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