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/capability.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 #include <linux/highmem.h>
 115
 116 #include <asm/uaccess.h>
 117 #include <asm/system.h>
 118
 119 #include <linux/netdevice.h>
 120 #include <net/protocol.h>
 121 #include <linux/skbuff.h>
 122 #include <net/request_sock.h>
 123 #include <net/sock.h>
 124 #include <net/xfrm.h>
 125 #include <linux/ipsec.h>
 126
 127 #include <linux/filter.h>
 128 #include <linux/vs_socket.h>
 129 #include <linux/vs_limit.h>
 130 #include <linux/vs_context.h>
 131
 132 #ifdef CONFIG_INET
 133 #include <net/tcp.h>
 134 #endif
 135
 136 /*
 137  * Each address family might have different locking rules, so we have
 138  * one slock key per address family:
 139  */
 140 static struct lock_class_key af_family_keys[AF_MAX];
 141 static struct lock_class_key af_family_slock_keys[AF_MAX];
 142
 143 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 144 /*
 145  * Make lock validator output more readable. (we pre-construct these
 146  * strings build-time, so that runtime initialization of socket
 147  * locks is fast):
 148  */
 149 static const char *af_family_key_strings[AF_MAX+1] = {
 150   "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX"     , "sk_lock-AF_INET"     ,
 151   "sk_lock-AF_AX25"  , "sk_lock-AF_IPX"      , "sk_lock-AF_APPLETALK",
 152   "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE"   , "sk_lock-AF_ATMPVC"   ,
 153   "sk_lock-AF_X25"   , "sk_lock-AF_INET6"    , "sk_lock-AF_ROSE"     ,
 154   "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI"  , "sk_lock-AF_SECURITY" ,
 155   "sk_lock-AF_KEY"   , "sk_lock-AF_NETLINK"  , "sk_lock-AF_PACKET"   ,
 156   "sk_lock-AF_ASH"   , "sk_lock-AF_ECONET"   , "sk_lock-AF_ATMSVC"   ,
 157   "sk_lock-21"       , "sk_lock-AF_SNA"      , "sk_lock-AF_IRDA"     ,
 158   "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE"  , "sk_lock-AF_LLC"      ,
 159   "sk_lock-27"       , "sk_lock-28"          , "sk_lock-29"          ,
 160   "sk_lock-AF_TIPC"  , "sk_lock-AF_BLUETOOTH", "sk_lock-AF_MAX"
 161 };
 162 static const char *af_family_slock_key_strings[AF_MAX+1] = {
 163   "slock-AF_UNSPEC", "slock-AF_UNIX"     , "slock-AF_INET"     ,
 164   "slock-AF_AX25"  , "slock-AF_IPX"      , "slock-AF_APPLETALK",
 165   "slock-AF_NETROM", "slock-AF_BRIDGE"   , "slock-AF_ATMPVC"   ,
 166   "slock-AF_X25"   , "slock-AF_INET6"    , "slock-AF_ROSE"     ,
 167   "slock-AF_DECnet", "slock-AF_NETBEUI"  , "slock-AF_SECURITY" ,
 168   "slock-AF_KEY"   , "slock-AF_NETLINK"  , "slock-AF_PACKET"   ,
 169   "slock-AF_ASH"   , "slock-AF_ECONET"   , "slock-AF_ATMSVC"   ,
 170   "slock-21"       , "slock-AF_SNA"      , "slock-AF_IRDA"     ,
 171   "slock-AF_PPPOX" , "slock-AF_WANPIPE"  , "slock-AF_LLC"      ,
 172   "slock-27"       , "slock-28"          , "slock-29"          ,
 173   "slock-AF_TIPC"  , "slock-AF_BLUETOOTH", "slock-AF_MAX"
 174 };
 175 #endif
 176
 177 /*
 178  * sk_callback_lock locking rules are per-address-family,
 179  * so split the lock classes by using a per-AF key:
 180  */
 181 static struct lock_class_key af_callback_keys[AF_MAX];
 182
 183 /* Take into consideration the size of the struct sk_buff overhead in the
 184  * determination of these values, since that is non-constant across
 185  * platforms.  This makes socket queueing behavior and performance
 186  * not depend upon such differences.
 187  */
 188 #define _SK_MEM_PACKETS         256
 189 #define _SK_MEM_OVERHEAD        (sizeof(struct sk_buff) + 256)
 190 #define SK_WMEM_MAX             (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
 191 #define SK_RMEM_MAX             (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
 192
 193 /* Run time adjustable parameters. */
 194 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
 195 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
 196 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
 197 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
 198
 199 /* Maximal space eaten by iovec or ancilliary data plus some space */
 200 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
 201
 202 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
 203 {
 204         struct timeval tv;
 205
 206         if (optlen < sizeof(tv))
 207                 return -EINVAL;
 208         if (copy_from_user(&tv, optval, sizeof(tv)))
 209                 return -EFAULT;
 210
 211         *timeo_p = MAX_SCHEDULE_TIMEOUT;
 212         if (tv.tv_sec == 0 && tv.tv_usec == 0)
 213                 return 0;
 214         if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
 215                 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
 216         return 0;
 217 }
 218
 219 static void sock_warn_obsolete_bsdism(const char *name)
 220 {
 221         static int warned;
 222         static char warncomm[TASK_COMM_LEN];
 223         if (strcmp(warncomm, current->comm) && warned < 5) {
 224                 strcpy(warncomm,  current->comm);
 225                 printk(KERN_WARNING "process `%s' is using obsolete "
 226                        "%s SO_BSDCOMPAT\n", warncomm, name);
 227                 warned++;
 228         }
 229 }
 230
 231 static void sock_disable_timestamp(struct sock *sk)
 232 {
 233         if (sock_flag(sk, SOCK_TIMESTAMP)) {
 234                 sock_reset_flag(sk, SOCK_TIMESTAMP);
 235                 net_disable_timestamp();
 236         }
 237 }
 238
 239
 240 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
 241 {
 242         int err = 0;
 243         int skb_len;
 244
 245         /* Cast skb->rcvbuf to unsigned... It's pointless, but reduces
 246            number of warnings when compiling with -W --ANK
 247          */
 248         if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
 249             (unsigned)sk->sk_rcvbuf) {
 250                 err = -ENOMEM;
 251                 goto out;
 252         }
 253
 254         err = sk_filter(sk, skb);
 255         if (err)
 256                 goto out;
 257
 258         skb->dev = NULL;
 259         skb_set_owner_r(skb, sk);
 260
 261         /* Cache the SKB length before we tack it onto the receive
 262          * queue.  Once it is added it no longer belongs to us and
 263          * may be freed by other threads of control pulling packets
 264          * from the queue.
 265          */
 266         skb_len = skb->len;
 267
 268         skb_queue_tail(&sk->sk_receive_queue, skb);
 269
 270         if (!sock_flag(sk, SOCK_DEAD))
 271                 sk->sk_data_ready(sk, skb_len);
 272 out:
 273         return err;
 274 }
 275 EXPORT_SYMBOL(sock_queue_rcv_skb);
 276
 277 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
 278 {
 279         int rc = NET_RX_SUCCESS;
 280
 281         if (sk_filter(sk, skb))
 282                 goto discard_and_relse;
 283
 284         skb->dev = NULL;
 285
 286         if (nested)
 287                 bh_lock_sock_nested(sk);
 288         else
 289                 bh_lock_sock(sk);
 290         if (!sock_owned_by_user(sk)) {
 291                 /*
 292                  * trylock + unlock semantics:
 293                  */
 294                 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
 295
 296                 rc = sk->sk_backlog_rcv(sk, skb);
 297
 298                 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
 299         } else
 300                 sk_add_backlog(sk, skb);
 301         bh_unlock_sock(sk);
 302 out:
 303         sock_put(sk);
 304         return rc;
 305 discard_and_relse:
 306         kfree_skb(skb);
 307         goto out;
 308 }
 309 EXPORT_SYMBOL(sk_receive_skb);
 310
 311 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
 312 {
 313         struct dst_entry *dst = sk->sk_dst_cache;
 314
 315         if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
 316                 sk->sk_dst_cache = NULL;
 317                 dst_release(dst);
 318                 return NULL;
 319         }
 320
 321         return dst;
 322 }
 323 EXPORT_SYMBOL(__sk_dst_check);
 324
 325 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
 326 {
 327         struct dst_entry *dst = sk_dst_get(sk);
 328
 329         if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
 330                 sk_dst_reset(sk);
 331                 dst_release(dst);
 332                 return NULL;
 333         }
 334
 335         return dst;
 336 }
 337 EXPORT_SYMBOL(sk_dst_check);
 338
 339 /*
 340  *      This is meant for all protocols to use and covers goings on
 341  *      at the socket level. Everything here is generic.
 342  */
 343
 344 int sock_setsockopt(struct socket *sock, int level, int optname,
 345                     char __user *optval, int optlen)
 346 {
 347         struct sock *sk=sock->sk;
 348         struct sk_filter *filter;
 349         int val;
 350         int valbool;
 351         struct linger ling;
 352         int ret = 0;
 353
 354         /*
 355          *      Options without arguments
 356          */
 357
 358 #ifdef SO_DONTLINGER            /* Compatibility item... */
 359         if (optname == SO_DONTLINGER) {
 360                 lock_sock(sk);
 361                 sock_reset_flag(sk, SOCK_LINGER);
 362                 release_sock(sk);
 363                 return 0;
 364         }
 365 #endif
 366
 367         if(optlen<sizeof(int))
 368                 return(-EINVAL);
 369
 370         if (get_user(val, (int __user *)optval))
 371                 return -EFAULT;
 372
 373         valbool = val?1:0;
 374
 375         lock_sock(sk);
 376
 377         switch(optname)
 378         {
 379                 case SO_DEBUG:
 380                         if(val && !capable(CAP_NET_ADMIN))
 381                         {
 382                                 ret = -EACCES;
 383                         }
 384                         else if (valbool)
 385                                 sock_set_flag(sk, SOCK_DBG);
 386                         else
 387                                 sock_reset_flag(sk, SOCK_DBG);
 388                         break;
 389                 case SO_REUSEADDR:
 390                         sk->sk_reuse = valbool;
 391                         break;
 392                 case SO_TYPE:
 393                 case SO_ERROR:
 394                         ret = -ENOPROTOOPT;
 395                         break;
 396                 case SO_DONTROUTE:
 397                         if (valbool)
 398                                 sock_set_flag(sk, SOCK_LOCALROUTE);
 399                         else
 400                                 sock_reset_flag(sk, SOCK_LOCALROUTE);
 401                         break;
 402                 case SO_BROADCAST:
 403                         sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
 404                         break;
 405                 case SO_SNDBUF:
 406                         /* Don't error on this BSD doesn't and if you think
 407                            about it this is right. Otherwise apps have to
 408                            play 'guess the biggest size' games. RCVBUF/SNDBUF
 409                            are treated in BSD as hints */
 410
 411                         if (val > sysctl_wmem_max)
 412                                 val = sysctl_wmem_max;
 413 set_sndbuf:
 414                         sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
 415                         if ((val * 2) < SOCK_MIN_SNDBUF)
 416                                 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
 417                         else
 418                                 sk->sk_sndbuf = val * 2;
 419
 420                         /*
 421                          *      Wake up sending tasks if we
 422                          *      upped the value.
 423                          */
 424                         sk->sk_write_space(sk);
 425                         break;
 426
 427                 case SO_SNDBUFFORCE:
 428                         if (!capable(CAP_NET_ADMIN)) {
 429                                 ret = -EPERM;
 430                                 break;
 431                         }
 432                         goto set_sndbuf;
 433
 434                 case SO_SETXID:
 435                         if (current->xid) {
 436                                 ret = -EPERM;
 437                                 break;
 438                         }
 439                         if (val < 0 || val > MAX_S_CONTEXT) {
 440                                 ret = -EINVAL;
 441                                 break;
 442                         }
 443                         sk->sk_xid = val;
 444                         break;
 445
 446                 case SO_RCVBUF:
 447                         /* Don't error on this BSD doesn't and if you think
 448                            about it this is right. Otherwise apps have to
 449                            play 'guess the biggest size' games. RCVBUF/SNDBUF
 450                            are treated in BSD as hints */
 451
 452                         if (val > sysctl_rmem_max)
 453                                 val = sysctl_rmem_max;
 454 set_rcvbuf:
 455                         sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
 456                         /*
 457                          * We double it on the way in to account for
 458                          * "struct sk_buff" etc. overhead.   Applications
 459                          * assume that the SO_RCVBUF setting they make will
 460                          * allow that much actual data to be received on that
 461                          * socket.
 462                          *
 463                          * Applications are unaware that "struct sk_buff" and
 464                          * other overheads allocate from the receive buffer
 465                          * during socket buffer allocation.
 466                          *
 467                          * And after considering the possible alternatives,
 468                          * returning the value we actually used in getsockopt
 469                          * is the most desirable behavior.
 470                          */
 471                         if ((val * 2) < SOCK_MIN_RCVBUF)
 472                                 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
 473                         else
 474                                 sk->sk_rcvbuf = val * 2;
 475                         break;
 476
 477                 case SO_RCVBUFFORCE:
 478                         if (!capable(CAP_NET_ADMIN)) {
 479                                 ret = -EPERM;
 480                                 break;
 481                         }
 482                         goto set_rcvbuf;
 483
 484                 case SO_KEEPALIVE:
 485 #ifdef CONFIG_INET
 486                         if (sk->sk_protocol == IPPROTO_TCP)
 487                                 tcp_set_keepalive(sk, valbool);
 488 #endif
 489                         sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
 490                         break;
 491
 492                 case SO_OOBINLINE:
 493                         sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
 494                         break;
 495
 496                 case SO_NO_CHECK:
 497                         sk->sk_no_check = valbool;
 498                         break;
 499
 500                 case SO_PRIORITY:
 501                         if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
 502                                 sk->sk_priority = val;
 503                         else
 504                                 ret = -EPERM;
 505                         break;
 506
 507                 case SO_LINGER:
 508                         if(optlen<sizeof(ling)) {
 509                                 ret = -EINVAL;  /* 1003.1g */
 510                                 break;
 511                         }
 512                         if (copy_from_user(&ling,optval,sizeof(ling))) {
 513                                 ret = -EFAULT;
 514                                 break;
 515                         }
 516                         if (!ling.l_onoff)
 517                                 sock_reset_flag(sk, SOCK_LINGER);
 518                         else {
 519 #if (BITS_PER_LONG == 32)
 520                                 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
 521                                         sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
 522                                 else
 523 #endif
 524                                         sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
 525                                 sock_set_flag(sk, SOCK_LINGER);
 526                         }
 527                         break;
 528
 529                 case SO_BSDCOMPAT:
 530                         sock_warn_obsolete_bsdism("setsockopt");
 531                         break;
 532
 533                 case SO_PASSCRED:
 534                         if (valbool)
 535                                 set_bit(SOCK_PASSCRED, &sock->flags);
 536                         else
 537                                 clear_bit(SOCK_PASSCRED, &sock->flags);
 538                         break;
 539
 540                 case SO_TIMESTAMP:
 541                         if (valbool)  {
 542                                 sock_set_flag(sk, SOCK_RCVTSTAMP);
 543                                 sock_enable_timestamp(sk);
 544                         } else
 545                                 sock_reset_flag(sk, SOCK_RCVTSTAMP);
 546                         break;
 547
 548                 case SO_RCVLOWAT:
 549                         if (val < 0)
 550                                 val = INT_MAX;
 551                         sk->sk_rcvlowat = val ? : 1;
 552                         break;
 553
 554                 case SO_RCVTIMEO:
 555                         ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
 556                         break;
 557
 558                 case SO_SNDTIMEO:
 559                         ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
 560                         break;
 561
 562 #ifdef CONFIG_NETDEVICES
 563                 case SO_BINDTODEVICE:
 564                 {
 565                         char devname[IFNAMSIZ];
 566
 567                         /* Sorry... */
 568                         if (!nx_capable(CAP_NET_RAW, NXC_RAW_SOCKET)) {
 569                                 ret = -EPERM;
 570                                 break;
 571                         }
 572
 573                         /* Bind this socket to a particular device like "eth0",
 574                          * as specified in the passed interface name. If the
 575                          * name is "" or the option length is zero the socket
 576                          * is not bound.
 577                          */
 578
 579                         if (!valbool) {
 580                                 sk->sk_bound_dev_if = 0;
 581                         } else {
 582                                 if (optlen > IFNAMSIZ - 1)
 583                                         optlen = IFNAMSIZ - 1;
 584                                 memset(devname, 0, sizeof(devname));
 585                                 if (copy_from_user(devname, optval, optlen)) {
 586                                         ret = -EFAULT;
 587                                         break;
 588                                 }
 589
 590                                 /* Remove any cached route for this socket. */
 591                                 sk_dst_reset(sk);
 592
 593                                 if (devname[0] == '\0') {
 594                                         sk->sk_bound_dev_if = 0;
 595                                 } else {
 596                                         struct net_device *dev = dev_get_by_name(devname);
 597                                         if (!dev) {
 598                                                 ret = -ENODEV;
 599                                                 break;
 600                                         }
 601                                         sk->sk_bound_dev_if = dev->ifindex;
 602                                         dev_put(dev);
 603                                 }
 604                         }
 605                         break;
 606                 }
 607 #endif
 608
 609
 610                 case SO_ATTACH_FILTER:
 611                         ret = -EINVAL;
 612                         if (optlen == sizeof(struct sock_fprog)) {
 613                                 struct sock_fprog fprog;
 614
 615                                 ret = -EFAULT;
 616                                 if (copy_from_user(&fprog, optval, sizeof(fprog)))
 617                                         break;
 618
 619                                 ret = sk_attach_filter(&fprog, sk);
 620                         }
 621                         break;
 622
 623                 case SO_DETACH_FILTER:
 624                         rcu_read_lock_bh();
 625                         filter = rcu_dereference(sk->sk_filter);
 626                         if (filter) {
 627                                 rcu_assign_pointer(sk->sk_filter, NULL);
 628                                 sk_filter_release(sk, filter);
 629                                 rcu_read_unlock_bh();
 630                                 break;
 631                         }
 632                         rcu_read_unlock_bh();
 633                         ret = -ENONET;
 634                         break;
 635
 636                 case SO_PASSSEC:
 637                         if (valbool)
 638                                 set_bit(SOCK_PASSSEC, &sock->flags);
 639                         else
 640                                 clear_bit(SOCK_PASSSEC, &sock->flags);
 641                         break;
 642
 643                 /* We implement the SO_SNDLOWAT etc to
 644                    not be settable (1003.1g 5.3) */
 645                 default:
 646                         ret = -ENOPROTOOPT;
 647                         break;
 648         }
 649         release_sock(sk);
 650         return ret;
 651 }
 652
 653
 654 int sock_getsockopt(struct socket *sock, int level, int optname,
 655                     char __user *optval, int __user *optlen)
 656 {
 657         struct sock *sk = sock->sk;
 658
 659         union
 660         {
 661                 int val;
 662                 struct linger ling;
 663                 struct timeval tm;
 664         } v;
 665
 666         unsigned int lv = sizeof(int);
 667         int len;
 668
 669         if(get_user(len,optlen))
 670                 return -EFAULT;
 671         if(len < 0)
 672                 return -EINVAL;
 673
 674         switch(optname)
 675         {
 676                 case SO_DEBUG:
 677                         v.val = sock_flag(sk, SOCK_DBG);
 678                         break;
 679
 680                 case SO_DONTROUTE:
 681                         v.val = sock_flag(sk, SOCK_LOCALROUTE);
 682                         break;
 683
 684                 case SO_BROADCAST:
 685                         v.val = !!sock_flag(sk, SOCK_BROADCAST);
 686                         break;
 687
 688                 case SO_SNDBUF:
 689                         v.val = sk->sk_sndbuf;
 690                         break;
 691
 692                 case SO_RCVBUF:
 693                         v.val = sk->sk_rcvbuf;
 694                         break;
 695
 696                 case SO_REUSEADDR:
 697                         v.val = sk->sk_reuse;
 698                         break;
 699
 700                 case SO_KEEPALIVE:
 701                         v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
 702                         break;
 703
 704                 case SO_TYPE:
 705                         v.val = sk->sk_type;
 706                         break;
 707
 708                 case SO_ERROR:
 709                         v.val = -sock_error(sk);
 710                         if(v.val==0)
 711                                 v.val = xchg(&sk->sk_err_soft, 0);
 712                         break;
 713
 714                 case SO_OOBINLINE:
 715                         v.val = !!sock_flag(sk, SOCK_URGINLINE);
 716                         break;
 717
 718                 case SO_NO_CHECK:
 719                         v.val = sk->sk_no_check;
 720                         break;
 721
 722                 case SO_PRIORITY:
 723                         v.val = sk->sk_priority;
 724                         break;
 725
 726                 case SO_LINGER:
 727                         lv              = sizeof(v.ling);
 728                         v.ling.l_onoff  = !!sock_flag(sk, SOCK_LINGER);
 729                         v.ling.l_linger = sk->sk_lingertime / HZ;
 730                         break;
 731
 732                 case SO_BSDCOMPAT:
 733                         sock_warn_obsolete_bsdism("getsockopt");
 734                         break;
 735
 736                 case SO_TIMESTAMP:
 737                         v.val = sock_flag(sk, SOCK_RCVTSTAMP);
 738                         break;
 739
 740                 case SO_RCVTIMEO:
 741                         lv=sizeof(struct timeval);
 742                         if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
 743                                 v.tm.tv_sec = 0;
 744                                 v.tm.tv_usec = 0;
 745                         } else {
 746                                 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
 747                                 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
 748                         }
 749                         break;
 750
 751                 case SO_SNDTIMEO:
 752                         lv=sizeof(struct timeval);
 753                         if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
 754                                 v.tm.tv_sec = 0;
 755                                 v.tm.tv_usec = 0;
 756                         } else {
 757                                 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
 758                                 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
 759                         }
 760                         break;
 761
 762                 case SO_RCVLOWAT:
 763                         v.val = sk->sk_rcvlowat;
 764                         break;
 765
 766                 case SO_SNDLOWAT:
 767                         v.val=1;
 768                         break;
 769
 770                 case SO_PASSCRED:
 771                         v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
 772                         break;
 773
 774                 case SO_PEERCRED:
 775                         if (len > sizeof(sk->sk_peercred))
 776                                 len = sizeof(sk->sk_peercred);
 777                         if (copy_to_user(optval, &sk->sk_peercred, len))
 778                                 return -EFAULT;
 779
 780                         goto lenout;
 781
 782                 case SO_PEERNAME:
 783                 {
 784                         char address[128];
 785
 786                         if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
 787                                 return -ENOTCONN;
 788                         if (lv < len)
 789                                 return -EINVAL;
 790                         if (copy_to_user(optval, address, len))
 791                                 return -EFAULT;
 792                         goto lenout;
 793                 }
 794
 795                 /* Dubious BSD thing... Probably nobody even uses it, but
 796                  * the UNIX standard wants it for whatever reason... -DaveM
 797                  */
 798                 case SO_ACCEPTCONN:
 799                         v.val = sk->sk_state == TCP_LISTEN;
 800                         break;
 801
 802                 case SO_PASSSEC:
 803                         v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
 804                         break;
 805
 806                 case SO_PEERSEC:
 807                         return security_socket_getpeersec_stream(sock, optval, optlen, len);
 808
 809                 default:
 810                         return(-ENOPROTOOPT);
 811         }
 812         if (len > lv)
 813                 len = lv;
 814         if (copy_to_user(optval, &v, len))
 815                 return -EFAULT;
 816 lenout:
 817         if (put_user(len, optlen))
 818                 return -EFAULT;
 819         return 0;
 820 }
 821
 822 /*
 823  * Initialize an sk_lock.
 824  *
 825  * (We also register the sk_lock with the lock validator.)
 826  */
 827 static void inline sock_lock_init(struct sock *sk)
 828 {
 829         sock_lock_init_class_and_name(sk,
 830                         af_family_slock_key_strings[sk->sk_family],
 831                         af_family_slock_keys + sk->sk_family,
 832                         af_family_key_strings[sk->sk_family],
 833                         af_family_keys + sk->sk_family);
 834 }
 835
 836 /**
 837  *      sk_alloc - All socket objects are allocated here
 838  *      @family: protocol family
 839  *      @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
 840  *      @prot: struct proto associated with this new sock instance
 841  *      @zero_it: if we should zero the newly allocated sock
 842  */
 843 struct sock *sk_alloc(int family, gfp_t priority,
 844                       struct proto *prot, int zero_it)
 845 {
 846         struct sock *sk = NULL;
 847         struct kmem_cache *slab = prot->slab;
 848
 849         if (slab != NULL)
 850                 sk = kmem_cache_alloc(slab, priority);
 851         else
 852                 sk = kmalloc(prot->obj_size, priority);
 853
 854         if (sk) {
 855                 if (zero_it) {
 856                         memset(sk, 0, prot->obj_size);
 857                         sk->sk_family = family;
 858                         /*
 859                          * See comment in struct sock definition to understand
 860                          * why we need sk_prot_creator -acme
 861                          */
 862                         sk->sk_prot = sk->sk_prot_creator = prot;
 863                         sock_lock_init(sk);
 864                 }
 865                 sock_vx_init(sk);
 866                 sock_nx_init(sk);
 867
 868                 if (security_sk_alloc(sk, family, priority))
 869                         goto out_free;
 870
 871                 if (!try_module_get(prot->owner))
 872                         goto out_free;
 873         }
 874         return sk;
 875
 876 out_free:
 877         if (slab != NULL)
 878                 kmem_cache_free(slab, sk);
 879         else
 880                 kfree(sk);
 881         return NULL;
 882 }
 883
 884 void sk_free(struct sock *sk)
 885 {
 886         struct sk_filter *filter;
 887         struct module *owner = sk->sk_prot_creator->owner;
 888
 889         if (sk->sk_destruct)
 890                 sk->sk_destruct(sk);
 891
 892         filter = rcu_dereference(sk->sk_filter);
 893         if (filter) {
 894                 sk_filter_release(sk, filter);
 895                 rcu_assign_pointer(sk->sk_filter, NULL);
 896         }
 897
 898         sock_disable_timestamp(sk);
 899
 900         if (atomic_read(&sk->sk_omem_alloc))
 901                 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
 902                        __FUNCTION__, atomic_read(&sk->sk_omem_alloc));
 903
 904         security_sk_free(sk);
 905         vx_sock_dec(sk);
 906         clr_vx_info(&sk->sk_vx_info);
 907         sk->sk_xid = -1;
 908         clr_nx_info(&sk->sk_nx_info);
 909         sk->sk_nid = -1;
 910         if (sk->sk_prot_creator->slab != NULL)
 911                 kmem_cache_free(sk->sk_prot_creator->slab, sk);
 912         else
 913                 kfree(sk);
 914         module_put(owner);
 915 }
 916
 917 struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
 918 {
 919         struct sock *newsk = sk_alloc(sk->sk_family, priority, sk->sk_prot, 0);
 920
 921         if (newsk != NULL) {
 922                 struct sk_filter *filter;
 923
 924                 sock_copy(newsk, sk);
 925
 926                 /* SANITY */
 927                 sock_vx_init(newsk);
 928                 sock_nx_init(newsk);
 929                 sk_node_init(&newsk->sk_node);
 930                 sock_lock_init(newsk);
 931                 bh_lock_sock(newsk);
 932
 933                 atomic_set(&newsk->sk_rmem_alloc, 0);
 934                 atomic_set(&newsk->sk_wmem_alloc, 0);
 935                 atomic_set(&newsk->sk_omem_alloc, 0);
 936                 skb_queue_head_init(&newsk->sk_receive_queue);
 937                 skb_queue_head_init(&newsk->sk_write_queue);
 938 #ifdef CONFIG_NET_DMA
 939                 skb_queue_head_init(&newsk->sk_async_wait_queue);
 940 #endif
 941
 942                 rwlock_init(&newsk->sk_dst_lock);
 943                 rwlock_init(&newsk->sk_callback_lock);
 944                 lockdep_set_class(&newsk->sk_callback_lock,
 945                                    af_callback_keys + newsk->sk_family);
 946
 947                 newsk->sk_dst_cache     = NULL;
 948                 newsk->sk_wmem_queued   = 0;
 949                 newsk->sk_forward_alloc = 0;
 950                 newsk->sk_send_head     = NULL;
 951                 newsk->sk_backlog.head  = newsk->sk_backlog.tail = NULL;
 952                 newsk->sk_userlocks     = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
 953
 954                 sock_reset_flag(newsk, SOCK_DONE);
 955                 skb_queue_head_init(&newsk->sk_error_queue);
 956
 957                 filter = newsk->sk_filter;
 958                 if (filter != NULL)
 959                         sk_filter_charge(newsk, filter);
 960
 961                 if (unlikely(xfrm_sk_clone_policy(newsk))) {
 962                         /* It is still raw copy of parent, so invalidate
 963                          * destructor and make plain sk_free() */
 964                         newsk->sk_destruct = NULL;
 965                         sk_free(newsk);
 966                         newsk = NULL;
 967                         goto out;
 968                 }
 969
 970                 newsk->sk_err      = 0;
 971                 newsk->sk_priority = 0;
 972                 atomic_set(&newsk->sk_refcnt, 2);
 973
 974                 set_vx_info(&newsk->sk_vx_info, sk->sk_vx_info);
 975                 newsk->sk_xid = sk->sk_xid;
 976                 vx_sock_inc(newsk);
 977                 set_nx_info(&newsk->sk_nx_info, sk->sk_nx_info);
 978                 newsk->sk_nid = sk->sk_nid;
 979
 980                 /*
 981                  * Increment the counter in the same struct proto as the master
 982                  * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
 983                  * is the same as sk->sk_prot->socks, as this field was copied
 984                  * with memcpy).
 985                  *
 986                  * This _changes_ the previous behaviour, where
 987                  * tcp_create_openreq_child always was incrementing the
 988                  * equivalent to tcp_prot->socks (inet_sock_nr), so this have
 989                  * to be taken into account in all callers. -acme
 990                  */
 991                 sk_refcnt_debug_inc(newsk);
 992                 newsk->sk_socket = NULL;
 993                 newsk->sk_sleep  = NULL;
 994
 995                 if (newsk->sk_prot->sockets_allocated)
 996                         atomic_inc(newsk->sk_prot->sockets_allocated);
 997         }
 998 out:
 999         return newsk;
1000 }
1001
1002 EXPORT_SYMBOL_GPL(sk_clone);
1003
1004 void __init sk_init(void)
1005 {
1006         if (num_physpages <= 4096) {
1007                 sysctl_wmem_max = 32767;
1008                 sysctl_rmem_max = 32767;
1009                 sysctl_wmem_default = 32767;
1010                 sysctl_rmem_default = 32767;
1011         } else if (num_physpages >= 131072) {
1012                 sysctl_wmem_max = 131071;
1013                 sysctl_rmem_max = 131071;
1014         }
1015 }
1016
1017 /*
1018  *      Simple resource managers for sockets.
1019  */
1020
1021
1022 /*
1023  * Write buffer destructor automatically called from kfree_skb.
1024  */
1025 void sock_wfree(struct sk_buff *skb)
1026 {
1027         struct sock *sk = skb->sk;
1028
1029         /* In case it might be waiting for more memory. */
1030         atomic_sub(skb->truesize, &sk->sk_wmem_alloc);
1031         if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE))
1032                 sk->sk_write_space(sk);
1033         sock_put(sk);
1034 }
1035
1036 /*
1037  * Read buffer destructor automatically called from kfree_skb.
1038  */
1039 void sock_rfree(struct sk_buff *skb)
1040 {
1041         struct sock *sk = skb->sk;
1042
1043         atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1044 }
1045
1046
1047 int sock_i_uid(struct sock *sk)
1048 {
1049         int uid;
1050
1051         read_lock(&sk->sk_callback_lock);
1052         uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1053         read_unlock(&sk->sk_callback_lock);
1054         return uid;
1055 }
1056
1057 unsigned long sock_i_ino(struct sock *sk)
1058 {
1059         unsigned long ino;
1060
1061         read_lock(&sk->sk_callback_lock);
1062         ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1063         read_unlock(&sk->sk_callback_lock);
1064         return ino;
1065 }
1066
1067 /*
1068  * Allocate a skb from the socket's send buffer.
1069  */
1070 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1071                              gfp_t priority)
1072 {
1073         if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1074                 struct sk_buff * skb = alloc_skb(size, priority);
1075                 if (skb) {
1076                         skb_set_owner_w(skb, sk);
1077                         return skb;
1078                 }
1079         }
1080         return NULL;
1081 }
1082
1083 /*
1084  * Allocate a skb from the socket's receive buffer.
1085  */
1086 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1087                              gfp_t priority)
1088 {
1089         if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1090                 struct sk_buff *skb = alloc_skb(size, priority);
1091                 if (skb) {
1092                         skb_set_owner_r(skb, sk);
1093                         return skb;
1094                 }
1095         }
1096         return NULL;
1097 }
1098
1099 /*
1100  * Allocate a memory block from the socket's option memory buffer.
1101  */
1102 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1103 {
1104         if ((unsigned)size <= sysctl_optmem_max &&
1105             atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1106                 void *mem;
1107                 /* First do the add, to avoid the race if kmalloc
1108                  * might sleep.
1109                  */
1110                 atomic_add(size, &sk->sk_omem_alloc);
1111                 mem = kmalloc(size, priority);
1112                 if (mem)
1113                         return mem;
1114                 atomic_sub(size, &sk->sk_omem_alloc);
1115         }
1116         return NULL;
1117 }
1118
1119 /*
1120  * Free an option memory block.
1121  */
1122 void sock_kfree_s(struct sock *sk, void *mem, int size)
1123 {
1124         kfree(mem);
1125         atomic_sub(size, &sk->sk_omem_alloc);
1126 }
1127
1128 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1129    I think, these locks should be removed for datagram sockets.
1130  */
1131 static long sock_wait_for_wmem(struct sock * sk, long timeo)
1132 {
1133         DEFINE_WAIT(wait);
1134
1135         clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1136         for (;;) {
1137                 if (!timeo)
1138                         break;
1139                 if (signal_pending(current))
1140                         break;
1141                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1142                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1143                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1144                         break;
1145                 if (sk->sk_shutdown & SEND_SHUTDOWN)
1146                         break;
1147                 if (sk->sk_err)
1148                         break;
1149                 timeo = schedule_timeout(timeo);
1150         }
1151         finish_wait(sk->sk_sleep, &wait);
1152         return timeo;
1153 }
1154
1155
1156 /*
1157  *      Generic send/receive buffer handlers
1158  */
1159
1160 static struct sk_buff *sock_alloc_send_pskb(struct sock *sk,
1161                                             unsigned long header_len,
1162                                             unsigned long data_len,
1163                                             int noblock, int *errcode)
1164 {
1165         struct sk_buff *skb;
1166         gfp_t gfp_mask;
1167         long timeo;
1168         int err;
1169
1170         gfp_mask = sk->sk_allocation;
1171         if (gfp_mask & __GFP_WAIT)
1172                 gfp_mask |= __GFP_REPEAT;
1173
1174         timeo = sock_sndtimeo(sk, noblock);
1175         while (1) {
1176                 err = sock_error(sk);
1177                 if (err != 0)
1178                         goto failure;
1179
1180                 err = -EPIPE;
1181                 if (sk->sk_shutdown & SEND_SHUTDOWN)
1182                         goto failure;
1183
1184                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1185                         skb = alloc_skb(header_len, gfp_mask);
1186                         if (skb) {
1187                                 int npages;
1188                                 int i;
1189
1190                                 /* No pages, we're done... */
1191                                 if (!data_len)
1192                                         break;
1193
1194                                 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1195                                 skb->truesize += data_len;
1196                                 skb_shinfo(skb)->nr_frags = npages;
1197                                 for (i = 0; i < npages; i++) {
1198                                         struct page *page;
1199                                         skb_frag_t *frag;
1200
1201                                         page = alloc_pages(sk->sk_allocation, 0);
1202                                         if (!page) {
1203                                                 err = -ENOBUFS;
1204                                                 skb_shinfo(skb)->nr_frags = i;
1205                                                 kfree_skb(skb);
1206                                                 goto failure;
1207                                         }
1208
1209                                         frag = &skb_shinfo(skb)->frags[i];
1210                                         frag->page = page;
1211                                         frag->page_offset = 0;
1212                                         frag->size = (data_len >= PAGE_SIZE ?
1213                                                       PAGE_SIZE :
1214                                                       data_len);
1215                                         data_len -= PAGE_SIZE;
1216                                 }
1217
1218                                 /* Full success... */
1219                                 break;
1220                         }
1221                         err = -ENOBUFS;
1222                         goto failure;
1223                 }
1224                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1225                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1226                 err = -EAGAIN;
1227                 if (!timeo)
1228                         goto failure;
1229                 if (signal_pending(current))
1230                         goto interrupted;
1231                 timeo = sock_wait_for_wmem(sk, timeo);
1232         }
1233
1234         skb_set_owner_w(skb, sk);
1235         return skb;
1236
1237 interrupted:
1238         err = sock_intr_errno(timeo);
1239 failure:
1240         *errcode = err;
1241         return NULL;
1242 }
1243
1244 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1245                                     int noblock, int *errcode)
1246 {
1247         return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1248 }
1249
1250 static void __lock_sock(struct sock *sk)
1251 {
1252         DEFINE_WAIT(wait);
1253
1254         for(;;) {
1255                 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1256                                         TASK_UNINTERRUPTIBLE);
1257                 spin_unlock_bh(&sk->sk_lock.slock);
1258                 schedule();
1259                 spin_lock_bh(&sk->sk_lock.slock);
1260                 if(!sock_owned_by_user(sk))
1261                         break;
1262         }
1263         finish_wait(&sk->sk_lock.wq, &wait);
1264 }
1265
1266 static void __release_sock(struct sock *sk)
1267 {
1268         struct sk_buff *skb = sk->sk_backlog.head;
1269
1270         do {
1271                 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1272                 bh_unlock_sock(sk);
1273
1274                 do {
1275                         struct sk_buff *next = skb->next;
1276
1277                         skb->next = NULL;
1278                         sk->sk_backlog_rcv(sk, skb);
1279
1280                         /*
1281                          * We are in process context here with softirqs
1282                          * disabled, use cond_resched_softirq() to preempt.
1283                          * This is safe to do because we've taken the backlog
1284                          * queue private:
1285                          */
1286                         cond_resched_softirq();
1287
1288                         skb = next;
1289                 } while (skb != NULL);
1290
1291                 bh_lock_sock(sk);
1292         } while((skb = sk->sk_backlog.head) != NULL);
1293 }
1294
1295 /**
1296  * sk_wait_data - wait for data to arrive at sk_receive_queue
1297  * @sk:    sock to wait on
1298  * @timeo: for how long
1299  *
1300  * Now socket state including sk->sk_err is changed only under lock,
1301  * hence we may omit checks after joining wait queue.
1302  * We check receive queue before schedule() only as optimization;
1303  * it is very likely that release_sock() added new data.
1304  */
1305 int sk_wait_data(struct sock *sk, long *timeo)
1306 {
1307         int rc;
1308         DEFINE_WAIT(wait);
1309
1310         prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1311         set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1312         rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1313         clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1314         finish_wait(sk->sk_sleep, &wait);
1315         return rc;
1316 }
1317
1318 EXPORT_SYMBOL(sk_wait_data);
1319
1320 /*
1321  * Set of default routines for initialising struct proto_ops when
1322  * the protocol does not support a particular function. In certain
1323  * cases where it makes no sense for a protocol to have a "do nothing"
1324  * function, some default processing is provided.
1325  */
1326
1327 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1328 {
1329         return -EOPNOTSUPP;
1330 }
1331
1332 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1333                     int len, int flags)
1334 {
1335         return -EOPNOTSUPP;
1336 }
1337
1338 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1339 {
1340         return -EOPNOTSUPP;
1341 }
1342
1343 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1344 {
1345         return -EOPNOTSUPP;
1346 }
1347
1348 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1349                     int *len, int peer)
1350 {
1351         return -EOPNOTSUPP;
1352 }
1353
1354 unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt)
1355 {
1356         return 0;
1357 }
1358
1359 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1360 {
1361         return -EOPNOTSUPP;
1362 }
1363
1364 int sock_no_listen(struct socket *sock, int backlog)
1365 {
1366         return -EOPNOTSUPP;
1367 }
1368
1369 int sock_no_shutdown(struct socket *sock, int how)
1370 {
1371         return -EOPNOTSUPP;
1372 }
1373
1374 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1375                     char __user *optval, int optlen)
1376 {
1377         return -EOPNOTSUPP;
1378 }
1379
1380 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1381                     char __user *optval, int __user *optlen)
1382 {
1383         return -EOPNOTSUPP;
1384 }
1385
1386 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1387                     size_t len)
1388 {
1389         return -EOPNOTSUPP;
1390 }
1391
1392 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1393                     size_t len, int flags)
1394 {
1395         return -EOPNOTSUPP;
1396 }
1397
1398 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1399 {
1400         /* Mirror missing mmap method error code */
1401         return -ENODEV;
1402 }
1403
1404 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1405 {
1406         ssize_t res;
1407         struct msghdr msg = {.msg_flags = flags};
1408         struct kvec iov;
1409         char *kaddr = kmap(page);
1410         iov.iov_base = kaddr + offset;
1411         iov.iov_len = size;
1412         res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1413         kunmap(page);
1414         return res;
1415 }
1416
1417 /*
1418  *      Default Socket Callbacks
1419  */
1420
1421 static void sock_def_wakeup(struct sock *sk)
1422 {
1423         read_lock(&sk->sk_callback_lock);
1424         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1425                 wake_up_interruptible_all(sk->sk_sleep);
1426         read_unlock(&sk->sk_callback_lock);
1427 }
1428
1429 static void sock_def_error_report(struct sock *sk)
1430 {
1431         read_lock(&sk->sk_callback_lock);
1432         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1433                 wake_up_interruptible(sk->sk_sleep);
1434         sk_wake_async(sk,0,POLL_ERR);
1435         read_unlock(&sk->sk_callback_lock);
1436 }
1437
1438 static void sock_def_readable(struct sock *sk, int len)
1439 {
1440         read_lock(&sk->sk_callback_lock);
1441         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1442                 wake_up_interruptible(sk->sk_sleep);
1443         sk_wake_async(sk,1,POLL_IN);
1444         read_unlock(&sk->sk_callback_lock);
1445 }
1446
1447 static void sock_def_write_space(struct sock *sk)
1448 {
1449         read_lock(&sk->sk_callback_lock);
1450
1451         /* Do not wake up a writer until he can make "significant"
1452          * progress.  --DaveM
1453          */
1454         if((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1455                 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1456                         wake_up_interruptible(sk->sk_sleep);
1457
1458                 /* Should agree with poll, otherwise some programs break */
1459                 if (sock_writeable(sk))
1460                         sk_wake_async(sk, 2, POLL_OUT);
1461         }
1462
1463         read_unlock(&sk->sk_callback_lock);
1464 }
1465
1466 static void sock_def_destruct(struct sock *sk)
1467 {
1468         kfree(sk->sk_protinfo);
1469 }
1470
1471 void sk_send_sigurg(struct sock *sk)
1472 {
1473         if (sk->sk_socket && sk->sk_socket->file)
1474                 if (send_sigurg(&sk->sk_socket->file->f_owner))
1475                         sk_wake_async(sk, 3, POLL_PRI);
1476 }
1477
1478 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1479                     unsigned long expires)
1480 {
1481         if (!mod_timer(timer, expires))
1482                 sock_hold(sk);
1483 }
1484
1485 EXPORT_SYMBOL(sk_reset_timer);
1486
1487 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1488 {
1489         if (timer_pending(timer) && del_timer(timer))
1490                 __sock_put(sk);
1491 }
1492
1493 EXPORT_SYMBOL(sk_stop_timer);
1494
1495 void sock_init_data(struct socket *sock, struct sock *sk)
1496 {
1497         skb_queue_head_init(&sk->sk_receive_queue);
1498         skb_queue_head_init(&sk->sk_write_queue);
1499         skb_queue_head_init(&sk->sk_error_queue);
1500 #ifdef CONFIG_NET_DMA
1501         skb_queue_head_init(&sk->sk_async_wait_queue);
1502 #endif
1503
1504         sk->sk_send_head        =       NULL;
1505
1506         init_timer(&sk->sk_timer);
1507
1508         sk->sk_allocation       =       GFP_KERNEL;
1509         sk->sk_rcvbuf           =       sysctl_rmem_default;
1510         sk->sk_sndbuf           =       sysctl_wmem_default;
1511         sk->sk_state            =       TCP_CLOSE;
1512         sk->sk_socket           =       sock;
1513
1514         sock_set_flag(sk, SOCK_ZAPPED);
1515
1516         if(sock)
1517         {
1518                 sk->sk_type     =       sock->type;
1519                 sk->sk_sleep    =       &sock->wait;
1520                 sock->sk        =       sk;
1521         } else
1522                 sk->sk_sleep    =       NULL;
1523
1524         rwlock_init(&sk->sk_dst_lock);
1525         rwlock_init(&sk->sk_callback_lock);
1526         lockdep_set_class(&sk->sk_callback_lock,
1527                            af_callback_keys + sk->sk_family);
1528
1529         sk->sk_state_change     =       sock_def_wakeup;
1530         sk->sk_data_ready       =       sock_def_readable;
1531         sk->sk_write_space      =       sock_def_write_space;
1532         sk->sk_error_report     =       sock_def_error_report;
1533         sk->sk_destruct         =       sock_def_destruct;
1534
1535         sk->sk_sndmsg_page      =       NULL;
1536         sk->sk_sndmsg_off       =       0;
1537
1538         sk->sk_peercred.pid     =       0;
1539         sk->sk_peercred.uid     =       -1;
1540         sk->sk_peercred.gid     =       -1;
1541         sk->sk_write_pending    =       0;
1542         sk->sk_rcvlowat         =       1;
1543         sk->sk_rcvtimeo         =       MAX_SCHEDULE_TIMEOUT;
1544         sk->sk_sndtimeo         =       MAX_SCHEDULE_TIMEOUT;
1545
1546         sk->sk_stamp.tv_sec     = -1L;
1547         sk->sk_stamp.tv_usec    = -1L;
1548
1549         set_vx_info(&sk->sk_vx_info, current->vx_info);
1550         sk->sk_xid = vx_current_xid();
1551         vx_sock_inc(sk);
1552         set_nx_info(&sk->sk_nx_info, current->nx_info);
1553         sk->sk_nid = nx_current_nid();
1554         atomic_set(&sk->sk_refcnt, 1);
1555 }
1556
1557 void fastcall lock_sock_nested(struct sock *sk, int subclass)
1558 {
1559         might_sleep();
1560         spin_lock_bh(&sk->sk_lock.slock);
1561         if (sk->sk_lock.owner)
1562                 __lock_sock(sk);
1563         sk->sk_lock.owner = (void *)1;
1564         spin_unlock(&sk->sk_lock.slock);
1565         /*
1566          * The sk_lock has mutex_lock() semantics here:
1567          */
1568         mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
1569         local_bh_enable();
1570 }
1571
1572 EXPORT_SYMBOL(lock_sock_nested);
1573
1574 void fastcall release_sock(struct sock *sk)
1575 {
1576         /*
1577          * The sk_lock has mutex_unlock() semantics:
1578          */
1579         mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1580
1581         spin_lock_bh(&sk->sk_lock.slock);
1582         if (sk->sk_backlog.tail)
1583                 __release_sock(sk);
1584         sk->sk_lock.owner = NULL;
1585         if (waitqueue_active(&sk->sk_lock.wq))
1586                 wake_up(&sk->sk_lock.wq);
1587         spin_unlock_bh(&sk->sk_lock.slock);
1588 }
1589 EXPORT_SYMBOL(release_sock);
1590
1591 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1592 {
1593         if (!sock_flag(sk, SOCK_TIMESTAMP))
1594                 sock_enable_timestamp(sk);
1595         if (sk->sk_stamp.tv_sec == -1)
1596                 return -ENOENT;
1597         if (sk->sk_stamp.tv_sec == 0)
1598                 do_gettimeofday(&sk->sk_stamp);
1599         return copy_to_user(userstamp, &sk->sk_stamp, sizeof(struct timeval)) ?
1600                 -EFAULT : 0;
1601 }
1602 EXPORT_SYMBOL(sock_get_timestamp);
1603
1604 void sock_enable_timestamp(struct sock *sk)
1605 {
1606         if (!sock_flag(sk, SOCK_TIMESTAMP)) {
1607                 sock_set_flag(sk, SOCK_TIMESTAMP);
1608                 net_enable_timestamp();
1609         }
1610 }
1611 EXPORT_SYMBOL(sock_enable_timestamp);
1612
1613 /*
1614  *      Get a socket option on an socket.
1615  *
1616  *      FIX: POSIX 1003.1g is very ambiguous here. It states that
1617  *      asynchronous errors should be reported by getsockopt. We assume
1618  *      this means if you specify SO_ERROR (otherwise whats the point of it).
1619  */
1620 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1621                            char __user *optval, int __user *optlen)
1622 {
1623         struct sock *sk = sock->sk;
1624
1625         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1626 }
1627
1628 EXPORT_SYMBOL(sock_common_getsockopt);
1629
1630 #ifdef CONFIG_COMPAT
1631 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
1632                                   char __user *optval, int __user *optlen)
1633 {
1634         struct sock *sk = sock->sk;
1635
1636         if (sk->sk_prot->compat_getsockopt != NULL)
1637                 return sk->sk_prot->compat_getsockopt(sk, level, optname,
1638                                                       optval, optlen);
1639         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1640 }
1641 EXPORT_SYMBOL(compat_sock_common_getsockopt);
1642 #endif
1643
1644 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1645                         struct msghdr *msg, size_t size, int flags)
1646 {
1647         struct sock *sk = sock->sk;
1648         int addr_len = 0;
1649         int err;
1650
1651         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
1652                                    flags & ~MSG_DONTWAIT, &addr_len);
1653         if (err >= 0)
1654                 msg->msg_namelen = addr_len;
1655         return err;
1656 }
1657
1658 EXPORT_SYMBOL(sock_common_recvmsg);
1659
1660 /*
1661  *      Set socket options on an inet socket.
1662  */
1663 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1664                            char __user *optval, int optlen)
1665 {
1666         struct sock *sk = sock->sk;
1667
1668         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1669 }
1670
1671 EXPORT_SYMBOL(sock_common_setsockopt);
1672
1673 #ifdef CONFIG_COMPAT
1674 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
1675                                   char __user *optval, int optlen)
1676 {
1677         struct sock *sk = sock->sk;
1678
1679         if (sk->sk_prot->compat_setsockopt != NULL)
1680                 return sk->sk_prot->compat_setsockopt(sk, level, optname,
1681                                                       optval, optlen);
1682         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1683 }
1684 EXPORT_SYMBOL(compat_sock_common_setsockopt);
1685 #endif
1686
1687 void sk_common_release(struct sock *sk)
1688 {
1689         if (sk->sk_prot->destroy)
1690                 sk->sk_prot->destroy(sk);
1691
1692         /*
1693          * Observation: when sock_common_release is called, processes have
1694          * no access to socket. But net still has.
1695          * Step one, detach it from networking:
1696          *
1697          * A. Remove from hash tables.
1698          */
1699
1700         sk->sk_prot->unhash(sk);
1701
1702         /*
1703          * In this point socket cannot receive new packets, but it is possible
1704          * that some packets are in flight because some CPU runs receiver and
1705          * did hash table lookup before we unhashed socket. They will achieve
1706          * receive queue and will be purged by socket destructor.
1707          *
1708          * Also we still have packets pending on receive queue and probably,
1709          * our own packets waiting in device queues. sock_destroy will drain
1710          * receive queue, but transmitted packets will delay socket destruction
1711          * until the last reference will be released.
1712          */
1713
1714         sock_orphan(sk);
1715
1716         xfrm_sk_free_policy(sk);
1717
1718         sk_refcnt_debug_release(sk);
1719         sock_put(sk);
1720 }
1721
1722 EXPORT_SYMBOL(sk_common_release);
1723
1724 static DEFINE_RWLOCK(proto_list_lock);
1725 static LIST_HEAD(proto_list);
1726
1727 int proto_register(struct proto *prot, int alloc_slab)
1728 {
1729         char *request_sock_slab_name = NULL;
1730         char *timewait_sock_slab_name;
1731         int rc = -ENOBUFS;
1732
1733         if (alloc_slab) {
1734                 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
1735                                                SLAB_HWCACHE_ALIGN, NULL, NULL);
1736
1737                 if (prot->slab == NULL) {
1738                         printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
1739                                prot->name);
1740                         goto out;
1741                 }
1742
1743                 if (prot->rsk_prot != NULL) {
1744                         static const char mask[] = "request_sock_%s";
1745
1746                         request_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
1747                         if (request_sock_slab_name == NULL)
1748                                 goto out_free_sock_slab;
1749
1750                         sprintf(request_sock_slab_name, mask, prot->name);
1751                         prot->rsk_prot->slab = kmem_cache_create(request_sock_slab_name,
1752                                                                  prot->rsk_prot->obj_size, 0,
1753                                                                  SLAB_HWCACHE_ALIGN, NULL, NULL);
1754
1755                         if (prot->rsk_prot->slab == NULL) {
1756                                 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
1757                                        prot->name);
1758                                 goto out_free_request_sock_slab_name;
1759                         }
1760                 }
1761
1762                 if (prot->twsk_prot != NULL) {
1763                         static const char mask[] = "tw_sock_%s";
1764
1765                         timewait_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
1766
1767                         if (timewait_sock_slab_name == NULL)
1768                                 goto out_free_request_sock_slab;
1769
1770                         sprintf(timewait_sock_slab_name, mask, prot->name);
1771                         prot->twsk_prot->twsk_slab =
1772                                 kmem_cache_create(timewait_sock_slab_name,
1773                                                   prot->twsk_prot->twsk_obj_size,
1774                                                   0, SLAB_HWCACHE_ALIGN,
1775                                                   NULL, NULL);
1776                         if (prot->twsk_prot->twsk_slab == NULL)
1777                                 goto out_free_timewait_sock_slab_name;
1778                 }
1779         }
1780
1781         write_lock(&proto_list_lock);
1782         list_add(&prot->node, &proto_list);
1783         write_unlock(&proto_list_lock);
1784         rc = 0;
1785 out:
1786         return rc;
1787 out_free_timewait_sock_slab_name:
1788         kfree(timewait_sock_slab_name);
1789 out_free_request_sock_slab:
1790         if (prot->rsk_prot && prot->rsk_prot->slab) {
1791                 kmem_cache_destroy(prot->rsk_prot->slab);
1792                 prot->rsk_prot->slab = NULL;
1793         }
1794 out_free_request_sock_slab_name:
1795         kfree(request_sock_slab_name);
1796 out_free_sock_slab:
1797         kmem_cache_destroy(prot->slab);
1798         prot->slab = NULL;
1799         goto out;
1800 }
1801
1802 EXPORT_SYMBOL(proto_register);
1803
1804 void proto_unregister(struct proto *prot)
1805 {
1806         write_lock(&proto_list_lock);
1807         list_del(&prot->node);
1808         write_unlock(&proto_list_lock);
1809
1810         if (prot->slab != NULL) {
1811                 kmem_cache_destroy(prot->slab);
1812                 prot->slab = NULL;
1813         }
1814
1815         if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
1816                 const char *name = kmem_cache_name(prot->rsk_prot->slab);
1817
1818                 kmem_cache_destroy(prot->rsk_prot->slab);
1819                 kfree(name);
1820                 prot->rsk_prot->slab = NULL;
1821         }
1822
1823         if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
1824                 const char *name = kmem_cache_name(prot->twsk_prot->twsk_slab);
1825
1826                 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
1827                 kfree(name);
1828                 prot->twsk_prot->twsk_slab = NULL;
1829         }
1830 }
1831
1832 EXPORT_SYMBOL(proto_unregister);
1833
1834 #ifdef CONFIG_PROC_FS
1835 static inline struct proto *__proto_head(void)
1836 {
1837         return list_entry(proto_list.next, struct proto, node);
1838 }
1839
1840 static inline struct proto *proto_head(void)
1841 {
1842         return list_empty(&proto_list) ? NULL : __proto_head();
1843 }
1844
1845 static inline struct proto *proto_next(struct proto *proto)
1846 {
1847         return proto->node.next == &proto_list ? NULL :
1848                 list_entry(proto->node.next, struct proto, node);
1849 }
1850
1851 static inline struct proto *proto_get_idx(loff_t pos)
1852 {
1853         struct proto *proto;
1854         loff_t i = 0;
1855
1856         list_for_each_entry(proto, &proto_list, node)
1857                 if (i++ == pos)
1858                         goto out;
1859
1860         proto = NULL;
1861 out:
1862         return proto;
1863 }
1864
1865 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
1866 {
1867         read_lock(&proto_list_lock);
1868         return *pos ? proto_get_idx(*pos - 1) : SEQ_START_TOKEN;
1869 }
1870
1871 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1872 {
1873         ++*pos;
1874         return v == SEQ_START_TOKEN ? proto_head() : proto_next(v);
1875 }
1876
1877 static void proto_seq_stop(struct seq_file *seq, void *v)
1878 {
1879         read_unlock(&proto_list_lock);
1880 }
1881
1882 static char proto_method_implemented(const void *method)
1883 {
1884         return method == NULL ? 'n' : 'y';
1885 }
1886
1887 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
1888 {
1889         seq_printf(seq, "%-9s %4u %6d  %6d   %-3s %6u   %-3s  %-10s "
1890                         "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
1891                    proto->name,
1892                    proto->obj_size,
1893                    proto->sockets_allocated != NULL ? atomic_read(proto->sockets_allocated) : -1,
1894                    proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
1895                    proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
1896                    proto->max_header,
1897                    proto->slab == NULL ? "no" : "yes",
1898                    module_name(proto->owner),
1899                    proto_method_implemented(proto->close),
1900                    proto_method_implemented(proto->connect),
1901                    proto_method_implemented(proto->disconnect),
1902                    proto_method_implemented(proto->accept),
1903                    proto_method_implemented(proto->ioctl),
1904                    proto_method_implemented(proto->init),
1905                    proto_method_implemented(proto->destroy),
1906                    proto_method_implemented(proto->shutdown),
1907                    proto_method_implemented(proto->setsockopt),
1908                    proto_method_implemented(proto->getsockopt),
1909                    proto_method_implemented(proto->sendmsg),
1910                    proto_method_implemented(proto->recvmsg),
1911                    proto_method_implemented(proto->sendpage),
1912                    proto_method_implemented(proto->bind),
1913                    proto_method_implemented(proto->backlog_rcv),
1914                    proto_method_implemented(proto->hash),
1915                    proto_method_implemented(proto->unhash),
1916                    proto_method_implemented(proto->get_port),
1917                    proto_method_implemented(proto->enter_memory_pressure));
1918 }
1919
1920 static int proto_seq_show(struct seq_file *seq, void *v)
1921 {
1922         if (v == SEQ_START_TOKEN)
1923                 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
1924                            "protocol",
1925                            "size",
1926                            "sockets",
1927                            "memory",
1928                            "press",
1929                            "maxhdr",
1930                            "slab",
1931                            "module",
1932                            "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
1933         else
1934                 proto_seq_printf(seq, v);
1935         return 0;
1936 }
1937
1938 static struct seq_operations proto_seq_ops = {
1939         .start  = proto_seq_start,
1940         .next   = proto_seq_next,
1941         .stop   = proto_seq_stop,
1942         .show   = proto_seq_show,
1943 };
1944
1945 static int proto_seq_open(struct inode *inode, struct file *file)
1946 {
1947         return seq_open(file, &proto_seq_ops);
1948 }
1949
1950 static struct file_operations proto_seq_fops = {
1951         .owner          = THIS_MODULE,
1952         .open           = proto_seq_open,
1953         .read           = seq_read,
1954         .llseek         = seq_lseek,
1955         .release        = seq_release,
1956 };
1957
1958 static int __init proto_init(void)
1959 {
1960         /* register /proc/net/protocols */
1961         return proc_net_fops_create("protocols", S_IRUGO, &proto_seq_fops) == NULL ? -ENOBUFS : 0;
1962 }
1963
1964 subsys_initcall(proto_init);
1965
1966 #endif /* PROC_FS */
1967
1968 EXPORT_SYMBOL(sk_alloc);
1969 EXPORT_SYMBOL(sk_free);
1970 EXPORT_SYMBOL(sk_send_sigurg);
1971 EXPORT_SYMBOL(sock_alloc_send_skb);
1972 EXPORT_SYMBOL(sock_init_data);
1973 EXPORT_SYMBOL(sock_kfree_s);
1974 EXPORT_SYMBOL(sock_kmalloc);
1975 EXPORT_SYMBOL(sock_no_accept);
1976 EXPORT_SYMBOL(sock_no_bind);
1977 EXPORT_SYMBOL(sock_no_connect);
1978 EXPORT_SYMBOL(sock_no_getname);
1979 EXPORT_SYMBOL(sock_no_getsockopt);
1980 EXPORT_SYMBOL(sock_no_ioctl);
1981 EXPORT_SYMBOL(sock_no_listen);
1982 EXPORT_SYMBOL(sock_no_mmap);
1983 EXPORT_SYMBOL(sock_no_poll);
1984 EXPORT_SYMBOL(sock_no_recvmsg);
1985 EXPORT_SYMBOL(sock_no_sendmsg);
1986 EXPORT_SYMBOL(sock_no_sendpage);
1987 EXPORT_SYMBOL(sock_no_setsockopt);
1988 EXPORT_SYMBOL(sock_no_shutdown);
1989 EXPORT_SYMBOL(sock_no_socketpair);
1990 EXPORT_SYMBOL(sock_rfree);
1991 EXPORT_SYMBOL(sock_setsockopt);
1992 EXPORT_SYMBOL(sock_wfree);
1993 EXPORT_SYMBOL(sock_wmalloc);
1994 EXPORT_SYMBOL(sock_i_uid);
1995 EXPORT_SYMBOL(sock_i_ino);
1996 EXPORT_SYMBOL(sysctl_optmem_max);
1997 #ifdef CONFIG_SYSCTL
1998 EXPORT_SYMBOL(sysctl_rmem_max);
1999 EXPORT_SYMBOL(sysctl_wmem_max);
2000 #endif