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 set_sndbuf:
 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_SNDBUFFORCE:
 278                         if (!capable(CAP_NET_ADMIN)) {
 279                                 ret = -EPERM;
 280                                 break;
 281                         }
 282                         goto set_sndbuf;
 283
 284                 case SO_RCVBUF:
 285                         /* Don't error on this BSD doesn't and if you think
 286                            about it this is right. Otherwise apps have to
 287                            play 'guess the biggest size' games. RCVBUF/SNDBUF
 288                            are treated in BSD as hints */
 289
 290                         if (val > sysctl_rmem_max)
 291                                 val = sysctl_rmem_max;
 292 set_rcvbuf:
 293                         sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
 294                         /* FIXME: is this lower bound the right one? */
 295                         if ((val * 2) < SOCK_MIN_RCVBUF)
 296                                 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
 297                         else
 298                                 sk->sk_rcvbuf = val * 2;
 299                         break;
 300
 301                 case SO_RCVBUFFORCE:
 302                         if (!capable(CAP_NET_ADMIN)) {
 303                                 ret = -EPERM;
 304                                 break;
 305                         }
 306                         goto set_rcvbuf;
 307
 308                 case SO_KEEPALIVE:
 309 #ifdef CONFIG_INET
 310                         if (sk->sk_protocol == IPPROTO_TCP)
 311                                 tcp_set_keepalive(sk, valbool);
 312 #endif
 313                         sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
 314                         break;
 315
 316                 case SO_OOBINLINE:
 317                         sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
 318                         break;
 319
 320                 case SO_NO_CHECK:
 321                         sk->sk_no_check = valbool;
 322                         break;
 323
 324                 case SO_PRIORITY:
 325                         if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
 326                                 sk->sk_priority = val;
 327                         else
 328                                 ret = -EPERM;
 329                         break;
 330
 331                 case SO_LINGER:
 332                         if(optlen<sizeof(ling)) {
 333                                 ret = -EINVAL;  /* 1003.1g */
 334                                 break;
 335                         }
 336                         if (copy_from_user(&ling,optval,sizeof(ling))) {
 337                                 ret = -EFAULT;
 338                                 break;
 339                         }
 340                         if (!ling.l_onoff)
 341                                 sock_reset_flag(sk, SOCK_LINGER);
 342                         else {
 343 #if (BITS_PER_LONG == 32)
 344                                 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
 345                                         sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
 346                                 else
 347 #endif
 348                                         sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
 349                                 sock_set_flag(sk, SOCK_LINGER);
 350                         }
 351                         break;
 352
 353                 case SO_BSDCOMPAT:
 354                         sock_warn_obsolete_bsdism("setsockopt");
 355                         break;
 356
 357                 case SO_PASSCRED:
 358                         if (valbool)
 359                                 set_bit(SOCK_PASSCRED, &sock->flags);
 360                         else
 361                                 clear_bit(SOCK_PASSCRED, &sock->flags);
 362                         break;
 363
 364                 case SO_TIMESTAMP:
 365                         if (valbool)  {
 366                                 sock_set_flag(sk, SOCK_RCVTSTAMP);
 367                                 sock_enable_timestamp(sk);
 368                         } else
 369                                 sock_reset_flag(sk, SOCK_RCVTSTAMP);
 370                         break;
 371
 372                 case SO_RCVLOWAT:
 373                         if (val < 0)
 374                                 val = INT_MAX;
 375                         sk->sk_rcvlowat = val ? : 1;
 376                         break;
 377
 378                 case SO_RCVTIMEO:
 379                         ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
 380                         break;
 381
 382                 case SO_SNDTIMEO:
 383                         ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
 384                         break;
 385
 386 #ifdef CONFIG_NETDEVICES
 387                 case SO_BINDTODEVICE:
 388                 {
 389                         char devname[IFNAMSIZ];
 390
 391                         /* Sorry... */
 392                         if (!capable(CAP_NET_RAW)) {
 393                                 ret = -EPERM;
 394                                 break;
 395                         }
 396
 397                         /* Bind this socket to a particular device like "eth0",
 398                          * as specified in the passed interface name. If the
 399                          * name is "" or the option length is zero the socket
 400                          * is not bound.
 401                          */
 402
 403                         if (!valbool) {
 404                                 sk->sk_bound_dev_if = 0;
 405                         } else {
 406                                 if (optlen > IFNAMSIZ)
 407                                         optlen = IFNAMSIZ;
 408                                 if (copy_from_user(devname, optval, optlen)) {
 409                                         ret = -EFAULT;
 410                                         break;
 411                                 }
 412
 413                                 /* Remove any cached route for this socket. */
 414                                 sk_dst_reset(sk);
 415
 416                                 if (devname[0] == '\0') {
 417                                         sk->sk_bound_dev_if = 0;
 418                                 } else {
 419                                         struct net_device *dev = dev_get_by_name(devname);
 420                                         if (!dev) {
 421                                                 ret = -ENODEV;
 422                                                 break;
 423                                         }
 424                                         sk->sk_bound_dev_if = dev->ifindex;
 425                                         dev_put(dev);
 426                                 }
 427                         }
 428                         break;
 429                 }
 430 #endif
 431
 432
 433                 case SO_ATTACH_FILTER:
 434                         ret = -EINVAL;
 435                         if (optlen == sizeof(struct sock_fprog)) {
 436                                 struct sock_fprog fprog;
 437
 438                                 ret = -EFAULT;
 439                                 if (copy_from_user(&fprog, optval, sizeof(fprog)))
 440                                         break;
 441
 442                                 ret = sk_attach_filter(&fprog, sk);
 443                         }
 444                         break;
 445
 446                 case SO_DETACH_FILTER:
 447                         spin_lock_bh(&sk->sk_lock.slock);
 448                         filter = sk->sk_filter;
 449                         if (filter) {
 450                                 sk->sk_filter = NULL;
 451                                 spin_unlock_bh(&sk->sk_lock.slock);
 452                                 sk_filter_release(sk, filter);
 453                                 break;
 454                         }
 455                         spin_unlock_bh(&sk->sk_lock.slock);
 456                         ret = -ENONET;
 457                         break;
 458
 459                 /* We implement the SO_SNDLOWAT etc to
 460                    not be settable (1003.1g 5.3) */
 461                 default:
 462                         ret = -ENOPROTOOPT;
 463                         break;
 464         }
 465         release_sock(sk);
 466         return ret;
 467 }
 468
 469
 470 int sock_getsockopt(struct socket *sock, int level, int optname,
 471                     char __user *optval, int __user *optlen)
 472 {
 473         struct sock *sk = sock->sk;
 474
 475         union
 476         {
 477                 int val;
 478                 struct linger ling;
 479                 struct timeval tm;
 480         } v;
 481
 482         unsigned int lv = sizeof(int);
 483         int len;
 484
 485         if(get_user(len,optlen))
 486                 return -EFAULT;
 487         if(len < 0)
 488                 return -EINVAL;
 489
 490         switch(optname)
 491         {
 492                 case SO_DEBUG:
 493                         v.val = sock_flag(sk, SOCK_DBG);
 494                         break;
 495
 496                 case SO_DONTROUTE:
 497                         v.val = sock_flag(sk, SOCK_LOCALROUTE);
 498                         break;
 499
 500                 case SO_BROADCAST:
 501                         v.val = !!sock_flag(sk, SOCK_BROADCAST);
 502                         break;
 503
 504                 case SO_SNDBUF:
 505                         v.val = sk->sk_sndbuf;
 506                         break;
 507
 508                 case SO_RCVBUF:
 509                         v.val = sk->sk_rcvbuf;
 510                         break;
 511
 512                 case SO_REUSEADDR:
 513                         v.val = sk->sk_reuse;
 514                         break;
 515
 516                 case SO_KEEPALIVE:
 517                         v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
 518                         break;
 519
 520                 case SO_TYPE:
 521                         v.val = sk->sk_type;
 522                         break;
 523
 524                 case SO_ERROR:
 525                         v.val = -sock_error(sk);
 526                         if(v.val==0)
 527                                 v.val = xchg(&sk->sk_err_soft, 0);
 528                         break;
 529
 530                 case SO_OOBINLINE:
 531                         v.val = !!sock_flag(sk, SOCK_URGINLINE);
 532                         break;
 533
 534                 case SO_NO_CHECK:
 535                         v.val = sk->sk_no_check;
 536                         break;
 537
 538                 case SO_PRIORITY:
 539                         v.val = sk->sk_priority;
 540                         break;
 541
 542                 case SO_LINGER:
 543                         lv              = sizeof(v.ling);
 544                         v.ling.l_onoff  = !!sock_flag(sk, SOCK_LINGER);
 545                         v.ling.l_linger = sk->sk_lingertime / HZ;
 546                         break;
 547
 548                 case SO_BSDCOMPAT:
 549                         sock_warn_obsolete_bsdism("getsockopt");
 550                         break;
 551
 552                 case SO_TIMESTAMP:
 553                         v.val = sock_flag(sk, SOCK_RCVTSTAMP);
 554                         break;
 555
 556                 case SO_RCVTIMEO:
 557                         lv=sizeof(struct timeval);
 558                         if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
 559                                 v.tm.tv_sec = 0;
 560                                 v.tm.tv_usec = 0;
 561                         } else {
 562                                 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
 563                                 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
 564                         }
 565                         break;
 566
 567                 case SO_SNDTIMEO:
 568                         lv=sizeof(struct timeval);
 569                         if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
 570                                 v.tm.tv_sec = 0;
 571                                 v.tm.tv_usec = 0;
 572                         } else {
 573                                 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
 574                                 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
 575                         }
 576                         break;
 577
 578                 case SO_RCVLOWAT:
 579                         v.val = sk->sk_rcvlowat;
 580                         break;
 581
 582                 case SO_SNDLOWAT:
 583                         v.val=1;
 584                         break;
 585
 586                 case SO_PASSCRED:
 587                         v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
 588                         break;
 589
 590                 case SO_PEERCRED:
 591                         if (len > sizeof(sk->sk_peercred))
 592                                 len = sizeof(sk->sk_peercred);
 593                         if (copy_to_user(optval, &sk->sk_peercred, len))
 594                                 return -EFAULT;
 595                         goto lenout;
 596
 597                 case SO_PEERNAME:
 598                 {
 599                         char address[128];
 600
 601                         if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
 602                                 return -ENOTCONN;
 603                         if (lv < len)
 604                                 return -EINVAL;
 605                         if (copy_to_user(optval, address, len))
 606                                 return -EFAULT;
 607                         goto lenout;
 608                 }
 609
 610                 /* Dubious BSD thing... Probably nobody even uses it, but
 611                  * the UNIX standard wants it for whatever reason... -DaveM
 612                  */
 613                 case SO_ACCEPTCONN:
 614                         v.val = sk->sk_state == TCP_LISTEN;
 615                         break;
 616
 617                 case SO_PEERSEC:
 618                         return security_socket_getpeersec(sock, optval, optlen, len);
 619
 620                 default:
 621                         return(-ENOPROTOOPT);
 622         }
 623         if (len > lv)
 624                 len = lv;
 625         if (copy_to_user(optval, &v, len))
 626                 return -EFAULT;
 627 lenout:
 628         if (put_user(len, optlen))
 629                 return -EFAULT;
 630         return 0;
 631 }
 632
 633 /**
 634  *      sk_alloc - All socket objects are allocated here
 635  *      @family: protocol family
 636  *      @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
 637  *      @prot: struct proto associated with this new sock instance
 638  *      @zero_it: if we should zero the newly allocated sock
 639  */
 640 struct sock *sk_alloc(int family, unsigned int __nocast priority,
 641                       struct proto *prot, int zero_it)
 642 {
 643         struct sock *sk = NULL;
 644         kmem_cache_t *slab = prot->slab;
 645
 646         if (slab != NULL)
 647                 sk = kmem_cache_alloc(slab, priority);
 648         else
 649                 sk = kmalloc(prot->obj_size, priority);
 650
 651         if (sk) {
 652                 if (zero_it) {
 653                         memset(sk, 0, prot->obj_size);
 654                         sk->sk_family = family;
 655                         /*
 656                          * See comment in struct sock definition to understand
 657                          * why we need sk_prot_creator -acme
 658                          */
 659                         sk->sk_prot = sk->sk_prot_creator = prot;
 660                         sock_lock_init(sk);
 661                 }
 662
 663                 if (security_sk_alloc(sk, family, priority))
 664                         goto out_free;
 665
 666                 if (!try_module_get(prot->owner))
 667                         goto out_free;
 668         }
 669         return sk;
 670
 671 out_free:
 672         if (slab != NULL)
 673                 kmem_cache_free(slab, sk);
 674         else
 675                 kfree(sk);
 676         return NULL;
 677 }
 678
 679 void sk_free(struct sock *sk)
 680 {
 681         struct sk_filter *filter;
 682         struct module *owner = sk->sk_prot_creator->owner;
 683
 684         if (sk->sk_destruct)
 685                 sk->sk_destruct(sk);
 686
 687         filter = sk->sk_filter;
 688         if (filter) {
 689                 sk_filter_release(sk, filter);
 690                 sk->sk_filter = NULL;
 691         }
 692
 693         sock_disable_timestamp(sk);
 694
 695         if (atomic_read(&sk->sk_omem_alloc))
 696                 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
 697                        __FUNCTION__, atomic_read(&sk->sk_omem_alloc));
 698
 699         security_sk_free(sk);
 700         if (sk->sk_prot_creator->slab != NULL)
 701                 kmem_cache_free(sk->sk_prot_creator->slab, sk);
 702         else
 703                 kfree(sk);
 704         module_put(owner);
 705 }
 706
 707 struct sock *sk_clone(const struct sock *sk, const unsigned int __nocast priority)
 708 {
 709         struct sock *newsk = sk_alloc(sk->sk_family, priority, sk->sk_prot, 0);
 710
 711         if (newsk != NULL) {
 712                 struct sk_filter *filter;
 713
 714                 memcpy(newsk, sk, sk->sk_prot->obj_size);
 715
 716                 /* SANITY */
 717                 sk_node_init(&newsk->sk_node);
 718                 sock_lock_init(newsk);
 719                 bh_lock_sock(newsk);
 720
 721                 atomic_set(&newsk->sk_rmem_alloc, 0);
 722                 atomic_set(&newsk->sk_wmem_alloc, 0);
 723                 atomic_set(&newsk->sk_omem_alloc, 0);
 724                 skb_queue_head_init(&newsk->sk_receive_queue);
 725                 skb_queue_head_init(&newsk->sk_write_queue);
 726
 727                 rwlock_init(&newsk->sk_dst_lock);
 728                 rwlock_init(&newsk->sk_callback_lock);
 729
 730                 newsk->sk_dst_cache     = NULL;
 731                 newsk->sk_wmem_queued   = 0;
 732                 newsk->sk_forward_alloc = 0;
 733                 newsk->sk_send_head     = NULL;
 734                 newsk->sk_backlog.head  = newsk->sk_backlog.tail = NULL;
 735                 newsk->sk_userlocks     = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
 736
 737                 sock_reset_flag(newsk, SOCK_DONE);
 738                 skb_queue_head_init(&newsk->sk_error_queue);
 739
 740                 filter = newsk->sk_filter;
 741                 if (filter != NULL)
 742                         sk_filter_charge(newsk, filter);
 743
 744                 if (unlikely(xfrm_sk_clone_policy(newsk))) {
 745                         /* It is still raw copy of parent, so invalidate
 746                          * destructor and make plain sk_free() */
 747                         newsk->sk_destruct = NULL;
 748                         sk_free(newsk);
 749                         newsk = NULL;
 750                         goto out;
 751                 }
 752
 753                 newsk->sk_err      = 0;
 754                 newsk->sk_priority = 0;
 755                 atomic_set(&newsk->sk_refcnt, 2);
 756
 757                 /*
 758                  * Increment the counter in the same struct proto as the master
 759                  * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
 760                  * is the same as sk->sk_prot->socks, as this field was copied
 761                  * with memcpy).
 762                  *
 763                  * This _changes_ the previous behaviour, where
 764                  * tcp_create_openreq_child always was incrementing the
 765                  * equivalent to tcp_prot->socks (inet_sock_nr), so this have
 766                  * to be taken into account in all callers. -acme
 767                  */
 768                 sk_refcnt_debug_inc(newsk);
 769                 newsk->sk_socket = NULL;
 770                 newsk->sk_sleep  = NULL;
 771
 772                 if (newsk->sk_prot->sockets_allocated)
 773                         atomic_inc(newsk->sk_prot->sockets_allocated);
 774         }
 775 out:
 776         return newsk;
 777 }
 778
 779 EXPORT_SYMBOL_GPL(sk_clone);
 780
 781 void __init sk_init(void)
 782 {
 783         if (num_physpages <= 4096) {
 784                 sysctl_wmem_max = 32767;
 785                 sysctl_rmem_max = 32767;
 786                 sysctl_wmem_default = 32767;
 787                 sysctl_rmem_default = 32767;
 788         } else if (num_physpages >= 131072) {
 789                 sysctl_wmem_max = 131071;
 790                 sysctl_rmem_max = 131071;
 791         }
 792 }
 793
 794 /*
 795  *      Simple resource managers for sockets.
 796  */
 797
 798
 799 /*
 800  * Write buffer destructor automatically called from kfree_skb.
 801  */
 802 void sock_wfree(struct sk_buff *skb)
 803 {
 804         struct sock *sk = skb->sk;
 805
 806         /* In case it might be waiting for more memory. */
 807         atomic_sub(skb->truesize, &sk->sk_wmem_alloc);
 808         if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE))
 809                 sk->sk_write_space(sk);
 810         sock_put(sk);
 811 }
 812
 813 /*
 814  * Read buffer destructor automatically called from kfree_skb.
 815  */
 816 void sock_rfree(struct sk_buff *skb)
 817 {
 818         struct sock *sk = skb->sk;
 819
 820         atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
 821 }
 822
 823
 824 int sock_i_uid(struct sock *sk)
 825 {
 826         int uid;
 827
 828         read_lock(&sk->sk_callback_lock);
 829         uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
 830         read_unlock(&sk->sk_callback_lock);
 831         return uid;
 832 }
 833
 834 unsigned long sock_i_ino(struct sock *sk)
 835 {
 836         unsigned long ino;
 837
 838         read_lock(&sk->sk_callback_lock);
 839         ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
 840         read_unlock(&sk->sk_callback_lock);
 841         return ino;
 842 }
 843
 844 /*
 845  * Allocate a skb from the socket's send buffer.
 846  */
 847 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
 848                              unsigned int __nocast priority)
 849 {
 850         if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
 851                 struct sk_buff * skb = alloc_skb(size, priority);
 852                 if (skb) {
 853                         skb_set_owner_w(skb, sk);
 854                         return skb;
 855                 }
 856         }
 857         return NULL;
 858 }
 859
 860 /*
 861  * Allocate a skb from the socket's receive buffer.
 862  */
 863 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
 864                              unsigned int __nocast priority)
 865 {
 866         if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
 867                 struct sk_buff *skb = alloc_skb(size, priority);
 868                 if (skb) {
 869                         skb_set_owner_r(skb, sk);
 870                         return skb;
 871                 }
 872         }
 873         return NULL;
 874 }
 875
 876 /*
 877  * Allocate a memory block from the socket's option memory buffer.
 878  */
 879 void *sock_kmalloc(struct sock *sk, int size, unsigned int __nocast priority)
 880 {
 881         if ((unsigned)size <= sysctl_optmem_max &&
 882             atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
 883                 void *mem;
 884                 /* First do the add, to avoid the race if kmalloc
 885                  * might sleep.
 886                  */
 887                 atomic_add(size, &sk->sk_omem_alloc);
 888                 mem = kmalloc(size, priority);
 889                 if (mem)
 890                         return mem;
 891                 atomic_sub(size, &sk->sk_omem_alloc);
 892         }
 893         return NULL;
 894 }
 895
 896 /*
 897  * Free an option memory block.
 898  */
 899 void sock_kfree_s(struct sock *sk, void *mem, int size)
 900 {
 901         kfree(mem);
 902         atomic_sub(size, &sk->sk_omem_alloc);
 903 }
 904
 905 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
 906    I think, these locks should be removed for datagram sockets.
 907  */
 908 static long sock_wait_for_wmem(struct sock * sk, long timeo)
 909 {
 910         DEFINE_WAIT(wait);
 911
 912         clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
 913         for (;;) {
 914                 if (!timeo)
 915                         break;
 916                 if (signal_pending(current))
 917                         break;
 918                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
 919                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
 920                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
 921                         break;
 922                 if (sk->sk_shutdown & SEND_SHUTDOWN)
 923                         break;
 924                 if (sk->sk_err)
 925                         break;
 926                 timeo = schedule_timeout(timeo);
 927         }
 928         finish_wait(sk->sk_sleep, &wait);
 929         return timeo;
 930 }
 931
 932
 933 /*
 934  *      Generic send/receive buffer handlers
 935  */
 936
 937 static struct sk_buff *sock_alloc_send_pskb(struct sock *sk,
 938                                             unsigned long header_len,
 939                                             unsigned long data_len,
 940                                             int noblock, int *errcode)
 941 {
 942         struct sk_buff *skb;
 943         unsigned int gfp_mask;
 944         long timeo;
 945         int err;
 946
 947         gfp_mask = sk->sk_allocation;
 948         if (gfp_mask & __GFP_WAIT)
 949                 gfp_mask |= __GFP_REPEAT;
 950
 951         timeo = sock_sndtimeo(sk, noblock);
 952         while (1) {
 953                 err = sock_error(sk);
 954                 if (err != 0)
 955                         goto failure;
 956
 957                 err = -EPIPE;
 958                 if (sk->sk_shutdown & SEND_SHUTDOWN)
 959                         goto failure;
 960
 961                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
 962                         skb = alloc_skb(header_len, sk->sk_allocation);
 963                         if (skb) {
 964                                 int npages;
 965                                 int i;
 966
 967                                 /* No pages, we're done... */
 968                                 if (!data_len)
 969                                         break;
 970
 971                                 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
 972                                 skb->truesize += data_len;
 973                                 skb_shinfo(skb)->nr_frags = npages;
 974                                 for (i = 0; i < npages; i++) {
 975                                         struct page *page;
 976                                         skb_frag_t *frag;
 977
 978                                         page = alloc_pages(sk->sk_allocation, 0);
 979                                         if (!page) {
 980                                                 err = -ENOBUFS;
 981                                                 skb_shinfo(skb)->nr_frags = i;
 982                                                 kfree_skb(skb);
 983                                                 goto failure;
 984                                         }
 985
 986                                         frag = &skb_shinfo(skb)->frags[i];
 987                                         frag->page = page;
 988                                         frag->page_offset = 0;
 989                                         frag->size = (data_len >= PAGE_SIZE ?
 990                                                       PAGE_SIZE :
 991                                                       data_len);
 992                                         data_len -= PAGE_SIZE;
 993                                 }
 994
 995                                 /* Full success... */
 996                                 break;
 997                         }
 998                         err = -ENOBUFS;
 999                         goto failure;
1000                 }
1001                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1002                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1003                 err = -EAGAIN;
1004                 if (!timeo)
1005                         goto failure;
1006                 if (signal_pending(current))
1007                         goto interrupted;
1008                 timeo = sock_wait_for_wmem(sk, timeo);
1009         }
1010
1011         skb_set_owner_w(skb, sk);
1012         return skb;
1013
1014 interrupted:
1015         err = sock_intr_errno(timeo);
1016 failure:
1017         *errcode = err;
1018         return NULL;
1019 }
1020
1021 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1022                                     int noblock, int *errcode)
1023 {
1024         return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1025 }
1026
1027 static void __lock_sock(struct sock *sk)
1028 {
1029         DEFINE_WAIT(wait);
1030
1031         for(;;) {
1032                 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1033                                         TASK_UNINTERRUPTIBLE);
1034                 spin_unlock_bh(&sk->sk_lock.slock);
1035                 schedule();
1036                 spin_lock_bh(&sk->sk_lock.slock);
1037                 if(!sock_owned_by_user(sk))
1038                         break;
1039         }
1040         finish_wait(&sk->sk_lock.wq, &wait);
1041 }
1042
1043 static void __release_sock(struct sock *sk)
1044 {
1045         struct sk_buff *skb = sk->sk_backlog.head;
1046
1047         do {
1048                 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1049                 bh_unlock_sock(sk);
1050
1051                 do {
1052                         struct sk_buff *next = skb->next;
1053
1054                         skb->next = NULL;
1055                         sk->sk_backlog_rcv(sk, skb);
1056
1057                         /*
1058                          * We are in process context here with softirqs
1059                          * disabled, use cond_resched_softirq() to preempt.
1060                          * This is safe to do because we've taken the backlog
1061                          * queue private:
1062                          */
1063                         cond_resched_softirq();
1064
1065                         skb = next;
1066                 } while (skb != NULL);
1067
1068                 bh_lock_sock(sk);
1069         } while((skb = sk->sk_backlog.head) != NULL);
1070 }
1071
1072 /**
1073  * sk_wait_data - wait for data to arrive at sk_receive_queue
1074  * @sk:    sock to wait on
1075  * @timeo: for how long
1076  *
1077  * Now socket state including sk->sk_err is changed only under lock,
1078  * hence we may omit checks after joining wait queue.
1079  * We check receive queue before schedule() only as optimization;
1080  * it is very likely that release_sock() added new data.
1081  */
1082 int sk_wait_data(struct sock *sk, long *timeo)
1083 {
1084         int rc;
1085         DEFINE_WAIT(wait);
1086
1087         prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1088         set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1089         rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1090         clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1091         finish_wait(sk->sk_sleep, &wait);
1092         return rc;
1093 }
1094
1095 EXPORT_SYMBOL(sk_wait_data);
1096
1097 /*
1098  * Set of default routines for initialising struct proto_ops when
1099  * the protocol does not support a particular function. In certain
1100  * cases where it makes no sense for a protocol to have a "do nothing"
1101  * function, some default processing is provided.
1102  */
1103
1104 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1105 {
1106         return -EOPNOTSUPP;
1107 }
1108
1109 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1110                     int len, int flags)
1111 {
1112         return -EOPNOTSUPP;
1113 }
1114
1115 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1116 {
1117         return -EOPNOTSUPP;
1118 }
1119
1120 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1121 {
1122         return -EOPNOTSUPP;
1123 }
1124
1125 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1126                     int *len, int peer)
1127 {
1128         return -EOPNOTSUPP;
1129 }
1130
1131 unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt)
1132 {
1133         return 0;
1134 }
1135
1136 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1137 {
1138         return -EOPNOTSUPP;
1139 }
1140
1141 int sock_no_listen(struct socket *sock, int backlog)
1142 {
1143         return -EOPNOTSUPP;
1144 }
1145
1146 int sock_no_shutdown(struct socket *sock, int how)
1147 {
1148         return -EOPNOTSUPP;
1149 }
1150
1151 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1152                     char __user *optval, int optlen)
1153 {
1154         return -EOPNOTSUPP;
1155 }
1156
1157 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1158                     char __user *optval, int __user *optlen)
1159 {
1160         return -EOPNOTSUPP;
1161 }
1162
1163 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1164                     size_t len)
1165 {
1166         return -EOPNOTSUPP;
1167 }
1168
1169 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1170                     size_t len, int flags)
1171 {
1172         return -EOPNOTSUPP;
1173 }
1174
1175 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1176 {
1177         /* Mirror missing mmap method error code */
1178         return -ENODEV;
1179 }
1180
1181 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1182 {
1183         ssize_t res;
1184         struct msghdr msg = {.msg_flags = flags};
1185         struct kvec iov;
1186         char *kaddr = kmap(page);
1187         iov.iov_base = kaddr + offset;
1188         iov.iov_len = size;
1189         res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1190         kunmap(page);
1191         return res;
1192 }
1193
1194 /*
1195  *      Default Socket Callbacks
1196  */
1197
1198 static void sock_def_wakeup(struct sock *sk)
1199 {
1200         read_lock(&sk->sk_callback_lock);
1201         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1202                 wake_up_interruptible_all(sk->sk_sleep);
1203         read_unlock(&sk->sk_callback_lock);
1204 }
1205
1206 static void sock_def_error_report(struct sock *sk)
1207 {
1208         read_lock(&sk->sk_callback_lock);
1209         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1210                 wake_up_interruptible(sk->sk_sleep);
1211         sk_wake_async(sk,0,POLL_ERR);
1212         read_unlock(&sk->sk_callback_lock);
1213 }
1214
1215 static void sock_def_readable(struct sock *sk, int len)
1216 {
1217         read_lock(&sk->sk_callback_lock);
1218         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1219                 wake_up_interruptible(sk->sk_sleep);
1220         sk_wake_async(sk,1,POLL_IN);
1221         read_unlock(&sk->sk_callback_lock);
1222 }
1223
1224 static void sock_def_write_space(struct sock *sk)
1225 {
1226         read_lock(&sk->sk_callback_lock);
1227
1228         /* Do not wake up a writer until he can make "significant"
1229          * progress.  --DaveM
1230          */
1231         if((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1232                 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1233                         wake_up_interruptible(sk->sk_sleep);
1234
1235                 /* Should agree with poll, otherwise some programs break */
1236                 if (sock_writeable(sk))
1237                         sk_wake_async(sk, 2, POLL_OUT);
1238         }
1239
1240         read_unlock(&sk->sk_callback_lock);
1241 }
1242
1243 static void sock_def_destruct(struct sock *sk)
1244 {
1245         if (sk->sk_protinfo)
1246                 kfree(sk->sk_protinfo);
1247 }
1248
1249 void sk_send_sigurg(struct sock *sk)
1250 {
1251         if (sk->sk_socket && sk->sk_socket->file)
1252                 if (send_sigurg(&sk->sk_socket->file->f_owner))
1253                         sk_wake_async(sk, 3, POLL_PRI);
1254 }
1255
1256 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1257                     unsigned long expires)
1258 {
1259         if (!mod_timer(timer, expires))
1260                 sock_hold(sk);
1261 }
1262
1263 EXPORT_SYMBOL(sk_reset_timer);
1264
1265 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1266 {
1267         if (timer_pending(timer) && del_timer(timer))
1268                 __sock_put(sk);
1269 }
1270
1271 EXPORT_SYMBOL(sk_stop_timer);
1272
1273 void sock_init_data(struct socket *sock, struct sock *sk)
1274 {
1275         skb_queue_head_init(&sk->sk_receive_queue);
1276         skb_queue_head_init(&sk->sk_write_queue);
1277         skb_queue_head_init(&sk->sk_error_queue);
1278
1279         sk->sk_send_head        =       NULL;
1280
1281         init_timer(&sk->sk_timer);
1282
1283         sk->sk_allocation       =       GFP_KERNEL;
1284         sk->sk_rcvbuf           =       sysctl_rmem_default;
1285         sk->sk_sndbuf           =       sysctl_wmem_default;
1286         sk->sk_state            =       TCP_CLOSE;
1287         sk->sk_socket           =       sock;
1288
1289         sock_set_flag(sk, SOCK_ZAPPED);
1290
1291         if(sock)
1292         {
1293                 sk->sk_type     =       sock->type;
1294                 sk->sk_sleep    =       &sock->wait;
1295                 sock->sk        =       sk;
1296         } else
1297                 sk->sk_sleep    =       NULL;
1298
1299         rwlock_init(&sk->sk_dst_lock);
1300         rwlock_init(&sk->sk_callback_lock);
1301
1302         sk->sk_state_change     =       sock_def_wakeup;
1303         sk->sk_data_ready       =       sock_def_readable;
1304         sk->sk_write_space      =       sock_def_write_space;
1305         sk->sk_error_report     =       sock_def_error_report;
1306         sk->sk_destruct         =       sock_def_destruct;
1307
1308         sk->sk_sndmsg_page      =       NULL;
1309         sk->sk_sndmsg_off       =       0;
1310
1311         sk->sk_peercred.pid     =       0;
1312         sk->sk_peercred.uid     =       -1;
1313         sk->sk_peercred.gid     =       -1;
1314         sk->sk_write_pending    =       0;
1315         sk->sk_rcvlowat         =       1;
1316         sk->sk_rcvtimeo         =       MAX_SCHEDULE_TIMEOUT;
1317         sk->sk_sndtimeo         =       MAX_SCHEDULE_TIMEOUT;
1318
1319         sk->sk_stamp.tv_sec     = -1L;
1320         sk->sk_stamp.tv_usec    = -1L;
1321
1322         atomic_set(&sk->sk_refcnt, 1);
1323 }
1324
1325 void fastcall lock_sock(struct sock *sk)
1326 {
1327         might_sleep();
1328         spin_lock_bh(&(sk->sk_lock.slock));
1329         if (sk->sk_lock.owner)
1330                 __lock_sock(sk);
1331         sk->sk_lock.owner = (void *)1;
1332         spin_unlock_bh(&(sk->sk_lock.slock));
1333 }
1334
1335 EXPORT_SYMBOL(lock_sock);
1336
1337 void fastcall release_sock(struct sock *sk)
1338 {
1339         spin_lock_bh(&(sk->sk_lock.slock));
1340         if (sk->sk_backlog.tail)
1341                 __release_sock(sk);
1342         sk->sk_lock.owner = NULL;
1343         if (waitqueue_active(&(sk->sk_lock.wq)))
1344                 wake_up(&(sk->sk_lock.wq));
1345         spin_unlock_bh(&(sk->sk_lock.slock));
1346 }
1347 EXPORT_SYMBOL(release_sock);
1348
1349 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1350 {
1351         if (!sock_flag(sk, SOCK_TIMESTAMP))
1352                 sock_enable_timestamp(sk);
1353         if (sk->sk_stamp.tv_sec == -1)
1354                 return -ENOENT;
1355         if (sk->sk_stamp.tv_sec == 0)
1356                 do_gettimeofday(&sk->sk_stamp);
1357         return copy_to_user(userstamp, &sk->sk_stamp, sizeof(struct timeval)) ?
1358                 -EFAULT : 0;
1359 }
1360 EXPORT_SYMBOL(sock_get_timestamp);
1361
1362 void sock_enable_timestamp(struct sock *sk)
1363 {
1364         if (!sock_flag(sk, SOCK_TIMESTAMP)) {
1365                 sock_set_flag(sk, SOCK_TIMESTAMP);
1366                 net_enable_timestamp();
1367         }
1368 }
1369 EXPORT_SYMBOL(sock_enable_timestamp);
1370
1371 /*
1372  *      Get a socket option on an socket.
1373  *
1374  *      FIX: POSIX 1003.1g is very ambiguous here. It states that
1375  *      asynchronous errors should be reported by getsockopt. We assume
1376  *      this means if you specify SO_ERROR (otherwise whats the point of it).
1377  */
1378 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1379                            char __user *optval, int __user *optlen)
1380 {
1381         struct sock *sk = sock->sk;
1382
1383         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1384 }
1385
1386 EXPORT_SYMBOL(sock_common_getsockopt);
1387
1388 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1389                         struct msghdr *msg, size_t size, int flags)
1390 {
1391         struct sock *sk = sock->sk;
1392         int addr_len = 0;
1393         int err;
1394
1395         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
1396                                    flags & ~MSG_DONTWAIT, &addr_len);
1397         if (err >= 0)
1398                 msg->msg_namelen = addr_len;
1399         return err;
1400 }
1401
1402 EXPORT_SYMBOL(sock_common_recvmsg);
1403
1404 /*
1405  *      Set socket options on an inet socket.
1406  */
1407 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1408                            char __user *optval, int optlen)
1409 {
1410         struct sock *sk = sock->sk;
1411
1412         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1413 }
1414
1415 EXPORT_SYMBOL(sock_common_setsockopt);
1416
1417 void sk_common_release(struct sock *sk)
1418 {
1419         if (sk->sk_prot->destroy)
1420                 sk->sk_prot->destroy(sk);
1421
1422         /*
1423          * Observation: when sock_common_release is called, processes have
1424          * no access to socket. But net still has.
1425          * Step one, detach it from networking:
1426          *
1427          * A. Remove from hash tables.
1428          */
1429
1430         sk->sk_prot->unhash(sk);
1431
1432         /*
1433          * In this point socket cannot receive new packets, but it is possible
1434          * that some packets are in flight because some CPU runs receiver and
1435          * did hash table lookup before we unhashed socket. They will achieve
1436          * receive queue and will be purged by socket destructor.
1437          *
1438          * Also we still have packets pending on receive queue and probably,
1439          * our own packets waiting in device queues. sock_destroy will drain
1440          * receive queue, but transmitted packets will delay socket destruction
1441          * until the last reference will be released.
1442          */
1443
1444         sock_orphan(sk);
1445
1446         xfrm_sk_free_policy(sk);
1447
1448         sk_refcnt_debug_release(sk);
1449         sock_put(sk);
1450 }
1451
1452 EXPORT_SYMBOL(sk_common_release);
1453
1454 static DEFINE_RWLOCK(proto_list_lock);
1455 static LIST_HEAD(proto_list);
1456
1457 int proto_register(struct proto *prot, int alloc_slab)
1458 {
1459         char *request_sock_slab_name = NULL;
1460         char *timewait_sock_slab_name;
1461         int rc = -ENOBUFS;
1462
1463         if (alloc_slab) {
1464                 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
1465                                                SLAB_HWCACHE_ALIGN, NULL, NULL);
1466
1467                 if (prot->slab == NULL) {
1468                         printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
1469                                prot->name);
1470                         goto out;
1471                 }
1472
1473                 if (prot->rsk_prot != NULL) {
1474                         static const char mask[] = "request_sock_%s";
1475
1476                         request_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
1477                         if (request_sock_slab_name == NULL)
1478                                 goto out_free_sock_slab;
1479
1480                         sprintf(request_sock_slab_name, mask, prot->name);
1481                         prot->rsk_prot->slab = kmem_cache_create(request_sock_slab_name,
1482                                                                  prot->rsk_prot->obj_size, 0,
1483                                                                  SLAB_HWCACHE_ALIGN, NULL, NULL);
1484
1485                         if (prot->rsk_prot->slab == NULL) {
1486                                 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
1487                                        prot->name);
1488                                 goto out_free_request_sock_slab_name;
1489                         }
1490                 }
1491
1492                 if (prot->twsk_obj_size) {
1493                         static const char mask[] = "tw_sock_%s";
1494
1495                         timewait_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
1496
1497                         if (timewait_sock_slab_name == NULL)
1498                                 goto out_free_request_sock_slab;
1499
1500                         sprintf(timewait_sock_slab_name, mask, prot->name);
1501                         prot->twsk_slab = kmem_cache_create(timewait_sock_slab_name,
1502                                                             prot->twsk_obj_size,
1503                                                             0, SLAB_HWCACHE_ALIGN,
1504                                                             NULL, NULL);
1505                         if (prot->twsk_slab == NULL)
1506                                 goto out_free_timewait_sock_slab_name;
1507                 }
1508         }
1509
1510         write_lock(&proto_list_lock);
1511         list_add(&prot->node, &proto_list);
1512         write_unlock(&proto_list_lock);
1513         rc = 0;
1514 out:
1515         return rc;
1516 out_free_timewait_sock_slab_name:
1517         kfree(timewait_sock_slab_name);
1518 out_free_request_sock_slab:
1519         if (prot->rsk_prot && prot->rsk_prot->slab) {
1520                 kmem_cache_destroy(prot->rsk_prot->slab);
1521                 prot->rsk_prot->slab = NULL;
1522         }
1523 out_free_request_sock_slab_name:
1524         kfree(request_sock_slab_name);
1525 out_free_sock_slab:
1526         kmem_cache_destroy(prot->slab);
1527         prot->slab = NULL;
1528         goto out;
1529 }
1530
1531 EXPORT_SYMBOL(proto_register);
1532
1533 void proto_unregister(struct proto *prot)
1534 {
1535         write_lock(&proto_list_lock);
1536         list_del(&prot->node);
1537         write_unlock(&proto_list_lock);
1538
1539         if (prot->slab != NULL) {
1540                 kmem_cache_destroy(prot->slab);
1541                 prot->slab = NULL;
1542         }
1543
1544         if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
1545                 const char *name = kmem_cache_name(prot->rsk_prot->slab);
1546
1547                 kmem_cache_destroy(prot->rsk_prot->slab);
1548                 kfree(name);
1549                 prot->rsk_prot->slab = NULL;
1550         }
1551
1552         if (prot->twsk_slab != NULL) {
1553                 const char *name = kmem_cache_name(prot->twsk_slab);
1554
1555                 kmem_cache_destroy(prot->twsk_slab);
1556                 kfree(name);
1557                 prot->twsk_slab = NULL;
1558         }
1559 }
1560
1561 EXPORT_SYMBOL(proto_unregister);
1562
1563 #ifdef CONFIG_PROC_FS
1564 static inline struct proto *__proto_head(void)
1565 {
1566         return list_entry(proto_list.next, struct proto, node);
1567 }
1568
1569 static inline struct proto *proto_head(void)
1570 {
1571         return list_empty(&proto_list) ? NULL : __proto_head();
1572 }
1573
1574 static inline struct proto *proto_next(struct proto *proto)
1575 {
1576         return proto->node.next == &proto_list ? NULL :
1577                 list_entry(proto->node.next, struct proto, node);
1578 }
1579
1580 static inline struct proto *proto_get_idx(loff_t pos)
1581 {
1582         struct proto *proto;
1583         loff_t i = 0;
1584
1585         list_for_each_entry(proto, &proto_list, node)
1586                 if (i++ == pos)
1587                         goto out;
1588
1589         proto = NULL;
1590 out:
1591         return proto;
1592 }
1593
1594 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
1595 {
1596         read_lock(&proto_list_lock);
1597         return *pos ? proto_get_idx(*pos - 1) : SEQ_START_TOKEN;
1598 }
1599
1600 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1601 {
1602         ++*pos;
1603         return v == SEQ_START_TOKEN ? proto_head() : proto_next(v);
1604 }
1605
1606 static void proto_seq_stop(struct seq_file *seq, void *v)
1607 {
1608         read_unlock(&proto_list_lock);
1609 }
1610
1611 static char proto_method_implemented(const void *method)
1612 {
1613         return method == NULL ? 'n' : 'y';
1614 }
1615
1616 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
1617 {
1618         seq_printf(seq, "%-9s %4u %6d  %6d   %-3s %6u   %-3s  %-10s "
1619                         "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
1620                    proto->name,
1621                    proto->obj_size,
1622                    proto->sockets_allocated != NULL ? atomic_read(proto->sockets_allocated) : -1,
1623                    proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
1624                    proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
1625                    proto->max_header,
1626                    proto->slab == NULL ? "no" : "yes",
1627                    module_name(proto->owner),
1628                    proto_method_implemented(proto->close),
1629                    proto_method_implemented(proto->connect),
1630                    proto_method_implemented(proto->disconnect),
1631                    proto_method_implemented(proto->accept),
1632                    proto_method_implemented(proto->ioctl),
1633                    proto_method_implemented(proto->init),
1634                    proto_method_implemented(proto->destroy),
1635                    proto_method_implemented(proto->shutdown),
1636                    proto_method_implemented(proto->setsockopt),
1637                    proto_method_implemented(proto->getsockopt),
1638                    proto_method_implemented(proto->sendmsg),
1639                    proto_method_implemented(proto->recvmsg),
1640                    proto_method_implemented(proto->sendpage),
1641                    proto_method_implemented(proto->bind),
1642                    proto_method_implemented(proto->backlog_rcv),
1643                    proto_method_implemented(proto->hash),
1644                    proto_method_implemented(proto->unhash),
1645                    proto_method_implemented(proto->get_port),
1646                    proto_method_implemented(proto->enter_memory_pressure));
1647 }
1648
1649 static int proto_seq_show(struct seq_file *seq, void *v)
1650 {
1651         if (v == SEQ_START_TOKEN)
1652                 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
1653                            "protocol",
1654                            "size",
1655                            "sockets",
1656                            "memory",
1657                            "press",
1658                            "maxhdr",
1659                            "slab",
1660                            "module",
1661                            "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
1662         else
1663                 proto_seq_printf(seq, v);
1664         return 0;
1665 }
1666
1667 static struct seq_operations proto_seq_ops = {
1668         .start  = proto_seq_start,
1669         .next   = proto_seq_next,
1670         .stop   = proto_seq_stop,
1671         .show   = proto_seq_show,
1672 };
1673
1674 static int proto_seq_open(struct inode *inode, struct file *file)
1675 {
1676         return seq_open(file, &proto_seq_ops);
1677 }
1678
1679 static struct file_operations proto_seq_fops = {
1680         .owner          = THIS_MODULE,
1681         .open           = proto_seq_open,
1682         .read           = seq_read,
1683         .llseek         = seq_lseek,
1684         .release        = seq_release,
1685 };
1686
1687 static int __init proto_init(void)
1688 {
1689         /* register /proc/net/protocols */
1690         return proc_net_fops_create("protocols", S_IRUGO, &proto_seq_fops) == NULL ? -ENOBUFS : 0;
1691 }
1692
1693 subsys_initcall(proto_init);
1694
1695 #endif /* PROC_FS */
1696
1697 EXPORT_SYMBOL(sk_alloc);
1698 EXPORT_SYMBOL(sk_free);
1699 EXPORT_SYMBOL(sk_send_sigurg);
1700 EXPORT_SYMBOL(sock_alloc_send_skb);
1701 EXPORT_SYMBOL(sock_init_data);
1702 EXPORT_SYMBOL(sock_kfree_s);
1703 EXPORT_SYMBOL(sock_kmalloc);
1704 EXPORT_SYMBOL(sock_no_accept);
1705 EXPORT_SYMBOL(sock_no_bind);
1706 EXPORT_SYMBOL(sock_no_connect);
1707 EXPORT_SYMBOL(sock_no_getname);
1708 EXPORT_SYMBOL(sock_no_getsockopt);
1709 EXPORT_SYMBOL(sock_no_ioctl);
1710 EXPORT_SYMBOL(sock_no_listen);
1711 EXPORT_SYMBOL(sock_no_mmap);
1712 EXPORT_SYMBOL(sock_no_poll);
1713 EXPORT_SYMBOL(sock_no_recvmsg);
1714 EXPORT_SYMBOL(sock_no_sendmsg);
1715 EXPORT_SYMBOL(sock_no_sendpage);
1716 EXPORT_SYMBOL(sock_no_setsockopt);
1717 EXPORT_SYMBOL(sock_no_shutdown);
1718 EXPORT_SYMBOL(sock_no_socketpair);
1719 EXPORT_SYMBOL(sock_rfree);
1720 EXPORT_SYMBOL(sock_setsockopt);
1721 EXPORT_SYMBOL(sock_wfree);
1722 EXPORT_SYMBOL(sock_wmalloc);
1723 EXPORT_SYMBOL(sock_i_uid);
1724 EXPORT_SYMBOL(sock_i_ino);
1725 EXPORT_SYMBOL(sysctl_optmem_max);
1726 #ifdef CONFIG_SYSCTL
1727 EXPORT_SYMBOL(sysctl_rmem_max);
1728 EXPORT_SYMBOL(sysctl_wmem_max);
1729 #endif