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