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 .uli_u = { .ports = 637 { .sport = inet->sport, 638 .dport = dport } } }; 639 struct net *net = sock_net(sk); 640 641 security_sk_classify_flow(sk, &fl); 642 err = ip_route_output_flow(net, &rt, &fl, sk, 1); 643 if (err) { 644 if (err == -ENETUNREACH) 645 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES); 646 goto out; 647 } 648 649 err = -EACCES; 650 if ((rt->rt_flags & RTCF_BROADCAST) && 651 !sock_flag(sk, SOCK_BROADCAST)) 652 goto out; 653 if (connected) 654 sk_dst_set(sk, dst_clone(&rt->u.dst)); 655 } 656 657 if (msg->msg_flags&MSG_CONFIRM) 658 goto do_confirm; 659 back_from_confirm: 660 661 saddr = rt->rt_src; 662 if (!ipc.addr) 663 daddr = ipc.addr = rt->rt_dst; 664 665 lock_sock(sk); 666 if (unlikely(up->pending)) { 667 /* The socket is already corked while preparing it. */ 668 /* ... which is an evident application bug. --ANK */ 669 release_sock(sk); 670 671 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n"); 672 err = -EINVAL; 673 goto out; 674 } 675 /* 676 * Now cork the socket to pend data. 677 */ 678 inet->cork.fl.fl4_dst = daddr; 679 inet->cork.fl.fl_ip_dport = dport; 680 inet->cork.fl.fl4_src = saddr; 681 inet->cork.fl.fl_ip_sport = inet->sport; 682 up->pending = AF_INET; 683 684 do_append_data: 685 up->len += ulen; 686 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 687 err = ip_append_data(sk, getfrag, msg->msg_iov, ulen, 688 sizeof(struct udphdr), &ipc, rt, 689 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 690 if (err) 691 udp_flush_pending_frames(sk); 692 else if (!corkreq) 693 err = udp_push_pending_frames(sk); 694 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 695 up->pending = 0; 696 release_sock(sk); 697 698 out: 699 ip_rt_put(rt); 700 if (free) 701 kfree(ipc.opt); 702 if (!err) 703 return len; 704 /* 705 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 706 * ENOBUFS might not be good (it's not tunable per se), but otherwise 707 * we don't have a good statistic (IpOutDiscards but it can be too many 708 * things). We could add another new stat but at least for now that 709 * seems like overkill. 710 */ 711 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 712 UDP_INC_STATS_USER(sock_net(sk), 713 UDP_MIB_SNDBUFERRORS, is_udplite); 714 } 715 return err; 716 717 do_confirm: 718 dst_confirm(&rt->u.dst); 719 if (!(msg->msg_flags&MSG_PROBE) || len) 720 goto back_from_confirm; 721 err = 0; 722 goto out; 723 } 724 725 int udp_sendpage(struct sock *sk, struct page *page, int offset, 726 size_t size, int flags) 727 { 728 struct udp_sock *up = udp_sk(sk); 729 int ret; 730 731 if (!up->pending) { 732 struct msghdr msg = { .msg_flags = flags|MSG_MORE }; 733 734 /* Call udp_sendmsg to specify destination address which 735 * sendpage interface can't pass. 736 * This will succeed only when the socket is connected. 737 */ 738 ret = udp_sendmsg(NULL, sk, &msg, 0); 739 if (ret < 0) 740 return ret; 741 } 742 743 lock_sock(sk); 744 745 if (unlikely(!up->pending)) { 746 release_sock(sk); 747 748 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n"); 749 return -EINVAL; 750 } 751 752 ret = ip_append_page(sk, page, offset, size, flags); 753 if (ret == -EOPNOTSUPP) { 754 release_sock(sk); 755 return sock_no_sendpage(sk->sk_socket, page, offset, 756 size, flags); 757 } 758 if (ret < 0) { 759 udp_flush_pending_frames(sk); 760 goto out; 761 } 762 763 up->len += size; 764 if (!(up->corkflag || (flags&MSG_MORE))) 765 ret = udp_push_pending_frames(sk); 766 if (!ret) 767 ret = size; 768 out: 769 release_sock(sk); 770 return ret; 771 } 772 773 /* 774 * IOCTL requests applicable to the UDP protocol 775 */ 776 777 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg) 778 { 779 switch (cmd) { 780 case SIOCOUTQ: 781 { 782 int amount = atomic_read(&sk->sk_wmem_alloc); 783 return put_user(amount, (int __user *)arg); 784 } 785 786 case SIOCINQ: 787 { 788 struct sk_buff *skb; 789 unsigned long amount; 790 791 amount = 0; 792 spin_lock_bh(&sk->sk_receive_queue.lock); 793 skb = skb_peek(&sk->sk_receive_queue); 794 if (skb != NULL) { 795 /* 796 * We will only return the amount 797 * of this packet since that is all 798 * that will be read. 799 */ 800 amount = skb->len - sizeof(struct udphdr); 801 } 802 spin_unlock_bh(&sk->sk_receive_queue.lock); 803 return put_user(amount, (int __user *)arg); 804 } 805 806 default: 807 return -ENOIOCTLCMD; 808 } 809 810 return 0; 811 } 812 813 /* 814 * This should be easy, if there is something there we 815 * return it, otherwise we block. 816 */ 817 818 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 819 size_t len, int noblock, int flags, int *addr_len) 820 { 821 struct inet_sock *inet = inet_sk(sk); 822 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name; 823 struct sk_buff *skb; 824 unsigned int ulen, copied; 825 int peeked; 826 int err; 827 int is_udplite = IS_UDPLITE(sk); 828 829 /* 830 * Check any passed addresses 831 */ 832 if (addr_len) 833 *addr_len=sizeof(*sin); 834 835 if (flags & MSG_ERRQUEUE) 836 return ip_recv_error(sk, msg, len); 837 838 try_again: 839 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0), 840 &peeked, &err); 841 if (!skb) 842 goto out; 843 844 ulen = skb->len - sizeof(struct udphdr); 845 copied = len; 846 if (copied > ulen) 847 copied = ulen; 848 else if (copied < ulen) 849 msg->msg_flags |= MSG_TRUNC; 850 851 /* 852 * If checksum is needed at all, try to do it while copying the 853 * data. If the data is truncated, or if we only want a partial 854 * coverage checksum (UDP-Lite), do it before the copy. 855 */ 856 857 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) { 858 if (udp_lib_checksum_complete(skb)) 859 goto csum_copy_err; 860 } 861 862 if (skb_csum_unnecessary(skb)) 863 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr), 864 msg->msg_iov, copied ); 865 else { 866 err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov); 867 868 if (err == -EINVAL) 869 goto csum_copy_err; 870 } 871 872 if (err) 873 goto out_free; 874 875 if (!peeked) 876 UDP_INC_STATS_USER(sock_net(sk), 877 UDP_MIB_INDATAGRAMS, is_udplite); 878 879 sock_recv_timestamp(msg, sk, skb); 880 881 /* Copy the address. */ 882 if (sin) 883 { 884 sin->sin_family = AF_INET; 885 sin->sin_port = udp_hdr(skb)->source; 886 sin->sin_addr.s_addr = ip_hdr(skb)->saddr; 887 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 888 } 889 if (inet->cmsg_flags) 890 ip_cmsg_recv(msg, skb); 891 892 err = copied; 893 if (flags & MSG_TRUNC) 894 err = ulen; 895 896 out_free: 897 lock_sock(sk); 898 skb_free_datagram(sk, skb); 899 release_sock(sk); 900 out: 901 return err; 902 903 csum_copy_err: 904 lock_sock(sk); 905 if (!skb_kill_datagram(sk, skb, flags)) 906 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 907 release_sock(sk); 908 909 if (noblock) 910 return -EAGAIN; 911 goto try_again; 912 } 913 914 915 int udp_disconnect(struct sock *sk, int flags) 916 { 917 struct inet_sock *inet = inet_sk(sk); 918 /* 919 * 1003.1g - break association. 920 */ 921 922 sk->sk_state = TCP_CLOSE; 923 inet->daddr = 0; 924 inet->dport = 0; 925 sk->sk_bound_dev_if = 0; 926 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) 927 inet_reset_saddr(sk); 928 929 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { 930 sk->sk_prot->unhash(sk); 931 inet->sport = 0; 932 } 933 sk_dst_reset(sk); 934 return 0; 935 } 936 937 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 938 { 939 int is_udplite = IS_UDPLITE(sk); 940 int rc; 941 942 if ((rc = sock_queue_rcv_skb(sk, skb)) < 0) { 943 /* Note that an ENOMEM error is charged twice */ 944 if (rc == -ENOMEM) 945 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, 946 is_udplite); 947 goto drop; 948 } 949 950 return 0; 951 952 drop: 953 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 954 kfree_skb(skb); 955 return -1; 956 } 957 958 /* returns: 959 * -1: error 960 * 0: success 961 * >0: "udp encap" protocol resubmission 962 * 963 * Note that in the success and error cases, the skb is assumed to 964 * have either been requeued or freed. 965 */ 966 int udp_queue_rcv_skb(struct sock * sk, struct sk_buff *skb) 967 { 968 struct udp_sock *up = udp_sk(sk); 969 int rc; 970 int is_udplite = IS_UDPLITE(sk); 971 972 /* 973 * Charge it to the socket, dropping if the queue is full. 974 */ 975 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) 976 goto drop; 977 nf_reset(skb); 978 979 if (up->encap_type) { 980 /* 981 * This is an encapsulation socket so pass the skb to 982 * the socket's udp_encap_rcv() hook. Otherwise, just 983 * fall through and pass this up the UDP socket. 984 * up->encap_rcv() returns the following value: 985 * =0 if skb was successfully passed to the encap 986 * handler or was discarded by it. 987 * >0 if skb should be passed on to UDP. 988 * <0 if skb should be resubmitted as proto -N 989 */ 990 991 /* if we're overly short, let UDP handle it */ 992 if (skb->len > sizeof(struct udphdr) && 993 up->encap_rcv != NULL) { 994 int ret; 995 996 ret = (*up->encap_rcv)(sk, skb); 997 if (ret <= 0) { 998 UDP_INC_STATS_BH(sock_net(sk), 999 UDP_MIB_INDATAGRAMS, 1000 is_udplite); 1001 return -ret; 1002 } 1003 } 1004 1005 /* FALLTHROUGH -- it's a UDP Packet */ 1006 } 1007 1008 /* 1009 * UDP-Lite specific tests, ignored on UDP sockets 1010 */ 1011 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 1012 1013 /* 1014 * MIB statistics other than incrementing the error count are 1015 * disabled for the following two types of errors: these depend 1016 * on the application settings, not on the functioning of the 1017 * protocol stack as such. 1018 * 1019 * RFC 3828 here recommends (sec 3.3): "There should also be a 1020 * way ... to ... at least let the receiving application block 1021 * delivery of packets with coverage values less than a value 1022 * provided by the application." 1023 */ 1024 if (up->pcrlen == 0) { /* full coverage was set */ 1025 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage " 1026 "%d while full coverage %d requested\n", 1027 UDP_SKB_CB(skb)->cscov, skb->len); 1028 goto drop; 1029 } 1030 /* The next case involves violating the min. coverage requested 1031 * by the receiver. This is subtle: if receiver wants x and x is 1032 * greater than the buffersize/MTU then receiver will complain 1033 * that it wants x while sender emits packets of smaller size y. 1034 * Therefore the above ...()->partial_cov statement is essential. 1035 */ 1036 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 1037 LIMIT_NETDEBUG(KERN_WARNING 1038 "UDPLITE: coverage %d too small, need min %d\n", 1039 UDP_SKB_CB(skb)->cscov, up->pcrlen); 1040 goto drop; 1041 } 1042 } 1043 1044 if (sk->sk_filter) { 1045 if (udp_lib_checksum_complete(skb)) 1046 goto drop; 1047 } 1048 1049 rc = 0; 1050 1051 bh_lock_sock(sk); 1052 if (!sock_owned_by_user(sk)) 1053 rc = __udp_queue_rcv_skb(sk, skb); 1054 else 1055 sk_add_backlog(sk, skb); 1056 bh_unlock_sock(sk); 1057 1058 return rc; 1059 1060 drop: 1061 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1062 kfree_skb(skb); 1063 return -1; 1064 } 1065 1066 /* 1067 * Multicasts and broadcasts go to each listener. 1068 * 1069 * Note: called only from the BH handler context, 1070 * so we don't need to lock the hashes. 1071 */ 1072 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 1073 struct udphdr *uh, 1074 __be32 saddr, __be32 daddr, 1075 struct hlist_head udptable[]) 1076 { 1077 struct sock *sk; 1078 int dif; 1079 1080 read_lock(&udp_hash_lock); 1081 sk = sk_head(&udptable[udp_hashfn(net, ntohs(uh->dest))]); 1082 dif = skb->dev->ifindex; 1083 sk = udp_v4_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif); 1084 if (sk) { 1085 struct sock *sknext = NULL; 1086 1087 do { 1088 struct sk_buff *skb1 = skb; 1089 1090 sknext = udp_v4_mcast_next(net, sk_next(sk), uh->dest, 1091 daddr, uh->source, saddr, 1092 dif); 1093 if (sknext) 1094 skb1 = skb_clone(skb, GFP_ATOMIC); 1095 1096 if (skb1) { 1097 int ret = udp_queue_rcv_skb(sk, skb1); 1098 if (ret > 0) 1099 /* we should probably re-process instead 1100 * of dropping packets here. */ 1101 kfree_skb(skb1); 1102 } 1103 sk = sknext; 1104 } while (sknext); 1105 } else 1106 kfree_skb(skb); 1107 read_unlock(&udp_hash_lock); 1108 return 0; 1109 } 1110 1111 /* Initialize UDP checksum. If exited with zero value (success), 1112 * CHECKSUM_UNNECESSARY means, that no more checks are required. 1113 * Otherwise, csum completion requires chacksumming packet body, 1114 * including udp header and folding it to skb->csum. 1115 */ 1116 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, 1117 int proto) 1118 { 1119 const struct iphdr *iph; 1120 int err; 1121 1122 UDP_SKB_CB(skb)->partial_cov = 0; 1123 UDP_SKB_CB(skb)->cscov = skb->len; 1124 1125 if (proto == IPPROTO_UDPLITE) { 1126 err = udplite_checksum_init(skb, uh); 1127 if (err) 1128 return err; 1129 } 1130 1131 iph = ip_hdr(skb); 1132 if (uh->check == 0) { 1133 skb->ip_summed = CHECKSUM_UNNECESSARY; 1134 } else if (skb->ip_summed == CHECKSUM_COMPLETE) { 1135 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len, 1136 proto, skb->csum)) 1137 skb->ip_summed = CHECKSUM_UNNECESSARY; 1138 } 1139 if (!skb_csum_unnecessary(skb)) 1140 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr, 1141 skb->len, proto, 0); 1142 /* Probably, we should checksum udp header (it should be in cache 1143 * in any case) and data in tiny packets (< rx copybreak). 1144 */ 1145 1146 return 0; 1147 } 1148 1149 /* 1150 * All we need to do is get the socket, and then do a checksum. 1151 */ 1152 1153 int __udp4_lib_rcv(struct sk_buff *skb, struct hlist_head udptable[], 1154 int proto) 1155 { 1156 struct sock *sk; 1157 struct udphdr *uh = udp_hdr(skb); 1158 unsigned short ulen; 1159 struct rtable *rt = (struct rtable*)skb->dst; 1160 __be32 saddr = ip_hdr(skb)->saddr; 1161 __be32 daddr = ip_hdr(skb)->daddr; 1162 struct net *net = dev_net(skb->dev); 1163 1164 /* 1165 * Validate the packet. 1166 */ 1167 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 1168 goto drop; /* No space for header. */ 1169 1170 ulen = ntohs(uh->len); 1171 if (ulen > skb->len) 1172 goto short_packet; 1173 1174 if (proto == IPPROTO_UDP) { 1175 /* UDP validates ulen. */ 1176 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) 1177 goto short_packet; 1178 uh = udp_hdr(skb); 1179 } 1180 1181 if (udp4_csum_init(skb, uh, proto)) 1182 goto csum_error; 1183 1184 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) 1185 return __udp4_lib_mcast_deliver(net, skb, uh, 1186 saddr, daddr, udptable); 1187 1188 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 1189 1190 if (sk != NULL) { 1191 int ret = udp_queue_rcv_skb(sk, skb); 1192 sock_put(sk); 1193 1194 /* a return value > 0 means to resubmit the input, but 1195 * it wants the return to be -protocol, or 0 1196 */ 1197 if (ret > 0) 1198 return -ret; 1199 return 0; 1200 } 1201 1202 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 1203 goto drop; 1204 nf_reset(skb); 1205 1206 /* No socket. Drop packet silently, if checksum is wrong */ 1207 if (udp_lib_checksum_complete(skb)) 1208 goto csum_error; 1209 1210 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 1211 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); 1212 1213 /* 1214 * Hmm. We got an UDP packet to a port to which we 1215 * don't wanna listen. Ignore it. 1216 */ 1217 kfree_skb(skb); 1218 return 0; 1219 1220 short_packet: 1221 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From " NIPQUAD_FMT ":%u %d/%d to " NIPQUAD_FMT ":%u\n", 1222 proto == IPPROTO_UDPLITE ? "-Lite" : "", 1223 NIPQUAD(saddr), 1224 ntohs(uh->source), 1225 ulen, 1226 skb->len, 1227 NIPQUAD(daddr), 1228 ntohs(uh->dest)); 1229 goto drop; 1230 1231 csum_error: 1232 /* 1233 * RFC1122: OK. Discards the bad packet silently (as far as 1234 * the network is concerned, anyway) as per 4.1.3.4 (MUST). 1235 */ 1236 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From " NIPQUAD_FMT ":%u to " NIPQUAD_FMT ":%u ulen %d\n", 1237 proto == IPPROTO_UDPLITE ? "-Lite" : "", 1238 NIPQUAD(saddr), 1239 ntohs(uh->source), 1240 NIPQUAD(daddr), 1241 ntohs(uh->dest), 1242 ulen); 1243 drop: 1244 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 1245 kfree_skb(skb); 1246 return 0; 1247 } 1248 1249 int udp_rcv(struct sk_buff *skb) 1250 { 1251 return __udp4_lib_rcv(skb, udp_hash, IPPROTO_UDP); 1252 } 1253 1254 void udp_destroy_sock(struct sock *sk) 1255 { 1256 lock_sock(sk); 1257 udp_flush_pending_frames(sk); 1258 release_sock(sk); 1259 } 1260 1261 /* 1262 * Socket option code for UDP 1263 */ 1264 int udp_lib_setsockopt(struct sock *sk, int level, int optname, 1265 char __user *optval, int optlen, 1266 int (*push_pending_frames)(struct sock *)) 1267 { 1268 struct udp_sock *up = udp_sk(sk); 1269 int val; 1270 int err = 0; 1271 int is_udplite = IS_UDPLITE(sk); 1272 1273 if (optlen<sizeof(int)) 1274 return -EINVAL; 1275 1276 if (get_user(val, (int __user *)optval)) 1277 return -EFAULT; 1278 1279 switch (optname) { 1280 case UDP_CORK: 1281 if (val != 0) { 1282 up->corkflag = 1; 1283 } else { 1284 up->corkflag = 0; 1285 lock_sock(sk); 1286 (*push_pending_frames)(sk); 1287 release_sock(sk); 1288 } 1289 break; 1290 1291 case UDP_ENCAP: 1292 switch (val) { 1293 case 0: 1294 case UDP_ENCAP_ESPINUDP: 1295 case UDP_ENCAP_ESPINUDP_NON_IKE: 1296 up->encap_rcv = xfrm4_udp_encap_rcv; 1297 /* FALLTHROUGH */ 1298 case UDP_ENCAP_L2TPINUDP: 1299 up->encap_type = val; 1300 break; 1301 default: 1302 err = -ENOPROTOOPT; 1303 break; 1304 } 1305 break; 1306 1307 /* 1308 * UDP-Lite's partial checksum coverage (RFC 3828). 1309 */ 1310 /* The sender sets actual checksum coverage length via this option. 1311 * The case coverage > packet length is handled by send module. */ 1312 case UDPLITE_SEND_CSCOV: 1313 if (!is_udplite) /* Disable the option on UDP sockets */ 1314 return -ENOPROTOOPT; 1315 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ 1316 val = 8; 1317 else if (val > USHORT_MAX) 1318 val = USHORT_MAX; 1319 up->pcslen = val; 1320 up->pcflag |= UDPLITE_SEND_CC; 1321 break; 1322 1323 /* The receiver specifies a minimum checksum coverage value. To make 1324 * sense, this should be set to at least 8 (as done below). If zero is 1325 * used, this again means full checksum coverage. */ 1326 case UDPLITE_RECV_CSCOV: 1327 if (!is_udplite) /* Disable the option on UDP sockets */ 1328 return -ENOPROTOOPT; 1329 if (val != 0 && val < 8) /* Avoid silly minimal values. */ 1330 val = 8; 1331 else if (val > USHORT_MAX) 1332 val = USHORT_MAX; 1333 up->pcrlen = val; 1334 up->pcflag |= UDPLITE_RECV_CC; 1335 break; 1336 1337 default: 1338 err = -ENOPROTOOPT; 1339 break; 1340 } 1341 1342 return err; 1343 } 1344 1345 int udp_setsockopt(struct sock *sk, int level, int optname, 1346 char __user *optval, int optlen) 1347 { 1348 if (level == SOL_UDP || level == SOL_UDPLITE) 1349 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1350 udp_push_pending_frames); 1351 return ip_setsockopt(sk, level, optname, optval, optlen); 1352 } 1353 1354 #ifdef CONFIG_COMPAT 1355 int compat_udp_setsockopt(struct sock *sk, int level, int optname, 1356 char __user *optval, int optlen) 1357 { 1358 if (level == SOL_UDP || level == SOL_UDPLITE) 1359 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1360 udp_push_pending_frames); 1361 return compat_ip_setsockopt(sk, level, optname, optval, optlen); 1362 } 1363 #endif 1364 1365 int udp_lib_getsockopt(struct sock *sk, int level, int optname, 1366 char __user *optval, int __user *optlen) 1367 { 1368 struct udp_sock *up = udp_sk(sk); 1369 int val, len; 1370 1371 if (get_user(len,optlen)) 1372 return -EFAULT; 1373 1374 len = min_t(unsigned int, len, sizeof(int)); 1375 1376 if (len < 0) 1377 return -EINVAL; 1378 1379 switch (optname) { 1380 case UDP_CORK: 1381 val = up->corkflag; 1382 break; 1383 1384 case UDP_ENCAP: 1385 val = up->encap_type; 1386 break; 1387 1388 /* The following two cannot be changed on UDP sockets, the return is 1389 * always 0 (which corresponds to the full checksum coverage of UDP). */ 1390 case UDPLITE_SEND_CSCOV: 1391 val = up->pcslen; 1392 break; 1393 1394 case UDPLITE_RECV_CSCOV: 1395 val = up->pcrlen; 1396 break; 1397 1398 default: 1399 return -ENOPROTOOPT; 1400 } 1401 1402 if (put_user(len, optlen)) 1403 return -EFAULT; 1404 if (copy_to_user(optval, &val,len)) 1405 return -EFAULT; 1406 return 0; 1407 } 1408 1409 int udp_getsockopt(struct sock *sk, int level, int optname, 1410 char __user *optval, int __user *optlen) 1411 { 1412 if (level == SOL_UDP || level == SOL_UDPLITE) 1413 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1414 return ip_getsockopt(sk, level, optname, optval, optlen); 1415 } 1416 1417 #ifdef CONFIG_COMPAT 1418 int compat_udp_getsockopt(struct sock *sk, int level, int optname, 1419 char __user *optval, int __user *optlen) 1420 { 1421 if (level == SOL_UDP || level == SOL_UDPLITE) 1422 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1423 return compat_ip_getsockopt(sk, level, optname, optval, optlen); 1424 } 1425 #endif 1426 /** 1427 * udp_poll - wait for a UDP event. 1428 * @file - file struct 1429 * @sock - socket 1430 * @wait - poll table 1431 * 1432 * This is same as datagram poll, except for the special case of 1433 * blocking sockets. If application is using a blocking fd 1434 * and a packet with checksum error is in the queue; 1435 * then it could get return from select indicating data available 1436 * but then block when reading it. Add special case code 1437 * to work around these arguably broken applications. 1438 */ 1439 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait) 1440 { 1441 unsigned int mask = datagram_poll(file, sock, wait); 1442 struct sock *sk = sock->sk; 1443 int is_lite = IS_UDPLITE(sk); 1444 1445 /* Check for false positives due to checksum errors */ 1446 if ( (mask & POLLRDNORM) && 1447 !(file->f_flags & O_NONBLOCK) && 1448 !(sk->sk_shutdown & RCV_SHUTDOWN)){ 1449 struct sk_buff_head *rcvq = &sk->sk_receive_queue; 1450 struct sk_buff *skb; 1451 1452 spin_lock_bh(&rcvq->lock); 1453 while ((skb = skb_peek(rcvq)) != NULL && 1454 udp_lib_checksum_complete(skb)) { 1455 UDP_INC_STATS_BH(sock_net(sk), 1456 UDP_MIB_INERRORS, is_lite); 1457 __skb_unlink(skb, rcvq); 1458 kfree_skb(skb); 1459 } 1460 spin_unlock_bh(&rcvq->lock); 1461 1462 /* nothing to see, move along */ 1463 if (skb == NULL) 1464 mask &= ~(POLLIN | POLLRDNORM); 1465 } 1466 1467 return mask; 1468 1469 } 1470 1471 struct proto udp_prot = { 1472 .name = "UDP", 1473 .owner = THIS_MODULE, 1474 .close = udp_lib_close, 1475 .connect = ip4_datagram_connect, 1476 .disconnect = udp_disconnect, 1477 .ioctl = udp_ioctl, 1478 .destroy = udp_destroy_sock, 1479 .setsockopt = udp_setsockopt, 1480 .getsockopt = udp_getsockopt, 1481 .sendmsg = udp_sendmsg, 1482 .recvmsg = udp_recvmsg, 1483 .sendpage = udp_sendpage, 1484 .backlog_rcv = __udp_queue_rcv_skb, 1485 .hash = udp_lib_hash, 1486 .unhash = udp_lib_unhash, 1487 .get_port = udp_v4_get_port, 1488 .memory_allocated = &udp_memory_allocated, 1489 .sysctl_mem = sysctl_udp_mem, 1490 .sysctl_wmem = &sysctl_udp_wmem_min, 1491 .sysctl_rmem = &sysctl_udp_rmem_min, 1492 .obj_size = sizeof(struct udp_sock), 1493 .h.udp_hash = udp_hash, 1494 #ifdef CONFIG_COMPAT 1495 .compat_setsockopt = compat_udp_setsockopt, 1496 .compat_getsockopt = compat_udp_getsockopt, 1497 #endif 1498 }; 1499 1500 /* ------------------------------------------------------------------------ */ 1501 #ifdef CONFIG_PROC_FS 1502 1503 static struct sock *udp_get_first(struct seq_file *seq) 1504 { 1505 struct sock *sk; 1506 struct udp_iter_state *state = seq->private; 1507 struct net *net = seq_file_net(seq); 1508 1509 for (state->bucket = 0; state->bucket < UDP_HTABLE_SIZE; ++state->bucket) { 1510 struct hlist_node *node; 1511 sk_for_each(sk, node, state->hashtable + state->bucket) { 1512 if (!net_eq(sock_net(sk), net)) 1513 continue; 1514 if (sk->sk_family == state->family) 1515 goto found; 1516 } 1517 } 1518 sk = NULL; 1519 found: 1520 return sk; 1521 } 1522 1523 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) 1524 { 1525 struct udp_iter_state *state = seq->private; 1526 struct net *net = seq_file_net(seq); 1527 1528 do { 1529 sk = sk_next(sk); 1530 try_again: 1531 ; 1532 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family)); 1533 1534 if (!sk && ++state->bucket < UDP_HTABLE_SIZE) { 1535 sk = sk_head(state->hashtable + state->bucket); 1536 goto try_again; 1537 } 1538 return sk; 1539 } 1540 1541 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) 1542 { 1543 struct sock *sk = udp_get_first(seq); 1544 1545 if (sk) 1546 while (pos && (sk = udp_get_next(seq, sk)) != NULL) 1547 --pos; 1548 return pos ? NULL : sk; 1549 } 1550 1551 static void *udp_seq_start(struct seq_file *seq, loff_t *pos) 1552 __acquires(udp_hash_lock) 1553 { 1554 read_lock(&udp_hash_lock); 1555 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN; 1556 } 1557 1558 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 1559 { 1560 struct sock *sk; 1561 1562 if (v == SEQ_START_TOKEN) 1563 sk = udp_get_idx(seq, 0); 1564 else 1565 sk = udp_get_next(seq, v); 1566 1567 ++*pos; 1568 return sk; 1569 } 1570 1571 static void udp_seq_stop(struct seq_file *seq, void *v) 1572 __releases(udp_hash_lock) 1573 { 1574 read_unlock(&udp_hash_lock); 1575 } 1576 1577 static int udp_seq_open(struct inode *inode, struct file *file) 1578 { 1579 struct udp_seq_afinfo *afinfo = PDE(inode)->data; 1580 struct udp_iter_state *s; 1581 int err; 1582 1583 err = seq_open_net(inode, file, &afinfo->seq_ops, 1584 sizeof(struct udp_iter_state)); 1585 if (err < 0) 1586 return err; 1587 1588 s = ((struct seq_file *)file->private_data)->private; 1589 s->family = afinfo->family; 1590 s->hashtable = afinfo->hashtable; 1591 return err; 1592 } 1593 1594 /* ------------------------------------------------------------------------ */ 1595 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo) 1596 { 1597 struct proc_dir_entry *p; 1598 int rc = 0; 1599 1600 afinfo->seq_fops.open = udp_seq_open; 1601 afinfo->seq_fops.read = seq_read; 1602 afinfo->seq_fops.llseek = seq_lseek; 1603 afinfo->seq_fops.release = seq_release_net; 1604 1605 afinfo->seq_ops.start = udp_seq_start; 1606 afinfo->seq_ops.next = udp_seq_next; 1607 afinfo->seq_ops.stop = udp_seq_stop; 1608 1609 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net, 1610 &afinfo->seq_fops, afinfo); 1611 if (!p) 1612 rc = -ENOMEM; 1613 return rc; 1614 } 1615 1616 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo) 1617 { 1618 proc_net_remove(net, afinfo->name); 1619 } 1620 1621 /* ------------------------------------------------------------------------ */ 1622 static void udp4_format_sock(struct sock *sp, struct seq_file *f, 1623 int bucket, int *len) 1624 { 1625 struct inet_sock *inet = inet_sk(sp); 1626 __be32 dest = inet->daddr; 1627 __be32 src = inet->rcv_saddr; 1628 __u16 destp = ntohs(inet->dport); 1629 __u16 srcp = ntohs(inet->sport); 1630 1631 seq_printf(f, "%4d: %08X:%04X %08X:%04X" 1632 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %d%n", 1633 bucket, src, srcp, dest, destp, sp->sk_state, 1634 atomic_read(&sp->sk_wmem_alloc), 1635 atomic_read(&sp->sk_rmem_alloc), 1636 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp), 1637 atomic_read(&sp->sk_refcnt), sp, 1638 atomic_read(&sp->sk_drops), len); 1639 } 1640 1641 int udp4_seq_show(struct seq_file *seq, void *v) 1642 { 1643 if (v == SEQ_START_TOKEN) 1644 seq_printf(seq, "%-127s\n", 1645 " sl local_address rem_address st tx_queue " 1646 "rx_queue tr tm->when retrnsmt uid timeout " 1647 "inode ref pointer drops"); 1648 else { 1649 struct udp_iter_state *state = seq->private; 1650 int len; 1651 1652 udp4_format_sock(v, seq, state->bucket, &len); 1653 seq_printf(seq, "%*s\n", 127 - len ,""); 1654 } 1655 return 0; 1656 } 1657 1658 /* ------------------------------------------------------------------------ */ 1659 static struct udp_seq_afinfo udp4_seq_afinfo = { 1660 .name = "udp", 1661 .family = AF_INET, 1662 .hashtable = udp_hash, 1663 .seq_fops = { 1664 .owner = THIS_MODULE, 1665 }, 1666 .seq_ops = { 1667 .show = udp4_seq_show, 1668 }, 1669 }; 1670 1671 static int udp4_proc_init_net(struct net *net) 1672 { 1673 return udp_proc_register(net, &udp4_seq_afinfo); 1674 } 1675 1676 static void udp4_proc_exit_net(struct net *net) 1677 { 1678 udp_proc_unregister(net, &udp4_seq_afinfo); 1679 } 1680 1681 static struct pernet_operations udp4_net_ops = { 1682 .init = udp4_proc_init_net, 1683 .exit = udp4_proc_exit_net, 1684 }; 1685 1686 int __init udp4_proc_init(void) 1687 { 1688 return register_pernet_subsys(&udp4_net_ops); 1689 } 1690 1691 void udp4_proc_exit(void) 1692 { 1693 unregister_pernet_subsys(&udp4_net_ops); 1694 } 1695 #endif /* CONFIG_PROC_FS */ 1696 1697 void __init udp_init(void) 1698 { 1699 unsigned long limit; 1700 1701 /* Set the pressure threshold up by the same strategy of TCP. It is a 1702 * fraction of global memory that is up to 1/2 at 256 MB, decreasing 1703 * toward zero with the amount of memory, with a floor of 128 pages. 1704 */ 1705 limit = min(nr_all_pages, 1UL<<(28-PAGE_SHIFT)) >> (20-PAGE_SHIFT); 1706 limit = (limit * (nr_all_pages >> (20-PAGE_SHIFT))) >> (PAGE_SHIFT-11); 1707 limit = max(limit, 128UL); 1708 sysctl_udp_mem[0] = limit / 4 * 3; 1709 sysctl_udp_mem[1] = limit; 1710 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2; 1711 1712 sysctl_udp_rmem_min = SK_MEM_QUANTUM; 1713 sysctl_udp_wmem_min = SK_MEM_QUANTUM; 1714 } 1715 1716 EXPORT_SYMBOL(udp_disconnect); 1717 EXPORT_SYMBOL(udp_hash); 1718 EXPORT_SYMBOL(udp_hash_lock); 1719 EXPORT_SYMBOL(udp_ioctl); 1720 EXPORT_SYMBOL(udp_prot); 1721 EXPORT_SYMBOL(udp_sendmsg); 1722 EXPORT_SYMBOL(udp_lib_getsockopt); 1723 EXPORT_SYMBOL(udp_lib_setsockopt); 1724 EXPORT_SYMBOL(udp_poll); 1725 EXPORT_SYMBOL(udp_lib_get_port); 1726 1727 #ifdef CONFIG_PROC_FS 1728 EXPORT_SYMBOL(udp_proc_register); 1729 EXPORT_SYMBOL(udp_proc_unregister); 1730 #endif 1731