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
 126 #ifdef CONFIG_INET
 127 #include <net/tcp.h>
 128 #endif
 129
 130 /* Take into consideration the size of the struct sk_buff overhead in the
 131  * determination of these values, since that is non-constant across
 132  * platforms.  This makes socket queueing behavior and performance
 133  * not depend upon such differences.
 134  */
 135 #define _SK_MEM_PACKETS         256
 136 #define _SK_MEM_OVERHEAD        (sizeof(struct sk_buff) + 256)
 137 #define SK_WMEM_MAX             (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
 138 #define SK_RMEM_MAX             (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
 139
 140 /* Run time adjustable parameters. */
 141 __u32 sysctl_wmem_max = SK_WMEM_MAX;
 142 __u32 sysctl_rmem_max = SK_RMEM_MAX;
 143 __u32 sysctl_wmem_default = SK_WMEM_MAX;
 144 __u32 sysctl_rmem_default = SK_RMEM_MAX;
 145
 146 /* Maximal space eaten by iovec or ancilliary data plus some space */
 147 int sysctl_optmem_max = sizeof(unsigned long)*(2*UIO_MAXIOV + 512);
 148
 149 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
 150 {
 151         struct timeval tv;
 152
 153         if (optlen < sizeof(tv))
 154                 return -EINVAL;
 155         if (copy_from_user(&tv, optval, sizeof(tv)))
 156                 return -EFAULT;
 157
 158         *timeo_p = MAX_SCHEDULE_TIMEOUT;
 159         if (tv.tv_sec == 0 && tv.tv_usec == 0)
 160                 return 0;
 161         if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
 162                 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
 163         return 0;
 164 }
 165
 166 static void sock_warn_obsolete_bsdism(const char *name)
 167 {
 168         static int warned;
 169         static char warncomm[16];
 170         if (strcmp(warncomm, current->comm) && warned < 5) {
 171                 strcpy(warncomm,  current->comm);
 172                 printk(KERN_WARNING "process `%s' is using obsolete "
 173                        "%s SO_BSDCOMPAT\n", warncomm, name);
 174                 warned++;
 175         }
 176 }
 177
 178 /*
 179  *      This is meant for all protocols to use and covers goings on
 180  *      at the socket level. Everything here is generic.
 181  */
 182
 183 int sock_setsockopt(struct socket *sock, int level, int optname,
 184                     char __user *optval, int optlen)
 185 {
 186         struct sock *sk=sock->sk;
 187         struct sk_filter *filter;
 188         int val;
 189         int valbool;
 190         struct linger ling;
 191         int ret = 0;
 192
 193         /*
 194          *      Options without arguments
 195          */
 196
 197 #ifdef SO_DONTLINGER            /* Compatibility item... */
 198         switch (optname) {
 199                 case SO_DONTLINGER:
 200                         sock_reset_flag(sk, SOCK_LINGER);
 201                         return 0;
 202         }
 203 #endif
 204
 205         if(optlen<sizeof(int))
 206                 return(-EINVAL);
 207
 208         if (get_user(val, (int __user *)optval))
 209                 return -EFAULT;
 210
 211         valbool = val?1:0;
 212
 213         lock_sock(sk);
 214
 215         switch(optname)
 216         {
 217                 case SO_DEBUG:
 218                         if(val && !capable(CAP_NET_ADMIN))
 219                         {
 220                                 ret = -EACCES;
 221                         }
 222                         else
 223                                 sk->sk_debug = valbool;
 224                         break;
 225                 case SO_REUSEADDR:
 226                         sk->sk_reuse = valbool;
 227                         break;
 228                 case SO_TYPE:
 229                 case SO_ERROR:
 230                         ret = -ENOPROTOOPT;
 231                         break;
 232                 case SO_DONTROUTE:
 233                         sk->sk_localroute = valbool;
 234                         break;
 235                 case SO_BROADCAST:
 236                         sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
 237                         break;
 238                 case SO_SNDBUF:
 239                         /* Don't error on this BSD doesn't and if you think
 240                            about it this is right. Otherwise apps have to
 241                            play 'guess the biggest size' games. RCVBUF/SNDBUF
 242                            are treated in BSD as hints */
 243
 244                         if (val > sysctl_wmem_max)
 245                                 val = sysctl_wmem_max;
 246
 247                         sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
 248                         if ((val * 2) < SOCK_MIN_SNDBUF)
 249                                 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
 250                         else
 251                                 sk->sk_sndbuf = val * 2;
 252
 253                         /*
 254                          *      Wake up sending tasks if we
 255                          *      upped the value.
 256                          */
 257                         sk->sk_write_space(sk);
 258                         break;
 259
 260                 case SO_RCVBUF:
 261                         /* Don't error on this BSD doesn't and if you think
 262                            about it this is right. Otherwise apps have to
 263                            play 'guess the biggest size' games. RCVBUF/SNDBUF
 264                            are treated in BSD as hints */
 265
 266                         if (val > sysctl_rmem_max)
 267                                 val = sysctl_rmem_max;
 268
 269                         sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
 270                         /* FIXME: is this lower bound the right one? */
 271                         if ((val * 2) < SOCK_MIN_RCVBUF)
 272                                 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
 273                         else
 274                                 sk->sk_rcvbuf = val * 2;
 275                         break;
 276
 277                 case SO_KEEPALIVE:
 278 #ifdef CONFIG_INET
 279                         if (sk->sk_protocol == IPPROTO_TCP)
 280                                 tcp_set_keepalive(sk, valbool);
 281 #endif
 282                         sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
 283                         break;
 284
 285                 case SO_OOBINLINE:
 286                         sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
 287                         break;
 288
 289                 case SO_NO_CHECK:
 290                         sk->sk_no_check = valbool;
 291                         break;
 292
 293                 case SO_PRIORITY:
 294                         if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
 295                                 sk->sk_priority = val;
 296                         else
 297                                 ret = -EPERM;
 298                         break;
 299
 300                 case SO_LINGER:
 301                         if(optlen<sizeof(ling)) {
 302                                 ret = -EINVAL;  /* 1003.1g */
 303                                 break;
 304                         }
 305                         if (copy_from_user(&ling,optval,sizeof(ling))) {
 306                                 ret = -EFAULT;
 307                                 break;
 308                         }
 309                         if (!ling.l_onoff)
 310                                 sock_reset_flag(sk, SOCK_LINGER);
 311                         else {
 312 #if (BITS_PER_LONG == 32)
 313                                 if (ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
 314                                         sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
 315                                 else
 316 #endif
 317                                         sk->sk_lingertime = ling.l_linger * HZ;
 318                                 sock_set_flag(sk, SOCK_LINGER);
 319                         }
 320                         break;
 321
 322                 case SO_BSDCOMPAT:
 323                         sock_warn_obsolete_bsdism("setsockopt");
 324                         break;
 325
 326                 case SO_PASSCRED:
 327                         sock->passcred = valbool;
 328                         break;
 329
 330                 case SO_TIMESTAMP:
 331                         sk->sk_rcvtstamp = valbool;
 332                         if (valbool)
 333                                 sock_enable_timestamp(sk);
 334                         break;
 335
 336                 case SO_RCVLOWAT:
 337                         if (val < 0)
 338                                 val = INT_MAX;
 339                         sk->sk_rcvlowat = val ? : 1;
 340                         break;
 341
 342                 case SO_RCVTIMEO:
 343                         ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
 344                         break;
 345
 346                 case SO_SNDTIMEO:
 347                         ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
 348                         break;
 349
 350 #ifdef CONFIG_NETDEVICES
 351                 case SO_BINDTODEVICE:
 352                 {
 353                         char devname[IFNAMSIZ];
 354
 355                         /* Sorry... */
 356                         if (!capable(CAP_NET_RAW)) {
 357                                 ret = -EPERM;
 358                                 break;
 359                         }
 360
 361                         /* Bind this socket to a particular device like "eth0",
 362                          * as specified in the passed interface name. If the
 363                          * name is "" or the option length is zero the socket
 364                          * is not bound.
 365                          */
 366
 367                         if (!valbool) {
 368                                 sk->sk_bound_dev_if = 0;
 369                         } else {
 370                                 if (optlen > IFNAMSIZ)
 371                                         optlen = IFNAMSIZ;
 372                                 if (copy_from_user(devname, optval, optlen)) {
 373                                         ret = -EFAULT;
 374                                         break;
 375                                 }
 376
 377                                 /* Remove any cached route for this socket. */
 378                                 sk_dst_reset(sk);
 379
 380                                 if (devname[0] == '\0') {
 381                                         sk->sk_bound_dev_if = 0;
 382                                 } else {
 383                                         struct net_device *dev = dev_get_by_name(devname);
 384                                         if (!dev) {
 385                                                 ret = -ENODEV;
 386                                                 break;
 387                                         }
 388                                         sk->sk_bound_dev_if = dev->ifindex;
 389                                         dev_put(dev);
 390                                 }
 391                         }
 392                         break;
 393                 }
 394 #endif
 395
 396
 397                 case SO_ATTACH_FILTER:
 398                         ret = -EINVAL;
 399                         if (optlen == sizeof(struct sock_fprog)) {
 400                                 struct sock_fprog fprog;
 401
 402                                 ret = -EFAULT;
 403                                 if (copy_from_user(&fprog, optval, sizeof(fprog)))
 404                                         break;
 405
 406                                 ret = sk_attach_filter(&fprog, sk);
 407                         }
 408                         break;
 409
 410                 case SO_DETACH_FILTER:
 411                         spin_lock_bh(&sk->sk_lock.slock);
 412                         filter = sk->sk_filter;
 413                         if (filter) {
 414                                 sk->sk_filter = NULL;
 415                                 spin_unlock_bh(&sk->sk_lock.slock);
 416                                 sk_filter_release(sk, filter);
 417                                 break;
 418                         }
 419                         spin_unlock_bh(&sk->sk_lock.slock);
 420                         ret = -ENONET;
 421                         break;
 422
 423                 /* We implement the SO_SNDLOWAT etc to
 424                    not be settable (1003.1g 5.3) */
 425                 default:
 426                         ret = -ENOPROTOOPT;
 427                         break;
 428         }
 429         release_sock(sk);
 430         return ret;
 431 }
 432
 433
 434 int sock_getsockopt(struct socket *sock, int level, int optname,
 435                     char __user *optval, int __user *optlen)
 436 {
 437         struct sock *sk = sock->sk;
 438
 439         union
 440         {
 441                 int val;
 442                 struct linger ling;
 443                 struct timeval tm;
 444         } v;
 445
 446         unsigned int lv = sizeof(int);
 447         int len;
 448
 449         if(get_user(len,optlen))
 450                 return -EFAULT;
 451         if(len < 0)
 452                 return -EINVAL;
 453
 454         switch(optname)
 455         {
 456                 case SO_DEBUG:
 457                         v.val = sk->sk_debug;
 458                         break;
 459
 460                 case SO_DONTROUTE:
 461                         v.val = sk->sk_localroute;
 462                         break;
 463
 464                 case SO_BROADCAST:
 465                         v.val = !!sock_flag(sk, SOCK_BROADCAST);
 466                         break;
 467
 468                 case SO_SNDBUF:
 469                         v.val = sk->sk_sndbuf;
 470                         break;
 471
 472                 case SO_RCVBUF:
 473                         v.val = sk->sk_rcvbuf;
 474                         break;
 475
 476                 case SO_REUSEADDR:
 477                         v.val = sk->sk_reuse;
 478                         break;
 479
 480                 case SO_KEEPALIVE:
 481                         v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
 482                         break;
 483
 484                 case SO_TYPE:
 485                         v.val = sk->sk_type;
 486                         break;
 487
 488                 case SO_ERROR:
 489                         v.val = -sock_error(sk);
 490                         if(v.val==0)
 491                                 v.val = xchg(&sk->sk_err_soft, 0);
 492                         break;
 493
 494                 case SO_OOBINLINE:
 495                         v.val = !!sock_flag(sk, SOCK_URGINLINE);
 496                         break;
 497
 498                 case SO_NO_CHECK:
 499                         v.val = sk->sk_no_check;
 500                         break;
 501
 502                 case SO_PRIORITY:
 503                         v.val = sk->sk_priority;
 504                         break;
 505
 506                 case SO_LINGER:
 507                         lv              = sizeof(v.ling);
 508                         v.ling.l_onoff  = !!sock_flag(sk, SOCK_LINGER);
 509                         v.ling.l_linger = sk->sk_lingertime / HZ;
 510                         break;
 511
 512                 case SO_BSDCOMPAT:
 513                         sock_warn_obsolete_bsdism("getsockopt");
 514                         break;
 515
 516                 case SO_TIMESTAMP:
 517                         v.val = sk->sk_rcvtstamp;
 518                         break;
 519
 520                 case SO_RCVTIMEO:
 521                         lv=sizeof(struct timeval);
 522                         if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
 523                                 v.tm.tv_sec = 0;
 524                                 v.tm.tv_usec = 0;
 525                         } else {
 526                                 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
 527                                 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
 528                         }
 529                         break;
 530
 531                 case SO_SNDTIMEO:
 532                         lv=sizeof(struct timeval);
 533                         if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
 534                                 v.tm.tv_sec = 0;
 535                                 v.tm.tv_usec = 0;
 536                         } else {
 537                                 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
 538                                 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
 539                         }
 540                         break;
 541
 542                 case SO_RCVLOWAT:
 543                         v.val = sk->sk_rcvlowat;
 544                         break;
 545
 546                 case SO_SNDLOWAT:
 547                         v.val=1;
 548                         break;
 549
 550                 case SO_PASSCRED:
 551                         v.val = sock->passcred;
 552                         break;
 553
 554                 case SO_PEERCRED:
 555                         if (len > sizeof(sk->sk_peercred))
 556                                 len = sizeof(sk->sk_peercred);
 557                         if (copy_to_user(optval, &sk->sk_peercred, len))
 558                                 return -EFAULT;
 559                         goto lenout;
 560
 561                 case SO_PEERNAME:
 562                 {
 563                         char address[128];
 564
 565                         if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
 566                                 return -ENOTCONN;
 567                         if (lv < len)
 568                                 return -EINVAL;
 569                         if (copy_to_user(optval, address, len))
 570                                 return -EFAULT;
 571                         goto lenout;
 572                 }
 573
 574                 /* Dubious BSD thing... Probably nobody even uses it, but
 575                  * the UNIX standard wants it for whatever reason... -DaveM
 576                  */
 577                 case SO_ACCEPTCONN:
 578                         v.val = sk->sk_state == TCP_LISTEN;
 579                         break;
 580
 581                 case SO_PEERSEC:
 582                         return security_socket_getpeersec(sock, optval, optlen, len);
 583
 584                 default:
 585                         return(-ENOPROTOOPT);
 586         }
 587         if (len > lv)
 588                 len = lv;
 589         if (copy_to_user(optval, &v, len))
 590                 return -EFAULT;
 591 lenout:
 592         if (put_user(len, optlen))
 593                 return -EFAULT;
 594         return 0;
 595 }
 596
 597 static kmem_cache_t *sk_cachep;
 598
 599 /**
 600  *      sk_alloc - All socket objects are allocated here
 601  *      @family - protocol family
 602  *      @priority - for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
 603  *      @zero_it - zeroes the allocated sock
 604  *      @slab - alternate slab
 605  *
 606  *      All socket objects are allocated here. If @zero_it is non-zero
 607  *      it should have the size of the are to be zeroed, because the
 608  *      private slabcaches have different sizes of the generic struct sock.
 609  *      1 has been kept as a way to say sizeof(struct sock).
 610  */
 611 struct sock *sk_alloc(int family, int priority, int zero_it, kmem_cache_t *slab)
 612 {
 613         struct sock *sk = NULL;
 614
 615         if (!slab)
 616                 slab = sk_cachep;
 617         sk = kmem_cache_alloc(slab, priority);
 618         if (sk) {
 619                 if (zero_it) {
 620                         memset(sk, 0,
 621                                zero_it == 1 ? sizeof(struct sock) : zero_it);
 622                         sk->sk_family = family;
 623                         sock_lock_init(sk);
 624                 }
 625                 sk->sk_slab = slab;
 626
 627                 if (security_sk_alloc(sk, family, priority)) {
 628                         kmem_cache_free(slab, sk);
 629                         sk = NULL;
 630                 }
 631         }
 632         return sk;
 633 }
 634
 635 void sk_free(struct sock *sk)
 636 {
 637         struct sk_filter *filter;
 638         struct module *owner = sk->sk_owner;
 639
 640         if (sk->sk_destruct)
 641                 sk->sk_destruct(sk);
 642
 643         filter = sk->sk_filter;
 644         if (filter) {
 645                 sk_filter_release(sk, filter);
 646                 sk->sk_filter = NULL;
 647         }
 648
 649         sock_disable_timestamp(sk);
 650
 651         if (atomic_read(&sk->sk_omem_alloc))
 652                 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
 653                        __FUNCTION__, atomic_read(&sk->sk_omem_alloc));
 654
 655         security_sk_free(sk);
 656         kmem_cache_free(sk->sk_slab, sk);
 657         module_put(owner);
 658 }
 659
 660 void __init sk_init(void)
 661 {
 662         sk_cachep = kmem_cache_create("sock", sizeof(struct sock), 0,
 663                                       SLAB_HWCACHE_ALIGN, NULL, NULL);
 664         if (!sk_cachep)
 665                 printk(KERN_CRIT "sk_init: Cannot create sock SLAB cache!");
 666
 667         if (num_physpages <= 4096) {
 668                 sysctl_wmem_max = 32767;
 669                 sysctl_rmem_max = 32767;
 670                 sysctl_wmem_default = 32767;
 671                 sysctl_rmem_default = 32767;
 672         } else if (num_physpages >= 131072) {
 673                 sysctl_wmem_max = 131071;
 674                 sysctl_rmem_max = 131071;
 675         }
 676 }
 677
 678 /*
 679  *      Simple resource managers for sockets.
 680  */
 681
 682
 683 /*
 684  * Write buffer destructor automatically called from kfree_skb.
 685  */
 686 void sock_wfree(struct sk_buff *skb)
 687 {
 688         struct sock *sk = skb->sk;
 689
 690         /* In case it might be waiting for more memory. */
 691         atomic_sub(skb->truesize, &sk->sk_wmem_alloc);
 692         if (!sk->sk_use_write_queue)
 693                 sk->sk_write_space(sk);
 694         sock_put(sk);
 695 }
 696
 697 /*
 698  * Read buffer destructor automatically called from kfree_skb.
 699  */
 700 void sock_rfree(struct sk_buff *skb)
 701 {
 702         struct sock *sk = skb->sk;
 703
 704         atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
 705 }
 706
 707
 708 int sock_i_uid(struct sock *sk)
 709 {
 710         int uid;
 711
 712         read_lock(&sk->sk_callback_lock);
 713         uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
 714         read_unlock(&sk->sk_callback_lock);
 715         return uid;
 716 }
 717
 718 unsigned long sock_i_ino(struct sock *sk)
 719 {
 720         unsigned long ino;
 721
 722         read_lock(&sk->sk_callback_lock);
 723         ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
 724         read_unlock(&sk->sk_callback_lock);
 725         return ino;
 726 }
 727
 728 /*
 729  * Allocate a skb from the socket's send buffer.
 730  */
 731 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, int priority)
 732 {
 733         if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
 734                 struct sk_buff * skb = alloc_skb(size, priority);
 735                 if (skb) {
 736                         skb_set_owner_w(skb, sk);
 737                         return skb;
 738                 }
 739         }
 740         return NULL;
 741 }
 742
 743 /*
 744  * Allocate a skb from the socket's receive buffer.
 745  */
 746 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force, int priority)
 747 {
 748         if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
 749                 struct sk_buff *skb = alloc_skb(size, priority);
 750                 if (skb) {
 751                         skb_set_owner_r(skb, sk);
 752                         return skb;
 753                 }
 754         }
 755         return NULL;
 756 }
 757
 758 /*
 759  * Allocate a memory block from the socket's option memory buffer.
 760  */
 761 void *sock_kmalloc(struct sock *sk, int size, int priority)
 762 {
 763         if ((unsigned)size <= sysctl_optmem_max &&
 764             atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
 765                 void *mem;
 766                 /* First do the add, to avoid the race if kmalloc
 767                  * might sleep.
 768                  */
 769                 atomic_add(size, &sk->sk_omem_alloc);
 770                 mem = kmalloc(size, priority);
 771                 if (mem)
 772                         return mem;
 773                 atomic_sub(size, &sk->sk_omem_alloc);
 774         }
 775         return NULL;
 776 }
 777
 778 /*
 779  * Free an option memory block.
 780  */
 781 void sock_kfree_s(struct sock *sk, void *mem, int size)
 782 {
 783         kfree(mem);
 784         atomic_sub(size, &sk->sk_omem_alloc);
 785 }
 786
 787 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
 788    I think, these locks should be removed for datagram sockets.
 789  */
 790 static long sock_wait_for_wmem(struct sock * sk, long timeo)
 791 {
 792         DEFINE_WAIT(wait);
 793
 794         clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
 795         for (;;) {
 796                 if (!timeo)
 797                         break;
 798                 if (signal_pending(current))
 799                         break;
 800                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
 801                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
 802                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
 803                         break;
 804                 if (sk->sk_shutdown & SEND_SHUTDOWN)
 805                         break;
 806                 if (sk->sk_err)
 807                         break;
 808                 timeo = schedule_timeout(timeo);
 809         }
 810         finish_wait(sk->sk_sleep, &wait);
 811         return timeo;
 812 }
 813
 814
 815 /*
 816  *      Generic send/receive buffer handlers
 817  */
 818
 819 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
 820                                      unsigned long data_len, int noblock, int *errcode)
 821 {
 822         struct sk_buff *skb;
 823         unsigned int gfp_mask;
 824         long timeo;
 825         int err;
 826
 827         gfp_mask = sk->sk_allocation;
 828         if (gfp_mask & __GFP_WAIT)
 829                 gfp_mask |= __GFP_REPEAT;
 830
 831         timeo = sock_sndtimeo(sk, noblock);
 832         while (1) {
 833                 err = sock_error(sk);
 834                 if (err != 0)
 835                         goto failure;
 836
 837                 err = -EPIPE;
 838                 if (sk->sk_shutdown & SEND_SHUTDOWN)
 839                         goto failure;
 840
 841                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
 842                         skb = alloc_skb(header_len, sk->sk_allocation);
 843                         if (skb) {
 844                                 int npages;
 845                                 int i;
 846
 847                                 /* No pages, we're done... */
 848                                 if (!data_len)
 849                                         break;
 850
 851                                 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
 852                                 skb->truesize += data_len;
 853                                 skb_shinfo(skb)->nr_frags = npages;
 854                                 for (i = 0; i < npages; i++) {
 855                                         struct page *page;
 856                                         skb_frag_t *frag;
 857
 858                                         page = alloc_pages(sk->sk_allocation, 0);
 859                                         if (!page) {
 860                                                 err = -ENOBUFS;
 861                                                 skb_shinfo(skb)->nr_frags = i;
 862                                                 kfree_skb(skb);
 863                                                 goto failure;
 864                                         }
 865
 866                                         frag = &skb_shinfo(skb)->frags[i];
 867                                         frag->page = page;
 868                                         frag->page_offset = 0;
 869                                         frag->size = (data_len >= PAGE_SIZE ?
 870                                                       PAGE_SIZE :
 871                                                       data_len);
 872                                         data_len -= PAGE_SIZE;
 873                                 }
 874
 875                                 /* Full success... */
 876                                 break;
 877                         }
 878                         err = -ENOBUFS;
 879                         goto failure;
 880                 }
 881                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
 882                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
 883                 err = -EAGAIN;
 884                 if (!timeo)
 885                         goto failure;
 886                 if (signal_pending(current))
 887                         goto interrupted;
 888                 timeo = sock_wait_for_wmem(sk, timeo);
 889         }
 890
 891         skb_set_owner_w(skb, sk);
 892         return skb;
 893
 894 interrupted:
 895         err = sock_intr_errno(timeo);
 896 failure:
 897         *errcode = err;
 898         return NULL;
 899 }
 900
 901 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
 902                                     int noblock, int *errcode)
 903 {
 904         return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
 905 }
 906
 907 void __lock_sock(struct sock *sk)
 908 {
 909         DEFINE_WAIT(wait);
 910
 911         for(;;) {
 912                 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
 913                                         TASK_UNINTERRUPTIBLE);
 914                 spin_unlock_bh(&sk->sk_lock.slock);
 915                 schedule();
 916                 spin_lock_bh(&sk->sk_lock.slock);
 917                 if(!sock_owned_by_user(sk))
 918                         break;
 919         }
 920         finish_wait(&sk->sk_lock.wq, &wait);
 921 }
 922
 923 void __release_sock(struct sock *sk)
 924 {
 925         struct sk_buff *skb = sk->sk_backlog.head;
 926
 927         do {
 928                 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
 929                 bh_unlock_sock(sk);
 930
 931                 do {
 932                         struct sk_buff *next = skb->next;
 933
 934                         skb->next = NULL;
 935                         sk->sk_backlog_rcv(sk, skb);
 936                         skb = next;
 937                 } while (skb != NULL);
 938
 939                 bh_lock_sock(sk);
 940         } while((skb = sk->sk_backlog.head) != NULL);
 941 }
 942
 943 /**
 944  * sk_wait_data - wait for data to arrive at sk_receive_queue
 945  * sk - sock to wait on
 946  * timeo - for how long
 947  *
 948  * Now socket state including sk->sk_err is changed only under lock,
 949  * hence we may omit checks after joining wait queue.
 950  * We check receive queue before schedule() only as optimization;
 951  * it is very likely that release_sock() added new data.
 952  */
 953 int sk_wait_data(struct sock *sk, long *timeo)
 954 {
 955         int rc;
 956         DEFINE_WAIT(wait);
 957
 958         prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
 959         set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
 960         rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
 961         clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
 962         finish_wait(sk->sk_sleep, &wait);
 963         return rc;
 964 }
 965
 966 EXPORT_SYMBOL(sk_wait_data);
 967
 968 /*
 969  * Set of default routines for initialising struct proto_ops when
 970  * the protocol does not support a particular function. In certain
 971  * cases where it makes no sense for a protocol to have a "do nothing"
 972  * function, some default processing is provided.
 973  */
 974
 975 int sock_no_release(struct socket *sock)
 976 {
 977         return 0;
 978 }
 979
 980 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
 981 {
 982         return -EOPNOTSUPP;
 983 }
 984
 985 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
 986                     int len, int flags)
 987 {
 988         return -EOPNOTSUPP;
 989 }
 990
 991 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
 992 {
 993         return -EOPNOTSUPP;
 994 }
 995
 996 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
 997 {
 998         return -EOPNOTSUPP;
 999 }
1000
1001 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1002                     int *len, int peer)
1003 {
1004         return -EOPNOTSUPP;
1005 }
1006
1007 unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt)
1008 {
1009         return 0;
1010 }
1011
1012 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1013 {
1014         return -EOPNOTSUPP;
1015 }
1016
1017 int sock_no_listen(struct socket *sock, int backlog)
1018 {
1019         return -EOPNOTSUPP;
1020 }
1021
1022 int sock_no_shutdown(struct socket *sock, int how)
1023 {
1024         return -EOPNOTSUPP;
1025 }
1026
1027 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1028                     char __user *optval, int optlen)
1029 {
1030         return -EOPNOTSUPP;
1031 }
1032
1033 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1034                     char __user *optval, int __user *optlen)
1035 {
1036         return -EOPNOTSUPP;
1037 }
1038
1039 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1040                     size_t len)
1041 {
1042         return -EOPNOTSUPP;
1043 }
1044
1045 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1046                     size_t len, int flags)
1047 {
1048         return -EOPNOTSUPP;
1049 }
1050
1051 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1052 {
1053         /* Mirror missing mmap method error code */
1054         return -ENODEV;
1055 }
1056
1057 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1058 {
1059         ssize_t res;
1060         struct msghdr msg = {.msg_flags = flags};
1061         struct kvec iov;
1062         char *kaddr = kmap(page);
1063         iov.iov_base = kaddr + offset;
1064         iov.iov_len = size;
1065         res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1066         kunmap(page);
1067         return res;
1068 }
1069
1070 /*
1071  *      Default Socket Callbacks
1072  */
1073
1074 void sock_def_wakeup(struct sock *sk)
1075 {
1076         read_lock(&sk->sk_callback_lock);
1077         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1078                 wake_up_interruptible_all(sk->sk_sleep);
1079         read_unlock(&sk->sk_callback_lock);
1080 }
1081
1082 void sock_def_error_report(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(sk->sk_sleep);
1087         sk_wake_async(sk,0,POLL_ERR);
1088         read_unlock(&sk->sk_callback_lock);
1089 }
1090
1091 void sock_def_readable(struct sock *sk, int len)
1092 {
1093         read_lock(&sk->sk_callback_lock);
1094         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1095                 wake_up_interruptible(sk->sk_sleep);
1096         sk_wake_async(sk,1,POLL_IN);
1097         read_unlock(&sk->sk_callback_lock);
1098 }
1099
1100 void sock_def_write_space(struct sock *sk)
1101 {
1102         read_lock(&sk->sk_callback_lock);
1103
1104         /* Do not wake up a writer until he can make "significant"
1105          * progress.  --DaveM
1106          */
1107         if((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1108                 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1109                         wake_up_interruptible(sk->sk_sleep);
1110
1111                 /* Should agree with poll, otherwise some programs break */
1112                 if (sock_writeable(sk))
1113                         sk_wake_async(sk, 2, POLL_OUT);
1114         }
1115
1116         read_unlock(&sk->sk_callback_lock);
1117 }
1118
1119 void sock_def_destruct(struct sock *sk)
1120 {
1121         if (sk->sk_protinfo)
1122                 kfree(sk->sk_protinfo);
1123 }
1124
1125 void sk_send_sigurg(struct sock *sk)
1126 {
1127         if (sk->sk_socket && sk->sk_socket->file)
1128                 if (send_sigurg(&sk->sk_socket->file->f_owner))
1129                         sk_wake_async(sk, 3, POLL_PRI);
1130 }
1131
1132 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1133                     unsigned long expires)
1134 {
1135         if (!mod_timer(timer, expires))
1136                 sock_hold(sk);
1137 }
1138
1139 EXPORT_SYMBOL(sk_reset_timer);
1140
1141 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1142 {
1143         if (timer_pending(timer) && del_timer(timer))
1144                 __sock_put(sk);
1145 }
1146
1147 EXPORT_SYMBOL(sk_stop_timer);
1148
1149 void sock_init_data(struct socket *sock, struct sock *sk)
1150 {
1151         skb_queue_head_init(&sk->sk_receive_queue);
1152         skb_queue_head_init(&sk->sk_write_queue);
1153         skb_queue_head_init(&sk->sk_error_queue);
1154
1155         sk->sk_send_head        =       NULL;
1156
1157         init_timer(&sk->sk_timer);
1158
1159         sk->sk_allocation       =       GFP_KERNEL;
1160         sk->sk_rcvbuf           =       sysctl_rmem_default;
1161         sk->sk_sndbuf           =       sysctl_wmem_default;
1162         sk->sk_state            =       TCP_CLOSE;
1163         sk->sk_zapped           =       1;
1164         sk->sk_socket           =       sock;
1165
1166         if(sock)
1167         {
1168                 sk->sk_type     =       sock->type;
1169                 sk->sk_sleep    =       &sock->wait;
1170                 sock->sk        =       sk;
1171         } else
1172                 sk->sk_sleep    =       NULL;
1173
1174         sk->sk_dst_lock         =       RW_LOCK_UNLOCKED;
1175         sk->sk_callback_lock    =       RW_LOCK_UNLOCKED;
1176
1177         sk->sk_state_change     =       sock_def_wakeup;
1178         sk->sk_data_ready       =       sock_def_readable;
1179         sk->sk_write_space      =       sock_def_write_space;
1180         sk->sk_error_report     =       sock_def_error_report;
1181         sk->sk_destruct         =       sock_def_destruct;
1182
1183         sk->sk_sndmsg_page      =       NULL;
1184         sk->sk_sndmsg_off       =       0;
1185
1186         sk->sk_peercred.pid     =       0;
1187         sk->sk_peercred.uid     =       -1;
1188         sk->sk_peercred.gid     =       -1;
1189         sk->sk_write_pending    =       0;
1190         sk->sk_rcvlowat         =       1;
1191         sk->sk_rcvtimeo         =       MAX_SCHEDULE_TIMEOUT;
1192         sk->sk_sndtimeo         =       MAX_SCHEDULE_TIMEOUT;
1193         sk->sk_owner            =       NULL;
1194
1195         sk->sk_stamp.tv_sec     = -1L;
1196         sk->sk_stamp.tv_usec    = -1L;
1197
1198         atomic_set(&sk->sk_refcnt, 1);
1199 }
1200
1201 void fastcall lock_sock(struct sock *sk)
1202 {
1203         might_sleep();
1204         spin_lock_bh(&(sk->sk_lock.slock));
1205         if (sk->sk_lock.owner)
1206                 __lock_sock(sk);
1207         sk->sk_lock.owner = (void *)1;
1208         spin_unlock_bh(&(sk->sk_lock.slock));
1209 }
1210
1211 EXPORT_SYMBOL(lock_sock);
1212
1213 void fastcall release_sock(struct sock *sk)
1214 {
1215         spin_lock_bh(&(sk->sk_lock.slock));
1216         if (sk->sk_backlog.tail)
1217                 __release_sock(sk);
1218         sk->sk_lock.owner = NULL;
1219         if (waitqueue_active(&(sk->sk_lock.wq)))
1220                 wake_up(&(sk->sk_lock.wq));
1221         spin_unlock_bh(&(sk->sk_lock.slock));
1222 }
1223 EXPORT_SYMBOL(release_sock);
1224
1225 /* When > 0 there are consumers of rx skb time stamps */
1226 atomic_t netstamp_needed = ATOMIC_INIT(0);
1227
1228 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1229 {
1230         if (!sock_flag(sk, SOCK_TIMESTAMP))
1231                 sock_enable_timestamp(sk);
1232         if (sk->sk_stamp.tv_sec == -1)
1233                 return -ENOENT;
1234         if (sk->sk_stamp.tv_sec == 0)
1235                 do_gettimeofday(&sk->sk_stamp);
1236         return copy_to_user(userstamp, &sk->sk_stamp, sizeof(struct timeval)) ?
1237                 -EFAULT : 0;
1238 }
1239 EXPORT_SYMBOL(sock_get_timestamp);
1240
1241 void sock_enable_timestamp(struct sock *sk)
1242 {
1243         if (!sock_flag(sk, SOCK_TIMESTAMP)) {
1244                 sock_set_flag(sk, SOCK_TIMESTAMP);
1245                 atomic_inc(&netstamp_needed);
1246         }
1247 }
1248 EXPORT_SYMBOL(sock_enable_timestamp);
1249
1250 void sock_disable_timestamp(struct sock *sk)
1251 {
1252         if (sock_flag(sk, SOCK_TIMESTAMP)) {
1253                 sock_reset_flag(sk, SOCK_TIMESTAMP);
1254                 atomic_dec(&netstamp_needed);
1255         }
1256 }
1257 EXPORT_SYMBOL(sock_disable_timestamp);
1258
1259 /*
1260  *      Get a socket option on an socket.
1261  *
1262  *      FIX: POSIX 1003.1g is very ambiguous here. It states that
1263  *      asynchronous errors should be reported by getsockopt. We assume
1264  *      this means if you specify SO_ERROR (otherwise whats the point of it).
1265  */
1266 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1267                            char __user *optval, int __user *optlen)
1268 {
1269         struct sock *sk = sock->sk;
1270
1271         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1272 }
1273
1274 EXPORT_SYMBOL(sock_common_getsockopt);
1275
1276 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1277                         struct msghdr *msg, size_t size, int flags)
1278 {
1279         struct sock *sk = sock->sk;
1280         int addr_len = 0;
1281         int err;
1282
1283         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
1284                                    flags & ~MSG_DONTWAIT, &addr_len);
1285         if (err >= 0)
1286                 msg->msg_namelen = addr_len;
1287         return err;
1288 }
1289
1290 EXPORT_SYMBOL(sock_common_recvmsg);
1291
1292 /*
1293  *      Set socket options on an inet socket.
1294  */
1295 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1296                            char __user *optval, int optlen)
1297 {
1298         struct sock *sk = sock->sk;
1299
1300         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1301 }
1302
1303 EXPORT_SYMBOL(sock_common_setsockopt);
1304
1305 void sk_common_release(struct sock *sk)
1306 {
1307         if (sk->sk_prot->destroy)
1308                 sk->sk_prot->destroy(sk);
1309
1310         /*
1311          * Observation: when sock_common_release is called, processes have
1312          * no access to socket. But net still has.
1313          * Step one, detach it from networking:
1314          *
1315          * A. Remove from hash tables.
1316          */
1317
1318         sk->sk_prot->unhash(sk);
1319
1320         /*
1321          * In this point socket cannot receive new packets, but it is possible
1322          * that some packets are in flight because some CPU runs receiver and
1323          * did hash table lookup before we unhashed socket. They will achieve
1324          * receive queue and will be purged by socket destructor.
1325          *
1326          * Also we still have packets pending on receive queue and probably,
1327          * our own packets waiting in device queues. sock_destroy will drain
1328          * receive queue, but transmitted packets will delay socket destruction
1329          * until the last reference will be released.
1330          */
1331
1332         sock_orphan(sk);
1333
1334         xfrm_sk_free_policy(sk);
1335
1336 #ifdef INET_REFCNT_DEBUG
1337         if (atomic_read(&sk->sk_refcnt) != 1)
1338                 printk(KERN_DEBUG "Destruction of the socket %p delayed, c=%d\n",
1339                        sk, atomic_read(&sk->sk_refcnt));
1340 #endif
1341         sock_put(sk);
1342 }
1343
1344 EXPORT_SYMBOL(sk_common_release);
1345
1346 int sk_alloc_slab(struct proto *prot, char *name)
1347 {
1348         prot->slab = kmem_cache_create(name,
1349                                        prot->slab_obj_size, 0,
1350                                        SLAB_HWCACHE_ALIGN, NULL, NULL);
1351
1352         return prot->slab != NULL ? 0 : -ENOBUFS;
1353 }
1354
1355 EXPORT_SYMBOL(sk_alloc_slab);
1356
1357 void sk_free_slab(struct proto *prot)
1358 {
1359         if (prot->slab != NULL) {
1360                 kmem_cache_destroy(prot->slab);
1361                 prot->slab = NULL;
1362         }
1363 }
1364
1365 EXPORT_SYMBOL(sk_free_slab);
1366
1367 EXPORT_SYMBOL(__lock_sock);
1368 EXPORT_SYMBOL(__release_sock);
1369 EXPORT_SYMBOL(sk_alloc);
1370 EXPORT_SYMBOL(sk_free);
1371 EXPORT_SYMBOL(sk_send_sigurg);
1372 EXPORT_SYMBOL(sock_alloc_send_pskb);
1373 EXPORT_SYMBOL(sock_alloc_send_skb);
1374 EXPORT_SYMBOL(sock_getsockopt);
1375 EXPORT_SYMBOL(sock_init_data);
1376 EXPORT_SYMBOL(sock_kfree_s);
1377 EXPORT_SYMBOL(sock_kmalloc);
1378 EXPORT_SYMBOL(sock_no_accept);
1379 EXPORT_SYMBOL(sock_no_bind);
1380 EXPORT_SYMBOL(sock_no_connect);
1381 EXPORT_SYMBOL(sock_no_getname);
1382 EXPORT_SYMBOL(sock_no_getsockopt);
1383 EXPORT_SYMBOL(sock_no_ioctl);
1384 EXPORT_SYMBOL(sock_no_listen);
1385 EXPORT_SYMBOL(sock_no_mmap);
1386 EXPORT_SYMBOL(sock_no_poll);
1387 EXPORT_SYMBOL(sock_no_recvmsg);
1388 EXPORT_SYMBOL(sock_no_release);
1389 EXPORT_SYMBOL(sock_no_sendmsg);
1390 EXPORT_SYMBOL(sock_no_sendpage);
1391 EXPORT_SYMBOL(sock_no_setsockopt);
1392 EXPORT_SYMBOL(sock_no_shutdown);
1393 EXPORT_SYMBOL(sock_no_socketpair);
1394 EXPORT_SYMBOL(sock_rfree);
1395 EXPORT_SYMBOL(sock_rmalloc);
1396 EXPORT_SYMBOL(sock_setsockopt);
1397 EXPORT_SYMBOL(sock_wfree);
1398 EXPORT_SYMBOL(sock_wmalloc);
1399 EXPORT_SYMBOL(sock_i_uid);
1400 EXPORT_SYMBOL(sock_i_ino);
1401 #ifdef CONFIG_SYSCTL
1402 EXPORT_SYMBOL(sysctl_optmem_max);
1403 EXPORT_SYMBOL(sysctl_rmem_max);
1404 EXPORT_SYMBOL(sysctl_wmem_max);
1405 #endif