linux-old/net/ipv4/udp.c
<<
>>
Prefs
   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 1.100.2.4 2002/03/05 12:47:34 davem Exp $
   9 *
  10 * Authors:     Ross Biro, <bir7@leland.Stanford.Edu>
  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 *
  15 * Fixes:
  16 *              Alan Cox        :       verify_area() calls
  17 *              Alan Cox        :       stopped close while in use off icmp
  18 *                                      messages. Not a fix but a botch that
  19 *                                      for udp at least is 'valid'.
  20 *              Alan Cox        :       Fixed icmp handling properly
  21 *              Alan Cox        :       Correct error for oversized datagrams
  22 *              Alan Cox        :       Tidied select() semantics. 
  23 *              Alan Cox        :       udp_err() fixed properly, also now 
  24 *                                      select and read wake correctly on errors
  25 *              Alan Cox        :       udp_send verify_area moved to avoid mem leak
  26 *              Alan Cox        :       UDP can count its memory
  27 *              Alan Cox        :       send to an unknown connection causes
  28 *                                      an ECONNREFUSED off the icmp, but
  29 *                                      does NOT close.
  30 *              Alan Cox        :       Switched to new sk_buff handlers. No more backlog!
  31 *              Alan Cox        :       Using generic datagram code. Even smaller and the PEEK
  32 *                                      bug no longer crashes it.
  33 *              Fred Van Kempen :       Net2e support for sk->broadcast.
  34 *              Alan Cox        :       Uses skb_free_datagram
  35 *              Alan Cox        :       Added get/set sockopt support.
  36 *              Alan Cox        :       Broadcasting without option set returns EACCES.
  37 *              Alan Cox        :       No wakeup calls. Instead we now use the callbacks.
  38 *              Alan Cox        :       Use ip_tos and ip_ttl
  39 *              Alan Cox        :       SNMP Mibs
  40 *              Alan Cox        :       MSG_DONTROUTE, and 0.0.0.0 support.
  41 *              Matt Dillon     :       UDP length checks.
  42 *              Alan Cox        :       Smarter af_inet used properly.
  43 *              Alan Cox        :       Use new kernel side addressing.
  44 *              Alan Cox        :       Incorrect return on truncated datagram receive.
  45 *      Arnt Gulbrandsen        :       New udp_send and stuff
  46 *              Alan Cox        :       Cache last socket
  47 *              Alan Cox        :       Route cache
  48 *              Jon Peatfield   :       Minor efficiency fix to sendto().
  49 *              Mike Shaver     :       RFC1122 checks.
  50 *              Alan Cox        :       Nonblocking error fix.
  51 *      Willy Konynenberg       :       Transparent proxying support.
  52 *              Mike McLagan    :       Routing by source
  53 *              David S. Miller :       New socket lookup architecture.
  54 *                                      Last socket cache retained as it
  55 *                                      does have a high hit rate.
  56 *              Olaf Kirch      :       Don't linearise iovec on sendmsg.
  57 *              Andi Kleen      :       Some cleanups, cache destination entry
  58 *                                      for connect. 
  59 *      Vitaly E. Lavrov        :       Transparent proxy revived after year coma.
  60 *              Melvin Smith    :       Check msg_name not msg_namelen in sendto(),
  61 *                                      return ENOTCONN for unconnected sockets (POSIX)
  62 *              Janos Farkas    :       don't deliver multi/broadcasts to a different
  63 *                                      bound-to-device socket
  64 *      YOSHIFUJI Hideaki @USAGI and:   Support IPV6_V6ONLY socket option, which
  65 *      Alexey Kuznetsov:               allow both IPv4 and IPv6 sockets to bind
  66 *                                      a single port at the same time.
  67 *
  68 *
  69 *              This program is free software; you can redistribute it and/or
  70 *              modify it under the terms of the GNU General Public License
  71 *              as published by the Free Software Foundation; either version
  72 *              2 of the License, or (at your option) any later version.
  73 */
  74 
  75#include <asm/system.h>
  76#include <asm/uaccess.h>
  77#include <asm/ioctls.h>
  78#include <linux/types.h>
  79#include <linux/fcntl.h>
  80#include <linux/socket.h>
  81#include <linux/sockios.h>
  82#include <linux/in.h>
  83#include <linux/errno.h>
  84#include <linux/timer.h>
  85#include <linux/mm.h>
  86#include <linux/config.h>
  87#include <linux/inet.h>
  88#include <linux/netdevice.h>
  89#include <net/snmp.h>
  90#include <net/ip.h>
  91#include <net/ipv6.h>
  92#include <net/protocol.h>
  93#include <linux/skbuff.h>
  94#include <net/sock.h>
  95#include <net/udp.h>
  96#include <net/icmp.h>
  97#include <net/route.h>
  98#include <net/inet_common.h>
  99#include <net/checksum.h>
 100
 101/*
 102 *      Snmp MIB for the UDP layer
 103 */
 104
 105struct udp_mib          udp_statistics[NR_CPUS*2];
 106
 107struct sock *udp_hash[UDP_HTABLE_SIZE];
 108rwlock_t udp_hash_lock = RW_LOCK_UNLOCKED;
 109
 110/* Shared by v4/v6 udp. */
 111int udp_port_rover;
 112
 113static int udp_v4_get_port(struct sock *sk, unsigned short snum)
 114{
 115        write_lock_bh(&udp_hash_lock);
 116        if (snum == 0) {
 117                int best_size_so_far, best, result, i;
 118
 119                if (udp_port_rover > sysctl_local_port_range[1] ||
 120                    udp_port_rover < sysctl_local_port_range[0])
 121                        udp_port_rover = sysctl_local_port_range[0];
 122                best_size_so_far = 32767;
 123                best = result = udp_port_rover;
 124                for (i = 0; i < UDP_HTABLE_SIZE; i++, result++) {
 125                        struct sock *sk;
 126                        int size;
 127
 128                        sk = udp_hash[result & (UDP_HTABLE_SIZE - 1)];
 129                        if (!sk) {
 130                                if (result > sysctl_local_port_range[1])
 131                                        result = sysctl_local_port_range[0] +
 132                                                ((result - sysctl_local_port_range[0]) &
 133                                                 (UDP_HTABLE_SIZE - 1));
 134                                goto gotit;
 135                        }
 136                        size = 0;
 137                        do {
 138                                if (++size >= best_size_so_far)
 139                                        goto next;
 140                        } while ((sk = sk->next) != NULL);
 141                        best_size_so_far = size;
 142                        best = result;
 143                next:;
 144                }
 145                result = best;
 146                for(i = 0; i < (1 << 16) / UDP_HTABLE_SIZE; i++, result += UDP_HTABLE_SIZE) {
 147                        if (result > sysctl_local_port_range[1])
 148                                result = sysctl_local_port_range[0]
 149                                        + ((result - sysctl_local_port_range[0]) &
 150                                           (UDP_HTABLE_SIZE - 1));
 151                        if (!udp_lport_inuse(result))
 152                                break;
 153                }
 154                if (i >= (1 << 16) / UDP_HTABLE_SIZE)
 155                        goto fail;
 156gotit:
 157                udp_port_rover = snum = result;
 158        } else {
 159                struct sock *sk2;
 160
 161                for (sk2 = udp_hash[snum & (UDP_HTABLE_SIZE - 1)];
 162                     sk2 != NULL;
 163                     sk2 = sk2->next) {
 164                        if (sk2->num == snum &&
 165                            sk2 != sk &&
 166                            !ipv6_only_sock(sk2) &&
 167                            (!sk2->bound_dev_if ||
 168                             !sk->bound_dev_if ||
 169                             sk2->bound_dev_if == sk->bound_dev_if) &&
 170                            (!sk2->rcv_saddr ||
 171                             !sk->rcv_saddr ||
 172                             sk2->rcv_saddr == sk->rcv_saddr) &&
 173                            (!sk2->reuse || !sk->reuse))
 174                                goto fail;
 175                }
 176        }
 177        sk->num = snum;
 178        if (sk->pprev == NULL) {
 179                struct sock **skp = &udp_hash[snum & (UDP_HTABLE_SIZE - 1)];
 180                if ((sk->next = *skp) != NULL)
 181                        (*skp)->pprev = &sk->next;
 182                *skp = sk;
 183                sk->pprev = skp;
 184                sock_prot_inc_use(sk->prot);
 185                sock_hold(sk);
 186        }
 187        write_unlock_bh(&udp_hash_lock);
 188        return 0;
 189
 190fail:
 191        write_unlock_bh(&udp_hash_lock);
 192        return 1;
 193}
 194
 195static void udp_v4_hash(struct sock *sk)
 196{
 197        BUG();
 198}
 199
 200static void udp_v4_unhash(struct sock *sk)
 201{
 202        write_lock_bh(&udp_hash_lock);
 203        if (sk->pprev) {
 204                if (sk->next)
 205                        sk->next->pprev = sk->pprev;
 206                *sk->pprev = sk->next;
 207                sk->pprev = NULL;
 208                sk->num = 0;
 209                sock_prot_dec_use(sk->prot);
 210                __sock_put(sk);
 211        }
 212        write_unlock_bh(&udp_hash_lock);
 213}
 214
 215/* UDP is nearly always wildcards out the wazoo, it makes no sense to try
 216 * harder than this. -DaveM
 217 */
 218struct sock *udp_v4_lookup_longway(u32 saddr, u16 sport, u32 daddr, u16 dport, int dif)
 219{
 220        struct sock *sk, *result = NULL;
 221        unsigned short hnum = ntohs(dport);
 222        int badness = -1;
 223
 224        for(sk = udp_hash[hnum & (UDP_HTABLE_SIZE - 1)]; sk != NULL; sk = sk->next) {
 225                if(sk->num == hnum && !ipv6_only_sock(sk)) {
 226                        int score;
 227#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 228                        score = sk->family == PF_INET ? 1 : 0;
 229#else
 230                        score = 1;
 231#endif
 232                        if(sk->rcv_saddr) {
 233                                if(sk->rcv_saddr != daddr)
 234                                        continue;
 235                                score+=2;
 236                        }
 237                        if(sk->daddr) {
 238                                if(sk->daddr != saddr)
 239                                        continue;
 240                                score+=2;
 241                        }
 242                        if(sk->dport) {
 243                                if(sk->dport != sport)
 244                                        continue;
 245                                score+=2;
 246                        }
 247                        if(sk->bound_dev_if) {
 248                                if(sk->bound_dev_if != dif)
 249                                        continue;
 250                                score+=2;
 251                        }
 252                        if(score == 9) {
 253                                result = sk;
 254                                break;
 255                        } else if(score > badness) {
 256                                result = sk;
 257                                badness = score;
 258                        }
 259                }
 260        }
 261        return result;
 262}
 263
 264__inline__ struct sock *udp_v4_lookup(u32 saddr, u16 sport, u32 daddr, u16 dport, int dif)
 265{
 266        struct sock *sk;
 267
 268        read_lock(&udp_hash_lock);
 269        sk = udp_v4_lookup_longway(saddr, sport, daddr, dport, dif);
 270        if (sk)
 271                sock_hold(sk);
 272        read_unlock(&udp_hash_lock);
 273        return sk;
 274}
 275
 276extern int ip_mc_sf_allow(struct sock *sk, u32 local, u32 rmt, int dif);
 277
 278static inline struct sock *udp_v4_mcast_next(struct sock *sk,
 279                                             u16 loc_port, u32 loc_addr,
 280                                             u16 rmt_port, u32 rmt_addr,
 281                                             int dif)
 282{
 283        struct sock *s = sk;
 284        unsigned short hnum = ntohs(loc_port);
 285        for(; s; s = s->next) {
 286                if ((s->num != hnum)                                    ||
 287                    (s->daddr && s->daddr!=rmt_addr)                    ||
 288                    (s->dport != rmt_port && s->dport != 0)                     ||
 289                    (s->rcv_saddr  && s->rcv_saddr != loc_addr)         ||
 290                    ipv6_only_sock(s)                                   ||
 291                    (s->bound_dev_if && s->bound_dev_if != dif))
 292                        continue;
 293                if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif))
 294                        continue;
 295                break;
 296        }
 297        return s;
 298}
 299
 300/*
 301 * This routine is called by the ICMP module when it gets some
 302 * sort of error condition.  If err < 0 then the socket should
 303 * be closed and the error returned to the user.  If err > 0
 304 * it's just the icmp type << 8 | icmp code.  
 305 * Header points to the ip header of the error packet. We move
 306 * on past this. Then (as it used to claim before adjustment)
 307 * header points to the first 8 bytes of the udp header.  We need
 308 * to find the appropriate port.
 309 */
 310
 311void udp_err(struct sk_buff *skb, u32 info)
 312{
 313        struct iphdr *iph = (struct iphdr*)skb->data;
 314        struct udphdr *uh = (struct udphdr*)(skb->data+(iph->ihl<<2));
 315        int type = skb->h.icmph->type;
 316        int code = skb->h.icmph->code;
 317        struct sock *sk;
 318        int harderr;
 319        int err;
 320
 321        sk = udp_v4_lookup(iph->daddr, uh->dest, iph->saddr, uh->source, skb->dev->ifindex);
 322        if (sk == NULL) {
 323                ICMP_INC_STATS_BH(IcmpInErrors);
 324                return; /* No socket for error */
 325        }
 326
 327        err = 0;
 328        harderr = 0;
 329
 330        switch (type) {
 331        default:
 332        case ICMP_TIME_EXCEEDED:
 333                err = EHOSTUNREACH;
 334                break;
 335        case ICMP_SOURCE_QUENCH:
 336                goto out;
 337        case ICMP_PARAMETERPROB:
 338                err = EPROTO;
 339                harderr = 1;
 340                break;
 341        case ICMP_DEST_UNREACH:
 342                if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
 343                        if (sk->protinfo.af_inet.pmtudisc != IP_PMTUDISC_DONT) {
 344                                err = EMSGSIZE;
 345                                harderr = 1;
 346                                break;
 347                        }
 348                        goto out;
 349                }
 350                err = EHOSTUNREACH;
 351                if (code <= NR_ICMP_UNREACH) {
 352                        harderr = icmp_err_convert[code].fatal;
 353                        err = icmp_err_convert[code].errno;
 354                }
 355                break;
 356        }
 357
 358        /*
 359         *      RFC1122: OK.  Passes ICMP errors back to application, as per 
 360         *      4.1.3.3.
 361         */
 362        if (!sk->protinfo.af_inet.recverr) {
 363                if (!harderr || sk->state != TCP_ESTABLISHED)
 364                        goto out;
 365        } else {
 366                ip_icmp_error(sk, skb, err, uh->dest, info, (u8*)(uh+1));
 367        }
 368        sk->err = err;
 369        sk->error_report(sk);
 370out:
 371        sock_put(sk);
 372}
 373
 374static unsigned short udp_check(struct udphdr *uh, int len, unsigned long saddr, unsigned long daddr, unsigned long base)
 375{
 376        return(csum_tcpudp_magic(saddr, daddr, len, IPPROTO_UDP, base));
 377}
 378
 379struct udpfakehdr 
 380{
 381        struct udphdr uh;
 382        u32 saddr;
 383        u32 daddr;
 384        struct iovec *iov;
 385        u32 wcheck;
 386};
 387
 388/*
 389 *      Copy and checksum a UDP packet from user space into a buffer.
 390 */
 391 
 392static int udp_getfrag(const void *p, char * to, unsigned int offset,
 393                       unsigned int fraglen, struct sk_buff *skb)
 394{
 395        struct udpfakehdr *ufh = (struct udpfakehdr *)p;
 396        if (offset==0) {
 397                if (csum_partial_copy_fromiovecend(to+sizeof(struct udphdr), ufh->iov, offset,
 398                                                   fraglen-sizeof(struct udphdr), &ufh->wcheck))
 399                        return -EFAULT;
 400                ufh->wcheck = csum_partial((char *)ufh, sizeof(struct udphdr),
 401                                           ufh->wcheck);
 402                ufh->uh.check = csum_tcpudp_magic(ufh->saddr, ufh->daddr, 
 403                                          ntohs(ufh->uh.len),
 404                                          IPPROTO_UDP, ufh->wcheck);
 405                if (ufh->uh.check == 0)
 406                        ufh->uh.check = -1;
 407                memcpy(to, ufh, sizeof(struct udphdr));
 408                return 0;
 409        }
 410        if (csum_partial_copy_fromiovecend(to, ufh->iov, offset-sizeof(struct udphdr),
 411                                           fraglen, &ufh->wcheck))
 412                return -EFAULT;
 413        return 0;
 414}
 415
 416/*
 417 *      Copy a UDP packet from user space into a buffer without checksumming.
 418 */
 419 
 420static int udp_getfrag_nosum(const void *p, char * to, unsigned int offset,
 421                             unsigned int fraglen, struct sk_buff *skb) 
 422{
 423        struct udpfakehdr *ufh = (struct udpfakehdr *)p;
 424
 425        if (offset==0) {
 426                memcpy(to, ufh, sizeof(struct udphdr));
 427                return memcpy_fromiovecend(to+sizeof(struct udphdr), ufh->iov, offset,
 428                                           fraglen-sizeof(struct udphdr));
 429        }
 430        return memcpy_fromiovecend(to, ufh->iov, offset-sizeof(struct udphdr),
 431                                   fraglen);
 432}
 433
 434int udp_sendmsg(struct sock *sk, struct msghdr *msg, int len)
 435{
 436        int ulen = len + sizeof(struct udphdr);
 437        struct ipcm_cookie ipc;
 438        struct udpfakehdr ufh;
 439        struct rtable *rt = NULL;
 440        int free = 0;
 441        int connected = 0;
 442        u32 daddr;
 443        u8  tos;
 444        int err;
 445
 446        /* This check is ONLY to check for arithmetic overflow
 447           on integer(!) len. Not more! Real check will be made
 448           in ip_build_xmit --ANK
 449
 450           BTW socket.c -> af_*.c -> ... make multiple
 451           invalid conversions size_t -> int. We MUST repair it f.e.
 452           by replacing all of them with size_t and revise all
 453           the places sort of len += sizeof(struct iphdr)
 454           If len was ULONG_MAX-10 it would be cathastrophe  --ANK
 455         */
 456
 457        if (len < 0 || len > 0xFFFF)
 458                return -EMSGSIZE;
 459
 460        /* 
 461         *      Check the flags.
 462         */
 463
 464        if (msg->msg_flags&MSG_OOB)     /* Mirror BSD error message compatibility */
 465                return -EOPNOTSUPP;
 466
 467        /*
 468         *      Get and verify the address. 
 469         */
 470         
 471        if (msg->msg_name) {
 472                struct sockaddr_in * usin = (struct sockaddr_in*)msg->msg_name;
 473                if (msg->msg_namelen < sizeof(*usin))
 474                        return -EINVAL;
 475                if (usin->sin_family != AF_INET) {
 476                        if (usin->sin_family != AF_UNSPEC)
 477                                return -EINVAL;
 478                }
 479
 480                ufh.daddr = usin->sin_addr.s_addr;
 481                ufh.uh.dest = usin->sin_port;
 482                if (ufh.uh.dest == 0)
 483                        return -EINVAL;
 484        } else {
 485                if (sk->state != TCP_ESTABLISHED)
 486                        return -EDESTADDRREQ;
 487                ufh.daddr = sk->daddr;
 488                ufh.uh.dest = sk->dport;
 489                /* Open fast path for connected socket.
 490                   Route will not be used, if at least one option is set.
 491                 */
 492                connected = 1;
 493        }
 494        ipc.addr = sk->saddr;
 495        ufh.uh.source = sk->sport;
 496
 497        ipc.opt = NULL;
 498        ipc.oif = sk->bound_dev_if;
 499        if (msg->msg_controllen) {
 500                err = ip_cmsg_send(msg, &ipc);
 501                if (err)
 502                        return err;
 503                if (ipc.opt)
 504                        free = 1;
 505                connected = 0;
 506        }
 507        if (!ipc.opt)
 508                ipc.opt = sk->protinfo.af_inet.opt;
 509
 510        ufh.saddr = ipc.addr;
 511        ipc.addr = daddr = ufh.daddr;
 512
 513        if (ipc.opt && ipc.opt->srr) {
 514                if (!daddr)
 515                        return -EINVAL;
 516                daddr = ipc.opt->faddr;
 517                connected = 0;
 518        }
 519        tos = RT_TOS(sk->protinfo.af_inet.tos);
 520        if (sk->localroute || (msg->msg_flags&MSG_DONTROUTE) || 
 521            (ipc.opt && ipc.opt->is_strictroute)) {
 522                tos |= RTO_ONLINK;
 523                connected = 0;
 524        }
 525
 526        if (MULTICAST(daddr)) {
 527                if (!ipc.oif)
 528                        ipc.oif = sk->protinfo.af_inet.mc_index;
 529                if (!ufh.saddr)
 530                        ufh.saddr = sk->protinfo.af_inet.mc_addr;
 531                connected = 0;
 532        }
 533
 534        if (connected)
 535                rt = (struct rtable*)sk_dst_check(sk, 0);
 536
 537        if (rt == NULL) {
 538                err = ip_route_output(&rt, daddr, ufh.saddr, tos, ipc.oif);
 539                if (err)
 540                        goto out;
 541
 542                err = -EACCES;
 543                if (rt->rt_flags&RTCF_BROADCAST && !sk->broadcast) 
 544                        goto out;
 545                if (connected)
 546                        sk_dst_set(sk, dst_clone(&rt->u.dst));
 547        }
 548
 549        if (msg->msg_flags&MSG_CONFIRM)
 550                goto do_confirm;
 551back_from_confirm:
 552
 553        ufh.saddr = rt->rt_src;
 554        if (!ipc.addr)
 555                ufh.daddr = ipc.addr = rt->rt_dst;
 556        ufh.uh.len = htons(ulen);
 557        ufh.uh.check = 0;
 558        ufh.iov = msg->msg_iov;
 559        ufh.wcheck = 0;
 560
 561        /* RFC1122: OK.  Provides the checksumming facility (MUST) as per */
 562        /* 4.1.3.4. It's configurable by the application via setsockopt() */
 563        /* (MAY) and it defaults to on (MUST). */
 564
 565        err = ip_build_xmit(sk,
 566                            (sk->no_check == UDP_CSUM_NOXMIT ?
 567                             udp_getfrag_nosum :
 568                             udp_getfrag),
 569                            &ufh, ulen, &ipc, rt, msg->msg_flags);
 570
 571out:
 572        ip_rt_put(rt);
 573        if (free)
 574                kfree(ipc.opt);
 575        if (!err) {
 576                UDP_INC_STATS_USER(UdpOutDatagrams);
 577                return len;
 578        }
 579        return err;
 580
 581do_confirm:
 582        dst_confirm(&rt->u.dst);
 583        if (!(msg->msg_flags&MSG_PROBE) || len)
 584                goto back_from_confirm;
 585        err = 0;
 586        goto out;
 587}
 588
 589/*
 590 *      IOCTL requests applicable to the UDP protocol
 591 */
 592 
 593int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
 594{
 595        switch(cmd) 
 596        {
 597                case SIOCOUTQ:
 598                {
 599                        int amount = atomic_read(&sk->wmem_alloc);
 600                        return put_user(amount, (int *)arg);
 601                }
 602
 603                case SIOCINQ:
 604                {
 605                        struct sk_buff *skb;
 606                        unsigned long amount;
 607
 608                        amount = 0;
 609                        spin_lock_irq(&sk->receive_queue.lock);
 610                        skb = skb_peek(&sk->receive_queue);
 611                        if (skb != NULL) {
 612                                /*
 613                                 * We will only return the amount
 614                                 * of this packet since that is all
 615                                 * that will be read.
 616                                 */
 617                                amount = skb->len - sizeof(struct udphdr);
 618                        }
 619                        spin_unlock_irq(&sk->receive_queue.lock);
 620                        return put_user(amount, (int *)arg);
 621                }
 622
 623                default:
 624                        return -ENOIOCTLCMD;
 625        }
 626        return(0);
 627}
 628
 629static __inline__ int __udp_checksum_complete(struct sk_buff *skb)
 630{
 631        return (unsigned short)csum_fold(skb_checksum(skb, 0, skb->len, skb->csum));
 632}
 633
 634static __inline__ int udp_checksum_complete(struct sk_buff *skb)
 635{
 636        return skb->ip_summed != CHECKSUM_UNNECESSARY &&
 637                __udp_checksum_complete(skb);
 638}
 639
 640
 641/**
 642 *      udp_poll - wait for a UDP event.
 643 *      @file - file struct
 644 *      @sock - socket
 645 *      @wait - poll table
 646 *
 647 *      This is same as datagram poll, except for the special case of 
 648 *      blocking sockets. If application is using a blocking fd
 649 *      and a packet with checksum error is in the queue;
 650 *      then it could get return from select indicating data available
 651 *      but then block when reading it. Add special case code
 652 *      to work around these arguably broken applications.
 653 */
 654unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
 655{
 656        unsigned int mask = datagram_poll(file, sock, wait);
 657        struct sock *sk = sock->sk;
 658        
 659        /* Check for false positives due to checksum errors */
 660        if ( (mask & POLLRDNORM) &&
 661             !(file->f_flags & O_NONBLOCK) &&
 662             !(sk->shutdown & RCV_SHUTDOWN)){
 663                struct sk_buff_head *rcvq = &sk->receive_queue;
 664                struct sk_buff *skb;
 665
 666                spin_lock_irq(&rcvq->lock);
 667                while ((skb = skb_peek(rcvq)) != NULL) {
 668                        if (udp_checksum_complete(skb)) {
 669                                UDP_INC_STATS_BH(UdpInErrors);
 670                                IP_INC_STATS_BH(IpInDiscards);
 671                                __skb_unlink(skb, rcvq);
 672                                kfree_skb(skb);
 673                        } else {
 674                                skb->ip_summed = CHECKSUM_UNNECESSARY;
 675                                break;
 676                        }
 677                }
 678                spin_unlock_irq(&rcvq->lock);
 679
 680                /* nothing to see, move along */
 681                if (skb == NULL)
 682                        mask &= ~(POLLIN | POLLRDNORM);
 683        }
 684
 685        return mask;
 686        
 687}
 688
 689/*
 690 *      This should be easy, if there is something there we
 691 *      return it, otherwise we block.
 692 */
 693
 694int udp_recvmsg(struct sock *sk, struct msghdr *msg, int len,
 695                int noblock, int flags, int *addr_len)
 696{
 697        struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
 698        struct sk_buff *skb;
 699        int copied, err;
 700
 701        /*
 702         *      Check any passed addresses
 703         */
 704        if (addr_len)
 705                *addr_len=sizeof(*sin);
 706
 707        if (flags & MSG_ERRQUEUE)
 708                return ip_recv_error(sk, msg, len);
 709
 710try_again:
 711        skb = skb_recv_datagram(sk, flags, noblock, &err);
 712        if (!skb)
 713                goto out;
 714  
 715        copied = skb->len - sizeof(struct udphdr);
 716        if (copied > len) {
 717                copied = len;
 718                msg->msg_flags |= MSG_TRUNC;
 719        }
 720
 721        if (skb->ip_summed==CHECKSUM_UNNECESSARY) {
 722                err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov,
 723                                              copied);
 724        } else if (msg->msg_flags&MSG_TRUNC) {
 725                if (__udp_checksum_complete(skb))
 726                        goto csum_copy_err;
 727                err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov,
 728                                              copied);
 729        } else {
 730                err = skb_copy_and_csum_datagram_iovec(skb, sizeof(struct udphdr), msg->msg_iov);
 731
 732                if (err == -EINVAL)
 733                        goto csum_copy_err;
 734        }
 735
 736        if (err)
 737                goto out_free;
 738
 739        sock_recv_timestamp(msg, sk, skb);
 740
 741        /* Copy the address. */
 742        if (sin)
 743        {
 744                sin->sin_family = AF_INET;
 745                sin->sin_port = skb->h.uh->source;
 746                sin->sin_addr.s_addr = skb->nh.iph->saddr;
 747                memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
 748        }
 749        if (sk->protinfo.af_inet.cmsg_flags)
 750                ip_cmsg_recv(msg, skb);
 751
 752        err = copied;
 753        if (flags & MSG_TRUNC)
 754                err = skb->len - sizeof(struct udphdr);
 755  
 756out_free:
 757        skb_free_datagram(sk, skb);
 758out:
 759        return err;
 760
 761csum_copy_err:
 762        UDP_INC_STATS_BH(UdpInErrors);
 763
 764        /* Clear queue. */
 765        if (flags&MSG_PEEK) {
 766                int clear = 0;
 767                spin_lock_irq(&sk->receive_queue.lock);
 768                if (skb == skb_peek(&sk->receive_queue)) {
 769                        __skb_unlink(skb, &sk->receive_queue);
 770                        clear = 1;
 771                }
 772                spin_unlock_irq(&sk->receive_queue.lock);
 773                if (clear)
 774                        kfree_skb(skb);
 775        }
 776
 777        skb_free_datagram(sk, skb);
 778
 779        if (noblock)
 780                return -EAGAIN; 
 781        goto try_again;
 782}
 783
 784int udp_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
 785{
 786        struct sockaddr_in *usin = (struct sockaddr_in *) uaddr;
 787        struct rtable *rt;
 788        u32 saddr;
 789        int oif;
 790        int err;
 791
 792        
 793        if (addr_len < sizeof(*usin)) 
 794                return -EINVAL;
 795
 796        if (usin->sin_family != AF_INET) 
 797                return -EAFNOSUPPORT;
 798
 799        sk_dst_reset(sk);
 800
 801        oif = sk->bound_dev_if;
 802        saddr = sk->saddr;
 803        if (MULTICAST(usin->sin_addr.s_addr)) {
 804                if (!oif)
 805                        oif = sk->protinfo.af_inet.mc_index;
 806                if (!saddr)
 807                        saddr = sk->protinfo.af_inet.mc_addr;
 808        }
 809        err = ip_route_connect(&rt, usin->sin_addr.s_addr, saddr,
 810                               RT_CONN_FLAGS(sk), oif);
 811        if (err)
 812                return err;
 813        if ((rt->rt_flags&RTCF_BROADCAST) && !sk->broadcast) {
 814                ip_rt_put(rt);
 815                return -EACCES;
 816        }
 817        if(!sk->saddr)
 818                sk->saddr = rt->rt_src;         /* Update source address */
 819        if(!sk->rcv_saddr)
 820                sk->rcv_saddr = rt->rt_src;
 821        sk->daddr = rt->rt_dst;
 822        sk->dport = usin->sin_port;
 823        sk->state = TCP_ESTABLISHED;
 824        sk->protinfo.af_inet.id = jiffies;
 825
 826        sk_dst_set(sk, &rt->u.dst);
 827        return(0);
 828}
 829
 830int udp_disconnect(struct sock *sk, int flags)
 831{
 832        /*
 833         *      1003.1g - break association.
 834         */
 835         
 836        sk->state = TCP_CLOSE;
 837        sk->daddr = 0;
 838        sk->dport = 0;
 839        sk->bound_dev_if = 0;
 840        if (!(sk->userlocks&SOCK_BINDADDR_LOCK)) {
 841                sk->rcv_saddr = 0;
 842                sk->saddr = 0;
 843#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 844                memset(&sk->net_pinfo.af_inet6.saddr, 0, 16);
 845                memset(&sk->net_pinfo.af_inet6.rcv_saddr, 0, 16);
 846#endif
 847        }
 848        if (!(sk->userlocks&SOCK_BINDPORT_LOCK)) {
 849                sk->prot->unhash(sk);
 850                sk->sport = 0;
 851        }
 852        sk_dst_reset(sk);
 853        return 0;
 854}
 855
 856static void udp_close(struct sock *sk, long timeout)
 857{
 858        inet_sock_release(sk);
 859}
 860
 861static int udp_queue_rcv_skb(struct sock * sk, struct sk_buff *skb)
 862{
 863        /*
 864         *      Charge it to the socket, dropping if the queue is full.
 865         */
 866
 867#if defined(CONFIG_FILTER)
 868        if (sk->filter && skb->ip_summed != CHECKSUM_UNNECESSARY) {
 869                if (__udp_checksum_complete(skb)) {
 870                        UDP_INC_STATS_BH(UdpInErrors);
 871                        IP_INC_STATS_BH(IpInDiscards);
 872                        ip_statistics[smp_processor_id()*2].IpInDelivers--;
 873                        kfree_skb(skb);
 874                        return -1;
 875                }
 876                skb->ip_summed = CHECKSUM_UNNECESSARY;
 877        }
 878#endif
 879
 880        if (sock_queue_rcv_skb(sk,skb)<0) {
 881                UDP_INC_STATS_BH(UdpInErrors);
 882                IP_INC_STATS_BH(IpInDiscards);
 883                ip_statistics[smp_processor_id()*2].IpInDelivers--;
 884                kfree_skb(skb);
 885                return -1;
 886        }
 887        UDP_INC_STATS_BH(UdpInDatagrams);
 888        return 0;
 889}
 890
 891/*
 892 *      Multicasts and broadcasts go to each listener.
 893 *
 894 *      Note: called only from the BH handler context,
 895 *      so we don't need to lock the hashes.
 896 */
 897static int udp_v4_mcast_deliver(struct sk_buff *skb, struct udphdr *uh,
 898                                 u32 saddr, u32 daddr)
 899{
 900        struct sock *sk;
 901        int dif;
 902
 903        read_lock(&udp_hash_lock);
 904        sk = udp_hash[ntohs(uh->dest) & (UDP_HTABLE_SIZE - 1)];
 905        dif = skb->dev->ifindex;
 906        sk = udp_v4_mcast_next(sk, uh->dest, daddr, uh->source, saddr, dif);
 907        if (sk) {
 908                struct sock *sknext = NULL;
 909
 910                do {
 911                        struct sk_buff *skb1 = skb;
 912
 913                        sknext = udp_v4_mcast_next(sk->next, uh->dest, daddr,
 914                                                   uh->source, saddr, dif);
 915                        if(sknext)
 916                                skb1 = skb_clone(skb, GFP_ATOMIC);
 917
 918                        if(skb1)
 919                                udp_queue_rcv_skb(sk, skb1);
 920                        sk = sknext;
 921                } while(sknext);
 922        } else
 923                kfree_skb(skb);
 924        read_unlock(&udp_hash_lock);
 925        return 0;
 926}
 927
 928/* Initialize UDP checksum. If exited with zero value (success),
 929 * CHECKSUM_UNNECESSARY means, that no more checks are required.
 930 * Otherwise, csum completion requires chacksumming packet body,
 931 * including udp header and folding it to skb->csum.
 932 */
 933static int udp_checksum_init(struct sk_buff *skb, struct udphdr *uh,
 934                             unsigned short ulen, u32 saddr, u32 daddr)
 935{
 936        if (uh->check == 0) {
 937                skb->ip_summed = CHECKSUM_UNNECESSARY;
 938        } else if (skb->ip_summed == CHECKSUM_HW) {
 939                skb->ip_summed = CHECKSUM_UNNECESSARY;
 940                if (!udp_check(uh, ulen, saddr, daddr, skb->csum))
 941                        return 0;
 942                NETDEBUG(if (net_ratelimit()) printk(KERN_DEBUG "udp v4 hw csum failure.\n"));
 943                skb->ip_summed = CHECKSUM_NONE;
 944        }
 945        if (skb->ip_summed != CHECKSUM_UNNECESSARY)
 946                skb->csum = csum_tcpudp_nofold(saddr, daddr, ulen, IPPROTO_UDP, 0);
 947        /* Probably, we should checksum udp header (it should be in cache
 948         * in any case) and data in tiny packets (< rx copybreak).
 949         */
 950        return 0;
 951}
 952
 953/*
 954 *      All we need to do is get the socket, and then do a checksum. 
 955 */
 956 
 957int udp_rcv(struct sk_buff *skb)
 958{
 959        struct sock *sk;
 960        struct udphdr *uh;
 961        unsigned short ulen;
 962        struct rtable *rt = (struct rtable*)skb->dst;
 963        u32 saddr = skb->nh.iph->saddr;
 964        u32 daddr = skb->nh.iph->daddr;
 965        int len = skb->len;
 966
 967        IP_INC_STATS_BH(IpInDelivers);
 968
 969        /*
 970         *      Validate the packet and the UDP length.
 971         */
 972        if (!pskb_may_pull(skb, sizeof(struct udphdr)))
 973                goto no_header;
 974
 975        uh = skb->h.uh;
 976
 977        ulen = ntohs(uh->len);
 978
 979        if (ulen > len || ulen < sizeof(*uh))
 980                goto short_packet;
 981
 982        if (pskb_trim(skb, ulen))
 983                goto short_packet;
 984
 985        if (udp_checksum_init(skb, uh, ulen, saddr, daddr) < 0)
 986                goto csum_error;
 987
 988        if(rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
 989                return udp_v4_mcast_deliver(skb, uh, saddr, daddr);
 990
 991        sk = udp_v4_lookup(saddr, uh->source, daddr, uh->dest, skb->dev->ifindex);
 992
 993        if (sk != NULL) {
 994                udp_queue_rcv_skb(sk, skb);
 995                sock_put(sk);
 996                return 0;
 997        }
 998
 999        /* No socket. Drop packet silently, if checksum is wrong */
1000        if (udp_checksum_complete(skb))
1001                goto csum_error;
1002
1003        UDP_INC_STATS_BH(UdpNoPorts);
1004        icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1005
1006        /*
1007         * Hmm.  We got an UDP packet to a port to which we
1008         * don't wanna listen.  Ignore it.
1009         */
1010        kfree_skb(skb);
1011        return(0);
1012
1013short_packet:
1014        NETDEBUG(if (net_ratelimit())
1015                 printk(KERN_DEBUG "UDP: short packet: %u.%u.%u.%u:%u %d/%d to %u.%u.%u.%u:%u\n",
1016                        NIPQUAD(saddr),
1017                        ntohs(uh->source),
1018                        ulen,
1019                        len,
1020                        NIPQUAD(daddr),
1021                        ntohs(uh->dest)));
1022no_header:
1023        UDP_INC_STATS_BH(UdpInErrors);
1024        kfree_skb(skb);
1025        return(0);
1026
1027csum_error:
1028        /* 
1029         * RFC1122: OK.  Discards the bad packet silently (as far as 
1030         * the network is concerned, anyway) as per 4.1.3.4 (MUST). 
1031         */
1032        NETDEBUG(if (net_ratelimit())
1033                 printk(KERN_DEBUG "UDP: bad checksum. From %d.%d.%d.%d:%d to %d.%d.%d.%d:%d ulen %d\n",
1034                        NIPQUAD(saddr),
1035                        ntohs(uh->source),
1036                        NIPQUAD(daddr),
1037                        ntohs(uh->dest),
1038                        ulen));
1039        UDP_INC_STATS_BH(UdpInErrors);
1040        kfree_skb(skb);
1041        return(0);
1042}
1043
1044static void get_udp_sock(struct sock *sp, char *tmpbuf, int i)
1045{
1046        unsigned int dest, src;
1047        __u16 destp, srcp;
1048
1049        dest  = sp->daddr;
1050        src   = sp->rcv_saddr;
1051        destp = ntohs(sp->dport);
1052        srcp  = ntohs(sp->sport);
1053        sprintf(tmpbuf, "%4d: %08X:%04X %08X:%04X"
1054                " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %p",
1055                i, src, srcp, dest, destp, sp->state, 
1056                atomic_read(&sp->wmem_alloc), atomic_read(&sp->rmem_alloc),
1057                0, 0L, 0,
1058                sock_i_uid(sp), 0,
1059                sock_i_ino(sp),
1060                atomic_read(&sp->refcnt), sp);
1061}
1062
1063int udp_get_info(char *buffer, char **start, off_t offset, int length)
1064{
1065        int len = 0, num = 0, i;
1066        off_t pos = 0;
1067        off_t begin;
1068        char tmpbuf[129];
1069
1070        if (offset < 128) 
1071                len += sprintf(buffer, "%-127s\n",
1072                               "  sl  local_address rem_address   st tx_queue "
1073                               "rx_queue tr tm->when retrnsmt   uid  timeout inode");
1074        pos = 128;
1075        read_lock(&udp_hash_lock);
1076        for (i = 0; i < UDP_HTABLE_SIZE; i++) {
1077                struct sock *sk;
1078
1079                for (sk = udp_hash[i]; sk; sk = sk->next, num++) {
1080                        if (sk->family != PF_INET)
1081                                continue;
1082                        pos += 128;
1083                        if (pos <= offset)
1084                                continue;
1085                        get_udp_sock(sk, tmpbuf, i);
1086                        len += sprintf(buffer+len, "%-127s\n", tmpbuf);
1087                        if(len >= length)
1088                                goto out;
1089                }
1090        }
1091out:
1092        read_unlock(&udp_hash_lock);
1093        begin = len - (pos - offset);
1094        *start = buffer + begin;
1095        len -= begin;
1096        if(len > length)
1097                len = length;
1098        if (len < 0)
1099                len = 0; 
1100        return len;
1101}
1102
1103struct proto udp_prot = {
1104        name:           "UDP",
1105        close:          udp_close,
1106        connect:        udp_connect,
1107        disconnect:     udp_disconnect,
1108        ioctl:          udp_ioctl,
1109        setsockopt:     ip_setsockopt,
1110        getsockopt:     ip_getsockopt,
1111        sendmsg:        udp_sendmsg,
1112        recvmsg:        udp_recvmsg,
1113        backlog_rcv:    udp_queue_rcv_skb,
1114        hash:           udp_v4_hash,
1115        unhash:         udp_v4_unhash,
1116        get_port:       udp_v4_get_port,
1117};
1118
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.