err.no Git - linux-2.6/blob - net/core/sock.c

   1 /*
   2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   3  *              operating system.  INET is implemented using the  BSD Socket
   4  *              interface as the means of communication with the user level.
   5  *
   6  *              Generic socket support routines. Memory allocators, socket lock/release
   7  *              handler for protocols to use and generic option handler.
   8  *
   9  *
  10  * Version:     $Id: sock.c,v 1.117 2002/02/01 22:01:03 davem Exp $
  11  *
  12  * Authors:     Ross Biro, <bir7@leland.Stanford.Edu>
  13  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  14  *              Florian La Roche, <flla@stud.uni-sb.de>
  15  *              Alan Cox, <A.Cox@swansea.ac.uk>
  16  *
  17  * Fixes:
  18  *              Alan Cox        :       Numerous verify_area() problems
  19  *              Alan Cox        :       Connecting on a connecting socket
  20  *                                      now returns an error for tcp.
  21  *              Alan Cox        :       sock->protocol is set correctly.
  22  *                                      and is not sometimes left as 0.
  23  *              Alan Cox        :       connect handles icmp errors on a
  24  *                                      connect properly. Unfortunately there
  25  *                                      is a restart syscall nasty there. I
  26  *                                      can't match BSD without hacking the C
  27  *                                      library. Ideas urgently sought!
  28  *              Alan Cox        :       Disallow bind() to addresses that are
  29  *                                      not ours - especially broadcast ones!!
  30  *              Alan Cox        :       Socket 1024 _IS_ ok for users. (fencepost)
  31  *              Alan Cox        :       sock_wfree/sock_rfree don't destroy sockets,
  32  *                                      instead they leave that for the DESTROY timer.
  33  *              Alan Cox        :       Clean up error flag in accept
  34  *              Alan Cox        :       TCP ack handling is buggy, the DESTROY timer
  35  *                                      was buggy. Put a remove_sock() in the handler
  36  *                                      for memory when we hit 0. Also altered the timer
  37  *                                      code. The ACK stuff can wait and needs major
  38  *                                      TCP layer surgery.
  39  *              Alan Cox        :       Fixed TCP ack bug, removed remove sock
  40  *                                      and fixed timer/inet_bh race.
  41  *              Alan Cox        :       Added zapped flag for TCP
  42  *              Alan Cox        :       Move kfree_skb into skbuff.c and tidied up surplus code
  43  *              Alan Cox        :       for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
  44  *              Alan Cox        :       kfree_s calls now are kfree_skbmem so we can track skb resources
  45  *              Alan Cox        :       Supports socket option broadcast now as does udp. Packet and raw need fixing.
  46  *              Alan Cox        :       Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
  47  *              Rick Sladkey    :       Relaxed UDP rules for matching packets.
  48  *              C.E.Hawkins     :       IFF_PROMISC/SIOCGHWADDR support
  49  *      Pauline Middelink       :       identd support
  50  *              Alan Cox        :       Fixed connect() taking signals I think.
  51  *              Alan Cox        :       SO_LINGER supported
  52  *              Alan Cox        :       Error reporting fixes
  53  *              Anonymous       :       inet_create tidied up (sk->reuse setting)
  54  *              Alan Cox        :       inet sockets don't set sk->type!
  55  *              Alan Cox        :       Split socket option code
  56  *              Alan Cox        :       Callbacks
  57  *              Alan Cox        :       Nagle flag for Charles & Johannes stuff
  58  *              Alex            :       Removed restriction on inet fioctl
  59  *              Alan Cox        :       Splitting INET from NET core
  60  *              Alan Cox        :       Fixed bogus SO_TYPE handling in getsockopt()
  61  *              Adam Caldwell   :       Missing return in SO_DONTROUTE/SO_DEBUG code
  62  *              Alan Cox        :       Split IP from generic code
  63  *              Alan Cox        :       New kfree_skbmem()
  64  *              Alan Cox        :       Make SO_DEBUG superuser only.
  65  *              Alan Cox        :       Allow anyone to clear SO_DEBUG
  66  *                                      (compatibility fix)
  67  *              Alan Cox        :       Added optimistic memory grabbing for AF_UNIX throughput.
  68  *              Alan Cox        :       Allocator for a socket is settable.
  69  *              Alan Cox        :       SO_ERROR includes soft errors.
  70  *              Alan Cox        :       Allow NULL arguments on some SO_ opts
  71  *              Alan Cox        :       Generic socket allocation to make hooks
  72  *                                      easier (suggested by Craig Metz).
  73  *              Michael Pall    :       SO_ERROR returns positive errno again
  74  *              Steve Whitehouse:       Added default destructor to free
  75  *                                      protocol private data.
  76  *              Steve Whitehouse:       Added various other default routines
  77  *                                      common to several socket families.
  78  *              Chris Evans     :       Call suser() check last on F_SETOWN
  79  *              Jay Schulist    :       Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
  80  *              Andi Kleen      :       Add sock_kmalloc()/sock_kfree_s()
  81  *              Andi Kleen      :       Fix write_space callback
  82  *              Chris Evans     :       Security fixes - signedness again
  83  *              Arnaldo C. Melo :       cleanups, use skb_queue_purge
  84  *
  85  * To Fix:
  86  *
  87  *
  88  *              This program is free software; you can redistribute it and/or
  89  *              modify it under the terms of the GNU General Public License
  90  *              as published by the Free Software Foundation; either version
  91  *              2 of the License, or (at your option) any later version.
  92  */
  93
  94 #include <linux/config.h>
  95 #include <linux/errno.h>
  96 #include <linux/types.h>
  97 #include <linux/socket.h>
  98 #include <linux/in.h>
  99 #include <linux/kernel.h>
 100 #include <linux/major.h>
 101 #include <linux/module.h>
 102 #include <linux/proc_fs.h>
 103 #include <linux/seq_file.h>
 104 #include <linux/sched.h>
 105 #include <linux/timer.h>
 106 #include <linux/string.h>
 107 #include <linux/sockios.h>
 108 #include <linux/net.h>
 109 #include <linux/mm.h>
 110 #include <linux/slab.h>
 111 #include <linux/interrupt.h>
 112 #include <linux/poll.h>
 113 #include <linux/tcp.h>
 114 #include <linux/init.h>
 115
 116 #include <asm/uaccess.h>
 117 #include <asm/system.h>
 118
 119 #include <linux/netdevice.h>
 120 #include <net/protocol.h>
 121 #include <linux/skbuff.h>
 122 #include <net/sock.h>
 123 #include <net/xfrm.h>
 124 #include <linux/ipsec.h>
 125
 126 #include <linux/filter.h>
 127
 128 #ifdef CONFIG_INET
 129 #include <net/tcp.h>
 130 #endif
 131
 132 /* Take into consideration the size of the struct sk_buff overhead in the
 133  * determination of these values, since that is non-constant across
 134  * platforms.  This makes socket queueing behavior and performance
 135  * not depend upon such differences.
 136  */
 137 #define _SK_MEM_PACKETS         256
 138 #define _SK_MEM_OVERHEAD        (sizeof(struct sk_buff) + 256)
 139 #define SK_WMEM_MAX             (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
 140 #define SK_RMEM_MAX             (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
 141
 142 /* Run time adjustable parameters. */
 143 __u32 sysctl_wmem_max = SK_WMEM_MAX;
 144 __u32 sysctl_rmem_max = SK_RMEM_MAX;
 145 __u32 sysctl_wmem_default = SK_WMEM_MAX;
 146 __u32 sysctl_rmem_default = SK_RMEM_MAX;
 147
 148 /* Maximal space eaten by iovec or ancilliary data plus some space */
 149 int sysctl_optmem_max = sizeof(unsigned long)*(2*UIO_MAXIOV + 512);
 150
 151 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
 152 {
 153         struct timeval tv;
 154
 155         if (optlen < sizeof(tv))
 156                 return -EINVAL;
 157         if (copy_from_user(&tv, optval, sizeof(tv)))
 158                 return -EFAULT;
 159
 160         *timeo_p = MAX_SCHEDULE_TIMEOUT;
 161         if (tv.tv_sec == 0 && tv.tv_usec == 0)
 162                 return 0;
 163         if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
 164                 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
 165         return 0;
 166 }
 167
 168 static void sock_warn_obsolete_bsdism(const char *name)
 169 {
 170         static int warned;
 171         static char warncomm[TASK_COMM_LEN];
 172         if (strcmp(warncomm, current->comm) && warned < 5) {
 173                 strcpy(warncomm,  current->comm);
 174                 printk(KERN_WARNING "process `%s' is using obsolete "
 175                        "%s SO_BSDCOMPAT\n", warncomm, name);
 176                 warned++;
 177         }
 178 }
 179
 180 static void sock_disable_timestamp(struct sock *sk)
 181 {
 182         if (sock_flag(sk, SOCK_TIMESTAMP)) {
 183                 sock_reset_flag(sk, SOCK_TIMESTAMP);
 184                 net_disable_timestamp();
 185         }
 186 }
 187
 188
 189 /*
 190  *      This is meant for all protocols to use and covers goings on
 191  *      at the socket level. Everything here is generic.
 192  */
 193
 194 int sock_setsockopt(struct socket *sock, int level, int optname,
 195                     char __user *optval, int optlen)
 196 {
 197         struct sock *sk=sock->sk;
 198         struct sk_filter *filter;
 199         int val;
 200         int valbool;
 201         struct linger ling;
 202         int ret = 0;
 203
 204         /*
 205          *      Options without arguments
 206          */
 207
 208 #ifdef SO_DONTLINGER            /* Compatibility item... */
 209         switch (optname) {
 210                 case SO_DONTLINGER:
 211                         sock_reset_flag(sk, SOCK_LINGER);
 212                         return 0;
 213         }
 214 #endif
 215
 216         if(optlen<sizeof(int))
 217                 return(-EINVAL);
 218
 219         if (get_user(val, (int __user *)optval))
 220                 return -EFAULT;
 221
 222         valbool = val?1:0;
 223
 224         lock_sock(sk);
 225
 226         switch(optname)
 227         {
 228                 case SO_DEBUG:
 229                         if(val && !capable(CAP_NET_ADMIN))
 230                         {
 231                                 ret = -EACCES;
 232                         }
 233                         else if (valbool)
 234                                 sock_set_flag(sk, SOCK_DBG);
 235                         else
 236                                 sock_reset_flag(sk, SOCK_DBG);
 237                         break;
 238                 case SO_REUSEADDR:
 239                         sk->sk_reuse = valbool;
 240                         break;
 241                 case SO_TYPE:
 242                 case SO_ERROR:
 243                         ret = -ENOPROTOOPT;
 244                         break;
 245                 case SO_DONTROUTE:
 246                         if (valbool)
 247                                 sock_set_flag(sk, SOCK_LOCALROUTE);
 248                         else
 249                                 sock_reset_flag(sk, SOCK_LOCALROUTE);
 250                         break;
 251                 case SO_BROADCAST:
 252                         sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
 253                         break;
 254                 case SO_SNDBUF:
 255                         /* Don't error on this BSD doesn't and if you think
 256                            about it this is right. Otherwise apps have to
 257                            play 'guess the biggest size' games. RCVBUF/SNDBUF
 258                            are treated in BSD as hints */
 259
 260                         if (val > sysctl_wmem_max)
 261                                 val = sysctl_wmem_max;
 262
 263                         sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
 264                         if ((val * 2) < SOCK_MIN_SNDBUF)
 265                                 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
 266                         else
 267                                 sk->sk_sndbuf = val * 2;
 268
 269                         /*
 270                          *      Wake up sending tasks if we
 271                          *      upped the value.
 272                          */
 273                         sk->sk_write_space(sk);
 274                         break;
 275
 276                 case SO_RCVBUF:
 277                         /* Don't error on this BSD doesn't and if you think
 278                            about it this is right. Otherwise apps have to
 279                            play 'guess the biggest size' games. RCVBUF/SNDBUF
 280                            are treated in BSD as hints */
 281
 282                         if (val > sysctl_rmem_max)
 283                                 val = sysctl_rmem_max;
 284
 285                         sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
 286                         /* FIXME: is this lower bound the right one? */
 287                         if ((val * 2) < SOCK_MIN_RCVBUF)
 288                                 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
 289                         else
 290                                 sk->sk_rcvbuf = val * 2;
 291                         break;
 292
 293                 case SO_KEEPALIVE:
 294 #ifdef CONFIG_INET
 295                         if (sk->sk_protocol == IPPROTO_TCP)
 296                                 tcp_set_keepalive(sk, valbool);
 297 #endif
 298                         sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
 299                         break;
 300
 301                 case SO_OOBINLINE:
 302                         sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
 303                         break;
 304
 305                 case SO_NO_CHECK:
 306                         sk->sk_no_check = valbool;
 307                         break;
 308
 309                 case SO_PRIORITY:
 310                         if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
 311                                 sk->sk_priority = val;
 312                         else
 313                                 ret = -EPERM;
 314                         break;
 315
 316                 case SO_LINGER:
 317                         if(optlen<sizeof(ling)) {
 318                                 ret = -EINVAL;  /* 1003.1g */
 319                                 break;
 320                         }
 321                         if (copy_from_user(&ling,optval,sizeof(ling))) {
 322                                 ret = -EFAULT;
 323                                 break;
 324                         }
 325                         if (!ling.l_onoff)
 326                                 sock_reset_flag(sk, SOCK_LINGER);
 327                         else {
 328 #if (BITS_PER_LONG == 32)
 329                                 if (ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
 330                                         sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
 331                                 else
 332 #endif
 333                                         sk->sk_lingertime = ling.l_linger * HZ;
 334                                 sock_set_flag(sk, SOCK_LINGER);
 335                         }
 336                         break;
 337
 338                 case SO_BSDCOMPAT:
 339                         sock_warn_obsolete_bsdism("setsockopt");
 340                         break;
 341
 342                 case SO_PASSCRED:
 343                         if (valbool)
 344                                 set_bit(SOCK_PASSCRED, &sock->flags);
 345                         else
 346                                 clear_bit(SOCK_PASSCRED, &sock->flags);
 347                         break;
 348
 349                 case SO_TIMESTAMP:
 350                         if (valbool)  {
 351                                 sock_set_flag(sk, SOCK_RCVTSTAMP);
 352                                 sock_enable_timestamp(sk);
 353                         } else
 354                                 sock_reset_flag(sk, SOCK_RCVTSTAMP);
 355                         break;
 356
 357                 case SO_RCVLOWAT:
 358                         if (val < 0)
 359                                 val = INT_MAX;
 360                         sk->sk_rcvlowat = val ? : 1;
 361                         break;
 362
 363                 case SO_RCVTIMEO:
 364                         ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
 365                         break;
 366
 367                 case SO_SNDTIMEO:
 368                         ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
 369                         break;
 370
 371 #ifdef CONFIG_NETDEVICES
 372                 case SO_BINDTODEVICE:
 373                 {
 374                         char devname[IFNAMSIZ];
 375
 376                         /* Sorry... */
 377                         if (!capable(CAP_NET_RAW)) {
 378                                 ret = -EPERM;
 379                                 break;
 380                         }
 381
 382                         /* Bind this socket to a particular device like "eth0",
 383                          * as specified in the passed interface name. If the
 384                          * name is "" or the option length is zero the socket
 385                          * is not bound.
 386                          */
 387
 388                         if (!valbool) {
 389                                 sk->sk_bound_dev_if = 0;
 390                         } else {
 391                                 if (optlen > IFNAMSIZ)
 392                                         optlen = IFNAMSIZ;
 393                                 if (copy_from_user(devname, optval, optlen)) {
 394                                         ret = -EFAULT;
 395                                         break;
 396                                 }
 397
 398                                 /* Remove any cached route for this socket. */
 399                                 sk_dst_reset(sk);
 400
 401                                 if (devname[0] == '\0') {
 402                                         sk->sk_bound_dev_if = 0;
 403                                 } else {
 404                                         struct net_device *dev = dev_get_by_name(devname);
 405                                         if (!dev) {
 406                                                 ret = -ENODEV;
 407                                                 break;
 408                                         }
 409                                         sk->sk_bound_dev_if = dev->ifindex;
 410                                         dev_put(dev);
 411                                 }
 412                         }
 413                         break;
 414                 }
 415 #endif
 416
 417
 418                 case SO_ATTACH_FILTER:
 419                         ret = -EINVAL;
 420                         if (optlen == sizeof(struct sock_fprog)) {
 421                                 struct sock_fprog fprog;
 422
 423                                 ret = -EFAULT;
 424                                 if (copy_from_user(&fprog, optval, sizeof(fprog)))
 425                                         break;
 426
 427                                 ret = sk_attach_filter(&fprog, sk);
 428                         }
 429                         break;
 430
 431                 case SO_DETACH_FILTER:
 432                         spin_lock_bh(&sk->sk_lock.slock);
 433                         filter = sk->sk_filter;
 434                         if (filter) {
 435                                 sk->sk_filter = NULL;
 436                                 spin_unlock_bh(&sk->sk_lock.slock);
 437                                 sk_filter_release(sk, filter);
 438                                 break;
 439                         }
 440                         spin_unlock_bh(&sk->sk_lock.slock);
 441                         ret = -ENONET;
 442                         break;
 443
 444                 /* We implement the SO_SNDLOWAT etc to
 445                    not be settable (1003.1g 5.3) */
 446                 default:
 447                         ret = -ENOPROTOOPT;
 448                         break;
 449         }
 450         release_sock(sk);
 451         return ret;
 452 }
 453
 454
 455 int sock_getsockopt(struct socket *sock, int level, int optname,
 456                     char __user *optval, int __user *optlen)
 457 {
 458         struct sock *sk = sock->sk;
 459
 460         union
 461         {
 462                 int val;
 463                 struct linger ling;
 464                 struct timeval tm;
 465         } v;
 466
 467         unsigned int lv = sizeof(int);
 468         int len;
 469
 470         if(get_user(len,optlen))
 471                 return -EFAULT;
 472         if(len < 0)
 473                 return -EINVAL;
 474
 475         switch(optname)
 476         {
 477                 case SO_DEBUG:
 478                         v.val = sock_flag(sk, SOCK_DBG);
 479                         break;
 480
 481                 case SO_DONTROUTE:
 482                         v.val = sock_flag(sk, SOCK_LOCALROUTE);
 483                         break;
 484
 485                 case SO_BROADCAST:
 486                         v.val = !!sock_flag(sk, SOCK_BROADCAST);
 487                         break;
 488
 489                 case SO_SNDBUF:
 490                         v.val = sk->sk_sndbuf;
 491                         break;
 492
 493                 case SO_RCVBUF:
 494                         v.val = sk->sk_rcvbuf;
 495                         break;
 496
 497                 case SO_REUSEADDR:
 498                         v.val = sk->sk_reuse;
 499                         break;
 500
 501                 case SO_KEEPALIVE:
 502                         v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
 503                         break;
 504
 505                 case SO_TYPE:
 506                         v.val = sk->sk_type;
 507                         break;
 508
 509                 case SO_ERROR:
 510                         v.val = -sock_error(sk);
 511                         if(v.val==0)
 512                                 v.val = xchg(&sk->sk_err_soft, 0);
 513                         break;
 514
 515                 case SO_OOBINLINE:
 516                         v.val = !!sock_flag(sk, SOCK_URGINLINE);
 517                         break;
 518
 519                 case SO_NO_CHECK:
 520                         v.val = sk->sk_no_check;
 521                         break;
 522
 523                 case SO_PRIORITY:
 524                         v.val = sk->sk_priority;
 525                         break;
 526
 527                 case SO_LINGER:
 528                         lv              = sizeof(v.ling);
 529                         v.ling.l_onoff  = !!sock_flag(sk, SOCK_LINGER);
 530                         v.ling.l_linger = sk->sk_lingertime / HZ;
 531                         break;
 532
 533                 case SO_BSDCOMPAT:
 534                         sock_warn_obsolete_bsdism("getsockopt");
 535                         break;
 536
 537                 case SO_TIMESTAMP:
 538                         v.val = sock_flag(sk, SOCK_RCVTSTAMP);
 539                         break;
 540
 541                 case SO_RCVTIMEO:
 542                         lv=sizeof(struct timeval);
 543                         if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
 544                                 v.tm.tv_sec = 0;
 545                                 v.tm.tv_usec = 0;
 546                         } else {
 547                                 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
 548                                 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
 549                         }
 550                         break;
 551
 552                 case SO_SNDTIMEO:
 553                         lv=sizeof(struct timeval);
 554                         if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
 555                                 v.tm.tv_sec = 0;
 556                                 v.tm.tv_usec = 0;
 557                         } else {
 558                                 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
 559                                 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
 560                         }
 561                         break;
 562
 563                 case SO_RCVLOWAT:
 564                         v.val = sk->sk_rcvlowat;
 565                         break;
 566
 567                 case SO_SNDLOWAT:
 568                         v.val=1;
 569                         break;
 570
 571                 case SO_PASSCRED:
 572                         v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
 573                         break;
 574
 575                 case SO_PEERCRED:
 576                         if (len > sizeof(sk->sk_peercred))
 577                                 len = sizeof(sk->sk_peercred);
 578                         if (copy_to_user(optval, &sk->sk_peercred, len))
 579                                 return -EFAULT;
 580                         goto lenout;
 581
 582                 case SO_PEERNAME:
 583                 {
 584                         char address[128];
 585
 586                         if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
 587                                 return -ENOTCONN;
 588                         if (lv < len)
 589                                 return -EINVAL;
 590                         if (copy_to_user(optval, address, len))
 591                                 return -EFAULT;
 592                         goto lenout;
 593                 }
 594
 595                 /* Dubious BSD thing... Probably nobody even uses it, but
 596                  * the UNIX standard wants it for whatever reason... -DaveM
 597                  */
 598                 case SO_ACCEPTCONN:
 599                         v.val = sk->sk_state == TCP_LISTEN;
 600                         break;
 601
 602                 case SO_PEERSEC:
 603                         return security_socket_getpeersec(sock, optval, optlen, len);
 604
 605                 default:
 606                         return(-ENOPROTOOPT);
 607         }
 608         if (len > lv)
 609                 len = lv;
 610         if (copy_to_user(optval, &v, len))
 611                 return -EFAULT;
 612 lenout:
 613         if (put_user(len, optlen))
 614                 return -EFAULT;
 615         return 0;
 616 }
 617
 618 /**
 619  *      sk_alloc - All socket objects are allocated here
 620  *      @family - protocol family
 621  *      @priority - for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
 622  *      @prot - struct proto associated with this new sock instance
 623  *      @zero_it - if we should zero the newly allocated sock
 624  */
 625 struct sock *sk_alloc(int family, int priority, struct proto *prot, int zero_it)
 626 {
 627         struct sock *sk = NULL;
 628         kmem_cache_t *slab = prot->slab;
 629
 630         if (slab != NULL)
 631                 sk = kmem_cache_alloc(slab, priority);
 632         else
 633                 sk = kmalloc(prot->obj_size, priority);
 634
 635         if (sk) {
 636                 if (zero_it) {
 637                         memset(sk, 0, prot->obj_size);
 638                         sk->sk_family = family;
 639                         sk->sk_prot = prot;
 640                         sock_lock_init(sk);
 641                 }
 642
 643                 if (security_sk_alloc(sk, family, priority)) {
 644                         if (slab != NULL)
 645                                 kmem_cache_free(slab, sk);
 646                         else
 647                                 kfree(sk);
 648                         sk = NULL;
 649                 } else
 650                         __module_get(prot->owner);
 651         }
 652         return sk;
 653 }
 654
 655 void sk_free(struct sock *sk)
 656 {
 657         struct sk_filter *filter;
 658         struct module *owner = sk->sk_prot->owner;
 659
 660         if (sk->sk_destruct)
 661                 sk->sk_destruct(sk);
 662
 663         filter = sk->sk_filter;
 664         if (filter) {
 665                 sk_filter_release(sk, filter);
 666                 sk->sk_filter = NULL;
 667         }
 668
 669         sock_disable_timestamp(sk);
 670
 671         if (atomic_read(&sk->sk_omem_alloc))
 672                 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
 673                        __FUNCTION__, atomic_read(&sk->sk_omem_alloc));
 674
 675         security_sk_free(sk);
 676         if (sk->sk_prot->slab != NULL)
 677                 kmem_cache_free(sk->sk_prot->slab, sk);
 678         else
 679                 kfree(sk);
 680         module_put(owner);
 681 }
 682
 683 void __init sk_init(void)
 684 {
 685         if (num_physpages <= 4096) {
 686                 sysctl_wmem_max = 32767;
 687                 sysctl_rmem_max = 32767;
 688                 sysctl_wmem_default = 32767;
 689                 sysctl_rmem_default = 32767;
 690         } else if (num_physpages >= 131072) {
 691                 sysctl_wmem_max = 131071;
 692                 sysctl_rmem_max = 131071;
 693         }
 694 }
 695
 696 /*
 697  *      Simple resource managers for sockets.
 698  */
 699
 700
 701 /*
 702  * Write buffer destructor automatically called from kfree_skb.
 703  */
 704 void sock_wfree(struct sk_buff *skb)
 705 {
 706         struct sock *sk = skb->sk;
 707
 708         /* In case it might be waiting for more memory. */
 709         atomic_sub(skb->truesize, &sk->sk_wmem_alloc);
 710         if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE))
 711                 sk->sk_write_space(sk);
 712         sock_put(sk);
 713 }
 714
 715 /*
 716  * Read buffer destructor automatically called from kfree_skb.
 717  */
 718 void sock_rfree(struct sk_buff *skb)
 719 {
 720         struct sock *sk = skb->sk;
 721
 722         atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
 723 }
 724
 725
 726 int sock_i_uid(struct sock *sk)
 727 {
 728         int uid;
 729
 730         read_lock(&sk->sk_callback_lock);
 731         uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
 732         read_unlock(&sk->sk_callback_lock);
 733         return uid;
 734 }
 735
 736 unsigned long sock_i_ino(struct sock *sk)
 737 {
 738         unsigned long ino;
 739
 740         read_lock(&sk->sk_callback_lock);
 741         ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
 742         read_unlock(&sk->sk_callback_lock);
 743         return ino;
 744 }
 745
 746 /*
 747  * Allocate a skb from the socket's send buffer.
 748  */
 749 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, int priority)
 750 {
 751         if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
 752                 struct sk_buff * skb = alloc_skb(size, priority);
 753                 if (skb) {
 754                         skb_set_owner_w(skb, sk);
 755                         return skb;
 756                 }
 757         }
 758         return NULL;
 759 }
 760
 761 /*
 762  * Allocate a skb from the socket's receive buffer.
 763  */
 764 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force, int priority)
 765 {
 766         if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
 767                 struct sk_buff *skb = alloc_skb(size, priority);
 768                 if (skb) {
 769                         skb_set_owner_r(skb, sk);
 770                         return skb;
 771                 }
 772         }
 773         return NULL;
 774 }
 775
 776 /*
 777  * Allocate a memory block from the socket's option memory buffer.
 778  */
 779 void *sock_kmalloc(struct sock *sk, int size, int priority)
 780 {
 781         if ((unsigned)size <= sysctl_optmem_max &&
 782             atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
 783                 void *mem;
 784                 /* First do the add, to avoid the race if kmalloc
 785                  * might sleep.
 786                  */
 787                 atomic_add(size, &sk->sk_omem_alloc);
 788                 mem = kmalloc(size, priority);
 789                 if (mem)
 790                         return mem;
 791                 atomic_sub(size, &sk->sk_omem_alloc);
 792         }
 793         return NULL;
 794 }
 795
 796 /*
 797  * Free an option memory block.
 798  */
 799 void sock_kfree_s(struct sock *sk, void *mem, int size)
 800 {
 801         kfree(mem);
 802         atomic_sub(size, &sk->sk_omem_alloc);
 803 }
 804
 805 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
 806    I think, these locks should be removed for datagram sockets.
 807  */
 808 static long sock_wait_for_wmem(struct sock * sk, long timeo)
 809 {
 810         DEFINE_WAIT(wait);
 811
 812         clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
 813         for (;;) {
 814                 if (!timeo)
 815                         break;
 816                 if (signal_pending(current))
 817                         break;
 818                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
 819                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
 820                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
 821                         break;
 822                 if (sk->sk_shutdown & SEND_SHUTDOWN)
 823                         break;
 824                 if (sk->sk_err)
 825                         break;
 826                 timeo = schedule_timeout(timeo);
 827         }
 828         finish_wait(sk->sk_sleep, &wait);
 829         return timeo;
 830 }
 831
 832
 833 /*
 834  *      Generic send/receive buffer handlers
 835  */
 836
 837 static struct sk_buff *sock_alloc_send_pskb(struct sock *sk,
 838                                             unsigned long header_len,
 839                                             unsigned long data_len,
 840                                             int noblock, int *errcode)
 841 {
 842         struct sk_buff *skb;
 843         unsigned int gfp_mask;
 844         long timeo;
 845         int err;
 846
 847         gfp_mask = sk->sk_allocation;
 848         if (gfp_mask & __GFP_WAIT)
 849                 gfp_mask |= __GFP_REPEAT;
 850
 851         timeo = sock_sndtimeo(sk, noblock);
 852         while (1) {
 853                 err = sock_error(sk);
 854                 if (err != 0)
 855                         goto failure;
 856
 857                 err = -EPIPE;
 858                 if (sk->sk_shutdown & SEND_SHUTDOWN)
 859                         goto failure;
 860
 861                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
 862                         skb = alloc_skb(header_len, sk->sk_allocation);
 863                         if (skb) {
 864                                 int npages;
 865                                 int i;
 866
 867                                 /* No pages, we're done... */
 868                                 if (!data_len)
 869                                         break;
 870
 871                                 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
 872                                 skb->truesize += data_len;
 873                                 skb_shinfo(skb)->nr_frags = npages;
 874                                 for (i = 0; i < npages; i++) {
 875                                         struct page *page;
 876                                         skb_frag_t *frag;
 877
 878                                         page = alloc_pages(sk->sk_allocation, 0);
 879                                         if (!page) {
 880                                                 err = -ENOBUFS;
 881                                                 skb_shinfo(skb)->nr_frags = i;
 882                                                 kfree_skb(skb);
 883                                                 goto failure;
 884                                         }
 885
 886                                         frag = &skb_shinfo(skb)->frags[i];
 887                                         frag->page = page;
 888                                         frag->page_offset = 0;
 889                                         frag->size = (data_len >= PAGE_SIZE ?
 890                                                       PAGE_SIZE :
 891                                                       data_len);
 892                                         data_len -= PAGE_SIZE;
 893                                 }
 894
 895                                 /* Full success... */
 896                                 break;
 897                         }
 898                         err = -ENOBUFS;
 899                         goto failure;
 900                 }
 901                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
 902                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
 903                 err = -EAGAIN;
 904                 if (!timeo)
 905                         goto failure;
 906                 if (signal_pending(current))
 907                         goto interrupted;
 908                 timeo = sock_wait_for_wmem(sk, timeo);
 909         }
 910
 911         skb_set_owner_w(skb, sk);
 912         return skb;
 913
 914 interrupted:
 915         err = sock_intr_errno(timeo);
 916 failure:
 917         *errcode = err;
 918         return NULL;
 919 }
 920
 921 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
 922                                     int noblock, int *errcode)
 923 {
 924         return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
 925 }
 926
 927 static void __lock_sock(struct sock *sk)
 928 {
 929         DEFINE_WAIT(wait);
 930
 931         for(;;) {
 932                 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
 933                                         TASK_UNINTERRUPTIBLE);
 934                 spin_unlock_bh(&sk->sk_lock.slock);
 935                 schedule();
 936                 spin_lock_bh(&sk->sk_lock.slock);
 937                 if(!sock_owned_by_user(sk))
 938                         break;
 939         }
 940         finish_wait(&sk->sk_lock.wq, &wait);
 941 }
 942
 943 static void __release_sock(struct sock *sk)
 944 {
 945         struct sk_buff *skb = sk->sk_backlog.head;
 946
 947         do {
 948                 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
 949                 bh_unlock_sock(sk);
 950
 951                 do {
 952                         struct sk_buff *next = skb->next;
 953
 954                         skb->next = NULL;
 955                         sk->sk_backlog_rcv(sk, skb);
 956
 957                         /*
 958                          * We are in process context here with softirqs
 959                          * disabled, use cond_resched_softirq() to preempt.
 960                          * This is safe to do because we've taken the backlog
 961                          * queue private:
 962                          */
 963                         cond_resched_softirq();
 964
 965                         skb = next;
 966                 } while (skb != NULL);
 967
 968                 bh_lock_sock(sk);
 969         } while((skb = sk->sk_backlog.head) != NULL);
 970 }
 971
 972 /**
 973  * sk_wait_data - wait for data to arrive at sk_receive_queue
 974  * sk - sock to wait on
 975  * timeo - for how long
 976  *
 977  * Now socket state including sk->sk_err is changed only under lock,
 978  * hence we may omit checks after joining wait queue.
 979  * We check receive queue before schedule() only as optimization;
 980  * it is very likely that release_sock() added new data.
 981  */
 982 int sk_wait_data(struct sock *sk, long *timeo)
 983 {
 984         int rc;
 985         DEFINE_WAIT(wait);
 986
 987         prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
 988         set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
 989         rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
 990         clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
 991         finish_wait(sk->sk_sleep, &wait);
 992         return rc;
 993 }
 994
 995 EXPORT_SYMBOL(sk_wait_data);
 996
 997 /*
 998  * Set of default routines for initialising struct proto_ops when
 999  * the protocol does not support a particular function. In certain
1000  * cases where it makes no sense for a protocol to have a "do nothing"
1001  * function, some default processing is provided.
1002  */
1003
1004 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1005 {
1006         return -EOPNOTSUPP;
1007 }
1008
1009 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1010                     int len, int flags)
1011 {
1012         return -EOPNOTSUPP;
1013 }
1014
1015 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1016 {
1017         return -EOPNOTSUPP;
1018 }
1019
1020 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1021 {
1022         return -EOPNOTSUPP;
1023 }
1024
1025 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1026                     int *len, int peer)
1027 {
1028         return -EOPNOTSUPP;
1029 }
1030
1031 unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt)
1032 {
1033         return 0;
1034 }
1035
1036 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1037 {
1038         return -EOPNOTSUPP;
1039 }
1040
1041 int sock_no_listen(struct socket *sock, int backlog)
1042 {
1043         return -EOPNOTSUPP;
1044 }
1045
1046 int sock_no_shutdown(struct socket *sock, int how)
1047 {
1048         return -EOPNOTSUPP;
1049 }
1050
1051 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1052                     char __user *optval, int optlen)
1053 {
1054         return -EOPNOTSUPP;
1055 }
1056
1057 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1058                     char __user *optval, int __user *optlen)
1059 {
1060         return -EOPNOTSUPP;
1061 }
1062
1063 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1064                     size_t len)
1065 {
1066         return -EOPNOTSUPP;
1067 }
1068
1069 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1070                     size_t len, int flags)
1071 {
1072         return -EOPNOTSUPP;
1073 }
1074
1075 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1076 {
1077         /* Mirror missing mmap method error code */
1078         return -ENODEV;
1079 }
1080
1081 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1082 {
1083         ssize_t res;
1084         struct msghdr msg = {.msg_flags = flags};
1085         struct kvec iov;
1086         char *kaddr = kmap(page);
1087         iov.iov_base = kaddr + offset;
1088         iov.iov_len = size;
1089         res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1090         kunmap(page);
1091         return res;
1092 }
1093
1094 /*
1095  *      Default Socket Callbacks
1096  */
1097
1098 static void sock_def_wakeup(struct sock *sk)
1099 {
1100         read_lock(&sk->sk_callback_lock);
1101         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1102                 wake_up_interruptible_all(sk->sk_sleep);
1103         read_unlock(&sk->sk_callback_lock);
1104 }
1105
1106 static void sock_def_error_report(struct sock *sk)
1107 {
1108         read_lock(&sk->sk_callback_lock);
1109         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1110                 wake_up_interruptible(sk->sk_sleep);
1111         sk_wake_async(sk,0,POLL_ERR);
1112         read_unlock(&sk->sk_callback_lock);
1113 }
1114
1115 static void sock_def_readable(struct sock *sk, int len)
1116 {
1117         read_lock(&sk->sk_callback_lock);
1118         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1119                 wake_up_interruptible(sk->sk_sleep);
1120         sk_wake_async(sk,1,POLL_IN);
1121         read_unlock(&sk->sk_callback_lock);
1122 }
1123
1124 static void sock_def_write_space(struct sock *sk)
1125 {
1126         read_lock(&sk->sk_callback_lock);
1127
1128         /* Do not wake up a writer until he can make "significant"
1129          * progress.  --DaveM
1130          */
1131         if((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1132                 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1133                         wake_up_interruptible(sk->sk_sleep);
1134
1135                 /* Should agree with poll, otherwise some programs break */
1136                 if (sock_writeable(sk))
1137                         sk_wake_async(sk, 2, POLL_OUT);
1138         }
1139
1140         read_unlock(&sk->sk_callback_lock);
1141 }
1142
1143 static void sock_def_destruct(struct sock *sk)
1144 {
1145         if (sk->sk_protinfo)
1146                 kfree(sk->sk_protinfo);
1147 }
1148
1149 void sk_send_sigurg(struct sock *sk)
1150 {
1151         if (sk->sk_socket && sk->sk_socket->file)
1152                 if (send_sigurg(&sk->sk_socket->file->f_owner))
1153                         sk_wake_async(sk, 3, POLL_PRI);
1154 }
1155
1156 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1157                     unsigned long expires)
1158 {
1159         if (!mod_timer(timer, expires))
1160                 sock_hold(sk);
1161 }
1162
1163 EXPORT_SYMBOL(sk_reset_timer);
1164
1165 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1166 {
1167         if (timer_pending(timer) && del_timer(timer))
1168                 __sock_put(sk);
1169 }
1170
1171 EXPORT_SYMBOL(sk_stop_timer);
1172
1173 void sock_init_data(struct socket *sock, struct sock *sk)
1174 {
1175         skb_queue_head_init(&sk->sk_receive_queue);
1176         skb_queue_head_init(&sk->sk_write_queue);
1177         skb_queue_head_init(&sk->sk_error_queue);
1178
1179         sk->sk_send_head        =       NULL;
1180
1181         init_timer(&sk->sk_timer);
1182
1183         sk->sk_allocation       =       GFP_KERNEL;
1184         sk->sk_rcvbuf           =       sysctl_rmem_default;
1185         sk->sk_sndbuf           =       sysctl_wmem_default;
1186         sk->sk_state            =       TCP_CLOSE;
1187         sk->sk_socket           =       sock;
1188
1189         sock_set_flag(sk, SOCK_ZAPPED);
1190
1191         if(sock)
1192         {
1193                 sk->sk_type     =       sock->type;
1194                 sk->sk_sleep    =       &sock->wait;
1195                 sock->sk        =       sk;
1196         } else
1197                 sk->sk_sleep    =       NULL;
1198
1199         rwlock_init(&sk->sk_dst_lock);
1200         rwlock_init(&sk->sk_callback_lock);
1201
1202         sk->sk_state_change     =       sock_def_wakeup;
1203         sk->sk_data_ready       =       sock_def_readable;
1204         sk->sk_write_space      =       sock_def_write_space;
1205         sk->sk_error_report     =       sock_def_error_report;
1206         sk->sk_destruct         =       sock_def_destruct;
1207
1208         sk->sk_sndmsg_page      =       NULL;
1209         sk->sk_sndmsg_off       =       0;
1210
1211         sk->sk_peercred.pid     =       0;
1212         sk->sk_peercred.uid     =       -1;
1213         sk->sk_peercred.gid     =       -1;
1214         sk->sk_write_pending    =       0;
1215         sk->sk_rcvlowat         =       1;
1216         sk->sk_rcvtimeo         =       MAX_SCHEDULE_TIMEOUT;
1217         sk->sk_sndtimeo         =       MAX_SCHEDULE_TIMEOUT;
1218
1219         sk->sk_stamp.tv_sec     = -1L;
1220         sk->sk_stamp.tv_usec    = -1L;
1221
1222         atomic_set(&sk->sk_refcnt, 1);
1223 }
1224
1225 void fastcall lock_sock(struct sock *sk)
1226 {
1227         might_sleep();
1228         spin_lock_bh(&(sk->sk_lock.slock));
1229         if (sk->sk_lock.owner)
1230                 __lock_sock(sk);
1231         sk->sk_lock.owner = (void *)1;
1232         spin_unlock_bh(&(sk->sk_lock.slock));
1233 }
1234
1235 EXPORT_SYMBOL(lock_sock);
1236
1237 void fastcall release_sock(struct sock *sk)
1238 {
1239         spin_lock_bh(&(sk->sk_lock.slock));
1240         if (sk->sk_backlog.tail)
1241                 __release_sock(sk);
1242         sk->sk_lock.owner = NULL;
1243         if (waitqueue_active(&(sk->sk_lock.wq)))
1244                 wake_up(&(sk->sk_lock.wq));
1245         spin_unlock_bh(&(sk->sk_lock.slock));
1246 }
1247 EXPORT_SYMBOL(release_sock);
1248
1249 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1250 {
1251         if (!sock_flag(sk, SOCK_TIMESTAMP))
1252                 sock_enable_timestamp(sk);
1253         if (sk->sk_stamp.tv_sec == -1)
1254                 return -ENOENT;
1255         if (sk->sk_stamp.tv_sec == 0)
1256                 do_gettimeofday(&sk->sk_stamp);
1257         return copy_to_user(userstamp, &sk->sk_stamp, sizeof(struct timeval)) ?
1258                 -EFAULT : 0;
1259 }
1260 EXPORT_SYMBOL(sock_get_timestamp);
1261
1262 void sock_enable_timestamp(struct sock *sk)
1263 {
1264         if (!sock_flag(sk, SOCK_TIMESTAMP)) {
1265                 sock_set_flag(sk, SOCK_TIMESTAMP);
1266                 net_enable_timestamp();
1267         }
1268 }
1269 EXPORT_SYMBOL(sock_enable_timestamp);
1270
1271 /*
1272  *      Get a socket option on an socket.
1273  *
1274  *      FIX: POSIX 1003.1g is very ambiguous here. It states that
1275  *      asynchronous errors should be reported by getsockopt. We assume
1276  *      this means if you specify SO_ERROR (otherwise whats the point of it).
1277  */
1278 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1279                            char __user *optval, int __user *optlen)
1280 {
1281         struct sock *sk = sock->sk;
1282
1283         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1284 }
1285
1286 EXPORT_SYMBOL(sock_common_getsockopt);
1287
1288 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1289                         struct msghdr *msg, size_t size, int flags)
1290 {
1291         struct sock *sk = sock->sk;
1292         int addr_len = 0;
1293         int err;
1294
1295         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
1296                                    flags & ~MSG_DONTWAIT, &addr_len);
1297         if (err >= 0)
1298                 msg->msg_namelen = addr_len;
1299         return err;
1300 }
1301
1302 EXPORT_SYMBOL(sock_common_recvmsg);
1303
1304 /*
1305  *      Set socket options on an inet socket.
1306  */
1307 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1308                            char __user *optval, int optlen)
1309 {
1310         struct sock *sk = sock->sk;
1311
1312         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1313 }
1314
1315 EXPORT_SYMBOL(sock_common_setsockopt);
1316
1317 void sk_common_release(struct sock *sk)
1318 {
1319         if (sk->sk_prot->destroy)
1320                 sk->sk_prot->destroy(sk);
1321
1322         /*
1323          * Observation: when sock_common_release is called, processes have
1324          * no access to socket. But net still has.
1325          * Step one, detach it from networking:
1326          *
1327          * A. Remove from hash tables.
1328          */
1329
1330         sk->sk_prot->unhash(sk);
1331
1332         /*
1333          * In this point socket cannot receive new packets, but it is possible
1334          * that some packets are in flight because some CPU runs receiver and
1335          * did hash table lookup before we unhashed socket. They will achieve
1336          * receive queue and will be purged by socket destructor.
1337          *
1338          * Also we still have packets pending on receive queue and probably,
1339          * our own packets waiting in device queues. sock_destroy will drain
1340          * receive queue, but transmitted packets will delay socket destruction
1341          * until the last reference will be released.
1342          */
1343
1344         sock_orphan(sk);
1345
1346         xfrm_sk_free_policy(sk);
1347
1348 #ifdef INET_REFCNT_DEBUG
1349         if (atomic_read(&sk->sk_refcnt) != 1)
1350                 printk(KERN_DEBUG "Destruction of the socket %p delayed, c=%d\n",
1351                        sk, atomic_read(&sk->sk_refcnt));
1352 #endif
1353         sock_put(sk);
1354 }
1355
1356 EXPORT_SYMBOL(sk_common_release);
1357
1358 static DEFINE_RWLOCK(proto_list_lock);
1359 static LIST_HEAD(proto_list);
1360
1361 int proto_register(struct proto *prot, int alloc_slab)
1362 {
1363         int rc = -ENOBUFS;
1364
1365         if (alloc_slab) {
1366                 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
1367                                                SLAB_HWCACHE_ALIGN, NULL, NULL);
1368
1369                 if (prot->slab == NULL) {
1370                         printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
1371                                prot->name);
1372                         goto out;
1373                 }
1374         }
1375
1376         write_lock(&proto_list_lock);
1377         list_add(&prot->node, &proto_list);
1378         write_unlock(&proto_list_lock);
1379         rc = 0;
1380 out:
1381         return rc;
1382 }
1383
1384 EXPORT_SYMBOL(proto_register);
1385
1386 void proto_unregister(struct proto *prot)
1387 {
1388         write_lock(&proto_list_lock);
1389
1390         if (prot->slab != NULL) {
1391                 kmem_cache_destroy(prot->slab);
1392                 prot->slab = NULL;
1393         }
1394
1395         list_del(&prot->node);
1396         write_unlock(&proto_list_lock);
1397 }
1398
1399 EXPORT_SYMBOL(proto_unregister);
1400
1401 #ifdef CONFIG_PROC_FS
1402 static inline struct proto *__proto_head(void)
1403 {
1404         return list_entry(proto_list.next, struct proto, node);
1405 }
1406
1407 static inline struct proto *proto_head(void)
1408 {
1409         return list_empty(&proto_list) ? NULL : __proto_head();
1410 }
1411
1412 static inline struct proto *proto_next(struct proto *proto)
1413 {
1414         return proto->node.next == &proto_list ? NULL :
1415                 list_entry(proto->node.next, struct proto, node);
1416 }
1417
1418 static inline struct proto *proto_get_idx(loff_t pos)
1419 {
1420         struct proto *proto;
1421         loff_t i = 0;
1422
1423         list_for_each_entry(proto, &proto_list, node)
1424                 if (i++ == pos)
1425                         goto out;
1426
1427         proto = NULL;
1428 out:
1429         return proto;
1430 }
1431
1432 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
1433 {
1434         read_lock(&proto_list_lock);
1435         return *pos ? proto_get_idx(*pos - 1) : SEQ_START_TOKEN;
1436 }
1437
1438 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1439 {
1440         ++*pos;
1441         return v == SEQ_START_TOKEN ? proto_head() : proto_next(v);
1442 }
1443
1444 static void proto_seq_stop(struct seq_file *seq, void *v)
1445 {
1446         read_unlock(&proto_list_lock);
1447 }
1448
1449 static char proto_method_implemented(const void *method)
1450 {
1451         return method == NULL ? 'n' : 'y';
1452 }
1453
1454 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
1455 {
1456         seq_printf(seq, "%-9s %4u %6d  %6d   %-3s %6u   %-3s  %-10s "
1457                         "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
1458                    proto->name,
1459                    proto->obj_size,
1460                    proto->sockets_allocated != NULL ? atomic_read(proto->sockets_allocated) : -1,
1461                    proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
1462                    proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
1463                    proto->max_header,
1464                    proto->slab == NULL ? "no" : "yes",
1465                    module_name(proto->owner),
1466                    proto_method_implemented(proto->close),
1467                    proto_method_implemented(proto->connect),
1468                    proto_method_implemented(proto->disconnect),
1469                    proto_method_implemented(proto->accept),
1470                    proto_method_implemented(proto->ioctl),
1471                    proto_method_implemented(proto->init),
1472                    proto_method_implemented(proto->destroy),
1473                    proto_method_implemented(proto->shutdown),
1474                    proto_method_implemented(proto->setsockopt),
1475                    proto_method_implemented(proto->getsockopt),
1476                    proto_method_implemented(proto->sendmsg),
1477                    proto_method_implemented(proto->recvmsg),
1478                    proto_method_implemented(proto->sendpage),
1479                    proto_method_implemented(proto->bind),
1480                    proto_method_implemented(proto->backlog_rcv),
1481                    proto_method_implemented(proto->hash),
1482                    proto_method_implemented(proto->unhash),
1483                    proto_method_implemented(proto->get_port),
1484                    proto_method_implemented(proto->enter_memory_pressure));
1485 }
1486
1487 static int proto_seq_show(struct seq_file *seq, void *v)
1488 {
1489         if (v == SEQ_START_TOKEN)
1490                 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
1491                            "protocol",
1492                            "size",
1493                            "sockets",
1494                            "memory",
1495                            "press",
1496                            "maxhdr",
1497                            "slab",
1498                            "module",
1499                            "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
1500         else
1501                 proto_seq_printf(seq, v);
1502         return 0;
1503 }
1504
1505 static struct seq_operations proto_seq_ops = {
1506         .start  = proto_seq_start,
1507         .next   = proto_seq_next,
1508         .stop   = proto_seq_stop,
1509         .show   = proto_seq_show,
1510 };
1511
1512 static int proto_seq_open(struct inode *inode, struct file *file)
1513 {
1514         return seq_open(file, &proto_seq_ops);
1515 }
1516
1517 static struct file_operations proto_seq_fops = {
1518         .owner          = THIS_MODULE,
1519         .open           = proto_seq_open,
1520         .read           = seq_read,
1521         .llseek         = seq_lseek,
1522         .release        = seq_release,
1523 };
1524
1525 static int __init proto_init(void)
1526 {
1527         /* register /proc/net/protocols */
1528         return proc_net_fops_create("protocols", S_IRUGO, &proto_seq_fops) == NULL ? -ENOBUFS : 0;
1529 }
1530
1531 subsys_initcall(proto_init);
1532
1533 #endif /* PROC_FS */
1534
1535 EXPORT_SYMBOL(sk_alloc);
1536 EXPORT_SYMBOL(sk_free);
1537 EXPORT_SYMBOL(sk_send_sigurg);
1538 EXPORT_SYMBOL(sock_alloc_send_skb);
1539 EXPORT_SYMBOL(sock_init_data);
1540 EXPORT_SYMBOL(sock_kfree_s);
1541 EXPORT_SYMBOL(sock_kmalloc);
1542 EXPORT_SYMBOL(sock_no_accept);
1543 EXPORT_SYMBOL(sock_no_bind);
1544 EXPORT_SYMBOL(sock_no_connect);
1545 EXPORT_SYMBOL(sock_no_getname);
1546 EXPORT_SYMBOL(sock_no_getsockopt);
1547 EXPORT_SYMBOL(sock_no_ioctl);
1548 EXPORT_SYMBOL(sock_no_listen);
1549 EXPORT_SYMBOL(sock_no_mmap);
1550 EXPORT_SYMBOL(sock_no_poll);
1551 EXPORT_SYMBOL(sock_no_recvmsg);
1552 EXPORT_SYMBOL(sock_no_sendmsg);
1553 EXPORT_SYMBOL(sock_no_sendpage);
1554 EXPORT_SYMBOL(sock_no_setsockopt);
1555 EXPORT_SYMBOL(sock_no_shutdown);
1556 EXPORT_SYMBOL(sock_no_socketpair);
1557 EXPORT_SYMBOL(sock_rfree);
1558 EXPORT_SYMBOL(sock_setsockopt);
1559 EXPORT_SYMBOL(sock_wfree);
1560 EXPORT_SYMBOL(sock_wmalloc);
1561 EXPORT_SYMBOL(sock_i_uid);
1562 EXPORT_SYMBOL(sock_i_ino);
1563 #ifdef CONFIG_SYSCTL
1564 EXPORT_SYMBOL(sysctl_optmem_max);
1565 EXPORT_SYMBOL(sysctl_rmem_max);
1566 EXPORT_SYMBOL(sysctl_wmem_max);
1567 #endif