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/sched.h>
 103 #include <linux/timer.h>
 104 #include <linux/string.h>
 105 #include <linux/sockios.h>
 106 #include <linux/net.h>
 107 #include <linux/mm.h>
 108 #include <linux/slab.h>
 109 #include <linux/interrupt.h>
 110 #include <linux/poll.h>
 111 #include <linux/tcp.h>
 112 #include <linux/init.h>
 113
 114 #include <asm/uaccess.h>
 115 #include <asm/system.h>
 116
 117 #include <linux/netdevice.h>
 118 #include <net/protocol.h>
 119 #include <linux/skbuff.h>
 120 #include <net/sock.h>
 121 #include <linux/ipsec.h>
 122
 123 #include <linux/filter.h>
 124
 125 #ifdef CONFIG_INET
 126 #include <net/tcp.h>
 127 #endif
 128
 129 /* Take into consideration the size of the struct sk_buff overhead in the
 130  * determination of these values, since that is non-constant across
 131  * platforms.  This makes socket queueing behavior and performance
 132  * not depend upon such differences.
 133  */
 134 #define _SK_MEM_PACKETS         256
 135 #define _SK_MEM_OVERHEAD        (sizeof(struct sk_buff) + 256)
 136 #define SK_WMEM_MAX             (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
 137 #define SK_RMEM_MAX             (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
 138
 139 /* Run time adjustable parameters. */
 140 __u32 sysctl_wmem_max = SK_WMEM_MAX;
 141 __u32 sysctl_rmem_max = SK_RMEM_MAX;
 142 __u32 sysctl_wmem_default = SK_WMEM_MAX;
 143 __u32 sysctl_rmem_default = SK_RMEM_MAX;
 144
 145 /* Maximal space eaten by iovec or ancilliary data plus some space */
 146 int sysctl_optmem_max = sizeof(unsigned long)*(2*UIO_MAXIOV + 512);
 147
 148 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
 149 {
 150         struct timeval tv;
 151
 152         if (optlen < sizeof(tv))
 153                 return -EINVAL;
 154         if (copy_from_user(&tv, optval, sizeof(tv)))
 155                 return -EFAULT;
 156
 157         *timeo_p = MAX_SCHEDULE_TIMEOUT;
 158         if (tv.tv_sec == 0 && tv.tv_usec == 0)
 159                 return 0;
 160         if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
 161                 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
 162         return 0;
 163 }
 164
 165 static void sock_warn_obsolete_bsdism(const char *name)
 166 {
 167         static int warned;
 168         static char warncomm[16];
 169         if (strcmp(warncomm, current->comm) && warned < 5) {
 170                 strcpy(warncomm,  current->comm);
 171                 printk(KERN_WARNING "process `%s' is using obsolete "
 172                        "%s SO_BSDCOMPAT\n", warncomm, name);
 173                 warned++;
 174         }
 175 }
 176
 177 /*
 178  *      This is meant for all protocols to use and covers goings on
 179  *      at the socket level. Everything here is generic.
 180  */
 181
 182 int sock_setsockopt(struct socket *sock, int level, int optname,
 183                     char __user *optval, int optlen)
 184 {
 185         struct sock *sk=sock->sk;
 186         struct sk_filter *filter;
 187         int val;
 188         int valbool;
 189         struct linger ling;
 190         int ret = 0;
 191
 192         /*
 193          *      Options without arguments
 194          */
 195
 196 #ifdef SO_DONTLINGER            /* Compatibility item... */
 197         switch (optname) {
 198                 case SO_DONTLINGER:
 199                         sock_reset_flag(sk, SOCK_LINGER);
 200                         return 0;
 201         }
 202 #endif
 203
 204         if(optlen<sizeof(int))
 205                 return(-EINVAL);
 206
 207         if (get_user(val, (int __user *)optval))
 208                 return -EFAULT;
 209
 210         valbool = val?1:0;
 211
 212         lock_sock(sk);
 213
 214         switch(optname)
 215         {
 216                 case SO_DEBUG:
 217                         if(val && !capable(CAP_NET_ADMIN))
 218                         {
 219                                 ret = -EACCES;
 220                         }
 221                         else
 222                                 sk->sk_debug = valbool;
 223                         break;
 224                 case SO_REUSEADDR:
 225                         sk->sk_reuse = valbool;
 226                         break;
 227                 case SO_TYPE:
 228                 case SO_ERROR:
 229                         ret = -ENOPROTOOPT;
 230                         break;
 231                 case SO_DONTROUTE:
 232                         sk->sk_localroute = valbool;
 233                         break;
 234                 case SO_BROADCAST:
 235                         sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
 236                         break;
 237                 case SO_SNDBUF:
 238                         /* Don't error on this BSD doesn't and if you think
 239                            about it this is right. Otherwise apps have to
 240                            play 'guess the biggest size' games. RCVBUF/SNDBUF
 241                            are treated in BSD as hints */
 242
 243                         if (val > sysctl_wmem_max)
 244                                 val = sysctl_wmem_max;
 245
 246                         sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
 247                         if ((val * 2) < SOCK_MIN_SNDBUF)
 248                                 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
 249                         else
 250                                 sk->sk_sndbuf = val * 2;
 251
 252                         /*
 253                          *      Wake up sending tasks if we
 254                          *      upped the value.
 255                          */
 256                         sk->sk_write_space(sk);
 257                         break;
 258
 259                 case SO_RCVBUF:
 260                         /* Don't error on this BSD doesn't and if you think
 261                            about it this is right. Otherwise apps have to
 262                            play 'guess the biggest size' games. RCVBUF/SNDBUF
 263                            are treated in BSD as hints */
 264
 265                         if (val > sysctl_rmem_max)
 266                                 val = sysctl_rmem_max;
 267
 268                         sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
 269                         /* FIXME: is this lower bound the right one? */
 270                         if ((val * 2) < SOCK_MIN_RCVBUF)
 271                                 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
 272                         else
 273                                 sk->sk_rcvbuf = val * 2;
 274                         break;
 275
 276                 case SO_KEEPALIVE:
 277 #ifdef CONFIG_INET
 278                         if (sk->sk_protocol == IPPROTO_TCP)
 279                                 tcp_set_keepalive(sk, valbool);
 280 #endif
 281                         sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
 282                         break;
 283
 284                 case SO_OOBINLINE:
 285                         sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
 286                         break;
 287
 288                 case SO_NO_CHECK:
 289                         sk->sk_no_check = valbool;
 290                         break;
 291
 292                 case SO_PRIORITY:
 293                         if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
 294                                 sk->sk_priority = val;
 295                         else
 296                                 ret = -EPERM;
 297                         break;
 298
 299                 case SO_LINGER:
 300                         if(optlen<sizeof(ling)) {
 301                                 ret = -EINVAL;  /* 1003.1g */
 302                                 break;
 303                         }
 304                         if (copy_from_user(&ling,optval,sizeof(ling))) {
 305                                 ret = -EFAULT;
 306                                 break;
 307                         }
 308                         if (!ling.l_onoff)
 309                                 sock_reset_flag(sk, SOCK_LINGER);
 310                         else {
 311 #if (BITS_PER_LONG == 32)
 312                                 if (ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
 313                                         sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
 314                                 else
 315 #endif
 316                                         sk->sk_lingertime = ling.l_linger * HZ;
 317                                 sock_set_flag(sk, SOCK_LINGER);
 318                         }
 319                         break;
 320
 321                 case SO_BSDCOMPAT:
 322                         sock_warn_obsolete_bsdism("setsockopt");
 323                         break;
 324
 325                 case SO_PASSCRED:
 326                         if (valbool)
 327                                 set_bit(SOCK_PASS_CRED, &sock->flags);
 328                         else
 329                                 clear_bit(SOCK_PASS_CRED, &sock->flags);
 330                         break;
 331
 332                 case SO_TIMESTAMP:
 333                         sk->sk_rcvtstamp = valbool;
 334                         if (valbool)
 335                                 sock_enable_timestamp(sk);
 336                         break;
 337
 338                 case SO_RCVLOWAT:
 339                         if (val < 0)
 340                                 val = INT_MAX;
 341                         sk->sk_rcvlowat = val ? : 1;
 342                         break;
 343
 344                 case SO_RCVTIMEO:
 345                         ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
 346                         break;
 347
 348                 case SO_SNDTIMEO:
 349                         ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
 350                         break;
 351
 352 #ifdef CONFIG_NETDEVICES
 353                 case SO_BINDTODEVICE:
 354                 {
 355                         char devname[IFNAMSIZ];
 356
 357                         /* Sorry... */
 358                         if (!capable(CAP_NET_RAW)) {
 359                                 ret = -EPERM;
 360                                 break;
 361                         }
 362
 363                         /* Bind this socket to a particular device like "eth0",
 364                          * as specified in the passed interface name. If the
 365                          * name is "" or the option length is zero the socket
 366                          * is not bound.
 367                          */
 368
 369                         if (!valbool) {
 370                                 sk->sk_bound_dev_if = 0;
 371                         } else {
 372                                 if (optlen > IFNAMSIZ)
 373                                         optlen = IFNAMSIZ;
 374                                 if (copy_from_user(devname, optval, optlen)) {
 375                                         ret = -EFAULT;
 376                                         break;
 377                                 }
 378
 379                                 /* Remove any cached route for this socket. */
 380                                 sk_dst_reset(sk);
 381
 382                                 if (devname[0] == '\0') {
 383                                         sk->sk_bound_dev_if = 0;
 384                                 } else {
 385                                         struct net_device *dev = dev_get_by_name(devname);
 386                                         if (!dev) {
 387                                                 ret = -ENODEV;
 388                                                 break;
 389                                         }
 390                                         sk->sk_bound_dev_if = dev->ifindex;
 391                                         dev_put(dev);
 392                                 }
 393                         }
 394                         break;
 395                 }
 396 #endif
 397
 398
 399                 case SO_ATTACH_FILTER:
 400                         ret = -EINVAL;
 401                         if (optlen == sizeof(struct sock_fprog)) {
 402                                 struct sock_fprog fprog;
 403
 404                                 ret = -EFAULT;
 405                                 if (copy_from_user(&fprog, optval, sizeof(fprog)))
 406                                         break;
 407
 408                                 ret = sk_attach_filter(&fprog, sk);
 409                         }
 410                         break;
 411
 412                 case SO_DETACH_FILTER:
 413                         spin_lock_bh(&sk->sk_lock.slock);
 414                         filter = sk->sk_filter;
 415                         if (filter) {
 416                                 sk->sk_filter = NULL;
 417                                 spin_unlock_bh(&sk->sk_lock.slock);
 418                                 sk_filter_release(sk, filter);
 419                                 break;
 420                         }
 421                         spin_unlock_bh(&sk->sk_lock.slock);
 422                         ret = -ENONET;
 423                         break;
 424
 425                 /* We implement the SO_SNDLOWAT etc to
 426                    not be settable (1003.1g 5.3) */
 427                 default:
 428                         ret = -ENOPROTOOPT;
 429                         break;
 430         }
 431         release_sock(sk);
 432         return ret;
 433 }
 434
 435
 436 int sock_getsockopt(struct socket *sock, int level, int optname,
 437                     char __user *optval, int __user *optlen)
 438 {
 439         struct sock *sk = sock->sk;
 440
 441         union
 442         {
 443                 int val;
 444                 struct linger ling;
 445                 struct timeval tm;
 446         } v;
 447
 448         unsigned int lv=sizeof(int),len;
 449
 450         if(get_user(len,optlen))
 451                 return -EFAULT;
 452         if(len < 0)
 453                 return -EINVAL;
 454
 455         switch(optname)
 456         {
 457                 case SO_DEBUG:
 458                         v.val = sk->sk_debug;
 459                         break;
 460
 461                 case SO_DONTROUTE:
 462                         v.val = sk->sk_localroute;
 463                         break;
 464
 465                 case SO_BROADCAST:
 466                         v.val = !!sock_flag(sk, SOCK_BROADCAST);
 467                         break;
 468
 469                 case SO_SNDBUF:
 470                         v.val = sk->sk_sndbuf;
 471                         break;
 472
 473                 case SO_RCVBUF:
 474                         v.val = sk->sk_rcvbuf;
 475                         break;
 476
 477                 case SO_REUSEADDR:
 478                         v.val = sk->sk_reuse;
 479                         break;
 480
 481                 case SO_KEEPALIVE:
 482                         v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
 483                         break;
 484
 485                 case SO_TYPE:
 486                         v.val = sk->sk_type;
 487                         break;
 488
 489                 case SO_ERROR:
 490                         v.val = -sock_error(sk);
 491                         if(v.val==0)
 492                                 v.val = xchg(&sk->sk_err_soft, 0);
 493                         break;
 494
 495                 case SO_OOBINLINE:
 496                         v.val = !!sock_flag(sk, SOCK_URGINLINE);
 497                         break;
 498
 499                 case SO_NO_CHECK:
 500                         v.val = sk->sk_no_check;
 501                         break;
 502
 503                 case SO_PRIORITY:
 504                         v.val = sk->sk_priority;
 505                         break;
 506
 507                 case SO_LINGER:
 508                         lv              = sizeof(v.ling);
 509                         v.ling.l_onoff  = !!sock_flag(sk, SOCK_LINGER);
 510                         v.ling.l_linger = sk->sk_lingertime / HZ;
 511                         break;
 512
 513                 case SO_BSDCOMPAT:
 514                         sock_warn_obsolete_bsdism("getsockopt");
 515                         break;
 516
 517                 case SO_TIMESTAMP:
 518                         v.val = sk->sk_rcvtstamp;
 519                         break;
 520
 521                 case SO_RCVTIMEO:
 522                         lv=sizeof(struct timeval);
 523                         if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
 524                                 v.tm.tv_sec = 0;
 525                                 v.tm.tv_usec = 0;
 526                         } else {
 527                                 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
 528                                 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000) / HZ;
 529                         }
 530                         break;
 531
 532                 case SO_SNDTIMEO:
 533                         lv=sizeof(struct timeval);
 534                         if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
 535                                 v.tm.tv_sec = 0;
 536                                 v.tm.tv_usec = 0;
 537                         } else {
 538                                 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
 539                                 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000) / HZ;
 540                         }
 541                         break;
 542
 543                 case SO_RCVLOWAT:
 544                         v.val = sk->sk_rcvlowat;
 545                         break;
 546
 547                 case SO_SNDLOWAT:
 548                         v.val=1;
 549                         break;
 550
 551                 case SO_PASSCRED:
 552                         v.val = test_bit(SOCK_PASS_CRED, &sock->flags)?1:0;
 553                         break;
 554
 555                 case SO_PEERCRED:
 556                         if (len > sizeof(sk->sk_peercred))
 557                                 len = sizeof(sk->sk_peercred);
 558                         if (copy_to_user(optval, &sk->sk_peercred, len))
 559                                 return -EFAULT;
 560                         goto lenout;
 561
 562                 case SO_PEERNAME:
 563                 {
 564                         char address[128];
 565
 566                         if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
 567                                 return -ENOTCONN;
 568                         if (lv < len)
 569                                 return -EINVAL;
 570                         if (copy_to_user(optval, address, len))
 571                                 return -EFAULT;
 572                         goto lenout;
 573                 }
 574
 575                 /* Dubious BSD thing... Probably nobody even uses it, but
 576                  * the UNIX standard wants it for whatever reason... -DaveM
 577                  */
 578                 case SO_ACCEPTCONN:
 579                         v.val = sk->sk_state == TCP_LISTEN;
 580                         break;
 581
 582                 case SO_PEERSEC:
 583                         return security_socket_getpeersec(sock, optval, optlen, len);
 584
 585                 default:
 586                         return(-ENOPROTOOPT);
 587         }
 588         if (len > lv)
 589                 len = lv;
 590         if (copy_to_user(optval, &v, len))
 591                 return -EFAULT;
 592 lenout:
 593         if (put_user(len, optlen))
 594                 return -EFAULT;
 595         return 0;
 596 }
 597
 598 static kmem_cache_t *sk_cachep;
 599
 600 /**
 601  *      sk_alloc - All socket objects are allocated here
 602  *      @family - protocol family
 603  *      @priority - for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
 604  *      @zero_it - zeroes the allocated sock
 605  *      @slab - alternate slab
 606  *
 607  *      All socket objects are allocated here. If @zero_it is non-zero
 608  *      it should have the size of the are to be zeroed, because the
 609  *      private slabcaches have different sizes of the generic struct sock.
 610  *      1 has been kept as a way to say sizeof(struct sock).
 611  */
 612 struct sock *sk_alloc(int family, int priority, int zero_it, kmem_cache_t *slab)
 613 {
 614         struct sock *sk = NULL;
 615
 616         if (!slab)
 617                 slab = sk_cachep;
 618         sk = kmem_cache_alloc(slab, priority);
 619         if (sk) {
 620                 if (zero_it) {
 621                         memset(sk, 0,
 622                                zero_it == 1 ? sizeof(struct sock) : zero_it);
 623                         sk->sk_family = family;
 624                         sock_lock_init(sk);
 625                 }
 626                 sk->sk_slab = slab;
 627                 sock_vx_init(sk);
 628                 sock_nx_init(sk);
 629
 630                 if (security_sk_alloc(sk, family, priority)) {
 631                         kmem_cache_free(slab, sk);
 632                         sk = NULL;
 633                 }
 634         }
 635         return sk;
 636 }
 637
 638 void sk_free(struct sock *sk)
 639 {
 640         struct sk_filter *filter;
 641         struct module *owner = sk->sk_owner;
 642
 643         if (sk->sk_destruct)
 644                 sk->sk_destruct(sk);
 645
 646         filter = sk->sk_filter;
 647         if (filter) {
 648                 sk_filter_release(sk, filter);
 649                 sk->sk_filter = NULL;
 650         }
 651
 652         sock_disable_timestamp(sk);
 653
 654         if (atomic_read(&sk->sk_omem_alloc))
 655                 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
 656                        __FUNCTION__, atomic_read(&sk->sk_omem_alloc));
 657
 658         security_sk_free(sk);
 659         BUG_ON(sk->sk_vx_info);
 660         BUG_ON(sk->sk_nx_info);
 661 /*      clr_vx_info(&sk->sk_vx_info);
 662         clr_nx_info(&sk->sk_nx_info);   */
 663         kmem_cache_free(sk->sk_slab, sk);
 664         module_put(owner);
 665 }
 666
 667 void __init sk_init(void)
 668 {
 669         sk_cachep = kmem_cache_create("sock", sizeof(struct sock), 0,
 670                                       SLAB_HWCACHE_ALIGN, 0, 0);
 671         if (!sk_cachep)
 672                 printk(KERN_CRIT "sk_init: Cannot create sock SLAB cache!");
 673
 674         if (num_physpages <= 4096) {
 675                 sysctl_wmem_max = 32767;
 676                 sysctl_rmem_max = 32767;
 677                 sysctl_wmem_default = 32767;
 678                 sysctl_rmem_default = 32767;
 679         } else if (num_physpages >= 131072) {
 680                 sysctl_wmem_max = 131071;
 681                 sysctl_rmem_max = 131071;
 682         }
 683 }
 684
 685 /*
 686  *      Simple resource managers for sockets.
 687  */
 688
 689
 690 /*
 691  * Write buffer destructor automatically called from kfree_skb.
 692  */
 693 void sock_wfree(struct sk_buff *skb)
 694 {
 695         struct sock *sk = skb->sk;
 696
 697         /* In case it might be waiting for more memory. */
 698         atomic_sub(skb->truesize, &sk->sk_wmem_alloc);
 699         if (!sk->sk_use_write_queue)
 700                 sk->sk_write_space(sk);
 701         sock_put(sk);
 702 }
 703
 704 /*
 705  * Read buffer destructor automatically called from kfree_skb.
 706  */
 707 void sock_rfree(struct sk_buff *skb)
 708 {
 709         struct sock *sk = skb->sk;
 710
 711         atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
 712 }
 713
 714 /*
 715  * Allocate a skb from the socket's send buffer.
 716  */
 717 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, int priority)
 718 {
 719         if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
 720                 struct sk_buff * skb = alloc_skb(size, priority);
 721                 if (skb) {
 722                         skb_set_owner_w(skb, sk);
 723                         return skb;
 724                 }
 725         }
 726         return NULL;
 727 }
 728
 729 /*
 730  * Allocate a skb from the socket's receive buffer.
 731  */
 732 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force, int priority)
 733 {
 734         if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
 735                 struct sk_buff *skb = alloc_skb(size, priority);
 736                 if (skb) {
 737                         skb_set_owner_r(skb, sk);
 738                         return skb;
 739                 }
 740         }
 741         return NULL;
 742 }
 743
 744 /*
 745  * Allocate a memory block from the socket's option memory buffer.
 746  */
 747 void *sock_kmalloc(struct sock *sk, int size, int priority)
 748 {
 749         if ((unsigned)size <= sysctl_optmem_max &&
 750             atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
 751                 void *mem;
 752                 /* First do the add, to avoid the race if kmalloc
 753                  * might sleep.
 754                  */
 755                 atomic_add(size, &sk->sk_omem_alloc);
 756                 mem = kmalloc(size, priority);
 757                 if (mem)
 758                         return mem;
 759                 atomic_sub(size, &sk->sk_omem_alloc);
 760         }
 761         return NULL;
 762 }
 763
 764 /*
 765  * Free an option memory block.
 766  */
 767 void sock_kfree_s(struct sock *sk, void *mem, int size)
 768 {
 769         kfree(mem);
 770         atomic_sub(size, &sk->sk_omem_alloc);
 771 }
 772
 773 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
 774    I think, these locks should be removed for datagram sockets.
 775  */
 776 static long sock_wait_for_wmem(struct sock * sk, long timeo)
 777 {
 778         DEFINE_WAIT(wait);
 779
 780         clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
 781         for (;;) {
 782                 if (!timeo)
 783                         break;
 784                 if (signal_pending(current))
 785                         break;
 786                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
 787                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
 788                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
 789                         break;
 790                 if (sk->sk_shutdown & SEND_SHUTDOWN)
 791                         break;
 792                 if (sk->sk_err)
 793                         break;
 794                 timeo = schedule_timeout(timeo);
 795         }
 796         finish_wait(sk->sk_sleep, &wait);
 797         return timeo;
 798 }
 799
 800
 801 /*
 802  *      Generic send/receive buffer handlers
 803  */
 804
 805 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
 806                                      unsigned long data_len, int noblock, int *errcode)
 807 {
 808         struct sk_buff *skb;
 809         unsigned int gfp_mask;
 810         long timeo;
 811         int err;
 812
 813         gfp_mask = sk->sk_allocation;
 814         if (gfp_mask & __GFP_WAIT)
 815                 gfp_mask |= __GFP_REPEAT;
 816
 817         timeo = sock_sndtimeo(sk, noblock);
 818         while (1) {
 819                 err = sock_error(sk);
 820                 if (err != 0)
 821                         goto failure;
 822
 823                 err = -EPIPE;
 824                 if (sk->sk_shutdown & SEND_SHUTDOWN)
 825                         goto failure;
 826
 827                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
 828                         skb = alloc_skb(header_len, sk->sk_allocation);
 829                         if (skb) {
 830                                 int npages;
 831                                 int i;
 832
 833                                 /* No pages, we're done... */
 834                                 if (!data_len)
 835                                         break;
 836
 837                                 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
 838                                 skb->truesize += data_len;
 839                                 skb_shinfo(skb)->nr_frags = npages;
 840                                 for (i = 0; i < npages; i++) {
 841                                         struct page *page;
 842                                         skb_frag_t *frag;
 843
 844                                         page = alloc_pages(sk->sk_allocation, 0);
 845                                         if (!page) {
 846                                                 err = -ENOBUFS;
 847                                                 skb_shinfo(skb)->nr_frags = i;
 848                                                 kfree_skb(skb);
 849                                                 goto failure;
 850                                         }
 851
 852                                         frag = &skb_shinfo(skb)->frags[i];
 853                                         frag->page = page;
 854                                         frag->page_offset = 0;
 855                                         frag->size = (data_len >= PAGE_SIZE ?
 856                                                       PAGE_SIZE :
 857                                                       data_len);
 858                                         data_len -= PAGE_SIZE;
 859                                 }
 860
 861                                 /* Full success... */
 862                                 break;
 863                         }
 864                         err = -ENOBUFS;
 865                         goto failure;
 866                 }
 867                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
 868                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
 869                 err = -EAGAIN;
 870                 if (!timeo)
 871                         goto failure;
 872                 if (signal_pending(current))
 873                         goto interrupted;
 874                 timeo = sock_wait_for_wmem(sk, timeo);
 875         }
 876
 877         skb_set_owner_w(skb, sk);
 878         return skb;
 879
 880 interrupted:
 881         err = sock_intr_errno(timeo);
 882 failure:
 883         *errcode = err;
 884         return NULL;
 885 }
 886
 887 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
 888                                     int noblock, int *errcode)
 889 {
 890         return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
 891 }
 892
 893 void __lock_sock(struct sock *sk)
 894 {
 895         DEFINE_WAIT(wait);
 896
 897         for(;;) {
 898                 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
 899                                         TASK_UNINTERRUPTIBLE);
 900                 spin_unlock_bh(&sk->sk_lock.slock);
 901                 schedule();
 902                 spin_lock_bh(&sk->sk_lock.slock);
 903                 if(!sock_owned_by_user(sk))
 904                         break;
 905         }
 906         finish_wait(&sk->sk_lock.wq, &wait);
 907 }
 908
 909 void __release_sock(struct sock *sk)
 910 {
 911         struct sk_buff *skb = sk->sk_backlog.head;
 912
 913         do {
 914                 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
 915                 bh_unlock_sock(sk);
 916
 917                 do {
 918                         struct sk_buff *next = skb->next;
 919
 920                         skb->next = NULL;
 921                         sk->sk_backlog_rcv(sk, skb);
 922                         skb = next;
 923                 } while (skb != NULL);
 924
 925                 bh_lock_sock(sk);
 926         } while((skb = sk->sk_backlog.head) != NULL);
 927 }
 928
 929 /*
 930  * Set of default routines for initialising struct proto_ops when
 931  * the protocol does not support a particular function. In certain
 932  * cases where it makes no sense for a protocol to have a "do nothing"
 933  * function, some default processing is provided.
 934  */
 935
 936 int sock_no_release(struct socket *sock)
 937 {
 938         return 0;
 939 }
 940
 941 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
 942 {
 943         return -EOPNOTSUPP;
 944 }
 945
 946 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
 947                     int len, int flags)
 948 {
 949         return -EOPNOTSUPP;
 950 }
 951
 952 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
 953 {
 954         return -EOPNOTSUPP;
 955 }
 956
 957 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
 958 {
 959         return -EOPNOTSUPP;
 960 }
 961
 962 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
 963                     int *len, int peer)
 964 {
 965         return -EOPNOTSUPP;
 966 }
 967
 968 unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt)
 969 {
 970         return 0;
 971 }
 972
 973 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
 974 {
 975         return -EOPNOTSUPP;
 976 }
 977
 978 int sock_no_listen(struct socket *sock, int backlog)
 979 {
 980         return -EOPNOTSUPP;
 981 }
 982
 983 int sock_no_shutdown(struct socket *sock, int how)
 984 {
 985         return -EOPNOTSUPP;
 986 }
 987
 988 int sock_no_setsockopt(struct socket *sock, int level, int optname,
 989                     char __user *optval, int optlen)
 990 {
 991         return -EOPNOTSUPP;
 992 }
 993
 994 int sock_no_getsockopt(struct socket *sock, int level, int optname,
 995                     char __user *optval, int __user *optlen)
 996 {
 997         return -EOPNOTSUPP;
 998 }
 999
1000 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1001                     size_t len)
1002 {
1003         return -EOPNOTSUPP;
1004 }
1005
1006 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1007                     size_t len, int flags)
1008 {
1009         return -EOPNOTSUPP;
1010 }
1011
1012 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1013 {
1014         /* Mirror missing mmap method error code */
1015         return -ENODEV;
1016 }
1017
1018 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1019 {
1020         ssize_t res;
1021         struct msghdr msg;
1022         struct iovec iov;
1023         mm_segment_t old_fs;
1024         char *kaddr;
1025
1026         kaddr = kmap(page);
1027
1028         msg.msg_name = NULL;
1029         msg.msg_namelen = 0;
1030         msg.msg_iov = &iov;
1031         msg.msg_iovlen = 1;
1032         msg.msg_control = NULL;
1033         msg.msg_controllen = 0;
1034         msg.msg_flags = flags;
1035
1036         /* This cast is ok because of the "set_fs(KERNEL_DS)" */
1037         iov.iov_base = (void __user *) (kaddr + offset);
1038         iov.iov_len = size;
1039
1040         old_fs = get_fs();
1041         set_fs(KERNEL_DS);
1042         res = sock_sendmsg(sock, &msg, size);
1043         set_fs(old_fs);
1044
1045         kunmap(page);
1046         return res;
1047 }
1048
1049 /*
1050  *      Default Socket Callbacks
1051  */
1052
1053 void sock_def_wakeup(struct sock *sk)
1054 {
1055         read_lock(&sk->sk_callback_lock);
1056         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1057                 wake_up_interruptible_all(sk->sk_sleep);
1058         read_unlock(&sk->sk_callback_lock);
1059 }
1060
1061 void sock_def_error_report(struct sock *sk)
1062 {
1063         read_lock(&sk->sk_callback_lock);
1064         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1065                 wake_up_interruptible(sk->sk_sleep);
1066         sk_wake_async(sk,0,POLL_ERR);
1067         read_unlock(&sk->sk_callback_lock);
1068 }
1069
1070 void sock_def_readable(struct sock *sk, int len)
1071 {
1072         read_lock(&sk->sk_callback_lock);
1073         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1074                 wake_up_interruptible(sk->sk_sleep);
1075         sk_wake_async(sk,1,POLL_IN);
1076         read_unlock(&sk->sk_callback_lock);
1077 }
1078
1079 void sock_def_write_space(struct sock *sk)
1080 {
1081         read_lock(&sk->sk_callback_lock);
1082
1083         /* Do not wake up a writer until he can make "significant"
1084          * progress.  --DaveM
1085          */
1086         if((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1087                 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1088                         wake_up_interruptible(sk->sk_sleep);
1089
1090                 /* Should agree with poll, otherwise some programs break */
1091                 if (sock_writeable(sk))
1092                         sk_wake_async(sk, 2, POLL_OUT);
1093         }
1094
1095         read_unlock(&sk->sk_callback_lock);
1096 }
1097
1098 void sock_def_destruct(struct sock *sk)
1099 {
1100         if (sk->sk_protinfo)
1101                 kfree(sk->sk_protinfo);
1102 }
1103
1104 void sk_send_sigurg(struct sock *sk)
1105 {
1106         if (sk->sk_socket && sk->sk_socket->file)
1107                 if (send_sigurg(&sk->sk_socket->file->f_owner))
1108                         sk_wake_async(sk, 3, POLL_PRI);
1109 }
1110
1111 void sock_init_data(struct socket *sock, struct sock *sk)
1112 {
1113         skb_queue_head_init(&sk->sk_receive_queue);
1114         skb_queue_head_init(&sk->sk_write_queue);
1115         skb_queue_head_init(&sk->sk_error_queue);
1116
1117         init_timer(&sk->sk_timer);
1118
1119         sk->sk_allocation       =       GFP_KERNEL;
1120         sk->sk_rcvbuf           =       sysctl_rmem_default;
1121         sk->sk_sndbuf           =       sysctl_wmem_default;
1122         sk->sk_state            =       TCP_CLOSE;
1123         sk->sk_zapped           =       1;
1124         sk->sk_socket           =       sock;
1125
1126         if(sock)
1127         {
1128                 sk->sk_type     =       sock->type;
1129                 sk->sk_sleep    =       &sock->wait;
1130                 sock->sk        =       sk;
1131         } else
1132                 sk->sk_sleep    =       NULL;
1133
1134         sk->sk_dst_lock         =       RW_LOCK_UNLOCKED;
1135         sk->sk_callback_lock    =       RW_LOCK_UNLOCKED;
1136
1137         sk->sk_state_change     =       sock_def_wakeup;
1138         sk->sk_data_ready       =       sock_def_readable;
1139         sk->sk_write_space      =       sock_def_write_space;
1140         sk->sk_error_report     =       sock_def_error_report;
1141         sk->sk_destruct         =       sock_def_destruct;
1142
1143         sk->sk_peercred.pid     =       0;
1144         sk->sk_peercred.uid     =       -1;
1145         sk->sk_peercred.gid     =       -1;
1146         sk->sk_rcvlowat         =       1;
1147         sk->sk_rcvtimeo         =       MAX_SCHEDULE_TIMEOUT;
1148         sk->sk_sndtimeo         =       MAX_SCHEDULE_TIMEOUT;
1149         sk->sk_owner            =       NULL;
1150
1151         sk->sk_stamp.tv_sec     = -1L;
1152         sk->sk_stamp.tv_usec    = -1L;
1153
1154         sk->sk_vx_info          =       NULL;
1155         sk->sk_xid              =       0;
1156         sk->sk_nx_info          =       NULL;
1157         sk->sk_nid              =       0;
1158
1159         atomic_set(&sk->sk_refcnt, 1);
1160 }
1161
1162 void fastcall lock_sock(struct sock *sk)
1163 {
1164         might_sleep();
1165         spin_lock_bh(&(sk->sk_lock.slock));
1166         if (sk->sk_lock.owner)
1167                 __lock_sock(sk);
1168         sk->sk_lock.owner = (void *)1;
1169         spin_unlock_bh(&(sk->sk_lock.slock));
1170 }
1171
1172 EXPORT_SYMBOL(lock_sock);
1173
1174 void fastcall release_sock(struct sock *sk)
1175 {
1176         spin_lock_bh(&(sk->sk_lock.slock));
1177         if (sk->sk_backlog.tail)
1178                 __release_sock(sk);
1179         sk->sk_lock.owner = NULL;
1180         if (waitqueue_active(&(sk->sk_lock.wq)))
1181                 wake_up(&(sk->sk_lock.wq));
1182         spin_unlock_bh(&(sk->sk_lock.slock));
1183 }
1184 EXPORT_SYMBOL(release_sock);
1185
1186 /* When > 0 there are consumers of rx skb time stamps */
1187 atomic_t netstamp_needed = ATOMIC_INIT(0);
1188
1189 int sock_get_timestamp(struct sock *sk, struct timeval *userstamp)
1190 {
1191         if (!sock_flag(sk, SOCK_TIMESTAMP))
1192                 sock_enable_timestamp(sk);
1193         if (sk->sk_stamp.tv_sec == -1)
1194                 return -ENOENT;
1195         if (sk->sk_stamp.tv_sec == 0)
1196                 do_gettimeofday(&sk->sk_stamp);
1197         return copy_to_user(userstamp, &sk->sk_stamp, sizeof(struct timeval)) ?
1198                 -EFAULT : 0;
1199 }
1200 EXPORT_SYMBOL(sock_get_timestamp);
1201
1202 void sock_enable_timestamp(struct sock *sk)
1203 {
1204         if (!sock_flag(sk, SOCK_TIMESTAMP)) {
1205                 sock_set_flag(sk, SOCK_TIMESTAMP);
1206                 atomic_inc(&netstamp_needed);
1207         }
1208 }
1209 EXPORT_SYMBOL(sock_enable_timestamp);
1210
1211 void sock_disable_timestamp(struct sock *sk)
1212 {
1213         if (sock_flag(sk, SOCK_TIMESTAMP)) {
1214                 sock_reset_flag(sk, SOCK_TIMESTAMP);
1215                 atomic_dec(&netstamp_needed);
1216         }
1217 }
1218 EXPORT_SYMBOL(sock_disable_timestamp);
1219
1220 EXPORT_SYMBOL(__lock_sock);
1221 EXPORT_SYMBOL(__release_sock);
1222 EXPORT_SYMBOL(sk_alloc);
1223 EXPORT_SYMBOL(sk_free);
1224 EXPORT_SYMBOL(sk_send_sigurg);
1225 EXPORT_SYMBOL(sock_alloc_send_pskb);
1226 EXPORT_SYMBOL(sock_alloc_send_skb);
1227 EXPORT_SYMBOL(sock_getsockopt);
1228 EXPORT_SYMBOL(sock_init_data);
1229 EXPORT_SYMBOL(sock_kfree_s);
1230 EXPORT_SYMBOL(sock_kmalloc);
1231 EXPORT_SYMBOL(sock_no_accept);
1232 EXPORT_SYMBOL(sock_no_bind);
1233 EXPORT_SYMBOL(sock_no_connect);
1234 EXPORT_SYMBOL(sock_no_getname);
1235 EXPORT_SYMBOL(sock_no_getsockopt);
1236 EXPORT_SYMBOL(sock_no_ioctl);
1237 EXPORT_SYMBOL(sock_no_listen);
1238 EXPORT_SYMBOL(sock_no_mmap);
1239 EXPORT_SYMBOL(sock_no_poll);
1240 EXPORT_SYMBOL(sock_no_recvmsg);
1241 EXPORT_SYMBOL(sock_no_release);
1242 EXPORT_SYMBOL(sock_no_sendmsg);
1243 EXPORT_SYMBOL(sock_no_sendpage);
1244 EXPORT_SYMBOL(sock_no_setsockopt);
1245 EXPORT_SYMBOL(sock_no_shutdown);
1246 EXPORT_SYMBOL(sock_no_socketpair);
1247 EXPORT_SYMBOL(sock_rfree);
1248 EXPORT_SYMBOL(sock_rmalloc);
1249 EXPORT_SYMBOL(sock_setsockopt);
1250 EXPORT_SYMBOL(sock_wfree);
1251 EXPORT_SYMBOL(sock_wmalloc);
1252 #ifdef CONFIG_SYSCTL
1253 EXPORT_SYMBOL(sysctl_optmem_max);
1254 EXPORT_SYMBOL(sysctl_rmem_max);
1255 EXPORT_SYMBOL(sysctl_wmem_max);
1256 #endif