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 | |
26 | #if IS_ENABLED(CONFIG_IPV6) |
27 | /* match_sk*_wildcard == true: IPV6_ADDR_ANY equals to any IPv6 addresses |
28 | * if IPv6 only, and any IPv4 addresses |
29 | * if not IPv6 only |
30 | * match_sk*_wildcard == false: addresses must be exactly the same, i.e. |
31 | * IPV6_ADDR_ANY only equals to IPV6_ADDR_ANY, |
32 | * and 0.0.0.0 equals to 0.0.0.0 only |
33 | */ |
34 | static bool ipv6_rcv_saddr_equal(const struct in6_addr *sk1_rcv_saddr6, |
35 | const struct in6_addr *sk2_rcv_saddr6, |
36 | __be32 sk1_rcv_saddr, __be32 sk2_rcv_saddr, |
37 | bool sk1_ipv6only, bool sk2_ipv6only, |
38 | bool match_sk1_wildcard, |
39 | bool match_sk2_wildcard) |
40 | { |
41 | int addr_type = ipv6_addr_type(addr: sk1_rcv_saddr6); |
42 | int addr_type2 = sk2_rcv_saddr6 ? ipv6_addr_type(addr: sk2_rcv_saddr6) : IPV6_ADDR_MAPPED; |
43 | |
44 | /* if both are mapped, treat as IPv4 */ |
45 | if (addr_type == IPV6_ADDR_MAPPED && addr_type2 == IPV6_ADDR_MAPPED) { |
46 | if (!sk2_ipv6only) { |
47 | if (sk1_rcv_saddr == sk2_rcv_saddr) |
48 | return true; |
49 | return (match_sk1_wildcard && !sk1_rcv_saddr) || |
50 | (match_sk2_wildcard && !sk2_rcv_saddr); |
51 | } |
52 | return false; |
53 | } |
54 | |
55 | if (addr_type == IPV6_ADDR_ANY && addr_type2 == IPV6_ADDR_ANY) |
56 | return true; |
57 | |
58 | if (addr_type2 == IPV6_ADDR_ANY && match_sk2_wildcard && |
59 | !(sk2_ipv6only && addr_type == IPV6_ADDR_MAPPED)) |
60 | return true; |
61 | |
62 | if (addr_type == IPV6_ADDR_ANY && match_sk1_wildcard && |
63 | !(sk1_ipv6only && addr_type2 == IPV6_ADDR_MAPPED)) |
64 | return true; |
65 | |
66 | if (sk2_rcv_saddr6 && |
67 | ipv6_addr_equal(a1: sk1_rcv_saddr6, a2: sk2_rcv_saddr6)) |
68 | return true; |
69 | |
70 | return false; |
71 | } |
72 | #endif |
73 | |
74 | /* match_sk*_wildcard == true: 0.0.0.0 equals to any IPv4 addresses |
75 | * match_sk*_wildcard == false: addresses must be exactly the same, i.e. |
76 | * 0.0.0.0 only equals to 0.0.0.0 |
77 | */ |
78 | static bool ipv4_rcv_saddr_equal(__be32 sk1_rcv_saddr, __be32 sk2_rcv_saddr, |
79 | bool sk2_ipv6only, bool match_sk1_wildcard, |
80 | bool match_sk2_wildcard) |
81 | { |
82 | if (!sk2_ipv6only) { |
83 | if (sk1_rcv_saddr == sk2_rcv_saddr) |
84 | return true; |
85 | return (match_sk1_wildcard && !sk1_rcv_saddr) || |
86 | (match_sk2_wildcard && !sk2_rcv_saddr); |
87 | } |
88 | return false; |
89 | } |
90 | |
91 | bool inet_rcv_saddr_equal(const struct sock *sk, const struct sock *sk2, |
92 | bool match_wildcard) |
93 | { |
94 | #if IS_ENABLED(CONFIG_IPV6) |
95 | if (sk->sk_family == AF_INET6) |
96 | return ipv6_rcv_saddr_equal(sk1_rcv_saddr6: &sk->sk_v6_rcv_saddr, |
97 | sk2_rcv_saddr6: inet6_rcv_saddr(sk: sk2), |
98 | sk1_rcv_saddr: sk->sk_rcv_saddr, |
99 | sk2_rcv_saddr: sk2->sk_rcv_saddr, |
100 | ipv6_only_sock(sk), |
101 | ipv6_only_sock(sk2), |
102 | match_sk1_wildcard: match_wildcard, |
103 | match_sk2_wildcard: match_wildcard); |
104 | #endif |
105 | return ipv4_rcv_saddr_equal(sk1_rcv_saddr: sk->sk_rcv_saddr, sk2_rcv_saddr: sk2->sk_rcv_saddr, |
106 | ipv6_only_sock(sk2), match_sk1_wildcard: match_wildcard, |
107 | match_sk2_wildcard: match_wildcard); |
108 | } |
109 | EXPORT_SYMBOL(inet_rcv_saddr_equal); |
110 | |
111 | bool inet_rcv_saddr_any(const struct sock *sk) |
112 | { |
113 | #if IS_ENABLED(CONFIG_IPV6) |
114 | if (sk->sk_family == AF_INET6) |
115 | return ipv6_addr_any(a: &sk->sk_v6_rcv_saddr); |
116 | #endif |
117 | return !sk->sk_rcv_saddr; |
118 | } |
119 | |
120 | /** |
121 | * inet_sk_get_local_port_range - fetch ephemeral ports range |
122 | * @sk: socket |
123 | * @low: pointer to low port |
124 | * @high: pointer to high port |
125 | * |
126 | * Fetch netns port range (/proc/sys/net/ipv4/ip_local_port_range) |
127 | * Range can be overridden if socket got IP_LOCAL_PORT_RANGE option. |
128 | * Returns true if IP_LOCAL_PORT_RANGE was set on this socket. |
129 | */ |
130 | bool inet_sk_get_local_port_range(const struct sock *sk, int *low, int *high) |
131 | { |
132 | int lo, hi, sk_lo, sk_hi; |
133 | bool local_range = false; |
134 | u32 sk_range; |
135 | |
136 | inet_get_local_port_range(net: sock_net(sk), low: &lo, high: &hi); |
137 | |
138 | sk_range = READ_ONCE(inet_sk(sk)->local_port_range); |
139 | if (unlikely(sk_range)) { |
140 | sk_lo = sk_range & 0xffff; |
141 | sk_hi = sk_range >> 16; |
142 | |
143 | if (lo <= sk_lo && sk_lo <= hi) |
144 | lo = sk_lo; |
145 | if (lo <= sk_hi && sk_hi <= hi) |
146 | hi = sk_hi; |
147 | local_range = true; |
148 | } |
149 | |
150 | *low = lo; |
151 | *high = hi; |
152 | return local_range; |
153 | } |
154 | EXPORT_SYMBOL(inet_sk_get_local_port_range); |
155 | |
156 | static bool inet_use_bhash2_on_bind(const struct sock *sk) |
157 | { |
158 | #if IS_ENABLED(CONFIG_IPV6) |
159 | if (sk->sk_family == AF_INET6) { |
160 | int addr_type = ipv6_addr_type(addr: &sk->sk_v6_rcv_saddr); |
161 | |
162 | if (addr_type == IPV6_ADDR_ANY) |
163 | return false; |
164 | |
165 | if (addr_type != IPV6_ADDR_MAPPED) |
166 | return true; |
167 | } |
168 | #endif |
169 | return sk->sk_rcv_saddr != htonl(INADDR_ANY); |
170 | } |
171 | |
172 | static bool inet_bind_conflict(const struct sock *sk, struct sock *sk2, |
173 | kuid_t sk_uid, bool relax, |
174 | bool reuseport_cb_ok, bool reuseport_ok) |
175 | { |
176 | int bound_dev_if2; |
177 | |
178 | if (sk == sk2) |
179 | return false; |
180 | |
181 | bound_dev_if2 = READ_ONCE(sk2->sk_bound_dev_if); |
182 | |
183 | if (!sk->sk_bound_dev_if || !bound_dev_if2 || |
184 | sk->sk_bound_dev_if == bound_dev_if2) { |
185 | if (sk->sk_reuse && sk2->sk_reuse && |
186 | sk2->sk_state != TCP_LISTEN) { |
187 | if (!relax || (!reuseport_ok && sk->sk_reuseport && |
188 | sk2->sk_reuseport && reuseport_cb_ok && |
189 | (sk2->sk_state == TCP_TIME_WAIT || |
190 | uid_eq(left: sk_uid, right: sock_i_uid(sk: sk2))))) |
191 | return true; |
192 | } else if (!reuseport_ok || !sk->sk_reuseport || |
193 | !sk2->sk_reuseport || !reuseport_cb_ok || |
194 | (sk2->sk_state != TCP_TIME_WAIT && |
195 | !uid_eq(left: sk_uid, right: sock_i_uid(sk: sk2)))) { |
196 | return true; |
197 | } |
198 | } |
199 | return false; |
200 | } |
201 | |
202 | static bool __inet_bhash2_conflict(const struct sock *sk, struct sock *sk2, |
203 | kuid_t sk_uid, bool relax, |
204 | bool reuseport_cb_ok, bool reuseport_ok) |
205 | { |
206 | if (ipv6_only_sock(sk2)) { |
207 | if (sk->sk_family == AF_INET) |
208 | return false; |
209 | |
210 | #if IS_ENABLED(CONFIG_IPV6) |
211 | if (ipv6_addr_v4mapped(a: &sk->sk_v6_rcv_saddr)) |
212 | return false; |
213 | #endif |
214 | } |
215 | |
216 | return inet_bind_conflict(sk, sk2, sk_uid, relax, |
217 | reuseport_cb_ok, reuseport_ok); |
218 | } |
219 | |
220 | static bool inet_bhash2_conflict(const struct sock *sk, |
221 | const struct inet_bind2_bucket *tb2, |
222 | kuid_t sk_uid, |
223 | bool relax, bool reuseport_cb_ok, |
224 | bool reuseport_ok) |
225 | { |
226 | struct sock *sk2; |
227 | |
228 | sk_for_each_bound(sk2, &tb2->owners) { |
229 | if (__inet_bhash2_conflict(sk, sk2, sk_uid, relax, |
230 | reuseport_cb_ok, reuseport_ok)) |
231 | return true; |
232 | } |
233 | |
234 | return false; |
235 | } |
236 | |
237 | #define sk_for_each_bound_bhash(__sk, __tb2, __tb) \ |
238 | hlist_for_each_entry(__tb2, &(__tb)->bhash2, bhash_node) \ |
239 | sk_for_each_bound(sk2, &(__tb2)->owners) |
240 | |
241 | /* This should be called only when the tb and tb2 hashbuckets' locks are held */ |
242 | static int inet_csk_bind_conflict(const struct sock *sk, |
243 | const struct inet_bind_bucket *tb, |
244 | const struct inet_bind2_bucket *tb2, /* may be null */ |
245 | bool relax, bool reuseport_ok) |
246 | { |
247 | kuid_t uid = sock_i_uid(sk: (struct sock *)sk); |
248 | struct sock_reuseport *reuseport_cb; |
249 | bool reuseport_cb_ok; |
250 | struct sock *sk2; |
251 | |
252 | rcu_read_lock(); |
253 | reuseport_cb = rcu_dereference(sk->sk_reuseport_cb); |
254 | /* paired with WRITE_ONCE() in __reuseport_(add|detach)_closed_sock */ |
255 | reuseport_cb_ok = !reuseport_cb || READ_ONCE(reuseport_cb->num_closed_socks); |
256 | rcu_read_unlock(); |
257 | |
258 | /* Conflicts with an existing IPV6_ADDR_ANY (if ipv6) or INADDR_ANY (if |
259 | * ipv4) should have been checked already. We need to do these two |
260 | * checks separately because their spinlocks have to be acquired/released |
261 | * independently of each other, to prevent possible deadlocks |
262 | */ |
263 | if (inet_use_bhash2_on_bind(sk)) |
264 | return tb2 && inet_bhash2_conflict(sk, tb2, sk_uid: uid, relax, |
265 | reuseport_cb_ok, reuseport_ok); |
266 | |
267 | /* Unlike other sk lookup places we do not check |
268 | * for sk_net here, since _all_ the socks listed |
269 | * in tb->owners and tb2->owners list belong |
270 | * to the same net - the one this bucket belongs to. |
271 | */ |
272 | sk_for_each_bound_bhash(sk2, tb2, tb) { |
273 | if (!inet_bind_conflict(sk, sk2, sk_uid: uid, relax, reuseport_cb_ok, reuseport_ok)) |
274 | continue; |
275 | |
276 | if (inet_rcv_saddr_equal(sk, sk2, true)) |
277 | return true; |
278 | } |
279 | |
280 | return false; |
281 | } |
282 | |
283 | /* Determine if there is a bind conflict with an existing IPV6_ADDR_ANY (if ipv6) or |
284 | * INADDR_ANY (if ipv4) socket. |
285 | * |
286 | * Caller must hold bhash hashbucket lock with local bh disabled, to protect |
287 | * against concurrent binds on the port for addr any |
288 | */ |
289 | static bool inet_bhash2_addr_any_conflict(const struct sock *sk, int port, int l3mdev, |
290 | bool relax, bool reuseport_ok) |
291 | { |
292 | kuid_t uid = sock_i_uid(sk: (struct sock *)sk); |
293 | const struct net *net = sock_net(sk); |
294 | struct sock_reuseport *reuseport_cb; |
295 | struct inet_bind_hashbucket *head2; |
296 | struct inet_bind2_bucket *tb2; |
297 | bool conflict = false; |
298 | bool reuseport_cb_ok; |
299 | |
300 | rcu_read_lock(); |
301 | reuseport_cb = rcu_dereference(sk->sk_reuseport_cb); |
302 | /* paired with WRITE_ONCE() in __reuseport_(add|detach)_closed_sock */ |
303 | reuseport_cb_ok = !reuseport_cb || READ_ONCE(reuseport_cb->num_closed_socks); |
304 | rcu_read_unlock(); |
305 | |
306 | head2 = inet_bhash2_addr_any_hashbucket(sk, net, port); |
307 | |
308 | spin_lock(lock: &head2->lock); |
309 | |
310 | inet_bind_bucket_for_each(tb2, &head2->chain) { |
311 | if (!inet_bind2_bucket_match_addr_any(tb: tb2, net, port, l3mdev, sk)) |
312 | continue; |
313 | |
314 | if (!inet_bhash2_conflict(sk, tb2, sk_uid: uid, relax, reuseport_cb_ok, reuseport_ok)) |
315 | continue; |
316 | |
317 | conflict = true; |
318 | break; |
319 | } |
320 | |
321 | spin_unlock(lock: &head2->lock); |
322 | |
323 | return conflict; |
324 | } |
325 | |
326 | /* |
327 | * Find an open port number for the socket. Returns with the |
328 | * inet_bind_hashbucket locks held if successful. |
329 | */ |
330 | static struct inet_bind_hashbucket * |
331 | inet_csk_find_open_port(const struct sock *sk, struct inet_bind_bucket **tb_ret, |
332 | struct inet_bind2_bucket **tb2_ret, |
333 | struct inet_bind_hashbucket **head2_ret, int *port_ret) |
334 | { |
335 | struct inet_hashinfo *hinfo = tcp_or_dccp_get_hashinfo(sk); |
336 | int i, low, high, attempt_half, port, l3mdev; |
337 | struct inet_bind_hashbucket *head, *head2; |
338 | struct net *net = sock_net(sk); |
339 | struct inet_bind2_bucket *tb2; |
340 | struct inet_bind_bucket *tb; |
341 | u32 remaining, offset; |
342 | bool relax = false; |
343 | |
344 | l3mdev = inet_sk_bound_l3mdev(sk); |
345 | ports_exhausted: |
346 | attempt_half = (sk->sk_reuse == SK_CAN_REUSE) ? 1 : 0; |
347 | other_half_scan: |
348 | inet_sk_get_local_port_range(sk, &low, &high); |
349 | high++; /* [32768, 60999] -> [32768, 61000[ */ |
350 | if (high - low < 4) |
351 | attempt_half = 0; |
352 | if (attempt_half) { |
353 | int half = low + (((high - low) >> 2) << 1); |
354 | |
355 | if (attempt_half == 1) |
356 | high = half; |
357 | else |
358 | low = half; |
359 | } |
360 | remaining = high - low; |
361 | if (likely(remaining > 1)) |
362 | remaining &= ~1U; |
363 | |
364 | offset = get_random_u32_below(ceil: remaining); |
365 | /* __inet_hash_connect() favors ports having @low parity |
366 | * We do the opposite to not pollute connect() users. |
367 | */ |
368 | offset |= 1U; |
369 | |
370 | other_parity_scan: |
371 | port = low + offset; |
372 | for (i = 0; i < remaining; i += 2, port += 2) { |
373 | if (unlikely(port >= high)) |
374 | port -= remaining; |
375 | if (inet_is_local_reserved_port(net, port)) |
376 | continue; |
377 | head = &hinfo->bhash[inet_bhashfn(net, lport: port, |
378 | bhash_size: hinfo->bhash_size)]; |
379 | spin_lock_bh(lock: &head->lock); |
380 | if (inet_use_bhash2_on_bind(sk)) { |
381 | if (inet_bhash2_addr_any_conflict(sk, port, l3mdev, relax, reuseport_ok: false)) |
382 | goto next_port; |
383 | } |
384 | |
385 | head2 = inet_bhashfn_portaddr(hinfo, sk, net, port); |
386 | spin_lock(lock: &head2->lock); |
387 | tb2 = inet_bind2_bucket_find(head: head2, net, port, l3mdev, sk); |
388 | inet_bind_bucket_for_each(tb, &head->chain) |
389 | if (inet_bind_bucket_match(tb, net, port, l3mdev)) { |
390 | if (!inet_csk_bind_conflict(sk, tb, tb2, |
391 | relax, reuseport_ok: false)) |
392 | goto success; |
393 | spin_unlock(lock: &head2->lock); |
394 | goto next_port; |
395 | } |
396 | tb = NULL; |
397 | goto success; |
398 | next_port: |
399 | spin_unlock_bh(lock: &head->lock); |
400 | cond_resched(); |
401 | } |
402 | |
403 | offset--; |
404 | if (!(offset & 1)) |
405 | goto other_parity_scan; |
406 | |
407 | if (attempt_half == 1) { |
408 | /* OK we now try the upper half of the range */ |
409 | attempt_half = 2; |
410 | goto other_half_scan; |
411 | } |
412 | |
413 | if (READ_ONCE(net->ipv4.sysctl_ip_autobind_reuse) && !relax) { |
414 | /* We still have a chance to connect to different destinations */ |
415 | relax = true; |
416 | goto ports_exhausted; |
417 | } |
418 | return NULL; |
419 | success: |
420 | *port_ret = port; |
421 | *tb_ret = tb; |
422 | *tb2_ret = tb2; |
423 | *head2_ret = head2; |
424 | return head; |
425 | } |
426 | |
427 | static inline int sk_reuseport_match(struct inet_bind_bucket *tb, |
428 | struct sock *sk) |
429 | { |
430 | kuid_t uid = sock_i_uid(sk); |
431 | |
432 | if (tb->fastreuseport <= 0) |
433 | return 0; |
434 | if (!sk->sk_reuseport) |
435 | return 0; |
436 | if (rcu_access_pointer(sk->sk_reuseport_cb)) |
437 | return 0; |
438 | if (!uid_eq(left: tb->fastuid, right: uid)) |
439 | return 0; |
440 | /* We only need to check the rcv_saddr if this tb was once marked |
441 | * without fastreuseport and then was reset, as we can only know that |
442 | * the fast_*rcv_saddr doesn't have any conflicts with the socks on the |
443 | * owners list. |
444 | */ |
445 | if (tb->fastreuseport == FASTREUSEPORT_ANY) |
446 | return 1; |
447 | #if IS_ENABLED(CONFIG_IPV6) |
448 | if (tb->fast_sk_family == AF_INET6) |
449 | return ipv6_rcv_saddr_equal(sk1_rcv_saddr6: &tb->fast_v6_rcv_saddr, |
450 | sk2_rcv_saddr6: inet6_rcv_saddr(sk), |
451 | sk1_rcv_saddr: tb->fast_rcv_saddr, |
452 | sk2_rcv_saddr: sk->sk_rcv_saddr, |
453 | sk1_ipv6only: tb->fast_ipv6_only, |
454 | ipv6_only_sock(sk), match_sk1_wildcard: true, match_sk2_wildcard: false); |
455 | #endif |
456 | return ipv4_rcv_saddr_equal(sk1_rcv_saddr: tb->fast_rcv_saddr, sk2_rcv_saddr: sk->sk_rcv_saddr, |
457 | ipv6_only_sock(sk), match_sk1_wildcard: true, match_sk2_wildcard: false); |
458 | } |
459 | |
460 | void inet_csk_update_fastreuse(struct inet_bind_bucket *tb, |
461 | struct sock *sk) |
462 | { |
463 | kuid_t uid = sock_i_uid(sk); |
464 | bool reuse = sk->sk_reuse && sk->sk_state != TCP_LISTEN; |
465 | |
466 | if (hlist_empty(h: &tb->bhash2)) { |
467 | tb->fastreuse = reuse; |
468 | if (sk->sk_reuseport) { |
469 | tb->fastreuseport = FASTREUSEPORT_ANY; |
470 | tb->fastuid = uid; |
471 | tb->fast_rcv_saddr = sk->sk_rcv_saddr; |
472 | tb->fast_ipv6_only = ipv6_only_sock(sk); |
473 | tb->fast_sk_family = sk->sk_family; |
474 | #if IS_ENABLED(CONFIG_IPV6) |
475 | tb->fast_v6_rcv_saddr = sk->sk_v6_rcv_saddr; |
476 | #endif |
477 | } else { |
478 | tb->fastreuseport = 0; |
479 | } |
480 | } else { |
481 | if (!reuse) |
482 | tb->fastreuse = 0; |
483 | if (sk->sk_reuseport) { |
484 | /* We didn't match or we don't have fastreuseport set on |
485 | * the tb, but we have sk_reuseport set on this socket |
486 | * and we know that there are no bind conflicts with |
487 | * this socket in this tb, so reset our tb's reuseport |
488 | * settings so that any subsequent sockets that match |
489 | * our current socket will be put on the fast path. |
490 | * |
491 | * If we reset we need to set FASTREUSEPORT_STRICT so we |
492 | * do extra checking for all subsequent sk_reuseport |
493 | * socks. |
494 | */ |
495 | if (!sk_reuseport_match(tb, sk)) { |
496 | tb->fastreuseport = FASTREUSEPORT_STRICT; |
497 | tb->fastuid = uid; |
498 | tb->fast_rcv_saddr = sk->sk_rcv_saddr; |
499 | tb->fast_ipv6_only = ipv6_only_sock(sk); |
500 | tb->fast_sk_family = sk->sk_family; |
501 | #if IS_ENABLED(CONFIG_IPV6) |
502 | tb->fast_v6_rcv_saddr = sk->sk_v6_rcv_saddr; |
503 | #endif |
504 | } |
505 | } else { |
506 | tb->fastreuseport = 0; |
507 | } |
508 | } |
509 | } |
510 | |
511 | /* Obtain a reference to a local port for the given sock, |
512 | * if snum is zero it means select any available local port. |
513 | * We try to allocate an odd port (and leave even ports for connect()) |
514 | */ |
515 | int inet_csk_get_port(struct sock *sk, unsigned short snum) |
516 | { |
517 | struct inet_hashinfo *hinfo = tcp_or_dccp_get_hashinfo(sk); |
518 | bool reuse = sk->sk_reuse && sk->sk_state != TCP_LISTEN; |
519 | bool found_port = false, check_bind_conflict = true; |
520 | bool bhash_created = false, bhash2_created = false; |
521 | int ret = -EADDRINUSE, port = snum, l3mdev; |
522 | struct inet_bind_hashbucket *head, *head2; |
523 | struct inet_bind2_bucket *tb2 = NULL; |
524 | struct inet_bind_bucket *tb = NULL; |
525 | bool head2_lock_acquired = false; |
526 | struct net *net = sock_net(sk); |
527 | |
528 | l3mdev = inet_sk_bound_l3mdev(sk); |
529 | |
530 | if (!port) { |
531 | head = inet_csk_find_open_port(sk, tb_ret: &tb, tb2_ret: &tb2, head2_ret: &head2, port_ret: &port); |
532 | if (!head) |
533 | return ret; |
534 | |
535 | head2_lock_acquired = true; |
536 | |
537 | if (tb && tb2) |
538 | goto success; |
539 | found_port = true; |
540 | } else { |
541 | head = &hinfo->bhash[inet_bhashfn(net, lport: port, |
542 | bhash_size: hinfo->bhash_size)]; |
543 | spin_lock_bh(lock: &head->lock); |
544 | inet_bind_bucket_for_each(tb, &head->chain) |
545 | if (inet_bind_bucket_match(tb, net, port, l3mdev)) |
546 | break; |
547 | } |
548 | |
549 | if (!tb) { |
550 | tb = inet_bind_bucket_create(cachep: hinfo->bind_bucket_cachep, net, |
551 | head, snum: port, l3mdev); |
552 | if (!tb) |
553 | goto fail_unlock; |
554 | bhash_created = true; |
555 | } |
556 | |
557 | if (!found_port) { |
558 | if (!hlist_empty(h: &tb->bhash2)) { |
559 | if (sk->sk_reuse == SK_FORCE_REUSE || |
560 | (tb->fastreuse > 0 && reuse) || |
561 | sk_reuseport_match(tb, sk)) |
562 | check_bind_conflict = false; |
563 | } |
564 | |
565 | if (check_bind_conflict && inet_use_bhash2_on_bind(sk)) { |
566 | if (inet_bhash2_addr_any_conflict(sk, port, l3mdev, relax: true, reuseport_ok: true)) |
567 | goto fail_unlock; |
568 | } |
569 | |
570 | head2 = inet_bhashfn_portaddr(hinfo, sk, net, port); |
571 | spin_lock(lock: &head2->lock); |
572 | head2_lock_acquired = true; |
573 | tb2 = inet_bind2_bucket_find(head: head2, net, port, l3mdev, sk); |
574 | } |
575 | |
576 | if (!tb2) { |
577 | tb2 = inet_bind2_bucket_create(cachep: hinfo->bind2_bucket_cachep, |
578 | net, head: head2, tb, sk); |
579 | if (!tb2) |
580 | goto fail_unlock; |
581 | bhash2_created = true; |
582 | } |
583 | |
584 | if (!found_port && check_bind_conflict) { |
585 | if (inet_csk_bind_conflict(sk, tb, tb2, relax: true, reuseport_ok: true)) |
586 | goto fail_unlock; |
587 | } |
588 | |
589 | success: |
590 | inet_csk_update_fastreuse(tb, sk); |
591 | |
592 | if (!inet_csk(sk)->icsk_bind_hash) |
593 | inet_bind_hash(sk, tb, tb2, port); |
594 | WARN_ON(inet_csk(sk)->icsk_bind_hash != tb); |
595 | WARN_ON(inet_csk(sk)->icsk_bind2_hash != tb2); |
596 | ret = 0; |
597 | |
598 | fail_unlock: |
599 | if (ret) { |
600 | if (bhash2_created) |
601 | inet_bind2_bucket_destroy(cachep: hinfo->bind2_bucket_cachep, tb: tb2); |
602 | if (bhash_created) |
603 | inet_bind_bucket_destroy(cachep: hinfo->bind_bucket_cachep, tb); |
604 | } |
605 | if (head2_lock_acquired) |
606 | spin_unlock(lock: &head2->lock); |
607 | spin_unlock_bh(lock: &head->lock); |
608 | return ret; |
609 | } |
610 | EXPORT_SYMBOL_GPL(inet_csk_get_port); |
611 | |
612 | /* |
613 | * Wait for an incoming connection, avoid race conditions. This must be called |
614 | * with the socket locked. |
615 | */ |
616 | static int inet_csk_wait_for_connect(struct sock *sk, long timeo) |
617 | { |
618 | struct inet_connection_sock *icsk = inet_csk(sk); |
619 | DEFINE_WAIT(wait); |
620 | int err; |
621 | |
622 | /* |
623 | * True wake-one mechanism for incoming connections: only |
624 | * one process gets woken up, not the 'whole herd'. |
625 | * Since we do not 'race & poll' for established sockets |
626 | * anymore, the common case will execute the loop only once. |
627 | * |
628 | * Subtle issue: "add_wait_queue_exclusive()" will be added |
629 | * after any current non-exclusive waiters, and we know that |
630 | * it will always _stay_ after any new non-exclusive waiters |
631 | * because all non-exclusive waiters are added at the |
632 | * beginning of the wait-queue. As such, it's ok to "drop" |
633 | * our exclusiveness temporarily when we get woken up without |
634 | * having to remove and re-insert us on the wait queue. |
635 | */ |
636 | for (;;) { |
637 | prepare_to_wait_exclusive(wq_head: sk_sleep(sk), wq_entry: &wait, |
638 | TASK_INTERRUPTIBLE); |
639 | release_sock(sk); |
640 | if (reqsk_queue_empty(queue: &icsk->icsk_accept_queue)) |
641 | timeo = schedule_timeout(timeout: timeo); |
642 | sched_annotate_sleep(); |
643 | lock_sock(sk); |
644 | err = 0; |
645 | if (!reqsk_queue_empty(queue: &icsk->icsk_accept_queue)) |
646 | break; |
647 | err = -EINVAL; |
648 | if (sk->sk_state != TCP_LISTEN) |
649 | break; |
650 | err = sock_intr_errno(timeo); |
651 | if (signal_pending(current)) |
652 | break; |
653 | err = -EAGAIN; |
654 | if (!timeo) |
655 | break; |
656 | } |
657 | finish_wait(wq_head: sk_sleep(sk), wq_entry: &wait); |
658 | return err; |
659 | } |
660 | |
661 | /* |
662 | * This will accept the next outstanding connection. |
663 | */ |
664 | struct sock *inet_csk_accept(struct sock *sk, int flags, int *err, bool kern) |
665 | { |
666 | struct inet_connection_sock *icsk = inet_csk(sk); |
667 | struct request_sock_queue *queue = &icsk->icsk_accept_queue; |
668 | struct request_sock *req; |
669 | struct sock *newsk; |
670 | int error; |
671 | |
672 | lock_sock(sk); |
673 | |
674 | /* We need to make sure that this socket is listening, |
675 | * and that it has something pending. |
676 | */ |
677 | error = -EINVAL; |
678 | if (sk->sk_state != TCP_LISTEN) |
679 | goto out_err; |
680 | |
681 | /* Find already established connection */ |
682 | if (reqsk_queue_empty(queue)) { |
683 | long timeo = sock_rcvtimeo(sk, noblock: flags & O_NONBLOCK); |
684 | |
685 | /* If this is a non blocking socket don't sleep */ |
686 | error = -EAGAIN; |
687 | if (!timeo) |
688 | goto out_err; |
689 | |
690 | error = inet_csk_wait_for_connect(sk, timeo); |
691 | if (error) |
692 | goto out_err; |
693 | } |
694 | req = reqsk_queue_remove(queue, parent: sk); |
695 | newsk = req->sk; |
696 | |
697 | if (sk->sk_protocol == IPPROTO_TCP && |
698 | tcp_rsk(req)->tfo_listener) { |
699 | spin_lock_bh(lock: &queue->fastopenq.lock); |
700 | if (tcp_rsk(req)->tfo_listener) { |
701 | /* We are still waiting for the final ACK from 3WHS |
702 | * so can't free req now. Instead, we set req->sk to |
703 | * NULL to signify that the child socket is taken |
704 | * so reqsk_fastopen_remove() will free the req |
705 | * when 3WHS finishes (or is aborted). |
706 | */ |
707 | req->sk = NULL; |
708 | req = NULL; |
709 | } |
710 | spin_unlock_bh(lock: &queue->fastopenq.lock); |
711 | } |
712 | |
713 | out: |
714 | release_sock(sk); |
715 | if (newsk && mem_cgroup_sockets_enabled) { |
716 | int amt = 0; |
717 | |
718 | /* atomically get the memory usage, set and charge the |
719 | * newsk->sk_memcg. |
720 | */ |
721 | lock_sock(sk: newsk); |
722 | |
723 | mem_cgroup_sk_alloc(sk: newsk); |
724 | if (newsk->sk_memcg) { |
725 | /* The socket has not been accepted yet, no need |
726 | * to look at newsk->sk_wmem_queued. |
727 | */ |
728 | amt = sk_mem_pages(amt: newsk->sk_forward_alloc + |
729 | atomic_read(v: &newsk->sk_rmem_alloc)); |
730 | } |
731 | |
732 | if (amt) |
733 | mem_cgroup_charge_skmem(memcg: newsk->sk_memcg, nr_pages: amt, |
734 | GFP_KERNEL | __GFP_NOFAIL); |
735 | |
736 | release_sock(sk: newsk); |
737 | } |
738 | if (req) |
739 | reqsk_put(req); |
740 | |
741 | if (newsk) |
742 | inet_init_csk_locks(sk: newsk); |
743 | |
744 | return newsk; |
745 | out_err: |
746 | newsk = NULL; |
747 | req = NULL; |
748 | *err = error; |
749 | goto out; |
750 | } |
751 | EXPORT_SYMBOL(inet_csk_accept); |
752 | |
753 | /* |
754 | * Using different timers for retransmit, delayed acks and probes |
755 | * We may wish use just one timer maintaining a list of expire jiffies |
756 | * to optimize. |
757 | */ |
758 | void inet_csk_init_xmit_timers(struct sock *sk, |
759 | void (*retransmit_handler)(struct timer_list *t), |
760 | void (*delack_handler)(struct timer_list *t), |
761 | void (*keepalive_handler)(struct timer_list *t)) |
762 | { |
763 | struct inet_connection_sock *icsk = inet_csk(sk); |
764 | |
765 | timer_setup(&icsk->icsk_retransmit_timer, retransmit_handler, 0); |
766 | timer_setup(&icsk->icsk_delack_timer, delack_handler, 0); |
767 | timer_setup(&sk->sk_timer, keepalive_handler, 0); |
768 | icsk->icsk_pending = icsk->icsk_ack.pending = 0; |
769 | } |
770 | EXPORT_SYMBOL(inet_csk_init_xmit_timers); |
771 | |
772 | void inet_csk_clear_xmit_timers(struct sock *sk) |
773 | { |
774 | struct inet_connection_sock *icsk = inet_csk(sk); |
775 | |
776 | icsk->icsk_pending = icsk->icsk_ack.pending = 0; |
777 | |
778 | sk_stop_timer(sk, timer: &icsk->icsk_retransmit_timer); |
779 | sk_stop_timer(sk, timer: &icsk->icsk_delack_timer); |
780 | sk_stop_timer(sk, timer: &sk->sk_timer); |
781 | } |
782 | EXPORT_SYMBOL(inet_csk_clear_xmit_timers); |
783 | |
784 | void inet_csk_clear_xmit_timers_sync(struct sock *sk) |
785 | { |
786 | struct inet_connection_sock *icsk = inet_csk(sk); |
787 | |
788 | /* ongoing timer handlers need to acquire socket lock. */ |
789 | sock_not_owned_by_me(sk); |
790 | |
791 | icsk->icsk_pending = icsk->icsk_ack.pending = 0; |
792 | |
793 | sk_stop_timer_sync(sk, timer: &icsk->icsk_retransmit_timer); |
794 | sk_stop_timer_sync(sk, timer: &icsk->icsk_delack_timer); |
795 | sk_stop_timer_sync(sk, timer: &sk->sk_timer); |
796 | } |
797 | |
798 | void inet_csk_delete_keepalive_timer(struct sock *sk) |
799 | { |
800 | sk_stop_timer(sk, timer: &sk->sk_timer); |
801 | } |
802 | EXPORT_SYMBOL(inet_csk_delete_keepalive_timer); |
803 | |
804 | void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len) |
805 | { |
806 | sk_reset_timer(sk, timer: &sk->sk_timer, expires: jiffies + len); |
807 | } |
808 | EXPORT_SYMBOL(inet_csk_reset_keepalive_timer); |
809 | |
810 | struct dst_entry *inet_csk_route_req(const struct sock *sk, |
811 | struct flowi4 *fl4, |
812 | const struct request_sock *req) |
813 | { |
814 | const struct inet_request_sock *ireq = inet_rsk(sk: req); |
815 | struct net *net = read_pnet(pnet: &ireq->ireq_net); |
816 | struct ip_options_rcu *opt; |
817 | struct rtable *rt; |
818 | |
819 | rcu_read_lock(); |
820 | opt = rcu_dereference(ireq->ireq_opt); |
821 | |
822 | flowi4_init_output(fl4, oif: ireq->ir_iif, mark: ireq->ir_mark, |
823 | tos: ip_sock_rt_tos(sk), scope: ip_sock_rt_scope(sk), |
824 | proto: sk->sk_protocol, flags: inet_sk_flowi_flags(sk), |
825 | daddr: (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr, |
826 | saddr: ireq->ir_loc_addr, dport: ireq->ir_rmt_port, |
827 | htons(ireq->ir_num), uid: sk->sk_uid); |
828 | security_req_classify_flow(req, flic: flowi4_to_flowi_common(fl4)); |
829 | rt = ip_route_output_flow(net, flp: fl4, sk); |
830 | if (IS_ERR(ptr: rt)) |
831 | goto no_route; |
832 | if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway) |
833 | goto route_err; |
834 | rcu_read_unlock(); |
835 | return &rt->dst; |
836 | |
837 | route_err: |
838 | ip_rt_put(rt); |
839 | no_route: |
840 | rcu_read_unlock(); |
841 | __IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); |
842 | return NULL; |
843 | } |
844 | EXPORT_SYMBOL_GPL(inet_csk_route_req); |
845 | |
846 | struct dst_entry *inet_csk_route_child_sock(const struct sock *sk, |
847 | struct sock *newsk, |
848 | const struct request_sock *req) |
849 | { |
850 | const struct inet_request_sock *ireq = inet_rsk(sk: req); |
851 | struct net *net = read_pnet(pnet: &ireq->ireq_net); |
852 | struct inet_sock *newinet = inet_sk(newsk); |
853 | struct ip_options_rcu *opt; |
854 | struct flowi4 *fl4; |
855 | struct rtable *rt; |
856 | |
857 | opt = rcu_dereference(ireq->ireq_opt); |
858 | fl4 = &newinet->cork.fl.u.ip4; |
859 | |
860 | flowi4_init_output(fl4, oif: ireq->ir_iif, mark: ireq->ir_mark, |
861 | tos: ip_sock_rt_tos(sk), scope: ip_sock_rt_scope(sk), |
862 | proto: sk->sk_protocol, flags: inet_sk_flowi_flags(sk), |
863 | daddr: (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr, |
864 | saddr: ireq->ir_loc_addr, dport: ireq->ir_rmt_port, |
865 | htons(ireq->ir_num), uid: sk->sk_uid); |
866 | security_req_classify_flow(req, flic: flowi4_to_flowi_common(fl4)); |
867 | rt = ip_route_output_flow(net, flp: fl4, sk); |
868 | if (IS_ERR(ptr: rt)) |
869 | goto no_route; |
870 | if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway) |
871 | goto route_err; |
872 | return &rt->dst; |
873 | |
874 | route_err: |
875 | ip_rt_put(rt); |
876 | no_route: |
877 | __IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); |
878 | return NULL; |
879 | } |
880 | EXPORT_SYMBOL_GPL(inet_csk_route_child_sock); |
881 | |
882 | /* Decide when to expire the request and when to resend SYN-ACK */ |
883 | static void syn_ack_recalc(struct request_sock *req, |
884 | const int max_syn_ack_retries, |
885 | const u8 rskq_defer_accept, |
886 | int *expire, int *resend) |
887 | { |
888 | if (!rskq_defer_accept) { |
889 | *expire = req->num_timeout >= max_syn_ack_retries; |
890 | *resend = 1; |
891 | return; |
892 | } |
893 | *expire = req->num_timeout >= max_syn_ack_retries && |
894 | (!inet_rsk(sk: req)->acked || req->num_timeout >= rskq_defer_accept); |
895 | /* Do not resend while waiting for data after ACK, |
896 | * start to resend on end of deferring period to give |
897 | * last chance for data or ACK to create established socket. |
898 | */ |
899 | *resend = !inet_rsk(sk: req)->acked || |
900 | req->num_timeout >= rskq_defer_accept - 1; |
901 | } |
902 | |
903 | int inet_rtx_syn_ack(const struct sock *parent, struct request_sock *req) |
904 | { |
905 | int err = req->rsk_ops->rtx_syn_ack(parent, req); |
906 | |
907 | if (!err) |
908 | req->num_retrans++; |
909 | return err; |
910 | } |
911 | EXPORT_SYMBOL(inet_rtx_syn_ack); |
912 | |
913 | static struct request_sock *inet_reqsk_clone(struct request_sock *req, |
914 | struct sock *sk) |
915 | { |
916 | struct sock *req_sk, *nreq_sk; |
917 | struct request_sock *nreq; |
918 | |
919 | nreq = kmem_cache_alloc(cachep: req->rsk_ops->slab, GFP_ATOMIC | __GFP_NOWARN); |
920 | if (!nreq) { |
921 | __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQFAILURE); |
922 | |
923 | /* paired with refcount_inc_not_zero() in reuseport_migrate_sock() */ |
924 | sock_put(sk); |
925 | return NULL; |
926 | } |
927 | |
928 | req_sk = req_to_sk(req); |
929 | nreq_sk = req_to_sk(req: nreq); |
930 | |
931 | memcpy(nreq_sk, req_sk, |
932 | offsetof(struct sock, sk_dontcopy_begin)); |
933 | unsafe_memcpy(&nreq_sk->sk_dontcopy_end, &req_sk->sk_dontcopy_end, |
934 | req->rsk_ops->obj_size - offsetof(struct sock, sk_dontcopy_end), |
935 | /* alloc is larger than struct, see above */); |
936 | |
937 | sk_node_init(node: &nreq_sk->sk_node); |
938 | nreq_sk->sk_tx_queue_mapping = req_sk->sk_tx_queue_mapping; |
939 | #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING |
940 | nreq_sk->sk_rx_queue_mapping = req_sk->sk_rx_queue_mapping; |
941 | #endif |
942 | nreq_sk->sk_incoming_cpu = req_sk->sk_incoming_cpu; |
943 | |
944 | nreq->rsk_listener = sk; |
945 | |
946 | /* We need not acquire fastopenq->lock |
947 | * because the child socket is locked in inet_csk_listen_stop(). |
948 | */ |
949 | if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(req: nreq)->tfo_listener) |
950 | rcu_assign_pointer(tcp_sk(nreq->sk)->fastopen_rsk, nreq); |
951 | |
952 | return nreq; |
953 | } |
954 | |
955 | static void reqsk_queue_migrated(struct request_sock_queue *queue, |
956 | const struct request_sock *req) |
957 | { |
958 | if (req->num_timeout == 0) |
959 | atomic_inc(v: &queue->young); |
960 | atomic_inc(v: &queue->qlen); |
961 | } |
962 | |
963 | static void reqsk_migrate_reset(struct request_sock *req) |
964 | { |
965 | req->saved_syn = NULL; |
966 | #if IS_ENABLED(CONFIG_IPV6) |
967 | inet_rsk(sk: req)->ipv6_opt = NULL; |
968 | inet_rsk(sk: req)->pktopts = NULL; |
969 | #else |
970 | inet_rsk(req)->ireq_opt = NULL; |
971 | #endif |
972 | } |
973 | |
974 | /* return true if req was found in the ehash table */ |
975 | static bool reqsk_queue_unlink(struct request_sock *req) |
976 | { |
977 | struct sock *sk = req_to_sk(req); |
978 | bool found = false; |
979 | |
980 | if (sk_hashed(sk)) { |
981 | struct inet_hashinfo *hashinfo = tcp_or_dccp_get_hashinfo(sk); |
982 | spinlock_t *lock = inet_ehash_lockp(hashinfo, hash: req->rsk_hash); |
983 | |
984 | spin_lock(lock); |
985 | found = __sk_nulls_del_node_init_rcu(sk); |
986 | spin_unlock(lock); |
987 | } |
988 | if (timer_pending(timer: &req->rsk_timer) && del_timer_sync(timer: &req->rsk_timer)) |
989 | reqsk_put(req); |
990 | return found; |
991 | } |
992 | |
993 | bool inet_csk_reqsk_queue_drop(struct sock *sk, struct request_sock *req) |
994 | { |
995 | bool unlinked = reqsk_queue_unlink(req); |
996 | |
997 | if (unlinked) { |
998 | reqsk_queue_removed(queue: &inet_csk(sk)->icsk_accept_queue, req); |
999 | reqsk_put(req); |
1000 | } |
1001 | return unlinked; |
1002 | } |
1003 | EXPORT_SYMBOL(inet_csk_reqsk_queue_drop); |
1004 | |
1005 | void inet_csk_reqsk_queue_drop_and_put(struct sock *sk, struct request_sock *req) |
1006 | { |
1007 | inet_csk_reqsk_queue_drop(sk, req); |
1008 | reqsk_put(req); |
1009 | } |
1010 | EXPORT_SYMBOL(inet_csk_reqsk_queue_drop_and_put); |
1011 | |
1012 | static void reqsk_timer_handler(struct timer_list *t) |
1013 | { |
1014 | struct request_sock *req = from_timer(req, t, rsk_timer); |
1015 | struct request_sock *nreq = NULL, *oreq = req; |
1016 | struct sock *sk_listener = req->rsk_listener; |
1017 | struct inet_connection_sock *icsk; |
1018 | struct request_sock_queue *queue; |
1019 | struct net *net; |
1020 | int max_syn_ack_retries, qlen, expire = 0, resend = 0; |
1021 | |
1022 | if (inet_sk_state_load(sk: sk_listener) != TCP_LISTEN) { |
1023 | struct sock *nsk; |
1024 | |
1025 | nsk = reuseport_migrate_sock(sk: sk_listener, migrating_sk: req_to_sk(req), NULL); |
1026 | if (!nsk) |
1027 | goto drop; |
1028 | |
1029 | nreq = inet_reqsk_clone(req, sk: nsk); |
1030 | if (!nreq) |
1031 | goto drop; |
1032 | |
1033 | /* The new timer for the cloned req can decrease the 2 |
1034 | * by calling inet_csk_reqsk_queue_drop_and_put(), so |
1035 | * hold another count to prevent use-after-free and |
1036 | * call reqsk_put() just before return. |
1037 | */ |
1038 | refcount_set(r: &nreq->rsk_refcnt, n: 2 + 1); |
1039 | timer_setup(&nreq->rsk_timer, reqsk_timer_handler, TIMER_PINNED); |
1040 | reqsk_queue_migrated(queue: &inet_csk(sk: nsk)->icsk_accept_queue, req); |
1041 | |
1042 | req = nreq; |
1043 | sk_listener = nsk; |
1044 | } |
1045 | |
1046 | icsk = inet_csk(sk: sk_listener); |
1047 | net = sock_net(sk: sk_listener); |
1048 | max_syn_ack_retries = READ_ONCE(icsk->icsk_syn_retries) ? : |
1049 | READ_ONCE(net->ipv4.sysctl_tcp_synack_retries); |
1050 | /* Normally all the openreqs are young and become mature |
1051 | * (i.e. converted to established socket) for first timeout. |
1052 | * If synack was not acknowledged for 1 second, it means |
1053 | * one of the following things: synack was lost, ack was lost, |
1054 | * rtt is high or nobody planned to ack (i.e. synflood). |
1055 | * When server is a bit loaded, queue is populated with old |
1056 | * open requests, reducing effective size of queue. |
1057 | * When server is well loaded, queue size reduces to zero |
1058 | * after several minutes of work. It is not synflood, |
1059 | * it is normal operation. The solution is pruning |
1060 | * too old entries overriding normal timeout, when |
1061 | * situation becomes dangerous. |
1062 | * |
1063 | * Essentially, we reserve half of room for young |
1064 | * embrions; and abort old ones without pity, if old |
1065 | * ones are about to clog our table. |
1066 | */ |
1067 | queue = &icsk->icsk_accept_queue; |
1068 | qlen = reqsk_queue_len(queue); |
1069 | if ((qlen << 1) > max(8U, READ_ONCE(sk_listener->sk_max_ack_backlog))) { |
1070 | int young = reqsk_queue_len_young(queue) << 1; |
1071 | |
1072 | while (max_syn_ack_retries > 2) { |
1073 | if (qlen < young) |
1074 | break; |
1075 | max_syn_ack_retries--; |
1076 | young <<= 1; |
1077 | } |
1078 | } |
1079 | syn_ack_recalc(req, max_syn_ack_retries, READ_ONCE(queue->rskq_defer_accept), |
1080 | expire: &expire, resend: &resend); |
1081 | req->rsk_ops->syn_ack_timeout(req); |
1082 | if (!expire && |
1083 | (!resend || |
1084 | !inet_rtx_syn_ack(sk_listener, req) || |
1085 | inet_rsk(sk: req)->acked)) { |
1086 | if (req->num_timeout++ == 0) |
1087 | atomic_dec(v: &queue->young); |
1088 | mod_timer(timer: &req->rsk_timer, expires: jiffies + reqsk_timeout(req, TCP_RTO_MAX)); |
1089 | |
1090 | if (!nreq) |
1091 | return; |
1092 | |
1093 | if (!inet_ehash_insert(sk: req_to_sk(req: nreq), osk: req_to_sk(req: oreq), NULL)) { |
1094 | /* delete timer */ |
1095 | inet_csk_reqsk_queue_drop(sk_listener, nreq); |
1096 | goto no_ownership; |
1097 | } |
1098 | |
1099 | __NET_INC_STATS(net, LINUX_MIB_TCPMIGRATEREQSUCCESS); |
1100 | reqsk_migrate_reset(req: oreq); |
1101 | reqsk_queue_removed(queue: &inet_csk(sk: oreq->rsk_listener)->icsk_accept_queue, req: oreq); |
1102 | reqsk_put(req: oreq); |
1103 | |
1104 | reqsk_put(req: nreq); |
1105 | return; |
1106 | } |
1107 | |
1108 | /* Even if we can clone the req, we may need not retransmit any more |
1109 | * SYN+ACKs (nreq->num_timeout > max_syn_ack_retries, etc), or another |
1110 | * CPU may win the "own_req" race so that inet_ehash_insert() fails. |
1111 | */ |
1112 | if (nreq) { |
1113 | __NET_INC_STATS(net, LINUX_MIB_TCPMIGRATEREQFAILURE); |
1114 | no_ownership: |
1115 | reqsk_migrate_reset(req: nreq); |
1116 | reqsk_queue_removed(queue, req: nreq); |
1117 | __reqsk_free(req: nreq); |
1118 | } |
1119 | |
1120 | drop: |
1121 | inet_csk_reqsk_queue_drop_and_put(oreq->rsk_listener, oreq); |
1122 | } |
1123 | |
1124 | static void reqsk_queue_hash_req(struct request_sock *req, |
1125 | unsigned long timeout) |
1126 | { |
1127 | timer_setup(&req->rsk_timer, reqsk_timer_handler, TIMER_PINNED); |
1128 | mod_timer(timer: &req->rsk_timer, expires: jiffies + timeout); |
1129 | |
1130 | inet_ehash_insert(sk: req_to_sk(req), NULL, NULL); |
1131 | /* before letting lookups find us, make sure all req fields |
1132 | * are committed to memory and refcnt initialized. |
1133 | */ |
1134 | smp_wmb(); |
1135 | refcount_set(r: &req->rsk_refcnt, n: 2 + 1); |
1136 | } |
1137 | |
1138 | void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req, |
1139 | unsigned long timeout) |
1140 | { |
1141 | reqsk_queue_hash_req(req, timeout); |
1142 | inet_csk_reqsk_queue_added(sk); |
1143 | } |
1144 | EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add); |
1145 | |
1146 | static void inet_clone_ulp(const struct request_sock *req, struct sock *newsk, |
1147 | const gfp_t priority) |
1148 | { |
1149 | struct inet_connection_sock *icsk = inet_csk(sk: newsk); |
1150 | |
1151 | if (!icsk->icsk_ulp_ops) |
1152 | return; |
1153 | |
1154 | icsk->icsk_ulp_ops->clone(req, newsk, priority); |
1155 | } |
1156 | |
1157 | /** |
1158 | * inet_csk_clone_lock - clone an inet socket, and lock its clone |
1159 | * @sk: the socket to clone |
1160 | * @req: request_sock |
1161 | * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) |
1162 | * |
1163 | * Caller must unlock socket even in error path (bh_unlock_sock(newsk)) |
1164 | */ |
1165 | struct sock *inet_csk_clone_lock(const struct sock *sk, |
1166 | const struct request_sock *req, |
1167 | const gfp_t priority) |
1168 | { |
1169 | struct sock *newsk = sk_clone_lock(sk, priority); |
1170 | |
1171 | if (newsk) { |
1172 | struct inet_connection_sock *newicsk = inet_csk(sk: newsk); |
1173 | |
1174 | inet_sk_set_state(sk: newsk, state: TCP_SYN_RECV); |
1175 | newicsk->icsk_bind_hash = NULL; |
1176 | newicsk->icsk_bind2_hash = NULL; |
1177 | |
1178 | inet_sk(newsk)->inet_dport = inet_rsk(sk: req)->ir_rmt_port; |
1179 | inet_sk(newsk)->inet_num = inet_rsk(sk: req)->ir_num; |
1180 | inet_sk(newsk)->inet_sport = htons(inet_rsk(req)->ir_num); |
1181 | |
1182 | /* listeners have SOCK_RCU_FREE, not the children */ |
1183 | sock_reset_flag(sk: newsk, flag: SOCK_RCU_FREE); |
1184 | |
1185 | inet_sk(newsk)->mc_list = NULL; |
1186 | |
1187 | newsk->sk_mark = inet_rsk(sk: req)->ir_mark; |
1188 | atomic64_set(v: &newsk->sk_cookie, |
1189 | i: atomic64_read(v: &inet_rsk(sk: req)->ir_cookie)); |
1190 | |
1191 | newicsk->icsk_retransmits = 0; |
1192 | newicsk->icsk_backoff = 0; |
1193 | newicsk->icsk_probes_out = 0; |
1194 | newicsk->icsk_probes_tstamp = 0; |
1195 | |
1196 | /* Deinitialize accept_queue to trap illegal accesses. */ |
1197 | memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue)); |
1198 | |
1199 | inet_clone_ulp(req, newsk, priority); |
1200 | |
1201 | security_inet_csk_clone(newsk, req); |
1202 | } |
1203 | return newsk; |
1204 | } |
1205 | EXPORT_SYMBOL_GPL(inet_csk_clone_lock); |
1206 | |
1207 | /* |
1208 | * At this point, there should be no process reference to this |
1209 | * socket, and thus no user references at all. Therefore we |
1210 | * can assume the socket waitqueue is inactive and nobody will |
1211 | * try to jump onto it. |
1212 | */ |
1213 | void inet_csk_destroy_sock(struct sock *sk) |
1214 | { |
1215 | WARN_ON(sk->sk_state != TCP_CLOSE); |
1216 | WARN_ON(!sock_flag(sk, SOCK_DEAD)); |
1217 | |
1218 | /* It cannot be in hash table! */ |
1219 | WARN_ON(!sk_unhashed(sk)); |
1220 | |
1221 | /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */ |
1222 | WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash); |
1223 | |
1224 | sk->sk_prot->destroy(sk); |
1225 | |
1226 | sk_stream_kill_queues(sk); |
1227 | |
1228 | xfrm_sk_free_policy(sk); |
1229 | |
1230 | this_cpu_dec(*sk->sk_prot->orphan_count); |
1231 | |
1232 | sock_put(sk); |
1233 | } |
1234 | EXPORT_SYMBOL(inet_csk_destroy_sock); |
1235 | |
1236 | /* This function allows to force a closure of a socket after the call to |
1237 | * tcp/dccp_create_openreq_child(). |
1238 | */ |
1239 | void inet_csk_prepare_forced_close(struct sock *sk) |
1240 | __releases(&sk->sk_lock.slock) |
1241 | { |
1242 | /* sk_clone_lock locked the socket and set refcnt to 2 */ |
1243 | bh_unlock_sock(sk); |
1244 | sock_put(sk); |
1245 | inet_csk_prepare_for_destroy_sock(sk); |
1246 | inet_sk(sk)->inet_num = 0; |
1247 | } |
1248 | EXPORT_SYMBOL(inet_csk_prepare_forced_close); |
1249 | |
1250 | static int inet_ulp_can_listen(const struct sock *sk) |
1251 | { |
1252 | const struct inet_connection_sock *icsk = inet_csk(sk); |
1253 | |
1254 | if (icsk->icsk_ulp_ops && !icsk->icsk_ulp_ops->clone) |
1255 | return -EINVAL; |
1256 | |
1257 | return 0; |
1258 | } |
1259 | |
1260 | int inet_csk_listen_start(struct sock *sk) |
1261 | { |
1262 | struct inet_connection_sock *icsk = inet_csk(sk); |
1263 | struct inet_sock *inet = inet_sk(sk); |
1264 | int err; |
1265 | |
1266 | err = inet_ulp_can_listen(sk); |
1267 | if (unlikely(err)) |
1268 | return err; |
1269 | |
1270 | reqsk_queue_alloc(queue: &icsk->icsk_accept_queue); |
1271 | |
1272 | sk->sk_ack_backlog = 0; |
1273 | inet_csk_delack_init(sk); |
1274 | |
1275 | /* There is race window here: we announce ourselves listening, |
1276 | * but this transition is still not validated by get_port(). |
1277 | * It is OK, because this socket enters to hash table only |
1278 | * after validation is complete. |
1279 | */ |
1280 | inet_sk_state_store(sk, newstate: TCP_LISTEN); |
1281 | err = sk->sk_prot->get_port(sk, inet->inet_num); |
1282 | if (!err) { |
1283 | inet->inet_sport = htons(inet->inet_num); |
1284 | |
1285 | sk_dst_reset(sk); |
1286 | err = sk->sk_prot->hash(sk); |
1287 | |
1288 | if (likely(!err)) |
1289 | return 0; |
1290 | } |
1291 | |
1292 | inet_sk_set_state(sk, state: TCP_CLOSE); |
1293 | return err; |
1294 | } |
1295 | EXPORT_SYMBOL_GPL(inet_csk_listen_start); |
1296 | |
1297 | static void inet_child_forget(struct sock *sk, struct request_sock *req, |
1298 | struct sock *child) |
1299 | { |
1300 | sk->sk_prot->disconnect(child, O_NONBLOCK); |
1301 | |
1302 | sock_orphan(sk: child); |
1303 | |
1304 | this_cpu_inc(*sk->sk_prot->orphan_count); |
1305 | |
1306 | if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(req)->tfo_listener) { |
1307 | BUG_ON(rcu_access_pointer(tcp_sk(child)->fastopen_rsk) != req); |
1308 | BUG_ON(sk != req->rsk_listener); |
1309 | |
1310 | /* Paranoid, to prevent race condition if |
1311 | * an inbound pkt destined for child is |
1312 | * blocked by sock lock in tcp_v4_rcv(). |
1313 | * Also to satisfy an assertion in |
1314 | * tcp_v4_destroy_sock(). |
1315 | */ |
1316 | RCU_INIT_POINTER(tcp_sk(child)->fastopen_rsk, NULL); |
1317 | } |
1318 | inet_csk_destroy_sock(child); |
1319 | } |
1320 | |
1321 | struct sock *inet_csk_reqsk_queue_add(struct sock *sk, |
1322 | struct request_sock *req, |
1323 | struct sock *child) |
1324 | { |
1325 | struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue; |
1326 | |
1327 | spin_lock(lock: &queue->rskq_lock); |
1328 | if (unlikely(sk->sk_state != TCP_LISTEN)) { |
1329 | inet_child_forget(sk, req, child); |
1330 | child = NULL; |
1331 | } else { |
1332 | req->sk = child; |
1333 | req->dl_next = NULL; |
1334 | if (queue->rskq_accept_head == NULL) |
1335 | WRITE_ONCE(queue->rskq_accept_head, req); |
1336 | else |
1337 | queue->rskq_accept_tail->dl_next = req; |
1338 | queue->rskq_accept_tail = req; |
1339 | sk_acceptq_added(sk); |
1340 | } |
1341 | spin_unlock(lock: &queue->rskq_lock); |
1342 | return child; |
1343 | } |
1344 | EXPORT_SYMBOL(inet_csk_reqsk_queue_add); |
1345 | |
1346 | struct sock *inet_csk_complete_hashdance(struct sock *sk, struct sock *child, |
1347 | struct request_sock *req, bool own_req) |
1348 | { |
1349 | if (own_req) { |
1350 | inet_csk_reqsk_queue_drop(req->rsk_listener, req); |
1351 | reqsk_queue_removed(queue: &inet_csk(sk: req->rsk_listener)->icsk_accept_queue, req); |
1352 | |
1353 | if (sk != req->rsk_listener) { |
1354 | /* another listening sk has been selected, |
1355 | * migrate the req to it. |
1356 | */ |
1357 | struct request_sock *nreq; |
1358 | |
1359 | /* hold a refcnt for the nreq->rsk_listener |
1360 | * which is assigned in inet_reqsk_clone() |
1361 | */ |
1362 | sock_hold(sk); |
1363 | nreq = inet_reqsk_clone(req, sk); |
1364 | if (!nreq) { |
1365 | inet_child_forget(sk, req, child); |
1366 | goto child_put; |
1367 | } |
1368 | |
1369 | refcount_set(r: &nreq->rsk_refcnt, n: 1); |
1370 | if (inet_csk_reqsk_queue_add(sk, nreq, child)) { |
1371 | __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQSUCCESS); |
1372 | reqsk_migrate_reset(req); |
1373 | reqsk_put(req); |
1374 | return child; |
1375 | } |
1376 | |
1377 | __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQFAILURE); |
1378 | reqsk_migrate_reset(req: nreq); |
1379 | __reqsk_free(req: nreq); |
1380 | } else if (inet_csk_reqsk_queue_add(sk, req, child)) { |
1381 | return child; |
1382 | } |
1383 | } |
1384 | /* Too bad, another child took ownership of the request, undo. */ |
1385 | child_put: |
1386 | bh_unlock_sock(child); |
1387 | sock_put(sk: child); |
1388 | return NULL; |
1389 | } |
1390 | EXPORT_SYMBOL(inet_csk_complete_hashdance); |
1391 | |
1392 | /* |
1393 | * This routine closes sockets which have been at least partially |
1394 | * opened, but not yet accepted. |
1395 | */ |
1396 | void inet_csk_listen_stop(struct sock *sk) |
1397 | { |
1398 | struct inet_connection_sock *icsk = inet_csk(sk); |
1399 | struct request_sock_queue *queue = &icsk->icsk_accept_queue; |
1400 | struct request_sock *next, *req; |
1401 | |
1402 | /* Following specs, it would be better either to send FIN |
1403 | * (and enter FIN-WAIT-1, it is normal close) |
1404 | * or to send active reset (abort). |
1405 | * Certainly, it is pretty dangerous while synflood, but it is |
1406 | * bad justification for our negligence 8) |
1407 | * To be honest, we are not able to make either |
1408 | * of the variants now. --ANK |
1409 | */ |
1410 | while ((req = reqsk_queue_remove(queue, parent: sk)) != NULL) { |
1411 | struct sock *child = req->sk, *nsk; |
1412 | struct request_sock *nreq; |
1413 | |
1414 | local_bh_disable(); |
1415 | bh_lock_sock(child); |
1416 | WARN_ON(sock_owned_by_user(child)); |
1417 | sock_hold(sk: child); |
1418 | |
1419 | nsk = reuseport_migrate_sock(sk, migrating_sk: child, NULL); |
1420 | if (nsk) { |
1421 | nreq = inet_reqsk_clone(req, sk: nsk); |
1422 | if (nreq) { |
1423 | refcount_set(r: &nreq->rsk_refcnt, n: 1); |
1424 | |
1425 | if (inet_csk_reqsk_queue_add(nsk, nreq, child)) { |
1426 | __NET_INC_STATS(sock_net(nsk), |
1427 | LINUX_MIB_TCPMIGRATEREQSUCCESS); |
1428 | reqsk_migrate_reset(req); |
1429 | } else { |
1430 | __NET_INC_STATS(sock_net(nsk), |
1431 | LINUX_MIB_TCPMIGRATEREQFAILURE); |
1432 | reqsk_migrate_reset(req: nreq); |
1433 | __reqsk_free(req: nreq); |
1434 | } |
1435 | |
1436 | /* inet_csk_reqsk_queue_add() has already |
1437 | * called inet_child_forget() on failure case. |
1438 | */ |
1439 | goto skip_child_forget; |
1440 | } |
1441 | } |
1442 | |
1443 | inet_child_forget(sk, req, child); |
1444 | skip_child_forget: |
1445 | reqsk_put(req); |
1446 | bh_unlock_sock(child); |
1447 | local_bh_enable(); |
1448 | sock_put(sk: child); |
1449 | |
1450 | cond_resched(); |
1451 | } |
1452 | if (queue->fastopenq.rskq_rst_head) { |
1453 | /* Free all the reqs queued in rskq_rst_head. */ |
1454 | spin_lock_bh(lock: &queue->fastopenq.lock); |
1455 | req = queue->fastopenq.rskq_rst_head; |
1456 | queue->fastopenq.rskq_rst_head = NULL; |
1457 | spin_unlock_bh(lock: &queue->fastopenq.lock); |
1458 | while (req != NULL) { |
1459 | next = req->dl_next; |
1460 | reqsk_put(req); |
1461 | req = next; |
1462 | } |
1463 | } |
1464 | WARN_ON_ONCE(sk->sk_ack_backlog); |
1465 | } |
1466 | EXPORT_SYMBOL_GPL(inet_csk_listen_stop); |
1467 | |
1468 | void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr) |
1469 | { |
1470 | struct sockaddr_in *sin = (struct sockaddr_in *)uaddr; |
1471 | const struct inet_sock *inet = inet_sk(sk); |
1472 | |
1473 | sin->sin_family = AF_INET; |
1474 | sin->sin_addr.s_addr = inet->inet_daddr; |
1475 | sin->sin_port = inet->inet_dport; |
1476 | } |
1477 | EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr); |
1478 | |
1479 | static struct dst_entry *inet_csk_rebuild_route(struct sock *sk, struct flowi *fl) |
1480 | { |
1481 | const struct inet_sock *inet = inet_sk(sk); |
1482 | const struct ip_options_rcu *inet_opt; |
1483 | __be32 daddr = inet->inet_daddr; |
1484 | struct flowi4 *fl4; |
1485 | struct rtable *rt; |
1486 | |
1487 | rcu_read_lock(); |
1488 | inet_opt = rcu_dereference(inet->inet_opt); |
1489 | if (inet_opt && inet_opt->opt.srr) |
1490 | daddr = inet_opt->opt.faddr; |
1491 | fl4 = &fl->u.ip4; |
1492 | rt = ip_route_output_ports(net: sock_net(sk), fl4, sk, daddr, |
1493 | saddr: inet->inet_saddr, dport: inet->inet_dport, |
1494 | sport: inet->inet_sport, proto: sk->sk_protocol, |
1495 | tos: ip_sock_rt_tos(sk), oif: sk->sk_bound_dev_if); |
1496 | if (IS_ERR(ptr: rt)) |
1497 | rt = NULL; |
1498 | if (rt) |
1499 | sk_setup_caps(sk, dst: &rt->dst); |
1500 | rcu_read_unlock(); |
1501 | |
1502 | return &rt->dst; |
1503 | } |
1504 | |
1505 | struct dst_entry *inet_csk_update_pmtu(struct sock *sk, u32 mtu) |
1506 | { |
1507 | struct dst_entry *dst = __sk_dst_check(sk, cookie: 0); |
1508 | struct inet_sock *inet = inet_sk(sk); |
1509 | |
1510 | if (!dst) { |
1511 | dst = inet_csk_rebuild_route(sk, fl: &inet->cork.fl); |
1512 | if (!dst) |
1513 | goto out; |
1514 | } |
1515 | dst->ops->update_pmtu(dst, sk, NULL, mtu, true); |
1516 | |
1517 | dst = __sk_dst_check(sk, cookie: 0); |
1518 | if (!dst) |
1519 | dst = inet_csk_rebuild_route(sk, fl: &inet->cork.fl); |
1520 | out: |
1521 | return dst; |
1522 | } |
1523 | EXPORT_SYMBOL_GPL(inet_csk_update_pmtu); |
1524 | |