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 * Authors: Ross Biro 9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 10 * Arnt Gulbrandsen, <agulbra@nvg.unit.no> 11 * Alan Cox, <Alan.Cox@linux.org> 12 * Hirokazu Takahashi, <taka@valinux.co.jp> 13 * 14 * Fixes: 15 * Alan Cox : verify_area() calls 16 * Alan Cox : stopped close while in use off icmp 17 * messages. Not a fix but a botch that 18 * for udp at least is 'valid'. 19 * Alan Cox : Fixed icmp handling properly 20 * Alan Cox : Correct error for oversized datagrams 21 * Alan Cox : Tidied select() semantics. 22 * Alan Cox : udp_err() fixed properly, also now 23 * select and read wake correctly on errors 24 * Alan Cox : udp_send verify_area moved to avoid mem leak 25 * Alan Cox : UDP can count its memory 26 * Alan Cox : send to an unknown connection causes 27 * an ECONNREFUSED off the icmp, but 28 * does NOT close. 29 * Alan Cox : Switched to new sk_buff handlers. No more backlog! 30 * Alan Cox : Using generic datagram code. Even smaller and the PEEK 31 * bug no longer crashes it. 32 * Fred Van Kempen : Net2e support for sk->broadcast. 33 * Alan Cox : Uses skb_free_datagram 34 * Alan Cox : Added get/set sockopt support. 35 * Alan Cox : Broadcasting without option set returns EACCES. 36 * Alan Cox : No wakeup calls. Instead we now use the callbacks. 37 * Alan Cox : Use ip_tos and ip_ttl 38 * Alan Cox : SNMP Mibs 39 * Alan Cox : MSG_DONTROUTE, and 0.0.0.0 support. 40 * Matt Dillon : UDP length checks. 41 * Alan Cox : Smarter af_inet used properly. 42 * Alan Cox : Use new kernel side addressing. 43 * Alan Cox : Incorrect return on truncated datagram receive. 44 * Arnt Gulbrandsen : New udp_send and stuff 45 * Alan Cox : Cache last socket 46 * Alan Cox : Route cache 47 * Jon Peatfield : Minor efficiency fix to sendto(). 48 * Mike Shaver : RFC1122 checks. 49 * Alan Cox : Nonblocking error fix. 50 * Willy Konynenberg : Transparent proxying support. 51 * Mike McLagan : Routing by source 52 * David S. Miller : New socket lookup architecture. 53 * Last socket cache retained as it 54 * does have a high hit rate. 55 * Olaf Kirch : Don't linearise iovec on sendmsg. 56 * Andi Kleen : Some cleanups, cache destination entry 57 * for connect. 58 * Vitaly E. Lavrov : Transparent proxy revived after year coma. 59 * Melvin Smith : Check msg_name not msg_namelen in sendto(), 60 * return ENOTCONN for unconnected sockets (POSIX) 61 * Janos Farkas : don't deliver multi/broadcasts to a different 62 * bound-to-device socket 63 * Hirokazu Takahashi : HW checksumming for outgoing UDP 64 * datagrams. 65 * Hirokazu Takahashi : sendfile() on UDP works now. 66 * Arnaldo C. Melo : convert /proc/net/udp to seq_file 67 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which 68 * Alexey Kuznetsov: allow both IPv4 and IPv6 sockets to bind 69 * a single port at the same time. 70 * Derek Atkins <derek@ihtfp.com>: Add Encapulation Support 71 * James Chapman : Add L2TP encapsulation type. 72 * 73 * 74 * This program is free software; you can redistribute it and/or 75 * modify it under the terms of the GNU General Public License 76 * as published by the Free Software Foundation; either version 77 * 2 of the License, or (at your option) any later version. 78 */ 79 80 #include <asm/system.h> 81 #include <asm/uaccess.h> 82 #include <asm/ioctls.h> 83 #include <linux/bootmem.h> 84 #include <linux/types.h> 85 #include <linux/fcntl.h> 86 #include <linux/module.h> 87 #include <linux/socket.h> 88 #include <linux/sockios.h> 89 #include <linux/igmp.h> 90 #include <linux/in.h> 91 #include <linux/errno.h> 92 #include <linux/timer.h> 93 #include <linux/mm.h> 94 #include <linux/inet.h> 95 #include <linux/netdevice.h> 96 #include <net/tcp_states.h> 97 #include <linux/skbuff.h> 98 #include <linux/proc_fs.h> 99 #include <linux/seq_file.h> 100 #include <net/net_namespace.h> 101 #include <net/icmp.h> 102 #include <net/route.h> 103 #include <net/checksum.h> 104 #include <net/xfrm.h> 105 #include "udp_impl.h" 106 107 /* 108 * Snmp MIB for the UDP layer 109 */ 110 111 DEFINE_SNMP_STAT(struct udp_mib, udp_stats_in6) __read_mostly; 112 EXPORT_SYMBOL(udp_stats_in6); 113 114 struct hlist_head udp_hash[UDP_HTABLE_SIZE]; 115 DEFINE_RWLOCK(udp_hash_lock); 116 117 int sysctl_udp_mem[3] __read_mostly; 118 int sysctl_udp_rmem_min __read_mostly; 119 int sysctl_udp_wmem_min __read_mostly; 120 121 EXPORT_SYMBOL(sysctl_udp_mem); 122 EXPORT_SYMBOL(sysctl_udp_rmem_min); 123 EXPORT_SYMBOL(sysctl_udp_wmem_min); 124 125 atomic_t udp_memory_allocated; 126 EXPORT_SYMBOL(udp_memory_allocated); 127 128 static inline int __udp_lib_lport_inuse(struct net *net, __u16 num, 129 const struct hlist_head udptable[]) 130 { 131 struct sock *sk; 132 struct hlist_node *node; 133 134 sk_for_each(sk, node, &udptable[udp_hashfn(net, num)]) 135 if (net_eq(sock_net(sk), net) && sk->sk_hash == num) 136 return 1; 137 return 0; 138 } 139 140 /** 141 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6 142 * 143 * @sk: socket struct in question 144 * @snum: port number to look up 145 * @saddr_comp: AF-dependent comparison of bound local IP addresses 146 */ 147 int udp_lib_get_port(struct sock *sk, unsigned short snum, 148 int (*saddr_comp)(const struct sock *sk1, 149 const struct sock *sk2 ) ) 150 { 151 struct hlist_head *udptable = sk->sk_prot->h.udp_hash; 152 struct hlist_node *node; 153 struct hlist_head *head; 154 struct sock *sk2; 155 int error = 1; 156 struct net *net = sock_net(sk); 157 158 write_lock_bh(&udp_hash_lock); 159 160 if (!snum) { 161 int i, low, high, remaining; 162 unsigned rover, best, best_size_so_far; 163 164 inet_get_local_port_range(&low, &high); 165 remaining = (high - low) + 1; 166 167 best_size_so_far = UINT_MAX; 168 best = rover = net_random() % remaining + low; 169 170 /* 1st pass: look for empty (or shortest) hash chain */ 171 for (i = 0; i < UDP_HTABLE_SIZE; i++) { 172 int size = 0; 173 174 head = &udptable[udp_hashfn(net, rover)]; 175 if (hlist_empty(head)) 176 goto gotit; 177 178 sk_for_each(sk2, node, head) { 179 if (++size >= best_size_so_far) 180 goto next; 181 } 182 best_size_so_far = size; 183 best = rover; 184 next: 185 /* fold back if end of range */ 186 if (++rover > high) 187 rover = low + ((rover - low) 188 & (UDP_HTABLE_SIZE - 1)); 189 190 191 } 192 193 /* 2nd pass: find hole in shortest hash chain */ 194 rover = best; 195 for (i = 0; i < (1 << 16) / UDP_HTABLE_SIZE; i++) { 196 if (! __udp_lib_lport_inuse(net, rover, udptable)) 197 goto gotit; 198 rover += UDP_HTABLE_SIZE; 199 if (rover > high) 200 rover = low + ((rover - low) 201 & (UDP_HTABLE_SIZE - 1)); 202 } 203 204 205 /* All ports in use! */ 206 goto fail; 207 208 gotit: 209 snum = rover; 210 } else { 211 head = &udptable[udp_hashfn(net, snum)]; 212 213 sk_for_each(sk2, node, head) 214 if (sk2->sk_hash == snum && 215 sk2 != sk && 216 net_eq(sock_net(sk2), net) && 217 (!sk2->sk_reuse || !sk->sk_reuse) && 218 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if 219 || sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 220 (*saddr_comp)(sk, sk2) ) 221 goto fail; 222 } 223 224 inet_sk(sk)->num = snum; 225 sk->sk_hash = snum; 226 if (sk_unhashed(sk)) { 227 head = &udptable[udp_hashfn(net, snum)]; 228 sk_add_node(sk, head); 229 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 230 } 231 error = 0; 232 fail: 233 write_unlock_bh(&udp_hash_lock); 234 return error; 235 } 236 237 static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2) 238 { 239 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2); 240 241 return ( !ipv6_only_sock(sk2) && 242 (!inet1->rcv_saddr || !inet2->rcv_saddr || 243 inet1->rcv_saddr == inet2->rcv_saddr )); 244 } 245 246 int udp_v4_get_port(struct sock *sk, unsigned short snum) 247 { 248 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal); 249 } 250 251 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try 252 * harder than this. -DaveM 253 */ 254 static struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, 255 __be16 sport, __be32 daddr, __be16 dport, 256 int dif, struct hlist_head udptable[]) 257 { 258 struct sock *sk, *result = NULL; 259 struct hlist_node *node; 260 unsigned short hnum = ntohs(dport); 261 int badness = -1; 262 263 read_lock(&udp_hash_lock); 264 sk_for_each(sk, node, &udptable[udp_hashfn(net, hnum)]) { 265 struct inet_sock *inet = inet_sk(sk); 266 267 if (net_eq(sock_net(sk), net) && sk->sk_hash == hnum && 268 !ipv6_only_sock(sk)) { 269 int score = (sk->sk_family == PF_INET ? 1 : 0); 270 if (inet->rcv_saddr) { 271 if (inet->rcv_saddr != daddr) 272 continue; 273 score+=2; 274 } 275 if (inet->daddr) { 276 if (inet->daddr != saddr) 277 continue; 278 score+=2; 279 } 280 if (inet->dport) { 281 if (inet->dport != sport) 282 continue; 283 score+=2; 284 } 285 if (sk->sk_bound_dev_if) { 286 if (sk->sk_bound_dev_if != dif) 287 continue; 288 score+=2; 289 } 290 if (score == 9) { 291 result = sk; 292 break; 293 } else if (score > badness) { 294 result = sk; 295 badness = score; 296 } 297 } 298 } 299 if (result) 300 sock_hold(result); 301 read_unlock(&udp_hash_lock); 302 return result; 303 } 304 305 static inline struct sock *udp_v4_mcast_next(struct sock *sk, 306 __be16 loc_port, __be32 loc_addr, 307 __be16 rmt_port, __be32 rmt_addr, 308 int dif) 309 { 310 struct hlist_node *node; 311 struct sock *s = sk; 312 unsigned short hnum = ntohs(loc_port); 313 314 sk_for_each_from(s, node) { 315 struct inet_sock *inet = inet_sk(s); 316 317 if (s->sk_hash != hnum || 318 (inet->daddr && inet->daddr != rmt_addr) || 319 (inet->dport != rmt_port && inet->dport) || 320 (inet->rcv_saddr && inet->rcv_saddr != loc_addr) || 321 ipv6_only_sock(s) || 322 (s->sk_bound_dev_if && s->sk_bound_dev_if != dif)) 323 continue; 324 if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif)) 325 continue; 326 goto found; 327 } 328 s = NULL; 329 found: 330 return s; 331 } 332 333 /* 334 * This routine is called by the ICMP module when it gets some 335 * sort of error condition. If err < 0 then the socket should 336 * be closed and the error returned to the user. If err > 0 337 * it's just the icmp type << 8 | icmp code. 338 * Header points to the ip header of the error packet. We move 339 * on past this. Then (as it used to claim before adjustment) 340 * header points to the first 8 bytes of the udp header. We need 341 * to find the appropriate port. 342 */ 343 344 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct hlist_head udptable[]) 345 { 346 struct inet_sock *inet; 347 struct iphdr *iph = (struct iphdr*)skb->data; 348 struct udphdr *uh = (struct udphdr*)(skb->data+(iph->ihl<<2)); 349 const int type = icmp_hdr(skb)->type; 350 const int code = icmp_hdr(skb)->code; 351 struct sock *sk; 352 int harderr; 353 int err; 354 struct net *net = dev_net(skb->dev); 355 356 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest, 357 iph->saddr, uh->source, skb->dev->ifindex, udptable); 358 if (sk == NULL) { 359 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS); 360 return; /* No socket for error */ 361 } 362 363 err = 0; 364 harderr = 0; 365 inet = inet_sk(sk); 366 367 switch (type) { 368 default: 369 case ICMP_TIME_EXCEEDED: 370 err = EHOSTUNREACH; 371 break; 372 case ICMP_SOURCE_QUENCH: 373 goto out; 374 case ICMP_PARAMETERPROB: 375 err = EPROTO; 376 harderr = 1; 377 break; 378 case ICMP_DEST_UNREACH: 379 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */ 380 if (inet->pmtudisc != IP_PMTUDISC_DONT) { 381 err = EMSGSIZE; 382 harderr = 1; 383 break; 384 } 385 goto out; 386 } 387 err = EHOSTUNREACH; 388 if (code <= NR_ICMP_UNREACH) { 389 harderr = icmp_err_convert[code].fatal; 390 err = icmp_err_convert[code].errno; 391 } 392 break; 393 } 394 395 /* 396 * RFC1122: OK. Passes ICMP errors back to application, as per 397 * 4.1.3.3. 398 */ 399 if (!inet->recverr) { 400 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 401 goto out; 402 } else { 403 ip_icmp_error(sk, skb, err, uh->dest, info, (u8*)(uh+1)); 404 } 405 sk->sk_err = err; 406 sk->sk_error_report(sk); 407 out: 408 sock_put(sk); 409 } 410 411 void udp_err(struct sk_buff *skb, u32 info) 412 { 413 __udp4_lib_err(skb, info, udp_hash); 414 } 415 416 /* 417 * Throw away all pending data and cancel the corking. Socket is locked. 418 */ 419 void udp_flush_pending_frames(struct sock *sk) 420 { 421 struct udp_sock *up = udp_sk(sk); 422 423 if (up->pending) { 424 up->len = 0; 425 up->pending = 0; 426 ip_flush_pending_frames(sk); 427 } 428 } 429 EXPORT_SYMBOL(udp_flush_pending_frames); 430 431 /** 432 * udp4_hwcsum_outgoing - handle outgoing HW checksumming 433 * @sk: socket we are sending on 434 * @skb: sk_buff containing the filled-in UDP header 435 * (checksum field must be zeroed out) 436 */ 437 static void udp4_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb, 438 __be32 src, __be32 dst, int len ) 439 { 440 unsigned int offset; 441 struct udphdr *uh = udp_hdr(skb); 442 __wsum csum = 0; 443 444 if (skb_queue_len(&sk->sk_write_queue) == 1) { 445 /* 446 * Only one fragment on the socket. 447 */ 448 skb->csum_start = skb_transport_header(skb) - skb->head; 449 skb->csum_offset = offsetof(struct udphdr, check); 450 uh->check = ~csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, 0); 451 } else { 452 /* 453 * HW-checksum won't work as there are two or more 454 * fragments on the socket so that all csums of sk_buffs 455 * should be together 456 */ 457 offset = skb_transport_offset(skb); 458 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0); 459 460 skb->ip_summed = CHECKSUM_NONE; 461 462 skb_queue_walk(&sk->sk_write_queue, skb) { 463 csum = csum_add(csum, skb->csum); 464 } 465 466 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); 467 if (uh->check == 0) 468 uh->check = CSUM_MANGLED_0; 469 } 470 } 471 472 /* 473 * Push out all pending data as one UDP datagram. Socket is locked. 474 */ 475 static int udp_push_pending_frames(struct sock *sk) 476 { 477 struct udp_sock *up = udp_sk(sk); 478 struct inet_sock *inet = inet_sk(sk); 479 struct flowi *fl = &inet->cork.fl; 480 struct sk_buff *skb; 481 struct udphdr *uh; 482 int err = 0; 483 int is_udplite = IS_UDPLITE(sk); 484 __wsum csum = 0; 485 486 /* Grab the skbuff where UDP header space exists. */ 487 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL) 488 goto out; 489 490 /* 491 * Create a UDP header 492 */ 493 uh = udp_hdr(skb); 494 uh->source = fl->fl_ip_sport; 495 uh->dest = fl->fl_ip_dport; 496 uh->len = htons(up->len); 497 uh->check = 0; 498 499 if (is_udplite) /* UDP-Lite */ 500 csum = udplite_csum_outgoing(sk, skb); 501 502 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */ 503 504 skb->ip_summed = CHECKSUM_NONE; 505 goto send; 506 507 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 508 509 udp4_hwcsum_outgoing(sk, skb, fl->fl4_src,fl->fl4_dst, up->len); 510 goto send; 511 512 } else /* `normal' UDP */ 513 csum = udp_csum_outgoing(sk, skb); 514 515 /* add protocol-dependent pseudo-header */ 516 uh->check = csum_tcpudp_magic(fl->fl4_src, fl->fl4_dst, up->len, 517 sk->sk_protocol, csum ); 518 if (uh->check == 0) 519 uh->check = CSUM_MANGLED_0; 520 521 send: 522 err = ip_push_pending_frames(sk); 523 out: 524 up->len = 0; 525 up->pending = 0; 526 if (!err) 527 UDP_INC_STATS_USER(sock_net(sk), 528 UDP_MIB_OUTDATAGRAMS, is_udplite); 529 return err; 530 } 531 532 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 533 size_t len) 534 { 535 struct inet_sock *inet = inet_sk(sk); 536 struct udp_sock *up = udp_sk(sk); 537 int ulen = len; 538 struct ipcm_cookie ipc; 539 struct rtable *rt = NULL; 540 int free = 0; 541 int connected = 0; 542 __be32 daddr, faddr, saddr; 543 __be16 dport; 544 u8 tos; 545 int err, is_udplite = IS_UDPLITE(sk); 546 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; 547 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 548 549 if (len > 0xFFFF) 550 return -EMSGSIZE; 551 552 /* 553 * Check the flags. 554 */ 555 556 if (msg->msg_flags&MSG_OOB) /* Mirror BSD error message compatibility */ 557 return -EOPNOTSUPP; 558 559 ipc.opt = NULL; 560 561 if (up->pending) { 562 /* 563 * There are pending frames. 564 * The socket lock must be held while it's corked. 565 */ 566 lock_sock(sk); 567 if (likely(up->pending)) { 568 if (unlikely(up->pending != AF_INET)) { 569 release_sock(sk); 570 return -EINVAL; 571 } 572 goto do_append_data; 573 } 574 release_sock(sk); 575 } 576 ulen += sizeof(struct udphdr); 577 578 /* 579 * Get and verify the address. 580 */ 581 if (msg->msg_name) { 582 struct sockaddr_in * usin = (struct sockaddr_in*)msg->msg_name; 583 if (msg->msg_namelen < sizeof(*usin)) 584 return -EINVAL; 585 if (usin->sin_family != AF_INET) { 586 if (usin->sin_family != AF_UNSPEC) 587 return -EAFNOSUPPORT; 588 } 589 590 daddr = usin->sin_addr.s_addr; 591 dport = usin->sin_port; 592 if (dport == 0) 593 return -EINVAL; 594 } else { 595 if (sk->sk_state != TCP_ESTABLISHED) 596 return -EDESTADDRREQ; 597 daddr = inet->daddr; 598 dport = inet->dport; 599 /* Open fast path for connected socket. 600 Route will not be used, if at least one option is set. 601 */ 602 connected = 1; 603 } 604 ipc.addr = inet->saddr; 605 606 ipc.oif = sk->sk_bound_dev_if; 607 if (msg->msg_controllen) { 608 err = ip_cmsg_send(sock_net(sk), msg, &ipc); 609 if (err) 610 return err; 611 if (ipc.opt) 612 free = 1; 613 connected = 0; 614 } 615 if (!ipc.opt) 616 ipc.opt = inet->opt; 617 618 saddr = ipc.addr; 619 ipc.addr = faddr = daddr; 620 621 if (ipc.opt && ipc.opt->srr) { 622 if (!daddr) 623 return -EINVAL; 624 faddr = ipc.opt->faddr; 625 connected = 0; 626 } 627 tos = RT_TOS(inet->tos); 628 if (sock_flag(sk, SOCK_LOCALROUTE) || 629 (msg->msg_flags & MSG_DONTROUTE) || 630 (ipc.opt && ipc.opt->is_strictroute)) { 631 tos |= RTO_ONLINK; 632 connected = 0; 633 } 634 635 if (ipv4_is_multicast(daddr)) { 636 if (!ipc.oif) 637 ipc.oif = inet->mc_index; 638 if (!saddr) 639 saddr = inet->mc_addr; 640 connected = 0; 641 } 642 643 if (connected) 644 rt = (struct rtable*)sk_dst_check(sk, 0); 645 646 if (rt == NULL) { 647 struct flowi fl = { .oif = ipc.oif, 648 .nl_u = { .ip4_u = 649 { .daddr = faddr, 650 .saddr = saddr, 651 .tos = tos } }, 652 .proto = sk->sk_protocol, 653 .uli_u = { .ports = 654 { .sport = inet->sport, 655 .dport = dport } } }; 656 struct net *net = sock_net(sk); 657 658 security_sk_classify_flow(sk, &fl); 659 err = ip_route_output_flow(net, &rt, &fl, sk, 1); 660 if (err) { 661 if (err == -ENETUNREACH) 662 IP_INC_STATS_BH(net, 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(sock_net(sk), 730 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(sock_net(sk), 894 UDP_MIB_INDATAGRAMS, is_udplite); 895 896 sock_recv_timestamp(msg, sk, skb); 897 898 /* Copy the address. */ 899 if (sin) 900 { 901 sin->sin_family = AF_INET; 902 sin->sin_port = udp_hdr(skb)->source; 903 sin->sin_addr.s_addr = ip_hdr(skb)->saddr; 904 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 905 } 906 if (inet->cmsg_flags) 907 ip_cmsg_recv(msg, skb); 908 909 err = copied; 910 if (flags & MSG_TRUNC) 911 err = ulen; 912 913 out_free: 914 lock_sock(sk); 915 skb_free_datagram(sk, skb); 916 release_sock(sk); 917 out: 918 return err; 919 920 csum_copy_err: 921 lock_sock(sk); 922 if (!skb_kill_datagram(sk, skb, flags)) 923 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 924 release_sock(sk); 925 926 if (noblock) 927 return -EAGAIN; 928 goto try_again; 929 } 930 931 932 int udp_disconnect(struct sock *sk, int flags) 933 { 934 struct inet_sock *inet = inet_sk(sk); 935 /* 936 * 1003.1g - break association. 937 */ 938 939 sk->sk_state = TCP_CLOSE; 940 inet->daddr = 0; 941 inet->dport = 0; 942 sk->sk_bound_dev_if = 0; 943 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) 944 inet_reset_saddr(sk); 945 946 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { 947 sk->sk_prot->unhash(sk); 948 inet->sport = 0; 949 } 950 sk_dst_reset(sk); 951 return 0; 952 } 953 954 /* returns: 955 * -1: error 956 * 0: success 957 * >0: "udp encap" protocol resubmission 958 * 959 * Note that in the success and error cases, the skb is assumed to 960 * have either been requeued or freed. 961 */ 962 int udp_queue_rcv_skb(struct sock * sk, struct sk_buff *skb) 963 { 964 struct udp_sock *up = udp_sk(sk); 965 int rc; 966 int is_udplite = IS_UDPLITE(sk); 967 968 /* 969 * Charge it to the socket, dropping if the queue is full. 970 */ 971 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) 972 goto drop; 973 nf_reset(skb); 974 975 if (up->encap_type) { 976 /* 977 * This is an encapsulation socket so pass the skb to 978 * the socket's udp_encap_rcv() hook. Otherwise, just 979 * fall through and pass this up the UDP socket. 980 * up->encap_rcv() returns the following value: 981 * =0 if skb was successfully passed to the encap 982 * handler or was discarded by it. 983 * >0 if skb should be passed on to UDP. 984 * <0 if skb should be resubmitted as proto -N 985 */ 986 987 /* if we're overly short, let UDP handle it */ 988 if (skb->len > sizeof(struct udphdr) && 989 up->encap_rcv != NULL) { 990 int ret; 991 992 ret = (*up->encap_rcv)(sk, skb); 993 if (ret <= 0) { 994 UDP_INC_STATS_BH(sock_net(sk), 995 UDP_MIB_INDATAGRAMS, 996 is_udplite); 997 return -ret; 998 } 999 } 1000 1001 /* FALLTHROUGH -- it's a UDP Packet */ 1002 } 1003 1004 /* 1005 * UDP-Lite specific tests, ignored on UDP sockets 1006 */ 1007 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 1008 1009 /* 1010 * MIB statistics other than incrementing the error count are 1011 * disabled for the following two types of errors: these depend 1012 * on the application settings, not on the functioning of the 1013 * protocol stack as such. 1014 * 1015 * RFC 3828 here recommends (sec 3.3): "There should also be a 1016 * way ... to ... at least let the receiving application block 1017 * delivery of packets with coverage values less than a value 1018 * provided by the application." 1019 */ 1020 if (up->pcrlen == 0) { /* full coverage was set */ 1021 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage " 1022 "%d while full coverage %d requested\n", 1023 UDP_SKB_CB(skb)->cscov, skb->len); 1024 goto drop; 1025 } 1026 /* The next case involves violating the min. coverage requested 1027 * by the receiver. This is subtle: if receiver wants x and x is 1028 * greater than the buffersize/MTU then receiver will complain 1029 * that it wants x while sender emits packets of smaller size y. 1030 * Therefore the above ...()->partial_cov statement is essential. 1031 */ 1032 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 1033 LIMIT_NETDEBUG(KERN_WARNING 1034 "UDPLITE: coverage %d too small, need min %d\n", 1035 UDP_SKB_CB(skb)->cscov, up->pcrlen); 1036 goto drop; 1037 } 1038 } 1039 1040 if (sk->sk_filter) { 1041 if (udp_lib_checksum_complete(skb)) 1042 goto drop; 1043 } 1044 1045 if ((rc = sock_queue_rcv_skb(sk,skb)) < 0) { 1046 /* Note that an ENOMEM error is charged twice */ 1047 if (rc == -ENOMEM) { 1048 UDP_INC_STATS_BH(sock_net(sk), 1049 UDP_MIB_RCVBUFERRORS, is_udplite); 1050 atomic_inc(&sk->sk_drops); 1051 } 1052 goto drop; 1053 } 1054 1055 return 0; 1056 1057 drop: 1058 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1059 kfree_skb(skb); 1060 return -1; 1061 } 1062 1063 /* 1064 * Multicasts and broadcasts go to each listener. 1065 * 1066 * Note: called only from the BH handler context, 1067 * so we don't need to lock the hashes. 1068 */ 1069 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 1070 struct udphdr *uh, 1071 __be32 saddr, __be32 daddr, 1072 struct hlist_head udptable[]) 1073 { 1074 struct sock *sk; 1075 int dif; 1076 1077 read_lock(&udp_hash_lock); 1078 sk = sk_head(&udptable[udp_hashfn(net, ntohs(uh->dest))]); 1079 dif = skb->dev->ifindex; 1080 sk = udp_v4_mcast_next(sk, uh->dest, daddr, uh->source, saddr, dif); 1081 if (sk) { 1082 struct sock *sknext = NULL; 1083 1084 do { 1085 struct sk_buff *skb1 = skb; 1086 1087 sknext = udp_v4_mcast_next(sk_next(sk), uh->dest, daddr, 1088 uh->source, saddr, dif); 1089 if (sknext) 1090 skb1 = skb_clone(skb, GFP_ATOMIC); 1091 1092 if (skb1) { 1093 int ret = 0; 1094 1095 bh_lock_sock_nested(sk); 1096 if (!sock_owned_by_user(sk)) 1097 ret = udp_queue_rcv_skb(sk, skb1); 1098 else 1099 sk_add_backlog(sk, skb1); 1100 bh_unlock_sock(sk); 1101 1102 if (ret > 0) 1103 /* we should probably re-process instead 1104 * of dropping packets here. */ 1105 kfree_skb(skb1); 1106 } 1107 sk = sknext; 1108 } while (sknext); 1109 } else 1110 kfree_skb(skb); 1111 read_unlock(&udp_hash_lock); 1112 return 0; 1113 } 1114 1115 /* Initialize UDP checksum. If exited with zero value (success), 1116 * CHECKSUM_UNNECESSARY means, that no more checks are required. 1117 * Otherwise, csum completion requires chacksumming packet body, 1118 * including udp header and folding it to skb->csum. 1119 */ 1120 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, 1121 int proto) 1122 { 1123 const struct iphdr *iph; 1124 int err; 1125 1126 UDP_SKB_CB(skb)->partial_cov = 0; 1127 UDP_SKB_CB(skb)->cscov = skb->len; 1128 1129 if (proto == IPPROTO_UDPLITE) { 1130 err = udplite_checksum_init(skb, uh); 1131 if (err) 1132 return err; 1133 } 1134 1135 iph = ip_hdr(skb); 1136 if (uh->check == 0) { 1137 skb->ip_summed = CHECKSUM_UNNECESSARY; 1138 } else if (skb->ip_summed == CHECKSUM_COMPLETE) { 1139 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len, 1140 proto, skb->csum)) 1141 skb->ip_summed = CHECKSUM_UNNECESSARY; 1142 } 1143 if (!skb_csum_unnecessary(skb)) 1144 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr, 1145 skb->len, proto, 0); 1146 /* Probably, we should checksum udp header (it should be in cache 1147 * in any case) and data in tiny packets (< rx copybreak). 1148 */ 1149 1150 return 0; 1151 } 1152 1153 /* 1154 * All we need to do is get the socket, and then do a checksum. 1155 */ 1156 1157 int __udp4_lib_rcv(struct sk_buff *skb, struct hlist_head udptable[], 1158 int proto) 1159 { 1160 struct sock *sk; 1161 struct udphdr *uh = udp_hdr(skb); 1162 unsigned short ulen; 1163 struct rtable *rt = (struct rtable*)skb->dst; 1164 __be32 saddr = ip_hdr(skb)->saddr; 1165 __be32 daddr = ip_hdr(skb)->daddr; 1166 struct net *net = dev_net(skb->dev); 1167 1168 /* 1169 * Validate the packet. 1170 */ 1171 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 1172 goto drop; /* No space for header. */ 1173 1174 ulen = ntohs(uh->len); 1175 if (ulen > skb->len) 1176 goto short_packet; 1177 1178 if (proto == IPPROTO_UDP) { 1179 /* UDP validates ulen. */ 1180 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) 1181 goto short_packet; 1182 uh = udp_hdr(skb); 1183 } 1184 1185 if (udp4_csum_init(skb, uh, proto)) 1186 goto csum_error; 1187 1188 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) 1189 return __udp4_lib_mcast_deliver(net, skb, uh, 1190 saddr, daddr, udptable); 1191 1192 sk = __udp4_lib_lookup(net, saddr, uh->source, daddr, 1193 uh->dest, inet_iif(skb), udptable); 1194 1195 if (sk != NULL) { 1196 int ret = 0; 1197 bh_lock_sock_nested(sk); 1198 if (!sock_owned_by_user(sk)) 1199 ret = udp_queue_rcv_skb(sk, skb); 1200 else 1201 sk_add_backlog(sk, skb); 1202 bh_unlock_sock(sk); 1203 sock_put(sk); 1204 1205 /* a return value > 0 means to resubmit the input, but 1206 * it wants the return to be -protocol, or 0 1207 */ 1208 if (ret > 0) 1209 return -ret; 1210 return 0; 1211 } 1212 1213 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 1214 goto drop; 1215 nf_reset(skb); 1216 1217 /* No socket. Drop packet silently, if checksum is wrong */ 1218 if (udp_lib_checksum_complete(skb)) 1219 goto csum_error; 1220 1221 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 1222 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); 1223 1224 /* 1225 * Hmm. We got an UDP packet to a port to which we 1226 * don't wanna listen. Ignore it. 1227 */ 1228 kfree_skb(skb); 1229 return 0; 1230 1231 short_packet: 1232 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From " NIPQUAD_FMT ":%u %d/%d to " NIPQUAD_FMT ":%u\n", 1233 proto == IPPROTO_UDPLITE ? "-Lite" : "", 1234 NIPQUAD(saddr), 1235 ntohs(uh->source), 1236 ulen, 1237 skb->len, 1238 NIPQUAD(daddr), 1239 ntohs(uh->dest)); 1240 goto drop; 1241 1242 csum_error: 1243 /* 1244 * RFC1122: OK. Discards the bad packet silently (as far as 1245 * the network is concerned, anyway) as per 4.1.3.4 (MUST). 1246 */ 1247 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From " NIPQUAD_FMT ":%u to " NIPQUAD_FMT ":%u ulen %d\n", 1248 proto == IPPROTO_UDPLITE ? "-Lite" : "", 1249 NIPQUAD(saddr), 1250 ntohs(uh->source), 1251 NIPQUAD(daddr), 1252 ntohs(uh->dest), 1253 ulen); 1254 drop: 1255 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 1256 kfree_skb(skb); 1257 return 0; 1258 } 1259 1260 int udp_rcv(struct sk_buff *skb) 1261 { 1262 return __udp4_lib_rcv(skb, udp_hash, IPPROTO_UDP); 1263 } 1264 1265 void udp_destroy_sock(struct sock *sk) 1266 { 1267 lock_sock(sk); 1268 udp_flush_pending_frames(sk); 1269 release_sock(sk); 1270 } 1271 1272 /* 1273 * Socket option code for UDP 1274 */ 1275 int udp_lib_setsockopt(struct sock *sk, int level, int optname, 1276 char __user *optval, int optlen, 1277 int (*push_pending_frames)(struct sock *)) 1278 { 1279 struct udp_sock *up = udp_sk(sk); 1280 int val; 1281 int err = 0; 1282 int is_udplite = IS_UDPLITE(sk); 1283 1284 if (optlen<sizeof(int)) 1285 return -EINVAL; 1286 1287 if (get_user(val, (int __user *)optval)) 1288 return -EFAULT; 1289 1290 switch (optname) { 1291 case UDP_CORK: 1292 if (val != 0) { 1293 up->corkflag = 1; 1294 } else { 1295 up->corkflag = 0; 1296 lock_sock(sk); 1297 (*push_pending_frames)(sk); 1298 release_sock(sk); 1299 } 1300 break; 1301 1302 case UDP_ENCAP: 1303 switch (val) { 1304 case 0: 1305 case UDP_ENCAP_ESPINUDP: 1306 case UDP_ENCAP_ESPINUDP_NON_IKE: 1307 up->encap_rcv = xfrm4_udp_encap_rcv; 1308 /* FALLTHROUGH */ 1309 case UDP_ENCAP_L2TPINUDP: 1310 up->encap_type = val; 1311 break; 1312 default: 1313 err = -ENOPROTOOPT; 1314 break; 1315 } 1316 break; 1317 1318 /* 1319 * UDP-Lite's partial checksum coverage (RFC 3828). 1320 */ 1321 /* The sender sets actual checksum coverage length via this option. 1322 * The case coverage > packet length is handled by send module. */ 1323 case UDPLITE_SEND_CSCOV: 1324 if (!is_udplite) /* Disable the option on UDP sockets */ 1325 return -ENOPROTOOPT; 1326 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ 1327 val = 8; 1328 else if (val > USHORT_MAX) 1329 val = USHORT_MAX; 1330 up->pcslen = val; 1331 up->pcflag |= UDPLITE_SEND_CC; 1332 break; 1333 1334 /* The receiver specifies a minimum checksum coverage value. To make 1335 * sense, this should be set to at least 8 (as done below). If zero is 1336 * used, this again means full checksum coverage. */ 1337 case UDPLITE_RECV_CSCOV: 1338 if (!is_udplite) /* Disable the option on UDP sockets */ 1339 return -ENOPROTOOPT; 1340 if (val != 0 && val < 8) /* Avoid silly minimal values. */ 1341 val = 8; 1342 else if (val > USHORT_MAX) 1343 val = USHORT_MAX; 1344 up->pcrlen = val; 1345 up->pcflag |= UDPLITE_RECV_CC; 1346 break; 1347 1348 default: 1349 err = -ENOPROTOOPT; 1350 break; 1351 } 1352 1353 return err; 1354 } 1355 1356 int 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 ip_setsockopt(sk, level, optname, optval, optlen); 1363 } 1364 1365 #ifdef CONFIG_COMPAT 1366 int compat_udp_setsockopt(struct sock *sk, int level, int optname, 1367 char __user *optval, int optlen) 1368 { 1369 if (level == SOL_UDP || level == SOL_UDPLITE) 1370 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1371 udp_push_pending_frames); 1372 return compat_ip_setsockopt(sk, level, optname, optval, optlen); 1373 } 1374 #endif 1375 1376 int udp_lib_getsockopt(struct sock *sk, int level, int optname, 1377 char __user *optval, int __user *optlen) 1378 { 1379 struct udp_sock *up = udp_sk(sk); 1380 int val, len; 1381 1382 if (get_user(len,optlen)) 1383 return -EFAULT; 1384 1385 len = min_t(unsigned int, len, sizeof(int)); 1386 1387 if (len < 0) 1388 return -EINVAL; 1389 1390 switch (optname) { 1391 case UDP_CORK: 1392 val = up->corkflag; 1393 break; 1394 1395 case UDP_ENCAP: 1396 val = up->encap_type; 1397 break; 1398 1399 /* The following two cannot be changed on UDP sockets, the return is 1400 * always 0 (which corresponds to the full checksum coverage of UDP). */ 1401 case UDPLITE_SEND_CSCOV: 1402 val = up->pcslen; 1403 break; 1404 1405 case UDPLITE_RECV_CSCOV: 1406 val = up->pcrlen; 1407 break; 1408 1409 default: 1410 return -ENOPROTOOPT; 1411 } 1412 1413 if (put_user(len, optlen)) 1414 return -EFAULT; 1415 if (copy_to_user(optval, &val,len)) 1416 return -EFAULT; 1417 return 0; 1418 } 1419 1420 int udp_getsockopt(struct sock *sk, int level, int optname, 1421 char __user *optval, int __user *optlen) 1422 { 1423 if (level == SOL_UDP || level == SOL_UDPLITE) 1424 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1425 return ip_getsockopt(sk, level, optname, optval, optlen); 1426 } 1427 1428 #ifdef CONFIG_COMPAT 1429 int compat_udp_getsockopt(struct sock *sk, int level, int optname, 1430 char __user *optval, int __user *optlen) 1431 { 1432 if (level == SOL_UDP || level == SOL_UDPLITE) 1433 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1434 return compat_ip_getsockopt(sk, level, optname, optval, optlen); 1435 } 1436 #endif 1437 /** 1438 * udp_poll - wait for a UDP event. 1439 * @file - file struct 1440 * @sock - socket 1441 * @wait - poll table 1442 * 1443 * This is same as datagram poll, except for the special case of 1444 * blocking sockets. If application is using a blocking fd 1445 * and a packet with checksum error is in the queue; 1446 * then it could get return from select indicating data available 1447 * but then block when reading it. Add special case code 1448 * to work around these arguably broken applications. 1449 */ 1450 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait) 1451 { 1452 unsigned int mask = datagram_poll(file, sock, wait); 1453 struct sock *sk = sock->sk; 1454 int is_lite = IS_UDPLITE(sk); 1455 1456 /* Check for false positives due to checksum errors */ 1457 if ( (mask & POLLRDNORM) && 1458 !(file->f_flags & O_NONBLOCK) && 1459 !(sk->sk_shutdown & RCV_SHUTDOWN)){ 1460 struct sk_buff_head *rcvq = &sk->sk_receive_queue; 1461 struct sk_buff *skb; 1462 1463 spin_lock_bh(&rcvq->lock); 1464 while ((skb = skb_peek(rcvq)) != NULL && 1465 udp_lib_checksum_complete(skb)) { 1466 UDP_INC_STATS_BH(sock_net(sk), 1467 UDP_MIB_INERRORS, is_lite); 1468 __skb_unlink(skb, rcvq); 1469 kfree_skb(skb); 1470 } 1471 spin_unlock_bh(&rcvq->lock); 1472 1473 /* nothing to see, move along */ 1474 if (skb == NULL) 1475 mask &= ~(POLLIN | POLLRDNORM); 1476 } 1477 1478 return mask; 1479 1480 } 1481 1482 struct proto udp_prot = { 1483 .name = "UDP", 1484 .owner = THIS_MODULE, 1485 .close = udp_lib_close, 1486 .connect = ip4_datagram_connect, 1487 .disconnect = udp_disconnect, 1488 .ioctl = udp_ioctl, 1489 .destroy = udp_destroy_sock, 1490 .setsockopt = udp_setsockopt, 1491 .getsockopt = udp_getsockopt, 1492 .sendmsg = udp_sendmsg, 1493 .recvmsg = udp_recvmsg, 1494 .sendpage = udp_sendpage, 1495 .backlog_rcv = udp_queue_rcv_skb, 1496 .hash = udp_lib_hash, 1497 .unhash = udp_lib_unhash, 1498 .get_port = udp_v4_get_port, 1499 .memory_allocated = &udp_memory_allocated, 1500 .sysctl_mem = sysctl_udp_mem, 1501 .sysctl_wmem = &sysctl_udp_wmem_min, 1502 .sysctl_rmem = &sysctl_udp_rmem_min, 1503 .obj_size = sizeof(struct udp_sock), 1504 .h.udp_hash = udp_hash, 1505 #ifdef CONFIG_COMPAT 1506 .compat_setsockopt = compat_udp_setsockopt, 1507 .compat_getsockopt = compat_udp_getsockopt, 1508 #endif 1509 }; 1510 1511 /* ------------------------------------------------------------------------ */ 1512 #ifdef CONFIG_PROC_FS 1513 1514 static struct sock *udp_get_first(struct seq_file *seq) 1515 { 1516 struct sock *sk; 1517 struct udp_iter_state *state = seq->private; 1518 struct net *net = seq_file_net(seq); 1519 1520 for (state->bucket = 0; state->bucket < UDP_HTABLE_SIZE; ++state->bucket) { 1521 struct hlist_node *node; 1522 sk_for_each(sk, node, state->hashtable + state->bucket) { 1523 if (!net_eq(sock_net(sk), net)) 1524 continue; 1525 if (sk->sk_family == state->family) 1526 goto found; 1527 } 1528 } 1529 sk = NULL; 1530 found: 1531 return sk; 1532 } 1533 1534 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) 1535 { 1536 struct udp_iter_state *state = seq->private; 1537 struct net *net = seq_file_net(seq); 1538 1539 do { 1540 sk = sk_next(sk); 1541 try_again: 1542 ; 1543 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family)); 1544 1545 if (!sk && ++state->bucket < UDP_HTABLE_SIZE) { 1546 sk = sk_head(state->hashtable + state->bucket); 1547 goto try_again; 1548 } 1549 return sk; 1550 } 1551 1552 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) 1553 { 1554 struct sock *sk = udp_get_first(seq); 1555 1556 if (sk) 1557 while (pos && (sk = udp_get_next(seq, sk)) != NULL) 1558 --pos; 1559 return pos ? NULL : sk; 1560 } 1561 1562 static void *udp_seq_start(struct seq_file *seq, loff_t *pos) 1563 __acquires(udp_hash_lock) 1564 { 1565 read_lock(&udp_hash_lock); 1566 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN; 1567 } 1568 1569 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 1570 { 1571 struct sock *sk; 1572 1573 if (v == SEQ_START_TOKEN) 1574 sk = udp_get_idx(seq, 0); 1575 else 1576 sk = udp_get_next(seq, v); 1577 1578 ++*pos; 1579 return sk; 1580 } 1581 1582 static void udp_seq_stop(struct seq_file *seq, void *v) 1583 __releases(udp_hash_lock) 1584 { 1585 read_unlock(&udp_hash_lock); 1586 } 1587 1588 static int udp_seq_open(struct inode *inode, struct file *file) 1589 { 1590 struct udp_seq_afinfo *afinfo = PDE(inode)->data; 1591 struct udp_iter_state *s; 1592 int err; 1593 1594 err = seq_open_net(inode, file, &afinfo->seq_ops, 1595 sizeof(struct udp_iter_state)); 1596 if (err < 0) 1597 return err; 1598 1599 s = ((struct seq_file *)file->private_data)->private; 1600 s->family = afinfo->family; 1601 s->hashtable = afinfo->hashtable; 1602 return err; 1603 } 1604 1605 /* ------------------------------------------------------------------------ */ 1606 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo) 1607 { 1608 struct proc_dir_entry *p; 1609 int rc = 0; 1610 1611 afinfo->seq_fops.open = udp_seq_open; 1612 afinfo->seq_fops.read = seq_read; 1613 afinfo->seq_fops.llseek = seq_lseek; 1614 afinfo->seq_fops.release = seq_release_net; 1615 1616 afinfo->seq_ops.start = udp_seq_start; 1617 afinfo->seq_ops.next = udp_seq_next; 1618 afinfo->seq_ops.stop = udp_seq_stop; 1619 1620 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net, 1621 &afinfo->seq_fops, afinfo); 1622 if (!p) 1623 rc = -ENOMEM; 1624 return rc; 1625 } 1626 1627 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo) 1628 { 1629 proc_net_remove(net, afinfo->name); 1630 } 1631 1632 /* ------------------------------------------------------------------------ */ 1633 static void udp4_format_sock(struct sock *sp, struct seq_file *f, 1634 int bucket, int *len) 1635 { 1636 struct inet_sock *inet = inet_sk(sp); 1637 __be32 dest = inet->daddr; 1638 __be32 src = inet->rcv_saddr; 1639 __u16 destp = ntohs(inet->dport); 1640 __u16 srcp = ntohs(inet->sport); 1641 1642 seq_printf(f, "%4d: %08X:%04X %08X:%04X" 1643 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %d%n", 1644 bucket, src, srcp, dest, destp, sp->sk_state, 1645 atomic_read(&sp->sk_wmem_alloc), 1646 atomic_read(&sp->sk_rmem_alloc), 1647 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp), 1648 atomic_read(&sp->sk_refcnt), sp, 1649 atomic_read(&sp->sk_drops), len); 1650 } 1651 1652 int udp4_seq_show(struct seq_file *seq, void *v) 1653 { 1654 if (v == SEQ_START_TOKEN) 1655 seq_printf(seq, "%-127s\n", 1656 " sl local_address rem_address st tx_queue " 1657 "rx_queue tr tm->when retrnsmt uid timeout " 1658 "inode ref pointer drops"); 1659 else { 1660 struct udp_iter_state *state = seq->private; 1661 int len; 1662 1663 udp4_format_sock(v, seq, state->bucket, &len); 1664 seq_printf(seq, "%*s\n", 127 - len ,""); 1665 } 1666 return 0; 1667 } 1668 1669 /* ------------------------------------------------------------------------ */ 1670 static struct udp_seq_afinfo udp4_seq_afinfo = { 1671 .name = "udp", 1672 .family = AF_INET, 1673 .hashtable = udp_hash, 1674 .seq_fops = { 1675 .owner = THIS_MODULE, 1676 }, 1677 .seq_ops = { 1678 .show = udp4_seq_show, 1679 }, 1680 }; 1681 1682 static int udp4_proc_init_net(struct net *net) 1683 { 1684 return udp_proc_register(net, &udp4_seq_afinfo); 1685 } 1686 1687 static void udp4_proc_exit_net(struct net *net) 1688 { 1689 udp_proc_unregister(net, &udp4_seq_afinfo); 1690 } 1691 1692 static struct pernet_operations udp4_net_ops = { 1693 .init = udp4_proc_init_net, 1694 .exit = udp4_proc_exit_net, 1695 }; 1696 1697 int __init udp4_proc_init(void) 1698 { 1699 return register_pernet_subsys(&udp4_net_ops); 1700 } 1701 1702 void udp4_proc_exit(void) 1703 { 1704 unregister_pernet_subsys(&udp4_net_ops); 1705 } 1706 #endif /* CONFIG_PROC_FS */ 1707 1708 void __init udp_init(void) 1709 { 1710 unsigned long limit; 1711 1712 /* Set the pressure threshold up by the same strategy of TCP. It is a 1713 * fraction of global memory that is up to 1/2 at 256 MB, decreasing 1714 * toward zero with the amount of memory, with a floor of 128 pages. 1715 */ 1716 limit = min(nr_all_pages, 1UL<<(28-PAGE_SHIFT)) >> (20-PAGE_SHIFT); 1717 limit = (limit * (nr_all_pages >> (20-PAGE_SHIFT))) >> (PAGE_SHIFT-11); 1718 limit = max(limit, 128UL); 1719 sysctl_udp_mem[0] = limit / 4 * 3; 1720 sysctl_udp_mem[1] = limit; 1721 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2; 1722 1723 sysctl_udp_rmem_min = SK_MEM_QUANTUM; 1724 sysctl_udp_wmem_min = SK_MEM_QUANTUM; 1725 } 1726 1727 EXPORT_SYMBOL(udp_disconnect); 1728 EXPORT_SYMBOL(udp_hash); 1729 EXPORT_SYMBOL(udp_hash_lock); 1730 EXPORT_SYMBOL(udp_ioctl); 1731 EXPORT_SYMBOL(udp_prot); 1732 EXPORT_SYMBOL(udp_sendmsg); 1733 EXPORT_SYMBOL(udp_lib_getsockopt); 1734 EXPORT_SYMBOL(udp_lib_setsockopt); 1735 EXPORT_SYMBOL(udp_poll); 1736 EXPORT_SYMBOL(udp_lib_get_port); 1737 1738 #ifdef CONFIG_PROC_FS 1739 EXPORT_SYMBOL(udp_proc_register); 1740 EXPORT_SYMBOL(udp_proc_unregister); 1741 #endif 1742