inet6_hashtables.c

  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 INET6 transport hashtables
  8 *
  9 * Authors:	Lotsa people, from code originally in tcp, generalised here
 10 *		by Arnaldo Carvalho de Melo <acme@mandriva.com>
 11 */
 12
 13#include <linux/module.h>
 14#include <linux/random.h>
 15
 16#include <net/addrconf.h>
 17#include <net/hotdata.h>
 18#include <net/inet_connection_sock.h>
 19#include <net/inet_hashtables.h>
 20#include <net/inet6_hashtables.h>
 21#include <net/secure_seq.h>
 22#include <net/ip.h>
 23#include <net/sock_reuseport.h>
 24#include <net/tcp.h>
 25
 26u32 inet6_ehashfn(const struct net *net,
 27		  const struct in6_addr *laddr, const u16 lport,
 28		  const struct in6_addr *faddr, const __be16 fport)
 29{
 30	u32 lhash, fhash;
 31
 32	net_get_random_once(&inet6_ehash_secret, sizeof(inet6_ehash_secret));
 33	net_get_random_once(&tcp_ipv6_hash_secret, sizeof(tcp_ipv6_hash_secret));
 34
 35	lhash = (__force u32)laddr->s6_addr32[3];
 36	fhash = __ipv6_addr_jhash(faddr, tcp_ipv6_hash_secret);
 37
 38	return lport + __inet6_ehashfn(lhash, 0, fhash, fport,
 39				       inet6_ehash_secret + net_hash_mix(net));
 40}
 41EXPORT_SYMBOL_GPL(inet6_ehashfn);
 42
 43/*
 44 * Sockets in TCP_CLOSE state are _always_ taken out of the hash, so
 45 * we need not check it for TCP lookups anymore, thanks Alexey. -DaveM
 46 *
 47 * The sockhash lock must be held as a reader here.
 48 */
 49struct sock *__inet6_lookup_established(const struct net *net,
 50					const struct in6_addr *saddr,
 51					const __be16 sport,
 52					const struct in6_addr *daddr,
 53					const u16 hnum,
 54					const int dif, const int sdif)
 55{
 56	const __portpair ports = INET_COMBINED_PORTS(sport, hnum);
 57	const struct hlist_nulls_node *node;
 58	struct inet_ehash_bucket *head;
 59	struct inet_hashinfo *hashinfo;
 60	unsigned int hash, slot;
 61	struct sock *sk;
 62
 63	hashinfo = net->ipv4.tcp_death_row.hashinfo;
 64	hash = inet6_ehashfn(net, daddr, hnum, saddr, sport);
 65	slot = hash & hashinfo->ehash_mask;
 66	head = &hashinfo->ehash[slot];
 67begin:
 68	sk_nulls_for_each_rcu(sk, node, &head->chain) {
 69		if (sk->sk_hash != hash)
 70			continue;
 71		if (!inet6_match(net, sk, saddr, daddr, ports, dif, sdif))
 72			continue;
 73		if (unlikely(!refcount_inc_not_zero(&sk->sk_refcnt)))
 74			goto out;
 75
 76		if (unlikely(!inet6_match(net, sk, saddr, daddr, ports, dif, sdif))) {
 77			sock_gen_put(sk);
 78			goto begin;
 79		}
 80		goto found;
 81	}
 82	if (get_nulls_value(node) != slot)
 83		goto begin;
 84out:
 85	sk = NULL;
 86found:
 87	return sk;
 88}
 89EXPORT_SYMBOL(__inet6_lookup_established);
 90
 91static inline int compute_score(struct sock *sk, const struct net *net,
 92				const unsigned short hnum,
 93				const struct in6_addr *daddr,
 94				const int dif, const int sdif)
 95{
 96	int score = -1;
 97
 98	if (net_eq(sock_net(sk), net) &&
 99	    READ_ONCE(inet_sk(sk)->inet_num) == hnum &&
100	    sk->sk_family == PF_INET6) {
101		if (!ipv6_addr_equal(&sk->sk_v6_rcv_saddr, daddr))
102			return -1;
103
104		if (!inet_sk_bound_dev_eq(net, sk->sk_bound_dev_if, dif, sdif))
105			return -1;
106
107		score =  sk->sk_bound_dev_if ? 2 : 1;
108		if (READ_ONCE(sk->sk_incoming_cpu) == raw_smp_processor_id())
109			score++;
110	}
111	return score;
112}
113
114/**
115 * inet6_lookup_reuseport() - execute reuseport logic on AF_INET6 socket if necessary.
116 * @net: network namespace.
117 * @sk: AF_INET6 socket, must be in TCP_LISTEN state for TCP or TCP_CLOSE for UDP.
118 * @skb: context for a potential SK_REUSEPORT program.
119 * @doff: header offset.
120 * @saddr: source address.
121 * @sport: source port.
122 * @daddr: destination address.
123 * @hnum: destination port in host byte order.
124 * @ehashfn: hash function used to generate the fallback hash.
125 *
126 * Return: NULL if sk doesn't have SO_REUSEPORT set, otherwise a pointer to
127 *         the selected sock or an error.
128 */
129struct sock *inet6_lookup_reuseport(const struct net *net, struct sock *sk,
130				    struct sk_buff *skb, int doff,
131				    const struct in6_addr *saddr,
132				    __be16 sport,
133				    const struct in6_addr *daddr,
134				    unsigned short hnum,
135				    inet6_ehashfn_t *ehashfn)
136{
137	struct sock *reuse_sk = NULL;
138	u32 phash;
139
140	if (sk->sk_reuseport) {
141		phash = INDIRECT_CALL_INET(ehashfn, udp6_ehashfn, inet6_ehashfn,
142					   net, daddr, hnum, saddr, sport);
143		reuse_sk = reuseport_select_sock(sk, phash, skb, doff);
144	}
145	return reuse_sk;
146}
147EXPORT_SYMBOL_GPL(inet6_lookup_reuseport);
148
149/* called with rcu_read_lock() */
150static struct sock *inet6_lhash2_lookup(const struct net *net,
151		struct inet_listen_hashbucket *ilb2,
152		struct sk_buff *skb, int doff,
153		const struct in6_addr *saddr,
154		const __be16 sport, const struct in6_addr *daddr,
155		const unsigned short hnum, const int dif, const int sdif)
156{
157	struct sock *sk, *result = NULL;
158	struct hlist_nulls_node *node;
159	int score, hiscore = 0;
160
161	sk_nulls_for_each_rcu(sk, node, &ilb2->nulls_head) {
162		score = compute_score(sk, net, hnum, daddr, dif, sdif);
163		if (score > hiscore) {
164			result = inet6_lookup_reuseport(net, sk, skb, doff,
165							saddr, sport, daddr, hnum, inet6_ehashfn);
166			if (result)
167				return result;
168
169			result = sk;
170			hiscore = score;
171		}
172	}
173
174	return result;
175}
176
177struct sock *inet6_lookup_run_sk_lookup(const struct net *net,
178					int protocol,
179					struct sk_buff *skb, int doff,
180					const struct in6_addr *saddr,
181					const __be16 sport,
182					const struct in6_addr *daddr,
183					const u16 hnum, const int dif,
184					inet6_ehashfn_t *ehashfn)
185{
186	struct sock *sk, *reuse_sk;
187	bool no_reuseport;
188
189	no_reuseport = bpf_sk_lookup_run_v6(net, protocol, saddr, sport,
190					    daddr, hnum, dif, &sk);
191	if (no_reuseport || IS_ERR_OR_NULL(sk))
192		return sk;
193
194	reuse_sk = inet6_lookup_reuseport(net, sk, skb, doff,
195					  saddr, sport, daddr, hnum, ehashfn);
196	if (reuse_sk)
197		sk = reuse_sk;
198	return sk;
199}
200EXPORT_SYMBOL_GPL(inet6_lookup_run_sk_lookup);
201
202struct sock *inet6_lookup_listener(const struct net *net,
203				   struct sk_buff *skb, int doff,
204				   const struct in6_addr *saddr,
205				   const __be16 sport,
206				   const struct in6_addr *daddr,
207				   const unsigned short hnum,
208				   const int dif, const int sdif)
209{
210	struct inet_listen_hashbucket *ilb2;
211	struct inet_hashinfo *hashinfo;
212	struct sock *result = NULL;
213	unsigned int hash2;
214
215	/* Lookup redirect from BPF */
216	if (static_branch_unlikely(&bpf_sk_lookup_enabled)) {
217		result = inet6_lookup_run_sk_lookup(net, IPPROTO_TCP, skb, doff,
218						    saddr, sport, daddr, hnum, dif,
219						    inet6_ehashfn);
220		if (result)
221			goto done;
222	}
223
224	hashinfo = net->ipv4.tcp_death_row.hashinfo;
225	hash2 = ipv6_portaddr_hash(net, daddr, hnum);
226	ilb2 = inet_lhash2_bucket(hashinfo, hash2);
227
228	result = inet6_lhash2_lookup(net, ilb2, skb, doff,
229				     saddr, sport, daddr, hnum,
230				     dif, sdif);
231	if (result)
232		goto done;
233
234	/* Lookup lhash2 with in6addr_any */
235	hash2 = ipv6_portaddr_hash(net, &in6addr_any, hnum);
236	ilb2 = inet_lhash2_bucket(hashinfo, hash2);
237
238	result = inet6_lhash2_lookup(net, ilb2, skb, doff,
239				     saddr, sport, &in6addr_any, hnum,
240				     dif, sdif);
241done:
242	if (IS_ERR(result))
243		return NULL;
244	return result;
245}
246EXPORT_SYMBOL_GPL(inet6_lookup_listener);
247
248struct sock *inet6_lookup(const struct net *net,
249			  struct sk_buff *skb, int doff,
250			  const struct in6_addr *saddr, const __be16 sport,
251			  const struct in6_addr *daddr, const __be16 dport,
252			  const int dif)
253{
254	struct sock *sk;
255	bool refcounted;
256
257	sk = __inet6_lookup(net, skb, doff, saddr, sport, daddr,
258			    ntohs(dport), dif, 0, &refcounted);
259	if (sk && !refcounted && !refcount_inc_not_zero(&sk->sk_refcnt))
260		sk = NULL;
261	return sk;
262}
263EXPORT_SYMBOL_GPL(inet6_lookup);
264
265static int __inet6_check_established(struct inet_timewait_death_row *death_row,
266				     struct sock *sk, const __u16 lport,
267				     struct inet_timewait_sock **twp,
268				     bool rcu_lookup,
269				     u32 hash)
270{
271	struct inet_hashinfo *hinfo = death_row->hashinfo;
272	struct inet_sock *inet = inet_sk(sk);
273	const struct in6_addr *daddr = &sk->sk_v6_rcv_saddr;
274	const struct in6_addr *saddr = &sk->sk_v6_daddr;
275	const int dif = sk->sk_bound_dev_if;
276	struct net *net = sock_net(sk);
277	const int sdif = l3mdev_master_ifindex_by_index(net, dif);
278	const __portpair ports = INET_COMBINED_PORTS(inet->inet_dport, lport);
279	struct inet_ehash_bucket *head = inet_ehash_bucket(hinfo, hash);
280	struct inet_timewait_sock *tw = NULL;
281	const struct hlist_nulls_node *node;
282	struct sock *sk2;
283	spinlock_t *lock;
284
285	if (rcu_lookup) {
286		sk_nulls_for_each(sk2, node, &head->chain) {
287			if (sk2->sk_hash != hash ||
288			    !inet6_match(net, sk2, saddr, daddr,
289					 ports, dif, sdif))
290				continue;
291			if (sk2->sk_state == TCP_TIME_WAIT)
292				break;
293			return -EADDRNOTAVAIL;
294		}
295		return 0;
296	}
297
298	lock = inet_ehash_lockp(hinfo, hash);
299	spin_lock(lock);
300
301	sk_nulls_for_each(sk2, node, &head->chain) {
302		if (sk2->sk_hash != hash)
303			continue;
304
305		if (likely(inet6_match(net, sk2, saddr, daddr, ports,
306				       dif, sdif))) {
307			if (sk2->sk_state == TCP_TIME_WAIT) {
308				tw = inet_twsk(sk2);
309				if (tcp_twsk_unique(sk, sk2, twp))
310					break;
311			}
312			goto not_unique;
313		}
314	}
315
316	/* Must record num and sport now. Otherwise we will see
317	 * in hash table socket with a funny identity.
318	 */
319	inet->inet_num = lport;
320	inet->inet_sport = htons(lport);
321	sk->sk_hash = hash;
322	WARN_ON(!sk_unhashed(sk));
323	__sk_nulls_add_node_rcu(sk, &head->chain);
324	if (tw) {
325		sk_nulls_del_node_init_rcu((struct sock *)tw);
326		__NET_INC_STATS(net, LINUX_MIB_TIMEWAITRECYCLED);
327	}
328	spin_unlock(lock);
329	sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
330
331	if (twp) {
332		*twp = tw;
333	} else if (tw) {
334		/* Silly. Should hash-dance instead... */
335		inet_twsk_deschedule_put(tw);
336	}
337	return 0;
338
339not_unique:
340	spin_unlock(lock);
341	return -EADDRNOTAVAIL;
342}
343
344static u64 inet6_sk_port_offset(const struct sock *sk)
345{
346	const struct inet_sock *inet = inet_sk(sk);
347
348	return secure_ipv6_port_ephemeral(sk->sk_v6_rcv_saddr.s6_addr32,
349					  sk->sk_v6_daddr.s6_addr32,
350					  inet->inet_dport);
351}
352
353int inet6_hash_connect(struct inet_timewait_death_row *death_row,
354		       struct sock *sk)
355{
356	const struct in6_addr *daddr = &sk->sk_v6_rcv_saddr;
357	const struct in6_addr *saddr = &sk->sk_v6_daddr;
358	const struct inet_sock *inet = inet_sk(sk);
359	const struct net *net = sock_net(sk);
360	u64 port_offset = 0;
361	u32 hash_port0;
362
363	if (!inet_sk(sk)->inet_num)
364		port_offset = inet6_sk_port_offset(sk);
365
366	hash_port0 = inet6_ehashfn(net, daddr, 0, saddr, inet->inet_dport);
367
368	return __inet_hash_connect(death_row, sk, port_offset, hash_port0,
369				   __inet6_check_established);
370}
371EXPORT_SYMBOL_GPL(inet6_hash_connect);