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_SETXID:
 335                         if (current->xid) {
 336                                 ret = -EPERM;
 337                                 break;
 338                         }
 339                         if (val < 0 || val > MAX_S_CONTEXT) {
 340                                 ret = -EINVAL;
 341                                 break;
 342                         }
 343                         sk->sk_xid = val;
 344                         break;
 345
 346                 case SO_TIMESTAMP:
 347                         sk->sk_rcvtstamp = valbool;
 348                         if (valbool)
 349                                 sock_enable_timestamp(sk);
 350                         break;
 351
 352                 case SO_RCVLOWAT:
 353                         if (val < 0)
 354                                 val = INT_MAX;
 355                         sk->sk_rcvlowat = val ? : 1;
 356                         break;
 357
 358                 case SO_RCVTIMEO:
 359                         ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
 360                         break;
 361
 362                 case SO_SNDTIMEO:
 363                         ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
 364                         break;
 365
 366 #ifdef CONFIG_NETDEVICES
 367                 case SO_BINDTODEVICE:
 368                 {
 369                         char devname[IFNAMSIZ];
 370
 371                         /* Sorry... */
 372                         if (!capable(CAP_NET_RAW)) {
 373                                 ret = -EPERM;
 374                                 break;
 375                         }
 376
 377                         /* Bind this socket to a particular device like "eth0",
 378                          * as specified in the passed interface name. If the
 379                          * name is "" or the option length is zero the socket
 380                          * is not bound.
 381                          */
 382
 383                         if (!valbool) {
 384                                 sk->sk_bound_dev_if = 0;
 385                         } else {
 386                                 if (optlen > IFNAMSIZ)
 387                                         optlen = IFNAMSIZ;
 388                                 if (copy_from_user(devname, optval, optlen)) {
 389                                         ret = -EFAULT;
 390                                         break;
 391                                 }
 392
 393                                 /* Remove any cached route for this socket. */
 394                                 sk_dst_reset(sk);
 395
 396                                 if (devname[0] == '\0') {
 397                                         sk->sk_bound_dev_if = 0;
 398                                 } else {
 399                                         struct net_device *dev = dev_get_by_name(devname);
 400                                         if (!dev) {
 401                                                 ret = -ENODEV;
 402                                                 break;
 403                                         }
 404                                         sk->sk_bound_dev_if = dev->ifindex;
 405                                         dev_put(dev);
 406                                 }
 407                         }
 408                         break;
 409                 }
 410 #endif
 411
 412
 413                 case SO_ATTACH_FILTER:
 414                         ret = -EINVAL;
 415                         if (optlen == sizeof(struct sock_fprog)) {
 416                                 struct sock_fprog fprog;
 417
 418                                 ret = -EFAULT;
 419                                 if (copy_from_user(&fprog, optval, sizeof(fprog)))
 420                                         break;
 421
 422                                 ret = sk_attach_filter(&fprog, sk);
 423                         }
 424                         break;
 425
 426                 case SO_DETACH_FILTER:
 427                         spin_lock_bh(&sk->sk_lock.slock);
 428                         filter = sk->sk_filter;
 429                         if (filter) {
 430                                 sk->sk_filter = NULL;
 431                                 spin_unlock_bh(&sk->sk_lock.slock);
 432                                 sk_filter_release(sk, filter);
 433                                 break;
 434                         }
 435                         spin_unlock_bh(&sk->sk_lock.slock);
 436                         ret = -ENONET;
 437                         break;
 438
 439                 /* We implement the SO_SNDLOWAT etc to
 440                    not be settable (1003.1g 5.3) */
 441                 default:
 442                         ret = -ENOPROTOOPT;
 443                         break;
 444         }
 445         release_sock(sk);
 446         return ret;
 447 }
 448
 449
 450 int sock_getsockopt(struct socket *sock, int level, int optname,
 451                     char __user *optval, int __user *optlen)
 452 {
 453         struct sock *sk = sock->sk;
 454
 455         union
 456         {
 457                 int val;
 458                 struct linger ling;
 459                 struct timeval tm;
 460         } v;
 461
 462         unsigned int lv = sizeof(int);
 463         int len;
 464
 465         if(get_user(len,optlen))
 466                 return -EFAULT;
 467         if(len < 0)
 468                 return -EINVAL;
 469
 470         switch(optname)
 471         {
 472                 case SO_DEBUG:
 473                         v.val = sk->sk_debug;
 474                         break;
 475
 476                 case SO_DONTROUTE:
 477                         v.val = sk->sk_localroute;
 478                         break;
 479
 480                 case SO_BROADCAST:
 481                         v.val = !!sock_flag(sk, SOCK_BROADCAST);
 482                         break;
 483
 484                 case SO_SNDBUF:
 485                         v.val = sk->sk_sndbuf;
 486                         break;
 487
 488                 case SO_RCVBUF:
 489                         v.val = sk->sk_rcvbuf;
 490                         break;
 491
 492                 case SO_REUSEADDR:
 493                         v.val = sk->sk_reuse;
 494                         break;
 495
 496                 case SO_KEEPALIVE:
 497                         v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
 498                         break;
 499
 500                 case SO_TYPE:
 501                         v.val = sk->sk_type;
 502                         break;
 503
 504                 case SO_ERROR:
 505                         v.val = -sock_error(sk);
 506                         if(v.val==0)
 507                                 v.val = xchg(&sk->sk_err_soft, 0);
 508                         break;
 509
 510                 case SO_OOBINLINE:
 511                         v.val = !!sock_flag(sk, SOCK_URGINLINE);
 512                         break;
 513
 514                 case SO_NO_CHECK:
 515                         v.val = sk->sk_no_check;
 516                         break;
 517
 518                 case SO_PRIORITY:
 519                         v.val = sk->sk_priority;
 520                         break;
 521
 522                 case SO_LINGER:
 523                         lv              = sizeof(v.ling);
 524                         v.ling.l_onoff  = !!sock_flag(sk, SOCK_LINGER);
 525                         v.ling.l_linger = sk->sk_lingertime / HZ;
 526                         break;
 527
 528                 case SO_BSDCOMPAT:
 529                         sock_warn_obsolete_bsdism("getsockopt");
 530                         break;
 531
 532                 case SO_TIMESTAMP:
 533                         v.val = sk->sk_rcvtstamp;
 534                         break;
 535
 536                 case SO_RCVTIMEO:
 537                         lv=sizeof(struct timeval);
 538                         if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
 539                                 v.tm.tv_sec = 0;
 540                                 v.tm.tv_usec = 0;
 541                         } else {
 542                                 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
 543                                 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
 544                         }
 545                         break;
 546
 547                 case SO_SNDTIMEO:
 548                         lv=sizeof(struct timeval);
 549                         if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
 550                                 v.tm.tv_sec = 0;
 551                                 v.tm.tv_usec = 0;
 552                         } else {
 553                                 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
 554                                 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
 555                         }
 556                         break;
 557
 558                 case SO_RCVLOWAT:
 559                         v.val = sk->sk_rcvlowat;
 560                         break;
 561
 562                 case SO_SNDLOWAT:
 563                         v.val=1;
 564                         break;
 565
 566                 case SO_PASSCRED:
 567                         v.val = test_bit(SOCK_PASS_CRED, &sock->flags)?1:0;
 568                         break;
 569
 570                 case SO_PEERCRED:
 571                         if (len > sizeof(sk->sk_peercred))
 572                                 len = sizeof(sk->sk_peercred);
 573                         if (copy_to_user(optval, &sk->sk_peercred, len))
 574                                 return -EFAULT;
 575                         goto lenout;
 576
 577                 case SO_PEERNAME:
 578                 {
 579                         char address[128];
 580
 581                         if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
 582                                 return -ENOTCONN;
 583                         if (lv < len)
 584                                 return -EINVAL;
 585                         if (copy_to_user(optval, address, len))
 586                                 return -EFAULT;
 587                         goto lenout;
 588                 }
 589
 590                 /* Dubious BSD thing... Probably nobody even uses it, but
 591                  * the UNIX standard wants it for whatever reason... -DaveM
 592                  */
 593                 case SO_ACCEPTCONN:
 594                         v.val = sk->sk_state == TCP_LISTEN;
 595                         break;
 596
 597                 case SO_PEERSEC:
 598                         return security_socket_getpeersec(sock, optval, optlen, len);
 599
 600                 default:
 601                         return(-ENOPROTOOPT);
 602         }
 603         if (len > lv)
 604                 len = lv;
 605         if (copy_to_user(optval, &v, len))
 606                 return -EFAULT;
 607 lenout:
 608         if (put_user(len, optlen))
 609                 return -EFAULT;
 610         return 0;
 611 }
 612
 613 static kmem_cache_t *sk_cachep;
 614
 615 /**
 616  *      sk_alloc - All socket objects are allocated here
 617  *      @family - protocol family
 618  *      @priority - for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
 619  *      @zero_it - zeroes the allocated sock
 620  *      @slab - alternate slab
 621  *
 622  *      All socket objects are allocated here. If @zero_it is non-zero
 623  *      it should have the size of the are to be zeroed, because the
 624  *      private slabcaches have different sizes of the generic struct sock.
 625  *      1 has been kept as a way to say sizeof(struct sock).
 626  */
 627 struct sock *sk_alloc(int family, int priority, int zero_it, kmem_cache_t *slab)
 628 {
 629         struct sock *sk = NULL;
 630
 631         if (!slab)
 632                 slab = sk_cachep;
 633         sk = kmem_cache_alloc(slab, priority);
 634         if (sk) {
 635                 if (zero_it) {
 636                         memset(sk, 0,
 637                                zero_it == 1 ? sizeof(struct sock) : zero_it);
 638                         sk->sk_family = family;
 639                         sock_lock_init(sk);
 640                 }
 641                 sk->sk_slab = slab;
 642                 sock_vx_init(sk);
 643                 sock_nx_init(sk);
 644
 645                 if (security_sk_alloc(sk, family, priority)) {
 646                         kmem_cache_free(slab, sk);
 647                         sk = NULL;
 648                 }
 649         }
 650         return sk;
 651 }
 652
 653 void sk_free(struct sock *sk)
 654 {
 655         struct sk_filter *filter;
 656         struct module *owner = sk->sk_owner;
 657
 658         if (sk->sk_destruct)
 659                 sk->sk_destruct(sk);
 660
 661         filter = sk->sk_filter;
 662         if (filter) {
 663                 sk_filter_release(sk, filter);
 664                 sk->sk_filter = NULL;
 665         }
 666
 667         sock_disable_timestamp(sk);
 668
 669         if (atomic_read(&sk->sk_omem_alloc))
 670                 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
 671                        __FUNCTION__, atomic_read(&sk->sk_omem_alloc));
 672
 673         security_sk_free(sk);
 674         BUG_ON(sk->sk_vx_info);
 675         BUG_ON(sk->sk_nx_info);
 676         kmem_cache_free(sk->sk_slab, sk);
 677         module_put(owner);
 678 }
 679
 680 void __init sk_init(void)
 681 {
 682         sk_cachep = kmem_cache_create("sock", sizeof(struct sock), 0,
 683                                       SLAB_HWCACHE_ALIGN, NULL, NULL);
 684         if (!sk_cachep)
 685                 printk(KERN_CRIT "sk_init: Cannot create sock SLAB cache!");
 686
 687         if (num_physpages <= 4096) {
 688                 sysctl_wmem_max = 32767;
 689                 sysctl_rmem_max = 32767;
 690                 sysctl_wmem_default = 32767;
 691                 sysctl_rmem_default = 32767;
 692         } else if (num_physpages >= 131072) {
 693                 sysctl_wmem_max = 131071;
 694                 sysctl_rmem_max = 131071;
 695         }
 696 }
 697
 698 /*
 699  *      Simple resource managers for sockets.
 700  */
 701
 702
 703 /*
 704  * Write buffer destructor automatically called from kfree_skb.
 705  */
 706 void sock_wfree(struct sk_buff *skb)
 707 {
 708         struct sock *sk = skb->sk;
 709
 710         /* In case it might be waiting for more memory. */
 711         atomic_sub(skb->truesize, &sk->sk_wmem_alloc);
 712         if (!sk->sk_use_write_queue)
 713                 sk->sk_write_space(sk);
 714         sock_put(sk);
 715 }
 716
 717 /*
 718  * Read buffer destructor automatically called from kfree_skb.
 719  */
 720 void sock_rfree(struct sk_buff *skb)
 721 {
 722         struct sock *sk = skb->sk;
 723
 724         atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
 725 }
 726
 727
 728 int sock_i_uid(struct sock *sk)
 729 {
 730         int uid;
 731
 732         read_lock(&sk->sk_callback_lock);
 733         uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
 734         read_unlock(&sk->sk_callback_lock);
 735         return uid;
 736 }
 737
 738 unsigned long sock_i_ino(struct sock *sk)
 739 {
 740         unsigned long ino;
 741
 742         read_lock(&sk->sk_callback_lock);
 743         ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
 744         read_unlock(&sk->sk_callback_lock);
 745         return ino;
 746 }
 747
 748 /*
 749  * Allocate a skb from the socket's send buffer.
 750  */
 751 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, int priority)
 752 {
 753         if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
 754                 struct sk_buff * skb = alloc_skb(size, priority);
 755                 if (skb) {
 756                         skb_set_owner_w(skb, sk);
 757                         return skb;
 758                 }
 759         }
 760         return NULL;
 761 }
 762
 763 /*
 764  * Allocate a skb from the socket's receive buffer.
 765  */
 766 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force, int priority)
 767 {
 768         if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
 769                 struct sk_buff *skb = alloc_skb(size, priority);
 770                 if (skb) {
 771                         skb_set_owner_r(skb, sk);
 772                         return skb;
 773                 }
 774         }
 775         return NULL;
 776 }
 777
 778 /*
 779  * Allocate a memory block from the socket's option memory buffer.
 780  */
 781 void *sock_kmalloc(struct sock *sk, int size, int priority)
 782 {
 783         if ((unsigned)size <= sysctl_optmem_max &&
 784             atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
 785                 void *mem;
 786                 /* First do the add, to avoid the race if kmalloc
 787                  * might sleep.
 788                  */
 789                 atomic_add(size, &sk->sk_omem_alloc);
 790                 mem = kmalloc(size, priority);
 791                 if (mem)
 792                         return mem;
 793                 atomic_sub(size, &sk->sk_omem_alloc);
 794         }
 795         return NULL;
 796 }
 797
 798 /*
 799  * Free an option memory block.
 800  */
 801 void sock_kfree_s(struct sock *sk, void *mem, int size)
 802 {
 803         kfree(mem);
 804         atomic_sub(size, &sk->sk_omem_alloc);
 805 }
 806
 807 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
 808    I think, these locks should be removed for datagram sockets.
 809  */
 810 static long sock_wait_for_wmem(struct sock * sk, long timeo)
 811 {
 812         DEFINE_WAIT(wait);
 813
 814         clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
 815         for (;;) {
 816                 if (!timeo)
 817                         break;
 818                 if (signal_pending(current))
 819                         break;
 820                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
 821                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
 822                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
 823                         break;
 824                 if (sk->sk_shutdown & SEND_SHUTDOWN)
 825                         break;
 826                 if (sk->sk_err)
 827                         break;
 828                 timeo = schedule_timeout(timeo);
 829         }
 830         finish_wait(sk->sk_sleep, &wait);
 831         return timeo;
 832 }
 833
 834
 835 /*
 836  *      Generic send/receive buffer handlers
 837  */
 838
 839 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
 840                                      unsigned long data_len, int noblock, int *errcode)
 841 {
 842         struct sk_buff *skb;
 843         unsigned int gfp_mask;
 844         long timeo;
 845         int err;
 846
 847         gfp_mask = sk->sk_allocation;
 848         if (gfp_mask & __GFP_WAIT)
 849                 gfp_mask |= __GFP_REPEAT;
 850
 851         timeo = sock_sndtimeo(sk, noblock);
 852         while (1) {
 853                 err = sock_error(sk);
 854                 if (err != 0)
 855                         goto failure;
 856
 857                 err = -EPIPE;
 858                 if (sk->sk_shutdown & SEND_SHUTDOWN)
 859                         goto failure;
 860
 861                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
 862                         skb = alloc_skb(header_len, sk->sk_allocation);
 863                         if (skb) {
 864                                 int npages;
 865                                 int i;
 866
 867                                 /* No pages, we're done... */
 868                                 if (!data_len)
 869                                         break;
 870
 871                                 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
 872                                 skb->truesize += data_len;
 873                                 skb_shinfo(skb)->nr_frags = npages;
 874                                 for (i = 0; i < npages; i++) {
 875                                         struct page *page;
 876                                         skb_frag_t *frag;
 877
 878                                         page = alloc_pages(sk->sk_allocation, 0);
 879                                         if (!page) {
 880                                                 err = -ENOBUFS;
 881                                                 skb_shinfo(skb)->nr_frags = i;
 882                                                 kfree_skb(skb);
 883                                                 goto failure;
 884                                         }
 885
 886                                         frag = &skb_shinfo(skb)->frags[i];
 887                                         frag->page = page;
 888                                         frag->page_offset = 0;
 889                                         frag->size = (data_len >= PAGE_SIZE ?
 890                                                       PAGE_SIZE :
 891                                                       data_len);
 892                                         data_len -= PAGE_SIZE;
 893                                 }
 894
 895                                 /* Full success... */
 896                                 break;
 897                         }
 898                         err = -ENOBUFS;
 899                         goto failure;
 900                 }
 901                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
 902                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
 903                 err = -EAGAIN;
 904                 if (!timeo)
 905                         goto failure;
 906                 if (signal_pending(current))
 907                         goto interrupted;
 908                 timeo = sock_wait_for_wmem(sk, timeo);
 909         }
 910
 911         skb_set_owner_w(skb, sk);
 912         return skb;
 913
 914 interrupted:
 915         err = sock_intr_errno(timeo);
 916 failure:
 917         *errcode = err;
 918         return NULL;
 919 }
 920
 921 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
 922                                     int noblock, int *errcode)
 923 {
 924         return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
 925 }
 926
 927 void __lock_sock(struct sock *sk)
 928 {
 929         DEFINE_WAIT(wait);
 930
 931         for(;;) {
 932                 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
 933                                         TASK_UNINTERRUPTIBLE);
 934                 spin_unlock_bh(&sk->sk_lock.slock);
 935                 schedule();
 936                 spin_lock_bh(&sk->sk_lock.slock);
 937                 if(!sock_owned_by_user(sk))
 938                         break;
 939         }
 940         finish_wait(&sk->sk_lock.wq, &wait);
 941 }
 942
 943 void __release_sock(struct sock *sk)
 944 {
 945         struct sk_buff *skb = sk->sk_backlog.head;
 946
 947         do {
 948                 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
 949                 bh_unlock_sock(sk);
 950
 951                 do {
 952                         struct sk_buff *next = skb->next;
 953
 954                         skb->next = NULL;
 955                         sk->sk_backlog_rcv(sk, skb);
 956                         skb = next;
 957                 } while (skb != NULL);
 958
 959                 bh_lock_sock(sk);
 960         } while((skb = sk->sk_backlog.head) != NULL);
 961 }
 962
 963 /**
 964  * sk_wait_data - wait for data to arrive at sk_receive_queue
 965  * sk - sock to wait on
 966  * timeo - for how long
 967  *
 968  * Now socket state including sk->sk_err is changed only under lock,
 969  * hence we may omit checks after joining wait queue.
 970  * We check receive queue before schedule() only as optimization;
 971  * it is very likely that release_sock() added new data.
 972  */
 973 int sk_wait_data(struct sock *sk, long *timeo)
 974 {
 975         int rc;
 976         DEFINE_WAIT(wait);
 977
 978         prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
 979         set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
 980         rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
 981         clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
 982         finish_wait(sk->sk_sleep, &wait);
 983         return rc;
 984 }
 985
 986 EXPORT_SYMBOL(sk_wait_data);
 987
 988 /*
 989  * Set of default routines for initialising struct proto_ops when
 990  * the protocol does not support a particular function. In certain
 991  * cases where it makes no sense for a protocol to have a "do nothing"
 992  * function, some default processing is provided.
 993  */
 994
 995 int sock_no_release(struct socket *sock)
 996 {
 997         return 0;
 998 }
 999
1000 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1001 {
1002         return -EOPNOTSUPP;
1003 }
1004
1005 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1006                     int len, int flags)
1007 {
1008         return -EOPNOTSUPP;
1009 }
1010
1011 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1012 {
1013         return -EOPNOTSUPP;
1014 }
1015
1016 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1017 {
1018         return -EOPNOTSUPP;
1019 }
1020
1021 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1022                     int *len, int peer)
1023 {
1024         return -EOPNOTSUPP;
1025 }
1026
1027 unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt)
1028 {
1029         return 0;
1030 }
1031
1032 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1033 {
1034         return -EOPNOTSUPP;
1035 }
1036
1037 int sock_no_listen(struct socket *sock, int backlog)
1038 {
1039         return -EOPNOTSUPP;
1040 }
1041
1042 int sock_no_shutdown(struct socket *sock, int how)
1043 {
1044         return -EOPNOTSUPP;
1045 }
1046
1047 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1048                     char __user *optval, int optlen)
1049 {
1050         return -EOPNOTSUPP;
1051 }
1052
1053 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1054                     char __user *optval, int __user *optlen)
1055 {
1056         return -EOPNOTSUPP;
1057 }
1058
1059 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1060                     size_t len)
1061 {
1062         return -EOPNOTSUPP;
1063 }
1064
1065 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1066                     size_t len, int flags)
1067 {
1068         return -EOPNOTSUPP;
1069 }
1070
1071 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1072 {
1073         /* Mirror missing mmap method error code */
1074         return -ENODEV;
1075 }
1076
1077 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1078 {
1079         ssize_t res;
1080         struct msghdr msg = {.msg_flags = flags};
1081         struct kvec iov;
1082         char *kaddr = kmap(page);
1083         iov.iov_base = kaddr + offset;
1084         iov.iov_len = size;
1085         res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1086         kunmap(page);
1087         return res;
1088 }
1089
1090 /*
1091  *      Default Socket Callbacks
1092  */
1093
1094 void sock_def_wakeup(struct sock *sk)
1095 {
1096         read_lock(&sk->sk_callback_lock);
1097         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1098                 wake_up_interruptible_all(sk->sk_sleep);
1099         read_unlock(&sk->sk_callback_lock);
1100 }
1101
1102 void sock_def_error_report(struct sock *sk)
1103 {
1104         read_lock(&sk->sk_callback_lock);
1105         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1106                 wake_up_interruptible(sk->sk_sleep);
1107         sk_wake_async(sk,0,POLL_ERR);
1108         read_unlock(&sk->sk_callback_lock);
1109 }
1110
1111 void sock_def_readable(struct sock *sk, int len)
1112 {
1113         read_lock(&sk->sk_callback_lock);
1114         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1115                 wake_up_interruptible(sk->sk_sleep);
1116         sk_wake_async(sk,1,POLL_IN);
1117         read_unlock(&sk->sk_callback_lock);
1118 }
1119
1120 void sock_def_write_space(struct sock *sk)
1121 {
1122         read_lock(&sk->sk_callback_lock);
1123
1124         /* Do not wake up a writer until he can make "significant"
1125          * progress.  --DaveM
1126          */
1127         if((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1128                 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1129                         wake_up_interruptible(sk->sk_sleep);
1130
1131                 /* Should agree with poll, otherwise some programs break */
1132                 if (sock_writeable(sk))
1133                         sk_wake_async(sk, 2, POLL_OUT);
1134         }
1135
1136         read_unlock(&sk->sk_callback_lock);
1137 }
1138
1139 void sock_def_destruct(struct sock *sk)
1140 {
1141         if (sk->sk_protinfo)
1142                 kfree(sk->sk_protinfo);
1143 }
1144
1145 void sk_send_sigurg(struct sock *sk)
1146 {
1147         if (sk->sk_socket && sk->sk_socket->file)
1148                 if (send_sigurg(&sk->sk_socket->file->f_owner))
1149                         sk_wake_async(sk, 3, POLL_PRI);
1150 }
1151
1152 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1153                     unsigned long expires)
1154 {
1155         if (!mod_timer(timer, expires))
1156                 sock_hold(sk);
1157 }
1158
1159 EXPORT_SYMBOL(sk_reset_timer);
1160
1161 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1162 {
1163         if (timer_pending(timer) && del_timer(timer))
1164                 __sock_put(sk);
1165 }
1166
1167 EXPORT_SYMBOL(sk_stop_timer);
1168
1169 void sock_init_data(struct socket *sock, struct sock *sk)
1170 {
1171         skb_queue_head_init(&sk->sk_receive_queue);
1172         skb_queue_head_init(&sk->sk_write_queue);
1173         skb_queue_head_init(&sk->sk_error_queue);
1174
1175         sk->sk_send_head        =       NULL;
1176
1177         init_timer(&sk->sk_timer);
1178
1179         sk->sk_allocation       =       GFP_KERNEL;
1180         sk->sk_rcvbuf           =       sysctl_rmem_default;
1181         sk->sk_sndbuf           =       sysctl_wmem_default;
1182         sk->sk_state            =       TCP_CLOSE;
1183         sk->sk_zapped           =       1;
1184         sk->sk_socket           =       sock;
1185
1186         if(sock)
1187         {
1188                 sk->sk_type     =       sock->type;
1189                 sk->sk_sleep    =       &sock->wait;
1190                 sock->sk        =       sk;
1191         } else
1192                 sk->sk_sleep    =       NULL;
1193
1194         sk->sk_dst_lock         =       RW_LOCK_UNLOCKED;
1195         sk->sk_callback_lock    =       RW_LOCK_UNLOCKED;
1196
1197         sk->sk_state_change     =       sock_def_wakeup;
1198         sk->sk_data_ready       =       sock_def_readable;
1199         sk->sk_write_space      =       sock_def_write_space;
1200         sk->sk_error_report     =       sock_def_error_report;
1201         sk->sk_destruct         =       sock_def_destruct;
1202
1203         sk->sk_sndmsg_page      =       NULL;
1204         sk->sk_sndmsg_off       =       0;
1205
1206         sk->sk_peercred.pid     =       0;
1207         sk->sk_peercred.uid     =       -1;
1208         sk->sk_peercred.gid     =       -1;
1209         sk->sk_write_pending    =       0;
1210         sk->sk_rcvlowat         =       1;
1211         sk->sk_rcvtimeo         =       MAX_SCHEDULE_TIMEOUT;
1212         sk->sk_sndtimeo         =       MAX_SCHEDULE_TIMEOUT;
1213         sk->sk_owner            =       NULL;
1214
1215         sk->sk_stamp.tv_sec     = -1L;
1216         sk->sk_stamp.tv_usec    = -1L;
1217
1218         sk->sk_vx_info          =       NULL;
1219         sk->sk_xid              =       0;
1220         sk->sk_nx_info          =       NULL;
1221         sk->sk_nid              =       0;
1222
1223         atomic_set(&sk->sk_refcnt, 1);
1224 }
1225
1226 void fastcall lock_sock(struct sock *sk)
1227 {
1228         might_sleep();
1229         spin_lock_bh(&(sk->sk_lock.slock));
1230         if (sk->sk_lock.owner)
1231                 __lock_sock(sk);
1232         sk->sk_lock.owner = (void *)1;
1233         spin_unlock_bh(&(sk->sk_lock.slock));
1234 }
1235
1236 EXPORT_SYMBOL(lock_sock);
1237
1238 void fastcall release_sock(struct sock *sk)
1239 {
1240         spin_lock_bh(&(sk->sk_lock.slock));
1241         if (sk->sk_backlog.tail)
1242                 __release_sock(sk);
1243         sk->sk_lock.owner = NULL;
1244         if (waitqueue_active(&(sk->sk_lock.wq)))
1245                 wake_up(&(sk->sk_lock.wq));
1246         spin_unlock_bh(&(sk->sk_lock.slock));
1247 }
1248 EXPORT_SYMBOL(release_sock);
1249
1250 /* When > 0 there are consumers of rx skb time stamps */
1251 atomic_t netstamp_needed = ATOMIC_INIT(0);
1252
1253 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1254 {
1255         if (!sock_flag(sk, SOCK_TIMESTAMP))
1256                 sock_enable_timestamp(sk);
1257         if (sk->sk_stamp.tv_sec == -1)
1258                 return -ENOENT;
1259         if (sk->sk_stamp.tv_sec == 0)
1260                 do_gettimeofday(&sk->sk_stamp);
1261         return copy_to_user(userstamp, &sk->sk_stamp, sizeof(struct timeval)) ?
1262                 -EFAULT : 0;
1263 }
1264 EXPORT_SYMBOL(sock_get_timestamp);
1265
1266 void sock_enable_timestamp(struct sock *sk)
1267 {
1268         if (!sock_flag(sk, SOCK_TIMESTAMP)) {
1269                 sock_set_flag(sk, SOCK_TIMESTAMP);
1270                 atomic_inc(&netstamp_needed);
1271         }
1272 }
1273 EXPORT_SYMBOL(sock_enable_timestamp);
1274
1275 void sock_disable_timestamp(struct sock *sk)
1276 {
1277         if (sock_flag(sk, SOCK_TIMESTAMP)) {
1278                 sock_reset_flag(sk, SOCK_TIMESTAMP);
1279                 atomic_dec(&netstamp_needed);
1280         }
1281 }
1282 EXPORT_SYMBOL(sock_disable_timestamp);
1283
1284 /*
1285  *      Get a socket option on an socket.
1286  *
1287  *      FIX: POSIX 1003.1g is very ambiguous here. It states that
1288  *      asynchronous errors should be reported by getsockopt. We assume
1289  *      this means if you specify SO_ERROR (otherwise whats the point of it).
1290  */
1291 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1292                            char __user *optval, int __user *optlen)
1293 {
1294         struct sock *sk = sock->sk;
1295
1296         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1297 }
1298
1299 EXPORT_SYMBOL(sock_common_getsockopt);
1300
1301 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1302                         struct msghdr *msg, size_t size, int flags)
1303 {
1304         struct sock *sk = sock->sk;
1305         int addr_len = 0;
1306         int err;
1307
1308         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
1309                                    flags & ~MSG_DONTWAIT, &addr_len);
1310         if (err >= 0)
1311                 msg->msg_namelen = addr_len;
1312         return err;
1313 }
1314
1315 EXPORT_SYMBOL(sock_common_recvmsg);
1316
1317 /*
1318  *      Set socket options on an inet socket.
1319  */
1320 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1321                            char __user *optval, int optlen)
1322 {
1323         struct sock *sk = sock->sk;
1324
1325         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1326 }
1327
1328 EXPORT_SYMBOL(sock_common_setsockopt);
1329
1330 void sk_common_release(struct sock *sk)
1331 {
1332         if (sk->sk_prot->destroy)
1333                 sk->sk_prot->destroy(sk);
1334
1335         /*
1336          * Observation: when sock_common_release is called, processes have
1337          * no access to socket. But net still has.
1338          * Step one, detach it from networking:
1339          *
1340          * A. Remove from hash tables.
1341          */
1342
1343         sk->sk_prot->unhash(sk);
1344
1345         /*
1346          * In this point socket cannot receive new packets, but it is possible
1347          * that some packets are in flight because some CPU runs receiver and
1348          * did hash table lookup before we unhashed socket. They will achieve
1349          * receive queue and will be purged by socket destructor.
1350          *
1351          * Also we still have packets pending on receive queue and probably,
1352          * our own packets waiting in device queues. sock_destroy will drain
1353          * receive queue, but transmitted packets will delay socket destruction
1354          * until the last reference will be released.
1355          */
1356
1357         sock_orphan(sk);
1358
1359         xfrm_sk_free_policy(sk);
1360
1361 #ifdef INET_REFCNT_DEBUG
1362         if (atomic_read(&sk->sk_refcnt) != 1)
1363                 printk(KERN_DEBUG "Destruction of the socket %p delayed, c=%d\n",
1364                        sk, atomic_read(&sk->sk_refcnt));
1365 #endif
1366         sock_put(sk);
1367 }
1368
1369 EXPORT_SYMBOL(sk_common_release);
1370
1371 EXPORT_SYMBOL(__lock_sock);
1372 EXPORT_SYMBOL(__release_sock);
1373 EXPORT_SYMBOL(sk_alloc);
1374 EXPORT_SYMBOL(sk_free);
1375 EXPORT_SYMBOL(sk_send_sigurg);
1376 EXPORT_SYMBOL(sock_alloc_send_pskb);
1377 EXPORT_SYMBOL(sock_alloc_send_skb);
1378 EXPORT_SYMBOL(sock_getsockopt);
1379 EXPORT_SYMBOL(sock_init_data);
1380 EXPORT_SYMBOL(sock_kfree_s);
1381 EXPORT_SYMBOL(sock_kmalloc);
1382 EXPORT_SYMBOL(sock_no_accept);
1383 EXPORT_SYMBOL(sock_no_bind);
1384 EXPORT_SYMBOL(sock_no_connect);
1385 EXPORT_SYMBOL(sock_no_getname);
1386 EXPORT_SYMBOL(sock_no_getsockopt);
1387 EXPORT_SYMBOL(sock_no_ioctl);
1388 EXPORT_SYMBOL(sock_no_listen);
1389 EXPORT_SYMBOL(sock_no_mmap);
1390 EXPORT_SYMBOL(sock_no_poll);
1391 EXPORT_SYMBOL(sock_no_recvmsg);
1392 EXPORT_SYMBOL(sock_no_release);
1393 EXPORT_SYMBOL(sock_no_sendmsg);
1394 EXPORT_SYMBOL(sock_no_sendpage);
1395 EXPORT_SYMBOL(sock_no_setsockopt);
1396 EXPORT_SYMBOL(sock_no_shutdown);
1397 EXPORT_SYMBOL(sock_no_socketpair);
1398 EXPORT_SYMBOL(sock_rfree);
1399 EXPORT_SYMBOL(sock_rmalloc);
1400 EXPORT_SYMBOL(sock_setsockopt);
1401 EXPORT_SYMBOL(sock_wfree);
1402 EXPORT_SYMBOL(sock_wmalloc);
1403 EXPORT_SYMBOL(sock_i_uid);
1404 EXPORT_SYMBOL(sock_i_ino);
1405 #ifdef CONFIG_SYSCTL
1406 EXPORT_SYMBOL(sysctl_optmem_max);
1407 EXPORT_SYMBOL(sysctl_rmem_max);
1408 EXPORT_SYMBOL(sysctl_wmem_max);
1409 #endif