1 /* 2 * INET An implementation of the TCP/IP protocol suite for the LINUX 3 * operating system. INET is implemented using the BSD Socket 4 * interface as the means of communication with the user level. 5 * 6 * PF_INET protocol family socket handler. 7 * 8 * Authors: Ross Biro 9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 10 * Florian La Roche, <flla@stud.uni-sb.de> 11 * Alan Cox, <A.Cox@swansea.ac.uk> 12 * 13 * Changes (see also sock.c) 14 * 15 * piggy, 16 * Karl Knutson : Socket protocol table 17 * A.N.Kuznetsov : Socket death error in accept(). 18 * John Richardson : Fix non blocking error in connect() 19 * so sockets that fail to connect 20 * don't return -EINPROGRESS. 21 * Alan Cox : Asynchronous I/O support 22 * Alan Cox : Keep correct socket pointer on sock 23 * structures 24 * when accept() ed 25 * Alan Cox : Semantics of SO_LINGER aren't state 26 * moved to close when you look carefully. 27 * With this fixed and the accept bug fixed 28 * some RPC stuff seems happier. 29 * Niibe Yutaka : 4.4BSD style write async I/O 30 * Alan Cox, 31 * Tony Gale : Fixed reuse semantics. 32 * Alan Cox : bind() shouldn't abort existing but dead 33 * sockets. Stops FTP netin:.. I hope. 34 * Alan Cox : bind() works correctly for RAW sockets. 35 * Note that FreeBSD at least was broken 36 * in this respect so be careful with 37 * compatibility tests... 38 * Alan Cox : routing cache support 39 * Alan Cox : memzero the socket structure for 40 * compactness. 41 * Matt Day : nonblock connect error handler 42 * Alan Cox : Allow large numbers of pending sockets 43 * (eg for big web sites), but only if 44 * specifically application requested. 45 * Alan Cox : New buffering throughout IP. Used 46 * dumbly. 47 * Alan Cox : New buffering now used smartly. 48 * Alan Cox : BSD rather than common sense 49 * interpretation of listen. 50 * Germano Caronni : Assorted small races. 51 * Alan Cox : sendmsg/recvmsg basic support. 52 * Alan Cox : Only sendmsg/recvmsg now supported. 53 * Alan Cox : Locked down bind (see security list). 54 * Alan Cox : Loosened bind a little. 55 * Mike McLagan : ADD/DEL DLCI Ioctls 56 * Willy Konynenberg : Transparent proxying support. 57 * David S. Miller : New socket lookup architecture. 58 * Some other random speedups. 59 * Cyrus Durgin : Cleaned up file for kmod hacks. 60 * Andi Kleen : Fix inet_stream_connect TCP race. 61 * 62 * This program is free software; you can redistribute it and/or 63 * modify it under the terms of the GNU General Public License 64 * as published by the Free Software Foundation; either version 65 * 2 of the License, or (at your option) any later version. 66 */ 67 68 #define pr_fmt(fmt) "IPv4: " fmt 69 70 #include <linux/err.h> 71 #include <linux/errno.h> 72 #include <linux/types.h> 73 #include <linux/socket.h> 74 #include <linux/in.h> 75 #include <linux/kernel.h> 76 #include <linux/kmod.h> 77 #include <linux/sched.h> 78 #include <linux/timer.h> 79 #include <linux/string.h> 80 #include <linux/sockios.h> 81 #include <linux/net.h> 82 #include <linux/capability.h> 83 #include <linux/fcntl.h> 84 #include <linux/mm.h> 85 #include <linux/interrupt.h> 86 #include <linux/stat.h> 87 #include <linux/init.h> 88 #include <linux/poll.h> 89 #include <linux/netfilter_ipv4.h> 90 #include <linux/random.h> 91 #include <linux/slab.h> 92 93 #include <asm/uaccess.h> 94 95 #include <linux/inet.h> 96 #include <linux/igmp.h> 97 #include <linux/inetdevice.h> 98 #include <linux/netdevice.h> 99 #include <net/checksum.h> 100 #include <net/ip.h> 101 #include <net/protocol.h> 102 #include <net/arp.h> 103 #include <net/route.h> 104 #include <net/ip_fib.h> 105 #include <net/inet_connection_sock.h> 106 #include <net/tcp.h> 107 #include <net/udp.h> 108 #include <net/udplite.h> 109 #include <net/ping.h> 110 #include <linux/skbuff.h> 111 #include <net/sock.h> 112 #include <net/raw.h> 113 #include <net/icmp.h> 114 #include <net/inet_common.h> 115 #include <net/ip_tunnels.h> 116 #include <net/xfrm.h> 117 #include <net/net_namespace.h> 118 #include <net/secure_seq.h> 119 #ifdef CONFIG_IP_MROUTE 120 #include <linux/mroute.h> 121 #endif 122 #include <net/l3mdev.h> 123 124 125 /* The inetsw table contains everything that inet_create needs to 126 * build a new socket. 127 */ 128 static struct list_head inetsw[SOCK_MAX]; 129 static DEFINE_SPINLOCK(inetsw_lock); 130 131 /* New destruction routine */ 132 133 void inet_sock_destruct(struct sock *sk) 134 { 135 struct inet_sock *inet = inet_sk(sk); 136 137 __skb_queue_purge(&sk->sk_receive_queue); 138 __skb_queue_purge(&sk->sk_error_queue); 139 140 sk_mem_reclaim(sk); 141 142 if (sk->sk_type == SOCK_STREAM && sk->sk_state != TCP_CLOSE) { 143 pr_err("Attempt to release TCP socket in state %d %p\n", 144 sk->sk_state, sk); 145 return; 146 } 147 if (!sock_flag(sk, SOCK_DEAD)) { 148 pr_err("Attempt to release alive inet socket %p\n", sk); 149 return; 150 } 151 152 WARN_ON(atomic_read(&sk->sk_rmem_alloc)); 153 WARN_ON(atomic_read(&sk->sk_wmem_alloc)); 154 WARN_ON(sk->sk_wmem_queued); 155 WARN_ON(sk->sk_forward_alloc); 156 157 kfree(rcu_dereference_protected(inet->inet_opt, 1)); 158 dst_release(rcu_dereference_check(sk->sk_dst_cache, 1)); 159 dst_release(sk->sk_rx_dst); 160 sk_refcnt_debug_dec(sk); 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 /* 192 * Move a socket into listening state. 193 */ 194 int inet_listen(struct socket *sock, int backlog) 195 { 196 struct sock *sk = sock->sk; 197 unsigned char old_state; 198 int err; 199 200 lock_sock(sk); 201 202 err = -EINVAL; 203 if (sock->state != SS_UNCONNECTED || sock->type != SOCK_STREAM) 204 goto out; 205 206 old_state = sk->sk_state; 207 if (!((1 << old_state) & (TCPF_CLOSE | TCPF_LISTEN))) 208 goto out; 209 210 /* Really, if the socket is already in listen state 211 * we can only allow the backlog to be adjusted. 212 */ 213 if (old_state != TCP_LISTEN) { 214 /* Enable TFO w/o requiring TCP_FASTOPEN socket option. 215 * Note that only TCP sockets (SOCK_STREAM) will reach here. 216 * Also fastopen backlog may already been set via the option 217 * because the socket was in TCP_LISTEN state previously but 218 * was shutdown() rather than close(). 219 */ 220 if ((sysctl_tcp_fastopen & TFO_SERVER_WO_SOCKOPT1) && 221 (sysctl_tcp_fastopen & TFO_SERVER_ENABLE) && 222 !inet_csk(sk)->icsk_accept_queue.fastopenq.max_qlen) { 223 fastopen_queue_tune(sk, backlog); 224 tcp_fastopen_init_key_once(true); 225 } 226 227 err = inet_csk_listen_start(sk, backlog); 228 if (err) 229 goto out; 230 } 231 sk->sk_max_ack_backlog = backlog; 232 err = 0; 233 234 out: 235 release_sock(sk); 236 return err; 237 } 238 EXPORT_SYMBOL(inet_listen); 239 240 /* 241 * Create an inet socket. 242 */ 243 244 static int inet_create(struct net *net, struct socket *sock, int protocol, 245 int kern) 246 { 247 struct sock *sk; 248 struct inet_protosw *answer; 249 struct inet_sock *inet; 250 struct proto *answer_prot; 251 unsigned char answer_flags; 252 int try_loading_module = 0; 253 int err; 254 255 if (protocol < 0 || protocol >= IPPROTO_MAX) 256 return -EINVAL; 257 258 sock->state = SS_UNCONNECTED; 259 260 /* Look for the requested type/protocol pair. */ 261 lookup_protocol: 262 err = -ESOCKTNOSUPPORT; 263 rcu_read_lock(); 264 list_for_each_entry_rcu(answer, &inetsw[sock->type], list) { 265 266 err = 0; 267 /* Check the non-wild match. */ 268 if (protocol == answer->protocol) { 269 if (protocol != IPPROTO_IP) 270 break; 271 } else { 272 /* Check for the two wild cases. */ 273 if (IPPROTO_IP == protocol) { 274 protocol = answer->protocol; 275 break; 276 } 277 if (IPPROTO_IP == answer->protocol) 278 break; 279 } 280 err = -EPROTONOSUPPORT; 281 } 282 283 if (unlikely(err)) { 284 if (try_loading_module < 2) { 285 rcu_read_unlock(); 286 /* 287 * Be more specific, e.g. net-pf-2-proto-132-type-1 288 * (net-pf-PF_INET-proto-IPPROTO_SCTP-type-SOCK_STREAM) 289 */ 290 if (++try_loading_module == 1) 291 request_module("net-pf-%d-proto-%d-type-%d", 292 PF_INET, protocol, sock->type); 293 /* 294 * Fall back to generic, e.g. net-pf-2-proto-132 295 * (net-pf-PF_INET-proto-IPPROTO_SCTP) 296 */ 297 else 298 request_module("net-pf-%d-proto-%d", 299 PF_INET, protocol); 300 goto lookup_protocol; 301 } else 302 goto out_rcu_unlock; 303 } 304 305 err = -EPERM; 306 if (sock->type == SOCK_RAW && !kern && 307 !ns_capable(net->user_ns, CAP_NET_RAW)) 308 goto out_rcu_unlock; 309 310 sock->ops = answer->ops; 311 answer_prot = answer->prot; 312 answer_flags = answer->flags; 313 rcu_read_unlock(); 314 315 WARN_ON(!answer_prot->slab); 316 317 err = -ENOBUFS; 318 sk = sk_alloc(net, PF_INET, GFP_KERNEL, answer_prot, kern); 319 if (!sk) 320 goto out; 321 322 err = 0; 323 if (INET_PROTOSW_REUSE & answer_flags) 324 sk->sk_reuse = SK_CAN_REUSE; 325 326 inet = inet_sk(sk); 327 inet->is_icsk = (INET_PROTOSW_ICSK & answer_flags) != 0; 328 329 inet->nodefrag = 0; 330 331 if (SOCK_RAW == sock->type) { 332 inet->inet_num = protocol; 333 if (IPPROTO_RAW == protocol) 334 inet->hdrincl = 1; 335 } 336 337 if (net->ipv4.sysctl_ip_no_pmtu_disc) 338 inet->pmtudisc = IP_PMTUDISC_DONT; 339 else 340 inet->pmtudisc = IP_PMTUDISC_WANT; 341 342 inet->inet_id = 0; 343 344 sock_init_data(sock, sk); 345 346 sk->sk_destruct = inet_sock_destruct; 347 sk->sk_protocol = protocol; 348 sk->sk_backlog_rcv = sk->sk_prot->backlog_rcv; 349 350 inet->uc_ttl = -1; 351 inet->mc_loop = 1; 352 inet->mc_ttl = 1; 353 inet->mc_all = 1; 354 inet->mc_index = 0; 355 inet->mc_list = NULL; 356 inet->rcv_tos = 0; 357 358 sk_refcnt_debug_inc(sk); 359 360 if (inet->inet_num) { 361 /* It assumes that any protocol which allows 362 * the user to assign a number at socket 363 * creation time automatically 364 * shares. 365 */ 366 inet->inet_sport = htons(inet->inet_num); 367 /* Add to protocol hash chains. */ 368 err = sk->sk_prot->hash(sk); 369 if (err) { 370 sk_common_release(sk); 371 goto out; 372 } 373 } 374 375 if (sk->sk_prot->init) { 376 err = sk->sk_prot->init(sk); 377 if (err) 378 sk_common_release(sk); 379 } 380 out: 381 return err; 382 out_rcu_unlock: 383 rcu_read_unlock(); 384 goto out; 385 } 386 387 388 /* 389 * The peer socket should always be NULL (or else). When we call this 390 * function we are destroying the object and from then on nobody 391 * should refer to it. 392 */ 393 int inet_release(struct socket *sock) 394 { 395 struct sock *sk = sock->sk; 396 397 if (sk) { 398 long timeout; 399 400 /* Applications forget to leave groups before exiting */ 401 ip_mc_drop_socket(sk); 402 403 /* If linger is set, we don't return until the close 404 * is complete. Otherwise we return immediately. The 405 * actually closing is done the same either way. 406 * 407 * If the close is due to the process exiting, we never 408 * linger.. 409 */ 410 timeout = 0; 411 if (sock_flag(sk, SOCK_LINGER) && 412 !(current->flags & PF_EXITING)) 413 timeout = sk->sk_lingertime; 414 sock->sk = NULL; 415 sk->sk_prot->close(sk, timeout); 416 } 417 return 0; 418 } 419 EXPORT_SYMBOL(inet_release); 420 421 int inet_bind(struct socket *sock, struct sockaddr *uaddr, int addr_len) 422 { 423 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr; 424 struct sock *sk = sock->sk; 425 struct inet_sock *inet = inet_sk(sk); 426 struct net *net = sock_net(sk); 427 unsigned short snum; 428 int chk_addr_ret; 429 u32 tb_id = RT_TABLE_LOCAL; 430 int err; 431 432 /* If the socket has its own bind function then use it. (RAW) */ 433 if (sk->sk_prot->bind) { 434 err = sk->sk_prot->bind(sk, uaddr, addr_len); 435 goto out; 436 } 437 err = -EINVAL; 438 if (addr_len < sizeof(struct sockaddr_in)) 439 goto out; 440 441 if (addr->sin_family != AF_INET) { 442 /* Compatibility games : accept AF_UNSPEC (mapped to AF_INET) 443 * only if s_addr is INADDR_ANY. 444 */ 445 err = -EAFNOSUPPORT; 446 if (addr->sin_family != AF_UNSPEC || 447 addr->sin_addr.s_addr != htonl(INADDR_ANY)) 448 goto out; 449 } 450 451 tb_id = l3mdev_fib_table_by_index(net, sk->sk_bound_dev_if) ? : tb_id; 452 chk_addr_ret = inet_addr_type_table(net, addr->sin_addr.s_addr, tb_id); 453 454 /* Not specified by any standard per-se, however it breaks too 455 * many applications when removed. It is unfortunate since 456 * allowing applications to make a non-local bind solves 457 * several problems with systems using dynamic addressing. 458 * (ie. your servers still start up even if your ISDN link 459 * is temporarily down) 460 */ 461 err = -EADDRNOTAVAIL; 462 if (!net->ipv4.sysctl_ip_nonlocal_bind && 463 !(inet->freebind || inet->transparent) && 464 addr->sin_addr.s_addr != htonl(INADDR_ANY) && 465 chk_addr_ret != RTN_LOCAL && 466 chk_addr_ret != RTN_MULTICAST && 467 chk_addr_ret != RTN_BROADCAST) 468 goto out; 469 470 snum = ntohs(addr->sin_port); 471 err = -EACCES; 472 if (snum && snum < PROT_SOCK && 473 !ns_capable(net->user_ns, CAP_NET_BIND_SERVICE)) 474 goto out; 475 476 /* We keep a pair of addresses. rcv_saddr is the one 477 * used by hash lookups, and saddr is used for transmit. 478 * 479 * In the BSD API these are the same except where it 480 * would be illegal to use them (multicast/broadcast) in 481 * which case the sending device address is used. 482 */ 483 lock_sock(sk); 484 485 /* Check these errors (active socket, double bind). */ 486 err = -EINVAL; 487 if (sk->sk_state != TCP_CLOSE || inet->inet_num) 488 goto out_release_sock; 489 490 inet->inet_rcv_saddr = inet->inet_saddr = addr->sin_addr.s_addr; 491 if (chk_addr_ret == RTN_MULTICAST || chk_addr_ret == RTN_BROADCAST) 492 inet->inet_saddr = 0; /* Use device */ 493 494 /* Make sure we are allowed to bind here. */ 495 if ((snum || !inet->bind_address_no_port) && 496 sk->sk_prot->get_port(sk, snum)) { 497 inet->inet_saddr = inet->inet_rcv_saddr = 0; 498 err = -EADDRINUSE; 499 goto out_release_sock; 500 } 501 502 if (inet->inet_rcv_saddr) 503 sk->sk_userlocks |= SOCK_BINDADDR_LOCK; 504 if (snum) 505 sk->sk_userlocks |= SOCK_BINDPORT_LOCK; 506 inet->inet_sport = htons(inet->inet_num); 507 inet->inet_daddr = 0; 508 inet->inet_dport = 0; 509 sk_dst_reset(sk); 510 err = 0; 511 out_release_sock: 512 release_sock(sk); 513 out: 514 return err; 515 } 516 EXPORT_SYMBOL(inet_bind); 517 518 int inet_dgram_connect(struct socket *sock, struct sockaddr *uaddr, 519 int addr_len, int flags) 520 { 521 struct sock *sk = sock->sk; 522 523 if (addr_len < sizeof(uaddr->sa_family)) 524 return -EINVAL; 525 if (uaddr->sa_family == AF_UNSPEC) 526 return sk->sk_prot->disconnect(sk, flags); 527 528 if (!inet_sk(sk)->inet_num && inet_autobind(sk)) 529 return -EAGAIN; 530 return sk->sk_prot->connect(sk, uaddr, addr_len); 531 } 532 EXPORT_SYMBOL(inet_dgram_connect); 533 534 static long inet_wait_for_connect(struct sock *sk, long timeo, int writebias) 535 { 536 DEFINE_WAIT_FUNC(wait, woken_wake_function); 537 538 add_wait_queue(sk_sleep(sk), &wait); 539 sk->sk_write_pending += writebias; 540 541 /* Basic assumption: if someone sets sk->sk_err, he _must_ 542 * change state of the socket from TCP_SYN_*. 543 * Connect() does not allow to get error notifications 544 * without closing the socket. 545 */ 546 while ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 547 release_sock(sk); 548 timeo = wait_woken(&wait, TASK_INTERRUPTIBLE, timeo); 549 lock_sock(sk); 550 if (signal_pending(current) || !timeo) 551 break; 552 } 553 remove_wait_queue(sk_sleep(sk), &wait); 554 sk->sk_write_pending -= writebias; 555 return timeo; 556 } 557 558 /* 559 * Connect to a remote host. There is regrettably still a little 560 * TCP 'magic' in here. 561 */ 562 int __inet_stream_connect(struct socket *sock, struct sockaddr *uaddr, 563 int addr_len, int flags) 564 { 565 struct sock *sk = sock->sk; 566 int err; 567 long timeo; 568 569 if (addr_len < sizeof(uaddr->sa_family)) 570 return -EINVAL; 571 572 if (uaddr->sa_family == AF_UNSPEC) { 573 err = sk->sk_prot->disconnect(sk, flags); 574 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED; 575 goto out; 576 } 577 578 switch (sock->state) { 579 default: 580 err = -EINVAL; 581 goto out; 582 case SS_CONNECTED: 583 err = -EISCONN; 584 goto out; 585 case SS_CONNECTING: 586 err = -EALREADY; 587 /* Fall out of switch with err, set for this state */ 588 break; 589 case SS_UNCONNECTED: 590 err = -EISCONN; 591 if (sk->sk_state != TCP_CLOSE) 592 goto out; 593 594 err = sk->sk_prot->connect(sk, uaddr, addr_len); 595 if (err < 0) 596 goto out; 597 598 sock->state = SS_CONNECTING; 599 600 /* Just entered SS_CONNECTING state; the only 601 * difference is that return value in non-blocking 602 * case is EINPROGRESS, rather than EALREADY. 603 */ 604 err = -EINPROGRESS; 605 break; 606 } 607 608 timeo = sock_sndtimeo(sk, flags & O_NONBLOCK); 609 610 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 611 int writebias = (sk->sk_protocol == IPPROTO_TCP) && 612 tcp_sk(sk)->fastopen_req && 613 tcp_sk(sk)->fastopen_req->data ? 1 : 0; 614 615 /* Error code is set above */ 616 if (!timeo || !inet_wait_for_connect(sk, timeo, writebias)) 617 goto out; 618 619 err = sock_intr_errno(timeo); 620 if (signal_pending(current)) 621 goto out; 622 } 623 624 /* Connection was closed by RST, timeout, ICMP error 625 * or another process disconnected us. 626 */ 627 if (sk->sk_state == TCP_CLOSE) 628 goto sock_error; 629 630 /* sk->sk_err may be not zero now, if RECVERR was ordered by user 631 * and error was received after socket entered established state. 632 * Hence, it is handled normally after connect() return successfully. 633 */ 634 635 sock->state = SS_CONNECTED; 636 err = 0; 637 out: 638 return err; 639 640 sock_error: 641 err = sock_error(sk) ? : -ECONNABORTED; 642 sock->state = SS_UNCONNECTED; 643 if (sk->sk_prot->disconnect(sk, flags)) 644 sock->state = SS_DISCONNECTING; 645 goto out; 646 } 647 EXPORT_SYMBOL(__inet_stream_connect); 648 649 int inet_stream_connect(struct socket *sock, struct sockaddr *uaddr, 650 int addr_len, int flags) 651 { 652 int err; 653 654 lock_sock(sock->sk); 655 err = __inet_stream_connect(sock, uaddr, addr_len, flags); 656 release_sock(sock->sk); 657 return err; 658 } 659 EXPORT_SYMBOL(inet_stream_connect); 660 661 /* 662 * Accept a pending connection. The TCP layer now gives BSD semantics. 663 */ 664 665 int inet_accept(struct socket *sock, struct socket *newsock, int flags) 666 { 667 struct sock *sk1 = sock->sk; 668 int err = -EINVAL; 669 struct sock *sk2 = sk1->sk_prot->accept(sk1, flags, &err); 670 671 if (!sk2) 672 goto do_err; 673 674 lock_sock(sk2); 675 676 sock_rps_record_flow(sk2); 677 WARN_ON(!((1 << sk2->sk_state) & 678 (TCPF_ESTABLISHED | TCPF_SYN_RECV | 679 TCPF_CLOSE_WAIT | TCPF_CLOSE))); 680 681 sock_graft(sk2, newsock); 682 683 newsock->state = SS_CONNECTED; 684 err = 0; 685 release_sock(sk2); 686 do_err: 687 return err; 688 } 689 EXPORT_SYMBOL(inet_accept); 690 691 692 /* 693 * This does both peername and sockname. 694 */ 695 int inet_getname(struct socket *sock, struct sockaddr *uaddr, 696 int *uaddr_len, int peer) 697 { 698 struct sock *sk = sock->sk; 699 struct inet_sock *inet = inet_sk(sk); 700 DECLARE_SOCKADDR(struct sockaddr_in *, sin, uaddr); 701 702 sin->sin_family = AF_INET; 703 if (peer) { 704 if (!inet->inet_dport || 705 (((1 << sk->sk_state) & (TCPF_CLOSE | TCPF_SYN_SENT)) && 706 peer == 1)) 707 return -ENOTCONN; 708 sin->sin_port = inet->inet_dport; 709 sin->sin_addr.s_addr = inet->inet_daddr; 710 } else { 711 __be32 addr = inet->inet_rcv_saddr; 712 if (!addr) 713 addr = inet->inet_saddr; 714 sin->sin_port = inet->inet_sport; 715 sin->sin_addr.s_addr = addr; 716 } 717 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 718 *uaddr_len = sizeof(*sin); 719 return 0; 720 } 721 EXPORT_SYMBOL(inet_getname); 722 723 int inet_sendmsg(struct socket *sock, struct msghdr *msg, size_t size) 724 { 725 struct sock *sk = sock->sk; 726 727 sock_rps_record_flow(sk); 728 729 /* We may need to bind the socket. */ 730 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind && 731 inet_autobind(sk)) 732 return -EAGAIN; 733 734 return sk->sk_prot->sendmsg(sk, msg, size); 735 } 736 EXPORT_SYMBOL(inet_sendmsg); 737 738 ssize_t inet_sendpage(struct socket *sock, struct page *page, int offset, 739 size_t size, int flags) 740 { 741 struct sock *sk = sock->sk; 742 743 sock_rps_record_flow(sk); 744 745 /* We may need to bind the socket. */ 746 if (!inet_sk(sk)->inet_num && !sk->sk_prot->no_autobind && 747 inet_autobind(sk)) 748 return -EAGAIN; 749 750 if (sk->sk_prot->sendpage) 751 return sk->sk_prot->sendpage(sk, page, offset, size, flags); 752 return sock_no_sendpage(sock, page, offset, size, flags); 753 } 754 EXPORT_SYMBOL(inet_sendpage); 755 756 int inet_recvmsg(struct socket *sock, struct msghdr *msg, size_t size, 757 int flags) 758 { 759 struct sock *sk = sock->sk; 760 int addr_len = 0; 761 int err; 762 763 sock_rps_record_flow(sk); 764 765 err = sk->sk_prot->recvmsg(sk, msg, size, flags & MSG_DONTWAIT, 766 flags & ~MSG_DONTWAIT, &addr_len); 767 if (err >= 0) 768 msg->msg_namelen = addr_len; 769 return err; 770 } 771 EXPORT_SYMBOL(inet_recvmsg); 772 773 int inet_shutdown(struct socket *sock, int how) 774 { 775 struct sock *sk = sock->sk; 776 int err = 0; 777 778 /* This should really check to make sure 779 * the socket is a TCP socket. (WHY AC...) 780 */ 781 how++; /* maps 0->1 has the advantage of making bit 1 rcvs and 782 1->2 bit 2 snds. 783 2->3 */ 784 if ((how & ~SHUTDOWN_MASK) || !how) /* MAXINT->0 */ 785 return -EINVAL; 786 787 lock_sock(sk); 788 if (sock->state == SS_CONNECTING) { 789 if ((1 << sk->sk_state) & 790 (TCPF_SYN_SENT | TCPF_SYN_RECV | TCPF_CLOSE)) 791 sock->state = SS_DISCONNECTING; 792 else 793 sock->state = SS_CONNECTED; 794 } 795 796 switch (sk->sk_state) { 797 case TCP_CLOSE: 798 err = -ENOTCONN; 799 /* Hack to wake up other listeners, who can poll for 800 POLLHUP, even on eg. unconnected UDP sockets -- RR */ 801 default: 802 sk->sk_shutdown |= how; 803 if (sk->sk_prot->shutdown) 804 sk->sk_prot->shutdown(sk, how); 805 break; 806 807 /* Remaining two branches are temporary solution for missing 808 * close() in multithreaded environment. It is _not_ a good idea, 809 * but we have no choice until close() is repaired at VFS level. 810 */ 811 case TCP_LISTEN: 812 if (!(how & RCV_SHUTDOWN)) 813 break; 814 /* Fall through */ 815 case TCP_SYN_SENT: 816 err = sk->sk_prot->disconnect(sk, O_NONBLOCK); 817 sock->state = err ? SS_DISCONNECTING : SS_UNCONNECTED; 818 break; 819 } 820 821 /* Wake up anyone sleeping in poll. */ 822 sk->sk_state_change(sk); 823 release_sock(sk); 824 return err; 825 } 826 EXPORT_SYMBOL(inet_shutdown); 827 828 /* 829 * ioctl() calls you can issue on an INET socket. Most of these are 830 * device configuration and stuff and very rarely used. Some ioctls 831 * pass on to the socket itself. 832 * 833 * NOTE: I like the idea of a module for the config stuff. ie ifconfig 834 * loads the devconfigure module does its configuring and unloads it. 835 * There's a good 20K of config code hanging around the kernel. 836 */ 837 838 int inet_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 839 { 840 struct sock *sk = sock->sk; 841 int err = 0; 842 struct net *net = sock_net(sk); 843 844 switch (cmd) { 845 case SIOCGSTAMP: 846 err = sock_get_timestamp(sk, (struct timeval __user *)arg); 847 break; 848 case SIOCGSTAMPNS: 849 err = sock_get_timestampns(sk, (struct timespec __user *)arg); 850 break; 851 case SIOCADDRT: 852 case SIOCDELRT: 853 case SIOCRTMSG: 854 err = ip_rt_ioctl(net, cmd, (void __user *)arg); 855 break; 856 case SIOCDARP: 857 case SIOCGARP: 858 case SIOCSARP: 859 err = arp_ioctl(net, cmd, (void __user *)arg); 860 break; 861 case SIOCGIFADDR: 862 case SIOCSIFADDR: 863 case SIOCGIFBRDADDR: 864 case SIOCSIFBRDADDR: 865 case SIOCGIFNETMASK: 866 case SIOCSIFNETMASK: 867 case SIOCGIFDSTADDR: 868 case SIOCSIFDSTADDR: 869 case SIOCSIFPFLAGS: 870 case SIOCGIFPFLAGS: 871 case SIOCSIFFLAGS: 872 err = devinet_ioctl(net, cmd, (void __user *)arg); 873 break; 874 default: 875 if (sk->sk_prot->ioctl) 876 err = sk->sk_prot->ioctl(sk, cmd, arg); 877 else 878 err = -ENOIOCTLCMD; 879 break; 880 } 881 return err; 882 } 883 EXPORT_SYMBOL(inet_ioctl); 884 885 #ifdef CONFIG_COMPAT 886 static int inet_compat_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg) 887 { 888 struct sock *sk = sock->sk; 889 int err = -ENOIOCTLCMD; 890 891 if (sk->sk_prot->compat_ioctl) 892 err = sk->sk_prot->compat_ioctl(sk, cmd, arg); 893 894 return err; 895 } 896 #endif 897 898 const struct proto_ops inet_stream_ops = { 899 .family = PF_INET, 900 .owner = THIS_MODULE, 901 .release = inet_release, 902 .bind = inet_bind, 903 .connect = inet_stream_connect, 904 .socketpair = sock_no_socketpair, 905 .accept = inet_accept, 906 .getname = inet_getname, 907 .poll = tcp_poll, 908 .ioctl = inet_ioctl, 909 .listen = inet_listen, 910 .shutdown = inet_shutdown, 911 .setsockopt = sock_common_setsockopt, 912 .getsockopt = sock_common_getsockopt, 913 .sendmsg = inet_sendmsg, 914 .recvmsg = inet_recvmsg, 915 .mmap = sock_no_mmap, 916 .sendpage = inet_sendpage, 917 .splice_read = tcp_splice_read, 918 .read_sock = tcp_read_sock, 919 .peek_len = tcp_peek_len, 920 #ifdef CONFIG_COMPAT 921 .compat_setsockopt = compat_sock_common_setsockopt, 922 .compat_getsockopt = compat_sock_common_getsockopt, 923 .compat_ioctl = inet_compat_ioctl, 924 #endif 925 }; 926 EXPORT_SYMBOL(inet_stream_ops); 927 928 const struct proto_ops inet_dgram_ops = { 929 .family = PF_INET, 930 .owner = THIS_MODULE, 931 .release = inet_release, 932 .bind = inet_bind, 933 .connect = inet_dgram_connect, 934 .socketpair = sock_no_socketpair, 935 .accept = sock_no_accept, 936 .getname = inet_getname, 937 .poll = udp_poll, 938 .ioctl = inet_ioctl, 939 .listen = sock_no_listen, 940 .shutdown = inet_shutdown, 941 .setsockopt = sock_common_setsockopt, 942 .getsockopt = sock_common_getsockopt, 943 .sendmsg = inet_sendmsg, 944 .recvmsg = inet_recvmsg, 945 .mmap = sock_no_mmap, 946 .sendpage = inet_sendpage, 947 .set_peek_off = sk_set_peek_off, 948 #ifdef CONFIG_COMPAT 949 .compat_setsockopt = compat_sock_common_setsockopt, 950 .compat_getsockopt = compat_sock_common_getsockopt, 951 .compat_ioctl = inet_compat_ioctl, 952 #endif 953 }; 954 EXPORT_SYMBOL(inet_dgram_ops); 955 956 /* 957 * For SOCK_RAW sockets; should be the same as inet_dgram_ops but without 958 * udp_poll 959 */ 960 static const struct proto_ops inet_sockraw_ops = { 961 .family = PF_INET, 962 .owner = THIS_MODULE, 963 .release = inet_release, 964 .bind = inet_bind, 965 .connect = inet_dgram_connect, 966 .socketpair = sock_no_socketpair, 967 .accept = sock_no_accept, 968 .getname = inet_getname, 969 .poll = datagram_poll, 970 .ioctl = inet_ioctl, 971 .listen = sock_no_listen, 972 .shutdown = inet_shutdown, 973 .setsockopt = sock_common_setsockopt, 974 .getsockopt = sock_common_getsockopt, 975 .sendmsg = inet_sendmsg, 976 .recvmsg = inet_recvmsg, 977 .mmap = sock_no_mmap, 978 .sendpage = inet_sendpage, 979 #ifdef CONFIG_COMPAT 980 .compat_setsockopt = compat_sock_common_setsockopt, 981 .compat_getsockopt = compat_sock_common_getsockopt, 982 .compat_ioctl = inet_compat_ioctl, 983 #endif 984 }; 985 986 static const struct net_proto_family inet_family_ops = { 987 .family = PF_INET, 988 .create = inet_create, 989 .owner = THIS_MODULE, 990 }; 991 992 /* Upon startup we insert all the elements in inetsw_array[] into 993 * the linked list inetsw. 994 */ 995 static struct inet_protosw inetsw_array[] = 996 { 997 { 998 .type = SOCK_STREAM, 999 .protocol = IPPROTO_TCP, 1000 .prot = &tcp_prot, 1001 .ops = &inet_stream_ops, 1002 .flags = INET_PROTOSW_PERMANENT | 1003 INET_PROTOSW_ICSK, 1004 }, 1005 1006 { 1007 .type = SOCK_DGRAM, 1008 .protocol = IPPROTO_UDP, 1009 .prot = &udp_prot, 1010 .ops = &inet_dgram_ops, 1011 .flags = INET_PROTOSW_PERMANENT, 1012 }, 1013 1014 { 1015 .type = SOCK_DGRAM, 1016 .protocol = IPPROTO_ICMP, 1017 .prot = &ping_prot, 1018 .ops = &inet_dgram_ops, 1019 .flags = INET_PROTOSW_REUSE, 1020 }, 1021 1022 { 1023 .type = SOCK_RAW, 1024 .protocol = IPPROTO_IP, /* wild card */ 1025 .prot = &raw_prot, 1026 .ops = &inet_sockraw_ops, 1027 .flags = INET_PROTOSW_REUSE, 1028 } 1029 }; 1030 1031 #define INETSW_ARRAY_LEN ARRAY_SIZE(inetsw_array) 1032 1033 void inet_register_protosw(struct inet_protosw *p) 1034 { 1035 struct list_head *lh; 1036 struct inet_protosw *answer; 1037 int protocol = p->protocol; 1038 struct list_head *last_perm; 1039 1040 spin_lock_bh(&inetsw_lock); 1041 1042 if (p->type >= SOCK_MAX) 1043 goto out_illegal; 1044 1045 /* If we are trying to override a permanent protocol, bail. */ 1046 last_perm = &inetsw[p->type]; 1047 list_for_each(lh, &inetsw[p->type]) { 1048 answer = list_entry(lh, struct inet_protosw, list); 1049 /* Check only the non-wild match. */ 1050 if ((INET_PROTOSW_PERMANENT & answer->flags) == 0) 1051 break; 1052 if (protocol == answer->protocol) 1053 goto out_permanent; 1054 last_perm = lh; 1055 } 1056 1057 /* Add the new entry after the last permanent entry if any, so that 1058 * the new entry does not override a permanent entry when matched with 1059 * a wild-card protocol. But it is allowed to override any existing 1060 * non-permanent entry. This means that when we remove this entry, the 1061 * system automatically returns to the old behavior. 1062 */ 1063 list_add_rcu(&p->list, last_perm); 1064 out: 1065 spin_unlock_bh(&inetsw_lock); 1066 1067 return; 1068 1069 out_permanent: 1070 pr_err("Attempt to override permanent protocol %d\n", protocol); 1071 goto out; 1072 1073 out_illegal: 1074 pr_err("Ignoring attempt to register invalid socket type %d\n", 1075 p->type); 1076 goto out; 1077 } 1078 EXPORT_SYMBOL(inet_register_protosw); 1079 1080 void inet_unregister_protosw(struct inet_protosw *p) 1081 { 1082 if (INET_PROTOSW_PERMANENT & p->flags) { 1083 pr_err("Attempt to unregister permanent protocol %d\n", 1084 p->protocol); 1085 } else { 1086 spin_lock_bh(&inetsw_lock); 1087 list_del_rcu(&p->list); 1088 spin_unlock_bh(&inetsw_lock); 1089 1090 synchronize_net(); 1091 } 1092 } 1093 EXPORT_SYMBOL(inet_unregister_protosw); 1094 1095 static int inet_sk_reselect_saddr(struct sock *sk) 1096 { 1097 struct inet_sock *inet = inet_sk(sk); 1098 __be32 old_saddr = inet->inet_saddr; 1099 __be32 daddr = inet->inet_daddr; 1100 struct flowi4 *fl4; 1101 struct rtable *rt; 1102 __be32 new_saddr; 1103 struct ip_options_rcu *inet_opt; 1104 1105 inet_opt = rcu_dereference_protected(inet->inet_opt, 1106 lockdep_sock_is_held(sk)); 1107 if (inet_opt && inet_opt->opt.srr) 1108 daddr = inet_opt->opt.faddr; 1109 1110 /* Query new route. */ 1111 fl4 = &inet->cork.fl.u.ip4; 1112 rt = ip_route_connect(fl4, daddr, 0, RT_CONN_FLAGS(sk), 1113 sk->sk_bound_dev_if, sk->sk_protocol, 1114 inet->inet_sport, inet->inet_dport, sk); 1115 if (IS_ERR(rt)) 1116 return PTR_ERR(rt); 1117 1118 sk_setup_caps(sk, &rt->dst); 1119 1120 new_saddr = fl4->saddr; 1121 1122 if (new_saddr == old_saddr) 1123 return 0; 1124 1125 if (sock_net(sk)->ipv4.sysctl_ip_dynaddr > 1) { 1126 pr_info("%s(): shifting inet->saddr from %pI4 to %pI4\n", 1127 __func__, &old_saddr, &new_saddr); 1128 } 1129 1130 inet->inet_saddr = inet->inet_rcv_saddr = new_saddr; 1131 1132 /* 1133 * XXX The only one ugly spot where we need to 1134 * XXX really change the sockets identity after 1135 * XXX it has entered the hashes. -DaveM 1136 * 1137 * Besides that, it does not check for connection 1138 * uniqueness. Wait for troubles. 1139 */ 1140 return __sk_prot_rehash(sk); 1141 } 1142 1143 int inet_sk_rebuild_header(struct sock *sk) 1144 { 1145 struct inet_sock *inet = inet_sk(sk); 1146 struct rtable *rt = (struct rtable *)__sk_dst_check(sk, 0); 1147 __be32 daddr; 1148 struct ip_options_rcu *inet_opt; 1149 struct flowi4 *fl4; 1150 int err; 1151 1152 /* Route is OK, nothing to do. */ 1153 if (rt) 1154 return 0; 1155 1156 /* Reroute. */ 1157 rcu_read_lock(); 1158 inet_opt = rcu_dereference(inet->inet_opt); 1159 daddr = inet->inet_daddr; 1160 if (inet_opt && inet_opt->opt.srr) 1161 daddr = inet_opt->opt.faddr; 1162 rcu_read_unlock(); 1163 fl4 = &inet->cork.fl.u.ip4; 1164 rt = ip_route_output_ports(sock_net(sk), fl4, sk, daddr, inet->inet_saddr, 1165 inet->inet_dport, inet->inet_sport, 1166 sk->sk_protocol, RT_CONN_FLAGS(sk), 1167 sk->sk_bound_dev_if); 1168 if (!IS_ERR(rt)) { 1169 err = 0; 1170 sk_setup_caps(sk, &rt->dst); 1171 } else { 1172 err = PTR_ERR(rt); 1173 1174 /* Routing failed... */ 1175 sk->sk_route_caps = 0; 1176 /* 1177 * Other protocols have to map its equivalent state to TCP_SYN_SENT. 1178 * DCCP maps its DCCP_REQUESTING state to TCP_SYN_SENT. -acme 1179 */ 1180 if (!sock_net(sk)->ipv4.sysctl_ip_dynaddr || 1181 sk->sk_state != TCP_SYN_SENT || 1182 (sk->sk_userlocks & SOCK_BINDADDR_LOCK) || 1183 (err = inet_sk_reselect_saddr(sk)) != 0) 1184 sk->sk_err_soft = -err; 1185 } 1186 1187 return err; 1188 } 1189 EXPORT_SYMBOL(inet_sk_rebuild_header); 1190 1191 struct sk_buff *inet_gso_segment(struct sk_buff *skb, 1192 netdev_features_t features) 1193 { 1194 bool udpfrag = false, fixedid = false, gso_partial, encap; 1195 struct sk_buff *segs = ERR_PTR(-EINVAL); 1196 const struct net_offload *ops; 1197 unsigned int offset = 0; 1198 struct iphdr *iph; 1199 int proto, tot_len; 1200 int nhoff; 1201 int ihl; 1202 int id; 1203 1204 skb_reset_network_header(skb); 1205 nhoff = skb_network_header(skb) - skb_mac_header(skb); 1206 if (unlikely(!pskb_may_pull(skb, sizeof(*iph)))) 1207 goto out; 1208 1209 iph = ip_hdr(skb); 1210 ihl = iph->ihl * 4; 1211 if (ihl < sizeof(*iph)) 1212 goto out; 1213 1214 id = ntohs(iph->id); 1215 proto = iph->protocol; 1216 1217 /* Warning: after this point, iph might be no longer valid */ 1218 if (unlikely(!pskb_may_pull(skb, ihl))) 1219 goto out; 1220 __skb_pull(skb, ihl); 1221 1222 encap = SKB_GSO_CB(skb)->encap_level > 0; 1223 if (encap) 1224 features &= skb->dev->hw_enc_features; 1225 SKB_GSO_CB(skb)->encap_level += ihl; 1226 1227 skb_reset_transport_header(skb); 1228 1229 segs = ERR_PTR(-EPROTONOSUPPORT); 1230 1231 if (!skb->encapsulation || encap) { 1232 udpfrag = !!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP); 1233 fixedid = !!(skb_shinfo(skb)->gso_type & SKB_GSO_TCP_FIXEDID); 1234 1235 /* fixed ID is invalid if DF bit is not set */ 1236 if (fixedid && !(iph->frag_off & htons(IP_DF))) 1237 goto out; 1238 } 1239 1240 ops = rcu_dereference(inet_offloads[proto]); 1241 if (likely(ops && ops->callbacks.gso_segment)) 1242 segs = ops->callbacks.gso_segment(skb, features); 1243 1244 if (IS_ERR_OR_NULL(segs)) 1245 goto out; 1246 1247 gso_partial = !!(skb_shinfo(segs)->gso_type & SKB_GSO_PARTIAL); 1248 1249 skb = segs; 1250 do { 1251 iph = (struct iphdr *)(skb_mac_header(skb) + nhoff); 1252 if (udpfrag) { 1253 iph->frag_off = htons(offset >> 3); 1254 if (skb->next) 1255 iph->frag_off |= htons(IP_MF); 1256 offset += skb->len - nhoff - ihl; 1257 tot_len = skb->len - nhoff; 1258 } else if (skb_is_gso(skb)) { 1259 if (!fixedid) { 1260 iph->id = htons(id); 1261 id += skb_shinfo(skb)->gso_segs; 1262 } 1263 1264 if (gso_partial) 1265 tot_len = skb_shinfo(skb)->gso_size + 1266 SKB_GSO_CB(skb)->data_offset + 1267 skb->head - (unsigned char *)iph; 1268 else 1269 tot_len = skb->len - nhoff; 1270 } else { 1271 if (!fixedid) 1272 iph->id = htons(id++); 1273 tot_len = skb->len - nhoff; 1274 } 1275 iph->tot_len = htons(tot_len); 1276 ip_send_check(iph); 1277 if (encap) 1278 skb_reset_inner_headers(skb); 1279 skb->network_header = (u8 *)iph - skb->head; 1280 } while ((skb = skb->next)); 1281 1282 out: 1283 return segs; 1284 } 1285 EXPORT_SYMBOL(inet_gso_segment); 1286 1287 struct sk_buff **inet_gro_receive(struct sk_buff **head, struct sk_buff *skb) 1288 { 1289 const struct net_offload *ops; 1290 struct sk_buff **pp = NULL; 1291 struct sk_buff *p; 1292 const struct iphdr *iph; 1293 unsigned int hlen; 1294 unsigned int off; 1295 unsigned int id; 1296 int flush = 1; 1297 int proto; 1298 1299 off = skb_gro_offset(skb); 1300 hlen = off + sizeof(*iph); 1301 iph = skb_gro_header_fast(skb, off); 1302 if (skb_gro_header_hard(skb, hlen)) { 1303 iph = skb_gro_header_slow(skb, hlen, off); 1304 if (unlikely(!iph)) 1305 goto out; 1306 } 1307 1308 proto = iph->protocol; 1309 1310 rcu_read_lock(); 1311 ops = rcu_dereference(inet_offloads[proto]); 1312 if (!ops || !ops->callbacks.gro_receive) 1313 goto out_unlock; 1314 1315 if (*(u8 *)iph != 0x45) 1316 goto out_unlock; 1317 1318 if (unlikely(ip_fast_csum((u8 *)iph, 5))) 1319 goto out_unlock; 1320 1321 id = ntohl(*(__be32 *)&iph->id); 1322 flush = (u16)((ntohl(*(__be32 *)iph) ^ skb_gro_len(skb)) | (id & ~IP_DF)); 1323 id >>= 16; 1324 1325 for (p = *head; p; p = p->next) { 1326 struct iphdr *iph2; 1327 u16 flush_id; 1328 1329 if (!NAPI_GRO_CB(p)->same_flow) 1330 continue; 1331 1332 iph2 = (struct iphdr *)(p->data + off); 1333 /* The above works because, with the exception of the top 1334 * (inner most) layer, we only aggregate pkts with the same 1335 * hdr length so all the hdrs we'll need to verify will start 1336 * at the same offset. 1337 */ 1338 if ((iph->protocol ^ iph2->protocol) | 1339 ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) | 1340 ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) { 1341 NAPI_GRO_CB(p)->same_flow = 0; 1342 continue; 1343 } 1344 1345 /* All fields must match except length and checksum. */ 1346 NAPI_GRO_CB(p)->flush |= 1347 (iph->ttl ^ iph2->ttl) | 1348 (iph->tos ^ iph2->tos) | 1349 ((iph->frag_off ^ iph2->frag_off) & htons(IP_DF)); 1350 1351 NAPI_GRO_CB(p)->flush |= flush; 1352 1353 /* We need to store of the IP ID check to be included later 1354 * when we can verify that this packet does in fact belong 1355 * to a given flow. 1356 */ 1357 flush_id = (u16)(id - ntohs(iph2->id)); 1358 1359 /* This bit of code makes it much easier for us to identify 1360 * the cases where we are doing atomic vs non-atomic IP ID 1361 * checks. Specifically an atomic check can return IP ID 1362 * values 0 - 0xFFFF, while a non-atomic check can only 1363 * return 0 or 0xFFFF. 1364 */ 1365 if (!NAPI_GRO_CB(p)->is_atomic || 1366 !(iph->frag_off & htons(IP_DF))) { 1367 flush_id ^= NAPI_GRO_CB(p)->count; 1368 flush_id = flush_id ? 0xFFFF : 0; 1369 } 1370 1371 /* If the previous IP ID value was based on an atomic 1372 * datagram we can overwrite the value and ignore it. 1373 */ 1374 if (NAPI_GRO_CB(skb)->is_atomic) 1375 NAPI_GRO_CB(p)->flush_id = flush_id; 1376 else 1377 NAPI_GRO_CB(p)->flush_id |= flush_id; 1378 } 1379 1380 NAPI_GRO_CB(skb)->is_atomic = !!(iph->frag_off & htons(IP_DF)); 1381 NAPI_GRO_CB(skb)->flush |= flush; 1382 skb_set_network_header(skb, off); 1383 /* The above will be needed by the transport layer if there is one 1384 * immediately following this IP hdr. 1385 */ 1386 1387 /* Note : No need to call skb_gro_postpull_rcsum() here, 1388 * as we already checked checksum over ipv4 header was 0 1389 */ 1390 skb_gro_pull(skb, sizeof(*iph)); 1391 skb_set_transport_header(skb, skb_gro_offset(skb)); 1392 1393 pp = call_gro_receive(ops->callbacks.gro_receive, head, skb); 1394 1395 out_unlock: 1396 rcu_read_unlock(); 1397 1398 out: 1399 NAPI_GRO_CB(skb)->flush |= flush; 1400 1401 return pp; 1402 } 1403 EXPORT_SYMBOL(inet_gro_receive); 1404 1405 static struct sk_buff **ipip_gro_receive(struct sk_buff **head, 1406 struct sk_buff *skb) 1407 { 1408 if (NAPI_GRO_CB(skb)->encap_mark) { 1409 NAPI_GRO_CB(skb)->flush = 1; 1410 return NULL; 1411 } 1412 1413 NAPI_GRO_CB(skb)->encap_mark = 1; 1414 1415 return inet_gro_receive(head, skb); 1416 } 1417 1418 #define SECONDS_PER_DAY 86400 1419 1420 /* inet_current_timestamp - Return IP network timestamp 1421 * 1422 * Return milliseconds since midnight in network byte order. 1423 */ 1424 __be32 inet_current_timestamp(void) 1425 { 1426 u32 secs; 1427 u32 msecs; 1428 struct timespec64 ts; 1429 1430 ktime_get_real_ts64(&ts); 1431 1432 /* Get secs since midnight. */ 1433 (void)div_u64_rem(ts.tv_sec, SECONDS_PER_DAY, &secs); 1434 /* Convert to msecs. */ 1435 msecs = secs * MSEC_PER_SEC; 1436 /* Convert nsec to msec. */ 1437 msecs += (u32)ts.tv_nsec / NSEC_PER_MSEC; 1438 1439 /* Convert to network byte order. */ 1440 return htonl(msecs); 1441 } 1442 EXPORT_SYMBOL(inet_current_timestamp); 1443 1444 int inet_recv_error(struct sock *sk, struct msghdr *msg, int len, int *addr_len) 1445 { 1446 if (sk->sk_family == AF_INET) 1447 return ip_recv_error(sk, msg, len, addr_len); 1448 #if IS_ENABLED(CONFIG_IPV6) 1449 if (sk->sk_family == AF_INET6) 1450 return pingv6_ops.ipv6_recv_error(sk, msg, len, addr_len); 1451 #endif 1452 return -EINVAL; 1453 } 1454 1455 int inet_gro_complete(struct sk_buff *skb, int nhoff) 1456 { 1457 __be16 newlen = htons(skb->len - nhoff); 1458 struct iphdr *iph = (struct iphdr *)(skb->data + nhoff); 1459 const struct net_offload *ops; 1460 int proto = iph->protocol; 1461 int err = -ENOSYS; 1462 1463 if (skb->encapsulation) 1464 skb_set_inner_network_header(skb, nhoff); 1465 1466 csum_replace2(&iph->check, iph->tot_len, newlen); 1467 iph->tot_len = newlen; 1468 1469 rcu_read_lock(); 1470 ops = rcu_dereference(inet_offloads[proto]); 1471 if (WARN_ON(!ops || !ops->callbacks.gro_complete)) 1472 goto out_unlock; 1473 1474 /* Only need to add sizeof(*iph) to get to the next hdr below 1475 * because any hdr with option will have been flushed in 1476 * inet_gro_receive(). 1477 */ 1478 err = ops->callbacks.gro_complete(skb, nhoff + sizeof(*iph)); 1479 1480 out_unlock: 1481 rcu_read_unlock(); 1482 1483 return err; 1484 } 1485 EXPORT_SYMBOL(inet_gro_complete); 1486 1487 static int ipip_gro_complete(struct sk_buff *skb, int nhoff) 1488 { 1489 skb->encapsulation = 1; 1490 skb_shinfo(skb)->gso_type |= SKB_GSO_IPXIP4; 1491 return inet_gro_complete(skb, nhoff); 1492 } 1493 1494 int inet_ctl_sock_create(struct sock **sk, unsigned short family, 1495 unsigned short type, unsigned char protocol, 1496 struct net *net) 1497 { 1498 struct socket *sock; 1499 int rc = sock_create_kern(net, family, type, protocol, &sock); 1500 1501 if (rc == 0) { 1502 *sk = sock->sk; 1503 (*sk)->sk_allocation = GFP_ATOMIC; 1504 /* 1505 * Unhash it so that IP input processing does not even see it, 1506 * we do not wish this socket to see incoming packets. 1507 */ 1508 (*sk)->sk_prot->unhash(*sk); 1509 } 1510 return rc; 1511 } 1512 EXPORT_SYMBOL_GPL(inet_ctl_sock_create); 1513 1514 u64 snmp_get_cpu_field(void __percpu *mib, int cpu, int offt) 1515 { 1516 return *(((unsigned long *)per_cpu_ptr(mib, cpu)) + offt); 1517 } 1518 EXPORT_SYMBOL_GPL(snmp_get_cpu_field); 1519 1520 unsigned long snmp_fold_field(void __percpu *mib, int offt) 1521 { 1522 unsigned long res = 0; 1523 int i; 1524 1525 for_each_possible_cpu(i) 1526 res += snmp_get_cpu_field(mib, i, offt); 1527 return res; 1528 } 1529 EXPORT_SYMBOL_GPL(snmp_fold_field); 1530 1531 #if BITS_PER_LONG==32 1532 1533 u64 snmp_get_cpu_field64(void __percpu *mib, int cpu, int offt, 1534 size_t syncp_offset) 1535 { 1536 void *bhptr; 1537 struct u64_stats_sync *syncp; 1538 u64 v; 1539 unsigned int start; 1540 1541 bhptr = per_cpu_ptr(mib, cpu); 1542 syncp = (struct u64_stats_sync *)(bhptr + syncp_offset); 1543 do { 1544 start = u64_stats_fetch_begin_irq(syncp); 1545 v = *(((u64 *)bhptr) + offt); 1546 } while (u64_stats_fetch_retry_irq(syncp, start)); 1547 1548 return v; 1549 } 1550 EXPORT_SYMBOL_GPL(snmp_get_cpu_field64); 1551 1552 u64 snmp_fold_field64(void __percpu *mib, int offt, size_t syncp_offset) 1553 { 1554 u64 res = 0; 1555 int cpu; 1556 1557 for_each_possible_cpu(cpu) { 1558 res += snmp_get_cpu_field64(mib, cpu, offt, syncp_offset); 1559 } 1560 return res; 1561 } 1562 EXPORT_SYMBOL_GPL(snmp_fold_field64); 1563 #endif 1564 1565 #ifdef CONFIG_IP_MULTICAST 1566 static const struct net_protocol igmp_protocol = { 1567 .handler = igmp_rcv, 1568 .netns_ok = 1, 1569 }; 1570 #endif 1571 1572 static const struct net_protocol tcp_protocol = { 1573 .early_demux = tcp_v4_early_demux, 1574 .handler = tcp_v4_rcv, 1575 .err_handler = tcp_v4_err, 1576 .no_policy = 1, 1577 .netns_ok = 1, 1578 .icmp_strict_tag_validation = 1, 1579 }; 1580 1581 static const struct net_protocol udp_protocol = { 1582 .early_demux = udp_v4_early_demux, 1583 .handler = udp_rcv, 1584 .err_handler = udp_err, 1585 .no_policy = 1, 1586 .netns_ok = 1, 1587 }; 1588 1589 static const struct net_protocol icmp_protocol = { 1590 .handler = icmp_rcv, 1591 .err_handler = icmp_err, 1592 .no_policy = 1, 1593 .netns_ok = 1, 1594 }; 1595 1596 static __net_init int ipv4_mib_init_net(struct net *net) 1597 { 1598 int i; 1599 1600 net->mib.tcp_statistics = alloc_percpu(struct tcp_mib); 1601 if (!net->mib.tcp_statistics) 1602 goto err_tcp_mib; 1603 net->mib.ip_statistics = alloc_percpu(struct ipstats_mib); 1604 if (!net->mib.ip_statistics) 1605 goto err_ip_mib; 1606 1607 for_each_possible_cpu(i) { 1608 struct ipstats_mib *af_inet_stats; 1609 af_inet_stats = per_cpu_ptr(net->mib.ip_statistics, i); 1610 u64_stats_init(&af_inet_stats->syncp); 1611 } 1612 1613 net->mib.net_statistics = alloc_percpu(struct linux_mib); 1614 if (!net->mib.net_statistics) 1615 goto err_net_mib; 1616 net->mib.udp_statistics = alloc_percpu(struct udp_mib); 1617 if (!net->mib.udp_statistics) 1618 goto err_udp_mib; 1619 net->mib.udplite_statistics = alloc_percpu(struct udp_mib); 1620 if (!net->mib.udplite_statistics) 1621 goto err_udplite_mib; 1622 net->mib.icmp_statistics = alloc_percpu(struct icmp_mib); 1623 if (!net->mib.icmp_statistics) 1624 goto err_icmp_mib; 1625 net->mib.icmpmsg_statistics = kzalloc(sizeof(struct icmpmsg_mib), 1626 GFP_KERNEL); 1627 if (!net->mib.icmpmsg_statistics) 1628 goto err_icmpmsg_mib; 1629 1630 tcp_mib_init(net); 1631 return 0; 1632 1633 err_icmpmsg_mib: 1634 free_percpu(net->mib.icmp_statistics); 1635 err_icmp_mib: 1636 free_percpu(net->mib.udplite_statistics); 1637 err_udplite_mib: 1638 free_percpu(net->mib.udp_statistics); 1639 err_udp_mib: 1640 free_percpu(net->mib.net_statistics); 1641 err_net_mib: 1642 free_percpu(net->mib.ip_statistics); 1643 err_ip_mib: 1644 free_percpu(net->mib.tcp_statistics); 1645 err_tcp_mib: 1646 return -ENOMEM; 1647 } 1648 1649 static __net_exit void ipv4_mib_exit_net(struct net *net) 1650 { 1651 kfree(net->mib.icmpmsg_statistics); 1652 free_percpu(net->mib.icmp_statistics); 1653 free_percpu(net->mib.udplite_statistics); 1654 free_percpu(net->mib.udp_statistics); 1655 free_percpu(net->mib.net_statistics); 1656 free_percpu(net->mib.ip_statistics); 1657 free_percpu(net->mib.tcp_statistics); 1658 } 1659 1660 static __net_initdata struct pernet_operations ipv4_mib_ops = { 1661 .init = ipv4_mib_init_net, 1662 .exit = ipv4_mib_exit_net, 1663 }; 1664 1665 static int __init init_ipv4_mibs(void) 1666 { 1667 return register_pernet_subsys(&ipv4_mib_ops); 1668 } 1669 1670 static __net_init int inet_init_net(struct net *net) 1671 { 1672 /* 1673 * Set defaults for local port range 1674 */ 1675 seqlock_init(&net->ipv4.ip_local_ports.lock); 1676 net->ipv4.ip_local_ports.range[0] = 32768; 1677 net->ipv4.ip_local_ports.range[1] = 60999; 1678 1679 seqlock_init(&net->ipv4.ping_group_range.lock); 1680 /* 1681 * Sane defaults - nobody may create ping sockets. 1682 * Boot scripts should set this to distro-specific group. 1683 */ 1684 net->ipv4.ping_group_range.range[0] = make_kgid(&init_user_ns, 1); 1685 net->ipv4.ping_group_range.range[1] = make_kgid(&init_user_ns, 0); 1686 1687 /* Default values for sysctl-controlled parameters. 1688 * We set them here, in case sysctl is not compiled. 1689 */ 1690 net->ipv4.sysctl_ip_default_ttl = IPDEFTTL; 1691 net->ipv4.sysctl_ip_dynaddr = 0; 1692 net->ipv4.sysctl_ip_early_demux = 1; 1693 1694 return 0; 1695 } 1696 1697 static __net_exit void inet_exit_net(struct net *net) 1698 { 1699 } 1700 1701 static __net_initdata struct pernet_operations af_inet_ops = { 1702 .init = inet_init_net, 1703 .exit = inet_exit_net, 1704 }; 1705 1706 static int __init init_inet_pernet_ops(void) 1707 { 1708 return register_pernet_subsys(&af_inet_ops); 1709 } 1710 1711 static int ipv4_proc_init(void); 1712 1713 /* 1714 * IP protocol layer initialiser 1715 */ 1716 1717 static struct packet_offload ip_packet_offload __read_mostly = { 1718 .type = cpu_to_be16(ETH_P_IP), 1719 .callbacks = { 1720 .gso_segment = inet_gso_segment, 1721 .gro_receive = inet_gro_receive, 1722 .gro_complete = inet_gro_complete, 1723 }, 1724 }; 1725 1726 static const struct net_offload ipip_offload = { 1727 .callbacks = { 1728 .gso_segment = inet_gso_segment, 1729 .gro_receive = ipip_gro_receive, 1730 .gro_complete = ipip_gro_complete, 1731 }, 1732 }; 1733 1734 static int __init ipv4_offload_init(void) 1735 { 1736 /* 1737 * Add offloads 1738 */ 1739 if (udpv4_offload_init() < 0) 1740 pr_crit("%s: Cannot add UDP protocol offload\n", __func__); 1741 if (tcpv4_offload_init() < 0) 1742 pr_crit("%s: Cannot add TCP protocol offload\n", __func__); 1743 1744 dev_add_offload(&ip_packet_offload); 1745 inet_add_offload(&ipip_offload, IPPROTO_IPIP); 1746 return 0; 1747 } 1748 1749 fs_initcall(ipv4_offload_init); 1750 1751 static struct packet_type ip_packet_type __read_mostly = { 1752 .type = cpu_to_be16(ETH_P_IP), 1753 .func = ip_rcv, 1754 }; 1755 1756 static int __init inet_init(void) 1757 { 1758 struct inet_protosw *q; 1759 struct list_head *r; 1760 int rc = -EINVAL; 1761 1762 sock_skb_cb_check_size(sizeof(struct inet_skb_parm)); 1763 1764 rc = proto_register(&tcp_prot, 1); 1765 if (rc) 1766 goto out; 1767 1768 rc = proto_register(&udp_prot, 1); 1769 if (rc) 1770 goto out_unregister_tcp_proto; 1771 1772 rc = proto_register(&raw_prot, 1); 1773 if (rc) 1774 goto out_unregister_udp_proto; 1775 1776 rc = proto_register(&ping_prot, 1); 1777 if (rc) 1778 goto out_unregister_raw_proto; 1779 1780 /* 1781 * Tell SOCKET that we are alive... 1782 */ 1783 1784 (void)sock_register(&inet_family_ops); 1785 1786 #ifdef CONFIG_SYSCTL 1787 ip_static_sysctl_init(); 1788 #endif 1789 1790 /* 1791 * Add all the base protocols. 1792 */ 1793 1794 if (inet_add_protocol(&icmp_protocol, IPPROTO_ICMP) < 0) 1795 pr_crit("%s: Cannot add ICMP protocol\n", __func__); 1796 if (inet_add_protocol(&udp_protocol, IPPROTO_UDP) < 0) 1797 pr_crit("%s: Cannot add UDP protocol\n", __func__); 1798 if (inet_add_protocol(&tcp_protocol, IPPROTO_TCP) < 0) 1799 pr_crit("%s: Cannot add TCP protocol\n", __func__); 1800 #ifdef CONFIG_IP_MULTICAST 1801 if (inet_add_protocol(&igmp_protocol, IPPROTO_IGMP) < 0) 1802 pr_crit("%s: Cannot add IGMP protocol\n", __func__); 1803 #endif 1804 1805 /* Register the socket-side information for inet_create. */ 1806 for (r = &inetsw[0]; r < &inetsw[SOCK_MAX]; ++r) 1807 INIT_LIST_HEAD(r); 1808 1809 for (q = inetsw_array; q < &inetsw_array[INETSW_ARRAY_LEN]; ++q) 1810 inet_register_protosw(q); 1811 1812 /* 1813 * Set the ARP module up 1814 */ 1815 1816 arp_init(); 1817 1818 /* 1819 * Set the IP module up 1820 */ 1821 1822 ip_init(); 1823 1824 tcp_v4_init(); 1825 1826 /* Setup TCP slab cache for open requests. */ 1827 tcp_init(); 1828 1829 /* Setup UDP memory threshold */ 1830 udp_init(); 1831 1832 /* Add UDP-Lite (RFC 3828) */ 1833 udplite4_register(); 1834 1835 ping_init(); 1836 1837 /* 1838 * Set the ICMP layer up 1839 */ 1840 1841 if (icmp_init() < 0) 1842 panic("Failed to create the ICMP control socket.\n"); 1843 1844 /* 1845 * Initialise the multicast router 1846 */ 1847 #if defined(CONFIG_IP_MROUTE) 1848 if (ip_mr_init()) 1849 pr_crit("%s: Cannot init ipv4 mroute\n", __func__); 1850 #endif 1851 1852 if (init_inet_pernet_ops()) 1853 pr_crit("%s: Cannot init ipv4 inet pernet ops\n", __func__); 1854 /* 1855 * Initialise per-cpu ipv4 mibs 1856 */ 1857 1858 if (init_ipv4_mibs()) 1859 pr_crit("%s: Cannot init ipv4 mibs\n", __func__); 1860 1861 ipv4_proc_init(); 1862 1863 ipfrag_init(); 1864 1865 dev_add_pack(&ip_packet_type); 1866 1867 ip_tunnel_core_init(); 1868 1869 rc = 0; 1870 out: 1871 return rc; 1872 out_unregister_raw_proto: 1873 proto_unregister(&raw_prot); 1874 out_unregister_udp_proto: 1875 proto_unregister(&udp_prot); 1876 out_unregister_tcp_proto: 1877 proto_unregister(&tcp_prot); 1878 goto out; 1879 } 1880 1881 fs_initcall(inet_init); 1882 1883 /* ------------------------------------------------------------------------ */ 1884 1885 #ifdef CONFIG_PROC_FS 1886 static int __init ipv4_proc_init(void) 1887 { 1888 int rc = 0; 1889 1890 if (raw_proc_init()) 1891 goto out_raw; 1892 if (tcp4_proc_init()) 1893 goto out_tcp; 1894 if (udp4_proc_init()) 1895 goto out_udp; 1896 if (ping_proc_init()) 1897 goto out_ping; 1898 if (ip_misc_proc_init()) 1899 goto out_misc; 1900 out: 1901 return rc; 1902 out_misc: 1903 ping_proc_exit(); 1904 out_ping: 1905 udp4_proc_exit(); 1906 out_udp: 1907 tcp4_proc_exit(); 1908 out_tcp: 1909 raw_proc_exit(); 1910 out_raw: 1911 rc = -ENOMEM; 1912 goto out; 1913 } 1914 1915 #else /* CONFIG_PROC_FS */ 1916 static int __init ipv4_proc_init(void) 1917 { 1918 return 0; 1919 } 1920 #endif /* CONFIG_PROC_FS */ 1921