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
121 /**
122 * inet_sk_get_local_port_range - fetch ephemeral ports range
123 * @sk: socket
124 * @low: pointer to low port
125 * @high: pointer to high port
126 *
127 * Fetch netns port range (/proc/sys/net/ipv4/ip_local_port_range)
128 * Range can be overridden if socket got IP_LOCAL_PORT_RANGE option.
129 * Returns true if IP_LOCAL_PORT_RANGE was set on this socket.
130 */
inet_sk_get_local_port_range(const struct sock * sk,int * low,int * high)131 bool inet_sk_get_local_port_range(const struct sock *sk, int *low, int *high)
132 {
133 int lo, hi, sk_lo, sk_hi;
134 bool local_range = false;
135 u32 sk_range;
136
137 inet_get_local_port_range(sock_net(sk), &lo, &hi);
138
139 sk_range = READ_ONCE(inet_sk(sk)->local_port_range);
140 if (unlikely(sk_range)) {
141 sk_lo = sk_range & 0xffff;
142 sk_hi = sk_range >> 16;
143
144 if (lo <= sk_lo && sk_lo <= hi)
145 lo = sk_lo;
146 if (lo <= sk_hi && sk_hi <= hi)
147 hi = sk_hi;
148 local_range = true;
149 }
150
151 *low = lo;
152 *high = hi;
153 return local_range;
154 }
155 EXPORT_SYMBOL(inet_sk_get_local_port_range);
156
inet_use_bhash2_on_bind(const struct sock * sk)157 static bool inet_use_bhash2_on_bind(const struct sock *sk)
158 {
159 #if IS_ENABLED(CONFIG_IPV6)
160 if (sk->sk_family == AF_INET6) {
161 int addr_type = ipv6_addr_type(&sk->sk_v6_rcv_saddr);
162
163 if (addr_type == IPV6_ADDR_ANY)
164 return false;
165
166 if (addr_type != IPV6_ADDR_MAPPED)
167 return true;
168 }
169 #endif
170 return sk->sk_rcv_saddr != htonl(INADDR_ANY);
171 }
172
inet_bind_conflict(const struct sock * sk,struct sock * sk2,kuid_t sk_uid,bool relax,bool reuseport_cb_ok,bool reuseport_ok)173 static bool inet_bind_conflict(const struct sock *sk, struct sock *sk2,
174 kuid_t sk_uid, bool relax,
175 bool reuseport_cb_ok, bool reuseport_ok)
176 {
177 int bound_dev_if2;
178
179 if (sk == sk2)
180 return false;
181
182 bound_dev_if2 = READ_ONCE(sk2->sk_bound_dev_if);
183
184 if (!sk->sk_bound_dev_if || !bound_dev_if2 ||
185 sk->sk_bound_dev_if == bound_dev_if2) {
186 if (sk->sk_reuse && sk2->sk_reuse &&
187 sk2->sk_state != TCP_LISTEN) {
188 if (!relax || (!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 } else if (!reuseport_ok || !sk->sk_reuseport ||
194 !sk2->sk_reuseport || !reuseport_cb_ok ||
195 (sk2->sk_state != TCP_TIME_WAIT &&
196 !uid_eq(sk_uid, sock_i_uid(sk2)))) {
197 return true;
198 }
199 }
200 return false;
201 }
202
__inet_bhash2_conflict(const struct sock * sk,struct sock * sk2,kuid_t sk_uid,bool relax,bool reuseport_cb_ok,bool reuseport_ok)203 static bool __inet_bhash2_conflict(const struct sock *sk, struct sock *sk2,
204 kuid_t sk_uid, bool relax,
205 bool reuseport_cb_ok, bool reuseport_ok)
206 {
207 if (ipv6_only_sock(sk2)) {
208 if (sk->sk_family == AF_INET)
209 return false;
210
211 #if IS_ENABLED(CONFIG_IPV6)
212 if (ipv6_addr_v4mapped(&sk->sk_v6_rcv_saddr))
213 return false;
214 #endif
215 }
216
217 return inet_bind_conflict(sk, sk2, sk_uid, relax,
218 reuseport_cb_ok, reuseport_ok);
219 }
220
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)221 static bool inet_bhash2_conflict(const struct sock *sk,
222 const struct inet_bind2_bucket *tb2,
223 kuid_t sk_uid,
224 bool relax, bool reuseport_cb_ok,
225 bool reuseport_ok)
226 {
227 struct sock *sk2;
228
229 sk_for_each_bound(sk2, &tb2->owners) {
230 if (__inet_bhash2_conflict(sk, sk2, sk_uid, relax,
231 reuseport_cb_ok, reuseport_ok))
232 return true;
233 }
234
235 return false;
236 }
237
238 #define sk_for_each_bound_bhash(__sk, __tb2, __tb) \
239 hlist_for_each_entry(__tb2, &(__tb)->bhash2, bhash_node) \
240 sk_for_each_bound((__sk), &(__tb2)->owners)
241
242 /* 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)243 static int inet_csk_bind_conflict(const struct sock *sk,
244 const struct inet_bind_bucket *tb,
245 const struct inet_bind2_bucket *tb2, /* may be null */
246 bool relax, bool reuseport_ok)
247 {
248 kuid_t uid = sock_i_uid((struct sock *)sk);
249 struct sock_reuseport *reuseport_cb;
250 bool reuseport_cb_ok;
251 struct sock *sk2;
252
253 rcu_read_lock();
254 reuseport_cb = rcu_dereference(sk->sk_reuseport_cb);
255 /* paired with WRITE_ONCE() in __reuseport_(add|detach)_closed_sock */
256 reuseport_cb_ok = !reuseport_cb || READ_ONCE(reuseport_cb->num_closed_socks);
257 rcu_read_unlock();
258
259 /* Conflicts with an existing IPV6_ADDR_ANY (if ipv6) or INADDR_ANY (if
260 * ipv4) should have been checked already. We need to do these two
261 * checks separately because their spinlocks have to be acquired/released
262 * independently of each other, to prevent possible deadlocks
263 */
264 if (inet_use_bhash2_on_bind(sk))
265 return tb2 && inet_bhash2_conflict(sk, tb2, uid, relax,
266 reuseport_cb_ok, reuseport_ok);
267
268 /* Unlike other sk lookup places we do not check
269 * for sk_net here, since _all_ the socks listed
270 * in tb->owners and tb2->owners list belong
271 * to the same net - the one this bucket belongs to.
272 */
273 sk_for_each_bound_bhash(sk2, tb2, tb) {
274 if (!inet_bind_conflict(sk, sk2, uid, relax, reuseport_cb_ok, reuseport_ok))
275 continue;
276
277 if (inet_rcv_saddr_equal(sk, sk2, true))
278 return true;
279 }
280
281 return false;
282 }
283
284 /* Determine if there is a bind conflict with an existing IPV6_ADDR_ANY (if ipv6) or
285 * INADDR_ANY (if ipv4) socket.
286 *
287 * Caller must hold bhash hashbucket lock with local bh disabled, to protect
288 * against concurrent binds on the port for addr any
289 */
inet_bhash2_addr_any_conflict(const struct sock * sk,int port,int l3mdev,bool relax,bool reuseport_ok)290 static bool inet_bhash2_addr_any_conflict(const struct sock *sk, int port, int l3mdev,
291 bool relax, bool reuseport_ok)
292 {
293 kuid_t uid = sock_i_uid((struct sock *)sk);
294 const struct net *net = sock_net(sk);
295 struct sock_reuseport *reuseport_cb;
296 struct inet_bind_hashbucket *head2;
297 struct inet_bind2_bucket *tb2;
298 bool conflict = false;
299 bool reuseport_cb_ok;
300
301 rcu_read_lock();
302 reuseport_cb = rcu_dereference(sk->sk_reuseport_cb);
303 /* paired with WRITE_ONCE() in __reuseport_(add|detach)_closed_sock */
304 reuseport_cb_ok = !reuseport_cb || READ_ONCE(reuseport_cb->num_closed_socks);
305 rcu_read_unlock();
306
307 head2 = inet_bhash2_addr_any_hashbucket(sk, net, port);
308
309 spin_lock(&head2->lock);
310
311 inet_bind_bucket_for_each(tb2, &head2->chain) {
312 if (!inet_bind2_bucket_match_addr_any(tb2, net, port, l3mdev, sk))
313 continue;
314
315 if (!inet_bhash2_conflict(sk, tb2, uid, relax, reuseport_cb_ok, reuseport_ok))
316 continue;
317
318 conflict = true;
319 break;
320 }
321
322 spin_unlock(&head2->lock);
323
324 return conflict;
325 }
326
327 /*
328 * Find an open port number for the socket. Returns with the
329 * inet_bind_hashbucket locks held if successful.
330 */
331 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)332 inet_csk_find_open_port(const struct sock *sk, struct inet_bind_bucket **tb_ret,
333 struct inet_bind2_bucket **tb2_ret,
334 struct inet_bind_hashbucket **head2_ret, int *port_ret)
335 {
336 struct inet_hashinfo *hinfo = tcp_or_dccp_get_hashinfo(sk);
337 int i, low, high, attempt_half, port, l3mdev;
338 struct inet_bind_hashbucket *head, *head2;
339 struct net *net = sock_net(sk);
340 struct inet_bind2_bucket *tb2;
341 struct inet_bind_bucket *tb;
342 u32 remaining, offset;
343 bool relax = false;
344
345 l3mdev = inet_sk_bound_l3mdev(sk);
346 ports_exhausted:
347 attempt_half = (sk->sk_reuse == SK_CAN_REUSE) ? 1 : 0;
348 other_half_scan:
349 inet_sk_get_local_port_range(sk, &low, &high);
350 high++; /* [32768, 60999] -> [32768, 61000[ */
351 if (high - low < 4)
352 attempt_half = 0;
353 if (attempt_half) {
354 int half = low + (((high - low) >> 2) << 1);
355
356 if (attempt_half == 1)
357 high = half;
358 else
359 low = half;
360 }
361 remaining = high - low;
362 if (likely(remaining > 1))
363 remaining &= ~1U;
364
365 offset = get_random_u32_below(remaining);
366 /* __inet_hash_connect() favors ports having @low parity
367 * We do the opposite to not pollute connect() users.
368 */
369 offset |= 1U;
370
371 other_parity_scan:
372 port = low + offset;
373 for (i = 0; i < remaining; i += 2, port += 2) {
374 if (unlikely(port >= high))
375 port -= remaining;
376 if (inet_is_local_reserved_port(net, port))
377 continue;
378 head = &hinfo->bhash[inet_bhashfn(net, port,
379 hinfo->bhash_size)];
380 spin_lock_bh(&head->lock);
381 if (inet_use_bhash2_on_bind(sk)) {
382 if (inet_bhash2_addr_any_conflict(sk, port, l3mdev, relax, false))
383 goto next_port;
384 }
385
386 head2 = inet_bhashfn_portaddr(hinfo, sk, net, port);
387 spin_lock(&head2->lock);
388 tb2 = inet_bind2_bucket_find(head2, net, port, l3mdev, sk);
389 inet_bind_bucket_for_each(tb, &head->chain)
390 if (inet_bind_bucket_match(tb, net, port, l3mdev)) {
391 if (!inet_csk_bind_conflict(sk, tb, tb2,
392 relax, false))
393 goto success;
394 spin_unlock(&head2->lock);
395 goto next_port;
396 }
397 tb = NULL;
398 goto success;
399 next_port:
400 spin_unlock_bh(&head->lock);
401 cond_resched();
402 }
403
404 offset--;
405 if (!(offset & 1))
406 goto other_parity_scan;
407
408 if (attempt_half == 1) {
409 /* OK we now try the upper half of the range */
410 attempt_half = 2;
411 goto other_half_scan;
412 }
413
414 if (READ_ONCE(net->ipv4.sysctl_ip_autobind_reuse) && !relax) {
415 /* We still have a chance to connect to different destinations */
416 relax = true;
417 goto ports_exhausted;
418 }
419 return NULL;
420 success:
421 *port_ret = port;
422 *tb_ret = tb;
423 *tb2_ret = tb2;
424 *head2_ret = head2;
425 return head;
426 }
427
sk_reuseport_match(struct inet_bind_bucket * tb,struct sock * sk)428 static inline int sk_reuseport_match(struct inet_bind_bucket *tb,
429 struct sock *sk)
430 {
431 kuid_t uid = sock_i_uid(sk);
432
433 if (tb->fastreuseport <= 0)
434 return 0;
435 if (!sk->sk_reuseport)
436 return 0;
437 if (rcu_access_pointer(sk->sk_reuseport_cb))
438 return 0;
439 if (!uid_eq(tb->fastuid, uid))
440 return 0;
441 /* We only need to check the rcv_saddr if this tb was once marked
442 * without fastreuseport and then was reset, as we can only know that
443 * the fast_*rcv_saddr doesn't have any conflicts with the socks on the
444 * owners list.
445 */
446 if (tb->fastreuseport == FASTREUSEPORT_ANY)
447 return 1;
448 #if IS_ENABLED(CONFIG_IPV6)
449 if (tb->fast_sk_family == AF_INET6)
450 return ipv6_rcv_saddr_equal(&tb->fast_v6_rcv_saddr,
451 inet6_rcv_saddr(sk),
452 tb->fast_rcv_saddr,
453 sk->sk_rcv_saddr,
454 tb->fast_ipv6_only,
455 ipv6_only_sock(sk), true, false);
456 #endif
457 return ipv4_rcv_saddr_equal(tb->fast_rcv_saddr, sk->sk_rcv_saddr,
458 ipv6_only_sock(sk), true, false);
459 }
460
inet_csk_update_fastreuse(struct inet_bind_bucket * tb,struct sock * sk)461 void inet_csk_update_fastreuse(struct inet_bind_bucket *tb,
462 struct sock *sk)
463 {
464 kuid_t uid = sock_i_uid(sk);
465 bool reuse = sk->sk_reuse && sk->sk_state != TCP_LISTEN;
466
467 if (hlist_empty(&tb->bhash2)) {
468 tb->fastreuse = reuse;
469 if (sk->sk_reuseport) {
470 tb->fastreuseport = FASTREUSEPORT_ANY;
471 tb->fastuid = uid;
472 tb->fast_rcv_saddr = sk->sk_rcv_saddr;
473 tb->fast_ipv6_only = ipv6_only_sock(sk);
474 tb->fast_sk_family = sk->sk_family;
475 #if IS_ENABLED(CONFIG_IPV6)
476 tb->fast_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
477 #endif
478 } else {
479 tb->fastreuseport = 0;
480 }
481 } else {
482 if (!reuse)
483 tb->fastreuse = 0;
484 if (sk->sk_reuseport) {
485 /* We didn't match or we don't have fastreuseport set on
486 * the tb, but we have sk_reuseport set on this socket
487 * and we know that there are no bind conflicts with
488 * this socket in this tb, so reset our tb's reuseport
489 * settings so that any subsequent sockets that match
490 * our current socket will be put on the fast path.
491 *
492 * If we reset we need to set FASTREUSEPORT_STRICT so we
493 * do extra checking for all subsequent sk_reuseport
494 * socks.
495 */
496 if (!sk_reuseport_match(tb, sk)) {
497 tb->fastreuseport = FASTREUSEPORT_STRICT;
498 tb->fastuid = uid;
499 tb->fast_rcv_saddr = sk->sk_rcv_saddr;
500 tb->fast_ipv6_only = ipv6_only_sock(sk);
501 tb->fast_sk_family = sk->sk_family;
502 #if IS_ENABLED(CONFIG_IPV6)
503 tb->fast_v6_rcv_saddr = sk->sk_v6_rcv_saddr;
504 #endif
505 }
506 } else {
507 tb->fastreuseport = 0;
508 }
509 }
510 }
511
512 /* Obtain a reference to a local port for the given sock,
513 * if snum is zero it means select any available local port.
514 * We try to allocate an odd port (and leave even ports for connect())
515 */
inet_csk_get_port(struct sock * sk,unsigned short snum)516 int inet_csk_get_port(struct sock *sk, unsigned short snum)
517 {
518 struct inet_hashinfo *hinfo = tcp_or_dccp_get_hashinfo(sk);
519 bool reuse = sk->sk_reuse && sk->sk_state != TCP_LISTEN;
520 bool found_port = false, check_bind_conflict = true;
521 bool bhash_created = false, bhash2_created = false;
522 int ret = -EADDRINUSE, port = snum, l3mdev;
523 struct inet_bind_hashbucket *head, *head2;
524 struct inet_bind2_bucket *tb2 = NULL;
525 struct inet_bind_bucket *tb = NULL;
526 bool head2_lock_acquired = false;
527 struct net *net = sock_net(sk);
528
529 l3mdev = inet_sk_bound_l3mdev(sk);
530
531 if (!port) {
532 head = inet_csk_find_open_port(sk, &tb, &tb2, &head2, &port);
533 if (!head)
534 return ret;
535
536 head2_lock_acquired = true;
537
538 if (tb && tb2)
539 goto success;
540 found_port = true;
541 } else {
542 head = &hinfo->bhash[inet_bhashfn(net, port,
543 hinfo->bhash_size)];
544 spin_lock_bh(&head->lock);
545 inet_bind_bucket_for_each(tb, &head->chain)
546 if (inet_bind_bucket_match(tb, net, port, l3mdev))
547 break;
548 }
549
550 if (!tb) {
551 tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep, net,
552 head, port, l3mdev);
553 if (!tb)
554 goto fail_unlock;
555 bhash_created = true;
556 }
557
558 if (!found_port) {
559 if (!hlist_empty(&tb->bhash2)) {
560 if (sk->sk_reuse == SK_FORCE_REUSE ||
561 (tb->fastreuse > 0 && reuse) ||
562 sk_reuseport_match(tb, sk))
563 check_bind_conflict = false;
564 }
565
566 if (check_bind_conflict && inet_use_bhash2_on_bind(sk)) {
567 if (inet_bhash2_addr_any_conflict(sk, port, l3mdev, true, true))
568 goto fail_unlock;
569 }
570
571 head2 = inet_bhashfn_portaddr(hinfo, sk, net, port);
572 spin_lock(&head2->lock);
573 head2_lock_acquired = true;
574 tb2 = inet_bind2_bucket_find(head2, net, port, l3mdev, sk);
575 }
576
577 if (!tb2) {
578 tb2 = inet_bind2_bucket_create(hinfo->bind2_bucket_cachep,
579 net, head2, tb, sk);
580 if (!tb2)
581 goto fail_unlock;
582 bhash2_created = true;
583 }
584
585 if (!found_port && check_bind_conflict) {
586 if (inet_csk_bind_conflict(sk, tb, tb2, true, true))
587 goto fail_unlock;
588 }
589
590 success:
591 inet_csk_update_fastreuse(tb, sk);
592
593 if (!inet_csk(sk)->icsk_bind_hash)
594 inet_bind_hash(sk, tb, tb2, port);
595 WARN_ON(inet_csk(sk)->icsk_bind_hash != tb);
596 WARN_ON(inet_csk(sk)->icsk_bind2_hash != tb2);
597 ret = 0;
598
599 fail_unlock:
600 if (ret) {
601 if (bhash2_created)
602 inet_bind2_bucket_destroy(hinfo->bind2_bucket_cachep, tb2);
603 if (bhash_created)
604 inet_bind_bucket_destroy(hinfo->bind_bucket_cachep, tb);
605 }
606 if (head2_lock_acquired)
607 spin_unlock(&head2->lock);
608 spin_unlock_bh(&head->lock);
609 return ret;
610 }
611 EXPORT_SYMBOL_GPL(inet_csk_get_port);
612
613 /*
614 * Wait for an incoming connection, avoid race conditions. This must be called
615 * with the socket locked.
616 */
inet_csk_wait_for_connect(struct sock * sk,long timeo)617 static int inet_csk_wait_for_connect(struct sock *sk, long timeo)
618 {
619 struct inet_connection_sock *icsk = inet_csk(sk);
620 DEFINE_WAIT(wait);
621 int err;
622
623 /*
624 * True wake-one mechanism for incoming connections: only
625 * one process gets woken up, not the 'whole herd'.
626 * Since we do not 'race & poll' for established sockets
627 * anymore, the common case will execute the loop only once.
628 *
629 * Subtle issue: "add_wait_queue_exclusive()" will be added
630 * after any current non-exclusive waiters, and we know that
631 * it will always _stay_ after any new non-exclusive waiters
632 * because all non-exclusive waiters are added at the
633 * beginning of the wait-queue. As such, it's ok to "drop"
634 * our exclusiveness temporarily when we get woken up without
635 * having to remove and re-insert us on the wait queue.
636 */
637 for (;;) {
638 prepare_to_wait_exclusive(sk_sleep(sk), &wait,
639 TASK_INTERRUPTIBLE);
640 release_sock(sk);
641 if (reqsk_queue_empty(&icsk->icsk_accept_queue))
642 timeo = schedule_timeout(timeo);
643 sched_annotate_sleep();
644 lock_sock(sk);
645 err = 0;
646 if (!reqsk_queue_empty(&icsk->icsk_accept_queue))
647 break;
648 err = -EINVAL;
649 if (sk->sk_state != TCP_LISTEN)
650 break;
651 err = sock_intr_errno(timeo);
652 if (signal_pending(current))
653 break;
654 err = -EAGAIN;
655 if (!timeo)
656 break;
657 }
658 finish_wait(sk_sleep(sk), &wait);
659 return err;
660 }
661
662 /*
663 * This will accept the next outstanding connection.
664 */
inet_csk_accept(struct sock * sk,struct proto_accept_arg * arg)665 struct sock *inet_csk_accept(struct sock *sk, struct proto_accept_arg *arg)
666 {
667 struct inet_connection_sock *icsk = inet_csk(sk);
668 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
669 struct request_sock *req;
670 struct sock *newsk;
671 int error;
672
673 lock_sock(sk);
674
675 /* We need to make sure that this socket is listening,
676 * and that it has something pending.
677 */
678 error = -EINVAL;
679 if (sk->sk_state != TCP_LISTEN)
680 goto out_err;
681
682 /* Find already established connection */
683 if (reqsk_queue_empty(queue)) {
684 long timeo = sock_rcvtimeo(sk, arg->flags & O_NONBLOCK);
685
686 /* If this is a non blocking socket don't sleep */
687 error = -EAGAIN;
688 if (!timeo)
689 goto out_err;
690
691 error = inet_csk_wait_for_connect(sk, timeo);
692 if (error)
693 goto out_err;
694 }
695 req = reqsk_queue_remove(queue, sk);
696 arg->is_empty = reqsk_queue_empty(queue);
697 newsk = req->sk;
698
699 if (sk->sk_protocol == IPPROTO_TCP &&
700 tcp_rsk(req)->tfo_listener) {
701 spin_lock_bh(&queue->fastopenq.lock);
702 if (tcp_rsk(req)->tfo_listener) {
703 /* We are still waiting for the final ACK from 3WHS
704 * so can't free req now. Instead, we set req->sk to
705 * NULL to signify that the child socket is taken
706 * so reqsk_fastopen_remove() will free the req
707 * when 3WHS finishes (or is aborted).
708 */
709 req->sk = NULL;
710 req = NULL;
711 }
712 spin_unlock_bh(&queue->fastopenq.lock);
713 }
714
715 out:
716 release_sock(sk);
717 if (newsk && mem_cgroup_sockets_enabled) {
718 gfp_t gfp = GFP_KERNEL | __GFP_NOFAIL;
719 int amt = 0;
720
721 /* atomically get the memory usage, set and charge the
722 * newsk->sk_memcg.
723 */
724 lock_sock(newsk);
725
726 mem_cgroup_sk_alloc(newsk);
727 if (newsk->sk_memcg) {
728 /* The socket has not been accepted yet, no need
729 * to look at newsk->sk_wmem_queued.
730 */
731 amt = sk_mem_pages(newsk->sk_forward_alloc +
732 atomic_read(&newsk->sk_rmem_alloc));
733 }
734
735 if (amt)
736 mem_cgroup_charge_skmem(newsk->sk_memcg, amt, gfp);
737 kmem_cache_charge(newsk, gfp);
738
739 release_sock(newsk);
740 }
741 if (req)
742 reqsk_put(req);
743
744 if (newsk)
745 inet_init_csk_locks(newsk);
746
747 return newsk;
748 out_err:
749 newsk = NULL;
750 req = NULL;
751 arg->err = error;
752 goto out;
753 }
754 EXPORT_SYMBOL(inet_csk_accept);
755
756 /*
757 * Using different timers for retransmit, delayed acks and probes
758 * We may wish use just one timer maintaining a list of expire jiffies
759 * to optimize.
760 */
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))761 void inet_csk_init_xmit_timers(struct sock *sk,
762 void (*retransmit_handler)(struct timer_list *t),
763 void (*delack_handler)(struct timer_list *t),
764 void (*keepalive_handler)(struct timer_list *t))
765 {
766 struct inet_connection_sock *icsk = inet_csk(sk);
767
768 timer_setup(&icsk->icsk_retransmit_timer, retransmit_handler, 0);
769 timer_setup(&icsk->icsk_delack_timer, delack_handler, 0);
770 timer_setup(&sk->sk_timer, keepalive_handler, 0);
771 icsk->icsk_pending = icsk->icsk_ack.pending = 0;
772 }
773 EXPORT_SYMBOL(inet_csk_init_xmit_timers);
774
inet_csk_clear_xmit_timers(struct sock * sk)775 void inet_csk_clear_xmit_timers(struct sock *sk)
776 {
777 struct inet_connection_sock *icsk = inet_csk(sk);
778
779 icsk->icsk_pending = icsk->icsk_ack.pending = 0;
780
781 sk_stop_timer(sk, &icsk->icsk_retransmit_timer);
782 sk_stop_timer(sk, &icsk->icsk_delack_timer);
783 sk_stop_timer(sk, &sk->sk_timer);
784 }
785 EXPORT_SYMBOL(inet_csk_clear_xmit_timers);
786
inet_csk_clear_xmit_timers_sync(struct sock * sk)787 void inet_csk_clear_xmit_timers_sync(struct sock *sk)
788 {
789 struct inet_connection_sock *icsk = inet_csk(sk);
790
791 /* ongoing timer handlers need to acquire socket lock. */
792 sock_not_owned_by_me(sk);
793
794 icsk->icsk_pending = icsk->icsk_ack.pending = 0;
795
796 sk_stop_timer_sync(sk, &icsk->icsk_retransmit_timer);
797 sk_stop_timer_sync(sk, &icsk->icsk_delack_timer);
798 sk_stop_timer_sync(sk, &sk->sk_timer);
799 }
800
inet_csk_delete_keepalive_timer(struct sock * sk)801 void inet_csk_delete_keepalive_timer(struct sock *sk)
802 {
803 sk_stop_timer(sk, &sk->sk_timer);
804 }
805 EXPORT_SYMBOL(inet_csk_delete_keepalive_timer);
806
inet_csk_reset_keepalive_timer(struct sock * sk,unsigned long len)807 void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len)
808 {
809 sk_reset_timer(sk, &sk->sk_timer, jiffies + len);
810 }
811 EXPORT_SYMBOL(inet_csk_reset_keepalive_timer);
812
inet_csk_route_req(const struct sock * sk,struct flowi4 * fl4,const struct request_sock * req)813 struct dst_entry *inet_csk_route_req(const struct sock *sk,
814 struct flowi4 *fl4,
815 const struct request_sock *req)
816 {
817 const struct inet_request_sock *ireq = inet_rsk(req);
818 struct net *net = read_pnet(&ireq->ireq_net);
819 struct ip_options_rcu *opt;
820 struct rtable *rt;
821
822 rcu_read_lock();
823 opt = rcu_dereference(ireq->ireq_opt);
824
825 flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
826 ip_sock_rt_tos(sk), ip_sock_rt_scope(sk),
827 sk->sk_protocol, inet_sk_flowi_flags(sk),
828 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
829 ireq->ir_loc_addr, ireq->ir_rmt_port,
830 htons(ireq->ir_num), sk->sk_uid);
831 security_req_classify_flow(req, flowi4_to_flowi_common(fl4));
832 rt = ip_route_output_flow(net, fl4, sk);
833 if (IS_ERR(rt))
834 goto no_route;
835 if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
836 goto route_err;
837 rcu_read_unlock();
838 return &rt->dst;
839
840 route_err:
841 ip_rt_put(rt);
842 no_route:
843 rcu_read_unlock();
844 __IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
845 return NULL;
846 }
847 EXPORT_SYMBOL_GPL(inet_csk_route_req);
848
inet_csk_route_child_sock(const struct sock * sk,struct sock * newsk,const struct request_sock * req)849 struct dst_entry *inet_csk_route_child_sock(const struct sock *sk,
850 struct sock *newsk,
851 const struct request_sock *req)
852 {
853 const struct inet_request_sock *ireq = inet_rsk(req);
854 struct net *net = read_pnet(&ireq->ireq_net);
855 struct inet_sock *newinet = inet_sk(newsk);
856 struct ip_options_rcu *opt;
857 struct flowi4 *fl4;
858 struct rtable *rt;
859
860 opt = rcu_dereference(ireq->ireq_opt);
861 fl4 = &newinet->cork.fl.u.ip4;
862
863 flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark,
864 ip_sock_rt_tos(sk), ip_sock_rt_scope(sk),
865 sk->sk_protocol, inet_sk_flowi_flags(sk),
866 (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr,
867 ireq->ir_loc_addr, ireq->ir_rmt_port,
868 htons(ireq->ir_num), sk->sk_uid);
869 security_req_classify_flow(req, flowi4_to_flowi_common(fl4));
870 rt = ip_route_output_flow(net, fl4, sk);
871 if (IS_ERR(rt))
872 goto no_route;
873 if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway)
874 goto route_err;
875 return &rt->dst;
876
877 route_err:
878 ip_rt_put(rt);
879 no_route:
880 __IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES);
881 return NULL;
882 }
883 EXPORT_SYMBOL_GPL(inet_csk_route_child_sock);
884
885 /* 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)886 static void syn_ack_recalc(struct request_sock *req,
887 const int max_syn_ack_retries,
888 const u8 rskq_defer_accept,
889 int *expire, int *resend)
890 {
891 if (!rskq_defer_accept) {
892 *expire = req->num_timeout >= max_syn_ack_retries;
893 *resend = 1;
894 return;
895 }
896 *expire = req->num_timeout >= max_syn_ack_retries &&
897 (!inet_rsk(req)->acked || req->num_timeout >= rskq_defer_accept);
898 /* Do not resend while waiting for data after ACK,
899 * start to resend on end of deferring period to give
900 * last chance for data or ACK to create established socket.
901 */
902 *resend = !inet_rsk(req)->acked ||
903 req->num_timeout >= rskq_defer_accept - 1;
904 }
905
inet_rtx_syn_ack(const struct sock * parent,struct request_sock * req)906 int inet_rtx_syn_ack(const struct sock *parent, struct request_sock *req)
907 {
908 int err = req->rsk_ops->rtx_syn_ack(parent, req);
909
910 if (!err)
911 req->num_retrans++;
912 return err;
913 }
914 EXPORT_SYMBOL(inet_rtx_syn_ack);
915
916 static struct request_sock *
reqsk_alloc_noprof(const struct request_sock_ops * ops,struct sock * sk_listener,bool attach_listener)917 reqsk_alloc_noprof(const struct request_sock_ops *ops, struct sock *sk_listener,
918 bool attach_listener)
919 {
920 struct request_sock *req;
921
922 req = kmem_cache_alloc_noprof(ops->slab, GFP_ATOMIC | __GFP_NOWARN);
923 if (!req)
924 return NULL;
925 req->rsk_listener = NULL;
926 if (attach_listener) {
927 if (unlikely(!refcount_inc_not_zero(&sk_listener->sk_refcnt))) {
928 kmem_cache_free(ops->slab, req);
929 return NULL;
930 }
931 req->rsk_listener = sk_listener;
932 }
933 req->rsk_ops = ops;
934 req_to_sk(req)->sk_prot = sk_listener->sk_prot;
935 sk_node_init(&req_to_sk(req)->sk_node);
936 sk_tx_queue_clear(req_to_sk(req));
937 req->saved_syn = NULL;
938 req->syncookie = 0;
939 req->timeout = 0;
940 req->num_timeout = 0;
941 req->num_retrans = 0;
942 req->sk = NULL;
943 refcount_set(&req->rsk_refcnt, 0);
944
945 return req;
946 }
947 #define reqsk_alloc(...) alloc_hooks(reqsk_alloc_noprof(__VA_ARGS__))
948
inet_reqsk_alloc(const struct request_sock_ops * ops,struct sock * sk_listener,bool attach_listener)949 struct request_sock *inet_reqsk_alloc(const struct request_sock_ops *ops,
950 struct sock *sk_listener,
951 bool attach_listener)
952 {
953 struct request_sock *req = reqsk_alloc(ops, sk_listener,
954 attach_listener);
955
956 if (req) {
957 struct inet_request_sock *ireq = inet_rsk(req);
958
959 ireq->ireq_opt = NULL;
960 #if IS_ENABLED(CONFIG_IPV6)
961 ireq->pktopts = NULL;
962 #endif
963 atomic64_set(&ireq->ir_cookie, 0);
964 ireq->ireq_state = TCP_NEW_SYN_RECV;
965 write_pnet(&ireq->ireq_net, sock_net(sk_listener));
966 ireq->ireq_family = sk_listener->sk_family;
967 req->timeout = TCP_TIMEOUT_INIT;
968 }
969
970 return req;
971 }
972 EXPORT_SYMBOL(inet_reqsk_alloc);
973
inet_reqsk_clone(struct request_sock * req,struct sock * sk)974 static struct request_sock *inet_reqsk_clone(struct request_sock *req,
975 struct sock *sk)
976 {
977 struct sock *req_sk, *nreq_sk;
978 struct request_sock *nreq;
979
980 nreq = kmem_cache_alloc(req->rsk_ops->slab, GFP_ATOMIC | __GFP_NOWARN);
981 if (!nreq) {
982 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQFAILURE);
983
984 /* paired with refcount_inc_not_zero() in reuseport_migrate_sock() */
985 sock_put(sk);
986 return NULL;
987 }
988
989 req_sk = req_to_sk(req);
990 nreq_sk = req_to_sk(nreq);
991
992 memcpy(nreq_sk, req_sk,
993 offsetof(struct sock, sk_dontcopy_begin));
994 unsafe_memcpy(&nreq_sk->sk_dontcopy_end, &req_sk->sk_dontcopy_end,
995 req->rsk_ops->obj_size - offsetof(struct sock, sk_dontcopy_end),
996 /* alloc is larger than struct, see above */);
997
998 sk_node_init(&nreq_sk->sk_node);
999 nreq_sk->sk_tx_queue_mapping = req_sk->sk_tx_queue_mapping;
1000 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING
1001 nreq_sk->sk_rx_queue_mapping = req_sk->sk_rx_queue_mapping;
1002 #endif
1003 nreq_sk->sk_incoming_cpu = req_sk->sk_incoming_cpu;
1004
1005 nreq->rsk_listener = sk;
1006
1007 /* We need not acquire fastopenq->lock
1008 * because the child socket is locked in inet_csk_listen_stop().
1009 */
1010 if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(nreq)->tfo_listener)
1011 rcu_assign_pointer(tcp_sk(nreq->sk)->fastopen_rsk, nreq);
1012
1013 return nreq;
1014 }
1015
reqsk_queue_migrated(struct request_sock_queue * queue,const struct request_sock * req)1016 static void reqsk_queue_migrated(struct request_sock_queue *queue,
1017 const struct request_sock *req)
1018 {
1019 if (req->num_timeout == 0)
1020 atomic_inc(&queue->young);
1021 atomic_inc(&queue->qlen);
1022 }
1023
reqsk_migrate_reset(struct request_sock * req)1024 static void reqsk_migrate_reset(struct request_sock *req)
1025 {
1026 req->saved_syn = NULL;
1027 #if IS_ENABLED(CONFIG_IPV6)
1028 inet_rsk(req)->ipv6_opt = NULL;
1029 inet_rsk(req)->pktopts = NULL;
1030 #else
1031 inet_rsk(req)->ireq_opt = NULL;
1032 #endif
1033 }
1034
1035 /* return true if req was found in the ehash table */
reqsk_queue_unlink(struct request_sock * req)1036 static bool reqsk_queue_unlink(struct request_sock *req)
1037 {
1038 struct sock *sk = req_to_sk(req);
1039 bool found = false;
1040
1041 if (sk_hashed(sk)) {
1042 struct inet_hashinfo *hashinfo = tcp_or_dccp_get_hashinfo(sk);
1043 spinlock_t *lock = inet_ehash_lockp(hashinfo, req->rsk_hash);
1044
1045 spin_lock(lock);
1046 found = __sk_nulls_del_node_init_rcu(sk);
1047 spin_unlock(lock);
1048 }
1049
1050 return found;
1051 }
1052
__inet_csk_reqsk_queue_drop(struct sock * sk,struct request_sock * req,bool from_timer)1053 static bool __inet_csk_reqsk_queue_drop(struct sock *sk,
1054 struct request_sock *req,
1055 bool from_timer)
1056 {
1057 bool unlinked = reqsk_queue_unlink(req);
1058
1059 if (!from_timer && timer_delete_sync(&req->rsk_timer))
1060 reqsk_put(req);
1061
1062 if (unlinked) {
1063 reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req);
1064 reqsk_put(req);
1065 }
1066
1067 return unlinked;
1068 }
1069
inet_csk_reqsk_queue_drop(struct sock * sk,struct request_sock * req)1070 bool inet_csk_reqsk_queue_drop(struct sock *sk, struct request_sock *req)
1071 {
1072 return __inet_csk_reqsk_queue_drop(sk, req, false);
1073 }
1074 EXPORT_SYMBOL(inet_csk_reqsk_queue_drop);
1075
inet_csk_reqsk_queue_drop_and_put(struct sock * sk,struct request_sock * req)1076 void inet_csk_reqsk_queue_drop_and_put(struct sock *sk, struct request_sock *req)
1077 {
1078 inet_csk_reqsk_queue_drop(sk, req);
1079 reqsk_put(req);
1080 }
1081 EXPORT_SYMBOL(inet_csk_reqsk_queue_drop_and_put);
1082
reqsk_timer_handler(struct timer_list * t)1083 static void reqsk_timer_handler(struct timer_list *t)
1084 {
1085 struct request_sock *req = from_timer(req, t, rsk_timer);
1086 struct request_sock *nreq = NULL, *oreq = req;
1087 struct sock *sk_listener = req->rsk_listener;
1088 struct inet_connection_sock *icsk;
1089 struct request_sock_queue *queue;
1090 struct net *net;
1091 int max_syn_ack_retries, qlen, expire = 0, resend = 0;
1092
1093 if (inet_sk_state_load(sk_listener) != TCP_LISTEN) {
1094 struct sock *nsk;
1095
1096 nsk = reuseport_migrate_sock(sk_listener, req_to_sk(req), NULL);
1097 if (!nsk)
1098 goto drop;
1099
1100 nreq = inet_reqsk_clone(req, nsk);
1101 if (!nreq)
1102 goto drop;
1103
1104 /* The new timer for the cloned req can decrease the 2
1105 * by calling inet_csk_reqsk_queue_drop_and_put(), so
1106 * hold another count to prevent use-after-free and
1107 * call reqsk_put() just before return.
1108 */
1109 refcount_set(&nreq->rsk_refcnt, 2 + 1);
1110 timer_setup(&nreq->rsk_timer, reqsk_timer_handler, TIMER_PINNED);
1111 reqsk_queue_migrated(&inet_csk(nsk)->icsk_accept_queue, req);
1112
1113 req = nreq;
1114 sk_listener = nsk;
1115 }
1116
1117 icsk = inet_csk(sk_listener);
1118 net = sock_net(sk_listener);
1119 max_syn_ack_retries = READ_ONCE(icsk->icsk_syn_retries) ? :
1120 READ_ONCE(net->ipv4.sysctl_tcp_synack_retries);
1121 /* Normally all the openreqs are young and become mature
1122 * (i.e. converted to established socket) for first timeout.
1123 * If synack was not acknowledged for 1 second, it means
1124 * one of the following things: synack was lost, ack was lost,
1125 * rtt is high or nobody planned to ack (i.e. synflood).
1126 * When server is a bit loaded, queue is populated with old
1127 * open requests, reducing effective size of queue.
1128 * When server is well loaded, queue size reduces to zero
1129 * after several minutes of work. It is not synflood,
1130 * it is normal operation. The solution is pruning
1131 * too old entries overriding normal timeout, when
1132 * situation becomes dangerous.
1133 *
1134 * Essentially, we reserve half of room for young
1135 * embrions; and abort old ones without pity, if old
1136 * ones are about to clog our table.
1137 */
1138 queue = &icsk->icsk_accept_queue;
1139 qlen = reqsk_queue_len(queue);
1140 if ((qlen << 1) > max(8U, READ_ONCE(sk_listener->sk_max_ack_backlog))) {
1141 int young = reqsk_queue_len_young(queue) << 1;
1142
1143 while (max_syn_ack_retries > 2) {
1144 if (qlen < young)
1145 break;
1146 max_syn_ack_retries--;
1147 young <<= 1;
1148 }
1149 }
1150 syn_ack_recalc(req, max_syn_ack_retries, READ_ONCE(queue->rskq_defer_accept),
1151 &expire, &resend);
1152 req->rsk_ops->syn_ack_timeout(req);
1153 if (!expire &&
1154 (!resend ||
1155 !inet_rtx_syn_ack(sk_listener, req) ||
1156 inet_rsk(req)->acked)) {
1157 if (req->num_timeout++ == 0)
1158 atomic_dec(&queue->young);
1159 mod_timer(&req->rsk_timer, jiffies + reqsk_timeout(req, TCP_RTO_MAX));
1160
1161 if (!nreq)
1162 return;
1163
1164 if (!inet_ehash_insert(req_to_sk(nreq), req_to_sk(oreq), NULL)) {
1165 /* delete timer */
1166 __inet_csk_reqsk_queue_drop(sk_listener, nreq, true);
1167 goto no_ownership;
1168 }
1169
1170 __NET_INC_STATS(net, LINUX_MIB_TCPMIGRATEREQSUCCESS);
1171 reqsk_migrate_reset(oreq);
1172 reqsk_queue_removed(&inet_csk(oreq->rsk_listener)->icsk_accept_queue, oreq);
1173 reqsk_put(oreq);
1174
1175 reqsk_put(nreq);
1176 return;
1177 }
1178
1179 /* Even if we can clone the req, we may need not retransmit any more
1180 * SYN+ACKs (nreq->num_timeout > max_syn_ack_retries, etc), or another
1181 * CPU may win the "own_req" race so that inet_ehash_insert() fails.
1182 */
1183 if (nreq) {
1184 __NET_INC_STATS(net, LINUX_MIB_TCPMIGRATEREQFAILURE);
1185 no_ownership:
1186 reqsk_migrate_reset(nreq);
1187 reqsk_queue_removed(queue, nreq);
1188 __reqsk_free(nreq);
1189 }
1190
1191 drop:
1192 __inet_csk_reqsk_queue_drop(sk_listener, oreq, true);
1193 reqsk_put(oreq);
1194 }
1195
reqsk_queue_hash_req(struct request_sock * req,unsigned long timeout)1196 static bool reqsk_queue_hash_req(struct request_sock *req,
1197 unsigned long timeout)
1198 {
1199 bool found_dup_sk = false;
1200
1201 if (!inet_ehash_insert(req_to_sk(req), NULL, &found_dup_sk))
1202 return false;
1203
1204 /* The timer needs to be setup after a successful insertion. */
1205 timer_setup(&req->rsk_timer, reqsk_timer_handler, TIMER_PINNED);
1206 mod_timer(&req->rsk_timer, jiffies + timeout);
1207
1208 /* before letting lookups find us, make sure all req fields
1209 * are committed to memory and refcnt initialized.
1210 */
1211 smp_wmb();
1212 refcount_set(&req->rsk_refcnt, 2 + 1);
1213 return true;
1214 }
1215
inet_csk_reqsk_queue_hash_add(struct sock * sk,struct request_sock * req,unsigned long timeout)1216 bool inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req,
1217 unsigned long timeout)
1218 {
1219 if (!reqsk_queue_hash_req(req, timeout))
1220 return false;
1221
1222 inet_csk_reqsk_queue_added(sk);
1223 return true;
1224 }
1225 EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add);
1226
inet_clone_ulp(const struct request_sock * req,struct sock * newsk,const gfp_t priority)1227 static void inet_clone_ulp(const struct request_sock *req, struct sock *newsk,
1228 const gfp_t priority)
1229 {
1230 struct inet_connection_sock *icsk = inet_csk(newsk);
1231
1232 if (!icsk->icsk_ulp_ops)
1233 return;
1234
1235 icsk->icsk_ulp_ops->clone(req, newsk, priority);
1236 }
1237
1238 /**
1239 * inet_csk_clone_lock - clone an inet socket, and lock its clone
1240 * @sk: the socket to clone
1241 * @req: request_sock
1242 * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1243 *
1244 * Caller must unlock socket even in error path (bh_unlock_sock(newsk))
1245 */
inet_csk_clone_lock(const struct sock * sk,const struct request_sock * req,const gfp_t priority)1246 struct sock *inet_csk_clone_lock(const struct sock *sk,
1247 const struct request_sock *req,
1248 const gfp_t priority)
1249 {
1250 struct sock *newsk = sk_clone_lock(sk, priority);
1251
1252 if (newsk) {
1253 struct inet_connection_sock *newicsk = inet_csk(newsk);
1254
1255 inet_sk_set_state(newsk, TCP_SYN_RECV);
1256 newicsk->icsk_bind_hash = NULL;
1257 newicsk->icsk_bind2_hash = NULL;
1258
1259 inet_sk(newsk)->inet_dport = inet_rsk(req)->ir_rmt_port;
1260 inet_sk(newsk)->inet_num = inet_rsk(req)->ir_num;
1261 inet_sk(newsk)->inet_sport = htons(inet_rsk(req)->ir_num);
1262
1263 /* listeners have SOCK_RCU_FREE, not the children */
1264 sock_reset_flag(newsk, SOCK_RCU_FREE);
1265
1266 inet_sk(newsk)->mc_list = NULL;
1267
1268 newsk->sk_mark = inet_rsk(req)->ir_mark;
1269 atomic64_set(&newsk->sk_cookie,
1270 atomic64_read(&inet_rsk(req)->ir_cookie));
1271
1272 newicsk->icsk_retransmits = 0;
1273 newicsk->icsk_backoff = 0;
1274 newicsk->icsk_probes_out = 0;
1275 newicsk->icsk_probes_tstamp = 0;
1276
1277 /* Deinitialize accept_queue to trap illegal accesses. */
1278 memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue));
1279
1280 inet_clone_ulp(req, newsk, priority);
1281
1282 security_inet_csk_clone(newsk, req);
1283
1284 trace_android_vh_inet_csk_clone_lock(newsk, req);
1285 }
1286 return newsk;
1287 }
1288 EXPORT_SYMBOL_GPL(inet_csk_clone_lock);
1289
1290 /*
1291 * At this point, there should be no process reference to this
1292 * socket, and thus no user references at all. Therefore we
1293 * can assume the socket waitqueue is inactive and nobody will
1294 * try to jump onto it.
1295 */
inet_csk_destroy_sock(struct sock * sk)1296 void inet_csk_destroy_sock(struct sock *sk)
1297 {
1298 WARN_ON(sk->sk_state != TCP_CLOSE);
1299 WARN_ON(!sock_flag(sk, SOCK_DEAD));
1300
1301 /* It cannot be in hash table! */
1302 WARN_ON(!sk_unhashed(sk));
1303
1304 /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */
1305 WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash);
1306
1307 sk->sk_prot->destroy(sk);
1308
1309 sk_stream_kill_queues(sk);
1310
1311 xfrm_sk_free_policy(sk);
1312
1313 this_cpu_dec(*sk->sk_prot->orphan_count);
1314
1315 sock_put(sk);
1316 }
1317 EXPORT_SYMBOL(inet_csk_destroy_sock);
1318
1319 /* This function allows to force a closure of a socket after the call to
1320 * tcp/dccp_create_openreq_child().
1321 */
inet_csk_prepare_forced_close(struct sock * sk)1322 void inet_csk_prepare_forced_close(struct sock *sk)
1323 __releases(&sk->sk_lock.slock)
1324 {
1325 /* sk_clone_lock locked the socket and set refcnt to 2 */
1326 bh_unlock_sock(sk);
1327 sock_put(sk);
1328 inet_csk_prepare_for_destroy_sock(sk);
1329 inet_sk(sk)->inet_num = 0;
1330 }
1331 EXPORT_SYMBOL(inet_csk_prepare_forced_close);
1332
inet_ulp_can_listen(const struct sock * sk)1333 static int inet_ulp_can_listen(const struct sock *sk)
1334 {
1335 const struct inet_connection_sock *icsk = inet_csk(sk);
1336
1337 if (icsk->icsk_ulp_ops && !icsk->icsk_ulp_ops->clone)
1338 return -EINVAL;
1339
1340 return 0;
1341 }
1342
inet_csk_listen_start(struct sock * sk)1343 int inet_csk_listen_start(struct sock *sk)
1344 {
1345 struct inet_connection_sock *icsk = inet_csk(sk);
1346 struct inet_sock *inet = inet_sk(sk);
1347 int err;
1348
1349 err = inet_ulp_can_listen(sk);
1350 if (unlikely(err))
1351 return err;
1352
1353 reqsk_queue_alloc(&icsk->icsk_accept_queue);
1354
1355 sk->sk_ack_backlog = 0;
1356 inet_csk_delack_init(sk);
1357
1358 /* There is race window here: we announce ourselves listening,
1359 * but this transition is still not validated by get_port().
1360 * It is OK, because this socket enters to hash table only
1361 * after validation is complete.
1362 */
1363 inet_sk_state_store(sk, TCP_LISTEN);
1364 err = sk->sk_prot->get_port(sk, inet->inet_num);
1365 if (!err) {
1366 inet->inet_sport = htons(inet->inet_num);
1367
1368 sk_dst_reset(sk);
1369 err = sk->sk_prot->hash(sk);
1370
1371 if (likely(!err))
1372 return 0;
1373 }
1374
1375 inet_sk_set_state(sk, TCP_CLOSE);
1376 return err;
1377 }
1378 EXPORT_SYMBOL_GPL(inet_csk_listen_start);
1379
inet_child_forget(struct sock * sk,struct request_sock * req,struct sock * child)1380 static void inet_child_forget(struct sock *sk, struct request_sock *req,
1381 struct sock *child)
1382 {
1383 sk->sk_prot->disconnect(child, O_NONBLOCK);
1384
1385 sock_orphan(child);
1386
1387 this_cpu_inc(*sk->sk_prot->orphan_count);
1388
1389 if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(req)->tfo_listener) {
1390 BUG_ON(rcu_access_pointer(tcp_sk(child)->fastopen_rsk) != req);
1391 BUG_ON(sk != req->rsk_listener);
1392
1393 /* Paranoid, to prevent race condition if
1394 * an inbound pkt destined for child is
1395 * blocked by sock lock in tcp_v4_rcv().
1396 * Also to satisfy an assertion in
1397 * tcp_v4_destroy_sock().
1398 */
1399 RCU_INIT_POINTER(tcp_sk(child)->fastopen_rsk, NULL);
1400 }
1401 inet_csk_destroy_sock(child);
1402 }
1403
inet_csk_reqsk_queue_add(struct sock * sk,struct request_sock * req,struct sock * child)1404 struct sock *inet_csk_reqsk_queue_add(struct sock *sk,
1405 struct request_sock *req,
1406 struct sock *child)
1407 {
1408 struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue;
1409
1410 spin_lock(&queue->rskq_lock);
1411 if (unlikely(sk->sk_state != TCP_LISTEN)) {
1412 inet_child_forget(sk, req, child);
1413 child = NULL;
1414 } else {
1415 req->sk = child;
1416 req->dl_next = NULL;
1417 if (queue->rskq_accept_head == NULL)
1418 WRITE_ONCE(queue->rskq_accept_head, req);
1419 else
1420 queue->rskq_accept_tail->dl_next = req;
1421 queue->rskq_accept_tail = req;
1422 sk_acceptq_added(sk);
1423 }
1424 spin_unlock(&queue->rskq_lock);
1425 return child;
1426 }
1427 EXPORT_SYMBOL(inet_csk_reqsk_queue_add);
1428
inet_csk_complete_hashdance(struct sock * sk,struct sock * child,struct request_sock * req,bool own_req)1429 struct sock *inet_csk_complete_hashdance(struct sock *sk, struct sock *child,
1430 struct request_sock *req, bool own_req)
1431 {
1432 if (own_req) {
1433 inet_csk_reqsk_queue_drop(req->rsk_listener, req);
1434 reqsk_queue_removed(&inet_csk(req->rsk_listener)->icsk_accept_queue, req);
1435
1436 if (sk != req->rsk_listener) {
1437 /* another listening sk has been selected,
1438 * migrate the req to it.
1439 */
1440 struct request_sock *nreq;
1441
1442 /* hold a refcnt for the nreq->rsk_listener
1443 * which is assigned in inet_reqsk_clone()
1444 */
1445 sock_hold(sk);
1446 nreq = inet_reqsk_clone(req, sk);
1447 if (!nreq) {
1448 inet_child_forget(sk, req, child);
1449 goto child_put;
1450 }
1451
1452 refcount_set(&nreq->rsk_refcnt, 1);
1453 if (inet_csk_reqsk_queue_add(sk, nreq, child)) {
1454 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQSUCCESS);
1455 reqsk_migrate_reset(req);
1456 reqsk_put(req);
1457 return child;
1458 }
1459
1460 __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQFAILURE);
1461 reqsk_migrate_reset(nreq);
1462 __reqsk_free(nreq);
1463 } else if (inet_csk_reqsk_queue_add(sk, req, child)) {
1464 return child;
1465 }
1466 }
1467 /* Too bad, another child took ownership of the request, undo. */
1468 child_put:
1469 bh_unlock_sock(child);
1470 sock_put(child);
1471 return NULL;
1472 }
1473 EXPORT_SYMBOL(inet_csk_complete_hashdance);
1474
1475 /*
1476 * This routine closes sockets which have been at least partially
1477 * opened, but not yet accepted.
1478 */
inet_csk_listen_stop(struct sock * sk)1479 void inet_csk_listen_stop(struct sock *sk)
1480 {
1481 struct inet_connection_sock *icsk = inet_csk(sk);
1482 struct request_sock_queue *queue = &icsk->icsk_accept_queue;
1483 struct request_sock *next, *req;
1484
1485 /* Following specs, it would be better either to send FIN
1486 * (and enter FIN-WAIT-1, it is normal close)
1487 * or to send active reset (abort).
1488 * Certainly, it is pretty dangerous while synflood, but it is
1489 * bad justification for our negligence 8)
1490 * To be honest, we are not able to make either
1491 * of the variants now. --ANK
1492 */
1493 while ((req = reqsk_queue_remove(queue, sk)) != NULL) {
1494 struct sock *child = req->sk, *nsk;
1495 struct request_sock *nreq;
1496
1497 local_bh_disable();
1498 bh_lock_sock(child);
1499 WARN_ON(sock_owned_by_user(child));
1500 sock_hold(child);
1501
1502 nsk = reuseport_migrate_sock(sk, child, NULL);
1503 if (nsk) {
1504 nreq = inet_reqsk_clone(req, nsk);
1505 if (nreq) {
1506 refcount_set(&nreq->rsk_refcnt, 1);
1507
1508 if (inet_csk_reqsk_queue_add(nsk, nreq, child)) {
1509 __NET_INC_STATS(sock_net(nsk),
1510 LINUX_MIB_TCPMIGRATEREQSUCCESS);
1511 reqsk_migrate_reset(req);
1512 } else {
1513 __NET_INC_STATS(sock_net(nsk),
1514 LINUX_MIB_TCPMIGRATEREQFAILURE);
1515 reqsk_migrate_reset(nreq);
1516 __reqsk_free(nreq);
1517 }
1518
1519 /* inet_csk_reqsk_queue_add() has already
1520 * called inet_child_forget() on failure case.
1521 */
1522 goto skip_child_forget;
1523 }
1524 }
1525
1526 inet_child_forget(sk, req, child);
1527 skip_child_forget:
1528 reqsk_put(req);
1529 bh_unlock_sock(child);
1530 local_bh_enable();
1531 sock_put(child);
1532
1533 cond_resched();
1534 }
1535 if (queue->fastopenq.rskq_rst_head) {
1536 /* Free all the reqs queued in rskq_rst_head. */
1537 spin_lock_bh(&queue->fastopenq.lock);
1538 req = queue->fastopenq.rskq_rst_head;
1539 queue->fastopenq.rskq_rst_head = NULL;
1540 spin_unlock_bh(&queue->fastopenq.lock);
1541 while (req != NULL) {
1542 next = req->dl_next;
1543 reqsk_put(req);
1544 req = next;
1545 }
1546 }
1547 WARN_ON_ONCE(sk->sk_ack_backlog);
1548 }
1549 EXPORT_SYMBOL_GPL(inet_csk_listen_stop);
1550
inet_csk_addr2sockaddr(struct sock * sk,struct sockaddr * uaddr)1551 void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr)
1552 {
1553 struct sockaddr_in *sin = (struct sockaddr_in *)uaddr;
1554 const struct inet_sock *inet = inet_sk(sk);
1555
1556 sin->sin_family = AF_INET;
1557 sin->sin_addr.s_addr = inet->inet_daddr;
1558 sin->sin_port = inet->inet_dport;
1559 }
1560 EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr);
1561
inet_csk_rebuild_route(struct sock * sk,struct flowi * fl)1562 static struct dst_entry *inet_csk_rebuild_route(struct sock *sk, struct flowi *fl)
1563 {
1564 const struct inet_sock *inet = inet_sk(sk);
1565 const struct ip_options_rcu *inet_opt;
1566 __be32 daddr = inet->inet_daddr;
1567 struct flowi4 *fl4;
1568 struct rtable *rt;
1569
1570 rcu_read_lock();
1571 inet_opt = rcu_dereference(inet->inet_opt);
1572 if (inet_opt && inet_opt->opt.srr)
1573 daddr = inet_opt->opt.faddr;
1574 fl4 = &fl->u.ip4;
1575 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr,
1576 inet->inet_saddr, inet->inet_dport,
1577 inet->inet_sport, sk->sk_protocol,
1578 ip_sock_rt_tos(sk), sk->sk_bound_dev_if);
1579 if (IS_ERR(rt))
1580 rt = NULL;
1581 if (rt)
1582 sk_setup_caps(sk, &rt->dst);
1583 rcu_read_unlock();
1584
1585 return &rt->dst;
1586 }
1587
inet_csk_update_pmtu(struct sock * sk,u32 mtu)1588 struct dst_entry *inet_csk_update_pmtu(struct sock *sk, u32 mtu)
1589 {
1590 struct dst_entry *dst = __sk_dst_check(sk, 0);
1591 struct inet_sock *inet = inet_sk(sk);
1592
1593 if (!dst) {
1594 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
1595 if (!dst)
1596 goto out;
1597 }
1598 dst->ops->update_pmtu(dst, sk, NULL, mtu, true);
1599
1600 dst = __sk_dst_check(sk, 0);
1601 if (!dst)
1602 dst = inet_csk_rebuild_route(sk, &inet->cork.fl);
1603 out:
1604 return dst;
1605 }
1606 EXPORT_SYMBOL_GPL(inet_csk_update_pmtu);
1607