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 * The User Datagram Protocol (UDP). 7 * 8 * Version: $Id: udp.c,v 1.102 2002/02/01 22:01:04 davem Exp $ 9 * 10 * Authors: Ross Biro 11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 12 * Arnt Gulbrandsen, <agulbra@nvg.unit.no> 13 * Alan Cox, <Alan.Cox@linux.org> 14 * Hirokazu Takahashi, <taka@valinux.co.jp> 15 * 16 * Fixes: 17 * Alan Cox : verify_area() calls 18 * Alan Cox : stopped close while in use off icmp 19 * messages. Not a fix but a botch that 20 * for udp at least is 'valid'. 21 * Alan Cox : Fixed icmp handling properly 22 * Alan Cox : Correct error for oversized datagrams 23 * Alan Cox : Tidied select() semantics. 24 * Alan Cox : udp_err() fixed properly, also now 25 * select and read wake correctly on errors 26 * Alan Cox : udp_send verify_area moved to avoid mem leak 27 * Alan Cox : UDP can count its memory 28 * Alan Cox : send to an unknown connection causes 29 * an ECONNREFUSED off the icmp, but 30 * does NOT close. 31 * Alan Cox : Switched to new sk_buff handlers. No more backlog! 32 * Alan Cox : Using generic datagram code. Even smaller and the PEEK 33 * bug no longer crashes it. 34 * Fred Van Kempen : Net2e support for sk->broadcast. 35 * Alan Cox : Uses skb_free_datagram 36 * Alan Cox : Added get/set sockopt support. 37 * Alan Cox : Broadcasting without option set returns EACCES. 38 * Alan Cox : No wakeup calls. Instead we now use the callbacks. 39 * Alan Cox : Use ip_tos and ip_ttl 40 * Alan Cox : SNMP Mibs 41 * Alan Cox : MSG_DONTROUTE, and 0.0.0.0 support. 42 * Matt Dillon : UDP length checks. 43 * Alan Cox : Smarter af_inet used properly. 44 * Alan Cox : Use new kernel side addressing. 45 * Alan Cox : Incorrect return on truncated datagram receive. 46 * Arnt Gulbrandsen : New udp_send and stuff 47 * Alan Cox : Cache last socket 48 * Alan Cox : Route cache 49 * Jon Peatfield : Minor efficiency fix to sendto(). 50 * Mike Shaver : RFC1122 checks. 51 * Alan Cox : Nonblocking error fix. 52 * Willy Konynenberg : Transparent proxying support. 53 * Mike McLagan : Routing by source 54 * David S. Miller : New socket lookup architecture. 55 * Last socket cache retained as it 56 * does have a high hit rate. 57 * Olaf Kirch : Don't linearise iovec on sendmsg. 58 * Andi Kleen : Some cleanups, cache destination entry 59 * for connect. 60 * Vitaly E. Lavrov : Transparent proxy revived after year coma. 61 * Melvin Smith : Check msg_name not msg_namelen in sendto(), 62 * return ENOTCONN for unconnected sockets (POSIX) 63 * Janos Farkas : don't deliver multi/broadcasts to a different 64 * bound-to-device socket 65 * Hirokazu Takahashi : HW checksumming for outgoing UDP 66 * datagrams. 67 * Hirokazu Takahashi : sendfile() on UDP works now. 68 * Arnaldo C. Melo : convert /proc/net/udp to seq_file 69 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which 70 * Alexey Kuznetsov: allow both IPv4 and IPv6 sockets to bind 71 * a single port at the same time. 72 * Derek Atkins <derek@ihtfp.com>: Add Encapulation Support 73 * James Chapman : Add L2TP encapsulation type. 74 * 75 * 76 * This program is free software; you can redistribute it and/or 77 * modify it under the terms of the GNU General Public License 78 * as published by the Free Software Foundation; either version 79 * 2 of the License, or (at your option) any later version. 80 */ 81 82 #include <asm/system.h> 83 #include <asm/uaccess.h> 84 #include <asm/ioctls.h> 85 #include <linux/bootmem.h> 86 #include <linux/types.h> 87 #include <linux/fcntl.h> 88 #include <linux/module.h> 89 #include <linux/socket.h> 90 #include <linux/sockios.h> 91 #include <linux/igmp.h> 92 #include <linux/in.h> 93 #include <linux/errno.h> 94 #include <linux/timer.h> 95 #include <linux/mm.h> 96 #include <linux/inet.h> 97 #include <linux/netdevice.h> 98 #include <net/tcp_states.h> 99 #include <linux/skbuff.h> 100 #include <linux/proc_fs.h> 101 #include <linux/seq_file.h> 102 #include <net/net_namespace.h> 103 #include <net/icmp.h> 104 #include <net/route.h> 105 #include <net/checksum.h> 106 #include <net/xfrm.h> 107 #include "udp_impl.h" 108 109 /* 110 * Snmp MIB for the UDP layer 111 */ 112 113 DEFINE_SNMP_STAT(struct udp_mib, udp_statistics) __read_mostly; 114 EXPORT_SYMBOL(udp_statistics); 115 116 DEFINE_SNMP_STAT(struct udp_mib, udp_stats_in6) __read_mostly; 117 EXPORT_SYMBOL(udp_stats_in6); 118 119 struct hlist_head udp_hash[UDP_HTABLE_SIZE]; 120 DEFINE_RWLOCK(udp_hash_lock); 121 122 int sysctl_udp_mem[3] __read_mostly; 123 int sysctl_udp_rmem_min __read_mostly; 124 int sysctl_udp_wmem_min __read_mostly; 125 126 EXPORT_SYMBOL(sysctl_udp_mem); 127 EXPORT_SYMBOL(sysctl_udp_rmem_min); 128 EXPORT_SYMBOL(sysctl_udp_wmem_min); 129 130 atomic_t udp_memory_allocated; 131 EXPORT_SYMBOL(udp_memory_allocated); 132 133 static inline int __udp_lib_lport_inuse(struct net *net, __u16 num, 134 const struct hlist_head udptable[]) 135 { 136 struct sock *sk; 137 struct hlist_node *node; 138 139 sk_for_each(sk, node, &udptable[num & (UDP_HTABLE_SIZE - 1)]) 140 if (net_eq(sock_net(sk), net) && sk->sk_hash == num) 141 return 1; 142 return 0; 143 } 144 145 /** 146 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6 147 * 148 * @sk: socket struct in question 149 * @snum: port number to look up 150 * @saddr_comp: AF-dependent comparison of bound local IP addresses 151 */ 152 int udp_lib_get_port(struct sock *sk, unsigned short snum, 153 int (*saddr_comp)(const struct sock *sk1, 154 const struct sock *sk2 ) ) 155 { 156 struct hlist_head *udptable = sk->sk_prot->h.udp_hash; 157 struct hlist_node *node; 158 struct hlist_head *head; 159 struct sock *sk2; 160 int error = 1; 161 struct net *net = sock_net(sk); 162 163 write_lock_bh(&udp_hash_lock); 164 165 if (!snum) { 166 int i, low, high, remaining; 167 unsigned rover, best, best_size_so_far; 168 169 inet_get_local_port_range(&low, &high); 170 remaining = (high - low) + 1; 171 172 best_size_so_far = UINT_MAX; 173 best = rover = net_random() % remaining + low; 174 175 /* 1st pass: look for empty (or shortest) hash chain */ 176 for (i = 0; i < UDP_HTABLE_SIZE; i++) { 177 int size = 0; 178 179 head = &udptable[rover & (UDP_HTABLE_SIZE - 1)]; 180 if (hlist_empty(head)) 181 goto gotit; 182 183 sk_for_each(sk2, node, head) { 184 if (++size >= best_size_so_far) 185 goto next; 186 } 187 best_size_so_far = size; 188 best = rover; 189 next: 190 /* fold back if end of range */ 191 if (++rover > high) 192 rover = low + ((rover - low) 193 & (UDP_HTABLE_SIZE - 1)); 194 195 196 } 197 198 /* 2nd pass: find hole in shortest hash chain */ 199 rover = best; 200 for (i = 0; i < (1 << 16) / UDP_HTABLE_SIZE; i++) { 201 if (! __udp_lib_lport_inuse(net, rover, udptable)) 202 goto gotit; 203 rover += UDP_HTABLE_SIZE; 204 if (rover > high) 205 rover = low + ((rover - low) 206 & (UDP_HTABLE_SIZE - 1)); 207 } 208 209 210 /* All ports in use! */ 211 goto fail; 212 213 gotit: 214 snum = rover; 215 } else { 216 head = &udptable[snum & (UDP_HTABLE_SIZE - 1)]; 217 218 sk_for_each(sk2, node, head) 219 if (sk2->sk_hash == snum && 220 sk2 != sk && 221 net_eq(sock_net(sk2), net) && 222 (!sk2->sk_reuse || !sk->sk_reuse) && 223 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if 224 || sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 225 (*saddr_comp)(sk, sk2) ) 226 goto fail; 227 } 228 229 inet_sk(sk)->num = snum; 230 sk->sk_hash = snum; 231 if (sk_unhashed(sk)) { 232 head = &udptable[snum & (UDP_HTABLE_SIZE - 1)]; 233 sk_add_node(sk, head); 234 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 235 } 236 error = 0; 237 fail: 238 write_unlock_bh(&udp_hash_lock); 239 return error; 240 } 241 242 static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2) 243 { 244 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2); 245 246 return ( !ipv6_only_sock(sk2) && 247 (!inet1->rcv_saddr || !inet2->rcv_saddr || 248 inet1->rcv_saddr == inet2->rcv_saddr )); 249 } 250 251 int udp_v4_get_port(struct sock *sk, unsigned short snum) 252 { 253 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal); 254 } 255 256 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try 257 * harder than this. -DaveM 258 */ 259 static struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, 260 __be16 sport, __be32 daddr, __be16 dport, 261 int dif, struct hlist_head udptable[]) 262 { 263 struct sock *sk, *result = NULL; 264 struct hlist_node *node; 265 unsigned short hnum = ntohs(dport); 266 int badness = -1; 267 268 read_lock(&udp_hash_lock); 269 sk_for_each(sk, node, &udptable[hnum & (UDP_HTABLE_SIZE - 1)]) { 270 struct inet_sock *inet = inet_sk(sk); 271 272 if (net_eq(sock_net(sk), net) && sk->sk_hash == hnum && 273 !ipv6_only_sock(sk)) { 274 int score = (sk->sk_family == PF_INET ? 1 : 0); 275 if (inet->rcv_saddr) { 276 if (inet->rcv_saddr != daddr) 277 continue; 278 score+=2; 279 } 280 if (inet->daddr) { 281 if (inet->daddr != saddr) 282 continue; 283 score+=2; 284 } 285 if (inet->dport) { 286 if (inet->dport != sport) 287 continue; 288 score+=2; 289 } 290 if (sk->sk_bound_dev_if) { 291 if (sk->sk_bound_dev_if != dif) 292 continue; 293 score+=2; 294 } 295 if (score == 9) { 296 result = sk; 297 break; 298 } else if (score > badness) { 299 result = sk; 300 badness = score; 301 } 302 } 303 } 304 if (result) 305 sock_hold(result); 306 read_unlock(&udp_hash_lock); 307 return result; 308 } 309 310 static inline struct sock *udp_v4_mcast_next(struct sock *sk, 311 __be16 loc_port, __be32 loc_addr, 312 __be16 rmt_port, __be32 rmt_addr, 313 int dif) 314 { 315 struct hlist_node *node; 316 struct sock *s = sk; 317 unsigned short hnum = ntohs(loc_port); 318 319 sk_for_each_from(s, node) { 320 struct inet_sock *inet = inet_sk(s); 321 322 if (s->sk_hash != hnum || 323 (inet->daddr && inet->daddr != rmt_addr) || 324 (inet->dport != rmt_port && inet->dport) || 325 (inet->rcv_saddr && inet->rcv_saddr != loc_addr) || 326 ipv6_only_sock(s) || 327 (s->sk_bound_dev_if && s->sk_bound_dev_if != dif)) 328 continue; 329 if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif)) 330 continue; 331 goto found; 332 } 333 s = NULL; 334 found: 335 return s; 336 } 337 338 /* 339 * This routine is called by the ICMP module when it gets some 340 * sort of error condition. If err < 0 then the socket should 341 * be closed and the error returned to the user. If err > 0 342 * it's just the icmp type << 8 | icmp code. 343 * Header points to the ip header of the error packet. We move 344 * on past this. Then (as it used to claim before adjustment) 345 * header points to the first 8 bytes of the udp header. We need 346 * to find the appropriate port. 347 */ 348 349 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct hlist_head udptable[]) 350 { 351 struct inet_sock *inet; 352 struct iphdr *iph = (struct iphdr*)skb->data; 353 struct udphdr *uh = (struct udphdr*)(skb->data+(iph->ihl<<2)); 354 const int type = icmp_hdr(skb)->type; 355 const int code = icmp_hdr(skb)->code; 356 struct sock *sk; 357 int harderr; 358 int err; 359 360 sk = __udp4_lib_lookup(dev_net(skb->dev), iph->daddr, uh->dest, 361 iph->saddr, uh->source, skb->dev->ifindex, udptable); 362 if (sk == NULL) { 363 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS); 364 return; /* No socket for error */ 365 } 366 367 err = 0; 368 harderr = 0; 369 inet = inet_sk(sk); 370 371 switch (type) { 372 default: 373 case ICMP_TIME_EXCEEDED: 374 err = EHOSTUNREACH; 375 break; 376 case ICMP_SOURCE_QUENCH: 377 goto out; 378 case ICMP_PARAMETERPROB: 379 err = EPROTO; 380 harderr = 1; 381 break; 382 case ICMP_DEST_UNREACH: 383 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */ 384 if (inet->pmtudisc != IP_PMTUDISC_DONT) { 385 err = EMSGSIZE; 386 harderr = 1; 387 break; 388 } 389 goto out; 390 } 391 err = EHOSTUNREACH; 392 if (code <= NR_ICMP_UNREACH) { 393 harderr = icmp_err_convert[code].fatal; 394 err = icmp_err_convert[code].errno; 395 } 396 break; 397 } 398 399 /* 400 * RFC1122: OK. Passes ICMP errors back to application, as per 401 * 4.1.3.3. 402 */ 403 if (!inet->recverr) { 404 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 405 goto out; 406 } else { 407 ip_icmp_error(sk, skb, err, uh->dest, info, (u8*)(uh+1)); 408 } 409 sk->sk_err = err; 410 sk->sk_error_report(sk); 411 out: 412 sock_put(sk); 413 } 414 415 void udp_err(struct sk_buff *skb, u32 info) 416 { 417 __udp4_lib_err(skb, info, udp_hash); 418 } 419 420 /* 421 * Throw away all pending data and cancel the corking. Socket is locked. 422 */ 423 void udp_flush_pending_frames(struct sock *sk) 424 { 425 struct udp_sock *up = udp_sk(sk); 426 427 if (up->pending) { 428 up->len = 0; 429 up->pending = 0; 430 ip_flush_pending_frames(sk); 431 } 432 } 433 EXPORT_SYMBOL(udp_flush_pending_frames); 434 435 /** 436 * udp4_hwcsum_outgoing - handle outgoing HW checksumming 437 * @sk: socket we are sending on 438 * @skb: sk_buff containing the filled-in UDP header 439 * (checksum field must be zeroed out) 440 */ 441 static void udp4_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb, 442 __be32 src, __be32 dst, int len ) 443 { 444 unsigned int offset; 445 struct udphdr *uh = udp_hdr(skb); 446 __wsum csum = 0; 447 448 if (skb_queue_len(&sk->sk_write_queue) == 1) { 449 /* 450 * Only one fragment on the socket. 451 */ 452 skb->csum_start = skb_transport_header(skb) - skb->head; 453 skb->csum_offset = offsetof(struct udphdr, check); 454 uh->check = ~csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, 0); 455 } else { 456 /* 457 * HW-checksum won't work as there are two or more 458 * fragments on the socket so that all csums of sk_buffs 459 * should be together 460 */ 461 offset = skb_transport_offset(skb); 462 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0); 463 464 skb->ip_summed = CHECKSUM_NONE; 465 466 skb_queue_walk(&sk->sk_write_queue, skb) { 467 csum = csum_add(csum, skb->csum); 468 } 469 470 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); 471 if (uh->check == 0) 472 uh->check = CSUM_MANGLED_0; 473 } 474 } 475 476 /* 477 * Push out all pending data as one UDP datagram. Socket is locked. 478 */ 479 static int udp_push_pending_frames(struct sock *sk) 480 { 481 struct udp_sock *up = udp_sk(sk); 482 struct inet_sock *inet = inet_sk(sk); 483 struct flowi *fl = &inet->cork.fl; 484 struct sk_buff *skb; 485 struct udphdr *uh; 486 int err = 0; 487 int is_udplite = IS_UDPLITE(sk); 488 __wsum csum = 0; 489 490 /* Grab the skbuff where UDP header space exists. */ 491 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL) 492 goto out; 493 494 /* 495 * Create a UDP header 496 */ 497 uh = udp_hdr(skb); 498 uh->source = fl->fl_ip_sport; 499 uh->dest = fl->fl_ip_dport; 500 uh->len = htons(up->len); 501 uh->check = 0; 502 503 if (is_udplite) /* UDP-Lite */ 504 csum = udplite_csum_outgoing(sk, skb); 505 506 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */ 507 508 skb->ip_summed = CHECKSUM_NONE; 509 goto send; 510 511 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 512 513 udp4_hwcsum_outgoing(sk, skb, fl->fl4_src,fl->fl4_dst, up->len); 514 goto send; 515 516 } else /* `normal' UDP */ 517 csum = udp_csum_outgoing(sk, skb); 518 519 /* add protocol-dependent pseudo-header */ 520 uh->check = csum_tcpudp_magic(fl->fl4_src, fl->fl4_dst, up->len, 521 sk->sk_protocol, csum ); 522 if (uh->check == 0) 523 uh->check = CSUM_MANGLED_0; 524 525 send: 526 err = ip_push_pending_frames(sk); 527 out: 528 up->len = 0; 529 up->pending = 0; 530 if (!err) 531 UDP_INC_STATS_USER(UDP_MIB_OUTDATAGRAMS, is_udplite); 532 return err; 533 } 534 535 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 536 size_t len) 537 { 538 struct inet_sock *inet = inet_sk(sk); 539 struct udp_sock *up = udp_sk(sk); 540 int ulen = len; 541 struct ipcm_cookie ipc; 542 struct rtable *rt = NULL; 543 int free = 0; 544 int connected = 0; 545 __be32 daddr, faddr, saddr; 546 __be16 dport; 547 u8 tos; 548 int err, is_udplite = IS_UDPLITE(sk); 549 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; 550 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 551 552 if (len > 0xFFFF) 553 return -EMSGSIZE; 554 555 /* 556 * Check the flags. 557 */ 558 559 if (msg->msg_flags&MSG_OOB) /* Mirror BSD error message compatibility */ 560 return -EOPNOTSUPP; 561 562 ipc.opt = NULL; 563 564 if (up->pending) { 565 /* 566 * There are pending frames. 567 * The socket lock must be held while it's corked. 568 */ 569 lock_sock(sk); 570 if (likely(up->pending)) { 571 if (unlikely(up->pending != AF_INET)) { 572 release_sock(sk); 573 return -EINVAL; 574 } 575 goto do_append_data; 576 } 577 release_sock(sk); 578 } 579 ulen += sizeof(struct udphdr); 580 581 /* 582 * Get and verify the address. 583 */ 584 if (msg->msg_name) { 585 struct sockaddr_in * usin = (struct sockaddr_in*)msg->msg_name; 586 if (msg->msg_namelen < sizeof(*usin)) 587 return -EINVAL; 588 if (usin->sin_family != AF_INET) { 589 if (usin->sin_family != AF_UNSPEC) 590 return -EAFNOSUPPORT; 591 } 592 593 daddr = usin->sin_addr.s_addr; 594 dport = usin->sin_port; 595 if (dport == 0) 596 return -EINVAL; 597 } else { 598 if (sk->sk_state != TCP_ESTABLISHED) 599 return -EDESTADDRREQ; 600 daddr = inet->daddr; 601 dport = inet->dport; 602 /* Open fast path for connected socket. 603 Route will not be used, if at least one option is set. 604 */ 605 connected = 1; 606 } 607 ipc.addr = inet->saddr; 608 609 ipc.oif = sk->sk_bound_dev_if; 610 if (msg->msg_controllen) { 611 err = ip_cmsg_send(sock_net(sk), msg, &ipc); 612 if (err) 613 return err; 614 if (ipc.opt) 615 free = 1; 616 connected = 0; 617 } 618 if (!ipc.opt) 619 ipc.opt = inet->opt; 620 621 saddr = ipc.addr; 622 ipc.addr = faddr = daddr; 623 624 if (ipc.opt && ipc.opt->srr) { 625 if (!daddr) 626 return -EINVAL; 627 faddr = ipc.opt->faddr; 628 connected = 0; 629 } 630 tos = RT_TOS(inet->tos); 631 if (sock_flag(sk, SOCK_LOCALROUTE) || 632 (msg->msg_flags & MSG_DONTROUTE) || 633 (ipc.opt && ipc.opt->is_strictroute)) { 634 tos |= RTO_ONLINK; 635 connected = 0; 636 } 637 638 if (ipv4_is_multicast(daddr)) { 639 if (!ipc.oif) 640 ipc.oif = inet->mc_index; 641 if (!saddr) 642 saddr = inet->mc_addr; 643 connected = 0; 644 } 645 646 if (connected) 647 rt = (struct rtable*)sk_dst_check(sk, 0); 648 649 if (rt == NULL) { 650 struct flowi fl = { .oif = ipc.oif, 651 .nl_u = { .ip4_u = 652 { .daddr = faddr, 653 .saddr = saddr, 654 .tos = tos } }, 655 .proto = sk->sk_protocol, 656 .uli_u = { .ports = 657 { .sport = inet->sport, 658 .dport = dport } } }; 659 security_sk_classify_flow(sk, &fl); 660 err = ip_route_output_flow(sock_net(sk), &rt, &fl, sk, 1); 661 if (err) { 662 if (err == -ENETUNREACH) 663 IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES); 664 goto out; 665 } 666 667 err = -EACCES; 668 if ((rt->rt_flags & RTCF_BROADCAST) && 669 !sock_flag(sk, SOCK_BROADCAST)) 670 goto out; 671 if (connected) 672 sk_dst_set(sk, dst_clone(&rt->u.dst)); 673 } 674 675 if (msg->msg_flags&MSG_CONFIRM) 676 goto do_confirm; 677 back_from_confirm: 678 679 saddr = rt->rt_src; 680 if (!ipc.addr) 681 daddr = ipc.addr = rt->rt_dst; 682 683 lock_sock(sk); 684 if (unlikely(up->pending)) { 685 /* The socket is already corked while preparing it. */ 686 /* ... which is an evident application bug. --ANK */ 687 release_sock(sk); 688 689 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n"); 690 err = -EINVAL; 691 goto out; 692 } 693 /* 694 * Now cork the socket to pend data. 695 */ 696 inet->cork.fl.fl4_dst = daddr; 697 inet->cork.fl.fl_ip_dport = dport; 698 inet->cork.fl.fl4_src = saddr; 699 inet->cork.fl.fl_ip_sport = inet->sport; 700 up->pending = AF_INET; 701 702 do_append_data: 703 up->len += ulen; 704 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 705 err = ip_append_data(sk, getfrag, msg->msg_iov, ulen, 706 sizeof(struct udphdr), &ipc, rt, 707 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 708 if (err) 709 udp_flush_pending_frames(sk); 710 else if (!corkreq) 711 err = udp_push_pending_frames(sk); 712 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 713 up->pending = 0; 714 release_sock(sk); 715 716 out: 717 ip_rt_put(rt); 718 if (free) 719 kfree(ipc.opt); 720 if (!err) 721 return len; 722 /* 723 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 724 * ENOBUFS might not be good (it's not tunable per se), but otherwise 725 * we don't have a good statistic (IpOutDiscards but it can be too many 726 * things). We could add another new stat but at least for now that 727 * seems like overkill. 728 */ 729 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 730 UDP_INC_STATS_USER(UDP_MIB_SNDBUFERRORS, is_udplite); 731 } 732 return err; 733 734 do_confirm: 735 dst_confirm(&rt->u.dst); 736 if (!(msg->msg_flags&MSG_PROBE) || len) 737 goto back_from_confirm; 738 err = 0; 739 goto out; 740 } 741 742 int udp_sendpage(struct sock *sk, struct page *page, int offset, 743 size_t size, int flags) 744 { 745 struct udp_sock *up = udp_sk(sk); 746 int ret; 747 748 if (!up->pending) { 749 struct msghdr msg = { .msg_flags = flags|MSG_MORE }; 750 751 /* Call udp_sendmsg to specify destination address which 752 * sendpage interface can't pass. 753 * This will succeed only when the socket is connected. 754 */ 755 ret = udp_sendmsg(NULL, sk, &msg, 0); 756 if (ret < 0) 757 return ret; 758 } 759 760 lock_sock(sk); 761 762 if (unlikely(!up->pending)) { 763 release_sock(sk); 764 765 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n"); 766 return -EINVAL; 767 } 768 769 ret = ip_append_page(sk, page, offset, size, flags); 770 if (ret == -EOPNOTSUPP) { 771 release_sock(sk); 772 return sock_no_sendpage(sk->sk_socket, page, offset, 773 size, flags); 774 } 775 if (ret < 0) { 776 udp_flush_pending_frames(sk); 777 goto out; 778 } 779 780 up->len += size; 781 if (!(up->corkflag || (flags&MSG_MORE))) 782 ret = udp_push_pending_frames(sk); 783 if (!ret) 784 ret = size; 785 out: 786 release_sock(sk); 787 return ret; 788 } 789 790 /* 791 * IOCTL requests applicable to the UDP protocol 792 */ 793 794 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg) 795 { 796 switch (cmd) { 797 case SIOCOUTQ: 798 { 799 int amount = atomic_read(&sk->sk_wmem_alloc); 800 return put_user(amount, (int __user *)arg); 801 } 802 803 case SIOCINQ: 804 { 805 struct sk_buff *skb; 806 unsigned long amount; 807 808 amount = 0; 809 spin_lock_bh(&sk->sk_receive_queue.lock); 810 skb = skb_peek(&sk->sk_receive_queue); 811 if (skb != NULL) { 812 /* 813 * We will only return the amount 814 * of this packet since that is all 815 * that will be read. 816 */ 817 amount = skb->len - sizeof(struct udphdr); 818 } 819 spin_unlock_bh(&sk->sk_receive_queue.lock); 820 return put_user(amount, (int __user *)arg); 821 } 822 823 default: 824 return -ENOIOCTLCMD; 825 } 826 827 return 0; 828 } 829 830 /* 831 * This should be easy, if there is something there we 832 * return it, otherwise we block. 833 */ 834 835 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 836 size_t len, int noblock, int flags, int *addr_len) 837 { 838 struct inet_sock *inet = inet_sk(sk); 839 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name; 840 struct sk_buff *skb; 841 unsigned int ulen, copied; 842 int peeked; 843 int err; 844 int is_udplite = IS_UDPLITE(sk); 845 846 /* 847 * Check any passed addresses 848 */ 849 if (addr_len) 850 *addr_len=sizeof(*sin); 851 852 if (flags & MSG_ERRQUEUE) 853 return ip_recv_error(sk, msg, len); 854 855 try_again: 856 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0), 857 &peeked, &err); 858 if (!skb) 859 goto out; 860 861 ulen = skb->len - sizeof(struct udphdr); 862 copied = len; 863 if (copied > ulen) 864 copied = ulen; 865 else if (copied < ulen) 866 msg->msg_flags |= MSG_TRUNC; 867 868 /* 869 * If checksum is needed at all, try to do it while copying the 870 * data. If the data is truncated, or if we only want a partial 871 * coverage checksum (UDP-Lite), do it before the copy. 872 */ 873 874 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) { 875 if (udp_lib_checksum_complete(skb)) 876 goto csum_copy_err; 877 } 878 879 if (skb_csum_unnecessary(skb)) 880 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr), 881 msg->msg_iov, copied ); 882 else { 883 err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov); 884 885 if (err == -EINVAL) 886 goto csum_copy_err; 887 } 888 889 if (err) 890 goto out_free; 891 892 if (!peeked) 893 UDP_INC_STATS_USER(UDP_MIB_INDATAGRAMS, is_udplite); 894 895 sock_recv_timestamp(msg, sk, skb); 896 897 /* Copy the address. */ 898 if (sin) 899 { 900 sin->sin_family = AF_INET; 901 sin->sin_port = udp_hdr(skb)->source; 902 sin->sin_addr.s_addr = ip_hdr(skb)->saddr; 903 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 904 } 905 if (inet->cmsg_flags) 906 ip_cmsg_recv(msg, skb); 907 908 err = copied; 909 if (flags & MSG_TRUNC) 910 err = ulen; 911 912 out_free: 913 lock_sock(sk); 914 skb_free_datagram(sk, skb); 915 release_sock(sk); 916 out: 917 return err; 918 919 csum_copy_err: 920 lock_sock(sk); 921 if (!skb_kill_datagram(sk, skb, flags)) 922 UDP_INC_STATS_USER(UDP_MIB_INERRORS, is_udplite); 923 release_sock(sk); 924 925 if (noblock) 926 return -EAGAIN; 927 goto try_again; 928 } 929 930 931 int udp_disconnect(struct sock *sk, int flags) 932 { 933 struct inet_sock *inet = inet_sk(sk); 934 /* 935 * 1003.1g - break association. 936 */ 937 938 sk->sk_state = TCP_CLOSE; 939 inet->daddr = 0; 940 inet->dport = 0; 941 sk->sk_bound_dev_if = 0; 942 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) 943 inet_reset_saddr(sk); 944 945 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { 946 sk->sk_prot->unhash(sk); 947 inet->sport = 0; 948 } 949 sk_dst_reset(sk); 950 return 0; 951 } 952 953 /* returns: 954 * -1: error 955 * 0: success 956 * >0: "udp encap" protocol resubmission 957 * 958 * Note that in the success and error cases, the skb is assumed to 959 * have either been requeued or freed. 960 */ 961 int udp_queue_rcv_skb(struct sock * sk, struct sk_buff *skb) 962 { 963 struct udp_sock *up = udp_sk(sk); 964 int rc; 965 int is_udplite = IS_UDPLITE(sk); 966 967 /* 968 * Charge it to the socket, dropping if the queue is full. 969 */ 970 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) 971 goto drop; 972 nf_reset(skb); 973 974 if (up->encap_type) { 975 /* 976 * This is an encapsulation socket so pass the skb to 977 * the socket's udp_encap_rcv() hook. Otherwise, just 978 * fall through and pass this up the UDP socket. 979 * up->encap_rcv() returns the following value: 980 * =0 if skb was successfully passed to the encap 981 * handler or was discarded by it. 982 * >0 if skb should be passed on to UDP. 983 * <0 if skb should be resubmitted as proto -N 984 */ 985 986 /* if we're overly short, let UDP handle it */ 987 if (skb->len > sizeof(struct udphdr) && 988 up->encap_rcv != NULL) { 989 int ret; 990 991 ret = (*up->encap_rcv)(sk, skb); 992 if (ret <= 0) { 993 UDP_INC_STATS_BH(UDP_MIB_INDATAGRAMS, 994 is_udplite); 995 return -ret; 996 } 997 } 998 999 /* FALLTHROUGH -- it's a UDP Packet */ 1000 } 1001 1002 /* 1003 * UDP-Lite specific tests, ignored on UDP sockets 1004 */ 1005 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 1006 1007 /* 1008 * MIB statistics other than incrementing the error count are 1009 * disabled for the following two types of errors: these depend 1010 * on the application settings, not on the functioning of the 1011 * protocol stack as such. 1012 * 1013 * RFC 3828 here recommends (sec 3.3): "There should also be a 1014 * way ... to ... at least let the receiving application block 1015 * delivery of packets with coverage values less than a value 1016 * provided by the application." 1017 */ 1018 if (up->pcrlen == 0) { /* full coverage was set */ 1019 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage " 1020 "%d while full coverage %d requested\n", 1021 UDP_SKB_CB(skb)->cscov, skb->len); 1022 goto drop; 1023 } 1024 /* The next case involves violating the min. coverage requested 1025 * by the receiver. This is subtle: if receiver wants x and x is 1026 * greater than the buffersize/MTU then receiver will complain 1027 * that it wants x while sender emits packets of smaller size y. 1028 * Therefore the above ...()->partial_cov statement is essential. 1029 */ 1030 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 1031 LIMIT_NETDEBUG(KERN_WARNING 1032 "UDPLITE: coverage %d too small, need min %d\n", 1033 UDP_SKB_CB(skb)->cscov, up->pcrlen); 1034 goto drop; 1035 } 1036 } 1037 1038 if (sk->sk_filter) { 1039 if (udp_lib_checksum_complete(skb)) 1040 goto drop; 1041 } 1042 1043 if ((rc = sock_queue_rcv_skb(sk,skb)) < 0) { 1044 /* Note that an ENOMEM error is charged twice */ 1045 if (rc == -ENOMEM) 1046 UDP_INC_STATS_BH(UDP_MIB_RCVBUFERRORS, is_udplite); 1047 goto drop; 1048 } 1049 1050 return 0; 1051 1052 drop: 1053 UDP_INC_STATS_BH(UDP_MIB_INERRORS, is_udplite); 1054 kfree_skb(skb); 1055 return -1; 1056 } 1057 1058 /* 1059 * Multicasts and broadcasts go to each listener. 1060 * 1061 * Note: called only from the BH handler context, 1062 * so we don't need to lock the hashes. 1063 */ 1064 static int __udp4_lib_mcast_deliver(struct sk_buff *skb, 1065 struct udphdr *uh, 1066 __be32 saddr, __be32 daddr, 1067 struct hlist_head udptable[]) 1068 { 1069 struct sock *sk; 1070 int dif; 1071 1072 read_lock(&udp_hash_lock); 1073 sk = sk_head(&udptable[ntohs(uh->dest) & (UDP_HTABLE_SIZE - 1)]); 1074 dif = skb->dev->ifindex; 1075 sk = udp_v4_mcast_next(sk, uh->dest, daddr, uh->source, saddr, dif); 1076 if (sk) { 1077 struct sock *sknext = NULL; 1078 1079 do { 1080 struct sk_buff *skb1 = skb; 1081 1082 sknext = udp_v4_mcast_next(sk_next(sk), uh->dest, daddr, 1083 uh->source, saddr, dif); 1084 if (sknext) 1085 skb1 = skb_clone(skb, GFP_ATOMIC); 1086 1087 if (skb1) { 1088 int ret = 0; 1089 1090 bh_lock_sock_nested(sk); 1091 if (!sock_owned_by_user(sk)) 1092 ret = udp_queue_rcv_skb(sk, skb1); 1093 else 1094 sk_add_backlog(sk, skb1); 1095 bh_unlock_sock(sk); 1096 1097 if (ret > 0) 1098 /* we should probably re-process instead 1099 * of dropping packets here. */ 1100 kfree_skb(skb1); 1101 } 1102 sk = sknext; 1103 } while (sknext); 1104 } else 1105 kfree_skb(skb); 1106 read_unlock(&udp_hash_lock); 1107 return 0; 1108 } 1109 1110 /* Initialize UDP checksum. If exited with zero value (success), 1111 * CHECKSUM_UNNECESSARY means, that no more checks are required. 1112 * Otherwise, csum completion requires chacksumming packet body, 1113 * including udp header and folding it to skb->csum. 1114 */ 1115 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, 1116 int proto) 1117 { 1118 const struct iphdr *iph; 1119 int err; 1120 1121 UDP_SKB_CB(skb)->partial_cov = 0; 1122 UDP_SKB_CB(skb)->cscov = skb->len; 1123 1124 if (proto == IPPROTO_UDPLITE) { 1125 err = udplite_checksum_init(skb, uh); 1126 if (err) 1127 return err; 1128 } 1129 1130 iph = ip_hdr(skb); 1131 if (uh->check == 0) { 1132 skb->ip_summed = CHECKSUM_UNNECESSARY; 1133 } else if (skb->ip_summed == CHECKSUM_COMPLETE) { 1134 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len, 1135 proto, skb->csum)) 1136 skb->ip_summed = CHECKSUM_UNNECESSARY; 1137 } 1138 if (!skb_csum_unnecessary(skb)) 1139 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr, 1140 skb->len, proto, 0); 1141 /* Probably, we should checksum udp header (it should be in cache 1142 * in any case) and data in tiny packets (< rx copybreak). 1143 */ 1144 1145 return 0; 1146 } 1147 1148 /* 1149 * All we need to do is get the socket, and then do a checksum. 1150 */ 1151 1152 int __udp4_lib_rcv(struct sk_buff *skb, struct hlist_head udptable[], 1153 int proto) 1154 { 1155 struct sock *sk; 1156 struct udphdr *uh = udp_hdr(skb); 1157 unsigned short ulen; 1158 struct rtable *rt = (struct rtable*)skb->dst; 1159 __be32 saddr = ip_hdr(skb)->saddr; 1160 __be32 daddr = ip_hdr(skb)->daddr; 1161 1162 /* 1163 * Validate the packet. 1164 */ 1165 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 1166 goto drop; /* No space for header. */ 1167 1168 ulen = ntohs(uh->len); 1169 if (ulen > skb->len) 1170 goto short_packet; 1171 1172 if (proto == IPPROTO_UDP) { 1173 /* UDP validates ulen. */ 1174 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) 1175 goto short_packet; 1176 uh = udp_hdr(skb); 1177 } 1178 1179 if (udp4_csum_init(skb, uh, proto)) 1180 goto csum_error; 1181 1182 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) 1183 return __udp4_lib_mcast_deliver(skb, uh, saddr, daddr, udptable); 1184 1185 sk = __udp4_lib_lookup(dev_net(skb->dev), saddr, uh->source, daddr, 1186 uh->dest, inet_iif(skb), udptable); 1187 1188 if (sk != NULL) { 1189 int ret = 0; 1190 bh_lock_sock_nested(sk); 1191 if (!sock_owned_by_user(sk)) 1192 ret = udp_queue_rcv_skb(sk, skb); 1193 else 1194 sk_add_backlog(sk, skb); 1195 bh_unlock_sock(sk); 1196 sock_put(sk); 1197 1198 /* a return value > 0 means to resubmit the input, but 1199 * it wants the return to be -protocol, or 0 1200 */ 1201 if (ret > 0) 1202 return -ret; 1203 return 0; 1204 } 1205 1206 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 1207 goto drop; 1208 nf_reset(skb); 1209 1210 /* No socket. Drop packet silently, if checksum is wrong */ 1211 if (udp_lib_checksum_complete(skb)) 1212 goto csum_error; 1213 1214 UDP_INC_STATS_BH(UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 1215 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); 1216 1217 /* 1218 * Hmm. We got an UDP packet to a port to which we 1219 * don't wanna listen. Ignore it. 1220 */ 1221 kfree_skb(skb); 1222 return 0; 1223 1224 short_packet: 1225 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From " NIPQUAD_FMT ":%u %d/%d to " NIPQUAD_FMT ":%u\n", 1226 proto == IPPROTO_UDPLITE ? "-Lite" : "", 1227 NIPQUAD(saddr), 1228 ntohs(uh->source), 1229 ulen, 1230 skb->len, 1231 NIPQUAD(daddr), 1232 ntohs(uh->dest)); 1233 goto drop; 1234 1235 csum_error: 1236 /* 1237 * RFC1122: OK. Discards the bad packet silently (as far as 1238 * the network is concerned, anyway) as per 4.1.3.4 (MUST). 1239 */ 1240 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From " NIPQUAD_FMT ":%u to " NIPQUAD_FMT ":%u ulen %d\n", 1241 proto == IPPROTO_UDPLITE ? "-Lite" : "", 1242 NIPQUAD(saddr), 1243 ntohs(uh->source), 1244 NIPQUAD(daddr), 1245 ntohs(uh->dest), 1246 ulen); 1247 drop: 1248 UDP_INC_STATS_BH(UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 1249 kfree_skb(skb); 1250 return 0; 1251 } 1252 1253 int udp_rcv(struct sk_buff *skb) 1254 { 1255 return __udp4_lib_rcv(skb, udp_hash, IPPROTO_UDP); 1256 } 1257 1258 int udp_destroy_sock(struct sock *sk) 1259 { 1260 lock_sock(sk); 1261 udp_flush_pending_frames(sk); 1262 release_sock(sk); 1263 return 0; 1264 } 1265 1266 /* 1267 * Socket option code for UDP 1268 */ 1269 int udp_lib_setsockopt(struct sock *sk, int level, int optname, 1270 char __user *optval, int optlen, 1271 int (*push_pending_frames)(struct sock *)) 1272 { 1273 struct udp_sock *up = udp_sk(sk); 1274 int val; 1275 int err = 0; 1276 int is_udplite = IS_UDPLITE(sk); 1277 1278 if (optlen<sizeof(int)) 1279 return -EINVAL; 1280 1281 if (get_user(val, (int __user *)optval)) 1282 return -EFAULT; 1283 1284 switch (optname) { 1285 case UDP_CORK: 1286 if (val != 0) { 1287 up->corkflag = 1; 1288 } else { 1289 up->corkflag = 0; 1290 lock_sock(sk); 1291 (*push_pending_frames)(sk); 1292 release_sock(sk); 1293 } 1294 break; 1295 1296 case UDP_ENCAP: 1297 switch (val) { 1298 case 0: 1299 case UDP_ENCAP_ESPINUDP: 1300 case UDP_ENCAP_ESPINUDP_NON_IKE: 1301 up->encap_rcv = xfrm4_udp_encap_rcv; 1302 /* FALLTHROUGH */ 1303 case UDP_ENCAP_L2TPINUDP: 1304 up->encap_type = val; 1305 break; 1306 default: 1307 err = -ENOPROTOOPT; 1308 break; 1309 } 1310 break; 1311 1312 /* 1313 * UDP-Lite's partial checksum coverage (RFC 3828). 1314 */ 1315 /* The sender sets actual checksum coverage length via this option. 1316 * The case coverage > packet length is handled by send module. */ 1317 case UDPLITE_SEND_CSCOV: 1318 if (!is_udplite) /* Disable the option on UDP sockets */ 1319 return -ENOPROTOOPT; 1320 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ 1321 val = 8; 1322 up->pcslen = val; 1323 up->pcflag |= UDPLITE_SEND_CC; 1324 break; 1325 1326 /* The receiver specifies a minimum checksum coverage value. To make 1327 * sense, this should be set to at least 8 (as done below). If zero is 1328 * used, this again means full checksum coverage. */ 1329 case UDPLITE_RECV_CSCOV: 1330 if (!is_udplite) /* Disable the option on UDP sockets */ 1331 return -ENOPROTOOPT; 1332 if (val != 0 && val < 8) /* Avoid silly minimal values. */ 1333 val = 8; 1334 up->pcrlen = val; 1335 up->pcflag |= UDPLITE_RECV_CC; 1336 break; 1337 1338 default: 1339 err = -ENOPROTOOPT; 1340 break; 1341 } 1342 1343 return err; 1344 } 1345 1346 int udp_setsockopt(struct sock *sk, int level, int optname, 1347 char __user *optval, int optlen) 1348 { 1349 if (level == SOL_UDP || level == SOL_UDPLITE) 1350 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1351 udp_push_pending_frames); 1352 return ip_setsockopt(sk, level, optname, optval, optlen); 1353 } 1354 1355 #ifdef CONFIG_COMPAT 1356 int compat_udp_setsockopt(struct sock *sk, int level, int optname, 1357 char __user *optval, int optlen) 1358 { 1359 if (level == SOL_UDP || level == SOL_UDPLITE) 1360 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1361 udp_push_pending_frames); 1362 return compat_ip_setsockopt(sk, level, optname, optval, optlen); 1363 } 1364 #endif 1365 1366 int udp_lib_getsockopt(struct sock *sk, int level, int optname, 1367 char __user *optval, int __user *optlen) 1368 { 1369 struct udp_sock *up = udp_sk(sk); 1370 int val, len; 1371 1372 if (get_user(len,optlen)) 1373 return -EFAULT; 1374 1375 len = min_t(unsigned int, len, sizeof(int)); 1376 1377 if (len < 0) 1378 return -EINVAL; 1379 1380 switch (optname) { 1381 case UDP_CORK: 1382 val = up->corkflag; 1383 break; 1384 1385 case UDP_ENCAP: 1386 val = up->encap_type; 1387 break; 1388 1389 /* The following two cannot be changed on UDP sockets, the return is 1390 * always 0 (which corresponds to the full checksum coverage of UDP). */ 1391 case UDPLITE_SEND_CSCOV: 1392 val = up->pcslen; 1393 break; 1394 1395 case UDPLITE_RECV_CSCOV: 1396 val = up->pcrlen; 1397 break; 1398 1399 default: 1400 return -ENOPROTOOPT; 1401 } 1402 1403 if (put_user(len, optlen)) 1404 return -EFAULT; 1405 if (copy_to_user(optval, &val,len)) 1406 return -EFAULT; 1407 return 0; 1408 } 1409 1410 int udp_getsockopt(struct sock *sk, int level, int optname, 1411 char __user *optval, int __user *optlen) 1412 { 1413 if (level == SOL_UDP || level == SOL_UDPLITE) 1414 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1415 return ip_getsockopt(sk, level, optname, optval, optlen); 1416 } 1417 1418 #ifdef CONFIG_COMPAT 1419 int compat_udp_getsockopt(struct sock *sk, int level, int optname, 1420 char __user *optval, int __user *optlen) 1421 { 1422 if (level == SOL_UDP || level == SOL_UDPLITE) 1423 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1424 return compat_ip_getsockopt(sk, level, optname, optval, optlen); 1425 } 1426 #endif 1427 /** 1428 * udp_poll - wait for a UDP event. 1429 * @file - file struct 1430 * @sock - socket 1431 * @wait - poll table 1432 * 1433 * This is same as datagram poll, except for the special case of 1434 * blocking sockets. If application is using a blocking fd 1435 * and a packet with checksum error is in the queue; 1436 * then it could get return from select indicating data available 1437 * but then block when reading it. Add special case code 1438 * to work around these arguably broken applications. 1439 */ 1440 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait) 1441 { 1442 unsigned int mask = datagram_poll(file, sock, wait); 1443 struct sock *sk = sock->sk; 1444 int is_lite = IS_UDPLITE(sk); 1445 1446 /* Check for false positives due to checksum errors */ 1447 if ( (mask & POLLRDNORM) && 1448 !(file->f_flags & O_NONBLOCK) && 1449 !(sk->sk_shutdown & RCV_SHUTDOWN)){ 1450 struct sk_buff_head *rcvq = &sk->sk_receive_queue; 1451 struct sk_buff *skb; 1452 1453 spin_lock_bh(&rcvq->lock); 1454 while ((skb = skb_peek(rcvq)) != NULL && 1455 udp_lib_checksum_complete(skb)) { 1456 UDP_INC_STATS_BH(UDP_MIB_INERRORS, is_lite); 1457 __skb_unlink(skb, rcvq); 1458 kfree_skb(skb); 1459 } 1460 spin_unlock_bh(&rcvq->lock); 1461 1462 /* nothing to see, move along */ 1463 if (skb == NULL) 1464 mask &= ~(POLLIN | POLLRDNORM); 1465 } 1466 1467 return mask; 1468 1469 } 1470 1471 struct proto udp_prot = { 1472 .name = "UDP", 1473 .owner = THIS_MODULE, 1474 .close = udp_lib_close, 1475 .connect = ip4_datagram_connect, 1476 .disconnect = udp_disconnect, 1477 .ioctl = udp_ioctl, 1478 .destroy = udp_destroy_sock, 1479 .setsockopt = udp_setsockopt, 1480 .getsockopt = udp_getsockopt, 1481 .sendmsg = udp_sendmsg, 1482 .recvmsg = udp_recvmsg, 1483 .sendpage = udp_sendpage, 1484 .backlog_rcv = udp_queue_rcv_skb, 1485 .hash = udp_lib_hash, 1486 .unhash = udp_lib_unhash, 1487 .get_port = udp_v4_get_port, 1488 .memory_allocated = &udp_memory_allocated, 1489 .sysctl_mem = sysctl_udp_mem, 1490 .sysctl_wmem = &sysctl_udp_wmem_min, 1491 .sysctl_rmem = &sysctl_udp_rmem_min, 1492 .obj_size = sizeof(struct udp_sock), 1493 .h.udp_hash = udp_hash, 1494 #ifdef CONFIG_COMPAT 1495 .compat_setsockopt = compat_udp_setsockopt, 1496 .compat_getsockopt = compat_udp_getsockopt, 1497 #endif 1498 }; 1499 1500 /* ------------------------------------------------------------------------ */ 1501 #ifdef CONFIG_PROC_FS 1502 1503 static struct sock *udp_get_first(struct seq_file *seq) 1504 { 1505 struct sock *sk; 1506 struct udp_iter_state *state = seq->private; 1507 struct net *net = seq_file_net(seq); 1508 1509 for (state->bucket = 0; state->bucket < UDP_HTABLE_SIZE; ++state->bucket) { 1510 struct hlist_node *node; 1511 sk_for_each(sk, node, state->hashtable + state->bucket) { 1512 if (!net_eq(sock_net(sk), net)) 1513 continue; 1514 if (sk->sk_family == state->family) 1515 goto found; 1516 } 1517 } 1518 sk = NULL; 1519 found: 1520 return sk; 1521 } 1522 1523 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) 1524 { 1525 struct udp_iter_state *state = seq->private; 1526 struct net *net = seq_file_net(seq); 1527 1528 do { 1529 sk = sk_next(sk); 1530 try_again: 1531 ; 1532 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family)); 1533 1534 if (!sk && ++state->bucket < UDP_HTABLE_SIZE) { 1535 sk = sk_head(state->hashtable + state->bucket); 1536 goto try_again; 1537 } 1538 return sk; 1539 } 1540 1541 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) 1542 { 1543 struct sock *sk = udp_get_first(seq); 1544 1545 if (sk) 1546 while (pos && (sk = udp_get_next(seq, sk)) != NULL) 1547 --pos; 1548 return pos ? NULL : sk; 1549 } 1550 1551 static void *udp_seq_start(struct seq_file *seq, loff_t *pos) 1552 __acquires(udp_hash_lock) 1553 { 1554 read_lock(&udp_hash_lock); 1555 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN; 1556 } 1557 1558 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 1559 { 1560 struct sock *sk; 1561 1562 if (v == SEQ_START_TOKEN) 1563 sk = udp_get_idx(seq, 0); 1564 else 1565 sk = udp_get_next(seq, v); 1566 1567 ++*pos; 1568 return sk; 1569 } 1570 1571 static void udp_seq_stop(struct seq_file *seq, void *v) 1572 __releases(udp_hash_lock) 1573 { 1574 read_unlock(&udp_hash_lock); 1575 } 1576 1577 static int udp_seq_open(struct inode *inode, struct file *file) 1578 { 1579 struct udp_seq_afinfo *afinfo = PDE(inode)->data; 1580 struct udp_iter_state *s; 1581 int err; 1582 1583 err = seq_open_net(inode, file, &afinfo->seq_ops, 1584 sizeof(struct udp_iter_state)); 1585 if (err < 0) 1586 return err; 1587 1588 s = ((struct seq_file *)file->private_data)->private; 1589 s->family = afinfo->family; 1590 s->hashtable = afinfo->hashtable; 1591 return err; 1592 } 1593 1594 /* ------------------------------------------------------------------------ */ 1595 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo) 1596 { 1597 struct proc_dir_entry *p; 1598 int rc = 0; 1599 1600 afinfo->seq_fops.open = udp_seq_open; 1601 afinfo->seq_fops.read = seq_read; 1602 afinfo->seq_fops.llseek = seq_lseek; 1603 afinfo->seq_fops.release = seq_release_net; 1604 1605 afinfo->seq_ops.start = udp_seq_start; 1606 afinfo->seq_ops.next = udp_seq_next; 1607 afinfo->seq_ops.stop = udp_seq_stop; 1608 1609 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net, 1610 &afinfo->seq_fops, afinfo); 1611 if (!p) 1612 rc = -ENOMEM; 1613 return rc; 1614 } 1615 1616 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo) 1617 { 1618 proc_net_remove(net, afinfo->name); 1619 } 1620 1621 /* ------------------------------------------------------------------------ */ 1622 static void udp4_format_sock(struct sock *sp, struct seq_file *f, 1623 int bucket, int *len) 1624 { 1625 struct inet_sock *inet = inet_sk(sp); 1626 __be32 dest = inet->daddr; 1627 __be32 src = inet->rcv_saddr; 1628 __u16 destp = ntohs(inet->dport); 1629 __u16 srcp = ntohs(inet->sport); 1630 1631 seq_printf(f, "%4d: %08X:%04X %08X:%04X" 1632 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p%n", 1633 bucket, src, srcp, dest, destp, sp->sk_state, 1634 atomic_read(&sp->sk_wmem_alloc), 1635 atomic_read(&sp->sk_rmem_alloc), 1636 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp), 1637 atomic_read(&sp->sk_refcnt), sp, len); 1638 } 1639 1640 int udp4_seq_show(struct seq_file *seq, void *v) 1641 { 1642 if (v == SEQ_START_TOKEN) 1643 seq_printf(seq, "%-127s\n", 1644 " sl local_address rem_address st tx_queue " 1645 "rx_queue tr tm->when retrnsmt uid timeout " 1646 "inode"); 1647 else { 1648 struct udp_iter_state *state = seq->private; 1649 int len; 1650 1651 udp4_format_sock(v, seq, state->bucket, &len); 1652 seq_printf(seq, "%*s\n", 127 - len ,""); 1653 } 1654 return 0; 1655 } 1656 1657 /* ------------------------------------------------------------------------ */ 1658 static struct udp_seq_afinfo udp4_seq_afinfo = { 1659 .name = "udp", 1660 .family = AF_INET, 1661 .hashtable = udp_hash, 1662 .seq_fops = { 1663 .owner = THIS_MODULE, 1664 }, 1665 .seq_ops = { 1666 .show = udp4_seq_show, 1667 }, 1668 }; 1669 1670 static int udp4_proc_init_net(struct net *net) 1671 { 1672 return udp_proc_register(net, &udp4_seq_afinfo); 1673 } 1674 1675 static void udp4_proc_exit_net(struct net *net) 1676 { 1677 udp_proc_unregister(net, &udp4_seq_afinfo); 1678 } 1679 1680 static struct pernet_operations udp4_net_ops = { 1681 .init = udp4_proc_init_net, 1682 .exit = udp4_proc_exit_net, 1683 }; 1684 1685 int __init udp4_proc_init(void) 1686 { 1687 return register_pernet_subsys(&udp4_net_ops); 1688 } 1689 1690 void udp4_proc_exit(void) 1691 { 1692 unregister_pernet_subsys(&udp4_net_ops); 1693 } 1694 #endif /* CONFIG_PROC_FS */ 1695 1696 void __init udp_init(void) 1697 { 1698 unsigned long limit; 1699 1700 /* Set the pressure threshold up by the same strategy of TCP. It is a 1701 * fraction of global memory that is up to 1/2 at 256 MB, decreasing 1702 * toward zero with the amount of memory, with a floor of 128 pages. 1703 */ 1704 limit = min(nr_all_pages, 1UL<<(28-PAGE_SHIFT)) >> (20-PAGE_SHIFT); 1705 limit = (limit * (nr_all_pages >> (20-PAGE_SHIFT))) >> (PAGE_SHIFT-11); 1706 limit = max(limit, 128UL); 1707 sysctl_udp_mem[0] = limit / 4 * 3; 1708 sysctl_udp_mem[1] = limit; 1709 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2; 1710 1711 sysctl_udp_rmem_min = SK_MEM_QUANTUM; 1712 sysctl_udp_wmem_min = SK_MEM_QUANTUM; 1713 } 1714 1715 EXPORT_SYMBOL(udp_disconnect); 1716 EXPORT_SYMBOL(udp_hash); 1717 EXPORT_SYMBOL(udp_hash_lock); 1718 EXPORT_SYMBOL(udp_ioctl); 1719 EXPORT_SYMBOL(udp_prot); 1720 EXPORT_SYMBOL(udp_sendmsg); 1721 EXPORT_SYMBOL(udp_lib_getsockopt); 1722 EXPORT_SYMBOL(udp_lib_setsockopt); 1723 EXPORT_SYMBOL(udp_poll); 1724 EXPORT_SYMBOL(udp_lib_get_port); 1725 1726 #ifdef CONFIG_PROC_FS 1727 EXPORT_SYMBOL(udp_proc_register); 1728 EXPORT_SYMBOL(udp_proc_unregister); 1729 #endif 1730