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