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 static void sock_disable_timestamp(struct sock *sk)
 180 {
 181         if (sock_flag(sk, SOCK_TIMESTAMP)) {
 182                 sock_reset_flag(sk, SOCK_TIMESTAMP);
 183                 atomic_dec(&netstamp_needed);
 184         }
 185 }
 186
 187
 188 /*
 189  *      This is meant for all protocols to use and covers goings on
 190  *      at the socket level. Everything here is generic.
 191  */
 192
 193 int sock_setsockopt(struct socket *sock, int level, int optname,
 194                     char __user *optval, int optlen)
 195 {
 196         struct sock *sk=sock->sk;
 197         struct sk_filter *filter;
 198         int val;
 199         int valbool;
 200         struct linger ling;
 201         int ret = 0;
 202
 203         /*
 204          *      Options without arguments
 205          */
 206
 207 #ifdef SO_DONTLINGER            /* Compatibility item... */
 208         switch (optname) {
 209                 case SO_DONTLINGER:
 210                         sock_reset_flag(sk, SOCK_LINGER);
 211                         return 0;
 212         }
 213 #endif
 214
 215         if(optlen<sizeof(int))
 216                 return(-EINVAL);
 217
 218         if (get_user(val, (int __user *)optval))
 219                 return -EFAULT;
 220
 221         valbool = val?1:0;
 222
 223         lock_sock(sk);
 224
 225         switch(optname)
 226         {
 227                 case SO_DEBUG:
 228                         if(val && !capable(CAP_NET_ADMIN))
 229                         {
 230                                 ret = -EACCES;
 231                         }
 232                         else
 233                                 sk->sk_debug = valbool;
 234                         break;
 235                 case SO_REUSEADDR:
 236                         sk->sk_reuse = valbool;
 237                         break;
 238                 case SO_TYPE:
 239                 case SO_ERROR:
 240                         ret = -ENOPROTOOPT;
 241                         break;
 242                 case SO_DONTROUTE:
 243                         sk->sk_localroute = valbool;
 244                         break;
 245                 case SO_BROADCAST:
 246                         sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
 247                         break;
 248                 case SO_SNDBUF:
 249                         /* Don't error on this BSD doesn't and if you think
 250                            about it this is right. Otherwise apps have to
 251                            play 'guess the biggest size' games. RCVBUF/SNDBUF
 252                            are treated in BSD as hints */
 253
 254                         if (val > sysctl_wmem_max)
 255                                 val = sysctl_wmem_max;
 256
 257                         sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
 258                         if ((val * 2) < SOCK_MIN_SNDBUF)
 259                                 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
 260                         else
 261                                 sk->sk_sndbuf = val * 2;
 262
 263                         /*
 264                          *      Wake up sending tasks if we
 265                          *      upped the value.
 266                          */
 267                         sk->sk_write_space(sk);
 268                         break;
 269
 270                 case SO_RCVBUF:
 271                         /* Don't error on this BSD doesn't and if you think
 272                            about it this is right. Otherwise apps have to
 273                            play 'guess the biggest size' games. RCVBUF/SNDBUF
 274                            are treated in BSD as hints */
 275
 276                         if (val > sysctl_rmem_max)
 277                                 val = sysctl_rmem_max;
 278
 279                         sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
 280                         /* FIXME: is this lower bound the right one? */
 281                         if ((val * 2) < SOCK_MIN_RCVBUF)
 282                                 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
 283                         else
 284                                 sk->sk_rcvbuf = val * 2;
 285                         break;
 286
 287                 case SO_KEEPALIVE:
 288 #ifdef CONFIG_INET
 289                         if (sk->sk_protocol == IPPROTO_TCP)
 290                                 tcp_set_keepalive(sk, valbool);
 291 #endif
 292                         sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
 293                         break;
 294
 295                 case SO_OOBINLINE:
 296                         sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
 297                         break;
 298
 299                 case SO_NO_CHECK:
 300                         sk->sk_no_check = valbool;
 301                         break;
 302
 303                 case SO_PRIORITY:
 304                         if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
 305                                 sk->sk_priority = val;
 306                         else
 307                                 ret = -EPERM;
 308                         break;
 309
 310                 case SO_LINGER:
 311                         if(optlen<sizeof(ling)) {
 312                                 ret = -EINVAL;  /* 1003.1g */
 313                                 break;
 314                         }
 315                         if (copy_from_user(&ling,optval,sizeof(ling))) {
 316                                 ret = -EFAULT;
 317                                 break;
 318                         }
 319                         if (!ling.l_onoff)
 320                                 sock_reset_flag(sk, SOCK_LINGER);
 321                         else {
 322 #if (BITS_PER_LONG == 32)
 323                                 if (ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
 324                                         sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
 325                                 else
 326 #endif
 327                                         sk->sk_lingertime = ling.l_linger * HZ;
 328                                 sock_set_flag(sk, SOCK_LINGER);
 329                         }
 330                         break;
 331
 332                 case SO_BSDCOMPAT:
 333                         sock_warn_obsolete_bsdism("setsockopt");
 334                         break;
 335
 336                 case SO_PASSCRED:
 337                         if (valbool)
 338                                 set_bit(SOCK_PASS_CRED, &sock->flags);
 339                         else
 340                                 clear_bit(SOCK_PASS_CRED, &sock->flags);
 341                         break;
 342
 343 #if defined(CONFIG_VNET) || defined(CONFIG_VNET_MODULE)
 344                 case SO_SETXID:
 345                         if (current->xid) {
 346                                 ret = -EPERM;
 347                                 break;
 348                         }
 349                         if (val < 0 || val > MAX_S_CONTEXT) {
 350                                 ret = -EINVAL;
 351                                 break;
 352                         }
 353                         sk->sk_xid = val;
 354                         break;
 355 #endif
 356
 357                 case SO_TIMESTAMP:
 358                         sk->sk_rcvtstamp = valbool;
 359                         if (valbool)
 360                                 sock_enable_timestamp(sk);
 361                         break;
 362
 363                 case SO_RCVLOWAT:
 364                         if (val < 0)
 365                                 val = INT_MAX;
 366                         sk->sk_rcvlowat = val ? : 1;
 367                         break;
 368
 369                 case SO_RCVTIMEO:
 370                         ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
 371                         break;
 372
 373                 case SO_SNDTIMEO:
 374                         ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
 375                         break;
 376
 377 #ifdef CONFIG_NETDEVICES
 378                 case SO_BINDTODEVICE:
 379                 {
 380                         char devname[IFNAMSIZ];
 381
 382                         /* Sorry... */
 383                         if (!capable(CAP_NET_RAW)) {
 384                                 ret = -EPERM;
 385                                 break;
 386                         }
 387
 388                         /* Bind this socket to a particular device like "eth0",
 389                          * as specified in the passed interface name. If the
 390                          * name is "" or the option length is zero the socket
 391                          * is not bound.
 392                          */
 393
 394                         if (!valbool) {
 395                                 sk->sk_bound_dev_if = 0;
 396                         } else {
 397                                 if (optlen > IFNAMSIZ)
 398                                         optlen = IFNAMSIZ;
 399                                 if (copy_from_user(devname, optval, optlen)) {
 400                                         ret = -EFAULT;
 401                                         break;
 402                                 }
 403
 404                                 /* Remove any cached route for this socket. */
 405                                 sk_dst_reset(sk);
 406
 407                                 if (devname[0] == '\0') {
 408                                         sk->sk_bound_dev_if = 0;
 409                                 } else {
 410                                         struct net_device *dev = dev_get_by_name(devname);
 411                                         if (!dev) {
 412                                                 ret = -ENODEV;
 413                                                 break;
 414                                         }
 415                                         sk->sk_bound_dev_if = dev->ifindex;
 416                                         dev_put(dev);
 417                                 }
 418                         }
 419                         break;
 420                 }
 421 #endif
 422
 423
 424                 case SO_ATTACH_FILTER:
 425                         ret = -EINVAL;
 426                         if (optlen == sizeof(struct sock_fprog)) {
 427                                 struct sock_fprog fprog;
 428
 429                                 ret = -EFAULT;
 430                                 if (copy_from_user(&fprog, optval, sizeof(fprog)))
 431                                         break;
 432
 433                                 ret = sk_attach_filter(&fprog, sk);
 434                         }
 435                         break;
 436
 437                 case SO_DETACH_FILTER:
 438                         spin_lock_bh(&sk->sk_lock.slock);
 439                         filter = sk->sk_filter;
 440                         if (filter) {
 441                                 sk->sk_filter = NULL;
 442                                 spin_unlock_bh(&sk->sk_lock.slock);
 443                                 sk_filter_release(sk, filter);
 444                                 break;
 445                         }
 446                         spin_unlock_bh(&sk->sk_lock.slock);
 447                         ret = -ENONET;
 448                         break;
 449
 450                 /* We implement the SO_SNDLOWAT etc to
 451                    not be settable (1003.1g 5.3) */
 452                 default:
 453                         ret = -ENOPROTOOPT;
 454                         break;
 455         }
 456         release_sock(sk);
 457         return ret;
 458 }
 459
 460
 461 int sock_getsockopt(struct socket *sock, int level, int optname,
 462                     char __user *optval, int __user *optlen)
 463 {
 464         struct sock *sk = sock->sk;
 465
 466         union
 467         {
 468                 int val;
 469                 struct linger ling;
 470                 struct timeval tm;
 471         } v;
 472
 473         unsigned int lv = sizeof(int);
 474         int len;
 475
 476         if(get_user(len,optlen))
 477                 return -EFAULT;
 478         if(len < 0)
 479                 return -EINVAL;
 480
 481         switch(optname)
 482         {
 483                 case SO_DEBUG:
 484                         v.val = sk->sk_debug;
 485                         break;
 486
 487                 case SO_DONTROUTE:
 488                         v.val = sk->sk_localroute;
 489                         break;
 490
 491                 case SO_BROADCAST:
 492                         v.val = !!sock_flag(sk, SOCK_BROADCAST);
 493                         break;
 494
 495                 case SO_SNDBUF:
 496                         v.val = sk->sk_sndbuf;
 497                         break;
 498
 499                 case SO_RCVBUF:
 500                         v.val = sk->sk_rcvbuf;
 501                         break;
 502
 503                 case SO_REUSEADDR:
 504                         v.val = sk->sk_reuse;
 505                         break;
 506
 507                 case SO_KEEPALIVE:
 508                         v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
 509                         break;
 510
 511                 case SO_TYPE:
 512                         v.val = sk->sk_type;
 513                         break;
 514
 515                 case SO_ERROR:
 516                         v.val = -sock_error(sk);
 517                         if(v.val==0)
 518                                 v.val = xchg(&sk->sk_err_soft, 0);
 519                         break;
 520
 521                 case SO_OOBINLINE:
 522                         v.val = !!sock_flag(sk, SOCK_URGINLINE);
 523                         break;
 524
 525                 case SO_NO_CHECK:
 526                         v.val = sk->sk_no_check;
 527                         break;
 528
 529                 case SO_PRIORITY:
 530                         v.val = sk->sk_priority;
 531                         break;
 532
 533                 case SO_LINGER:
 534                         lv              = sizeof(v.ling);
 535                         v.ling.l_onoff  = !!sock_flag(sk, SOCK_LINGER);
 536                         v.ling.l_linger = sk->sk_lingertime / HZ;
 537                         break;
 538
 539                 case SO_BSDCOMPAT:
 540                         sock_warn_obsolete_bsdism("getsockopt");
 541                         break;
 542
 543                 case SO_TIMESTAMP:
 544                         v.val = sk->sk_rcvtstamp;
 545                         break;
 546
 547                 case SO_RCVTIMEO:
 548                         lv=sizeof(struct timeval);
 549                         if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
 550                                 v.tm.tv_sec = 0;
 551                                 v.tm.tv_usec = 0;
 552                         } else {
 553                                 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
 554                                 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
 555                         }
 556                         break;
 557
 558                 case SO_SNDTIMEO:
 559                         lv=sizeof(struct timeval);
 560                         if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
 561                                 v.tm.tv_sec = 0;
 562                                 v.tm.tv_usec = 0;
 563                         } else {
 564                                 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
 565                                 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
 566                         }
 567                         break;
 568
 569                 case SO_RCVLOWAT:
 570                         v.val = sk->sk_rcvlowat;
 571                         break;
 572
 573                 case SO_SNDLOWAT:
 574                         v.val=1;
 575                         break;
 576
 577                 case SO_PASSCRED:
 578                         v.val = test_bit(SOCK_PASS_CRED, &sock->flags)?1:0;
 579                         break;
 580
 581                 case SO_PEERCRED:
 582                         if (len > sizeof(sk->sk_peercred))
 583                                 len = sizeof(sk->sk_peercred);
 584                         if (copy_to_user(optval, &sk->sk_peercred, len))
 585                                 return -EFAULT;
 586                         goto lenout;
 587
 588                 case SO_PEERNAME:
 589                 {
 590                         char address[128];
 591
 592                         if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
 593                                 return -ENOTCONN;
 594                         if (lv < len)
 595                                 return -EINVAL;
 596                         if (copy_to_user(optval, address, len))
 597                                 return -EFAULT;
 598                         goto lenout;
 599                 }
 600
 601                 /* Dubious BSD thing... Probably nobody even uses it, but
 602                  * the UNIX standard wants it for whatever reason... -DaveM
 603                  */
 604                 case SO_ACCEPTCONN:
 605                         v.val = sk->sk_state == TCP_LISTEN;
 606                         break;
 607
 608                 case SO_PEERSEC:
 609                         return security_socket_getpeersec(sock, optval, optlen, len);
 610
 611                 default:
 612                         return(-ENOPROTOOPT);
 613         }
 614         if (len > lv)
 615                 len = lv;
 616         if (copy_to_user(optval, &v, len))
 617                 return -EFAULT;
 618 lenout:
 619         if (put_user(len, optlen))
 620                 return -EFAULT;
 621         return 0;
 622 }
 623
 624 static kmem_cache_t *sk_cachep;
 625
 626 /**
 627  *      sk_alloc - All socket objects are allocated here
 628  *      @family - protocol family
 629  *      @priority - for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
 630  *      @zero_it - zeroes the allocated sock
 631  *      @slab - alternate slab
 632  *
 633  *      All socket objects are allocated here. If @zero_it is non-zero
 634  *      it should have the size of the are to be zeroed, because the
 635  *      private slabcaches have different sizes of the generic struct sock.
 636  *      1 has been kept as a way to say sizeof(struct sock).
 637  */
 638 struct sock *sk_alloc(int family, int priority, int zero_it, kmem_cache_t *slab)
 639 {
 640         struct sock *sk = NULL;
 641
 642         if (!slab)
 643                 slab = sk_cachep;
 644         sk = kmem_cache_alloc(slab, priority);
 645         if (sk) {
 646                 if (zero_it) {
 647                         memset(sk, 0,
 648                                zero_it == 1 ? sizeof(struct sock) : zero_it);
 649                         sk->sk_family = family;
 650                         sock_lock_init(sk);
 651                 }
 652                 sk->sk_slab = slab;
 653                 sock_vx_init(sk);
 654                 sock_nx_init(sk);
 655
 656                 if (security_sk_alloc(sk, family, priority)) {
 657                         kmem_cache_free(slab, sk);
 658                         sk = NULL;
 659                 }
 660         }
 661         return sk;
 662 }
 663
 664 void sk_free(struct sock *sk)
 665 {
 666         struct sk_filter *filter;
 667         struct module *owner = sk->sk_owner;
 668
 669         if (sk->sk_destruct)
 670                 sk->sk_destruct(sk);
 671
 672         filter = sk->sk_filter;
 673         if (filter) {
 674                 sk_filter_release(sk, filter);
 675                 sk->sk_filter = NULL;
 676         }
 677
 678         sock_disable_timestamp(sk);
 679
 680         if (atomic_read(&sk->sk_omem_alloc))
 681                 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
 682                        __FUNCTION__, atomic_read(&sk->sk_omem_alloc));
 683
 684         security_sk_free(sk);
 685         BUG_ON(sk->sk_vx_info);
 686         BUG_ON(sk->sk_nx_info);
 687         kmem_cache_free(sk->sk_slab, sk);
 688         module_put(owner);
 689 }
 690
 691 void __init sk_init(void)
 692 {
 693         sk_cachep = kmem_cache_create("sock", sizeof(struct sock), 0,
 694                                       SLAB_HWCACHE_ALIGN, NULL, NULL);
 695         if (!sk_cachep)
 696                 printk(KERN_CRIT "sk_init: Cannot create sock SLAB cache!");
 697
 698         if (num_physpages <= 4096) {
 699                 sysctl_wmem_max = 32767;
 700                 sysctl_rmem_max = 32767;
 701                 sysctl_wmem_default = 32767;
 702                 sysctl_rmem_default = 32767;
 703         } else if (num_physpages >= 131072) {
 704                 sysctl_wmem_max = 131071;
 705                 sysctl_rmem_max = 131071;
 706         }
 707 }
 708
 709 /*
 710  *      Simple resource managers for sockets.
 711  */
 712
 713
 714 /*
 715  * Write buffer destructor automatically called from kfree_skb.
 716  */
 717 void sock_wfree(struct sk_buff *skb)
 718 {
 719         struct sock *sk = skb->sk;
 720
 721         /* In case it might be waiting for more memory. */
 722         atomic_sub(skb->truesize, &sk->sk_wmem_alloc);
 723         if (!sk->sk_use_write_queue)
 724                 sk->sk_write_space(sk);
 725         sock_put(sk);
 726 }
 727
 728 /*
 729  * Read buffer destructor automatically called from kfree_skb.
 730  */
 731 void sock_rfree(struct sk_buff *skb)
 732 {
 733         struct sock *sk = skb->sk;
 734
 735         atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
 736 }
 737
 738
 739 int sock_i_uid(struct sock *sk)
 740 {
 741         int uid;
 742
 743         read_lock(&sk->sk_callback_lock);
 744         uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
 745         read_unlock(&sk->sk_callback_lock);
 746         return uid;
 747 }
 748
 749 unsigned long sock_i_ino(struct sock *sk)
 750 {
 751         unsigned long ino;
 752
 753         read_lock(&sk->sk_callback_lock);
 754         ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
 755         read_unlock(&sk->sk_callback_lock);
 756         return ino;
 757 }
 758
 759 /*
 760  * Allocate a skb from the socket's send buffer.
 761  */
 762 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, int priority)
 763 {
 764         if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
 765                 struct sk_buff * skb = alloc_skb(size, priority);
 766                 if (skb) {
 767                         skb_set_owner_w(skb, sk);
 768                         return skb;
 769                 }
 770         }
 771         return NULL;
 772 }
 773
 774 /*
 775  * Allocate a skb from the socket's receive buffer.
 776  */
 777 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force, int priority)
 778 {
 779         if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
 780                 struct sk_buff *skb = alloc_skb(size, priority);
 781                 if (skb) {
 782                         skb_set_owner_r(skb, sk);
 783                         return skb;
 784                 }
 785         }
 786         return NULL;
 787 }
 788
 789 /*
 790  * Allocate a memory block from the socket's option memory buffer.
 791  */
 792 void *sock_kmalloc(struct sock *sk, int size, int priority)
 793 {
 794         if ((unsigned)size <= sysctl_optmem_max &&
 795             atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
 796                 void *mem;
 797                 /* First do the add, to avoid the race if kmalloc
 798                  * might sleep.
 799                  */
 800                 atomic_add(size, &sk->sk_omem_alloc);
 801                 mem = kmalloc(size, priority);
 802                 if (mem)
 803                         return mem;
 804                 atomic_sub(size, &sk->sk_omem_alloc);
 805         }
 806         return NULL;
 807 }
 808
 809 /*
 810  * Free an option memory block.
 811  */
 812 void sock_kfree_s(struct sock *sk, void *mem, int size)
 813 {
 814         kfree(mem);
 815         atomic_sub(size, &sk->sk_omem_alloc);
 816 }
 817
 818 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
 819    I think, these locks should be removed for datagram sockets.
 820  */
 821 static long sock_wait_for_wmem(struct sock * sk, long timeo)
 822 {
 823         DEFINE_WAIT(wait);
 824
 825         clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
 826         for (;;) {
 827                 if (!timeo)
 828                         break;
 829                 if (signal_pending(current))
 830                         break;
 831                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
 832                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
 833                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
 834                         break;
 835                 if (sk->sk_shutdown & SEND_SHUTDOWN)
 836                         break;
 837                 if (sk->sk_err)
 838                         break;
 839                 timeo = schedule_timeout(timeo);
 840         }
 841         finish_wait(sk->sk_sleep, &wait);
 842         return timeo;
 843 }
 844
 845
 846 /*
 847  *      Generic send/receive buffer handlers
 848  */
 849
 850 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
 851                                      unsigned long data_len, int noblock, int *errcode)
 852 {
 853         struct sk_buff *skb;
 854         unsigned int gfp_mask;
 855         long timeo;
 856         int err;
 857
 858         gfp_mask = sk->sk_allocation;
 859         if (gfp_mask & __GFP_WAIT)
 860                 gfp_mask |= __GFP_REPEAT;
 861
 862         timeo = sock_sndtimeo(sk, noblock);
 863         while (1) {
 864                 err = sock_error(sk);
 865                 if (err != 0)
 866                         goto failure;
 867
 868                 err = -EPIPE;
 869                 if (sk->sk_shutdown & SEND_SHUTDOWN)
 870                         goto failure;
 871
 872                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
 873                         skb = alloc_skb(header_len, sk->sk_allocation);
 874                         if (skb) {
 875                                 int npages;
 876                                 int i;
 877
 878                                 /* No pages, we're done... */
 879                                 if (!data_len)
 880                                         break;
 881
 882                                 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
 883                                 skb->truesize += data_len;
 884                                 skb_shinfo(skb)->nr_frags = npages;
 885                                 for (i = 0; i < npages; i++) {
 886                                         struct page *page;
 887                                         skb_frag_t *frag;
 888
 889                                         page = alloc_pages(sk->sk_allocation, 0);
 890                                         if (!page) {
 891                                                 err = -ENOBUFS;
 892                                                 skb_shinfo(skb)->nr_frags = i;
 893                                                 kfree_skb(skb);
 894                                                 goto failure;
 895                                         }
 896
 897                                         frag = &skb_shinfo(skb)->frags[i];
 898                                         frag->page = page;
 899                                         frag->page_offset = 0;
 900                                         frag->size = (data_len >= PAGE_SIZE ?
 901                                                       PAGE_SIZE :
 902                                                       data_len);
 903                                         data_len -= PAGE_SIZE;
 904                                 }
 905
 906                                 /* Full success... */
 907                                 break;
 908                         }
 909                         err = -ENOBUFS;
 910                         goto failure;
 911                 }
 912                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
 913                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
 914                 err = -EAGAIN;
 915                 if (!timeo)
 916                         goto failure;
 917                 if (signal_pending(current))
 918                         goto interrupted;
 919                 timeo = sock_wait_for_wmem(sk, timeo);
 920         }
 921
 922         skb_set_owner_w(skb, sk);
 923         return skb;
 924
 925 interrupted:
 926         err = sock_intr_errno(timeo);
 927 failure:
 928         *errcode = err;
 929         return NULL;
 930 }
 931
 932 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
 933                                     int noblock, int *errcode)
 934 {
 935         return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
 936 }
 937
 938 static void __lock_sock(struct sock *sk)
 939 {
 940         DEFINE_WAIT(wait);
 941
 942         for(;;) {
 943                 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
 944                                         TASK_UNINTERRUPTIBLE);
 945                 spin_unlock_bh(&sk->sk_lock.slock);
 946                 schedule();
 947                 spin_lock_bh(&sk->sk_lock.slock);
 948                 if(!sock_owned_by_user(sk))
 949                         break;
 950         }
 951         finish_wait(&sk->sk_lock.wq, &wait);
 952 }
 953
 954 static void __release_sock(struct sock *sk)
 955 {
 956         struct sk_buff *skb = sk->sk_backlog.head;
 957
 958         do {
 959                 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
 960                 bh_unlock_sock(sk);
 961
 962                 do {
 963                         struct sk_buff *next = skb->next;
 964
 965                         skb->next = NULL;
 966                         sk->sk_backlog_rcv(sk, skb);
 967                         skb = next;
 968                 } while (skb != NULL);
 969
 970                 bh_lock_sock(sk);
 971         } while((skb = sk->sk_backlog.head) != NULL);
 972 }
 973
 974 /**
 975  * sk_wait_data - wait for data to arrive at sk_receive_queue
 976  * sk - sock to wait on
 977  * timeo - for how long
 978  *
 979  * Now socket state including sk->sk_err is changed only under lock,
 980  * hence we may omit checks after joining wait queue.
 981  * We check receive queue before schedule() only as optimization;
 982  * it is very likely that release_sock() added new data.
 983  */
 984 int sk_wait_data(struct sock *sk, long *timeo)
 985 {
 986         int rc;
 987         DEFINE_WAIT(wait);
 988
 989         prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
 990         set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
 991         rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
 992         clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
 993         finish_wait(sk->sk_sleep, &wait);
 994         return rc;
 995 }
 996
 997 EXPORT_SYMBOL(sk_wait_data);
 998
 999 /*
1000  * Set of default routines for initialising struct proto_ops when
1001  * the protocol does not support a particular function. In certain
1002  * cases where it makes no sense for a protocol to have a "do nothing"
1003  * function, some default processing is provided.
1004  */
1005
1006 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1007 {
1008         return -EOPNOTSUPP;
1009 }
1010
1011 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1012                     int len, int flags)
1013 {
1014         return -EOPNOTSUPP;
1015 }
1016
1017 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1018 {
1019         return -EOPNOTSUPP;
1020 }
1021
1022 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1023 {
1024         return -EOPNOTSUPP;
1025 }
1026
1027 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1028                     int *len, int peer)
1029 {
1030         return -EOPNOTSUPP;
1031 }
1032
1033 unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt)
1034 {
1035         return 0;
1036 }
1037
1038 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1039 {
1040         return -EOPNOTSUPP;
1041 }
1042
1043 int sock_no_listen(struct socket *sock, int backlog)
1044 {
1045         return -EOPNOTSUPP;
1046 }
1047
1048 int sock_no_shutdown(struct socket *sock, int how)
1049 {
1050         return -EOPNOTSUPP;
1051 }
1052
1053 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1054                     char __user *optval, int optlen)
1055 {
1056         return -EOPNOTSUPP;
1057 }
1058
1059 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1060                     char __user *optval, int __user *optlen)
1061 {
1062         return -EOPNOTSUPP;
1063 }
1064
1065 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1066                     size_t len)
1067 {
1068         return -EOPNOTSUPP;
1069 }
1070
1071 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1072                     size_t len, int flags)
1073 {
1074         return -EOPNOTSUPP;
1075 }
1076
1077 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1078 {
1079         /* Mirror missing mmap method error code */
1080         return -ENODEV;
1081 }
1082
1083 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1084 {
1085         ssize_t res;
1086         struct msghdr msg = {.msg_flags = flags};
1087         struct kvec iov;
1088         char *kaddr = kmap(page);
1089         iov.iov_base = kaddr + offset;
1090         iov.iov_len = size;
1091         res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1092         kunmap(page);
1093         return res;
1094 }
1095
1096 /*
1097  *      Default Socket Callbacks
1098  */
1099
1100 void sock_def_wakeup(struct sock *sk)
1101 {
1102         read_lock(&sk->sk_callback_lock);
1103         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1104                 wake_up_interruptible_all(sk->sk_sleep);
1105         read_unlock(&sk->sk_callback_lock);
1106 }
1107
1108 void sock_def_error_report(struct sock *sk)
1109 {
1110         read_lock(&sk->sk_callback_lock);
1111         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1112                 wake_up_interruptible(sk->sk_sleep);
1113         sk_wake_async(sk,0,POLL_ERR);
1114         read_unlock(&sk->sk_callback_lock);
1115 }
1116
1117 void sock_def_readable(struct sock *sk, int len)
1118 {
1119         read_lock(&sk->sk_callback_lock);
1120         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1121                 wake_up_interruptible(sk->sk_sleep);
1122         sk_wake_async(sk,1,POLL_IN);
1123         read_unlock(&sk->sk_callback_lock);
1124 }
1125
1126 void sock_def_write_space(struct sock *sk)
1127 {
1128         read_lock(&sk->sk_callback_lock);
1129
1130         /* Do not wake up a writer until he can make "significant"
1131          * progress.  --DaveM
1132          */
1133         if((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1134                 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1135                         wake_up_interruptible(sk->sk_sleep);
1136
1137                 /* Should agree with poll, otherwise some programs break */
1138                 if (sock_writeable(sk))
1139                         sk_wake_async(sk, 2, POLL_OUT);
1140         }
1141
1142         read_unlock(&sk->sk_callback_lock);
1143 }
1144
1145 void sock_def_destruct(struct sock *sk)
1146 {
1147         if (sk->sk_protinfo)
1148                 kfree(sk->sk_protinfo);
1149 }
1150
1151 void sk_send_sigurg(struct sock *sk)
1152 {
1153         if (sk->sk_socket && sk->sk_socket->file)
1154                 if (send_sigurg(&sk->sk_socket->file->f_owner))
1155                         sk_wake_async(sk, 3, POLL_PRI);
1156 }
1157
1158 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1159                     unsigned long expires)
1160 {
1161         if (!mod_timer(timer, expires))
1162                 sock_hold(sk);
1163 }
1164
1165 EXPORT_SYMBOL(sk_reset_timer);
1166
1167 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1168 {
1169         if (timer_pending(timer) && del_timer(timer))
1170                 __sock_put(sk);
1171 }
1172
1173 EXPORT_SYMBOL(sk_stop_timer);
1174
1175 void sock_init_data(struct socket *sock, struct sock *sk)
1176 {
1177         skb_queue_head_init(&sk->sk_receive_queue);
1178         skb_queue_head_init(&sk->sk_write_queue);
1179         skb_queue_head_init(&sk->sk_error_queue);
1180
1181         sk->sk_send_head        =       NULL;
1182
1183         init_timer(&sk->sk_timer);
1184
1185         sk->sk_allocation       =       GFP_KERNEL;
1186         sk->sk_rcvbuf           =       sysctl_rmem_default;
1187         sk->sk_sndbuf           =       sysctl_wmem_default;
1188         sk->sk_state            =       TCP_CLOSE;
1189         sk->sk_zapped           =       1;
1190         sk->sk_socket           =       sock;
1191
1192         if(sock)
1193         {
1194                 sk->sk_type     =       sock->type;
1195                 sk->sk_sleep    =       &sock->wait;
1196                 sock->sk        =       sk;
1197         } else
1198                 sk->sk_sleep    =       NULL;
1199
1200         sk->sk_dst_lock         =       RW_LOCK_UNLOCKED;
1201         sk->sk_callback_lock    =       RW_LOCK_UNLOCKED;
1202
1203         sk->sk_state_change     =       sock_def_wakeup;
1204         sk->sk_data_ready       =       sock_def_readable;
1205         sk->sk_write_space      =       sock_def_write_space;
1206         sk->sk_error_report     =       sock_def_error_report;
1207         sk->sk_destruct         =       sock_def_destruct;
1208
1209         sk->sk_sndmsg_page      =       NULL;
1210         sk->sk_sndmsg_off       =       0;
1211
1212         sk->sk_peercred.pid     =       0;
1213         sk->sk_peercred.uid     =       -1;
1214         sk->sk_peercred.gid     =       -1;
1215         sk->sk_write_pending    =       0;
1216         sk->sk_rcvlowat         =       1;
1217         sk->sk_rcvtimeo         =       MAX_SCHEDULE_TIMEOUT;
1218         sk->sk_sndtimeo         =       MAX_SCHEDULE_TIMEOUT;
1219         sk->sk_owner            =       NULL;
1220
1221         sk->sk_stamp.tv_sec     = -1L;
1222         sk->sk_stamp.tv_usec    = -1L;
1223
1224         sk->sk_vx_info          =       NULL;
1225         sk->sk_xid              =       0;
1226         sk->sk_nx_info          =       NULL;
1227         sk->sk_nid              =       0;
1228
1229         atomic_set(&sk->sk_refcnt, 1);
1230 }
1231
1232 void fastcall lock_sock(struct sock *sk)
1233 {
1234         might_sleep();
1235         spin_lock_bh(&(sk->sk_lock.slock));
1236         if (sk->sk_lock.owner)
1237                 __lock_sock(sk);
1238         sk->sk_lock.owner = (void *)1;
1239         spin_unlock_bh(&(sk->sk_lock.slock));
1240 }
1241
1242 EXPORT_SYMBOL(lock_sock);
1243
1244 void fastcall release_sock(struct sock *sk)
1245 {
1246         spin_lock_bh(&(sk->sk_lock.slock));
1247         if (sk->sk_backlog.tail)
1248                 __release_sock(sk);
1249         sk->sk_lock.owner = NULL;
1250         if (waitqueue_active(&(sk->sk_lock.wq)))
1251                 wake_up(&(sk->sk_lock.wq));
1252         spin_unlock_bh(&(sk->sk_lock.slock));
1253 }
1254 EXPORT_SYMBOL(release_sock);
1255
1256 /* When > 0 there are consumers of rx skb time stamps */
1257 atomic_t netstamp_needed = ATOMIC_INIT(0);
1258
1259 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1260 {
1261         if (!sock_flag(sk, SOCK_TIMESTAMP))
1262                 sock_enable_timestamp(sk);
1263         if (sk->sk_stamp.tv_sec == -1)
1264                 return -ENOENT;
1265         if (sk->sk_stamp.tv_sec == 0)
1266                 do_gettimeofday(&sk->sk_stamp);
1267         return copy_to_user(userstamp, &sk->sk_stamp, sizeof(struct timeval)) ?
1268                 -EFAULT : 0;
1269 }
1270 EXPORT_SYMBOL(sock_get_timestamp);
1271
1272 void sock_enable_timestamp(struct sock *sk)
1273 {
1274         if (!sock_flag(sk, SOCK_TIMESTAMP)) {
1275                 sock_set_flag(sk, SOCK_TIMESTAMP);
1276                 atomic_inc(&netstamp_needed);
1277         }
1278 }
1279 EXPORT_SYMBOL(sock_enable_timestamp);
1280
1281 /*
1282  *      Get a socket option on an socket.
1283  *
1284  *      FIX: POSIX 1003.1g is very ambiguous here. It states that
1285  *      asynchronous errors should be reported by getsockopt. We assume
1286  *      this means if you specify SO_ERROR (otherwise whats the point of it).
1287  */
1288 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1289                            char __user *optval, int __user *optlen)
1290 {
1291         struct sock *sk = sock->sk;
1292
1293         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1294 }
1295
1296 EXPORT_SYMBOL(sock_common_getsockopt);
1297
1298 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1299                         struct msghdr *msg, size_t size, int flags)
1300 {
1301         struct sock *sk = sock->sk;
1302         int addr_len = 0;
1303         int err;
1304
1305         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
1306                                    flags & ~MSG_DONTWAIT, &addr_len);
1307         if (err >= 0)
1308                 msg->msg_namelen = addr_len;
1309         return err;
1310 }
1311
1312 EXPORT_SYMBOL(sock_common_recvmsg);
1313
1314 /*
1315  *      Set socket options on an inet socket.
1316  */
1317 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1318                            char __user *optval, int optlen)
1319 {
1320         struct sock *sk = sock->sk;
1321
1322         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1323 }
1324
1325 EXPORT_SYMBOL(sock_common_setsockopt);
1326
1327 void sk_common_release(struct sock *sk)
1328 {
1329         if (sk->sk_prot->destroy)
1330                 sk->sk_prot->destroy(sk);
1331
1332         /*
1333          * Observation: when sock_common_release is called, processes have
1334          * no access to socket. But net still has.
1335          * Step one, detach it from networking:
1336          *
1337          * A. Remove from hash tables.
1338          */
1339
1340         sk->sk_prot->unhash(sk);
1341
1342         /*
1343          * In this point socket cannot receive new packets, but it is possible
1344          * that some packets are in flight because some CPU runs receiver and
1345          * did hash table lookup before we unhashed socket. They will achieve
1346          * receive queue and will be purged by socket destructor.
1347          *
1348          * Also we still have packets pending on receive queue and probably,
1349          * our own packets waiting in device queues. sock_destroy will drain
1350          * receive queue, but transmitted packets will delay socket destruction
1351          * until the last reference will be released.
1352          */
1353
1354         sock_orphan(sk);
1355
1356         xfrm_sk_free_policy(sk);
1357
1358 #ifdef INET_REFCNT_DEBUG
1359         if (atomic_read(&sk->sk_refcnt) != 1)
1360                 printk(KERN_DEBUG "Destruction of the socket %p delayed, c=%d\n",
1361                        sk, atomic_read(&sk->sk_refcnt));
1362 #endif
1363         sock_put(sk);
1364 }
1365
1366 EXPORT_SYMBOL(sk_common_release);
1367
1368 int sk_alloc_slab(struct proto *prot, char *name)
1369 {
1370         prot->slab = kmem_cache_create(name,
1371                                        prot->slab_obj_size, 0,
1372                                        SLAB_HWCACHE_ALIGN, NULL, NULL);
1373
1374         return prot->slab != NULL ? 0 : -ENOBUFS;
1375 }
1376
1377 EXPORT_SYMBOL(sk_alloc_slab);
1378
1379 void sk_free_slab(struct proto *prot)
1380 {
1381         if (prot->slab != NULL) {
1382                 kmem_cache_destroy(prot->slab);
1383                 prot->slab = NULL;
1384         }
1385 }
1386
1387 EXPORT_SYMBOL(sk_free_slab);
1388 EXPORT_SYMBOL(sk_alloc);
1389 EXPORT_SYMBOL(sk_free);
1390 EXPORT_SYMBOL(sk_send_sigurg);
1391 EXPORT_SYMBOL(sock_alloc_send_pskb);
1392 EXPORT_SYMBOL(sock_alloc_send_skb);
1393 EXPORT_SYMBOL(sock_init_data);
1394 EXPORT_SYMBOL(sock_kfree_s);
1395 EXPORT_SYMBOL(sock_kmalloc);
1396 EXPORT_SYMBOL(sock_no_accept);
1397 EXPORT_SYMBOL(sock_no_bind);
1398 EXPORT_SYMBOL(sock_no_connect);
1399 EXPORT_SYMBOL(sock_no_getname);
1400 EXPORT_SYMBOL(sock_no_getsockopt);
1401 EXPORT_SYMBOL(sock_no_ioctl);
1402 EXPORT_SYMBOL(sock_no_listen);
1403 EXPORT_SYMBOL(sock_no_mmap);
1404 EXPORT_SYMBOL(sock_no_poll);
1405 EXPORT_SYMBOL(sock_no_recvmsg);
1406 EXPORT_SYMBOL(sock_no_sendmsg);
1407 EXPORT_SYMBOL(sock_no_sendpage);
1408 EXPORT_SYMBOL(sock_no_setsockopt);
1409 EXPORT_SYMBOL(sock_no_shutdown);
1410 EXPORT_SYMBOL(sock_no_socketpair);
1411 EXPORT_SYMBOL(sock_rfree);
1412 EXPORT_SYMBOL(sock_setsockopt);
1413 EXPORT_SYMBOL(sock_wfree);
1414 EXPORT_SYMBOL(sock_wmalloc);
1415 EXPORT_SYMBOL(sock_i_uid);
1416 EXPORT_SYMBOL(sock_i_ino);
1417 #ifdef CONFIG_SYSCTL
1418 EXPORT_SYMBOL(sysctl_optmem_max);
1419 EXPORT_SYMBOL(sysctl_rmem_max);
1420 EXPORT_SYMBOL(sysctl_wmem_max);
1421 #endif