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 static 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 } 433 434 /** 435 * udp4_hwcsum_outgoing - handle outgoing HW checksumming 436 * @sk: socket we are sending on 437 * @skb: sk_buff containing the filled-in UDP header 438 * (checksum field must be zeroed out) 439 */ 440 static void udp4_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb, 441 __be32 src, __be32 dst, int len ) 442 { 443 unsigned int offset; 444 struct udphdr *uh = udp_hdr(skb); 445 __wsum csum = 0; 446 447 if (skb_queue_len(&sk->sk_write_queue) == 1) { 448 /* 449 * Only one fragment on the socket. 450 */ 451 skb->csum_start = skb_transport_header(skb) - skb->head; 452 skb->csum_offset = offsetof(struct udphdr, check); 453 uh->check = ~csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, 0); 454 } else { 455 /* 456 * HW-checksum won't work as there are two or more 457 * fragments on the socket so that all csums of sk_buffs 458 * should be together 459 */ 460 offset = skb_transport_offset(skb); 461 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0); 462 463 skb->ip_summed = CHECKSUM_NONE; 464 465 skb_queue_walk(&sk->sk_write_queue, skb) { 466 csum = csum_add(csum, skb->csum); 467 } 468 469 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); 470 if (uh->check == 0) 471 uh->check = CSUM_MANGLED_0; 472 } 473 } 474 475 /* 476 * Push out all pending data as one UDP datagram. Socket is locked. 477 */ 478 static int udp_push_pending_frames(struct sock *sk) 479 { 480 struct udp_sock *up = udp_sk(sk); 481 struct inet_sock *inet = inet_sk(sk); 482 struct flowi *fl = &inet->cork.fl; 483 struct sk_buff *skb; 484 struct udphdr *uh; 485 int err = 0; 486 int is_udplite = IS_UDPLITE(sk); 487 __wsum csum = 0; 488 489 /* Grab the skbuff where UDP header space exists. */ 490 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL) 491 goto out; 492 493 /* 494 * Create a UDP header 495 */ 496 uh = udp_hdr(skb); 497 uh->source = fl->fl_ip_sport; 498 uh->dest = fl->fl_ip_dport; 499 uh->len = htons(up->len); 500 uh->check = 0; 501 502 if (is_udplite) /* UDP-Lite */ 503 csum = udplite_csum_outgoing(sk, skb); 504 505 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */ 506 507 skb->ip_summed = CHECKSUM_NONE; 508 goto send; 509 510 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 511 512 udp4_hwcsum_outgoing(sk, skb, fl->fl4_src,fl->fl4_dst, up->len); 513 goto send; 514 515 } else /* `normal' UDP */ 516 csum = udp_csum_outgoing(sk, skb); 517 518 /* add protocol-dependent pseudo-header */ 519 uh->check = csum_tcpudp_magic(fl->fl4_src, fl->fl4_dst, up->len, 520 sk->sk_protocol, csum ); 521 if (uh->check == 0) 522 uh->check = CSUM_MANGLED_0; 523 524 send: 525 err = ip_push_pending_frames(sk); 526 out: 527 up->len = 0; 528 up->pending = 0; 529 if (!err) 530 UDP_INC_STATS_USER(UDP_MIB_OUTDATAGRAMS, is_udplite); 531 return err; 532 } 533 534 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 535 size_t len) 536 { 537 struct inet_sock *inet = inet_sk(sk); 538 struct udp_sock *up = udp_sk(sk); 539 int ulen = len; 540 struct ipcm_cookie ipc; 541 struct rtable *rt = NULL; 542 int free = 0; 543 int connected = 0; 544 __be32 daddr, faddr, saddr; 545 __be16 dport; 546 u8 tos; 547 int err, is_udplite = IS_UDPLITE(sk); 548 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; 549 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 550 551 if (len > 0xFFFF) 552 return -EMSGSIZE; 553 554 /* 555 * Check the flags. 556 */ 557 558 if (msg->msg_flags&MSG_OOB) /* Mirror BSD error message compatibility */ 559 return -EOPNOTSUPP; 560 561 ipc.opt = NULL; 562 563 if (up->pending) { 564 /* 565 * There are pending frames. 566 * The socket lock must be held while it's corked. 567 */ 568 lock_sock(sk); 569 if (likely(up->pending)) { 570 if (unlikely(up->pending != AF_INET)) { 571 release_sock(sk); 572 return -EINVAL; 573 } 574 goto do_append_data; 575 } 576 release_sock(sk); 577 } 578 ulen += sizeof(struct udphdr); 579 580 /* 581 * Get and verify the address. 582 */ 583 if (msg->msg_name) { 584 struct sockaddr_in * usin = (struct sockaddr_in*)msg->msg_name; 585 if (msg->msg_namelen < sizeof(*usin)) 586 return -EINVAL; 587 if (usin->sin_family != AF_INET) { 588 if (usin->sin_family != AF_UNSPEC) 589 return -EAFNOSUPPORT; 590 } 591 592 daddr = usin->sin_addr.s_addr; 593 dport = usin->sin_port; 594 if (dport == 0) 595 return -EINVAL; 596 } else { 597 if (sk->sk_state != TCP_ESTABLISHED) 598 return -EDESTADDRREQ; 599 daddr = inet->daddr; 600 dport = inet->dport; 601 /* Open fast path for connected socket. 602 Route will not be used, if at least one option is set. 603 */ 604 connected = 1; 605 } 606 ipc.addr = inet->saddr; 607 608 ipc.oif = sk->sk_bound_dev_if; 609 if (msg->msg_controllen) { 610 err = ip_cmsg_send(sock_net(sk), msg, &ipc); 611 if (err) 612 return err; 613 if (ipc.opt) 614 free = 1; 615 connected = 0; 616 } 617 if (!ipc.opt) 618 ipc.opt = inet->opt; 619 620 saddr = ipc.addr; 621 ipc.addr = faddr = daddr; 622 623 if (ipc.opt && ipc.opt->srr) { 624 if (!daddr) 625 return -EINVAL; 626 faddr = ipc.opt->faddr; 627 connected = 0; 628 } 629 tos = RT_TOS(inet->tos); 630 if (sock_flag(sk, SOCK_LOCALROUTE) || 631 (msg->msg_flags & MSG_DONTROUTE) || 632 (ipc.opt && ipc.opt->is_strictroute)) { 633 tos |= RTO_ONLINK; 634 connected = 0; 635 } 636 637 if (ipv4_is_multicast(daddr)) { 638 if (!ipc.oif) 639 ipc.oif = inet->mc_index; 640 if (!saddr) 641 saddr = inet->mc_addr; 642 connected = 0; 643 } 644 645 if (connected) 646 rt = (struct rtable*)sk_dst_check(sk, 0); 647 648 if (rt == NULL) { 649 struct flowi fl = { .oif = ipc.oif, 650 .nl_u = { .ip4_u = 651 { .daddr = faddr, 652 .saddr = saddr, 653 .tos = tos } }, 654 .proto = sk->sk_protocol, 655 .uli_u = { .ports = 656 { .sport = inet->sport, 657 .dport = dport } } }; 658 security_sk_classify_flow(sk, &fl); 659 err = ip_route_output_flow(sock_net(sk), &rt, &fl, sk, 1); 660 if (err) { 661 if (err == -ENETUNREACH) 662 IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES); 663 goto out; 664 } 665 666 err = -EACCES; 667 if ((rt->rt_flags & RTCF_BROADCAST) && 668 !sock_flag(sk, SOCK_BROADCAST)) 669 goto out; 670 if (connected) 671 sk_dst_set(sk, dst_clone(&rt->u.dst)); 672 } 673 674 if (msg->msg_flags&MSG_CONFIRM) 675 goto do_confirm; 676 back_from_confirm: 677 678 saddr = rt->rt_src; 679 if (!ipc.addr) 680 daddr = ipc.addr = rt->rt_dst; 681 682 lock_sock(sk); 683 if (unlikely(up->pending)) { 684 /* The socket is already corked while preparing it. */ 685 /* ... which is an evident application bug. --ANK */ 686 release_sock(sk); 687 688 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n"); 689 err = -EINVAL; 690 goto out; 691 } 692 /* 693 * Now cork the socket to pend data. 694 */ 695 inet->cork.fl.fl4_dst = daddr; 696 inet->cork.fl.fl_ip_dport = dport; 697 inet->cork.fl.fl4_src = saddr; 698 inet->cork.fl.fl_ip_sport = inet->sport; 699 up->pending = AF_INET; 700 701 do_append_data: 702 up->len += ulen; 703 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 704 err = ip_append_data(sk, getfrag, msg->msg_iov, ulen, 705 sizeof(struct udphdr), &ipc, rt, 706 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 707 if (err) 708 udp_flush_pending_frames(sk); 709 else if (!corkreq) 710 err = udp_push_pending_frames(sk); 711 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 712 up->pending = 0; 713 release_sock(sk); 714 715 out: 716 ip_rt_put(rt); 717 if (free) 718 kfree(ipc.opt); 719 if (!err) 720 return len; 721 /* 722 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 723 * ENOBUFS might not be good (it's not tunable per se), but otherwise 724 * we don't have a good statistic (IpOutDiscards but it can be too many 725 * things). We could add another new stat but at least for now that 726 * seems like overkill. 727 */ 728 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 729 UDP_INC_STATS_USER(UDP_MIB_SNDBUFERRORS, is_udplite); 730 } 731 return err; 732 733 do_confirm: 734 dst_confirm(&rt->u.dst); 735 if (!(msg->msg_flags&MSG_PROBE) || len) 736 goto back_from_confirm; 737 err = 0; 738 goto out; 739 } 740 741 int udp_sendpage(struct sock *sk, struct page *page, int offset, 742 size_t size, int flags) 743 { 744 struct udp_sock *up = udp_sk(sk); 745 int ret; 746 747 if (!up->pending) { 748 struct msghdr msg = { .msg_flags = flags|MSG_MORE }; 749 750 /* Call udp_sendmsg to specify destination address which 751 * sendpage interface can't pass. 752 * This will succeed only when the socket is connected. 753 */ 754 ret = udp_sendmsg(NULL, sk, &msg, 0); 755 if (ret < 0) 756 return ret; 757 } 758 759 lock_sock(sk); 760 761 if (unlikely(!up->pending)) { 762 release_sock(sk); 763 764 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n"); 765 return -EINVAL; 766 } 767 768 ret = ip_append_page(sk, page, offset, size, flags); 769 if (ret == -EOPNOTSUPP) { 770 release_sock(sk); 771 return sock_no_sendpage(sk->sk_socket, page, offset, 772 size, flags); 773 } 774 if (ret < 0) { 775 udp_flush_pending_frames(sk); 776 goto out; 777 } 778 779 up->len += size; 780 if (!(up->corkflag || (flags&MSG_MORE))) 781 ret = udp_push_pending_frames(sk); 782 if (!ret) 783 ret = size; 784 out: 785 release_sock(sk); 786 return ret; 787 } 788 789 /* 790 * IOCTL requests applicable to the UDP protocol 791 */ 792 793 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg) 794 { 795 switch (cmd) { 796 case SIOCOUTQ: 797 { 798 int amount = atomic_read(&sk->sk_wmem_alloc); 799 return put_user(amount, (int __user *)arg); 800 } 801 802 case SIOCINQ: 803 { 804 struct sk_buff *skb; 805 unsigned long amount; 806 807 amount = 0; 808 spin_lock_bh(&sk->sk_receive_queue.lock); 809 skb = skb_peek(&sk->sk_receive_queue); 810 if (skb != NULL) { 811 /* 812 * We will only return the amount 813 * of this packet since that is all 814 * that will be read. 815 */ 816 amount = skb->len - sizeof(struct udphdr); 817 } 818 spin_unlock_bh(&sk->sk_receive_queue.lock); 819 return put_user(amount, (int __user *)arg); 820 } 821 822 default: 823 return -ENOIOCTLCMD; 824 } 825 826 return 0; 827 } 828 829 /* 830 * This should be easy, if there is something there we 831 * return it, otherwise we block. 832 */ 833 834 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 835 size_t len, int noblock, int flags, int *addr_len) 836 { 837 struct inet_sock *inet = inet_sk(sk); 838 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name; 839 struct sk_buff *skb; 840 unsigned int ulen, copied; 841 int peeked; 842 int err; 843 int is_udplite = IS_UDPLITE(sk); 844 845 /* 846 * Check any passed addresses 847 */ 848 if (addr_len) 849 *addr_len=sizeof(*sin); 850 851 if (flags & MSG_ERRQUEUE) 852 return ip_recv_error(sk, msg, len); 853 854 try_again: 855 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0), 856 &peeked, &err); 857 if (!skb) 858 goto out; 859 860 ulen = skb->len - sizeof(struct udphdr); 861 copied = len; 862 if (copied > ulen) 863 copied = ulen; 864 else if (copied < ulen) 865 msg->msg_flags |= MSG_TRUNC; 866 867 /* 868 * If checksum is needed at all, try to do it while copying the 869 * data. If the data is truncated, or if we only want a partial 870 * coverage checksum (UDP-Lite), do it before the copy. 871 */ 872 873 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) { 874 if (udp_lib_checksum_complete(skb)) 875 goto csum_copy_err; 876 } 877 878 if (skb_csum_unnecessary(skb)) 879 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr), 880 msg->msg_iov, copied ); 881 else { 882 err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov); 883 884 if (err == -EINVAL) 885 goto csum_copy_err; 886 } 887 888 if (err) 889 goto out_free; 890 891 if (!peeked) 892 UDP_INC_STATS_USER(UDP_MIB_INDATAGRAMS, is_udplite); 893 894 sock_recv_timestamp(msg, sk, skb); 895 896 /* Copy the address. */ 897 if (sin) 898 { 899 sin->sin_family = AF_INET; 900 sin->sin_port = udp_hdr(skb)->source; 901 sin->sin_addr.s_addr = ip_hdr(skb)->saddr; 902 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 903 } 904 if (inet->cmsg_flags) 905 ip_cmsg_recv(msg, skb); 906 907 err = copied; 908 if (flags & MSG_TRUNC) 909 err = ulen; 910 911 out_free: 912 lock_sock(sk); 913 skb_free_datagram(sk, skb); 914 release_sock(sk); 915 out: 916 return err; 917 918 csum_copy_err: 919 lock_sock(sk); 920 if (!skb_kill_datagram(sk, skb, flags)) 921 UDP_INC_STATS_USER(UDP_MIB_INERRORS, is_udplite); 922 release_sock(sk); 923 924 if (noblock) 925 return -EAGAIN; 926 goto try_again; 927 } 928 929 930 int udp_disconnect(struct sock *sk, int flags) 931 { 932 struct inet_sock *inet = inet_sk(sk); 933 /* 934 * 1003.1g - break association. 935 */ 936 937 sk->sk_state = TCP_CLOSE; 938 inet->daddr = 0; 939 inet->dport = 0; 940 sk->sk_bound_dev_if = 0; 941 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) 942 inet_reset_saddr(sk); 943 944 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { 945 sk->sk_prot->unhash(sk); 946 inet->sport = 0; 947 } 948 sk_dst_reset(sk); 949 return 0; 950 } 951 952 /* returns: 953 * -1: error 954 * 0: success 955 * >0: "udp encap" protocol resubmission 956 * 957 * Note that in the success and error cases, the skb is assumed to 958 * have either been requeued or freed. 959 */ 960 int udp_queue_rcv_skb(struct sock * sk, struct sk_buff *skb) 961 { 962 struct udp_sock *up = udp_sk(sk); 963 int rc; 964 int is_udplite = IS_UDPLITE(sk); 965 966 /* 967 * Charge it to the socket, dropping if the queue is full. 968 */ 969 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) 970 goto drop; 971 nf_reset(skb); 972 973 if (up->encap_type) { 974 /* 975 * This is an encapsulation socket so pass the skb to 976 * the socket's udp_encap_rcv() hook. Otherwise, just 977 * fall through and pass this up the UDP socket. 978 * up->encap_rcv() returns the following value: 979 * =0 if skb was successfully passed to the encap 980 * handler or was discarded by it. 981 * >0 if skb should be passed on to UDP. 982 * <0 if skb should be resubmitted as proto -N 983 */ 984 985 /* if we're overly short, let UDP handle it */ 986 if (skb->len > sizeof(struct udphdr) && 987 up->encap_rcv != NULL) { 988 int ret; 989 990 ret = (*up->encap_rcv)(sk, skb); 991 if (ret <= 0) { 992 UDP_INC_STATS_BH(UDP_MIB_INDATAGRAMS, 993 is_udplite); 994 return -ret; 995 } 996 } 997 998 /* FALLTHROUGH -- it's a UDP Packet */ 999 } 1000 1001 /* 1002 * UDP-Lite specific tests, ignored on UDP sockets 1003 */ 1004 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 1005 1006 /* 1007 * MIB statistics other than incrementing the error count are 1008 * disabled for the following two types of errors: these depend 1009 * on the application settings, not on the functioning of the 1010 * protocol stack as such. 1011 * 1012 * RFC 3828 here recommends (sec 3.3): "There should also be a 1013 * way ... to ... at least let the receiving application block 1014 * delivery of packets with coverage values less than a value 1015 * provided by the application." 1016 */ 1017 if (up->pcrlen == 0) { /* full coverage was set */ 1018 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage " 1019 "%d while full coverage %d requested\n", 1020 UDP_SKB_CB(skb)->cscov, skb->len); 1021 goto drop; 1022 } 1023 /* The next case involves violating the min. coverage requested 1024 * by the receiver. This is subtle: if receiver wants x and x is 1025 * greater than the buffersize/MTU then receiver will complain 1026 * that it wants x while sender emits packets of smaller size y. 1027 * Therefore the above ...()->partial_cov statement is essential. 1028 */ 1029 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 1030 LIMIT_NETDEBUG(KERN_WARNING 1031 "UDPLITE: coverage %d too small, need min %d\n", 1032 UDP_SKB_CB(skb)->cscov, up->pcrlen); 1033 goto drop; 1034 } 1035 } 1036 1037 if (sk->sk_filter) { 1038 if (udp_lib_checksum_complete(skb)) 1039 goto drop; 1040 } 1041 1042 if ((rc = sock_queue_rcv_skb(sk,skb)) < 0) { 1043 /* Note that an ENOMEM error is charged twice */ 1044 if (rc == -ENOMEM) 1045 UDP_INC_STATS_BH(UDP_MIB_RCVBUFERRORS, is_udplite); 1046 goto drop; 1047 } 1048 1049 return 0; 1050 1051 drop: 1052 UDP_INC_STATS_BH(UDP_MIB_INERRORS, is_udplite); 1053 kfree_skb(skb); 1054 return -1; 1055 } 1056 1057 /* 1058 * Multicasts and broadcasts go to each listener. 1059 * 1060 * Note: called only from the BH handler context, 1061 * so we don't need to lock the hashes. 1062 */ 1063 static int __udp4_lib_mcast_deliver(struct sk_buff *skb, 1064 struct udphdr *uh, 1065 __be32 saddr, __be32 daddr, 1066 struct hlist_head udptable[]) 1067 { 1068 struct sock *sk; 1069 int dif; 1070 1071 read_lock(&udp_hash_lock); 1072 sk = sk_head(&udptable[ntohs(uh->dest) & (UDP_HTABLE_SIZE - 1)]); 1073 dif = skb->dev->ifindex; 1074 sk = udp_v4_mcast_next(sk, uh->dest, daddr, uh->source, saddr, dif); 1075 if (sk) { 1076 struct sock *sknext = NULL; 1077 1078 do { 1079 struct sk_buff *skb1 = skb; 1080 1081 sknext = udp_v4_mcast_next(sk_next(sk), uh->dest, daddr, 1082 uh->source, saddr, dif); 1083 if (sknext) 1084 skb1 = skb_clone(skb, GFP_ATOMIC); 1085 1086 if (skb1) { 1087 int ret = 0; 1088 1089 bh_lock_sock_nested(sk); 1090 if (!sock_owned_by_user(sk)) 1091 ret = udp_queue_rcv_skb(sk, skb1); 1092 else 1093 sk_add_backlog(sk, skb1); 1094 bh_unlock_sock(sk); 1095 1096 if (ret > 0) 1097 /* we should probably re-process instead 1098 * of dropping packets here. */ 1099 kfree_skb(skb1); 1100 } 1101 sk = sknext; 1102 } while (sknext); 1103 } else 1104 kfree_skb(skb); 1105 read_unlock(&udp_hash_lock); 1106 return 0; 1107 } 1108 1109 /* Initialize UDP checksum. If exited with zero value (success), 1110 * CHECKSUM_UNNECESSARY means, that no more checks are required. 1111 * Otherwise, csum completion requires chacksumming packet body, 1112 * including udp header and folding it to skb->csum. 1113 */ 1114 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, 1115 int proto) 1116 { 1117 const struct iphdr *iph; 1118 int err; 1119 1120 UDP_SKB_CB(skb)->partial_cov = 0; 1121 UDP_SKB_CB(skb)->cscov = skb->len; 1122 1123 if (proto == IPPROTO_UDPLITE) { 1124 err = udplite_checksum_init(skb, uh); 1125 if (err) 1126 return err; 1127 } 1128 1129 iph = ip_hdr(skb); 1130 if (uh->check == 0) { 1131 skb->ip_summed = CHECKSUM_UNNECESSARY; 1132 } else if (skb->ip_summed == CHECKSUM_COMPLETE) { 1133 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len, 1134 proto, skb->csum)) 1135 skb->ip_summed = CHECKSUM_UNNECESSARY; 1136 } 1137 if (!skb_csum_unnecessary(skb)) 1138 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr, 1139 skb->len, proto, 0); 1140 /* Probably, we should checksum udp header (it should be in cache 1141 * in any case) and data in tiny packets (< rx copybreak). 1142 */ 1143 1144 return 0; 1145 } 1146 1147 /* 1148 * All we need to do is get the socket, and then do a checksum. 1149 */ 1150 1151 int __udp4_lib_rcv(struct sk_buff *skb, struct hlist_head udptable[], 1152 int proto) 1153 { 1154 struct sock *sk; 1155 struct udphdr *uh = udp_hdr(skb); 1156 unsigned short ulen; 1157 struct rtable *rt = (struct rtable*)skb->dst; 1158 __be32 saddr = ip_hdr(skb)->saddr; 1159 __be32 daddr = ip_hdr(skb)->daddr; 1160 1161 /* 1162 * Validate the packet. 1163 */ 1164 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 1165 goto drop; /* No space for header. */ 1166 1167 ulen = ntohs(uh->len); 1168 if (ulen > skb->len) 1169 goto short_packet; 1170 1171 if (proto == IPPROTO_UDP) { 1172 /* UDP validates ulen. */ 1173 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) 1174 goto short_packet; 1175 uh = udp_hdr(skb); 1176 } 1177 1178 if (udp4_csum_init(skb, uh, proto)) 1179 goto csum_error; 1180 1181 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) 1182 return __udp4_lib_mcast_deliver(skb, uh, saddr, daddr, udptable); 1183 1184 sk = __udp4_lib_lookup(dev_net(skb->dev), saddr, uh->source, daddr, 1185 uh->dest, inet_iif(skb), udptable); 1186 1187 if (sk != NULL) { 1188 int ret = 0; 1189 bh_lock_sock_nested(sk); 1190 if (!sock_owned_by_user(sk)) 1191 ret = udp_queue_rcv_skb(sk, skb); 1192 else 1193 sk_add_backlog(sk, skb); 1194 bh_unlock_sock(sk); 1195 sock_put(sk); 1196 1197 /* a return value > 0 means to resubmit the input, but 1198 * it wants the return to be -protocol, or 0 1199 */ 1200 if (ret > 0) 1201 return -ret; 1202 return 0; 1203 } 1204 1205 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 1206 goto drop; 1207 nf_reset(skb); 1208 1209 /* No socket. Drop packet silently, if checksum is wrong */ 1210 if (udp_lib_checksum_complete(skb)) 1211 goto csum_error; 1212 1213 UDP_INC_STATS_BH(UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 1214 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); 1215 1216 /* 1217 * Hmm. We got an UDP packet to a port to which we 1218 * don't wanna listen. Ignore it. 1219 */ 1220 kfree_skb(skb); 1221 return 0; 1222 1223 short_packet: 1224 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From " NIPQUAD_FMT ":%u %d/%d to " NIPQUAD_FMT ":%u\n", 1225 proto == IPPROTO_UDPLITE ? "-Lite" : "", 1226 NIPQUAD(saddr), 1227 ntohs(uh->source), 1228 ulen, 1229 skb->len, 1230 NIPQUAD(daddr), 1231 ntohs(uh->dest)); 1232 goto drop; 1233 1234 csum_error: 1235 /* 1236 * RFC1122: OK. Discards the bad packet silently (as far as 1237 * the network is concerned, anyway) as per 4.1.3.4 (MUST). 1238 */ 1239 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From " NIPQUAD_FMT ":%u to " NIPQUAD_FMT ":%u ulen %d\n", 1240 proto == IPPROTO_UDPLITE ? "-Lite" : "", 1241 NIPQUAD(saddr), 1242 ntohs(uh->source), 1243 NIPQUAD(daddr), 1244 ntohs(uh->dest), 1245 ulen); 1246 drop: 1247 UDP_INC_STATS_BH(UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 1248 kfree_skb(skb); 1249 return 0; 1250 } 1251 1252 int udp_rcv(struct sk_buff *skb) 1253 { 1254 return __udp4_lib_rcv(skb, udp_hash, IPPROTO_UDP); 1255 } 1256 1257 int udp_destroy_sock(struct sock *sk) 1258 { 1259 lock_sock(sk); 1260 udp_flush_pending_frames(sk); 1261 release_sock(sk); 1262 return 0; 1263 } 1264 1265 /* 1266 * Socket option code for UDP 1267 */ 1268 int udp_lib_setsockopt(struct sock *sk, int level, int optname, 1269 char __user *optval, int optlen, 1270 int (*push_pending_frames)(struct sock *)) 1271 { 1272 struct udp_sock *up = udp_sk(sk); 1273 int val; 1274 int err = 0; 1275 int is_udplite = IS_UDPLITE(sk); 1276 1277 if (optlen<sizeof(int)) 1278 return -EINVAL; 1279 1280 if (get_user(val, (int __user *)optval)) 1281 return -EFAULT; 1282 1283 switch (optname) { 1284 case UDP_CORK: 1285 if (val != 0) { 1286 up->corkflag = 1; 1287 } else { 1288 up->corkflag = 0; 1289 lock_sock(sk); 1290 (*push_pending_frames)(sk); 1291 release_sock(sk); 1292 } 1293 break; 1294 1295 case UDP_ENCAP: 1296 switch (val) { 1297 case 0: 1298 case UDP_ENCAP_ESPINUDP: 1299 case UDP_ENCAP_ESPINUDP_NON_IKE: 1300 up->encap_rcv = xfrm4_udp_encap_rcv; 1301 /* FALLTHROUGH */ 1302 case UDP_ENCAP_L2TPINUDP: 1303 up->encap_type = val; 1304 break; 1305 default: 1306 err = -ENOPROTOOPT; 1307 break; 1308 } 1309 break; 1310 1311 /* 1312 * UDP-Lite's partial checksum coverage (RFC 3828). 1313 */ 1314 /* The sender sets actual checksum coverage length via this option. 1315 * The case coverage > packet length is handled by send module. */ 1316 case UDPLITE_SEND_CSCOV: 1317 if (!is_udplite) /* Disable the option on UDP sockets */ 1318 return -ENOPROTOOPT; 1319 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ 1320 val = 8; 1321 up->pcslen = val; 1322 up->pcflag |= UDPLITE_SEND_CC; 1323 break; 1324 1325 /* The receiver specifies a minimum checksum coverage value. To make 1326 * sense, this should be set to at least 8 (as done below). If zero is 1327 * used, this again means full checksum coverage. */ 1328 case UDPLITE_RECV_CSCOV: 1329 if (!is_udplite) /* Disable the option on UDP sockets */ 1330 return -ENOPROTOOPT; 1331 if (val != 0 && val < 8) /* Avoid silly minimal values. */ 1332 val = 8; 1333 up->pcrlen = val; 1334 up->pcflag |= UDPLITE_RECV_CC; 1335 break; 1336 1337 default: 1338 err = -ENOPROTOOPT; 1339 break; 1340 } 1341 1342 return err; 1343 } 1344 1345 int udp_setsockopt(struct sock *sk, int level, int optname, 1346 char __user *optval, int optlen) 1347 { 1348 if (level == SOL_UDP || level == SOL_UDPLITE) 1349 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1350 udp_push_pending_frames); 1351 return ip_setsockopt(sk, level, optname, optval, optlen); 1352 } 1353 1354 #ifdef CONFIG_COMPAT 1355 int compat_udp_setsockopt(struct sock *sk, int level, int optname, 1356 char __user *optval, int optlen) 1357 { 1358 if (level == SOL_UDP || level == SOL_UDPLITE) 1359 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1360 udp_push_pending_frames); 1361 return compat_ip_setsockopt(sk, level, optname, optval, optlen); 1362 } 1363 #endif 1364 1365 int udp_lib_getsockopt(struct sock *sk, int level, int optname, 1366 char __user *optval, int __user *optlen) 1367 { 1368 struct udp_sock *up = udp_sk(sk); 1369 int val, len; 1370 1371 if (get_user(len,optlen)) 1372 return -EFAULT; 1373 1374 len = min_t(unsigned int, len, sizeof(int)); 1375 1376 if (len < 0) 1377 return -EINVAL; 1378 1379 switch (optname) { 1380 case UDP_CORK: 1381 val = up->corkflag; 1382 break; 1383 1384 case UDP_ENCAP: 1385 val = up->encap_type; 1386 break; 1387 1388 /* The following two cannot be changed on UDP sockets, the return is 1389 * always 0 (which corresponds to the full checksum coverage of UDP). */ 1390 case UDPLITE_SEND_CSCOV: 1391 val = up->pcslen; 1392 break; 1393 1394 case UDPLITE_RECV_CSCOV: 1395 val = up->pcrlen; 1396 break; 1397 1398 default: 1399 return -ENOPROTOOPT; 1400 } 1401 1402 if (put_user(len, optlen)) 1403 return -EFAULT; 1404 if (copy_to_user(optval, &val,len)) 1405 return -EFAULT; 1406 return 0; 1407 } 1408 1409 int udp_getsockopt(struct sock *sk, int level, int optname, 1410 char __user *optval, int __user *optlen) 1411 { 1412 if (level == SOL_UDP || level == SOL_UDPLITE) 1413 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1414 return ip_getsockopt(sk, level, optname, optval, optlen); 1415 } 1416 1417 #ifdef CONFIG_COMPAT 1418 int compat_udp_getsockopt(struct sock *sk, int level, int optname, 1419 char __user *optval, int __user *optlen) 1420 { 1421 if (level == SOL_UDP || level == SOL_UDPLITE) 1422 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1423 return compat_ip_getsockopt(sk, level, optname, optval, optlen); 1424 } 1425 #endif 1426 /** 1427 * udp_poll - wait for a UDP event. 1428 * @file - file struct 1429 * @sock - socket 1430 * @wait - poll table 1431 * 1432 * This is same as datagram poll, except for the special case of 1433 * blocking sockets. If application is using a blocking fd 1434 * and a packet with checksum error is in the queue; 1435 * then it could get return from select indicating data available 1436 * but then block when reading it. Add special case code 1437 * to work around these arguably broken applications. 1438 */ 1439 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait) 1440 { 1441 unsigned int mask = datagram_poll(file, sock, wait); 1442 struct sock *sk = sock->sk; 1443 int is_lite = IS_UDPLITE(sk); 1444 1445 /* Check for false positives due to checksum errors */ 1446 if ( (mask & POLLRDNORM) && 1447 !(file->f_flags & O_NONBLOCK) && 1448 !(sk->sk_shutdown & RCV_SHUTDOWN)){ 1449 struct sk_buff_head *rcvq = &sk->sk_receive_queue; 1450 struct sk_buff *skb; 1451 1452 spin_lock_bh(&rcvq->lock); 1453 while ((skb = skb_peek(rcvq)) != NULL && 1454 udp_lib_checksum_complete(skb)) { 1455 UDP_INC_STATS_BH(UDP_MIB_INERRORS, is_lite); 1456 __skb_unlink(skb, rcvq); 1457 kfree_skb(skb); 1458 } 1459 spin_unlock_bh(&rcvq->lock); 1460 1461 /* nothing to see, move along */ 1462 if (skb == NULL) 1463 mask &= ~(POLLIN | POLLRDNORM); 1464 } 1465 1466 return mask; 1467 1468 } 1469 1470 struct proto udp_prot = { 1471 .name = "UDP", 1472 .owner = THIS_MODULE, 1473 .close = udp_lib_close, 1474 .connect = ip4_datagram_connect, 1475 .disconnect = udp_disconnect, 1476 .ioctl = udp_ioctl, 1477 .destroy = udp_destroy_sock, 1478 .setsockopt = udp_setsockopt, 1479 .getsockopt = udp_getsockopt, 1480 .sendmsg = udp_sendmsg, 1481 .recvmsg = udp_recvmsg, 1482 .sendpage = udp_sendpage, 1483 .backlog_rcv = udp_queue_rcv_skb, 1484 .hash = udp_lib_hash, 1485 .unhash = udp_lib_unhash, 1486 .get_port = udp_v4_get_port, 1487 .memory_allocated = &udp_memory_allocated, 1488 .sysctl_mem = sysctl_udp_mem, 1489 .sysctl_wmem = &sysctl_udp_wmem_min, 1490 .sysctl_rmem = &sysctl_udp_rmem_min, 1491 .obj_size = sizeof(struct udp_sock), 1492 .h.udp_hash = udp_hash, 1493 #ifdef CONFIG_COMPAT 1494 .compat_setsockopt = compat_udp_setsockopt, 1495 .compat_getsockopt = compat_udp_getsockopt, 1496 #endif 1497 }; 1498 1499 /* ------------------------------------------------------------------------ */ 1500 #ifdef CONFIG_PROC_FS 1501 1502 static struct sock *udp_get_first(struct seq_file *seq) 1503 { 1504 struct sock *sk; 1505 struct udp_iter_state *state = seq->private; 1506 struct net *net = seq_file_net(seq); 1507 1508 for (state->bucket = 0; state->bucket < UDP_HTABLE_SIZE; ++state->bucket) { 1509 struct hlist_node *node; 1510 sk_for_each(sk, node, state->hashtable + state->bucket) { 1511 if (!net_eq(sock_net(sk), net)) 1512 continue; 1513 if (sk->sk_family == state->family) 1514 goto found; 1515 } 1516 } 1517 sk = NULL; 1518 found: 1519 return sk; 1520 } 1521 1522 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) 1523 { 1524 struct udp_iter_state *state = seq->private; 1525 struct net *net = seq_file_net(seq); 1526 1527 do { 1528 sk = sk_next(sk); 1529 try_again: 1530 ; 1531 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family)); 1532 1533 if (!sk && ++state->bucket < UDP_HTABLE_SIZE) { 1534 sk = sk_head(state->hashtable + state->bucket); 1535 goto try_again; 1536 } 1537 return sk; 1538 } 1539 1540 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) 1541 { 1542 struct sock *sk = udp_get_first(seq); 1543 1544 if (sk) 1545 while (pos && (sk = udp_get_next(seq, sk)) != NULL) 1546 --pos; 1547 return pos ? NULL : sk; 1548 } 1549 1550 static void *udp_seq_start(struct seq_file *seq, loff_t *pos) 1551 __acquires(udp_hash_lock) 1552 { 1553 read_lock(&udp_hash_lock); 1554 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN; 1555 } 1556 1557 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 1558 { 1559 struct sock *sk; 1560 1561 if (v == SEQ_START_TOKEN) 1562 sk = udp_get_idx(seq, 0); 1563 else 1564 sk = udp_get_next(seq, v); 1565 1566 ++*pos; 1567 return sk; 1568 } 1569 1570 static void udp_seq_stop(struct seq_file *seq, void *v) 1571 __releases(udp_hash_lock) 1572 { 1573 read_unlock(&udp_hash_lock); 1574 } 1575 1576 static int udp_seq_open(struct inode *inode, struct file *file) 1577 { 1578 struct udp_seq_afinfo *afinfo = PDE(inode)->data; 1579 struct udp_iter_state *s; 1580 int err; 1581 1582 err = seq_open_net(inode, file, &afinfo->seq_ops, 1583 sizeof(struct udp_iter_state)); 1584 if (err < 0) 1585 return err; 1586 1587 s = ((struct seq_file *)file->private_data)->private; 1588 s->family = afinfo->family; 1589 s->hashtable = afinfo->hashtable; 1590 return err; 1591 } 1592 1593 /* ------------------------------------------------------------------------ */ 1594 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo) 1595 { 1596 struct proc_dir_entry *p; 1597 int rc = 0; 1598 1599 afinfo->seq_fops.open = udp_seq_open; 1600 afinfo->seq_fops.read = seq_read; 1601 afinfo->seq_fops.llseek = seq_lseek; 1602 afinfo->seq_fops.release = seq_release_net; 1603 1604 afinfo->seq_ops.start = udp_seq_start; 1605 afinfo->seq_ops.next = udp_seq_next; 1606 afinfo->seq_ops.stop = udp_seq_stop; 1607 1608 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net, 1609 &afinfo->seq_fops, afinfo); 1610 if (!p) 1611 rc = -ENOMEM; 1612 return rc; 1613 } 1614 1615 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo) 1616 { 1617 proc_net_remove(net, afinfo->name); 1618 } 1619 1620 /* ------------------------------------------------------------------------ */ 1621 static void udp4_format_sock(struct sock *sp, struct seq_file *f, 1622 int bucket, int *len) 1623 { 1624 struct inet_sock *inet = inet_sk(sp); 1625 __be32 dest = inet->daddr; 1626 __be32 src = inet->rcv_saddr; 1627 __u16 destp = ntohs(inet->dport); 1628 __u16 srcp = ntohs(inet->sport); 1629 1630 seq_printf(f, "%4d: %08X:%04X %08X:%04X" 1631 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p%n", 1632 bucket, src, srcp, dest, destp, sp->sk_state, 1633 atomic_read(&sp->sk_wmem_alloc), 1634 atomic_read(&sp->sk_rmem_alloc), 1635 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp), 1636 atomic_read(&sp->sk_refcnt), sp, len); 1637 } 1638 1639 int udp4_seq_show(struct seq_file *seq, void *v) 1640 { 1641 if (v == SEQ_START_TOKEN) 1642 seq_printf(seq, "%-127s\n", 1643 " sl local_address rem_address st tx_queue " 1644 "rx_queue tr tm->when retrnsmt uid timeout " 1645 "inode"); 1646 else { 1647 struct udp_iter_state *state = seq->private; 1648 int len; 1649 1650 udp4_format_sock(v, seq, state->bucket, &len); 1651 seq_printf(seq, "%*s\n", 127 - len ,""); 1652 } 1653 return 0; 1654 } 1655 1656 /* ------------------------------------------------------------------------ */ 1657 static struct udp_seq_afinfo udp4_seq_afinfo = { 1658 .name = "udp", 1659 .family = AF_INET, 1660 .hashtable = udp_hash, 1661 .seq_fops = { 1662 .owner = THIS_MODULE, 1663 }, 1664 .seq_ops = { 1665 .show = udp4_seq_show, 1666 }, 1667 }; 1668 1669 static int udp4_proc_init_net(struct net *net) 1670 { 1671 return udp_proc_register(net, &udp4_seq_afinfo); 1672 } 1673 1674 static void udp4_proc_exit_net(struct net *net) 1675 { 1676 udp_proc_unregister(net, &udp4_seq_afinfo); 1677 } 1678 1679 static struct pernet_operations udp4_net_ops = { 1680 .init = udp4_proc_init_net, 1681 .exit = udp4_proc_exit_net, 1682 }; 1683 1684 int __init udp4_proc_init(void) 1685 { 1686 return register_pernet_subsys(&udp4_net_ops); 1687 } 1688 1689 void udp4_proc_exit(void) 1690 { 1691 unregister_pernet_subsys(&udp4_net_ops); 1692 } 1693 #endif /* CONFIG_PROC_FS */ 1694 1695 void __init udp_init(void) 1696 { 1697 unsigned long limit; 1698 1699 /* Set the pressure threshold up by the same strategy of TCP. It is a 1700 * fraction of global memory that is up to 1/2 at 256 MB, decreasing 1701 * toward zero with the amount of memory, with a floor of 128 pages. 1702 */ 1703 limit = min(nr_all_pages, 1UL<<(28-PAGE_SHIFT)) >> (20-PAGE_SHIFT); 1704 limit = (limit * (nr_all_pages >> (20-PAGE_SHIFT))) >> (PAGE_SHIFT-11); 1705 limit = max(limit, 128UL); 1706 sysctl_udp_mem[0] = limit / 4 * 3; 1707 sysctl_udp_mem[1] = limit; 1708 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2; 1709 1710 sysctl_udp_rmem_min = SK_MEM_QUANTUM; 1711 sysctl_udp_wmem_min = SK_MEM_QUANTUM; 1712 } 1713 1714 EXPORT_SYMBOL(udp_disconnect); 1715 EXPORT_SYMBOL(udp_hash); 1716 EXPORT_SYMBOL(udp_hash_lock); 1717 EXPORT_SYMBOL(udp_ioctl); 1718 EXPORT_SYMBOL(udp_prot); 1719 EXPORT_SYMBOL(udp_sendmsg); 1720 EXPORT_SYMBOL(udp_lib_getsockopt); 1721 EXPORT_SYMBOL(udp_lib_setsockopt); 1722 EXPORT_SYMBOL(udp_poll); 1723 EXPORT_SYMBOL(udp_lib_get_port); 1724 1725 #ifdef CONFIG_PROC_FS 1726 EXPORT_SYMBOL(udp_proc_register); 1727 EXPORT_SYMBOL(udp_proc_unregister); 1728 #endif 1729