1 /* 2 * INET An implementation of the TCP/IP protocol suite for the LINUX 3 * operating system. INET is implemented using the BSD Socket 4 * interface as the means of communication with the user level. 5 * 6 * The User Datagram Protocol (UDP). 7 * 8 * Version: $Id: udp.c,v 1.102 2002/02/01 22:01:04 davem Exp $ 9 * 10 * Authors: Ross Biro 11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> 12 * Arnt Gulbrandsen, <agulbra@nvg.unit.no> 13 * Alan Cox, <Alan.Cox@linux.org> 14 * Hirokazu Takahashi, <taka@valinux.co.jp> 15 * 16 * Fixes: 17 * Alan Cox : verify_area() calls 18 * Alan Cox : stopped close while in use off icmp 19 * messages. Not a fix but a botch that 20 * for udp at least is 'valid'. 21 * Alan Cox : Fixed icmp handling properly 22 * Alan Cox : Correct error for oversized datagrams 23 * Alan Cox : Tidied select() semantics. 24 * Alan Cox : udp_err() fixed properly, also now 25 * select and read wake correctly on errors 26 * Alan Cox : udp_send verify_area moved to avoid mem leak 27 * Alan Cox : UDP can count its memory 28 * Alan Cox : send to an unknown connection causes 29 * an ECONNREFUSED off the icmp, but 30 * does NOT close. 31 * Alan Cox : Switched to new sk_buff handlers. No more backlog! 32 * Alan Cox : Using generic datagram code. Even smaller and the PEEK 33 * bug no longer crashes it. 34 * Fred Van Kempen : Net2e support for sk->broadcast. 35 * Alan Cox : Uses skb_free_datagram 36 * Alan Cox : Added get/set sockopt support. 37 * Alan Cox : Broadcasting without option set returns EACCES. 38 * Alan Cox : No wakeup calls. Instead we now use the callbacks. 39 * Alan Cox : Use ip_tos and ip_ttl 40 * Alan Cox : SNMP Mibs 41 * Alan Cox : MSG_DONTROUTE, and 0.0.0.0 support. 42 * Matt Dillon : UDP length checks. 43 * Alan Cox : Smarter af_inet used properly. 44 * Alan Cox : Use new kernel side addressing. 45 * Alan Cox : Incorrect return on truncated datagram receive. 46 * Arnt Gulbrandsen : New udp_send and stuff 47 * Alan Cox : Cache last socket 48 * Alan Cox : Route cache 49 * Jon Peatfield : Minor efficiency fix to sendto(). 50 * Mike Shaver : RFC1122 checks. 51 * Alan Cox : Nonblocking error fix. 52 * Willy Konynenberg : Transparent proxying support. 53 * Mike McLagan : Routing by source 54 * David S. Miller : New socket lookup architecture. 55 * Last socket cache retained as it 56 * does have a high hit rate. 57 * Olaf Kirch : Don't linearise iovec on sendmsg. 58 * Andi Kleen : Some cleanups, cache destination entry 59 * for connect. 60 * Vitaly E. Lavrov : Transparent proxy revived after year coma. 61 * Melvin Smith : Check msg_name not msg_namelen in sendto(), 62 * return ENOTCONN for unconnected sockets (POSIX) 63 * Janos Farkas : don't deliver multi/broadcasts to a different 64 * bound-to-device socket 65 * Hirokazu Takahashi : HW checksumming for outgoing UDP 66 * datagrams. 67 * Hirokazu Takahashi : sendfile() on UDP works now. 68 * Arnaldo C. Melo : convert /proc/net/udp to seq_file 69 * YOSHIFUJI Hideaki @USAGI and: Support IPV6_V6ONLY socket option, which 70 * Alexey Kuznetsov: allow both IPv4 and IPv6 sockets to bind 71 * a single port at the same time. 72 * Derek Atkins <derek@ihtfp.com>: Add Encapulation Support 73 * 74 * 75 * This program is free software; you can redistribute it and/or 76 * modify it under the terms of the GNU General Public License 77 * as published by the Free Software Foundation; either version 78 * 2 of the License, or (at your option) any later version. 79 */ 80 81 #include <asm/system.h> 82 #include <asm/uaccess.h> 83 #include <asm/ioctls.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/icmp.h> 101 #include <net/route.h> 102 #include <net/checksum.h> 103 #include <net/xfrm.h> 104 #include "udp_impl.h" 105 106 /* 107 * Snmp MIB for the UDP layer 108 */ 109 110 DEFINE_SNMP_STAT(struct udp_mib, udp_statistics) __read_mostly; 111 112 struct hlist_head udp_hash[UDP_HTABLE_SIZE]; 113 DEFINE_RWLOCK(udp_hash_lock); 114 115 static int udp_port_rover; 116 117 /* 118 * Note about this hash function : 119 * Typical use is probably daddr = 0, only dport is going to vary hash 120 */ 121 static inline unsigned int hash_port_and_addr(__u16 port, __be32 addr) 122 { 123 addr ^= addr >> 16; 124 addr ^= addr >> 8; 125 return port ^ addr; 126 } 127 128 static inline int __udp_lib_port_inuse(unsigned int hash, int port, 129 __be32 daddr, struct hlist_head udptable[]) 130 { 131 struct sock *sk; 132 struct hlist_node *node; 133 struct inet_sock *inet; 134 135 sk_for_each(sk, node, &udptable[hash & (UDP_HTABLE_SIZE - 1)]) { 136 if (sk->sk_hash != hash) 137 continue; 138 inet = inet_sk(sk); 139 if (inet->num != port) 140 continue; 141 if (inet->rcv_saddr == daddr) 142 return 1; 143 } 144 return 0; 145 } 146 147 /** 148 * __udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6 149 * 150 * @sk: socket struct in question 151 * @snum: port number to look up 152 * @udptable: hash list table, must be of UDP_HTABLE_SIZE 153 * @port_rover: pointer to record of last unallocated port 154 * @saddr_comp: AF-dependent comparison of bound local IP addresses 155 */ 156 int __udp_lib_get_port(struct sock *sk, unsigned short snum, 157 struct hlist_head udptable[], int *port_rover, 158 int (*saddr_comp)(const struct sock *sk1, 159 const struct sock *sk2 ) ) 160 { 161 struct hlist_node *node; 162 struct hlist_head *head; 163 struct sock *sk2; 164 unsigned int hash; 165 int error = 1; 166 167 write_lock_bh(&udp_hash_lock); 168 if (snum == 0) { 169 int best_size_so_far, best, result, i; 170 171 if (*port_rover > sysctl_local_port_range[1] || 172 *port_rover < sysctl_local_port_range[0]) 173 *port_rover = sysctl_local_port_range[0]; 174 best_size_so_far = 32767; 175 best = result = *port_rover; 176 for (i = 0; i < UDP_HTABLE_SIZE; i++, result++) { 177 int size; 178 179 hash = hash_port_and_addr(result, 180 inet_sk(sk)->rcv_saddr); 181 head = &udptable[hash & (UDP_HTABLE_SIZE - 1)]; 182 if (hlist_empty(head)) { 183 if (result > sysctl_local_port_range[1]) 184 result = sysctl_local_port_range[0] + 185 ((result - sysctl_local_port_range[0]) & 186 (UDP_HTABLE_SIZE - 1)); 187 goto gotit; 188 } 189 size = 0; 190 sk_for_each(sk2, node, head) { 191 if (++size >= best_size_so_far) 192 goto next; 193 } 194 best_size_so_far = size; 195 best = result; 196 next: 197 ; 198 } 199 result = best; 200 for (i = 0; i < (1 << 16) / UDP_HTABLE_SIZE; 201 i++, result += UDP_HTABLE_SIZE) { 202 if (result > sysctl_local_port_range[1]) 203 result = sysctl_local_port_range[0] 204 + ((result - sysctl_local_port_range[0]) & 205 (UDP_HTABLE_SIZE - 1)); 206 hash = hash_port_and_addr(result, 0); 207 if (__udp_lib_port_inuse(hash, result, 208 0, udptable)) 209 continue; 210 if (!inet_sk(sk)->rcv_saddr) 211 break; 212 213 hash = hash_port_and_addr(result, 214 inet_sk(sk)->rcv_saddr); 215 if (! __udp_lib_port_inuse(hash, result, 216 inet_sk(sk)->rcv_saddr, udptable)) 217 break; 218 } 219 if (i >= (1 << 16) / UDP_HTABLE_SIZE) 220 goto fail; 221 gotit: 222 *port_rover = snum = result; 223 } else { 224 hash = hash_port_and_addr(snum, 0); 225 head = &udptable[hash & (UDP_HTABLE_SIZE - 1)]; 226 227 sk_for_each(sk2, node, head) 228 if (sk2->sk_hash == hash && 229 sk2 != sk && 230 inet_sk(sk2)->num == snum && 231 (!sk2->sk_reuse || !sk->sk_reuse) && 232 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if || 233 sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 234 (*saddr_comp)(sk, sk2)) 235 goto fail; 236 237 if (inet_sk(sk)->rcv_saddr) { 238 hash = hash_port_and_addr(snum, 239 inet_sk(sk)->rcv_saddr); 240 head = &udptable[hash & (UDP_HTABLE_SIZE - 1)]; 241 242 sk_for_each(sk2, node, head) 243 if (sk2->sk_hash == hash && 244 sk2 != sk && 245 inet_sk(sk2)->num == snum && 246 (!sk2->sk_reuse || !sk->sk_reuse) && 247 (!sk2->sk_bound_dev_if || 248 !sk->sk_bound_dev_if || 249 sk2->sk_bound_dev_if == 250 sk->sk_bound_dev_if) && 251 (*saddr_comp)(sk, sk2)) 252 goto fail; 253 } 254 } 255 inet_sk(sk)->num = snum; 256 sk->sk_hash = hash; 257 if (sk_unhashed(sk)) { 258 head = &udptable[hash & (UDP_HTABLE_SIZE - 1)]; 259 sk_add_node(sk, head); 260 sock_prot_inc_use(sk->sk_prot); 261 } 262 error = 0; 263 fail: 264 write_unlock_bh(&udp_hash_lock); 265 return error; 266 } 267 268 int udp_get_port(struct sock *sk, unsigned short snum, 269 int (*scmp)(const struct sock *, const struct sock *)) 270 { 271 return __udp_lib_get_port(sk, snum, udp_hash, &udp_port_rover, scmp); 272 } 273 274 int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2) 275 { 276 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2); 277 278 return ( !ipv6_only_sock(sk2) && 279 (!inet1->rcv_saddr || !inet2->rcv_saddr || 280 inet1->rcv_saddr == inet2->rcv_saddr )); 281 } 282 283 static inline int udp_v4_get_port(struct sock *sk, unsigned short snum) 284 { 285 return udp_get_port(sk, snum, ipv4_rcv_saddr_equal); 286 } 287 288 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try 289 * harder than this. -DaveM 290 */ 291 static struct sock *__udp4_lib_lookup(__be32 saddr, __be16 sport, 292 __be32 daddr, __be16 dport, 293 int dif, struct hlist_head udptable[]) 294 { 295 struct sock *sk, *result = NULL; 296 struct hlist_node *node; 297 unsigned int hash, hashwild; 298 int score, best = -1, hport = ntohs(dport); 299 300 hash = hash_port_and_addr(hport, daddr); 301 hashwild = hash_port_and_addr(hport, 0); 302 303 read_lock(&udp_hash_lock); 304 305 lookup: 306 307 sk_for_each(sk, node, &udptable[hash & (UDP_HTABLE_SIZE - 1)]) { 308 struct inet_sock *inet = inet_sk(sk); 309 310 if (sk->sk_hash != hash || ipv6_only_sock(sk) || 311 inet->num != hport) 312 continue; 313 314 score = (sk->sk_family == PF_INET ? 1 : 0); 315 if (inet->rcv_saddr) { 316 if (inet->rcv_saddr != daddr) 317 continue; 318 score+=2; 319 } 320 if (inet->daddr) { 321 if (inet->daddr != saddr) 322 continue; 323 score+=2; 324 } 325 if (inet->dport) { 326 if (inet->dport != sport) 327 continue; 328 score+=2; 329 } 330 if (sk->sk_bound_dev_if) { 331 if (sk->sk_bound_dev_if != dif) 332 continue; 333 score+=2; 334 } 335 if (score == 9) { 336 result = sk; 337 goto found; 338 } else if (score > best) { 339 result = sk; 340 best = score; 341 } 342 } 343 344 if (hash != hashwild) { 345 hash = hashwild; 346 goto lookup; 347 } 348 found: 349 if (result) 350 sock_hold(result); 351 read_unlock(&udp_hash_lock); 352 return result; 353 } 354 355 static inline struct sock *udp_v4_mcast_next(struct sock *sk, unsigned int hnum, 356 int hport, __be32 loc_addr, 357 __be16 rmt_port, __be32 rmt_addr, 358 int dif) 359 { 360 struct hlist_node *node; 361 struct sock *s = sk; 362 363 sk_for_each_from(s, node) { 364 struct inet_sock *inet = inet_sk(s); 365 366 if (s->sk_hash != hnum || 367 inet->num != hport || 368 (inet->daddr && inet->daddr != rmt_addr) || 369 (inet->dport != rmt_port && inet->dport) || 370 (inet->rcv_saddr && inet->rcv_saddr != loc_addr) || 371 ipv6_only_sock(s) || 372 (s->sk_bound_dev_if && s->sk_bound_dev_if != dif)) 373 continue; 374 if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif)) 375 continue; 376 goto found; 377 } 378 s = NULL; 379 found: 380 return s; 381 } 382 383 /* 384 * This routine is called by the ICMP module when it gets some 385 * sort of error condition. If err < 0 then the socket should 386 * be closed and the error returned to the user. If err > 0 387 * it's just the icmp type << 8 | icmp code. 388 * Header points to the ip header of the error packet. We move 389 * on past this. Then (as it used to claim before adjustment) 390 * header points to the first 8 bytes of the udp header. We need 391 * to find the appropriate port. 392 */ 393 394 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct hlist_head udptable[]) 395 { 396 struct inet_sock *inet; 397 struct iphdr *iph = (struct iphdr*)skb->data; 398 struct udphdr *uh = (struct udphdr*)(skb->data+(iph->ihl<<2)); 399 const int type = icmp_hdr(skb)->type; 400 const int code = icmp_hdr(skb)->code; 401 struct sock *sk; 402 int harderr; 403 int err; 404 405 sk = __udp4_lib_lookup(iph->daddr, uh->dest, iph->saddr, uh->source, 406 skb->dev->ifindex, udptable ); 407 if (sk == NULL) { 408 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS); 409 return; /* No socket for error */ 410 } 411 412 err = 0; 413 harderr = 0; 414 inet = inet_sk(sk); 415 416 switch (type) { 417 default: 418 case ICMP_TIME_EXCEEDED: 419 err = EHOSTUNREACH; 420 break; 421 case ICMP_SOURCE_QUENCH: 422 goto out; 423 case ICMP_PARAMETERPROB: 424 err = EPROTO; 425 harderr = 1; 426 break; 427 case ICMP_DEST_UNREACH: 428 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */ 429 if (inet->pmtudisc != IP_PMTUDISC_DONT) { 430 err = EMSGSIZE; 431 harderr = 1; 432 break; 433 } 434 goto out; 435 } 436 err = EHOSTUNREACH; 437 if (code <= NR_ICMP_UNREACH) { 438 harderr = icmp_err_convert[code].fatal; 439 err = icmp_err_convert[code].errno; 440 } 441 break; 442 } 443 444 /* 445 * RFC1122: OK. Passes ICMP errors back to application, as per 446 * 4.1.3.3. 447 */ 448 if (!inet->recverr) { 449 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 450 goto out; 451 } else { 452 ip_icmp_error(sk, skb, err, uh->dest, info, (u8*)(uh+1)); 453 } 454 sk->sk_err = err; 455 sk->sk_error_report(sk); 456 out: 457 sock_put(sk); 458 } 459 460 void udp_err(struct sk_buff *skb, u32 info) 461 { 462 return __udp4_lib_err(skb, info, udp_hash); 463 } 464 465 /* 466 * Throw away all pending data and cancel the corking. Socket is locked. 467 */ 468 static void udp_flush_pending_frames(struct sock *sk) 469 { 470 struct udp_sock *up = udp_sk(sk); 471 472 if (up->pending) { 473 up->len = 0; 474 up->pending = 0; 475 ip_flush_pending_frames(sk); 476 } 477 } 478 479 /** 480 * udp4_hwcsum_outgoing - handle outgoing HW checksumming 481 * @sk: socket we are sending on 482 * @skb: sk_buff containing the filled-in UDP header 483 * (checksum field must be zeroed out) 484 */ 485 static void udp4_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb, 486 __be32 src, __be32 dst, int len ) 487 { 488 unsigned int offset; 489 struct udphdr *uh = udp_hdr(skb); 490 __wsum csum = 0; 491 492 if (skb_queue_len(&sk->sk_write_queue) == 1) { 493 /* 494 * Only one fragment on the socket. 495 */ 496 skb->csum_start = skb_transport_header(skb) - skb->head; 497 skb->csum_offset = offsetof(struct udphdr, check); 498 uh->check = ~csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, 0); 499 } else { 500 /* 501 * HW-checksum won't work as there are two or more 502 * fragments on the socket so that all csums of sk_buffs 503 * should be together 504 */ 505 offset = skb_transport_offset(skb); 506 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0); 507 508 skb->ip_summed = CHECKSUM_NONE; 509 510 skb_queue_walk(&sk->sk_write_queue, skb) { 511 csum = csum_add(csum, skb->csum); 512 } 513 514 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); 515 if (uh->check == 0) 516 uh->check = CSUM_MANGLED_0; 517 } 518 } 519 520 /* 521 * Push out all pending data as one UDP datagram. Socket is locked. 522 */ 523 static int udp_push_pending_frames(struct sock *sk) 524 { 525 struct udp_sock *up = udp_sk(sk); 526 struct inet_sock *inet = inet_sk(sk); 527 struct flowi *fl = &inet->cork.fl; 528 struct sk_buff *skb; 529 struct udphdr *uh; 530 int err = 0; 531 __wsum csum = 0; 532 533 /* Grab the skbuff where UDP header space exists. */ 534 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL) 535 goto out; 536 537 /* 538 * Create a UDP header 539 */ 540 uh = udp_hdr(skb); 541 uh->source = fl->fl_ip_sport; 542 uh->dest = fl->fl_ip_dport; 543 uh->len = htons(up->len); 544 uh->check = 0; 545 546 if (up->pcflag) /* UDP-Lite */ 547 csum = udplite_csum_outgoing(sk, skb); 548 549 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */ 550 551 skb->ip_summed = CHECKSUM_NONE; 552 goto send; 553 554 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 555 556 udp4_hwcsum_outgoing(sk, skb, fl->fl4_src,fl->fl4_dst, up->len); 557 goto send; 558 559 } else /* `normal' UDP */ 560 csum = udp_csum_outgoing(sk, skb); 561 562 /* add protocol-dependent pseudo-header */ 563 uh->check = csum_tcpudp_magic(fl->fl4_src, fl->fl4_dst, up->len, 564 sk->sk_protocol, csum ); 565 if (uh->check == 0) 566 uh->check = CSUM_MANGLED_0; 567 568 send: 569 err = ip_push_pending_frames(sk); 570 out: 571 up->len = 0; 572 up->pending = 0; 573 return err; 574 } 575 576 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 577 size_t len) 578 { 579 struct inet_sock *inet = inet_sk(sk); 580 struct udp_sock *up = udp_sk(sk); 581 int ulen = len; 582 struct ipcm_cookie ipc; 583 struct rtable *rt = NULL; 584 int free = 0; 585 int connected = 0; 586 __be32 daddr, faddr, saddr; 587 __be16 dport; 588 u8 tos; 589 int err, is_udplite = up->pcflag; 590 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; 591 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 592 593 if (len > 0xFFFF) 594 return -EMSGSIZE; 595 596 /* 597 * Check the flags. 598 */ 599 600 if (msg->msg_flags&MSG_OOB) /* Mirror BSD error message compatibility */ 601 return -EOPNOTSUPP; 602 603 ipc.opt = NULL; 604 605 if (up->pending) { 606 /* 607 * There are pending frames. 608 * The socket lock must be held while it's corked. 609 */ 610 lock_sock(sk); 611 if (likely(up->pending)) { 612 if (unlikely(up->pending != AF_INET)) { 613 release_sock(sk); 614 return -EINVAL; 615 } 616 goto do_append_data; 617 } 618 release_sock(sk); 619 } 620 ulen += sizeof(struct udphdr); 621 622 /* 623 * Get and verify the address. 624 */ 625 if (msg->msg_name) { 626 struct sockaddr_in * usin = (struct sockaddr_in*)msg->msg_name; 627 if (msg->msg_namelen < sizeof(*usin)) 628 return -EINVAL; 629 if (usin->sin_family != AF_INET) { 630 if (usin->sin_family != AF_UNSPEC) 631 return -EAFNOSUPPORT; 632 } 633 634 daddr = usin->sin_addr.s_addr; 635 dport = usin->sin_port; 636 if (dport == 0) 637 return -EINVAL; 638 } else { 639 if (sk->sk_state != TCP_ESTABLISHED) 640 return -EDESTADDRREQ; 641 daddr = inet->daddr; 642 dport = inet->dport; 643 /* Open fast path for connected socket. 644 Route will not be used, if at least one option is set. 645 */ 646 connected = 1; 647 } 648 ipc.addr = inet->saddr; 649 650 ipc.oif = sk->sk_bound_dev_if; 651 if (msg->msg_controllen) { 652 err = ip_cmsg_send(msg, &ipc); 653 if (err) 654 return err; 655 if (ipc.opt) 656 free = 1; 657 connected = 0; 658 } 659 if (!ipc.opt) 660 ipc.opt = inet->opt; 661 662 saddr = ipc.addr; 663 ipc.addr = faddr = daddr; 664 665 if (ipc.opt && ipc.opt->srr) { 666 if (!daddr) 667 return -EINVAL; 668 faddr = ipc.opt->faddr; 669 connected = 0; 670 } 671 tos = RT_TOS(inet->tos); 672 if (sock_flag(sk, SOCK_LOCALROUTE) || 673 (msg->msg_flags & MSG_DONTROUTE) || 674 (ipc.opt && ipc.opt->is_strictroute)) { 675 tos |= RTO_ONLINK; 676 connected = 0; 677 } 678 679 if (MULTICAST(daddr)) { 680 if (!ipc.oif) 681 ipc.oif = inet->mc_index; 682 if (!saddr) 683 saddr = inet->mc_addr; 684 connected = 0; 685 } 686 687 if (connected) 688 rt = (struct rtable*)sk_dst_check(sk, 0); 689 690 if (rt == NULL) { 691 struct flowi fl = { .oif = ipc.oif, 692 .nl_u = { .ip4_u = 693 { .daddr = faddr, 694 .saddr = saddr, 695 .tos = tos } }, 696 .proto = sk->sk_protocol, 697 .uli_u = { .ports = 698 { .sport = inet->sport, 699 .dport = dport } } }; 700 security_sk_classify_flow(sk, &fl); 701 err = ip_route_output_flow(&rt, &fl, sk, 1); 702 if (err) 703 goto out; 704 705 err = -EACCES; 706 if ((rt->rt_flags & RTCF_BROADCAST) && 707 !sock_flag(sk, SOCK_BROADCAST)) 708 goto out; 709 if (connected) 710 sk_dst_set(sk, dst_clone(&rt->u.dst)); 711 } 712 713 if (msg->msg_flags&MSG_CONFIRM) 714 goto do_confirm; 715 back_from_confirm: 716 717 saddr = rt->rt_src; 718 if (!ipc.addr) 719 daddr = ipc.addr = rt->rt_dst; 720 721 lock_sock(sk); 722 if (unlikely(up->pending)) { 723 /* The socket is already corked while preparing it. */ 724 /* ... which is an evident application bug. --ANK */ 725 release_sock(sk); 726 727 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n"); 728 err = -EINVAL; 729 goto out; 730 } 731 /* 732 * Now cork the socket to pend data. 733 */ 734 inet->cork.fl.fl4_dst = daddr; 735 inet->cork.fl.fl_ip_dport = dport; 736 inet->cork.fl.fl4_src = saddr; 737 inet->cork.fl.fl_ip_sport = inet->sport; 738 up->pending = AF_INET; 739 740 do_append_data: 741 up->len += ulen; 742 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 743 err = ip_append_data(sk, getfrag, msg->msg_iov, ulen, 744 sizeof(struct udphdr), &ipc, rt, 745 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 746 if (err) 747 udp_flush_pending_frames(sk); 748 else if (!corkreq) 749 err = udp_push_pending_frames(sk); 750 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 751 up->pending = 0; 752 release_sock(sk); 753 754 out: 755 ip_rt_put(rt); 756 if (free) 757 kfree(ipc.opt); 758 if (!err) { 759 UDP_INC_STATS_USER(UDP_MIB_OUTDATAGRAMS, is_udplite); 760 return len; 761 } 762 /* 763 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 764 * ENOBUFS might not be good (it's not tunable per se), but otherwise 765 * we don't have a good statistic (IpOutDiscards but it can be too many 766 * things). We could add another new stat but at least for now that 767 * seems like overkill. 768 */ 769 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 770 UDP_INC_STATS_USER(UDP_MIB_SNDBUFERRORS, is_udplite); 771 } 772 return err; 773 774 do_confirm: 775 dst_confirm(&rt->u.dst); 776 if (!(msg->msg_flags&MSG_PROBE) || len) 777 goto back_from_confirm; 778 err = 0; 779 goto out; 780 } 781 782 int udp_sendpage(struct sock *sk, struct page *page, int offset, 783 size_t size, int flags) 784 { 785 struct udp_sock *up = udp_sk(sk); 786 int ret; 787 788 if (!up->pending) { 789 struct msghdr msg = { .msg_flags = flags|MSG_MORE }; 790 791 /* Call udp_sendmsg to specify destination address which 792 * sendpage interface can't pass. 793 * This will succeed only when the socket is connected. 794 */ 795 ret = udp_sendmsg(NULL, sk, &msg, 0); 796 if (ret < 0) 797 return ret; 798 } 799 800 lock_sock(sk); 801 802 if (unlikely(!up->pending)) { 803 release_sock(sk); 804 805 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n"); 806 return -EINVAL; 807 } 808 809 ret = ip_append_page(sk, page, offset, size, flags); 810 if (ret == -EOPNOTSUPP) { 811 release_sock(sk); 812 return sock_no_sendpage(sk->sk_socket, page, offset, 813 size, flags); 814 } 815 if (ret < 0) { 816 udp_flush_pending_frames(sk); 817 goto out; 818 } 819 820 up->len += size; 821 if (!(up->corkflag || (flags&MSG_MORE))) 822 ret = udp_push_pending_frames(sk); 823 if (!ret) 824 ret = size; 825 out: 826 release_sock(sk); 827 return ret; 828 } 829 830 /* 831 * IOCTL requests applicable to the UDP protocol 832 */ 833 834 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg) 835 { 836 switch (cmd) { 837 case SIOCOUTQ: 838 { 839 int amount = atomic_read(&sk->sk_wmem_alloc); 840 return put_user(amount, (int __user *)arg); 841 } 842 843 case SIOCINQ: 844 { 845 struct sk_buff *skb; 846 unsigned long amount; 847 848 amount = 0; 849 spin_lock_bh(&sk->sk_receive_queue.lock); 850 skb = skb_peek(&sk->sk_receive_queue); 851 if (skb != NULL) { 852 /* 853 * We will only return the amount 854 * of this packet since that is all 855 * that will be read. 856 */ 857 amount = skb->len - sizeof(struct udphdr); 858 } 859 spin_unlock_bh(&sk->sk_receive_queue.lock); 860 return put_user(amount, (int __user *)arg); 861 } 862 863 default: 864 return -ENOIOCTLCMD; 865 } 866 867 return 0; 868 } 869 870 /* 871 * This should be easy, if there is something there we 872 * return it, otherwise we block. 873 */ 874 875 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 876 size_t len, int noblock, int flags, int *addr_len) 877 { 878 struct inet_sock *inet = inet_sk(sk); 879 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name; 880 struct sk_buff *skb; 881 unsigned int ulen, copied; 882 int err; 883 int is_udplite = IS_UDPLITE(sk); 884 885 /* 886 * Check any passed addresses 887 */ 888 if (addr_len) 889 *addr_len=sizeof(*sin); 890 891 if (flags & MSG_ERRQUEUE) 892 return ip_recv_error(sk, msg, len); 893 894 try_again: 895 skb = skb_recv_datagram(sk, flags, noblock, &err); 896 if (!skb) 897 goto out; 898 899 ulen = skb->len - sizeof(struct udphdr); 900 copied = len; 901 if (copied > ulen) 902 copied = ulen; 903 else if (copied < ulen) 904 msg->msg_flags |= MSG_TRUNC; 905 906 /* 907 * If checksum is needed at all, try to do it while copying the 908 * data. If the data is truncated, or if we only want a partial 909 * coverage checksum (UDP-Lite), do it before the copy. 910 */ 911 912 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) { 913 if (udp_lib_checksum_complete(skb)) 914 goto csum_copy_err; 915 } 916 917 if (skb_csum_unnecessary(skb)) 918 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr), 919 msg->msg_iov, copied ); 920 else { 921 err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov); 922 923 if (err == -EINVAL) 924 goto csum_copy_err; 925 } 926 927 if (err) 928 goto out_free; 929 930 sock_recv_timestamp(msg, sk, skb); 931 932 /* Copy the address. */ 933 if (sin) 934 { 935 sin->sin_family = AF_INET; 936 sin->sin_port = udp_hdr(skb)->source; 937 sin->sin_addr.s_addr = ip_hdr(skb)->saddr; 938 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 939 } 940 if (inet->cmsg_flags) 941 ip_cmsg_recv(msg, skb); 942 943 err = copied; 944 if (flags & MSG_TRUNC) 945 err = ulen; 946 947 out_free: 948 skb_free_datagram(sk, skb); 949 out: 950 return err; 951 952 csum_copy_err: 953 UDP_INC_STATS_BH(UDP_MIB_INERRORS, is_udplite); 954 955 skb_kill_datagram(sk, skb, flags); 956 957 if (noblock) 958 return -EAGAIN; 959 goto try_again; 960 } 961 962 963 int udp_disconnect(struct sock *sk, int flags) 964 { 965 struct inet_sock *inet = inet_sk(sk); 966 /* 967 * 1003.1g - break association. 968 */ 969 970 sk->sk_state = TCP_CLOSE; 971 inet->daddr = 0; 972 inet->dport = 0; 973 sk->sk_bound_dev_if = 0; 974 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) 975 inet_reset_saddr(sk); 976 977 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { 978 sk->sk_prot->unhash(sk); 979 inet->sport = 0; 980 } 981 sk_dst_reset(sk); 982 return 0; 983 } 984 985 /* return: 986 * 1 if the the UDP system should process it 987 * 0 if we should drop this packet 988 * -1 if it should get processed by xfrm4_rcv_encap 989 */ 990 static int udp_encap_rcv(struct sock * sk, struct sk_buff *skb) 991 { 992 #ifndef CONFIG_XFRM 993 return 1; 994 #else 995 struct udp_sock *up = udp_sk(sk); 996 struct udphdr *uh; 997 struct iphdr *iph; 998 int iphlen, len; 999 1000 __u8 *udpdata; 1001 __be32 *udpdata32; 1002 __u16 encap_type = up->encap_type; 1003 1004 /* if we're overly short, let UDP handle it */ 1005 len = skb->len - sizeof(struct udphdr); 1006 if (len <= 0) 1007 return 1; 1008 1009 /* if this is not encapsulated socket, then just return now */ 1010 if (!encap_type) 1011 return 1; 1012 1013 /* If this is a paged skb, make sure we pull up 1014 * whatever data we need to look at. */ 1015 if (!pskb_may_pull(skb, sizeof(struct udphdr) + min(len, 8))) 1016 return 1; 1017 1018 /* Now we can get the pointers */ 1019 uh = udp_hdr(skb); 1020 udpdata = (__u8 *)uh + sizeof(struct udphdr); 1021 udpdata32 = (__be32 *)udpdata; 1022 1023 switch (encap_type) { 1024 default: 1025 case UDP_ENCAP_ESPINUDP: 1026 /* Check if this is a keepalive packet. If so, eat it. */ 1027 if (len == 1 && udpdata[0] == 0xff) { 1028 return 0; 1029 } else if (len > sizeof(struct ip_esp_hdr) && udpdata32[0] != 0) { 1030 /* ESP Packet without Non-ESP header */ 1031 len = sizeof(struct udphdr); 1032 } else 1033 /* Must be an IKE packet.. pass it through */ 1034 return 1; 1035 break; 1036 case UDP_ENCAP_ESPINUDP_NON_IKE: 1037 /* Check if this is a keepalive packet. If so, eat it. */ 1038 if (len == 1 && udpdata[0] == 0xff) { 1039 return 0; 1040 } else if (len > 2 * sizeof(u32) + sizeof(struct ip_esp_hdr) && 1041 udpdata32[0] == 0 && udpdata32[1] == 0) { 1042 1043 /* ESP Packet with Non-IKE marker */ 1044 len = sizeof(struct udphdr) + 2 * sizeof(u32); 1045 } else 1046 /* Must be an IKE packet.. pass it through */ 1047 return 1; 1048 break; 1049 } 1050 1051 /* At this point we are sure that this is an ESPinUDP packet, 1052 * so we need to remove 'len' bytes from the packet (the UDP 1053 * header and optional ESP marker bytes) and then modify the 1054 * protocol to ESP, and then call into the transform receiver. 1055 */ 1056 if (skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC)) 1057 return 0; 1058 1059 /* Now we can update and verify the packet length... */ 1060 iph = ip_hdr(skb); 1061 iphlen = iph->ihl << 2; 1062 iph->tot_len = htons(ntohs(iph->tot_len) - len); 1063 if (skb->len < iphlen + len) { 1064 /* packet is too small!?! */ 1065 return 0; 1066 } 1067 1068 /* pull the data buffer up to the ESP header and set the 1069 * transport header to point to ESP. Keep UDP on the stack 1070 * for later. 1071 */ 1072 __skb_pull(skb, len); 1073 skb_reset_transport_header(skb); 1074 1075 /* modify the protocol (it's ESP!) */ 1076 iph->protocol = IPPROTO_ESP; 1077 1078 /* and let the caller know to send this into the ESP processor... */ 1079 return -1; 1080 #endif 1081 } 1082 1083 /* returns: 1084 * -1: error 1085 * 0: success 1086 * >0: "udp encap" protocol resubmission 1087 * 1088 * Note that in the success and error cases, the skb is assumed to 1089 * have either been requeued or freed. 1090 */ 1091 int udp_queue_rcv_skb(struct sock * sk, struct sk_buff *skb) 1092 { 1093 struct udp_sock *up = udp_sk(sk); 1094 int rc; 1095 1096 /* 1097 * Charge it to the socket, dropping if the queue is full. 1098 */ 1099 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) 1100 goto drop; 1101 nf_reset(skb); 1102 1103 if (up->encap_type) { 1104 /* 1105 * This is an encapsulation socket, so let's see if this is 1106 * an encapsulated packet. 1107 * If it's a keepalive packet, then just eat it. 1108 * If it's an encapsulateed packet, then pass it to the 1109 * IPsec xfrm input and return the response 1110 * appropriately. Otherwise, just fall through and 1111 * pass this up the UDP socket. 1112 */ 1113 int ret; 1114 1115 ret = udp_encap_rcv(sk, skb); 1116 if (ret == 0) { 1117 /* Eat the packet .. */ 1118 kfree_skb(skb); 1119 return 0; 1120 } 1121 if (ret < 0) { 1122 /* process the ESP packet */ 1123 ret = xfrm4_rcv_encap(skb, up->encap_type); 1124 UDP_INC_STATS_BH(UDP_MIB_INDATAGRAMS, up->pcflag); 1125 return -ret; 1126 } 1127 /* FALLTHROUGH -- it's a UDP Packet */ 1128 } 1129 1130 /* 1131 * UDP-Lite specific tests, ignored on UDP sockets 1132 */ 1133 if ((up->pcflag & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 1134 1135 /* 1136 * MIB statistics other than incrementing the error count are 1137 * disabled for the following two types of errors: these depend 1138 * on the application settings, not on the functioning of the 1139 * protocol stack as such. 1140 * 1141 * RFC 3828 here recommends (sec 3.3): "There should also be a 1142 * way ... to ... at least let the receiving application block 1143 * delivery of packets with coverage values less than a value 1144 * provided by the application." 1145 */ 1146 if (up->pcrlen == 0) { /* full coverage was set */ 1147 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage " 1148 "%d while full coverage %d requested\n", 1149 UDP_SKB_CB(skb)->cscov, skb->len); 1150 goto drop; 1151 } 1152 /* The next case involves violating the min. coverage requested 1153 * by the receiver. This is subtle: if receiver wants x and x is 1154 * greater than the buffersize/MTU then receiver will complain 1155 * that it wants x while sender emits packets of smaller size y. 1156 * Therefore the above ...()->partial_cov statement is essential. 1157 */ 1158 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 1159 LIMIT_NETDEBUG(KERN_WARNING 1160 "UDPLITE: coverage %d too small, need min %d\n", 1161 UDP_SKB_CB(skb)->cscov, up->pcrlen); 1162 goto drop; 1163 } 1164 } 1165 1166 if (sk->sk_filter) { 1167 if (udp_lib_checksum_complete(skb)) 1168 goto drop; 1169 } 1170 1171 if ((rc = sock_queue_rcv_skb(sk,skb)) < 0) { 1172 /* Note that an ENOMEM error is charged twice */ 1173 if (rc == -ENOMEM) 1174 UDP_INC_STATS_BH(UDP_MIB_RCVBUFERRORS, up->pcflag); 1175 goto drop; 1176 } 1177 1178 UDP_INC_STATS_BH(UDP_MIB_INDATAGRAMS, up->pcflag); 1179 return 0; 1180 1181 drop: 1182 UDP_INC_STATS_BH(UDP_MIB_INERRORS, up->pcflag); 1183 kfree_skb(skb); 1184 return -1; 1185 } 1186 1187 /* 1188 * Multicasts and broadcasts go to each listener. 1189 * 1190 * Note: called only from the BH handler context, 1191 * so we don't need to lock the hashes. 1192 */ 1193 static int __udp4_lib_mcast_deliver(struct sk_buff *skb, 1194 struct udphdr *uh, 1195 __be32 saddr, __be32 daddr, 1196 struct hlist_head udptable[]) 1197 { 1198 struct sock *sk, *skw, *sknext; 1199 int dif; 1200 int hport = ntohs(uh->dest); 1201 unsigned int hash = hash_port_and_addr(hport, daddr); 1202 unsigned int hashwild = hash_port_and_addr(hport, 0); 1203 1204 dif = skb->dev->ifindex; 1205 1206 read_lock(&udp_hash_lock); 1207 1208 sk = sk_head(&udptable[hash & (UDP_HTABLE_SIZE - 1)]); 1209 skw = sk_head(&udptable[hashwild & (UDP_HTABLE_SIZE - 1)]); 1210 1211 sk = udp_v4_mcast_next(sk, hash, hport, daddr, uh->source, saddr, dif); 1212 if (!sk) { 1213 hash = hashwild; 1214 sk = udp_v4_mcast_next(skw, hash, hport, daddr, uh->source, 1215 saddr, dif); 1216 } 1217 if (sk) { 1218 do { 1219 struct sk_buff *skb1 = skb; 1220 sknext = udp_v4_mcast_next(sk_next(sk), hash, hport, 1221 daddr, uh->source, saddr, dif); 1222 if (!sknext && hash != hashwild) { 1223 hash = hashwild; 1224 sknext = udp_v4_mcast_next(skw, hash, hport, 1225 daddr, uh->source, saddr, dif); 1226 } 1227 if (sknext) 1228 skb1 = skb_clone(skb, GFP_ATOMIC); 1229 1230 if (skb1) { 1231 int ret = udp_queue_rcv_skb(sk, skb1); 1232 if (ret > 0) 1233 /* 1234 * we should probably re-process 1235 * instead of dropping packets here. 1236 */ 1237 kfree_skb(skb1); 1238 } 1239 sk = sknext; 1240 } while (sknext); 1241 } else 1242 kfree_skb(skb); 1243 read_unlock(&udp_hash_lock); 1244 return 0; 1245 } 1246 1247 /* Initialize UDP checksum. If exited with zero value (success), 1248 * CHECKSUM_UNNECESSARY means, that no more checks are required. 1249 * Otherwise, csum completion requires chacksumming packet body, 1250 * including udp header and folding it to skb->csum. 1251 */ 1252 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, 1253 int proto) 1254 { 1255 const struct iphdr *iph; 1256 int err; 1257 1258 UDP_SKB_CB(skb)->partial_cov = 0; 1259 UDP_SKB_CB(skb)->cscov = skb->len; 1260 1261 if (proto == IPPROTO_UDPLITE) { 1262 err = udplite_checksum_init(skb, uh); 1263 if (err) 1264 return err; 1265 } 1266 1267 iph = ip_hdr(skb); 1268 if (uh->check == 0) { 1269 skb->ip_summed = CHECKSUM_UNNECESSARY; 1270 } else if (skb->ip_summed == CHECKSUM_COMPLETE) { 1271 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len, 1272 proto, skb->csum)) 1273 skb->ip_summed = CHECKSUM_UNNECESSARY; 1274 } 1275 if (!skb_csum_unnecessary(skb)) 1276 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr, 1277 skb->len, proto, 0); 1278 /* Probably, we should checksum udp header (it should be in cache 1279 * in any case) and data in tiny packets (< rx copybreak). 1280 */ 1281 1282 return 0; 1283 } 1284 1285 /* 1286 * All we need to do is get the socket, and then do a checksum. 1287 */ 1288 1289 int __udp4_lib_rcv(struct sk_buff *skb, struct hlist_head udptable[], 1290 int proto) 1291 { 1292 struct sock *sk; 1293 struct udphdr *uh = udp_hdr(skb); 1294 unsigned short ulen; 1295 struct rtable *rt = (struct rtable*)skb->dst; 1296 __be32 saddr = ip_hdr(skb)->saddr; 1297 __be32 daddr = ip_hdr(skb)->daddr; 1298 1299 /* 1300 * Validate the packet. 1301 */ 1302 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 1303 goto drop; /* No space for header. */ 1304 1305 ulen = ntohs(uh->len); 1306 if (ulen > skb->len) 1307 goto short_packet; 1308 1309 if (proto == IPPROTO_UDP) { 1310 /* UDP validates ulen. */ 1311 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) 1312 goto short_packet; 1313 uh = udp_hdr(skb); 1314 } 1315 1316 if (udp4_csum_init(skb, uh, proto)) 1317 goto csum_error; 1318 1319 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) 1320 return __udp4_lib_mcast_deliver(skb, uh, saddr, daddr, udptable); 1321 1322 sk = __udp4_lib_lookup(saddr, uh->source, daddr, uh->dest, 1323 skb->dev->ifindex, udptable); 1324 1325 if (sk != NULL) { 1326 int ret = udp_queue_rcv_skb(sk, skb); 1327 sock_put(sk); 1328 1329 /* a return value > 0 means to resubmit the input, but 1330 * it wants the return to be -protocol, or 0 1331 */ 1332 if (ret > 0) 1333 return -ret; 1334 return 0; 1335 } 1336 1337 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 1338 goto drop; 1339 nf_reset(skb); 1340 1341 /* No socket. Drop packet silently, if checksum is wrong */ 1342 if (udp_lib_checksum_complete(skb)) 1343 goto csum_error; 1344 1345 UDP_INC_STATS_BH(UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 1346 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); 1347 1348 /* 1349 * Hmm. We got an UDP packet to a port to which we 1350 * don't wanna listen. Ignore it. 1351 */ 1352 kfree_skb(skb); 1353 return 0; 1354 1355 short_packet: 1356 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %u.%u.%u.%u:%u %d/%d to %u.%u.%u.%u:%u\n", 1357 proto == IPPROTO_UDPLITE ? "-Lite" : "", 1358 NIPQUAD(saddr), 1359 ntohs(uh->source), 1360 ulen, 1361 skb->len, 1362 NIPQUAD(daddr), 1363 ntohs(uh->dest)); 1364 goto drop; 1365 1366 csum_error: 1367 /* 1368 * RFC1122: OK. Discards the bad packet silently (as far as 1369 * the network is concerned, anyway) as per 4.1.3.4 (MUST). 1370 */ 1371 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %d.%d.%d.%d:%d to %d.%d.%d.%d:%d ulen %d\n", 1372 proto == IPPROTO_UDPLITE ? "-Lite" : "", 1373 NIPQUAD(saddr), 1374 ntohs(uh->source), 1375 NIPQUAD(daddr), 1376 ntohs(uh->dest), 1377 ulen); 1378 drop: 1379 UDP_INC_STATS_BH(UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 1380 kfree_skb(skb); 1381 return 0; 1382 } 1383 1384 int udp_rcv(struct sk_buff *skb) 1385 { 1386 return __udp4_lib_rcv(skb, udp_hash, IPPROTO_UDP); 1387 } 1388 1389 int udp_destroy_sock(struct sock *sk) 1390 { 1391 lock_sock(sk); 1392 udp_flush_pending_frames(sk); 1393 release_sock(sk); 1394 return 0; 1395 } 1396 1397 /* 1398 * Socket option code for UDP 1399 */ 1400 int udp_lib_setsockopt(struct sock *sk, int level, int optname, 1401 char __user *optval, int optlen, 1402 int (*push_pending_frames)(struct sock *)) 1403 { 1404 struct udp_sock *up = udp_sk(sk); 1405 int val; 1406 int err = 0; 1407 1408 if (optlen<sizeof(int)) 1409 return -EINVAL; 1410 1411 if (get_user(val, (int __user *)optval)) 1412 return -EFAULT; 1413 1414 switch (optname) { 1415 case UDP_CORK: 1416 if (val != 0) { 1417 up->corkflag = 1; 1418 } else { 1419 up->corkflag = 0; 1420 lock_sock(sk); 1421 (*push_pending_frames)(sk); 1422 release_sock(sk); 1423 } 1424 break; 1425 1426 case UDP_ENCAP: 1427 switch (val) { 1428 case 0: 1429 case UDP_ENCAP_ESPINUDP: 1430 case UDP_ENCAP_ESPINUDP_NON_IKE: 1431 up->encap_type = val; 1432 break; 1433 default: 1434 err = -ENOPROTOOPT; 1435 break; 1436 } 1437 break; 1438 1439 /* 1440 * UDP-Lite's partial checksum coverage (RFC 3828). 1441 */ 1442 /* The sender sets actual checksum coverage length via this option. 1443 * The case coverage > packet length is handled by send module. */ 1444 case UDPLITE_SEND_CSCOV: 1445 if (!up->pcflag) /* Disable the option on UDP sockets */ 1446 return -ENOPROTOOPT; 1447 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ 1448 val = 8; 1449 up->pcslen = val; 1450 up->pcflag |= UDPLITE_SEND_CC; 1451 break; 1452 1453 /* The receiver specifies a minimum checksum coverage value. To make 1454 * sense, this should be set to at least 8 (as done below). If zero is 1455 * used, this again means full checksum coverage. */ 1456 case UDPLITE_RECV_CSCOV: 1457 if (!up->pcflag) /* Disable the option on UDP sockets */ 1458 return -ENOPROTOOPT; 1459 if (val != 0 && val < 8) /* Avoid silly minimal values. */ 1460 val = 8; 1461 up->pcrlen = val; 1462 up->pcflag |= UDPLITE_RECV_CC; 1463 break; 1464 1465 default: 1466 err = -ENOPROTOOPT; 1467 break; 1468 } 1469 1470 return err; 1471 } 1472 1473 int udp_setsockopt(struct sock *sk, int level, int optname, 1474 char __user *optval, int optlen) 1475 { 1476 if (level == SOL_UDP || level == SOL_UDPLITE) 1477 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1478 udp_push_pending_frames); 1479 return ip_setsockopt(sk, level, optname, optval, optlen); 1480 } 1481 1482 #ifdef CONFIG_COMPAT 1483 int compat_udp_setsockopt(struct sock *sk, int level, int optname, 1484 char __user *optval, int optlen) 1485 { 1486 if (level == SOL_UDP || level == SOL_UDPLITE) 1487 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1488 udp_push_pending_frames); 1489 return compat_ip_setsockopt(sk, level, optname, optval, optlen); 1490 } 1491 #endif 1492 1493 int udp_lib_getsockopt(struct sock *sk, int level, int optname, 1494 char __user *optval, int __user *optlen) 1495 { 1496 struct udp_sock *up = udp_sk(sk); 1497 int val, len; 1498 1499 if (get_user(len,optlen)) 1500 return -EFAULT; 1501 1502 len = min_t(unsigned int, len, sizeof(int)); 1503 1504 if (len < 0) 1505 return -EINVAL; 1506 1507 switch (optname) { 1508 case UDP_CORK: 1509 val = up->corkflag; 1510 break; 1511 1512 case UDP_ENCAP: 1513 val = up->encap_type; 1514 break; 1515 1516 /* The following two cannot be changed on UDP sockets, the return is 1517 * always 0 (which corresponds to the full checksum coverage of UDP). */ 1518 case UDPLITE_SEND_CSCOV: 1519 val = up->pcslen; 1520 break; 1521 1522 case UDPLITE_RECV_CSCOV: 1523 val = up->pcrlen; 1524 break; 1525 1526 default: 1527 return -ENOPROTOOPT; 1528 } 1529 1530 if (put_user(len, optlen)) 1531 return -EFAULT; 1532 if (copy_to_user(optval, &val,len)) 1533 return -EFAULT; 1534 return 0; 1535 } 1536 1537 int udp_getsockopt(struct sock *sk, int level, int optname, 1538 char __user *optval, int __user *optlen) 1539 { 1540 if (level == SOL_UDP || level == SOL_UDPLITE) 1541 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1542 return ip_getsockopt(sk, level, optname, optval, optlen); 1543 } 1544 1545 #ifdef CONFIG_COMPAT 1546 int compat_udp_getsockopt(struct sock *sk, int level, int optname, 1547 char __user *optval, int __user *optlen) 1548 { 1549 if (level == SOL_UDP || level == SOL_UDPLITE) 1550 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1551 return compat_ip_getsockopt(sk, level, optname, optval, optlen); 1552 } 1553 #endif 1554 /** 1555 * udp_poll - wait for a UDP event. 1556 * @file - file struct 1557 * @sock - socket 1558 * @wait - poll table 1559 * 1560 * This is same as datagram poll, except for the special case of 1561 * blocking sockets. If application is using a blocking fd 1562 * and a packet with checksum error is in the queue; 1563 * then it could get return from select indicating data available 1564 * but then block when reading it. Add special case code 1565 * to work around these arguably broken applications. 1566 */ 1567 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait) 1568 { 1569 unsigned int mask = datagram_poll(file, sock, wait); 1570 struct sock *sk = sock->sk; 1571 int is_lite = IS_UDPLITE(sk); 1572 1573 /* Check for false positives due to checksum errors */ 1574 if ( (mask & POLLRDNORM) && 1575 !(file->f_flags & O_NONBLOCK) && 1576 !(sk->sk_shutdown & RCV_SHUTDOWN)){ 1577 struct sk_buff_head *rcvq = &sk->sk_receive_queue; 1578 struct sk_buff *skb; 1579 1580 spin_lock_bh(&rcvq->lock); 1581 while ((skb = skb_peek(rcvq)) != NULL && 1582 udp_lib_checksum_complete(skb)) { 1583 UDP_INC_STATS_BH(UDP_MIB_INERRORS, is_lite); 1584 __skb_unlink(skb, rcvq); 1585 kfree_skb(skb); 1586 } 1587 spin_unlock_bh(&rcvq->lock); 1588 1589 /* nothing to see, move along */ 1590 if (skb == NULL) 1591 mask &= ~(POLLIN | POLLRDNORM); 1592 } 1593 1594 return mask; 1595 1596 } 1597 1598 struct proto udp_prot = { 1599 .name = "UDP", 1600 .owner = THIS_MODULE, 1601 .close = udp_lib_close, 1602 .connect = ip4_datagram_connect, 1603 .disconnect = udp_disconnect, 1604 .ioctl = udp_ioctl, 1605 .destroy = udp_destroy_sock, 1606 .setsockopt = udp_setsockopt, 1607 .getsockopt = udp_getsockopt, 1608 .sendmsg = udp_sendmsg, 1609 .recvmsg = udp_recvmsg, 1610 .sendpage = udp_sendpage, 1611 .backlog_rcv = udp_queue_rcv_skb, 1612 .hash = udp_lib_hash, 1613 .unhash = udp_lib_unhash, 1614 .get_port = udp_v4_get_port, 1615 .obj_size = sizeof(struct udp_sock), 1616 #ifdef CONFIG_COMPAT 1617 .compat_setsockopt = compat_udp_setsockopt, 1618 .compat_getsockopt = compat_udp_getsockopt, 1619 #endif 1620 }; 1621 1622 /* ------------------------------------------------------------------------ */ 1623 #ifdef CONFIG_PROC_FS 1624 1625 static struct sock *udp_get_first(struct seq_file *seq) 1626 { 1627 struct sock *sk; 1628 struct udp_iter_state *state = seq->private; 1629 1630 for (state->bucket = 0; state->bucket < UDP_HTABLE_SIZE; ++state->bucket) { 1631 struct hlist_node *node; 1632 sk_for_each(sk, node, state->hashtable + state->bucket) { 1633 if (sk->sk_family == state->family) 1634 goto found; 1635 } 1636 } 1637 sk = NULL; 1638 found: 1639 return sk; 1640 } 1641 1642 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) 1643 { 1644 struct udp_iter_state *state = seq->private; 1645 1646 do { 1647 sk = sk_next(sk); 1648 try_again: 1649 ; 1650 } while (sk && sk->sk_family != state->family); 1651 1652 if (!sk && ++state->bucket < UDP_HTABLE_SIZE) { 1653 sk = sk_head(state->hashtable + state->bucket); 1654 goto try_again; 1655 } 1656 return sk; 1657 } 1658 1659 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) 1660 { 1661 struct sock *sk = udp_get_first(seq); 1662 1663 if (sk) 1664 while (pos && (sk = udp_get_next(seq, sk)) != NULL) 1665 --pos; 1666 return pos ? NULL : sk; 1667 } 1668 1669 static void *udp_seq_start(struct seq_file *seq, loff_t *pos) 1670 { 1671 read_lock(&udp_hash_lock); 1672 return *pos ? udp_get_idx(seq, *pos-1) : (void *)1; 1673 } 1674 1675 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 1676 { 1677 struct sock *sk; 1678 1679 if (v == (void *)1) 1680 sk = udp_get_idx(seq, 0); 1681 else 1682 sk = udp_get_next(seq, v); 1683 1684 ++*pos; 1685 return sk; 1686 } 1687 1688 static void udp_seq_stop(struct seq_file *seq, void *v) 1689 { 1690 read_unlock(&udp_hash_lock); 1691 } 1692 1693 static int udp_seq_open(struct inode *inode, struct file *file) 1694 { 1695 struct udp_seq_afinfo *afinfo = PDE(inode)->data; 1696 struct seq_file *seq; 1697 int rc = -ENOMEM; 1698 struct udp_iter_state *s = kzalloc(sizeof(*s), GFP_KERNEL); 1699 1700 if (!s) 1701 goto out; 1702 s->family = afinfo->family; 1703 s->hashtable = afinfo->hashtable; 1704 s->seq_ops.start = udp_seq_start; 1705 s->seq_ops.next = udp_seq_next; 1706 s->seq_ops.show = afinfo->seq_show; 1707 s->seq_ops.stop = udp_seq_stop; 1708 1709 rc = seq_open(file, &s->seq_ops); 1710 if (rc) 1711 goto out_kfree; 1712 1713 seq = file->private_data; 1714 seq->private = s; 1715 out: 1716 return rc; 1717 out_kfree: 1718 kfree(s); 1719 goto out; 1720 } 1721 1722 /* ------------------------------------------------------------------------ */ 1723 int udp_proc_register(struct udp_seq_afinfo *afinfo) 1724 { 1725 struct proc_dir_entry *p; 1726 int rc = 0; 1727 1728 if (!afinfo) 1729 return -EINVAL; 1730 afinfo->seq_fops->owner = afinfo->owner; 1731 afinfo->seq_fops->open = udp_seq_open; 1732 afinfo->seq_fops->read = seq_read; 1733 afinfo->seq_fops->llseek = seq_lseek; 1734 afinfo->seq_fops->release = seq_release_private; 1735 1736 p = proc_net_fops_create(afinfo->name, S_IRUGO, afinfo->seq_fops); 1737 if (p) 1738 p->data = afinfo; 1739 else 1740 rc = -ENOMEM; 1741 return rc; 1742 } 1743 1744 void udp_proc_unregister(struct udp_seq_afinfo *afinfo) 1745 { 1746 if (!afinfo) 1747 return; 1748 proc_net_remove(afinfo->name); 1749 memset(afinfo->seq_fops, 0, sizeof(*afinfo->seq_fops)); 1750 } 1751 1752 /* ------------------------------------------------------------------------ */ 1753 static void udp4_format_sock(struct sock *sp, char *tmpbuf, int bucket) 1754 { 1755 struct inet_sock *inet = inet_sk(sp); 1756 __be32 dest = inet->daddr; 1757 __be32 src = inet->rcv_saddr; 1758 __u16 destp = ntohs(inet->dport); 1759 __u16 srcp = ntohs(inet->sport); 1760 1761 sprintf(tmpbuf, "%4d: %08X:%04X %08X:%04X" 1762 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p", 1763 bucket, src, srcp, dest, destp, sp->sk_state, 1764 atomic_read(&sp->sk_wmem_alloc), 1765 atomic_read(&sp->sk_rmem_alloc), 1766 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp), 1767 atomic_read(&sp->sk_refcnt), sp); 1768 } 1769 1770 int udp4_seq_show(struct seq_file *seq, void *v) 1771 { 1772 if (v == SEQ_START_TOKEN) 1773 seq_printf(seq, "%-127s\n", 1774 " sl local_address rem_address st tx_queue " 1775 "rx_queue tr tm->when retrnsmt uid timeout " 1776 "inode"); 1777 else { 1778 char tmpbuf[129]; 1779 struct udp_iter_state *state = seq->private; 1780 1781 udp4_format_sock(v, tmpbuf, state->bucket); 1782 seq_printf(seq, "%-127s\n", tmpbuf); 1783 } 1784 return 0; 1785 } 1786 1787 /* ------------------------------------------------------------------------ */ 1788 static struct file_operations udp4_seq_fops; 1789 static struct udp_seq_afinfo udp4_seq_afinfo = { 1790 .owner = THIS_MODULE, 1791 .name = "udp", 1792 .family = AF_INET, 1793 .hashtable = udp_hash, 1794 .seq_show = udp4_seq_show, 1795 .seq_fops = &udp4_seq_fops, 1796 }; 1797 1798 int __init udp4_proc_init(void) 1799 { 1800 return udp_proc_register(&udp4_seq_afinfo); 1801 } 1802 1803 void udp4_proc_exit(void) 1804 { 1805 udp_proc_unregister(&udp4_seq_afinfo); 1806 } 1807 #endif /* CONFIG_PROC_FS */ 1808 1809 EXPORT_SYMBOL(udp_disconnect); 1810 EXPORT_SYMBOL(udp_hash); 1811 EXPORT_SYMBOL(udp_hash_lock); 1812 EXPORT_SYMBOL(udp_ioctl); 1813 EXPORT_SYMBOL(udp_get_port); 1814 EXPORT_SYMBOL(udp_prot); 1815 EXPORT_SYMBOL(udp_sendmsg); 1816 EXPORT_SYMBOL(udp_lib_getsockopt); 1817 EXPORT_SYMBOL(udp_lib_setsockopt); 1818 EXPORT_SYMBOL(udp_poll); 1819 1820 #ifdef CONFIG_PROC_FS 1821 EXPORT_SYMBOL(udp_proc_register); 1822 EXPORT_SYMBOL(udp_proc_unregister); 1823 #endif 1824