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