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
  13  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  14  *              Florian La Roche, <flla@stud.uni-sb.de>
  15  *              Alan Cox, <A.Cox@swansea.ac.uk>
  16  *
  17  * Fixes:
  18  *              Alan Cox        :       Numerous verify_area() problems
  19  *              Alan Cox        :       Connecting on a connecting socket
  20  *                                      now returns an error for tcp.
  21  *              Alan Cox        :       sock->protocol is set correctly.
  22  *                                      and is not sometimes left as 0.
  23  *              Alan Cox        :       connect handles icmp errors on a
  24  *                                      connect properly. Unfortunately there
  25  *                                      is a restart syscall nasty there. I
  26  *                                      can't match BSD without hacking the C
  27  *                                      library. Ideas urgently sought!
  28  *              Alan Cox        :       Disallow bind() to addresses that are
  29  *                                      not ours - especially broadcast ones!!
  30  *              Alan Cox        :       Socket 1024 _IS_ ok for users. (fencepost)
  31  *              Alan Cox        :       sock_wfree/sock_rfree don't destroy sockets,
  32  *                                      instead they leave that for the DESTROY timer.
  33  *              Alan Cox        :       Clean up error flag in accept
  34  *              Alan Cox        :       TCP ack handling is buggy, the DESTROY timer
  35  *                                      was buggy. Put a remove_sock() in the handler
  36  *                                      for memory when we hit 0. Also altered the timer
  37  *                                      code. The ACK stuff can wait and needs major
  38  *                                      TCP layer surgery.
  39  *              Alan Cox        :       Fixed TCP ack bug, removed remove sock
  40  *                                      and fixed timer/inet_bh race.
  41  *              Alan Cox        :       Added zapped flag for TCP
  42  *              Alan Cox        :       Move kfree_skb into skbuff.c and tidied up surplus code
  43  *              Alan Cox        :       for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
  44  *              Alan Cox        :       kfree_s calls now are kfree_skbmem so we can track skb resources
  45  *              Alan Cox        :       Supports socket option broadcast now as does udp. Packet and raw need fixing.
  46  *              Alan Cox        :       Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
  47  *              Rick Sladkey    :       Relaxed UDP rules for matching packets.
  48  *              C.E.Hawkins     :       IFF_PROMISC/SIOCGHWADDR support
  49  *      Pauline Middelink       :       identd support
  50  *              Alan Cox        :       Fixed connect() taking signals I think.
  51  *              Alan Cox        :       SO_LINGER supported
  52  *              Alan Cox        :       Error reporting fixes
  53  *              Anonymous       :       inet_create tidied up (sk->reuse setting)
  54  *              Alan Cox        :       inet sockets don't set sk->type!
  55  *              Alan Cox        :       Split socket option code
  56  *              Alan Cox        :       Callbacks
  57  *              Alan Cox        :       Nagle flag for Charles & Johannes stuff
  58  *              Alex            :       Removed restriction on inet fioctl
  59  *              Alan Cox        :       Splitting INET from NET core
  60  *              Alan Cox        :       Fixed bogus SO_TYPE handling in getsockopt()
  61  *              Adam Caldwell   :       Missing return in SO_DONTROUTE/SO_DEBUG code
  62  *              Alan Cox        :       Split IP from generic code
  63  *              Alan Cox        :       New kfree_skbmem()
  64  *              Alan Cox        :       Make SO_DEBUG superuser only.
  65  *              Alan Cox        :       Allow anyone to clear SO_DEBUG
  66  *                                      (compatibility fix)
  67  *              Alan Cox        :       Added optimistic memory grabbing for AF_UNIX throughput.
  68  *              Alan Cox        :       Allocator for a socket is settable.
  69  *              Alan Cox        :       SO_ERROR includes soft errors.
  70  *              Alan Cox        :       Allow NULL arguments on some SO_ opts
  71  *              Alan Cox        :       Generic socket allocation to make hooks
  72  *                                      easier (suggested by Craig Metz).
  73  *              Michael Pall    :       SO_ERROR returns positive errno again
  74  *              Steve Whitehouse:       Added default destructor to free
  75  *                                      protocol private data.
  76  *              Steve Whitehouse:       Added various other default routines
  77  *                                      common to several socket families.
  78  *              Chris Evans     :       Call suser() check last on F_SETOWN
  79  *              Jay Schulist    :       Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
  80  *              Andi Kleen      :       Add sock_kmalloc()/sock_kfree_s()
  81  *              Andi Kleen      :       Fix write_space callback
  82  *              Chris Evans     :       Security fixes - signedness again
  83  *              Arnaldo C. Melo :       cleanups, use skb_queue_purge
  84  *
  85  * To Fix:
  86  *
  87  *
  88  *              This program is free software; you can redistribute it and/or
  89  *              modify it under the terms of the GNU General Public License
  90  *              as published by the Free Software Foundation; either version
  91  *              2 of the License, or (at your option) any later version.
  92  */
  93
  94 #include <linux/config.h>
  95 #include <linux/errno.h>
  96 #include <linux/types.h>
  97 #include <linux/socket.h>
  98 #include <linux/in.h>
  99 #include <linux/kernel.h>
 100 #include <linux/module.h>
 101 #include <linux/proc_fs.h>
 102 #include <linux/seq_file.h>
 103 #include <linux/sched.h>
 104 #include <linux/timer.h>
 105 #include <linux/string.h>
 106 #include <linux/sockios.h>
 107 #include <linux/net.h>
 108 #include <linux/mm.h>
 109 #include <linux/slab.h>
 110 #include <linux/interrupt.h>
 111 #include <linux/poll.h>
 112 #include <linux/tcp.h>
 113 #include <linux/init.h>
 114
 115 #include <asm/uaccess.h>
 116 #include <asm/system.h>
 117
 118 #include <linux/netdevice.h>
 119 #include <net/protocol.h>
 120 #include <linux/skbuff.h>
 121 #include <net/request_sock.h>
 122 #include <net/sock.h>
 123 #include <net/xfrm.h>
 124 #include <linux/ipsec.h>
 125
 126 #include <linux/filter.h>
 127
 128 #ifdef CONFIG_INET
 129 #include <net/tcp.h>
 130 #endif
 131
 132 /* Take into consideration the size of the struct sk_buff overhead in the
 133  * determination of these values, since that is non-constant across
 134  * platforms.  This makes socket queueing behavior and performance
 135  * not depend upon such differences.
 136  */
 137 #define _SK_MEM_PACKETS         256
 138 #define _SK_MEM_OVERHEAD        (sizeof(struct sk_buff) + 256)
 139 #define SK_WMEM_MAX             (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
 140 #define SK_RMEM_MAX             (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
 141
 142 /* Run time adjustable parameters. */
 143 __u32 sysctl_wmem_max = SK_WMEM_MAX;
 144 __u32 sysctl_rmem_max = SK_RMEM_MAX;
 145 __u32 sysctl_wmem_default = SK_WMEM_MAX;
 146 __u32 sysctl_rmem_default = SK_RMEM_MAX;
 147
 148 /* Maximal space eaten by iovec or ancilliary data plus some space */
 149 int sysctl_optmem_max = sizeof(unsigned long)*(2*UIO_MAXIOV + 512);
 150
 151 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
 152 {
 153         struct timeval tv;
 154
 155         if (optlen < sizeof(tv))
 156                 return -EINVAL;
 157         if (copy_from_user(&tv, optval, sizeof(tv)))
 158                 return -EFAULT;
 159
 160         *timeo_p = MAX_SCHEDULE_TIMEOUT;
 161         if (tv.tv_sec == 0 && tv.tv_usec == 0)
 162                 return 0;
 163         if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
 164                 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
 165         return 0;
 166 }
 167
 168 static void sock_warn_obsolete_bsdism(const char *name)
 169 {
 170         static int warned;
 171         static char warncomm[TASK_COMM_LEN];
 172         if (strcmp(warncomm, current->comm) && warned < 5) {
 173                 strcpy(warncomm,  current->comm);
 174                 printk(KERN_WARNING "process `%s' is using obsolete "
 175                        "%s SO_BSDCOMPAT\n", warncomm, name);
 176                 warned++;
 177         }
 178 }
 179
 180 static void sock_disable_timestamp(struct sock *sk)
 181 {
 182         if (sock_flag(sk, SOCK_TIMESTAMP)) {
 183                 sock_reset_flag(sk, SOCK_TIMESTAMP);
 184                 net_disable_timestamp();
 185         }
 186 }
 187
 188
 189 /*
 190  *      This is meant for all protocols to use and covers goings on
 191  *      at the socket level. Everything here is generic.
 192  */
 193
 194 int sock_setsockopt(struct socket *sock, int level, int optname,
 195                     char __user *optval, int optlen)
 196 {
 197         struct sock *sk=sock->sk;
 198         struct sk_filter *filter;
 199         int val;
 200         int valbool;
 201         struct linger ling;
 202         int ret = 0;
 203
 204         /*
 205          *      Options without arguments
 206          */
 207
 208 #ifdef SO_DONTLINGER            /* Compatibility item... */
 209         if (optname == SO_DONTLINGER) {
 210                 lock_sock(sk);
 211                 sock_reset_flag(sk, SOCK_LINGER);
 212                 release_sock(sk);
 213                 return 0;
 214         }
 215 #endif
 216
 217         if(optlen<sizeof(int))
 218                 return(-EINVAL);
 219
 220         if (get_user(val, (int __user *)optval))
 221                 return -EFAULT;
 222
 223         valbool = val?1:0;
 224
 225         lock_sock(sk);
 226
 227         switch(optname)
 228         {
 229                 case SO_DEBUG:
 230                         if(val && !capable(CAP_NET_ADMIN))
 231                         {
 232                                 ret = -EACCES;
 233                         }
 234                         else if (valbool)
 235                                 sock_set_flag(sk, SOCK_DBG);
 236                         else
 237                                 sock_reset_flag(sk, SOCK_DBG);
 238                         break;
 239                 case SO_REUSEADDR:
 240                         sk->sk_reuse = valbool;
 241                         break;
 242                 case SO_TYPE:
 243                 case SO_ERROR:
 244                         ret = -ENOPROTOOPT;
 245                         break;
 246                 case SO_DONTROUTE:
 247                         if (valbool)
 248                                 sock_set_flag(sk, SOCK_LOCALROUTE);
 249                         else
 250                                 sock_reset_flag(sk, SOCK_LOCALROUTE);
 251                         break;
 252                 case SO_BROADCAST:
 253                         sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
 254                         break;
 255                 case SO_SNDBUF:
 256                         /* Don't error on this BSD doesn't and if you think
 257                            about it this is right. Otherwise apps have to
 258                            play 'guess the biggest size' games. RCVBUF/SNDBUF
 259                            are treated in BSD as hints */
 260
 261                         if (val > sysctl_wmem_max)
 262                                 val = sysctl_wmem_max;
 263 set_sndbuf:
 264                         sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
 265                         if ((val * 2) < SOCK_MIN_SNDBUF)
 266                                 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
 267                         else
 268                                 sk->sk_sndbuf = val * 2;
 269
 270                         /*
 271                          *      Wake up sending tasks if we
 272                          *      upped the value.
 273                          */
 274                         sk->sk_write_space(sk);
 275                         break;
 276
 277                 case SO_SNDBUFFORCE:
 278                         if (!capable(CAP_NET_ADMIN)) {
 279                                 ret = -EPERM;
 280                                 break;
 281                         }
 282                         goto set_sndbuf;
 283
 284                 case SO_RCVBUF:
 285                         /* Don't error on this BSD doesn't and if you think
 286                            about it this is right. Otherwise apps have to
 287                            play 'guess the biggest size' games. RCVBUF/SNDBUF
 288                            are treated in BSD as hints */
 289
 290                         if (val > sysctl_rmem_max)
 291                                 val = sysctl_rmem_max;
 292 set_rcvbuf:
 293                         sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
 294                         /* FIXME: is this lower bound the right one? */
 295                         if ((val * 2) < SOCK_MIN_RCVBUF)
 296                                 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
 297                         else
 298                                 sk->sk_rcvbuf = val * 2;
 299                         break;
 300
 301                 case SO_RCVBUFFORCE:
 302                         if (!capable(CAP_NET_ADMIN)) {
 303                                 ret = -EPERM;
 304                                 break;
 305                         }
 306                         goto set_rcvbuf;
 307
 308                 case SO_KEEPALIVE:
 309 #ifdef CONFIG_INET
 310                         if (sk->sk_protocol == IPPROTO_TCP)
 311                                 tcp_set_keepalive(sk, valbool);
 312 #endif
 313                         sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
 314                         break;
 315
 316                 case SO_OOBINLINE:
 317                         sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
 318                         break;
 319
 320                 case SO_NO_CHECK:
 321                         sk->sk_no_check = valbool;
 322                         break;
 323
 324                 case SO_PRIORITY:
 325                         if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
 326                                 sk->sk_priority = val;
 327                         else
 328                                 ret = -EPERM;
 329                         break;
 330
 331                 case SO_LINGER:
 332                         if(optlen<sizeof(ling)) {
 333                                 ret = -EINVAL;  /* 1003.1g */
 334                                 break;
 335                         }
 336                         if (copy_from_user(&ling,optval,sizeof(ling))) {
 337                                 ret = -EFAULT;
 338                                 break;
 339                         }
 340                         if (!ling.l_onoff)
 341                                 sock_reset_flag(sk, SOCK_LINGER);
 342                         else {
 343 #if (BITS_PER_LONG == 32)
 344                                 if (ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
 345                                         sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
 346                                 else
 347 #endif
 348                                         sk->sk_lingertime = ling.l_linger * HZ;
 349                                 sock_set_flag(sk, SOCK_LINGER);
 350                         }
 351                         break;
 352
 353                 case SO_BSDCOMPAT:
 354                         sock_warn_obsolete_bsdism("setsockopt");
 355                         break;
 356
 357                 case SO_PASSCRED:
 358                         if (valbool)
 359                                 set_bit(SOCK_PASSCRED, &sock->flags);
 360                         else
 361                                 clear_bit(SOCK_PASSCRED, &sock->flags);
 362                         break;
 363
 364                 case SO_TIMESTAMP:
 365                         if (valbool)  {
 366                                 sock_set_flag(sk, SOCK_RCVTSTAMP);
 367                                 sock_enable_timestamp(sk);
 368                         } else
 369                                 sock_reset_flag(sk, SOCK_RCVTSTAMP);
 370                         break;
 371
 372                 case SO_RCVLOWAT:
 373                         if (val < 0)
 374                                 val = INT_MAX;
 375                         sk->sk_rcvlowat = val ? : 1;
 376                         break;
 377
 378                 case SO_RCVTIMEO:
 379                         ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
 380                         break;
 381
 382                 case SO_SNDTIMEO:
 383                         ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
 384                         break;
 385
 386 #ifdef CONFIG_NETDEVICES
 387                 case SO_BINDTODEVICE:
 388                 {
 389                         char devname[IFNAMSIZ];
 390
 391                         /* Sorry... */
 392                         if (!capable(CAP_NET_RAW)) {
 393                                 ret = -EPERM;
 394                                 break;
 395                         }
 396
 397                         /* Bind this socket to a particular device like "eth0",
 398                          * as specified in the passed interface name. If the
 399                          * name is "" or the option length is zero the socket
 400                          * is not bound.
 401                          */
 402
 403                         if (!valbool) {
 404                                 sk->sk_bound_dev_if = 0;
 405                         } else {
 406                                 if (optlen > IFNAMSIZ)
 407                                         optlen = IFNAMSIZ;
 408                                 if (copy_from_user(devname, optval, optlen)) {
 409                                         ret = -EFAULT;
 410                                         break;
 411                                 }
 412
 413                                 /* Remove any cached route for this socket. */
 414                                 sk_dst_reset(sk);
 415
 416                                 if (devname[0] == '\0') {
 417                                         sk->sk_bound_dev_if = 0;
 418                                 } else {
 419                                         struct net_device *dev = dev_get_by_name(devname);
 420                                         if (!dev) {
 421                                                 ret = -ENODEV;
 422                                                 break;
 423                                         }
 424                                         sk->sk_bound_dev_if = dev->ifindex;
 425                                         dev_put(dev);
 426                                 }
 427                         }
 428                         break;
 429                 }
 430 #endif
 431
 432
 433                 case SO_ATTACH_FILTER:
 434                         ret = -EINVAL;
 435                         if (optlen == sizeof(struct sock_fprog)) {
 436                                 struct sock_fprog fprog;
 437
 438                                 ret = -EFAULT;
 439                                 if (copy_from_user(&fprog, optval, sizeof(fprog)))
 440                                         break;
 441
 442                                 ret = sk_attach_filter(&fprog, sk);
 443                         }
 444                         break;
 445
 446                 case SO_DETACH_FILTER:
 447                         spin_lock_bh(&sk->sk_lock.slock);
 448                         filter = sk->sk_filter;
 449                         if (filter) {
 450                                 sk->sk_filter = NULL;
 451                                 spin_unlock_bh(&sk->sk_lock.slock);
 452                                 sk_filter_release(sk, filter);
 453                                 break;
 454                         }
 455                         spin_unlock_bh(&sk->sk_lock.slock);
 456                         ret = -ENONET;
 457                         break;
 458
 459                 /* We implement the SO_SNDLOWAT etc to
 460                    not be settable (1003.1g 5.3) */
 461                 default:
 462                         ret = -ENOPROTOOPT;
 463                         break;
 464         }
 465         release_sock(sk);
 466         return ret;
 467 }
 468
 469
 470 int sock_getsockopt(struct socket *sock, int level, int optname,
 471                     char __user *optval, int __user *optlen)
 472 {
 473         struct sock *sk = sock->sk;
 474
 475         union
 476         {
 477                 int val;
 478                 struct linger ling;
 479                 struct timeval tm;
 480         } v;
 481
 482         unsigned int lv = sizeof(int);
 483         int len;
 484
 485         if(get_user(len,optlen))
 486                 return -EFAULT;
 487         if(len < 0)
 488                 return -EINVAL;
 489
 490         switch(optname)
 491         {
 492                 case SO_DEBUG:
 493                         v.val = sock_flag(sk, SOCK_DBG);
 494                         break;
 495
 496                 case SO_DONTROUTE:
 497                         v.val = sock_flag(sk, SOCK_LOCALROUTE);
 498                         break;
 499
 500                 case SO_BROADCAST:
 501                         v.val = !!sock_flag(sk, SOCK_BROADCAST);
 502                         break;
 503
 504                 case SO_SNDBUF:
 505                         v.val = sk->sk_sndbuf;
 506                         break;
 507
 508                 case SO_RCVBUF:
 509                         v.val = sk->sk_rcvbuf;
 510                         break;
 511
 512                 case SO_REUSEADDR:
 513                         v.val = sk->sk_reuse;
 514                         break;
 515
 516                 case SO_KEEPALIVE:
 517                         v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
 518                         break;
 519
 520                 case SO_TYPE:
 521                         v.val = sk->sk_type;
 522                         break;
 523
 524                 case SO_ERROR:
 525                         v.val = -sock_error(sk);
 526                         if(v.val==0)
 527                                 v.val = xchg(&sk->sk_err_soft, 0);
 528                         break;
 529
 530                 case SO_OOBINLINE:
 531                         v.val = !!sock_flag(sk, SOCK_URGINLINE);
 532                         break;
 533
 534                 case SO_NO_CHECK:
 535                         v.val = sk->sk_no_check;
 536                         break;
 537
 538                 case SO_PRIORITY:
 539                         v.val = sk->sk_priority;
 540                         break;
 541
 542                 case SO_LINGER:
 543                         lv              = sizeof(v.ling);
 544                         v.ling.l_onoff  = !!sock_flag(sk, SOCK_LINGER);
 545                         v.ling.l_linger = sk->sk_lingertime / HZ;
 546                         break;
 547
 548                 case SO_BSDCOMPAT:
 549                         sock_warn_obsolete_bsdism("getsockopt");
 550                         break;
 551
 552                 case SO_TIMESTAMP:
 553                         v.val = sock_flag(sk, SOCK_RCVTSTAMP);
 554                         break;
 555
 556                 case SO_RCVTIMEO:
 557                         lv=sizeof(struct timeval);
 558                         if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
 559                                 v.tm.tv_sec = 0;
 560                                 v.tm.tv_usec = 0;
 561                         } else {
 562                                 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
 563                                 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
 564                         }
 565                         break;
 566
 567                 case SO_SNDTIMEO:
 568                         lv=sizeof(struct timeval);
 569                         if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
 570                                 v.tm.tv_sec = 0;
 571                                 v.tm.tv_usec = 0;
 572                         } else {
 573                                 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
 574                                 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
 575                         }
 576                         break;
 577
 578                 case SO_RCVLOWAT:
 579                         v.val = sk->sk_rcvlowat;
 580                         break;
 581
 582                 case SO_SNDLOWAT:
 583                         v.val=1;
 584                         break;
 585
 586                 case SO_PASSCRED:
 587                         v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
 588                         break;
 589
 590                 case SO_PEERCRED:
 591                         if (len > sizeof(sk->sk_peercred))
 592                                 len = sizeof(sk->sk_peercred);
 593                         if (copy_to_user(optval, &sk->sk_peercred, len))
 594                                 return -EFAULT;
 595                         goto lenout;
 596
 597                 case SO_PEERNAME:
 598                 {
 599                         char address[128];
 600
 601                         if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
 602                                 return -ENOTCONN;
 603                         if (lv < len)
 604                                 return -EINVAL;
 605                         if (copy_to_user(optval, address, len))
 606                                 return -EFAULT;
 607                         goto lenout;
 608                 }
 609
 610                 /* Dubious BSD thing... Probably nobody even uses it, but
 611                  * the UNIX standard wants it for whatever reason... -DaveM
 612                  */
 613                 case SO_ACCEPTCONN:
 614                         v.val = sk->sk_state == TCP_LISTEN;
 615                         break;
 616
 617                 case SO_PEERSEC:
 618                         return security_socket_getpeersec(sock, optval, optlen, len);
 619
 620                 default:
 621                         return(-ENOPROTOOPT);
 622         }
 623         if (len > lv)
 624                 len = lv;
 625         if (copy_to_user(optval, &v, len))
 626                 return -EFAULT;
 627 lenout:
 628         if (put_user(len, optlen))
 629                 return -EFAULT;
 630         return 0;
 631 }
 632
 633 /**
 634  *      sk_alloc - All socket objects are allocated here
 635  *      @family: protocol family
 636  *      @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
 637  *      @prot: struct proto associated with this new sock instance
 638  *      @zero_it: if we should zero the newly allocated sock
 639  */
 640 struct sock *sk_alloc(int family, unsigned int __nocast priority,
 641                       struct proto *prot, int zero_it)
 642 {
 643         struct sock *sk = NULL;
 644         kmem_cache_t *slab = prot->slab;
 645
 646         if (slab != NULL)
 647                 sk = kmem_cache_alloc(slab, priority);
 648         else
 649                 sk = kmalloc(prot->obj_size, priority);
 650
 651         if (sk) {
 652                 if (zero_it) {
 653                         memset(sk, 0, prot->obj_size);
 654                         sk->sk_family = family;
 655                         /*
 656                          * See comment in struct sock definition to understand
 657                          * why we need sk_prot_creator -acme
 658                          */
 659                         sk->sk_prot = sk->sk_prot_creator = prot;
 660                         sock_lock_init(sk);
 661                 }
 662
 663                 if (security_sk_alloc(sk, family, priority)) {
 664                         if (slab != NULL)
 665                                 kmem_cache_free(slab, sk);
 666                         else
 667                                 kfree(sk);
 668                         sk = NULL;
 669                 } else
 670                         __module_get(prot->owner);
 671         }
 672         return sk;
 673 }
 674
 675 void sk_free(struct sock *sk)
 676 {
 677         struct sk_filter *filter;
 678         struct module *owner = sk->sk_prot_creator->owner;
 679
 680         if (sk->sk_destruct)
 681                 sk->sk_destruct(sk);
 682
 683         filter = sk->sk_filter;
 684         if (filter) {
 685                 sk_filter_release(sk, filter);
 686                 sk->sk_filter = NULL;
 687         }
 688
 689         sock_disable_timestamp(sk);
 690
 691         if (atomic_read(&sk->sk_omem_alloc))
 692                 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
 693                        __FUNCTION__, atomic_read(&sk->sk_omem_alloc));
 694
 695         security_sk_free(sk);
 696         if (sk->sk_prot_creator->slab != NULL)
 697                 kmem_cache_free(sk->sk_prot_creator->slab, sk);
 698         else
 699                 kfree(sk);
 700         module_put(owner);
 701 }
 702
 703 void __init sk_init(void)
 704 {
 705         if (num_physpages <= 4096) {
 706                 sysctl_wmem_max = 32767;
 707                 sysctl_rmem_max = 32767;
 708                 sysctl_wmem_default = 32767;
 709                 sysctl_rmem_default = 32767;
 710         } else if (num_physpages >= 131072) {
 711                 sysctl_wmem_max = 131071;
 712                 sysctl_rmem_max = 131071;
 713         }
 714 }
 715
 716 /*
 717  *      Simple resource managers for sockets.
 718  */
 719
 720
 721 /*
 722  * Write buffer destructor automatically called from kfree_skb.
 723  */
 724 void sock_wfree(struct sk_buff *skb)
 725 {
 726         struct sock *sk = skb->sk;
 727
 728         /* In case it might be waiting for more memory. */
 729         atomic_sub(skb->truesize, &sk->sk_wmem_alloc);
 730         if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE))
 731                 sk->sk_write_space(sk);
 732         sock_put(sk);
 733 }
 734
 735 /*
 736  * Read buffer destructor automatically called from kfree_skb.
 737  */
 738 void sock_rfree(struct sk_buff *skb)
 739 {
 740         struct sock *sk = skb->sk;
 741
 742         atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
 743 }
 744
 745
 746 int sock_i_uid(struct sock *sk)
 747 {
 748         int uid;
 749
 750         read_lock(&sk->sk_callback_lock);
 751         uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
 752         read_unlock(&sk->sk_callback_lock);
 753         return uid;
 754 }
 755
 756 unsigned long sock_i_ino(struct sock *sk)
 757 {
 758         unsigned long ino;
 759
 760         read_lock(&sk->sk_callback_lock);
 761         ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
 762         read_unlock(&sk->sk_callback_lock);
 763         return ino;
 764 }
 765
 766 /*
 767  * Allocate a skb from the socket's send buffer.
 768  */
 769 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
 770                              unsigned int __nocast priority)
 771 {
 772         if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
 773                 struct sk_buff * skb = alloc_skb(size, priority);
 774                 if (skb) {
 775                         skb_set_owner_w(skb, sk);
 776                         return skb;
 777                 }
 778         }
 779         return NULL;
 780 }
 781
 782 /*
 783  * Allocate a skb from the socket's receive buffer.
 784  */
 785 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
 786                              unsigned int __nocast priority)
 787 {
 788         if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
 789                 struct sk_buff *skb = alloc_skb(size, priority);
 790                 if (skb) {
 791                         skb_set_owner_r(skb, sk);
 792                         return skb;
 793                 }
 794         }
 795         return NULL;
 796 }
 797
 798 /*
 799  * Allocate a memory block from the socket's option memory buffer.
 800  */
 801 void *sock_kmalloc(struct sock *sk, int size, unsigned int __nocast priority)
 802 {
 803         if ((unsigned)size <= sysctl_optmem_max &&
 804             atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
 805                 void *mem;
 806                 /* First do the add, to avoid the race if kmalloc
 807                  * might sleep.
 808                  */
 809                 atomic_add(size, &sk->sk_omem_alloc);
 810                 mem = kmalloc(size, priority);
 811                 if (mem)
 812                         return mem;
 813                 atomic_sub(size, &sk->sk_omem_alloc);
 814         }
 815         return NULL;
 816 }
 817
 818 /*
 819  * Free an option memory block.
 820  */
 821 void sock_kfree_s(struct sock *sk, void *mem, int size)
 822 {
 823         kfree(mem);
 824         atomic_sub(size, &sk->sk_omem_alloc);
 825 }
 826
 827 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
 828    I think, these locks should be removed for datagram sockets.
 829  */
 830 static long sock_wait_for_wmem(struct sock * sk, long timeo)
 831 {
 832         DEFINE_WAIT(wait);
 833
 834         clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
 835         for (;;) {
 836                 if (!timeo)
 837                         break;
 838                 if (signal_pending(current))
 839                         break;
 840                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
 841                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
 842                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
 843                         break;
 844                 if (sk->sk_shutdown & SEND_SHUTDOWN)
 845                         break;
 846                 if (sk->sk_err)
 847                         break;
 848                 timeo = schedule_timeout(timeo);
 849         }
 850         finish_wait(sk->sk_sleep, &wait);
 851         return timeo;
 852 }
 853
 854
 855 /*
 856  *      Generic send/receive buffer handlers
 857  */
 858
 859 static struct sk_buff *sock_alloc_send_pskb(struct sock *sk,
 860                                             unsigned long header_len,
 861                                             unsigned long data_len,
 862                                             int noblock, int *errcode)
 863 {
 864         struct sk_buff *skb;
 865         unsigned int gfp_mask;
 866         long timeo;
 867         int err;
 868
 869         gfp_mask = sk->sk_allocation;
 870         if (gfp_mask & __GFP_WAIT)
 871                 gfp_mask |= __GFP_REPEAT;
 872
 873         timeo = sock_sndtimeo(sk, noblock);
 874         while (1) {
 875                 err = sock_error(sk);
 876                 if (err != 0)
 877                         goto failure;
 878
 879                 err = -EPIPE;
 880                 if (sk->sk_shutdown & SEND_SHUTDOWN)
 881                         goto failure;
 882
 883                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
 884                         skb = alloc_skb(header_len, sk->sk_allocation);
 885                         if (skb) {
 886                                 int npages;
 887                                 int i;
 888
 889                                 /* No pages, we're done... */
 890                                 if (!data_len)
 891                                         break;
 892
 893                                 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
 894                                 skb->truesize += data_len;
 895                                 skb_shinfo(skb)->nr_frags = npages;
 896                                 for (i = 0; i < npages; i++) {
 897                                         struct page *page;
 898                                         skb_frag_t *frag;
 899
 900                                         page = alloc_pages(sk->sk_allocation, 0);
 901                                         if (!page) {
 902                                                 err = -ENOBUFS;
 903                                                 skb_shinfo(skb)->nr_frags = i;
 904                                                 kfree_skb(skb);
 905                                                 goto failure;
 906                                         }
 907
 908                                         frag = &skb_shinfo(skb)->frags[i];
 909                                         frag->page = page;
 910                                         frag->page_offset = 0;
 911                                         frag->size = (data_len >= PAGE_SIZE ?
 912                                                       PAGE_SIZE :
 913                                                       data_len);
 914                                         data_len -= PAGE_SIZE;
 915                                 }
 916
 917                                 /* Full success... */
 918                                 break;
 919                         }
 920                         err = -ENOBUFS;
 921                         goto failure;
 922                 }
 923                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
 924                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
 925                 err = -EAGAIN;
 926                 if (!timeo)
 927                         goto failure;
 928                 if (signal_pending(current))
 929                         goto interrupted;
 930                 timeo = sock_wait_for_wmem(sk, timeo);
 931         }
 932
 933         skb_set_owner_w(skb, sk);
 934         return skb;
 935
 936 interrupted:
 937         err = sock_intr_errno(timeo);
 938 failure:
 939         *errcode = err;
 940         return NULL;
 941 }
 942
 943 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
 944                                     int noblock, int *errcode)
 945 {
 946         return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
 947 }
 948
 949 static void __lock_sock(struct sock *sk)
 950 {
 951         DEFINE_WAIT(wait);
 952
 953         for(;;) {
 954                 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
 955                                         TASK_UNINTERRUPTIBLE);
 956                 spin_unlock_bh(&sk->sk_lock.slock);
 957                 schedule();
 958                 spin_lock_bh(&sk->sk_lock.slock);
 959                 if(!sock_owned_by_user(sk))
 960                         break;
 961         }
 962         finish_wait(&sk->sk_lock.wq, &wait);
 963 }
 964
 965 static void __release_sock(struct sock *sk)
 966 {
 967         struct sk_buff *skb = sk->sk_backlog.head;
 968
 969         do {
 970                 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
 971                 bh_unlock_sock(sk);
 972
 973                 do {
 974                         struct sk_buff *next = skb->next;
 975
 976                         skb->next = NULL;
 977                         sk->sk_backlog_rcv(sk, skb);
 978
 979                         /*
 980                          * We are in process context here with softirqs
 981                          * disabled, use cond_resched_softirq() to preempt.
 982                          * This is safe to do because we've taken the backlog
 983                          * queue private:
 984                          */
 985                         cond_resched_softirq();
 986
 987                         skb = next;
 988                 } while (skb != NULL);
 989
 990                 bh_lock_sock(sk);
 991         } while((skb = sk->sk_backlog.head) != NULL);
 992 }
 993
 994 /**
 995  * sk_wait_data - wait for data to arrive at sk_receive_queue
 996  * @sk:    sock to wait on
 997  * @timeo: for how long
 998  *
 999  * Now socket state including sk->sk_err is changed only under lock,
1000  * hence we may omit checks after joining wait queue.
1001  * We check receive queue before schedule() only as optimization;
1002  * it is very likely that release_sock() added new data.
1003  */
1004 int sk_wait_data(struct sock *sk, long *timeo)
1005 {
1006         int rc;
1007         DEFINE_WAIT(wait);
1008
1009         prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1010         set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1011         rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1012         clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1013         finish_wait(sk->sk_sleep, &wait);
1014         return rc;
1015 }
1016
1017 EXPORT_SYMBOL(sk_wait_data);
1018
1019 /*
1020  * Set of default routines for initialising struct proto_ops when
1021  * the protocol does not support a particular function. In certain
1022  * cases where it makes no sense for a protocol to have a "do nothing"
1023  * function, some default processing is provided.
1024  */
1025
1026 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1027 {
1028         return -EOPNOTSUPP;
1029 }
1030
1031 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1032                     int len, int flags)
1033 {
1034         return -EOPNOTSUPP;
1035 }
1036
1037 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1038 {
1039         return -EOPNOTSUPP;
1040 }
1041
1042 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1043 {
1044         return -EOPNOTSUPP;
1045 }
1046
1047 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1048                     int *len, int peer)
1049 {
1050         return -EOPNOTSUPP;
1051 }
1052
1053 unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt)
1054 {
1055         return 0;
1056 }
1057
1058 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1059 {
1060         return -EOPNOTSUPP;
1061 }
1062
1063 int sock_no_listen(struct socket *sock, int backlog)
1064 {
1065         return -EOPNOTSUPP;
1066 }
1067
1068 int sock_no_shutdown(struct socket *sock, int how)
1069 {
1070         return -EOPNOTSUPP;
1071 }
1072
1073 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1074                     char __user *optval, int optlen)
1075 {
1076         return -EOPNOTSUPP;
1077 }
1078
1079 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1080                     char __user *optval, int __user *optlen)
1081 {
1082         return -EOPNOTSUPP;
1083 }
1084
1085 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1086                     size_t len)
1087 {
1088         return -EOPNOTSUPP;
1089 }
1090
1091 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1092                     size_t len, int flags)
1093 {
1094         return -EOPNOTSUPP;
1095 }
1096
1097 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1098 {
1099         /* Mirror missing mmap method error code */
1100         return -ENODEV;
1101 }
1102
1103 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1104 {
1105         ssize_t res;
1106         struct msghdr msg = {.msg_flags = flags};
1107         struct kvec iov;
1108         char *kaddr = kmap(page);
1109         iov.iov_base = kaddr + offset;
1110         iov.iov_len = size;
1111         res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1112         kunmap(page);
1113         return res;
1114 }
1115
1116 /*
1117  *      Default Socket Callbacks
1118  */
1119
1120 static void sock_def_wakeup(struct sock *sk)
1121 {
1122         read_lock(&sk->sk_callback_lock);
1123         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1124                 wake_up_interruptible_all(sk->sk_sleep);
1125         read_unlock(&sk->sk_callback_lock);
1126 }
1127
1128 static void sock_def_error_report(struct sock *sk)
1129 {
1130         read_lock(&sk->sk_callback_lock);
1131         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1132                 wake_up_interruptible(sk->sk_sleep);
1133         sk_wake_async(sk,0,POLL_ERR);
1134         read_unlock(&sk->sk_callback_lock);
1135 }
1136
1137 static void sock_def_readable(struct sock *sk, int len)
1138 {
1139         read_lock(&sk->sk_callback_lock);
1140         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1141                 wake_up_interruptible(sk->sk_sleep);
1142         sk_wake_async(sk,1,POLL_IN);
1143         read_unlock(&sk->sk_callback_lock);
1144 }
1145
1146 static void sock_def_write_space(struct sock *sk)
1147 {
1148         read_lock(&sk->sk_callback_lock);
1149
1150         /* Do not wake up a writer until he can make "significant"
1151          * progress.  --DaveM
1152          */
1153         if((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1154                 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1155                         wake_up_interruptible(sk->sk_sleep);
1156
1157                 /* Should agree with poll, otherwise some programs break */
1158                 if (sock_writeable(sk))
1159                         sk_wake_async(sk, 2, POLL_OUT);
1160         }
1161
1162         read_unlock(&sk->sk_callback_lock);
1163 }
1164
1165 static void sock_def_destruct(struct sock *sk)
1166 {
1167         if (sk->sk_protinfo)
1168                 kfree(sk->sk_protinfo);
1169 }
1170
1171 void sk_send_sigurg(struct sock *sk)
1172 {
1173         if (sk->sk_socket && sk->sk_socket->file)
1174                 if (send_sigurg(&sk->sk_socket->file->f_owner))
1175                         sk_wake_async(sk, 3, POLL_PRI);
1176 }
1177
1178 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1179                     unsigned long expires)
1180 {
1181         if (!mod_timer(timer, expires))
1182                 sock_hold(sk);
1183 }
1184
1185 EXPORT_SYMBOL(sk_reset_timer);
1186
1187 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1188 {
1189         if (timer_pending(timer) && del_timer(timer))
1190                 __sock_put(sk);
1191 }
1192
1193 EXPORT_SYMBOL(sk_stop_timer);
1194
1195 void sock_init_data(struct socket *sock, struct sock *sk)
1196 {
1197         skb_queue_head_init(&sk->sk_receive_queue);
1198         skb_queue_head_init(&sk->sk_write_queue);
1199         skb_queue_head_init(&sk->sk_error_queue);
1200
1201         sk->sk_send_head        =       NULL;
1202
1203         init_timer(&sk->sk_timer);
1204
1205         sk->sk_allocation       =       GFP_KERNEL;
1206         sk->sk_rcvbuf           =       sysctl_rmem_default;
1207         sk->sk_sndbuf           =       sysctl_wmem_default;
1208         sk->sk_state            =       TCP_CLOSE;
1209         sk->sk_socket           =       sock;
1210
1211         sock_set_flag(sk, SOCK_ZAPPED);
1212
1213         if(sock)
1214         {
1215                 sk->sk_type     =       sock->type;
1216                 sk->sk_sleep    =       &sock->wait;
1217                 sock->sk        =       sk;
1218         } else
1219                 sk->sk_sleep    =       NULL;
1220
1221         rwlock_init(&sk->sk_dst_lock);
1222         rwlock_init(&sk->sk_callback_lock);
1223
1224         sk->sk_state_change     =       sock_def_wakeup;
1225         sk->sk_data_ready       =       sock_def_readable;
1226         sk->sk_write_space      =       sock_def_write_space;
1227         sk->sk_error_report     =       sock_def_error_report;
1228         sk->sk_destruct         =       sock_def_destruct;
1229
1230         sk->sk_sndmsg_page      =       NULL;
1231         sk->sk_sndmsg_off       =       0;
1232
1233         sk->sk_peercred.pid     =       0;
1234         sk->sk_peercred.uid     =       -1;
1235         sk->sk_peercred.gid     =       -1;
1236         sk->sk_write_pending    =       0;
1237         sk->sk_rcvlowat         =       1;
1238         sk->sk_rcvtimeo         =       MAX_SCHEDULE_TIMEOUT;
1239         sk->sk_sndtimeo         =       MAX_SCHEDULE_TIMEOUT;
1240
1241         sk->sk_stamp.tv_sec     = -1L;
1242         sk->sk_stamp.tv_usec    = -1L;
1243
1244         atomic_set(&sk->sk_refcnt, 1);
1245 }
1246
1247 void fastcall lock_sock(struct sock *sk)
1248 {
1249         might_sleep();
1250         spin_lock_bh(&(sk->sk_lock.slock));
1251         if (sk->sk_lock.owner)
1252                 __lock_sock(sk);
1253         sk->sk_lock.owner = (void *)1;
1254         spin_unlock_bh(&(sk->sk_lock.slock));
1255 }
1256
1257 EXPORT_SYMBOL(lock_sock);
1258
1259 void fastcall release_sock(struct sock *sk)
1260 {
1261         spin_lock_bh(&(sk->sk_lock.slock));
1262         if (sk->sk_backlog.tail)
1263                 __release_sock(sk);
1264         sk->sk_lock.owner = NULL;
1265         if (waitqueue_active(&(sk->sk_lock.wq)))
1266                 wake_up(&(sk->sk_lock.wq));
1267         spin_unlock_bh(&(sk->sk_lock.slock));
1268 }
1269 EXPORT_SYMBOL(release_sock);
1270
1271 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1272 {
1273         if (!sock_flag(sk, SOCK_TIMESTAMP))
1274                 sock_enable_timestamp(sk);
1275         if (sk->sk_stamp.tv_sec == -1)
1276                 return -ENOENT;
1277         if (sk->sk_stamp.tv_sec == 0)
1278                 do_gettimeofday(&sk->sk_stamp);
1279         return copy_to_user(userstamp, &sk->sk_stamp, sizeof(struct timeval)) ?
1280                 -EFAULT : 0;
1281 }
1282 EXPORT_SYMBOL(sock_get_timestamp);
1283
1284 void sock_enable_timestamp(struct sock *sk)
1285 {
1286         if (!sock_flag(sk, SOCK_TIMESTAMP)) {
1287                 sock_set_flag(sk, SOCK_TIMESTAMP);
1288                 net_enable_timestamp();
1289         }
1290 }
1291 EXPORT_SYMBOL(sock_enable_timestamp);
1292
1293 /*
1294  *      Get a socket option on an socket.
1295  *
1296  *      FIX: POSIX 1003.1g is very ambiguous here. It states that
1297  *      asynchronous errors should be reported by getsockopt. We assume
1298  *      this means if you specify SO_ERROR (otherwise whats the point of it).
1299  */
1300 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1301                            char __user *optval, int __user *optlen)
1302 {
1303         struct sock *sk = sock->sk;
1304
1305         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1306 }
1307
1308 EXPORT_SYMBOL(sock_common_getsockopt);
1309
1310 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1311                         struct msghdr *msg, size_t size, int flags)
1312 {
1313         struct sock *sk = sock->sk;
1314         int addr_len = 0;
1315         int err;
1316
1317         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
1318                                    flags & ~MSG_DONTWAIT, &addr_len);
1319         if (err >= 0)
1320                 msg->msg_namelen = addr_len;
1321         return err;
1322 }
1323
1324 EXPORT_SYMBOL(sock_common_recvmsg);
1325
1326 /*
1327  *      Set socket options on an inet socket.
1328  */
1329 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1330                            char __user *optval, int optlen)
1331 {
1332         struct sock *sk = sock->sk;
1333
1334         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1335 }
1336
1337 EXPORT_SYMBOL(sock_common_setsockopt);
1338
1339 void sk_common_release(struct sock *sk)
1340 {
1341         if (sk->sk_prot->destroy)
1342                 sk->sk_prot->destroy(sk);
1343
1344         /*
1345          * Observation: when sock_common_release is called, processes have
1346          * no access to socket. But net still has.
1347          * Step one, detach it from networking:
1348          *
1349          * A. Remove from hash tables.
1350          */
1351
1352         sk->sk_prot->unhash(sk);
1353
1354         /*
1355          * In this point socket cannot receive new packets, but it is possible
1356          * that some packets are in flight because some CPU runs receiver and
1357          * did hash table lookup before we unhashed socket. They will achieve
1358          * receive queue and will be purged by socket destructor.
1359          *
1360          * Also we still have packets pending on receive queue and probably,
1361          * our own packets waiting in device queues. sock_destroy will drain
1362          * receive queue, but transmitted packets will delay socket destruction
1363          * until the last reference will be released.
1364          */
1365
1366         sock_orphan(sk);
1367
1368         xfrm_sk_free_policy(sk);
1369
1370 #ifdef INET_REFCNT_DEBUG
1371         if (atomic_read(&sk->sk_refcnt) != 1)
1372                 printk(KERN_DEBUG "Destruction of the socket %p delayed, c=%d\n",
1373                        sk, atomic_read(&sk->sk_refcnt));
1374 #endif
1375         sock_put(sk);
1376 }
1377
1378 EXPORT_SYMBOL(sk_common_release);
1379
1380 static DEFINE_RWLOCK(proto_list_lock);
1381 static LIST_HEAD(proto_list);
1382
1383 int proto_register(struct proto *prot, int alloc_slab)
1384 {
1385         char *request_sock_slab_name;
1386         int rc = -ENOBUFS;
1387
1388         if (alloc_slab) {
1389                 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
1390                                                SLAB_HWCACHE_ALIGN, NULL, NULL);
1391
1392                 if (prot->slab == NULL) {
1393                         printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
1394                                prot->name);
1395                         goto out;
1396                 }
1397
1398                 if (prot->rsk_prot != NULL) {
1399                         static const char mask[] = "request_sock_%s";
1400
1401                         request_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
1402                         if (request_sock_slab_name == NULL)
1403                                 goto out_free_sock_slab;
1404
1405                         sprintf(request_sock_slab_name, mask, prot->name);
1406                         prot->rsk_prot->slab = kmem_cache_create(request_sock_slab_name,
1407                                                                  prot->rsk_prot->obj_size, 0,
1408                                                                  SLAB_HWCACHE_ALIGN, NULL, NULL);
1409
1410                         if (prot->rsk_prot->slab == NULL) {
1411                                 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
1412                                        prot->name);
1413                                 goto out_free_request_sock_slab_name;
1414                         }
1415                 }
1416         }
1417
1418         write_lock(&proto_list_lock);
1419         list_add(&prot->node, &proto_list);
1420         write_unlock(&proto_list_lock);
1421         rc = 0;
1422 out:
1423         return rc;
1424 out_free_request_sock_slab_name:
1425         kfree(request_sock_slab_name);
1426 out_free_sock_slab:
1427         kmem_cache_destroy(prot->slab);
1428         prot->slab = NULL;
1429         goto out;
1430 }
1431
1432 EXPORT_SYMBOL(proto_register);
1433
1434 void proto_unregister(struct proto *prot)
1435 {
1436         write_lock(&proto_list_lock);
1437
1438         if (prot->slab != NULL) {
1439                 kmem_cache_destroy(prot->slab);
1440                 prot->slab = NULL;
1441         }
1442
1443         if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
1444                 const char *name = kmem_cache_name(prot->rsk_prot->slab);
1445
1446                 kmem_cache_destroy(prot->rsk_prot->slab);
1447                 kfree(name);
1448                 prot->rsk_prot->slab = NULL;
1449         }
1450
1451         list_del(&prot->node);
1452         write_unlock(&proto_list_lock);
1453 }
1454
1455 EXPORT_SYMBOL(proto_unregister);
1456
1457 #ifdef CONFIG_PROC_FS
1458 static inline struct proto *__proto_head(void)
1459 {
1460         return list_entry(proto_list.next, struct proto, node);
1461 }
1462
1463 static inline struct proto *proto_head(void)
1464 {
1465         return list_empty(&proto_list) ? NULL : __proto_head();
1466 }
1467
1468 static inline struct proto *proto_next(struct proto *proto)
1469 {
1470         return proto->node.next == &proto_list ? NULL :
1471                 list_entry(proto->node.next, struct proto, node);
1472 }
1473
1474 static inline struct proto *proto_get_idx(loff_t pos)
1475 {
1476         struct proto *proto;
1477         loff_t i = 0;
1478
1479         list_for_each_entry(proto, &proto_list, node)
1480                 if (i++ == pos)
1481                         goto out;
1482
1483         proto = NULL;
1484 out:
1485         return proto;
1486 }
1487
1488 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
1489 {
1490         read_lock(&proto_list_lock);
1491         return *pos ? proto_get_idx(*pos - 1) : SEQ_START_TOKEN;
1492 }
1493
1494 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1495 {
1496         ++*pos;
1497         return v == SEQ_START_TOKEN ? proto_head() : proto_next(v);
1498 }
1499
1500 static void proto_seq_stop(struct seq_file *seq, void *v)
1501 {
1502         read_unlock(&proto_list_lock);
1503 }
1504
1505 static char proto_method_implemented(const void *method)
1506 {
1507         return method == NULL ? 'n' : 'y';
1508 }
1509
1510 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
1511 {
1512         seq_printf(seq, "%-9s %4u %6d  %6d   %-3s %6u   %-3s  %-10s "
1513                         "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
1514                    proto->name,
1515                    proto->obj_size,
1516                    proto->sockets_allocated != NULL ? atomic_read(proto->sockets_allocated) : -1,
1517                    proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
1518                    proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
1519                    proto->max_header,
1520                    proto->slab == NULL ? "no" : "yes",
1521                    module_name(proto->owner),
1522                    proto_method_implemented(proto->close),
1523                    proto_method_implemented(proto->connect),
1524                    proto_method_implemented(proto->disconnect),
1525                    proto_method_implemented(proto->accept),
1526                    proto_method_implemented(proto->ioctl),
1527                    proto_method_implemented(proto->init),
1528                    proto_method_implemented(proto->destroy),
1529                    proto_method_implemented(proto->shutdown),
1530                    proto_method_implemented(proto->setsockopt),
1531                    proto_method_implemented(proto->getsockopt),
1532                    proto_method_implemented(proto->sendmsg),
1533                    proto_method_implemented(proto->recvmsg),
1534                    proto_method_implemented(proto->sendpage),
1535                    proto_method_implemented(proto->bind),
1536                    proto_method_implemented(proto->backlog_rcv),
1537                    proto_method_implemented(proto->hash),
1538                    proto_method_implemented(proto->unhash),
1539                    proto_method_implemented(proto->get_port),
1540                    proto_method_implemented(proto->enter_memory_pressure));
1541 }
1542
1543 static int proto_seq_show(struct seq_file *seq, void *v)
1544 {
1545         if (v == SEQ_START_TOKEN)
1546                 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
1547                            "protocol",
1548                            "size",
1549                            "sockets",
1550                            "memory",
1551                            "press",
1552                            "maxhdr",
1553                            "slab",
1554                            "module",
1555                            "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
1556         else
1557                 proto_seq_printf(seq, v);
1558         return 0;
1559 }
1560
1561 static struct seq_operations proto_seq_ops = {
1562         .start  = proto_seq_start,
1563         .next   = proto_seq_next,
1564         .stop   = proto_seq_stop,
1565         .show   = proto_seq_show,
1566 };
1567
1568 static int proto_seq_open(struct inode *inode, struct file *file)
1569 {
1570         return seq_open(file, &proto_seq_ops);
1571 }
1572
1573 static struct file_operations proto_seq_fops = {
1574         .owner          = THIS_MODULE,
1575         .open           = proto_seq_open,
1576         .read           = seq_read,
1577         .llseek         = seq_lseek,
1578         .release        = seq_release,
1579 };
1580
1581 static int __init proto_init(void)
1582 {
1583         /* register /proc/net/protocols */
1584         return proc_net_fops_create("protocols", S_IRUGO, &proto_seq_fops) == NULL ? -ENOBUFS : 0;
1585 }
1586
1587 subsys_initcall(proto_init);
1588
1589 #endif /* PROC_FS */
1590
1591 EXPORT_SYMBOL(sk_alloc);
1592 EXPORT_SYMBOL(sk_free);
1593 EXPORT_SYMBOL(sk_send_sigurg);
1594 EXPORT_SYMBOL(sock_alloc_send_skb);
1595 EXPORT_SYMBOL(sock_init_data);
1596 EXPORT_SYMBOL(sock_kfree_s);
1597 EXPORT_SYMBOL(sock_kmalloc);
1598 EXPORT_SYMBOL(sock_no_accept);
1599 EXPORT_SYMBOL(sock_no_bind);
1600 EXPORT_SYMBOL(sock_no_connect);
1601 EXPORT_SYMBOL(sock_no_getname);
1602 EXPORT_SYMBOL(sock_no_getsockopt);
1603 EXPORT_SYMBOL(sock_no_ioctl);
1604 EXPORT_SYMBOL(sock_no_listen);
1605 EXPORT_SYMBOL(sock_no_mmap);
1606 EXPORT_SYMBOL(sock_no_poll);
1607 EXPORT_SYMBOL(sock_no_recvmsg);
1608 EXPORT_SYMBOL(sock_no_sendmsg);
1609 EXPORT_SYMBOL(sock_no_sendpage);
1610 EXPORT_SYMBOL(sock_no_setsockopt);
1611 EXPORT_SYMBOL(sock_no_shutdown);
1612 EXPORT_SYMBOL(sock_no_socketpair);
1613 EXPORT_SYMBOL(sock_rfree);
1614 EXPORT_SYMBOL(sock_setsockopt);
1615 EXPORT_SYMBOL(sock_wfree);
1616 EXPORT_SYMBOL(sock_wmalloc);
1617 EXPORT_SYMBOL(sock_i_uid);
1618 EXPORT_SYMBOL(sock_i_ino);
1619 #ifdef CONFIG_SYSCTL
1620 EXPORT_SYMBOL(sysctl_optmem_max);
1621 EXPORT_SYMBOL(sysctl_rmem_max);
1622 EXPORT_SYMBOL(sysctl_wmem_max);
1623 #endif