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@lxorguk.ukuu.org.uk> 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 struct hlist_head udp_hash[UDP_HTABLE_SIZE]; 112 DEFINE_RWLOCK(udp_hash_lock); 113 114 int sysctl_udp_mem[3] __read_mostly; 115 int sysctl_udp_rmem_min __read_mostly; 116 int sysctl_udp_wmem_min __read_mostly; 117 118 EXPORT_SYMBOL(sysctl_udp_mem); 119 EXPORT_SYMBOL(sysctl_udp_rmem_min); 120 EXPORT_SYMBOL(sysctl_udp_wmem_min); 121 122 atomic_t udp_memory_allocated; 123 EXPORT_SYMBOL(udp_memory_allocated); 124 125 static int udp_lib_lport_inuse(struct net *net, __u16 num, 126 const struct hlist_head udptable[], 127 struct sock *sk, 128 int (*saddr_comp)(const struct sock *sk1, 129 const struct sock *sk2)) 130 { 131 struct sock *sk2; 132 struct hlist_node *node; 133 134 sk_for_each(sk2, node, &udptable[udp_hashfn(net, num)]) 135 if (net_eq(sock_net(sk2), net) && 136 sk2 != sk && 137 sk2->sk_hash == num && 138 (!sk2->sk_reuse || !sk->sk_reuse) && 139 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if 140 || sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 141 (*saddr_comp)(sk, sk2)) 142 return 1; 143 return 0; 144 } 145 146 /** 147 * udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6 148 * 149 * @sk: socket struct in question 150 * @snum: port number to look up 151 * @saddr_comp: AF-dependent comparison of bound local IP addresses 152 */ 153 int udp_lib_get_port(struct sock *sk, unsigned short snum, 154 int (*saddr_comp)(const struct sock *sk1, 155 const struct sock *sk2 ) ) 156 { 157 struct hlist_head *udptable = sk->sk_prot->h.udp_hash; 158 int error = 1; 159 struct net *net = sock_net(sk); 160 161 write_lock_bh(&udp_hash_lock); 162 163 if (!snum) { 164 int low, high, remaining; 165 unsigned rand; 166 unsigned short first; 167 168 inet_get_local_port_range(&low, &high); 169 remaining = (high - low) + 1; 170 171 rand = net_random(); 172 snum = first = rand % remaining + low; 173 rand |= 1; 174 while (udp_lib_lport_inuse(net, snum, udptable, sk, 175 saddr_comp)) { 176 do { 177 snum = snum + rand; 178 } while (snum < low || snum > high); 179 if (snum == first) 180 goto fail; 181 } 182 } else if (udp_lib_lport_inuse(net, snum, udptable, sk, saddr_comp)) 183 goto fail; 184 185 inet_sk(sk)->num = snum; 186 sk->sk_hash = snum; 187 if (sk_unhashed(sk)) { 188 sk_add_node(sk, &udptable[udp_hashfn(net, snum)]); 189 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 190 } 191 error = 0; 192 fail: 193 write_unlock_bh(&udp_hash_lock); 194 return error; 195 } 196 197 static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2) 198 { 199 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2); 200 201 return ( !ipv6_only_sock(sk2) && 202 (!inet1->rcv_saddr || !inet2->rcv_saddr || 203 inet1->rcv_saddr == inet2->rcv_saddr )); 204 } 205 206 int udp_v4_get_port(struct sock *sk, unsigned short snum) 207 { 208 return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal); 209 } 210 211 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try 212 * harder than this. -DaveM 213 */ 214 static struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr, 215 __be16 sport, __be32 daddr, __be16 dport, 216 int dif, struct hlist_head udptable[]) 217 { 218 struct sock *sk, *result = NULL; 219 struct hlist_node *node; 220 unsigned short hnum = ntohs(dport); 221 int badness = -1; 222 223 read_lock(&udp_hash_lock); 224 sk_for_each(sk, node, &udptable[udp_hashfn(net, hnum)]) { 225 struct inet_sock *inet = inet_sk(sk); 226 227 if (net_eq(sock_net(sk), net) && sk->sk_hash == hnum && 228 !ipv6_only_sock(sk)) { 229 int score = (sk->sk_family == PF_INET ? 1 : 0); 230 if (inet->rcv_saddr) { 231 if (inet->rcv_saddr != daddr) 232 continue; 233 score+=2; 234 } 235 if (inet->daddr) { 236 if (inet->daddr != saddr) 237 continue; 238 score+=2; 239 } 240 if (inet->dport) { 241 if (inet->dport != sport) 242 continue; 243 score+=2; 244 } 245 if (sk->sk_bound_dev_if) { 246 if (sk->sk_bound_dev_if != dif) 247 continue; 248 score+=2; 249 } 250 if (score == 9) { 251 result = sk; 252 break; 253 } else if (score > badness) { 254 result = sk; 255 badness = score; 256 } 257 } 258 } 259 if (result) 260 sock_hold(result); 261 read_unlock(&udp_hash_lock); 262 return result; 263 } 264 265 static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb, 266 __be16 sport, __be16 dport, 267 struct hlist_head udptable[]) 268 { 269 struct sock *sk; 270 const struct iphdr *iph = ip_hdr(skb); 271 272 if (unlikely(sk = skb_steal_sock(skb))) 273 return sk; 274 else 275 return __udp4_lib_lookup(dev_net(skb->dst->dev), iph->saddr, sport, 276 iph->daddr, dport, inet_iif(skb), 277 udptable); 278 } 279 280 struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport, 281 __be32 daddr, __be16 dport, int dif) 282 { 283 return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, udp_hash); 284 } 285 EXPORT_SYMBOL_GPL(udp4_lib_lookup); 286 287 static inline struct sock *udp_v4_mcast_next(struct net *net, struct sock *sk, 288 __be16 loc_port, __be32 loc_addr, 289 __be16 rmt_port, __be32 rmt_addr, 290 int dif) 291 { 292 struct hlist_node *node; 293 struct sock *s = sk; 294 unsigned short hnum = ntohs(loc_port); 295 296 sk_for_each_from(s, node) { 297 struct inet_sock *inet = inet_sk(s); 298 299 if (!net_eq(sock_net(s), net) || 300 s->sk_hash != hnum || 301 (inet->daddr && inet->daddr != rmt_addr) || 302 (inet->dport != rmt_port && inet->dport) || 303 (inet->rcv_saddr && inet->rcv_saddr != loc_addr) || 304 ipv6_only_sock(s) || 305 (s->sk_bound_dev_if && s->sk_bound_dev_if != dif)) 306 continue; 307 if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif)) 308 continue; 309 goto found; 310 } 311 s = NULL; 312 found: 313 return s; 314 } 315 316 /* 317 * This routine is called by the ICMP module when it gets some 318 * sort of error condition. If err < 0 then the socket should 319 * be closed and the error returned to the user. If err > 0 320 * it's just the icmp type << 8 | icmp code. 321 * Header points to the ip header of the error packet. We move 322 * on past this. Then (as it used to claim before adjustment) 323 * header points to the first 8 bytes of the udp header. We need 324 * to find the appropriate port. 325 */ 326 327 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct hlist_head udptable[]) 328 { 329 struct inet_sock *inet; 330 struct iphdr *iph = (struct iphdr*)skb->data; 331 struct udphdr *uh = (struct udphdr*)(skb->data+(iph->ihl<<2)); 332 const int type = icmp_hdr(skb)->type; 333 const int code = icmp_hdr(skb)->code; 334 struct sock *sk; 335 int harderr; 336 int err; 337 struct net *net = dev_net(skb->dev); 338 339 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest, 340 iph->saddr, uh->source, skb->dev->ifindex, udptable); 341 if (sk == NULL) { 342 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS); 343 return; /* No socket for error */ 344 } 345 346 err = 0; 347 harderr = 0; 348 inet = inet_sk(sk); 349 350 switch (type) { 351 default: 352 case ICMP_TIME_EXCEEDED: 353 err = EHOSTUNREACH; 354 break; 355 case ICMP_SOURCE_QUENCH: 356 goto out; 357 case ICMP_PARAMETERPROB: 358 err = EPROTO; 359 harderr = 1; 360 break; 361 case ICMP_DEST_UNREACH: 362 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */ 363 if (inet->pmtudisc != IP_PMTUDISC_DONT) { 364 err = EMSGSIZE; 365 harderr = 1; 366 break; 367 } 368 goto out; 369 } 370 err = EHOSTUNREACH; 371 if (code <= NR_ICMP_UNREACH) { 372 harderr = icmp_err_convert[code].fatal; 373 err = icmp_err_convert[code].errno; 374 } 375 break; 376 } 377 378 /* 379 * RFC1122: OK. Passes ICMP errors back to application, as per 380 * 4.1.3.3. 381 */ 382 if (!inet->recverr) { 383 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 384 goto out; 385 } else { 386 ip_icmp_error(sk, skb, err, uh->dest, info, (u8*)(uh+1)); 387 } 388 sk->sk_err = err; 389 sk->sk_error_report(sk); 390 out: 391 sock_put(sk); 392 } 393 394 void udp_err(struct sk_buff *skb, u32 info) 395 { 396 __udp4_lib_err(skb, info, udp_hash); 397 } 398 399 /* 400 * Throw away all pending data and cancel the corking. Socket is locked. 401 */ 402 void udp_flush_pending_frames(struct sock *sk) 403 { 404 struct udp_sock *up = udp_sk(sk); 405 406 if (up->pending) { 407 up->len = 0; 408 up->pending = 0; 409 ip_flush_pending_frames(sk); 410 } 411 } 412 EXPORT_SYMBOL(udp_flush_pending_frames); 413 414 /** 415 * udp4_hwcsum_outgoing - handle outgoing HW checksumming 416 * @sk: socket we are sending on 417 * @skb: sk_buff containing the filled-in UDP header 418 * (checksum field must be zeroed out) 419 */ 420 static void udp4_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb, 421 __be32 src, __be32 dst, int len ) 422 { 423 unsigned int offset; 424 struct udphdr *uh = udp_hdr(skb); 425 __wsum csum = 0; 426 427 if (skb_queue_len(&sk->sk_write_queue) == 1) { 428 /* 429 * Only one fragment on the socket. 430 */ 431 skb->csum_start = skb_transport_header(skb) - skb->head; 432 skb->csum_offset = offsetof(struct udphdr, check); 433 uh->check = ~csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, 0); 434 } else { 435 /* 436 * HW-checksum won't work as there are two or more 437 * fragments on the socket so that all csums of sk_buffs 438 * should be together 439 */ 440 offset = skb_transport_offset(skb); 441 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0); 442 443 skb->ip_summed = CHECKSUM_NONE; 444 445 skb_queue_walk(&sk->sk_write_queue, skb) { 446 csum = csum_add(csum, skb->csum); 447 } 448 449 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); 450 if (uh->check == 0) 451 uh->check = CSUM_MANGLED_0; 452 } 453 } 454 455 /* 456 * Push out all pending data as one UDP datagram. Socket is locked. 457 */ 458 static int udp_push_pending_frames(struct sock *sk) 459 { 460 struct udp_sock *up = udp_sk(sk); 461 struct inet_sock *inet = inet_sk(sk); 462 struct flowi *fl = &inet->cork.fl; 463 struct sk_buff *skb; 464 struct udphdr *uh; 465 int err = 0; 466 int is_udplite = IS_UDPLITE(sk); 467 __wsum csum = 0; 468 469 /* Grab the skbuff where UDP header space exists. */ 470 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL) 471 goto out; 472 473 /* 474 * Create a UDP header 475 */ 476 uh = udp_hdr(skb); 477 uh->source = fl->fl_ip_sport; 478 uh->dest = fl->fl_ip_dport; 479 uh->len = htons(up->len); 480 uh->check = 0; 481 482 if (is_udplite) /* UDP-Lite */ 483 csum = udplite_csum_outgoing(sk, skb); 484 485 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */ 486 487 skb->ip_summed = CHECKSUM_NONE; 488 goto send; 489 490 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 491 492 udp4_hwcsum_outgoing(sk, skb, fl->fl4_src,fl->fl4_dst, up->len); 493 goto send; 494 495 } else /* `normal' UDP */ 496 csum = udp_csum_outgoing(sk, skb); 497 498 /* add protocol-dependent pseudo-header */ 499 uh->check = csum_tcpudp_magic(fl->fl4_src, fl->fl4_dst, up->len, 500 sk->sk_protocol, csum ); 501 if (uh->check == 0) 502 uh->check = CSUM_MANGLED_0; 503 504 send: 505 err = ip_push_pending_frames(sk); 506 out: 507 up->len = 0; 508 up->pending = 0; 509 if (!err) 510 UDP_INC_STATS_USER(sock_net(sk), 511 UDP_MIB_OUTDATAGRAMS, is_udplite); 512 return err; 513 } 514 515 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 516 size_t len) 517 { 518 struct inet_sock *inet = inet_sk(sk); 519 struct udp_sock *up = udp_sk(sk); 520 int ulen = len; 521 struct ipcm_cookie ipc; 522 struct rtable *rt = NULL; 523 int free = 0; 524 int connected = 0; 525 __be32 daddr, faddr, saddr; 526 __be16 dport; 527 u8 tos; 528 int err, is_udplite = IS_UDPLITE(sk); 529 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; 530 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 531 532 if (len > 0xFFFF) 533 return -EMSGSIZE; 534 535 /* 536 * Check the flags. 537 */ 538 539 if (msg->msg_flags&MSG_OOB) /* Mirror BSD error message compatibility */ 540 return -EOPNOTSUPP; 541 542 ipc.opt = NULL; 543 544 if (up->pending) { 545 /* 546 * There are pending frames. 547 * The socket lock must be held while it's corked. 548 */ 549 lock_sock(sk); 550 if (likely(up->pending)) { 551 if (unlikely(up->pending != AF_INET)) { 552 release_sock(sk); 553 return -EINVAL; 554 } 555 goto do_append_data; 556 } 557 release_sock(sk); 558 } 559 ulen += sizeof(struct udphdr); 560 561 /* 562 * Get and verify the address. 563 */ 564 if (msg->msg_name) { 565 struct sockaddr_in * usin = (struct sockaddr_in*)msg->msg_name; 566 if (msg->msg_namelen < sizeof(*usin)) 567 return -EINVAL; 568 if (usin->sin_family != AF_INET) { 569 if (usin->sin_family != AF_UNSPEC) 570 return -EAFNOSUPPORT; 571 } 572 573 daddr = usin->sin_addr.s_addr; 574 dport = usin->sin_port; 575 if (dport == 0) 576 return -EINVAL; 577 } else { 578 if (sk->sk_state != TCP_ESTABLISHED) 579 return -EDESTADDRREQ; 580 daddr = inet->daddr; 581 dport = inet->dport; 582 /* Open fast path for connected socket. 583 Route will not be used, if at least one option is set. 584 */ 585 connected = 1; 586 } 587 ipc.addr = inet->saddr; 588 589 ipc.oif = sk->sk_bound_dev_if; 590 if (msg->msg_controllen) { 591 err = ip_cmsg_send(sock_net(sk), msg, &ipc); 592 if (err) 593 return err; 594 if (ipc.opt) 595 free = 1; 596 connected = 0; 597 } 598 if (!ipc.opt) 599 ipc.opt = inet->opt; 600 601 saddr = ipc.addr; 602 ipc.addr = faddr = daddr; 603 604 if (ipc.opt && ipc.opt->srr) { 605 if (!daddr) 606 return -EINVAL; 607 faddr = ipc.opt->faddr; 608 connected = 0; 609 } 610 tos = RT_TOS(inet->tos); 611 if (sock_flag(sk, SOCK_LOCALROUTE) || 612 (msg->msg_flags & MSG_DONTROUTE) || 613 (ipc.opt && ipc.opt->is_strictroute)) { 614 tos |= RTO_ONLINK; 615 connected = 0; 616 } 617 618 if (ipv4_is_multicast(daddr)) { 619 if (!ipc.oif) 620 ipc.oif = inet->mc_index; 621 if (!saddr) 622 saddr = inet->mc_addr; 623 connected = 0; 624 } 625 626 if (connected) 627 rt = (struct rtable*)sk_dst_check(sk, 0); 628 629 if (rt == NULL) { 630 struct flowi fl = { .oif = ipc.oif, 631 .nl_u = { .ip4_u = 632 { .daddr = faddr, 633 .saddr = saddr, 634 .tos = tos } }, 635 .proto = sk->sk_protocol, 636 .flags = inet_sk_flowi_flags(sk), 637 .uli_u = { .ports = 638 { .sport = inet->sport, 639 .dport = dport } } }; 640 struct net *net = sock_net(sk); 641 642 security_sk_classify_flow(sk, &fl); 643 err = ip_route_output_flow(net, &rt, &fl, sk, 1); 644 if (err) { 645 if (err == -ENETUNREACH) 646 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES); 647 goto out; 648 } 649 650 err = -EACCES; 651 if ((rt->rt_flags & RTCF_BROADCAST) && 652 !sock_flag(sk, SOCK_BROADCAST)) 653 goto out; 654 if (connected) 655 sk_dst_set(sk, dst_clone(&rt->u.dst)); 656 } 657 658 if (msg->msg_flags&MSG_CONFIRM) 659 goto do_confirm; 660 back_from_confirm: 661 662 saddr = rt->rt_src; 663 if (!ipc.addr) 664 daddr = ipc.addr = rt->rt_dst; 665 666 lock_sock(sk); 667 if (unlikely(up->pending)) { 668 /* The socket is already corked while preparing it. */ 669 /* ... which is an evident application bug. --ANK */ 670 release_sock(sk); 671 672 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n"); 673 err = -EINVAL; 674 goto out; 675 } 676 /* 677 * Now cork the socket to pend data. 678 */ 679 inet->cork.fl.fl4_dst = daddr; 680 inet->cork.fl.fl_ip_dport = dport; 681 inet->cork.fl.fl4_src = saddr; 682 inet->cork.fl.fl_ip_sport = inet->sport; 683 up->pending = AF_INET; 684 685 do_append_data: 686 up->len += ulen; 687 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 688 err = ip_append_data(sk, getfrag, msg->msg_iov, ulen, 689 sizeof(struct udphdr), &ipc, rt, 690 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 691 if (err) 692 udp_flush_pending_frames(sk); 693 else if (!corkreq) 694 err = udp_push_pending_frames(sk); 695 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 696 up->pending = 0; 697 release_sock(sk); 698 699 out: 700 ip_rt_put(rt); 701 if (free) 702 kfree(ipc.opt); 703 if (!err) 704 return len; 705 /* 706 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 707 * ENOBUFS might not be good (it's not tunable per se), but otherwise 708 * we don't have a good statistic (IpOutDiscards but it can be too many 709 * things). We could add another new stat but at least for now that 710 * seems like overkill. 711 */ 712 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 713 UDP_INC_STATS_USER(sock_net(sk), 714 UDP_MIB_SNDBUFERRORS, is_udplite); 715 } 716 return err; 717 718 do_confirm: 719 dst_confirm(&rt->u.dst); 720 if (!(msg->msg_flags&MSG_PROBE) || len) 721 goto back_from_confirm; 722 err = 0; 723 goto out; 724 } 725 726 int udp_sendpage(struct sock *sk, struct page *page, int offset, 727 size_t size, int flags) 728 { 729 struct udp_sock *up = udp_sk(sk); 730 int ret; 731 732 if (!up->pending) { 733 struct msghdr msg = { .msg_flags = flags|MSG_MORE }; 734 735 /* Call udp_sendmsg to specify destination address which 736 * sendpage interface can't pass. 737 * This will succeed only when the socket is connected. 738 */ 739 ret = udp_sendmsg(NULL, sk, &msg, 0); 740 if (ret < 0) 741 return ret; 742 } 743 744 lock_sock(sk); 745 746 if (unlikely(!up->pending)) { 747 release_sock(sk); 748 749 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n"); 750 return -EINVAL; 751 } 752 753 ret = ip_append_page(sk, page, offset, size, flags); 754 if (ret == -EOPNOTSUPP) { 755 release_sock(sk); 756 return sock_no_sendpage(sk->sk_socket, page, offset, 757 size, flags); 758 } 759 if (ret < 0) { 760 udp_flush_pending_frames(sk); 761 goto out; 762 } 763 764 up->len += size; 765 if (!(up->corkflag || (flags&MSG_MORE))) 766 ret = udp_push_pending_frames(sk); 767 if (!ret) 768 ret = size; 769 out: 770 release_sock(sk); 771 return ret; 772 } 773 774 /* 775 * IOCTL requests applicable to the UDP protocol 776 */ 777 778 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg) 779 { 780 switch (cmd) { 781 case SIOCOUTQ: 782 { 783 int amount = atomic_read(&sk->sk_wmem_alloc); 784 return put_user(amount, (int __user *)arg); 785 } 786 787 case SIOCINQ: 788 { 789 struct sk_buff *skb; 790 unsigned long amount; 791 792 amount = 0; 793 spin_lock_bh(&sk->sk_receive_queue.lock); 794 skb = skb_peek(&sk->sk_receive_queue); 795 if (skb != NULL) { 796 /* 797 * We will only return the amount 798 * of this packet since that is all 799 * that will be read. 800 */ 801 amount = skb->len - sizeof(struct udphdr); 802 } 803 spin_unlock_bh(&sk->sk_receive_queue.lock); 804 return put_user(amount, (int __user *)arg); 805 } 806 807 default: 808 return -ENOIOCTLCMD; 809 } 810 811 return 0; 812 } 813 814 /* 815 * This should be easy, if there is something there we 816 * return it, otherwise we block. 817 */ 818 819 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 820 size_t len, int noblock, int flags, int *addr_len) 821 { 822 struct inet_sock *inet = inet_sk(sk); 823 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name; 824 struct sk_buff *skb; 825 unsigned int ulen, copied; 826 int peeked; 827 int err; 828 int is_udplite = IS_UDPLITE(sk); 829 830 /* 831 * Check any passed addresses 832 */ 833 if (addr_len) 834 *addr_len=sizeof(*sin); 835 836 if (flags & MSG_ERRQUEUE) 837 return ip_recv_error(sk, msg, len); 838 839 try_again: 840 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0), 841 &peeked, &err); 842 if (!skb) 843 goto out; 844 845 ulen = skb->len - sizeof(struct udphdr); 846 copied = len; 847 if (copied > ulen) 848 copied = ulen; 849 else if (copied < ulen) 850 msg->msg_flags |= MSG_TRUNC; 851 852 /* 853 * If checksum is needed at all, try to do it while copying the 854 * data. If the data is truncated, or if we only want a partial 855 * coverage checksum (UDP-Lite), do it before the copy. 856 */ 857 858 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) { 859 if (udp_lib_checksum_complete(skb)) 860 goto csum_copy_err; 861 } 862 863 if (skb_csum_unnecessary(skb)) 864 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr), 865 msg->msg_iov, copied ); 866 else { 867 err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov); 868 869 if (err == -EINVAL) 870 goto csum_copy_err; 871 } 872 873 if (err) 874 goto out_free; 875 876 if (!peeked) 877 UDP_INC_STATS_USER(sock_net(sk), 878 UDP_MIB_INDATAGRAMS, is_udplite); 879 880 sock_recv_timestamp(msg, sk, skb); 881 882 /* Copy the address. */ 883 if (sin) 884 { 885 sin->sin_family = AF_INET; 886 sin->sin_port = udp_hdr(skb)->source; 887 sin->sin_addr.s_addr = ip_hdr(skb)->saddr; 888 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 889 } 890 if (inet->cmsg_flags) 891 ip_cmsg_recv(msg, skb); 892 893 err = copied; 894 if (flags & MSG_TRUNC) 895 err = ulen; 896 897 out_free: 898 lock_sock(sk); 899 skb_free_datagram(sk, skb); 900 release_sock(sk); 901 out: 902 return err; 903 904 csum_copy_err: 905 lock_sock(sk); 906 if (!skb_kill_datagram(sk, skb, flags)) 907 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 908 release_sock(sk); 909 910 if (noblock) 911 return -EAGAIN; 912 goto try_again; 913 } 914 915 916 int udp_disconnect(struct sock *sk, int flags) 917 { 918 struct inet_sock *inet = inet_sk(sk); 919 /* 920 * 1003.1g - break association. 921 */ 922 923 sk->sk_state = TCP_CLOSE; 924 inet->daddr = 0; 925 inet->dport = 0; 926 sk->sk_bound_dev_if = 0; 927 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) 928 inet_reset_saddr(sk); 929 930 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { 931 sk->sk_prot->unhash(sk); 932 inet->sport = 0; 933 } 934 sk_dst_reset(sk); 935 return 0; 936 } 937 938 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 939 { 940 int is_udplite = IS_UDPLITE(sk); 941 int rc; 942 943 if ((rc = sock_queue_rcv_skb(sk, skb)) < 0) { 944 /* Note that an ENOMEM error is charged twice */ 945 if (rc == -ENOMEM) 946 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, 947 is_udplite); 948 goto drop; 949 } 950 951 return 0; 952 953 drop: 954 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 955 kfree_skb(skb); 956 return -1; 957 } 958 959 /* returns: 960 * -1: error 961 * 0: success 962 * >0: "udp encap" protocol resubmission 963 * 964 * Note that in the success and error cases, the skb is assumed to 965 * have either been requeued or freed. 966 */ 967 int udp_queue_rcv_skb(struct sock * sk, struct sk_buff *skb) 968 { 969 struct udp_sock *up = udp_sk(sk); 970 int rc; 971 int is_udplite = IS_UDPLITE(sk); 972 973 /* 974 * Charge it to the socket, dropping if the queue is full. 975 */ 976 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) 977 goto drop; 978 nf_reset(skb); 979 980 if (up->encap_type) { 981 /* 982 * This is an encapsulation socket so pass the skb to 983 * the socket's udp_encap_rcv() hook. Otherwise, just 984 * fall through and pass this up the UDP socket. 985 * up->encap_rcv() returns the following value: 986 * =0 if skb was successfully passed to the encap 987 * handler or was discarded by it. 988 * >0 if skb should be passed on to UDP. 989 * <0 if skb should be resubmitted as proto -N 990 */ 991 992 /* if we're overly short, let UDP handle it */ 993 if (skb->len > sizeof(struct udphdr) && 994 up->encap_rcv != NULL) { 995 int ret; 996 997 ret = (*up->encap_rcv)(sk, skb); 998 if (ret <= 0) { 999 UDP_INC_STATS_BH(sock_net(sk), 1000 UDP_MIB_INDATAGRAMS, 1001 is_udplite); 1002 return -ret; 1003 } 1004 } 1005 1006 /* FALLTHROUGH -- it's a UDP Packet */ 1007 } 1008 1009 /* 1010 * UDP-Lite specific tests, ignored on UDP sockets 1011 */ 1012 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 1013 1014 /* 1015 * MIB statistics other than incrementing the error count are 1016 * disabled for the following two types of errors: these depend 1017 * on the application settings, not on the functioning of the 1018 * protocol stack as such. 1019 * 1020 * RFC 3828 here recommends (sec 3.3): "There should also be a 1021 * way ... to ... at least let the receiving application block 1022 * delivery of packets with coverage values less than a value 1023 * provided by the application." 1024 */ 1025 if (up->pcrlen == 0) { /* full coverage was set */ 1026 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage " 1027 "%d while full coverage %d requested\n", 1028 UDP_SKB_CB(skb)->cscov, skb->len); 1029 goto drop; 1030 } 1031 /* The next case involves violating the min. coverage requested 1032 * by the receiver. This is subtle: if receiver wants x and x is 1033 * greater than the buffersize/MTU then receiver will complain 1034 * that it wants x while sender emits packets of smaller size y. 1035 * Therefore the above ...()->partial_cov statement is essential. 1036 */ 1037 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 1038 LIMIT_NETDEBUG(KERN_WARNING 1039 "UDPLITE: coverage %d too small, need min %d\n", 1040 UDP_SKB_CB(skb)->cscov, up->pcrlen); 1041 goto drop; 1042 } 1043 } 1044 1045 if (sk->sk_filter) { 1046 if (udp_lib_checksum_complete(skb)) 1047 goto drop; 1048 } 1049 1050 rc = 0; 1051 1052 bh_lock_sock(sk); 1053 if (!sock_owned_by_user(sk)) 1054 rc = __udp_queue_rcv_skb(sk, skb); 1055 else 1056 sk_add_backlog(sk, skb); 1057 bh_unlock_sock(sk); 1058 1059 return rc; 1060 1061 drop: 1062 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1063 kfree_skb(skb); 1064 return -1; 1065 } 1066 1067 /* 1068 * Multicasts and broadcasts go to each listener. 1069 * 1070 * Note: called only from the BH handler context, 1071 * so we don't need to lock the hashes. 1072 */ 1073 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 1074 struct udphdr *uh, 1075 __be32 saddr, __be32 daddr, 1076 struct hlist_head udptable[]) 1077 { 1078 struct sock *sk; 1079 int dif; 1080 1081 read_lock(&udp_hash_lock); 1082 sk = sk_head(&udptable[udp_hashfn(net, ntohs(uh->dest))]); 1083 dif = skb->dev->ifindex; 1084 sk = udp_v4_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif); 1085 if (sk) { 1086 struct sock *sknext = NULL; 1087 1088 do { 1089 struct sk_buff *skb1 = skb; 1090 1091 sknext = udp_v4_mcast_next(net, sk_next(sk), uh->dest, 1092 daddr, uh->source, saddr, 1093 dif); 1094 if (sknext) 1095 skb1 = skb_clone(skb, GFP_ATOMIC); 1096 1097 if (skb1) { 1098 int ret = udp_queue_rcv_skb(sk, skb1); 1099 if (ret > 0) 1100 /* we should probably re-process instead 1101 * of dropping packets here. */ 1102 kfree_skb(skb1); 1103 } 1104 sk = sknext; 1105 } while (sknext); 1106 } else 1107 kfree_skb(skb); 1108 read_unlock(&udp_hash_lock); 1109 return 0; 1110 } 1111 1112 /* Initialize UDP checksum. If exited with zero value (success), 1113 * CHECKSUM_UNNECESSARY means, that no more checks are required. 1114 * Otherwise, csum completion requires chacksumming packet body, 1115 * including udp header and folding it to skb->csum. 1116 */ 1117 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, 1118 int proto) 1119 { 1120 const struct iphdr *iph; 1121 int err; 1122 1123 UDP_SKB_CB(skb)->partial_cov = 0; 1124 UDP_SKB_CB(skb)->cscov = skb->len; 1125 1126 if (proto == IPPROTO_UDPLITE) { 1127 err = udplite_checksum_init(skb, uh); 1128 if (err) 1129 return err; 1130 } 1131 1132 iph = ip_hdr(skb); 1133 if (uh->check == 0) { 1134 skb->ip_summed = CHECKSUM_UNNECESSARY; 1135 } else if (skb->ip_summed == CHECKSUM_COMPLETE) { 1136 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len, 1137 proto, skb->csum)) 1138 skb->ip_summed = CHECKSUM_UNNECESSARY; 1139 } 1140 if (!skb_csum_unnecessary(skb)) 1141 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr, 1142 skb->len, proto, 0); 1143 /* Probably, we should checksum udp header (it should be in cache 1144 * in any case) and data in tiny packets (< rx copybreak). 1145 */ 1146 1147 return 0; 1148 } 1149 1150 /* 1151 * All we need to do is get the socket, and then do a checksum. 1152 */ 1153 1154 int __udp4_lib_rcv(struct sk_buff *skb, struct hlist_head udptable[], 1155 int proto) 1156 { 1157 struct sock *sk; 1158 struct udphdr *uh = udp_hdr(skb); 1159 unsigned short ulen; 1160 struct rtable *rt = (struct rtable*)skb->dst; 1161 __be32 saddr = ip_hdr(skb)->saddr; 1162 __be32 daddr = ip_hdr(skb)->daddr; 1163 struct net *net = dev_net(skb->dev); 1164 1165 /* 1166 * Validate the packet. 1167 */ 1168 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 1169 goto drop; /* No space for header. */ 1170 1171 ulen = ntohs(uh->len); 1172 if (ulen > skb->len) 1173 goto short_packet; 1174 1175 if (proto == IPPROTO_UDP) { 1176 /* UDP validates ulen. */ 1177 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) 1178 goto short_packet; 1179 uh = udp_hdr(skb); 1180 } 1181 1182 if (udp4_csum_init(skb, uh, proto)) 1183 goto csum_error; 1184 1185 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) 1186 return __udp4_lib_mcast_deliver(net, skb, uh, 1187 saddr, daddr, udptable); 1188 1189 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 1190 1191 if (sk != NULL) { 1192 int ret = udp_queue_rcv_skb(sk, skb); 1193 sock_put(sk); 1194 1195 /* a return value > 0 means to resubmit the input, but 1196 * it wants the return to be -protocol, or 0 1197 */ 1198 if (ret > 0) 1199 return -ret; 1200 return 0; 1201 } 1202 1203 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 1204 goto drop; 1205 nf_reset(skb); 1206 1207 /* No socket. Drop packet silently, if checksum is wrong */ 1208 if (udp_lib_checksum_complete(skb)) 1209 goto csum_error; 1210 1211 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 1212 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); 1213 1214 /* 1215 * Hmm. We got an UDP packet to a port to which we 1216 * don't wanna listen. Ignore it. 1217 */ 1218 kfree_skb(skb); 1219 return 0; 1220 1221 short_packet: 1222 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From " NIPQUAD_FMT ":%u %d/%d to " NIPQUAD_FMT ":%u\n", 1223 proto == IPPROTO_UDPLITE ? "-Lite" : "", 1224 NIPQUAD(saddr), 1225 ntohs(uh->source), 1226 ulen, 1227 skb->len, 1228 NIPQUAD(daddr), 1229 ntohs(uh->dest)); 1230 goto drop; 1231 1232 csum_error: 1233 /* 1234 * RFC1122: OK. Discards the bad packet silently (as far as 1235 * the network is concerned, anyway) as per 4.1.3.4 (MUST). 1236 */ 1237 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From " NIPQUAD_FMT ":%u to " NIPQUAD_FMT ":%u ulen %d\n", 1238 proto == IPPROTO_UDPLITE ? "-Lite" : "", 1239 NIPQUAD(saddr), 1240 ntohs(uh->source), 1241 NIPQUAD(daddr), 1242 ntohs(uh->dest), 1243 ulen); 1244 drop: 1245 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 1246 kfree_skb(skb); 1247 return 0; 1248 } 1249 1250 int udp_rcv(struct sk_buff *skb) 1251 { 1252 return __udp4_lib_rcv(skb, udp_hash, IPPROTO_UDP); 1253 } 1254 1255 void udp_destroy_sock(struct sock *sk) 1256 { 1257 lock_sock(sk); 1258 udp_flush_pending_frames(sk); 1259 release_sock(sk); 1260 } 1261 1262 /* 1263 * Socket option code for UDP 1264 */ 1265 int udp_lib_setsockopt(struct sock *sk, int level, int optname, 1266 char __user *optval, int optlen, 1267 int (*push_pending_frames)(struct sock *)) 1268 { 1269 struct udp_sock *up = udp_sk(sk); 1270 int val; 1271 int err = 0; 1272 int is_udplite = IS_UDPLITE(sk); 1273 1274 if (optlen<sizeof(int)) 1275 return -EINVAL; 1276 1277 if (get_user(val, (int __user *)optval)) 1278 return -EFAULT; 1279 1280 switch (optname) { 1281 case UDP_CORK: 1282 if (val != 0) { 1283 up->corkflag = 1; 1284 } else { 1285 up->corkflag = 0; 1286 lock_sock(sk); 1287 (*push_pending_frames)(sk); 1288 release_sock(sk); 1289 } 1290 break; 1291 1292 case UDP_ENCAP: 1293 switch (val) { 1294 case 0: 1295 case UDP_ENCAP_ESPINUDP: 1296 case UDP_ENCAP_ESPINUDP_NON_IKE: 1297 up->encap_rcv = xfrm4_udp_encap_rcv; 1298 /* FALLTHROUGH */ 1299 case UDP_ENCAP_L2TPINUDP: 1300 up->encap_type = val; 1301 break; 1302 default: 1303 err = -ENOPROTOOPT; 1304 break; 1305 } 1306 break; 1307 1308 /* 1309 * UDP-Lite's partial checksum coverage (RFC 3828). 1310 */ 1311 /* The sender sets actual checksum coverage length via this option. 1312 * The case coverage > packet length is handled by send module. */ 1313 case UDPLITE_SEND_CSCOV: 1314 if (!is_udplite) /* Disable the option on UDP sockets */ 1315 return -ENOPROTOOPT; 1316 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ 1317 val = 8; 1318 else if (val > USHORT_MAX) 1319 val = USHORT_MAX; 1320 up->pcslen = val; 1321 up->pcflag |= UDPLITE_SEND_CC; 1322 break; 1323 1324 /* The receiver specifies a minimum checksum coverage value. To make 1325 * sense, this should be set to at least 8 (as done below). If zero is 1326 * used, this again means full checksum coverage. */ 1327 case UDPLITE_RECV_CSCOV: 1328 if (!is_udplite) /* Disable the option on UDP sockets */ 1329 return -ENOPROTOOPT; 1330 if (val != 0 && val < 8) /* Avoid silly minimal values. */ 1331 val = 8; 1332 else if (val > USHORT_MAX) 1333 val = USHORT_MAX; 1334 up->pcrlen = val; 1335 up->pcflag |= UDPLITE_RECV_CC; 1336 break; 1337 1338 default: 1339 err = -ENOPROTOOPT; 1340 break; 1341 } 1342 1343 return err; 1344 } 1345 1346 int udp_setsockopt(struct sock *sk, int level, int optname, 1347 char __user *optval, int optlen) 1348 { 1349 if (level == SOL_UDP || level == SOL_UDPLITE) 1350 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1351 udp_push_pending_frames); 1352 return ip_setsockopt(sk, level, optname, optval, optlen); 1353 } 1354 1355 #ifdef CONFIG_COMPAT 1356 int compat_udp_setsockopt(struct sock *sk, int level, int optname, 1357 char __user *optval, int optlen) 1358 { 1359 if (level == SOL_UDP || level == SOL_UDPLITE) 1360 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1361 udp_push_pending_frames); 1362 return compat_ip_setsockopt(sk, level, optname, optval, optlen); 1363 } 1364 #endif 1365 1366 int udp_lib_getsockopt(struct sock *sk, int level, int optname, 1367 char __user *optval, int __user *optlen) 1368 { 1369 struct udp_sock *up = udp_sk(sk); 1370 int val, len; 1371 1372 if (get_user(len,optlen)) 1373 return -EFAULT; 1374 1375 len = min_t(unsigned int, len, sizeof(int)); 1376 1377 if (len < 0) 1378 return -EINVAL; 1379 1380 switch (optname) { 1381 case UDP_CORK: 1382 val = up->corkflag; 1383 break; 1384 1385 case UDP_ENCAP: 1386 val = up->encap_type; 1387 break; 1388 1389 /* The following two cannot be changed on UDP sockets, the return is 1390 * always 0 (which corresponds to the full checksum coverage of UDP). */ 1391 case UDPLITE_SEND_CSCOV: 1392 val = up->pcslen; 1393 break; 1394 1395 case UDPLITE_RECV_CSCOV: 1396 val = up->pcrlen; 1397 break; 1398 1399 default: 1400 return -ENOPROTOOPT; 1401 } 1402 1403 if (put_user(len, optlen)) 1404 return -EFAULT; 1405 if (copy_to_user(optval, &val,len)) 1406 return -EFAULT; 1407 return 0; 1408 } 1409 1410 int udp_getsockopt(struct sock *sk, int level, int optname, 1411 char __user *optval, int __user *optlen) 1412 { 1413 if (level == SOL_UDP || level == SOL_UDPLITE) 1414 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1415 return ip_getsockopt(sk, level, optname, optval, optlen); 1416 } 1417 1418 #ifdef CONFIG_COMPAT 1419 int compat_udp_getsockopt(struct sock *sk, int level, int optname, 1420 char __user *optval, int __user *optlen) 1421 { 1422 if (level == SOL_UDP || level == SOL_UDPLITE) 1423 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1424 return compat_ip_getsockopt(sk, level, optname, optval, optlen); 1425 } 1426 #endif 1427 /** 1428 * udp_poll - wait for a UDP event. 1429 * @file - file struct 1430 * @sock - socket 1431 * @wait - poll table 1432 * 1433 * This is same as datagram poll, except for the special case of 1434 * blocking sockets. If application is using a blocking fd 1435 * and a packet with checksum error is in the queue; 1436 * then it could get return from select indicating data available 1437 * but then block when reading it. Add special case code 1438 * to work around these arguably broken applications. 1439 */ 1440 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait) 1441 { 1442 unsigned int mask = datagram_poll(file, sock, wait); 1443 struct sock *sk = sock->sk; 1444 int is_lite = IS_UDPLITE(sk); 1445 1446 /* Check for false positives due to checksum errors */ 1447 if ( (mask & POLLRDNORM) && 1448 !(file->f_flags & O_NONBLOCK) && 1449 !(sk->sk_shutdown & RCV_SHUTDOWN)){ 1450 struct sk_buff_head *rcvq = &sk->sk_receive_queue; 1451 struct sk_buff *skb; 1452 1453 spin_lock_bh(&rcvq->lock); 1454 while ((skb = skb_peek(rcvq)) != NULL && 1455 udp_lib_checksum_complete(skb)) { 1456 UDP_INC_STATS_BH(sock_net(sk), 1457 UDP_MIB_INERRORS, is_lite); 1458 __skb_unlink(skb, rcvq); 1459 kfree_skb(skb); 1460 } 1461 spin_unlock_bh(&rcvq->lock); 1462 1463 /* nothing to see, move along */ 1464 if (skb == NULL) 1465 mask &= ~(POLLIN | POLLRDNORM); 1466 } 1467 1468 return mask; 1469 1470 } 1471 1472 struct proto udp_prot = { 1473 .name = "UDP", 1474 .owner = THIS_MODULE, 1475 .close = udp_lib_close, 1476 .connect = ip4_datagram_connect, 1477 .disconnect = udp_disconnect, 1478 .ioctl = udp_ioctl, 1479 .destroy = udp_destroy_sock, 1480 .setsockopt = udp_setsockopt, 1481 .getsockopt = udp_getsockopt, 1482 .sendmsg = udp_sendmsg, 1483 .recvmsg = udp_recvmsg, 1484 .sendpage = udp_sendpage, 1485 .backlog_rcv = __udp_queue_rcv_skb, 1486 .hash = udp_lib_hash, 1487 .unhash = udp_lib_unhash, 1488 .get_port = udp_v4_get_port, 1489 .memory_allocated = &udp_memory_allocated, 1490 .sysctl_mem = sysctl_udp_mem, 1491 .sysctl_wmem = &sysctl_udp_wmem_min, 1492 .sysctl_rmem = &sysctl_udp_rmem_min, 1493 .obj_size = sizeof(struct udp_sock), 1494 .h.udp_hash = udp_hash, 1495 #ifdef CONFIG_COMPAT 1496 .compat_setsockopt = compat_udp_setsockopt, 1497 .compat_getsockopt = compat_udp_getsockopt, 1498 #endif 1499 }; 1500 1501 /* ------------------------------------------------------------------------ */ 1502 #ifdef CONFIG_PROC_FS 1503 1504 static struct sock *udp_get_first(struct seq_file *seq) 1505 { 1506 struct sock *sk; 1507 struct udp_iter_state *state = seq->private; 1508 struct net *net = seq_file_net(seq); 1509 1510 for (state->bucket = 0; state->bucket < UDP_HTABLE_SIZE; ++state->bucket) { 1511 struct hlist_node *node; 1512 sk_for_each(sk, node, state->hashtable + state->bucket) { 1513 if (!net_eq(sock_net(sk), net)) 1514 continue; 1515 if (sk->sk_family == state->family) 1516 goto found; 1517 } 1518 } 1519 sk = NULL; 1520 found: 1521 return sk; 1522 } 1523 1524 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) 1525 { 1526 struct udp_iter_state *state = seq->private; 1527 struct net *net = seq_file_net(seq); 1528 1529 do { 1530 sk = sk_next(sk); 1531 try_again: 1532 ; 1533 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family)); 1534 1535 if (!sk && ++state->bucket < UDP_HTABLE_SIZE) { 1536 sk = sk_head(state->hashtable + state->bucket); 1537 goto try_again; 1538 } 1539 return sk; 1540 } 1541 1542 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) 1543 { 1544 struct sock *sk = udp_get_first(seq); 1545 1546 if (sk) 1547 while (pos && (sk = udp_get_next(seq, sk)) != NULL) 1548 --pos; 1549 return pos ? NULL : sk; 1550 } 1551 1552 static void *udp_seq_start(struct seq_file *seq, loff_t *pos) 1553 __acquires(udp_hash_lock) 1554 { 1555 read_lock(&udp_hash_lock); 1556 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN; 1557 } 1558 1559 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 1560 { 1561 struct sock *sk; 1562 1563 if (v == SEQ_START_TOKEN) 1564 sk = udp_get_idx(seq, 0); 1565 else 1566 sk = udp_get_next(seq, v); 1567 1568 ++*pos; 1569 return sk; 1570 } 1571 1572 static void udp_seq_stop(struct seq_file *seq, void *v) 1573 __releases(udp_hash_lock) 1574 { 1575 read_unlock(&udp_hash_lock); 1576 } 1577 1578 static int udp_seq_open(struct inode *inode, struct file *file) 1579 { 1580 struct udp_seq_afinfo *afinfo = PDE(inode)->data; 1581 struct udp_iter_state *s; 1582 int err; 1583 1584 err = seq_open_net(inode, file, &afinfo->seq_ops, 1585 sizeof(struct udp_iter_state)); 1586 if (err < 0) 1587 return err; 1588 1589 s = ((struct seq_file *)file->private_data)->private; 1590 s->family = afinfo->family; 1591 s->hashtable = afinfo->hashtable; 1592 return err; 1593 } 1594 1595 /* ------------------------------------------------------------------------ */ 1596 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo) 1597 { 1598 struct proc_dir_entry *p; 1599 int rc = 0; 1600 1601 afinfo->seq_fops.open = udp_seq_open; 1602 afinfo->seq_fops.read = seq_read; 1603 afinfo->seq_fops.llseek = seq_lseek; 1604 afinfo->seq_fops.release = seq_release_net; 1605 1606 afinfo->seq_ops.start = udp_seq_start; 1607 afinfo->seq_ops.next = udp_seq_next; 1608 afinfo->seq_ops.stop = udp_seq_stop; 1609 1610 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net, 1611 &afinfo->seq_fops, afinfo); 1612 if (!p) 1613 rc = -ENOMEM; 1614 return rc; 1615 } 1616 1617 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo) 1618 { 1619 proc_net_remove(net, afinfo->name); 1620 } 1621 1622 /* ------------------------------------------------------------------------ */ 1623 static void udp4_format_sock(struct sock *sp, struct seq_file *f, 1624 int bucket, int *len) 1625 { 1626 struct inet_sock *inet = inet_sk(sp); 1627 __be32 dest = inet->daddr; 1628 __be32 src = inet->rcv_saddr; 1629 __u16 destp = ntohs(inet->dport); 1630 __u16 srcp = ntohs(inet->sport); 1631 1632 seq_printf(f, "%4d: %08X:%04X %08X:%04X" 1633 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %d%n", 1634 bucket, src, srcp, dest, destp, sp->sk_state, 1635 atomic_read(&sp->sk_wmem_alloc), 1636 atomic_read(&sp->sk_rmem_alloc), 1637 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp), 1638 atomic_read(&sp->sk_refcnt), sp, 1639 atomic_read(&sp->sk_drops), len); 1640 } 1641 1642 int udp4_seq_show(struct seq_file *seq, void *v) 1643 { 1644 if (v == SEQ_START_TOKEN) 1645 seq_printf(seq, "%-127s\n", 1646 " sl local_address rem_address st tx_queue " 1647 "rx_queue tr tm->when retrnsmt uid timeout " 1648 "inode ref pointer drops"); 1649 else { 1650 struct udp_iter_state *state = seq->private; 1651 int len; 1652 1653 udp4_format_sock(v, seq, state->bucket, &len); 1654 seq_printf(seq, "%*s\n", 127 - len ,""); 1655 } 1656 return 0; 1657 } 1658 1659 /* ------------------------------------------------------------------------ */ 1660 static struct udp_seq_afinfo udp4_seq_afinfo = { 1661 .name = "udp", 1662 .family = AF_INET, 1663 .hashtable = udp_hash, 1664 .seq_fops = { 1665 .owner = THIS_MODULE, 1666 }, 1667 .seq_ops = { 1668 .show = udp4_seq_show, 1669 }, 1670 }; 1671 1672 static int udp4_proc_init_net(struct net *net) 1673 { 1674 return udp_proc_register(net, &udp4_seq_afinfo); 1675 } 1676 1677 static void udp4_proc_exit_net(struct net *net) 1678 { 1679 udp_proc_unregister(net, &udp4_seq_afinfo); 1680 } 1681 1682 static struct pernet_operations udp4_net_ops = { 1683 .init = udp4_proc_init_net, 1684 .exit = udp4_proc_exit_net, 1685 }; 1686 1687 int __init udp4_proc_init(void) 1688 { 1689 return register_pernet_subsys(&udp4_net_ops); 1690 } 1691 1692 void udp4_proc_exit(void) 1693 { 1694 unregister_pernet_subsys(&udp4_net_ops); 1695 } 1696 #endif /* CONFIG_PROC_FS */ 1697 1698 void __init udp_init(void) 1699 { 1700 unsigned long limit; 1701 1702 /* Set the pressure threshold up by the same strategy of TCP. It is a 1703 * fraction of global memory that is up to 1/2 at 256 MB, decreasing 1704 * toward zero with the amount of memory, with a floor of 128 pages. 1705 */ 1706 limit = min(nr_all_pages, 1UL<<(28-PAGE_SHIFT)) >> (20-PAGE_SHIFT); 1707 limit = (limit * (nr_all_pages >> (20-PAGE_SHIFT))) >> (PAGE_SHIFT-11); 1708 limit = max(limit, 128UL); 1709 sysctl_udp_mem[0] = limit / 4 * 3; 1710 sysctl_udp_mem[1] = limit; 1711 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2; 1712 1713 sysctl_udp_rmem_min = SK_MEM_QUANTUM; 1714 sysctl_udp_wmem_min = SK_MEM_QUANTUM; 1715 } 1716 1717 EXPORT_SYMBOL(udp_disconnect); 1718 EXPORT_SYMBOL(udp_hash); 1719 EXPORT_SYMBOL(udp_hash_lock); 1720 EXPORT_SYMBOL(udp_ioctl); 1721 EXPORT_SYMBOL(udp_prot); 1722 EXPORT_SYMBOL(udp_sendmsg); 1723 EXPORT_SYMBOL(udp_lib_getsockopt); 1724 EXPORT_SYMBOL(udp_lib_setsockopt); 1725 EXPORT_SYMBOL(udp_poll); 1726 EXPORT_SYMBOL(udp_lib_get_port); 1727 1728 #ifdef CONFIG_PROC_FS 1729 EXPORT_SYMBOL(udp_proc_register); 1730 EXPORT_SYMBOL(udp_proc_unregister); 1731 #endif 1732