	struct sock *sk, *reuse_sk;
313 	bool no_reuseport;
314 
315 	if (hashinfo != &tcp_hashinfo)
316 		return NULL; /* only TCP is supported */
317 
318 	no_reuseport = bpf_sk_lookup_run_v4(net, IPPROTO_TCP,
319 					    saddr, sport, daddr, hnum, &sk);
320 	if (no_reuseport || IS_ERR_OR_NULL(sk))
321 		return sk;
322 
323 	reuse_sk = lookup_reuseport(net, sk, skb, doff, saddr, sport, daddr, hnum);
324 	if (reuse_sk)
325 		sk = reuse_sk;
326 	return sk;
327 }
328 
__inet_lookup_listener(struct net * net,struct inet_hashinfo * hashinfo,struct sk_buff * skb,int doff,const __be32 saddr,__be16 sport,const __be32 daddr,const unsigned short hnum,const int dif,const int sdif)329 struct sock *__inet_lookup_listener(struct net *net,
330 				    struct inet_hashinfo *hashinfo,
331 				    struct sk_buff *skb, int doff,
332 				    const __be32 saddr, __be16 sport,
333 				    const __be32 daddr, const unsigned short hnum,
334 				    const int dif, const int sdif)
335 {
336 	struct inet_listen_hashbucket *ilb2;
337 	struct sock *result = NULL;
338 	unsigned int hash2;
339 
340 	/* Lookup redirect from BPF */
341 	if (static_branch_unlikely(&bpf_sk_lookup_enabled)) {
342 		result = inet_lookup_run_bpf(net, hashinfo, skb, doff,
343 					     saddr, sport, daddr, hnum);
344 		if (result)
345 			goto done;
346 	}
347 
348 	hash2 = ipv4_portaddr_hash(net, daddr, hnum);
349 	ilb2 = inet_lhash2_bucket(hashinfo, hash2);
350 
351 	result = inet_lhash2_lookup(net, ilb2, skb, doff,
352 				    saddr, sport, daddr, hnum,
353 				    dif, sdif);
354 	if (result)
355 		goto done;
356 
357 	/* Lookup lhash2 with INADDR_ANY */
358 	hash2 = ipv4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
359 	ilb2 = inet_lhash2_bucket(hashinfo, hash2);
360 
361 	result = inet_lhash2_lookup(net, ilb2, skb, doff,
362 				    saddr, sport, htonl(INADDR_ANY), hnum,
363 				    dif, sdif);
364 done:
365 	if (IS_ERR(result))
366 		return NULL;
367 	return result;
368 }
369 EXPORT_SYMBOL_GPL(__inet_lookup_listener);
370 
371 /* All sockets share common refcount, but have different destructors */
sock_gen_put(struct sock * sk)372 void sock_gen_put(struct sock *sk)
373 {
374 	if (!refcount_dec_and_test(&sk->sk_refcnt))
375 		return;
376 
377 	if (sk->sk_state == TCP_TIME_WAIT)
378 		inet_twsk_free(inet_twsk(sk));
379 	else if (sk->sk_state == TCP_NEW_SYN_RECV)
380 		reqsk_free(inet_reqsk(sk));
381 	else
382 		sk_free(sk);
383 }
384 EXPORT_SYMBOL_GPL(sock_gen_put);
385 
sock_edemux(struct sk_buff * skb)386 void sock_edemux(struct sk_buff *skb)
387 {
388 	sock_gen_put(skb->sk);
389 }
390 EXPORT_SYMBOL(sock_edemux);
391 
__inet_lookup_established(struct net * net,struct inet_hashinfo * hashinfo,const __be32 saddr,const __be16 sport,const __be32 daddr,const u16 hnum,const int dif,const int sdif)392 struct sock *__inet_lookup_established(struct net *net,
393 				  struct inet_hashinfo *hashinfo,
394 				  const __be32 saddr, const __be16 sport,
395 				  const __be32 daddr, const u16 hnum,
396 				  const int dif, const int sdif)
397 {
398 	INET_ADDR_COOKIE(acookie, saddr, daddr);
399 	const __portpair ports = INET_COMBINED_PORTS(sport, hnum);
400 	struct sock *sk;
401 	const struct hlist_nulls_node *node;
402 	/* Optimize here for direct hit, only listening connections can
403 	 * have wildcards anyways.
404 	 */
405 	unsigned int hash = inet_ehashfn(net, daddr, hnum, saddr, sport);
406 	unsigned int slot = hash & hashinfo->ehash_mask;
407 	struct inet_ehash_bucket *head = &hashinfo->ehash[slot];
408 
409 begin:
410 	sk_nulls_for_each_rcu(sk, node, &head->chain) {
411 		if (sk->sk_hash != hash)
412 			continue;
413 		if (likely(INET_MATCH(sk, net, acookie,
414 				      saddr, daddr, ports, dif, sdif))) {
415 			if (unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
416 				goto out;
417 			if (unlikely(!INET_MATCH(sk, net, acookie,
418 						 saddr, daddr, ports,
419 						 dif, sdif))) {
420 				sock_gen_put(sk);
421 				goto begin;
422 			}
423 			goto found;
424 		}
425 	}
426 	/*
427 	 * if the nulls value we got at the end of this lookup is
428 	 * not the expected one, we must restart lookup.
429 	 * We probably met an item that was moved to another chain.
430 	 */
431 	if (get_nulls_value(node) != slot)
432 		goto begin;
433 out:
434 	sk = NULL;
435 found:
436 	return sk;
437 }
438 EXPORT_SYMBOL_GPL(__inet_lookup_established);
439 
440 /* called with local bh disabled */
__inet_check_established(struct inet_timewait_death_row * death_row,struct sock * sk,__u16 lport,struct inet_timewait_sock ** twp)441 static int __inet_check_established(struct inet_timewait_death_row *death_row,
442 				    struct sock *sk, __u16 lport,
443 				    struct inet_timewait_sock **twp)
444 {
445 	struct inet_hashinfo *hinfo = death_row->hashinfo;
446 	struct inet_sock *inet = inet_sk(sk);
447 	__be32 daddr = inet->inet_rcv_saddr;
448 	__be32 saddr = inet->inet_daddr;
449 	int dif = sk->sk_bound_dev_if;
450 	struct net *net = sock_net(sk);
451 	int sdif = l3mdev_master_ifindex_by_index(net, dif);
452 	INET_ADDR_COOKIE(acookie, saddr, daddr);
453 	const __portpair ports = INET_COMBINED_PORTS(inet->inet_dport, lport);
454 	unsigned int hash = inet_ehashfn(net, daddr, lport,
455 					 saddr, inet->inet_dport);
456 	struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash);
457 	spinlock_t *lock = inet_ehash_lockp(hinfo, hash);
458 	struct sock *sk2;
459 	const struct hlist_nulls_node *node;
460 	struct inet_timewait_sock *tw = NULL;
461 
462 	spin_lock(lock);
463 
464 	sk_nulls_for_each(sk2, node, &head->chain) {
465 		if (sk2->sk_hash != hash)
466 			continue;
467 
468 		if (likely(INET_MATCH(sk2, net, acookie,
469 					 saddr, daddr, ports, dif, sdif))) {
470 			if (sk2->sk_state == TCP_TIME_WAIT) {
471 				tw = inet_twsk(sk2);
472 				if (twsk_unique(sk, sk2, twp))
473 					break;
474 			}
475 			goto not_unique;
476 		}
477 	}
478 
479 	/* Must record num and sport now. Otherwise we will see
480 	 * in hash table socket with a funny identity.
481 	 */
482 	inet->inet_num = lport;
483 	inet->inet_sport = htons(lport);
484 	sk->sk_hash = hash;
485 	WARN_ON(!sk_unhashed(sk));
486 	__sk_nulls_add_node_rcu(sk, &head->chain);
487 	if (tw) {
488 		sk_nulls_del_node_init_rcu((struct sock *)tw);
489 		__NET_INC_STATS(net, LINUX_MIB_TIMEWAITRECYCLED);
490 	}
491 	spin_unlock(lock);
492 	sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
493 
494 	if (twp) {
495 		*twp = tw;
496 	} else if (tw) {
497 		/* Silly. Should hash-dance instead... */
498 		inet_twsk_deschedule_put(tw);
499 	}
500 	return 0;
501 
502 not_unique:
503 	spin_unlock(lock);
504 	return -EADDRNOTAVAIL;
505 }
506 
inet_sk_port_offset(const struct sock * sk)507 static u32 inet_sk_port_offset(const struct sock *sk)
508 {
509 	const struct inet_sock *inet = inet_sk(sk);
510 
511 	return secure_ipv4_port_ephemeral(inet->inet_rcv_saddr,
512 					  inet->inet_daddr,
513 					  inet->inet_dport);
514 }
515 
516 /* Searches for an exsiting socket in the ehash bucket list.
517  * Returns true if found, false otherwise.
518  */
inet_ehash_lookup_by_sk(struct sock * sk,struct hlist_nulls_head * list)519 static bool inet_ehash_lookup_by_sk(struct sock *sk,
520 				    struct hlist_nulls_head *list)
521 {
522 	const __portpair ports = INET_COMBINED_PORTS(sk->sk_dport, sk->sk_num);
523 	const int sdif = sk->sk_bound_dev_if;
524 	const int dif = sk->sk_bound_dev_if;
525 	const struct hlist_nulls_node *node;
526 	struct net *net = sock_net(sk);
527 	struct sock *esk;
528 
529 	INET_ADDR_COOKIE(acookie, sk->sk_daddr, sk->sk_rcv_saddr);
530 
531 	sk_nulls_for_each_rcu(esk, node, list) {
532 		if (esk->sk_hash != sk->sk_hash)
533 			continue;
534 		if (sk->sk_family == AF_INET) {
535 			if (unlikely(INET_MATCH(esk, net, acookie,
536 						sk->sk_daddr,
537 						sk->sk_rcv_saddr,
538 						ports, dif, sdif))) {
539 				return true;
540 			}
541 		}
542 #if IS_ENABLED(CONFIG_IPV6)
543 		else if (sk->sk_family == AF_INET6) {
544 			if (unlikely(INET6_MATCH(esk, net,
545 						 &sk->sk_v6_daddr,
546 						 &sk->sk_v6_rcv_saddr,
547 						 ports, dif, sdif))) {
548 				return true;
549 			}
550 		}
551 #endif
552 	}
553 	return false;
554 }
555 
556 /* Insert a socket into ehash, and eventually remove another one
557  * (The another one can be a SYN_RECV or TIMEWAIT)
558  * If an existing socket already exists, socket sk is not inserted,
559  * and sets found_dup_sk parameter to true.
560  */
inet_ehash_insert(struct sock * sk,struct sock * osk,bool * found_dup_sk)561 bool inet_ehash_insert(struct sock *sk, struct sock *osk, bool *found_dup_sk)
562 {
563 	struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
564 	struct hlist_nulls_head *list;
565 	struct inet_ehash_bucket *head;
566 	spinlock_t *lock;
567 	bool ret = true;
568 
569 	WARN_ON_ONCE(!sk_unhashed(sk));
570 
571 	sk->sk_hash = sk_ehashfn(sk);
572 	head = inet_ehash_bucket(hashinfo, sk->sk_hash);
573 	list = &head->chain;
574 	lock = inet_ehash_lockp(hashinfo, sk->sk_hash);
575 
576 	spin_lock(lock);
577 	if (osk) {
578 		WARN_ON_ONCE(sk->sk_hash != osk->sk_hash);
579 		ret = sk_nulls_del_node_init_rcu(osk);
580 	} else if (found_dup_sk) {
581 		*found_dup_sk = inet_ehash_lookup_by_sk(sk, list);
582 		if (*found_dup_sk)
583 			ret = false;
584 	}
585 
586 	if (ret)
587 		__sk_nulls_add_node_rcu(sk, list);
588 
589 	spin_unlock(lock);
590 
591 	return ret;
592 }
593 
inet_ehash_nolisten(struct sock * sk,struct sock * osk,bool * found_dup_sk)594 bool inet_ehash_nolisten(struct sock *sk, struct sock *osk, bool *found_dup_sk)
595 {
596 	bool ok = inet_ehash_insert(sk, osk, found_dup_sk);
597 
598 	if (ok) {
599 		sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
600 	} else {
601 		this_cpu_inc(*sk->sk_prot->orphan_count);
602 		inet_sk_set_state(sk, TCP_CLOSE);
603 		sock_set_flag(sk, SOCK_DEAD);
604 		inet_csk_destroy_sock(sk);
605 	}
606 	return ok;
607 }
608 EXPORT_SYMBOL_GPL(inet_ehash_nolisten);
609 
inet_reuseport_add_sock(struct sock * sk,struct inet_listen_hashbucket * ilb)610 static int inet_reuseport_add_sock(struct sock *sk,
611 				   struct inet_listen_hashbucket *ilb)
612 {
613 	struct inet_bind_bucket *tb = inet_csk(sk)->icsk_bind_hash;
614 	const struct hlist_nulls_node *node;
615 	struct sock *sk2;
616 	kuid_t uid = sock_i_uid(sk);
617 
618 	sk_nulls_for_each_rcu(sk2, node, &ilb->nulls_head) {
619 		if (sk2 != sk &&
620 		    sk2->sk_family == sk->sk_family &&
621 		    ipv6_only_sock(sk2) == ipv6_only_sock(sk) &&
622 		    sk2->sk_bound_dev_if == sk->sk_bound_dev_if &&
623 		    inet_csk(sk2)->icsk_bind_hash == tb &&
624 		    sk2->sk_reuseport && uid_eq(uid, sock_i_uid(sk2)) &&
625 		    inet_rcv_saddr_equal(sk, sk2, false))
626 			return reuseport_add_sock(sk, sk2,
627 						  inet_rcv_saddr_any(sk));
628 	}
629 
630 	return reuseport_alloc(sk, inet_rcv_saddr_any(sk));
631 }
632 
__inet_hash(struct sock * sk,struct sock * osk)633 int __inet_hash(struct sock *sk, struct sock *osk)
634 {
635 	struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
636 	struct inet_listen_hashbucket *ilb;
637 	int err = 0;
638 
639 	if (sk->sk_state != TCP_LISTEN) {
640 		inet_ehash_nolisten(sk, osk, NULL);
641 		return 0;
642 	}
643 	WARN_ON(!sk_unhashed(sk));
644 	ilb = &hashinfo->listening_hash[inet_sk_listen_hashfn(sk)];
645 
646 	spin_lock(&ilb->lock);
647 	if (sk->sk_reuseport) {
648 		err = inet_reuseport_add_sock(sk, ilb);
649 		if (err)
650 			goto unlock;
651 	}
652 	if (IS_ENABLED(CONFIG_IPV6) && sk->sk_reuseport &&
653 		sk->sk_family == AF_INET6)
654 		__sk_nulls_add_node_tail_rcu(sk, &ilb->nulls_head);
655 	else
656 		__sk_nulls_add_node_rcu(sk, &ilb->nulls_head);
657 	inet_hash2(hashinfo, sk);
658 	ilb->count++;
659 	sock_set_flag(sk, SOCK_RCU_FREE);
660 	sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
661 unlock:
662 	spin_unlock(&ilb->lock);
663 
664 	return err;
665 }
666 EXPORT_SYMBOL(__inet_hash);
667 
inet_hash(struct sock * sk)668 int inet_hash(struct sock *sk)
669 {
670 	int err = 0;
671 
672 	if (sk->sk_state != TCP_CLOSE) {
673 		local_bh_disable();
674 		err = __inet_hash(sk, NULL);
675 		local_bh_enable();
676 	}
677 
678 	return err;
679 }
680 EXPORT_SYMBOL_GPL(inet_hash);
681 
inet_unhash(struct sock * sk)682 void inet_unhash(struct sock *sk)
683 {
684 	struct inet_hashinfo *hashinfo = sk->sk_prot->h.hashinfo;
685 	struct inet_listen_hashbucket *ilb = NULL;
686 	spinlock_t *lock;
687 
688 	if (sk_unhashed(sk))
689 		return;
690 
691 	if (sk->sk_state == TCP_LISTEN) {
692 		ilb = &hashinfo->listening_hash[inet_sk_listen_hashfn(sk)];
693 		lock = &ilb->lock;
694 	} else {
695 		lock = inet_ehash_lockp(hashinfo, sk->sk_hash);
696 	}
697 	spin_lock_bh(lock);
698 	if (sk_unhashed(sk))
699 		goto unlock;
700 
701 	if (rcu_access_pointer(sk->sk_reuseport_cb))
702 		reuseport_stop_listen_sock(sk);
703 	if (ilb) {
704 		inet_unhash2(hashinfo, sk);
705 		ilb->count--;
706 	}
707 	__sk_nulls_del_node_init_rcu(sk);
708 	sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
709 unlock:
710 	spin_unlock_bh(lock);
711 }
712 EXPORT_SYMBOL_GPL(inet_unhash);
713 
714 /* RFC 6056 3.3.4.  Algorithm 4: Double-Hash Port Selection Algorithm
715  * Note that we use 32bit integers (vs RFC 'short integers')
716  * because 2^16 is not a multiple of num_ephemeral and this
717  * property might be used by clever attacker.
718  * RFC claims using TABLE_LENGTH=10 buckets gives an improvement,
719  * we use 256 instead to really give more isolation and
720  * privacy, this only consumes 1 KB of kernel memory.
721  */
722 #define INET_TABLE_PERTURB_SHIFT 8
723 static u32 table_perturb[1 << INET_TABLE_PERTURB_SHIFT];
724 
__inet_hash_connect(struct inet_timewait_death_row * death_row,struct sock * sk,u32 port_offset,int (* check_established)(struct inet_timewait_death_row *,struct sock *,__u16,struct inet_timewait_sock **))725 int __inet_hash_connect(struct inet_timewait_death_row *death_row,
726 		struct sock *sk, u32 port_offset,
727 		int (*check_established)(struct inet_timewait_death_row *,
728 			struct sock *, __u16, struct inet_timewait_sock **))
729 {
730 	struct inet_hashinfo *hinfo = death_row->hashinfo;
731 	struct inet_timewait_sock *tw = NULL;
732 	struct inet_bind_hashbucket *head;
733 	int port = inet_sk(sk)->inet_num;
734 	struct net *net = sock_net(sk);
735 	struct inet_bind_bucket *tb;
736 	u32 remaining, offset;
737 	int ret, i, low, high;
738 	int l3mdev;
739 	u32 index;
740 
741 	if (port) {
742 		head = &hinfo->bhash[inet_bhashfn(net, port,
743 						  hinfo->bhash_size)];
744 		tb = inet_csk(sk)->icsk_bind_hash;
745 		spin_lock_bh(&head->lock);
746 		if (sk_head(&tb->owners) == sk && !sk->sk_bind_node.next) {
747 			inet_ehash_nolisten(sk, NULL, NULL);
748 			spin_unlock_bh(&head->lock);
749 			return 0;
750 		}
751 		spin_unlock(&head->lock);
752 		/* No definite answer... Walk to established hash table */
753 		ret = check_established(death_row, sk, port, NULL);
754 		local_bh_enable();
755 		return ret;
756 	}
757 
758 	l3mdev = inet_sk_bound_l3mdev(sk);
759 
760 	inet_get_local_port_range(net, &low, &high);
761 	high++; /* [32768, 60999] -> [32768, 61000[ */
762 	remaining = high - low;
763 	if (likely(remaining > 1))
764 		remaining &= ~1U;
765 
766 	net_get_random_once(table_perturb, sizeof(table_perturb));
767 	index = hash_32(port_offset, INET_TABLE_PERTURB_SHIFT);
768 
769 	offset = (READ_ONCE(table_perturb[index]) + port_offset) % remaining;
770 	/* In first pass we try ports of @low parity.
771 	 * inet_csk_get_port() does the opposite choice.
772 	 */
773 	offset &= ~1U;
774 other_parity_scan:
775 	port = low + offset;
776 	for (i = 0; i < remaining; i += 2, port += 2) {
777 		if (unlikely(port >= high))
778 			port -= remaining;
779 		if (inet_is_local_reserved_port(net, port))
780 			continue;
781 		head = &hinfo->bhash[inet_bhashfn(net, port,
782 						  hinfo->bhash_size)];
783 		spin_lock_bh(&head->lock);
784 
785 		/* Does not bother with rcv_saddr checks, because
786 		 * the established check is already unique enough.
787 		 */
788 		inet_bind_bucket_for_each(tb, &head->chain) {
789 			if (net_eq(ib_net(tb), net) && tb->l3mdev == l3mdev &&
790 			    tb->port == port) {
791 				if (tb->fastreuse >= 0 ||
792 				    tb->fastreuseport >= 0)
793 					goto next_port;
794 				WARN_ON(hlist_empty(&tb->owners));
795 				if (!check_established(death_row, sk,
796 						       port, &tw))
797 					goto ok;
798 				goto next_port;
799 			}
800 		}
801 
802 		tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep,
803 					     net, head, port, l3mdev);
804 		if (!tb) {
805 			spin_unlock_bh(&head->lock);
806 			return -ENOMEM;
807 		}
808 		tb->fastreuse = -1;
809 		tb->fastreuseport = -1;
810 		goto ok;
811 next_port:
812 		spin_unlock_bh(&head->lock);
813 		cond_resched();
814 	}
815 
816 	offset++;
817 	if ((offset & 1) && remaining > 1)
818 		goto other_parity_scan;
819 
820 	return -EADDRNOTAVAIL;
821 
822 ok:
823 	/* If our first attempt found a candidate, skip next candidate
824 	 * in 1/16 of cases to add some noise.
825 	 */
826 	if (!i && !(prandom_u32() % 16))
827 		i = 2;
828 	WRITE_ONCE(table_perturb[index], READ_ONCE(table_perturb[index]) + i + 2);
829 
830 	/* Head lock still held and bh's disabled */
831 	inet_bind_hash(sk, tb, port);
832 	if (sk_unhashed(sk)) {
833 		inet_sk(sk)->inet_sport = htons(port);
834 		inet_ehash_nolisten(sk, (struct sock *)tw, NULL);
835 	}
836 	if (tw)
837 		inet_twsk_bind_unhash(tw, hinfo);
838 	spin_unlock(&head->lock);
839 	if (tw)
840 		inet_twsk_deschedule_put(tw);
841 	local_bh_enable();
842 	return 0;
843 }
844 
845 /*
846  * Bind a port for a connect operation and hash it.
847  */
inet_hash_connect(struct inet_timewait_death_row * death_row,struct sock * sk)848 int inet_hash_connect(struct inet_timewait_death_row *death_row,
849 		      struct sock *sk)
850 {
851 	u32 port_offset = 0;
852 
853 	if (!inet_sk(sk)->inet_num)
854 		port_offset = inet_sk_port_offset(sk);
855 	return __inet_hash_connect(death_row, sk, port_offset,
856 				   __inet_check_established);
857 }
858 EXPORT_SYMBOL_GPL(inet_hash_connect);
859 
inet_hashinfo_init(struct inet_hashinfo * h)860 void inet_hashinfo_init(struct inet_hashinfo *h)
861 {
862 	int i;
863 
864 	for (i = 0; i < INET_LHTABLE_SIZE; i++) {
865 		spin_lock_init(&h->listening_hash[i].lock);
866 		INIT_HLIST_NULLS_HEAD(&h->listening_hash[i].nulls_head,
867 				      i + LISTENING_NULLS_BASE);
868 		h->listening_hash[i].count = 0;
869 	}
870 
871 	h->lhash2 = NULL;
872 }
873 EXPORT_SYMBOL_GPL(inet_hashinfo_init);
874 
init_hashinfo_lhash2(struct inet_hashinfo * h)875 static void init_hashinfo_lhash2(struct inet_hashinfo *h)
876 {
877 	int i;
878 
879 	for (i = 0; i <= h->lhash2_mask; i++) {
880 		spin_lock_init(&h->lhash2[i].lock);
881 		INIT_HLIST_HEAD(&h->lhash2[i].head);
882 		h->lhash2[i].count = 0;
883 	}
884 }
885 
inet_hashinfo2_init(struct inet_hashinfo * h,const char * name,unsigned long numentries,int scale,unsigned long low_limit,unsigned long high_limit)886 void __init inet_hashinfo2_init(struct inet_hashinfo *h, const char *name,
887 				unsigned long numentries, int scale,
888 				unsigned long low_limit,
889 				unsigned long high_limit)
890 {
891 	h->lhash2 = alloc_large_system_hash(name,
892 					    sizeof(*h->lhash2),
893 					    numentries,
894 					    scale,
895 					    0,
896 					    NULL,
897 					    &h->lhash2_mask,
898 					    low_limit,
899 					    high_limit);
900 	init_hashinfo_lhash2(h);
901 }
902 
inet_hashinfo2_init_mod(struct inet_hashinfo * h)903 int inet_hashinfo2_init_mod(struct inet_hashinfo *h)
904 {
905 	h->lhash2 = kmalloc_array(INET_LHTABLE_SIZE, sizeof(*h->lhash2), GFP_KERNEL);
906 	if (!h->lhash2)
907 		return -ENOMEM;
908 
909 	h->lhash2_mask = INET_LHTABLE_SIZE - 1;
910 	/* INET_LHTABLE_SIZE must be a power of 2 */
911 	BUG_ON(INET_LHTABLE_SIZE & h->lhash2_mask);
912 
913 	init_hashinfo_lhash2(h);
914 	return 0;
915 }
916 EXPORT_SYMBOL_GPL(inet_hashinfo2_init_mod);
917 
inet_ehash_locks_alloc(struct inet_hashinfo * hashinfo)918 int inet_ehash_locks_alloc(struct inet_hashinfo *hashinfo)
919 {
920 	unsigned int locksz = sizeof(spinlock_t);
921 	unsigned int i, nblocks = 1;
922 
923 	if (locksz != 0) {
924 		/* allocate 2 cache lines or at least one spinlock per cpu */
925 		nblocks = max(2U * L1_CACHE_BYTES / locksz, 1U);
926 		nblocks = roundup_pow_of_two(nblocks * num_possible_cpus());
927 
928 		/* no more locks than number of hash buckets */
929 		nblocks = min(nblocks, hashinfo->ehash_mask + 1);
930 
931 		hashinfo->ehash_locks = kvmalloc_array(nblocks, locksz, GFP_KERNEL);
932 		if (!hashinfo->ehash_locks)
933 			return -ENOMEM;
934 
935 		for (i = 0; i < nblocks; i++)
936 			spin_lock_init(&hashinfo->ehash_locks[i]);
937 	}
938 	hashinfo->ehash_locks_mask = nblocks - 1;
939 	return 0;
940 }
941 EXPORT_SYMBOL_GPL(inet_ehash_locks_alloc);
942