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 | * PF_INET protocol family socket handler. |
8 | * |
9 | * Authors: Ross Biro |
10 | * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> |
11 | * Florian La Roche, <flla@stud.uni-sb.de> |
12 | * Alan Cox, <A.Cox@swansea.ac.uk> |
13 | * |
14 | * Changes (see also sock.c) |
15 | * |
16 | * piggy, |
17 | * Karl Knutson : Socket protocol table |
18 | * A.N.Kuznetsov : Socket death error in accept(). |
19 | * John Richardson : Fix non blocking error in connect() |
20 | * so sockets that fail to connect |
21 | * don't return -EINPROGRESS. |
22 | * Alan Cox : Asynchronous I/O support |
23 | * Alan Cox : Keep correct socket pointer on sock |
24 | * structures |
25 | * when accept() ed |
26 | * Alan Cox : Semantics of SO_LINGER aren't state |
27 | * moved to close when you look carefully. |
28 | * With this fixed and the accept bug fixed |
29 | * some RPC stuff seems happier. |
30 | * Niibe Yutaka : 4.4BSD style write async I/O |
31 | * Alan Cox, |
32 | * Tony Gale : Fixed reuse semantics. |
33 | * Alan Cox : bind() shouldn't abort existing but dead |
34 | * sockets. Stops FTP netin:.. I hope. |
35 | * Alan Cox : bind() works correctly for RAW sockets. |
36 | * Note that FreeBSD at least was broken |
37 | * in this respect so be careful with |
38 | * compatibility tests... |
39 | * Alan Cox : routing cache support |
40 | * Alan Cox : memzero the socket structure for |
41 | * compactness. |
42 | * Matt Day : nonblock connect error handler |
43 | * Alan Cox : Allow large numbers of pending sockets |
44 | * (eg for big web sites), but only if |
45 | * specifically application requested. |
46 | * Alan Cox : New buffering throughout IP. Used |
47 | * dumbly. |
48 | * Alan Cox : New buffering now used smartly. |
49 | * Alan Cox : BSD rather than common sense |
50 | * interpretation of listen. |
51 | * Germano Caronni : Assorted small races. |
52 | * Alan Cox : sendmsg/recvmsg basic support. |
53 | * Alan Cox : Only sendmsg/recvmsg now supported. |
54 | * Alan Cox : Locked down bind (see security list). |
55 | * Alan Cox : Loosened bind a little. |
56 | * Mike McLagan : ADD/DEL DLCI Ioctls |
57 | * Willy Konynenberg : Transparent proxying support. |
58 | * David S. Miller : New socket lookup architecture. |
59 | * Some other random speedups. |
60 | * Cyrus Durgin : Cleaned up file for kmod hacks. |
61 | * Andi Kleen : Fix inet_stream_connect TCP race. |
62 | */ |
63 | |
64 | #define pr_fmt(fmt) "IPv4: " fmt |
65 | |
66 | #include <linux/err.h> |
67 | #include <linux/errno.h> |
68 | #include <linux/types.h> |
69 | #include <linux/socket.h> |
70 | #include <linux/in.h> |
71 | #include <linux/kernel.h> |
72 | #include <linux/kmod.h> |
73 | #include <linux/sched.h> |
74 | #include <linux/timer.h> |
75 | #include <linux/string.h> |
76 | #include <linux/sockios.h> |
77 | #include <linux/net.h> |
78 | #include <linux/capability.h> |
79 | #include <linux/fcntl.h> |
80 | #include <linux/mm.h> |
81 | #include <linux/interrupt.h> |
82 | #include <linux/stat.h> |
83 | #include <linux/init.h> |
84 | #include <linux/poll.h> |
85 | #include <linux/netfilter_ipv4.h> |
86 | #include <linux/random.h> |
87 | #include <linux/slab.h> |
88 | |
89 | #include <linux/uaccess.h> |
90 | |
91 | #include <linux/inet.h> |
92 | #include <linux/igmp.h> |
93 | #include <linux/inetdevice.h> |
94 | #include <linux/netdevice.h> |
95 | #include <net/checksum.h> |
96 | #include <net/ip.h> |
97 | #include <net/protocol.h> |
98 | #include <net/arp.h> |
99 | #include <net/route.h> |
100 | #include <net/ip_fib.h> |
101 | #include <net/inet_connection_sock.h> |
102 | #include <net/gro.h> |
103 | #include <net/gso.h> |
104 | #include <net/tcp.h> |
105 | #include <net/udp.h> |
106 | #include <net/udplite.h> |
107 | #include <net/ping.h> |
108 | #include <linux/skbuff.h> |
109 | #include <net/sock.h> |
110 | #include <net/raw.h> |
111 | #include <net/icmp.h> |
112 | #include <net/inet_common.h> |
113 | #include <net/ip_tunnels.h> |
114 | #include <net/xfrm.h> |
115 | #include <net/net_namespace.h> |
116 | #include <net/secure_seq.h> |
117 | #ifdef CONFIG_IP_MROUTE |
118 | #include <linux/mroute.h> |
119 | #endif |
120 | #include <net/l3mdev.h> |
121 | #include <net/compat.h> |
122 | #include <net/rps.h> |
123 | |
124 | #include <trace/events/sock.h> |
125 | |
126 | /* The inetsw table contains everything that inet_create needs to |
127 | * build a new socket. |
128 | */ |
129 | static struct list_head inetsw[SOCK_MAX]; |
130 | static DEFINE_SPINLOCK(inetsw_lock); |
131 | |
132 | /* New destruction routine */ |
133 | |
134 | void inet_sock_destruct(struct sock *sk) |
135 | { |
136 | struct inet_sock *inet = inet_sk(sk); |
137 | |
138 | __skb_queue_purge(list: &sk->sk_receive_queue); |
139 | __skb_queue_purge(list: &sk->sk_error_queue); |
140 | |
141 | sk_mem_reclaim_final(sk); |
142 | |
143 | if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) { |
144 | pr_err("Attempt to release TCP socket in state %d %p\n" , |
145 | sk->sk_state, sk); |
146 | return; |
147 | } |
148 | if (!sock_flag(sk, flag: SOCK_DEAD)) { |
149 | pr_err("Attempt to release alive inet socket %p\n" , sk); |
150 | return; |
151 | } |
152 | |
153 | WARN_ON_ONCE(atomic_read(&sk->sk_rmem_alloc)); |
154 | WARN_ON_ONCE(refcount_read(&sk->sk_wmem_alloc)); |
155 | WARN_ON_ONCE(sk->sk_wmem_queued); |
156 | WARN_ON_ONCE(sk_forward_alloc_get(sk)); |
157 | |
158 | kfree(rcu_dereference_protected(inet->inet_opt, 1)); |
159 | dst_release(rcu_dereference_protected(sk->sk_dst_cache, 1)); |
160 | dst_release(rcu_dereference_protected(sk->sk_rx_dst, 1)); |
161 | } |
162 | EXPORT_SYMBOL(inet_sock_destruct); |
163 | |
164 | /* |
165 | * The routines beyond this point handle the behaviour of an AF_INET |
166 | * socket object. Mostly it punts to the subprotocols of IP to do |
167 | * the work. |
168 | */ |
169 | |
170 | /* |
171 | * Automatically bind an unbound socket. |
172 | */ |
173 | |
174 | static int inet_autobind(struct sock *sk) |
175 | { |
176 | struct inet_sock *inet; |
177 | /* We may need to bind the socket. */ |
178 | lock_sock(sk); |
179 | inet = inet_sk(sk); |
180 | if (!inet->inet_num) { |
181 | if (sk->sk_prot->get_port(sk, 0)) { |
182 | release_sock(sk); |
183 | return -EAGAIN; |
184 | } |
185 | inet->inet_sport = htons(inet->inet_num); |
186 | } |
187 | release_sock(sk); |
188 | return 0; |
189 | } |
190 | |
191 | int __inet_listen_sk(struct sock *sk, int backlog) |
192 | { |
193 | unsigned char old_state = sk->sk_state; |
194 | int err, tcp_fastopen; |
195 | |
196 | if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN))) |
197 | return -EINVAL; |
198 | |
199 | WRITE_ONCE(sk->sk_max_ack_backlog, backlog); |
200 | /* Really, if the socket is already in listen state |
201 | * we can only allow the backlog to be adjusted. |
202 | */ |
203 | if (old_state != TCP_LISTEN) { |
204 | /* Enable TFO w/o requiring TCP_FASTOPEN socket option. |
205 | * Note that only TCP sockets (SOCK_STREAM) will reach here. |
206 | * Also fastopen backlog may already been set via the option |
207 | * because the socket was in TCP_LISTEN state previously but |
208 | * was shutdown() rather than close(). |
209 | */ |
210 | tcp_fastopen = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fastopen); |
211 | if ((tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) && |
212 | (tcp_fastopen & TFO_SERVER_ENABLE) && |
213 | !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) { |
214 | fastopen_queue_tune(sk, backlog); |
215 | tcp_fastopen_init_key_once(net: sock_net(sk)); |
216 | } |
217 | |
218 | err = inet_csk_listen_start(sk); |
219 | if (err) |
220 | return err; |
221 | |
222 | tcp_call_bpf(sk, op: BPF_SOCK_OPS_TCP_LISTEN_CB, nargs: 0, NULL); |
223 | } |
224 | return 0; |
225 | } |
226 | |
227 | /* |
228 | * Move a socket into listening state. |
229 | */ |
230 | int inet_listen(struct socket *sock, int backlog) |
231 | { |
232 | struct sock *sk = sock->sk; |
233 | int err = -EINVAL; |
234 | |
235 | lock_sock(sk); |
236 | |
237 | if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM) |
238 | goto out; |
239 | |
240 | err = __inet_listen_sk(sk, backlog); |
241 | |
242 | out: |
243 | release_sock(sk); |
244 | return err; |
245 | } |
246 | EXPORT_SYMBOL(inet_listen); |
247 | |
248 | /* |
249 | * Create an inet socket. |
250 | */ |
251 | |
252 | static int inet_create(struct net *net, struct socket *sock, int protocol, |
253 | int kern) |
254 | { |
255 | struct sock *sk; |
256 | struct inet_protosw *answer; |
257 | struct inet_sock *inet; |
258 | struct proto *answer_prot; |
259 | unsigned char answer_flags; |
260 | int try_loading_module = 0; |
261 | int err; |
262 | |
263 | if (protocol < 0 || protocol >= IPPROTO_MAX) |
264 | return -EINVAL; |
265 | |
266 | sock->state = SS_UNCONNECTED; |
267 | |
268 | /* Look for the requested type/protocol pair. */ |
269 | lookup_protocol: |
270 | err = -ESOCKTNOSUPPORT; |
271 | rcu_read_lock(); |
272 | list_for_each_entry_rcu(answer, &inetsw[sock->type], list) { |
273 | |
274 | err = 0; |
275 | /* Check the non-wild match. */ |
276 | if (protocol == answer->protocol) { |
277 | if (protocol != IPPROTO_IP) |
278 | break; |
279 | } else { |
280 | /* Check for the two wild cases. */ |
281 | if (IPPROTO_IP == protocol) { |
282 | protocol = answer->protocol; |
283 | break; |
284 | } |
285 | if (IPPROTO_IP == answer->protocol) |
286 | break; |
287 | } |
288 | err = -EPROTONOSUPPORT; |
289 | } |
290 | |
291 | if (unlikely(err)) { |
292 | if (try_loading_module < 2) { |
293 | rcu_read_unlock(); |
294 | /* |
295 | * Be more specific, e.g. net-pf-2-proto-132-type-1 |
296 | * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM) |
297 | */ |
298 | if (++try_loading_module == 1) |
299 | request_module("net-pf-%d-proto-%d-type-%d" , |
300 | PF_INET, protocol, sock->type); |
301 | /* |
302 | * Fall back to generic, e.g. net-pf-2-proto-132 |
303 | * (net-pf-PF_INET-proto-IPPROTO_SCTP) |
304 | */ |
305 | else |
306 | request_module("net-pf-%d-proto-%d" , |
307 | PF_INET, protocol); |
308 | goto lookup_protocol; |
309 | } else |
310 | goto out_rcu_unlock; |
311 | } |
312 | |
313 | err = -EPERM; |
314 | if (sock->type == SOCK_RAW && !kern && |
315 | !ns_capable(ns: net->user_ns, CAP_NET_RAW)) |
316 | goto out_rcu_unlock; |
317 | |
318 | sock->ops = answer->ops; |
319 | answer_prot = answer->prot; |
320 | answer_flags = answer->flags; |
321 | rcu_read_unlock(); |
322 | |
323 | WARN_ON(!answer_prot->slab); |
324 | |
325 | err = -ENOMEM; |
326 | sk = sk_alloc(net, PF_INET, GFP_KERNEL, prot: answer_prot, kern); |
327 | if (!sk) |
328 | goto out; |
329 | |
330 | err = 0; |
331 | if (INET_PROTOSW_REUSE & answer_flags) |
332 | sk->sk_reuse = SK_CAN_REUSE; |
333 | |
334 | if (INET_PROTOSW_ICSK & answer_flags) |
335 | inet_init_csk_locks(sk); |
336 | |
337 | inet = inet_sk(sk); |
338 | inet_assign_bit(IS_ICSK, sk, INET_PROTOSW_ICSK & answer_flags); |
339 | |
340 | inet_clear_bit(NODEFRAG, sk); |
341 | |
342 | if (SOCK_RAW == sock->type) { |
343 | inet->inet_num = protocol; |
344 | if (IPPROTO_RAW == protocol) |
345 | inet_set_bit(HDRINCL, sk); |
346 | } |
347 | |
348 | if (READ_ONCE(net->ipv4.sysctl_ip_no_pmtu_disc)) |
349 | inet->pmtudisc = IP_PMTUDISC_DONT; |
350 | else |
351 | inet->pmtudisc = IP_PMTUDISC_WANT; |
352 | |
353 | atomic_set(v: &inet->inet_id, i: 0); |
354 | |
355 | sock_init_data(sock, sk); |
356 | |
357 | sk->sk_destruct = inet_sock_destruct; |
358 | sk->sk_protocol = protocol; |
359 | sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv; |
360 | sk->sk_txrehash = READ_ONCE(net->core.sysctl_txrehash); |
361 | |
362 | inet->uc_ttl = -1; |
363 | inet_set_bit(MC_LOOP, sk); |
364 | inet->mc_ttl = 1; |
365 | inet_set_bit(MC_ALL, sk); |
366 | inet->mc_index = 0; |
367 | inet->mc_list = NULL; |
368 | inet->rcv_tos = 0; |
369 | |
370 | if (inet->inet_num) { |
371 | /* It assumes that any protocol which allows |
372 | * the user to assign a number at socket |
373 | * creation time automatically |
374 | * shares. |
375 | */ |
376 | inet->inet_sport = htons(inet->inet_num); |
377 | /* Add to protocol hash chains. */ |
378 | err = sk->sk_prot->hash(sk); |
379 | if (err) { |
380 | sk_common_release(sk); |
381 | goto out; |
382 | } |
383 | } |
384 | |
385 | if (sk->sk_prot->init) { |
386 | err = sk->sk_prot->init(sk); |
387 | if (err) { |
388 | sk_common_release(sk); |
389 | goto out; |
390 | } |
391 | } |
392 | |
393 | if (!kern) { |
394 | err = BPF_CGROUP_RUN_PROG_INET_SOCK(sk); |
395 | if (err) { |
396 | sk_common_release(sk); |
397 | goto out; |
398 | } |
399 | } |
400 | out: |
401 | return err; |
402 | out_rcu_unlock: |
403 | rcu_read_unlock(); |
404 | goto out; |
405 | } |
406 | |
407 | |
408 | /* |
409 | * The peer socket should always be NULL (or else). When we call this |
410 | * function we are destroying the object and from then on nobody |
411 | * should refer to it. |
412 | */ |
413 | int inet_release(struct socket *sock) |
414 | { |
415 | struct sock *sk = sock->sk; |
416 | |
417 | if (sk) { |
418 | long timeout; |
419 | |
420 | if (!sk->sk_kern_sock) |
421 | BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE(sk); |
422 | |
423 | /* Applications forget to leave groups before exiting */ |
424 | ip_mc_drop_socket(sk); |
425 | |
426 | /* If linger is set, we don't return until the close |
427 | * is complete. Otherwise we return immediately. The |
428 | * actually closing is done the same either way. |
429 | * |
430 | * If the close is due to the process exiting, we never |
431 | * linger.. |
432 | */ |
433 | timeout = 0; |
434 | if (sock_flag(sk, flag: SOCK_LINGER) && |
435 | !(current->flags & PF_EXITING)) |
436 | timeout = sk->sk_lingertime; |
437 | sk->sk_prot->close(sk, timeout); |
438 | sock->sk = NULL; |
439 | } |
440 | return 0; |
441 | } |
442 | EXPORT_SYMBOL(inet_release); |
443 | |
444 | int inet_bind_sk(struct sock *sk, struct sockaddr *uaddr, int addr_len) |
445 | { |
446 | u32 flags = BIND_WITH_LOCK; |
447 | int err; |
448 | |
449 | /* If the socket has its own bind function then use it. (RAW) */ |
450 | if (sk->sk_prot->bind) { |
451 | return sk->sk_prot->bind(sk, uaddr, addr_len); |
452 | } |
453 | if (addr_len < sizeof(struct sockaddr_in)) |
454 | return -EINVAL; |
455 | |
456 | /* BPF prog is run before any checks are done so that if the prog |
457 | * changes context in a wrong way it will be caught. |
458 | */ |
459 | err = BPF_CGROUP_RUN_PROG_INET_BIND_LOCK(sk, uaddr, &addr_len, |
460 | CGROUP_INET4_BIND, &flags); |
461 | if (err) |
462 | return err; |
463 | |
464 | return __inet_bind(sk, uaddr, addr_len, flags); |
465 | } |
466 | |
467 | int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len) |
468 | { |
469 | return inet_bind_sk(sk: sock->sk, uaddr, addr_len); |
470 | } |
471 | EXPORT_SYMBOL(inet_bind); |
472 | |
473 | int __inet_bind(struct sock *sk, struct sockaddr *uaddr, int addr_len, |
474 | u32 flags) |
475 | { |
476 | struct sockaddr_in *addr = (struct sockaddr_in *)uaddr; |
477 | struct inet_sock *inet = inet_sk(sk); |
478 | struct net *net = sock_net(sk); |
479 | unsigned short snum; |
480 | int chk_addr_ret; |
481 | u32 tb_id = RT_TABLE_LOCAL; |
482 | int err; |
483 | |
484 | if (addr->sin_family != AF_INET) { |
485 | /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET) |
486 | * only if s_addr is INADDR_ANY. |
487 | */ |
488 | err = -EAFNOSUPPORT; |
489 | if (addr->sin_family != AF_UNSPEC || |
490 | addr->sin_addr.s_addr != htonl(INADDR_ANY)) |
491 | goto out; |
492 | } |
493 | |
494 | tb_id = l3mdev_fib_table_by_index(net, ifindex: sk->sk_bound_dev_if) ? : tb_id; |
495 | chk_addr_ret = inet_addr_type_table(net, addr: addr->sin_addr.s_addr, tb_id); |
496 | |
497 | /* Not specified by any standard per-se, however it breaks too |
498 | * many applications when removed. It is unfortunate since |
499 | * allowing applications to make a non-local bind solves |
500 | * several problems with systems using dynamic addressing. |
501 | * (ie. your servers still start up even if your ISDN link |
502 | * is temporarily down) |
503 | */ |
504 | err = -EADDRNOTAVAIL; |
505 | if (!inet_addr_valid_or_nonlocal(net, inet, addr: addr->sin_addr.s_addr, |
506 | addr_type: chk_addr_ret)) |
507 | goto out; |
508 | |
509 | snum = ntohs(addr->sin_port); |
510 | err = -EACCES; |
511 | if (!(flags & BIND_NO_CAP_NET_BIND_SERVICE) && |
512 | snum && inet_port_requires_bind_service(net, port: snum) && |
513 | !ns_capable(ns: net->user_ns, CAP_NET_BIND_SERVICE)) |
514 | goto out; |
515 | |
516 | /* We keep a pair of addresses. rcv_saddr is the one |
517 | * used by hash lookups, and saddr is used for transmit. |
518 | * |
519 | * In the BSD API these are the same except where it |
520 | * would be illegal to use them (multicast/broadcast) in |
521 | * which case the sending device address is used. |
522 | */ |
523 | if (flags & BIND_WITH_LOCK) |
524 | lock_sock(sk); |
525 | |
526 | /* Check these errors (active socket, double bind). */ |
527 | err = -EINVAL; |
528 | if (sk->sk_state != TCP_CLOSE || inet->inet_num) |
529 | goto out_release_sock; |
530 | |
531 | inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr; |
532 | if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST) |
533 | inet->inet_saddr = 0; /* Use device */ |
534 | |
535 | /* Make sure we are allowed to bind here. */ |
536 | if (snum || !(inet_test_bit(BIND_ADDRESS_NO_PORT, sk) || |
537 | (flags & BIND_FORCE_ADDRESS_NO_PORT))) { |
538 | err = sk->sk_prot->get_port(sk, snum); |
539 | if (err) { |
540 | inet->inet_saddr = inet->inet_rcv_saddr = 0; |
541 | goto out_release_sock; |
542 | } |
543 | if (!(flags & BIND_FROM_BPF)) { |
544 | err = BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk); |
545 | if (err) { |
546 | inet->inet_saddr = inet->inet_rcv_saddr = 0; |
547 | if (sk->sk_prot->put_port) |
548 | sk->sk_prot->put_port(sk); |
549 | goto out_release_sock; |
550 | } |
551 | } |
552 | } |
553 | |
554 | if (inet->inet_rcv_saddr) |
555 | sk->sk_userlocks |= SOCK_BINDADDR_LOCK; |
556 | if (snum) |
557 | sk->sk_userlocks |= SOCK_BINDPORT_LOCK; |
558 | inet->inet_sport = htons(inet->inet_num); |
559 | inet->inet_daddr = 0; |
560 | inet->inet_dport = 0; |
561 | sk_dst_reset(sk); |
562 | err = 0; |
563 | out_release_sock: |
564 | if (flags & BIND_WITH_LOCK) |
565 | release_sock(sk); |
566 | out: |
567 | return err; |
568 | } |
569 | |
570 | int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr, |
571 | int addr_len, int flags) |
572 | { |
573 | struct sock *sk = sock->sk; |
574 | const struct proto *prot; |
575 | int err; |
576 | |
577 | if (addr_len < sizeof(uaddr->sa_family)) |
578 | return -EINVAL; |
579 | |
580 | /* IPV6_ADDRFORM can change sk->sk_prot under us. */ |
581 | prot = READ_ONCE(sk->sk_prot); |
582 | |
583 | if (uaddr->sa_family == AF_UNSPEC) |
584 | return prot->disconnect(sk, flags); |
585 | |
586 | if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) { |
587 | err = prot->pre_connect(sk, uaddr, addr_len); |
588 | if (err) |
589 | return err; |
590 | } |
591 | |
592 | if (data_race(!inet_sk(sk)->inet_num) && inet_autobind(sk)) |
593 | return -EAGAIN; |
594 | return prot->connect(sk, uaddr, addr_len); |
595 | } |
596 | EXPORT_SYMBOL(inet_dgram_connect); |
597 | |
598 | static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias) |
599 | { |
600 | DEFINE_WAIT_FUNC(wait, woken_wake_function); |
601 | |
602 | add_wait_queue(wq_head: sk_sleep(sk), wq_entry: &wait); |
603 | sk->sk_write_pending += writebias; |
604 | |
605 | /* Basic assumption: if someone sets sk->sk_err, he _must_ |
606 | * change state of the socket from TCP_SYN_*. |
607 | * Connect() does not allow to get error notifications |
608 | * without closing the socket. |
609 | */ |
610 | while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { |
611 | release_sock(sk); |
612 | timeo = wait_woken(wq_entry: &wait, TASK_INTERRUPTIBLE, timeout: timeo); |
613 | lock_sock(sk); |
614 | if (signal_pending(current) || !timeo) |
615 | break; |
616 | } |
617 | remove_wait_queue(wq_head: sk_sleep(sk), wq_entry: &wait); |
618 | sk->sk_write_pending -= writebias; |
619 | return timeo; |
620 | } |
621 | |
622 | /* |
623 | * Connect to a remote host. There is regrettably still a little |
624 | * TCP 'magic' in here. |
625 | */ |
626 | int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr, |
627 | int addr_len, int flags, int is_sendmsg) |
628 | { |
629 | struct sock *sk = sock->sk; |
630 | int err; |
631 | long timeo; |
632 | |
633 | /* |
634 | * uaddr can be NULL and addr_len can be 0 if: |
635 | * sk is a TCP fastopen active socket and |
636 | * TCP_FASTOPEN_CONNECT sockopt is set and |
637 | * we already have a valid cookie for this socket. |
638 | * In this case, user can call write() after connect(). |
639 | * write() will invoke tcp_sendmsg_fastopen() which calls |
640 | * __inet_stream_connect(). |
641 | */ |
642 | if (uaddr) { |
643 | if (addr_len < sizeof(uaddr->sa_family)) |
644 | return -EINVAL; |
645 | |
646 | if (uaddr->sa_family == AF_UNSPEC) { |
647 | sk->sk_disconnects++; |
648 | err = sk->sk_prot->disconnect(sk, flags); |
649 | sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED; |
650 | goto out; |
651 | } |
652 | } |
653 | |
654 | switch (sock->state) { |
655 | default: |
656 | err = -EINVAL; |
657 | goto out; |
658 | case SS_CONNECTED: |
659 | err = -EISCONN; |
660 | goto out; |
661 | case SS_CONNECTING: |
662 | if (inet_test_bit(DEFER_CONNECT, sk)) |
663 | err = is_sendmsg ? -EINPROGRESS : -EISCONN; |
664 | else |
665 | err = -EALREADY; |
666 | /* Fall out of switch with err, set for this state */ |
667 | break; |
668 | case SS_UNCONNECTED: |
669 | err = -EISCONN; |
670 | if (sk->sk_state != TCP_CLOSE) |
671 | goto out; |
672 | |
673 | if (BPF_CGROUP_PRE_CONNECT_ENABLED(sk)) { |
674 | err = sk->sk_prot->pre_connect(sk, uaddr, addr_len); |
675 | if (err) |
676 | goto out; |
677 | } |
678 | |
679 | err = sk->sk_prot->connect(sk, uaddr, addr_len); |
680 | if (err < 0) |
681 | goto out; |
682 | |
683 | sock->state = SS_CONNECTING; |
684 | |
685 | if (!err && inet_test_bit(DEFER_CONNECT, sk)) |
686 | goto out; |
687 | |
688 | /* Just entered SS_CONNECTING state; the only |
689 | * difference is that return value in non-blocking |
690 | * case is EINPROGRESS, rather than EALREADY. |
691 | */ |
692 | err = -EINPROGRESS; |
693 | break; |
694 | } |
695 | |
696 | timeo = sock_sndtimeo(sk, noblock: flags & O_NONBLOCK); |
697 | |
698 | if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { |
699 | int writebias = (sk->sk_protocol == IPPROTO_TCP) && |
700 | tcp_sk(sk)->fastopen_req && |
701 | tcp_sk(sk)->fastopen_req->data ? 1 : 0; |
702 | int dis = sk->sk_disconnects; |
703 | |
704 | /* Error code is set above */ |
705 | if (!timeo || !inet_wait_for_connect(sk, timeo, writebias)) |
706 | goto out; |
707 | |
708 | err = sock_intr_errno(timeo); |
709 | if (signal_pending(current)) |
710 | goto out; |
711 | |
712 | if (dis != sk->sk_disconnects) { |
713 | err = -EPIPE; |
714 | goto out; |
715 | } |
716 | } |
717 | |
718 | /* Connection was closed by RST, timeout, ICMP error |
719 | * or another process disconnected us. |
720 | */ |
721 | if (sk->sk_state == TCP_CLOSE) |
722 | goto sock_error; |
723 | |
724 | /* sk->sk_err may be not zero now, if RECVERR was ordered by user |
725 | * and error was received after socket entered established state. |
726 | * Hence, it is handled normally after connect() return successfully. |
727 | */ |
728 | |
729 | sock->state = SS_CONNECTED; |
730 | err = 0; |
731 | out: |
732 | return err; |
733 | |
734 | sock_error: |
735 | err = sock_error(sk) ? : -ECONNABORTED; |
736 | sock->state = SS_UNCONNECTED; |
737 | sk->sk_disconnects++; |
738 | if (sk->sk_prot->disconnect(sk, flags)) |
739 | sock->state = SS_DISCONNECTING; |
740 | goto out; |
741 | } |
742 | EXPORT_SYMBOL(__inet_stream_connect); |
743 | |
744 | int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr, |
745 | int addr_len, int flags) |
746 | { |
747 | int err; |
748 | |
749 | lock_sock(sk: sock->sk); |
750 | err = __inet_stream_connect(sock, uaddr, addr_len, flags, 0); |
751 | release_sock(sk: sock->sk); |
752 | return err; |
753 | } |
754 | EXPORT_SYMBOL(inet_stream_connect); |
755 | |
756 | void __inet_accept(struct socket *sock, struct socket *newsock, struct sock *newsk) |
757 | { |
758 | sock_rps_record_flow(sk: newsk); |
759 | WARN_ON(!((1 << newsk->sk_state) & |
760 | (TCPF_ESTABLISHED | TCPF_SYN_RECV | |
761 | TCPF_CLOSE_WAIT | TCPF_CLOSE))); |
762 | |
763 | if (test_bit(SOCK_SUPPORT_ZC, &sock->flags)) |
764 | set_bit(SOCK_SUPPORT_ZC, addr: &newsock->flags); |
765 | sock_graft(sk: newsk, parent: newsock); |
766 | |
767 | newsock->state = SS_CONNECTED; |
768 | } |
769 | |
770 | /* |
771 | * Accept a pending connection. The TCP layer now gives BSD semantics. |
772 | */ |
773 | |
774 | int inet_accept(struct socket *sock, struct socket *newsock, int flags, |
775 | bool kern) |
776 | { |
777 | struct sock *sk1 = sock->sk, *sk2; |
778 | int err = -EINVAL; |
779 | |
780 | /* IPV6_ADDRFORM can change sk->sk_prot under us. */ |
781 | sk2 = READ_ONCE(sk1->sk_prot)->accept(sk1, flags, &err, kern); |
782 | if (!sk2) |
783 | return err; |
784 | |
785 | lock_sock(sk: sk2); |
786 | __inet_accept(sock, newsock, newsk: sk2); |
787 | release_sock(sk: sk2); |
788 | return 0; |
789 | } |
790 | EXPORT_SYMBOL(inet_accept); |
791 | |
792 | /* |
793 | * This does both peername and sockname. |
794 | */ |
795 | int inet_getname(struct socket *sock, struct sockaddr *uaddr, |
796 | int peer) |
797 | { |
798 | struct sock *sk = sock->sk; |
799 | struct inet_sock *inet = inet_sk(sk); |
800 | DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr); |
801 | int sin_addr_len = sizeof(*sin); |
802 | |
803 | sin->sin_family = AF_INET; |
804 | lock_sock(sk); |
805 | if (peer) { |
806 | if (!inet->inet_dport || |
807 | (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) && |
808 | peer == 1)) { |
809 | release_sock(sk); |
810 | return -ENOTCONN; |
811 | } |
812 | sin->sin_port = inet->inet_dport; |
813 | sin->sin_addr.s_addr = inet->inet_daddr; |
814 | BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin, &sin_addr_len, |
815 | CGROUP_INET4_GETPEERNAME); |
816 | } else { |
817 | __be32 addr = inet->inet_rcv_saddr; |
818 | if (!addr) |
819 | addr = inet->inet_saddr; |
820 | sin->sin_port = inet->inet_sport; |
821 | sin->sin_addr.s_addr = addr; |
822 | BPF_CGROUP_RUN_SA_PROG(sk, (struct sockaddr *)sin, &sin_addr_len, |
823 | CGROUP_INET4_GETSOCKNAME); |
824 | } |
825 | release_sock(sk); |
826 | memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); |
827 | return sin_addr_len; |
828 | } |
829 | EXPORT_SYMBOL(inet_getname); |
830 | |
831 | int inet_send_prepare(struct sock *sk) |
832 | { |
833 | sock_rps_record_flow(sk); |
834 | |
835 | /* We may need to bind the socket. */ |
836 | if (data_race(!inet_sk(sk)->inet_num) && !sk->sk_prot->no_autobind && |
837 | inet_autobind(sk)) |
838 | return -EAGAIN; |
839 | |
840 | return 0; |
841 | } |
842 | EXPORT_SYMBOL_GPL(inet_send_prepare); |
843 | |
844 | int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size) |
845 | { |
846 | struct sock *sk = sock->sk; |
847 | |
848 | if (unlikely(inet_send_prepare(sk))) |
849 | return -EAGAIN; |
850 | |
851 | return INDIRECT_CALL_2(sk->sk_prot->sendmsg, tcp_sendmsg, udp_sendmsg, |
852 | sk, msg, size); |
853 | } |
854 | EXPORT_SYMBOL(inet_sendmsg); |
855 | |
856 | void inet_splice_eof(struct socket *sock) |
857 | { |
858 | const struct proto *prot; |
859 | struct sock *sk = sock->sk; |
860 | |
861 | if (unlikely(inet_send_prepare(sk))) |
862 | return; |
863 | |
864 | /* IPV6_ADDRFORM can change sk->sk_prot under us. */ |
865 | prot = READ_ONCE(sk->sk_prot); |
866 | if (prot->splice_eof) |
867 | prot->splice_eof(sock); |
868 | } |
869 | EXPORT_SYMBOL_GPL(inet_splice_eof); |
870 | |
871 | INDIRECT_CALLABLE_DECLARE(int udp_recvmsg(struct sock *, struct msghdr *, |
872 | size_t, int, int *)); |
873 | int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, |
874 | int flags) |
875 | { |
876 | struct sock *sk = sock->sk; |
877 | int addr_len = 0; |
878 | int err; |
879 | |
880 | if (likely(!(flags & MSG_ERRQUEUE))) |
881 | sock_rps_record_flow(sk); |
882 | |
883 | err = INDIRECT_CALL_2(sk->sk_prot->recvmsg, tcp_recvmsg, udp_recvmsg, |
884 | sk, msg, size, flags, &addr_len); |
885 | if (err >= 0) |
886 | msg->msg_namelen = addr_len; |
887 | return err; |
888 | } |
889 | EXPORT_SYMBOL(inet_recvmsg); |
890 | |
891 | int inet_shutdown(struct socket *sock, int how) |
892 | { |
893 | struct sock *sk = sock->sk; |
894 | int err = 0; |
895 | |
896 | /* This should really check to make sure |
897 | * the socket is a TCP socket. (WHY AC...) |
898 | */ |
899 | how++; /* maps 0->1 has the advantage of making bit 1 rcvs and |
900 | 1->2 bit 2 snds. |
901 | 2->3 */ |
902 | if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */ |
903 | return -EINVAL; |
904 | |
905 | lock_sock(sk); |
906 | if (sock->state == SS_CONNECTING) { |
907 | if ((1 << sk->sk_state) & |
908 | (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE)) |
909 | sock->state = SS_DISCONNECTING; |
910 | else |
911 | sock->state = SS_CONNECTED; |
912 | } |
913 | |
914 | switch (sk->sk_state) { |
915 | case TCP_CLOSE: |
916 | err = -ENOTCONN; |
917 | /* Hack to wake up other listeners, who can poll for |
918 | EPOLLHUP, even on eg. unconnected UDP sockets -- RR */ |
919 | fallthrough; |
920 | default: |
921 | WRITE_ONCE(sk->sk_shutdown, sk->sk_shutdown | how); |
922 | if (sk->sk_prot->shutdown) |
923 | sk->sk_prot->shutdown(sk, how); |
924 | break; |
925 | |
926 | /* Remaining two branches are temporary solution for missing |
927 | * close() in multithreaded environment. It is _not_ a good idea, |
928 | * but we have no choice until close() is repaired at VFS level. |
929 | */ |
930 | case TCP_LISTEN: |
931 | if (!(how & RCV_SHUTDOWN)) |
932 | break; |
933 | fallthrough; |
934 | case TCP_SYN_SENT: |
935 | err = sk->sk_prot->disconnect(sk, O_NONBLOCK); |
936 | sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED; |
937 | break; |
938 | } |
939 | |
940 | /* Wake up anyone sleeping in poll. */ |
941 | sk->sk_state_change(sk); |
942 | release_sock(sk); |
943 | return err; |
944 | } |
945 | EXPORT_SYMBOL(inet_shutdown); |
946 | |
947 | /* |
948 | * ioctl() calls you can issue on an INET socket. Most of these are |
949 | * device configuration and stuff and very rarely used. Some ioctls |
950 | * pass on to the socket itself. |
951 | * |
952 | * NOTE: I like the idea of a module for the config stuff. ie ifconfig |
953 | * loads the devconfigure module does its configuring and unloads it. |
954 | * There's a good 20K of config code hanging around the kernel. |
955 | */ |
956 | |
957 | int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) |
958 | { |
959 | struct sock *sk = sock->sk; |
960 | int err = 0; |
961 | struct net *net = sock_net(sk); |
962 | void __user *p = (void __user *)arg; |
963 | struct ifreq ifr; |
964 | struct rtentry rt; |
965 | |
966 | switch (cmd) { |
967 | case SIOCADDRT: |
968 | case SIOCDELRT: |
969 | if (copy_from_user(to: &rt, from: p, n: sizeof(struct rtentry))) |
970 | return -EFAULT; |
971 | err = ip_rt_ioctl(net, cmd, rt: &rt); |
972 | break; |
973 | case SIOCRTMSG: |
974 | err = -EINVAL; |
975 | break; |
976 | case SIOCDARP: |
977 | case SIOCGARP: |
978 | case SIOCSARP: |
979 | err = arp_ioctl(net, cmd, arg: (void __user *)arg); |
980 | break; |
981 | case SIOCGIFADDR: |
982 | case SIOCGIFBRDADDR: |
983 | case SIOCGIFNETMASK: |
984 | case SIOCGIFDSTADDR: |
985 | case SIOCGIFPFLAGS: |
986 | if (get_user_ifreq(ifr: &ifr, NULL, arg: p)) |
987 | return -EFAULT; |
988 | err = devinet_ioctl(net, cmd, &ifr); |
989 | if (!err && put_user_ifreq(ifr: &ifr, arg: p)) |
990 | err = -EFAULT; |
991 | break; |
992 | |
993 | case SIOCSIFADDR: |
994 | case SIOCSIFBRDADDR: |
995 | case SIOCSIFNETMASK: |
996 | case SIOCSIFDSTADDR: |
997 | case SIOCSIFPFLAGS: |
998 | case SIOCSIFFLAGS: |
999 | if (get_user_ifreq(ifr: &ifr, NULL, arg: p)) |
1000 | return -EFAULT; |
1001 | err = devinet_ioctl(net, cmd, &ifr); |
1002 | break; |
1003 | default: |
1004 | if (sk->sk_prot->ioctl) |
1005 | err = sk_ioctl(sk, cmd, arg: (void __user *)arg); |
1006 | else |
1007 | err = -ENOIOCTLCMD; |
1008 | break; |
1009 | } |
1010 | return err; |
1011 | } |
1012 | EXPORT_SYMBOL(inet_ioctl); |
1013 | |
1014 | #ifdef CONFIG_COMPAT |
1015 | static int inet_compat_routing_ioctl(struct sock *sk, unsigned int cmd, |
1016 | struct compat_rtentry __user *ur) |
1017 | { |
1018 | compat_uptr_t rtdev; |
1019 | struct rtentry rt; |
1020 | |
1021 | if (copy_from_user(to: &rt.rt_dst, from: &ur->rt_dst, |
1022 | n: 3 * sizeof(struct sockaddr)) || |
1023 | get_user(rt.rt_flags, &ur->rt_flags) || |
1024 | get_user(rt.rt_metric, &ur->rt_metric) || |
1025 | get_user(rt.rt_mtu, &ur->rt_mtu) || |
1026 | get_user(rt.rt_window, &ur->rt_window) || |
1027 | get_user(rt.rt_irtt, &ur->rt_irtt) || |
1028 | get_user(rtdev, &ur->rt_dev)) |
1029 | return -EFAULT; |
1030 | |
1031 | rt.rt_dev = compat_ptr(uptr: rtdev); |
1032 | return ip_rt_ioctl(sock_net(sk), cmd, rt: &rt); |
1033 | } |
1034 | |
1035 | static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) |
1036 | { |
1037 | void __user *argp = compat_ptr(uptr: arg); |
1038 | struct sock *sk = sock->sk; |
1039 | |
1040 | switch (cmd) { |
1041 | case SIOCADDRT: |
1042 | case SIOCDELRT: |
1043 | return inet_compat_routing_ioctl(sk, cmd, ur: argp); |
1044 | default: |
1045 | if (!sk->sk_prot->compat_ioctl) |
1046 | return -ENOIOCTLCMD; |
1047 | return sk->sk_prot->compat_ioctl(sk, cmd, arg); |
1048 | } |
1049 | } |
1050 | #endif /* CONFIG_COMPAT */ |
1051 | |
1052 | const struct proto_ops inet_stream_ops = { |
1053 | .family = PF_INET, |
1054 | .owner = THIS_MODULE, |
1055 | .release = inet_release, |
1056 | .bind = inet_bind, |
1057 | .connect = inet_stream_connect, |
1058 | .socketpair = sock_no_socketpair, |
1059 | .accept = inet_accept, |
1060 | .getname = inet_getname, |
1061 | .poll = tcp_poll, |
1062 | .ioctl = inet_ioctl, |
1063 | .gettstamp = sock_gettstamp, |
1064 | .listen = inet_listen, |
1065 | .shutdown = inet_shutdown, |
1066 | .setsockopt = sock_common_setsockopt, |
1067 | .getsockopt = sock_common_getsockopt, |
1068 | .sendmsg = inet_sendmsg, |
1069 | .recvmsg = inet_recvmsg, |
1070 | #ifdef CONFIG_MMU |
1071 | .mmap = tcp_mmap, |
1072 | #endif |
1073 | .splice_eof = inet_splice_eof, |
1074 | .splice_read = tcp_splice_read, |
1075 | .read_sock = tcp_read_sock, |
1076 | .read_skb = tcp_read_skb, |
1077 | .sendmsg_locked = tcp_sendmsg_locked, |
1078 | .peek_len = tcp_peek_len, |
1079 | #ifdef CONFIG_COMPAT |
1080 | .compat_ioctl = inet_compat_ioctl, |
1081 | #endif |
1082 | .set_rcvlowat = tcp_set_rcvlowat, |
1083 | }; |
1084 | EXPORT_SYMBOL(inet_stream_ops); |
1085 | |
1086 | const struct proto_ops inet_dgram_ops = { |
1087 | .family = PF_INET, |
1088 | .owner = THIS_MODULE, |
1089 | .release = inet_release, |
1090 | .bind = inet_bind, |
1091 | .connect = inet_dgram_connect, |
1092 | .socketpair = sock_no_socketpair, |
1093 | .accept = sock_no_accept, |
1094 | .getname = inet_getname, |
1095 | .poll = udp_poll, |
1096 | .ioctl = inet_ioctl, |
1097 | .gettstamp = sock_gettstamp, |
1098 | .listen = sock_no_listen, |
1099 | .shutdown = inet_shutdown, |
1100 | .setsockopt = sock_common_setsockopt, |
1101 | .getsockopt = sock_common_getsockopt, |
1102 | .sendmsg = inet_sendmsg, |
1103 | .read_skb = udp_read_skb, |
1104 | .recvmsg = inet_recvmsg, |
1105 | .mmap = sock_no_mmap, |
1106 | .splice_eof = inet_splice_eof, |
1107 | .set_peek_off = udp_set_peek_off, |
1108 | #ifdef CONFIG_COMPAT |
1109 | .compat_ioctl = inet_compat_ioctl, |
1110 | #endif |
1111 | }; |
1112 | EXPORT_SYMBOL(inet_dgram_ops); |
1113 | |
1114 | /* |
1115 | * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without |
1116 | * udp_poll |
1117 | */ |
1118 | static const struct proto_ops inet_sockraw_ops = { |
1119 | .family = PF_INET, |
1120 | .owner = THIS_MODULE, |
1121 | .release = inet_release, |
1122 | .bind = inet_bind, |
1123 | .connect = inet_dgram_connect, |
1124 | .socketpair = sock_no_socketpair, |
1125 | .accept = sock_no_accept, |
1126 | .getname = inet_getname, |
1127 | .poll = datagram_poll, |
1128 | .ioctl = inet_ioctl, |
1129 | .gettstamp = sock_gettstamp, |
1130 | .listen = sock_no_listen, |
1131 | .shutdown = inet_shutdown, |
1132 | .setsockopt = sock_common_setsockopt, |
1133 | .getsockopt = sock_common_getsockopt, |
1134 | .sendmsg = inet_sendmsg, |
1135 | .recvmsg = inet_recvmsg, |
1136 | .mmap = sock_no_mmap, |
1137 | .splice_eof = inet_splice_eof, |
1138 | #ifdef CONFIG_COMPAT |
1139 | .compat_ioctl = inet_compat_ioctl, |
1140 | #endif |
1141 | }; |
1142 | |
1143 | static const struct net_proto_family inet_family_ops = { |
1144 | .family = PF_INET, |
1145 | .create = inet_create, |
1146 | .owner = THIS_MODULE, |
1147 | }; |
1148 | |
1149 | /* Upon startup we insert all the elements in inetsw_array[] into |
1150 | * the linked list inetsw. |
1151 | */ |
1152 | static struct inet_protosw inetsw_array[] = |
1153 | { |
1154 | { |
1155 | .type = SOCK_STREAM, |
1156 | .protocol = IPPROTO_TCP, |
1157 | .prot = &tcp_prot, |
1158 | .ops = &inet_stream_ops, |
1159 | .flags = INET_PROTOSW_PERMANENT | |
1160 | INET_PROTOSW_ICSK, |
1161 | }, |
1162 | |
1163 | { |
1164 | .type = SOCK_DGRAM, |
1165 | .protocol = IPPROTO_UDP, |
1166 | .prot = &udp_prot, |
1167 | .ops = &inet_dgram_ops, |
1168 | .flags = INET_PROTOSW_PERMANENT, |
1169 | }, |
1170 | |
1171 | { |
1172 | .type = SOCK_DGRAM, |
1173 | .protocol = IPPROTO_ICMP, |
1174 | .prot = &ping_prot, |
1175 | .ops = &inet_sockraw_ops, |
1176 | .flags = INET_PROTOSW_REUSE, |
1177 | }, |
1178 | |
1179 | { |
1180 | .type = SOCK_RAW, |
1181 | .protocol = IPPROTO_IP, /* wild card */ |
1182 | .prot = &raw_prot, |
1183 | .ops = &inet_sockraw_ops, |
1184 | .flags = INET_PROTOSW_REUSE, |
1185 | } |
1186 | }; |
1187 | |
1188 | #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array) |
1189 | |
1190 | void inet_register_protosw(struct inet_protosw *p) |
1191 | { |
1192 | struct list_head *lh; |
1193 | struct inet_protosw *answer; |
1194 | int protocol = p->protocol; |
1195 | struct list_head *last_perm; |
1196 | |
1197 | spin_lock_bh(lock: &inetsw_lock); |
1198 | |
1199 | if (p->type >= SOCK_MAX) |
1200 | goto out_illegal; |
1201 | |
1202 | /* If we are trying to override a permanent protocol, bail. */ |
1203 | last_perm = &inetsw[p->type]; |
1204 | list_for_each(lh, &inetsw[p->type]) { |
1205 | answer = list_entry(lh, struct inet_protosw, list); |
1206 | /* Check only the non-wild match. */ |
1207 | if ((INET_PROTOSW_PERMANENT & answer->flags) == 0) |
1208 | break; |
1209 | if (protocol == answer->protocol) |
1210 | goto out_permanent; |
1211 | last_perm = lh; |
1212 | } |
1213 | |
1214 | /* Add the new entry after the last permanent entry if any, so that |
1215 | * the new entry does not override a permanent entry when matched with |
1216 | * a wild-card protocol. But it is allowed to override any existing |
1217 | * non-permanent entry. This means that when we remove this entry, the |
1218 | * system automatically returns to the old behavior. |
1219 | */ |
1220 | list_add_rcu(new: &p->list, head: last_perm); |
1221 | out: |
1222 | spin_unlock_bh(lock: &inetsw_lock); |
1223 | |
1224 | return; |
1225 | |
1226 | out_permanent: |
1227 | pr_err("Attempt to override permanent protocol %d\n" , protocol); |
1228 | goto out; |
1229 | |
1230 | out_illegal: |
1231 | pr_err("Ignoring attempt to register invalid socket type %d\n" , |
1232 | p->type); |
1233 | goto out; |
1234 | } |
1235 | EXPORT_SYMBOL(inet_register_protosw); |
1236 | |
1237 | void inet_unregister_protosw(struct inet_protosw *p) |
1238 | { |
1239 | if (INET_PROTOSW_PERMANENT & p->flags) { |
1240 | pr_err("Attempt to unregister permanent protocol %d\n" , |
1241 | p->protocol); |
1242 | } else { |
1243 | spin_lock_bh(lock: &inetsw_lock); |
1244 | list_del_rcu(entry: &p->list); |
1245 | spin_unlock_bh(lock: &inetsw_lock); |
1246 | |
1247 | synchronize_net(); |
1248 | } |
1249 | } |
1250 | EXPORT_SYMBOL(inet_unregister_protosw); |
1251 | |
1252 | static int inet_sk_reselect_saddr(struct sock *sk) |
1253 | { |
1254 | struct inet_sock *inet = inet_sk(sk); |
1255 | __be32 old_saddr = inet->inet_saddr; |
1256 | __be32 daddr = inet->inet_daddr; |
1257 | struct flowi4 *fl4; |
1258 | struct rtable *rt; |
1259 | __be32 new_saddr; |
1260 | struct ip_options_rcu *inet_opt; |
1261 | int err; |
1262 | |
1263 | inet_opt = rcu_dereference_protected(inet->inet_opt, |
1264 | lockdep_sock_is_held(sk)); |
1265 | if (inet_opt && inet_opt->opt.srr) |
1266 | daddr = inet_opt->opt.faddr; |
1267 | |
1268 | /* Query new route. */ |
1269 | fl4 = &inet->cork.fl.u.ip4; |
1270 | rt = ip_route_connect(fl4, dst: daddr, src: 0, oif: sk->sk_bound_dev_if, |
1271 | protocol: sk->sk_protocol, sport: inet->inet_sport, |
1272 | dport: inet->inet_dport, sk); |
1273 | if (IS_ERR(ptr: rt)) |
1274 | return PTR_ERR(ptr: rt); |
1275 | |
1276 | new_saddr = fl4->saddr; |
1277 | |
1278 | if (new_saddr == old_saddr) { |
1279 | sk_setup_caps(sk, dst: &rt->dst); |
1280 | return 0; |
1281 | } |
1282 | |
1283 | err = inet_bhash2_update_saddr(sk, saddr: &new_saddr, AF_INET); |
1284 | if (err) { |
1285 | ip_rt_put(rt); |
1286 | return err; |
1287 | } |
1288 | |
1289 | sk_setup_caps(sk, dst: &rt->dst); |
1290 | |
1291 | if (READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) > 1) { |
1292 | pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n" , |
1293 | __func__, &old_saddr, &new_saddr); |
1294 | } |
1295 | |
1296 | /* |
1297 | * XXX The only one ugly spot where we need to |
1298 | * XXX really change the sockets identity after |
1299 | * XXX it has entered the hashes. -DaveM |
1300 | * |
1301 | * Besides that, it does not check for connection |
1302 | * uniqueness. Wait for troubles. |
1303 | */ |
1304 | return __sk_prot_rehash(sk); |
1305 | } |
1306 | |
1307 | int (struct sock *sk) |
1308 | { |
1309 | struct inet_sock *inet = inet_sk(sk); |
1310 | struct rtable *rt = (struct rtable *)__sk_dst_check(sk, cookie: 0); |
1311 | __be32 daddr; |
1312 | struct ip_options_rcu *inet_opt; |
1313 | struct flowi4 *fl4; |
1314 | int err; |
1315 | |
1316 | /* Route is OK, nothing to do. */ |
1317 | if (rt) |
1318 | return 0; |
1319 | |
1320 | /* Reroute. */ |
1321 | rcu_read_lock(); |
1322 | inet_opt = rcu_dereference(inet->inet_opt); |
1323 | daddr = inet->inet_daddr; |
1324 | if (inet_opt && inet_opt->opt.srr) |
1325 | daddr = inet_opt->opt.faddr; |
1326 | rcu_read_unlock(); |
1327 | fl4 = &inet->cork.fl.u.ip4; |
1328 | rt = ip_route_output_ports(net: sock_net(sk), fl4, sk, daddr, saddr: inet->inet_saddr, |
1329 | dport: inet->inet_dport, sport: inet->inet_sport, |
1330 | proto: sk->sk_protocol, tos: ip_sock_rt_tos(sk), |
1331 | oif: sk->sk_bound_dev_if); |
1332 | if (!IS_ERR(ptr: rt)) { |
1333 | err = 0; |
1334 | sk_setup_caps(sk, dst: &rt->dst); |
1335 | } else { |
1336 | err = PTR_ERR(ptr: rt); |
1337 | |
1338 | /* Routing failed... */ |
1339 | sk->sk_route_caps = 0; |
1340 | /* |
1341 | * Other protocols have to map its equivalent state to TCP_SYN_SENT. |
1342 | * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme |
1343 | */ |
1344 | if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_ip_dynaddr) || |
1345 | sk->sk_state != TCP_SYN_SENT || |
1346 | (sk->sk_userlocks & SOCK_BINDADDR_LOCK) || |
1347 | (err = inet_sk_reselect_saddr(sk)) != 0) |
1348 | WRITE_ONCE(sk->sk_err_soft, -err); |
1349 | } |
1350 | |
1351 | return err; |
1352 | } |
1353 | EXPORT_SYMBOL(inet_sk_rebuild_header); |
1354 | |
1355 | void inet_sk_set_state(struct sock *sk, int state) |
1356 | { |
1357 | trace_inet_sock_set_state(sk, oldstate: sk->sk_state, newstate: state); |
1358 | sk->sk_state = state; |
1359 | } |
1360 | EXPORT_SYMBOL(inet_sk_set_state); |
1361 | |
1362 | void inet_sk_state_store(struct sock *sk, int newstate) |
1363 | { |
1364 | trace_inet_sock_set_state(sk, oldstate: sk->sk_state, newstate); |
1365 | smp_store_release(&sk->sk_state, newstate); |
1366 | } |
1367 | |
1368 | struct sk_buff *inet_gso_segment(struct sk_buff *skb, |
1369 | netdev_features_t features) |
1370 | { |
1371 | bool udpfrag = false, fixedid = false, gso_partial, encap; |
1372 | struct sk_buff *segs = ERR_PTR(error: -EINVAL); |
1373 | const struct net_offload *ops; |
1374 | unsigned int offset = 0; |
1375 | struct iphdr *iph; |
1376 | int proto, tot_len; |
1377 | int nhoff; |
1378 | int ihl; |
1379 | int id; |
1380 | |
1381 | skb_reset_network_header(skb); |
1382 | nhoff = skb_network_header(skb) - skb_mac_header(skb); |
1383 | if (unlikely(!pskb_may_pull(skb, sizeof(*iph)))) |
1384 | goto out; |
1385 | |
1386 | iph = ip_hdr(skb); |
1387 | ihl = iph->ihl * 4; |
1388 | if (ihl < sizeof(*iph)) |
1389 | goto out; |
1390 | |
1391 | id = ntohs(iph->id); |
1392 | proto = iph->protocol; |
1393 | |
1394 | /* Warning: after this point, iph might be no longer valid */ |
1395 | if (unlikely(!pskb_may_pull(skb, ihl))) |
1396 | goto out; |
1397 | __skb_pull(skb, len: ihl); |
1398 | |
1399 | encap = SKB_GSO_CB(skb)->encap_level > 0; |
1400 | if (encap) |
1401 | features &= skb->dev->hw_enc_features; |
1402 | SKB_GSO_CB(skb)->encap_level += ihl; |
1403 | |
1404 | skb_reset_transport_header(skb); |
1405 | |
1406 | segs = ERR_PTR(error: -EPROTONOSUPPORT); |
1407 | |
1408 | if (!skb->encapsulation || encap) { |
1409 | udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP); |
1410 | fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID); |
1411 | |
1412 | /* fixed ID is invalid if DF bit is not set */ |
1413 | if (fixedid && !(ip_hdr(skb)->frag_off & htons(IP_DF))) |
1414 | goto out; |
1415 | } |
1416 | |
1417 | ops = rcu_dereference(inet_offloads[proto]); |
1418 | if (likely(ops && ops->callbacks.gso_segment)) { |
1419 | segs = ops->callbacks.gso_segment(skb, features); |
1420 | if (!segs) |
1421 | skb->network_header = skb_mac_header(skb) + nhoff - skb->head; |
1422 | } |
1423 | |
1424 | if (IS_ERR_OR_NULL(ptr: segs)) |
1425 | goto out; |
1426 | |
1427 | gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL); |
1428 | |
1429 | skb = segs; |
1430 | do { |
1431 | iph = (struct iphdr *)(skb_mac_header(skb) + nhoff); |
1432 | if (udpfrag) { |
1433 | iph->frag_off = htons(offset >> 3); |
1434 | if (skb->next) |
1435 | iph->frag_off |= htons(IP_MF); |
1436 | offset += skb->len - nhoff - ihl; |
1437 | tot_len = skb->len - nhoff; |
1438 | } else if (skb_is_gso(skb)) { |
1439 | if (!fixedid) { |
1440 | iph->id = htons(id); |
1441 | id += skb_shinfo(skb)->gso_segs; |
1442 | } |
1443 | |
1444 | if (gso_partial) |
1445 | tot_len = skb_shinfo(skb)->gso_size + |
1446 | SKB_GSO_CB(skb)->data_offset + |
1447 | skb->head - (unsigned char *)iph; |
1448 | else |
1449 | tot_len = skb->len - nhoff; |
1450 | } else { |
1451 | if (!fixedid) |
1452 | iph->id = htons(id++); |
1453 | tot_len = skb->len - nhoff; |
1454 | } |
1455 | iph->tot_len = htons(tot_len); |
1456 | ip_send_check(ip: iph); |
1457 | if (encap) |
1458 | skb_reset_inner_headers(skb); |
1459 | skb->network_header = (u8 *)iph - skb->head; |
1460 | skb_reset_mac_len(skb); |
1461 | } while ((skb = skb->next)); |
1462 | |
1463 | out: |
1464 | return segs; |
1465 | } |
1466 | |
1467 | static struct sk_buff *ipip_gso_segment(struct sk_buff *skb, |
1468 | netdev_features_t features) |
1469 | { |
1470 | if (!(skb_shinfo(skb)->gso_type & SKB_GSO_IPXIP4)) |
1471 | return ERR_PTR(error: -EINVAL); |
1472 | |
1473 | return inet_gso_segment(skb, features); |
1474 | } |
1475 | |
1476 | struct sk_buff *inet_gro_receive(struct list_head *head, struct sk_buff *skb) |
1477 | { |
1478 | const struct net_offload *ops; |
1479 | struct sk_buff *pp = NULL; |
1480 | const struct iphdr *iph; |
1481 | struct sk_buff *p; |
1482 | unsigned int hlen; |
1483 | unsigned int off; |
1484 | unsigned int id; |
1485 | int flush = 1; |
1486 | int proto; |
1487 | |
1488 | off = skb_gro_offset(skb); |
1489 | hlen = off + sizeof(*iph); |
1490 | iph = skb_gro_header(skb, hlen, offset: off); |
1491 | if (unlikely(!iph)) |
1492 | goto out; |
1493 | |
1494 | proto = iph->protocol; |
1495 | |
1496 | ops = rcu_dereference(inet_offloads[proto]); |
1497 | if (!ops || !ops->callbacks.gro_receive) |
1498 | goto out; |
1499 | |
1500 | if (*(u8 *)iph != 0x45) |
1501 | goto out; |
1502 | |
1503 | if (ip_is_fragment(iph)) |
1504 | goto out; |
1505 | |
1506 | if (unlikely(ip_fast_csum((u8 *)iph, 5))) |
1507 | goto out; |
1508 | |
1509 | NAPI_GRO_CB(skb)->proto = proto; |
1510 | id = ntohl(*(__be32 *)&iph->id); |
1511 | flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF)); |
1512 | id >>= 16; |
1513 | |
1514 | list_for_each_entry(p, head, list) { |
1515 | struct iphdr *iph2; |
1516 | u16 flush_id; |
1517 | |
1518 | if (!NAPI_GRO_CB(p)->same_flow) |
1519 | continue; |
1520 | |
1521 | iph2 = (struct iphdr *)(p->data + off); |
1522 | /* The above works because, with the exception of the top |
1523 | * (inner most) layer, we only aggregate pkts with the same |
1524 | * hdr length so all the hdrs we'll need to verify will start |
1525 | * at the same offset. |
1526 | */ |
1527 | if ((iph->protocol ^ iph2->protocol) | |
1528 | ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) | |
1529 | ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) { |
1530 | NAPI_GRO_CB(p)->same_flow = 0; |
1531 | continue; |
1532 | } |
1533 | |
1534 | /* All fields must match except length and checksum. */ |
1535 | NAPI_GRO_CB(p)->flush |= |
1536 | (iph->ttl ^ iph2->ttl) | |
1537 | (iph->tos ^ iph2->tos) | |
1538 | ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF)); |
1539 | |
1540 | NAPI_GRO_CB(p)->flush |= flush; |
1541 | |
1542 | /* We need to store of the IP ID check to be included later |
1543 | * when we can verify that this packet does in fact belong |
1544 | * to a given flow. |
1545 | */ |
1546 | flush_id = (u16)(id - ntohs(iph2->id)); |
1547 | |
1548 | /* This bit of code makes it much easier for us to identify |
1549 | * the cases where we are doing atomic vs non-atomic IP ID |
1550 | * checks. Specifically an atomic check can return IP ID |
1551 | * values 0 - 0xFFFF, while a non-atomic check can only |
1552 | * return 0 or 0xFFFF. |
1553 | */ |
1554 | if (!NAPI_GRO_CB(p)->is_atomic || |
1555 | !(iph->frag_off & htons(IP_DF))) { |
1556 | flush_id ^= NAPI_GRO_CB(p)->count; |
1557 | flush_id = flush_id ? 0xFFFF : 0; |
1558 | } |
1559 | |
1560 | /* If the previous IP ID value was based on an atomic |
1561 | * datagram we can overwrite the value and ignore it. |
1562 | */ |
1563 | if (NAPI_GRO_CB(skb)->is_atomic) |
1564 | NAPI_GRO_CB(p)->flush_id = flush_id; |
1565 | else |
1566 | NAPI_GRO_CB(p)->flush_id |= flush_id; |
1567 | } |
1568 | |
1569 | NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF)); |
1570 | NAPI_GRO_CB(skb)->flush |= flush; |
1571 | skb_set_network_header(skb, offset: off); |
1572 | /* The above will be needed by the transport layer if there is one |
1573 | * immediately following this IP hdr. |
1574 | */ |
1575 | |
1576 | /* Note : No need to call skb_gro_postpull_rcsum() here, |
1577 | * as we already checked checksum over ipv4 header was 0 |
1578 | */ |
1579 | skb_gro_pull(skb, len: sizeof(*iph)); |
1580 | skb_set_transport_header(skb, offset: skb_gro_offset(skb)); |
1581 | |
1582 | pp = indirect_call_gro_receive(tcp4_gro_receive, udp4_gro_receive, |
1583 | ops->callbacks.gro_receive, head, skb); |
1584 | |
1585 | out: |
1586 | skb_gro_flush_final(skb, pp, flush); |
1587 | |
1588 | return pp; |
1589 | } |
1590 | |
1591 | static struct sk_buff *ipip_gro_receive(struct list_head *head, |
1592 | struct sk_buff *skb) |
1593 | { |
1594 | if (NAPI_GRO_CB(skb)->encap_mark) { |
1595 | NAPI_GRO_CB(skb)->flush = 1; |
1596 | return NULL; |
1597 | } |
1598 | |
1599 | NAPI_GRO_CB(skb)->encap_mark = 1; |
1600 | |
1601 | return inet_gro_receive(head, skb); |
1602 | } |
1603 | |
1604 | #define SECONDS_PER_DAY 86400 |
1605 | |
1606 | /* inet_current_timestamp - Return IP network timestamp |
1607 | * |
1608 | * Return milliseconds since midnight in network byte order. |
1609 | */ |
1610 | __be32 inet_current_timestamp(void) |
1611 | { |
1612 | u32 secs; |
1613 | u32 msecs; |
1614 | struct timespec64 ts; |
1615 | |
1616 | ktime_get_real_ts64(tv: &ts); |
1617 | |
1618 | /* Get secs since midnight. */ |
1619 | (void)div_u64_rem(dividend: ts.tv_sec, SECONDS_PER_DAY, remainder: &secs); |
1620 | /* Convert to msecs. */ |
1621 | msecs = secs * MSEC_PER_SEC; |
1622 | /* Convert nsec to msec. */ |
1623 | msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC; |
1624 | |
1625 | /* Convert to network byte order. */ |
1626 | return htonl(msecs); |
1627 | } |
1628 | EXPORT_SYMBOL(inet_current_timestamp); |
1629 | |
1630 | int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len) |
1631 | { |
1632 | unsigned int family = READ_ONCE(sk->sk_family); |
1633 | |
1634 | if (family == AF_INET) |
1635 | return ip_recv_error(sk, msg, len, addr_len); |
1636 | #if IS_ENABLED(CONFIG_IPV6) |
1637 | if (family == AF_INET6) |
1638 | return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len); |
1639 | #endif |
1640 | return -EINVAL; |
1641 | } |
1642 | EXPORT_SYMBOL(inet_recv_error); |
1643 | |
1644 | int inet_gro_complete(struct sk_buff *skb, int nhoff) |
1645 | { |
1646 | struct iphdr *iph = (struct iphdr *)(skb->data + nhoff); |
1647 | const struct net_offload *ops; |
1648 | __be16 totlen = iph->tot_len; |
1649 | int proto = iph->protocol; |
1650 | int err = -ENOSYS; |
1651 | |
1652 | if (skb->encapsulation) { |
1653 | skb_set_inner_protocol(skb, cpu_to_be16(ETH_P_IP)); |
1654 | skb_set_inner_network_header(skb, offset: nhoff); |
1655 | } |
1656 | |
1657 | iph_set_totlen(iph, len: skb->len - nhoff); |
1658 | csum_replace2(sum: &iph->check, old: totlen, new: iph->tot_len); |
1659 | |
1660 | ops = rcu_dereference(inet_offloads[proto]); |
1661 | if (WARN_ON(!ops || !ops->callbacks.gro_complete)) |
1662 | goto out; |
1663 | |
1664 | /* Only need to add sizeof(*iph) to get to the next hdr below |
1665 | * because any hdr with option will have been flushed in |
1666 | * inet_gro_receive(). |
1667 | */ |
1668 | err = INDIRECT_CALL_2(ops->callbacks.gro_complete, |
1669 | tcp4_gro_complete, udp4_gro_complete, |
1670 | skb, nhoff + sizeof(*iph)); |
1671 | |
1672 | out: |
1673 | return err; |
1674 | } |
1675 | |
1676 | static int ipip_gro_complete(struct sk_buff *skb, int nhoff) |
1677 | { |
1678 | skb->encapsulation = 1; |
1679 | skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4; |
1680 | return inet_gro_complete(skb, nhoff); |
1681 | } |
1682 | |
1683 | int inet_ctl_sock_create(struct sock **sk, unsigned short family, |
1684 | unsigned short type, unsigned char protocol, |
1685 | struct net *net) |
1686 | { |
1687 | struct socket *sock; |
1688 | int rc = sock_create_kern(net, family, type, proto: protocol, res: &sock); |
1689 | |
1690 | if (rc == 0) { |
1691 | *sk = sock->sk; |
1692 | (*sk)->sk_allocation = GFP_ATOMIC; |
1693 | (*sk)->sk_use_task_frag = false; |
1694 | /* |
1695 | * Unhash it so that IP input processing does not even see it, |
1696 | * we do not wish this socket to see incoming packets. |
1697 | */ |
1698 | (*sk)->sk_prot->unhash(*sk); |
1699 | } |
1700 | return rc; |
1701 | } |
1702 | EXPORT_SYMBOL_GPL(inet_ctl_sock_create); |
1703 | |
1704 | unsigned long snmp_fold_field(void __percpu *mib, int offt) |
1705 | { |
1706 | unsigned long res = 0; |
1707 | int i; |
1708 | |
1709 | for_each_possible_cpu(i) |
1710 | res += snmp_get_cpu_field(mib, cpu: i, offt); |
1711 | return res; |
1712 | } |
1713 | EXPORT_SYMBOL_GPL(snmp_fold_field); |
1714 | |
1715 | #if BITS_PER_LONG==32 |
1716 | |
1717 | u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt, |
1718 | size_t syncp_offset) |
1719 | { |
1720 | void *bhptr; |
1721 | struct u64_stats_sync *syncp; |
1722 | u64 v; |
1723 | unsigned int start; |
1724 | |
1725 | bhptr = per_cpu_ptr(mib, cpu); |
1726 | syncp = (struct u64_stats_sync *)(bhptr + syncp_offset); |
1727 | do { |
1728 | start = u64_stats_fetch_begin(syncp); |
1729 | v = *(((u64 *)bhptr) + offt); |
1730 | } while (u64_stats_fetch_retry(syncp, start)); |
1731 | |
1732 | return v; |
1733 | } |
1734 | EXPORT_SYMBOL_GPL(snmp_get_cpu_field64); |
1735 | |
1736 | u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset) |
1737 | { |
1738 | u64 res = 0; |
1739 | int cpu; |
1740 | |
1741 | for_each_possible_cpu(cpu) { |
1742 | res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset); |
1743 | } |
1744 | return res; |
1745 | } |
1746 | EXPORT_SYMBOL_GPL(snmp_fold_field64); |
1747 | #endif |
1748 | |
1749 | #ifdef CONFIG_IP_MULTICAST |
1750 | static const struct net_protocol igmp_protocol = { |
1751 | .handler = igmp_rcv, |
1752 | }; |
1753 | #endif |
1754 | |
1755 | static const struct net_protocol icmp_protocol = { |
1756 | .handler = icmp_rcv, |
1757 | .err_handler = icmp_err, |
1758 | .no_policy = 1, |
1759 | }; |
1760 | |
1761 | static __net_init int ipv4_mib_init_net(struct net *net) |
1762 | { |
1763 | int i; |
1764 | |
1765 | net->mib.tcp_statistics = alloc_percpu(struct tcp_mib); |
1766 | if (!net->mib.tcp_statistics) |
1767 | goto err_tcp_mib; |
1768 | net->mib.ip_statistics = alloc_percpu(struct ipstats_mib); |
1769 | if (!net->mib.ip_statistics) |
1770 | goto err_ip_mib; |
1771 | |
1772 | for_each_possible_cpu(i) { |
1773 | struct ipstats_mib *af_inet_stats; |
1774 | af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i); |
1775 | u64_stats_init(syncp: &af_inet_stats->syncp); |
1776 | } |
1777 | |
1778 | net->mib.net_statistics = alloc_percpu(struct linux_mib); |
1779 | if (!net->mib.net_statistics) |
1780 | goto err_net_mib; |
1781 | net->mib.udp_statistics = alloc_percpu(struct udp_mib); |
1782 | if (!net->mib.udp_statistics) |
1783 | goto err_udp_mib; |
1784 | net->mib.udplite_statistics = alloc_percpu(struct udp_mib); |
1785 | if (!net->mib.udplite_statistics) |
1786 | goto err_udplite_mib; |
1787 | net->mib.icmp_statistics = alloc_percpu(struct icmp_mib); |
1788 | if (!net->mib.icmp_statistics) |
1789 | goto err_icmp_mib; |
1790 | net->mib.icmpmsg_statistics = kzalloc(size: sizeof(struct icmpmsg_mib), |
1791 | GFP_KERNEL); |
1792 | if (!net->mib.icmpmsg_statistics) |
1793 | goto err_icmpmsg_mib; |
1794 | |
1795 | tcp_mib_init(net); |
1796 | return 0; |
1797 | |
1798 | err_icmpmsg_mib: |
1799 | free_percpu(pdata: net->mib.icmp_statistics); |
1800 | err_icmp_mib: |
1801 | free_percpu(pdata: net->mib.udplite_statistics); |
1802 | err_udplite_mib: |
1803 | free_percpu(pdata: net->mib.udp_statistics); |
1804 | err_udp_mib: |
1805 | free_percpu(pdata: net->mib.net_statistics); |
1806 | err_net_mib: |
1807 | free_percpu(pdata: net->mib.ip_statistics); |
1808 | err_ip_mib: |
1809 | free_percpu(pdata: net->mib.tcp_statistics); |
1810 | err_tcp_mib: |
1811 | return -ENOMEM; |
1812 | } |
1813 | |
1814 | static __net_exit void ipv4_mib_exit_net(struct net *net) |
1815 | { |
1816 | kfree(objp: net->mib.icmpmsg_statistics); |
1817 | free_percpu(pdata: net->mib.icmp_statistics); |
1818 | free_percpu(pdata: net->mib.udplite_statistics); |
1819 | free_percpu(pdata: net->mib.udp_statistics); |
1820 | free_percpu(pdata: net->mib.net_statistics); |
1821 | free_percpu(pdata: net->mib.ip_statistics); |
1822 | free_percpu(pdata: net->mib.tcp_statistics); |
1823 | #ifdef CONFIG_MPTCP |
1824 | /* allocated on demand, see mptcp_init_sock() */ |
1825 | free_percpu(pdata: net->mib.mptcp_statistics); |
1826 | #endif |
1827 | } |
1828 | |
1829 | static __net_initdata struct pernet_operations ipv4_mib_ops = { |
1830 | .init = ipv4_mib_init_net, |
1831 | .exit = ipv4_mib_exit_net, |
1832 | }; |
1833 | |
1834 | static int __init init_ipv4_mibs(void) |
1835 | { |
1836 | return register_pernet_subsys(&ipv4_mib_ops); |
1837 | } |
1838 | |
1839 | static __net_init int inet_init_net(struct net *net) |
1840 | { |
1841 | /* |
1842 | * Set defaults for local port range |
1843 | */ |
1844 | net->ipv4.ip_local_ports.range = 60999u << 16 | 32768u; |
1845 | |
1846 | seqlock_init(&net->ipv4.ping_group_range.lock); |
1847 | /* |
1848 | * Sane defaults - nobody may create ping sockets. |
1849 | * Boot scripts should set this to distro-specific group. |
1850 | */ |
1851 | net->ipv4.ping_group_range.range[0] = make_kgid(from: &init_user_ns, gid: 1); |
1852 | net->ipv4.ping_group_range.range[1] = make_kgid(from: &init_user_ns, gid: 0); |
1853 | |
1854 | /* Default values for sysctl-controlled parameters. |
1855 | * We set them here, in case sysctl is not compiled. |
1856 | */ |
1857 | net->ipv4.sysctl_ip_default_ttl = IPDEFTTL; |
1858 | net->ipv4.sysctl_ip_fwd_update_priority = 1; |
1859 | net->ipv4.sysctl_ip_dynaddr = 0; |
1860 | net->ipv4.sysctl_ip_early_demux = 1; |
1861 | net->ipv4.sysctl_udp_early_demux = 1; |
1862 | net->ipv4.sysctl_tcp_early_demux = 1; |
1863 | net->ipv4.sysctl_nexthop_compat_mode = 1; |
1864 | #ifdef CONFIG_SYSCTL |
1865 | net->ipv4.sysctl_ip_prot_sock = PROT_SOCK; |
1866 | #endif |
1867 | |
1868 | /* Some igmp sysctl, whose values are always used */ |
1869 | net->ipv4.sysctl_igmp_max_memberships = 20; |
1870 | net->ipv4.sysctl_igmp_max_msf = 10; |
1871 | /* IGMP reports for link-local multicast groups are enabled by default */ |
1872 | net->ipv4.sysctl_igmp_llm_reports = 1; |
1873 | net->ipv4.sysctl_igmp_qrv = 2; |
1874 | |
1875 | net->ipv4.sysctl_fib_notify_on_flag_change = 0; |
1876 | |
1877 | return 0; |
1878 | } |
1879 | |
1880 | static __net_initdata struct pernet_operations af_inet_ops = { |
1881 | .init = inet_init_net, |
1882 | }; |
1883 | |
1884 | static int __init init_inet_pernet_ops(void) |
1885 | { |
1886 | return register_pernet_subsys(&af_inet_ops); |
1887 | } |
1888 | |
1889 | static int ipv4_proc_init(void); |
1890 | |
1891 | /* |
1892 | * IP protocol layer initialiser |
1893 | */ |
1894 | |
1895 | |
1896 | static const struct net_offload ipip_offload = { |
1897 | .callbacks = { |
1898 | .gso_segment = ipip_gso_segment, |
1899 | .gro_receive = ipip_gro_receive, |
1900 | .gro_complete = ipip_gro_complete, |
1901 | }, |
1902 | }; |
1903 | |
1904 | static int __init ipip_offload_init(void) |
1905 | { |
1906 | return inet_add_offload(prot: &ipip_offload, IPPROTO_IPIP); |
1907 | } |
1908 | |
1909 | static int __init ipv4_offload_init(void) |
1910 | { |
1911 | /* |
1912 | * Add offloads |
1913 | */ |
1914 | if (udpv4_offload_init() < 0) |
1915 | pr_crit("%s: Cannot add UDP protocol offload\n" , __func__); |
1916 | if (tcpv4_offload_init() < 0) |
1917 | pr_crit("%s: Cannot add TCP protocol offload\n" , __func__); |
1918 | if (ipip_offload_init() < 0) |
1919 | pr_crit("%s: Cannot add IPIP protocol offload\n" , __func__); |
1920 | |
1921 | net_hotdata.ip_packet_offload = (struct packet_offload) { |
1922 | .type = cpu_to_be16(ETH_P_IP), |
1923 | .callbacks = { |
1924 | .gso_segment = inet_gso_segment, |
1925 | .gro_receive = inet_gro_receive, |
1926 | .gro_complete = inet_gro_complete, |
1927 | }, |
1928 | }; |
1929 | dev_add_offload(po: &net_hotdata.ip_packet_offload); |
1930 | return 0; |
1931 | } |
1932 | |
1933 | fs_initcall(ipv4_offload_init); |
1934 | |
1935 | static struct packet_type ip_packet_type __read_mostly = { |
1936 | .type = cpu_to_be16(ETH_P_IP), |
1937 | .func = ip_rcv, |
1938 | .list_func = ip_list_rcv, |
1939 | }; |
1940 | |
1941 | static int __init inet_init(void) |
1942 | { |
1943 | struct inet_protosw *q; |
1944 | struct list_head *r; |
1945 | int rc; |
1946 | |
1947 | sock_skb_cb_check_size(sizeof(struct inet_skb_parm)); |
1948 | |
1949 | raw_hashinfo_init(hashinfo: &raw_v4_hashinfo); |
1950 | |
1951 | rc = proto_register(prot: &tcp_prot, alloc_slab: 1); |
1952 | if (rc) |
1953 | goto out; |
1954 | |
1955 | rc = proto_register(prot: &udp_prot, alloc_slab: 1); |
1956 | if (rc) |
1957 | goto out_unregister_tcp_proto; |
1958 | |
1959 | rc = proto_register(prot: &raw_prot, alloc_slab: 1); |
1960 | if (rc) |
1961 | goto out_unregister_udp_proto; |
1962 | |
1963 | rc = proto_register(prot: &ping_prot, alloc_slab: 1); |
1964 | if (rc) |
1965 | goto out_unregister_raw_proto; |
1966 | |
1967 | /* |
1968 | * Tell SOCKET that we are alive... |
1969 | */ |
1970 | |
1971 | (void)sock_register(fam: &inet_family_ops); |
1972 | |
1973 | #ifdef CONFIG_SYSCTL |
1974 | ip_static_sysctl_init(); |
1975 | #endif |
1976 | |
1977 | /* |
1978 | * Add all the base protocols. |
1979 | */ |
1980 | |
1981 | if (inet_add_protocol(prot: &icmp_protocol, IPPROTO_ICMP) < 0) |
1982 | pr_crit("%s: Cannot add ICMP protocol\n" , __func__); |
1983 | |
1984 | net_hotdata.udp_protocol = (struct net_protocol) { |
1985 | .handler = udp_rcv, |
1986 | .err_handler = udp_err, |
1987 | .no_policy = 1, |
1988 | }; |
1989 | if (inet_add_protocol(prot: &net_hotdata.udp_protocol, IPPROTO_UDP) < 0) |
1990 | pr_crit("%s: Cannot add UDP protocol\n" , __func__); |
1991 | |
1992 | net_hotdata.tcp_protocol = (struct net_protocol) { |
1993 | .handler = tcp_v4_rcv, |
1994 | .err_handler = tcp_v4_err, |
1995 | .no_policy = 1, |
1996 | .icmp_strict_tag_validation = 1, |
1997 | }; |
1998 | if (inet_add_protocol(prot: &net_hotdata.tcp_protocol, IPPROTO_TCP) < 0) |
1999 | pr_crit("%s: Cannot add TCP protocol\n" , __func__); |
2000 | #ifdef CONFIG_IP_MULTICAST |
2001 | if (inet_add_protocol(prot: &igmp_protocol, IPPROTO_IGMP) < 0) |
2002 | pr_crit("%s: Cannot add IGMP protocol\n" , __func__); |
2003 | #endif |
2004 | |
2005 | /* Register the socket-side information for inet_create. */ |
2006 | for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r) |
2007 | INIT_LIST_HEAD(list: r); |
2008 | |
2009 | for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q) |
2010 | inet_register_protosw(q); |
2011 | |
2012 | /* |
2013 | * Set the ARP module up |
2014 | */ |
2015 | |
2016 | arp_init(); |
2017 | |
2018 | /* |
2019 | * Set the IP module up |
2020 | */ |
2021 | |
2022 | ip_init(); |
2023 | |
2024 | /* Initialise per-cpu ipv4 mibs */ |
2025 | if (init_ipv4_mibs()) |
2026 | panic(fmt: "%s: Cannot init ipv4 mibs\n" , __func__); |
2027 | |
2028 | /* Setup TCP slab cache for open requests. */ |
2029 | tcp_init(); |
2030 | |
2031 | /* Setup UDP memory threshold */ |
2032 | udp_init(); |
2033 | |
2034 | /* Add UDP-Lite (RFC 3828) */ |
2035 | udplite4_register(); |
2036 | |
2037 | raw_init(); |
2038 | |
2039 | ping_init(); |
2040 | |
2041 | /* |
2042 | * Set the ICMP layer up |
2043 | */ |
2044 | |
2045 | if (icmp_init() < 0) |
2046 | panic(fmt: "Failed to create the ICMP control socket.\n" ); |
2047 | |
2048 | /* |
2049 | * Initialise the multicast router |
2050 | */ |
2051 | #if defined(CONFIG_IP_MROUTE) |
2052 | if (ip_mr_init()) |
2053 | pr_crit("%s: Cannot init ipv4 mroute\n" , __func__); |
2054 | #endif |
2055 | |
2056 | if (init_inet_pernet_ops()) |
2057 | pr_crit("%s: Cannot init ipv4 inet pernet ops\n" , __func__); |
2058 | |
2059 | ipv4_proc_init(); |
2060 | |
2061 | ipfrag_init(); |
2062 | |
2063 | dev_add_pack(pt: &ip_packet_type); |
2064 | |
2065 | ip_tunnel_core_init(); |
2066 | |
2067 | rc = 0; |
2068 | out: |
2069 | return rc; |
2070 | out_unregister_raw_proto: |
2071 | proto_unregister(prot: &raw_prot); |
2072 | out_unregister_udp_proto: |
2073 | proto_unregister(prot: &udp_prot); |
2074 | out_unregister_tcp_proto: |
2075 | proto_unregister(prot: &tcp_prot); |
2076 | goto out; |
2077 | } |
2078 | |
2079 | fs_initcall(inet_init); |
2080 | |
2081 | /* ------------------------------------------------------------------------ */ |
2082 | |
2083 | #ifdef CONFIG_PROC_FS |
2084 | static int __init ipv4_proc_init(void) |
2085 | { |
2086 | int rc = 0; |
2087 | |
2088 | if (raw_proc_init()) |
2089 | goto out_raw; |
2090 | if (tcp4_proc_init()) |
2091 | goto out_tcp; |
2092 | if (udp4_proc_init()) |
2093 | goto out_udp; |
2094 | if (ping_proc_init()) |
2095 | goto out_ping; |
2096 | if (ip_misc_proc_init()) |
2097 | goto out_misc; |
2098 | out: |
2099 | return rc; |
2100 | out_misc: |
2101 | ping_proc_exit(); |
2102 | out_ping: |
2103 | udp4_proc_exit(); |
2104 | out_udp: |
2105 | tcp4_proc_exit(); |
2106 | out_tcp: |
2107 | raw_proc_exit(); |
2108 | out_raw: |
2109 | rc = -ENOMEM; |
2110 | goto out; |
2111 | } |
2112 | |
2113 | #else /* CONFIG_PROC_FS */ |
2114 | static int __init ipv4_proc_init(void) |
2115 | { |
2116 | return 0; |
2117 | } |
2118 | #endif /* CONFIG_PROC_FS */ |
2119 | |