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/icmp.h> 102 #include <net/route.h> 103 #include <net/checksum.h> 104 #include <net/xfrm.h> 105 #include "udp_impl.h" 106 107 /* 108 * Snmp MIB for the UDP layer 109 */ 110 111 DEFINE_SNMP_STAT(struct udp_mib, udp_statistics) __read_mostly; 112 113 struct hlist_head udp_hash[UDP_HTABLE_SIZE]; 114 DEFINE_RWLOCK(udp_hash_lock); 115 116 static int udp_port_rover; 117 118 static inline int __udp_lib_lport_inuse(__u16 num, 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 * @port_rover: pointer to record of last unallocated port 136 * @saddr_comp: AF-dependent comparison of bound local IP addresses 137 */ 138 int __udp_lib_get_port(struct sock *sk, unsigned short snum, 139 struct hlist_head udptable[], int *port_rover, 140 int (*saddr_comp)(const struct sock *sk1, 141 const struct sock *sk2 ) ) 142 { 143 struct hlist_node *node; 144 struct hlist_head *head; 145 struct sock *sk2; 146 int error = 1; 147 148 write_lock_bh(&udp_hash_lock); 149 if (snum == 0) { 150 int best_size_so_far, best, result, i; 151 152 if (*port_rover > sysctl_local_port_range[1] || 153 *port_rover < sysctl_local_port_range[0]) 154 *port_rover = sysctl_local_port_range[0]; 155 best_size_so_far = 32767; 156 best = result = *port_rover; 157 for (i = 0; i < UDP_HTABLE_SIZE; i++, result++) { 158 int size; 159 160 head = &udptable[result & (UDP_HTABLE_SIZE - 1)]; 161 if (hlist_empty(head)) { 162 if (result > sysctl_local_port_range[1]) 163 result = sysctl_local_port_range[0] + 164 ((result - sysctl_local_port_range[0]) & 165 (UDP_HTABLE_SIZE - 1)); 166 goto gotit; 167 } 168 size = 0; 169 sk_for_each(sk2, node, head) { 170 if (++size >= best_size_so_far) 171 goto next; 172 } 173 best_size_so_far = size; 174 best = result; 175 next: 176 ; 177 } 178 result = best; 179 for (i = 0; i < (1 << 16) / UDP_HTABLE_SIZE; 180 i++, result += UDP_HTABLE_SIZE) { 181 if (result > sysctl_local_port_range[1]) 182 result = sysctl_local_port_range[0] 183 + ((result - sysctl_local_port_range[0]) & 184 (UDP_HTABLE_SIZE - 1)); 185 if (! __udp_lib_lport_inuse(result, udptable)) 186 break; 187 } 188 if (i >= (1 << 16) / UDP_HTABLE_SIZE) 189 goto fail; 190 gotit: 191 *port_rover = snum = result; 192 } else { 193 head = &udptable[snum & (UDP_HTABLE_SIZE - 1)]; 194 195 sk_for_each(sk2, node, head) 196 if (sk2->sk_hash == snum && 197 sk2 != sk && 198 (!sk2->sk_reuse || !sk->sk_reuse) && 199 (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if 200 || sk2->sk_bound_dev_if == sk->sk_bound_dev_if) && 201 (*saddr_comp)(sk, sk2) ) 202 goto fail; 203 } 204 inet_sk(sk)->num = snum; 205 sk->sk_hash = snum; 206 if (sk_unhashed(sk)) { 207 head = &udptable[snum & (UDP_HTABLE_SIZE - 1)]; 208 sk_add_node(sk, head); 209 sock_prot_inc_use(sk->sk_prot); 210 } 211 error = 0; 212 fail: 213 write_unlock_bh(&udp_hash_lock); 214 return error; 215 } 216 217 int udp_get_port(struct sock *sk, unsigned short snum, 218 int (*scmp)(const struct sock *, const struct sock *)) 219 { 220 return __udp_lib_get_port(sk, snum, udp_hash, &udp_port_rover, scmp); 221 } 222 223 int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2) 224 { 225 struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2); 226 227 return ( !ipv6_only_sock(sk2) && 228 (!inet1->rcv_saddr || !inet2->rcv_saddr || 229 inet1->rcv_saddr == inet2->rcv_saddr )); 230 } 231 232 static inline int udp_v4_get_port(struct sock *sk, unsigned short snum) 233 { 234 return udp_get_port(sk, snum, ipv4_rcv_saddr_equal); 235 } 236 237 /* UDP is nearly always wildcards out the wazoo, it makes no sense to try 238 * harder than this. -DaveM 239 */ 240 static struct sock *__udp4_lib_lookup(__be32 saddr, __be16 sport, 241 __be32 daddr, __be16 dport, 242 int dif, struct hlist_head udptable[]) 243 { 244 struct sock *sk, *result = NULL; 245 struct hlist_node *node; 246 unsigned short hnum = ntohs(dport); 247 int badness = -1; 248 249 read_lock(&udp_hash_lock); 250 sk_for_each(sk, node, &udptable[hnum & (UDP_HTABLE_SIZE - 1)]) { 251 struct inet_sock *inet = inet_sk(sk); 252 253 if (sk->sk_hash == hnum && !ipv6_only_sock(sk)) { 254 int score = (sk->sk_family == PF_INET ? 1 : 0); 255 if (inet->rcv_saddr) { 256 if (inet->rcv_saddr != daddr) 257 continue; 258 score+=2; 259 } 260 if (inet->daddr) { 261 if (inet->daddr != saddr) 262 continue; 263 score+=2; 264 } 265 if (inet->dport) { 266 if (inet->dport != sport) 267 continue; 268 score+=2; 269 } 270 if (sk->sk_bound_dev_if) { 271 if (sk->sk_bound_dev_if != dif) 272 continue; 273 score+=2; 274 } 275 if (score == 9) { 276 result = sk; 277 break; 278 } else if (score > badness) { 279 result = sk; 280 badness = score; 281 } 282 } 283 } 284 if (result) 285 sock_hold(result); 286 read_unlock(&udp_hash_lock); 287 return result; 288 } 289 290 static inline struct sock *udp_v4_mcast_next(struct sock *sk, 291 __be16 loc_port, __be32 loc_addr, 292 __be16 rmt_port, __be32 rmt_addr, 293 int dif) 294 { 295 struct hlist_node *node; 296 struct sock *s = sk; 297 unsigned short hnum = ntohs(loc_port); 298 299 sk_for_each_from(s, node) { 300 struct inet_sock *inet = inet_sk(s); 301 302 if (s->sk_hash != hnum || 303 (inet->daddr && inet->daddr != rmt_addr) || 304 (inet->dport != rmt_port && inet->dport) || 305 (inet->rcv_saddr && inet->rcv_saddr != loc_addr) || 306 ipv6_only_sock(s) || 307 (s->sk_bound_dev_if && s->sk_bound_dev_if != dif)) 308 continue; 309 if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif)) 310 continue; 311 goto found; 312 } 313 s = NULL; 314 found: 315 return s; 316 } 317 318 /* 319 * This routine is called by the ICMP module when it gets some 320 * sort of error condition. If err < 0 then the socket should 321 * be closed and the error returned to the user. If err > 0 322 * it's just the icmp type << 8 | icmp code. 323 * Header points to the ip header of the error packet. We move 324 * on past this. Then (as it used to claim before adjustment) 325 * header points to the first 8 bytes of the udp header. We need 326 * to find the appropriate port. 327 */ 328 329 void __udp4_lib_err(struct sk_buff *skb, u32 info, struct hlist_head udptable[]) 330 { 331 struct inet_sock *inet; 332 struct iphdr *iph = (struct iphdr*)skb->data; 333 struct udphdr *uh = (struct udphdr*)(skb->data+(iph->ihl<<2)); 334 const int type = icmp_hdr(skb)->type; 335 const int code = icmp_hdr(skb)->code; 336 struct sock *sk; 337 int harderr; 338 int err; 339 340 sk = __udp4_lib_lookup(iph->daddr, uh->dest, iph->saddr, uh->source, 341 skb->dev->ifindex, udptable ); 342 if (sk == NULL) { 343 ICMP_INC_STATS_BH(ICMP_MIB_INERRORS); 344 return; /* No socket for error */ 345 } 346 347 err = 0; 348 harderr = 0; 349 inet = inet_sk(sk); 350 351 switch (type) { 352 default: 353 case ICMP_TIME_EXCEEDED: 354 err = EHOSTUNREACH; 355 break; 356 case ICMP_SOURCE_QUENCH: 357 goto out; 358 case ICMP_PARAMETERPROB: 359 err = EPROTO; 360 harderr = 1; 361 break; 362 case ICMP_DEST_UNREACH: 363 if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */ 364 if (inet->pmtudisc != IP_PMTUDISC_DONT) { 365 err = EMSGSIZE; 366 harderr = 1; 367 break; 368 } 369 goto out; 370 } 371 err = EHOSTUNREACH; 372 if (code <= NR_ICMP_UNREACH) { 373 harderr = icmp_err_convert[code].fatal; 374 err = icmp_err_convert[code].errno; 375 } 376 break; 377 } 378 379 /* 380 * RFC1122: OK. Passes ICMP errors back to application, as per 381 * 4.1.3.3. 382 */ 383 if (!inet->recverr) { 384 if (!harderr || sk->sk_state != TCP_ESTABLISHED) 385 goto out; 386 } else { 387 ip_icmp_error(sk, skb, err, uh->dest, info, (u8*)(uh+1)); 388 } 389 sk->sk_err = err; 390 sk->sk_error_report(sk); 391 out: 392 sock_put(sk); 393 } 394 395 void udp_err(struct sk_buff *skb, u32 info) 396 { 397 return __udp4_lib_err(skb, info, udp_hash); 398 } 399 400 /* 401 * Throw away all pending data and cancel the corking. Socket is locked. 402 */ 403 static void udp_flush_pending_frames(struct sock *sk) 404 { 405 struct udp_sock *up = udp_sk(sk); 406 407 if (up->pending) { 408 up->len = 0; 409 up->pending = 0; 410 ip_flush_pending_frames(sk); 411 } 412 } 413 414 /** 415 * udp4_hwcsum_outgoing - handle outgoing HW checksumming 416 * @sk: socket we are sending on 417 * @skb: sk_buff containing the filled-in UDP header 418 * (checksum field must be zeroed out) 419 */ 420 static void udp4_hwcsum_outgoing(struct sock *sk, struct sk_buff *skb, 421 __be32 src, __be32 dst, int len ) 422 { 423 unsigned int offset; 424 struct udphdr *uh = udp_hdr(skb); 425 __wsum csum = 0; 426 427 if (skb_queue_len(&sk->sk_write_queue) == 1) { 428 /* 429 * Only one fragment on the socket. 430 */ 431 skb->csum_start = skb_transport_header(skb) - skb->head; 432 skb->csum_offset = offsetof(struct udphdr, check); 433 uh->check = ~csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, 0); 434 } else { 435 /* 436 * HW-checksum won't work as there are two or more 437 * fragments on the socket so that all csums of sk_buffs 438 * should be together 439 */ 440 offset = skb_transport_offset(skb); 441 skb->csum = skb_checksum(skb, offset, skb->len - offset, 0); 442 443 skb->ip_summed = CHECKSUM_NONE; 444 445 skb_queue_walk(&sk->sk_write_queue, skb) { 446 csum = csum_add(csum, skb->csum); 447 } 448 449 uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum); 450 if (uh->check == 0) 451 uh->check = CSUM_MANGLED_0; 452 } 453 } 454 455 /* 456 * Push out all pending data as one UDP datagram. Socket is locked. 457 */ 458 static int udp_push_pending_frames(struct sock *sk) 459 { 460 struct udp_sock *up = udp_sk(sk); 461 struct inet_sock *inet = inet_sk(sk); 462 struct flowi *fl = &inet->cork.fl; 463 struct sk_buff *skb; 464 struct udphdr *uh; 465 int err = 0; 466 __wsum csum = 0; 467 468 /* Grab the skbuff where UDP header space exists. */ 469 if ((skb = skb_peek(&sk->sk_write_queue)) == NULL) 470 goto out; 471 472 /* 473 * Create a UDP header 474 */ 475 uh = udp_hdr(skb); 476 uh->source = fl->fl_ip_sport; 477 uh->dest = fl->fl_ip_dport; 478 uh->len = htons(up->len); 479 uh->check = 0; 480 481 if (up->pcflag) /* UDP-Lite */ 482 csum = udplite_csum_outgoing(sk, skb); 483 484 else if (sk->sk_no_check == UDP_CSUM_NOXMIT) { /* UDP csum disabled */ 485 486 skb->ip_summed = CHECKSUM_NONE; 487 goto send; 488 489 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */ 490 491 udp4_hwcsum_outgoing(sk, skb, fl->fl4_src,fl->fl4_dst, up->len); 492 goto send; 493 494 } else /* `normal' UDP */ 495 csum = udp_csum_outgoing(sk, skb); 496 497 /* add protocol-dependent pseudo-header */ 498 uh->check = csum_tcpudp_magic(fl->fl4_src, fl->fl4_dst, up->len, 499 sk->sk_protocol, csum ); 500 if (uh->check == 0) 501 uh->check = CSUM_MANGLED_0; 502 503 send: 504 err = ip_push_pending_frames(sk); 505 out: 506 up->len = 0; 507 up->pending = 0; 508 return err; 509 } 510 511 int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 512 size_t len) 513 { 514 struct inet_sock *inet = inet_sk(sk); 515 struct udp_sock *up = udp_sk(sk); 516 int ulen = len; 517 struct ipcm_cookie ipc; 518 struct rtable *rt = NULL; 519 int free = 0; 520 int connected = 0; 521 __be32 daddr, faddr, saddr; 522 __be16 dport; 523 u8 tos; 524 int err, is_udplite = up->pcflag; 525 int corkreq = up->corkflag || msg->msg_flags&MSG_MORE; 526 int (*getfrag)(void *, char *, int, int, int, struct sk_buff *); 527 528 if (len > 0xFFFF) 529 return -EMSGSIZE; 530 531 /* 532 * Check the flags. 533 */ 534 535 if (msg->msg_flags&MSG_OOB) /* Mirror BSD error message compatibility */ 536 return -EOPNOTSUPP; 537 538 ipc.opt = NULL; 539 540 if (up->pending) { 541 /* 542 * There are pending frames. 543 * The socket lock must be held while it's corked. 544 */ 545 lock_sock(sk); 546 if (likely(up->pending)) { 547 if (unlikely(up->pending != AF_INET)) { 548 release_sock(sk); 549 return -EINVAL; 550 } 551 goto do_append_data; 552 } 553 release_sock(sk); 554 } 555 ulen += sizeof(struct udphdr); 556 557 /* 558 * Get and verify the address. 559 */ 560 if (msg->msg_name) { 561 struct sockaddr_in * usin = (struct sockaddr_in*)msg->msg_name; 562 if (msg->msg_namelen < sizeof(*usin)) 563 return -EINVAL; 564 if (usin->sin_family != AF_INET) { 565 if (usin->sin_family != AF_UNSPEC) 566 return -EAFNOSUPPORT; 567 } 568 569 daddr = usin->sin_addr.s_addr; 570 dport = usin->sin_port; 571 if (dport == 0) 572 return -EINVAL; 573 } else { 574 if (sk->sk_state != TCP_ESTABLISHED) 575 return -EDESTADDRREQ; 576 daddr = inet->daddr; 577 dport = inet->dport; 578 /* Open fast path for connected socket. 579 Route will not be used, if at least one option is set. 580 */ 581 connected = 1; 582 } 583 ipc.addr = inet->saddr; 584 585 ipc.oif = sk->sk_bound_dev_if; 586 if (msg->msg_controllen) { 587 err = ip_cmsg_send(msg, &ipc); 588 if (err) 589 return err; 590 if (ipc.opt) 591 free = 1; 592 connected = 0; 593 } 594 if (!ipc.opt) 595 ipc.opt = inet->opt; 596 597 saddr = ipc.addr; 598 ipc.addr = faddr = daddr; 599 600 if (ipc.opt && ipc.opt->srr) { 601 if (!daddr) 602 return -EINVAL; 603 faddr = ipc.opt->faddr; 604 connected = 0; 605 } 606 tos = RT_TOS(inet->tos); 607 if (sock_flag(sk, SOCK_LOCALROUTE) || 608 (msg->msg_flags & MSG_DONTROUTE) || 609 (ipc.opt && ipc.opt->is_strictroute)) { 610 tos |= RTO_ONLINK; 611 connected = 0; 612 } 613 614 if (MULTICAST(daddr)) { 615 if (!ipc.oif) 616 ipc.oif = inet->mc_index; 617 if (!saddr) 618 saddr = inet->mc_addr; 619 connected = 0; 620 } 621 622 if (connected) 623 rt = (struct rtable*)sk_dst_check(sk, 0); 624 625 if (rt == NULL) { 626 struct flowi fl = { .oif = ipc.oif, 627 .nl_u = { .ip4_u = 628 { .daddr = faddr, 629 .saddr = saddr, 630 .tos = tos } }, 631 .proto = sk->sk_protocol, 632 .uli_u = { .ports = 633 { .sport = inet->sport, 634 .dport = dport } } }; 635 security_sk_classify_flow(sk, &fl); 636 err = ip_route_output_flow(&rt, &fl, sk, 1); 637 if (err) { 638 if (err == -ENETUNREACH) 639 IP_INC_STATS_BH(IPSTATS_MIB_OUTNOROUTES); 640 goto out; 641 } 642 643 err = -EACCES; 644 if ((rt->rt_flags & RTCF_BROADCAST) && 645 !sock_flag(sk, SOCK_BROADCAST)) 646 goto out; 647 if (connected) 648 sk_dst_set(sk, dst_clone(&rt->u.dst)); 649 } 650 651 if (msg->msg_flags&MSG_CONFIRM) 652 goto do_confirm; 653 back_from_confirm: 654 655 saddr = rt->rt_src; 656 if (!ipc.addr) 657 daddr = ipc.addr = rt->rt_dst; 658 659 lock_sock(sk); 660 if (unlikely(up->pending)) { 661 /* The socket is already corked while preparing it. */ 662 /* ... which is an evident application bug. --ANK */ 663 release_sock(sk); 664 665 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 2\n"); 666 err = -EINVAL; 667 goto out; 668 } 669 /* 670 * Now cork the socket to pend data. 671 */ 672 inet->cork.fl.fl4_dst = daddr; 673 inet->cork.fl.fl_ip_dport = dport; 674 inet->cork.fl.fl4_src = saddr; 675 inet->cork.fl.fl_ip_sport = inet->sport; 676 up->pending = AF_INET; 677 678 do_append_data: 679 up->len += ulen; 680 getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag; 681 err = ip_append_data(sk, getfrag, msg->msg_iov, ulen, 682 sizeof(struct udphdr), &ipc, rt, 683 corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags); 684 if (err) 685 udp_flush_pending_frames(sk); 686 else if (!corkreq) 687 err = udp_push_pending_frames(sk); 688 else if (unlikely(skb_queue_empty(&sk->sk_write_queue))) 689 up->pending = 0; 690 release_sock(sk); 691 692 out: 693 ip_rt_put(rt); 694 if (free) 695 kfree(ipc.opt); 696 if (!err) { 697 UDP_INC_STATS_USER(UDP_MIB_OUTDATAGRAMS, is_udplite); 698 return len; 699 } 700 /* 701 * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space. Reporting 702 * ENOBUFS might not be good (it's not tunable per se), but otherwise 703 * we don't have a good statistic (IpOutDiscards but it can be too many 704 * things). We could add another new stat but at least for now that 705 * seems like overkill. 706 */ 707 if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) { 708 UDP_INC_STATS_USER(UDP_MIB_SNDBUFERRORS, is_udplite); 709 } 710 return err; 711 712 do_confirm: 713 dst_confirm(&rt->u.dst); 714 if (!(msg->msg_flags&MSG_PROBE) || len) 715 goto back_from_confirm; 716 err = 0; 717 goto out; 718 } 719 720 int udp_sendpage(struct sock *sk, struct page *page, int offset, 721 size_t size, int flags) 722 { 723 struct udp_sock *up = udp_sk(sk); 724 int ret; 725 726 if (!up->pending) { 727 struct msghdr msg = { .msg_flags = flags|MSG_MORE }; 728 729 /* Call udp_sendmsg to specify destination address which 730 * sendpage interface can't pass. 731 * This will succeed only when the socket is connected. 732 */ 733 ret = udp_sendmsg(NULL, sk, &msg, 0); 734 if (ret < 0) 735 return ret; 736 } 737 738 lock_sock(sk); 739 740 if (unlikely(!up->pending)) { 741 release_sock(sk); 742 743 LIMIT_NETDEBUG(KERN_DEBUG "udp cork app bug 3\n"); 744 return -EINVAL; 745 } 746 747 ret = ip_append_page(sk, page, offset, size, flags); 748 if (ret == -EOPNOTSUPP) { 749 release_sock(sk); 750 return sock_no_sendpage(sk->sk_socket, page, offset, 751 size, flags); 752 } 753 if (ret < 0) { 754 udp_flush_pending_frames(sk); 755 goto out; 756 } 757 758 up->len += size; 759 if (!(up->corkflag || (flags&MSG_MORE))) 760 ret = udp_push_pending_frames(sk); 761 if (!ret) 762 ret = size; 763 out: 764 release_sock(sk); 765 return ret; 766 } 767 768 /* 769 * IOCTL requests applicable to the UDP protocol 770 */ 771 772 int udp_ioctl(struct sock *sk, int cmd, unsigned long arg) 773 { 774 switch (cmd) { 775 case SIOCOUTQ: 776 { 777 int amount = atomic_read(&sk->sk_wmem_alloc); 778 return put_user(amount, (int __user *)arg); 779 } 780 781 case SIOCINQ: 782 { 783 struct sk_buff *skb; 784 unsigned long amount; 785 786 amount = 0; 787 spin_lock_bh(&sk->sk_receive_queue.lock); 788 skb = skb_peek(&sk->sk_receive_queue); 789 if (skb != NULL) { 790 /* 791 * We will only return the amount 792 * of this packet since that is all 793 * that will be read. 794 */ 795 amount = skb->len - sizeof(struct udphdr); 796 } 797 spin_unlock_bh(&sk->sk_receive_queue.lock); 798 return put_user(amount, (int __user *)arg); 799 } 800 801 default: 802 return -ENOIOCTLCMD; 803 } 804 805 return 0; 806 } 807 808 /* 809 * This should be easy, if there is something there we 810 * return it, otherwise we block. 811 */ 812 813 int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg, 814 size_t len, int noblock, int flags, int *addr_len) 815 { 816 struct inet_sock *inet = inet_sk(sk); 817 struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name; 818 struct sk_buff *skb; 819 unsigned int ulen, copied; 820 int err; 821 int is_udplite = IS_UDPLITE(sk); 822 823 /* 824 * Check any passed addresses 825 */ 826 if (addr_len) 827 *addr_len=sizeof(*sin); 828 829 if (flags & MSG_ERRQUEUE) 830 return ip_recv_error(sk, msg, len); 831 832 try_again: 833 skb = skb_recv_datagram(sk, flags, noblock, &err); 834 if (!skb) 835 goto out; 836 837 ulen = skb->len - sizeof(struct udphdr); 838 copied = len; 839 if (copied > ulen) 840 copied = ulen; 841 else if (copied < ulen) 842 msg->msg_flags |= MSG_TRUNC; 843 844 /* 845 * If checksum is needed at all, try to do it while copying the 846 * data. If the data is truncated, or if we only want a partial 847 * coverage checksum (UDP-Lite), do it before the copy. 848 */ 849 850 if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) { 851 if (udp_lib_checksum_complete(skb)) 852 goto csum_copy_err; 853 } 854 855 if (skb_csum_unnecessary(skb)) 856 err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr), 857 msg->msg_iov, copied ); 858 else { 859 err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov); 860 861 if (err == -EINVAL) 862 goto csum_copy_err; 863 } 864 865 if (err) 866 goto out_free; 867 868 sock_recv_timestamp(msg, sk, skb); 869 870 /* Copy the address. */ 871 if (sin) 872 { 873 sin->sin_family = AF_INET; 874 sin->sin_port = udp_hdr(skb)->source; 875 sin->sin_addr.s_addr = ip_hdr(skb)->saddr; 876 memset(sin->sin_zero, 0, sizeof(sin->sin_zero)); 877 } 878 if (inet->cmsg_flags) 879 ip_cmsg_recv(msg, skb); 880 881 err = copied; 882 if (flags & MSG_TRUNC) 883 err = ulen; 884 885 out_free: 886 skb_free_datagram(sk, skb); 887 out: 888 return err; 889 890 csum_copy_err: 891 UDP_INC_STATS_BH(UDP_MIB_INERRORS, is_udplite); 892 893 skb_kill_datagram(sk, skb, flags); 894 895 if (noblock) 896 return -EAGAIN; 897 goto try_again; 898 } 899 900 901 int udp_disconnect(struct sock *sk, int flags) 902 { 903 struct inet_sock *inet = inet_sk(sk); 904 /* 905 * 1003.1g - break association. 906 */ 907 908 sk->sk_state = TCP_CLOSE; 909 inet->daddr = 0; 910 inet->dport = 0; 911 sk->sk_bound_dev_if = 0; 912 if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK)) 913 inet_reset_saddr(sk); 914 915 if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) { 916 sk->sk_prot->unhash(sk); 917 inet->sport = 0; 918 } 919 sk_dst_reset(sk); 920 return 0; 921 } 922 923 /* returns: 924 * -1: error 925 * 0: success 926 * >0: "udp encap" protocol resubmission 927 * 928 * Note that in the success and error cases, the skb is assumed to 929 * have either been requeued or freed. 930 */ 931 int udp_queue_rcv_skb(struct sock * sk, struct sk_buff *skb) 932 { 933 struct udp_sock *up = udp_sk(sk); 934 int rc; 935 936 /* 937 * Charge it to the socket, dropping if the queue is full. 938 */ 939 if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb)) 940 goto drop; 941 nf_reset(skb); 942 943 if (up->encap_type) { 944 /* 945 * This is an encapsulation socket so pass the skb to 946 * the socket's udp_encap_rcv() hook. Otherwise, just 947 * fall through and pass this up the UDP socket. 948 * up->encap_rcv() returns the following value: 949 * =0 if skb was successfully passed to the encap 950 * handler or was discarded by it. 951 * >0 if skb should be passed on to UDP. 952 * <0 if skb should be resubmitted as proto -N 953 */ 954 955 /* if we're overly short, let UDP handle it */ 956 if (skb->len > sizeof(struct udphdr) && 957 up->encap_rcv != NULL) { 958 int ret; 959 960 ret = (*up->encap_rcv)(sk, skb); 961 if (ret <= 0) { 962 UDP_INC_STATS_BH(UDP_MIB_INDATAGRAMS, up->pcflag); 963 return -ret; 964 } 965 } 966 967 /* FALLTHROUGH -- it's a UDP Packet */ 968 } 969 970 /* 971 * UDP-Lite specific tests, ignored on UDP sockets 972 */ 973 if ((up->pcflag & UDPLITE_RECV_CC) && UDP_SKB_CB(skb)->partial_cov) { 974 975 /* 976 * MIB statistics other than incrementing the error count are 977 * disabled for the following two types of errors: these depend 978 * on the application settings, not on the functioning of the 979 * protocol stack as such. 980 * 981 * RFC 3828 here recommends (sec 3.3): "There should also be a 982 * way ... to ... at least let the receiving application block 983 * delivery of packets with coverage values less than a value 984 * provided by the application." 985 */ 986 if (up->pcrlen == 0) { /* full coverage was set */ 987 LIMIT_NETDEBUG(KERN_WARNING "UDPLITE: partial coverage " 988 "%d while full coverage %d requested\n", 989 UDP_SKB_CB(skb)->cscov, skb->len); 990 goto drop; 991 } 992 /* The next case involves violating the min. coverage requested 993 * by the receiver. This is subtle: if receiver wants x and x is 994 * greater than the buffersize/MTU then receiver will complain 995 * that it wants x while sender emits packets of smaller size y. 996 * Therefore the above ...()->partial_cov statement is essential. 997 */ 998 if (UDP_SKB_CB(skb)->cscov < up->pcrlen) { 999 LIMIT_NETDEBUG(KERN_WARNING 1000 "UDPLITE: coverage %d too small, need min %d\n", 1001 UDP_SKB_CB(skb)->cscov, up->pcrlen); 1002 goto drop; 1003 } 1004 } 1005 1006 if (sk->sk_filter) { 1007 if (udp_lib_checksum_complete(skb)) 1008 goto drop; 1009 } 1010 1011 if ((rc = sock_queue_rcv_skb(sk,skb)) < 0) { 1012 /* Note that an ENOMEM error is charged twice */ 1013 if (rc == -ENOMEM) 1014 UDP_INC_STATS_BH(UDP_MIB_RCVBUFERRORS, up->pcflag); 1015 goto drop; 1016 } 1017 1018 UDP_INC_STATS_BH(UDP_MIB_INDATAGRAMS, up->pcflag); 1019 return 0; 1020 1021 drop: 1022 UDP_INC_STATS_BH(UDP_MIB_INERRORS, up->pcflag); 1023 kfree_skb(skb); 1024 return -1; 1025 } 1026 1027 /* 1028 * Multicasts and broadcasts go to each listener. 1029 * 1030 * Note: called only from the BH handler context, 1031 * so we don't need to lock the hashes. 1032 */ 1033 static int __udp4_lib_mcast_deliver(struct sk_buff *skb, 1034 struct udphdr *uh, 1035 __be32 saddr, __be32 daddr, 1036 struct hlist_head udptable[]) 1037 { 1038 struct sock *sk; 1039 int dif; 1040 1041 read_lock(&udp_hash_lock); 1042 sk = sk_head(&udptable[ntohs(uh->dest) & (UDP_HTABLE_SIZE - 1)]); 1043 dif = skb->dev->ifindex; 1044 sk = udp_v4_mcast_next(sk, uh->dest, daddr, uh->source, saddr, dif); 1045 if (sk) { 1046 struct sock *sknext = NULL; 1047 1048 do { 1049 struct sk_buff *skb1 = skb; 1050 1051 sknext = udp_v4_mcast_next(sk_next(sk), uh->dest, daddr, 1052 uh->source, saddr, dif); 1053 if (sknext) 1054 skb1 = skb_clone(skb, GFP_ATOMIC); 1055 1056 if (skb1) { 1057 int ret = udp_queue_rcv_skb(sk, skb1); 1058 if (ret > 0) 1059 /* we should probably re-process instead 1060 * of dropping packets here. */ 1061 kfree_skb(skb1); 1062 } 1063 sk = sknext; 1064 } while (sknext); 1065 } else 1066 kfree_skb(skb); 1067 read_unlock(&udp_hash_lock); 1068 return 0; 1069 } 1070 1071 /* Initialize UDP checksum. If exited with zero value (success), 1072 * CHECKSUM_UNNECESSARY means, that no more checks are required. 1073 * Otherwise, csum completion requires chacksumming packet body, 1074 * including udp header and folding it to skb->csum. 1075 */ 1076 static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh, 1077 int proto) 1078 { 1079 const struct iphdr *iph; 1080 int err; 1081 1082 UDP_SKB_CB(skb)->partial_cov = 0; 1083 UDP_SKB_CB(skb)->cscov = skb->len; 1084 1085 if (proto == IPPROTO_UDPLITE) { 1086 err = udplite_checksum_init(skb, uh); 1087 if (err) 1088 return err; 1089 } 1090 1091 iph = ip_hdr(skb); 1092 if (uh->check == 0) { 1093 skb->ip_summed = CHECKSUM_UNNECESSARY; 1094 } else if (skb->ip_summed == CHECKSUM_COMPLETE) { 1095 if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len, 1096 proto, skb->csum)) 1097 skb->ip_summed = CHECKSUM_UNNECESSARY; 1098 } 1099 if (!skb_csum_unnecessary(skb)) 1100 skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr, 1101 skb->len, proto, 0); 1102 /* Probably, we should checksum udp header (it should be in cache 1103 * in any case) and data in tiny packets (< rx copybreak). 1104 */ 1105 1106 return 0; 1107 } 1108 1109 /* 1110 * All we need to do is get the socket, and then do a checksum. 1111 */ 1112 1113 int __udp4_lib_rcv(struct sk_buff *skb, struct hlist_head udptable[], 1114 int proto) 1115 { 1116 struct sock *sk; 1117 struct udphdr *uh = udp_hdr(skb); 1118 unsigned short ulen; 1119 struct rtable *rt = (struct rtable*)skb->dst; 1120 __be32 saddr = ip_hdr(skb)->saddr; 1121 __be32 daddr = ip_hdr(skb)->daddr; 1122 1123 /* 1124 * Validate the packet. 1125 */ 1126 if (!pskb_may_pull(skb, sizeof(struct udphdr))) 1127 goto drop; /* No space for header. */ 1128 1129 ulen = ntohs(uh->len); 1130 if (ulen > skb->len) 1131 goto short_packet; 1132 1133 if (proto == IPPROTO_UDP) { 1134 /* UDP validates ulen. */ 1135 if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen)) 1136 goto short_packet; 1137 uh = udp_hdr(skb); 1138 } 1139 1140 if (udp4_csum_init(skb, uh, proto)) 1141 goto csum_error; 1142 1143 if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST)) 1144 return __udp4_lib_mcast_deliver(skb, uh, saddr, daddr, udptable); 1145 1146 sk = __udp4_lib_lookup(saddr, uh->source, daddr, uh->dest, 1147 skb->dev->ifindex, udptable ); 1148 1149 if (sk != NULL) { 1150 int ret = udp_queue_rcv_skb(sk, skb); 1151 sock_put(sk); 1152 1153 /* a return value > 0 means to resubmit the input, but 1154 * it wants the return to be -protocol, or 0 1155 */ 1156 if (ret > 0) 1157 return -ret; 1158 return 0; 1159 } 1160 1161 if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb)) 1162 goto drop; 1163 nf_reset(skb); 1164 1165 /* No socket. Drop packet silently, if checksum is wrong */ 1166 if (udp_lib_checksum_complete(skb)) 1167 goto csum_error; 1168 1169 UDP_INC_STATS_BH(UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE); 1170 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0); 1171 1172 /* 1173 * Hmm. We got an UDP packet to a port to which we 1174 * don't wanna listen. Ignore it. 1175 */ 1176 kfree_skb(skb); 1177 return 0; 1178 1179 short_packet: 1180 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %u.%u.%u.%u:%u %d/%d to %u.%u.%u.%u:%u\n", 1181 proto == IPPROTO_UDPLITE ? "-Lite" : "", 1182 NIPQUAD(saddr), 1183 ntohs(uh->source), 1184 ulen, 1185 skb->len, 1186 NIPQUAD(daddr), 1187 ntohs(uh->dest)); 1188 goto drop; 1189 1190 csum_error: 1191 /* 1192 * RFC1122: OK. Discards the bad packet silently (as far as 1193 * the network is concerned, anyway) as per 4.1.3.4 (MUST). 1194 */ 1195 LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %d.%d.%d.%d:%d to %d.%d.%d.%d:%d ulen %d\n", 1196 proto == IPPROTO_UDPLITE ? "-Lite" : "", 1197 NIPQUAD(saddr), 1198 ntohs(uh->source), 1199 NIPQUAD(daddr), 1200 ntohs(uh->dest), 1201 ulen); 1202 drop: 1203 UDP_INC_STATS_BH(UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE); 1204 kfree_skb(skb); 1205 return 0; 1206 } 1207 1208 int udp_rcv(struct sk_buff *skb) 1209 { 1210 return __udp4_lib_rcv(skb, udp_hash, IPPROTO_UDP); 1211 } 1212 1213 int udp_destroy_sock(struct sock *sk) 1214 { 1215 lock_sock(sk); 1216 udp_flush_pending_frames(sk); 1217 release_sock(sk); 1218 return 0; 1219 } 1220 1221 /* 1222 * Socket option code for UDP 1223 */ 1224 int udp_lib_setsockopt(struct sock *sk, int level, int optname, 1225 char __user *optval, int optlen, 1226 int (*push_pending_frames)(struct sock *)) 1227 { 1228 struct udp_sock *up = udp_sk(sk); 1229 int val; 1230 int err = 0; 1231 1232 if (optlen<sizeof(int)) 1233 return -EINVAL; 1234 1235 if (get_user(val, (int __user *)optval)) 1236 return -EFAULT; 1237 1238 switch (optname) { 1239 case UDP_CORK: 1240 if (val != 0) { 1241 up->corkflag = 1; 1242 } else { 1243 up->corkflag = 0; 1244 lock_sock(sk); 1245 (*push_pending_frames)(sk); 1246 release_sock(sk); 1247 } 1248 break; 1249 1250 case UDP_ENCAP: 1251 switch (val) { 1252 case 0: 1253 case UDP_ENCAP_ESPINUDP: 1254 case UDP_ENCAP_ESPINUDP_NON_IKE: 1255 up->encap_rcv = xfrm4_udp_encap_rcv; 1256 /* FALLTHROUGH */ 1257 case UDP_ENCAP_L2TPINUDP: 1258 up->encap_type = val; 1259 break; 1260 default: 1261 err = -ENOPROTOOPT; 1262 break; 1263 } 1264 break; 1265 1266 /* 1267 * UDP-Lite's partial checksum coverage (RFC 3828). 1268 */ 1269 /* The sender sets actual checksum coverage length via this option. 1270 * The case coverage > packet length is handled by send module. */ 1271 case UDPLITE_SEND_CSCOV: 1272 if (!up->pcflag) /* Disable the option on UDP sockets */ 1273 return -ENOPROTOOPT; 1274 if (val != 0 && val < 8) /* Illegal coverage: use default (8) */ 1275 val = 8; 1276 up->pcslen = val; 1277 up->pcflag |= UDPLITE_SEND_CC; 1278 break; 1279 1280 /* The receiver specifies a minimum checksum coverage value. To make 1281 * sense, this should be set to at least 8 (as done below). If zero is 1282 * used, this again means full checksum coverage. */ 1283 case UDPLITE_RECV_CSCOV: 1284 if (!up->pcflag) /* Disable the option on UDP sockets */ 1285 return -ENOPROTOOPT; 1286 if (val != 0 && val < 8) /* Avoid silly minimal values. */ 1287 val = 8; 1288 up->pcrlen = val; 1289 up->pcflag |= UDPLITE_RECV_CC; 1290 break; 1291 1292 default: 1293 err = -ENOPROTOOPT; 1294 break; 1295 } 1296 1297 return err; 1298 } 1299 1300 int udp_setsockopt(struct sock *sk, int level, int optname, 1301 char __user *optval, int optlen) 1302 { 1303 if (level == SOL_UDP || level == SOL_UDPLITE) 1304 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1305 udp_push_pending_frames); 1306 return ip_setsockopt(sk, level, optname, optval, optlen); 1307 } 1308 1309 #ifdef CONFIG_COMPAT 1310 int compat_udp_setsockopt(struct sock *sk, int level, int optname, 1311 char __user *optval, int optlen) 1312 { 1313 if (level == SOL_UDP || level == SOL_UDPLITE) 1314 return udp_lib_setsockopt(sk, level, optname, optval, optlen, 1315 udp_push_pending_frames); 1316 return compat_ip_setsockopt(sk, level, optname, optval, optlen); 1317 } 1318 #endif 1319 1320 int udp_lib_getsockopt(struct sock *sk, int level, int optname, 1321 char __user *optval, int __user *optlen) 1322 { 1323 struct udp_sock *up = udp_sk(sk); 1324 int val, len; 1325 1326 if (get_user(len,optlen)) 1327 return -EFAULT; 1328 1329 len = min_t(unsigned int, len, sizeof(int)); 1330 1331 if (len < 0) 1332 return -EINVAL; 1333 1334 switch (optname) { 1335 case UDP_CORK: 1336 val = up->corkflag; 1337 break; 1338 1339 case UDP_ENCAP: 1340 val = up->encap_type; 1341 break; 1342 1343 /* The following two cannot be changed on UDP sockets, the return is 1344 * always 0 (which corresponds to the full checksum coverage of UDP). */ 1345 case UDPLITE_SEND_CSCOV: 1346 val = up->pcslen; 1347 break; 1348 1349 case UDPLITE_RECV_CSCOV: 1350 val = up->pcrlen; 1351 break; 1352 1353 default: 1354 return -ENOPROTOOPT; 1355 } 1356 1357 if (put_user(len, optlen)) 1358 return -EFAULT; 1359 if (copy_to_user(optval, &val,len)) 1360 return -EFAULT; 1361 return 0; 1362 } 1363 1364 int udp_getsockopt(struct sock *sk, int level, int optname, 1365 char __user *optval, int __user *optlen) 1366 { 1367 if (level == SOL_UDP || level == SOL_UDPLITE) 1368 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1369 return ip_getsockopt(sk, level, optname, optval, optlen); 1370 } 1371 1372 #ifdef CONFIG_COMPAT 1373 int compat_udp_getsockopt(struct sock *sk, int level, int optname, 1374 char __user *optval, int __user *optlen) 1375 { 1376 if (level == SOL_UDP || level == SOL_UDPLITE) 1377 return udp_lib_getsockopt(sk, level, optname, optval, optlen); 1378 return compat_ip_getsockopt(sk, level, optname, optval, optlen); 1379 } 1380 #endif 1381 /** 1382 * udp_poll - wait for a UDP event. 1383 * @file - file struct 1384 * @sock - socket 1385 * @wait - poll table 1386 * 1387 * This is same as datagram poll, except for the special case of 1388 * blocking sockets. If application is using a blocking fd 1389 * and a packet with checksum error is in the queue; 1390 * then it could get return from select indicating data available 1391 * but then block when reading it. Add special case code 1392 * to work around these arguably broken applications. 1393 */ 1394 unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait) 1395 { 1396 unsigned int mask = datagram_poll(file, sock, wait); 1397 struct sock *sk = sock->sk; 1398 int is_lite = IS_UDPLITE(sk); 1399 1400 /* Check for false positives due to checksum errors */ 1401 if ( (mask & POLLRDNORM) && 1402 !(file->f_flags & O_NONBLOCK) && 1403 !(sk->sk_shutdown & RCV_SHUTDOWN)){ 1404 struct sk_buff_head *rcvq = &sk->sk_receive_queue; 1405 struct sk_buff *skb; 1406 1407 spin_lock_bh(&rcvq->lock); 1408 while ((skb = skb_peek(rcvq)) != NULL && 1409 udp_lib_checksum_complete(skb)) { 1410 UDP_INC_STATS_BH(UDP_MIB_INERRORS, is_lite); 1411 __skb_unlink(skb, rcvq); 1412 kfree_skb(skb); 1413 } 1414 spin_unlock_bh(&rcvq->lock); 1415 1416 /* nothing to see, move along */ 1417 if (skb == NULL) 1418 mask &= ~(POLLIN | POLLRDNORM); 1419 } 1420 1421 return mask; 1422 1423 } 1424 1425 struct proto udp_prot = { 1426 .name = "UDP", 1427 .owner = THIS_MODULE, 1428 .close = udp_lib_close, 1429 .connect = ip4_datagram_connect, 1430 .disconnect = udp_disconnect, 1431 .ioctl = udp_ioctl, 1432 .destroy = udp_destroy_sock, 1433 .setsockopt = udp_setsockopt, 1434 .getsockopt = udp_getsockopt, 1435 .sendmsg = udp_sendmsg, 1436 .recvmsg = udp_recvmsg, 1437 .sendpage = udp_sendpage, 1438 .backlog_rcv = udp_queue_rcv_skb, 1439 .hash = udp_lib_hash, 1440 .unhash = udp_lib_unhash, 1441 .get_port = udp_v4_get_port, 1442 .obj_size = sizeof(struct udp_sock), 1443 #ifdef CONFIG_COMPAT 1444 .compat_setsockopt = compat_udp_setsockopt, 1445 .compat_getsockopt = compat_udp_getsockopt, 1446 #endif 1447 }; 1448 1449 /* ------------------------------------------------------------------------ */ 1450 #ifdef CONFIG_PROC_FS 1451 1452 static struct sock *udp_get_first(struct seq_file *seq) 1453 { 1454 struct sock *sk; 1455 struct udp_iter_state *state = seq->private; 1456 1457 for (state->bucket = 0; state->bucket < UDP_HTABLE_SIZE; ++state->bucket) { 1458 struct hlist_node *node; 1459 sk_for_each(sk, node, state->hashtable + state->bucket) { 1460 if (sk->sk_family == state->family) 1461 goto found; 1462 } 1463 } 1464 sk = NULL; 1465 found: 1466 return sk; 1467 } 1468 1469 static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk) 1470 { 1471 struct udp_iter_state *state = seq->private; 1472 1473 do { 1474 sk = sk_next(sk); 1475 try_again: 1476 ; 1477 } while (sk && sk->sk_family != state->family); 1478 1479 if (!sk && ++state->bucket < UDP_HTABLE_SIZE) { 1480 sk = sk_head(state->hashtable + state->bucket); 1481 goto try_again; 1482 } 1483 return sk; 1484 } 1485 1486 static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos) 1487 { 1488 struct sock *sk = udp_get_first(seq); 1489 1490 if (sk) 1491 while (pos && (sk = udp_get_next(seq, sk)) != NULL) 1492 --pos; 1493 return pos ? NULL : sk; 1494 } 1495 1496 static void *udp_seq_start(struct seq_file *seq, loff_t *pos) 1497 { 1498 read_lock(&udp_hash_lock); 1499 return *pos ? udp_get_idx(seq, *pos-1) : (void *)1; 1500 } 1501 1502 static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos) 1503 { 1504 struct sock *sk; 1505 1506 if (v == (void *)1) 1507 sk = udp_get_idx(seq, 0); 1508 else 1509 sk = udp_get_next(seq, v); 1510 1511 ++*pos; 1512 return sk; 1513 } 1514 1515 static void udp_seq_stop(struct seq_file *seq, void *v) 1516 { 1517 read_unlock(&udp_hash_lock); 1518 } 1519 1520 static int udp_seq_open(struct inode *inode, struct file *file) 1521 { 1522 struct udp_seq_afinfo *afinfo = PDE(inode)->data; 1523 struct seq_file *seq; 1524 int rc = -ENOMEM; 1525 struct udp_iter_state *s = kzalloc(sizeof(*s), GFP_KERNEL); 1526 1527 if (!s) 1528 goto out; 1529 s->family = afinfo->family; 1530 s->hashtable = afinfo->hashtable; 1531 s->seq_ops.start = udp_seq_start; 1532 s->seq_ops.next = udp_seq_next; 1533 s->seq_ops.show = afinfo->seq_show; 1534 s->seq_ops.stop = udp_seq_stop; 1535 1536 rc = seq_open(file, &s->seq_ops); 1537 if (rc) 1538 goto out_kfree; 1539 1540 seq = file->private_data; 1541 seq->private = s; 1542 out: 1543 return rc; 1544 out_kfree: 1545 kfree(s); 1546 goto out; 1547 } 1548 1549 /* ------------------------------------------------------------------------ */ 1550 int udp_proc_register(struct udp_seq_afinfo *afinfo) 1551 { 1552 struct proc_dir_entry *p; 1553 int rc = 0; 1554 1555 if (!afinfo) 1556 return -EINVAL; 1557 afinfo->seq_fops->owner = afinfo->owner; 1558 afinfo->seq_fops->open = udp_seq_open; 1559 afinfo->seq_fops->read = seq_read; 1560 afinfo->seq_fops->llseek = seq_lseek; 1561 afinfo->seq_fops->release = seq_release_private; 1562 1563 p = proc_net_fops_create(afinfo->name, S_IRUGO, afinfo->seq_fops); 1564 if (p) 1565 p->data = afinfo; 1566 else 1567 rc = -ENOMEM; 1568 return rc; 1569 } 1570 1571 void udp_proc_unregister(struct udp_seq_afinfo *afinfo) 1572 { 1573 if (!afinfo) 1574 return; 1575 proc_net_remove(afinfo->name); 1576 memset(afinfo->seq_fops, 0, sizeof(*afinfo->seq_fops)); 1577 } 1578 1579 /* ------------------------------------------------------------------------ */ 1580 static void udp4_format_sock(struct sock *sp, char *tmpbuf, int bucket) 1581 { 1582 struct inet_sock *inet = inet_sk(sp); 1583 __be32 dest = inet->daddr; 1584 __be32 src = inet->rcv_saddr; 1585 __u16 destp = ntohs(inet->dport); 1586 __u16 srcp = ntohs(inet->sport); 1587 1588 sprintf(tmpbuf, "%4d: %08X:%04X %08X:%04X" 1589 " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p", 1590 bucket, src, srcp, dest, destp, sp->sk_state, 1591 atomic_read(&sp->sk_wmem_alloc), 1592 atomic_read(&sp->sk_rmem_alloc), 1593 0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp), 1594 atomic_read(&sp->sk_refcnt), sp); 1595 } 1596 1597 int udp4_seq_show(struct seq_file *seq, void *v) 1598 { 1599 if (v == SEQ_START_TOKEN) 1600 seq_printf(seq, "%-127s\n", 1601 " sl local_address rem_address st tx_queue " 1602 "rx_queue tr tm->when retrnsmt uid timeout " 1603 "inode"); 1604 else { 1605 char tmpbuf[129]; 1606 struct udp_iter_state *state = seq->private; 1607 1608 udp4_format_sock(v, tmpbuf, state->bucket); 1609 seq_printf(seq, "%-127s\n", tmpbuf); 1610 } 1611 return 0; 1612 } 1613 1614 /* ------------------------------------------------------------------------ */ 1615 static struct file_operations udp4_seq_fops; 1616 static struct udp_seq_afinfo udp4_seq_afinfo = { 1617 .owner = THIS_MODULE, 1618 .name = "udp", 1619 .family = AF_INET, 1620 .hashtable = udp_hash, 1621 .seq_show = udp4_seq_show, 1622 .seq_fops = &udp4_seq_fops, 1623 }; 1624 1625 int __init udp4_proc_init(void) 1626 { 1627 return udp_proc_register(&udp4_seq_afinfo); 1628 } 1629 1630 void udp4_proc_exit(void) 1631 { 1632 udp_proc_unregister(&udp4_seq_afinfo); 1633 } 1634 #endif /* CONFIG_PROC_FS */ 1635 1636 EXPORT_SYMBOL(udp_disconnect); 1637 EXPORT_SYMBOL(udp_hash); 1638 EXPORT_SYMBOL(udp_hash_lock); 1639 EXPORT_SYMBOL(udp_ioctl); 1640 EXPORT_SYMBOL(udp_get_port); 1641 EXPORT_SYMBOL(udp_prot); 1642 EXPORT_SYMBOL(udp_sendmsg); 1643 EXPORT_SYMBOL(udp_lib_getsockopt); 1644 EXPORT_SYMBOL(udp_lib_setsockopt); 1645 EXPORT_SYMBOL(udp_poll); 1646 1647 #ifdef CONFIG_PROC_FS 1648 EXPORT_SYMBOL(udp_proc_register); 1649 EXPORT_SYMBOL(udp_proc_unregister); 1650 #endif 1651