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 * Support for INET connection oriented protocols.
8 *
9 * Authors: See the TCP sources
10 */
11
12 #include <linux/module.h>
13 #include <linux/jhash.h>
14
15 #include <net/inet_connection_sock.h>
16 #include <net/inet_hashtables.h>
17 #include <net/inet_timewait_sock.h>
18 #include <net/ip.h>
19 #include <net/route.h>
20 #include <net/tcp_states.h>
21 #include <net/xfrm.h>
22 #include <net/tcp.h>
23 #include <net/sock_reuseport.h>
24 #include <net/addrconf.h>
25 #include <trace/hooks/net.h>
26
27 #if IS_ENABLED(CONFIG_IPV6)
28 /* match_sk*_wildcard == true: IPV6_ADDR_ANY equals to any IPv6 addresses
29 * if IPv6 only, and any IPv4 addresses
30 * if not IPv6 only
31 * match_sk*_wildcard == false: addresses must be exactly the same, i.e.
32 * IPV6_ADDR_ANY only equals to IPV6_ADDR_ANY,
33 * and 0.0.0.0 equals to 0.0.0.0 only
34 */
ipv6_rcv_saddr_equal(const struct in6_addr * sk1_rcv_saddr6,const struct in6_addr * sk2_rcv_saddr6,__be32 sk1_rcv_saddr,__be32 sk2_rcv_saddr,bool sk1_ipv6only,bool sk2_ipv6only,bool match_sk1_wildcard,bool match_sk2_wildcard)35 static bool ipv6_rcv_saddr_equal(const struct in6_addr *sk1_rcv_saddr6,
36 const struct in6_addr *sk2_rcv_saddr6,
37 __be32 sk1_rcv_saddr, __be32 sk2_rcv_saddr,
38 bool sk1_ipv6only, bool sk2_ipv6only,
39 bool match_sk1_wildcard,
40 bool match_sk2_wildcard)
41 {
42 int addr_type = ipv6_addr_type(sk1_rcv_saddr6);
43 int addr_type2 = sk2_rcv_saddr6 ? ipv6_addr_type(sk2_rcv_saddr6) : IPV6_ADDR_MAPPED;
44
45 /* if both are mapped, treat as IPv4 */
46 if (addr_type == IPV6_ADDR_MAPPED && addr_type2 == IPV6_ADDR_MAPPED) {
47 if (!sk2_ipv6only) {
48 if (sk1_rcv_saddr == sk2_rcv_saddr)
49 return true;
50 return (match_sk1_wildcard && !sk1_rcv_saddr) ||
51 (match_sk2_wildcard && !sk2_rcv_saddr);
52 }
53 return false;
54 }
55
56 if (addr_type == IPV6_ADDR_ANY && addr_type2 == IPV6_ADDR_ANY)
57 return true;
58
59 if (addr_type2 == IPV6_ADDR_ANY && match_sk2_wildcard &&
60 !(sk2_ipv6only && addr_type == IPV6_ADDR_MAPPED))
61 return true;
62
63 if (addr_type == IPV6_ADDR_ANY && match_sk1_wildcard &&
64 !(sk1_ipv6only && addr_type2 == IPV6_ADDR_MAPPED))
65 return true;
66
67 if (sk2_rcv_saddr6 &&
68 ipv6_addr_equal(sk1_rcv_saddr6, sk2_rcv_saddr6))
69 return true;
70
71 return false;
72 }
73 #endif
74
75 /* match_sk*_wildcard == true: 0.0.0.0 equals to any IPv4 addresses
76 * match_sk*_wildcard == false: addresses must be exactly the same, i.e.
77 * 0.0.0.0 only equals to 0.0.0.0
78 */
ipv4_rcv_saddr_equal(__be32 sk1_rcv_saddr,__be32 sk2_rcv_saddr,bool sk2_ipv6only,bool match_sk1_wildcard,bool match_sk2_wildcard)79 static bool ipv4_rcv_saddr_equal(__be32 sk1_rcv_saddr, __be32 sk2_rcv_saddr,
80 bool sk2_ipv6only, bool match_sk1_wildcard,
81 bool match_sk2_wildcard)
82 {
83 if (!sk2_ipv6only) {
84 if (sk1_rcv_saddr == sk2_rcv_saddr)
85 return true;
86 return (match_sk1_wildcard && !sk1_rcv_saddr) ||
87 (match_sk2_wildcard && !sk2_rcv_saddr);
88 }
89 return false;
90 }
91
inet_rcv_saddr_equal(const struct sock * sk,const struct sock * sk2,bool match_wildcard)92 bool inet_rcv_saddr_equal(const struct sock *sk, const struct sock *sk2,
93 bool match_wildcard)
94 {
95 #if IS_ENABLED(CONFIG_IPV6)
96 if (sk->sk_family == AF_INET6)
97 return ipv6_rcv_saddr_equal(&sk->sk_v6_rcv_saddr,
98 inet6_rcv_saddr(sk2),
99 sk->sk_rcv_saddr,
100 sk2->sk_rcv_saddr,
101 ipv6_only_sock(sk),
102 ipv6_only_sock(sk2),
103 match_wildcard,
104 match_wildcard);
105 #endif
106 return ipv4_rcv_saddr_equal(sk->sk_rcv_saddr, sk2->sk_rcv_saddr,
107 ipv6_only_sock(sk2), match_wildcard,
108 match_wildcard);
109 }
110 EXPORT_SYMBOL(inet_rcv_saddr_equal);
111
inet_rcv_saddr_any(const struct sock * sk)112 bool inet_rcv_saddr_any(const struct sock *sk)
113 {
114 #if IS_ENABLED(CONFIG_IPV6)
115 if (sk->sk_family == AF_INET6)
116 return ipv6_addr_any(&sk->sk_v6_rcv_saddr);
117 #endif
118 return !sk->sk_rcv_saddr;
119 }
120
inet_get_local_port_range(const struct net * net,int * low,int * high)121 void inet_get_local_port_range(const struct net *net, int *low, int *high)
122 {
123 unsigned int seq;
124
125 do {
126 seq = read_seqbegin(&net->ipv4.ip_local_ports.lock);
127
128 *low = net->ipv4.ip_local_ports.range[0];
129 *high = net->ipv4.ip_local_ports.range[1];
130 } while (read_seqretry(&net->ipv4.ip_local_ports.lock, seq));
131 }
132 EXPORT_SYMBOL(inet_get_local_port_range);
133
inet_sk_get_local_port_range(const struct sock * sk,int * low,int * high)134 void inet_sk_get_local_port_range(const struct sock *sk, int *low, int *high)
135 {
136 const struct inet_sock *inet = inet_sk(sk);
137 const struct net *net = sock_net(sk);
138 int lo, hi, sk_lo, sk_hi;
139
140 inet_get_local_port_range(net, &lo, &hi);
141
142 sk_lo = inet->local_port_range.lo;
143 sk_hi = inet->local_port_range.hi;
144
145 if (unlikely(lo <= sk_lo && sk_lo <= hi))
146 lo = sk_lo;
147 if (unlikely(lo <= sk_hi && sk_hi <= hi))
148 hi = sk_hi;
149
150 *low = lo;
151 *high = hi;
152 }
153 EXPORT_SYMBOL(inet_sk_get_local_port_range);
154
inet_use_bhash2_on_bind(const struct sock * sk)155 static bool inet_use_bhash2_on_bind(const struct sock *sk)
156 {
157 #if IS_ENABLED(CONFIG_IPV6)
158 if (sk->sk_family == AF_INET6) {
159 int addr_type = ipv6_addr_type(&sk->sk_v6_rcv_saddr);
160
161 return addr_type != IPV6_ADDR_ANY &&
162 addr_type != IPV6_ADDR_MAPPED;
163 }
164 #endif
165 return sk->sk_rcv_saddr != htonl(INADDR_ANY);
166 }
167
inet_bind_conflict(const struct sock * sk,struct sock * sk2,kuid_t sk_uid,bool relax,bool reuseport_cb_ok,bool reuseport_ok)168 static bool inet_bind_conflict(const struct sock *sk, struct sock *sk2,
169 kuid_t sk_uid, bool relax,
170 bool reuseport_cb_ok, bool reuseport_ok)
171 {
172 int bound_dev_if2;
173
174 if (sk == sk2)
175 return false;
176
177 bound_dev_if2 = READ_ONCE(sk2->sk_bound_dev_if);
178
179 if (!sk->sk_bound_dev_if || !bound_dev_if2 ||
180 sk->sk_bound_dev_if == bound_dev_if2) {
181 if (sk->sk_reuse && sk2->sk_reuse &&
182 sk2->sk_state != TCP_LISTEN) {
183 if (!relax || (!reuseport_ok && sk->sk_reuseport &&
184 sk2->sk_reuseport && reuseport_cb_ok &&
185 (sk2->sk_state == TCP_TIME_WAIT ||
186 uid_eq(sk_uid, sock_i_uid(sk2)))))
187 return true;
188 } else if (!reuseport_ok || !sk->sk_reuseport ||
189 !sk2->sk_reuseport || !reuseport_cb_ok ||
190 (sk2->sk_state != TCP_TIME_WAIT &&
191 !uid_eq(sk_uid, sock_i_uid(sk2)))) {
192 return true;
193 }
194 }
195 return false;
196 }
197
__inet_bhash2_conflict(const struct sock * sk,struct sock * sk2,kuid_t sk_uid,bool relax,bool reuseport_cb_ok,bool reuseport_ok)198 static bool __inet_bhash2_conflict(const struct sock *sk, struct sock *sk2,
199 kuid_t sk_uid, bool relax,
200 bool reuseport_cb_ok, bool reuseport_ok)
201 {
202 if (sk->sk_family == AF_INET && ipv6_only_sock(sk2))
203 return false;
204
205 return inet_bind_conflict(sk, sk2, sk_uid, relax,
206 reuseport_cb_ok, reuseport_ok);
207 }
208
inet_bhash2_conflict(const struct sock * sk,const struct inet_bind2_bucket * tb2,kuid_t sk_uid,bool relax,bool reuseport_cb_ok,bool reuseport_ok)209 static bool inet_bhash2_conflict(const struct sock *sk,
210 const struct inet_bind2_bucket *tb2,
211 kuid_t sk_uid,
212 bool relax, bool reuseport_cb_ok,
213 bool reuseport_ok)
214 {
215 struct inet_timewait_sock *tw2;
216 struct sock *sk2;
217
218 sk_for_each_bound_bhash2(sk2, &tb2->owners) {
219 if (__inet_bhash2_conflict(sk, sk2, sk_uid, relax,
220 reuseport_cb_ok, reuseport_ok))
221 return true;
222 }
223
224 twsk_for_each_bound_bhash2(tw2, &tb2->deathrow) {
225 sk2 = (struct sock *)tw2;
226
227 if (__inet_bhash2_conflict(sk, sk2, sk_uid, relax,
228 reuseport_cb_ok, reuseport_ok))
229 return true;
230 }
231
232 return false;
233 }
234
235 /* This should be called only when the tb and tb2 hashbuckets' locks are held */
inet_csk_bind_conflict(const struct sock * sk,const struct inet_bind_bucket * tb,const struct inet_bind2_bucket * tb2,bool relax,bool reuseport_ok)236 static int inet_csk_bind_conflict(const struct sock *sk,
237 const struct inet_bind_bucket *tb,
238 const struct inet_bind2_bucket *tb2, /* may be null */
239 bool relax, bool reuseport_ok)
240 {
241 bool reuseport_cb_ok;
242 struct sock_reuseport *reuseport_cb;
243 kuid_t uid = sock_i_uid((struct sock *)sk);
244
245 rcu_read_lock();
246 reuseport_cb = rcu_dereference(sk->sk_reuseport_cb);
247 /* paired with WRITE_ONCE() in __reuseport_(add|detach)_closed_sock */
248 reuseport_cb_ok = !reuseport_cb || READ_ONCE(reuseport_cb->num_closed_socks);
249 rcu_read_unlock();
250
251 /*
252 * Unlike other sk lookup places we do not check
253 * for sk_net here, since _all_ the socks listed
254 * in tb->owners and tb2->owners list belong
255 * to the same net - the one this bucket belongs to.
256 */
257
258 if (!inet_use_bhash2_on_bind(sk)) {
259 struct sock *sk2;
260
261 sk_for_each_bound(sk2, &tb->owners)
262 if (inet_bind_conflict(sk, sk2, uid, relax,
263 reuseport_cb_ok, reuseport_ok) &&
264 inet_rcv_saddr_equal(sk, sk2, true))
265 return true;
266
267 return false;
268 }
269
270 /* Conflicts with an existing IPV6_ADDR_ANY (if ipv6) or INADDR_ANY (if
271 * ipv4) should have been checked already. We need to do these two
272 * checks separately because their spinlocks have to be acquired/released
273 * independently of each other, to prevent possible deadlocks
274 */
275 return tb2 && inet_bhash2_conflict(sk, tb2, uid, relax, reuseport_cb_ok,
276 reuseport_ok);
277 }
278
279 /* Determine if there is a bind conflict with an existing IPV6_ADDR_ANY (if ipv6) or
280 * INADDR_ANY (if ipv4) socket.
281 *
282 * Caller must hold bhash hashbucket lock with local bh disabled, to protect
283 * against concurrent binds on the port for addr any
284 */
inet_bhash2_addr_any_conflict(const struct sock * sk,int port,int l3mdev,bool relax,bool reuseport_ok)285 static bool inet_bhash2_addr_any_conflict(const struct sock *sk, int port, int l3mdev,
286 bool relax, bool reuseport_ok)
287 {
288 kuid_t uid = sock_i_uid((struct sock *)sk);
289 const struct net *net = sock_net(sk);
290 struct sock_reuseport *reuseport_cb;
291 struct inet_bind_hashbucket *head2;
292 struct inet_bind2_bucket *tb2;
293 bool conflict = false;
294 bool reuseport_cb_ok;
295
296 rcu_read_lock();
297 reuseport_cb = rcu_dereference(sk->sk_reuseport_cb);
298 /* paired with WRITE_ONCE() in __reuseport_(add|detach)_closed_sock */
299 reuseport_cb_ok = !reuseport_cb || READ_ONCE(reuseport_cb->num_closed_socks);
300 rcu_read_unlock();
301
302 head2 = inet_bhash2_addr_any_hashbucket(sk, net, port);
303
304 spin_lock(&head2->lock);
305
306 inet_bind_bucket_for_each(tb2, &head2->chain) {
307 if (!inet_bind2_bucket_match_addr_any(tb2, net, port, l3mdev, sk))
308 continue;
309
310 if (!inet_bhash2_conflict(sk, tb2, uid, relax, reuseport_cb_ok, reuseport_ok))
311 continue;
312
313 conflict = true;
314 break;
315 }
316
317 spin_unlock(&head2->lock);
318
319 return conflict;
320 }
321
322 /*
323 * Find an open port number for the socket. Returns with the
324 * inet_bind_hashbucket locks held if successful.
325 */
326 static struct inet_bind_hashbucket *
inet_csk_find_open_port(const struct sock * sk,struct inet_bind_bucket ** tb_ret,struct inet_bind2_bucket ** tb2_ret,struct inet_bind_hashbucket ** head2_ret,int * port_ret)327 inet_csk_find_open_port(const struct sock *sk, struct inet_bind_bucket **tb_ret,
328 struct inet_bind2_bucket **tb2_ret,
329 struct inet_bind_hashbucket **head2_ret, int *port_ret)
330 {
331 struct inet_hashinfo *hinfo = tcp_or_dccp_get_hashinfo(sk);
332 int i, low, high, attempt_half, port, l3mdev;
333 struct inet_bind_hashbucket *head, *head2;
334 struct net *net = sock_net(sk);
335 struct inet_bind2_bucket *tb2;
336 struct inet_bind_bucket *tb;
337 u32 remaining, offset;
338 bool relax = false;
339
340 l3mdev = inet_sk_bound_l3mdev(sk);
341 ports_exhausted:
342 attempt_half = (sk->sk_reuse == SK_CAN_REUSE) ? 1 : 0;
343 other_half_scan:
344 inet_sk_get_local_port_range(sk, &low, &high);
345 high++; /* [32768, 60999] -> [32768, 61000[ */
346 if (high - low < 4)
347 attempt_half = 0;
348 if (attempt_half) {
349 int half = low + (((high - low) >> 2) << 1);
350
351 if (attempt_half == 1)
352 high = half;
353 else
354 low = half;
355 }
356 remaining = high - low;
357 if (likely(remaining > 1))
358 remaining &= ~1U;
359
360 offset = get_random_u32_below(remaining);
361 /* __inet_hash_connect() favors ports having @low parity
362 * We do the opposite to not pollute connect() users.
363 */
364 offset |= 1U;
365
366 other_parity_scan:
367 port = low + offset;
368 for (i = 0; i < remaining; i += 2, port += 2) {
369 if (unlikely(port >= high))
370 port -= remaining;
371 if (inet_is_local_reserved_port(net, port))
372 continue;
373 head = &hinfo->bhash[inet_bhashfn(net, port,
374 hinfo->bhash_size)];
375 spin_lock_bh(&head->lock);
376 if (inet_use_bhash2_on_bind(sk)) {
377 if (inet_bhash2_addr_any_conflict(sk, port, l3mdev, relax, false))
378 goto next_port;
379 }
380
381 head2 = inet_bhashfn_portaddr(hinfo, sk, net, port);
382 spin_lock(&head2->lock);
383 tb2 = inet_bind2_bucket_find(head2, net, port, l3mdev, sk);
384 inet_bind_bucket_for_each(tb, &head->chain)
385 if (inet_bind_bucket_match(tb, net, port, l3mdev)) {
386 if (!inet_csk_bind_conflict(sk, tb, tb2,
387 relax, false))
388 goto success;
389 spin_unlock(&head2->lock);
390 goto next_port;
391 }
392 tb = NULL;
393 goto success;
394 next_port:
395 spin_unlock_bh(&head->lock);
396 cond_resched();
397 }
398
399 offset--;
400 if (!(offset & 1))
401 goto other_parity_scan;
402
403 if (attempt_half == 1) {
404 /* OK we now try the upper half of the range */
405 attempt_half = 2;
406 goto other_half_scan;
407 }
408
409 if (READ_ONCE(net->ipv4.sysctl_ip_autobind_reuse) && !relax) {
410 /* We still have a chance to connect to different destinations */
411 relax = true;
412 goto ports_exhausted;
413 }
414 return NULL;
415 success:
416 *port_ret = port;
417 *tb_ret = tb;
418 *tb2_ret = tb2;
419 *head2_ret = head2;
420 return head;
421 }
422
sk_reuseport_match(struct inet_bind_bucket * tb,struct sock * sk)423 static inline int sk_reuseport_match(struct inet_bind_bucket *tb,
424 struct sock *sk)
425 {
426 kuid_t uid = sock_i_uid(sk);
427
428 if (tb->fastreuseport <= 0)
429 return 0;
430 if (!sk->sk_reuseport)
431 return 0;
432 if (rcu_access_pointer(sk->sk_reuseport_cb))
433 return 0;
434 if (!uid_eq(tb->fastuid, uid))
435 return 0;
436 /* We only need to check the rcv_saddr if this tb was once marked
437 * without fastreuseport and then was reset, as we can only know that
438 * the fast_*rcv_saddr doesn't have any conflicts with the socks on the
439 * owners list.
440 */
441 if (tb->fastreuseport == FASTREUSEPORT_ANY)
442 return 1;
443 #if IS_ENABLED(CONFIG_IPV6)
444 if (tb->fast_sk_family == AF_INET6)
445 return ipv6_rcv_saddr_equal(&tb->fast_v6_rcv_saddr,
446 inet6_rcv_saddr(sk),
447 tb->fast_rcv_saddr,
448 sk->sk_rcv_saddr,
449 tb->fast_ipv6_only,
450 ipv6_only_sock(sk), true, false);
451 #endif
452 return ipv4_rcv_saddr_equal(tb->fast_rcv_saddr, sk->sk_rcv_saddr,
453 ipv6_only_sock(sk), true, false);
454 }
455
inet_csk_update_fastreuse(struct inet_bind_bucket * tb,struct sock * sk)456 void inet_csk_update_fastreuse(struct inet_bind_bucket *tb,
457 struct sock *sk)
458 {
459 kuid_t uid = sock_i_uid(sk);
460 bool reuse = sk->sk_reuse && sk->sk_state != TCP_LISTEN;
461
462 if (hlist_empty(&tb->owners)) {
463 tb->fastreuse = reuse;
464 if (sk->sk_reuseport) {
465 tb->fastreuseport = FASTREUSEPORT_ANY;
466 tb->fastuid = uid;
467 tb->fast_rcv_saddr = sk->sk_rcv_saddr;
468 tb->fast_ipv6_only = ipv6_only_sock(sk);
469 tb->fast_sk_family = sk->sk_family;
470 #if IS_ENABLED(CONFIG_IPV6)
471 tb->fast_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
472 #endif
473 } else {
474 tb->fastreuseport = 0;
475 }
476 } else {
477 if (!reuse)
478 tb->fastreuse = 0;
479 if (sk->sk_reuseport) {
480 /* We didn't match or we don't have fastreuseport set on
481 * the tb, but we have sk_reuseport set on this socket
482 * and we know that there are no bind conflicts with
483 * this socket in this tb, so reset our tb's reuseport
484 * settings so that any subsequent sockets that match
485 * our current socket will be put on the fast path.
486 *
487 * If we reset we need to set FASTREUSEPORT_STRICT so we
488 * do extra checking for all subsequent sk_reuseport
489 * socks.
490 */
491 if (!sk_reuseport_match(tb, sk)) {
492 tb->fastreuseport = FASTREUSEPORT_STRICT;
493 tb->fastuid = uid;
494 tb->fast_rcv_saddr = sk->sk_rcv_saddr;
495 tb->fast_ipv6_only = ipv6_only_sock(sk);
496 tb->fast_sk_family = sk->sk_family;
497 #if IS_ENABLED(CONFIG_IPV6)
498 tb->fast_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
499 #endif
500 }
501 } else {
502 tb->fastreuseport = 0;
503 }
504 }
505 }
506
507 /* Obtain a reference to a local port for the given sock,
508 * if snum is zero it means select any available local port.
509 * We try to allocate an odd port (and leave even ports for connect())
510 */
inet_csk_get_port(struct sock * sk,unsigned short snum)511 int inet_csk_get_port(struct sock *sk, unsigned short snum)
512 {
513 struct inet_hashinfo *hinfo = tcp_or_dccp_get_hashinfo(sk);
514 bool reuse = sk->sk_reuse && sk->sk_state != TCP_LISTEN;
515 bool found_port = false, check_bind_conflict = true;
516 bool bhash_created = false, bhash2_created = false;
517 int ret = -EADDRINUSE, port = snum, l3mdev;
518 struct inet_bind_hashbucket *head, *head2;
519 struct inet_bind2_bucket *tb2 = NULL;
520 struct inet_bind_bucket *tb = NULL;
521 bool head2_lock_acquired = false;
522 struct net *net = sock_net(sk);
523
524 l3mdev = inet_sk_bound_l3mdev(sk);
525
526 if (!port) {
527 head = inet_csk_find_open_port(sk, &tb, &tb2, &head2, &port);
528 if (!head)
529 return ret;
530
531 head2_lock_acquired = true;
532
533 if (tb && tb2)
534 goto success;
535 found_port = true;
536 } else {
537 head = &hinfo->bhash[inet_bhashfn(net, port,
538 hinfo->bhash_size)];
539 spin_lock_bh(&head->lock);
540 inet_bind_bucket_for_each(tb, &head->chain)
541 if (inet_bind_bucket_match(tb, net, port, l3mdev))
542 break;
543 }
544
545 if (!tb) {
546 tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep, net,
547 head, port, l3mdev);
548 if (!tb)
549 goto fail_unlock;
550 bhash_created = true;
551 }
552
553 if (!found_port) {
554 if (!hlist_empty(&tb->owners)) {
555 if (sk->sk_reuse == SK_FORCE_REUSE ||
556 (tb->fastreuse > 0 && reuse) ||
557 sk_reuseport_match(tb, sk))
558 check_bind_conflict = false;
559 }
560
561 if (check_bind_conflict && inet_use_bhash2_on_bind(sk)) {
562 if (inet_bhash2_addr_any_conflict(sk, port, l3mdev, true, true))
563 goto fail_unlock;
564 }
565
566 head2 = inet_bhashfn_portaddr(hinfo, sk, net, port);
567 spin_lock(&head2->lock);
568 head2_lock_acquired = true;
569 tb2 = inet_bind2_bucket_find(head2, net, port, l3mdev, sk);
570 }
571
572 if (!tb2) {
573 tb2 = inet_bind2_bucket_create(hinfo->bind2_bucket_cachep,
574 net, head2, port, l3mdev, sk);
575 if (!tb2)
576 goto fail_unlock;
577 bhash2_created = true;
578 }
579
580 if (!found_port && check_bind_conflict) {
581 if (inet_csk_bind_conflict(sk, tb, tb2, true, true))
582 goto fail_unlock;
583 }
584
585 success:
586 inet_csk_update_fastreuse(tb, sk);
587
588 if (!inet_csk(sk)->icsk_bind_hash)
589 inet_bind_hash(sk, tb, tb2, port);
590 WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
591 WARN_ON(inet_csk(sk)->icsk_bind2_hash != tb2);
592 ret = 0;
593
594 fail_unlock:
595 if (ret) {
596 if (bhash_created)
597 inet_bind_bucket_destroy(hinfo->bind_bucket_cachep, tb);
598 if (bhash2_created)
599 inet_bind2_bucket_destroy(hinfo->bind2_bucket_cachep,
600 tb2);
601 }
602 if (head2_lock_acquired)
603 spin_unlock(&head2->lock);
604 spin_unlock_bh(&head->lock);
605 return ret;
606 }
607 EXPORT_SYMBOL_GPL(inet_csk_get_port);
608
609 /*
610 * Wait for an incoming connection, avoid race conditions. This must be called
611 * with the socket locked.
612 */
inet_csk_wait_for_connect(struct sock * sk,long timeo)613 static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
614 {
615 struct inet_connection_sock *icsk = inet_csk(sk);
616 DEFINE_WAIT(wait);
617 int err;
618
619 /*
620 * True wake-one mechanism for incoming connections: only
621 * one process gets woken up, not the 'whole herd'.
622 * Since we do not 'race & poll' for established sockets
623 * anymore, the common case will execute the loop only once.
624 *
625 * Subtle issue: "add_wait_queue_exclusive()" will be added
626 * after any current non-exclusive waiters, and we know that
627 * it will always _stay_ after any new non-exclusive waiters
628 * because all non-exclusive waiters are added at the
629 * beginning of the wait-queue. As such, it's ok to "drop"
630 * our exclusiveness temporarily when we get woken up without
631 * having to remove and re-insert us on the wait queue.
632 */
633 for (;;) {
634 prepare_to_wait_exclusive(sk_sleep(sk), &wait,
635 TASK_INTERRUPTIBLE);
636 release_sock(sk);
637 if (reqsk_queue_empty(&icsk->icsk_accept_queue))
638 timeo = schedule_timeout(timeo);
639 sched_annotate_sleep();
640 lock_sock(sk);
641 err = 0;
642 if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
643 break;
644 err = -EINVAL;
645 if (sk->sk_state != TCP_LISTEN)
646 break;
647 err = sock_intr_errno(timeo);
648 if (signal_pending(current))
649 break;
650 err = -EAGAIN;
651 if (!timeo)
652 break;
653 }
654 finish_wait(sk_sleep(sk), &wait);
655 return err;
656 }
657
658 /*
659 * This will accept the next outstanding connection.
660 */
inet_csk_accept(struct sock * sk,int flags,int * err,bool kern)661 struct sock *inet_csk_accept(struct sock *sk, int flags, int *err, bool kern)
662 {
663 struct inet_connection_sock *icsk = inet_csk(sk);
664 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
665 struct request_sock *req;
666 struct sock *newsk;
667 int error;
668
669 lock_sock(sk);
670
671 /* We need to make sure that this socket is listening,
672 * and that it has something pending.
673 */
674 error = -EINVAL;
675 if (sk->sk_state != TCP_LISTEN)
676 goto out_err;
677
678 /* Find already established connection */
679 if (reqsk_queue_empty(queue)) {
680 long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
681
682 /* If this is a non blocking socket don't sleep */
683 error = -EAGAIN;
684 if (!timeo)
685 goto out_err;
686
687 error = inet_csk_wait_for_connect(sk, timeo);
688 if (error)
689 goto out_err;
690 }
691 req = reqsk_queue_remove(queue, sk);
692 newsk = req->sk;
693
694 if (sk->sk_protocol == IPPROTO_TCP &&
695 tcp_rsk(req)->tfo_listener) {
696 spin_lock_bh(&queue->fastopenq.lock);
697 if (tcp_rsk(req)->tfo_listener) {
698 /* We are still waiting for the final ACK from 3WHS
699 * so can't free req now. Instead, we set req->sk to
700 * NULL to signify that the child socket is taken
701 * so reqsk_fastopen_remove() will free the req
702 * when 3WHS finishes (or is aborted).
703 */
704 req->sk = NULL;
705 req = NULL;
706 }
707 spin_unlock_bh(&queue->fastopenq.lock);
708 }
709
710 out:
711 release_sock(sk);
712 if (newsk && mem_cgroup_sockets_enabled) {
713 int amt = 0;
714
715 /* atomically get the memory usage, set and charge the
716 * newsk->sk_memcg.
717 */
718 lock_sock(newsk);
719
720 mem_cgroup_sk_alloc(newsk);
721 if (newsk->sk_memcg) {
722 /* The socket has not been accepted yet, no need
723 * to look at newsk->sk_wmem_queued.
724 */
725 amt = sk_mem_pages(newsk->sk_forward_alloc +
726 atomic_read(&newsk->sk_rmem_alloc));
727 }
728
729 if (amt)
730 mem_cgroup_charge_skmem(newsk->sk_memcg, amt,
731 GFP_KERNEL | __GFP_NOFAIL);
732
733 release_sock(newsk);
734 }
735 if (req)
736 reqsk_put(req);
737
738 if (newsk)
739 inet_init_csk_locks(newsk);
740
741 return newsk;
742 out_err:
743 newsk = NULL;
744 req = NULL;
745 *err = error;
746 goto out;
747 }
748 EXPORT_SYMBOL(inet_csk_accept);
749
750 /*
751 * Using different timers for retransmit, delayed acks and probes
752 * We may wish use just one timer maintaining a list of expire jiffies
753 * to optimize.
754 */
inet_csk_init_xmit_timers(struct sock * sk,void (* retransmit_handler)(struct timer_list * t),void (* delack_handler)(struct timer_list * t),void (* keepalive_handler)(struct timer_list * t))755 void inet_csk_init_xmit_timers(struct sock *sk,
756 void (*retransmit_handler)(struct timer_list *t),
757 void (*delack_handler)(struct timer_list *t),
758 void (*keepalive_handler)(struct timer_list *t))
759 {
760 struct inet_connection_sock *icsk = inet_csk(sk);
761
762 timer_setup(&icsk->icsk_retransmit_timer, retransmit_handler, 0);
763 timer_setup(&icsk->icsk_delack_timer, delack_handler, 0);
764 timer_setup(&sk->sk_timer, keepalive_handler, 0);
765 icsk->icsk_pending = icsk->icsk_ack.pending = 0;
766 }
767 EXPORT_SYMBOL(inet_csk_init_xmit_timers);
768
inet_csk_clear_xmit_timers(struct sock * sk)769 void inet_csk_clear_xmit_timers(struct sock *sk)
770 {
771 struct inet_connection_sock *icsk = inet_csk(sk);
772
773 icsk->icsk_pending = icsk->icsk_ack.pending = 0;
774
775 sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
776 sk_stop_timer(sk, &icsk->icsk_delack_timer);
777 sk_stop_timer(sk, &sk->sk_timer);
778 }
779 EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
780
inet_csk_clear_xmit_timers_sync(struct sock * sk)781 void inet_csk_clear_xmit_timers_sync(struct sock *sk)
782 {
783 struct inet_connection_sock *icsk = inet_csk(sk);
784
785 /* ongoing timer handlers need to acquire socket lock. */
786 sock_not_owned_by_me(sk);
787
788 icsk->icsk_pending = icsk->icsk_ack.pending = 0;
789
790 sk_stop_timer_sync(sk, &icsk->icsk_retransmit_timer);
791 sk_stop_timer_sync(sk, &icsk->icsk_delack_timer);
792 sk_stop_timer_sync(sk, &sk->sk_timer);
793 }
794
inet_csk_delete_keepalive_timer(struct sock * sk)795 void inet_csk_delete_keepalive_timer(struct sock *sk)
796 {
797 sk_stop_timer(sk, &sk->sk_timer);
798 }
799 EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
800
inet_csk_reset_keepalive_timer(struct sock * sk,unsigned long len)801 void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
802 {
803 sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
804 }
805 EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
806
inet_csk_route_req(const struct sock * sk,struct flowi4 * fl4,const struct request_sock * req)807 struct dst_entry *inet_csk_route_req(const struct sock *sk,
808 struct flowi4 *fl4,
809 const struct request_sock *req)
810 {
811 const struct inet_request_sock *ireq = inet_rsk(req);
812 struct net *net = read_pnet(&ireq->ireq_net);
813 struct ip_options_rcu *opt;
814 struct rtable *rt;
815
816 rcu_read_lock();
817 opt = rcu_dereference(ireq->ireq_opt);
818
819 flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
820 ip_sock_rt_tos(sk), ip_sock_rt_scope(sk),
821 sk->sk_protocol, inet_sk_flowi_flags(sk),
822 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
823 ireq->ir_loc_addr, ireq->ir_rmt_port,
824 htons(ireq->ir_num), sk->sk_uid);
825 security_req_classify_flow(req, flowi4_to_flowi_common(fl4));
826 rt = ip_route_output_flow(net, fl4, sk);
827 if (IS_ERR(rt))
828 goto no_route;
829 if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
830 goto route_err;
831 rcu_read_unlock();
832 return &rt->dst;
833
834 route_err:
835 ip_rt_put(rt);
836 no_route:
837 rcu_read_unlock();
838 __IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
839 return NULL;
840 }
841 EXPORT_SYMBOL_GPL(inet_csk_route_req);
842
inet_csk_route_child_sock(const struct sock * sk,struct sock * newsk,const struct request_sock * req)843 struct dst_entry *inet_csk_route_child_sock(const struct sock *sk,
844 struct sock *newsk,
845 const struct request_sock *req)
846 {
847 const struct inet_request_sock *ireq = inet_rsk(req);
848 struct net *net = read_pnet(&ireq->ireq_net);
849 struct inet_sock *newinet = inet_sk(newsk);
850 struct ip_options_rcu *opt;
851 struct flowi4 *fl4;
852 struct rtable *rt;
853
854 opt = rcu_dereference(ireq->ireq_opt);
855 fl4 = &newinet->cork.fl.u.ip4;
856
857 flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
858 ip_sock_rt_tos(sk), ip_sock_rt_scope(sk),
859 sk->sk_protocol, inet_sk_flowi_flags(sk),
860 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
861 ireq->ir_loc_addr, ireq->ir_rmt_port,
862 htons(ireq->ir_num), sk->sk_uid);
863 security_req_classify_flow(req, flowi4_to_flowi_common(fl4));
864 rt = ip_route_output_flow(net, fl4, sk);
865 if (IS_ERR(rt))
866 goto no_route;
867 if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
868 goto route_err;
869 return &rt->dst;
870
871 route_err:
872 ip_rt_put(rt);
873 no_route:
874 __IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
875 return NULL;
876 }
877 EXPORT_SYMBOL_GPL(inet_csk_route_child_sock);
878
879 /* Decide when to expire the request and when to resend SYN-ACK */
syn_ack_recalc(struct request_sock * req,const int max_syn_ack_retries,const u8 rskq_defer_accept,int * expire,int * resend)880 static void syn_ack_recalc(struct request_sock *req,
881 const int max_syn_ack_retries,
882 const u8 rskq_defer_accept,
883 int *expire, int *resend)
884 {
885 if (!rskq_defer_accept) {
886 *expire = req->num_timeout >= max_syn_ack_retries;
887 *resend = 1;
888 return;
889 }
890 *expire = req->num_timeout >= max_syn_ack_retries &&
891 (!inet_rsk(req)->acked || req->num_timeout >= rskq_defer_accept);
892 /* Do not resend while waiting for data after ACK,
893 * start to resend on end of deferring period to give
894 * last chance for data or ACK to create established socket.
895 */
896 *resend = !inet_rsk(req)->acked ||
897 req->num_timeout >= rskq_defer_accept - 1;
898 }
899
inet_rtx_syn_ack(const struct sock * parent,struct request_sock * req)900 int inet_rtx_syn_ack(const struct sock *parent, struct request_sock *req)
901 {
902 int err = req->rsk_ops->rtx_syn_ack(parent, req);
903
904 if (!err)
905 req->num_retrans++;
906 return err;
907 }
908 EXPORT_SYMBOL(inet_rtx_syn_ack);
909
inet_reqsk_clone(struct request_sock * req,struct sock * sk)910 static struct request_sock *inet_reqsk_clone(struct request_sock *req,
911 struct sock *sk)
912 {
913 struct sock *req_sk, *nreq_sk;
914 struct request_sock *nreq;
915
916 nreq = kmem_cache_alloc(req->rsk_ops->slab, GFP_ATOMIC | __GFP_NOWARN);
917 if (!nreq) {
918 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQFAILURE);
919
920 /* paired with refcount_inc_not_zero() in reuseport_migrate_sock() */
921 sock_put(sk);
922 return NULL;
923 }
924
925 req_sk = req_to_sk(req);
926 nreq_sk = req_to_sk(nreq);
927
928 memcpy(nreq_sk, req_sk,
929 offsetof(struct sock, sk_dontcopy_begin));
930 memcpy(&nreq_sk->sk_dontcopy_end, &req_sk->sk_dontcopy_end,
931 req->rsk_ops->obj_size - offsetof(struct sock, sk_dontcopy_end));
932
933 sk_node_init(&nreq_sk->sk_node);
934 nreq_sk->sk_tx_queue_mapping = req_sk->sk_tx_queue_mapping;
935 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
936 nreq_sk->sk_rx_queue_mapping = req_sk->sk_rx_queue_mapping;
937 #endif
938 nreq_sk->sk_incoming_cpu = req_sk->sk_incoming_cpu;
939
940 nreq->rsk_listener = sk;
941
942 /* We need not acquire fastopenq->lock
943 * because the child socket is locked in inet_csk_listen_stop().
944 */
945 if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(nreq)->tfo_listener)
946 rcu_assign_pointer(tcp_sk(nreq->sk)->fastopen_rsk, nreq);
947
948 return nreq;
949 }
950
reqsk_queue_migrated(struct request_sock_queue * queue,const struct request_sock * req)951 static void reqsk_queue_migrated(struct request_sock_queue *queue,
952 const struct request_sock *req)
953 {
954 if (req->num_timeout == 0)
955 atomic_inc(&queue->young);
956 atomic_inc(&queue->qlen);
957 }
958
reqsk_migrate_reset(struct request_sock * req)959 static void reqsk_migrate_reset(struct request_sock *req)
960 {
961 req->saved_syn = NULL;
962 #if IS_ENABLED(CONFIG_IPV6)
963 inet_rsk(req)->ipv6_opt = NULL;
964 inet_rsk(req)->pktopts = NULL;
965 #else
966 inet_rsk(req)->ireq_opt = NULL;
967 #endif
968 }
969
970 /* return true if req was found in the ehash table */
reqsk_queue_unlink(struct request_sock * req)971 static bool reqsk_queue_unlink(struct request_sock *req)
972 {
973 struct sock *sk = req_to_sk(req);
974 bool found = false;
975
976 if (sk_hashed(sk)) {
977 struct inet_hashinfo *hashinfo = tcp_or_dccp_get_hashinfo(sk);
978 spinlock_t *lock = inet_ehash_lockp(hashinfo, req->rsk_hash);
979
980 spin_lock(lock);
981 found = __sk_nulls_del_node_init_rcu(sk);
982 spin_unlock(lock);
983 }
984 if (timer_pending(&req->rsk_timer) && del_timer_sync(&req->rsk_timer))
985 reqsk_put(req);
986 return found;
987 }
988
inet_csk_reqsk_queue_drop(struct sock * sk,struct request_sock * req)989 bool inet_csk_reqsk_queue_drop(struct sock *sk, struct request_sock *req)
990 {
991 bool unlinked = reqsk_queue_unlink(req);
992
993 if (unlinked) {
994 reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req);
995 reqsk_put(req);
996 }
997 return unlinked;
998 }
999 EXPORT_SYMBOL(inet_csk_reqsk_queue_drop);
1000
inet_csk_reqsk_queue_drop_and_put(struct sock * sk,struct request_sock * req)1001 void inet_csk_reqsk_queue_drop_and_put(struct sock *sk, struct request_sock *req)
1002 {
1003 inet_csk_reqsk_queue_drop(sk, req);
1004 reqsk_put(req);
1005 }
1006 EXPORT_SYMBOL(inet_csk_reqsk_queue_drop_and_put);
1007
reqsk_timer_handler(struct timer_list * t)1008 static void reqsk_timer_handler(struct timer_list *t)
1009 {
1010 struct request_sock *req = from_timer(req, t, rsk_timer);
1011 struct request_sock *nreq = NULL, *oreq = req;
1012 struct sock *sk_listener = req->rsk_listener;
1013 struct inet_connection_sock *icsk;
1014 struct request_sock_queue *queue;
1015 struct net *net;
1016 int max_syn_ack_retries, qlen, expire = 0, resend = 0;
1017
1018 if (inet_sk_state_load(sk_listener) != TCP_LISTEN) {
1019 struct sock *nsk;
1020
1021 nsk = reuseport_migrate_sock(sk_listener, req_to_sk(req), NULL);
1022 if (!nsk)
1023 goto drop;
1024
1025 nreq = inet_reqsk_clone(req, nsk);
1026 if (!nreq)
1027 goto drop;
1028
1029 /* The new timer for the cloned req can decrease the 2
1030 * by calling inet_csk_reqsk_queue_drop_and_put(), so
1031 * hold another count to prevent use-after-free and
1032 * call reqsk_put() just before return.
1033 */
1034 refcount_set(&nreq->rsk_refcnt, 2 + 1);
1035 timer_setup(&nreq->rsk_timer, reqsk_timer_handler, TIMER_PINNED);
1036 reqsk_queue_migrated(&inet_csk(nsk)->icsk_accept_queue, req);
1037
1038 req = nreq;
1039 sk_listener = nsk;
1040 }
1041
1042 icsk = inet_csk(sk_listener);
1043 net = sock_net(sk_listener);
1044 max_syn_ack_retries = READ_ONCE(icsk->icsk_syn_retries) ? :
1045 READ_ONCE(net->ipv4.sysctl_tcp_synack_retries);
1046 /* Normally all the openreqs are young and become mature
1047 * (i.e. converted to established socket) for first timeout.
1048 * If synack was not acknowledged for 1 second, it means
1049 * one of the following things: synack was lost, ack was lost,
1050 * rtt is high or nobody planned to ack (i.e. synflood).
1051 * When server is a bit loaded, queue is populated with old
1052 * open requests, reducing effective size of queue.
1053 * When server is well loaded, queue size reduces to zero
1054 * after several minutes of work. It is not synflood,
1055 * it is normal operation. The solution is pruning
1056 * too old entries overriding normal timeout, when
1057 * situation becomes dangerous.
1058 *
1059 * Essentially, we reserve half of room for young
1060 * embrions; and abort old ones without pity, if old
1061 * ones are about to clog our table.
1062 */
1063 queue = &icsk->icsk_accept_queue;
1064 qlen = reqsk_queue_len(queue);
1065 if ((qlen << 1) > max(8U, READ_ONCE(sk_listener->sk_max_ack_backlog))) {
1066 int young = reqsk_queue_len_young(queue) << 1;
1067
1068 while (max_syn_ack_retries > 2) {
1069 if (qlen < young)
1070 break;
1071 max_syn_ack_retries--;
1072 young <<= 1;
1073 }
1074 }
1075 syn_ack_recalc(req, max_syn_ack_retries, READ_ONCE(queue->rskq_defer_accept),
1076 &expire, &resend);
1077 req->rsk_ops->syn_ack_timeout(req);
1078 if (!expire &&
1079 (!resend ||
1080 !inet_rtx_syn_ack(sk_listener, req) ||
1081 inet_rsk(req)->acked)) {
1082 if (req->num_timeout++ == 0)
1083 atomic_dec(&queue->young);
1084 mod_timer(&req->rsk_timer, jiffies + reqsk_timeout(req, TCP_RTO_MAX));
1085
1086 if (!nreq)
1087 return;
1088
1089 if (!inet_ehash_insert(req_to_sk(nreq), req_to_sk(oreq), NULL)) {
1090 /* delete timer */
1091 inet_csk_reqsk_queue_drop(sk_listener, nreq);
1092 goto no_ownership;
1093 }
1094
1095 __NET_INC_STATS(net, LINUX_MIB_TCPMIGRATEREQSUCCESS);
1096 reqsk_migrate_reset(oreq);
1097 reqsk_queue_removed(&inet_csk(oreq->rsk_listener)->icsk_accept_queue, oreq);
1098 reqsk_put(oreq);
1099
1100 reqsk_put(nreq);
1101 return;
1102 }
1103
1104 /* Even if we can clone the req, we may need not retransmit any more
1105 * SYN+ACKs (nreq->num_timeout > max_syn_ack_retries, etc), or another
1106 * CPU may win the "own_req" race so that inet_ehash_insert() fails.
1107 */
1108 if (nreq) {
1109 __NET_INC_STATS(net, LINUX_MIB_TCPMIGRATEREQFAILURE);
1110 no_ownership:
1111 reqsk_migrate_reset(nreq);
1112 reqsk_queue_removed(queue, nreq);
1113 __reqsk_free(nreq);
1114 }
1115
1116 drop:
1117 inet_csk_reqsk_queue_drop_and_put(oreq->rsk_listener, oreq);
1118 }
1119
reqsk_queue_hash_req(struct request_sock * req,unsigned long timeout)1120 static bool reqsk_queue_hash_req(struct request_sock *req,
1121 unsigned long timeout)
1122 {
1123 bool found_dup_sk = false;
1124
1125 if (!inet_ehash_insert(req_to_sk(req), NULL, &found_dup_sk))
1126 return false;
1127
1128 /* The timer needs to be setup after a successful insertion. */
1129 timer_setup(&req->rsk_timer, reqsk_timer_handler, TIMER_PINNED);
1130 mod_timer(&req->rsk_timer, jiffies + timeout);
1131
1132 /* before letting lookups find us, make sure all req fields
1133 * are committed to memory and refcnt initialized.
1134 */
1135 smp_wmb();
1136 refcount_set(&req->rsk_refcnt, 2 + 1);
1137 return true;
1138 }
1139
inet_csk_reqsk_queue_hash_add(struct sock * sk,struct request_sock * req,unsigned long timeout)1140 bool inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
1141 unsigned long timeout)
1142 {
1143 if (!reqsk_queue_hash_req(req, timeout))
1144 return false;
1145
1146 inet_csk_reqsk_queue_added(sk);
1147 return true;
1148 }
1149 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
1150
inet_clone_ulp(const struct request_sock * req,struct sock * newsk,const gfp_t priority)1151 static void inet_clone_ulp(const struct request_sock *req, struct sock *newsk,
1152 const gfp_t priority)
1153 {
1154 struct inet_connection_sock *icsk = inet_csk(newsk);
1155
1156 if (!icsk->icsk_ulp_ops)
1157 return;
1158
1159 icsk->icsk_ulp_ops->clone(req, newsk, priority);
1160 }
1161
1162 /**
1163 * inet_csk_clone_lock - clone an inet socket, and lock its clone
1164 * @sk: the socket to clone
1165 * @req: request_sock
1166 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1167 *
1168 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1169 */
inet_csk_clone_lock(const struct sock * sk,const struct request_sock * req,const gfp_t priority)1170 struct sock *inet_csk_clone_lock(const struct sock *sk,
1171 const struct request_sock *req,
1172 const gfp_t priority)
1173 {
1174 struct sock *newsk = sk_clone_lock(sk, priority);
1175
1176 if (newsk) {
1177 struct inet_connection_sock *newicsk = inet_csk(newsk);
1178
1179 inet_sk_set_state(newsk, TCP_SYN_RECV);
1180 newicsk->icsk_bind_hash = NULL;
1181 newicsk->icsk_bind2_hash = NULL;
1182
1183 inet_sk(newsk)->inet_dport = inet_rsk(req)->ir_rmt_port;
1184 inet_sk(newsk)->inet_num = inet_rsk(req)->ir_num;
1185 inet_sk(newsk)->inet_sport = htons(inet_rsk(req)->ir_num);
1186
1187 /* listeners have SOCK_RCU_FREE, not the children */
1188 sock_reset_flag(newsk, SOCK_RCU_FREE);
1189
1190 inet_sk(newsk)->mc_list = NULL;
1191
1192 newsk->sk_mark = inet_rsk(req)->ir_mark;
1193 atomic64_set(&newsk->sk_cookie,
1194 atomic64_read(&inet_rsk(req)->ir_cookie));
1195
1196 newicsk->icsk_retransmits = 0;
1197 newicsk->icsk_backoff = 0;
1198 newicsk->icsk_probes_out = 0;
1199 newicsk->icsk_probes_tstamp = 0;
1200
1201 /* Deinitialize accept_queue to trap illegal accesses. */
1202 memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
1203
1204 inet_clone_ulp(req, newsk, priority);
1205
1206 security_inet_csk_clone(newsk, req);
1207
1208 trace_android_vh_inet_csk_clone_lock(newsk, req);
1209 }
1210 return newsk;
1211 }
1212 EXPORT_SYMBOL_GPL(inet_csk_clone_lock);
1213
1214 /*
1215 * At this point, there should be no process reference to this
1216 * socket, and thus no user references at all. Therefore we
1217 * can assume the socket waitqueue is inactive and nobody will
1218 * try to jump onto it.
1219 */
inet_csk_destroy_sock(struct sock * sk)1220 void inet_csk_destroy_sock(struct sock *sk)
1221 {
1222 WARN_ON(sk->sk_state != TCP_CLOSE);
1223 WARN_ON(!sock_flag(sk, SOCK_DEAD));
1224
1225 /* It cannot be in hash table! */
1226 WARN_ON(!sk_unhashed(sk));
1227
1228 /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */
1229 WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash);
1230
1231 sk->sk_prot->destroy(sk);
1232
1233 sk_stream_kill_queues(sk);
1234
1235 xfrm_sk_free_policy(sk);
1236
1237 this_cpu_dec(*sk->sk_prot->orphan_count);
1238
1239 sock_put(sk);
1240 }
1241 EXPORT_SYMBOL(inet_csk_destroy_sock);
1242
1243 /* This function allows to force a closure of a socket after the call to
1244 * tcp/dccp_create_openreq_child().
1245 */
inet_csk_prepare_forced_close(struct sock * sk)1246 void inet_csk_prepare_forced_close(struct sock *sk)
1247 __releases(&sk->sk_lock.slock)
1248 {
1249 /* sk_clone_lock locked the socket and set refcnt to 2 */
1250 bh_unlock_sock(sk);
1251 sock_put(sk);
1252 inet_csk_prepare_for_destroy_sock(sk);
1253 inet_sk(sk)->inet_num = 0;
1254 }
1255 EXPORT_SYMBOL(inet_csk_prepare_forced_close);
1256
inet_ulp_can_listen(const struct sock * sk)1257 static int inet_ulp_can_listen(const struct sock *sk)
1258 {
1259 const struct inet_connection_sock *icsk = inet_csk(sk);
1260
1261 if (icsk->icsk_ulp_ops && !icsk->icsk_ulp_ops->clone)
1262 return -EINVAL;
1263
1264 return 0;
1265 }
1266
inet_csk_listen_start(struct sock * sk)1267 int inet_csk_listen_start(struct sock *sk)
1268 {
1269 struct inet_connection_sock *icsk = inet_csk(sk);
1270 struct inet_sock *inet = inet_sk(sk);
1271 int err;
1272
1273 err = inet_ulp_can_listen(sk);
1274 if (unlikely(err))
1275 return err;
1276
1277 reqsk_queue_alloc(&icsk->icsk_accept_queue);
1278
1279 sk->sk_ack_backlog = 0;
1280 inet_csk_delack_init(sk);
1281
1282 /* There is race window here: we announce ourselves listening,
1283 * but this transition is still not validated by get_port().
1284 * It is OK, because this socket enters to hash table only
1285 * after validation is complete.
1286 */
1287 inet_sk_state_store(sk, TCP_LISTEN);
1288 err = sk->sk_prot->get_port(sk, inet->inet_num);
1289 if (!err) {
1290 inet->inet_sport = htons(inet->inet_num);
1291
1292 sk_dst_reset(sk);
1293 err = sk->sk_prot->hash(sk);
1294
1295 if (likely(!err))
1296 return 0;
1297 }
1298
1299 inet_sk_set_state(sk, TCP_CLOSE);
1300 return err;
1301 }
1302 EXPORT_SYMBOL_GPL(inet_csk_listen_start);
1303
inet_child_forget(struct sock * sk,struct request_sock * req,struct sock * child)1304 static void inet_child_forget(struct sock *sk, struct request_sock *req,
1305 struct sock *child)
1306 {
1307 sk->sk_prot->disconnect(child, O_NONBLOCK);
1308
1309 sock_orphan(child);
1310
1311 this_cpu_inc(*sk->sk_prot->orphan_count);
1312
1313 if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(req)->tfo_listener) {
1314 BUG_ON(rcu_access_pointer(tcp_sk(child)->fastopen_rsk) != req);
1315 BUG_ON(sk != req->rsk_listener);
1316
1317 /* Paranoid, to prevent race condition if
1318 * an inbound pkt destined for child is
1319 * blocked by sock lock in tcp_v4_rcv().
1320 * Also to satisfy an assertion in
1321 * tcp_v4_destroy_sock().
1322 */
1323 RCU_INIT_POINTER(tcp_sk(child)->fastopen_rsk, NULL);
1324 }
1325 inet_csk_destroy_sock(child);
1326 }
1327
inet_csk_reqsk_queue_add(struct sock * sk,struct request_sock * req,struct sock * child)1328 struct sock *inet_csk_reqsk_queue_add(struct sock *sk,
1329 struct request_sock *req,
1330 struct sock *child)
1331 {
1332 struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;
1333
1334 spin_lock(&queue->rskq_lock);
1335 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1336 inet_child_forget(sk, req, child);
1337 child = NULL;
1338 } else {
1339 req->sk = child;
1340 req->dl_next = NULL;
1341 if (queue->rskq_accept_head == NULL)
1342 WRITE_ONCE(queue->rskq_accept_head, req);
1343 else
1344 queue->rskq_accept_tail->dl_next = req;
1345 queue->rskq_accept_tail = req;
1346 sk_acceptq_added(sk);
1347 }
1348 spin_unlock(&queue->rskq_lock);
1349 return child;
1350 }
1351 EXPORT_SYMBOL(inet_csk_reqsk_queue_add);
1352
inet_csk_complete_hashdance(struct sock * sk,struct sock * child,struct request_sock * req,bool own_req)1353 struct sock *inet_csk_complete_hashdance(struct sock *sk, struct sock *child,
1354 struct request_sock *req, bool own_req)
1355 {
1356 if (own_req) {
1357 inet_csk_reqsk_queue_drop(req->rsk_listener, req);
1358 reqsk_queue_removed(&inet_csk(req->rsk_listener)->icsk_accept_queue, req);
1359
1360 if (sk != req->rsk_listener) {
1361 /* another listening sk has been selected,
1362 * migrate the req to it.
1363 */
1364 struct request_sock *nreq;
1365
1366 /* hold a refcnt for the nreq->rsk_listener
1367 * which is assigned in inet_reqsk_clone()
1368 */
1369 sock_hold(sk);
1370 nreq = inet_reqsk_clone(req, sk);
1371 if (!nreq) {
1372 inet_child_forget(sk, req, child);
1373 goto child_put;
1374 }
1375
1376 refcount_set(&nreq->rsk_refcnt, 1);
1377 if (inet_csk_reqsk_queue_add(sk, nreq, child)) {
1378 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQSUCCESS);
1379 reqsk_migrate_reset(req);
1380 reqsk_put(req);
1381 return child;
1382 }
1383
1384 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQFAILURE);
1385 reqsk_migrate_reset(nreq);
1386 __reqsk_free(nreq);
1387 } else if (inet_csk_reqsk_queue_add(sk, req, child)) {
1388 return child;
1389 }
1390 }
1391 /* Too bad, another child took ownership of the request, undo. */
1392 child_put:
1393 bh_unlock_sock(child);
1394 sock_put(child);
1395 return NULL;
1396 }
1397 EXPORT_SYMBOL(inet_csk_complete_hashdance);
1398
1399 /*
1400 * This routine closes sockets which have been at least partially
1401 * opened, but not yet accepted.
1402 */
inet_csk_listen_stop(struct sock * sk)1403 void inet_csk_listen_stop(struct sock *sk)
1404 {
1405 struct inet_connection_sock *icsk = inet_csk(sk);
1406 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
1407 struct request_sock *next, *req;
1408
1409 /* Following specs, it would be better either to send FIN
1410 * (and enter FIN-WAIT-1, it is normal close)
1411 * or to send active reset (abort).
1412 * Certainly, it is pretty dangerous while synflood, but it is
1413 * bad justification for our negligence 8)
1414 * To be honest, we are not able to make either
1415 * of the variants now. --ANK
1416 */
1417 while ((req = reqsk_queue_remove(queue, sk)) != NULL) {
1418 struct sock *child = req->sk, *nsk;
1419 struct request_sock *nreq;
1420
1421 local_bh_disable();
1422 bh_lock_sock(child);
1423 WARN_ON(sock_owned_by_user(child));
1424 sock_hold(child);
1425
1426 nsk = reuseport_migrate_sock(sk, child, NULL);
1427 if (nsk) {
1428 nreq = inet_reqsk_clone(req, nsk);
1429 if (nreq) {
1430 refcount_set(&nreq->rsk_refcnt, 1);
1431
1432 if (inet_csk_reqsk_queue_add(nsk, nreq, child)) {
1433 __NET_INC_STATS(sock_net(nsk),
1434 LINUX_MIB_TCPMIGRATEREQSUCCESS);
1435 reqsk_migrate_reset(req);
1436 } else {
1437 __NET_INC_STATS(sock_net(nsk),
1438 LINUX_MIB_TCPMIGRATEREQFAILURE);
1439 reqsk_migrate_reset(nreq);
1440 __reqsk_free(nreq);
1441 }
1442
1443 /* inet_csk_reqsk_queue_add() has already
1444 * called inet_child_forget() on failure case.
1445 */
1446 goto skip_child_forget;
1447 }
1448 }
1449
1450 inet_child_forget(sk, req, child);
1451 skip_child_forget:
1452 reqsk_put(req);
1453 bh_unlock_sock(child);
1454 local_bh_enable();
1455 sock_put(child);
1456
1457 cond_resched();
1458 }
1459 if (queue->fastopenq.rskq_rst_head) {
1460 /* Free all the reqs queued in rskq_rst_head. */
1461 spin_lock_bh(&queue->fastopenq.lock);
1462 req = queue->fastopenq.rskq_rst_head;
1463 queue->fastopenq.rskq_rst_head = NULL;
1464 spin_unlock_bh(&queue->fastopenq.lock);
1465 while (req != NULL) {
1466 next = req->dl_next;
1467 reqsk_put(req);
1468 req = next;
1469 }
1470 }
1471 WARN_ON_ONCE(sk->sk_ack_backlog);
1472 }
1473 EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
1474
inet_csk_addr2sockaddr(struct sock * sk,struct sockaddr * uaddr)1475 void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
1476 {
1477 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
1478 const struct inet_sock *inet = inet_sk(sk);
1479
1480 sin->sin_family = AF_INET;
1481 sin->sin_addr.s_addr = inet->inet_daddr;
1482 sin->sin_port = inet->inet_dport;
1483 }
1484 EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
1485
inet_csk_rebuild_route(struct sock * sk,struct flowi * fl)1486 static struct dst_entry *inet_csk_rebuild_route(struct sock *sk, struct flowi *fl)
1487 {
1488 const struct inet_sock *inet = inet_sk(sk);
1489 const struct ip_options_rcu *inet_opt;
1490 __be32 daddr = inet->inet_daddr;
1491 struct flowi4 *fl4;
1492 struct rtable *rt;
1493
1494 rcu_read_lock();
1495 inet_opt = rcu_dereference(inet->inet_opt);
1496 if (inet_opt && inet_opt->opt.srr)
1497 daddr = inet_opt->opt.faddr;
1498 fl4 = &fl->u.ip4;
1499 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr,
1500 inet->inet_saddr, inet->inet_dport,
1501 inet->inet_sport, sk->sk_protocol,
1502 RT_CONN_FLAGS(sk), sk->sk_bound_dev_if);
1503 if (IS_ERR(rt))
1504 rt = NULL;
1505 if (rt)
1506 sk_setup_caps(sk, &rt->dst);
1507 rcu_read_unlock();
1508
1509 return &rt->dst;
1510 }
1511
inet_csk_update_pmtu(struct sock * sk,u32 mtu)1512 struct dst_entry *inet_csk_update_pmtu(struct sock *sk, u32 mtu)
1513 {
1514 struct dst_entry *dst = __sk_dst_check(sk, 0);
1515 struct inet_sock *inet = inet_sk(sk);
1516
1517 if (!dst) {
1518 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
1519 if (!dst)
1520 goto out;
1521 }
1522 dst->ops->update_pmtu(dst, sk, NULL, mtu, true);
1523
1524 dst = __sk_dst_check(sk, 0);
1525 if (!dst)
1526 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
1527 out:
1528 return dst;
1529 }
1530 EXPORT_SYMBOL_GPL(inet_csk_update_pmtu);
1531