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
  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/module.h>
 101 #include <linux/proc_fs.h>
 102 #include <linux/seq_file.h>
 103 #include <linux/sched.h>
 104 #include <linux/timer.h>
 105 #include <linux/string.h>
 106 #include <linux/sockios.h>
 107 #include <linux/net.h>
 108 #include <linux/mm.h>
 109 #include <linux/slab.h>
 110 #include <linux/interrupt.h>
 111 #include <linux/poll.h>
 112 #include <linux/tcp.h>
 113 #include <linux/init.h>
 114
 115 #include <asm/uaccess.h>
 116 #include <asm/system.h>
 117
 118 #include <linux/netdevice.h>
 119 #include <net/protocol.h>
 120 #include <linux/skbuff.h>
 121 #include <net/request_sock.h>
 122 #include <net/sock.h>
 123 #include <net/xfrm.h>
 124 #include <linux/ipsec.h>
 125
 126 #include <linux/filter.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 if (valbool)
 234                                 sock_set_flag(sk, SOCK_DBG);
 235                         else
 236                                 sock_reset_flag(sk, SOCK_DBG);
 237                         break;
 238                 case SO_REUSEADDR:
 239                         sk->sk_reuse = valbool;
 240                         break;
 241                 case SO_TYPE:
 242                 case SO_ERROR:
 243                         ret = -ENOPROTOOPT;
 244                         break;
 245                 case SO_DONTROUTE:
 246                         if (valbool)
 247                                 sock_set_flag(sk, SOCK_LOCALROUTE);
 248                         else
 249                                 sock_reset_flag(sk, SOCK_LOCALROUTE);
 250                         break;
 251                 case SO_BROADCAST:
 252                         sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
 253                         break;
 254                 case SO_SNDBUF:
 255                         /* Don't error on this BSD doesn't and if you think
 256                            about it this is right. Otherwise apps have to
 257                            play 'guess the biggest size' games. RCVBUF/SNDBUF
 258                            are treated in BSD as hints */
 259
 260                         if (val > sysctl_wmem_max)
 261                                 val = sysctl_wmem_max;
 262
 263                         sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
 264                         if ((val * 2) < SOCK_MIN_SNDBUF)
 265                                 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
 266                         else
 267                                 sk->sk_sndbuf = val * 2;
 268
 269                         /*
 270                          *      Wake up sending tasks if we
 271                          *      upped the value.
 272                          */
 273                         sk->sk_write_space(sk);
 274                         break;
 275
 276                 case SO_RCVBUF:
 277                         /* Don't error on this BSD doesn't and if you think
 278                            about it this is right. Otherwise apps have to
 279                            play 'guess the biggest size' games. RCVBUF/SNDBUF
 280                            are treated in BSD as hints */
 281
 282                         if (val > sysctl_rmem_max)
 283                                 val = sysctl_rmem_max;
 284
 285                         sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
 286                         /* FIXME: is this lower bound the right one? */
 287                         if ((val * 2) < SOCK_MIN_RCVBUF)
 288                                 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
 289                         else
 290                                 sk->sk_rcvbuf = val * 2;
 291                         break;
 292
 293                 case SO_KEEPALIVE:
 294 #ifdef CONFIG_INET
 295                         if (sk->sk_protocol == IPPROTO_TCP)
 296                                 tcp_set_keepalive(sk, valbool);
 297 #endif
 298                         sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
 299                         break;
 300
 301                 case SO_OOBINLINE:
 302                         sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
 303                         break;
 304
 305                 case SO_NO_CHECK:
 306                         sk->sk_no_check = valbool;
 307                         break;
 308
 309                 case SO_PRIORITY:
 310                         if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
 311                                 sk->sk_priority = val;
 312                         else
 313                                 ret = -EPERM;
 314                         break;
 315
 316                 case SO_LINGER:
 317                         if(optlen<sizeof(ling)) {
 318                                 ret = -EINVAL;  /* 1003.1g */
 319                                 break;
 320                         }
 321                         if (copy_from_user(&ling,optval,sizeof(ling))) {
 322                                 ret = -EFAULT;
 323                                 break;
 324                         }
 325                         if (!ling.l_onoff)
 326                                 sock_reset_flag(sk, SOCK_LINGER);
 327                         else {
 328 #if (BITS_PER_LONG == 32)
 329                                 if (ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
 330                                         sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
 331                                 else
 332 #endif
 333                                         sk->sk_lingertime = ling.l_linger * HZ;
 334                                 sock_set_flag(sk, SOCK_LINGER);
 335                         }
 336                         break;
 337
 338                 case SO_BSDCOMPAT:
 339                         sock_warn_obsolete_bsdism("setsockopt");
 340                         break;
 341
 342                 case SO_PASSCRED:
 343                         if (valbool)
 344                                 set_bit(SOCK_PASSCRED, &sock->flags);
 345                         else
 346                                 clear_bit(SOCK_PASSCRED, &sock->flags);
 347                         break;
 348
 349                 case SO_TIMESTAMP:
 350                         if (valbool)  {
 351                                 sock_set_flag(sk, SOCK_RCVTSTAMP);
 352                                 sock_enable_timestamp(sk);
 353                         } else
 354                                 sock_reset_flag(sk, SOCK_RCVTSTAMP);
 355                         break;
 356
 357                 case SO_RCVLOWAT:
 358                         if (val < 0)
 359                                 val = INT_MAX;
 360                         sk->sk_rcvlowat = val ? : 1;
 361                         break;
 362
 363                 case SO_RCVTIMEO:
 364                         ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
 365                         break;
 366
 367                 case SO_SNDTIMEO:
 368                         ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
 369                         break;
 370
 371 #ifdef CONFIG_NETDEVICES
 372                 case SO_BINDTODEVICE:
 373                 {
 374                         char devname[IFNAMSIZ];
 375
 376                         /* Sorry... */
 377                         if (!capable(CAP_NET_RAW)) {
 378                                 ret = -EPERM;
 379                                 break;
 380                         }
 381
 382                         /* Bind this socket to a particular device like "eth0",
 383                          * as specified in the passed interface name. If the
 384                          * name is "" or the option length is zero the socket
 385                          * is not bound.
 386                          */
 387
 388                         if (!valbool) {
 389                                 sk->sk_bound_dev_if = 0;
 390                         } else {
 391                                 if (optlen > IFNAMSIZ)
 392                                         optlen = IFNAMSIZ;
 393                                 if (copy_from_user(devname, optval, optlen)) {
 394                                         ret = -EFAULT;
 395                                         break;
 396                                 }
 397
 398                                 /* Remove any cached route for this socket. */
 399                                 sk_dst_reset(sk);
 400
 401                                 if (devname[0] == '\0') {
 402                                         sk->sk_bound_dev_if = 0;
 403                                 } else {
 404                                         struct net_device *dev = dev_get_by_name(devname);
 405                                         if (!dev) {
 406                                                 ret = -ENODEV;
 407                                                 break;
 408                                         }
 409                                         sk->sk_bound_dev_if = dev->ifindex;
 410                                         dev_put(dev);
 411                                 }
 412                         }
 413                         break;
 414                 }
 415 #endif
 416
 417
 418                 case SO_ATTACH_FILTER:
 419                         ret = -EINVAL;
 420                         if (optlen == sizeof(struct sock_fprog)) {
 421                                 struct sock_fprog fprog;
 422
 423                                 ret = -EFAULT;
 424                                 if (copy_from_user(&fprog, optval, sizeof(fprog)))
 425                                         break;
 426
 427                                 ret = sk_attach_filter(&fprog, sk);
 428                         }
 429                         break;
 430
 431                 case SO_DETACH_FILTER:
 432                         spin_lock_bh(&sk->sk_lock.slock);
 433                         filter = sk->sk_filter;
 434                         if (filter) {
 435                                 sk->sk_filter = NULL;
 436                                 spin_unlock_bh(&sk->sk_lock.slock);
 437                                 sk_filter_release(sk, filter);
 438                                 break;
 439                         }
 440                         spin_unlock_bh(&sk->sk_lock.slock);
 441                         ret = -ENONET;
 442                         break;
 443
 444                 /* We implement the SO_SNDLOWAT etc to
 445                    not be settable (1003.1g 5.3) */
 446                 default:
 447                         ret = -ENOPROTOOPT;
 448                         break;
 449         }
 450         release_sock(sk);
 451         return ret;
 452 }
 453
 454
 455 int sock_getsockopt(struct socket *sock, int level, int optname,
 456                     char __user *optval, int __user *optlen)
 457 {
 458         struct sock *sk = sock->sk;
 459
 460         union
 461         {
 462                 int val;
 463                 struct linger ling;
 464                 struct timeval tm;
 465         } v;
 466
 467         unsigned int lv = sizeof(int);
 468         int len;
 469
 470         if(get_user(len,optlen))
 471                 return -EFAULT;
 472         if(len < 0)
 473                 return -EINVAL;
 474
 475         switch(optname)
 476         {
 477                 case SO_DEBUG:
 478                         v.val = sock_flag(sk, SOCK_DBG);
 479                         break;
 480
 481                 case SO_DONTROUTE:
 482                         v.val = sock_flag(sk, SOCK_LOCALROUTE);
 483                         break;
 484
 485                 case SO_BROADCAST:
 486                         v.val = !!sock_flag(sk, SOCK_BROADCAST);
 487                         break;
 488
 489                 case SO_SNDBUF:
 490                         v.val = sk->sk_sndbuf;
 491                         break;
 492
 493                 case SO_RCVBUF:
 494                         v.val = sk->sk_rcvbuf;
 495                         break;
 496
 497                 case SO_REUSEADDR:
 498                         v.val = sk->sk_reuse;
 499                         break;
 500
 501                 case SO_KEEPALIVE:
 502                         v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
 503                         break;
 504
 505                 case SO_TYPE:
 506                         v.val = sk->sk_type;
 507                         break;
 508
 509                 case SO_ERROR:
 510                         v.val = -sock_error(sk);
 511                         if(v.val==0)
 512                                 v.val = xchg(&sk->sk_err_soft, 0);
 513                         break;
 514
 515                 case SO_OOBINLINE:
 516                         v.val = !!sock_flag(sk, SOCK_URGINLINE);
 517                         break;
 518
 519                 case SO_NO_CHECK:
 520                         v.val = sk->sk_no_check;
 521                         break;
 522
 523                 case SO_PRIORITY:
 524                         v.val = sk->sk_priority;
 525                         break;
 526
 527                 case SO_LINGER:
 528                         lv              = sizeof(v.ling);
 529                         v.ling.l_onoff  = !!sock_flag(sk, SOCK_LINGER);
 530                         v.ling.l_linger = sk->sk_lingertime / HZ;
 531                         break;
 532
 533                 case SO_BSDCOMPAT:
 534                         sock_warn_obsolete_bsdism("getsockopt");
 535                         break;
 536
 537                 case SO_TIMESTAMP:
 538                         v.val = sock_flag(sk, SOCK_RCVTSTAMP);
 539                         break;
 540
 541                 case SO_RCVTIMEO:
 542                         lv=sizeof(struct timeval);
 543                         if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
 544                                 v.tm.tv_sec = 0;
 545                                 v.tm.tv_usec = 0;
 546                         } else {
 547                                 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
 548                                 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
 549                         }
 550                         break;
 551
 552                 case SO_SNDTIMEO:
 553                         lv=sizeof(struct timeval);
 554                         if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
 555                                 v.tm.tv_sec = 0;
 556                                 v.tm.tv_usec = 0;
 557                         } else {
 558                                 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
 559                                 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
 560                         }
 561                         break;
 562
 563                 case SO_RCVLOWAT:
 564                         v.val = sk->sk_rcvlowat;
 565                         break;
 566
 567                 case SO_SNDLOWAT:
 568                         v.val=1;
 569                         break;
 570
 571                 case SO_PASSCRED:
 572                         v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
 573                         break;
 574
 575                 case SO_PEERCRED:
 576                         if (len > sizeof(sk->sk_peercred))
 577                                 len = sizeof(sk->sk_peercred);
 578                         if (copy_to_user(optval, &sk->sk_peercred, len))
 579                                 return -EFAULT;
 580                         goto lenout;
 581
 582                 case SO_PEERNAME:
 583                 {
 584                         char address[128];
 585
 586                         if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
 587                                 return -ENOTCONN;
 588                         if (lv < len)
 589                                 return -EINVAL;
 590                         if (copy_to_user(optval, address, len))
 591                                 return -EFAULT;
 592                         goto lenout;
 593                 }
 594
 595                 /* Dubious BSD thing... Probably nobody even uses it, but
 596                  * the UNIX standard wants it for whatever reason... -DaveM
 597                  */
 598                 case SO_ACCEPTCONN:
 599                         v.val = sk->sk_state == TCP_LISTEN;
 600                         break;
 601
 602                 case SO_PEERSEC:
 603                         return security_socket_getpeersec(sock, optval, optlen, len);
 604
 605                 default:
 606                         return(-ENOPROTOOPT);
 607         }
 608         if (len > lv)
 609                 len = lv;
 610         if (copy_to_user(optval, &v, len))
 611                 return -EFAULT;
 612 lenout:
 613         if (put_user(len, optlen))
 614                 return -EFAULT;
 615         return 0;
 616 }
 617
 618 /**
 619  *      sk_alloc - All socket objects are allocated here
 620  *      @family: protocol family
 621  *      @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
 622  *      @prot: struct proto associated with this new sock instance
 623  *      @zero_it: if we should zero the newly allocated sock
 624  */
 625 struct sock *sk_alloc(int family, unsigned int __nocast priority,
 626                       struct proto *prot, int zero_it)
 627 {
 628         struct sock *sk = NULL;
 629         kmem_cache_t *slab = prot->slab;
 630
 631         if (slab != NULL)
 632                 sk = kmem_cache_alloc(slab, priority);
 633         else
 634                 sk = kmalloc(prot->obj_size, priority);
 635
 636         if (sk) {
 637                 if (zero_it) {
 638                         memset(sk, 0, prot->obj_size);
 639                         sk->sk_family = family;
 640                         /*
 641                          * See comment in struct sock definition to understand
 642                          * why we need sk_prot_creator -acme
 643                          */
 644                         sk->sk_prot = sk->sk_prot_creator = prot;
 645                         sock_lock_init(sk);
 646                 }
 647
 648                 if (security_sk_alloc(sk, family, priority)) {
 649                         if (slab != NULL)
 650                                 kmem_cache_free(slab, sk);
 651                         else
 652                                 kfree(sk);
 653                         sk = NULL;
 654                 } else
 655                         __module_get(prot->owner);
 656         }
 657         return sk;
 658 }
 659
 660 void sk_free(struct sock *sk)
 661 {
 662         struct sk_filter *filter;
 663         struct module *owner = sk->sk_prot_creator->owner;
 664
 665         if (sk->sk_destruct)
 666                 sk->sk_destruct(sk);
 667
 668         filter = sk->sk_filter;
 669         if (filter) {
 670                 sk_filter_release(sk, filter);
 671                 sk->sk_filter = NULL;
 672         }
 673
 674         sock_disable_timestamp(sk);
 675
 676         if (atomic_read(&sk->sk_omem_alloc))
 677                 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
 678                        __FUNCTION__, atomic_read(&sk->sk_omem_alloc));
 679
 680         security_sk_free(sk);
 681         if (sk->sk_prot_creator->slab != NULL)
 682                 kmem_cache_free(sk->sk_prot_creator->slab, sk);
 683         else
 684                 kfree(sk);
 685         module_put(owner);
 686 }
 687
 688 void __init sk_init(void)
 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 (!sock_flag(sk, SOCK_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,
 755                              unsigned int __nocast priority)
 756 {
 757         if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
 758                 struct sk_buff * skb = alloc_skb(size, priority);
 759                 if (skb) {
 760                         skb_set_owner_w(skb, sk);
 761                         return skb;
 762                 }
 763         }
 764         return NULL;
 765 }
 766
 767 /*
 768  * Allocate a skb from the socket's receive buffer.
 769  */
 770 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
 771                              unsigned int __nocast priority)
 772 {
 773         if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
 774                 struct sk_buff *skb = alloc_skb(size, priority);
 775                 if (skb) {
 776                         skb_set_owner_r(skb, sk);
 777                         return skb;
 778                 }
 779         }
 780         return NULL;
 781 }
 782
 783 /*
 784  * Allocate a memory block from the socket's option memory buffer.
 785  */
 786 void *sock_kmalloc(struct sock *sk, int size, unsigned int __nocast priority)
 787 {
 788         if ((unsigned)size <= sysctl_optmem_max &&
 789             atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
 790                 void *mem;
 791                 /* First do the add, to avoid the race if kmalloc
 792                  * might sleep.
 793                  */
 794                 atomic_add(size, &sk->sk_omem_alloc);
 795                 mem = kmalloc(size, priority);
 796                 if (mem)
 797                         return mem;
 798                 atomic_sub(size, &sk->sk_omem_alloc);
 799         }
 800         return NULL;
 801 }
 802
 803 /*
 804  * Free an option memory block.
 805  */
 806 void sock_kfree_s(struct sock *sk, void *mem, int size)
 807 {
 808         kfree(mem);
 809         atomic_sub(size, &sk->sk_omem_alloc);
 810 }
 811
 812 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
 813    I think, these locks should be removed for datagram sockets.
 814  */
 815 static long sock_wait_for_wmem(struct sock * sk, long timeo)
 816 {
 817         DEFINE_WAIT(wait);
 818
 819         clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
 820         for (;;) {
 821                 if (!timeo)
 822                         break;
 823                 if (signal_pending(current))
 824                         break;
 825                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
 826                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
 827                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
 828                         break;
 829                 if (sk->sk_shutdown & SEND_SHUTDOWN)
 830                         break;
 831                 if (sk->sk_err)
 832                         break;
 833                 timeo = schedule_timeout(timeo);
 834         }
 835         finish_wait(sk->sk_sleep, &wait);
 836         return timeo;
 837 }
 838
 839
 840 /*
 841  *      Generic send/receive buffer handlers
 842  */
 843
 844 static struct sk_buff *sock_alloc_send_pskb(struct sock *sk,
 845                                             unsigned long header_len,
 846                                             unsigned long data_len,
 847                                             int noblock, int *errcode)
 848 {
 849         struct sk_buff *skb;
 850         unsigned int gfp_mask;
 851         long timeo;
 852         int err;
 853
 854         gfp_mask = sk->sk_allocation;
 855         if (gfp_mask & __GFP_WAIT)
 856                 gfp_mask |= __GFP_REPEAT;
 857
 858         timeo = sock_sndtimeo(sk, noblock);
 859         while (1) {
 860                 err = sock_error(sk);
 861                 if (err != 0)
 862                         goto failure;
 863
 864                 err = -EPIPE;
 865                 if (sk->sk_shutdown & SEND_SHUTDOWN)
 866                         goto failure;
 867
 868                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
 869                         skb = alloc_skb(header_len, sk->sk_allocation);
 870                         if (skb) {
 871                                 int npages;
 872                                 int i;
 873
 874                                 /* No pages, we're done... */
 875                                 if (!data_len)
 876                                         break;
 877
 878                                 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
 879                                 skb->truesize += data_len;
 880                                 skb_shinfo(skb)->nr_frags = npages;
 881                                 for (i = 0; i < npages; i++) {
 882                                         struct page *page;
 883                                         skb_frag_t *frag;
 884
 885                                         page = alloc_pages(sk->sk_allocation, 0);
 886                                         if (!page) {
 887                                                 err = -ENOBUFS;
 888                                                 skb_shinfo(skb)->nr_frags = i;
 889                                                 kfree_skb(skb);
 890                                                 goto failure;
 891                                         }
 892
 893                                         frag = &skb_shinfo(skb)->frags[i];
 894                                         frag->page = page;
 895                                         frag->page_offset = 0;
 896                                         frag->size = (data_len >= PAGE_SIZE ?
 897                                                       PAGE_SIZE :
 898                                                       data_len);
 899                                         data_len -= PAGE_SIZE;
 900                                 }
 901
 902                                 /* Full success... */
 903                                 break;
 904                         }
 905                         err = -ENOBUFS;
 906                         goto failure;
 907                 }
 908                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
 909                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
 910                 err = -EAGAIN;
 911                 if (!timeo)
 912                         goto failure;
 913                 if (signal_pending(current))
 914                         goto interrupted;
 915                 timeo = sock_wait_for_wmem(sk, timeo);
 916         }
 917
 918         skb_set_owner_w(skb, sk);
 919         return skb;
 920
 921 interrupted:
 922         err = sock_intr_errno(timeo);
 923 failure:
 924         *errcode = err;
 925         return NULL;
 926 }
 927
 928 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
 929                                     int noblock, int *errcode)
 930 {
 931         return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
 932 }
 933
 934 static void __lock_sock(struct sock *sk)
 935 {
 936         DEFINE_WAIT(wait);
 937
 938         for(;;) {
 939                 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
 940                                         TASK_UNINTERRUPTIBLE);
 941                 spin_unlock_bh(&sk->sk_lock.slock);
 942                 schedule();
 943                 spin_lock_bh(&sk->sk_lock.slock);
 944                 if(!sock_owned_by_user(sk))
 945                         break;
 946         }
 947         finish_wait(&sk->sk_lock.wq, &wait);
 948 }
 949
 950 static void __release_sock(struct sock *sk)
 951 {
 952         struct sk_buff *skb = sk->sk_backlog.head;
 953
 954         do {
 955                 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
 956                 bh_unlock_sock(sk);
 957
 958                 do {
 959                         struct sk_buff *next = skb->next;
 960
 961                         skb->next = NULL;
 962                         sk->sk_backlog_rcv(sk, skb);
 963
 964                         /*
 965                          * We are in process context here with softirqs
 966                          * disabled, use cond_resched_softirq() to preempt.
 967                          * This is safe to do because we've taken the backlog
 968                          * queue private:
 969                          */
 970                         cond_resched_softirq();
 971
 972                         skb = next;
 973                 } while (skb != NULL);
 974
 975                 bh_lock_sock(sk);
 976         } while((skb = sk->sk_backlog.head) != NULL);
 977 }
 978
 979 /**
 980  * sk_wait_data - wait for data to arrive at sk_receive_queue
 981  * @sk:    sock to wait on
 982  * @timeo: for how long
 983  *
 984  * Now socket state including sk->sk_err is changed only under lock,
 985  * hence we may omit checks after joining wait queue.
 986  * We check receive queue before schedule() only as optimization;
 987  * it is very likely that release_sock() added new data.
 988  */
 989 int sk_wait_data(struct sock *sk, long *timeo)
 990 {
 991         int rc;
 992         DEFINE_WAIT(wait);
 993
 994         prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
 995         set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
 996         rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
 997         clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
 998         finish_wait(sk->sk_sleep, &wait);
 999         return rc;
1000 }
1001
1002 EXPORT_SYMBOL(sk_wait_data);
1003
1004 /*
1005  * Set of default routines for initialising struct proto_ops when
1006  * the protocol does not support a particular function. In certain
1007  * cases where it makes no sense for a protocol to have a "do nothing"
1008  * function, some default processing is provided.
1009  */
1010
1011 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1012 {
1013         return -EOPNOTSUPP;
1014 }
1015
1016 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1017                     int len, int flags)
1018 {
1019         return -EOPNOTSUPP;
1020 }
1021
1022 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1023 {
1024         return -EOPNOTSUPP;
1025 }
1026
1027 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1028 {
1029         return -EOPNOTSUPP;
1030 }
1031
1032 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1033                     int *len, int peer)
1034 {
1035         return -EOPNOTSUPP;
1036 }
1037
1038 unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt)
1039 {
1040         return 0;
1041 }
1042
1043 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1044 {
1045         return -EOPNOTSUPP;
1046 }
1047
1048 int sock_no_listen(struct socket *sock, int backlog)
1049 {
1050         return -EOPNOTSUPP;
1051 }
1052
1053 int sock_no_shutdown(struct socket *sock, int how)
1054 {
1055         return -EOPNOTSUPP;
1056 }
1057
1058 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1059                     char __user *optval, int optlen)
1060 {
1061         return -EOPNOTSUPP;
1062 }
1063
1064 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1065                     char __user *optval, int __user *optlen)
1066 {
1067         return -EOPNOTSUPP;
1068 }
1069
1070 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1071                     size_t len)
1072 {
1073         return -EOPNOTSUPP;
1074 }
1075
1076 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1077                     size_t len, int flags)
1078 {
1079         return -EOPNOTSUPP;
1080 }
1081
1082 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1083 {
1084         /* Mirror missing mmap method error code */
1085         return -ENODEV;
1086 }
1087
1088 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1089 {
1090         ssize_t res;
1091         struct msghdr msg = {.msg_flags = flags};
1092         struct kvec iov;
1093         char *kaddr = kmap(page);
1094         iov.iov_base = kaddr + offset;
1095         iov.iov_len = size;
1096         res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1097         kunmap(page);
1098         return res;
1099 }
1100
1101 /*
1102  *      Default Socket Callbacks
1103  */
1104
1105 static void sock_def_wakeup(struct sock *sk)
1106 {
1107         read_lock(&sk->sk_callback_lock);
1108         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1109                 wake_up_interruptible_all(sk->sk_sleep);
1110         read_unlock(&sk->sk_callback_lock);
1111 }
1112
1113 static void sock_def_error_report(struct sock *sk)
1114 {
1115         read_lock(&sk->sk_callback_lock);
1116         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1117                 wake_up_interruptible(sk->sk_sleep);
1118         sk_wake_async(sk,0,POLL_ERR);
1119         read_unlock(&sk->sk_callback_lock);
1120 }
1121
1122 static void sock_def_readable(struct sock *sk, int len)
1123 {
1124         read_lock(&sk->sk_callback_lock);
1125         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1126                 wake_up_interruptible(sk->sk_sleep);
1127         sk_wake_async(sk,1,POLL_IN);
1128         read_unlock(&sk->sk_callback_lock);
1129 }
1130
1131 static void sock_def_write_space(struct sock *sk)
1132 {
1133         read_lock(&sk->sk_callback_lock);
1134
1135         /* Do not wake up a writer until he can make "significant"
1136          * progress.  --DaveM
1137          */
1138         if((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1139                 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1140                         wake_up_interruptible(sk->sk_sleep);
1141
1142                 /* Should agree with poll, otherwise some programs break */
1143                 if (sock_writeable(sk))
1144                         sk_wake_async(sk, 2, POLL_OUT);
1145         }
1146
1147         read_unlock(&sk->sk_callback_lock);
1148 }
1149
1150 static void sock_def_destruct(struct sock *sk)
1151 {
1152         if (sk->sk_protinfo)
1153                 kfree(sk->sk_protinfo);
1154 }
1155
1156 void sk_send_sigurg(struct sock *sk)
1157 {
1158         if (sk->sk_socket && sk->sk_socket->file)
1159                 if (send_sigurg(&sk->sk_socket->file->f_owner))
1160                         sk_wake_async(sk, 3, POLL_PRI);
1161 }
1162
1163 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1164                     unsigned long expires)
1165 {
1166         if (!mod_timer(timer, expires))
1167                 sock_hold(sk);
1168 }
1169
1170 EXPORT_SYMBOL(sk_reset_timer);
1171
1172 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1173 {
1174         if (timer_pending(timer) && del_timer(timer))
1175                 __sock_put(sk);
1176 }
1177
1178 EXPORT_SYMBOL(sk_stop_timer);
1179
1180 void sock_init_data(struct socket *sock, struct sock *sk)
1181 {
1182         skb_queue_head_init(&sk->sk_receive_queue);
1183         skb_queue_head_init(&sk->sk_write_queue);
1184         skb_queue_head_init(&sk->sk_error_queue);
1185
1186         sk->sk_send_head        =       NULL;
1187
1188         init_timer(&sk->sk_timer);
1189
1190         sk->sk_allocation       =       GFP_KERNEL;
1191         sk->sk_rcvbuf           =       sysctl_rmem_default;
1192         sk->sk_sndbuf           =       sysctl_wmem_default;
1193         sk->sk_state            =       TCP_CLOSE;
1194         sk->sk_socket           =       sock;
1195
1196         sock_set_flag(sk, SOCK_ZAPPED);
1197
1198         if(sock)
1199         {
1200                 sk->sk_type     =       sock->type;
1201                 sk->sk_sleep    =       &sock->wait;
1202                 sock->sk        =       sk;
1203         } else
1204                 sk->sk_sleep    =       NULL;
1205
1206         rwlock_init(&sk->sk_dst_lock);
1207         rwlock_init(&sk->sk_callback_lock);
1208
1209         sk->sk_state_change     =       sock_def_wakeup;
1210         sk->sk_data_ready       =       sock_def_readable;
1211         sk->sk_write_space      =       sock_def_write_space;
1212         sk->sk_error_report     =       sock_def_error_report;
1213         sk->sk_destruct         =       sock_def_destruct;
1214
1215         sk->sk_sndmsg_page      =       NULL;
1216         sk->sk_sndmsg_off       =       0;
1217
1218         sk->sk_peercred.pid     =       0;
1219         sk->sk_peercred.uid     =       -1;
1220         sk->sk_peercred.gid     =       -1;
1221         sk->sk_write_pending    =       0;
1222         sk->sk_rcvlowat         =       1;
1223         sk->sk_rcvtimeo         =       MAX_SCHEDULE_TIMEOUT;
1224         sk->sk_sndtimeo         =       MAX_SCHEDULE_TIMEOUT;
1225
1226         sk->sk_stamp.tv_sec     = -1L;
1227         sk->sk_stamp.tv_usec    = -1L;
1228
1229         atomic_set(&sk->sk_refcnt, 1);
1230 }
1231
1232 void fastcall lock_sock(struct sock *sk)
1233 {
1234         might_sleep();
1235         spin_lock_bh(&(sk->sk_lock.slock));
1236         if (sk->sk_lock.owner)
1237                 __lock_sock(sk);
1238         sk->sk_lock.owner = (void *)1;
1239         spin_unlock_bh(&(sk->sk_lock.slock));
1240 }
1241
1242 EXPORT_SYMBOL(lock_sock);
1243
1244 void fastcall release_sock(struct sock *sk)
1245 {
1246         spin_lock_bh(&(sk->sk_lock.slock));
1247         if (sk->sk_backlog.tail)
1248                 __release_sock(sk);
1249         sk->sk_lock.owner = NULL;
1250         if (waitqueue_active(&(sk->sk_lock.wq)))
1251                 wake_up(&(sk->sk_lock.wq));
1252         spin_unlock_bh(&(sk->sk_lock.slock));
1253 }
1254 EXPORT_SYMBOL(release_sock);
1255
1256 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1257 {
1258         if (!sock_flag(sk, SOCK_TIMESTAMP))
1259                 sock_enable_timestamp(sk);
1260         if (sk->sk_stamp.tv_sec == -1)
1261                 return -ENOENT;
1262         if (sk->sk_stamp.tv_sec == 0)
1263                 do_gettimeofday(&sk->sk_stamp);
1264         return copy_to_user(userstamp, &sk->sk_stamp, sizeof(struct timeval)) ?
1265                 -EFAULT : 0;
1266 }
1267 EXPORT_SYMBOL(sock_get_timestamp);
1268
1269 void sock_enable_timestamp(struct sock *sk)
1270 {
1271         if (!sock_flag(sk, SOCK_TIMESTAMP)) {
1272                 sock_set_flag(sk, SOCK_TIMESTAMP);
1273                 net_enable_timestamp();
1274         }
1275 }
1276 EXPORT_SYMBOL(sock_enable_timestamp);
1277
1278 /*
1279  *      Get a socket option on an socket.
1280  *
1281  *      FIX: POSIX 1003.1g is very ambiguous here. It states that
1282  *      asynchronous errors should be reported by getsockopt. We assume
1283  *      this means if you specify SO_ERROR (otherwise whats the point of it).
1284  */
1285 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1286                            char __user *optval, int __user *optlen)
1287 {
1288         struct sock *sk = sock->sk;
1289
1290         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1291 }
1292
1293 EXPORT_SYMBOL(sock_common_getsockopt);
1294
1295 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1296                         struct msghdr *msg, size_t size, int flags)
1297 {
1298         struct sock *sk = sock->sk;
1299         int addr_len = 0;
1300         int err;
1301
1302         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
1303                                    flags & ~MSG_DONTWAIT, &addr_len);
1304         if (err >= 0)
1305                 msg->msg_namelen = addr_len;
1306         return err;
1307 }
1308
1309 EXPORT_SYMBOL(sock_common_recvmsg);
1310
1311 /*
1312  *      Set socket options on an inet socket.
1313  */
1314 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1315                            char __user *optval, int optlen)
1316 {
1317         struct sock *sk = sock->sk;
1318
1319         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1320 }
1321
1322 EXPORT_SYMBOL(sock_common_setsockopt);
1323
1324 void sk_common_release(struct sock *sk)
1325 {
1326         if (sk->sk_prot->destroy)
1327                 sk->sk_prot->destroy(sk);
1328
1329         /*
1330          * Observation: when sock_common_release is called, processes have
1331          * no access to socket. But net still has.
1332          * Step one, detach it from networking:
1333          *
1334          * A. Remove from hash tables.
1335          */
1336
1337         sk->sk_prot->unhash(sk);
1338
1339         /*
1340          * In this point socket cannot receive new packets, but it is possible
1341          * that some packets are in flight because some CPU runs receiver and
1342          * did hash table lookup before we unhashed socket. They will achieve
1343          * receive queue and will be purged by socket destructor.
1344          *
1345          * Also we still have packets pending on receive queue and probably,
1346          * our own packets waiting in device queues. sock_destroy will drain
1347          * receive queue, but transmitted packets will delay socket destruction
1348          * until the last reference will be released.
1349          */
1350
1351         sock_orphan(sk);
1352
1353         xfrm_sk_free_policy(sk);
1354
1355 #ifdef INET_REFCNT_DEBUG
1356         if (atomic_read(&sk->sk_refcnt) != 1)
1357                 printk(KERN_DEBUG "Destruction of the socket %p delayed, c=%d\n",
1358                        sk, atomic_read(&sk->sk_refcnt));
1359 #endif
1360         sock_put(sk);
1361 }
1362
1363 EXPORT_SYMBOL(sk_common_release);
1364
1365 static DEFINE_RWLOCK(proto_list_lock);
1366 static LIST_HEAD(proto_list);
1367
1368 int proto_register(struct proto *prot, int alloc_slab)
1369 {
1370         char *request_sock_slab_name;
1371         int rc = -ENOBUFS;
1372
1373         if (alloc_slab) {
1374                 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
1375                                                SLAB_HWCACHE_ALIGN, NULL, NULL);
1376
1377                 if (prot->slab == NULL) {
1378                         printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
1379                                prot->name);
1380                         goto out;
1381                 }
1382
1383                 if (prot->rsk_prot != NULL) {
1384                         static const char mask[] = "request_sock_%s";
1385
1386                         request_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
1387                         if (request_sock_slab_name == NULL)
1388                                 goto out_free_sock_slab;
1389
1390                         sprintf(request_sock_slab_name, mask, prot->name);
1391                         prot->rsk_prot->slab = kmem_cache_create(request_sock_slab_name,
1392                                                                  prot->rsk_prot->obj_size, 0,
1393                                                                  SLAB_HWCACHE_ALIGN, NULL, NULL);
1394
1395                         if (prot->rsk_prot->slab == NULL) {
1396                                 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
1397                                        prot->name);
1398                                 goto out_free_request_sock_slab_name;
1399                         }
1400                 }
1401         }
1402
1403         write_lock(&proto_list_lock);
1404         list_add(&prot->node, &proto_list);
1405         write_unlock(&proto_list_lock);
1406         rc = 0;
1407 out:
1408         return rc;
1409 out_free_request_sock_slab_name:
1410         kfree(request_sock_slab_name);
1411 out_free_sock_slab:
1412         kmem_cache_destroy(prot->slab);
1413         prot->slab = NULL;
1414         goto out;
1415 }
1416
1417 EXPORT_SYMBOL(proto_register);
1418
1419 void proto_unregister(struct proto *prot)
1420 {
1421         write_lock(&proto_list_lock);
1422
1423         if (prot->slab != NULL) {
1424                 kmem_cache_destroy(prot->slab);
1425                 prot->slab = NULL;
1426         }
1427
1428         if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
1429                 const char *name = kmem_cache_name(prot->rsk_prot->slab);
1430
1431                 kmem_cache_destroy(prot->rsk_prot->slab);
1432                 kfree(name);
1433                 prot->rsk_prot->slab = NULL;
1434         }
1435
1436         list_del(&prot->node);
1437         write_unlock(&proto_list_lock);
1438 }
1439
1440 EXPORT_SYMBOL(proto_unregister);
1441
1442 #ifdef CONFIG_PROC_FS
1443 static inline struct proto *__proto_head(void)
1444 {
1445         return list_entry(proto_list.next, struct proto, node);
1446 }
1447
1448 static inline struct proto *proto_head(void)
1449 {
1450         return list_empty(&proto_list) ? NULL : __proto_head();
1451 }
1452
1453 static inline struct proto *proto_next(struct proto *proto)
1454 {
1455         return proto->node.next == &proto_list ? NULL :
1456                 list_entry(proto->node.next, struct proto, node);
1457 }
1458
1459 static inline struct proto *proto_get_idx(loff_t pos)
1460 {
1461         struct proto *proto;
1462         loff_t i = 0;
1463
1464         list_for_each_entry(proto, &proto_list, node)
1465                 if (i++ == pos)
1466                         goto out;
1467
1468         proto = NULL;
1469 out:
1470         return proto;
1471 }
1472
1473 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
1474 {
1475         read_lock(&proto_list_lock);
1476         return *pos ? proto_get_idx(*pos - 1) : SEQ_START_TOKEN;
1477 }
1478
1479 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1480 {
1481         ++*pos;
1482         return v == SEQ_START_TOKEN ? proto_head() : proto_next(v);
1483 }
1484
1485 static void proto_seq_stop(struct seq_file *seq, void *v)
1486 {
1487         read_unlock(&proto_list_lock);
1488 }
1489
1490 static char proto_method_implemented(const void *method)
1491 {
1492         return method == NULL ? 'n' : 'y';
1493 }
1494
1495 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
1496 {
1497         seq_printf(seq, "%-9s %4u %6d  %6d   %-3s %6u   %-3s  %-10s "
1498                         "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
1499                    proto->name,
1500                    proto->obj_size,
1501                    proto->sockets_allocated != NULL ? atomic_read(proto->sockets_allocated) : -1,
1502                    proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
1503                    proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
1504                    proto->max_header,
1505                    proto->slab == NULL ? "no" : "yes",
1506                    module_name(proto->owner),
1507                    proto_method_implemented(proto->close),
1508                    proto_method_implemented(proto->connect),
1509                    proto_method_implemented(proto->disconnect),
1510                    proto_method_implemented(proto->accept),
1511                    proto_method_implemented(proto->ioctl),
1512                    proto_method_implemented(proto->init),
1513                    proto_method_implemented(proto->destroy),
1514                    proto_method_implemented(proto->shutdown),
1515                    proto_method_implemented(proto->setsockopt),
1516                    proto_method_implemented(proto->getsockopt),
1517                    proto_method_implemented(proto->sendmsg),
1518                    proto_method_implemented(proto->recvmsg),
1519                    proto_method_implemented(proto->sendpage),
1520                    proto_method_implemented(proto->bind),
1521                    proto_method_implemented(proto->backlog_rcv),
1522                    proto_method_implemented(proto->hash),
1523                    proto_method_implemented(proto->unhash),
1524                    proto_method_implemented(proto->get_port),
1525                    proto_method_implemented(proto->enter_memory_pressure));
1526 }
1527
1528 static int proto_seq_show(struct seq_file *seq, void *v)
1529 {
1530         if (v == SEQ_START_TOKEN)
1531                 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
1532                            "protocol",
1533                            "size",
1534                            "sockets",
1535                            "memory",
1536                            "press",
1537                            "maxhdr",
1538                            "slab",
1539                            "module",
1540                            "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
1541         else
1542                 proto_seq_printf(seq, v);
1543         return 0;
1544 }
1545
1546 static struct seq_operations proto_seq_ops = {
1547         .start  = proto_seq_start,
1548         .next   = proto_seq_next,
1549         .stop   = proto_seq_stop,
1550         .show   = proto_seq_show,
1551 };
1552
1553 static int proto_seq_open(struct inode *inode, struct file *file)
1554 {
1555         return seq_open(file, &proto_seq_ops);
1556 }
1557
1558 static struct file_operations proto_seq_fops = {
1559         .owner          = THIS_MODULE,
1560         .open           = proto_seq_open,
1561         .read           = seq_read,
1562         .llseek         = seq_lseek,
1563         .release        = seq_release,
1564 };
1565
1566 static int __init proto_init(void)
1567 {
1568         /* register /proc/net/protocols */
1569         return proc_net_fops_create("protocols", S_IRUGO, &proto_seq_fops) == NULL ? -ENOBUFS : 0;
1570 }
1571
1572 subsys_initcall(proto_init);
1573
1574 #endif /* PROC_FS */
1575
1576 EXPORT_SYMBOL(sk_alloc);
1577 EXPORT_SYMBOL(sk_free);
1578 EXPORT_SYMBOL(sk_send_sigurg);
1579 EXPORT_SYMBOL(sock_alloc_send_skb);
1580 EXPORT_SYMBOL(sock_init_data);
1581 EXPORT_SYMBOL(sock_kfree_s);
1582 EXPORT_SYMBOL(sock_kmalloc);
1583 EXPORT_SYMBOL(sock_no_accept);
1584 EXPORT_SYMBOL(sock_no_bind);
1585 EXPORT_SYMBOL(sock_no_connect);
1586 EXPORT_SYMBOL(sock_no_getname);
1587 EXPORT_SYMBOL(sock_no_getsockopt);
1588 EXPORT_SYMBOL(sock_no_ioctl);
1589 EXPORT_SYMBOL(sock_no_listen);
1590 EXPORT_SYMBOL(sock_no_mmap);
1591 EXPORT_SYMBOL(sock_no_poll);
1592 EXPORT_SYMBOL(sock_no_recvmsg);
1593 EXPORT_SYMBOL(sock_no_sendmsg);
1594 EXPORT_SYMBOL(sock_no_sendpage);
1595 EXPORT_SYMBOL(sock_no_setsockopt);
1596 EXPORT_SYMBOL(sock_no_shutdown);
1597 EXPORT_SYMBOL(sock_no_socketpair);
1598 EXPORT_SYMBOL(sock_rfree);
1599 EXPORT_SYMBOL(sock_setsockopt);
1600 EXPORT_SYMBOL(sock_wfree);
1601 EXPORT_SYMBOL(sock_wmalloc);
1602 EXPORT_SYMBOL(sock_i_uid);
1603 EXPORT_SYMBOL(sock_i_ino);
1604 #ifdef CONFIG_SYSCTL
1605 EXPORT_SYMBOL(sysctl_optmem_max);
1606 EXPORT_SYMBOL(sysctl_rmem_max);
1607 EXPORT_SYMBOL(sysctl_wmem_max);
1608 #endif