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 #include <linux/vs_limit.h>
 127
 128 #ifdef CONFIG_INET
 129 #include <net/tcp.h>
 130 #endif
 131
 132 /* Take into consideration the size of the struct sk_buff overhead in the
 133  * determination of these values, since that is non-constant across
 134  * platforms.  This makes socket queueing behavior and performance
 135  * not depend upon such differences.
 136  */
 137 #define _SK_MEM_PACKETS         256
 138 #define _SK_MEM_OVERHEAD        (sizeof(struct sk_buff) + 256)
 139 #define SK_WMEM_MAX             (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
 140 #define SK_RMEM_MAX             (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
 141
 142 /* Run time adjustable parameters. */
 143 __u32 sysctl_wmem_max = SK_WMEM_MAX;
 144 __u32 sysctl_rmem_max = SK_RMEM_MAX;
 145 __u32 sysctl_wmem_default = SK_WMEM_MAX;
 146 __u32 sysctl_rmem_default = SK_RMEM_MAX;
 147
 148 /* Maximal space eaten by iovec or ancilliary data plus some space */
 149 int sysctl_optmem_max = sizeof(unsigned long)*(2*UIO_MAXIOV + 512);
 150
 151 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
 152 {
 153         struct timeval tv;
 154
 155         if (optlen < sizeof(tv))
 156                 return -EINVAL;
 157         if (copy_from_user(&tv, optval, sizeof(tv)))
 158                 return -EFAULT;
 159
 160         *timeo_p = MAX_SCHEDULE_TIMEOUT;
 161         if (tv.tv_sec == 0 && tv.tv_usec == 0)
 162                 return 0;
 163         if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
 164                 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
 165         return 0;
 166 }
 167
 168 static void sock_warn_obsolete_bsdism(const char *name)
 169 {
 170         static int warned;
 171         static char warncomm[TASK_COMM_LEN];
 172         if (strcmp(warncomm, current->comm) && warned < 5) {
 173                 strcpy(warncomm,  current->comm);
 174                 printk(KERN_WARNING "process `%s' is using obsolete "
 175                        "%s SO_BSDCOMPAT\n", warncomm, name);
 176                 warned++;
 177         }
 178 }
 179
 180 static void sock_disable_timestamp(struct sock *sk)
 181 {
 182         if (sock_flag(sk, SOCK_TIMESTAMP)) {
 183                 sock_reset_flag(sk, SOCK_TIMESTAMP);
 184                 net_disable_timestamp();
 185         }
 186 }
 187
 188
 189 /*
 190  *      This is meant for all protocols to use and covers goings on
 191  *      at the socket level. Everything here is generic.
 192  */
 193
 194 int sock_setsockopt(struct socket *sock, int level, int optname,
 195                     char __user *optval, int optlen)
 196 {
 197         struct sock *sk=sock->sk;
 198         struct sk_filter *filter;
 199         int val;
 200         int valbool;
 201         struct linger ling;
 202         int ret = 0;
 203
 204         /*
 205          *      Options without arguments
 206          */
 207
 208 #ifdef SO_DONTLINGER            /* Compatibility item... */
 209         switch (optname) {
 210                 case SO_DONTLINGER:
 211                         sock_reset_flag(sk, SOCK_LINGER);
 212                         return 0;
 213         }
 214 #endif
 215
 216         if(optlen<sizeof(int))
 217                 return(-EINVAL);
 218
 219         if (get_user(val, (int __user *)optval))
 220                 return -EFAULT;
 221
 222         valbool = val?1:0;
 223
 224         lock_sock(sk);
 225
 226         switch(optname)
 227         {
 228                 case SO_DEBUG:
 229                         if(val && !capable(CAP_NET_ADMIN))
 230                         {
 231                                 ret = -EACCES;
 232                         }
 233                         else
 234                                 sk->sk_debug = valbool;
 235                         break;
 236                 case SO_REUSEADDR:
 237                         sk->sk_reuse = valbool;
 238                         break;
 239                 case SO_TYPE:
 240                 case SO_ERROR:
 241                         ret = -ENOPROTOOPT;
 242                         break;
 243                 case SO_DONTROUTE:
 244                         sk->sk_localroute = valbool;
 245                         break;
 246                 case SO_BROADCAST:
 247                         sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
 248                         break;
 249                 case SO_SNDBUF:
 250                         /* Don't error on this BSD doesn't and if you think
 251                            about it this is right. Otherwise apps have to
 252                            play 'guess the biggest size' games. RCVBUF/SNDBUF
 253                            are treated in BSD as hints */
 254
 255                         if (val > sysctl_wmem_max)
 256                                 val = sysctl_wmem_max;
 257
 258                         sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
 259                         if ((val * 2) < SOCK_MIN_SNDBUF)
 260                                 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
 261                         else
 262                                 sk->sk_sndbuf = val * 2;
 263
 264                         /*
 265                          *      Wake up sending tasks if we
 266                          *      upped the value.
 267                          */
 268                         sk->sk_write_space(sk);
 269                         break;
 270
 271                 case SO_RCVBUF:
 272                         /* Don't error on this BSD doesn't and if you think
 273                            about it this is right. Otherwise apps have to
 274                            play 'guess the biggest size' games. RCVBUF/SNDBUF
 275                            are treated in BSD as hints */
 276
 277                         if (val > sysctl_rmem_max)
 278                                 val = sysctl_rmem_max;
 279
 280                         sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
 281                         /* FIXME: is this lower bound the right one? */
 282                         if ((val * 2) < SOCK_MIN_RCVBUF)
 283                                 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
 284                         else
 285                                 sk->sk_rcvbuf = val * 2;
 286                         break;
 287
 288                 case SO_KEEPALIVE:
 289 #ifdef CONFIG_INET
 290                         if (sk->sk_protocol == IPPROTO_TCP)
 291                                 tcp_set_keepalive(sk, valbool);
 292 #endif
 293                         sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
 294                         break;
 295
 296                 case SO_OOBINLINE:
 297                         sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
 298                         break;
 299
 300                 case SO_NO_CHECK:
 301                         sk->sk_no_check = valbool;
 302                         break;
 303
 304                 case SO_PRIORITY:
 305                         if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
 306                                 sk->sk_priority = val;
 307                         else
 308                                 ret = -EPERM;
 309                         break;
 310
 311                 case SO_LINGER:
 312                         if(optlen<sizeof(ling)) {
 313                                 ret = -EINVAL;  /* 1003.1g */
 314                                 break;
 315                         }
 316                         if (copy_from_user(&ling,optval,sizeof(ling))) {
 317                                 ret = -EFAULT;
 318                                 break;
 319                         }
 320                         if (!ling.l_onoff)
 321                                 sock_reset_flag(sk, SOCK_LINGER);
 322                         else {
 323 #if (BITS_PER_LONG == 32)
 324                                 if (ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
 325                                         sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
 326                                 else
 327 #endif
 328                                         sk->sk_lingertime = ling.l_linger * HZ;
 329                                 sock_set_flag(sk, SOCK_LINGER);
 330                         }
 331                         break;
 332
 333                 case SO_BSDCOMPAT:
 334                         sock_warn_obsolete_bsdism("setsockopt");
 335                         break;
 336
 337                 case SO_PASSCRED:
 338                         if (valbool)
 339                                 set_bit(SOCK_PASS_CRED, &sock->flags);
 340                         else
 341                                 clear_bit(SOCK_PASS_CRED, &sock->flags);
 342                         break;
 343
 344                 case SO_TIMESTAMP:
 345                         sk->sk_rcvtstamp = valbool;
 346                         if (valbool)
 347                                 sock_enable_timestamp(sk);
 348                         break;
 349
 350                 case SO_RCVLOWAT:
 351                         if (val < 0)
 352                                 val = INT_MAX;
 353                         sk->sk_rcvlowat = val ? : 1;
 354                         break;
 355
 356                 case SO_RCVTIMEO:
 357                         ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
 358                         break;
 359
 360                 case SO_SNDTIMEO:
 361                         ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
 362                         break;
 363
 364 #ifdef CONFIG_NETDEVICES
 365                 case SO_BINDTODEVICE:
 366                 {
 367                         char devname[IFNAMSIZ];
 368
 369                         /* Sorry... */
 370                         if (!capable(CAP_NET_RAW)) {
 371                                 ret = -EPERM;
 372                                 break;
 373                         }
 374
 375                         /* Bind this socket to a particular device like "eth0",
 376                          * as specified in the passed interface name. If the
 377                          * name is "" or the option length is zero the socket
 378                          * is not bound.
 379                          */
 380
 381                         if (!valbool) {
 382                                 sk->sk_bound_dev_if = 0;
 383                         } else {
 384                                 if (optlen > IFNAMSIZ)
 385                                         optlen = IFNAMSIZ;
 386                                 if (copy_from_user(devname, optval, optlen)) {
 387                                         ret = -EFAULT;
 388                                         break;
 389                                 }
 390
 391                                 /* Remove any cached route for this socket. */
 392                                 sk_dst_reset(sk);
 393
 394                                 if (devname[0] == '\0') {
 395                                         sk->sk_bound_dev_if = 0;
 396                                 } else {
 397                                         struct net_device *dev = dev_get_by_name(devname);
 398                                         if (!dev) {
 399                                                 ret = -ENODEV;
 400                                                 break;
 401                                         }
 402                                         sk->sk_bound_dev_if = dev->ifindex;
 403                                         dev_put(dev);
 404                                 }
 405                         }
 406                         break;
 407                 }
 408 #endif
 409
 410
 411                 case SO_ATTACH_FILTER:
 412                         ret = -EINVAL;
 413                         if (optlen == sizeof(struct sock_fprog)) {
 414                                 struct sock_fprog fprog;
 415
 416                                 ret = -EFAULT;
 417                                 if (copy_from_user(&fprog, optval, sizeof(fprog)))
 418                                         break;
 419
 420                                 ret = sk_attach_filter(&fprog, sk);
 421                         }
 422                         break;
 423
 424                 case SO_DETACH_FILTER:
 425                         spin_lock_bh(&sk->sk_lock.slock);
 426                         filter = sk->sk_filter;
 427                         if (filter) {
 428                                 sk->sk_filter = NULL;
 429                                 spin_unlock_bh(&sk->sk_lock.slock);
 430                                 sk_filter_release(sk, filter);
 431                                 break;
 432                         }
 433                         spin_unlock_bh(&sk->sk_lock.slock);
 434                         ret = -ENONET;
 435                         break;
 436
 437                 /* We implement the SO_SNDLOWAT etc to
 438                    not be settable (1003.1g 5.3) */
 439                 default:
 440                         ret = -ENOPROTOOPT;
 441                         break;
 442         }
 443         release_sock(sk);
 444         return ret;
 445 }
 446
 447
 448 int sock_getsockopt(struct socket *sock, int level, int optname,
 449                     char __user *optval, int __user *optlen)
 450 {
 451         struct sock *sk = sock->sk;
 452
 453         union
 454         {
 455                 int val;
 456                 struct linger ling;
 457                 struct timeval tm;
 458         } v;
 459
 460         unsigned int lv = sizeof(int);
 461         int len;
 462
 463         if(get_user(len,optlen))
 464                 return -EFAULT;
 465         if(len < 0)
 466                 return -EINVAL;
 467
 468         switch(optname)
 469         {
 470                 case SO_DEBUG:
 471                         v.val = sk->sk_debug;
 472                         break;
 473
 474                 case SO_DONTROUTE:
 475                         v.val = sk->sk_localroute;
 476                         break;
 477
 478                 case SO_BROADCAST:
 479                         v.val = !!sock_flag(sk, SOCK_BROADCAST);
 480                         break;
 481
 482                 case SO_SNDBUF:
 483                         v.val = sk->sk_sndbuf;
 484                         break;
 485
 486                 case SO_RCVBUF:
 487                         v.val = sk->sk_rcvbuf;
 488                         break;
 489
 490                 case SO_REUSEADDR:
 491                         v.val = sk->sk_reuse;
 492                         break;
 493
 494                 case SO_KEEPALIVE:
 495                         v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
 496                         break;
 497
 498                 case SO_TYPE:
 499                         v.val = sk->sk_type;
 500                         break;
 501
 502                 case SO_ERROR:
 503                         v.val = -sock_error(sk);
 504                         if(v.val==0)
 505                                 v.val = xchg(&sk->sk_err_soft, 0);
 506                         break;
 507
 508                 case SO_OOBINLINE:
 509                         v.val = !!sock_flag(sk, SOCK_URGINLINE);
 510                         break;
 511
 512                 case SO_NO_CHECK:
 513                         v.val = sk->sk_no_check;
 514                         break;
 515
 516                 case SO_PRIORITY:
 517                         v.val = sk->sk_priority;
 518                         break;
 519
 520                 case SO_LINGER:
 521                         lv              = sizeof(v.ling);
 522                         v.ling.l_onoff  = !!sock_flag(sk, SOCK_LINGER);
 523                         v.ling.l_linger = sk->sk_lingertime / HZ;
 524                         break;
 525
 526                 case SO_BSDCOMPAT:
 527                         sock_warn_obsolete_bsdism("getsockopt");
 528                         break;
 529
 530                 case SO_TIMESTAMP:
 531                         v.val = sk->sk_rcvtstamp;
 532                         break;
 533
 534                 case SO_RCVTIMEO:
 535                         lv=sizeof(struct timeval);
 536                         if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
 537                                 v.tm.tv_sec = 0;
 538                                 v.tm.tv_usec = 0;
 539                         } else {
 540                                 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
 541                                 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
 542                         }
 543                         break;
 544
 545                 case SO_SNDTIMEO:
 546                         lv=sizeof(struct timeval);
 547                         if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
 548                                 v.tm.tv_sec = 0;
 549                                 v.tm.tv_usec = 0;
 550                         } else {
 551                                 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
 552                                 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
 553                         }
 554                         break;
 555
 556                 case SO_RCVLOWAT:
 557                         v.val = sk->sk_rcvlowat;
 558                         break;
 559
 560                 case SO_SNDLOWAT:
 561                         v.val=1;
 562                         break;
 563
 564                 case SO_PASSCRED:
 565                         v.val = test_bit(SOCK_PASS_CRED, &sock->flags)?1:0;
 566                         break;
 567
 568                 case SO_PEERCRED:
 569                         if (len > sizeof(sk->sk_peercred))
 570                                 len = sizeof(sk->sk_peercred);
 571                         if (copy_to_user(optval, &sk->sk_peercred, len))
 572                                 return -EFAULT;
 573                         goto lenout;
 574
 575                 case SO_PEERNAME:
 576                 {
 577                         char address[128];
 578
 579                         if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
 580                                 return -ENOTCONN;
 581                         if (lv < len)
 582                                 return -EINVAL;
 583                         if (copy_to_user(optval, address, len))
 584                                 return -EFAULT;
 585                         goto lenout;
 586                 }
 587
 588                 /* Dubious BSD thing... Probably nobody even uses it, but
 589                  * the UNIX standard wants it for whatever reason... -DaveM
 590                  */
 591                 case SO_ACCEPTCONN:
 592                         v.val = sk->sk_state == TCP_LISTEN;
 593                         break;
 594
 595                 case SO_PEERSEC:
 596                         return security_socket_getpeersec(sock, optval, optlen, len);
 597
 598                 default:
 599                         return(-ENOPROTOOPT);
 600         }
 601         if (len > lv)
 602                 len = lv;
 603         if (copy_to_user(optval, &v, len))
 604                 return -EFAULT;
 605 lenout:
 606         if (put_user(len, optlen))
 607                 return -EFAULT;
 608         return 0;
 609 }
 610
 611 static kmem_cache_t *sk_cachep;
 612
 613 /**
 614  *      sk_alloc - All socket objects are allocated here
 615  *      @family - protocol family
 616  *      @priority - for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
 617  *      @zero_it - zeroes the allocated sock
 618  *      @slab - alternate slab
 619  *
 620  *      All socket objects are allocated here. If @zero_it is non-zero
 621  *      it should have the size of the are to be zeroed, because the
 622  *      private slabcaches have different sizes of the generic struct sock.
 623  *      1 has been kept as a way to say sizeof(struct sock).
 624  */
 625 struct sock *sk_alloc(int family, int priority, int zero_it, kmem_cache_t *slab)
 626 {
 627         struct sock *sk = NULL;
 628
 629         if (!slab)
 630                 slab = sk_cachep;
 631         sk = kmem_cache_alloc(slab, priority);
 632         if (sk) {
 633                 if (zero_it) {
 634                         memset(sk, 0,
 635                                zero_it == 1 ? sizeof(struct sock) : zero_it);
 636                         sk->sk_family = family;
 637                         sock_lock_init(sk);
 638                 }
 639                 sk->sk_slab = slab;
 640                 sock_vx_init(sk);
 641                 sock_nx_init(sk);
 642
 643                 if (security_sk_alloc(sk, family, priority)) {
 644                         kmem_cache_free(slab, sk);
 645                         sk = NULL;
 646                 }
 647         }
 648         return sk;
 649 }
 650
 651 void sk_free(struct sock *sk)
 652 {
 653         struct sk_filter *filter;
 654         struct module *owner = sk->sk_owner;
 655
 656         if (sk->sk_destruct)
 657                 sk->sk_destruct(sk);
 658
 659         filter = sk->sk_filter;
 660         if (filter) {
 661                 sk_filter_release(sk, filter);
 662                 sk->sk_filter = NULL;
 663         }
 664
 665         sock_disable_timestamp(sk);
 666
 667         if (atomic_read(&sk->sk_omem_alloc))
 668                 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
 669                        __FUNCTION__, atomic_read(&sk->sk_omem_alloc));
 670
 671         security_sk_free(sk);
 672         vx_sock_dec(sk);
 673         // BUG_ON(sk->sk_vx_info);
 674         clr_vx_info(&sk->sk_vx_info);
 675         sk->sk_xid = -1;
 676         // BUG_ON(sk->sk_nx_info);
 677         clr_nx_info(&sk->sk_nx_info);
 678         sk->sk_nid = -1;
 679         kmem_cache_free(sk->sk_slab, sk);
 680         module_put(owner);
 681 }
 682
 683 void __init sk_init(void)
 684 {
 685         sk_cachep = kmem_cache_create("sock", sizeof(struct sock), 0,
 686                                       SLAB_HWCACHE_ALIGN, NULL, NULL);
 687         if (!sk_cachep)
 688                 printk(KERN_CRIT "sk_init: Cannot create sock SLAB cache!");
 689
 690         if (num_physpages <= 4096) {
 691                 sysctl_wmem_max = 32767;
 692                 sysctl_rmem_max = 32767;
 693                 sysctl_wmem_default = 32767;
 694                 sysctl_rmem_default = 32767;
 695         } else if (num_physpages >= 131072) {
 696                 sysctl_wmem_max = 131071;
 697                 sysctl_rmem_max = 131071;
 698         }
 699 }
 700
 701 /*
 702  *      Simple resource managers for sockets.
 703  */
 704
 705
 706 /*
 707  * Write buffer destructor automatically called from kfree_skb.
 708  */
 709 void sock_wfree(struct sk_buff *skb)
 710 {
 711         struct sock *sk = skb->sk;
 712
 713         /* In case it might be waiting for more memory. */
 714         atomic_sub(skb->truesize, &sk->sk_wmem_alloc);
 715         if (!sk->sk_use_write_queue)
 716                 sk->sk_write_space(sk);
 717         sock_put(sk);
 718 }
 719
 720 /*
 721  * Read buffer destructor automatically called from kfree_skb.
 722  */
 723 void sock_rfree(struct sk_buff *skb)
 724 {
 725         struct sock *sk = skb->sk;
 726
 727         atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
 728 }
 729
 730
 731 int sock_i_uid(struct sock *sk)
 732 {
 733         int uid;
 734
 735         read_lock(&sk->sk_callback_lock);
 736         uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
 737         read_unlock(&sk->sk_callback_lock);
 738         return uid;
 739 }
 740
 741 unsigned long sock_i_ino(struct sock *sk)
 742 {
 743         unsigned long ino;
 744
 745         read_lock(&sk->sk_callback_lock);
 746         ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
 747         read_unlock(&sk->sk_callback_lock);
 748         return ino;
 749 }
 750
 751 /*
 752  * Allocate a skb from the socket's send buffer.
 753  */
 754 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, int priority)
 755 {
 756         if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
 757                 struct sk_buff * skb = alloc_skb(size, priority);
 758                 if (skb) {
 759                         skb_set_owner_w(skb, sk);
 760                         return skb;
 761                 }
 762         }
 763         return NULL;
 764 }
 765
 766 /*
 767  * Allocate a skb from the socket's receive buffer.
 768  */
 769 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force, int priority)
 770 {
 771         if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
 772                 struct sk_buff *skb = alloc_skb(size, priority);
 773                 if (skb) {
 774                         skb_set_owner_r(skb, sk);
 775                         return skb;
 776                 }
 777         }
 778         return NULL;
 779 }
 780
 781 /*
 782  * Allocate a memory block from the socket's option memory buffer.
 783  */
 784 void *sock_kmalloc(struct sock *sk, int size, int priority)
 785 {
 786         if ((unsigned)size <= sysctl_optmem_max &&
 787             atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
 788                 void *mem;
 789                 /* First do the add, to avoid the race if kmalloc
 790                  * might sleep.
 791                  */
 792                 atomic_add(size, &sk->sk_omem_alloc);
 793                 mem = kmalloc(size, priority);
 794                 if (mem)
 795                         return mem;
 796                 atomic_sub(size, &sk->sk_omem_alloc);
 797         }
 798         return NULL;
 799 }
 800
 801 /*
 802  * Free an option memory block.
 803  */
 804 void sock_kfree_s(struct sock *sk, void *mem, int size)
 805 {
 806         kfree(mem);
 807         atomic_sub(size, &sk->sk_omem_alloc);
 808 }
 809
 810 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
 811    I think, these locks should be removed for datagram sockets.
 812  */
 813 static long sock_wait_for_wmem(struct sock * sk, long timeo)
 814 {
 815         DEFINE_WAIT(wait);
 816
 817         clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
 818         for (;;) {
 819                 if (!timeo)
 820                         break;
 821                 if (signal_pending(current))
 822                         break;
 823                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
 824                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
 825                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
 826                         break;
 827                 if (sk->sk_shutdown & SEND_SHUTDOWN)
 828                         break;
 829                 if (sk->sk_err)
 830                         break;
 831                 timeo = schedule_timeout(timeo);
 832         }
 833         finish_wait(sk->sk_sleep, &wait);
 834         return timeo;
 835 }
 836
 837
 838 /*
 839  *      Generic send/receive buffer handlers
 840  */
 841
 842 static struct sk_buff *sock_alloc_send_pskb(struct sock *sk,
 843                                             unsigned long header_len,
 844                                             unsigned long data_len,
 845                                             int noblock, int *errcode)
 846 {
 847         struct sk_buff *skb;
 848         unsigned int gfp_mask;
 849         long timeo;
 850         int err;
 851
 852         gfp_mask = sk->sk_allocation;
 853         if (gfp_mask & __GFP_WAIT)
 854                 gfp_mask |= __GFP_REPEAT;
 855
 856         timeo = sock_sndtimeo(sk, noblock);
 857         while (1) {
 858                 err = sock_error(sk);
 859                 if (err != 0)
 860                         goto failure;
 861
 862                 err = -EPIPE;
 863                 if (sk->sk_shutdown & SEND_SHUTDOWN)
 864                         goto failure;
 865
 866                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
 867                         skb = alloc_skb(header_len, sk->sk_allocation);
 868                         if (skb) {
 869                                 int npages;
 870                                 int i;
 871
 872                                 /* No pages, we're done... */
 873                                 if (!data_len)
 874                                         break;
 875
 876                                 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
 877                                 skb->truesize += data_len;
 878                                 skb_shinfo(skb)->nr_frags = npages;
 879                                 for (i = 0; i < npages; i++) {
 880                                         struct page *page;
 881                                         skb_frag_t *frag;
 882
 883                                         page = alloc_pages(sk->sk_allocation, 0);
 884                                         if (!page) {
 885                                                 err = -ENOBUFS;
 886                                                 skb_shinfo(skb)->nr_frags = i;
 887                                                 kfree_skb(skb);
 888                                                 goto failure;
 889                                         }
 890
 891                                         frag = &skb_shinfo(skb)->frags[i];
 892                                         frag->page = page;
 893                                         frag->page_offset = 0;
 894                                         frag->size = (data_len >= PAGE_SIZE ?
 895                                                       PAGE_SIZE :
 896                                                       data_len);
 897                                         data_len -= PAGE_SIZE;
 898                                 }
 899
 900                                 /* Full success... */
 901                                 break;
 902                         }
 903                         err = -ENOBUFS;
 904                         goto failure;
 905                 }
 906                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
 907                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
 908                 err = -EAGAIN;
 909                 if (!timeo)
 910                         goto failure;
 911                 if (signal_pending(current))
 912                         goto interrupted;
 913                 timeo = sock_wait_for_wmem(sk, timeo);
 914         }
 915
 916         skb_set_owner_w(skb, sk);
 917         return skb;
 918
 919 interrupted:
 920         err = sock_intr_errno(timeo);
 921 failure:
 922         *errcode = err;
 923         return NULL;
 924 }
 925
 926 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
 927                                     int noblock, int *errcode)
 928 {
 929         return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
 930 }
 931
 932 static void __lock_sock(struct sock *sk)
 933 {
 934         DEFINE_WAIT(wait);
 935
 936         for(;;) {
 937                 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
 938                                         TASK_UNINTERRUPTIBLE);
 939                 spin_unlock_bh(&sk->sk_lock.slock);
 940                 schedule();
 941                 spin_lock_bh(&sk->sk_lock.slock);
 942                 if(!sock_owned_by_user(sk))
 943                         break;
 944         }
 945         finish_wait(&sk->sk_lock.wq, &wait);
 946 }
 947
 948 static void __release_sock(struct sock *sk)
 949 {
 950         struct sk_buff *skb = sk->sk_backlog.head;
 951
 952         do {
 953                 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
 954                 bh_unlock_sock(sk);
 955
 956                 do {
 957                         struct sk_buff *next = skb->next;
 958
 959                         skb->next = NULL;
 960                         sk->sk_backlog_rcv(sk, skb);
 961
 962                         /*
 963                          * We are in process context here with softirqs
 964                          * disabled, use cond_resched_softirq() to preempt.
 965                          * This is safe to do because we've taken the backlog
 966                          * queue private:
 967                          */
 968                         cond_resched_softirq();
 969
 970                         skb = next;
 971                 } while (skb != NULL);
 972
 973                 bh_lock_sock(sk);
 974         } while((skb = sk->sk_backlog.head) != NULL);
 975 }
 976
 977 /**
 978  * sk_wait_data - wait for data to arrive at sk_receive_queue
 979  * sk - sock to wait on
 980  * timeo - for how long
 981  *
 982  * Now socket state including sk->sk_err is changed only under lock,
 983  * hence we may omit checks after joining wait queue.
 984  * We check receive queue before schedule() only as optimization;
 985  * it is very likely that release_sock() added new data.
 986  */
 987 int sk_wait_data(struct sock *sk, long *timeo)
 988 {
 989         int rc;
 990         DEFINE_WAIT(wait);
 991
 992         prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
 993         set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
 994         rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
 995         clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
 996         finish_wait(sk->sk_sleep, &wait);
 997         return rc;
 998 }
 999
1000 EXPORT_SYMBOL(sk_wait_data);
1001
1002 /*
1003  * Set of default routines for initialising struct proto_ops when
1004  * the protocol does not support a particular function. In certain
1005  * cases where it makes no sense for a protocol to have a "do nothing"
1006  * function, some default processing is provided.
1007  */
1008
1009 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1010 {
1011         return -EOPNOTSUPP;
1012 }
1013
1014 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1015                     int len, int flags)
1016 {
1017         return -EOPNOTSUPP;
1018 }
1019
1020 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1021 {
1022         return -EOPNOTSUPP;
1023 }
1024
1025 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1026 {
1027         return -EOPNOTSUPP;
1028 }
1029
1030 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1031                     int *len, int peer)
1032 {
1033         return -EOPNOTSUPP;
1034 }
1035
1036 unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt)
1037 {
1038         return 0;
1039 }
1040
1041 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1042 {
1043         return -EOPNOTSUPP;
1044 }
1045
1046 int sock_no_listen(struct socket *sock, int backlog)
1047 {
1048         return -EOPNOTSUPP;
1049 }
1050
1051 int sock_no_shutdown(struct socket *sock, int how)
1052 {
1053         return -EOPNOTSUPP;
1054 }
1055
1056 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1057                     char __user *optval, int optlen)
1058 {
1059         return -EOPNOTSUPP;
1060 }
1061
1062 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1063                     char __user *optval, int __user *optlen)
1064 {
1065         return -EOPNOTSUPP;
1066 }
1067
1068 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1069                     size_t len)
1070 {
1071         return -EOPNOTSUPP;
1072 }
1073
1074 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1075                     size_t len, int flags)
1076 {
1077         return -EOPNOTSUPP;
1078 }
1079
1080 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1081 {
1082         /* Mirror missing mmap method error code */
1083         return -ENODEV;
1084 }
1085
1086 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1087 {
1088         ssize_t res;
1089         struct msghdr msg = {.msg_flags = flags};
1090         struct kvec iov;
1091         char *kaddr = kmap(page);
1092         iov.iov_base = kaddr + offset;
1093         iov.iov_len = size;
1094         res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1095         kunmap(page);
1096         return res;
1097 }
1098
1099 /*
1100  *      Default Socket Callbacks
1101  */
1102
1103 static void sock_def_wakeup(struct sock *sk)
1104 {
1105         read_lock(&sk->sk_callback_lock);
1106         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1107                 wake_up_interruptible_all(sk->sk_sleep);
1108         read_unlock(&sk->sk_callback_lock);
1109 }
1110
1111 static void sock_def_error_report(struct sock *sk)
1112 {
1113         read_lock(&sk->sk_callback_lock);
1114         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1115                 wake_up_interruptible(sk->sk_sleep);
1116         sk_wake_async(sk,0,POLL_ERR);
1117         read_unlock(&sk->sk_callback_lock);
1118 }
1119
1120 static void sock_def_readable(struct sock *sk, int len)
1121 {
1122         read_lock(&sk->sk_callback_lock);
1123         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1124                 wake_up_interruptible(sk->sk_sleep);
1125         sk_wake_async(sk,1,POLL_IN);
1126         read_unlock(&sk->sk_callback_lock);
1127 }
1128
1129 static void sock_def_write_space(struct sock *sk)
1130 {
1131         read_lock(&sk->sk_callback_lock);
1132
1133         /* Do not wake up a writer until he can make "significant"
1134          * progress.  --DaveM
1135          */
1136         if((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1137                 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1138                         wake_up_interruptible(sk->sk_sleep);
1139
1140                 /* Should agree with poll, otherwise some programs break */
1141                 if (sock_writeable(sk))
1142                         sk_wake_async(sk, 2, POLL_OUT);
1143         }
1144
1145         read_unlock(&sk->sk_callback_lock);
1146 }
1147
1148 static void sock_def_destruct(struct sock *sk)
1149 {
1150         if (sk->sk_protinfo)
1151                 kfree(sk->sk_protinfo);
1152 }
1153
1154 void sk_send_sigurg(struct sock *sk)
1155 {
1156         if (sk->sk_socket && sk->sk_socket->file)
1157                 if (send_sigurg(&sk->sk_socket->file->f_owner))
1158                         sk_wake_async(sk, 3, POLL_PRI);
1159 }
1160
1161 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1162                     unsigned long expires)
1163 {
1164         if (!mod_timer(timer, expires))
1165                 sock_hold(sk);
1166 }
1167
1168 EXPORT_SYMBOL(sk_reset_timer);
1169
1170 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1171 {
1172         if (timer_pending(timer) && del_timer(timer))
1173                 __sock_put(sk);
1174 }
1175
1176 EXPORT_SYMBOL(sk_stop_timer);
1177
1178 void sock_init_data(struct socket *sock, struct sock *sk)
1179 {
1180         skb_queue_head_init(&sk->sk_receive_queue);
1181         skb_queue_head_init(&sk->sk_write_queue);
1182         skb_queue_head_init(&sk->sk_error_queue);
1183
1184         sk->sk_send_head        =       NULL;
1185
1186         init_timer(&sk->sk_timer);
1187
1188         sk->sk_allocation       =       GFP_KERNEL;
1189         sk->sk_rcvbuf           =       sysctl_rmem_default;
1190         sk->sk_sndbuf           =       sysctl_wmem_default;
1191         sk->sk_state            =       TCP_CLOSE;
1192         sk->sk_zapped           =       1;
1193         sk->sk_socket           =       sock;
1194
1195         if(sock)
1196         {
1197                 sk->sk_type     =       sock->type;
1198                 sk->sk_sleep    =       &sock->wait;
1199                 sock->sk        =       sk;
1200         } else
1201                 sk->sk_sleep    =       NULL;
1202
1203         rwlock_init(&sk->sk_dst_lock);
1204         rwlock_init(&sk->sk_callback_lock);
1205
1206         sk->sk_state_change     =       sock_def_wakeup;
1207         sk->sk_data_ready       =       sock_def_readable;
1208         sk->sk_write_space      =       sock_def_write_space;
1209         sk->sk_error_report     =       sock_def_error_report;
1210         sk->sk_destruct         =       sock_def_destruct;
1211
1212         sk->sk_sndmsg_page      =       NULL;
1213         sk->sk_sndmsg_off       =       0;
1214
1215         sk->sk_peercred.pid     =       0;
1216         sk->sk_peercred.uid     =       -1;
1217         sk->sk_peercred.gid     =       -1;
1218         sk->sk_write_pending    =       0;
1219         sk->sk_rcvlowat         =       1;
1220         sk->sk_rcvtimeo         =       MAX_SCHEDULE_TIMEOUT;
1221         sk->sk_sndtimeo         =       MAX_SCHEDULE_TIMEOUT;
1222         sk->sk_owner            =       NULL;
1223
1224         sk->sk_stamp.tv_sec     = -1L;
1225         sk->sk_stamp.tv_usec    = -1L;
1226
1227         set_vx_info(&sk->sk_vx_info, current->vx_info);
1228         sk->sk_xid = vx_current_xid();
1229         vx_sock_inc(sk);
1230         set_nx_info(&sk->sk_nx_info, current->nx_info);
1231         sk->sk_nid = nx_current_nid();
1232         atomic_set(&sk->sk_refcnt, 1);
1233 }
1234
1235 void fastcall lock_sock(struct sock *sk)
1236 {
1237         might_sleep();
1238         spin_lock_bh(&(sk->sk_lock.slock));
1239         if (sk->sk_lock.owner)
1240                 __lock_sock(sk);
1241         sk->sk_lock.owner = (void *)1;
1242         spin_unlock_bh(&(sk->sk_lock.slock));
1243 }
1244
1245 EXPORT_SYMBOL(lock_sock);
1246
1247 void fastcall release_sock(struct sock *sk)
1248 {
1249         spin_lock_bh(&(sk->sk_lock.slock));
1250         if (sk->sk_backlog.tail)
1251                 __release_sock(sk);
1252         sk->sk_lock.owner = NULL;
1253         if (waitqueue_active(&(sk->sk_lock.wq)))
1254                 wake_up(&(sk->sk_lock.wq));
1255         spin_unlock_bh(&(sk->sk_lock.slock));
1256 }
1257 EXPORT_SYMBOL(release_sock);
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                 net_enable_timestamp();
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
1389 EXPORT_SYMBOL(sk_alloc);
1390 EXPORT_SYMBOL(sk_free);
1391 EXPORT_SYMBOL(sk_send_sigurg);
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