linux/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 * Authors:     Ross Biro
   9 *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  10 *              Arnt Gulbrandsen, <agulbra@nvg.unit.no>
  11 *              Alan Cox, <alan@lxorguk.ukuu.org.uk>
  12 *              Hirokazu Takahashi, <taka@valinux.co.jp>
  13 *
  14 * Fixes:
  15 *              Alan Cox        :       verify_area() calls
  16 *              Alan Cox        :       stopped close while in use off icmp
  17 *                                      messages. Not a fix but a botch that
  18 *                                      for udp at least is 'valid'.
  19 *              Alan Cox        :       Fixed icmp handling properly
  20 *              Alan Cox        :       Correct error for oversized datagrams
  21 *              Alan Cox        :       Tidied select() semantics.
  22 *              Alan Cox        :       udp_err() fixed properly, also now
  23 *                                      select and read wake correctly on errors
  24 *              Alan Cox        :       udp_send verify_area moved to avoid mem leak
  25 *              Alan Cox        :       UDP can count its memory
  26 *              Alan Cox        :       send to an unknown connection causes
  27 *                                      an ECONNREFUSED off the icmp, but
  28 *                                      does NOT close.
  29 *              Alan Cox        :       Switched to new sk_buff handlers. No more backlog!
  30 *              Alan Cox        :       Using generic datagram code. Even smaller and the PEEK
  31 *                                      bug no longer crashes it.
  32 *              Fred Van Kempen :       Net2e support for sk->broadcast.
  33 *              Alan Cox        :       Uses skb_free_datagram
  34 *              Alan Cox        :       Added get/set sockopt support.
  35 *              Alan Cox        :       Broadcasting without option set returns EACCES.
  36 *              Alan Cox        :       No wakeup calls. Instead we now use the callbacks.
  37 *              Alan Cox        :       Use ip_tos and ip_ttl
  38 *              Alan Cox        :       SNMP Mibs
  39 *              Alan Cox        :       MSG_DONTROUTE, and 0.0.0.0 support.
  40 *              Matt Dillon     :       UDP length checks.
  41 *              Alan Cox        :       Smarter af_inet used properly.
  42 *              Alan Cox        :       Use new kernel side addressing.
  43 *              Alan Cox        :       Incorrect return on truncated datagram receive.
  44 *      Arnt Gulbrandsen        :       New udp_send and stuff
  45 *              Alan Cox        :       Cache last socket
  46 *              Alan Cox        :       Route cache
  47 *              Jon Peatfield   :       Minor efficiency fix to sendto().
  48 *              Mike Shaver     :       RFC1122 checks.
  49 *              Alan Cox        :       Nonblocking error fix.
  50 *      Willy Konynenberg       :       Transparent proxying support.
  51 *              Mike McLagan    :       Routing by source
  52 *              David S. Miller :       New socket lookup architecture.
  53 *                                      Last socket cache retained as it
  54 *                                      does have a high hit rate.
  55 *              Olaf Kirch      :       Don't linearise iovec on sendmsg.
  56 *              Andi Kleen      :       Some cleanups, cache destination entry
  57 *                                      for connect.
  58 *      Vitaly E. Lavrov        :       Transparent proxy revived after year coma.
  59 *              Melvin Smith    :       Check msg_name not msg_namelen in sendto(),
  60 *                                      return ENOTCONN for unconnected sockets (POSIX)
  61 *              Janos Farkas    :       don't deliver multi/broadcasts to a different
  62 *                                      bound-to-device socket
  63 *      Hirokazu Takahashi      :       HW checksumming for outgoing UDP
  64 *                                      datagrams.
  65 *      Hirokazu Takahashi      :       sendfile() on UDP works now.
  66 *              Arnaldo C. Melo :       convert /proc/net/udp to seq_file
  67 *      YOSHIFUJI Hideaki @USAGI and:   Support IPV6_V6ONLY socket option, which
  68 *      Alexey Kuznetsov:               allow both IPv4 and IPv6 sockets to bind
  69 *                                      a single port at the same time.
  70 *      Derek Atkins <derek@ihtfp.com>: Add Encapulation Support
  71 *      James Chapman           :       Add L2TP encapsulation type.
  72 *
  73 *
  74 *              This program is free software; you can redistribute it and/or
  75 *              modify it under the terms of the GNU General Public License
  76 *              as published by the Free Software Foundation; either version
  77 *              2 of the License, or (at your option) any later version.
  78 */
  79
  80#define pr_fmt(fmt) "UDP: " fmt
  81
  82#include <asm/uaccess.h>
  83#include <asm/ioctls.h>
  84#include <linux/bootmem.h>
  85#include <linux/highmem.h>
  86#include <linux/swap.h>
  87#include <linux/types.h>
  88#include <linux/fcntl.h>
  89#include <linux/module.h>
  90#include <linux/socket.h>
  91#include <linux/sockios.h>
  92#include <linux/igmp.h>
  93#include <linux/in.h>
  94#include <linux/errno.h>
  95#include <linux/timer.h>
  96#include <linux/mm.h>
  97#include <linux/inet.h>
  98#include <linux/netdevice.h>
  99#include <linux/slab.h>
 100#include <net/tcp_states.h>
 101#include <linux/skbuff.h>
 102#include <linux/proc_fs.h>
 103#include <linux/seq_file.h>
 104#include <net/net_namespace.h>
 105#include <net/icmp.h>
 106#include <net/route.h>
 107#include <net/checksum.h>
 108#include <net/xfrm.h>
 109#include <trace/events/udp.h>
 110#include <linux/static_key.h>
 111#include <trace/events/skb.h>
 112#include "udp_impl.h"
 113
 114struct udp_table udp_table __read_mostly;
 115EXPORT_SYMBOL(udp_table);
 116
 117long sysctl_udp_mem[3] __read_mostly;
 118EXPORT_SYMBOL(sysctl_udp_mem);
 119
 120int sysctl_udp_rmem_min __read_mostly;
 121EXPORT_SYMBOL(sysctl_udp_rmem_min);
 122
 123int sysctl_udp_wmem_min __read_mostly;
 124EXPORT_SYMBOL(sysctl_udp_wmem_min);
 125
 126atomic_long_t udp_memory_allocated;
 127EXPORT_SYMBOL(udp_memory_allocated);
 128
 129#define MAX_UDP_PORTS 65536
 130#define PORTS_PER_CHAIN (MAX_UDP_PORTS / UDP_HTABLE_SIZE_MIN)
 131
 132static int udp_lib_lport_inuse(struct net *net, __u16 num,
 133                               const struct udp_hslot *hslot,
 134                               unsigned long *bitmap,
 135                               struct sock *sk,
 136                               int (*saddr_comp)(const struct sock *sk1,
 137                                                 const struct sock *sk2),
 138                               unsigned int log)
 139{
 140        struct sock *sk2;
 141        struct hlist_nulls_node *node;
 142
 143        sk_nulls_for_each(sk2, node, &hslot->head)
 144                if (net_eq(sock_net(sk2), net) &&
 145                    sk2 != sk &&
 146                    (bitmap || udp_sk(sk2)->udp_port_hash == num) &&
 147                    (!sk2->sk_reuse || !sk->sk_reuse) &&
 148                    (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
 149                     sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
 150                    (*saddr_comp)(sk, sk2)) {
 151                        if (bitmap)
 152                                __set_bit(udp_sk(sk2)->udp_port_hash >> log,
 153                                          bitmap);
 154                        else
 155                                return 1;
 156                }
 157        return 0;
 158}
 159
 160/*
 161 * Note: we still hold spinlock of primary hash chain, so no other writer
 162 * can insert/delete a socket with local_port == num
 163 */
 164static int udp_lib_lport_inuse2(struct net *net, __u16 num,
 165                               struct udp_hslot *hslot2,
 166                               struct sock *sk,
 167                               int (*saddr_comp)(const struct sock *sk1,
 168                                                 const struct sock *sk2))
 169{
 170        struct sock *sk2;
 171        struct hlist_nulls_node *node;
 172        int res = 0;
 173
 174        spin_lock(&hslot2->lock);
 175        udp_portaddr_for_each_entry(sk2, node, &hslot2->head)
 176                if (net_eq(sock_net(sk2), net) &&
 177                    sk2 != sk &&
 178                    (udp_sk(sk2)->udp_port_hash == num) &&
 179                    (!sk2->sk_reuse || !sk->sk_reuse) &&
 180                    (!sk2->sk_bound_dev_if || !sk->sk_bound_dev_if ||
 181                     sk2->sk_bound_dev_if == sk->sk_bound_dev_if) &&
 182                    (*saddr_comp)(sk, sk2)) {
 183                        res = 1;
 184                        break;
 185                }
 186        spin_unlock(&hslot2->lock);
 187        return res;
 188}
 189
 190/**
 191 *  udp_lib_get_port  -  UDP/-Lite port lookup for IPv4 and IPv6
 192 *
 193 *  @sk:          socket struct in question
 194 *  @snum:        port number to look up
 195 *  @saddr_comp:  AF-dependent comparison of bound local IP addresses
 196 *  @hash2_nulladdr: AF-dependent hash value in secondary hash chains,
 197 *                   with NULL address
 198 */
 199int udp_lib_get_port(struct sock *sk, unsigned short snum,
 200                       int (*saddr_comp)(const struct sock *sk1,
 201                                         const struct sock *sk2),
 202                     unsigned int hash2_nulladdr)
 203{
 204        struct udp_hslot *hslot, *hslot2;
 205        struct udp_table *udptable = sk->sk_prot->h.udp_table;
 206        int    error = 1;
 207        struct net *net = sock_net(sk);
 208
 209        if (!snum) {
 210                int low, high, remaining;
 211                unsigned int rand;
 212                unsigned short first, last;
 213                DECLARE_BITMAP(bitmap, PORTS_PER_CHAIN);
 214
 215                inet_get_local_port_range(&low, &high);
 216                remaining = (high - low) + 1;
 217
 218                rand = net_random();
 219                first = (((u64)rand * remaining) >> 32) + low;
 220                /*
 221                 * force rand to be an odd multiple of UDP_HTABLE_SIZE
 222                 */
 223                rand = (rand | 1) * (udptable->mask + 1);
 224                last = first + udptable->mask + 1;
 225                do {
 226                        hslot = udp_hashslot(udptable, net, first);
 227                        bitmap_zero(bitmap, PORTS_PER_CHAIN);
 228                        spin_lock_bh(&hslot->lock);
 229                        udp_lib_lport_inuse(net, snum, hslot, bitmap, sk,
 230                                            saddr_comp, udptable->log);
 231
 232                        snum = first;
 233                        /*
 234                         * Iterate on all possible values of snum for this hash.
 235                         * Using steps of an odd multiple of UDP_HTABLE_SIZE
 236                         * give us randomization and full range coverage.
 237                         */
 238                        do {
 239                                if (low <= snum && snum <= high &&
 240                                    !test_bit(snum >> udptable->log, bitmap) &&
 241                                    !inet_is_reserved_local_port(snum))
 242                                        goto found;
 243                                snum += rand;
 244                        } while (snum != first);
 245                        spin_unlock_bh(&hslot->lock);
 246                } while (++first != last);
 247                goto fail;
 248        } else {
 249                hslot = udp_hashslot(udptable, net, snum);
 250                spin_lock_bh(&hslot->lock);
 251                if (hslot->count > 10) {
 252                        int exist;
 253                        unsigned int slot2 = udp_sk(sk)->udp_portaddr_hash ^ snum;
 254
 255                        slot2          &= udptable->mask;
 256                        hash2_nulladdr &= udptable->mask;
 257
 258                        hslot2 = udp_hashslot2(udptable, slot2);
 259                        if (hslot->count < hslot2->count)
 260                                goto scan_primary_hash;
 261
 262                        exist = udp_lib_lport_inuse2(net, snum, hslot2,
 263                                                     sk, saddr_comp);
 264                        if (!exist && (hash2_nulladdr != slot2)) {
 265                                hslot2 = udp_hashslot2(udptable, hash2_nulladdr);
 266                                exist = udp_lib_lport_inuse2(net, snum, hslot2,
 267                                                             sk, saddr_comp);
 268                        }
 269                        if (exist)
 270                                goto fail_unlock;
 271                        else
 272                                goto found;
 273                }
 274scan_primary_hash:
 275                if (udp_lib_lport_inuse(net, snum, hslot, NULL, sk,
 276                                        saddr_comp, 0))
 277                        goto fail_unlock;
 278        }
 279found:
 280        inet_sk(sk)->inet_num = snum;
 281        udp_sk(sk)->udp_port_hash = snum;
 282        udp_sk(sk)->udp_portaddr_hash ^= snum;
 283        if (sk_unhashed(sk)) {
 284                sk_nulls_add_node_rcu(sk, &hslot->head);
 285                hslot->count++;
 286                sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
 287
 288                hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
 289                spin_lock(&hslot2->lock);
 290                hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
 291                                         &hslot2->head);
 292                hslot2->count++;
 293                spin_unlock(&hslot2->lock);
 294        }
 295        error = 0;
 296fail_unlock:
 297        spin_unlock_bh(&hslot->lock);
 298fail:
 299        return error;
 300}
 301EXPORT_SYMBOL(udp_lib_get_port);
 302
 303static int ipv4_rcv_saddr_equal(const struct sock *sk1, const struct sock *sk2)
 304{
 305        struct inet_sock *inet1 = inet_sk(sk1), *inet2 = inet_sk(sk2);
 306
 307        return  (!ipv6_only_sock(sk2)  &&
 308                 (!inet1->inet_rcv_saddr || !inet2->inet_rcv_saddr ||
 309                   inet1->inet_rcv_saddr == inet2->inet_rcv_saddr));
 310}
 311
 312static unsigned int udp4_portaddr_hash(struct net *net, __be32 saddr,
 313                                       unsigned int port)
 314{
 315        return jhash_1word((__force u32)saddr, net_hash_mix(net)) ^ port;
 316}
 317
 318int udp_v4_get_port(struct sock *sk, unsigned short snum)
 319{
 320        unsigned int hash2_nulladdr =
 321                udp4_portaddr_hash(sock_net(sk), htonl(INADDR_ANY), snum);
 322        unsigned int hash2_partial =
 323                udp4_portaddr_hash(sock_net(sk), inet_sk(sk)->inet_rcv_saddr, 0);
 324
 325        /* precompute partial secondary hash */
 326        udp_sk(sk)->udp_portaddr_hash = hash2_partial;
 327        return udp_lib_get_port(sk, snum, ipv4_rcv_saddr_equal, hash2_nulladdr);
 328}
 329
 330static inline int compute_score(struct sock *sk, struct net *net, __be32 saddr,
 331                         unsigned short hnum,
 332                         __be16 sport, __be32 daddr, __be16 dport, int dif)
 333{
 334        int score = -1;
 335
 336        if (net_eq(sock_net(sk), net) && udp_sk(sk)->udp_port_hash == hnum &&
 337                        !ipv6_only_sock(sk)) {
 338                struct inet_sock *inet = inet_sk(sk);
 339
 340                score = (sk->sk_family == PF_INET ? 1 : 0);
 341                if (inet->inet_rcv_saddr) {
 342                        if (inet->inet_rcv_saddr != daddr)
 343                                return -1;
 344                        score += 2;
 345                }
 346                if (inet->inet_daddr) {
 347                        if (inet->inet_daddr != saddr)
 348                                return -1;
 349                        score += 2;
 350                }
 351                if (inet->inet_dport) {
 352                        if (inet->inet_dport != sport)
 353                                return -1;
 354                        score += 2;
 355                }
 356                if (sk->sk_bound_dev_if) {
 357                        if (sk->sk_bound_dev_if != dif)
 358                                return -1;
 359                        score += 2;
 360                }
 361        }
 362        return score;
 363}
 364
 365/*
 366 * In this second variant, we check (daddr, dport) matches (inet_rcv_sadd, inet_num)
 367 */
 368#define SCORE2_MAX (1 + 2 + 2 + 2)
 369static inline int compute_score2(struct sock *sk, struct net *net,
 370                                 __be32 saddr, __be16 sport,
 371                                 __be32 daddr, unsigned int hnum, int dif)
 372{
 373        int score = -1;
 374
 375        if (net_eq(sock_net(sk), net) && !ipv6_only_sock(sk)) {
 376                struct inet_sock *inet = inet_sk(sk);
 377
 378                if (inet->inet_rcv_saddr != daddr)
 379                        return -1;
 380                if (inet->inet_num != hnum)
 381                        return -1;
 382
 383                score = (sk->sk_family == PF_INET ? 1 : 0);
 384                if (inet->inet_daddr) {
 385                        if (inet->inet_daddr != saddr)
 386                                return -1;
 387                        score += 2;
 388                }
 389                if (inet->inet_dport) {
 390                        if (inet->inet_dport != sport)
 391                                return -1;
 392                        score += 2;
 393                }
 394                if (sk->sk_bound_dev_if) {
 395                        if (sk->sk_bound_dev_if != dif)
 396                                return -1;
 397                        score += 2;
 398                }
 399        }
 400        return score;
 401}
 402
 403
 404/* called with read_rcu_lock() */
 405static struct sock *udp4_lib_lookup2(struct net *net,
 406                __be32 saddr, __be16 sport,
 407                __be32 daddr, unsigned int hnum, int dif,
 408                struct udp_hslot *hslot2, unsigned int slot2)
 409{
 410        struct sock *sk, *result;
 411        struct hlist_nulls_node *node;
 412        int score, badness;
 413
 414begin:
 415        result = NULL;
 416        badness = -1;
 417        udp_portaddr_for_each_entry_rcu(sk, node, &hslot2->head) {
 418                score = compute_score2(sk, net, saddr, sport,
 419                                      daddr, hnum, dif);
 420                if (score > badness) {
 421                        result = sk;
 422                        badness = score;
 423                        if (score == SCORE2_MAX)
 424                                goto exact_match;
 425                }
 426        }
 427        /*
 428         * if the nulls value we got at the end of this lookup is
 429         * not the expected one, we must restart lookup.
 430         * We probably met an item that was moved to another chain.
 431         */
 432        if (get_nulls_value(node) != slot2)
 433                goto begin;
 434
 435        if (result) {
 436exact_match:
 437                if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
 438                        result = NULL;
 439                else if (unlikely(compute_score2(result, net, saddr, sport,
 440                                  daddr, hnum, dif) < badness)) {
 441                        sock_put(result);
 442                        goto begin;
 443                }
 444        }
 445        return result;
 446}
 447
 448/* UDP is nearly always wildcards out the wazoo, it makes no sense to try
 449 * harder than this. -DaveM
 450 */
 451struct sock *__udp4_lib_lookup(struct net *net, __be32 saddr,
 452                __be16 sport, __be32 daddr, __be16 dport,
 453                int dif, struct udp_table *udptable)
 454{
 455        struct sock *sk, *result;
 456        struct hlist_nulls_node *node;
 457        unsigned short hnum = ntohs(dport);
 458        unsigned int hash2, slot2, slot = udp_hashfn(net, hnum, udptable->mask);
 459        struct udp_hslot *hslot2, *hslot = &udptable->hash[slot];
 460        int score, badness;
 461
 462        rcu_read_lock();
 463        if (hslot->count > 10) {
 464                hash2 = udp4_portaddr_hash(net, daddr, hnum);
 465                slot2 = hash2 & udptable->mask;
 466                hslot2 = &udptable->hash2[slot2];
 467                if (hslot->count < hslot2->count)
 468                        goto begin;
 469
 470                result = udp4_lib_lookup2(net, saddr, sport,
 471                                          daddr, hnum, dif,
 472                                          hslot2, slot2);
 473                if (!result) {
 474                        hash2 = udp4_portaddr_hash(net, htonl(INADDR_ANY), hnum);
 475                        slot2 = hash2 & udptable->mask;
 476                        hslot2 = &udptable->hash2[slot2];
 477                        if (hslot->count < hslot2->count)
 478                                goto begin;
 479
 480                        result = udp4_lib_lookup2(net, saddr, sport,
 481                                                  htonl(INADDR_ANY), hnum, dif,
 482                                                  hslot2, slot2);
 483                }
 484                rcu_read_unlock();
 485                return result;
 486        }
 487begin:
 488        result = NULL;
 489        badness = -1;
 490        sk_nulls_for_each_rcu(sk, node, &hslot->head) {
 491                score = compute_score(sk, net, saddr, hnum, sport,
 492                                      daddr, dport, dif);
 493                if (score > badness) {
 494                        result = sk;
 495                        badness = score;
 496                }
 497        }
 498        /*
 499         * if the nulls value we got at the end of this lookup is
 500         * not the expected one, we must restart lookup.
 501         * We probably met an item that was moved to another chain.
 502         */
 503        if (get_nulls_value(node) != slot)
 504                goto begin;
 505
 506        if (result) {
 507                if (unlikely(!atomic_inc_not_zero_hint(&result->sk_refcnt, 2)))
 508                        result = NULL;
 509                else if (unlikely(compute_score(result, net, saddr, hnum, sport,
 510                                  daddr, dport, dif) < badness)) {
 511                        sock_put(result);
 512                        goto begin;
 513                }
 514        }
 515        rcu_read_unlock();
 516        return result;
 517}
 518EXPORT_SYMBOL_GPL(__udp4_lib_lookup);
 519
 520static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
 521                                                 __be16 sport, __be16 dport,
 522                                                 struct udp_table *udptable)
 523{
 524        struct sock *sk;
 525        const struct iphdr *iph = ip_hdr(skb);
 526
 527        if (unlikely(sk = skb_steal_sock(skb)))
 528                return sk;
 529        else
 530                return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport,
 531                                         iph->daddr, dport, inet_iif(skb),
 532                                         udptable);
 533}
 534
 535struct sock *udp4_lib_lookup(struct net *net, __be32 saddr, __be16 sport,
 536                             __be32 daddr, __be16 dport, int dif)
 537{
 538        return __udp4_lib_lookup(net, saddr, sport, daddr, dport, dif, &udp_table);
 539}
 540EXPORT_SYMBOL_GPL(udp4_lib_lookup);
 541
 542static inline struct sock *udp_v4_mcast_next(struct net *net, struct sock *sk,
 543                                             __be16 loc_port, __be32 loc_addr,
 544                                             __be16 rmt_port, __be32 rmt_addr,
 545                                             int dif)
 546{
 547        struct hlist_nulls_node *node;
 548        struct sock *s = sk;
 549        unsigned short hnum = ntohs(loc_port);
 550
 551        sk_nulls_for_each_from(s, node) {
 552                struct inet_sock *inet = inet_sk(s);
 553
 554                if (!net_eq(sock_net(s), net) ||
 555                    udp_sk(s)->udp_port_hash != hnum ||
 556                    (inet->inet_daddr && inet->inet_daddr != rmt_addr) ||
 557                    (inet->inet_dport != rmt_port && inet->inet_dport) ||
 558                    (inet->inet_rcv_saddr &&
 559                     inet->inet_rcv_saddr != loc_addr) ||
 560                    ipv6_only_sock(s) ||
 561                    (s->sk_bound_dev_if && s->sk_bound_dev_if != dif))
 562                        continue;
 563                if (!ip_mc_sf_allow(s, loc_addr, rmt_addr, dif))
 564                        continue;
 565                goto found;
 566        }
 567        s = NULL;
 568found:
 569        return s;
 570}
 571
 572/*
 573 * This routine is called by the ICMP module when it gets some
 574 * sort of error condition.  If err < 0 then the socket should
 575 * be closed and the error returned to the user.  If err > 0
 576 * it's just the icmp type << 8 | icmp code.
 577 * Header points to the ip header of the error packet. We move
 578 * on past this. Then (as it used to claim before adjustment)
 579 * header points to the first 8 bytes of the udp header.  We need
 580 * to find the appropriate port.
 581 */
 582
 583void __udp4_lib_err(struct sk_buff *skb, u32 info, struct udp_table *udptable)
 584{
 585        struct inet_sock *inet;
 586        const struct iphdr *iph = (const struct iphdr *)skb->data;
 587        struct udphdr *uh = (struct udphdr *)(skb->data+(iph->ihl<<2));
 588        const int type = icmp_hdr(skb)->type;
 589        const int code = icmp_hdr(skb)->code;
 590        struct sock *sk;
 591        int harderr;
 592        int err;
 593        struct net *net = dev_net(skb->dev);
 594
 595        sk = __udp4_lib_lookup(net, iph->daddr, uh->dest,
 596                        iph->saddr, uh->source, skb->dev->ifindex, udptable);
 597        if (sk == NULL) {
 598                ICMP_INC_STATS_BH(net, ICMP_MIB_INERRORS);
 599                return; /* No socket for error */
 600        }
 601
 602        err = 0;
 603        harderr = 0;
 604        inet = inet_sk(sk);
 605
 606        switch (type) {
 607        default:
 608        case ICMP_TIME_EXCEEDED:
 609                err = EHOSTUNREACH;
 610                break;
 611        case ICMP_SOURCE_QUENCH:
 612                goto out;
 613        case ICMP_PARAMETERPROB:
 614                err = EPROTO;
 615                harderr = 1;
 616                break;
 617        case ICMP_DEST_UNREACH:
 618                if (code == ICMP_FRAG_NEEDED) { /* Path MTU discovery */
 619                        ipv4_sk_update_pmtu(skb, sk, info);
 620                        if (inet->pmtudisc != IP_PMTUDISC_DONT) {
 621                                err = EMSGSIZE;
 622                                harderr = 1;
 623                                break;
 624                        }
 625                        goto out;
 626                }
 627                err = EHOSTUNREACH;
 628                if (code <= NR_ICMP_UNREACH) {
 629                        harderr = icmp_err_convert[code].fatal;
 630                        err = icmp_err_convert[code].errno;
 631                }
 632                break;
 633        case ICMP_REDIRECT:
 634                ipv4_sk_redirect(skb, sk);
 635                break;
 636        }
 637
 638        /*
 639         *      RFC1122: OK.  Passes ICMP errors back to application, as per
 640         *      4.1.3.3.
 641         */
 642        if (!inet->recverr) {
 643                if (!harderr || sk->sk_state != TCP_ESTABLISHED)
 644                        goto out;
 645        } else
 646                ip_icmp_error(sk, skb, err, uh->dest, info, (u8 *)(uh+1));
 647
 648        sk->sk_err = err;
 649        sk->sk_error_report(sk);
 650out:
 651        sock_put(sk);
 652}
 653
 654void udp_err(struct sk_buff *skb, u32 info)
 655{
 656        __udp4_lib_err(skb, info, &udp_table);
 657}
 658
 659/*
 660 * Throw away all pending data and cancel the corking. Socket is locked.
 661 */
 662void udp_flush_pending_frames(struct sock *sk)
 663{
 664        struct udp_sock *up = udp_sk(sk);
 665
 666        if (up->pending) {
 667                up->len = 0;
 668                up->pending = 0;
 669                ip_flush_pending_frames(sk);
 670        }
 671}
 672EXPORT_SYMBOL(udp_flush_pending_frames);
 673
 674/**
 675 *      udp4_hwcsum  -  handle outgoing HW checksumming
 676 *      @skb:   sk_buff containing the filled-in UDP header
 677 *              (checksum field must be zeroed out)
 678 *      @src:   source IP address
 679 *      @dst:   destination IP address
 680 */
 681static void udp4_hwcsum(struct sk_buff *skb, __be32 src, __be32 dst)
 682{
 683        struct udphdr *uh = udp_hdr(skb);
 684        struct sk_buff *frags = skb_shinfo(skb)->frag_list;
 685        int offset = skb_transport_offset(skb);
 686        int len = skb->len - offset;
 687        int hlen = len;
 688        __wsum csum = 0;
 689
 690        if (!frags) {
 691                /*
 692                 * Only one fragment on the socket.
 693                 */
 694                skb->csum_start = skb_transport_header(skb) - skb->head;
 695                skb->csum_offset = offsetof(struct udphdr, check);
 696                uh->check = ~csum_tcpudp_magic(src, dst, len,
 697                                               IPPROTO_UDP, 0);
 698        } else {
 699                /*
 700                 * HW-checksum won't work as there are two or more
 701                 * fragments on the socket so that all csums of sk_buffs
 702                 * should be together
 703                 */
 704                do {
 705                        csum = csum_add(csum, frags->csum);
 706                        hlen -= frags->len;
 707                } while ((frags = frags->next));
 708
 709                csum = skb_checksum(skb, offset, hlen, csum);
 710                skb->ip_summed = CHECKSUM_NONE;
 711
 712                uh->check = csum_tcpudp_magic(src, dst, len, IPPROTO_UDP, csum);
 713                if (uh->check == 0)
 714                        uh->check = CSUM_MANGLED_0;
 715        }
 716}
 717
 718static int udp_send_skb(struct sk_buff *skb, struct flowi4 *fl4)
 719{
 720        struct sock *sk = skb->sk;
 721        struct inet_sock *inet = inet_sk(sk);
 722        struct udphdr *uh;
 723        int err = 0;
 724        int is_udplite = IS_UDPLITE(sk);
 725        int offset = skb_transport_offset(skb);
 726        int len = skb->len - offset;
 727        __wsum csum = 0;
 728
 729        /*
 730         * Create a UDP header
 731         */
 732        uh = udp_hdr(skb);
 733        uh->source = inet->inet_sport;
 734        uh->dest = fl4->fl4_dport;
 735        uh->len = htons(len);
 736        uh->check = 0;
 737
 738        if (is_udplite)                                  /*     UDP-Lite      */
 739                csum = udplite_csum(skb);
 740
 741        else if (sk->sk_no_check == UDP_CSUM_NOXMIT) {   /* UDP csum disabled */
 742
 743                skb->ip_summed = CHECKSUM_NONE;
 744                goto send;
 745
 746        } else if (skb->ip_summed == CHECKSUM_PARTIAL) { /* UDP hardware csum */
 747
 748                udp4_hwcsum(skb, fl4->saddr, fl4->daddr);
 749                goto send;
 750
 751        } else
 752                csum = udp_csum(skb);
 753
 754        /* add protocol-dependent pseudo-header */
 755        uh->check = csum_tcpudp_magic(fl4->saddr, fl4->daddr, len,
 756                                      sk->sk_protocol, csum);
 757        if (uh->check == 0)
 758                uh->check = CSUM_MANGLED_0;
 759
 760send:
 761        err = ip_send_skb(sock_net(sk), skb);
 762        if (err) {
 763                if (err == -ENOBUFS && !inet->recverr) {
 764                        UDP_INC_STATS_USER(sock_net(sk),
 765                                           UDP_MIB_SNDBUFERRORS, is_udplite);
 766                        err = 0;
 767                }
 768        } else
 769                UDP_INC_STATS_USER(sock_net(sk),
 770                                   UDP_MIB_OUTDATAGRAMS, is_udplite);
 771        return err;
 772}
 773
 774/*
 775 * Push out all pending data as one UDP datagram. Socket is locked.
 776 */
 777static int udp_push_pending_frames(struct sock *sk)
 778{
 779        struct udp_sock  *up = udp_sk(sk);
 780        struct inet_sock *inet = inet_sk(sk);
 781        struct flowi4 *fl4 = &inet->cork.fl.u.ip4;
 782        struct sk_buff *skb;
 783        int err = 0;
 784
 785        skb = ip_finish_skb(sk, fl4);
 786        if (!skb)
 787                goto out;
 788
 789        err = udp_send_skb(skb, fl4);
 790
 791out:
 792        up->len = 0;
 793        up->pending = 0;
 794        return err;
 795}
 796
 797int udp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
 798                size_t len)
 799{
 800        struct inet_sock *inet = inet_sk(sk);
 801        struct udp_sock *up = udp_sk(sk);
 802        struct flowi4 fl4_stack;
 803        struct flowi4 *fl4;
 804        int ulen = len;
 805        struct ipcm_cookie ipc;
 806        struct rtable *rt = NULL;
 807        int free = 0;
 808        int connected = 0;
 809        __be32 daddr, faddr, saddr;
 810        __be16 dport;
 811        u8  tos;
 812        int err, is_udplite = IS_UDPLITE(sk);
 813        int corkreq = up->corkflag || msg->msg_flags&MSG_MORE;
 814        int (*getfrag)(void *, char *, int, int, int, struct sk_buff *);
 815        struct sk_buff *skb;
 816        struct ip_options_data opt_copy;
 817
 818        if (len > 0xFFFF)
 819                return -EMSGSIZE;
 820
 821        /*
 822         *      Check the flags.
 823         */
 824
 825        if (msg->msg_flags & MSG_OOB) /* Mirror BSD error message compatibility */
 826                return -EOPNOTSUPP;
 827
 828        ipc.opt = NULL;
 829        ipc.tx_flags = 0;
 830
 831        getfrag = is_udplite ? udplite_getfrag : ip_generic_getfrag;
 832
 833        fl4 = &inet->cork.fl.u.ip4;
 834        if (up->pending) {
 835                /*
 836                 * There are pending frames.
 837                 * The socket lock must be held while it's corked.
 838                 */
 839                lock_sock(sk);
 840                if (likely(up->pending)) {
 841                        if (unlikely(up->pending != AF_INET)) {
 842                                release_sock(sk);
 843                                return -EINVAL;
 844                        }
 845                        goto do_append_data;
 846                }
 847                release_sock(sk);
 848        }
 849        ulen += sizeof(struct udphdr);
 850
 851        /*
 852         *      Get and verify the address.
 853         */
 854        if (msg->msg_name) {
 855                struct sockaddr_in *usin = (struct sockaddr_in *)msg->msg_name;
 856                if (msg->msg_namelen < sizeof(*usin))
 857                        return -EINVAL;
 858                if (usin->sin_family != AF_INET) {
 859                        if (usin->sin_family != AF_UNSPEC)
 860                                return -EAFNOSUPPORT;
 861                }
 862
 863                daddr = usin->sin_addr.s_addr;
 864                dport = usin->sin_port;
 865                if (dport == 0)
 866                        return -EINVAL;
 867        } else {
 868                if (sk->sk_state != TCP_ESTABLISHED)
 869                        return -EDESTADDRREQ;
 870                daddr = inet->inet_daddr;
 871                dport = inet->inet_dport;
 872                /* Open fast path for connected socket.
 873                   Route will not be used, if at least one option is set.
 874                 */
 875                connected = 1;
 876        }
 877        ipc.addr = inet->inet_saddr;
 878
 879        ipc.oif = sk->sk_bound_dev_if;
 880        err = sock_tx_timestamp(sk, &ipc.tx_flags);
 881        if (err)
 882                return err;
 883        if (msg->msg_controllen) {
 884                err = ip_cmsg_send(sock_net(sk), msg, &ipc);
 885                if (err)
 886                        return err;
 887                if (ipc.opt)
 888                        free = 1;
 889                connected = 0;
 890        }
 891        if (!ipc.opt) {
 892                struct ip_options_rcu *inet_opt;
 893
 894                rcu_read_lock();
 895                inet_opt = rcu_dereference(inet->inet_opt);
 896                if (inet_opt) {
 897                        memcpy(&opt_copy, inet_opt,
 898                               sizeof(*inet_opt) + inet_opt->opt.optlen);
 899                        ipc.opt = &opt_copy.opt;
 900                }
 901                rcu_read_unlock();
 902        }
 903
 904        saddr = ipc.addr;
 905        ipc.addr = faddr = daddr;
 906
 907        if (ipc.opt && ipc.opt->opt.srr) {
 908                if (!daddr)
 909                        return -EINVAL;
 910                faddr = ipc.opt->opt.faddr;
 911                connected = 0;
 912        }
 913        tos = RT_TOS(inet->tos);
 914        if (sock_flag(sk, SOCK_LOCALROUTE) ||
 915            (msg->msg_flags & MSG_DONTROUTE) ||
 916            (ipc.opt && ipc.opt->opt.is_strictroute)) {
 917                tos |= RTO_ONLINK;
 918                connected = 0;
 919        }
 920
 921        if (ipv4_is_multicast(daddr)) {
 922                if (!ipc.oif)
 923                        ipc.oif = inet->mc_index;
 924                if (!saddr)
 925                        saddr = inet->mc_addr;
 926                connected = 0;
 927        } else if (!ipc.oif)
 928                ipc.oif = inet->uc_index;
 929
 930        if (connected)
 931                rt = (struct rtable *)sk_dst_check(sk, 0);
 932
 933        if (rt == NULL) {
 934                struct net *net = sock_net(sk);
 935
 936                fl4 = &fl4_stack;
 937                flowi4_init_output(fl4, ipc.oif, sk->sk_mark, tos,
 938                                   RT_SCOPE_UNIVERSE, sk->sk_protocol,
 939                                   inet_sk_flowi_flags(sk)|FLOWI_FLAG_CAN_SLEEP,
 940                                   faddr, saddr, dport, inet->inet_sport);
 941
 942                security_sk_classify_flow(sk, flowi4_to_flowi(fl4));
 943                rt = ip_route_output_flow(net, fl4, sk);
 944                if (IS_ERR(rt)) {
 945                        err = PTR_ERR(rt);
 946                        rt = NULL;
 947                        if (err == -ENETUNREACH)
 948                                IP_INC_STATS_BH(net, IPSTATS_MIB_OUTNOROUTES);
 949                        goto out;
 950                }
 951
 952                err = -EACCES;
 953                if ((rt->rt_flags & RTCF_BROADCAST) &&
 954                    !sock_flag(sk, SOCK_BROADCAST))
 955                        goto out;
 956                if (connected)
 957                        sk_dst_set(sk, dst_clone(&rt->dst));
 958        }
 959
 960        if (msg->msg_flags&MSG_CONFIRM)
 961                goto do_confirm;
 962back_from_confirm:
 963
 964        saddr = fl4->saddr;
 965        if (!ipc.addr)
 966                daddr = ipc.addr = fl4->daddr;
 967
 968        /* Lockless fast path for the non-corking case. */
 969        if (!corkreq) {
 970                skb = ip_make_skb(sk, fl4, getfrag, msg->msg_iov, ulen,
 971                                  sizeof(struct udphdr), &ipc, &rt,
 972                                  msg->msg_flags);
 973                err = PTR_ERR(skb);
 974                if (skb && !IS_ERR(skb))
 975                        err = udp_send_skb(skb, fl4);
 976                goto out;
 977        }
 978
 979        lock_sock(sk);
 980        if (unlikely(up->pending)) {
 981                /* The socket is already corked while preparing it. */
 982                /* ... which is an evident application bug. --ANK */
 983                release_sock(sk);
 984
 985                LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("cork app bug 2\n"));
 986                err = -EINVAL;
 987                goto out;
 988        }
 989        /*
 990         *      Now cork the socket to pend data.
 991         */
 992        fl4 = &inet->cork.fl.u.ip4;
 993        fl4->daddr = daddr;
 994        fl4->saddr = saddr;
 995        fl4->fl4_dport = dport;
 996        fl4->fl4_sport = inet->inet_sport;
 997        up->pending = AF_INET;
 998
 999do_append_data:
1000        up->len += ulen;
1001        err = ip_append_data(sk, fl4, getfrag, msg->msg_iov, ulen,
1002                             sizeof(struct udphdr), &ipc, &rt,
1003                             corkreq ? msg->msg_flags|MSG_MORE : msg->msg_flags);
1004        if (err)
1005                udp_flush_pending_frames(sk);
1006        else if (!corkreq)
1007                err = udp_push_pending_frames(sk);
1008        else if (unlikely(skb_queue_empty(&sk->sk_write_queue)))
1009                up->pending = 0;
1010        release_sock(sk);
1011
1012out:
1013        ip_rt_put(rt);
1014        if (free)
1015                kfree(ipc.opt);
1016        if (!err)
1017                return len;
1018        /*
1019         * ENOBUFS = no kernel mem, SOCK_NOSPACE = no sndbuf space.  Reporting
1020         * ENOBUFS might not be good (it's not tunable per se), but otherwise
1021         * we don't have a good statistic (IpOutDiscards but it can be too many
1022         * things).  We could add another new stat but at least for now that
1023         * seems like overkill.
1024         */
1025        if (err == -ENOBUFS || test_bit(SOCK_NOSPACE, &sk->sk_socket->flags)) {
1026                UDP_INC_STATS_USER(sock_net(sk),
1027                                UDP_MIB_SNDBUFERRORS, is_udplite);
1028        }
1029        return err;
1030
1031do_confirm:
1032        dst_confirm(&rt->dst);
1033        if (!(msg->msg_flags&MSG_PROBE) || len)
1034                goto back_from_confirm;
1035        err = 0;
1036        goto out;
1037}
1038EXPORT_SYMBOL(udp_sendmsg);
1039
1040int udp_sendpage(struct sock *sk, struct page *page, int offset,
1041                 size_t size, int flags)
1042{
1043        struct inet_sock *inet = inet_sk(sk);
1044        struct udp_sock *up = udp_sk(sk);
1045        int ret;
1046
1047        if (!up->pending) {
1048                struct msghdr msg = {   .msg_flags = flags|MSG_MORE };
1049
1050                /* Call udp_sendmsg to specify destination address which
1051                 * sendpage interface can't pass.
1052                 * This will succeed only when the socket is connected.
1053                 */
1054                ret = udp_sendmsg(NULL, sk, &msg, 0);
1055                if (ret < 0)
1056                        return ret;
1057        }
1058
1059        lock_sock(sk);
1060
1061        if (unlikely(!up->pending)) {
1062                release_sock(sk);
1063
1064                LIMIT_NETDEBUG(KERN_DEBUG pr_fmt("udp cork app bug 3\n"));
1065                return -EINVAL;
1066        }
1067
1068        ret = ip_append_page(sk, &inet->cork.fl.u.ip4,
1069                             page, offset, size, flags);
1070        if (ret == -EOPNOTSUPP) {
1071                release_sock(sk);
1072                return sock_no_sendpage(sk->sk_socket, page, offset,
1073                                        size, flags);
1074        }
1075        if (ret < 0) {
1076                udp_flush_pending_frames(sk);
1077                goto out;
1078        }
1079
1080        up->len += size;
1081        if (!(up->corkflag || (flags&MSG_MORE)))
1082                ret = udp_push_pending_frames(sk);
1083        if (!ret)
1084                ret = size;
1085out:
1086        release_sock(sk);
1087        return ret;
1088}
1089
1090
1091/**
1092 *      first_packet_length     - return length of first packet in receive queue
1093 *      @sk: socket
1094 *
1095 *      Drops all bad checksum frames, until a valid one is found.
1096 *      Returns the length of found skb, or 0 if none is found.
1097 */
1098static unsigned int first_packet_length(struct sock *sk)
1099{
1100        struct sk_buff_head list_kill, *rcvq = &sk->sk_receive_queue;
1101        struct sk_buff *skb;
1102        unsigned int res;
1103
1104        __skb_queue_head_init(&list_kill);
1105
1106        spin_lock_bh(&rcvq->lock);
1107        while ((skb = skb_peek(rcvq)) != NULL &&
1108                udp_lib_checksum_complete(skb)) {
1109                UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1110                                 IS_UDPLITE(sk));
1111                atomic_inc(&sk->sk_drops);
1112                __skb_unlink(skb, rcvq);
1113                __skb_queue_tail(&list_kill, skb);
1114        }
1115        res = skb ? skb->len : 0;
1116        spin_unlock_bh(&rcvq->lock);
1117
1118        if (!skb_queue_empty(&list_kill)) {
1119                bool slow = lock_sock_fast(sk);
1120
1121                __skb_queue_purge(&list_kill);
1122                sk_mem_reclaim_partial(sk);
1123                unlock_sock_fast(sk, slow);
1124        }
1125        return res;
1126}
1127
1128/*
1129 *      IOCTL requests applicable to the UDP protocol
1130 */
1131
1132int udp_ioctl(struct sock *sk, int cmd, unsigned long arg)
1133{
1134        switch (cmd) {
1135        case SIOCOUTQ:
1136        {
1137                int amount = sk_wmem_alloc_get(sk);
1138
1139                return put_user(amount, (int __user *)arg);
1140        }
1141
1142        case SIOCINQ:
1143        {
1144                unsigned int amount = first_packet_length(sk);
1145
1146                if (amount)
1147                        /*
1148                         * We will only return the amount
1149                         * of this packet since that is all
1150                         * that will be read.
1151                         */
1152                        amount -= sizeof(struct udphdr);
1153
1154                return put_user(amount, (int __user *)arg);
1155        }
1156
1157        default:
1158                return -ENOIOCTLCMD;
1159        }
1160
1161        return 0;
1162}
1163EXPORT_SYMBOL(udp_ioctl);
1164
1165/*
1166 *      This should be easy, if there is something there we
1167 *      return it, otherwise we block.
1168 */
1169
1170int udp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
1171                size_t len, int noblock, int flags, int *addr_len)
1172{
1173        struct inet_sock *inet = inet_sk(sk);
1174        struct sockaddr_in *sin = (struct sockaddr_in *)msg->msg_name;
1175        struct sk_buff *skb;
1176        unsigned int ulen, copied;
1177        int peeked, off = 0;
1178        int err;
1179        int is_udplite = IS_UDPLITE(sk);
1180        bool slow;
1181
1182        /*
1183         *      Check any passed addresses
1184         */
1185        if (addr_len)
1186                *addr_len = sizeof(*sin);
1187
1188        if (flags & MSG_ERRQUEUE)
1189                return ip_recv_error(sk, msg, len);
1190
1191try_again:
1192        skb = __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
1193                                  &peeked, &off, &err);
1194        if (!skb)
1195                goto out;
1196
1197        ulen = skb->len - sizeof(struct udphdr);
1198        copied = len;
1199        if (copied > ulen)
1200                copied = ulen;
1201        else if (copied < ulen)
1202                msg->msg_flags |= MSG_TRUNC;
1203
1204        /*
1205         * If checksum is needed at all, try to do it while copying the
1206         * data.  If the data is truncated, or if we only want a partial
1207         * coverage checksum (UDP-Lite), do it before the copy.
1208         */
1209
1210        if (copied < ulen || UDP_SKB_CB(skb)->partial_cov) {
1211                if (udp_lib_checksum_complete(skb))
1212                        goto csum_copy_err;
1213        }
1214
1215        if (skb_csum_unnecessary(skb))
1216                err = skb_copy_datagram_iovec(skb, sizeof(struct udphdr),
1217                                              msg->msg_iov, copied);
1218        else {
1219                err = skb_copy_and_csum_datagram_iovec(skb,
1220                                                       sizeof(struct udphdr),
1221                                                       msg->msg_iov);
1222
1223                if (err == -EINVAL)
1224                        goto csum_copy_err;
1225        }
1226
1227        if (unlikely(err)) {
1228                trace_kfree_skb(skb, udp_recvmsg);
1229                if (!peeked) {
1230                        atomic_inc(&sk->sk_drops);
1231                        UDP_INC_STATS_USER(sock_net(sk),
1232                                           UDP_MIB_INERRORS, is_udplite);
1233                }
1234                goto out_free;
1235        }
1236
1237        if (!peeked)
1238                UDP_INC_STATS_USER(sock_net(sk),
1239                                UDP_MIB_INDATAGRAMS, is_udplite);
1240
1241        sock_recv_ts_and_drops(msg, sk, skb);
1242
1243        /* Copy the address. */
1244        if (sin) {
1245                sin->sin_family = AF_INET;
1246                sin->sin_port = udp_hdr(skb)->source;
1247                sin->sin_addr.s_addr = ip_hdr(skb)->saddr;
1248                memset(sin->sin_zero, 0, sizeof(sin->sin_zero));
1249        }
1250        if (inet->cmsg_flags)
1251                ip_cmsg_recv(msg, skb);
1252
1253        err = copied;
1254        if (flags & MSG_TRUNC)
1255                err = ulen;
1256
1257out_free:
1258        skb_free_datagram_locked(sk, skb);
1259out:
1260        return err;
1261
1262csum_copy_err:
1263        slow = lock_sock_fast(sk);
1264        if (!skb_kill_datagram(sk, skb, flags))
1265                UDP_INC_STATS_USER(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1266        unlock_sock_fast(sk, slow);
1267
1268        if (noblock)
1269                return -EAGAIN;
1270
1271        /* starting over for a new packet */
1272        msg->msg_flags &= ~MSG_TRUNC;
1273        goto try_again;
1274}
1275
1276
1277int udp_disconnect(struct sock *sk, int flags)
1278{
1279        struct inet_sock *inet = inet_sk(sk);
1280        /*
1281         *      1003.1g - break association.
1282         */
1283
1284        sk->sk_state = TCP_CLOSE;
1285        inet->inet_daddr = 0;
1286        inet->inet_dport = 0;
1287        sock_rps_reset_rxhash(sk);
1288        sk->sk_bound_dev_if = 0;
1289        if (!(sk->sk_userlocks & SOCK_BINDADDR_LOCK))
1290                inet_reset_saddr(sk);
1291
1292        if (!(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) {
1293                sk->sk_prot->unhash(sk);
1294                inet->inet_sport = 0;
1295        }
1296        sk_dst_reset(sk);
1297        return 0;
1298}
1299EXPORT_SYMBOL(udp_disconnect);
1300
1301void udp_lib_unhash(struct sock *sk)
1302{
1303        if (sk_hashed(sk)) {
1304                struct udp_table *udptable = sk->sk_prot->h.udp_table;
1305                struct udp_hslot *hslot, *hslot2;
1306
1307                hslot  = udp_hashslot(udptable, sock_net(sk),
1308                                      udp_sk(sk)->udp_port_hash);
1309                hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1310
1311                spin_lock_bh(&hslot->lock);
1312                if (sk_nulls_del_node_init_rcu(sk)) {
1313                        hslot->count--;
1314                        inet_sk(sk)->inet_num = 0;
1315                        sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
1316
1317                        spin_lock(&hslot2->lock);
1318                        hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1319                        hslot2->count--;
1320                        spin_unlock(&hslot2->lock);
1321                }
1322                spin_unlock_bh(&hslot->lock);
1323        }
1324}
1325EXPORT_SYMBOL(udp_lib_unhash);
1326
1327/*
1328 * inet_rcv_saddr was changed, we must rehash secondary hash
1329 */
1330void udp_lib_rehash(struct sock *sk, u16 newhash)
1331{
1332        if (sk_hashed(sk)) {
1333                struct udp_table *udptable = sk->sk_prot->h.udp_table;
1334                struct udp_hslot *hslot, *hslot2, *nhslot2;
1335
1336                hslot2 = udp_hashslot2(udptable, udp_sk(sk)->udp_portaddr_hash);
1337                nhslot2 = udp_hashslot2(udptable, newhash);
1338                udp_sk(sk)->udp_portaddr_hash = newhash;
1339                if (hslot2 != nhslot2) {
1340                        hslot = udp_hashslot(udptable, sock_net(sk),
1341                                             udp_sk(sk)->udp_port_hash);
1342                        /* we must lock primary chain too */
1343                        spin_lock_bh(&hslot->lock);
1344
1345                        spin_lock(&hslot2->lock);
1346                        hlist_nulls_del_init_rcu(&udp_sk(sk)->udp_portaddr_node);
1347                        hslot2->count--;
1348                        spin_unlock(&hslot2->lock);
1349
1350                        spin_lock(&nhslot2->lock);
1351                        hlist_nulls_add_head_rcu(&udp_sk(sk)->udp_portaddr_node,
1352                                                 &nhslot2->head);
1353                        nhslot2->count++;
1354                        spin_unlock(&nhslot2->lock);
1355
1356                        spin_unlock_bh(&hslot->lock);
1357                }
1358        }
1359}
1360EXPORT_SYMBOL(udp_lib_rehash);
1361
1362static void udp_v4_rehash(struct sock *sk)
1363{
1364        u16 new_hash = udp4_portaddr_hash(sock_net(sk),
1365                                          inet_sk(sk)->inet_rcv_saddr,
1366                                          inet_sk(sk)->inet_num);
1367        udp_lib_rehash(sk, new_hash);
1368}
1369
1370static int __udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1371{
1372        int rc;
1373
1374        if (inet_sk(sk)->inet_daddr)
1375                sock_rps_save_rxhash(sk, skb);
1376
1377        rc = sock_queue_rcv_skb(sk, skb);
1378        if (rc < 0) {
1379                int is_udplite = IS_UDPLITE(sk);
1380
1381                /* Note that an ENOMEM error is charged twice */
1382                if (rc == -ENOMEM)
1383                        UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1384                                         is_udplite);
1385                UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1386                kfree_skb(skb);
1387                trace_udp_fail_queue_rcv_skb(rc, sk);
1388                return -1;
1389        }
1390
1391        return 0;
1392
1393}
1394
1395static struct static_key udp_encap_needed __read_mostly;
1396void udp_encap_enable(void)
1397{
1398        if (!static_key_enabled(&udp_encap_needed))
1399                static_key_slow_inc(&udp_encap_needed);
1400}
1401EXPORT_SYMBOL(udp_encap_enable);
1402
1403/* returns:
1404 *  -1: error
1405 *   0: success
1406 *  >0: "udp encap" protocol resubmission
1407 *
1408 * Note that in the success and error cases, the skb is assumed to
1409 * have either been requeued or freed.
1410 */
1411int udp_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
1412{
1413        struct udp_sock *up = udp_sk(sk);
1414        int rc;
1415        int is_udplite = IS_UDPLITE(sk);
1416
1417        /*
1418         *      Charge it to the socket, dropping if the queue is full.
1419         */
1420        if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))
1421                goto drop;
1422        nf_reset(skb);
1423
1424        if (static_key_false(&udp_encap_needed) && up->encap_type) {
1425                int (*encap_rcv)(struct sock *sk, struct sk_buff *skb);
1426
1427                /*
1428                 * This is an encapsulation socket so pass the skb to
1429                 * the socket's udp_encap_rcv() hook. Otherwise, just
1430                 * fall through and pass this up the UDP socket.
1431                 * up->encap_rcv() returns the following value:
1432                 * =0 if skb was successfully passed to the encap
1433                 *    handler or was discarded by it.
1434                 * >0 if skb should be passed on to UDP.
1435                 * <0 if skb should be resubmitted as proto -N
1436                 */
1437
1438                /* if we're overly short, let UDP handle it */
1439                encap_rcv = ACCESS_ONCE(up->encap_rcv);
1440                if (skb->len > sizeof(struct udphdr) && encap_rcv != NULL) {
1441                        int ret;
1442
1443                        ret = encap_rcv(sk, skb);
1444                        if (ret <= 0) {
1445                                UDP_INC_STATS_BH(sock_net(sk),
1446                                                 UDP_MIB_INDATAGRAMS,
1447                                                 is_udplite);
1448                                return -ret;
1449                        }
1450                }
1451
1452                /* FALLTHROUGH -- it's a UDP Packet */
1453        }
1454
1455        /*
1456         *      UDP-Lite specific tests, ignored on UDP sockets
1457         */
1458        if ((is_udplite & UDPLITE_RECV_CC)  &&  UDP_SKB_CB(skb)->partial_cov) {
1459
1460                /*
1461                 * MIB statistics other than incrementing the error count are
1462                 * disabled for the following two types of errors: these depend
1463                 * on the application settings, not on the functioning of the
1464                 * protocol stack as such.
1465                 *
1466                 * RFC 3828 here recommends (sec 3.3): "There should also be a
1467                 * way ... to ... at least let the receiving application block
1468                 * delivery of packets with coverage values less than a value
1469                 * provided by the application."
1470                 */
1471                if (up->pcrlen == 0) {          /* full coverage was set  */
1472                        LIMIT_NETDEBUG(KERN_WARNING "UDPLite: partial coverage %d while full coverage %d requested\n",
1473                                       UDP_SKB_CB(skb)->cscov, skb->len);
1474                        goto drop;
1475                }
1476                /* The next case involves violating the min. coverage requested
1477                 * by the receiver. This is subtle: if receiver wants x and x is
1478                 * greater than the buffersize/MTU then receiver will complain
1479                 * that it wants x while sender emits packets of smaller size y.
1480                 * Therefore the above ...()->partial_cov statement is essential.
1481                 */
1482                if (UDP_SKB_CB(skb)->cscov  <  up->pcrlen) {
1483                        LIMIT_NETDEBUG(KERN_WARNING "UDPLite: coverage %d too small, need min %d\n",
1484                                       UDP_SKB_CB(skb)->cscov, up->pcrlen);
1485                        goto drop;
1486                }
1487        }
1488
1489        if (rcu_access_pointer(sk->sk_filter) &&
1490            udp_lib_checksum_complete(skb))
1491                goto drop;
1492
1493
1494        if (sk_rcvqueues_full(sk, skb, sk->sk_rcvbuf))
1495                goto drop;
1496
1497        rc = 0;
1498
1499        ipv4_pktinfo_prepare(skb);
1500        bh_lock_sock(sk);
1501        if (!sock_owned_by_user(sk))
1502                rc = __udp_queue_rcv_skb(sk, skb);
1503        else if (sk_add_backlog(sk, skb, sk->sk_rcvbuf)) {
1504                bh_unlock_sock(sk);
1505                goto drop;
1506        }
1507        bh_unlock_sock(sk);
1508
1509        return rc;
1510
1511drop:
1512        UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS, is_udplite);
1513        atomic_inc(&sk->sk_drops);
1514        kfree_skb(skb);
1515        return -1;
1516}
1517
1518
1519static void flush_stack(struct sock **stack, unsigned int count,
1520                        struct sk_buff *skb, unsigned int final)
1521{
1522        unsigned int i;
1523        struct sk_buff *skb1 = NULL;
1524        struct sock *sk;
1525
1526        for (i = 0; i < count; i++) {
1527                sk = stack[i];
1528                if (likely(skb1 == NULL))
1529                        skb1 = (i == final) ? skb : skb_clone(skb, GFP_ATOMIC);
1530
1531                if (!skb1) {
1532                        atomic_inc(&sk->sk_drops);
1533                        UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_RCVBUFERRORS,
1534                                         IS_UDPLITE(sk));
1535                        UDP_INC_STATS_BH(sock_net(sk), UDP_MIB_INERRORS,
1536                                         IS_UDPLITE(sk));
1537                }
1538
1539                if (skb1 && udp_queue_rcv_skb(sk, skb1) <= 0)
1540                        skb1 = NULL;
1541        }
1542        if (unlikely(skb1))
1543                kfree_skb(skb1);
1544}
1545
1546/*
1547 *      Multicasts and broadcasts go to each listener.
1548 *
1549 *      Note: called only from the BH handler context.
1550 */
1551static int __udp4_lib_mcast_deliver(struct net *net, struct sk_buff *skb,
1552                                    struct udphdr  *uh,
1553                                    __be32 saddr, __be32 daddr,
1554                                    struct udp_table *udptable)
1555{
1556        struct sock *sk, *stack[256 / sizeof(struct sock *)];
1557        struct udp_hslot *hslot = udp_hashslot(udptable, net, ntohs(uh->dest));
1558        int dif;
1559        unsigned int i, count = 0;
1560
1561        spin_lock(&hslot->lock);
1562        sk = sk_nulls_head(&hslot->head);
1563        dif = skb->dev->ifindex;
1564        sk = udp_v4_mcast_next(net, sk, uh->dest, daddr, uh->source, saddr, dif);
1565        while (sk) {
1566                stack[count++] = sk;
1567                sk = udp_v4_mcast_next(net, sk_nulls_next(sk), uh->dest,
1568                                       daddr, uh->source, saddr, dif);
1569                if (unlikely(count == ARRAY_SIZE(stack))) {
1570                        if (!sk)
1571                                break;
1572                        flush_stack(stack, count, skb, ~0);
1573                        count = 0;
1574                }
1575        }
1576        /*
1577         * before releasing chain lock, we must take a reference on sockets
1578         */
1579        for (i = 0; i < count; i++)
1580                sock_hold(stack[i]);
1581
1582        spin_unlock(&hslot->lock);
1583
1584        /*
1585         * do the slow work with no lock held
1586         */
1587        if (count) {
1588                flush_stack(stack, count, skb, count - 1);
1589
1590                for (i = 0; i < count; i++)
1591                        sock_put(stack[i]);
1592        } else {
1593                kfree_skb(skb);
1594        }
1595        return 0;
1596}
1597
1598/* Initialize UDP checksum. If exited with zero value (success),
1599 * CHECKSUM_UNNECESSARY means, that no more checks are required.
1600 * Otherwise, csum completion requires chacksumming packet body,
1601 * including udp header and folding it to skb->csum.
1602 */
1603static inline int udp4_csum_init(struct sk_buff *skb, struct udphdr *uh,
1604                                 int proto)
1605{
1606        const struct iphdr *iph;
1607        int err;
1608
1609        UDP_SKB_CB(skb)->partial_cov = 0;
1610        UDP_SKB_CB(skb)->cscov = skb->len;
1611
1612        if (proto == IPPROTO_UDPLITE) {
1613                err = udplite_checksum_init(skb, uh);
1614                if (err)
1615                        return err;
1616        }
1617
1618        iph = ip_hdr(skb);
1619        if (uh->check == 0) {
1620                skb->ip_summed = CHECKSUM_UNNECESSARY;
1621        } else if (skb->ip_summed == CHECKSUM_COMPLETE) {
1622                if (!csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
1623                                      proto, skb->csum))
1624                        skb->ip_summed = CHECKSUM_UNNECESSARY;
1625        }
1626        if (!skb_csum_unnecessary(skb))
1627                skb->csum = csum_tcpudp_nofold(iph->saddr, iph->daddr,
1628                                               skb->len, proto, 0);
1629        /* Probably, we should checksum udp header (it should be in cache
1630         * in any case) and data in tiny packets (< rx copybreak).
1631         */
1632
1633        return 0;
1634}
1635
1636/*
1637 *      All we need to do is get the socket, and then do a checksum.
1638 */
1639
1640int __udp4_lib_rcv(struct sk_buff *skb, struct udp_table *udptable,
1641                   int proto)
1642{
1643        struct sock *sk;
1644        struct udphdr *uh;
1645        unsigned short ulen;
1646        struct rtable *rt = skb_rtable(skb);
1647        __be32 saddr, daddr;
1648        struct net *net = dev_net(skb->dev);
1649
1650        /*
1651         *  Validate the packet.
1652         */
1653        if (!pskb_may_pull(skb, sizeof(struct udphdr)))
1654                goto drop;              /* No space for header. */
1655
1656        uh   = udp_hdr(skb);
1657        ulen = ntohs(uh->len);
1658        saddr = ip_hdr(skb)->saddr;
1659        daddr = ip_hdr(skb)->daddr;
1660
1661        if (ulen > skb->len)
1662                goto short_packet;
1663
1664        if (proto == IPPROTO_UDP) {
1665                /* UDP validates ulen. */
1666                if (ulen < sizeof(*uh) || pskb_trim_rcsum(skb, ulen))
1667                        goto short_packet;
1668                uh = udp_hdr(skb);
1669        }
1670
1671        if (udp4_csum_init(skb, uh, proto))
1672                goto csum_error;
1673
1674        if (rt->rt_flags & (RTCF_BROADCAST|RTCF_MULTICAST))
1675                return __udp4_lib_mcast_deliver(net, skb, uh,
1676                                saddr, daddr, udptable);
1677
1678        sk = __udp4_lib_lookup_skb(skb, uh->source, uh->dest, udptable);
1679
1680        if (sk != NULL) {
1681                int ret = udp_queue_rcv_skb(sk, skb);
1682                sock_put(sk);
1683
1684                /* a return value > 0 means to resubmit the input, but
1685                 * it wants the return to be -protocol, or 0
1686                 */
1687                if (ret > 0)
1688                        return -ret;
1689                return 0;
1690        }
1691
1692        if (!xfrm4_policy_check(NULL, XFRM_POLICY_IN, skb))
1693                goto drop;
1694        nf_reset(skb);
1695
1696        /* No socket. Drop packet silently, if checksum is wrong */
1697        if (udp_lib_checksum_complete(skb))
1698                goto csum_error;
1699
1700        UDP_INC_STATS_BH(net, UDP_MIB_NOPORTS, proto == IPPROTO_UDPLITE);
1701        icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
1702
1703        /*
1704         * Hmm.  We got an UDP packet to a port to which we
1705         * don't wanna listen.  Ignore it.
1706         */
1707        kfree_skb(skb);
1708        return 0;
1709
1710short_packet:
1711        LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: short packet: From %pI4:%u %d/%d to %pI4:%u\n",
1712                       proto == IPPROTO_UDPLITE ? "Lite" : "",
1713                       &saddr, ntohs(uh->source),
1714                       ulen, skb->len,
1715                       &daddr, ntohs(uh->dest));
1716        goto drop;
1717
1718csum_error:
1719        /*
1720         * RFC1122: OK.  Discards the bad packet silently (as far as
1721         * the network is concerned, anyway) as per 4.1.3.4 (MUST).
1722         */
1723        LIMIT_NETDEBUG(KERN_DEBUG "UDP%s: bad checksum. From %pI4:%u to %pI4:%u ulen %d\n",
1724                       proto == IPPROTO_UDPLITE ? "Lite" : "",
1725                       &saddr, ntohs(uh->source), &daddr, ntohs(uh->dest),
1726                       ulen);
1727drop:
1728        UDP_INC_STATS_BH(net, UDP_MIB_INERRORS, proto == IPPROTO_UDPLITE);
1729        kfree_skb(skb);
1730        return 0;
1731}
1732
1733int udp_rcv(struct sk_buff *skb)
1734{
1735        return __udp4_lib_rcv(skb, &udp_table, IPPROTO_UDP);
1736}
1737
1738void udp_destroy_sock(struct sock *sk)
1739{
1740        bool slow = lock_sock_fast(sk);
1741        udp_flush_pending_frames(sk);
1742        unlock_sock_fast(sk, slow);
1743}
1744
1745/*
1746 *      Socket option code for UDP
1747 */
1748int udp_lib_setsockopt(struct sock *sk, int level, int optname,
1749                       char __user *optval, unsigned int optlen,
1750                       int (*push_pending_frames)(struct sock *))
1751{
1752        struct udp_sock *up = udp_sk(sk);
1753        int val;
1754        int err = 0;
1755        int is_udplite = IS_UDPLITE(sk);
1756
1757        if (optlen < sizeof(int))
1758                return -EINVAL;
1759
1760        if (get_user(val, (int __user *)optval))
1761                return -EFAULT;
1762
1763        switch (optname) {
1764        case UDP_CORK:
1765                if (val != 0) {
1766                        up->corkflag = 1;
1767                } else {
1768                        up->corkflag = 0;
1769                        lock_sock(sk);
1770                        (*push_pending_frames)(sk);
1771                        release_sock(sk);
1772                }
1773                break;
1774
1775        case UDP_ENCAP:
1776                switch (val) {
1777                case 0:
1778                case UDP_ENCAP_ESPINUDP:
1779                case UDP_ENCAP_ESPINUDP_NON_IKE:
1780                        up->encap_rcv = xfrm4_udp_encap_rcv;
1781                        /* FALLTHROUGH */
1782                case UDP_ENCAP_L2TPINUDP:
1783                        up->encap_type = val;
1784                        udp_encap_enable();
1785                        break;
1786                default:
1787                        err = -ENOPROTOOPT;
1788                        break;
1789                }
1790                break;
1791
1792        /*
1793         *      UDP-Lite's partial checksum coverage (RFC 3828).
1794         */
1795        /* The sender sets actual checksum coverage length via this option.
1796         * The case coverage > packet length is handled by send module. */
1797        case UDPLITE_SEND_CSCOV:
1798                if (!is_udplite)         /* Disable the option on UDP sockets */
1799                        return -ENOPROTOOPT;
1800                if (val != 0 && val < 8) /* Illegal coverage: use default (8) */
1801                        val = 8;
1802                else if (val > USHRT_MAX)
1803                        val = USHRT_MAX;
1804                up->pcslen = val;
1805                up->pcflag |= UDPLITE_SEND_CC;
1806                break;
1807
1808        /* The receiver specifies a minimum checksum coverage value. To make
1809         * sense, this should be set to at least 8 (as done below). If zero is
1810         * used, this again means full checksum coverage.                     */
1811        case UDPLITE_RECV_CSCOV:
1812                if (!is_udplite)         /* Disable the option on UDP sockets */
1813                        return -ENOPROTOOPT;
1814                if (val != 0 && val < 8) /* Avoid silly minimal values.       */
1815                        val = 8;
1816                else if (val > USHRT_MAX)
1817                        val = USHRT_MAX;
1818                up->pcrlen = val;
1819                up->pcflag |= UDPLITE_RECV_CC;
1820                break;
1821
1822        default:
1823                err = -ENOPROTOOPT;
1824                break;
1825        }
1826
1827        return err;
1828}
1829EXPORT_SYMBOL(udp_lib_setsockopt);
1830
1831int udp_setsockopt(struct sock *sk, int level, int optname,
1832                   char __user *optval, unsigned int optlen)
1833{
1834        if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1835                return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1836                                          udp_push_pending_frames);
1837        return ip_setsockopt(sk, level, optname, optval, optlen);
1838}
1839
1840#ifdef CONFIG_COMPAT
1841int compat_udp_setsockopt(struct sock *sk, int level, int optname,
1842                          char __user *optval, unsigned int optlen)
1843{
1844        if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1845                return udp_lib_setsockopt(sk, level, optname, optval, optlen,
1846                                          udp_push_pending_frames);
1847        return compat_ip_setsockopt(sk, level, optname, optval, optlen);
1848}
1849#endif
1850
1851int udp_lib_getsockopt(struct sock *sk, int level, int optname,
1852                       char __user *optval, int __user *optlen)
1853{
1854        struct udp_sock *up = udp_sk(sk);
1855        int val, len;
1856
1857        if (get_user(len, optlen))
1858                return -EFAULT;
1859
1860        len = min_t(unsigned int, len, sizeof(int));
1861
1862        if (len < 0)
1863                return -EINVAL;
1864
1865        switch (optname) {
1866        case UDP_CORK:
1867                val = up->corkflag;
1868                break;
1869
1870        case UDP_ENCAP:
1871                val = up->encap_type;
1872                break;
1873
1874        /* The following two cannot be changed on UDP sockets, the return is
1875         * always 0 (which corresponds to the full checksum coverage of UDP). */
1876        case UDPLITE_SEND_CSCOV:
1877                val = up->pcslen;
1878                break;
1879
1880        case UDPLITE_RECV_CSCOV:
1881                val = up->pcrlen;
1882                break;
1883
1884        default:
1885                return -ENOPROTOOPT;
1886        }
1887
1888        if (put_user(len, optlen))
1889                return -EFAULT;
1890        if (copy_to_user(optval, &val, len))
1891                return -EFAULT;
1892        return 0;
1893}
1894EXPORT_SYMBOL(udp_lib_getsockopt);
1895
1896int udp_getsockopt(struct sock *sk, int level, int optname,
1897                   char __user *optval, int __user *optlen)
1898{
1899        if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1900                return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1901        return ip_getsockopt(sk, level, optname, optval, optlen);
1902}
1903
1904#ifdef CONFIG_COMPAT
1905int compat_udp_getsockopt(struct sock *sk, int level, int optname,
1906                                 char __user *optval, int __user *optlen)
1907{
1908        if (level == SOL_UDP  ||  level == SOL_UDPLITE)
1909                return udp_lib_getsockopt(sk, level, optname, optval, optlen);
1910        return compat_ip_getsockopt(sk, level, optname, optval, optlen);
1911}
1912#endif
1913/**
1914 *      udp_poll - wait for a UDP event.
1915 *      @file - file struct
1916 *      @sock - socket
1917 *      @wait - poll table
1918 *
1919 *      This is same as datagram poll, except for the special case of
1920 *      blocking sockets. If application is using a blocking fd
1921 *      and a packet with checksum error is in the queue;
1922 *      then it could get return from select indicating data available
1923 *      but then block when reading it. Add special case code
1924 *      to work around these arguably broken applications.
1925 */
1926unsigned int udp_poll(struct file *file, struct socket *sock, poll_table *wait)
1927{
1928        unsigned int mask = datagram_poll(file, sock, wait);
1929        struct sock *sk = sock->sk;
1930
1931        /* Check for false positives due to checksum errors */
1932        if ((mask & POLLRDNORM) && !(file->f_flags & O_NONBLOCK) &&
1933            !(sk->sk_shutdown & RCV_SHUTDOWN) && !first_packet_length(sk))
1934                mask &= ~(POLLIN | POLLRDNORM);
1935
1936        return mask;
1937
1938}
1939EXPORT_SYMBOL(udp_poll);
1940
1941struct proto udp_prot = {
1942        .name              = "UDP",
1943        .owner             = THIS_MODULE,
1944        .close             = udp_lib_close,
1945        .connect           = ip4_datagram_connect,
1946        .disconnect        = udp_disconnect,
1947        .ioctl             = udp_ioctl,
1948        .destroy           = udp_destroy_sock,
1949        .setsockopt        = udp_setsockopt,
1950        .getsockopt        = udp_getsockopt,
1951        .sendmsg           = udp_sendmsg,
1952        .recvmsg           = udp_recvmsg,
1953        .sendpage          = udp_sendpage,
1954        .backlog_rcv       = __udp_queue_rcv_skb,
1955        .hash              = udp_lib_hash,
1956        .unhash            = udp_lib_unhash,
1957        .rehash            = udp_v4_rehash,
1958        .get_port          = udp_v4_get_port,
1959        .memory_allocated  = &udp_memory_allocated,
1960        .sysctl_mem        = sysctl_udp_mem,
1961        .sysctl_wmem       = &sysctl_udp_wmem_min,
1962        .sysctl_rmem       = &sysctl_udp_rmem_min,
1963        .obj_size          = sizeof(struct udp_sock),
1964        .slab_flags        = SLAB_DESTROY_BY_RCU,
1965        .h.udp_table       = &udp_table,
1966#ifdef CONFIG_COMPAT
1967        .compat_setsockopt = compat_udp_setsockopt,
1968        .compat_getsockopt = compat_udp_getsockopt,
1969#endif
1970        .clear_sk          = sk_prot_clear_portaddr_nulls,
1971};
1972EXPORT_SYMBOL(udp_prot);
1973
1974/* ------------------------------------------------------------------------ */
1975#ifdef CONFIG_PROC_FS
1976
1977static struct sock *udp_get_first(struct seq_file *seq, int start)
1978{
1979        struct sock *sk;
1980        struct udp_iter_state *state = seq->private;
1981        struct net *net = seq_file_net(seq);
1982
1983        for (state->bucket = start; state->bucket <= state->udp_table->mask;
1984             ++state->bucket) {
1985                struct hlist_nulls_node *node;
1986                struct udp_hslot *hslot = &state->udp_table->hash[state->bucket];
1987
1988                if (hlist_nulls_empty(&hslot->head))
1989                        continue;
1990
1991                spin_lock_bh(&hslot->lock);
1992                sk_nulls_for_each(sk, node, &hslot->head) {
1993                        if (!net_eq(sock_net(sk), net))
1994                                continue;
1995                        if (sk->sk_family == state->family)
1996                                goto found;
1997                }
1998                spin_unlock_bh(&hslot->lock);
1999        }
2000        sk = NULL;
2001found:
2002        return sk;
2003}
2004
2005static struct sock *udp_get_next(struct seq_file *seq, struct sock *sk)
2006{
2007        struct udp_iter_state *state = seq->private;
2008        struct net *net = seq_file_net(seq);
2009
2010        do {
2011                sk = sk_nulls_next(sk);
2012        } while (sk && (!net_eq(sock_net(sk), net) || sk->sk_family != state->family));
2013
2014        if (!sk) {
2015                if (state->bucket <= state->udp_table->mask)
2016                        spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2017                return udp_get_first(seq, state->bucket + 1);
2018        }
2019        return sk;
2020}
2021
2022static struct sock *udp_get_idx(struct seq_file *seq, loff_t pos)
2023{
2024        struct sock *sk = udp_get_first(seq, 0);
2025
2026        if (sk)
2027                while (pos && (sk = udp_get_next(seq, sk)) != NULL)
2028                        --pos;
2029        return pos ? NULL : sk;
2030}
2031
2032static void *udp_seq_start(struct seq_file *seq, loff_t *pos)
2033{
2034        struct udp_iter_state *state = seq->private;
2035        state->bucket = MAX_UDP_PORTS;
2036
2037        return *pos ? udp_get_idx(seq, *pos-1) : SEQ_START_TOKEN;
2038}
2039
2040static void *udp_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2041{
2042        struct sock *sk;
2043
2044        if (v == SEQ_START_TOKEN)
2045                sk = udp_get_idx(seq, 0);
2046        else
2047                sk = udp_get_next(seq, v);
2048
2049        ++*pos;
2050        return sk;
2051}
2052
2053static void udp_seq_stop(struct seq_file *seq, void *v)
2054{
2055        struct udp_iter_state *state = seq->private;
2056
2057        if (state->bucket <= state->udp_table->mask)
2058                spin_unlock_bh(&state->udp_table->hash[state->bucket].lock);
2059}
2060
2061int udp_seq_open(struct inode *inode, struct file *file)
2062{
2063        struct udp_seq_afinfo *afinfo = PDE(inode)->data;
2064        struct udp_iter_state *s;
2065        int err;
2066
2067        err = seq_open_net(inode, file, &afinfo->seq_ops,
2068                           sizeof(struct udp_iter_state));
2069        if (err < 0)
2070                return err;
2071
2072        s = ((struct seq_file *)file->private_data)->private;
2073        s->family               = afinfo->family;
2074        s->udp_table            = afinfo->udp_table;
2075        return err;
2076}
2077EXPORT_SYMBOL(udp_seq_open);
2078
2079/* ------------------------------------------------------------------------ */
2080int udp_proc_register(struct net *net, struct udp_seq_afinfo *afinfo)
2081{
2082        struct proc_dir_entry *p;
2083        int rc = 0;
2084
2085        afinfo->seq_ops.start           = udp_seq_start;
2086        afinfo->seq_ops.next            = udp_seq_next;
2087        afinfo->seq_ops.stop            = udp_seq_stop;
2088
2089        p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
2090                             afinfo->seq_fops, afinfo);
2091        if (!p)
2092                rc = -ENOMEM;
2093        return rc;
2094}
2095EXPORT_SYMBOL(udp_proc_register);
2096
2097void udp_proc_unregister(struct net *net, struct udp_seq_afinfo *afinfo)
2098{
2099        proc_net_remove(net, afinfo->name);
2100}
2101EXPORT_SYMBOL(udp_proc_unregister);
2102
2103/* ------------------------------------------------------------------------ */
2104static void udp4_format_sock(struct sock *sp, struct seq_file *f,
2105                int bucket, int *len)
2106{
2107        struct inet_sock *inet = inet_sk(sp);
2108        __be32 dest = inet->inet_daddr;
2109        __be32 src  = inet->inet_rcv_saddr;
2110        __u16 destp       = ntohs(inet->inet_dport);
2111        __u16 srcp        = ntohs(inet->inet_sport);
2112
2113        seq_printf(f, "%5d: %08X:%04X %08X:%04X"
2114                " %02X %08X:%08X %02X:%08lX %08X %5d %8d %lu %d %pK %d%n",
2115                bucket, src, srcp, dest, destp, sp->sk_state,
2116                sk_wmem_alloc_get(sp),
2117                sk_rmem_alloc_get(sp),
2118                0, 0L, 0, sock_i_uid(sp), 0, sock_i_ino(sp),
2119                atomic_read(&sp->sk_refcnt), sp,
2120                atomic_read(&sp->sk_drops), len);
2121}
2122
2123int udp4_seq_show(struct seq_file *seq, void *v)
2124{
2125        if (v == SEQ_START_TOKEN)
2126                seq_printf(seq, "%-127s\n",
2127                           "  sl  local_address rem_address   st tx_queue "
2128                           "rx_queue tr tm->when retrnsmt   uid  timeout "
2129                           "inode ref pointer drops");
2130        else {
2131                struct udp_iter_state *state = seq->private;
2132                int len;
2133
2134                udp4_format_sock(v, seq, state->bucket, &len);
2135                seq_printf(seq, "%*s\n", 127 - len, "");
2136        }
2137        return 0;
2138}
2139
2140static const struct file_operations udp_afinfo_seq_fops = {
2141        .owner    = THIS_MODULE,
2142        .open     = udp_seq_open,
2143        .read     = seq_read,
2144        .llseek   = seq_lseek,
2145        .release  = seq_release_net
2146};
2147
2148/* ------------------------------------------------------------------------ */
2149static struct udp_seq_afinfo udp4_seq_afinfo = {
2150        .name           = "udp",
2151        .family         = AF_INET,
2152        .udp_table      = &udp_table,
2153        .seq_fops       = &udp_afinfo_seq_fops,
2154        .seq_ops        = {
2155                .show           = udp4_seq_show,
2156        },
2157};
2158
2159static int __net_init udp4_proc_init_net(struct net *net)
2160{
2161        return udp_proc_register(net, &udp4_seq_afinfo);
2162}
2163
2164static void __net_exit udp4_proc_exit_net(struct net *net)
2165{
2166        udp_proc_unregister(net, &udp4_seq_afinfo);
2167}
2168
2169static struct pernet_operations udp4_net_ops = {
2170        .init = udp4_proc_init_net,
2171        .exit = udp4_proc_exit_net,
2172};
2173
2174int __init udp4_proc_init(void)
2175{
2176        return register_pernet_subsys(&udp4_net_ops);
2177}
2178
2179void udp4_proc_exit(void)
2180{
2181        unregister_pernet_subsys(&udp4_net_ops);
2182}
2183#endif /* CONFIG_PROC_FS */
2184
2185static __initdata unsigned long uhash_entries;
2186static int __init set_uhash_entries(char *str)
2187{
2188        ssize_t ret;
2189
2190        if (!str)
2191                return 0;
2192
2193        ret = kstrtoul(str, 0, &uhash_entries);
2194        if (ret)
2195                return 0;
2196
2197        if (uhash_entries && uhash_entries < UDP_HTABLE_SIZE_MIN)
2198                uhash_entries = UDP_HTABLE_SIZE_MIN;
2199        return 1;
2200}
2201__setup("uhash_entries=", set_uhash_entries);
2202
2203void __init udp_table_init(struct udp_table *table, const char *name)
2204{
2205        unsigned int i;
2206
2207        table->hash = alloc_large_system_hash(name,
2208                                              2 * sizeof(struct udp_hslot),
2209                                              uhash_entries,
2210                                              21, /* one slot per 2 MB */
2211                                              0,
2212                                              &table->log,
2213                                              &table->mask,
2214                                              UDP_HTABLE_SIZE_MIN,
2215                                              64 * 1024);
2216
2217        table->hash2 = table->hash + (table->mask + 1);
2218        for (i = 0; i <= table->mask; i++) {
2219                INIT_HLIST_NULLS_HEAD(&table->hash[i].head, i);
2220                table->hash[i].count = 0;
2221                spin_lock_init(&table->hash[i].lock);
2222        }
2223        for (i = 0; i <= table->mask; i++) {
2224                INIT_HLIST_NULLS_HEAD(&table->hash2[i].head, i);
2225                table->hash2[i].count = 0;
2226                spin_lock_init(&table->hash2[i].lock);
2227        }
2228}
2229
2230void __init udp_init(void)
2231{
2232        unsigned long limit;
2233
2234        udp_table_init(&udp_table, "UDP");
2235        limit = nr_free_buffer_pages() / 8;
2236        limit = max(limit, 128UL);
2237        sysctl_udp_mem[0] = limit / 4 * 3;
2238        sysctl_udp_mem[1] = limit;
2239        sysctl_udp_mem[2] = sysctl_udp_mem[0] * 2;
2240
2241        sysctl_udp_rmem_min = SK_MEM_QUANTUM;
2242        sysctl_udp_wmem_min = SK_MEM_QUANTUM;
2243}
2244
2245int udp4_ufo_send_check(struct sk_buff *skb)
2246{
2247        const struct iphdr *iph;
2248        struct udphdr *uh;
2249
2250        if (!pskb_may_pull(skb, sizeof(*uh)))
2251                return -EINVAL;
2252
2253        iph = ip_hdr(skb);
2254        uh = udp_hdr(skb);
2255
2256        uh->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, skb->len,
2257                                       IPPROTO_UDP, 0);
2258        skb->csum_start = skb_transport_header(skb) - skb->head;
2259        skb->csum_offset = offsetof(struct udphdr, check);
2260        skb->ip_summed = CHECKSUM_PARTIAL;
2261        return 0;
2262}
2263
2264struct sk_buff *udp4_ufo_fragment(struct sk_buff *skb,
2265        netdev_features_t features)
2266{
2267        struct sk_buff *segs = ERR_PTR(-EINVAL);
2268        unsigned int mss;
2269        int offset;
2270        __wsum csum;
2271
2272        mss = skb_shinfo(skb)->gso_size;
2273        if (unlikely(skb->len <= mss))
2274                goto out;
2275
2276        if (skb_gso_ok(skb, features | NETIF_F_GSO_ROBUST)) {
2277                /* Packet is from an untrusted source, reset gso_segs. */
2278                int type = skb_shinfo(skb)->gso_type;
2279
2280                if (unlikely(type & ~(SKB_GSO_UDP | SKB_GSO_DODGY) ||
2281                             !(type & (SKB_GSO_UDP))))
2282                        goto out;
2283
2284                skb_shinfo(skb)->gso_segs = DIV_ROUND_UP(skb->len, mss);
2285
2286                segs = NULL;
2287                goto out;
2288        }
2289
2290        /* Do software UFO. Complete and fill in the UDP checksum as HW cannot
2291         * do checksum of UDP packets sent as multiple IP fragments.
2292         */
2293        offset = skb_checksum_start_offset(skb);
2294        csum = skb_checksum(skb, offset, skb->len - offset, 0);
2295        offset += skb->csum_offset;
2296        *(__sum16 *)(skb->data + offset) = csum_fold(csum);
2297        skb->ip_summed = CHECKSUM_NONE;
2298
2299        /* Fragment the skb. IP headers of the fragments are updated in
2300         * inet_gso_segment()
2301         */
2302        segs = skb_segment(skb, features);
2303out:
2304        return segs;
2305}
2306
2307
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.