1 /* 2 * INET An implementation of the TCP/IP protocol suite for the LINUX 3 * operating system. INET is implemented using the BSD Socket 4 * interface as the means of communication with the user level. 5 * 6 * The User Datagram Protocol (UDP). 7 * 8 * Authors: Ross Biro 9 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 10 * Arnt Gulbrandsen, <agulbra@nvg.unit.no> 11 * Alan Cox, <Alan.Cox@linux.org> 12 * Hirokazu Takahashi, <taka@valinux.co.jp> 13 * 14 * Fixes: 15 * Alan Cox : verify_area() calls 16 * Alan Cox : stopped close while in use off icmp 17 * messages. Not a fix but a botch that 18 * for udp at least is 'valid'. 19 * Alan Cox : Fixed icmp handling properly 20 * Alan Cox : Correct error for oversized datagrams 21 * Alan Cox : Tidied select() semantics. 22 * Alan Cox : udp_err() fixed properly, also now 23 * select and read wake correctly on errors 24 * Alan Cox : udp_send verify_area moved to avoid mem leak 25 * Alan Cox : UDP can count its memory 26 * Alan Cox : send to an unknown connection causes 27 * an ECONNREFUSED off the icmp, but 28 * does NOT close. 29 * Alan Cox : Switched to new sk_buff handlers. No more backlog! 30 * Alan Cox : Using generic datagram code. Even smaller and the PEEK 31 * bug no longer crashes it. 32 * Fred Van Kempen : Net2e support for sk->broadcast. 33 * Alan Cox : Uses skb_free_datagram 34 * Alan Cox : Added get/set sockopt support. 35 * Alan Cox : Broadcasting without option set returns EACCES. 36 * Alan Cox : No wakeup calls. Instead we now use the callbacks. 37 * Alan Cox : Use ip_tos and ip_ttl 38 * Alan Cox : SNMP Mibs 39 * Alan Cox : MSG_DONTROUTE, and 0.0.0.0 support. 40 * Matt Dillon : UDP length checks. 41 * Alan Cox : Smarter af_inet used properly. 42 * Alan Cox : Use new kernel side addressing. 43 * Alan Cox : Incorrect return on truncated datagram receive. 44 * Arnt Gulbrandsen : New udp_send and stuff 45 * Alan Cox : Cache last socket 46 * Alan Cox : Route cache 47 * Jon Peatfield : Minor efficiency fix to sendto(). 48 * Mike Shaver : RFC1122 checks. 49 * Alan Cox : Nonblocking error fix. 50 * Willy Konynenberg : Transparent proxying support. 51 * Mike McLagan : Routing by source 52 * David S. Miller : New socket lookup architecture. 53 * Last socket cache retained as it 54 * does have a high hit rate. 55 * Olaf Kirch : Don't linearise iovec on sendmsg. 56 * Andi Kleen : Some cleanups, cache destination entry 57 * for connect. 58 * Vitaly E. Lavrov : Transparent proxy revived after year coma. 59 * Melvin Smith : Check msg_name not msg_namelen in sendto(), 60 * return ENOTCONN for unconnected sockets (POSIX) 61 * Janos Farkas : don't deliver multi/broadcasts to a different 62 * bound-to-device socket 63 * Hirokazu Takahashi : HW checksumming for outgoing UDP 64 * datagrams. 65 * Hirokazu Takahashi : sendfile() on UDP works now. 66 * Arnaldo C. Melo : convert /proc/net/udp to seq_file 67 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which 68 * Alexey Kuznetsov: allow both IPv4 and IPv6 sockets to bind 69 * a single port at the same time. 70 * Derek Atkins <derek@ihtfp.com>: Add Encapulation Support 71 * James Chapman : Add L2TP encapsulation type. 72 * 73 * 74 * This program is free software; you can redistribute it and/or 75 * modify it under the terms of the GNU General Public License 76 * as published by the Free Software Foundation; either version 77 * 2 of the License, or (at your option) any later version. 78 */ 79 80 #include <asm/system.h> 81 #include <asm/uaccess.h> 82 #include <asm/ioctls.h> 83 #include <linux/bootmem.h> 84 #include <linux/types.h> 85 #include <linux/fcntl.h> 86 #include <linux/module.h> 87 #include <linux/socket.h> 88 #include <linux/sockios.h> 89 #include <linux/igmp.h> 90 #include <linux/in.h> 91 #include <linux/errno.h> 92 #include <linux/timer.h> 93 #include <linux/mm.h> 94 #include <linux/inet.h> 95 #include <linux/netdevice.h> 96 #include <net/tcp_states.h> 97 #include <linux/skbuff.h> 98 #include <linux/proc_fs.h> 99 #include <linux/seq_file.h> 100 #include <net/net_namespace.h> 101 #include <net/icmp.h> 102 #include <net/route.h> 103 #include <net/checksum.h> 104 #include <net/xfrm.h> 105 #include "udp_impl.h" 106 107 /* 108 * Snmp MIB for the UDP layer 109 */ 110 111 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 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 (s->sk_hash != hnum || 300 (inet->daddr && inet->daddr != rmt_addr) || 301 (inet->dport != rmt_port && inet->dport) || 302 (inet->rcv_saddr && inet->rcv_saddr != loc_addr) || 303 ipv6_only_sock(s) || 304 (s->sk_bound_dev_if && s->sk_bound_dev_if != dif)) 305 continue; 306 if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif)) 307 continue; 308 goto found; 309 } 310 s = NULL; 311 found: 312 return s; 313 } 314 315 /* 316 * This routine is called by the ICMP module when it gets some 317 * sort of error condition. If err < 0 then the socket should 318 * be closed and the error returned to the user. If err > 0 319 * it's just the icmp type << 8 | icmp code. 320 * Header points to the ip header of the error packet. We move 321 * on past this. Then (as it used to claim before adjustment) 322 * header points to the first 8 bytes of the udp header. We need 323 * to find the appropriate port. 324 */ 325 326 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct hlist_head udptable[]) 327 { 328 struct inet_sock *inet; 329 struct iphdr *iph = (struct iphdr*)skb->data; 330 struct udphdr *uh = (struct udphdr*)(skb->data+(iph->ihl<<2)); 331 const int type = icmp_hdr(skb)->type; 332 const int code = icmp_hdr(skb)->code; 333 struct sock *sk; 334 int harderr; 335 int err; 336 struct net *net = dev_net(skb->dev); 337 338 sk = __udp4_lib_lookup(net, iph->daddr, uh->dest, 339 iph->saddr, uh->source, skb->dev->ifindex, udptable); 340 if (sk == NULL) { 341 ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS); 342 return; /* No socket for error */ 343 } 344 345 err = 0; 346 harderr = 0; 347 inet = inet_sk(sk); 348 349 switch (type) { 350 default: 351 case ICMP_TIME_EXCEEDED: 352 err = EHOSTUNREACH; 353 break; 354 case ICMP_SOURCE_QUENCH: 355 goto out; 356 case ICMP_PARAMETERPROB: 357 err = EPROTO; 358 harderr = 1; 359 break; 360 case ICMP_DEST_UNREACH: 361 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */ 362 if (inet->pmtudisc != IP_PMTUDISC_DONT) { 363 err = EMSGSIZE; 364 harderr = 1; 365 break; 366 } 367 goto out; 368 } 369 err = EHOSTUNREACH; 370 if (code <= NR_ICMP_UNREACH) { 371 harderr = icmp_err_convert[code].fatal; 372 err = icmp_err_convert[code].errno; 373 } 374 break; 375 } 376 377 /* 378 * RFC1122: OK. Passes ICMP errors back to application, as per 379 * 4.1.3.3. 380 */ 381 if (!inet->recverr) { 382 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 383 goto out; 384 } else { 385 ip_icmp_error(sk, skb, err, uh->dest, info, (u8*)(uh+1)); 386 } 387 sk->sk_err = err; 388 sk->sk_error_report(sk); 389 out: 390 sock_put(sk); 391 } 392 393 void udp_err(struct sk_buff *skb, u32 info) 394 { 395 __udp4_lib_err(skb, info, udp_hash); 396 } 397 398 /* 399 * Throw away all pending data and cancel the corking. Socket is locked. 400 */ 401 void udp_flush_pending_frames(struct sock *sk) 402 { 403 struct udp_sock *up = udp_sk(sk); 404 405 if (up->pending) { 406 up->len = 0; 407 up->pending = 0; 408 ip_flush_pending_frames(sk); 409 } 410 } 411 EXPORT_SYMBOL(udp_flush_pending_frames); 412 413 /** 414 * udp4_hwcsum_outgoing - handle outgoing HW checksumming 415 * @sk: socket we are sending on 416 * @skb: sk_buff containing the filled-in UDP header 417 * (checksum field must be zeroed out) 418 */ 419 static void udp4_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb, 420 __be32 src, __be32 dst, int len ) 421 { 422 unsigned int offset; 423 struct udphdr *uh = udp_hdr(skb); 424 __wsum csum = 0; 425 426 if (skb_queue_len(&sk->sk_write_queue) == 1) { 427 /* 428 * Only one fragment on the socket. 429 */ 430 skb->csum_start = skb_transport_header(skb) - skb->head; 431 skb->csum_offset = offsetof(struct udphdr, check); 432 uh->check = ~csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, 0); 433 } else { 434 /* 435 * HW-checksum won't work as there are two or more 436 * fragments on the socket so that all csums of sk_buffs 437 * should be together 438 */ 439 offset = skb_transport_offset(skb); 440 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0); 441 442 skb->ip_summed = CHECKSUM_NONE; 443 444 skb_queue_walk(&sk->sk_write_queue, skb) { 445 csum = csum_add(csum, skb->csum); 446 } 447 448 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); 449 if (uh->check == 0) 450 uh->check = CSUM_MANGLED_0; 451 } 452 } 453 454 /* 455 * Push out all pending data as one UDP datagram. Socket is locked. 456 */ 457 static int udp_push_pending_frames(struct sock *sk) 458 { 459 struct udp_sock *up = udp_sk(sk); 460 struct inet_sock *inet = inet_sk(sk); 461 struct flowi *fl = &inet->cork.fl; 462 struct sk_buff *skb; 463 struct udphdr *uh; 464 int err = 0; 465 int is_udplite = IS_UDPLITE(sk); 466 __wsum csum = 0; 467 468 /* Grab the skbuff where UDP header space exists. */ 469 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL) 470 goto out; 471 472 /* 473 * Create a UDP header 474 */ 475 uh = udp_hdr(skb); 476 uh->source = fl->fl_ip_sport; 477 uh->dest = fl->fl_ip_dport; 478 uh->len = htons(up->len); 479 uh->check = 0; 480 481 if (is_udplite) /* UDP-Lite */ 482 csum = udplite_csum_outgoing(sk, skb); 483 484 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */ 485 486 skb->ip_summed = CHECKSUM_NONE; 487 goto send; 488 489 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 490 491 udp4_hwcsum_outgoing(sk, skb, fl->fl4_src,fl->fl4_dst, up->len); 492 goto send; 493 494 } else /* `normal' UDP */ 495 csum = udp_csum_outgoing(sk, skb); 496 497 /* add protocol-dependent pseudo-header */ 498 uh->check = csum_tcpudp_magic(fl->fl4_src, fl->fl4_dst, up->len, 499 sk->sk_protocol, csum ); 500 if (uh->check == 0) 501 uh->check = CSUM_MANGLED_0; 502 503 send: 504 err = ip_push_pending_frames(sk); 505 out: 506 up->len = 0; 507 up->pending = 0; 508 if (!err) 509 UDP_INC_STATS_USER(sock_net(sk), 510 UDP_MIB_OUTDATAGRAMS, is_udplite); 511 return err; 512 } 513 514 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 515 size_t len) 516 { 517 struct inet_sock *inet = inet_sk(sk); 518 struct udp_sock *up = udp_sk(sk); 519 int ulen = len; 520 struct ipcm_cookie ipc; 521 struct rtable *rt = NULL; 522 int free = 0; 523 int connected = 0; 524 __be32 daddr, faddr, saddr; 525 __be16 dport; 526 u8 tos; 527 int err, is_udplite = IS_UDPLITE(sk); 528 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; 529 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 530 531 if (len > 0xFFFF) 532 return -EMSGSIZE; 533 534 /* 535 * Check the flags. 536 */ 537 538 if (msg->msg_flags&MSG_OOB) /* Mirror BSD error message compatibility */ 539 return -EOPNOTSUPP; 540 541 ipc.opt = NULL; 542 543 if (up->pending) { 544 /* 545 * There are pending frames. 546 * The socket lock must be held while it's corked. 547 */ 548 lock_sock(sk); 549 if (likely(up->pending)) { 550 if (unlikely(up->pending != AF_INET)) { 551 release_sock(sk); 552 return -EINVAL; 553 } 554 goto do_append_data; 555 } 556 release_sock(sk); 557 } 558 ulen += sizeof(struct udphdr); 559 560 /* 561 * Get and verify the address. 562 */ 563 if (msg->msg_name) { 564 struct sockaddr_in * usin = (struct sockaddr_in*)msg->msg_name; 565 if (msg->msg_namelen < sizeof(*usin)) 566 return -EINVAL; 567 if (usin->sin_family != AF_INET) { 568 if (usin->sin_family != AF_UNSPEC) 569 return -EAFNOSUPPORT; 570 } 571 572 daddr = usin->sin_addr.s_addr; 573 dport = usin->sin_port; 574 if (dport == 0) 575 return -EINVAL; 576 } else { 577 if (sk->sk_state != TCP_ESTABLISHED) 578 return -EDESTADDRREQ; 579 daddr = inet->daddr; 580 dport = inet->dport; 581 /* Open fast path for connected socket. 582 Route will not be used, if at least one option is set. 583 */ 584 connected = 1; 585 } 586 ipc.addr = inet->saddr; 587 588 ipc.oif = sk->sk_bound_dev_if; 589 if (msg->msg_controllen) { 590 err = ip_cmsg_send(sock_net(sk), msg, &ipc); 591 if (err) 592 return err; 593 if (ipc.opt) 594 free = 1; 595 connected = 0; 596 } 597 if (!ipc.opt) 598 ipc.opt = inet->opt; 599 600 saddr = ipc.addr; 601 ipc.addr = faddr = daddr; 602 603 if (ipc.opt && ipc.opt->srr) { 604 if (!daddr) 605 return -EINVAL; 606 faddr = ipc.opt->faddr; 607 connected = 0; 608 } 609 tos = RT_TOS(inet->tos); 610 if (sock_flag(sk, SOCK_LOCALROUTE) || 611 (msg->msg_flags & MSG_DONTROUTE) || 612 (ipc.opt && ipc.opt->is_strictroute)) { 613 tos |= RTO_ONLINK; 614 connected = 0; 615 } 616 617 if (ipv4_is_multicast(daddr)) { 618 if (!ipc.oif) 619 ipc.oif = inet->mc_index; 620 if (!saddr) 621 saddr = inet->mc_addr; 622 connected = 0; 623 } 624 625 if (connected) 626 rt = (struct rtable*)sk_dst_check(sk, 0); 627 628 if (rt == NULL) { 629 struct flowi fl = { .oif = ipc.oif, 630 .nl_u = { .ip4_u = 631 { .daddr = faddr, 632 .saddr = saddr, 633 .tos = tos } }, 634 .proto = sk->sk_protocol, 635 .uli_u = { .ports = 636 { .sport = inet->sport, 637 .dport = dport } } }; 638 struct net *net = sock_net(sk); 639 640 security_sk_classify_flow(sk, &fl); 641 err = ip_route_output_flow(net, &rt, &fl, sk, 1); 642 if (err) { 643 if (err == -ENETUNREACH) 644 IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES); 645 goto out; 646 } 647 648 err = -EACCES; 649 if ((rt->rt_flags & RTCF_BROADCAST) && 650 !sock_flag(sk, SOCK_BROADCAST)) 651 goto out; 652 if (connected) 653 sk_dst_set(sk, dst_clone(&rt->u.dst)); 654 } 655 656 if (msg->msg_flags&MSG_CONFIRM) 657 goto do_confirm; 658 back_from_confirm: 659 660 saddr = rt->rt_src; 661 if (!ipc.addr) 662 daddr = ipc.addr = rt->rt_dst; 663 664 lock_sock(sk); 665 if (unlikely(up->pending)) { 666 /* The socket is already corked while preparing it. */ 667 /* ... which is an evident application bug. --ANK */ 668 release_sock(sk); 669 670 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n"); 671 err = -EINVAL; 672 goto out; 673 } 674 /* 675 * Now cork the socket to pend data. 676 */ 677 inet->cork.fl.fl4_dst = daddr; 678 inet->cork.fl.fl_ip_dport = dport; 679 inet->cork.fl.fl4_src = saddr; 680 inet->cork.fl.fl_ip_sport = inet->sport; 681 up->pending = AF_INET; 682 683 do_append_data: 684 up->len += ulen; 685 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 686 err = ip_append_data(sk, getfrag, msg->msg_iov, ulen, 687 sizeof(struct udphdr), &ipc, rt, 688 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 689 if (err) 690 udp_flush_pending_frames(sk); 691 else if (!corkreq) 692 err = udp_push_pending_frames(sk); 693 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 694 up->pending = 0; 695 release_sock(sk); 696 697 out: 698 ip_rt_put(rt); 699 if (free) 700 kfree(ipc.opt); 701 if (!err) 702 return len; 703 /* 704 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 705 * ENOBUFS might not be good (it's not tunable per se), but otherwise 706 * we don't have a good statistic (IpOutDiscards but it can be too many 707 * things). We could add another new stat but at least for now that 708 * seems like overkill. 709 */ 710 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 711 UDP_INC_STATS_USER(sock_net(sk), 712 UDP_MIB_SNDBUFERRORS, is_udplite); 713 } 714 return err; 715 716 do_confirm: 717 dst_confirm(&rt->u.dst); 718 if (!(msg->msg_flags&MSG_PROBE) || len) 719 goto back_from_confirm; 720 err = 0; 721 goto out; 722 } 723 724 int udp_sendpage(struct sock *sk, struct page *page, int offset, 725 size_t size, int flags) 726 { 727 struct udp_sock *up = udp_sk(sk); 728 int ret; 729 730 if (!up->pending) { 731 struct msghdr msg = { .msg_flags = flags|MSG_MORE }; 732 733 /* Call udp_sendmsg to specify destination address which 734 * sendpage interface can't pass. 735 * This will succeed only when the socket is connected. 736 */ 737 ret = udp_sendmsg(NULL, sk, &msg, 0); 738 if (ret < 0) 739 return ret; 740 } 741 742 lock_sock(sk); 743 744 if (unlikely(!up->pending)) { 745 release_sock(sk); 746 747 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n"); 748 return -EINVAL; 749 } 750 751 ret = ip_append_page(sk, page, offset, size, flags); 752 if (ret == -EOPNOTSUPP) { 753 release_sock(sk); 754 return sock_no_sendpage(sk->sk_socket, page, offset, 755 size, flags); 756 } 757 if (ret < 0) { 758 udp_flush_pending_frames(sk); 759 goto out; 760 } 761 762 up->len += size; 763 if (!(up->corkflag || (flags&MSG_MORE))) 764 ret = udp_push_pending_frames(sk); 765 if (!ret) 766 ret = size; 767 out: 768 release_sock(sk); 769 return ret; 770 } 771 772 /* 773 * IOCTL requests applicable to the UDP protocol 774 */ 775 776 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg) 777 { 778 switch (cmd) { 779 case SIOCOUTQ: 780 { 781 int amount = atomic_read(&sk->sk_wmem_alloc); 782 return put_user(amount, (int __user *)arg); 783 } 784 785 case SIOCINQ: 786 { 787 struct sk_buff *skb; 788 unsigned long amount; 789 790 amount = 0; 791 spin_lock_bh(&sk->sk_receive_queue.lock); 792 skb = skb_peek(&sk->sk_receive_queue); 793 if (skb != NULL) { 794 /* 795 * We will only return the amount 796 * of this packet since that is all 797 * that will be read. 798 */ 799 amount = skb->len - sizeof(struct udphdr); 800 } 801 spin_unlock_bh(&sk->sk_receive_queue.lock); 802 return put_user(amount, (int __user *)arg); 803 } 804 805 default: 806 return -ENOIOCTLCMD; 807 } 808 809 return 0; 810 } 811 812 /* 813 * This should be easy, if there is something there we 814 * return it, otherwise we block. 815 */ 816 817 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 818 size_t len, int noblock, int flags, int *addr_len) 819 { 820 struct inet_sock *inet = inet_sk(sk); 821 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name; 822 struct sk_buff *skb; 823 unsigned int ulen, copied; 824 int peeked; 825 int err; 826 int is_udplite = IS_UDPLITE(sk); 827 828 /* 829 * Check any passed addresses 830 */ 831 if (addr_len) 832 *addr_len=sizeof(*sin); 833 834 if (flags & MSG_ERRQUEUE) 835 return ip_recv_error(sk, msg, len); 836 837 try_again: 838 skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0), 839 &peeked, &err); 840 if (!skb) 841 goto out; 842 843 ulen = skb->len - sizeof(struct udphdr); 844 copied = len; 845 if (copied > ulen) 846 copied = ulen; 847 else if (copied < ulen) 848 msg->msg_flags |= MSG_TRUNC; 849 850 /* 851 * If checksum is needed at all, try to do it while copying the 852 * data. If the data is truncated, or if we only want a partial 853 * coverage checksum (UDP-Lite), do it before the copy. 854 */ 855 856 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) { 857 if (udp_lib_checksum_complete(skb)) 858 goto csum_copy_err; 859 } 860 861 if (skb_csum_unnecessary(skb)) 862 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr), 863 msg->msg_iov, copied ); 864 else { 865 err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov); 866 867 if (err == -EINVAL) 868 goto csum_copy_err; 869 } 870 871 if (err) 872 goto out_free; 873 874 if (!peeked) 875 UDP_INC_STATS_USER(sock_net(sk), 876 UDP_MIB_INDATAGRAMS, is_udplite); 877 878 sock_recv_timestamp(msg, sk, skb); 879 880 /* Copy the address. */ 881 if (sin) 882 { 883 sin->sin_family = AF_INET; 884 sin->sin_port = udp_hdr(skb)->source; 885 sin->sin_addr.s_addr = ip_hdr(skb)->saddr; 886 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 887 } 888 if (inet->cmsg_flags) 889 ip_cmsg_recv(msg, skb); 890 891 err = copied; 892 if (flags & MSG_TRUNC) 893 err = ulen; 894 895 out_free: 896 lock_sock(sk); 897 skb_free_datagram(sk, skb); 898 release_sock(sk); 899 out: 900 return err; 901 902 csum_copy_err: 903 lock_sock(sk); 904 if (!skb_kill_datagram(sk, skb, flags)) 905 UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 906 release_sock(sk); 907 908 if (noblock) 909 return -EAGAIN; 910 goto try_again; 911 } 912 913 914 int udp_disconnect(struct sock *sk, int flags) 915 { 916 struct inet_sock *inet = inet_sk(sk); 917 /* 918 * 1003.1g - break association. 919 */ 920 921 sk->sk_state = TCP_CLOSE; 922 inet->daddr = 0; 923 inet->dport = 0; 924 sk->sk_bound_dev_if = 0; 925 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) 926 inet_reset_saddr(sk); 927 928 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { 929 sk->sk_prot->unhash(sk); 930 inet->sport = 0; 931 } 932 sk_dst_reset(sk); 933 return 0; 934 } 935 936 static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb) 937 { 938 int is_udplite = IS_UDPLITE(sk); 939 int rc; 940 941 if ((rc = sock_queue_rcv_skb(sk, skb)) < 0) { 942 /* Note that an ENOMEM error is charged twice */ 943 if (rc == -ENOMEM) 944 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS, 945 is_udplite); 946 goto drop; 947 } 948 949 return 0; 950 951 drop: 952 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 953 kfree_skb(skb); 954 return -1; 955 } 956 957 /* returns: 958 * -1: error 959 * 0: success 960 * >0: "udp encap" protocol resubmission 961 * 962 * Note that in the success and error cases, the skb is assumed to 963 * have either been requeued or freed. 964 */ 965 int udp_queue_rcv_skb(struct sock * sk, struct sk_buff *skb) 966 { 967 struct udp_sock *up = udp_sk(sk); 968 int rc; 969 int is_udplite = IS_UDPLITE(sk); 970 971 /* 972 * Charge it to the socket, dropping if the queue is full. 973 */ 974 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) 975 goto drop; 976 nf_reset(skb); 977 978 if (up->encap_type) { 979 /* 980 * This is an encapsulation socket so pass the skb to 981 * the socket's udp_encap_rcv() hook. Otherwise, just 982 * fall through and pass this up the UDP socket. 983 * up->encap_rcv() returns the following value: 984 * =0 if skb was successfully passed to the encap 985 * handler or was discarded by it. 986 * >0 if skb should be passed on to UDP. 987 * <0 if skb should be resubmitted as proto -N 988 */ 989 990 /* if we're overly short, let UDP handle it */ 991 if (skb->len > sizeof(struct udphdr) && 992 up->encap_rcv != NULL) { 993 int ret; 994 995 ret = (*up->encap_rcv)(sk, skb); 996 if (ret <= 0) { 997 UDP_INC_STATS_BH(sock_net(sk), 998 UDP_MIB_INDATAGRAMS, 999 is_udplite); 1000 return -ret; 1001 } 1002 } 1003 1004 /* FALLTHROUGH -- it's a UDP Packet */ 1005 } 1006 1007 /* 1008 * UDP-Lite specific tests, ignored on UDP sockets 1009 */ 1010 if ((is_udplite & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 1011 1012 /* 1013 * MIB statistics other than incrementing the error count are 1014 * disabled for the following two types of errors: these depend 1015 * on the application settings, not on the functioning of the 1016 * protocol stack as such. 1017 * 1018 * RFC 3828 here recommends (sec 3.3): "There should also be a 1019 * way ... to ... at least let the receiving application block 1020 * delivery of packets with coverage values less than a value 1021 * provided by the application." 1022 */ 1023 if (up->pcrlen == 0) { /* full coverage was set */ 1024 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage " 1025 "%d while full coverage %d requested\n", 1026 UDP_SKB_CB(skb)->cscov, skb->len); 1027 goto drop; 1028 } 1029 /* The next case involves violating the min. coverage requested 1030 * by the receiver. This is subtle: if receiver wants x and x is 1031 * greater than the buffersize/MTU then receiver will complain 1032 * that it wants x while sender emits packets of smaller size y. 1033 * Therefore the above ...()->partial_cov statement is essential. 1034 */ 1035 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 1036 LIMIT_NETDEBUG(KERN_WARNING 1037 "UDPLITE: coverage %d too small, need min %d\n", 1038 UDP_SKB_CB(skb)->cscov, up->pcrlen); 1039 goto drop; 1040 } 1041 } 1042 1043 if (sk->sk_filter) { 1044 if (udp_lib_checksum_complete(skb)) 1045 goto drop; 1046 } 1047 1048 rc = 0; 1049 1050 bh_lock_sock(sk); 1051 if (!sock_owned_by_user(sk)) 1052 rc = __udp_queue_rcv_skb(sk, skb); 1053 else 1054 sk_add_backlog(sk, skb); 1055 bh_unlock_sock(sk); 1056 1057 return rc; 1058 1059 drop: 1060 UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite); 1061 kfree_skb(skb); 1062 return -1; 1063 } 1064 1065 /* 1066 * Multicasts and broadcasts go to each listener. 1067 * 1068 * Note: called only from the BH handler context, 1069 * so we don't need to lock the hashes. 1070 */ 1071 static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb, 1072 struct udphdr *uh, 1073 __be32 saddr, __be32 daddr, 1074 struct hlist_head udptable[]) 1075 { 1076 struct sock *sk; 1077 int dif; 1078 1079 read_lock(&udp_hash_lock); 1080 sk = sk_head(&udptable[udp_hashfn(net, ntohs(uh->dest))]); 1081 dif = skb->dev->ifindex; 1082 sk = udp_v4_mcast_next(sk, uh->dest, daddr, uh->source, saddr, dif); 1083 if (sk) { 1084 struct sock *sknext = NULL; 1085 1086 do { 1087 struct sk_buff *skb1 = skb; 1088 1089 sknext = udp_v4_mcast_next(sk_next(sk), uh->dest, daddr, 1090 uh->source, saddr, dif); 1091 if (sknext) 1092 skb1 = skb_clone(skb, GFP_ATOMIC); 1093 1094 if (skb1) { 1095 int ret = udp_queue_rcv_skb(sk, skb1); 1096 if (ret > 0) 1097 /* we should probably re-process instead 1098 * of dropping packets here. */ 1099 kfree_skb(skb1); 1100 } 1101 sk = sknext; 1102 } while (sknext); 1103 } else 1104 kfree_skb(skb); 1105 read_unlock(&udp_hash_lock); 1106 return 0; 1107 } 1108 1109 /* Initialize UDP checksum. If exited with zero value (success), 1110 * CHECKSUM_UNNECESSARY means, that no more checks are required. 1111 * Otherwise, csum completion requires chacksumming packet body, 1112 * including udp header and folding it to skb->csum. 1113 */ 1114 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, 1115 int proto) 1116 { 1117 const struct iphdr *iph; 1118 int err; 1119 1120 UDP_SKB_CB(skb)->partial_cov = 0; 1121 UDP_SKB_CB(skb)->cscov = skb->len; 1122 1123 if (proto == IPPROTO_UDPLITE) { 1124 err = udplite_checksum_init(skb, uh); 1125 if (err) 1126 return err; 1127 } 1128 1129 iph = ip_hdr(skb); 1130 if (uh->check == 0) { 1131 skb->ip_summed = CHECKSUM_UNNECESSARY; 1132 } else if (skb->ip_summed == CHECKSUM_COMPLETE) { 1133 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len, 1134 proto, skb->csum)) 1135 skb->ip_summed = CHECKSUM_UNNECESSARY; 1136 } 1137 if (!skb_csum_unnecessary(skb)) 1138 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr, 1139 skb->len, proto, 0); 1140 /* Probably, we should checksum udp header (it should be in cache 1141 * in any case) and data in tiny packets (< rx copybreak). 1142 */ 1143 1144 return 0; 1145 } 1146 1147 /* 1148 * All we need to do is get the socket, and then do a checksum. 1149 */ 1150 1151 int __udp4_lib_rcv(struct sk_buff *skb, struct hlist_head udptable[], 1152 int proto) 1153 { 1154 struct sock *sk; 1155 struct udphdr *uh = udp_hdr(skb); 1156 unsigned short ulen; 1157 struct rtable *rt = (struct rtable*)skb->dst; 1158 __be32 saddr = ip_hdr(skb)->saddr; 1159 __be32 daddr = ip_hdr(skb)->daddr; 1160 struct net *net = dev_net(skb->dev); 1161 1162 /* 1163 * Validate the packet. 1164 */ 1165 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 1166 goto drop; /* No space for header. */ 1167 1168 ulen = ntohs(uh->len); 1169 if (ulen > skb->len) 1170 goto short_packet; 1171 1172 if (proto == IPPROTO_UDP) { 1173 /* UDP validates ulen. */ 1174 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) 1175 goto short_packet; 1176 uh = udp_hdr(skb); 1177 } 1178 1179 if (udp4_csum_init(skb, uh, proto)) 1180 goto csum_error; 1181 1182 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) 1183 return __udp4_lib_mcast_deliver(net, skb, uh, 1184 saddr, daddr, udptable); 1185 1186 sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable); 1187 1188 if (sk != NULL) { 1189 int ret = udp_queue_rcv_skb(sk, skb); 1190 sock_put(sk); 1191 1192 /* a return value > 0 means to resubmit the input, but 1193 * it wants the return to be -protocol, or 0 1194 */ 1195 if (ret > 0) 1196 return -ret; 1197 return 0; 1198 } 1199 1200 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 1201 goto drop; 1202 nf_reset(skb); 1203 1204 /* No socket. Drop packet silently, if checksum is wrong */ 1205 if (udp_lib_checksum_complete(skb)) 1206 goto csum_error; 1207 1208 UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 1209 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); 1210 1211 /* 1212 * Hmm. We got an UDP packet to a port to which we 1213 * don't wanna listen. Ignore it. 1214 */ 1215 kfree_skb(skb); 1216 return 0; 1217 1218 short_packet: 1219 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From " NIPQUAD_FMT ":%u %d/%d to " NIPQUAD_FMT ":%u\n", 1220 proto == IPPROTO_UDPLITE ? "-Lite" : "", 1221 NIPQUAD(saddr), 1222 ntohs(uh->source), 1223 ulen, 1224 skb->len, 1225 NIPQUAD(daddr), 1226 ntohs(uh->dest)); 1227 goto drop; 1228 1229 csum_error: 1230 /* 1231 * RFC1122: OK. Discards the bad packet silently (as far as 1232 * the network is concerned, anyway) as per 4.1.3.4 (MUST). 1233 */ 1234 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From " NIPQUAD_FMT ":%u to " NIPQUAD_FMT ":%u ulen %d\n", 1235 proto == IPPROTO_UDPLITE ? "-Lite" : "", 1236 NIPQUAD(saddr), 1237 ntohs(uh->source), 1238 NIPQUAD(daddr), 1239 ntohs(uh->dest), 1240 ulen); 1241 drop: 1242 UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 1243 kfree_skb(skb); 1244 return 0; 1245 } 1246 1247 int udp_rcv(struct sk_buff *skb) 1248 { 1249 return __udp4_lib_rcv(skb, udp_hash, IPPROTO_UDP); 1250 } 1251 1252 void udp_destroy_sock(struct sock *sk) 1253 { 1254 lock_sock(sk); 1255 udp_flush_pending_frames(sk); 1256 release_sock(sk); 1257 } 1258 1259 /* 1260 * Socket option code for UDP 1261 */ 1262 int udp_lib_setsockopt(struct sock *sk, int level, int optname, 1263 char __user *optval, int optlen, 1264 int (*push_pending_frames)(struct sock *)) 1265 { 1266 struct udp_sock *up = udp_sk(sk); 1267 int val; 1268 int err = 0; 1269 int is_udplite = IS_UDPLITE(sk); 1270 1271 if (optlen<sizeof(int)) 1272 return -EINVAL; 1273 1274 if (get_user(val, (int __user *)optval)) 1275 return -EFAULT; 1276 1277 switch (optname) { 1278 case UDP_CORK: 1279 if (val != 0) { 1280 up->corkflag = 1; 1281 } else { 1282 up->corkflag = 0; 1283 lock_sock(sk); 1284 (*push_pending_frames)(sk); 1285 release_sock(sk); 1286 } 1287 break; 1288 1289 case UDP_ENCAP: 1290 switch (val) { 1291 case 0: 1292 case UDP_ENCAP_ESPINUDP: 1293 case UDP_ENCAP_ESPINUDP_NON_IKE: 1294 up->encap_rcv = xfrm4_udp_encap_rcv; 1295 /* FALLTHROUGH */ 1296 case UDP_ENCAP_L2TPINUDP: 1297 up->encap_type = val; 1298 break; 1299 default: 1300 err = -ENOPROTOOPT; 1301 break; 1302 } 1303 break; 1304 1305 /* 1306 * UDP-Lite's partial checksum coverage (RFC 3828). 1307 */ 1308 /* The sender sets actual checksum coverage length via this option. 1309 * The case coverage > packet length is handled by send module. */ 1310 case UDPLITE_SEND_CSCOV: 1311 if (!is_udplite) /* Disable the option on UDP sockets */ 1312 return -ENOPROTOOPT; 1313 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ 1314 val = 8; 1315 else if (val > USHORT_MAX) 1316 val = USHORT_MAX; 1317 up->pcslen = val; 1318 up->pcflag |= UDPLITE_SEND_CC; 1319 break; 1320 1321 /* The receiver specifies a minimum checksum coverage value. To make 1322 * sense, this should be set to at least 8 (as done below). If zero is 1323 * used, this again means full checksum coverage. */ 1324 case UDPLITE_RECV_CSCOV: 1325 if (!is_udplite) /* Disable the option on UDP sockets */ 1326 return -ENOPROTOOPT; 1327 if (val != 0 && val < 8) /* Avoid silly minimal values. */ 1328 val = 8; 1329 else if (val > USHORT_MAX) 1330 val = USHORT_MAX; 1331 up->pcrlen = val; 1332 up->pcflag |= UDPLITE_RECV_CC; 1333 break; 1334 1335 default: 1336 err = -ENOPROTOOPT; 1337 break; 1338 } 1339 1340 return err; 1341 } 1342 1343 int udp_setsockopt(struct sock *sk, int level, int optname, 1344 char __user *optval, int optlen) 1345 { 1346 if (level == SOL_UDP || level == SOL_UDPLITE) 1347 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1348 udp_push_pending_frames); 1349 return ip_setsockopt(sk, level, optname, optval, optlen); 1350 } 1351 1352 #ifdef CONFIG_COMPAT 1353 int compat_udp_setsockopt(struct sock *sk, int level, int optname, 1354 char __user *optval, int optlen) 1355 { 1356 if (level == SOL_UDP || level == SOL_UDPLITE) 1357 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1358 udp_push_pending_frames); 1359 return compat_ip_setsockopt(sk, level, optname, optval, optlen); 1360 } 1361 #endif 1362 1363 int udp_lib_getsockopt(struct sock *sk, int level, int optname, 1364 char __user *optval, int __user *optlen) 1365 { 1366 struct udp_sock *up = udp_sk(sk); 1367 int val, len; 1368 1369 if (get_user(len,optlen)) 1370 return -EFAULT; 1371 1372 len = min_t(unsigned int, len, sizeof(int)); 1373 1374 if (len < 0) 1375 return -EINVAL; 1376 1377 switch (optname) { 1378 case UDP_CORK: 1379 val = up->corkflag; 1380 break; 1381 1382 case UDP_ENCAP: 1383 val = up->encap_type; 1384 break; 1385 1386 /* The following two cannot be changed on UDP sockets, the return is 1387 * always 0 (which corresponds to the full checksum coverage of UDP). */ 1388 case UDPLITE_SEND_CSCOV: 1389 val = up->pcslen; 1390 break; 1391 1392 case UDPLITE_RECV_CSCOV: 1393 val = up->pcrlen; 1394 break; 1395 1396 default: 1397 return -ENOPROTOOPT; 1398 } 1399 1400 if (put_user(len, optlen)) 1401 return -EFAULT; 1402 if (copy_to_user(optval, &val,len)) 1403 return -EFAULT; 1404 return 0; 1405 } 1406 1407 int udp_getsockopt(struct sock *sk, int level, int optname, 1408 char __user *optval, int __user *optlen) 1409 { 1410 if (level == SOL_UDP || level == SOL_UDPLITE) 1411 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1412 return ip_getsockopt(sk, level, optname, optval, optlen); 1413 } 1414 1415 #ifdef CONFIG_COMPAT 1416 int compat_udp_getsockopt(struct sock *sk, int level, int optname, 1417 char __user *optval, int __user *optlen) 1418 { 1419 if (level == SOL_UDP || level == SOL_UDPLITE) 1420 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1421 return compat_ip_getsockopt(sk, level, optname, optval, optlen); 1422 } 1423 #endif 1424 /** 1425 * udp_poll - wait for a UDP event. 1426 * @file - file struct 1427 * @sock - socket 1428 * @wait - poll table 1429 * 1430 * This is same as datagram poll, except for the special case of 1431 * blocking sockets. If application is using a blocking fd 1432 * and a packet with checksum error is in the queue; 1433 * then it could get return from select indicating data available 1434 * but then block when reading it. Add special case code 1435 * to work around these arguably broken applications. 1436 */ 1437 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait) 1438 { 1439 unsigned int mask = datagram_poll(file, sock, wait); 1440 struct sock *sk = sock->sk; 1441 int is_lite = IS_UDPLITE(sk); 1442 1443 /* Check for false positives due to checksum errors */ 1444 if ( (mask & POLLRDNORM) && 1445 !(file->f_flags & O_NONBLOCK) && 1446 !(sk->sk_shutdown & RCV_SHUTDOWN)){ 1447 struct sk_buff_head *rcvq = &sk->sk_receive_queue; 1448 struct sk_buff *skb; 1449 1450 spin_lock_bh(&rcvq->lock); 1451 while ((skb = skb_peek(rcvq)) != NULL && 1452 udp_lib_checksum_complete(skb)) { 1453 UDP_INC_STATS_BH(sock_net(sk), 1454 UDP_MIB_INERRORS, is_lite); 1455 __skb_unlink(skb, rcvq); 1456 kfree_skb(skb); 1457 } 1458 spin_unlock_bh(&rcvq->lock); 1459 1460 /* nothing to see, move along */ 1461 if (skb == NULL) 1462 mask &= ~(POLLIN | POLLRDNORM); 1463 } 1464 1465 return mask; 1466 1467 } 1468 1469 struct proto udp_prot = { 1470 .name = "UDP", 1471 .owner = THIS_MODULE, 1472 .close = udp_lib_close, 1473 .connect = ip4_datagram_connect, 1474 .disconnect = udp_disconnect, 1475 .ioctl = udp_ioctl, 1476 .destroy = udp_destroy_sock, 1477 .setsockopt = udp_setsockopt, 1478 .getsockopt = udp_getsockopt, 1479 .sendmsg = udp_sendmsg, 1480 .recvmsg = udp_recvmsg, 1481 .sendpage = udp_sendpage, 1482 .backlog_rcv = __udp_queue_rcv_skb, 1483 .hash = udp_lib_hash, 1484 .unhash = udp_lib_unhash, 1485 .get_port = udp_v4_get_port, 1486 .memory_allocated = &udp_memory_allocated, 1487 .sysctl_mem = sysctl_udp_mem, 1488 .sysctl_wmem = &sysctl_udp_wmem_min, 1489 .sysctl_rmem = &sysctl_udp_rmem_min, 1490 .obj_size = sizeof(struct udp_sock), 1491 .h.udp_hash = udp_hash, 1492 #ifdef CONFIG_COMPAT 1493 .compat_setsockopt = compat_udp_setsockopt, 1494 .compat_getsockopt = compat_udp_getsockopt, 1495 #endif 1496 }; 1497 1498 /* ------------------------------------------------------------------------ */ 1499 #ifdef CONFIG_PROC_FS 1500 1501 static struct sock *udp_get_first(struct seq_file *seq) 1502 { 1503 struct sock *sk; 1504 struct udp_iter_state *state = seq->private; 1505 struct net *net = seq_file_net(seq); 1506 1507 for (state->bucket = 0; state->bucket < UDP_HTABLE_SIZE; ++state->bucket) { 1508 struct hlist_node *node; 1509 sk_for_each(sk, node, state->hashtable + state->bucket) { 1510 if (!net_eq(sock_net(sk), net)) 1511 continue; 1512 if (sk->sk_family == state->family) 1513 goto found; 1514 } 1515 } 1516 sk = NULL; 1517 found: 1518 return sk; 1519 } 1520 1521 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) 1522 { 1523 struct udp_iter_state *state = seq->private; 1524 struct net *net = seq_file_net(seq); 1525 1526 do { 1527 sk = sk_next(sk); 1528 try_again: 1529 ; 1530 } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family)); 1531 1532 if (!sk && ++state->bucket < UDP_HTABLE_SIZE) { 1533 sk = sk_head(state->hashtable + state->bucket); 1534 goto try_again; 1535 } 1536 return sk; 1537 } 1538 1539 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) 1540 { 1541 struct sock *sk = udp_get_first(seq); 1542 1543 if (sk) 1544 while (pos && (sk = udp_get_next(seq, sk)) != NULL) 1545 --pos; 1546 return pos ? NULL : sk; 1547 } 1548 1549 static void *udp_seq_start(struct seq_file *seq, loff_t *pos) 1550 __acquires(udp_hash_lock) 1551 { 1552 read_lock(&udp_hash_lock); 1553 return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN; 1554 } 1555 1556 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 1557 { 1558 struct sock *sk; 1559 1560 if (v == SEQ_START_TOKEN) 1561 sk = udp_get_idx(seq, 0); 1562 else 1563 sk = udp_get_next(seq, v); 1564 1565 ++*pos; 1566 return sk; 1567 } 1568 1569 static void udp_seq_stop(struct seq_file *seq, void *v) 1570 __releases(udp_hash_lock) 1571 { 1572 read_unlock(&udp_hash_lock); 1573 } 1574 1575 static int udp_seq_open(struct inode *inode, struct file *file) 1576 { 1577 struct udp_seq_afinfo *afinfo = PDE(inode)->data; 1578 struct udp_iter_state *s; 1579 int err; 1580 1581 err = seq_open_net(inode, file, &afinfo->seq_ops, 1582 sizeof(struct udp_iter_state)); 1583 if (err < 0) 1584 return err; 1585 1586 s = ((struct seq_file *)file->private_data)->private; 1587 s->family = afinfo->family; 1588 s->hashtable = afinfo->hashtable; 1589 return err; 1590 } 1591 1592 /* ------------------------------------------------------------------------ */ 1593 int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo) 1594 { 1595 struct proc_dir_entry *p; 1596 int rc = 0; 1597 1598 afinfo->seq_fops.open = udp_seq_open; 1599 afinfo->seq_fops.read = seq_read; 1600 afinfo->seq_fops.llseek = seq_lseek; 1601 afinfo->seq_fops.release = seq_release_net; 1602 1603 afinfo->seq_ops.start = udp_seq_start; 1604 afinfo->seq_ops.next = udp_seq_next; 1605 afinfo->seq_ops.stop = udp_seq_stop; 1606 1607 p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net, 1608 &afinfo->seq_fops, afinfo); 1609 if (!p) 1610 rc = -ENOMEM; 1611 return rc; 1612 } 1613 1614 void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo) 1615 { 1616 proc_net_remove(net, afinfo->name); 1617 } 1618 1619 /* ------------------------------------------------------------------------ */ 1620 static void udp4_format_sock(struct sock *sp, struct seq_file *f, 1621 int bucket, int *len) 1622 { 1623 struct inet_sock *inet = inet_sk(sp); 1624 __be32 dest = inet->daddr; 1625 __be32 src = inet->rcv_saddr; 1626 __u16 destp = ntohs(inet->dport); 1627 __u16 srcp = ntohs(inet->sport); 1628 1629 seq_printf(f, "%4d: %08X:%04X %08X:%04X" 1630 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p %d%n", 1631 bucket, src, srcp, dest, destp, sp->sk_state, 1632 atomic_read(&sp->sk_wmem_alloc), 1633 atomic_read(&sp->sk_rmem_alloc), 1634 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp), 1635 atomic_read(&sp->sk_refcnt), sp, 1636 atomic_read(&sp->sk_drops), len); 1637 } 1638 1639 int udp4_seq_show(struct seq_file *seq, void *v) 1640 { 1641 if (v == SEQ_START_TOKEN) 1642 seq_printf(seq, "%-127s\n", 1643 " sl local_address rem_address st tx_queue " 1644 "rx_queue tr tm->when retrnsmt uid timeout " 1645 "inode ref pointer drops"); 1646 else { 1647 struct udp_iter_state *state = seq->private; 1648 int len; 1649 1650 udp4_format_sock(v, seq, state->bucket, &len); 1651 seq_printf(seq, "%*s\n", 127 - len ,""); 1652 } 1653 return 0; 1654 } 1655 1656 /* ------------------------------------------------------------------------ */ 1657 static struct udp_seq_afinfo udp4_seq_afinfo = { 1658 .name = "udp", 1659 .family = AF_INET, 1660 .hashtable = udp_hash, 1661 .seq_fops = { 1662 .owner = THIS_MODULE, 1663 }, 1664 .seq_ops = { 1665 .show = udp4_seq_show, 1666 }, 1667 }; 1668 1669 static int udp4_proc_init_net(struct net *net) 1670 { 1671 return udp_proc_register(net, &udp4_seq_afinfo); 1672 } 1673 1674 static void udp4_proc_exit_net(struct net *net) 1675 { 1676 udp_proc_unregister(net, &udp4_seq_afinfo); 1677 } 1678 1679 static struct pernet_operations udp4_net_ops = { 1680 .init = udp4_proc_init_net, 1681 .exit = udp4_proc_exit_net, 1682 }; 1683 1684 int __init udp4_proc_init(void) 1685 { 1686 return register_pernet_subsys(&udp4_net_ops); 1687 } 1688 1689 void udp4_proc_exit(void) 1690 { 1691 unregister_pernet_subsys(&udp4_net_ops); 1692 } 1693 #endif /* CONFIG_PROC_FS */ 1694 1695 void __init udp_init(void) 1696 { 1697 unsigned long limit; 1698 1699 /* Set the pressure threshold up by the same strategy of TCP. It is a 1700 * fraction of global memory that is up to 1/2 at 256 MB, decreasing 1701 * toward zero with the amount of memory, with a floor of 128 pages. 1702 */ 1703 limit = min(nr_all_pages, 1UL<<(28-PAGE_SHIFT)) >> (20-PAGE_SHIFT); 1704 limit = (limit * (nr_all_pages >> (20-PAGE_SHIFT))) >> (PAGE_SHIFT-11); 1705 limit = max(limit, 128UL); 1706 sysctl_udp_mem[0] = limit / 4 * 3; 1707 sysctl_udp_mem[1] = limit; 1708 sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2; 1709 1710 sysctl_udp_rmem_min = SK_MEM_QUANTUM; 1711 sysctl_udp_wmem_min = SK_MEM_QUANTUM; 1712 } 1713 1714 EXPORT_SYMBOL(udp_disconnect); 1715 EXPORT_SYMBOL(udp_hash); 1716 EXPORT_SYMBOL(udp_hash_lock); 1717 EXPORT_SYMBOL(udp_ioctl); 1718 EXPORT_SYMBOL(udp_prot); 1719 EXPORT_SYMBOL(udp_sendmsg); 1720 EXPORT_SYMBOL(udp_lib_getsockopt); 1721 EXPORT_SYMBOL(udp_lib_setsockopt); 1722 EXPORT_SYMBOL(udp_poll); 1723 EXPORT_SYMBOL(udp_lib_get_port); 1724 1725 #ifdef CONFIG_PROC_FS 1726 EXPORT_SYMBOL(udp_proc_register); 1727 EXPORT_SYMBOL(udp_proc_unregister); 1728 #endif 1729