]>
Commit | Line | Data |
---|---|---|
2874c5fd | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
3f421baa ACM |
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 | |
3f421baa ACM |
10 | */ |
11 | ||
3f421baa ACM |
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> | |
a019d6fe | 21 | #include <net/xfrm.h> |
fa76ce73 | 22 | #include <net/tcp.h> |
c125e80b | 23 | #include <net/sock_reuseport.h> |
9691724e | 24 | #include <net/addrconf.h> |
3f421baa | 25 | |
fe38d2a1 | 26 | #if IS_ENABLED(CONFIG_IPV6) |
88d7fcfa MKL |
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 | |
fe38d2a1 | 33 | */ |
7016e062 JP |
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, | |
88d7fcfa MKL |
38 | bool match_sk1_wildcard, |
39 | bool match_sk2_wildcard) | |
fe38d2a1 | 40 | { |
637bc8bb | 41 | int addr_type = ipv6_addr_type(sk1_rcv_saddr6); |
fe38d2a1 JB |
42 | int addr_type2 = sk2_rcv_saddr6 ? ipv6_addr_type(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) { | |
637bc8bb | 47 | if (sk1_rcv_saddr == sk2_rcv_saddr) |
7016e062 | 48 | return true; |
88d7fcfa MKL |
49 | return (match_sk1_wildcard && !sk1_rcv_saddr) || |
50 | (match_sk2_wildcard && !sk2_rcv_saddr); | |
fe38d2a1 | 51 | } |
7016e062 | 52 | return false; |
fe38d2a1 JB |
53 | } |
54 | ||
55 | if (addr_type == IPV6_ADDR_ANY && addr_type2 == IPV6_ADDR_ANY) | |
7016e062 | 56 | return true; |
fe38d2a1 | 57 | |
88d7fcfa | 58 | if (addr_type2 == IPV6_ADDR_ANY && match_sk2_wildcard && |
fe38d2a1 | 59 | !(sk2_ipv6only && addr_type == IPV6_ADDR_MAPPED)) |
7016e062 | 60 | return true; |
fe38d2a1 | 61 | |
88d7fcfa | 62 | if (addr_type == IPV6_ADDR_ANY && match_sk1_wildcard && |
637bc8bb | 63 | !(sk1_ipv6only && addr_type2 == IPV6_ADDR_MAPPED)) |
7016e062 | 64 | return true; |
fe38d2a1 JB |
65 | |
66 | if (sk2_rcv_saddr6 && | |
637bc8bb | 67 | ipv6_addr_equal(sk1_rcv_saddr6, sk2_rcv_saddr6)) |
7016e062 | 68 | return true; |
fe38d2a1 | 69 | |
7016e062 | 70 | return false; |
fe38d2a1 JB |
71 | } |
72 | #endif | |
73 | ||
88d7fcfa MKL |
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 | |
fe38d2a1 | 77 | */ |
7016e062 | 78 | static bool ipv4_rcv_saddr_equal(__be32 sk1_rcv_saddr, __be32 sk2_rcv_saddr, |
88d7fcfa MKL |
79 | bool sk2_ipv6only, bool match_sk1_wildcard, |
80 | bool match_sk2_wildcard) | |
fe38d2a1 | 81 | { |
637bc8bb JB |
82 | if (!sk2_ipv6only) { |
83 | if (sk1_rcv_saddr == sk2_rcv_saddr) | |
7016e062 | 84 | return true; |
88d7fcfa MKL |
85 | return (match_sk1_wildcard && !sk1_rcv_saddr) || |
86 | (match_sk2_wildcard && !sk2_rcv_saddr); | |
fe38d2a1 | 87 | } |
7016e062 | 88 | return false; |
fe38d2a1 JB |
89 | } |
90 | ||
7016e062 JP |
91 | bool inet_rcv_saddr_equal(const struct sock *sk, const struct sock *sk2, |
92 | bool match_wildcard) | |
fe38d2a1 JB |
93 | { |
94 | #if IS_ENABLED(CONFIG_IPV6) | |
95 | if (sk->sk_family == AF_INET6) | |
637bc8bb | 96 | return ipv6_rcv_saddr_equal(&sk->sk_v6_rcv_saddr, |
319554f2 | 97 | inet6_rcv_saddr(sk2), |
637bc8bb JB |
98 | sk->sk_rcv_saddr, |
99 | sk2->sk_rcv_saddr, | |
100 | ipv6_only_sock(sk), | |
101 | ipv6_only_sock(sk2), | |
88d7fcfa | 102 | match_wildcard, |
637bc8bb | 103 | match_wildcard); |
fe38d2a1 | 104 | #endif |
637bc8bb | 105 | return ipv4_rcv_saddr_equal(sk->sk_rcv_saddr, sk2->sk_rcv_saddr, |
88d7fcfa MKL |
106 | ipv6_only_sock(sk2), match_wildcard, |
107 | match_wildcard); | |
fe38d2a1 JB |
108 | } |
109 | EXPORT_SYMBOL(inet_rcv_saddr_equal); | |
110 | ||
2dbb9b9e MKL |
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(&sk->sk_v6_rcv_saddr); | |
116 | #endif | |
117 | return !sk->sk_rcv_saddr; | |
118 | } | |
119 | ||
0bbf87d8 | 120 | void inet_get_local_port_range(struct net *net, int *low, int *high) |
227b60f5 | 121 | { |
95c96174 ED |
122 | unsigned int seq; |
123 | ||
227b60f5 | 124 | do { |
c9d8f1a6 | 125 | seq = read_seqbegin(&net->ipv4.ip_local_ports.lock); |
227b60f5 | 126 | |
c9d8f1a6 CW |
127 | *low = net->ipv4.ip_local_ports.range[0]; |
128 | *high = net->ipv4.ip_local_ports.range[1]; | |
129 | } while (read_seqretry(&net->ipv4.ip_local_ports.lock, seq)); | |
227b60f5 SH |
130 | } |
131 | EXPORT_SYMBOL(inet_get_local_port_range); | |
3f421baa | 132 | |
593d1ebe JK |
133 | static int inet_csk_bind_conflict(const struct sock *sk, |
134 | const struct inet_bind_bucket *tb, | |
d5a42de8 JK |
135 | bool relax, bool reuseport_ok) |
136 | { | |
d5a42de8 | 137 | struct sock *sk2; |
593d1ebe JK |
138 | bool reuseport_cb_ok; |
139 | bool reuse = sk->sk_reuse; | |
140 | bool reuseport = !!sk->sk_reuseport; | |
141 | struct sock_reuseport *reuseport_cb; | |
142 | kuid_t uid = sock_i_uid((struct sock *)sk); | |
3f421baa | 143 | |
333bb73f KI |
144 | rcu_read_lock(); |
145 | reuseport_cb = rcu_dereference(sk->sk_reuseport_cb); | |
146 | /* paired with WRITE_ONCE() in __reuseport_(add|detach)_closed_sock */ | |
147 | reuseport_cb_ok = !reuseport_cb || READ_ONCE(reuseport_cb->num_closed_socks); | |
148 | rcu_read_unlock(); | |
149 | ||
7477fd2e PE |
150 | /* |
151 | * Unlike other sk lookup places we do not check | |
152 | * for sk_net here, since _all_ the socks listed | |
593d1ebe JK |
153 | * in tb->owners list belong to the same net - the |
154 | * one this bucket belongs to. | |
7477fd2e PE |
155 | */ |
156 | ||
593d1ebe JK |
157 | sk_for_each_bound(sk2, &tb->owners) { |
158 | int bound_dev_if2; | |
d2c13561 | 159 | |
593d1ebe JK |
160 | if (sk == sk2) |
161 | continue; | |
162 | bound_dev_if2 = READ_ONCE(sk2->sk_bound_dev_if); | |
163 | if ((!sk->sk_bound_dev_if || | |
164 | !bound_dev_if2 || | |
165 | sk->sk_bound_dev_if == bound_dev_if2)) { | |
166 | if (reuse && sk2->sk_reuse && | |
167 | sk2->sk_state != TCP_LISTEN) { | |
168 | if ((!relax || | |
169 | (!reuseport_ok && | |
170 | reuseport && sk2->sk_reuseport && | |
171 | reuseport_cb_ok && | |
172 | (sk2->sk_state == TCP_TIME_WAIT || | |
173 | uid_eq(uid, sock_i_uid(sk2))))) && | |
174 | inet_rcv_saddr_equal(sk, sk2, true)) | |
175 | break; | |
176 | } else if (!reuseport_ok || | |
177 | !reuseport || !sk2->sk_reuseport || | |
178 | !reuseport_cb_ok || | |
179 | (sk2->sk_state != TCP_TIME_WAIT && | |
180 | !uid_eq(uid, sock_i_uid(sk2)))) { | |
181 | if (inet_rcv_saddr_equal(sk, sk2, true)) | |
182 | break; | |
183 | } | |
184 | } | |
3f421baa | 185 | } |
593d1ebe | 186 | return sk2 != NULL; |
3f421baa | 187 | } |
971af18b | 188 | |
289141b7 JB |
189 | /* |
190 | * Find an open port number for the socket. Returns with the | |
191 | * inet_bind_hashbucket lock held. | |
3f421baa | 192 | */ |
289141b7 | 193 | static struct inet_bind_hashbucket * |
593d1ebe | 194 | inet_csk_find_open_port(struct sock *sk, struct inet_bind_bucket **tb_ret, int *port_ret) |
3f421baa | 195 | { |
ea8add2b | 196 | struct inet_hashinfo *hinfo = sk->sk_prot->h.hashinfo; |
593d1ebe | 197 | int port = 0; |
3f421baa | 198 | struct inet_bind_hashbucket *head; |
3b1e0a65 | 199 | struct net *net = sock_net(sk); |
593d1ebe | 200 | bool relax = false; |
ea8add2b ED |
201 | int i, low, high, attempt_half; |
202 | struct inet_bind_bucket *tb; | |
ea8add2b | 203 | u32 remaining, offset; |
3c82a21f | 204 | int l3mdev; |
3f421baa | 205 | |
3c82a21f | 206 | l3mdev = inet_sk_bound_l3mdev(sk); |
4b01a967 | 207 | ports_exhausted: |
ea8add2b ED |
208 | attempt_half = (sk->sk_reuse == SK_CAN_REUSE) ? 1 : 0; |
209 | other_half_scan: | |
210 | inet_get_local_port_range(net, &low, &high); | |
211 | high++; /* [32768, 60999] -> [32768, 61000[ */ | |
212 | if (high - low < 4) | |
213 | attempt_half = 0; | |
214 | if (attempt_half) { | |
215 | int half = low + (((high - low) >> 2) << 1); | |
216 | ||
217 | if (attempt_half == 1) | |
218 | high = half; | |
219 | else | |
220 | low = half; | |
221 | } | |
222 | remaining = high - low; | |
223 | if (likely(remaining > 1)) | |
224 | remaining &= ~1U; | |
3f421baa | 225 | |
ea8add2b ED |
226 | offset = prandom_u32() % remaining; |
227 | /* __inet_hash_connect() favors ports having @low parity | |
228 | * We do the opposite to not pollute connect() users. | |
229 | */ | |
230 | offset |= 1U; | |
ea8add2b ED |
231 | |
232 | other_parity_scan: | |
233 | port = low + offset; | |
234 | for (i = 0; i < remaining; i += 2, port += 2) { | |
235 | if (unlikely(port >= high)) | |
236 | port -= remaining; | |
237 | if (inet_is_local_reserved_port(net, port)) | |
238 | continue; | |
239 | head = &hinfo->bhash[inet_bhashfn(net, port, | |
240 | hinfo->bhash_size)]; | |
241 | spin_lock_bh(&head->lock); | |
242 | inet_bind_bucket_for_each(tb, &head->chain) | |
593d1ebe JK |
243 | if (net_eq(ib_net(tb), net) && tb->l3mdev == l3mdev && |
244 | tb->port == port) { | |
245 | if (!inet_csk_bind_conflict(sk, tb, relax, false)) | |
6cd66616 | 246 | goto success; |
ea8add2b | 247 | goto next_port; |
946f9eb2 | 248 | } |
289141b7 JB |
249 | tb = NULL; |
250 | goto success; | |
ea8add2b ED |
251 | next_port: |
252 | spin_unlock_bh(&head->lock); | |
253 | cond_resched(); | |
254 | } | |
255 | ||
ea8add2b ED |
256 | offset--; |
257 | if (!(offset & 1)) | |
258 | goto other_parity_scan; | |
259 | ||
260 | if (attempt_half == 1) { | |
261 | /* OK we now try the upper half of the range */ | |
262 | attempt_half = 2; | |
263 | goto other_half_scan; | |
264 | } | |
4b01a967 KI |
265 | |
266 | if (net->ipv4.sysctl_ip_autobind_reuse && !relax) { | |
267 | /* We still have a chance to connect to different destinations */ | |
268 | relax = true; | |
269 | goto ports_exhausted; | |
270 | } | |
289141b7 JB |
271 | return NULL; |
272 | success: | |
273 | *port_ret = port; | |
274 | *tb_ret = tb; | |
275 | return head; | |
276 | } | |
ea8add2b | 277 | |
637bc8bb JB |
278 | static inline int sk_reuseport_match(struct inet_bind_bucket *tb, |
279 | struct sock *sk) | |
280 | { | |
281 | kuid_t uid = sock_i_uid(sk); | |
282 | ||
283 | if (tb->fastreuseport <= 0) | |
284 | return 0; | |
285 | if (!sk->sk_reuseport) | |
286 | return 0; | |
287 | if (rcu_access_pointer(sk->sk_reuseport_cb)) | |
288 | return 0; | |
289 | if (!uid_eq(tb->fastuid, uid)) | |
290 | return 0; | |
291 | /* We only need to check the rcv_saddr if this tb was once marked | |
292 | * without fastreuseport and then was reset, as we can only know that | |
293 | * the fast_*rcv_saddr doesn't have any conflicts with the socks on the | |
294 | * owners list. | |
295 | */ | |
296 | if (tb->fastreuseport == FASTREUSEPORT_ANY) | |
297 | return 1; | |
298 | #if IS_ENABLED(CONFIG_IPV6) | |
299 | if (tb->fast_sk_family == AF_INET6) | |
300 | return ipv6_rcv_saddr_equal(&tb->fast_v6_rcv_saddr, | |
7a56673b | 301 | inet6_rcv_saddr(sk), |
637bc8bb JB |
302 | tb->fast_rcv_saddr, |
303 | sk->sk_rcv_saddr, | |
304 | tb->fast_ipv6_only, | |
88d7fcfa | 305 | ipv6_only_sock(sk), true, false); |
637bc8bb JB |
306 | #endif |
307 | return ipv4_rcv_saddr_equal(tb->fast_rcv_saddr, sk->sk_rcv_saddr, | |
88d7fcfa | 308 | ipv6_only_sock(sk), true, false); |
637bc8bb JB |
309 | } |
310 | ||
62ffc589 TF |
311 | void inet_csk_update_fastreuse(struct inet_bind_bucket *tb, |
312 | struct sock *sk) | |
313 | { | |
314 | kuid_t uid = sock_i_uid(sk); | |
315 | bool reuse = sk->sk_reuse && sk->sk_state != TCP_LISTEN; | |
316 | ||
317 | if (hlist_empty(&tb->owners)) { | |
318 | tb->fastreuse = reuse; | |
319 | if (sk->sk_reuseport) { | |
320 | tb->fastreuseport = FASTREUSEPORT_ANY; | |
321 | tb->fastuid = uid; | |
322 | tb->fast_rcv_saddr = sk->sk_rcv_saddr; | |
323 | tb->fast_ipv6_only = ipv6_only_sock(sk); | |
324 | tb->fast_sk_family = sk->sk_family; | |
325 | #if IS_ENABLED(CONFIG_IPV6) | |
326 | tb->fast_v6_rcv_saddr = sk->sk_v6_rcv_saddr; | |
327 | #endif | |
328 | } else { | |
329 | tb->fastreuseport = 0; | |
330 | } | |
331 | } else { | |
332 | if (!reuse) | |
333 | tb->fastreuse = 0; | |
334 | if (sk->sk_reuseport) { | |
335 | /* We didn't match or we don't have fastreuseport set on | |
336 | * the tb, but we have sk_reuseport set on this socket | |
337 | * and we know that there are no bind conflicts with | |
338 | * this socket in this tb, so reset our tb's reuseport | |
339 | * settings so that any subsequent sockets that match | |
340 | * our current socket will be put on the fast path. | |
341 | * | |
342 | * If we reset we need to set FASTREUSEPORT_STRICT so we | |
343 | * do extra checking for all subsequent sk_reuseport | |
344 | * socks. | |
345 | */ | |
346 | if (!sk_reuseport_match(tb, sk)) { | |
347 | tb->fastreuseport = FASTREUSEPORT_STRICT; | |
348 | tb->fastuid = uid; | |
349 | tb->fast_rcv_saddr = sk->sk_rcv_saddr; | |
350 | tb->fast_ipv6_only = ipv6_only_sock(sk); | |
351 | tb->fast_sk_family = sk->sk_family; | |
352 | #if IS_ENABLED(CONFIG_IPV6) | |
353 | tb->fast_v6_rcv_saddr = sk->sk_v6_rcv_saddr; | |
354 | #endif | |
355 | } | |
356 | } else { | |
357 | tb->fastreuseport = 0; | |
358 | } | |
359 | } | |
360 | } | |
361 | ||
289141b7 JB |
362 | /* Obtain a reference to a local port for the given sock, |
363 | * if snum is zero it means select any available local port. | |
364 | * We try to allocate an odd port (and leave even ports for connect()) | |
365 | */ | |
366 | int inet_csk_get_port(struct sock *sk, unsigned short snum) | |
367 | { | |
368 | bool reuse = sk->sk_reuse && sk->sk_state != TCP_LISTEN; | |
369 | struct inet_hashinfo *hinfo = sk->sk_prot->h.hashinfo; | |
593d1ebe | 370 | int ret = 1, port = snum; |
289141b7 JB |
371 | struct inet_bind_hashbucket *head; |
372 | struct net *net = sock_net(sk); | |
593d1ebe | 373 | struct inet_bind_bucket *tb = NULL; |
3c82a21f RS |
374 | int l3mdev; |
375 | ||
376 | l3mdev = inet_sk_bound_l3mdev(sk); | |
289141b7 JB |
377 | |
378 | if (!port) { | |
593d1ebe | 379 | head = inet_csk_find_open_port(sk, &tb, &port); |
289141b7 JB |
380 | if (!head) |
381 | return ret; | |
382 | if (!tb) | |
593d1ebe JK |
383 | goto tb_not_found; |
384 | goto success; | |
d5a42de8 | 385 | } |
593d1ebe JK |
386 | head = &hinfo->bhash[inet_bhashfn(net, port, |
387 | hinfo->bhash_size)]; | |
388 | spin_lock_bh(&head->lock); | |
389 | inet_bind_bucket_for_each(tb, &head->chain) | |
390 | if (net_eq(ib_net(tb), net) && tb->l3mdev == l3mdev && | |
391 | tb->port == port) | |
392 | goto tb_found; | |
393 | tb_not_found: | |
394 | tb = inet_bind_bucket_create(hinfo->bind_bucket_cachep, | |
395 | net, head, port, l3mdev); | |
396 | if (!tb) | |
397 | goto fail_unlock; | |
398 | tb_found: | |
399 | if (!hlist_empty(&tb->owners)) { | |
4a17fd52 PE |
400 | if (sk->sk_reuse == SK_FORCE_REUSE) |
401 | goto success; | |
402 | ||
b9470c27 | 403 | if ((tb->fastreuse > 0 && reuse) || |
637bc8bb | 404 | sk_reuseport_match(tb, sk)) |
3f421baa | 405 | goto success; |
593d1ebe | 406 | if (inet_csk_bind_conflict(sk, tb, true, true)) |
ea8add2b | 407 | goto fail_unlock; |
6cd66616 JB |
408 | } |
409 | success: | |
62ffc589 TF |
410 | inet_csk_update_fastreuse(tb, sk); |
411 | ||
3f421baa | 412 | if (!inet_csk(sk)->icsk_bind_hash) |
593d1ebe | 413 | inet_bind_hash(sk, tb, port); |
547b792c | 414 | WARN_ON(inet_csk(sk)->icsk_bind_hash != tb); |
e905a9ed | 415 | ret = 0; |
3f421baa ACM |
416 | |
417 | fail_unlock: | |
ea8add2b | 418 | spin_unlock_bh(&head->lock); |
3f421baa ACM |
419 | return ret; |
420 | } | |
3f421baa ACM |
421 | EXPORT_SYMBOL_GPL(inet_csk_get_port); |
422 | ||
423 | /* | |
424 | * Wait for an incoming connection, avoid race conditions. This must be called | |
425 | * with the socket locked. | |
426 | */ | |
427 | static int inet_csk_wait_for_connect(struct sock *sk, long timeo) | |
428 | { | |
429 | struct inet_connection_sock *icsk = inet_csk(sk); | |
430 | DEFINE_WAIT(wait); | |
431 | int err; | |
432 | ||
433 | /* | |
434 | * True wake-one mechanism for incoming connections: only | |
435 | * one process gets woken up, not the 'whole herd'. | |
436 | * Since we do not 'race & poll' for established sockets | |
437 | * anymore, the common case will execute the loop only once. | |
438 | * | |
439 | * Subtle issue: "add_wait_queue_exclusive()" will be added | |
440 | * after any current non-exclusive waiters, and we know that | |
441 | * it will always _stay_ after any new non-exclusive waiters | |
442 | * because all non-exclusive waiters are added at the | |
443 | * beginning of the wait-queue. As such, it's ok to "drop" | |
444 | * our exclusiveness temporarily when we get woken up without | |
445 | * having to remove and re-insert us on the wait queue. | |
446 | */ | |
447 | for (;;) { | |
aa395145 | 448 | prepare_to_wait_exclusive(sk_sleep(sk), &wait, |
3f421baa ACM |
449 | TASK_INTERRUPTIBLE); |
450 | release_sock(sk); | |
451 | if (reqsk_queue_empty(&icsk->icsk_accept_queue)) | |
452 | timeo = schedule_timeout(timeo); | |
cb7cf8a3 | 453 | sched_annotate_sleep(); |
3f421baa ACM |
454 | lock_sock(sk); |
455 | err = 0; | |
456 | if (!reqsk_queue_empty(&icsk->icsk_accept_queue)) | |
457 | break; | |
458 | err = -EINVAL; | |
459 | if (sk->sk_state != TCP_LISTEN) | |
460 | break; | |
461 | err = sock_intr_errno(timeo); | |
462 | if (signal_pending(current)) | |
463 | break; | |
464 | err = -EAGAIN; | |
465 | if (!timeo) | |
466 | break; | |
467 | } | |
aa395145 | 468 | finish_wait(sk_sleep(sk), &wait); |
3f421baa ACM |
469 | return err; |
470 | } | |
471 | ||
472 | /* | |
473 | * This will accept the next outstanding connection. | |
474 | */ | |
cdfbabfb | 475 | struct sock *inet_csk_accept(struct sock *sk, int flags, int *err, bool kern) |
3f421baa ACM |
476 | { |
477 | struct inet_connection_sock *icsk = inet_csk(sk); | |
8336886f | 478 | struct request_sock_queue *queue = &icsk->icsk_accept_queue; |
8336886f | 479 | struct request_sock *req; |
e3d95ad7 | 480 | struct sock *newsk; |
3f421baa ACM |
481 | int error; |
482 | ||
483 | lock_sock(sk); | |
484 | ||
485 | /* We need to make sure that this socket is listening, | |
486 | * and that it has something pending. | |
487 | */ | |
488 | error = -EINVAL; | |
489 | if (sk->sk_state != TCP_LISTEN) | |
490 | goto out_err; | |
491 | ||
492 | /* Find already established connection */ | |
8336886f | 493 | if (reqsk_queue_empty(queue)) { |
3f421baa ACM |
494 | long timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK); |
495 | ||
496 | /* If this is a non blocking socket don't sleep */ | |
497 | error = -EAGAIN; | |
498 | if (!timeo) | |
499 | goto out_err; | |
500 | ||
501 | error = inet_csk_wait_for_connect(sk, timeo); | |
502 | if (error) | |
503 | goto out_err; | |
504 | } | |
fff1f300 | 505 | req = reqsk_queue_remove(queue, sk); |
8336886f JC |
506 | newsk = req->sk; |
507 | ||
e3d95ad7 | 508 | if (sk->sk_protocol == IPPROTO_TCP && |
0536fcc0 ED |
509 | tcp_rsk(req)->tfo_listener) { |
510 | spin_lock_bh(&queue->fastopenq.lock); | |
9439ce00 | 511 | if (tcp_rsk(req)->tfo_listener) { |
8336886f JC |
512 | /* We are still waiting for the final ACK from 3WHS |
513 | * so can't free req now. Instead, we set req->sk to | |
514 | * NULL to signify that the child socket is taken | |
515 | * so reqsk_fastopen_remove() will free the req | |
516 | * when 3WHS finishes (or is aborted). | |
517 | */ | |
518 | req->sk = NULL; | |
519 | req = NULL; | |
520 | } | |
0536fcc0 | 521 | spin_unlock_bh(&queue->fastopenq.lock); |
8336886f | 522 | } |
d752a498 | 523 | |
3f421baa ACM |
524 | out: |
525 | release_sock(sk); | |
06669ea3 | 526 | if (newsk && mem_cgroup_sockets_enabled) { |
d752a498 SB |
527 | int amt; |
528 | ||
529 | /* atomically get the memory usage, set and charge the | |
06669ea3 | 530 | * newsk->sk_memcg. |
d752a498 SB |
531 | */ |
532 | lock_sock(newsk); | |
533 | ||
06669ea3 ED |
534 | /* The socket has not been accepted yet, no need to look at |
535 | * newsk->sk_wmem_queued. | |
d752a498 SB |
536 | */ |
537 | amt = sk_mem_pages(newsk->sk_forward_alloc + | |
06669ea3 | 538 | atomic_read(&newsk->sk_rmem_alloc)); |
d752a498 SB |
539 | mem_cgroup_sk_alloc(newsk); |
540 | if (newsk->sk_memcg && amt) | |
4b1327be WW |
541 | mem_cgroup_charge_skmem(newsk->sk_memcg, amt, |
542 | GFP_KERNEL | __GFP_NOFAIL); | |
d752a498 SB |
543 | |
544 | release_sock(newsk); | |
545 | } | |
8336886f | 546 | if (req) |
13854e5a | 547 | reqsk_put(req); |
3f421baa ACM |
548 | return newsk; |
549 | out_err: | |
550 | newsk = NULL; | |
8336886f | 551 | req = NULL; |
3f421baa ACM |
552 | *err = error; |
553 | goto out; | |
554 | } | |
3f421baa ACM |
555 | EXPORT_SYMBOL(inet_csk_accept); |
556 | ||
557 | /* | |
558 | * Using different timers for retransmit, delayed acks and probes | |
e905a9ed | 559 | * We may wish use just one timer maintaining a list of expire jiffies |
3f421baa ACM |
560 | * to optimize. |
561 | */ | |
562 | void inet_csk_init_xmit_timers(struct sock *sk, | |
59f379f9 KC |
563 | void (*retransmit_handler)(struct timer_list *t), |
564 | void (*delack_handler)(struct timer_list *t), | |
565 | void (*keepalive_handler)(struct timer_list *t)) | |
3f421baa ACM |
566 | { |
567 | struct inet_connection_sock *icsk = inet_csk(sk); | |
568 | ||
59f379f9 KC |
569 | timer_setup(&icsk->icsk_retransmit_timer, retransmit_handler, 0); |
570 | timer_setup(&icsk->icsk_delack_timer, delack_handler, 0); | |
571 | timer_setup(&sk->sk_timer, keepalive_handler, 0); | |
3f421baa ACM |
572 | icsk->icsk_pending = icsk->icsk_ack.pending = 0; |
573 | } | |
3f421baa ACM |
574 | EXPORT_SYMBOL(inet_csk_init_xmit_timers); |
575 | ||
576 | void inet_csk_clear_xmit_timers(struct sock *sk) | |
577 | { | |
578 | struct inet_connection_sock *icsk = inet_csk(sk); | |
579 | ||
b6b6d653 | 580 | icsk->icsk_pending = icsk->icsk_ack.pending = 0; |
3f421baa ACM |
581 | |
582 | sk_stop_timer(sk, &icsk->icsk_retransmit_timer); | |
583 | sk_stop_timer(sk, &icsk->icsk_delack_timer); | |
584 | sk_stop_timer(sk, &sk->sk_timer); | |
585 | } | |
3f421baa ACM |
586 | EXPORT_SYMBOL(inet_csk_clear_xmit_timers); |
587 | ||
588 | void inet_csk_delete_keepalive_timer(struct sock *sk) | |
589 | { | |
590 | sk_stop_timer(sk, &sk->sk_timer); | |
591 | } | |
3f421baa ACM |
592 | EXPORT_SYMBOL(inet_csk_delete_keepalive_timer); |
593 | ||
594 | void inet_csk_reset_keepalive_timer(struct sock *sk, unsigned long len) | |
595 | { | |
596 | sk_reset_timer(sk, &sk->sk_timer, jiffies + len); | |
597 | } | |
3f421baa ACM |
598 | EXPORT_SYMBOL(inet_csk_reset_keepalive_timer); |
599 | ||
e5895bc6 | 600 | struct dst_entry *inet_csk_route_req(const struct sock *sk, |
6bd023f3 | 601 | struct flowi4 *fl4, |
ba3f7f04 | 602 | const struct request_sock *req) |
3f421baa | 603 | { |
3f421baa | 604 | const struct inet_request_sock *ireq = inet_rsk(req); |
8b929ab1 | 605 | struct net *net = read_pnet(&ireq->ireq_net); |
c92e8c02 | 606 | struct ip_options_rcu *opt; |
8b929ab1 | 607 | struct rtable *rt; |
3f421baa | 608 | |
2ab2ddd3 ED |
609 | rcu_read_lock(); |
610 | opt = rcu_dereference(ireq->ireq_opt); | |
06f877d6 | 611 | |
8b929ab1 | 612 | flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark, |
e79d9bc7 | 613 | RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE, |
8b929ab1 | 614 | sk->sk_protocol, inet_sk_flowi_flags(sk), |
634fb979 | 615 | (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr, |
8b929ab1 | 616 | ireq->ir_loc_addr, ireq->ir_rmt_port, |
e2d118a1 | 617 | htons(ireq->ir_num), sk->sk_uid); |
3df98d79 | 618 | security_req_classify_flow(req, flowi4_to_flowi_common(fl4)); |
6bd023f3 | 619 | rt = ip_route_output_flow(net, fl4, sk); |
b23dd4fe | 620 | if (IS_ERR(rt)) |
857a6e0a | 621 | goto no_route; |
77d5bc7e | 622 | if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway) |
857a6e0a | 623 | goto route_err; |
2ab2ddd3 | 624 | rcu_read_unlock(); |
d8d1f30b | 625 | return &rt->dst; |
857a6e0a IJ |
626 | |
627 | route_err: | |
628 | ip_rt_put(rt); | |
629 | no_route: | |
2ab2ddd3 | 630 | rcu_read_unlock(); |
b45386ef | 631 | __IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); |
857a6e0a | 632 | return NULL; |
3f421baa | 633 | } |
3f421baa ACM |
634 | EXPORT_SYMBOL_GPL(inet_csk_route_req); |
635 | ||
a2432c4f | 636 | struct dst_entry *inet_csk_route_child_sock(const struct sock *sk, |
77357a95 DM |
637 | struct sock *newsk, |
638 | const struct request_sock *req) | |
639 | { | |
640 | const struct inet_request_sock *ireq = inet_rsk(req); | |
8b929ab1 | 641 | struct net *net = read_pnet(&ireq->ireq_net); |
77357a95 | 642 | struct inet_sock *newinet = inet_sk(newsk); |
1a7b27c9 | 643 | struct ip_options_rcu *opt; |
77357a95 DM |
644 | struct flowi4 *fl4; |
645 | struct rtable *rt; | |
646 | ||
c92e8c02 | 647 | opt = rcu_dereference(ireq->ireq_opt); |
77357a95 | 648 | fl4 = &newinet->cork.fl.u.ip4; |
1a7b27c9 | 649 | |
8b929ab1 | 650 | flowi4_init_output(fl4, ireq->ir_iif, ireq->ir_mark, |
77357a95 DM |
651 | RT_CONN_FLAGS(sk), RT_SCOPE_UNIVERSE, |
652 | sk->sk_protocol, inet_sk_flowi_flags(sk), | |
634fb979 | 653 | (opt && opt->opt.srr) ? opt->opt.faddr : ireq->ir_rmt_addr, |
8b929ab1 | 654 | ireq->ir_loc_addr, ireq->ir_rmt_port, |
e2d118a1 | 655 | htons(ireq->ir_num), sk->sk_uid); |
3df98d79 | 656 | security_req_classify_flow(req, flowi4_to_flowi_common(fl4)); |
77357a95 DM |
657 | rt = ip_route_output_flow(net, fl4, sk); |
658 | if (IS_ERR(rt)) | |
659 | goto no_route; | |
77d5bc7e | 660 | if (opt && opt->opt.is_strictroute && rt->rt_uses_gateway) |
77357a95 DM |
661 | goto route_err; |
662 | return &rt->dst; | |
663 | ||
664 | route_err: | |
665 | ip_rt_put(rt); | |
666 | no_route: | |
b45386ef | 667 | __IP_INC_STATS(net, IPSTATS_MIB_OUTNOROUTES); |
77357a95 DM |
668 | return NULL; |
669 | } | |
670 | EXPORT_SYMBOL_GPL(inet_csk_route_child_sock); | |
671 | ||
0c3d79bc | 672 | /* Decide when to expire the request and when to resend SYN-ACK */ |
a594920f KI |
673 | static void syn_ack_recalc(struct request_sock *req, |
674 | const int max_syn_ack_retries, | |
675 | const u8 rskq_defer_accept, | |
676 | int *expire, int *resend) | |
0c3d79bc JA |
677 | { |
678 | if (!rskq_defer_accept) { | |
a594920f | 679 | *expire = req->num_timeout >= max_syn_ack_retries; |
0c3d79bc JA |
680 | *resend = 1; |
681 | return; | |
682 | } | |
a594920f KI |
683 | *expire = req->num_timeout >= max_syn_ack_retries && |
684 | (!inet_rsk(req)->acked || req->num_timeout >= rskq_defer_accept); | |
685 | /* Do not resend while waiting for data after ACK, | |
0c3d79bc JA |
686 | * start to resend on end of deferring period to give |
687 | * last chance for data or ACK to create established socket. | |
688 | */ | |
689 | *resend = !inet_rsk(req)->acked || | |
e6c022a4 | 690 | req->num_timeout >= rskq_defer_accept - 1; |
0c3d79bc JA |
691 | } |
692 | ||
1b70e977 | 693 | int inet_rtx_syn_ack(const struct sock *parent, struct request_sock *req) |
e6c022a4 | 694 | { |
1a2c6181 | 695 | int err = req->rsk_ops->rtx_syn_ack(parent, req); |
e6c022a4 ED |
696 | |
697 | if (!err) | |
698 | req->num_retrans++; | |
699 | return err; | |
700 | } | |
701 | EXPORT_SYMBOL(inet_rtx_syn_ack); | |
702 | ||
54b92e84 KI |
703 | static struct request_sock *inet_reqsk_clone(struct request_sock *req, |
704 | struct sock *sk) | |
705 | { | |
706 | struct sock *req_sk, *nreq_sk; | |
707 | struct request_sock *nreq; | |
708 | ||
709 | nreq = kmem_cache_alloc(req->rsk_ops->slab, GFP_ATOMIC | __GFP_NOWARN); | |
710 | if (!nreq) { | |
55d444b3 KI |
711 | __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQFAILURE); |
712 | ||
54b92e84 KI |
713 | /* paired with refcount_inc_not_zero() in reuseport_migrate_sock() */ |
714 | sock_put(sk); | |
715 | return NULL; | |
716 | } | |
717 | ||
718 | req_sk = req_to_sk(req); | |
719 | nreq_sk = req_to_sk(nreq); | |
720 | ||
721 | memcpy(nreq_sk, req_sk, | |
722 | offsetof(struct sock, sk_dontcopy_begin)); | |
723 | memcpy(&nreq_sk->sk_dontcopy_end, &req_sk->sk_dontcopy_end, | |
724 | req->rsk_ops->obj_size - offsetof(struct sock, sk_dontcopy_end)); | |
725 | ||
726 | sk_node_init(&nreq_sk->sk_node); | |
727 | nreq_sk->sk_tx_queue_mapping = req_sk->sk_tx_queue_mapping; | |
a9418924 | 728 | #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING |
54b92e84 KI |
729 | nreq_sk->sk_rx_queue_mapping = req_sk->sk_rx_queue_mapping; |
730 | #endif | |
731 | nreq_sk->sk_incoming_cpu = req_sk->sk_incoming_cpu; | |
732 | ||
733 | nreq->rsk_listener = sk; | |
734 | ||
735 | /* We need not acquire fastopenq->lock | |
736 | * because the child socket is locked in inet_csk_listen_stop(). | |
737 | */ | |
738 | if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(nreq)->tfo_listener) | |
739 | rcu_assign_pointer(tcp_sk(nreq->sk)->fastopen_rsk, nreq); | |
740 | ||
741 | return nreq; | |
742 | } | |
743 | ||
c905dee6 KI |
744 | static void reqsk_queue_migrated(struct request_sock_queue *queue, |
745 | const struct request_sock *req) | |
746 | { | |
747 | if (req->num_timeout == 0) | |
748 | atomic_inc(&queue->young); | |
749 | atomic_inc(&queue->qlen); | |
750 | } | |
751 | ||
54b92e84 KI |
752 | static void reqsk_migrate_reset(struct request_sock *req) |
753 | { | |
c905dee6 | 754 | req->saved_syn = NULL; |
54b92e84 KI |
755 | #if IS_ENABLED(CONFIG_IPV6) |
756 | inet_rsk(req)->ipv6_opt = NULL; | |
c905dee6 KI |
757 | inet_rsk(req)->pktopts = NULL; |
758 | #else | |
759 | inet_rsk(req)->ireq_opt = NULL; | |
54b92e84 KI |
760 | #endif |
761 | } | |
762 | ||
079096f1 | 763 | /* return true if req was found in the ehash table */ |
8b5e07d7 | 764 | static bool reqsk_queue_unlink(struct request_sock *req) |
b357a364 | 765 | { |
079096f1 | 766 | struct inet_hashinfo *hashinfo = req_to_sk(req)->sk_prot->h.hashinfo; |
5e0724d0 | 767 | bool found = false; |
b357a364 | 768 | |
5e0724d0 ED |
769 | if (sk_hashed(req_to_sk(req))) { |
770 | spinlock_t *lock = inet_ehash_lockp(hashinfo, req->rsk_hash); | |
b357a364 | 771 | |
5e0724d0 ED |
772 | spin_lock(lock); |
773 | found = __sk_nulls_del_node_init_rcu(req_to_sk(req)); | |
774 | spin_unlock(lock); | |
775 | } | |
83fccfc3 | 776 | if (timer_pending(&req->rsk_timer) && del_timer_sync(&req->rsk_timer)) |
b357a364 ED |
777 | reqsk_put(req); |
778 | return found; | |
779 | } | |
780 | ||
7233da86 | 781 | bool inet_csk_reqsk_queue_drop(struct sock *sk, struct request_sock *req) |
b357a364 | 782 | { |
7233da86 AO |
783 | bool unlinked = reqsk_queue_unlink(req); |
784 | ||
785 | if (unlinked) { | |
b357a364 ED |
786 | reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req); |
787 | reqsk_put(req); | |
788 | } | |
7233da86 | 789 | return unlinked; |
b357a364 ED |
790 | } |
791 | EXPORT_SYMBOL(inet_csk_reqsk_queue_drop); | |
792 | ||
f03f2e15 ED |
793 | void inet_csk_reqsk_queue_drop_and_put(struct sock *sk, struct request_sock *req) |
794 | { | |
795 | inet_csk_reqsk_queue_drop(sk, req); | |
796 | reqsk_put(req); | |
797 | } | |
798 | EXPORT_SYMBOL(inet_csk_reqsk_queue_drop_and_put); | |
799 | ||
59f379f9 | 800 | static void reqsk_timer_handler(struct timer_list *t) |
a019d6fe | 801 | { |
59f379f9 | 802 | struct request_sock *req = from_timer(req, t, rsk_timer); |
c905dee6 | 803 | struct request_sock *nreq = NULL, *oreq = req; |
fa76ce73 | 804 | struct sock *sk_listener = req->rsk_listener; |
c905dee6 KI |
805 | struct inet_connection_sock *icsk; |
806 | struct request_sock_queue *queue; | |
807 | struct net *net; | |
a594920f | 808 | int max_syn_ack_retries, qlen, expire = 0, resend = 0; |
a019d6fe | 809 | |
c905dee6 KI |
810 | if (inet_sk_state_load(sk_listener) != TCP_LISTEN) { |
811 | struct sock *nsk; | |
812 | ||
813 | nsk = reuseport_migrate_sock(sk_listener, req_to_sk(req), NULL); | |
814 | if (!nsk) | |
815 | goto drop; | |
816 | ||
817 | nreq = inet_reqsk_clone(req, nsk); | |
818 | if (!nreq) | |
819 | goto drop; | |
820 | ||
821 | /* The new timer for the cloned req can decrease the 2 | |
822 | * by calling inet_csk_reqsk_queue_drop_and_put(), so | |
823 | * hold another count to prevent use-after-free and | |
824 | * call reqsk_put() just before return. | |
825 | */ | |
826 | refcount_set(&nreq->rsk_refcnt, 2 + 1); | |
827 | timer_setup(&nreq->rsk_timer, reqsk_timer_handler, TIMER_PINNED); | |
828 | reqsk_queue_migrated(&inet_csk(nsk)->icsk_accept_queue, req); | |
829 | ||
830 | req = nreq; | |
831 | sk_listener = nsk; | |
832 | } | |
a019d6fe | 833 | |
c905dee6 KI |
834 | icsk = inet_csk(sk_listener); |
835 | net = sock_net(sk_listener); | |
a594920f | 836 | max_syn_ack_retries = icsk->icsk_syn_retries ? : net->ipv4.sysctl_tcp_synack_retries; |
a019d6fe ACM |
837 | /* Normally all the openreqs are young and become mature |
838 | * (i.e. converted to established socket) for first timeout. | |
fd4f2cea | 839 | * If synack was not acknowledged for 1 second, it means |
a019d6fe ACM |
840 | * one of the following things: synack was lost, ack was lost, |
841 | * rtt is high or nobody planned to ack (i.e. synflood). | |
842 | * When server is a bit loaded, queue is populated with old | |
843 | * open requests, reducing effective size of queue. | |
844 | * When server is well loaded, queue size reduces to zero | |
845 | * after several minutes of work. It is not synflood, | |
846 | * it is normal operation. The solution is pruning | |
847 | * too old entries overriding normal timeout, when | |
848 | * situation becomes dangerous. | |
849 | * | |
850 | * Essentially, we reserve half of room for young | |
851 | * embrions; and abort old ones without pity, if old | |
852 | * ones are about to clog our table. | |
853 | */ | |
c905dee6 | 854 | queue = &icsk->icsk_accept_queue; |
aac065c5 | 855 | qlen = reqsk_queue_len(queue); |
099ecf59 | 856 | if ((qlen << 1) > max(8U, READ_ONCE(sk_listener->sk_max_ack_backlog))) { |
aac065c5 | 857 | int young = reqsk_queue_len_young(queue) << 1; |
a019d6fe | 858 | |
a594920f | 859 | while (max_syn_ack_retries > 2) { |
2b41fab7 | 860 | if (qlen < young) |
a019d6fe | 861 | break; |
a594920f | 862 | max_syn_ack_retries--; |
a019d6fe ACM |
863 | young <<= 1; |
864 | } | |
865 | } | |
a594920f | 866 | syn_ack_recalc(req, max_syn_ack_retries, READ_ONCE(queue->rskq_defer_accept), |
fa76ce73 | 867 | &expire, &resend); |
42cb80a2 | 868 | req->rsk_ops->syn_ack_timeout(req); |
fa76ce73 ED |
869 | if (!expire && |
870 | (!resend || | |
871 | !inet_rtx_syn_ack(sk_listener, req) || | |
872 | inet_rsk(req)->acked)) { | |
fa76ce73 | 873 | if (req->num_timeout++ == 0) |
aac065c5 | 874 | atomic_dec(&queue->young); |
5903123f | 875 | mod_timer(&req->rsk_timer, jiffies + reqsk_timeout(req, TCP_RTO_MAX)); |
c905dee6 KI |
876 | |
877 | if (!nreq) | |
878 | return; | |
879 | ||
880 | if (!inet_ehash_insert(req_to_sk(nreq), req_to_sk(oreq), NULL)) { | |
881 | /* delete timer */ | |
882 | inet_csk_reqsk_queue_drop(sk_listener, nreq); | |
55d444b3 | 883 | goto no_ownership; |
c905dee6 KI |
884 | } |
885 | ||
55d444b3 | 886 | __NET_INC_STATS(net, LINUX_MIB_TCPMIGRATEREQSUCCESS); |
c905dee6 KI |
887 | reqsk_migrate_reset(oreq); |
888 | reqsk_queue_removed(&inet_csk(oreq->rsk_listener)->icsk_accept_queue, oreq); | |
889 | reqsk_put(oreq); | |
890 | ||
891 | reqsk_put(nreq); | |
fa76ce73 ED |
892 | return; |
893 | } | |
c905dee6 | 894 | |
c905dee6 KI |
895 | /* Even if we can clone the req, we may need not retransmit any more |
896 | * SYN+ACKs (nreq->num_timeout > max_syn_ack_retries, etc), or another | |
897 | * CPU may win the "own_req" race so that inet_ehash_insert() fails. | |
898 | */ | |
899 | if (nreq) { | |
55d444b3 KI |
900 | __NET_INC_STATS(net, LINUX_MIB_TCPMIGRATEREQFAILURE); |
901 | no_ownership: | |
c905dee6 KI |
902 | reqsk_migrate_reset(nreq); |
903 | reqsk_queue_removed(queue, nreq); | |
904 | __reqsk_free(nreq); | |
905 | } | |
906 | ||
55d444b3 | 907 | drop: |
c905dee6 | 908 | inet_csk_reqsk_queue_drop_and_put(oreq->rsk_listener, oreq); |
fa76ce73 | 909 | } |
ec0a1966 | 910 | |
079096f1 ED |
911 | static void reqsk_queue_hash_req(struct request_sock *req, |
912 | unsigned long timeout) | |
fa76ce73 | 913 | { |
59f379f9 | 914 | timer_setup(&req->rsk_timer, reqsk_timer_handler, TIMER_PINNED); |
f3438bc7 | 915 | mod_timer(&req->rsk_timer, jiffies + timeout); |
29c68526 | 916 | |
01770a16 | 917 | inet_ehash_insert(req_to_sk(req), NULL, NULL); |
fa76ce73 ED |
918 | /* before letting lookups find us, make sure all req fields |
919 | * are committed to memory and refcnt initialized. | |
920 | */ | |
921 | smp_wmb(); | |
41c6d650 | 922 | refcount_set(&req->rsk_refcnt, 2 + 1); |
079096f1 | 923 | } |
a019d6fe | 924 | |
079096f1 ED |
925 | void inet_csk_reqsk_queue_hash_add(struct sock *sk, struct request_sock *req, |
926 | unsigned long timeout) | |
927 | { | |
928 | reqsk_queue_hash_req(req, timeout); | |
929 | inet_csk_reqsk_queue_added(sk); | |
a019d6fe | 930 | } |
079096f1 | 931 | EXPORT_SYMBOL_GPL(inet_csk_reqsk_queue_hash_add); |
a019d6fe | 932 | |
13230593 MM |
933 | static void inet_clone_ulp(const struct request_sock *req, struct sock *newsk, |
934 | const gfp_t priority) | |
935 | { | |
936 | struct inet_connection_sock *icsk = inet_csk(newsk); | |
937 | ||
938 | if (!icsk->icsk_ulp_ops) | |
939 | return; | |
940 | ||
941 | if (icsk->icsk_ulp_ops->clone) | |
942 | icsk->icsk_ulp_ops->clone(req, newsk, priority); | |
943 | } | |
944 | ||
e56c57d0 ED |
945 | /** |
946 | * inet_csk_clone_lock - clone an inet socket, and lock its clone | |
947 | * @sk: the socket to clone | |
948 | * @req: request_sock | |
949 | * @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc) | |
950 | * | |
951 | * Caller must unlock socket even in error path (bh_unlock_sock(newsk)) | |
952 | */ | |
953 | struct sock *inet_csk_clone_lock(const struct sock *sk, | |
954 | const struct request_sock *req, | |
955 | const gfp_t priority) | |
9f1d2604 | 956 | { |
e56c57d0 | 957 | struct sock *newsk = sk_clone_lock(sk, priority); |
9f1d2604 | 958 | |
00db4124 | 959 | if (newsk) { |
9f1d2604 ACM |
960 | struct inet_connection_sock *newicsk = inet_csk(newsk); |
961 | ||
563e0bb0 | 962 | inet_sk_set_state(newsk, TCP_SYN_RECV); |
9f1d2604 ACM |
963 | newicsk->icsk_bind_hash = NULL; |
964 | ||
634fb979 | 965 | inet_sk(newsk)->inet_dport = inet_rsk(req)->ir_rmt_port; |
b44084c2 ED |
966 | inet_sk(newsk)->inet_num = inet_rsk(req)->ir_num; |
967 | inet_sk(newsk)->inet_sport = htons(inet_rsk(req)->ir_num); | |
9f1d2604 | 968 | |
85017869 ED |
969 | /* listeners have SOCK_RCU_FREE, not the children */ |
970 | sock_reset_flag(newsk, SOCK_RCU_FREE); | |
971 | ||
657831ff ED |
972 | inet_sk(newsk)->mc_list = NULL; |
973 | ||
84f39b08 | 974 | newsk->sk_mark = inet_rsk(req)->ir_mark; |
33cf7c90 ED |
975 | atomic64_set(&newsk->sk_cookie, |
976 | atomic64_read(&inet_rsk(req)->ir_cookie)); | |
84f39b08 | 977 | |
9f1d2604 | 978 | newicsk->icsk_retransmits = 0; |
6687e988 ACM |
979 | newicsk->icsk_backoff = 0; |
980 | newicsk->icsk_probes_out = 0; | |
9d9b1ee0 | 981 | newicsk->icsk_probes_tstamp = 0; |
9f1d2604 ACM |
982 | |
983 | /* Deinitialize accept_queue to trap illegal accesses. */ | |
984 | memset(&newicsk->icsk_accept_queue, 0, sizeof(newicsk->icsk_accept_queue)); | |
4237c75c | 985 | |
13230593 MM |
986 | inet_clone_ulp(req, newsk, priority); |
987 | ||
4237c75c | 988 | security_inet_csk_clone(newsk, req); |
9f1d2604 ACM |
989 | } |
990 | return newsk; | |
991 | } | |
e56c57d0 | 992 | EXPORT_SYMBOL_GPL(inet_csk_clone_lock); |
a019d6fe ACM |
993 | |
994 | /* | |
995 | * At this point, there should be no process reference to this | |
996 | * socket, and thus no user references at all. Therefore we | |
997 | * can assume the socket waitqueue is inactive and nobody will | |
998 | * try to jump onto it. | |
999 | */ | |
1000 | void inet_csk_destroy_sock(struct sock *sk) | |
1001 | { | |
547b792c IJ |
1002 | WARN_ON(sk->sk_state != TCP_CLOSE); |
1003 | WARN_ON(!sock_flag(sk, SOCK_DEAD)); | |
a019d6fe ACM |
1004 | |
1005 | /* It cannot be in hash table! */ | |
547b792c | 1006 | WARN_ON(!sk_unhashed(sk)); |
a019d6fe | 1007 | |
c720c7e8 ED |
1008 | /* If it has not 0 inet_sk(sk)->inet_num, it must be bound */ |
1009 | WARN_ON(inet_sk(sk)->inet_num && !inet_csk(sk)->icsk_bind_hash); | |
a019d6fe ACM |
1010 | |
1011 | sk->sk_prot->destroy(sk); | |
1012 | ||
1013 | sk_stream_kill_queues(sk); | |
1014 | ||
1015 | xfrm_sk_free_policy(sk); | |
1016 | ||
1017 | sk_refcnt_debug_release(sk); | |
1018 | ||
19757ceb | 1019 | this_cpu_dec(*sk->sk_prot->orphan_count); |
c2a2efbb | 1020 | |
a019d6fe ACM |
1021 | sock_put(sk); |
1022 | } | |
a019d6fe ACM |
1023 | EXPORT_SYMBOL(inet_csk_destroy_sock); |
1024 | ||
e337e24d CP |
1025 | /* This function allows to force a closure of a socket after the call to |
1026 | * tcp/dccp_create_openreq_child(). | |
1027 | */ | |
1028 | void inet_csk_prepare_forced_close(struct sock *sk) | |
c10cb5fc | 1029 | __releases(&sk->sk_lock.slock) |
e337e24d CP |
1030 | { |
1031 | /* sk_clone_lock locked the socket and set refcnt to 2 */ | |
1032 | bh_unlock_sock(sk); | |
1033 | sock_put(sk); | |
2f8a397d | 1034 | inet_csk_prepare_for_destroy_sock(sk); |
6761893e | 1035 | inet_sk(sk)->inet_num = 0; |
e337e24d CP |
1036 | } |
1037 | EXPORT_SYMBOL(inet_csk_prepare_forced_close); | |
1038 | ||
e7049395 | 1039 | int inet_csk_listen_start(struct sock *sk) |
a019d6fe | 1040 | { |
a019d6fe | 1041 | struct inet_connection_sock *icsk = inet_csk(sk); |
10cbc8f1 | 1042 | struct inet_sock *inet = inet_sk(sk); |
086c653f | 1043 | int err = -EADDRINUSE; |
a019d6fe | 1044 | |
ef547f2a | 1045 | reqsk_queue_alloc(&icsk->icsk_accept_queue); |
a019d6fe | 1046 | |
a019d6fe ACM |
1047 | sk->sk_ack_backlog = 0; |
1048 | inet_csk_delack_init(sk); | |
1049 | ||
26859240 AK |
1050 | if (sk->sk_txrehash == SOCK_TXREHASH_DEFAULT) |
1051 | sk->sk_txrehash = READ_ONCE(sock_net(sk)->core.sysctl_txrehash); | |
1052 | ||
a019d6fe ACM |
1053 | /* There is race window here: we announce ourselves listening, |
1054 | * but this transition is still not validated by get_port(). | |
1055 | * It is OK, because this socket enters to hash table only | |
1056 | * after validation is complete. | |
1057 | */ | |
563e0bb0 | 1058 | inet_sk_state_store(sk, TCP_LISTEN); |
c720c7e8 ED |
1059 | if (!sk->sk_prot->get_port(sk, inet->inet_num)) { |
1060 | inet->inet_sport = htons(inet->inet_num); | |
a019d6fe ACM |
1061 | |
1062 | sk_dst_reset(sk); | |
086c653f | 1063 | err = sk->sk_prot->hash(sk); |
a019d6fe | 1064 | |
086c653f CG |
1065 | if (likely(!err)) |
1066 | return 0; | |
a019d6fe ACM |
1067 | } |
1068 | ||
563e0bb0 | 1069 | inet_sk_set_state(sk, TCP_CLOSE); |
086c653f | 1070 | return err; |
a019d6fe | 1071 | } |
a019d6fe ACM |
1072 | EXPORT_SYMBOL_GPL(inet_csk_listen_start); |
1073 | ||
ebb516af ED |
1074 | static void inet_child_forget(struct sock *sk, struct request_sock *req, |
1075 | struct sock *child) | |
1076 | { | |
1077 | sk->sk_prot->disconnect(child, O_NONBLOCK); | |
1078 | ||
1079 | sock_orphan(child); | |
1080 | ||
19757ceb | 1081 | this_cpu_inc(*sk->sk_prot->orphan_count); |
ebb516af ED |
1082 | |
1083 | if (sk->sk_protocol == IPPROTO_TCP && tcp_rsk(req)->tfo_listener) { | |
d983ea6f | 1084 | BUG_ON(rcu_access_pointer(tcp_sk(child)->fastopen_rsk) != req); |
ebb516af ED |
1085 | BUG_ON(sk != req->rsk_listener); |
1086 | ||
1087 | /* Paranoid, to prevent race condition if | |
1088 | * an inbound pkt destined for child is | |
1089 | * blocked by sock lock in tcp_v4_rcv(). | |
1090 | * Also to satisfy an assertion in | |
1091 | * tcp_v4_destroy_sock(). | |
1092 | */ | |
d983ea6f | 1093 | RCU_INIT_POINTER(tcp_sk(child)->fastopen_rsk, NULL); |
ebb516af ED |
1094 | } |
1095 | inet_csk_destroy_sock(child); | |
ebb516af ED |
1096 | } |
1097 | ||
7716682c ED |
1098 | struct sock *inet_csk_reqsk_queue_add(struct sock *sk, |
1099 | struct request_sock *req, | |
1100 | struct sock *child) | |
ebb516af ED |
1101 | { |
1102 | struct request_sock_queue *queue = &inet_csk(sk)->icsk_accept_queue; | |
1103 | ||
1104 | spin_lock(&queue->rskq_lock); | |
1105 | if (unlikely(sk->sk_state != TCP_LISTEN)) { | |
1106 | inet_child_forget(sk, req, child); | |
7716682c | 1107 | child = NULL; |
ebb516af ED |
1108 | } else { |
1109 | req->sk = child; | |
1110 | req->dl_next = NULL; | |
1111 | if (queue->rskq_accept_head == NULL) | |
60b173ca | 1112 | WRITE_ONCE(queue->rskq_accept_head, req); |
ebb516af ED |
1113 | else |
1114 | queue->rskq_accept_tail->dl_next = req; | |
1115 | queue->rskq_accept_tail = req; | |
1116 | sk_acceptq_added(sk); | |
1117 | } | |
1118 | spin_unlock(&queue->rskq_lock); | |
7716682c | 1119 | return child; |
ebb516af ED |
1120 | } |
1121 | EXPORT_SYMBOL(inet_csk_reqsk_queue_add); | |
1122 | ||
5e0724d0 ED |
1123 | struct sock *inet_csk_complete_hashdance(struct sock *sk, struct sock *child, |
1124 | struct request_sock *req, bool own_req) | |
1125 | { | |
1126 | if (own_req) { | |
d4f2c86b KI |
1127 | inet_csk_reqsk_queue_drop(req->rsk_listener, req); |
1128 | reqsk_queue_removed(&inet_csk(req->rsk_listener)->icsk_accept_queue, req); | |
1129 | ||
1130 | if (sk != req->rsk_listener) { | |
1131 | /* another listening sk has been selected, | |
1132 | * migrate the req to it. | |
1133 | */ | |
1134 | struct request_sock *nreq; | |
1135 | ||
1136 | /* hold a refcnt for the nreq->rsk_listener | |
1137 | * which is assigned in inet_reqsk_clone() | |
1138 | */ | |
1139 | sock_hold(sk); | |
1140 | nreq = inet_reqsk_clone(req, sk); | |
1141 | if (!nreq) { | |
1142 | inet_child_forget(sk, req, child); | |
1143 | goto child_put; | |
1144 | } | |
1145 | ||
1146 | refcount_set(&nreq->rsk_refcnt, 1); | |
1147 | if (inet_csk_reqsk_queue_add(sk, nreq, child)) { | |
55d444b3 | 1148 | __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQSUCCESS); |
d4f2c86b KI |
1149 | reqsk_migrate_reset(req); |
1150 | reqsk_put(req); | |
1151 | return child; | |
1152 | } | |
1153 | ||
55d444b3 | 1154 | __NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPMIGRATEREQFAILURE); |
d4f2c86b KI |
1155 | reqsk_migrate_reset(nreq); |
1156 | __reqsk_free(nreq); | |
1157 | } else if (inet_csk_reqsk_queue_add(sk, req, child)) { | |
7716682c | 1158 | return child; |
d4f2c86b | 1159 | } |
5e0724d0 ED |
1160 | } |
1161 | /* Too bad, another child took ownership of the request, undo. */ | |
d4f2c86b | 1162 | child_put: |
5e0724d0 ED |
1163 | bh_unlock_sock(child); |
1164 | sock_put(child); | |
1165 | return NULL; | |
1166 | } | |
1167 | EXPORT_SYMBOL(inet_csk_complete_hashdance); | |
1168 | ||
a019d6fe ACM |
1169 | /* |
1170 | * This routine closes sockets which have been at least partially | |
1171 | * opened, but not yet accepted. | |
1172 | */ | |
1173 | void inet_csk_listen_stop(struct sock *sk) | |
1174 | { | |
1175 | struct inet_connection_sock *icsk = inet_csk(sk); | |
8336886f | 1176 | struct request_sock_queue *queue = &icsk->icsk_accept_queue; |
fff1f300 | 1177 | struct request_sock *next, *req; |
a019d6fe ACM |
1178 | |
1179 | /* Following specs, it would be better either to send FIN | |
1180 | * (and enter FIN-WAIT-1, it is normal close) | |
1181 | * or to send active reset (abort). | |
1182 | * Certainly, it is pretty dangerous while synflood, but it is | |
1183 | * bad justification for our negligence 8) | |
1184 | * To be honest, we are not able to make either | |
1185 | * of the variants now. --ANK | |
1186 | */ | |
fff1f300 | 1187 | while ((req = reqsk_queue_remove(queue, sk)) != NULL) { |
54b92e84 KI |
1188 | struct sock *child = req->sk, *nsk; |
1189 | struct request_sock *nreq; | |
a019d6fe | 1190 | |
a019d6fe ACM |
1191 | local_bh_disable(); |
1192 | bh_lock_sock(child); | |
547b792c | 1193 | WARN_ON(sock_owned_by_user(child)); |
a019d6fe ACM |
1194 | sock_hold(child); |
1195 | ||
54b92e84 KI |
1196 | nsk = reuseport_migrate_sock(sk, child, NULL); |
1197 | if (nsk) { | |
1198 | nreq = inet_reqsk_clone(req, nsk); | |
1199 | if (nreq) { | |
1200 | refcount_set(&nreq->rsk_refcnt, 1); | |
1201 | ||
1202 | if (inet_csk_reqsk_queue_add(nsk, nreq, child)) { | |
55d444b3 KI |
1203 | __NET_INC_STATS(sock_net(nsk), |
1204 | LINUX_MIB_TCPMIGRATEREQSUCCESS); | |
54b92e84 KI |
1205 | reqsk_migrate_reset(req); |
1206 | } else { | |
55d444b3 KI |
1207 | __NET_INC_STATS(sock_net(nsk), |
1208 | LINUX_MIB_TCPMIGRATEREQFAILURE); | |
54b92e84 KI |
1209 | reqsk_migrate_reset(nreq); |
1210 | __reqsk_free(nreq); | |
1211 | } | |
1212 | ||
1213 | /* inet_csk_reqsk_queue_add() has already | |
1214 | * called inet_child_forget() on failure case. | |
1215 | */ | |
1216 | goto skip_child_forget; | |
1217 | } | |
1218 | } | |
1219 | ||
ebb516af | 1220 | inet_child_forget(sk, req, child); |
54b92e84 | 1221 | skip_child_forget: |
da8ab578 | 1222 | reqsk_put(req); |
a019d6fe ACM |
1223 | bh_unlock_sock(child); |
1224 | local_bh_enable(); | |
1225 | sock_put(child); | |
1226 | ||
92d6f176 | 1227 | cond_resched(); |
a019d6fe | 1228 | } |
0536fcc0 | 1229 | if (queue->fastopenq.rskq_rst_head) { |
8336886f | 1230 | /* Free all the reqs queued in rskq_rst_head. */ |
0536fcc0 | 1231 | spin_lock_bh(&queue->fastopenq.lock); |
fff1f300 | 1232 | req = queue->fastopenq.rskq_rst_head; |
0536fcc0 ED |
1233 | queue->fastopenq.rskq_rst_head = NULL; |
1234 | spin_unlock_bh(&queue->fastopenq.lock); | |
fff1f300 ED |
1235 | while (req != NULL) { |
1236 | next = req->dl_next; | |
13854e5a | 1237 | reqsk_put(req); |
fff1f300 | 1238 | req = next; |
8336886f JC |
1239 | } |
1240 | } | |
ebb516af | 1241 | WARN_ON_ONCE(sk->sk_ack_backlog); |
a019d6fe | 1242 | } |
a019d6fe | 1243 | EXPORT_SYMBOL_GPL(inet_csk_listen_stop); |
af05dc93 ACM |
1244 | |
1245 | void inet_csk_addr2sockaddr(struct sock *sk, struct sockaddr *uaddr) | |
1246 | { | |
1247 | struct sockaddr_in *sin = (struct sockaddr_in *)uaddr; | |
1248 | const struct inet_sock *inet = inet_sk(sk); | |
1249 | ||
1250 | sin->sin_family = AF_INET; | |
c720c7e8 ED |
1251 | sin->sin_addr.s_addr = inet->inet_daddr; |
1252 | sin->sin_port = inet->inet_dport; | |
af05dc93 | 1253 | } |
af05dc93 | 1254 | EXPORT_SYMBOL_GPL(inet_csk_addr2sockaddr); |
c4d93909 | 1255 | |
80d0a69f DM |
1256 | static struct dst_entry *inet_csk_rebuild_route(struct sock *sk, struct flowi *fl) |
1257 | { | |
5abf7f7e ED |
1258 | const struct inet_sock *inet = inet_sk(sk); |
1259 | const struct ip_options_rcu *inet_opt; | |
80d0a69f DM |
1260 | __be32 daddr = inet->inet_daddr; |
1261 | struct flowi4 *fl4; | |
1262 | struct rtable *rt; | |
1263 | ||
1264 | rcu_read_lock(); | |
1265 | inet_opt = rcu_dereference(inet->inet_opt); | |
1266 | if (inet_opt && inet_opt->opt.srr) | |
1267 | daddr = inet_opt->opt.faddr; | |
1268 | fl4 = &fl->u.ip4; | |
1269 | rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, | |
1270 | inet->inet_saddr, inet->inet_dport, | |
1271 | inet->inet_sport, sk->sk_protocol, | |
1272 | RT_CONN_FLAGS(sk), sk->sk_bound_dev_if); | |
1273 | if (IS_ERR(rt)) | |
1274 | rt = NULL; | |
1275 | if (rt) | |
1276 | sk_setup_caps(sk, &rt->dst); | |
1277 | rcu_read_unlock(); | |
1278 | ||
1279 | return &rt->dst; | |
1280 | } | |
1281 | ||
1282 | struct dst_entry *inet_csk_update_pmtu(struct sock *sk, u32 mtu) | |
1283 | { | |
1284 | struct dst_entry *dst = __sk_dst_check(sk, 0); | |
1285 | struct inet_sock *inet = inet_sk(sk); | |
1286 | ||
1287 | if (!dst) { | |
1288 | dst = inet_csk_rebuild_route(sk, &inet->cork.fl); | |
1289 | if (!dst) | |
1290 | goto out; | |
1291 | } | |
bd085ef6 | 1292 | dst->ops->update_pmtu(dst, sk, NULL, mtu, true); |
80d0a69f DM |
1293 | |
1294 | dst = __sk_dst_check(sk, 0); | |
1295 | if (!dst) | |
1296 | dst = inet_csk_rebuild_route(sk, &inet->cork.fl); | |
1297 | out: | |
1298 | return dst; | |
1299 | } | |
1300 | EXPORT_SYMBOL_GPL(inet_csk_update_pmtu); |