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