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                         kmem_cache_free(slab, sk);
 645                         sk = NULL;
 646                 } else
 647                         __module_get(prot->owner);
 648         }
 649         return sk;
 650 }
 651
 652 void sk_free(struct sock *sk)
 653 {
 654         struct sk_filter *filter;
 655         struct module *owner = sk->sk_prot->owner;
 656
 657         if (sk->sk_destruct)
 658                 sk->sk_destruct(sk);
 659
 660         filter = sk->sk_filter;
 661         if (filter) {
 662                 sk_filter_release(sk, filter);
 663                 sk->sk_filter = NULL;
 664         }
 665
 666         sock_disable_timestamp(sk);
 667
 668         if (atomic_read(&sk->sk_omem_alloc))
 669                 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
 670                        __FUNCTION__, atomic_read(&sk->sk_omem_alloc));
 671
 672         security_sk_free(sk);
 673         if (sk->sk_prot->slab != NULL)
 674                 kmem_cache_free(sk->sk_prot->slab, sk);
 675         else
 676                 kfree(sk);
 677         module_put(owner);
 678 }
 679
 680 void __init sk_init(void)
 681 {
 682         if (num_physpages <= 4096) {
 683                 sysctl_wmem_max = 32767;
 684                 sysctl_rmem_max = 32767;
 685                 sysctl_wmem_default = 32767;
 686                 sysctl_rmem_default = 32767;
 687         } else if (num_physpages >= 131072) {
 688                 sysctl_wmem_max = 131071;
 689                 sysctl_rmem_max = 131071;
 690         }
 691 }
 692
 693 /*
 694  *      Simple resource managers for sockets.
 695  */
 696
 697
 698 /*
 699  * Write buffer destructor automatically called from kfree_skb.
 700  */
 701 void sock_wfree(struct sk_buff *skb)
 702 {
 703         struct sock *sk = skb->sk;
 704
 705         /* In case it might be waiting for more memory. */
 706         atomic_sub(skb->truesize, &sk->sk_wmem_alloc);
 707         if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE))
 708                 sk->sk_write_space(sk);
 709         sock_put(sk);
 710 }
 711
 712 /*
 713  * Read buffer destructor automatically called from kfree_skb.
 714  */
 715 void sock_rfree(struct sk_buff *skb)
 716 {
 717         struct sock *sk = skb->sk;
 718
 719         atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
 720 }
 721
 722
 723 int sock_i_uid(struct sock *sk)
 724 {
 725         int uid;
 726
 727         read_lock(&sk->sk_callback_lock);
 728         uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
 729         read_unlock(&sk->sk_callback_lock);
 730         return uid;
 731 }
 732
 733 unsigned long sock_i_ino(struct sock *sk)
 734 {
 735         unsigned long ino;
 736
 737         read_lock(&sk->sk_callback_lock);
 738         ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
 739         read_unlock(&sk->sk_callback_lock);
 740         return ino;
 741 }
 742
 743 /*
 744  * Allocate a skb from the socket's send buffer.
 745  */
 746 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, int priority)
 747 {
 748         if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
 749                 struct sk_buff * skb = alloc_skb(size, priority);
 750                 if (skb) {
 751                         skb_set_owner_w(skb, sk);
 752                         return skb;
 753                 }
 754         }
 755         return NULL;
 756 }
 757
 758 /*
 759  * Allocate a skb from the socket's receive buffer.
 760  */
 761 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force, int priority)
 762 {
 763         if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
 764                 struct sk_buff *skb = alloc_skb(size, priority);
 765                 if (skb) {
 766                         skb_set_owner_r(skb, sk);
 767                         return skb;
 768                 }
 769         }
 770         return NULL;
 771 }
 772
 773 /*
 774  * Allocate a memory block from the socket's option memory buffer.
 775  */
 776 void *sock_kmalloc(struct sock *sk, int size, int priority)
 777 {
 778         if ((unsigned)size <= sysctl_optmem_max &&
 779             atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
 780                 void *mem;
 781                 /* First do the add, to avoid the race if kmalloc
 782                  * might sleep.
 783                  */
 784                 atomic_add(size, &sk->sk_omem_alloc);
 785                 mem = kmalloc(size, priority);
 786                 if (mem)
 787                         return mem;
 788                 atomic_sub(size, &sk->sk_omem_alloc);
 789         }
 790         return NULL;
 791 }
 792
 793 /*
 794  * Free an option memory block.
 795  */
 796 void sock_kfree_s(struct sock *sk, void *mem, int size)
 797 {
 798         kfree(mem);
 799         atomic_sub(size, &sk->sk_omem_alloc);
 800 }
 801
 802 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
 803    I think, these locks should be removed for datagram sockets.
 804  */
 805 static long sock_wait_for_wmem(struct sock * sk, long timeo)
 806 {
 807         DEFINE_WAIT(wait);
 808
 809         clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
 810         for (;;) {
 811                 if (!timeo)
 812                         break;
 813                 if (signal_pending(current))
 814                         break;
 815                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
 816                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
 817                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
 818                         break;
 819                 if (sk->sk_shutdown & SEND_SHUTDOWN)
 820                         break;
 821                 if (sk->sk_err)
 822                         break;
 823                 timeo = schedule_timeout(timeo);
 824         }
 825         finish_wait(sk->sk_sleep, &wait);
 826         return timeo;
 827 }
 828
 829
 830 /*
 831  *      Generic send/receive buffer handlers
 832  */
 833
 834 static struct sk_buff *sock_alloc_send_pskb(struct sock *sk,
 835                                             unsigned long header_len,
 836                                             unsigned long data_len,
 837                                             int noblock, int *errcode)
 838 {
 839         struct sk_buff *skb;
 840         unsigned int gfp_mask;
 841         long timeo;
 842         int err;
 843
 844         gfp_mask = sk->sk_allocation;
 845         if (gfp_mask & __GFP_WAIT)
 846                 gfp_mask |= __GFP_REPEAT;
 847
 848         timeo = sock_sndtimeo(sk, noblock);
 849         while (1) {
 850                 err = sock_error(sk);
 851                 if (err != 0)
 852                         goto failure;
 853
 854                 err = -EPIPE;
 855                 if (sk->sk_shutdown & SEND_SHUTDOWN)
 856                         goto failure;
 857
 858                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
 859                         skb = alloc_skb(header_len, sk->sk_allocation);
 860                         if (skb) {
 861                                 int npages;
 862                                 int i;
 863
 864                                 /* No pages, we're done... */
 865                                 if (!data_len)
 866                                         break;
 867
 868                                 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
 869                                 skb->truesize += data_len;
 870                                 skb_shinfo(skb)->nr_frags = npages;
 871                                 for (i = 0; i < npages; i++) {
 872                                         struct page *page;
 873                                         skb_frag_t *frag;
 874
 875                                         page = alloc_pages(sk->sk_allocation, 0);
 876                                         if (!page) {
 877                                                 err = -ENOBUFS;
 878                                                 skb_shinfo(skb)->nr_frags = i;
 879                                                 kfree_skb(skb);
 880                                                 goto failure;
 881                                         }
 882
 883                                         frag = &skb_shinfo(skb)->frags[i];
 884                                         frag->page = page;
 885                                         frag->page_offset = 0;
 886                                         frag->size = (data_len >= PAGE_SIZE ?
 887                                                       PAGE_SIZE :
 888                                                       data_len);
 889                                         data_len -= PAGE_SIZE;
 890                                 }
 891
 892                                 /* Full success... */
 893                                 break;
 894                         }
 895                         err = -ENOBUFS;
 896                         goto failure;
 897                 }
 898                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
 899                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
 900                 err = -EAGAIN;
 901                 if (!timeo)
 902                         goto failure;
 903                 if (signal_pending(current))
 904                         goto interrupted;
 905                 timeo = sock_wait_for_wmem(sk, timeo);
 906         }
 907
 908         skb_set_owner_w(skb, sk);
 909         return skb;
 910
 911 interrupted:
 912         err = sock_intr_errno(timeo);
 913 failure:
 914         *errcode = err;
 915         return NULL;
 916 }
 917
 918 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
 919                                     int noblock, int *errcode)
 920 {
 921         return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
 922 }
 923
 924 static void __lock_sock(struct sock *sk)
 925 {
 926         DEFINE_WAIT(wait);
 927
 928         for(;;) {
 929                 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
 930                                         TASK_UNINTERRUPTIBLE);
 931                 spin_unlock_bh(&sk->sk_lock.slock);
 932                 schedule();
 933                 spin_lock_bh(&sk->sk_lock.slock);
 934                 if(!sock_owned_by_user(sk))
 935                         break;
 936         }
 937         finish_wait(&sk->sk_lock.wq, &wait);
 938 }
 939
 940 static void __release_sock(struct sock *sk)
 941 {
 942         struct sk_buff *skb = sk->sk_backlog.head;
 943
 944         do {
 945                 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
 946                 bh_unlock_sock(sk);
 947
 948                 do {
 949                         struct sk_buff *next = skb->next;
 950
 951                         skb->next = NULL;
 952                         sk->sk_backlog_rcv(sk, skb);
 953
 954                         /*
 955                          * We are in process context here with softirqs
 956                          * disabled, use cond_resched_softirq() to preempt.
 957                          * This is safe to do because we've taken the backlog
 958                          * queue private:
 959                          */
 960                         cond_resched_softirq();
 961
 962                         skb = next;
 963                 } while (skb != NULL);
 964
 965                 bh_lock_sock(sk);
 966         } while((skb = sk->sk_backlog.head) != NULL);
 967 }
 968
 969 /**
 970  * sk_wait_data - wait for data to arrive at sk_receive_queue
 971  * sk - sock to wait on
 972  * timeo - for how long
 973  *
 974  * Now socket state including sk->sk_err is changed only under lock,
 975  * hence we may omit checks after joining wait queue.
 976  * We check receive queue before schedule() only as optimization;
 977  * it is very likely that release_sock() added new data.
 978  */
 979 int sk_wait_data(struct sock *sk, long *timeo)
 980 {
 981         int rc;
 982         DEFINE_WAIT(wait);
 983
 984         prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
 985         set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
 986         rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
 987         clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
 988         finish_wait(sk->sk_sleep, &wait);
 989         return rc;
 990 }
 991
 992 EXPORT_SYMBOL(sk_wait_data);
 993
 994 /*
 995  * Set of default routines for initialising struct proto_ops when
 996  * the protocol does not support a particular function. In certain
 997  * cases where it makes no sense for a protocol to have a "do nothing"
 998  * function, some default processing is provided.
 999  */
1000
1001 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1002 {
1003         return -EOPNOTSUPP;
1004 }
1005
1006 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1007                     int len, int flags)
1008 {
1009         return -EOPNOTSUPP;
1010 }
1011
1012 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1013 {
1014         return -EOPNOTSUPP;
1015 }
1016
1017 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1018 {
1019         return -EOPNOTSUPP;
1020 }
1021
1022 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1023                     int *len, int peer)
1024 {
1025         return -EOPNOTSUPP;
1026 }
1027
1028 unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt)
1029 {
1030         return 0;
1031 }
1032
1033 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1034 {
1035         return -EOPNOTSUPP;
1036 }
1037
1038 int sock_no_listen(struct socket *sock, int backlog)
1039 {
1040         return -EOPNOTSUPP;
1041 }
1042
1043 int sock_no_shutdown(struct socket *sock, int how)
1044 {
1045         return -EOPNOTSUPP;
1046 }
1047
1048 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1049                     char __user *optval, int optlen)
1050 {
1051         return -EOPNOTSUPP;
1052 }
1053
1054 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1055                     char __user *optval, int __user *optlen)
1056 {
1057         return -EOPNOTSUPP;
1058 }
1059
1060 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1061                     size_t len)
1062 {
1063         return -EOPNOTSUPP;
1064 }
1065
1066 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1067                     size_t len, int flags)
1068 {
1069         return -EOPNOTSUPP;
1070 }
1071
1072 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1073 {
1074         /* Mirror missing mmap method error code */
1075         return -ENODEV;
1076 }
1077
1078 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1079 {
1080         ssize_t res;
1081         struct msghdr msg = {.msg_flags = flags};
1082         struct kvec iov;
1083         char *kaddr = kmap(page);
1084         iov.iov_base = kaddr + offset;
1085         iov.iov_len = size;
1086         res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1087         kunmap(page);
1088         return res;
1089 }
1090
1091 /*
1092  *      Default Socket Callbacks
1093  */
1094
1095 static void sock_def_wakeup(struct sock *sk)
1096 {
1097         read_lock(&sk->sk_callback_lock);
1098         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1099                 wake_up_interruptible_all(sk->sk_sleep);
1100         read_unlock(&sk->sk_callback_lock);
1101 }
1102
1103 static void sock_def_error_report(struct sock *sk)
1104 {
1105         read_lock(&sk->sk_callback_lock);
1106         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1107                 wake_up_interruptible(sk->sk_sleep);
1108         sk_wake_async(sk,0,POLL_ERR);
1109         read_unlock(&sk->sk_callback_lock);
1110 }
1111
1112 static void sock_def_readable(struct sock *sk, int len)
1113 {
1114         read_lock(&sk->sk_callback_lock);
1115         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1116                 wake_up_interruptible(sk->sk_sleep);
1117         sk_wake_async(sk,1,POLL_IN);
1118         read_unlock(&sk->sk_callback_lock);
1119 }
1120
1121 static void sock_def_write_space(struct sock *sk)
1122 {
1123         read_lock(&sk->sk_callback_lock);
1124
1125         /* Do not wake up a writer until he can make "significant"
1126          * progress.  --DaveM
1127          */
1128         if((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1129                 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1130                         wake_up_interruptible(sk->sk_sleep);
1131
1132                 /* Should agree with poll, otherwise some programs break */
1133                 if (sock_writeable(sk))
1134                         sk_wake_async(sk, 2, POLL_OUT);
1135         }
1136
1137         read_unlock(&sk->sk_callback_lock);
1138 }
1139
1140 static void sock_def_destruct(struct sock *sk)
1141 {
1142         if (sk->sk_protinfo)
1143                 kfree(sk->sk_protinfo);
1144 }
1145
1146 void sk_send_sigurg(struct sock *sk)
1147 {
1148         if (sk->sk_socket && sk->sk_socket->file)
1149                 if (send_sigurg(&sk->sk_socket->file->f_owner))
1150                         sk_wake_async(sk, 3, POLL_PRI);
1151 }
1152
1153 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1154                     unsigned long expires)
1155 {
1156         if (!mod_timer(timer, expires))
1157                 sock_hold(sk);
1158 }
1159
1160 EXPORT_SYMBOL(sk_reset_timer);
1161
1162 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1163 {
1164         if (timer_pending(timer) && del_timer(timer))
1165                 __sock_put(sk);
1166 }
1167
1168 EXPORT_SYMBOL(sk_stop_timer);
1169
1170 void sock_init_data(struct socket *sock, struct sock *sk)
1171 {
1172         skb_queue_head_init(&sk->sk_receive_queue);
1173         skb_queue_head_init(&sk->sk_write_queue);
1174         skb_queue_head_init(&sk->sk_error_queue);
1175
1176         sk->sk_send_head        =       NULL;
1177
1178         init_timer(&sk->sk_timer);
1179
1180         sk->sk_allocation       =       GFP_KERNEL;
1181         sk->sk_rcvbuf           =       sysctl_rmem_default;
1182         sk->sk_sndbuf           =       sysctl_wmem_default;
1183         sk->sk_state            =       TCP_CLOSE;
1184         sk->sk_socket           =       sock;
1185
1186         sock_set_flag(sk, SOCK_ZAPPED);
1187
1188         if(sock)
1189         {
1190                 sk->sk_type     =       sock->type;
1191                 sk->sk_sleep    =       &sock->wait;
1192                 sock->sk        =       sk;
1193         } else
1194                 sk->sk_sleep    =       NULL;
1195
1196         rwlock_init(&sk->sk_dst_lock);
1197         rwlock_init(&sk->sk_callback_lock);
1198
1199         sk->sk_state_change     =       sock_def_wakeup;
1200         sk->sk_data_ready       =       sock_def_readable;
1201         sk->sk_write_space      =       sock_def_write_space;
1202         sk->sk_error_report     =       sock_def_error_report;
1203         sk->sk_destruct         =       sock_def_destruct;
1204
1205         sk->sk_sndmsg_page      =       NULL;
1206         sk->sk_sndmsg_off       =       0;
1207
1208         sk->sk_peercred.pid     =       0;
1209         sk->sk_peercred.uid     =       -1;
1210         sk->sk_peercred.gid     =       -1;
1211         sk->sk_write_pending    =       0;
1212         sk->sk_rcvlowat         =       1;
1213         sk->sk_rcvtimeo         =       MAX_SCHEDULE_TIMEOUT;
1214         sk->sk_sndtimeo         =       MAX_SCHEDULE_TIMEOUT;
1215
1216         sk->sk_stamp.tv_sec     = -1L;
1217         sk->sk_stamp.tv_usec    = -1L;
1218
1219         atomic_set(&sk->sk_refcnt, 1);
1220 }
1221
1222 void fastcall lock_sock(struct sock *sk)
1223 {
1224         might_sleep();
1225         spin_lock_bh(&(sk->sk_lock.slock));
1226         if (sk->sk_lock.owner)
1227                 __lock_sock(sk);
1228         sk->sk_lock.owner = (void *)1;
1229         spin_unlock_bh(&(sk->sk_lock.slock));
1230 }
1231
1232 EXPORT_SYMBOL(lock_sock);
1233
1234 void fastcall release_sock(struct sock *sk)
1235 {
1236         spin_lock_bh(&(sk->sk_lock.slock));
1237         if (sk->sk_backlog.tail)
1238                 __release_sock(sk);
1239         sk->sk_lock.owner = NULL;
1240         if (waitqueue_active(&(sk->sk_lock.wq)))
1241                 wake_up(&(sk->sk_lock.wq));
1242         spin_unlock_bh(&(sk->sk_lock.slock));
1243 }
1244 EXPORT_SYMBOL(release_sock);
1245
1246 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1247 {
1248         if (!sock_flag(sk, SOCK_TIMESTAMP))
1249                 sock_enable_timestamp(sk);
1250         if (sk->sk_stamp.tv_sec == -1)
1251                 return -ENOENT;
1252         if (sk->sk_stamp.tv_sec == 0)
1253                 do_gettimeofday(&sk->sk_stamp);
1254         return copy_to_user(userstamp, &sk->sk_stamp, sizeof(struct timeval)) ?
1255                 -EFAULT : 0;
1256 }
1257 EXPORT_SYMBOL(sock_get_timestamp);
1258
1259 void sock_enable_timestamp(struct sock *sk)
1260 {
1261         if (!sock_flag(sk, SOCK_TIMESTAMP)) {
1262                 sock_set_flag(sk, SOCK_TIMESTAMP);
1263                 net_enable_timestamp();
1264         }
1265 }
1266 EXPORT_SYMBOL(sock_enable_timestamp);
1267
1268 /*
1269  *      Get a socket option on an socket.
1270  *
1271  *      FIX: POSIX 1003.1g is very ambiguous here. It states that
1272  *      asynchronous errors should be reported by getsockopt. We assume
1273  *      this means if you specify SO_ERROR (otherwise whats the point of it).
1274  */
1275 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1276                            char __user *optval, int __user *optlen)
1277 {
1278         struct sock *sk = sock->sk;
1279
1280         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1281 }
1282
1283 EXPORT_SYMBOL(sock_common_getsockopt);
1284
1285 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1286                         struct msghdr *msg, size_t size, int flags)
1287 {
1288         struct sock *sk = sock->sk;
1289         int addr_len = 0;
1290         int err;
1291
1292         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
1293                                    flags & ~MSG_DONTWAIT, &addr_len);
1294         if (err >= 0)
1295                 msg->msg_namelen = addr_len;
1296         return err;
1297 }
1298
1299 EXPORT_SYMBOL(sock_common_recvmsg);
1300
1301 /*
1302  *      Set socket options on an inet socket.
1303  */
1304 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1305                            char __user *optval, int optlen)
1306 {
1307         struct sock *sk = sock->sk;
1308
1309         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1310 }
1311
1312 EXPORT_SYMBOL(sock_common_setsockopt);
1313
1314 void sk_common_release(struct sock *sk)
1315 {
1316         if (sk->sk_prot->destroy)
1317                 sk->sk_prot->destroy(sk);
1318
1319         /*
1320          * Observation: when sock_common_release is called, processes have
1321          * no access to socket. But net still has.
1322          * Step one, detach it from networking:
1323          *
1324          * A. Remove from hash tables.
1325          */
1326
1327         sk->sk_prot->unhash(sk);
1328
1329         /*
1330          * In this point socket cannot receive new packets, but it is possible
1331          * that some packets are in flight because some CPU runs receiver and
1332          * did hash table lookup before we unhashed socket. They will achieve
1333          * receive queue and will be purged by socket destructor.
1334          *
1335          * Also we still have packets pending on receive queue and probably,
1336          * our own packets waiting in device queues. sock_destroy will drain
1337          * receive queue, but transmitted packets will delay socket destruction
1338          * until the last reference will be released.
1339          */
1340
1341         sock_orphan(sk);
1342
1343         xfrm_sk_free_policy(sk);
1344
1345 #ifdef INET_REFCNT_DEBUG
1346         if (atomic_read(&sk->sk_refcnt) != 1)
1347                 printk(KERN_DEBUG "Destruction of the socket %p delayed, c=%d\n",
1348                        sk, atomic_read(&sk->sk_refcnt));
1349 #endif
1350         sock_put(sk);
1351 }
1352
1353 EXPORT_SYMBOL(sk_common_release);
1354
1355 static DEFINE_RWLOCK(proto_list_lock);
1356 static LIST_HEAD(proto_list);
1357
1358 int proto_register(struct proto *prot, int alloc_slab)
1359 {
1360         int rc = -ENOBUFS;
1361
1362         if (alloc_slab) {
1363                 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
1364                                                SLAB_HWCACHE_ALIGN, NULL, NULL);
1365
1366                 if (prot->slab == NULL) {
1367                         printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
1368                                prot->name);
1369                         goto out;
1370                 }
1371         }
1372
1373         write_lock(&proto_list_lock);
1374         list_add(&prot->node, &proto_list);
1375         write_unlock(&proto_list_lock);
1376         rc = 0;
1377 out:
1378         return rc;
1379 }
1380
1381 EXPORT_SYMBOL(proto_register);
1382
1383 void proto_unregister(struct proto *prot)
1384 {
1385         write_lock(&proto_list_lock);
1386
1387         if (prot->slab != NULL) {
1388                 kmem_cache_destroy(prot->slab);
1389                 prot->slab = NULL;
1390         }
1391
1392         list_del(&prot->node);
1393         write_unlock(&proto_list_lock);
1394 }
1395
1396 EXPORT_SYMBOL(proto_unregister);
1397
1398 #ifdef CONFIG_PROC_FS
1399 static inline struct proto *__proto_head(void)
1400 {
1401         return list_entry(proto_list.next, struct proto, node);
1402 }
1403
1404 static inline struct proto *proto_head(void)
1405 {
1406         return list_empty(&proto_list) ? NULL : __proto_head();
1407 }
1408
1409 static inline struct proto *proto_next(struct proto *proto)
1410 {
1411         return proto->node.next == &proto_list ? NULL :
1412                 list_entry(proto->node.next, struct proto, node);
1413 }
1414
1415 static inline struct proto *proto_get_idx(loff_t pos)
1416 {
1417         struct proto *proto;
1418         loff_t i = 0;
1419
1420         list_for_each_entry(proto, &proto_list, node)
1421                 if (i++ == pos)
1422                         goto out;
1423
1424         proto = NULL;
1425 out:
1426         return proto;
1427 }
1428
1429 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
1430 {
1431         read_lock(&proto_list_lock);
1432         return *pos ? proto_get_idx(*pos - 1) : SEQ_START_TOKEN;
1433 }
1434
1435 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1436 {
1437         ++*pos;
1438         return v == SEQ_START_TOKEN ? proto_head() : proto_next(v);
1439 }
1440
1441 static void proto_seq_stop(struct seq_file *seq, void *v)
1442 {
1443         read_unlock(&proto_list_lock);
1444 }
1445
1446 static char proto_method_implemented(const void *method)
1447 {
1448         return method == NULL ? 'n' : 'y';
1449 }
1450
1451 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
1452 {
1453         seq_printf(seq, "%-9s %4u %6d  %6d   %-3s %6u   %-3s  %-10s "
1454                         "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
1455                    proto->name,
1456                    proto->obj_size,
1457                    proto->sockets_allocated != NULL ? atomic_read(proto->sockets_allocated) : -1,
1458                    proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
1459                    proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
1460                    proto->max_header,
1461                    proto->slab == NULL ? "no" : "yes",
1462                    module_name(proto->owner),
1463                    proto_method_implemented(proto->close),
1464                    proto_method_implemented(proto->connect),
1465                    proto_method_implemented(proto->disconnect),
1466                    proto_method_implemented(proto->accept),
1467                    proto_method_implemented(proto->ioctl),
1468                    proto_method_implemented(proto->init),
1469                    proto_method_implemented(proto->destroy),
1470                    proto_method_implemented(proto->shutdown),
1471                    proto_method_implemented(proto->setsockopt),
1472                    proto_method_implemented(proto->getsockopt),
1473                    proto_method_implemented(proto->sendmsg),
1474                    proto_method_implemented(proto->recvmsg),
1475                    proto_method_implemented(proto->sendpage),
1476                    proto_method_implemented(proto->bind),
1477                    proto_method_implemented(proto->backlog_rcv),
1478                    proto_method_implemented(proto->hash),
1479                    proto_method_implemented(proto->unhash),
1480                    proto_method_implemented(proto->get_port),
1481                    proto_method_implemented(proto->enter_memory_pressure));
1482 }
1483
1484 static int proto_seq_show(struct seq_file *seq, void *v)
1485 {
1486         if (v == SEQ_START_TOKEN)
1487                 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
1488                            "protocol",
1489                            "size",
1490                            "sockets",
1491                            "memory",
1492                            "press",
1493                            "maxhdr",
1494                            "slab",
1495                            "module",
1496                            "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
1497         else
1498                 proto_seq_printf(seq, v);
1499         return 0;
1500 }
1501
1502 static struct seq_operations proto_seq_ops = {
1503         .start  = proto_seq_start,
1504         .next   = proto_seq_next,
1505         .stop   = proto_seq_stop,
1506         .show   = proto_seq_show,
1507 };
1508
1509 static int proto_seq_open(struct inode *inode, struct file *file)
1510 {
1511         return seq_open(file, &proto_seq_ops);
1512 }
1513
1514 static struct file_operations proto_seq_fops = {
1515         .owner          = THIS_MODULE,
1516         .open           = proto_seq_open,
1517         .read           = seq_read,
1518         .llseek         = seq_lseek,
1519         .release        = seq_release,
1520 };
1521
1522 static int __init proto_init(void)
1523 {
1524         /* register /proc/net/protocols */
1525         return proc_net_fops_create("protocols", S_IRUGO, &proto_seq_fops) == NULL ? -ENOBUFS : 0;
1526 }
1527
1528 subsys_initcall(proto_init);
1529
1530 #endif /* PROC_FS */
1531
1532 EXPORT_SYMBOL(sk_alloc);
1533 EXPORT_SYMBOL(sk_free);
1534 EXPORT_SYMBOL(sk_send_sigurg);
1535 EXPORT_SYMBOL(sock_alloc_send_skb);
1536 EXPORT_SYMBOL(sock_init_data);
1537 EXPORT_SYMBOL(sock_kfree_s);
1538 EXPORT_SYMBOL(sock_kmalloc);
1539 EXPORT_SYMBOL(sock_no_accept);
1540 EXPORT_SYMBOL(sock_no_bind);
1541 EXPORT_SYMBOL(sock_no_connect);
1542 EXPORT_SYMBOL(sock_no_getname);
1543 EXPORT_SYMBOL(sock_no_getsockopt);
1544 EXPORT_SYMBOL(sock_no_ioctl);
1545 EXPORT_SYMBOL(sock_no_listen);
1546 EXPORT_SYMBOL(sock_no_mmap);
1547 EXPORT_SYMBOL(sock_no_poll);
1548 EXPORT_SYMBOL(sock_no_recvmsg);
1549 EXPORT_SYMBOL(sock_no_sendmsg);
1550 EXPORT_SYMBOL(sock_no_sendpage);
1551 EXPORT_SYMBOL(sock_no_setsockopt);
1552 EXPORT_SYMBOL(sock_no_shutdown);
1553 EXPORT_SYMBOL(sock_no_socketpair);
1554 EXPORT_SYMBOL(sock_rfree);
1555 EXPORT_SYMBOL(sock_setsockopt);
1556 EXPORT_SYMBOL(sock_wfree);
1557 EXPORT_SYMBOL(sock_wmalloc);
1558 EXPORT_SYMBOL(sock_i_uid);
1559 EXPORT_SYMBOL(sock_i_ino);
1560 #ifdef CONFIG_SYSCTL
1561 EXPORT_SYMBOL(sysctl_optmem_max);
1562 EXPORT_SYMBOL(sysctl_rmem_max);
1563 EXPORT_SYMBOL(sysctl_wmem_max);
1564 #endif