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         if (optname == SO_DONTLINGER) {
 210                 lock_sock(sk);
 211                 sock_reset_flag(sk, SOCK_LINGER);
 212                 release_sock(sk);
 213                 return 0;
 214         }
 215 #endif
 216
 217         if(optlen<sizeof(int))
 218                 return(-EINVAL);
 219
 220         if (get_user(val, (int __user *)optval))
 221                 return -EFAULT;
 222
 223         valbool = val?1:0;
 224
 225         lock_sock(sk);
 226
 227         switch(optname)
 228         {
 229                 case SO_DEBUG:
 230                         if(val && !capable(CAP_NET_ADMIN))
 231                         {
 232                                 ret = -EACCES;
 233                         }
 234                         else if (valbool)
 235                                 sock_set_flag(sk, SOCK_DBG);
 236                         else
 237                                 sock_reset_flag(sk, SOCK_DBG);
 238                         break;
 239                 case SO_REUSEADDR:
 240                         sk->sk_reuse = valbool;
 241                         break;
 242                 case SO_TYPE:
 243                 case SO_ERROR:
 244                         ret = -ENOPROTOOPT;
 245                         break;
 246                 case SO_DONTROUTE:
 247                         if (valbool)
 248                                 sock_set_flag(sk, SOCK_LOCALROUTE);
 249                         else
 250                                 sock_reset_flag(sk, SOCK_LOCALROUTE);
 251                         break;
 252                 case SO_BROADCAST:
 253                         sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
 254                         break;
 255                 case SO_SNDBUF:
 256                         /* Don't error on this BSD doesn't and if you think
 257                            about it this is right. Otherwise apps have to
 258                            play 'guess the biggest size' games. RCVBUF/SNDBUF
 259                            are treated in BSD as hints */
 260
 261                         if (val > sysctl_wmem_max)
 262                                 val = sysctl_wmem_max;
 263
 264                         sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
 265                         if ((val * 2) < SOCK_MIN_SNDBUF)
 266                                 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
 267                         else
 268                                 sk->sk_sndbuf = val * 2;
 269
 270                         /*
 271                          *      Wake up sending tasks if we
 272                          *      upped the value.
 273                          */
 274                         sk->sk_write_space(sk);
 275                         break;
 276
 277                 case SO_RCVBUF:
 278                         /* Don't error on this BSD doesn't and if you think
 279                            about it this is right. Otherwise apps have to
 280                            play 'guess the biggest size' games. RCVBUF/SNDBUF
 281                            are treated in BSD as hints */
 282
 283                         if (val > sysctl_rmem_max)
 284                                 val = sysctl_rmem_max;
 285
 286                         sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
 287                         /* FIXME: is this lower bound the right one? */
 288                         if ((val * 2) < SOCK_MIN_RCVBUF)
 289                                 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
 290                         else
 291                                 sk->sk_rcvbuf = val * 2;
 292                         break;
 293
 294                 case SO_KEEPALIVE:
 295 #ifdef CONFIG_INET
 296                         if (sk->sk_protocol == IPPROTO_TCP)
 297                                 tcp_set_keepalive(sk, valbool);
 298 #endif
 299                         sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
 300                         break;
 301
 302                 case SO_OOBINLINE:
 303                         sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
 304                         break;
 305
 306                 case SO_NO_CHECK:
 307                         sk->sk_no_check = valbool;
 308                         break;
 309
 310                 case SO_PRIORITY:
 311                         if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
 312                                 sk->sk_priority = val;
 313                         else
 314                                 ret = -EPERM;
 315                         break;
 316
 317                 case SO_LINGER:
 318                         if(optlen<sizeof(ling)) {
 319                                 ret = -EINVAL;  /* 1003.1g */
 320                                 break;
 321                         }
 322                         if (copy_from_user(&ling,optval,sizeof(ling))) {
 323                                 ret = -EFAULT;
 324                                 break;
 325                         }
 326                         if (!ling.l_onoff)
 327                                 sock_reset_flag(sk, SOCK_LINGER);
 328                         else {
 329 #if (BITS_PER_LONG == 32)
 330                                 if (ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
 331                                         sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
 332                                 else
 333 #endif
 334                                         sk->sk_lingertime = ling.l_linger * HZ;
 335                                 sock_set_flag(sk, SOCK_LINGER);
 336                         }
 337                         break;
 338
 339                 case SO_BSDCOMPAT:
 340                         sock_warn_obsolete_bsdism("setsockopt");
 341                         break;
 342
 343                 case SO_PASSCRED:
 344                         if (valbool)
 345                                 set_bit(SOCK_PASSCRED, &sock->flags);
 346                         else
 347                                 clear_bit(SOCK_PASSCRED, &sock->flags);
 348                         break;
 349
 350                 case SO_TIMESTAMP:
 351                         if (valbool)  {
 352                                 sock_set_flag(sk, SOCK_RCVTSTAMP);
 353                                 sock_enable_timestamp(sk);
 354                         } else
 355                                 sock_reset_flag(sk, SOCK_RCVTSTAMP);
 356                         break;
 357
 358                 case SO_RCVLOWAT:
 359                         if (val < 0)
 360                                 val = INT_MAX;
 361                         sk->sk_rcvlowat = val ? : 1;
 362                         break;
 363
 364                 case SO_RCVTIMEO:
 365                         ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
 366                         break;
 367
 368                 case SO_SNDTIMEO:
 369                         ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
 370                         break;
 371
 372 #ifdef CONFIG_NETDEVICES
 373                 case SO_BINDTODEVICE:
 374                 {
 375                         char devname[IFNAMSIZ];
 376
 377                         /* Sorry... */
 378                         if (!capable(CAP_NET_RAW)) {
 379                                 ret = -EPERM;
 380                                 break;
 381                         }
 382
 383                         /* Bind this socket to a particular device like "eth0",
 384                          * as specified in the passed interface name. If the
 385                          * name is "" or the option length is zero the socket
 386                          * is not bound.
 387                          */
 388
 389                         if (!valbool) {
 390                                 sk->sk_bound_dev_if = 0;
 391                         } else {
 392                                 if (optlen > IFNAMSIZ)
 393                                         optlen = IFNAMSIZ;
 394                                 if (copy_from_user(devname, optval, optlen)) {
 395                                         ret = -EFAULT;
 396                                         break;
 397                                 }
 398
 399                                 /* Remove any cached route for this socket. */
 400                                 sk_dst_reset(sk);
 401
 402                                 if (devname[0] == '\0') {
 403                                         sk->sk_bound_dev_if = 0;
 404                                 } else {
 405                                         struct net_device *dev = dev_get_by_name(devname);
 406                                         if (!dev) {
 407                                                 ret = -ENODEV;
 408                                                 break;
 409                                         }
 410                                         sk->sk_bound_dev_if = dev->ifindex;
 411                                         dev_put(dev);
 412                                 }
 413                         }
 414                         break;
 415                 }
 416 #endif
 417
 418
 419                 case SO_ATTACH_FILTER:
 420                         ret = -EINVAL;
 421                         if (optlen == sizeof(struct sock_fprog)) {
 422                                 struct sock_fprog fprog;
 423
 424                                 ret = -EFAULT;
 425                                 if (copy_from_user(&fprog, optval, sizeof(fprog)))
 426                                         break;
 427
 428                                 ret = sk_attach_filter(&fprog, sk);
 429                         }
 430                         break;
 431
 432                 case SO_DETACH_FILTER:
 433                         spin_lock_bh(&sk->sk_lock.slock);
 434                         filter = sk->sk_filter;
 435                         if (filter) {
 436                                 sk->sk_filter = NULL;
 437                                 spin_unlock_bh(&sk->sk_lock.slock);
 438                                 sk_filter_release(sk, filter);
 439                                 break;
 440                         }
 441                         spin_unlock_bh(&sk->sk_lock.slock);
 442                         ret = -ENONET;
 443                         break;
 444
 445                 /* We implement the SO_SNDLOWAT etc to
 446                    not be settable (1003.1g 5.3) */
 447                 default:
 448                         ret = -ENOPROTOOPT;
 449                         break;
 450         }
 451         release_sock(sk);
 452         return ret;
 453 }
 454
 455
 456 int sock_getsockopt(struct socket *sock, int level, int optname,
 457                     char __user *optval, int __user *optlen)
 458 {
 459         struct sock *sk = sock->sk;
 460
 461         union
 462         {
 463                 int val;
 464                 struct linger ling;
 465                 struct timeval tm;
 466         } v;
 467
 468         unsigned int lv = sizeof(int);
 469         int len;
 470
 471         if(get_user(len,optlen))
 472                 return -EFAULT;
 473         if(len < 0)
 474                 return -EINVAL;
 475
 476         switch(optname)
 477         {
 478                 case SO_DEBUG:
 479                         v.val = sock_flag(sk, SOCK_DBG);
 480                         break;
 481
 482                 case SO_DONTROUTE:
 483                         v.val = sock_flag(sk, SOCK_LOCALROUTE);
 484                         break;
 485
 486                 case SO_BROADCAST:
 487                         v.val = !!sock_flag(sk, SOCK_BROADCAST);
 488                         break;
 489
 490                 case SO_SNDBUF:
 491                         v.val = sk->sk_sndbuf;
 492                         break;
 493
 494                 case SO_RCVBUF:
 495                         v.val = sk->sk_rcvbuf;
 496                         break;
 497
 498                 case SO_REUSEADDR:
 499                         v.val = sk->sk_reuse;
 500                         break;
 501
 502                 case SO_KEEPALIVE:
 503                         v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
 504                         break;
 505
 506                 case SO_TYPE:
 507                         v.val = sk->sk_type;
 508                         break;
 509
 510                 case SO_ERROR:
 511                         v.val = -sock_error(sk);
 512                         if(v.val==0)
 513                                 v.val = xchg(&sk->sk_err_soft, 0);
 514                         break;
 515
 516                 case SO_OOBINLINE:
 517                         v.val = !!sock_flag(sk, SOCK_URGINLINE);
 518                         break;
 519
 520                 case SO_NO_CHECK:
 521                         v.val = sk->sk_no_check;
 522                         break;
 523
 524                 case SO_PRIORITY:
 525                         v.val = sk->sk_priority;
 526                         break;
 527
 528                 case SO_LINGER:
 529                         lv              = sizeof(v.ling);
 530                         v.ling.l_onoff  = !!sock_flag(sk, SOCK_LINGER);
 531                         v.ling.l_linger = sk->sk_lingertime / HZ;
 532                         break;
 533
 534                 case SO_BSDCOMPAT:
 535                         sock_warn_obsolete_bsdism("getsockopt");
 536                         break;
 537
 538                 case SO_TIMESTAMP:
 539                         v.val = sock_flag(sk, SOCK_RCVTSTAMP);
 540                         break;
 541
 542                 case SO_RCVTIMEO:
 543                         lv=sizeof(struct timeval);
 544                         if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
 545                                 v.tm.tv_sec = 0;
 546                                 v.tm.tv_usec = 0;
 547                         } else {
 548                                 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
 549                                 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
 550                         }
 551                         break;
 552
 553                 case SO_SNDTIMEO:
 554                         lv=sizeof(struct timeval);
 555                         if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
 556                                 v.tm.tv_sec = 0;
 557                                 v.tm.tv_usec = 0;
 558                         } else {
 559                                 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
 560                                 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
 561                         }
 562                         break;
 563
 564                 case SO_RCVLOWAT:
 565                         v.val = sk->sk_rcvlowat;
 566                         break;
 567
 568                 case SO_SNDLOWAT:
 569                         v.val=1;
 570                         break;
 571
 572                 case SO_PASSCRED:
 573                         v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
 574                         break;
 575
 576                 case SO_PEERCRED:
 577                         if (len > sizeof(sk->sk_peercred))
 578                                 len = sizeof(sk->sk_peercred);
 579                         if (copy_to_user(optval, &sk->sk_peercred, len))
 580                                 return -EFAULT;
 581                         goto lenout;
 582
 583                 case SO_PEERNAME:
 584                 {
 585                         char address[128];
 586
 587                         if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
 588                                 return -ENOTCONN;
 589                         if (lv < len)
 590                                 return -EINVAL;
 591                         if (copy_to_user(optval, address, len))
 592                                 return -EFAULT;
 593                         goto lenout;
 594                 }
 595
 596                 /* Dubious BSD thing... Probably nobody even uses it, but
 597                  * the UNIX standard wants it for whatever reason... -DaveM
 598                  */
 599                 case SO_ACCEPTCONN:
 600                         v.val = sk->sk_state == TCP_LISTEN;
 601                         break;
 602
 603                 case SO_PEERSEC:
 604                         return security_socket_getpeersec(sock, optval, optlen, len);
 605
 606                 default:
 607                         return(-ENOPROTOOPT);
 608         }
 609         if (len > lv)
 610                 len = lv;
 611         if (copy_to_user(optval, &v, len))
 612                 return -EFAULT;
 613 lenout:
 614         if (put_user(len, optlen))
 615                 return -EFAULT;
 616         return 0;
 617 }
 618
 619 /**
 620  *      sk_alloc - All socket objects are allocated here
 621  *      @family: protocol family
 622  *      @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
 623  *      @prot: struct proto associated with this new sock instance
 624  *      @zero_it: if we should zero the newly allocated sock
 625  */
 626 struct sock *sk_alloc(int family, unsigned int __nocast priority,
 627                       struct proto *prot, int zero_it)
 628 {
 629         struct sock *sk = NULL;
 630         kmem_cache_t *slab = prot->slab;
 631
 632         if (slab != NULL)
 633                 sk = kmem_cache_alloc(slab, priority);
 634         else
 635                 sk = kmalloc(prot->obj_size, priority);
 636
 637         if (sk) {
 638                 if (zero_it) {
 639                         memset(sk, 0, prot->obj_size);
 640                         sk->sk_family = family;
 641                         /*
 642                          * See comment in struct sock definition to understand
 643                          * why we need sk_prot_creator -acme
 644                          */
 645                         sk->sk_prot = sk->sk_prot_creator = prot;
 646                         sock_lock_init(sk);
 647                 }
 648
 649                 if (security_sk_alloc(sk, family, priority)) {
 650                         if (slab != NULL)
 651                                 kmem_cache_free(slab, sk);
 652                         else
 653                                 kfree(sk);
 654                         sk = NULL;
 655                 } else
 656                         __module_get(prot->owner);
 657         }
 658         return sk;
 659 }
 660
 661 void sk_free(struct sock *sk)
 662 {
 663         struct sk_filter *filter;
 664         struct module *owner = sk->sk_prot_creator->owner;
 665
 666         if (sk->sk_destruct)
 667                 sk->sk_destruct(sk);
 668
 669         filter = sk->sk_filter;
 670         if (filter) {
 671                 sk_filter_release(sk, filter);
 672                 sk->sk_filter = NULL;
 673         }
 674
 675         sock_disable_timestamp(sk);
 676
 677         if (atomic_read(&sk->sk_omem_alloc))
 678                 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
 679                        __FUNCTION__, atomic_read(&sk->sk_omem_alloc));
 680
 681         security_sk_free(sk);
 682         if (sk->sk_prot_creator->slab != NULL)
 683                 kmem_cache_free(sk->sk_prot_creator->slab, sk);
 684         else
 685                 kfree(sk);
 686         module_put(owner);
 687 }
 688
 689 void __init sk_init(void)
 690 {
 691         if (num_physpages <= 4096) {
 692                 sysctl_wmem_max = 32767;
 693                 sysctl_rmem_max = 32767;
 694                 sysctl_wmem_default = 32767;
 695                 sysctl_rmem_default = 32767;
 696         } else if (num_physpages >= 131072) {
 697                 sysctl_wmem_max = 131071;
 698                 sysctl_rmem_max = 131071;
 699         }
 700 }
 701
 702 /*
 703  *      Simple resource managers for sockets.
 704  */
 705
 706
 707 /*
 708  * Write buffer destructor automatically called from kfree_skb.
 709  */
 710 void sock_wfree(struct sk_buff *skb)
 711 {
 712         struct sock *sk = skb->sk;
 713
 714         /* In case it might be waiting for more memory. */
 715         atomic_sub(skb->truesize, &sk->sk_wmem_alloc);
 716         if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE))
 717                 sk->sk_write_space(sk);
 718         sock_put(sk);
 719 }
 720
 721 /*
 722  * Read buffer destructor automatically called from kfree_skb.
 723  */
 724 void sock_rfree(struct sk_buff *skb)
 725 {
 726         struct sock *sk = skb->sk;
 727
 728         atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
 729 }
 730
 731
 732 int sock_i_uid(struct sock *sk)
 733 {
 734         int uid;
 735
 736         read_lock(&sk->sk_callback_lock);
 737         uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
 738         read_unlock(&sk->sk_callback_lock);
 739         return uid;
 740 }
 741
 742 unsigned long sock_i_ino(struct sock *sk)
 743 {
 744         unsigned long ino;
 745
 746         read_lock(&sk->sk_callback_lock);
 747         ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
 748         read_unlock(&sk->sk_callback_lock);
 749         return ino;
 750 }
 751
 752 /*
 753  * Allocate a skb from the socket's send buffer.
 754  */
 755 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
 756                              unsigned int __nocast priority)
 757 {
 758         if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
 759                 struct sk_buff * skb = alloc_skb(size, priority);
 760                 if (skb) {
 761                         skb_set_owner_w(skb, sk);
 762                         return skb;
 763                 }
 764         }
 765         return NULL;
 766 }
 767
 768 /*
 769  * Allocate a skb from the socket's receive buffer.
 770  */
 771 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
 772                              unsigned int __nocast priority)
 773 {
 774         if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
 775                 struct sk_buff *skb = alloc_skb(size, priority);
 776                 if (skb) {
 777                         skb_set_owner_r(skb, sk);
 778                         return skb;
 779                 }
 780         }
 781         return NULL;
 782 }
 783
 784 /*
 785  * Allocate a memory block from the socket's option memory buffer.
 786  */
 787 void *sock_kmalloc(struct sock *sk, int size, unsigned int __nocast priority)
 788 {
 789         if ((unsigned)size <= sysctl_optmem_max &&
 790             atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
 791                 void *mem;
 792                 /* First do the add, to avoid the race if kmalloc
 793                  * might sleep.
 794                  */
 795                 atomic_add(size, &sk->sk_omem_alloc);
 796                 mem = kmalloc(size, priority);
 797                 if (mem)
 798                         return mem;
 799                 atomic_sub(size, &sk->sk_omem_alloc);
 800         }
 801         return NULL;
 802 }
 803
 804 /*
 805  * Free an option memory block.
 806  */
 807 void sock_kfree_s(struct sock *sk, void *mem, int size)
 808 {
 809         kfree(mem);
 810         atomic_sub(size, &sk->sk_omem_alloc);
 811 }
 812
 813 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
 814    I think, these locks should be removed for datagram sockets.
 815  */
 816 static long sock_wait_for_wmem(struct sock * sk, long timeo)
 817 {
 818         DEFINE_WAIT(wait);
 819
 820         clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
 821         for (;;) {
 822                 if (!timeo)
 823                         break;
 824                 if (signal_pending(current))
 825                         break;
 826                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
 827                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
 828                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
 829                         break;
 830                 if (sk->sk_shutdown & SEND_SHUTDOWN)
 831                         break;
 832                 if (sk->sk_err)
 833                         break;
 834                 timeo = schedule_timeout(timeo);
 835         }
 836         finish_wait(sk->sk_sleep, &wait);
 837         return timeo;
 838 }
 839
 840
 841 /*
 842  *      Generic send/receive buffer handlers
 843  */
 844
 845 static struct sk_buff *sock_alloc_send_pskb(struct sock *sk,
 846                                             unsigned long header_len,
 847                                             unsigned long data_len,
 848                                             int noblock, int *errcode)
 849 {
 850         struct sk_buff *skb;
 851         unsigned int gfp_mask;
 852         long timeo;
 853         int err;
 854
 855         gfp_mask = sk->sk_allocation;
 856         if (gfp_mask & __GFP_WAIT)
 857                 gfp_mask |= __GFP_REPEAT;
 858
 859         timeo = sock_sndtimeo(sk, noblock);
 860         while (1) {
 861                 err = sock_error(sk);
 862                 if (err != 0)
 863                         goto failure;
 864
 865                 err = -EPIPE;
 866                 if (sk->sk_shutdown & SEND_SHUTDOWN)
 867                         goto failure;
 868
 869                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
 870                         skb = alloc_skb(header_len, sk->sk_allocation);
 871                         if (skb) {
 872                                 int npages;
 873                                 int i;
 874
 875                                 /* No pages, we're done... */
 876                                 if (!data_len)
 877                                         break;
 878
 879                                 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
 880                                 skb->truesize += data_len;
 881                                 skb_shinfo(skb)->nr_frags = npages;
 882                                 for (i = 0; i < npages; i++) {
 883                                         struct page *page;
 884                                         skb_frag_t *frag;
 885
 886                                         page = alloc_pages(sk->sk_allocation, 0);
 887                                         if (!page) {
 888                                                 err = -ENOBUFS;
 889                                                 skb_shinfo(skb)->nr_frags = i;
 890                                                 kfree_skb(skb);
 891                                                 goto failure;
 892                                         }
 893
 894                                         frag = &skb_shinfo(skb)->frags[i];
 895                                         frag->page = page;
 896                                         frag->page_offset = 0;
 897                                         frag->size = (data_len >= PAGE_SIZE ?
 898                                                       PAGE_SIZE :
 899                                                       data_len);
 900                                         data_len -= PAGE_SIZE;
 901                                 }
 902
 903                                 /* Full success... */
 904                                 break;
 905                         }
 906                         err = -ENOBUFS;
 907                         goto failure;
 908                 }
 909                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
 910                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
 911                 err = -EAGAIN;
 912                 if (!timeo)
 913                         goto failure;
 914                 if (signal_pending(current))
 915                         goto interrupted;
 916                 timeo = sock_wait_for_wmem(sk, timeo);
 917         }
 918
 919         skb_set_owner_w(skb, sk);
 920         return skb;
 921
 922 interrupted:
 923         err = sock_intr_errno(timeo);
 924 failure:
 925         *errcode = err;
 926         return NULL;
 927 }
 928
 929 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
 930                                     int noblock, int *errcode)
 931 {
 932         return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
 933 }
 934
 935 static void __lock_sock(struct sock *sk)
 936 {
 937         DEFINE_WAIT(wait);
 938
 939         for(;;) {
 940                 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
 941                                         TASK_UNINTERRUPTIBLE);
 942                 spin_unlock_bh(&sk->sk_lock.slock);
 943                 schedule();
 944                 spin_lock_bh(&sk->sk_lock.slock);
 945                 if(!sock_owned_by_user(sk))
 946                         break;
 947         }
 948         finish_wait(&sk->sk_lock.wq, &wait);
 949 }
 950
 951 static void __release_sock(struct sock *sk)
 952 {
 953         struct sk_buff *skb = sk->sk_backlog.head;
 954
 955         do {
 956                 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
 957                 bh_unlock_sock(sk);
 958
 959                 do {
 960                         struct sk_buff *next = skb->next;
 961
 962                         skb->next = NULL;
 963                         sk->sk_backlog_rcv(sk, skb);
 964
 965                         /*
 966                          * We are in process context here with softirqs
 967                          * disabled, use cond_resched_softirq() to preempt.
 968                          * This is safe to do because we've taken the backlog
 969                          * queue private:
 970                          */
 971                         cond_resched_softirq();
 972
 973                         skb = next;
 974                 } while (skb != NULL);
 975
 976                 bh_lock_sock(sk);
 977         } while((skb = sk->sk_backlog.head) != NULL);
 978 }
 979
 980 /**
 981  * sk_wait_data - wait for data to arrive at sk_receive_queue
 982  * @sk:    sock to wait on
 983  * @timeo: for how long
 984  *
 985  * Now socket state including sk->sk_err is changed only under lock,
 986  * hence we may omit checks after joining wait queue.
 987  * We check receive queue before schedule() only as optimization;
 988  * it is very likely that release_sock() added new data.
 989  */
 990 int sk_wait_data(struct sock *sk, long *timeo)
 991 {
 992         int rc;
 993         DEFINE_WAIT(wait);
 994
 995         prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
 996         set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
 997         rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
 998         clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
 999         finish_wait(sk->sk_sleep, &wait);
1000         return rc;
1001 }
1002
1003 EXPORT_SYMBOL(sk_wait_data);
1004
1005 /*
1006  * Set of default routines for initialising struct proto_ops when
1007  * the protocol does not support a particular function. In certain
1008  * cases where it makes no sense for a protocol to have a "do nothing"
1009  * function, some default processing is provided.
1010  */
1011
1012 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1013 {
1014         return -EOPNOTSUPP;
1015 }
1016
1017 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1018                     int len, int flags)
1019 {
1020         return -EOPNOTSUPP;
1021 }
1022
1023 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1024 {
1025         return -EOPNOTSUPP;
1026 }
1027
1028 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1029 {
1030         return -EOPNOTSUPP;
1031 }
1032
1033 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1034                     int *len, int peer)
1035 {
1036         return -EOPNOTSUPP;
1037 }
1038
1039 unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt)
1040 {
1041         return 0;
1042 }
1043
1044 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1045 {
1046         return -EOPNOTSUPP;
1047 }
1048
1049 int sock_no_listen(struct socket *sock, int backlog)
1050 {
1051         return -EOPNOTSUPP;
1052 }
1053
1054 int sock_no_shutdown(struct socket *sock, int how)
1055 {
1056         return -EOPNOTSUPP;
1057 }
1058
1059 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1060                     char __user *optval, int optlen)
1061 {
1062         return -EOPNOTSUPP;
1063 }
1064
1065 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1066                     char __user *optval, int __user *optlen)
1067 {
1068         return -EOPNOTSUPP;
1069 }
1070
1071 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1072                     size_t len)
1073 {
1074         return -EOPNOTSUPP;
1075 }
1076
1077 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1078                     size_t len, int flags)
1079 {
1080         return -EOPNOTSUPP;
1081 }
1082
1083 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1084 {
1085         /* Mirror missing mmap method error code */
1086         return -ENODEV;
1087 }
1088
1089 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1090 {
1091         ssize_t res;
1092         struct msghdr msg = {.msg_flags = flags};
1093         struct kvec iov;
1094         char *kaddr = kmap(page);
1095         iov.iov_base = kaddr + offset;
1096         iov.iov_len = size;
1097         res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1098         kunmap(page);
1099         return res;
1100 }
1101
1102 /*
1103  *      Default Socket Callbacks
1104  */
1105
1106 static void sock_def_wakeup(struct sock *sk)
1107 {
1108         read_lock(&sk->sk_callback_lock);
1109         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1110                 wake_up_interruptible_all(sk->sk_sleep);
1111         read_unlock(&sk->sk_callback_lock);
1112 }
1113
1114 static void sock_def_error_report(struct sock *sk)
1115 {
1116         read_lock(&sk->sk_callback_lock);
1117         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1118                 wake_up_interruptible(sk->sk_sleep);
1119         sk_wake_async(sk,0,POLL_ERR);
1120         read_unlock(&sk->sk_callback_lock);
1121 }
1122
1123 static void sock_def_readable(struct sock *sk, int len)
1124 {
1125         read_lock(&sk->sk_callback_lock);
1126         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1127                 wake_up_interruptible(sk->sk_sleep);
1128         sk_wake_async(sk,1,POLL_IN);
1129         read_unlock(&sk->sk_callback_lock);
1130 }
1131
1132 static void sock_def_write_space(struct sock *sk)
1133 {
1134         read_lock(&sk->sk_callback_lock);
1135
1136         /* Do not wake up a writer until he can make "significant"
1137          * progress.  --DaveM
1138          */
1139         if((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1140                 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1141                         wake_up_interruptible(sk->sk_sleep);
1142
1143                 /* Should agree with poll, otherwise some programs break */
1144                 if (sock_writeable(sk))
1145                         sk_wake_async(sk, 2, POLL_OUT);
1146         }
1147
1148         read_unlock(&sk->sk_callback_lock);
1149 }
1150
1151 static void sock_def_destruct(struct sock *sk)
1152 {
1153         if (sk->sk_protinfo)
1154                 kfree(sk->sk_protinfo);
1155 }
1156
1157 void sk_send_sigurg(struct sock *sk)
1158 {
1159         if (sk->sk_socket && sk->sk_socket->file)
1160                 if (send_sigurg(&sk->sk_socket->file->f_owner))
1161                         sk_wake_async(sk, 3, POLL_PRI);
1162 }
1163
1164 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1165                     unsigned long expires)
1166 {
1167         if (!mod_timer(timer, expires))
1168                 sock_hold(sk);
1169 }
1170
1171 EXPORT_SYMBOL(sk_reset_timer);
1172
1173 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1174 {
1175         if (timer_pending(timer) && del_timer(timer))
1176                 __sock_put(sk);
1177 }
1178
1179 EXPORT_SYMBOL(sk_stop_timer);
1180
1181 void sock_init_data(struct socket *sock, struct sock *sk)
1182 {
1183         skb_queue_head_init(&sk->sk_receive_queue);
1184         skb_queue_head_init(&sk->sk_write_queue);
1185         skb_queue_head_init(&sk->sk_error_queue);
1186
1187         sk->sk_send_head        =       NULL;
1188
1189         init_timer(&sk->sk_timer);
1190
1191         sk->sk_allocation       =       GFP_KERNEL;
1192         sk->sk_rcvbuf           =       sysctl_rmem_default;
1193         sk->sk_sndbuf           =       sysctl_wmem_default;
1194         sk->sk_state            =       TCP_CLOSE;
1195         sk->sk_socket           =       sock;
1196
1197         sock_set_flag(sk, SOCK_ZAPPED);
1198
1199         if(sock)
1200         {
1201                 sk->sk_type     =       sock->type;
1202                 sk->sk_sleep    =       &sock->wait;
1203                 sock->sk        =       sk;
1204         } else
1205                 sk->sk_sleep    =       NULL;
1206
1207         rwlock_init(&sk->sk_dst_lock);
1208         rwlock_init(&sk->sk_callback_lock);
1209
1210         sk->sk_state_change     =       sock_def_wakeup;
1211         sk->sk_data_ready       =       sock_def_readable;
1212         sk->sk_write_space      =       sock_def_write_space;
1213         sk->sk_error_report     =       sock_def_error_report;
1214         sk->sk_destruct         =       sock_def_destruct;
1215
1216         sk->sk_sndmsg_page      =       NULL;
1217         sk->sk_sndmsg_off       =       0;
1218
1219         sk->sk_peercred.pid     =       0;
1220         sk->sk_peercred.uid     =       -1;
1221         sk->sk_peercred.gid     =       -1;
1222         sk->sk_write_pending    =       0;
1223         sk->sk_rcvlowat         =       1;
1224         sk->sk_rcvtimeo         =       MAX_SCHEDULE_TIMEOUT;
1225         sk->sk_sndtimeo         =       MAX_SCHEDULE_TIMEOUT;
1226
1227         sk->sk_stamp.tv_sec     = -1L;
1228         sk->sk_stamp.tv_usec    = -1L;
1229
1230         atomic_set(&sk->sk_refcnt, 1);
1231 }
1232
1233 void fastcall lock_sock(struct sock *sk)
1234 {
1235         might_sleep();
1236         spin_lock_bh(&(sk->sk_lock.slock));
1237         if (sk->sk_lock.owner)
1238                 __lock_sock(sk);
1239         sk->sk_lock.owner = (void *)1;
1240         spin_unlock_bh(&(sk->sk_lock.slock));
1241 }
1242
1243 EXPORT_SYMBOL(lock_sock);
1244
1245 void fastcall release_sock(struct sock *sk)
1246 {
1247         spin_lock_bh(&(sk->sk_lock.slock));
1248         if (sk->sk_backlog.tail)
1249                 __release_sock(sk);
1250         sk->sk_lock.owner = NULL;
1251         if (waitqueue_active(&(sk->sk_lock.wq)))
1252                 wake_up(&(sk->sk_lock.wq));
1253         spin_unlock_bh(&(sk->sk_lock.slock));
1254 }
1255 EXPORT_SYMBOL(release_sock);
1256
1257 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1258 {
1259         if (!sock_flag(sk, SOCK_TIMESTAMP))
1260                 sock_enable_timestamp(sk);
1261         if (sk->sk_stamp.tv_sec == -1)
1262                 return -ENOENT;
1263         if (sk->sk_stamp.tv_sec == 0)
1264                 do_gettimeofday(&sk->sk_stamp);
1265         return copy_to_user(userstamp, &sk->sk_stamp, sizeof(struct timeval)) ?
1266                 -EFAULT : 0;
1267 }
1268 EXPORT_SYMBOL(sock_get_timestamp);
1269
1270 void sock_enable_timestamp(struct sock *sk)
1271 {
1272         if (!sock_flag(sk, SOCK_TIMESTAMP)) {
1273                 sock_set_flag(sk, SOCK_TIMESTAMP);
1274                 net_enable_timestamp();
1275         }
1276 }
1277 EXPORT_SYMBOL(sock_enable_timestamp);
1278
1279 /*
1280  *      Get a socket option on an socket.
1281  *
1282  *      FIX: POSIX 1003.1g is very ambiguous here. It states that
1283  *      asynchronous errors should be reported by getsockopt. We assume
1284  *      this means if you specify SO_ERROR (otherwise whats the point of it).
1285  */
1286 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1287                            char __user *optval, int __user *optlen)
1288 {
1289         struct sock *sk = sock->sk;
1290
1291         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1292 }
1293
1294 EXPORT_SYMBOL(sock_common_getsockopt);
1295
1296 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1297                         struct msghdr *msg, size_t size, int flags)
1298 {
1299         struct sock *sk = sock->sk;
1300         int addr_len = 0;
1301         int err;
1302
1303         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
1304                                    flags & ~MSG_DONTWAIT, &addr_len);
1305         if (err >= 0)
1306                 msg->msg_namelen = addr_len;
1307         return err;
1308 }
1309
1310 EXPORT_SYMBOL(sock_common_recvmsg);
1311
1312 /*
1313  *      Set socket options on an inet socket.
1314  */
1315 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1316                            char __user *optval, int optlen)
1317 {
1318         struct sock *sk = sock->sk;
1319
1320         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1321 }
1322
1323 EXPORT_SYMBOL(sock_common_setsockopt);
1324
1325 void sk_common_release(struct sock *sk)
1326 {
1327         if (sk->sk_prot->destroy)
1328                 sk->sk_prot->destroy(sk);
1329
1330         /*
1331          * Observation: when sock_common_release is called, processes have
1332          * no access to socket. But net still has.
1333          * Step one, detach it from networking:
1334          *
1335          * A. Remove from hash tables.
1336          */
1337
1338         sk->sk_prot->unhash(sk);
1339
1340         /*
1341          * In this point socket cannot receive new packets, but it is possible
1342          * that some packets are in flight because some CPU runs receiver and
1343          * did hash table lookup before we unhashed socket. They will achieve
1344          * receive queue and will be purged by socket destructor.
1345          *
1346          * Also we still have packets pending on receive queue and probably,
1347          * our own packets waiting in device queues. sock_destroy will drain
1348          * receive queue, but transmitted packets will delay socket destruction
1349          * until the last reference will be released.
1350          */
1351
1352         sock_orphan(sk);
1353
1354         xfrm_sk_free_policy(sk);
1355
1356 #ifdef INET_REFCNT_DEBUG
1357         if (atomic_read(&sk->sk_refcnt) != 1)
1358                 printk(KERN_DEBUG "Destruction of the socket %p delayed, c=%d\n",
1359                        sk, atomic_read(&sk->sk_refcnt));
1360 #endif
1361         sock_put(sk);
1362 }
1363
1364 EXPORT_SYMBOL(sk_common_release);
1365
1366 static DEFINE_RWLOCK(proto_list_lock);
1367 static LIST_HEAD(proto_list);
1368
1369 int proto_register(struct proto *prot, int alloc_slab)
1370 {
1371         char *request_sock_slab_name;
1372         int rc = -ENOBUFS;
1373
1374         if (alloc_slab) {
1375                 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
1376                                                SLAB_HWCACHE_ALIGN, NULL, NULL);
1377
1378                 if (prot->slab == NULL) {
1379                         printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
1380                                prot->name);
1381                         goto out;
1382                 }
1383
1384                 if (prot->rsk_prot != NULL) {
1385                         static const char mask[] = "request_sock_%s";
1386
1387                         request_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
1388                         if (request_sock_slab_name == NULL)
1389                                 goto out_free_sock_slab;
1390
1391                         sprintf(request_sock_slab_name, mask, prot->name);
1392                         prot->rsk_prot->slab = kmem_cache_create(request_sock_slab_name,
1393                                                                  prot->rsk_prot->obj_size, 0,
1394                                                                  SLAB_HWCACHE_ALIGN, NULL, NULL);
1395
1396                         if (prot->rsk_prot->slab == NULL) {
1397                                 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
1398                                        prot->name);
1399                                 goto out_free_request_sock_slab_name;
1400                         }
1401                 }
1402         }
1403
1404         write_lock(&proto_list_lock);
1405         list_add(&prot->node, &proto_list);
1406         write_unlock(&proto_list_lock);
1407         rc = 0;
1408 out:
1409         return rc;
1410 out_free_request_sock_slab_name:
1411         kfree(request_sock_slab_name);
1412 out_free_sock_slab:
1413         kmem_cache_destroy(prot->slab);
1414         prot->slab = NULL;
1415         goto out;
1416 }
1417
1418 EXPORT_SYMBOL(proto_register);
1419
1420 void proto_unregister(struct proto *prot)
1421 {
1422         write_lock(&proto_list_lock);
1423
1424         if (prot->slab != NULL) {
1425                 kmem_cache_destroy(prot->slab);
1426                 prot->slab = NULL;
1427         }
1428
1429         if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
1430                 const char *name = kmem_cache_name(prot->rsk_prot->slab);
1431
1432                 kmem_cache_destroy(prot->rsk_prot->slab);
1433                 kfree(name);
1434                 prot->rsk_prot->slab = NULL;
1435         }
1436
1437         list_del(&prot->node);
1438         write_unlock(&proto_list_lock);
1439 }
1440
1441 EXPORT_SYMBOL(proto_unregister);
1442
1443 #ifdef CONFIG_PROC_FS
1444 static inline struct proto *__proto_head(void)
1445 {
1446         return list_entry(proto_list.next, struct proto, node);
1447 }
1448
1449 static inline struct proto *proto_head(void)
1450 {
1451         return list_empty(&proto_list) ? NULL : __proto_head();
1452 }
1453
1454 static inline struct proto *proto_next(struct proto *proto)
1455 {
1456         return proto->node.next == &proto_list ? NULL :
1457                 list_entry(proto->node.next, struct proto, node);
1458 }
1459
1460 static inline struct proto *proto_get_idx(loff_t pos)
1461 {
1462         struct proto *proto;
1463         loff_t i = 0;
1464
1465         list_for_each_entry(proto, &proto_list, node)
1466                 if (i++ == pos)
1467                         goto out;
1468
1469         proto = NULL;
1470 out:
1471         return proto;
1472 }
1473
1474 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
1475 {
1476         read_lock(&proto_list_lock);
1477         return *pos ? proto_get_idx(*pos - 1) : SEQ_START_TOKEN;
1478 }
1479
1480 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1481 {
1482         ++*pos;
1483         return v == SEQ_START_TOKEN ? proto_head() : proto_next(v);
1484 }
1485
1486 static void proto_seq_stop(struct seq_file *seq, void *v)
1487 {
1488         read_unlock(&proto_list_lock);
1489 }
1490
1491 static char proto_method_implemented(const void *method)
1492 {
1493         return method == NULL ? 'n' : 'y';
1494 }
1495
1496 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
1497 {
1498         seq_printf(seq, "%-9s %4u %6d  %6d   %-3s %6u   %-3s  %-10s "
1499                         "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
1500                    proto->name,
1501                    proto->obj_size,
1502                    proto->sockets_allocated != NULL ? atomic_read(proto->sockets_allocated) : -1,
1503                    proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
1504                    proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
1505                    proto->max_header,
1506                    proto->slab == NULL ? "no" : "yes",
1507                    module_name(proto->owner),
1508                    proto_method_implemented(proto->close),
1509                    proto_method_implemented(proto->connect),
1510                    proto_method_implemented(proto->disconnect),
1511                    proto_method_implemented(proto->accept),
1512                    proto_method_implemented(proto->ioctl),
1513                    proto_method_implemented(proto->init),
1514                    proto_method_implemented(proto->destroy),
1515                    proto_method_implemented(proto->shutdown),
1516                    proto_method_implemented(proto->setsockopt),
1517                    proto_method_implemented(proto->getsockopt),
1518                    proto_method_implemented(proto->sendmsg),
1519                    proto_method_implemented(proto->recvmsg),
1520                    proto_method_implemented(proto->sendpage),
1521                    proto_method_implemented(proto->bind),
1522                    proto_method_implemented(proto->backlog_rcv),
1523                    proto_method_implemented(proto->hash),
1524                    proto_method_implemented(proto->unhash),
1525                    proto_method_implemented(proto->get_port),
1526                    proto_method_implemented(proto->enter_memory_pressure));
1527 }
1528
1529 static int proto_seq_show(struct seq_file *seq, void *v)
1530 {
1531         if (v == SEQ_START_TOKEN)
1532                 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
1533                            "protocol",
1534                            "size",
1535                            "sockets",
1536                            "memory",
1537                            "press",
1538                            "maxhdr",
1539                            "slab",
1540                            "module",
1541                            "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
1542         else
1543                 proto_seq_printf(seq, v);
1544         return 0;
1545 }
1546
1547 static struct seq_operations proto_seq_ops = {
1548         .start  = proto_seq_start,
1549         .next   = proto_seq_next,
1550         .stop   = proto_seq_stop,
1551         .show   = proto_seq_show,
1552 };
1553
1554 static int proto_seq_open(struct inode *inode, struct file *file)
1555 {
1556         return seq_open(file, &proto_seq_ops);
1557 }
1558
1559 static struct file_operations proto_seq_fops = {
1560         .owner          = THIS_MODULE,
1561         .open           = proto_seq_open,
1562         .read           = seq_read,
1563         .llseek         = seq_lseek,
1564         .release        = seq_release,
1565 };
1566
1567 static int __init proto_init(void)
1568 {
1569         /* register /proc/net/protocols */
1570         return proc_net_fops_create("protocols", S_IRUGO, &proto_seq_fops) == NULL ? -ENOBUFS : 0;
1571 }
1572
1573 subsys_initcall(proto_init);
1574
1575 #endif /* PROC_FS */
1576
1577 EXPORT_SYMBOL(sk_alloc);
1578 EXPORT_SYMBOL(sk_free);
1579 EXPORT_SYMBOL(sk_send_sigurg);
1580 EXPORT_SYMBOL(sock_alloc_send_skb);
1581 EXPORT_SYMBOL(sock_init_data);
1582 EXPORT_SYMBOL(sock_kfree_s);
1583 EXPORT_SYMBOL(sock_kmalloc);
1584 EXPORT_SYMBOL(sock_no_accept);
1585 EXPORT_SYMBOL(sock_no_bind);
1586 EXPORT_SYMBOL(sock_no_connect);
1587 EXPORT_SYMBOL(sock_no_getname);
1588 EXPORT_SYMBOL(sock_no_getsockopt);
1589 EXPORT_SYMBOL(sock_no_ioctl);
1590 EXPORT_SYMBOL(sock_no_listen);
1591 EXPORT_SYMBOL(sock_no_mmap);
1592 EXPORT_SYMBOL(sock_no_poll);
1593 EXPORT_SYMBOL(sock_no_recvmsg);
1594 EXPORT_SYMBOL(sock_no_sendmsg);
1595 EXPORT_SYMBOL(sock_no_sendpage);
1596 EXPORT_SYMBOL(sock_no_setsockopt);
1597 EXPORT_SYMBOL(sock_no_shutdown);
1598 EXPORT_SYMBOL(sock_no_socketpair);
1599 EXPORT_SYMBOL(sock_rfree);
1600 EXPORT_SYMBOL(sock_setsockopt);
1601 EXPORT_SYMBOL(sock_wfree);
1602 EXPORT_SYMBOL(sock_wmalloc);
1603 EXPORT_SYMBOL(sock_i_uid);
1604 EXPORT_SYMBOL(sock_i_ino);
1605 #ifdef CONFIG_SYSCTL
1606 EXPORT_SYMBOL(sysctl_optmem_max);
1607 EXPORT_SYMBOL(sysctl_rmem_max);
1608 EXPORT_SYMBOL(sysctl_wmem_max);
1609 #endif