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 * James Chapman : Add L2TP encapsulation type. 74 * 75 * 76 * This program is free software; you can redistribute it and/or 77 * modify it under the terms of the GNU General Public License 78 * as published by the Free Software Foundation; either version 79 * 2 of the License, or (at your option) any later version. 80 */ 81 82 #include <asm/system.h> 83 #include <asm/uaccess.h> 84 #include <asm/ioctls.h> 85 #include <linux/types.h> 86 #include <linux/fcntl.h> 87 #include <linux/module.h> 88 #include <linux/socket.h> 89 #include <linux/sockios.h> 90 #include <linux/igmp.h> 91 #include <linux/in.h> 92 #include <linux/errno.h> 93 #include <linux/timer.h> 94 #include <linux/mm.h> 95 #include <linux/inet.h> 96 #include <linux/netdevice.h> 97 #include <net/tcp_states.h> 98 #include <linux/skbuff.h> 99 #include <linux/proc_fs.h> 100 #include <linux/seq_file.h> 101 #include <net/net_namespace.h> 102 #include <net/icmp.h> 103 #include <net/route.h> 104 #include <net/checksum.h> 105 #include <net/xfrm.h> 106 #include "udp_impl.h" 107 108 /* 109 * Snmp MIB for the UDP layer 110 */ 111 112 DEFINE_SNMP_STAT(struct udp_mib, udp_statistics) __read_mostly; 113 114 struct hlist_head udp_hash[UDP_HTABLE_SIZE]; 115 DEFINE_RWLOCK(udp_hash_lock); 116 117 static inline int __udp_lib_lport_inuse(__u16 num, 118 const struct hlist_head udptable[]) 119 { 120 struct sock *sk; 121 struct hlist_node *node; 122 123 sk_for_each(sk, node, &udptable[num & (UDP_HTABLE_SIZE - 1)]) 124 if (sk->sk_hash == num) 125 return 1; 126 return 0; 127 } 128 129 /** 130 * __udp_lib_get_port - UDP/-Lite port lookup for IPv4 and IPv6 131 * 132 * @sk: socket struct in question 133 * @snum: port number to look up 134 * @udptable: hash list table, must be of UDP_HTABLE_SIZE 135 * @saddr_comp: AF-dependent comparison of bound local IP addresses 136 */ 137 int __udp_lib_get_port(struct sock *sk, unsigned short snum, 138 struct hlist_head udptable[], 139 int (*saddr_comp)(const struct sock *sk1, 140 const struct sock *sk2 ) ) 141 { 142 struct hlist_node *node; 143 struct hlist_head *head; 144 struct sock *sk2; 145 int error = 1; 146 147 write_lock_bh(&udp_hash_lock); 148 149 if (!snum) { 150 int i, low, high; 151 unsigned rover, best, best_size_so_far; 152 153 inet_get_local_port_range(&low, &high); 154 155 best_size_so_far = UINT_MAX; 156 best = rover = net_random() % (high - low) + low; 157 158 /* 1st pass: look for empty (or shortest) hash chain */ 159 for (i = 0; i < UDP_HTABLE_SIZE; i++) { 160 int size = 0; 161 162 head = &udptable[rover & (UDP_HTABLE_SIZE - 1)]; 163 if (hlist_empty(head)) 164 goto gotit; 165 166 sk_for_each(sk2, node, head) { 167 if (++size >= best_size_so_far) 168 goto next; 169 } 170 best_size_so_far = size; 171 best = rover; 172 next: 173 /* fold back if end of range */ 174 if (++rover > high) 175 rover = low + ((rover - low) 176 & (UDP_HTABLE_SIZE - 1)); 177 178 179 } 180 181 /* 2nd pass: find hole in shortest hash chain */ 182 rover = best; 183 for (i = 0; i < (1 << 16) / UDP_HTABLE_SIZE; i++) { 184 if (! __udp_lib_lport_inuse(rover, udptable)) 185 goto gotit; 186 rover += UDP_HTABLE_SIZE; 187 if (rover > high) 188 rover = low + ((rover - low) 189 & (UDP_HTABLE_SIZE - 1)); 190 } 191 192 193 /* All ports in use! */ 194 goto fail; 195 196 gotit: 197 snum = rover; 198 } else { 199 head = &udptable[snum & (UDP_HTABLE_SIZE - 1)]; 200 201 sk_for_each(sk2, node, head) 202 if (sk2->sk_hash == snum && 203 sk2 != sk && 204 (!sk2->sk_reuse || !sk->sk_reuse) && 205 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if 206 || sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 207 (*saddr_comp)(sk, sk2) ) 208 goto fail; 209 } 210 211 inet_sk(sk)->num = snum; 212 sk->sk_hash = snum; 213 if (sk_unhashed(sk)) { 214 head = &udptable[snum & (UDP_HTABLE_SIZE - 1)]; 215 sk_add_node(sk, head); 216 sock_prot_inc_use(sk->sk_prot); 217 } 218 error = 0; 219 fail: 220 write_unlock_bh(&udp_hash_lock); 221 return error; 222 } 223 224 int udp_get_port(struct sock *sk, unsigned short snum, 225 int (*scmp)(const struct sock *, const struct sock *)) 226 { 227 return __udp_lib_get_port(sk, snum, udp_hash, scmp); 228 } 229 230 int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2) 231 { 232 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2); 233 234 return ( !ipv6_only_sock(sk2) && 235 (!inet1->rcv_saddr || !inet2->rcv_saddr || 236 inet1->rcv_saddr == inet2->rcv_saddr )); 237 } 238 239 static inline int udp_v4_get_port(struct sock *sk, unsigned short snum) 240 { 241 return udp_get_port(sk, snum, ipv4_rcv_saddr_equal); 242 } 243 244 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try 245 * harder than this. -DaveM 246 */ 247 static struct sock *__udp4_lib_lookup(__be32 saddr, __be16 sport, 248 __be32 daddr, __be16 dport, 249 int dif, struct hlist_head udptable[]) 250 { 251 struct sock *sk, *result = NULL; 252 struct hlist_node *node; 253 unsigned short hnum = ntohs(dport); 254 int badness = -1; 255 256 read_lock(&udp_hash_lock); 257 sk_for_each(sk, node, &udptable[hnum & (UDP_HTABLE_SIZE - 1)]) { 258 struct inet_sock *inet = inet_sk(sk); 259 260 if (sk->sk_hash == hnum && !ipv6_only_sock(sk)) { 261 int score = (sk->sk_family == PF_INET ? 1 : 0); 262 if (inet->rcv_saddr) { 263 if (inet->rcv_saddr != daddr) 264 continue; 265 score+=2; 266 } 267 if (inet->daddr) { 268 if (inet->daddr != saddr) 269 continue; 270 score+=2; 271 } 272 if (inet->dport) { 273 if (inet->dport != sport) 274 continue; 275 score+=2; 276 } 277 if (sk->sk_bound_dev_if) { 278 if (sk->sk_bound_dev_if != dif) 279 continue; 280 score+=2; 281 } 282 if (score == 9) { 283 result = sk; 284 break; 285 } else if (score > badness) { 286 result = sk; 287 badness = score; 288 } 289 } 290 } 291 if (result) 292 sock_hold(result); 293 read_unlock(&udp_hash_lock); 294 return result; 295 } 296 297 static inline struct sock *udp_v4_mcast_next(struct sock *sk, 298 __be16 loc_port, __be32 loc_addr, 299 __be16 rmt_port, __be32 rmt_addr, 300 int dif) 301 { 302 struct hlist_node *node; 303 struct sock *s = sk; 304 unsigned short hnum = ntohs(loc_port); 305 306 sk_for_each_from(s, node) { 307 struct inet_sock *inet = inet_sk(s); 308 309 if (s->sk_hash != hnum || 310 (inet->daddr && inet->daddr != rmt_addr) || 311 (inet->dport != rmt_port && inet->dport) || 312 (inet->rcv_saddr && inet->rcv_saddr != loc_addr) || 313 ipv6_only_sock(s) || 314 (s->sk_bound_dev_if && s->sk_bound_dev_if != dif)) 315 continue; 316 if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif)) 317 continue; 318 goto found; 319 } 320 s = NULL; 321 found: 322 return s; 323 } 324 325 /* 326 * This routine is called by the ICMP module when it gets some 327 * sort of error condition. If err < 0 then the socket should 328 * be closed and the error returned to the user. If err > 0 329 * it's just the icmp type << 8 | icmp code. 330 * Header points to the ip header of the error packet. We move 331 * on past this. Then (as it used to claim before adjustment) 332 * header points to the first 8 bytes of the udp header. We need 333 * to find the appropriate port. 334 */ 335 336 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct hlist_head udptable[]) 337 { 338 struct inet_sock *inet; 339 struct iphdr *iph = (struct iphdr*)skb->data; 340 struct udphdr *uh = (struct udphdr*)(skb->data+(iph->ihl<<2)); 341 const int type = icmp_hdr(skb)->type; 342 const int code = icmp_hdr(skb)->code; 343 struct sock *sk; 344 int harderr; 345 int err; 346 347 sk = __udp4_lib_lookup(iph->daddr, uh->dest, iph->saddr, uh->source, 348 skb->dev->ifindex, udptable ); 349 if (sk == NULL) { 350 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS); 351 return; /* No socket for error */ 352 } 353 354 err = 0; 355 harderr = 0; 356 inet = inet_sk(sk); 357 358 switch (type) { 359 default: 360 case ICMP_TIME_EXCEEDED: 361 err = EHOSTUNREACH; 362 break; 363 case ICMP_SOURCE_QUENCH: 364 goto out; 365 case ICMP_PARAMETERPROB: 366 err = EPROTO; 367 harderr = 1; 368 break; 369 case ICMP_DEST_UNREACH: 370 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */ 371 if (inet->pmtudisc != IP_PMTUDISC_DONT) { 372 err = EMSGSIZE; 373 harderr = 1; 374 break; 375 } 376 goto out; 377 } 378 err = EHOSTUNREACH; 379 if (code <= NR_ICMP_UNREACH) { 380 harderr = icmp_err_convert[code].fatal; 381 err = icmp_err_convert[code].errno; 382 } 383 break; 384 } 385 386 /* 387 * RFC1122: OK. Passes ICMP errors back to application, as per 388 * 4.1.3.3. 389 */ 390 if (!inet->recverr) { 391 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 392 goto out; 393 } else { 394 ip_icmp_error(sk, skb, err, uh->dest, info, (u8*)(uh+1)); 395 } 396 sk->sk_err = err; 397 sk->sk_error_report(sk); 398 out: 399 sock_put(sk); 400 } 401 402 void udp_err(struct sk_buff *skb, u32 info) 403 { 404 return __udp4_lib_err(skb, info, udp_hash); 405 } 406 407 /* 408 * Throw away all pending data and cancel the corking. Socket is locked. 409 */ 410 static void udp_flush_pending_frames(struct sock *sk) 411 { 412 struct udp_sock *up = udp_sk(sk); 413 414 if (up->pending) { 415 up->len = 0; 416 up->pending = 0; 417 ip_flush_pending_frames(sk); 418 } 419 } 420 421 /** 422 * udp4_hwcsum_outgoing - handle outgoing HW checksumming 423 * @sk: socket we are sending on 424 * @skb: sk_buff containing the filled-in UDP header 425 * (checksum field must be zeroed out) 426 */ 427 static void udp4_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb, 428 __be32 src, __be32 dst, int len ) 429 { 430 unsigned int offset; 431 struct udphdr *uh = udp_hdr(skb); 432 __wsum csum = 0; 433 434 if (skb_queue_len(&sk->sk_write_queue) == 1) { 435 /* 436 * Only one fragment on the socket. 437 */ 438 skb->csum_start = skb_transport_header(skb) - skb->head; 439 skb->csum_offset = offsetof(struct udphdr, check); 440 uh->check = ~csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, 0); 441 } else { 442 /* 443 * HW-checksum won't work as there are two or more 444 * fragments on the socket so that all csums of sk_buffs 445 * should be together 446 */ 447 offset = skb_transport_offset(skb); 448 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0); 449 450 skb->ip_summed = CHECKSUM_NONE; 451 452 skb_queue_walk(&sk->sk_write_queue, skb) { 453 csum = csum_add(csum, skb->csum); 454 } 455 456 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); 457 if (uh->check == 0) 458 uh->check = CSUM_MANGLED_0; 459 } 460 } 461 462 /* 463 * Push out all pending data as one UDP datagram. Socket is locked. 464 */ 465 static int udp_push_pending_frames(struct sock *sk) 466 { 467 struct udp_sock *up = udp_sk(sk); 468 struct inet_sock *inet = inet_sk(sk); 469 struct flowi *fl = &inet->cork.fl; 470 struct sk_buff *skb; 471 struct udphdr *uh; 472 int err = 0; 473 __wsum csum = 0; 474 475 /* Grab the skbuff where UDP header space exists. */ 476 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL) 477 goto out; 478 479 /* 480 * Create a UDP header 481 */ 482 uh = udp_hdr(skb); 483 uh->source = fl->fl_ip_sport; 484 uh->dest = fl->fl_ip_dport; 485 uh->len = htons(up->len); 486 uh->check = 0; 487 488 if (up->pcflag) /* UDP-Lite */ 489 csum = udplite_csum_outgoing(sk, skb); 490 491 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */ 492 493 skb->ip_summed = CHECKSUM_NONE; 494 goto send; 495 496 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 497 498 udp4_hwcsum_outgoing(sk, skb, fl->fl4_src,fl->fl4_dst, up->len); 499 goto send; 500 501 } else /* `normal' UDP */ 502 csum = udp_csum_outgoing(sk, skb); 503 504 /* add protocol-dependent pseudo-header */ 505 uh->check = csum_tcpudp_magic(fl->fl4_src, fl->fl4_dst, up->len, 506 sk->sk_protocol, csum ); 507 if (uh->check == 0) 508 uh->check = CSUM_MANGLED_0; 509 510 send: 511 err = ip_push_pending_frames(sk); 512 out: 513 up->len = 0; 514 up->pending = 0; 515 if (!err) 516 UDP_INC_STATS_USER(UDP_MIB_OUTDATAGRAMS, up->pcflag); 517 return err; 518 } 519 520 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 521 size_t len) 522 { 523 struct inet_sock *inet = inet_sk(sk); 524 struct udp_sock *up = udp_sk(sk); 525 int ulen = len; 526 struct ipcm_cookie ipc; 527 struct rtable *rt = NULL; 528 int free = 0; 529 int connected = 0; 530 __be32 daddr, faddr, saddr; 531 __be16 dport; 532 u8 tos; 533 int err, is_udplite = up->pcflag; 534 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; 535 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 536 537 if (len > 0xFFFF) 538 return -EMSGSIZE; 539 540 /* 541 * Check the flags. 542 */ 543 544 if (msg->msg_flags&MSG_OOB) /* Mirror BSD error message compatibility */ 545 return -EOPNOTSUPP; 546 547 ipc.opt = NULL; 548 549 if (up->pending) { 550 /* 551 * There are pending frames. 552 * The socket lock must be held while it's corked. 553 */ 554 lock_sock(sk); 555 if (likely(up->pending)) { 556 if (unlikely(up->pending != AF_INET)) { 557 release_sock(sk); 558 return -EINVAL; 559 } 560 goto do_append_data; 561 } 562 release_sock(sk); 563 } 564 ulen += sizeof(struct udphdr); 565 566 /* 567 * Get and verify the address. 568 */ 569 if (msg->msg_name) { 570 struct sockaddr_in * usin = (struct sockaddr_in*)msg->msg_name; 571 if (msg->msg_namelen < sizeof(*usin)) 572 return -EINVAL; 573 if (usin->sin_family != AF_INET) { 574 if (usin->sin_family != AF_UNSPEC) 575 return -EAFNOSUPPORT; 576 } 577 578 daddr = usin->sin_addr.s_addr; 579 dport = usin->sin_port; 580 if (dport == 0) 581 return -EINVAL; 582 } else { 583 if (sk->sk_state != TCP_ESTABLISHED) 584 return -EDESTADDRREQ; 585 daddr = inet->daddr; 586 dport = inet->dport; 587 /* Open fast path for connected socket. 588 Route will not be used, if at least one option is set. 589 */ 590 connected = 1; 591 } 592 ipc.addr = inet->saddr; 593 594 ipc.oif = sk->sk_bound_dev_if; 595 if (msg->msg_controllen) { 596 err = ip_cmsg_send(msg, &ipc); 597 if (err) 598 return err; 599 if (ipc.opt) 600 free = 1; 601 connected = 0; 602 } 603 if (!ipc.opt) 604 ipc.opt = inet->opt; 605 606 saddr = ipc.addr; 607 ipc.addr = faddr = daddr; 608 609 if (ipc.opt && ipc.opt->srr) { 610 if (!daddr) 611 return -EINVAL; 612 faddr = ipc.opt->faddr; 613 connected = 0; 614 } 615 tos = RT_TOS(inet->tos); 616 if (sock_flag(sk, SOCK_LOCALROUTE) || 617 (msg->msg_flags & MSG_DONTROUTE) || 618 (ipc.opt && ipc.opt->is_strictroute)) { 619 tos |= RTO_ONLINK; 620 connected = 0; 621 } 622 623 if (MULTICAST(daddr)) { 624 if (!ipc.oif) 625 ipc.oif = inet->mc_index; 626 if (!saddr) 627 saddr = inet->mc_addr; 628 connected = 0; 629 } 630 631 if (connected) 632 rt = (struct rtable*)sk_dst_check(sk, 0); 633 634 if (rt == NULL) { 635 struct flowi fl = { .oif = ipc.oif, 636 .nl_u = { .ip4_u = 637 { .daddr = faddr, 638 .saddr = saddr, 639 .tos = tos } }, 640 .proto = sk->sk_protocol, 641 .uli_u = { .ports = 642 { .sport = inet->sport, 643 .dport = dport } } }; 644 security_sk_classify_flow(sk, &fl); 645 err = ip_route_output_flow(&rt, &fl, sk, 1); 646 if (err) { 647 if (err == -ENETUNREACH) 648 IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES); 649 goto out; 650 } 651 652 err = -EACCES; 653 if ((rt->rt_flags & RTCF_BROADCAST) && 654 !sock_flag(sk, SOCK_BROADCAST)) 655 goto out; 656 if (connected) 657 sk_dst_set(sk, dst_clone(&rt->u.dst)); 658 } 659 660 if (msg->msg_flags&MSG_CONFIRM) 661 goto do_confirm; 662 back_from_confirm: 663 664 saddr = rt->rt_src; 665 if (!ipc.addr) 666 daddr = ipc.addr = rt->rt_dst; 667 668 lock_sock(sk); 669 if (unlikely(up->pending)) { 670 /* The socket is already corked while preparing it. */ 671 /* ... which is an evident application bug. --ANK */ 672 release_sock(sk); 673 674 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n"); 675 err = -EINVAL; 676 goto out; 677 } 678 /* 679 * Now cork the socket to pend data. 680 */ 681 inet->cork.fl.fl4_dst = daddr; 682 inet->cork.fl.fl_ip_dport = dport; 683 inet->cork.fl.fl4_src = saddr; 684 inet->cork.fl.fl_ip_sport = inet->sport; 685 up->pending = AF_INET; 686 687 do_append_data: 688 up->len += ulen; 689 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 690 err = ip_append_data(sk, getfrag, msg->msg_iov, ulen, 691 sizeof(struct udphdr), &ipc, rt, 692 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 693 if (err) 694 udp_flush_pending_frames(sk); 695 else if (!corkreq) 696 err = udp_push_pending_frames(sk); 697 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 698 up->pending = 0; 699 release_sock(sk); 700 701 out: 702 ip_rt_put(rt); 703 if (free) 704 kfree(ipc.opt); 705 if (!err) 706 return len; 707 /* 708 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 709 * ENOBUFS might not be good (it's not tunable per se), but otherwise 710 * we don't have a good statistic (IpOutDiscards but it can be too many 711 * things). We could add another new stat but at least for now that 712 * seems like overkill. 713 */ 714 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 715 UDP_INC_STATS_USER(UDP_MIB_SNDBUFERRORS, is_udplite); 716 } 717 return err; 718 719 do_confirm: 720 dst_confirm(&rt->u.dst); 721 if (!(msg->msg_flags&MSG_PROBE) || len) 722 goto back_from_confirm; 723 err = 0; 724 goto out; 725 } 726 727 int udp_sendpage(struct sock *sk, struct page *page, int offset, 728 size_t size, int flags) 729 { 730 struct udp_sock *up = udp_sk(sk); 731 int ret; 732 733 if (!up->pending) { 734 struct msghdr msg = { .msg_flags = flags|MSG_MORE }; 735 736 /* Call udp_sendmsg to specify destination address which 737 * sendpage interface can't pass. 738 * This will succeed only when the socket is connected. 739 */ 740 ret = udp_sendmsg(NULL, sk, &msg, 0); 741 if (ret < 0) 742 return ret; 743 } 744 745 lock_sock(sk); 746 747 if (unlikely(!up->pending)) { 748 release_sock(sk); 749 750 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n"); 751 return -EINVAL; 752 } 753 754 ret = ip_append_page(sk, page, offset, size, flags); 755 if (ret == -EOPNOTSUPP) { 756 release_sock(sk); 757 return sock_no_sendpage(sk->sk_socket, page, offset, 758 size, flags); 759 } 760 if (ret < 0) { 761 udp_flush_pending_frames(sk); 762 goto out; 763 } 764 765 up->len += size; 766 if (!(up->corkflag || (flags&MSG_MORE))) 767 ret = udp_push_pending_frames(sk); 768 if (!ret) 769 ret = size; 770 out: 771 release_sock(sk); 772 return ret; 773 } 774 775 /* 776 * IOCTL requests applicable to the UDP protocol 777 */ 778 779 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg) 780 { 781 switch (cmd) { 782 case SIOCOUTQ: 783 { 784 int amount = atomic_read(&sk->sk_wmem_alloc); 785 return put_user(amount, (int __user *)arg); 786 } 787 788 case SIOCINQ: 789 { 790 struct sk_buff *skb; 791 unsigned long amount; 792 793 amount = 0; 794 spin_lock_bh(&sk->sk_receive_queue.lock); 795 skb = skb_peek(&sk->sk_receive_queue); 796 if (skb != NULL) { 797 /* 798 * We will only return the amount 799 * of this packet since that is all 800 * that will be read. 801 */ 802 amount = skb->len - sizeof(struct udphdr); 803 } 804 spin_unlock_bh(&sk->sk_receive_queue.lock); 805 return put_user(amount, (int __user *)arg); 806 } 807 808 default: 809 return -ENOIOCTLCMD; 810 } 811 812 return 0; 813 } 814 815 /* 816 * This should be easy, if there is something there we 817 * return it, otherwise we block. 818 */ 819 820 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 821 size_t len, int noblock, int flags, int *addr_len) 822 { 823 struct inet_sock *inet = inet_sk(sk); 824 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name; 825 struct sk_buff *skb; 826 unsigned int ulen, copied; 827 int err; 828 int is_udplite = IS_UDPLITE(sk); 829 830 /* 831 * Check any passed addresses 832 */ 833 if (addr_len) 834 *addr_len=sizeof(*sin); 835 836 if (flags & MSG_ERRQUEUE) 837 return ip_recv_error(sk, msg, len); 838 839 try_again: 840 skb = skb_recv_datagram(sk, flags, noblock, &err); 841 if (!skb) 842 goto out; 843 844 ulen = skb->len - sizeof(struct udphdr); 845 copied = len; 846 if (copied > ulen) 847 copied = ulen; 848 else if (copied < ulen) 849 msg->msg_flags |= MSG_TRUNC; 850 851 /* 852 * If checksum is needed at all, try to do it while copying the 853 * data. If the data is truncated, or if we only want a partial 854 * coverage checksum (UDP-Lite), do it before the copy. 855 */ 856 857 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) { 858 if (udp_lib_checksum_complete(skb)) 859 goto csum_copy_err; 860 } 861 862 if (skb_csum_unnecessary(skb)) 863 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr), 864 msg->msg_iov, copied ); 865 else { 866 err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov); 867 868 if (err == -EINVAL) 869 goto csum_copy_err; 870 } 871 872 if (err) 873 goto out_free; 874 875 sock_recv_timestamp(msg, sk, skb); 876 877 /* Copy the address. */ 878 if (sin) 879 { 880 sin->sin_family = AF_INET; 881 sin->sin_port = udp_hdr(skb)->source; 882 sin->sin_addr.s_addr = ip_hdr(skb)->saddr; 883 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 884 } 885 if (inet->cmsg_flags) 886 ip_cmsg_recv(msg, skb); 887 888 err = copied; 889 if (flags & MSG_TRUNC) 890 err = ulen; 891 892 out_free: 893 skb_free_datagram(sk, skb); 894 out: 895 return err; 896 897 csum_copy_err: 898 UDP_INC_STATS_BH(UDP_MIB_INERRORS, is_udplite); 899 900 skb_kill_datagram(sk, skb, flags); 901 902 if (noblock) 903 return -EAGAIN; 904 goto try_again; 905 } 906 907 908 int udp_disconnect(struct sock *sk, int flags) 909 { 910 struct inet_sock *inet = inet_sk(sk); 911 /* 912 * 1003.1g - break association. 913 */ 914 915 sk->sk_state = TCP_CLOSE; 916 inet->daddr = 0; 917 inet->dport = 0; 918 sk->sk_bound_dev_if = 0; 919 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) 920 inet_reset_saddr(sk); 921 922 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { 923 sk->sk_prot->unhash(sk); 924 inet->sport = 0; 925 } 926 sk_dst_reset(sk); 927 return 0; 928 } 929 930 /* returns: 931 * -1: error 932 * 0: success 933 * >0: "udp encap" protocol resubmission 934 * 935 * Note that in the success and error cases, the skb is assumed to 936 * have either been requeued or freed. 937 */ 938 int udp_queue_rcv_skb(struct sock * sk, struct sk_buff *skb) 939 { 940 struct udp_sock *up = udp_sk(sk); 941 int rc; 942 943 /* 944 * Charge it to the socket, dropping if the queue is full. 945 */ 946 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) 947 goto drop; 948 nf_reset(skb); 949 950 if (up->encap_type) { 951 /* 952 * This is an encapsulation socket so pass the skb to 953 * the socket's udp_encap_rcv() hook. Otherwise, just 954 * fall through and pass this up the UDP socket. 955 * up->encap_rcv() returns the following value: 956 * =0 if skb was successfully passed to the encap 957 * handler or was discarded by it. 958 * >0 if skb should be passed on to UDP. 959 * <0 if skb should be resubmitted as proto -N 960 */ 961 962 /* if we're overly short, let UDP handle it */ 963 if (skb->len > sizeof(struct udphdr) && 964 up->encap_rcv != NULL) { 965 int ret; 966 967 ret = (*up->encap_rcv)(sk, skb); 968 if (ret <= 0) { 969 UDP_INC_STATS_BH(UDP_MIB_INDATAGRAMS, up->pcflag); 970 return -ret; 971 } 972 } 973 974 /* FALLTHROUGH -- it's a UDP Packet */ 975 } 976 977 /* 978 * UDP-Lite specific tests, ignored on UDP sockets 979 */ 980 if ((up->pcflag & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 981 982 /* 983 * MIB statistics other than incrementing the error count are 984 * disabled for the following two types of errors: these depend 985 * on the application settings, not on the functioning of the 986 * protocol stack as such. 987 * 988 * RFC 3828 here recommends (sec 3.3): "There should also be a 989 * way ... to ... at least let the receiving application block 990 * delivery of packets with coverage values less than a value 991 * provided by the application." 992 */ 993 if (up->pcrlen == 0) { /* full coverage was set */ 994 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage " 995 "%d while full coverage %d requested\n", 996 UDP_SKB_CB(skb)->cscov, skb->len); 997 goto drop; 998 } 999 /* The next case involves violating the min. coverage requested 1000 * by the receiver. This is subtle: if receiver wants x and x is 1001 * greater than the buffersize/MTU then receiver will complain 1002 * that it wants x while sender emits packets of smaller size y. 1003 * Therefore the above ...()->partial_cov statement is essential. 1004 */ 1005 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 1006 LIMIT_NETDEBUG(KERN_WARNING 1007 "UDPLITE: coverage %d too small, need min %d\n", 1008 UDP_SKB_CB(skb)->cscov, up->pcrlen); 1009 goto drop; 1010 } 1011 } 1012 1013 if (sk->sk_filter) { 1014 if (udp_lib_checksum_complete(skb)) 1015 goto drop; 1016 } 1017 1018 if ((rc = sock_queue_rcv_skb(sk,skb)) < 0) { 1019 /* Note that an ENOMEM error is charged twice */ 1020 if (rc == -ENOMEM) 1021 UDP_INC_STATS_BH(UDP_MIB_RCVBUFERRORS, up->pcflag); 1022 goto drop; 1023 } 1024 1025 UDP_INC_STATS_BH(UDP_MIB_INDATAGRAMS, up->pcflag); 1026 return 0; 1027 1028 drop: 1029 UDP_INC_STATS_BH(UDP_MIB_INERRORS, up->pcflag); 1030 kfree_skb(skb); 1031 return -1; 1032 } 1033 1034 /* 1035 * Multicasts and broadcasts go to each listener. 1036 * 1037 * Note: called only from the BH handler context, 1038 * so we don't need to lock the hashes. 1039 */ 1040 static int __udp4_lib_mcast_deliver(struct sk_buff *skb, 1041 struct udphdr *uh, 1042 __be32 saddr, __be32 daddr, 1043 struct hlist_head udptable[]) 1044 { 1045 struct sock *sk; 1046 int dif; 1047 1048 read_lock(&udp_hash_lock); 1049 sk = sk_head(&udptable[ntohs(uh->dest) & (UDP_HTABLE_SIZE - 1)]); 1050 dif = skb->dev->ifindex; 1051 sk = udp_v4_mcast_next(sk, uh->dest, daddr, uh->source, saddr, dif); 1052 if (sk) { 1053 struct sock *sknext = NULL; 1054 1055 do { 1056 struct sk_buff *skb1 = skb; 1057 1058 sknext = udp_v4_mcast_next(sk_next(sk), uh->dest, daddr, 1059 uh->source, saddr, dif); 1060 if (sknext) 1061 skb1 = skb_clone(skb, GFP_ATOMIC); 1062 1063 if (skb1) { 1064 int ret = udp_queue_rcv_skb(sk, skb1); 1065 if (ret > 0) 1066 /* we should probably re-process instead 1067 * of dropping packets here. */ 1068 kfree_skb(skb1); 1069 } 1070 sk = sknext; 1071 } while (sknext); 1072 } else 1073 kfree_skb(skb); 1074 read_unlock(&udp_hash_lock); 1075 return 0; 1076 } 1077 1078 /* Initialize UDP checksum. If exited with zero value (success), 1079 * CHECKSUM_UNNECESSARY means, that no more checks are required. 1080 * Otherwise, csum completion requires chacksumming packet body, 1081 * including udp header and folding it to skb->csum. 1082 */ 1083 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, 1084 int proto) 1085 { 1086 const struct iphdr *iph; 1087 int err; 1088 1089 UDP_SKB_CB(skb)->partial_cov = 0; 1090 UDP_SKB_CB(skb)->cscov = skb->len; 1091 1092 if (proto == IPPROTO_UDPLITE) { 1093 err = udplite_checksum_init(skb, uh); 1094 if (err) 1095 return err; 1096 } 1097 1098 iph = ip_hdr(skb); 1099 if (uh->check == 0) { 1100 skb->ip_summed = CHECKSUM_UNNECESSARY; 1101 } else if (skb->ip_summed == CHECKSUM_COMPLETE) { 1102 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len, 1103 proto, skb->csum)) 1104 skb->ip_summed = CHECKSUM_UNNECESSARY; 1105 } 1106 if (!skb_csum_unnecessary(skb)) 1107 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr, 1108 skb->len, proto, 0); 1109 /* Probably, we should checksum udp header (it should be in cache 1110 * in any case) and data in tiny packets (< rx copybreak). 1111 */ 1112 1113 return 0; 1114 } 1115 1116 /* 1117 * All we need to do is get the socket, and then do a checksum. 1118 */ 1119 1120 int __udp4_lib_rcv(struct sk_buff *skb, struct hlist_head udptable[], 1121 int proto) 1122 { 1123 struct sock *sk; 1124 struct udphdr *uh = udp_hdr(skb); 1125 unsigned short ulen; 1126 struct rtable *rt = (struct rtable*)skb->dst; 1127 __be32 saddr = ip_hdr(skb)->saddr; 1128 __be32 daddr = ip_hdr(skb)->daddr; 1129 1130 /* 1131 * Validate the packet. 1132 */ 1133 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 1134 goto drop; /* No space for header. */ 1135 1136 ulen = ntohs(uh->len); 1137 if (ulen > skb->len) 1138 goto short_packet; 1139 1140 if (proto == IPPROTO_UDP) { 1141 /* UDP validates ulen. */ 1142 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) 1143 goto short_packet; 1144 uh = udp_hdr(skb); 1145 } 1146 1147 if (udp4_csum_init(skb, uh, proto)) 1148 goto csum_error; 1149 1150 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) 1151 return __udp4_lib_mcast_deliver(skb, uh, saddr, daddr, udptable); 1152 1153 sk = __udp4_lib_lookup(saddr, uh->source, daddr, uh->dest, 1154 skb->dev->ifindex, udptable ); 1155 1156 if (sk != NULL) { 1157 int ret = udp_queue_rcv_skb(sk, skb); 1158 sock_put(sk); 1159 1160 /* a return value > 0 means to resubmit the input, but 1161 * it wants the return to be -protocol, or 0 1162 */ 1163 if (ret > 0) 1164 return -ret; 1165 return 0; 1166 } 1167 1168 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 1169 goto drop; 1170 nf_reset(skb); 1171 1172 /* No socket. Drop packet silently, if checksum is wrong */ 1173 if (udp_lib_checksum_complete(skb)) 1174 goto csum_error; 1175 1176 UDP_INC_STATS_BH(UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 1177 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); 1178 1179 /* 1180 * Hmm. We got an UDP packet to a port to which we 1181 * don't wanna listen. Ignore it. 1182 */ 1183 kfree_skb(skb); 1184 return 0; 1185 1186 short_packet: 1187 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %u.%u.%u.%u:%u %d/%d to %u.%u.%u.%u:%u\n", 1188 proto == IPPROTO_UDPLITE ? "-Lite" : "", 1189 NIPQUAD(saddr), 1190 ntohs(uh->source), 1191 ulen, 1192 skb->len, 1193 NIPQUAD(daddr), 1194 ntohs(uh->dest)); 1195 goto drop; 1196 1197 csum_error: 1198 /* 1199 * RFC1122: OK. Discards the bad packet silently (as far as 1200 * the network is concerned, anyway) as per 4.1.3.4 (MUST). 1201 */ 1202 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %d.%d.%d.%d:%d to %d.%d.%d.%d:%d ulen %d\n", 1203 proto == IPPROTO_UDPLITE ? "-Lite" : "", 1204 NIPQUAD(saddr), 1205 ntohs(uh->source), 1206 NIPQUAD(daddr), 1207 ntohs(uh->dest), 1208 ulen); 1209 drop: 1210 UDP_INC_STATS_BH(UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 1211 kfree_skb(skb); 1212 return 0; 1213 } 1214 1215 int udp_rcv(struct sk_buff *skb) 1216 { 1217 return __udp4_lib_rcv(skb, udp_hash, IPPROTO_UDP); 1218 } 1219 1220 int udp_destroy_sock(struct sock *sk) 1221 { 1222 lock_sock(sk); 1223 udp_flush_pending_frames(sk); 1224 release_sock(sk); 1225 return 0; 1226 } 1227 1228 /* 1229 * Socket option code for UDP 1230 */ 1231 int udp_lib_setsockopt(struct sock *sk, int level, int optname, 1232 char __user *optval, int optlen, 1233 int (*push_pending_frames)(struct sock *)) 1234 { 1235 struct udp_sock *up = udp_sk(sk); 1236 int val; 1237 int err = 0; 1238 1239 if (optlen<sizeof(int)) 1240 return -EINVAL; 1241 1242 if (get_user(val, (int __user *)optval)) 1243 return -EFAULT; 1244 1245 switch (optname) { 1246 case UDP_CORK: 1247 if (val != 0) { 1248 up->corkflag = 1; 1249 } else { 1250 up->corkflag = 0; 1251 lock_sock(sk); 1252 (*push_pending_frames)(sk); 1253 release_sock(sk); 1254 } 1255 break; 1256 1257 case UDP_ENCAP: 1258 switch (val) { 1259 case 0: 1260 case UDP_ENCAP_ESPINUDP: 1261 case UDP_ENCAP_ESPINUDP_NON_IKE: 1262 up->encap_rcv = xfrm4_udp_encap_rcv; 1263 /* FALLTHROUGH */ 1264 case UDP_ENCAP_L2TPINUDP: 1265 up->encap_type = val; 1266 break; 1267 default: 1268 err = -ENOPROTOOPT; 1269 break; 1270 } 1271 break; 1272 1273 /* 1274 * UDP-Lite's partial checksum coverage (RFC 3828). 1275 */ 1276 /* The sender sets actual checksum coverage length via this option. 1277 * The case coverage > packet length is handled by send module. */ 1278 case UDPLITE_SEND_CSCOV: 1279 if (!up->pcflag) /* Disable the option on UDP sockets */ 1280 return -ENOPROTOOPT; 1281 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ 1282 val = 8; 1283 up->pcslen = val; 1284 up->pcflag |= UDPLITE_SEND_CC; 1285 break; 1286 1287 /* The receiver specifies a minimum checksum coverage value. To make 1288 * sense, this should be set to at least 8 (as done below). If zero is 1289 * used, this again means full checksum coverage. */ 1290 case UDPLITE_RECV_CSCOV: 1291 if (!up->pcflag) /* Disable the option on UDP sockets */ 1292 return -ENOPROTOOPT; 1293 if (val != 0 && val < 8) /* Avoid silly minimal values. */ 1294 val = 8; 1295 up->pcrlen = val; 1296 up->pcflag |= UDPLITE_RECV_CC; 1297 break; 1298 1299 default: 1300 err = -ENOPROTOOPT; 1301 break; 1302 } 1303 1304 return err; 1305 } 1306 1307 int udp_setsockopt(struct sock *sk, int level, int optname, 1308 char __user *optval, int optlen) 1309 { 1310 if (level == SOL_UDP || level == SOL_UDPLITE) 1311 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1312 udp_push_pending_frames); 1313 return ip_setsockopt(sk, level, optname, optval, optlen); 1314 } 1315 1316 #ifdef CONFIG_COMPAT 1317 int compat_udp_setsockopt(struct sock *sk, int level, int optname, 1318 char __user *optval, int optlen) 1319 { 1320 if (level == SOL_UDP || level == SOL_UDPLITE) 1321 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1322 udp_push_pending_frames); 1323 return compat_ip_setsockopt(sk, level, optname, optval, optlen); 1324 } 1325 #endif 1326 1327 int udp_lib_getsockopt(struct sock *sk, int level, int optname, 1328 char __user *optval, int __user *optlen) 1329 { 1330 struct udp_sock *up = udp_sk(sk); 1331 int val, len; 1332 1333 if (get_user(len,optlen)) 1334 return -EFAULT; 1335 1336 len = min_t(unsigned int, len, sizeof(int)); 1337 1338 if (len < 0) 1339 return -EINVAL; 1340 1341 switch (optname) { 1342 case UDP_CORK: 1343 val = up->corkflag; 1344 break; 1345 1346 case UDP_ENCAP: 1347 val = up->encap_type; 1348 break; 1349 1350 /* The following two cannot be changed on UDP sockets, the return is 1351 * always 0 (which corresponds to the full checksum coverage of UDP). */ 1352 case UDPLITE_SEND_CSCOV: 1353 val = up->pcslen; 1354 break; 1355 1356 case UDPLITE_RECV_CSCOV: 1357 val = up->pcrlen; 1358 break; 1359 1360 default: 1361 return -ENOPROTOOPT; 1362 } 1363 1364 if (put_user(len, optlen)) 1365 return -EFAULT; 1366 if (copy_to_user(optval, &val,len)) 1367 return -EFAULT; 1368 return 0; 1369 } 1370 1371 int udp_getsockopt(struct sock *sk, int level, int optname, 1372 char __user *optval, int __user *optlen) 1373 { 1374 if (level == SOL_UDP || level == SOL_UDPLITE) 1375 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1376 return ip_getsockopt(sk, level, optname, optval, optlen); 1377 } 1378 1379 #ifdef CONFIG_COMPAT 1380 int compat_udp_getsockopt(struct sock *sk, int level, int optname, 1381 char __user *optval, int __user *optlen) 1382 { 1383 if (level == SOL_UDP || level == SOL_UDPLITE) 1384 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1385 return compat_ip_getsockopt(sk, level, optname, optval, optlen); 1386 } 1387 #endif 1388 /** 1389 * udp_poll - wait for a UDP event. 1390 * @file - file struct 1391 * @sock - socket 1392 * @wait - poll table 1393 * 1394 * This is same as datagram poll, except for the special case of 1395 * blocking sockets. If application is using a blocking fd 1396 * and a packet with checksum error is in the queue; 1397 * then it could get return from select indicating data available 1398 * but then block when reading it. Add special case code 1399 * to work around these arguably broken applications. 1400 */ 1401 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait) 1402 { 1403 unsigned int mask = datagram_poll(file, sock, wait); 1404 struct sock *sk = sock->sk; 1405 int is_lite = IS_UDPLITE(sk); 1406 1407 /* Check for false positives due to checksum errors */ 1408 if ( (mask & POLLRDNORM) && 1409 !(file->f_flags & O_NONBLOCK) && 1410 !(sk->sk_shutdown & RCV_SHUTDOWN)){ 1411 struct sk_buff_head *rcvq = &sk->sk_receive_queue; 1412 struct sk_buff *skb; 1413 1414 spin_lock_bh(&rcvq->lock); 1415 while ((skb = skb_peek(rcvq)) != NULL && 1416 udp_lib_checksum_complete(skb)) { 1417 UDP_INC_STATS_BH(UDP_MIB_INERRORS, is_lite); 1418 __skb_unlink(skb, rcvq); 1419 kfree_skb(skb); 1420 } 1421 spin_unlock_bh(&rcvq->lock); 1422 1423 /* nothing to see, move along */ 1424 if (skb == NULL) 1425 mask &= ~(POLLIN | POLLRDNORM); 1426 } 1427 1428 return mask; 1429 1430 } 1431 1432 struct proto udp_prot = { 1433 .name = "UDP", 1434 .owner = THIS_MODULE, 1435 .close = udp_lib_close, 1436 .connect = ip4_datagram_connect, 1437 .disconnect = udp_disconnect, 1438 .ioctl = udp_ioctl, 1439 .destroy = udp_destroy_sock, 1440 .setsockopt = udp_setsockopt, 1441 .getsockopt = udp_getsockopt, 1442 .sendmsg = udp_sendmsg, 1443 .recvmsg = udp_recvmsg, 1444 .sendpage = udp_sendpage, 1445 .backlog_rcv = udp_queue_rcv_skb, 1446 .hash = udp_lib_hash, 1447 .unhash = udp_lib_unhash, 1448 .get_port = udp_v4_get_port, 1449 .obj_size = sizeof(struct udp_sock), 1450 #ifdef CONFIG_COMPAT 1451 .compat_setsockopt = compat_udp_setsockopt, 1452 .compat_getsockopt = compat_udp_getsockopt, 1453 #endif 1454 }; 1455 1456 /* ------------------------------------------------------------------------ */ 1457 #ifdef CONFIG_PROC_FS 1458 1459 static struct sock *udp_get_first(struct seq_file *seq) 1460 { 1461 struct sock *sk; 1462 struct udp_iter_state *state = seq->private; 1463 1464 for (state->bucket = 0; state->bucket < UDP_HTABLE_SIZE; ++state->bucket) { 1465 struct hlist_node *node; 1466 sk_for_each(sk, node, state->hashtable + state->bucket) { 1467 if (sk->sk_family == state->family) 1468 goto found; 1469 } 1470 } 1471 sk = NULL; 1472 found: 1473 return sk; 1474 } 1475 1476 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) 1477 { 1478 struct udp_iter_state *state = seq->private; 1479 1480 do { 1481 sk = sk_next(sk); 1482 try_again: 1483 ; 1484 } while (sk && sk->sk_family != state->family); 1485 1486 if (!sk && ++state->bucket < UDP_HTABLE_SIZE) { 1487 sk = sk_head(state->hashtable + state->bucket); 1488 goto try_again; 1489 } 1490 return sk; 1491 } 1492 1493 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) 1494 { 1495 struct sock *sk = udp_get_first(seq); 1496 1497 if (sk) 1498 while (pos && (sk = udp_get_next(seq, sk)) != NULL) 1499 --pos; 1500 return pos ? NULL : sk; 1501 } 1502 1503 static void *udp_seq_start(struct seq_file *seq, loff_t *pos) 1504 { 1505 read_lock(&udp_hash_lock); 1506 return *pos ? udp_get_idx(seq, *pos-1) : (void *)1; 1507 } 1508 1509 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 1510 { 1511 struct sock *sk; 1512 1513 if (v == (void *)1) 1514 sk = udp_get_idx(seq, 0); 1515 else 1516 sk = udp_get_next(seq, v); 1517 1518 ++*pos; 1519 return sk; 1520 } 1521 1522 static void udp_seq_stop(struct seq_file *seq, void *v) 1523 { 1524 read_unlock(&udp_hash_lock); 1525 } 1526 1527 static int udp_seq_open(struct inode *inode, struct file *file) 1528 { 1529 struct udp_seq_afinfo *afinfo = PDE(inode)->data; 1530 struct seq_file *seq; 1531 int rc = -ENOMEM; 1532 struct udp_iter_state *s = kzalloc(sizeof(*s), GFP_KERNEL); 1533 1534 if (!s) 1535 goto out; 1536 s->family = afinfo->family; 1537 s->hashtable = afinfo->hashtable; 1538 s->seq_ops.start = udp_seq_start; 1539 s->seq_ops.next = udp_seq_next; 1540 s->seq_ops.show = afinfo->seq_show; 1541 s->seq_ops.stop = udp_seq_stop; 1542 1543 rc = seq_open(file, &s->seq_ops); 1544 if (rc) 1545 goto out_kfree; 1546 1547 seq = file->private_data; 1548 seq->private = s; 1549 out: 1550 return rc; 1551 out_kfree: 1552 kfree(s); 1553 goto out; 1554 } 1555 1556 /* ------------------------------------------------------------------------ */ 1557 int udp_proc_register(struct udp_seq_afinfo *afinfo) 1558 { 1559 struct proc_dir_entry *p; 1560 int rc = 0; 1561 1562 if (!afinfo) 1563 return -EINVAL; 1564 afinfo->seq_fops->owner = afinfo->owner; 1565 afinfo->seq_fops->open = udp_seq_open; 1566 afinfo->seq_fops->read = seq_read; 1567 afinfo->seq_fops->llseek = seq_lseek; 1568 afinfo->seq_fops->release = seq_release_private; 1569 1570 p = proc_net_fops_create(&init_net, afinfo->name, S_IRUGO, afinfo->seq_fops); 1571 if (p) 1572 p->data = afinfo; 1573 else 1574 rc = -ENOMEM; 1575 return rc; 1576 } 1577 1578 void udp_proc_unregister(struct udp_seq_afinfo *afinfo) 1579 { 1580 if (!afinfo) 1581 return; 1582 proc_net_remove(&init_net, afinfo->name); 1583 memset(afinfo->seq_fops, 0, sizeof(*afinfo->seq_fops)); 1584 } 1585 1586 /* ------------------------------------------------------------------------ */ 1587 static void udp4_format_sock(struct sock *sp, char *tmpbuf, int bucket) 1588 { 1589 struct inet_sock *inet = inet_sk(sp); 1590 __be32 dest = inet->daddr; 1591 __be32 src = inet->rcv_saddr; 1592 __u16 destp = ntohs(inet->dport); 1593 __u16 srcp = ntohs(inet->sport); 1594 1595 sprintf(tmpbuf, "%4d: %08X:%04X %08X:%04X" 1596 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p", 1597 bucket, src, srcp, dest, destp, sp->sk_state, 1598 atomic_read(&sp->sk_wmem_alloc), 1599 atomic_read(&sp->sk_rmem_alloc), 1600 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp), 1601 atomic_read(&sp->sk_refcnt), sp); 1602 } 1603 1604 int udp4_seq_show(struct seq_file *seq, void *v) 1605 { 1606 if (v == SEQ_START_TOKEN) 1607 seq_printf(seq, "%-127s\n", 1608 " sl local_address rem_address st tx_queue " 1609 "rx_queue tr tm->when retrnsmt uid timeout " 1610 "inode"); 1611 else { 1612 char tmpbuf[129]; 1613 struct udp_iter_state *state = seq->private; 1614 1615 udp4_format_sock(v, tmpbuf, state->bucket); 1616 seq_printf(seq, "%-127s\n", tmpbuf); 1617 } 1618 return 0; 1619 } 1620 1621 /* ------------------------------------------------------------------------ */ 1622 static struct file_operations udp4_seq_fops; 1623 static struct udp_seq_afinfo udp4_seq_afinfo = { 1624 .owner = THIS_MODULE, 1625 .name = "udp", 1626 .family = AF_INET, 1627 .hashtable = udp_hash, 1628 .seq_show = udp4_seq_show, 1629 .seq_fops = &udp4_seq_fops, 1630 }; 1631 1632 int __init udp4_proc_init(void) 1633 { 1634 return udp_proc_register(&udp4_seq_afinfo); 1635 } 1636 1637 void udp4_proc_exit(void) 1638 { 1639 udp_proc_unregister(&udp4_seq_afinfo); 1640 } 1641 #endif /* CONFIG_PROC_FS */ 1642 1643 EXPORT_SYMBOL(udp_disconnect); 1644 EXPORT_SYMBOL(udp_hash); 1645 EXPORT_SYMBOL(udp_hash_lock); 1646 EXPORT_SYMBOL(udp_ioctl); 1647 EXPORT_SYMBOL(udp_get_port); 1648 EXPORT_SYMBOL(udp_prot); 1649 EXPORT_SYMBOL(udp_sendmsg); 1650 EXPORT_SYMBOL(udp_lib_getsockopt); 1651 EXPORT_SYMBOL(udp_lib_setsockopt); 1652 EXPORT_SYMBOL(udp_poll); 1653 1654 #ifdef CONFIG_PROC_FS 1655 EXPORT_SYMBOL(udp_proc_register); 1656 EXPORT_SYMBOL(udp_proc_unregister); 1657 #endif 1658