net/core/sock.c

   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  *              Generic socket support routines. Memory allocators, socket lock/release
   7  *              handler for protocols to use and generic option handler.
   8  *
   9  *
  10  * Version:     $Id: sock.c,v 1.117 2002/02/01 22:01:03 davem Exp $
  11  *
  12  * Authors:     Ross Biro, <bir7@leland.Stanford.Edu>
  13  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  14  *              Florian La Roche, <flla@stud.uni-sb.de>
  15  *              Alan Cox, <A.Cox@swansea.ac.uk>
  16  *
  17  * Fixes:
  18  *              Alan Cox        :       Numerous verify_area() problems
  19  *              Alan Cox        :       Connecting on a connecting socket
  20  *                                      now returns an error for tcp.
  21  *              Alan Cox        :       sock->protocol is set correctly.
  22  *                                      and is not sometimes left as 0.
  23  *              Alan Cox        :       connect handles icmp errors on a
  24  *                                      connect properly. Unfortunately there
  25  *                                      is a restart syscall nasty there. I
  26  *                                      can't match BSD without hacking the C
  27  *                                      library. Ideas urgently sought!
  28  *              Alan Cox        :       Disallow bind() to addresses that are
  29  *                                      not ours - especially broadcast ones!!
  30  *              Alan Cox        :       Socket 1024 _IS_ ok for users. (fencepost)
  31  *              Alan Cox        :       sock_wfree/sock_rfree don't destroy sockets,
  32  *                                      instead they leave that for the DESTROY timer.
  33  *              Alan Cox        :       Clean up error flag in accept
  34  *              Alan Cox        :       TCP ack handling is buggy, the DESTROY timer
  35  *                                      was buggy. Put a remove_sock() in the handler
  36  *                                      for memory when we hit 0. Also altered the timer
  37  *                                      code. The ACK stuff can wait and needs major
  38  *                                      TCP layer surgery.
  39  *              Alan Cox        :       Fixed TCP ack bug, removed remove sock
  40  *                                      and fixed timer/inet_bh race.
  41  *              Alan Cox        :       Added zapped flag for TCP
  42  *              Alan Cox        :       Move kfree_skb into skbuff.c and tidied up surplus code
  43  *              Alan Cox        :       for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
  44  *              Alan Cox        :       kfree_s calls now are kfree_skbmem so we can track skb resources
  45  *              Alan Cox        :       Supports socket option broadcast now as does udp. Packet and raw need fixing.
  46  *              Alan Cox        :       Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
  47  *              Rick Sladkey    :       Relaxed UDP rules for matching packets.
  48  *              C.E.Hawkins     :       IFF_PROMISC/SIOCGHWADDR support
  49  *      Pauline Middelink       :       identd support
  50  *              Alan Cox        :       Fixed connect() taking signals I think.
  51  *              Alan Cox        :       SO_LINGER supported
  52  *              Alan Cox        :       Error reporting fixes
  53  *              Anonymous       :       inet_create tidied up (sk->reuse setting)
  54  *              Alan Cox        :       inet sockets don't set sk->type!
  55  *              Alan Cox        :       Split socket option code
  56  *              Alan Cox        :       Callbacks
  57  *              Alan Cox        :       Nagle flag for Charles & Johannes stuff
  58  *              Alex            :       Removed restriction on inet fioctl
  59  *              Alan Cox        :       Splitting INET from NET core
  60  *              Alan Cox        :       Fixed bogus SO_TYPE handling in getsockopt()
  61  *              Adam Caldwell   :       Missing return in SO_DONTROUTE/SO_DEBUG code
  62  *              Alan Cox        :       Split IP from generic code
  63  *              Alan Cox        :       New kfree_skbmem()
  64  *              Alan Cox        :       Make SO_DEBUG superuser only.
  65  *              Alan Cox        :       Allow anyone to clear SO_DEBUG
  66  *                                      (compatibility fix)
  67  *              Alan Cox        :       Added optimistic memory grabbing for AF_UNIX throughput.
  68  *              Alan Cox        :       Allocator for a socket is settable.
  69  *              Alan Cox        :       SO_ERROR includes soft errors.
  70  *              Alan Cox        :       Allow NULL arguments on some SO_ opts
  71  *              Alan Cox        :       Generic socket allocation to make hooks
  72  *                                      easier (suggested by Craig Metz).
  73  *              Michael Pall    :       SO_ERROR returns positive errno again
  74  *              Steve Whitehouse:       Added default destructor to free
  75  *                                      protocol private data.
  76  *              Steve Whitehouse:       Added various other default routines
  77  *                                      common to several socket families.
  78  *              Chris Evans     :       Call suser() check last on F_SETOWN
  79  *              Jay Schulist    :       Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
  80  *              Andi Kleen      :       Add sock_kmalloc()/sock_kfree_s()
  81  *              Andi Kleen      :       Fix write_space callback
  82  *              Chris Evans     :       Security fixes - signedness again
  83  *              Arnaldo C. Melo :       cleanups, use skb_queue_purge
  84  *
  85  * To Fix:
  86  *
  87  *
  88  *              This program is free software; you can redistribute it and/or
  89  *              modify it under the terms of the GNU General Public License
  90  *              as published by the Free Software Foundation; either version
  91  *              2 of the License, or (at your option) any later version.
  92  */
  93
  94 #include <linux/config.h>
  95 #include <linux/errno.h>
  96 #include <linux/types.h>
  97 #include <linux/socket.h>
  98 #include <linux/in.h>
  99 #include <linux/kernel.h>
 100 #include <linux/major.h>
 101 #include <linux/module.h>
 102 #include <linux/sched.h>
 103 #include <linux/timer.h>
 104 #include <linux/string.h>
 105 #include <linux/sockios.h>
 106 #include <linux/net.h>
 107 #include <linux/mm.h>
 108 #include <linux/slab.h>
 109 #include <linux/interrupt.h>
 110 #include <linux/poll.h>
 111 #include <linux/tcp.h>
 112 #include <linux/init.h>
 113
 114 #include <asm/uaccess.h>
 115 #include <asm/system.h>
 116
 117 #include <linux/netdevice.h>
 118 #include <net/protocol.h>
 119 #include <linux/skbuff.h>
 120 #include <net/sock.h>
 121 #include <net/xfrm.h>
 122 #include <linux/ipsec.h>
 123
 124 #include <linux/filter.h>
 125 #include <linux/vs_socket.h>
 126
 127 #ifdef CONFIG_INET
 128 #include <net/tcp.h>
 129 #endif
 130
 131 /* Take into consideration the size of the struct sk_buff overhead in the
 132  * determination of these values, since that is non-constant across
 133  * platforms.  This makes socket queueing behavior and performance
 134  * not depend upon such differences.
 135  */
 136 #define _SK_MEM_PACKETS         256
 137 #define _SK_MEM_OVERHEAD        (sizeof(struct sk_buff) + 256)
 138 #define SK_WMEM_MAX             (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
 139 #define SK_RMEM_MAX             (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
 140
 141 /* Run time adjustable parameters. */
 142 __u32 sysctl_wmem_max = SK_WMEM_MAX;
 143 __u32 sysctl_rmem_max = SK_RMEM_MAX;
 144 __u32 sysctl_wmem_default = SK_WMEM_MAX;
 145 __u32 sysctl_rmem_default = SK_RMEM_MAX;
 146
 147 /* Maximal space eaten by iovec or ancilliary data plus some space */
 148 int sysctl_optmem_max = sizeof(unsigned long)*(2*UIO_MAXIOV + 512);
 149
 150 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
 151 {
 152         struct timeval tv;
 153
 154         if (optlen < sizeof(tv))
 155                 return -EINVAL;
 156         if (copy_from_user(&tv, optval, sizeof(tv)))
 157                 return -EFAULT;
 158
 159         *timeo_p = MAX_SCHEDULE_TIMEOUT;
 160         if (tv.tv_sec == 0 && tv.tv_usec == 0)
 161                 return 0;
 162         if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
 163                 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
 164         return 0;
 165 }
 166
 167 static void sock_warn_obsolete_bsdism(const char *name)
 168 {
 169         static int warned;
 170         static char warncomm[16];
 171         if (strcmp(warncomm, current->comm) && warned < 5) {
 172                 strcpy(warncomm,  current->comm);
 173                 printk(KERN_WARNING "process `%s' is using obsolete "
 174                        "%s SO_BSDCOMPAT\n", warncomm, name);
 175                 warned++;
 176         }
 177 }
 178
 179 /*
 180  *      This is meant for all protocols to use and covers goings on
 181  *      at the socket level. Everything here is generic.
 182  */
 183
 184 int sock_setsockopt(struct socket *sock, int level, int optname,
 185                     char __user *optval, int optlen)
 186 {
 187         struct sock *sk=sock->sk;
 188         struct sk_filter *filter;
 189         int val;
 190         int valbool;
 191         struct linger ling;
 192         int ret = 0;
 193
 194         /*
 195          *      Options without arguments
 196          */
 197
 198 #ifdef SO_DONTLINGER            /* Compatibility item... */
 199         switch (optname) {
 200                 case SO_DONTLINGER:
 201                         sock_reset_flag(sk, SOCK_LINGER);
 202                         return 0;
 203         }
 204 #endif
 205
 206         if(optlen<sizeof(int))
 207                 return(-EINVAL);
 208
 209         if (get_user(val, (int __user *)optval))
 210                 return -EFAULT;
 211
 212         valbool = val?1:0;
 213
 214         lock_sock(sk);
 215
 216         switch(optname)
 217         {
 218                 case SO_DEBUG:
 219                         if(val && !capable(CAP_NET_ADMIN))
 220                         {
 221                                 ret = -EACCES;
 222                         }
 223                         else
 224                                 sk->sk_debug = valbool;
 225                         break;
 226                 case SO_REUSEADDR:
 227                         sk->sk_reuse = valbool;
 228                         break;
 229                 case SO_TYPE:
 230                 case SO_ERROR:
 231                         ret = -ENOPROTOOPT;
 232                         break;
 233                 case SO_DONTROUTE:
 234                         sk->sk_localroute = valbool;
 235                         break;
 236                 case SO_BROADCAST:
 237                         sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
 238                         break;
 239                 case SO_SNDBUF:
 240                         /* Don't error on this BSD doesn't and if you think
 241                            about it this is right. Otherwise apps have to
 242                            play 'guess the biggest size' games. RCVBUF/SNDBUF
 243                            are treated in BSD as hints */
 244
 245                         if (val > sysctl_wmem_max)
 246                                 val = sysctl_wmem_max;
 247
 248                         sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
 249                         if ((val * 2) < SOCK_MIN_SNDBUF)
 250                                 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
 251                         else
 252                                 sk->sk_sndbuf = val * 2;
 253
 254                         /*
 255                          *      Wake up sending tasks if we
 256                          *      upped the value.
 257                          */
 258                         sk->sk_write_space(sk);
 259                         break;
 260
 261                 case SO_RCVBUF:
 262                         /* Don't error on this BSD doesn't and if you think
 263                            about it this is right. Otherwise apps have to
 264                            play 'guess the biggest size' games. RCVBUF/SNDBUF
 265                            are treated in BSD as hints */
 266
 267                         if (val > sysctl_rmem_max)
 268                                 val = sysctl_rmem_max;
 269
 270                         sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
 271                         /* FIXME: is this lower bound the right one? */
 272                         if ((val * 2) < SOCK_MIN_RCVBUF)
 273                                 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
 274                         else
 275                                 sk->sk_rcvbuf = val * 2;
 276                         break;
 277
 278                 case SO_KEEPALIVE:
 279 #ifdef CONFIG_INET
 280                         if (sk->sk_protocol == IPPROTO_TCP)
 281                                 tcp_set_keepalive(sk, valbool);
 282 #endif
 283                         sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
 284                         break;
 285
 286                 case SO_OOBINLINE:
 287                         sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
 288                         break;
 289
 290                 case SO_NO_CHECK:
 291                         sk->sk_no_check = valbool;
 292                         break;
 293
 294                 case SO_PRIORITY:
 295                         if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
 296                                 sk->sk_priority = val;
 297                         else
 298                                 ret = -EPERM;
 299                         break;
 300
 301                 case SO_LINGER:
 302                         if(optlen<sizeof(ling)) {
 303                                 ret = -EINVAL;  /* 1003.1g */
 304                                 break;
 305                         }
 306                         if (copy_from_user(&ling,optval,sizeof(ling))) {
 307                                 ret = -EFAULT;
 308                                 break;
 309                         }
 310                         if (!ling.l_onoff)
 311                                 sock_reset_flag(sk, SOCK_LINGER);
 312                         else {
 313 #if (BITS_PER_LONG == 32)
 314                                 if (ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
 315                                         sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
 316                                 else
 317 #endif
 318                                         sk->sk_lingertime = ling.l_linger * HZ;
 319                                 sock_set_flag(sk, SOCK_LINGER);
 320                         }
 321                         break;
 322
 323                 case SO_BSDCOMPAT:
 324                         sock_warn_obsolete_bsdism("setsockopt");
 325                         break;
 326
 327                 case SO_PASSCRED:
 328                         if (valbool)
 329                                 set_bit(SOCK_PASS_CRED, &sock->flags);
 330                         else
 331                                 clear_bit(SOCK_PASS_CRED, &sock->flags);
 332                         break;
 333
 334                 case SO_TIMESTAMP:
 335                         sk->sk_rcvtstamp = valbool;
 336                         if (valbool)
 337                                 sock_enable_timestamp(sk);
 338                         break;
 339
 340                 case SO_RCVLOWAT:
 341                         if (val < 0)
 342                                 val = INT_MAX;
 343                         sk->sk_rcvlowat = val ? : 1;
 344                         break;
 345
 346                 case SO_RCVTIMEO:
 347                         ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
 348                         break;
 349
 350                 case SO_SNDTIMEO:
 351                         ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
 352                         break;
 353
 354 #ifdef CONFIG_NETDEVICES
 355                 case SO_BINDTODEVICE:
 356                 {
 357                         char devname[IFNAMSIZ];
 358
 359                         /* Sorry... */
 360                         if (!capable(CAP_NET_RAW)) {
 361                                 ret = -EPERM;
 362                                 break;
 363                         }
 364
 365                         /* Bind this socket to a particular device like "eth0",
 366                          * as specified in the passed interface name. If the
 367                          * name is "" or the option length is zero the socket
 368                          * is not bound.
 369                          */
 370
 371                         if (!valbool) {
 372                                 sk->sk_bound_dev_if = 0;
 373                         } else {
 374                                 if (optlen > IFNAMSIZ)
 375                                         optlen = IFNAMSIZ;
 376                                 if (copy_from_user(devname, optval, optlen)) {
 377                                         ret = -EFAULT;
 378                                         break;
 379                                 }
 380
 381                                 /* Remove any cached route for this socket. */
 382                                 sk_dst_reset(sk);
 383
 384                                 if (devname[0] == '\0') {
 385                                         sk->sk_bound_dev_if = 0;
 386                                 } else {
 387                                         struct net_device *dev = dev_get_by_name(devname);
 388                                         if (!dev) {
 389                                                 ret = -ENODEV;
 390                                                 break;
 391                                         }
 392                                         sk->sk_bound_dev_if = dev->ifindex;
 393                                         dev_put(dev);
 394                                 }
 395                         }
 396                         break;
 397                 }
 398 #endif
 399
 400
 401                 case SO_ATTACH_FILTER:
 402                         ret = -EINVAL;
 403                         if (optlen == sizeof(struct sock_fprog)) {
 404                                 struct sock_fprog fprog;
 405
 406                                 ret = -EFAULT;
 407                                 if (copy_from_user(&fprog, optval, sizeof(fprog)))
 408                                         break;
 409
 410                                 ret = sk_attach_filter(&fprog, sk);
 411                         }
 412                         break;
 413
 414                 case SO_DETACH_FILTER:
 415                         spin_lock_bh(&sk->sk_lock.slock);
 416                         filter = sk->sk_filter;
 417                         if (filter) {
 418                                 sk->sk_filter = NULL;
 419                                 spin_unlock_bh(&sk->sk_lock.slock);
 420                                 sk_filter_release(sk, filter);
 421                                 break;
 422                         }
 423                         spin_unlock_bh(&sk->sk_lock.slock);
 424                         ret = -ENONET;
 425                         break;
 426
 427                 /* We implement the SO_SNDLOWAT etc to
 428                    not be settable (1003.1g 5.3) */
 429                 default:
 430                         ret = -ENOPROTOOPT;
 431                         break;
 432         }
 433         release_sock(sk);
 434         return ret;
 435 }
 436
 437
 438 int sock_getsockopt(struct socket *sock, int level, int optname,
 439                     char __user *optval, int __user *optlen)
 440 {
 441         struct sock *sk = sock->sk;
 442
 443         union
 444         {
 445                 int val;
 446                 struct linger ling;
 447                 struct timeval tm;
 448         } v;
 449
 450         unsigned int lv = sizeof(int);
 451         int len;
 452
 453         if(get_user(len,optlen))
 454                 return -EFAULT;
 455         if(len < 0)
 456                 return -EINVAL;
 457
 458         switch(optname)
 459         {
 460                 case SO_DEBUG:
 461                         v.val = sk->sk_debug;
 462                         break;
 463
 464                 case SO_DONTROUTE:
 465                         v.val = sk->sk_localroute;
 466                         break;
 467
 468                 case SO_BROADCAST:
 469                         v.val = !!sock_flag(sk, SOCK_BROADCAST);
 470                         break;
 471
 472                 case SO_SNDBUF:
 473                         v.val = sk->sk_sndbuf;
 474                         break;
 475
 476                 case SO_RCVBUF:
 477                         v.val = sk->sk_rcvbuf;
 478                         break;
 479
 480                 case SO_REUSEADDR:
 481                         v.val = sk->sk_reuse;
 482                         break;
 483
 484                 case SO_KEEPALIVE:
 485                         v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
 486                         break;
 487
 488                 case SO_TYPE:
 489                         v.val = sk->sk_type;
 490                         break;
 491
 492                 case SO_ERROR:
 493                         v.val = -sock_error(sk);
 494                         if(v.val==0)
 495                                 v.val = xchg(&sk->sk_err_soft, 0);
 496                         break;
 497
 498                 case SO_OOBINLINE:
 499                         v.val = !!sock_flag(sk, SOCK_URGINLINE);
 500                         break;
 501
 502                 case SO_NO_CHECK:
 503                         v.val = sk->sk_no_check;
 504                         break;
 505
 506                 case SO_PRIORITY:
 507                         v.val = sk->sk_priority;
 508                         break;
 509
 510                 case SO_LINGER:
 511                         lv              = sizeof(v.ling);
 512                         v.ling.l_onoff  = !!sock_flag(sk, SOCK_LINGER);
 513                         v.ling.l_linger = sk->sk_lingertime / HZ;
 514                         break;
 515
 516                 case SO_BSDCOMPAT:
 517                         sock_warn_obsolete_bsdism("getsockopt");
 518                         break;
 519
 520                 case SO_TIMESTAMP:
 521                         v.val = sk->sk_rcvtstamp;
 522                         break;
 523
 524                 case SO_RCVTIMEO:
 525                         lv=sizeof(struct timeval);
 526                         if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
 527                                 v.tm.tv_sec = 0;
 528                                 v.tm.tv_usec = 0;
 529                         } else {
 530                                 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
 531                                 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
 532                         }
 533                         break;
 534
 535                 case SO_SNDTIMEO:
 536                         lv=sizeof(struct timeval);
 537                         if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
 538                                 v.tm.tv_sec = 0;
 539                                 v.tm.tv_usec = 0;
 540                         } else {
 541                                 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
 542                                 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
 543                         }
 544                         break;
 545
 546                 case SO_RCVLOWAT:
 547                         v.val = sk->sk_rcvlowat;
 548                         break;
 549
 550                 case SO_SNDLOWAT:
 551                         v.val=1;
 552                         break;
 553
 554                 case SO_PASSCRED:
 555                         v.val = test_bit(SOCK_PASS_CRED, &sock->flags)?1:0;
 556                         break;
 557
 558                 case SO_PEERCRED:
 559                         if (len > sizeof(sk->sk_peercred))
 560                                 len = sizeof(sk->sk_peercred);
 561                         if (copy_to_user(optval, &sk->sk_peercred, len))
 562                                 return -EFAULT;
 563                         goto lenout;
 564
 565                 case SO_PEERNAME:
 566                 {
 567                         char address[128];
 568
 569                         if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
 570                                 return -ENOTCONN;
 571                         if (lv < len)
 572                                 return -EINVAL;
 573                         if (copy_to_user(optval, address, len))
 574                                 return -EFAULT;
 575                         goto lenout;
 576                 }
 577
 578                 /* Dubious BSD thing... Probably nobody even uses it, but
 579                  * the UNIX standard wants it for whatever reason... -DaveM
 580                  */
 581                 case SO_ACCEPTCONN:
 582                         v.val = sk->sk_state == TCP_LISTEN;
 583                         break;
 584
 585                 case SO_PEERSEC:
 586                         return security_socket_getpeersec(sock, optval, optlen, len);
 587
 588                 default:
 589                         return(-ENOPROTOOPT);
 590         }
 591         if (len > lv)
 592                 len = lv;
 593         if (copy_to_user(optval, &v, len))
 594                 return -EFAULT;
 595 lenout:
 596         if (put_user(len, optlen))
 597                 return -EFAULT;
 598         return 0;
 599 }
 600
 601 static kmem_cache_t *sk_cachep;
 602
 603 /**
 604  *      sk_alloc - All socket objects are allocated here
 605  *      @family - protocol family
 606  *      @priority - for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
 607  *      @zero_it - zeroes the allocated sock
 608  *      @slab - alternate slab
 609  *
 610  *      All socket objects are allocated here. If @zero_it is non-zero
 611  *      it should have the size of the are to be zeroed, because the
 612  *      private slabcaches have different sizes of the generic struct sock.
 613  *      1 has been kept as a way to say sizeof(struct sock).
 614  */
 615 struct sock *sk_alloc(int family, int priority, int zero_it, kmem_cache_t *slab)
 616 {
 617         struct sock *sk = NULL;
 618
 619         if (!slab)
 620                 slab = sk_cachep;
 621         sk = kmem_cache_alloc(slab, priority);
 622         if (sk) {
 623                 if (zero_it) {
 624                         memset(sk, 0,
 625                                zero_it == 1 ? sizeof(struct sock) : zero_it);
 626                         sk->sk_family = family;
 627                         sock_lock_init(sk);
 628                 }
 629                 sk->sk_slab = slab;
 630                 sock_vx_init(sk);
 631                 sock_nx_init(sk);
 632
 633                 if (security_sk_alloc(sk, family, priority)) {
 634                         kmem_cache_free(slab, sk);
 635                         sk = NULL;
 636                 }
 637         }
 638         return sk;
 639 }
 640
 641 void sk_free(struct sock *sk)
 642 {
 643         struct sk_filter *filter;
 644         struct module *owner = sk->sk_owner;
 645
 646         if (sk->sk_destruct)
 647                 sk->sk_destruct(sk);
 648
 649         filter = sk->sk_filter;
 650         if (filter) {
 651                 sk_filter_release(sk, filter);
 652                 sk->sk_filter = NULL;
 653         }
 654
 655         sock_disable_timestamp(sk);
 656
 657         if (atomic_read(&sk->sk_omem_alloc))
 658                 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
 659                        __FUNCTION__, atomic_read(&sk->sk_omem_alloc));
 660
 661         security_sk_free(sk);
 662         BUG_ON(sk->sk_vx_info);
 663         BUG_ON(sk->sk_nx_info);
 664         kmem_cache_free(sk->sk_slab, sk);
 665         module_put(owner);
 666 }
 667
 668 void __init sk_init(void)
 669 {
 670         sk_cachep = kmem_cache_create("sock", sizeof(struct sock), 0,
 671                                       SLAB_HWCACHE_ALIGN, NULL, NULL);
 672         if (!sk_cachep)
 673                 printk(KERN_CRIT "sk_init: Cannot create sock SLAB cache!");
 674
 675         if (num_physpages <= 4096) {
 676                 sysctl_wmem_max = 32767;
 677                 sysctl_rmem_max = 32767;
 678                 sysctl_wmem_default = 32767;
 679                 sysctl_rmem_default = 32767;
 680         } else if (num_physpages >= 131072) {
 681                 sysctl_wmem_max = 131071;
 682                 sysctl_rmem_max = 131071;
 683         }
 684 }
 685
 686 /*
 687  *      Simple resource managers for sockets.
 688  */
 689
 690
 691 /*
 692  * Write buffer destructor automatically called from kfree_skb.
 693  */
 694 void sock_wfree(struct sk_buff *skb)
 695 {
 696         struct sock *sk = skb->sk;
 697
 698         /* In case it might be waiting for more memory. */
 699         atomic_sub(skb->truesize, &sk->sk_wmem_alloc);
 700         if (!sk->sk_use_write_queue)
 701                 sk->sk_write_space(sk);
 702         sock_put(sk);
 703 }
 704
 705 /*
 706  * Read buffer destructor automatically called from kfree_skb.
 707  */
 708 void sock_rfree(struct sk_buff *skb)
 709 {
 710         struct sock *sk = skb->sk;
 711
 712         atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
 713 }
 714
 715
 716 int sock_i_uid(struct sock *sk)
 717 {
 718         int uid;
 719
 720         read_lock(&sk->sk_callback_lock);
 721         uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
 722         read_unlock(&sk->sk_callback_lock);
 723         return uid;
 724 }
 725
 726 unsigned long sock_i_ino(struct sock *sk)
 727 {
 728         unsigned long ino;
 729
 730         read_lock(&sk->sk_callback_lock);
 731         ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
 732         read_unlock(&sk->sk_callback_lock);
 733         return ino;
 734 }
 735
 736 /*
 737  * Allocate a skb from the socket's send buffer.
 738  */
 739 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, int priority)
 740 {
 741         if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
 742                 struct sk_buff * skb = alloc_skb(size, priority);
 743                 if (skb) {
 744                         skb_set_owner_w(skb, sk);
 745                         return skb;
 746                 }
 747         }
 748         return NULL;
 749 }
 750
 751 /*
 752  * Allocate a skb from the socket's receive buffer.
 753  */
 754 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force, int priority)
 755 {
 756         if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
 757                 struct sk_buff *skb = alloc_skb(size, priority);
 758                 if (skb) {
 759                         skb_set_owner_r(skb, sk);
 760                         return skb;
 761                 }
 762         }
 763         return NULL;
 764 }
 765
 766 /*
 767  * Allocate a memory block from the socket's option memory buffer.
 768  */
 769 void *sock_kmalloc(struct sock *sk, int size, int priority)
 770 {
 771         if ((unsigned)size <= sysctl_optmem_max &&
 772             atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
 773                 void *mem;
 774                 /* First do the add, to avoid the race if kmalloc
 775                  * might sleep.
 776                  */
 777                 atomic_add(size, &sk->sk_omem_alloc);
 778                 mem = kmalloc(size, priority);
 779                 if (mem)
 780                         return mem;
 781                 atomic_sub(size, &sk->sk_omem_alloc);
 782         }
 783         return NULL;
 784 }
 785
 786 /*
 787  * Free an option memory block.
 788  */
 789 void sock_kfree_s(struct sock *sk, void *mem, int size)
 790 {
 791         kfree(mem);
 792         atomic_sub(size, &sk->sk_omem_alloc);
 793 }
 794
 795 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
 796    I think, these locks should be removed for datagram sockets.
 797  */
 798 static long sock_wait_for_wmem(struct sock * sk, long timeo)
 799 {
 800         DEFINE_WAIT(wait);
 801
 802         clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
 803         for (;;) {
 804                 if (!timeo)
 805                         break;
 806                 if (signal_pending(current))
 807                         break;
 808                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
 809                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
 810                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
 811                         break;
 812                 if (sk->sk_shutdown & SEND_SHUTDOWN)
 813                         break;
 814                 if (sk->sk_err)
 815                         break;
 816                 timeo = schedule_timeout(timeo);
 817         }
 818         finish_wait(sk->sk_sleep, &wait);
 819         return timeo;
 820 }
 821
 822
 823 /*
 824  *      Generic send/receive buffer handlers
 825  */
 826
 827 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
 828                                      unsigned long data_len, int noblock, int *errcode)
 829 {
 830         struct sk_buff *skb;
 831         unsigned int gfp_mask;
 832         long timeo;
 833         int err;
 834
 835         gfp_mask = sk->sk_allocation;
 836         if (gfp_mask & __GFP_WAIT)
 837                 gfp_mask |= __GFP_REPEAT;
 838
 839         timeo = sock_sndtimeo(sk, noblock);
 840         while (1) {
 841                 err = sock_error(sk);
 842                 if (err != 0)
 843                         goto failure;
 844
 845                 err = -EPIPE;
 846                 if (sk->sk_shutdown & SEND_SHUTDOWN)
 847                         goto failure;
 848
 849                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
 850                         skb = alloc_skb(header_len, sk->sk_allocation);
 851                         if (skb) {
 852                                 int npages;
 853                                 int i;
 854
 855                                 /* No pages, we're done... */
 856                                 if (!data_len)
 857                                         break;
 858
 859                                 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
 860                                 skb->truesize += data_len;
 861                                 skb_shinfo(skb)->nr_frags = npages;
 862                                 for (i = 0; i < npages; i++) {
 863                                         struct page *page;
 864                                         skb_frag_t *frag;
 865
 866                                         page = alloc_pages(sk->sk_allocation, 0);
 867                                         if (!page) {
 868                                                 err = -ENOBUFS;
 869                                                 skb_shinfo(skb)->nr_frags = i;
 870                                                 kfree_skb(skb);
 871                                                 goto failure;
 872                                         }
 873
 874                                         frag = &skb_shinfo(skb)->frags[i];
 875                                         frag->page = page;
 876                                         frag->page_offset = 0;
 877                                         frag->size = (data_len >= PAGE_SIZE ?
 878                                                       PAGE_SIZE :
 879                                                       data_len);
 880                                         data_len -= PAGE_SIZE;
 881                                 }
 882
 883                                 /* Full success... */
 884                                 break;
 885                         }
 886                         err = -ENOBUFS;
 887                         goto failure;
 888                 }
 889                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
 890                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
 891                 err = -EAGAIN;
 892                 if (!timeo)
 893                         goto failure;
 894                 if (signal_pending(current))
 895                         goto interrupted;
 896                 timeo = sock_wait_for_wmem(sk, timeo);
 897         }
 898
 899         skb_set_owner_w(skb, sk);
 900         return skb;
 901
 902 interrupted:
 903         err = sock_intr_errno(timeo);
 904 failure:
 905         *errcode = err;
 906         return NULL;
 907 }
 908
 909 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
 910                                     int noblock, int *errcode)
 911 {
 912         return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
 913 }
 914
 915 void __lock_sock(struct sock *sk)
 916 {
 917         DEFINE_WAIT(wait);
 918
 919         for(;;) {
 920                 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
 921                                         TASK_UNINTERRUPTIBLE);
 922                 spin_unlock_bh(&sk->sk_lock.slock);
 923                 schedule();
 924                 spin_lock_bh(&sk->sk_lock.slock);
 925                 if(!sock_owned_by_user(sk))
 926                         break;
 927         }
 928         finish_wait(&sk->sk_lock.wq, &wait);
 929 }
 930
 931 void __release_sock(struct sock *sk)
 932 {
 933         struct sk_buff *skb = sk->sk_backlog.head;
 934
 935         do {
 936                 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
 937                 bh_unlock_sock(sk);
 938
 939                 do {
 940                         struct sk_buff *next = skb->next;
 941
 942                         skb->next = NULL;
 943                         sk->sk_backlog_rcv(sk, skb);
 944                         skb = next;
 945                 } while (skb != NULL);
 946
 947                 bh_lock_sock(sk);
 948         } while((skb = sk->sk_backlog.head) != NULL);
 949 }
 950
 951 /**
 952  * sk_wait_data - wait for data to arrive at sk_receive_queue
 953  * sk - sock to wait on
 954  * timeo - for how long
 955  *
 956  * Now socket state including sk->sk_err is changed only under lock,
 957  * hence we may omit checks after joining wait queue.
 958  * We check receive queue before schedule() only as optimization;
 959  * it is very likely that release_sock() added new data.
 960  */
 961 int sk_wait_data(struct sock *sk, long *timeo)
 962 {
 963         int rc;
 964         DEFINE_WAIT(wait);
 965
 966         prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
 967         set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
 968         rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
 969         clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
 970         finish_wait(sk->sk_sleep, &wait);
 971         return rc;
 972 }
 973
 974 EXPORT_SYMBOL(sk_wait_data);
 975
 976 /*
 977  * Set of default routines for initialising struct proto_ops when
 978  * the protocol does not support a particular function. In certain
 979  * cases where it makes no sense for a protocol to have a "do nothing"
 980  * function, some default processing is provided.
 981  */
 982
 983 int sock_no_release(struct socket *sock)
 984 {
 985         return 0;
 986 }
 987
 988 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
 989 {
 990         return -EOPNOTSUPP;
 991 }
 992
 993 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
 994                     int len, int flags)
 995 {
 996         return -EOPNOTSUPP;
 997 }
 998
 999 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1000 {
1001         return -EOPNOTSUPP;
1002 }
1003
1004 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1005 {
1006         return -EOPNOTSUPP;
1007 }
1008
1009 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1010                     int *len, int peer)
1011 {
1012         return -EOPNOTSUPP;
1013 }
1014
1015 unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt)
1016 {
1017         return 0;
1018 }
1019
1020 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1021 {
1022         return -EOPNOTSUPP;
1023 }
1024
1025 int sock_no_listen(struct socket *sock, int backlog)
1026 {
1027         return -EOPNOTSUPP;
1028 }
1029
1030 int sock_no_shutdown(struct socket *sock, int how)
1031 {
1032         return -EOPNOTSUPP;
1033 }
1034
1035 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1036                     char __user *optval, int optlen)
1037 {
1038         return -EOPNOTSUPP;
1039 }
1040
1041 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1042                     char __user *optval, int __user *optlen)
1043 {
1044         return -EOPNOTSUPP;
1045 }
1046
1047 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1048                     size_t len)
1049 {
1050         return -EOPNOTSUPP;
1051 }
1052
1053 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1054                     size_t len, int flags)
1055 {
1056         return -EOPNOTSUPP;
1057 }
1058
1059 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1060 {
1061         /* Mirror missing mmap method error code */
1062         return -ENODEV;
1063 }
1064
1065 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1066 {
1067         ssize_t res;
1068         struct msghdr msg = {.msg_flags = flags};
1069         struct kvec iov;
1070         char *kaddr = kmap(page);
1071         iov.iov_base = kaddr + offset;
1072         iov.iov_len = size;
1073         res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1074         kunmap(page);
1075         return res;
1076 }
1077
1078 /*
1079  *      Default Socket Callbacks
1080  */
1081
1082 void sock_def_wakeup(struct sock *sk)
1083 {
1084         read_lock(&sk->sk_callback_lock);
1085         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1086                 wake_up_interruptible_all(sk->sk_sleep);
1087         read_unlock(&sk->sk_callback_lock);
1088 }
1089
1090 void sock_def_error_report(struct sock *sk)
1091 {
1092         read_lock(&sk->sk_callback_lock);
1093         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1094                 wake_up_interruptible(sk->sk_sleep);
1095         sk_wake_async(sk,0,POLL_ERR);
1096         read_unlock(&sk->sk_callback_lock);
1097 }
1098
1099 void sock_def_readable(struct sock *sk, int len)
1100 {
1101         read_lock(&sk->sk_callback_lock);
1102         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1103                 wake_up_interruptible(sk->sk_sleep);
1104         sk_wake_async(sk,1,POLL_IN);
1105         read_unlock(&sk->sk_callback_lock);
1106 }
1107
1108 void sock_def_write_space(struct sock *sk)
1109 {
1110         read_lock(&sk->sk_callback_lock);
1111
1112         /* Do not wake up a writer until he can make "significant"
1113          * progress.  --DaveM
1114          */
1115         if((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1116                 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1117                         wake_up_interruptible(sk->sk_sleep);
1118
1119                 /* Should agree with poll, otherwise some programs break */
1120                 if (sock_writeable(sk))
1121                         sk_wake_async(sk, 2, POLL_OUT);
1122         }
1123
1124         read_unlock(&sk->sk_callback_lock);
1125 }
1126
1127 void sock_def_destruct(struct sock *sk)
1128 {
1129         if (sk->sk_protinfo)
1130                 kfree(sk->sk_protinfo);
1131 }
1132
1133 void sk_send_sigurg(struct sock *sk)
1134 {
1135         if (sk->sk_socket && sk->sk_socket->file)
1136                 if (send_sigurg(&sk->sk_socket->file->f_owner))
1137                         sk_wake_async(sk, 3, POLL_PRI);
1138 }
1139
1140 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1141                     unsigned long expires)
1142 {
1143         if (!mod_timer(timer, expires))
1144                 sock_hold(sk);
1145 }
1146
1147 EXPORT_SYMBOL(sk_reset_timer);
1148
1149 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1150 {
1151         if (timer_pending(timer) && del_timer(timer))
1152                 __sock_put(sk);
1153 }
1154
1155 EXPORT_SYMBOL(sk_stop_timer);
1156
1157 void sock_init_data(struct socket *sock, struct sock *sk)
1158 {
1159         skb_queue_head_init(&sk->sk_receive_queue);
1160         skb_queue_head_init(&sk->sk_write_queue);
1161         skb_queue_head_init(&sk->sk_error_queue);
1162
1163         sk->sk_send_head        =       NULL;
1164
1165         init_timer(&sk->sk_timer);
1166
1167         sk->sk_allocation       =       GFP_KERNEL;
1168         sk->sk_rcvbuf           =       sysctl_rmem_default;
1169         sk->sk_sndbuf           =       sysctl_wmem_default;
1170         sk->sk_state            =       TCP_CLOSE;
1171         sk->sk_zapped           =       1;
1172         sk->sk_socket           =       sock;
1173
1174         if(sock)
1175         {
1176                 sk->sk_type     =       sock->type;
1177                 sk->sk_sleep    =       &sock->wait;
1178                 sock->sk        =       sk;
1179         } else
1180                 sk->sk_sleep    =       NULL;
1181
1182         sk->sk_dst_lock         =       RW_LOCK_UNLOCKED;
1183         sk->sk_callback_lock    =       RW_LOCK_UNLOCKED;
1184
1185         sk->sk_state_change     =       sock_def_wakeup;
1186         sk->sk_data_ready       =       sock_def_readable;
1187         sk->sk_write_space      =       sock_def_write_space;
1188         sk->sk_error_report     =       sock_def_error_report;
1189         sk->sk_destruct         =       sock_def_destruct;
1190
1191         sk->sk_sndmsg_page      =       NULL;
1192         sk->sk_sndmsg_off       =       0;
1193
1194         sk->sk_peercred.pid     =       0;
1195         sk->sk_peercred.uid     =       -1;
1196         sk->sk_peercred.gid     =       -1;
1197         sk->sk_write_pending    =       0;
1198         sk->sk_rcvlowat         =       1;
1199         sk->sk_rcvtimeo         =       MAX_SCHEDULE_TIMEOUT;
1200         sk->sk_sndtimeo         =       MAX_SCHEDULE_TIMEOUT;
1201         sk->sk_owner            =       NULL;
1202
1203         sk->sk_stamp.tv_sec     = -1L;
1204         sk->sk_stamp.tv_usec    = -1L;
1205
1206         sk->sk_vx_info          =       NULL;
1207         sk->sk_xid              =       0;
1208         sk->sk_nx_info          =       NULL;
1209         sk->sk_nid              =       0;
1210
1211         atomic_set(&sk->sk_refcnt, 1);
1212 }
1213
1214 void fastcall lock_sock(struct sock *sk)
1215 {
1216         might_sleep();
1217         spin_lock_bh(&(sk->sk_lock.slock));
1218         if (sk->sk_lock.owner)
1219                 __lock_sock(sk);
1220         sk->sk_lock.owner = (void *)1;
1221         spin_unlock_bh(&(sk->sk_lock.slock));
1222 }
1223
1224 EXPORT_SYMBOL(lock_sock);
1225
1226 void fastcall release_sock(struct sock *sk)
1227 {
1228         spin_lock_bh(&(sk->sk_lock.slock));
1229         if (sk->sk_backlog.tail)
1230                 __release_sock(sk);
1231         sk->sk_lock.owner = NULL;
1232         if (waitqueue_active(&(sk->sk_lock.wq)))
1233                 wake_up(&(sk->sk_lock.wq));
1234         spin_unlock_bh(&(sk->sk_lock.slock));
1235 }
1236 EXPORT_SYMBOL(release_sock);
1237
1238 /* When > 0 there are consumers of rx skb time stamps */
1239 atomic_t netstamp_needed = ATOMIC_INIT(0);
1240
1241 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1242 {
1243         if (!sock_flag(sk, SOCK_TIMESTAMP))
1244                 sock_enable_timestamp(sk);
1245         if (sk->sk_stamp.tv_sec == -1)
1246                 return -ENOENT;
1247         if (sk->sk_stamp.tv_sec == 0)
1248                 do_gettimeofday(&sk->sk_stamp);
1249         return copy_to_user(userstamp, &sk->sk_stamp, sizeof(struct timeval)) ?
1250                 -EFAULT : 0;
1251 }
1252 EXPORT_SYMBOL(sock_get_timestamp);
1253
1254 void sock_enable_timestamp(struct sock *sk)
1255 {
1256         if (!sock_flag(sk, SOCK_TIMESTAMP)) {
1257                 sock_set_flag(sk, SOCK_TIMESTAMP);
1258                 atomic_inc(&netstamp_needed);
1259         }
1260 }
1261 EXPORT_SYMBOL(sock_enable_timestamp);
1262
1263 void sock_disable_timestamp(struct sock *sk)
1264 {
1265         if (sock_flag(sk, SOCK_TIMESTAMP)) {
1266                 sock_reset_flag(sk, SOCK_TIMESTAMP);
1267                 atomic_dec(&netstamp_needed);
1268         }
1269 }
1270 EXPORT_SYMBOL(sock_disable_timestamp);
1271
1272 /*
1273  *      Get a socket option on an socket.
1274  *
1275  *      FIX: POSIX 1003.1g is very ambiguous here. It states that
1276  *      asynchronous errors should be reported by getsockopt. We assume
1277  *      this means if you specify SO_ERROR (otherwise whats the point of it).
1278  */
1279 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1280                            char __user *optval, int __user *optlen)
1281 {
1282         struct sock *sk = sock->sk;
1283
1284         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1285 }
1286
1287 EXPORT_SYMBOL(sock_common_getsockopt);
1288
1289 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1290                         struct msghdr *msg, size_t size, int flags)
1291 {
1292         struct sock *sk = sock->sk;
1293         int addr_len = 0;
1294         int err;
1295
1296         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
1297                                    flags & ~MSG_DONTWAIT, &addr_len);
1298         if (err >= 0)
1299                 msg->msg_namelen = addr_len;
1300         return err;
1301 }
1302
1303 EXPORT_SYMBOL(sock_common_recvmsg);
1304
1305 /*
1306  *      Set socket options on an inet socket.
1307  */
1308 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1309                            char __user *optval, int optlen)
1310 {
1311         struct sock *sk = sock->sk;
1312
1313         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1314 }
1315
1316 EXPORT_SYMBOL(sock_common_setsockopt);
1317
1318 void sk_common_release(struct sock *sk)
1319 {
1320         if (sk->sk_prot->destroy)
1321                 sk->sk_prot->destroy(sk);
1322
1323         /*
1324          * Observation: when sock_common_release is called, processes have
1325          * no access to socket. But net still has.
1326          * Step one, detach it from networking:
1327          *
1328          * A. Remove from hash tables.
1329          */
1330
1331         sk->sk_prot->unhash(sk);
1332
1333         /*
1334          * In this point socket cannot receive new packets, but it is possible
1335          * that some packets are in flight because some CPU runs receiver and
1336          * did hash table lookup before we unhashed socket. They will achieve
1337          * receive queue and will be purged by socket destructor.
1338          *
1339          * Also we still have packets pending on receive queue and probably,
1340          * our own packets waiting in device queues. sock_destroy will drain
1341          * receive queue, but transmitted packets will delay socket destruction
1342          * until the last reference will be released.
1343          */
1344
1345         sock_orphan(sk);
1346
1347         xfrm_sk_free_policy(sk);
1348
1349 #ifdef INET_REFCNT_DEBUG
1350         if (atomic_read(&sk->sk_refcnt) != 1)
1351                 printk(KERN_DEBUG "Destruction of the socket %p delayed, c=%d\n",
1352                        sk, atomic_read(&sk->sk_refcnt));
1353 #endif
1354         sock_put(sk);
1355 }
1356
1357 EXPORT_SYMBOL(sk_common_release);
1358
1359 EXPORT_SYMBOL(__lock_sock);
1360 EXPORT_SYMBOL(__release_sock);
1361 EXPORT_SYMBOL(sk_alloc);
1362 EXPORT_SYMBOL(sk_free);
1363 EXPORT_SYMBOL(sk_send_sigurg);
1364 EXPORT_SYMBOL(sock_alloc_send_pskb);
1365 EXPORT_SYMBOL(sock_alloc_send_skb);
1366 EXPORT_SYMBOL(sock_getsockopt);
1367 EXPORT_SYMBOL(sock_init_data);
1368 EXPORT_SYMBOL(sock_kfree_s);
1369 EXPORT_SYMBOL(sock_kmalloc);
1370 EXPORT_SYMBOL(sock_no_accept);
1371 EXPORT_SYMBOL(sock_no_bind);
1372 EXPORT_SYMBOL(sock_no_connect);
1373 EXPORT_SYMBOL(sock_no_getname);
1374 EXPORT_SYMBOL(sock_no_getsockopt);
1375 EXPORT_SYMBOL(sock_no_ioctl);
1376 EXPORT_SYMBOL(sock_no_listen);
1377 EXPORT_SYMBOL(sock_no_mmap);
1378 EXPORT_SYMBOL(sock_no_poll);
1379 EXPORT_SYMBOL(sock_no_recvmsg);
1380 EXPORT_SYMBOL(sock_no_release);
1381 EXPORT_SYMBOL(sock_no_sendmsg);
1382 EXPORT_SYMBOL(sock_no_sendpage);
1383 EXPORT_SYMBOL(sock_no_setsockopt);
1384 EXPORT_SYMBOL(sock_no_shutdown);
1385 EXPORT_SYMBOL(sock_no_socketpair);
1386 EXPORT_SYMBOL(sock_rfree);
1387 EXPORT_SYMBOL(sock_rmalloc);
1388 EXPORT_SYMBOL(sock_setsockopt);
1389 EXPORT_SYMBOL(sock_wfree);
1390 EXPORT_SYMBOL(sock_wmalloc);
1391 EXPORT_SYMBOL(sock_i_uid);
1392 EXPORT_SYMBOL(sock_i_ino);
1393 #ifdef CONFIG_SYSCTL
1394 EXPORT_SYMBOL(sysctl_optmem_max);
1395 EXPORT_SYMBOL(sysctl_rmem_max);
1396 EXPORT_SYMBOL(sysctl_wmem_max);
1397 #endif