git.oblomov.eu Git - linux-2.6/blob - 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/net_namespace.h>
 123 #include <net/request_sock.h>
 124 #include <net/sock.h>
 125 #include <net/xfrm.h>
 126 #include <linux/ipsec.h>
 127
 128 #include <linux/filter.h>
 129
 130 #ifdef CONFIG_INET
 131 #include <net/tcp.h>
 132 #endif
 133
 134 /*
 135  * Each address family might have different locking rules, so we have
 136  * one slock key per address family:
 137  */
 138 static struct lock_class_key af_family_keys[AF_MAX];
 139 static struct lock_class_key af_family_slock_keys[AF_MAX];
 140
 141 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 142 /*
 143  * Make lock validator output more readable. (we pre-construct these
 144  * strings build-time, so that runtime initialization of socket
 145  * locks is fast):
 146  */
 147 static const char *af_family_key_strings[AF_MAX+1] = {
 148   "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX"     , "sk_lock-AF_INET"     ,
 149   "sk_lock-AF_AX25"  , "sk_lock-AF_IPX"      , "sk_lock-AF_APPLETALK",
 150   "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE"   , "sk_lock-AF_ATMPVC"   ,
 151   "sk_lock-AF_X25"   , "sk_lock-AF_INET6"    , "sk_lock-AF_ROSE"     ,
 152   "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI"  , "sk_lock-AF_SECURITY" ,
 153   "sk_lock-AF_KEY"   , "sk_lock-AF_NETLINK"  , "sk_lock-AF_PACKET"   ,
 154   "sk_lock-AF_ASH"   , "sk_lock-AF_ECONET"   , "sk_lock-AF_ATMSVC"   ,
 155   "sk_lock-21"       , "sk_lock-AF_SNA"      , "sk_lock-AF_IRDA"     ,
 156   "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE"  , "sk_lock-AF_LLC"      ,
 157   "sk_lock-27"       , "sk_lock-28"          , "sk_lock-29"          ,
 158   "sk_lock-AF_TIPC"  , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV"        ,
 159   "sk_lock-AF_RXRPC" , "sk_lock-AF_MAX"
 160 };
 161 static const char *af_family_slock_key_strings[AF_MAX+1] = {
 162   "slock-AF_UNSPEC", "slock-AF_UNIX"     , "slock-AF_INET"     ,
 163   "slock-AF_AX25"  , "slock-AF_IPX"      , "slock-AF_APPLETALK",
 164   "slock-AF_NETROM", "slock-AF_BRIDGE"   , "slock-AF_ATMPVC"   ,
 165   "slock-AF_X25"   , "slock-AF_INET6"    , "slock-AF_ROSE"     ,
 166   "slock-AF_DECnet", "slock-AF_NETBEUI"  , "slock-AF_SECURITY" ,
 167   "slock-AF_KEY"   , "slock-AF_NETLINK"  , "slock-AF_PACKET"   ,
 168   "slock-AF_ASH"   , "slock-AF_ECONET"   , "slock-AF_ATMSVC"   ,
 169   "slock-21"       , "slock-AF_SNA"      , "slock-AF_IRDA"     ,
 170   "slock-AF_PPPOX" , "slock-AF_WANPIPE"  , "slock-AF_LLC"      ,
 171   "slock-27"       , "slock-28"          , "slock-29"          ,
 172   "slock-AF_TIPC"  , "slock-AF_BLUETOOTH", "slock-AF_IUCV"     ,
 173   "slock-AF_RXRPC" , "slock-AF_MAX"
 174 };
 175 static const char *af_family_clock_key_strings[AF_MAX+1] = {
 176   "clock-AF_UNSPEC", "clock-AF_UNIX"     , "clock-AF_INET"     ,
 177   "clock-AF_AX25"  , "clock-AF_IPX"      , "clock-AF_APPLETALK",
 178   "clock-AF_NETROM", "clock-AF_BRIDGE"   , "clock-AF_ATMPVC"   ,
 179   "clock-AF_X25"   , "clock-AF_INET6"    , "clock-AF_ROSE"     ,
 180   "clock-AF_DECnet", "clock-AF_NETBEUI"  , "clock-AF_SECURITY" ,
 181   "clock-AF_KEY"   , "clock-AF_NETLINK"  , "clock-AF_PACKET"   ,
 182   "clock-AF_ASH"   , "clock-AF_ECONET"   , "clock-AF_ATMSVC"   ,
 183   "clock-21"       , "clock-AF_SNA"      , "clock-AF_IRDA"     ,
 184   "clock-AF_PPPOX" , "clock-AF_WANPIPE"  , "clock-AF_LLC"      ,
 185   "clock-27"       , "clock-28"          , "clock-29"          ,
 186   "clock-AF_TIPC"  , "clock-AF_BLUETOOTH", "clock-AF_IUCV"     ,
 187   "clock-AF_RXRPC" , "clock-AF_MAX"
 188 };
 189 #endif
 190
 191 /*
 192  * sk_callback_lock locking rules are per-address-family,
 193  * so split the lock classes by using a per-AF key:
 194  */
 195 static struct lock_class_key af_callback_keys[AF_MAX];
 196
 197 /* Take into consideration the size of the struct sk_buff overhead in the
 198  * determination of these values, since that is non-constant across
 199  * platforms.  This makes socket queueing behavior and performance
 200  * not depend upon such differences.
 201  */
 202 #define _SK_MEM_PACKETS         256
 203 #define _SK_MEM_OVERHEAD        (sizeof(struct sk_buff) + 256)
 204 #define SK_WMEM_MAX             (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
 205 #define SK_RMEM_MAX             (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
 206
 207 /* Run time adjustable parameters. */
 208 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
 209 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
 210 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
 211 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
 212
 213 /* Maximal space eaten by iovec or ancilliary data plus some space */
 214 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
 215
 216 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
 217 {
 218         struct timeval tv;
 219
 220         if (optlen < sizeof(tv))
 221                 return -EINVAL;
 222         if (copy_from_user(&tv, optval, sizeof(tv)))
 223                 return -EFAULT;
 224         if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
 225                 return -EDOM;
 226
 227         if (tv.tv_sec < 0) {
 228                 static int warned __read_mostly;
 229
 230                 *timeo_p = 0;
 231                 if (warned < 10 && net_ratelimit())
 232                         warned++;
 233                         printk(KERN_INFO "sock_set_timeout: `%s' (pid %d) "
 234                                "tries to set negative timeout\n",
 235                                 current->comm, task_pid_nr(current));
 236                 return 0;
 237         }
 238         *timeo_p = MAX_SCHEDULE_TIMEOUT;
 239         if (tv.tv_sec == 0 && tv.tv_usec == 0)
 240                 return 0;
 241         if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
 242                 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
 243         return 0;
 244 }
 245
 246 static void sock_warn_obsolete_bsdism(const char *name)
 247 {
 248         static int warned;
 249         static char warncomm[TASK_COMM_LEN];
 250         if (strcmp(warncomm, current->comm) && warned < 5) {
 251                 strcpy(warncomm,  current->comm);
 252                 printk(KERN_WARNING "process `%s' is using obsolete "
 253                        "%s SO_BSDCOMPAT\n", warncomm, name);
 254                 warned++;
 255         }
 256 }
 257
 258 static void sock_disable_timestamp(struct sock *sk)
 259 {
 260         if (sock_flag(sk, SOCK_TIMESTAMP)) {
 261                 sock_reset_flag(sk, SOCK_TIMESTAMP);
 262                 net_disable_timestamp();
 263         }
 264 }
 265
 266
 267 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
 268 {
 269         int err = 0;
 270         int skb_len;
 271
 272         /* Cast skb->rcvbuf to unsigned... It's pointless, but reduces
 273            number of warnings when compiling with -W --ANK
 274          */
 275         if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
 276             (unsigned)sk->sk_rcvbuf) {
 277                 err = -ENOMEM;
 278                 goto out;
 279         }
 280
 281         err = sk_filter(sk, skb);
 282         if (err)
 283                 goto out;
 284
 285         skb->dev = NULL;
 286         skb_set_owner_r(skb, sk);
 287
 288         /* Cache the SKB length before we tack it onto the receive
 289          * queue.  Once it is added it no longer belongs to us and
 290          * may be freed by other threads of control pulling packets
 291          * from the queue.
 292          */
 293         skb_len = skb->len;
 294
 295         skb_queue_tail(&sk->sk_receive_queue, skb);
 296
 297         if (!sock_flag(sk, SOCK_DEAD))
 298                 sk->sk_data_ready(sk, skb_len);
 299 out:
 300         return err;
 301 }
 302 EXPORT_SYMBOL(sock_queue_rcv_skb);
 303
 304 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
 305 {
 306         int rc = NET_RX_SUCCESS;
 307
 308         if (sk_filter(sk, skb))
 309                 goto discard_and_relse;
 310
 311         skb->dev = NULL;
 312
 313         if (nested)
 314                 bh_lock_sock_nested(sk);
 315         else
 316                 bh_lock_sock(sk);
 317         if (!sock_owned_by_user(sk)) {
 318                 /*
 319                  * trylock + unlock semantics:
 320                  */
 321                 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
 322
 323                 rc = sk->sk_backlog_rcv(sk, skb);
 324
 325                 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
 326         } else
 327                 sk_add_backlog(sk, skb);
 328         bh_unlock_sock(sk);
 329 out:
 330         sock_put(sk);
 331         return rc;
 332 discard_and_relse:
 333         kfree_skb(skb);
 334         goto out;
 335 }
 336 EXPORT_SYMBOL(sk_receive_skb);
 337
 338 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
 339 {
 340         struct dst_entry *dst = sk->sk_dst_cache;
 341
 342         if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
 343                 sk->sk_dst_cache = NULL;
 344                 dst_release(dst);
 345                 return NULL;
 346         }
 347
 348         return dst;
 349 }
 350 EXPORT_SYMBOL(__sk_dst_check);
 351
 352 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
 353 {
 354         struct dst_entry *dst = sk_dst_get(sk);
 355
 356         if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
 357                 sk_dst_reset(sk);
 358                 dst_release(dst);
 359                 return NULL;
 360         }
 361
 362         return dst;
 363 }
 364 EXPORT_SYMBOL(sk_dst_check);
 365
 366 static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
 367 {
 368         int ret = -ENOPROTOOPT;
 369 #ifdef CONFIG_NETDEVICES
 370         struct net *net = sk->sk_net;
 371         char devname[IFNAMSIZ];
 372         int index;
 373
 374         /* Sorry... */
 375         ret = -EPERM;
 376         if (!capable(CAP_NET_RAW))
 377                 goto out;
 378
 379         ret = -EINVAL;
 380         if (optlen < 0)
 381                 goto out;
 382
 383         /* Bind this socket to a particular device like "eth0",
 384          * as specified in the passed interface name. If the
 385          * name is "" or the option length is zero the socket
 386          * is not bound.
 387          */
 388         if (optlen > IFNAMSIZ - 1)
 389                 optlen = IFNAMSIZ - 1;
 390         memset(devname, 0, sizeof(devname));
 391
 392         ret = -EFAULT;
 393         if (copy_from_user(devname, optval, optlen))
 394                 goto out;
 395
 396         if (devname[0] == '\0') {
 397                 index = 0;
 398         } else {
 399                 struct net_device *dev = dev_get_by_name(net, devname);
 400
 401                 ret = -ENODEV;
 402                 if (!dev)
 403                         goto out;
 404
 405                 index = dev->ifindex;
 406                 dev_put(dev);
 407         }
 408
 409         lock_sock(sk);
 410         sk->sk_bound_dev_if = index;
 411         sk_dst_reset(sk);
 412         release_sock(sk);
 413
 414         ret = 0;
 415
 416 out:
 417 #endif
 418
 419         return ret;
 420 }
 421
 422 /*
 423  *      This is meant for all protocols to use and covers goings on
 424  *      at the socket level. Everything here is generic.
 425  */
 426
 427 int sock_setsockopt(struct socket *sock, int level, int optname,
 428                     char __user *optval, int optlen)
 429 {
 430         struct sock *sk=sock->sk;
 431         int val;
 432         int valbool;
 433         struct linger ling;
 434         int ret = 0;
 435
 436         /*
 437          *      Options without arguments
 438          */
 439
 440 #ifdef SO_DONTLINGER            /* Compatibility item... */
 441         if (optname == SO_DONTLINGER) {
 442                 lock_sock(sk);
 443                 sock_reset_flag(sk, SOCK_LINGER);
 444                 release_sock(sk);
 445                 return 0;
 446         }
 447 #endif
 448
 449         if (optname == SO_BINDTODEVICE)
 450                 return sock_bindtodevice(sk, optval, optlen);
 451
 452         if (optlen < sizeof(int))
 453                 return -EINVAL;
 454
 455         if (get_user(val, (int __user *)optval))
 456                 return -EFAULT;
 457
 458         valbool = val?1:0;
 459
 460         lock_sock(sk);
 461
 462         switch(optname) {
 463         case SO_DEBUG:
 464                 if (val && !capable(CAP_NET_ADMIN)) {
 465                         ret = -EACCES;
 466                 }
 467                 else if (valbool)
 468                         sock_set_flag(sk, SOCK_DBG);
 469                 else
 470                         sock_reset_flag(sk, SOCK_DBG);
 471                 break;
 472         case SO_REUSEADDR:
 473                 sk->sk_reuse = valbool;
 474                 break;
 475         case SO_TYPE:
 476         case SO_ERROR:
 477                 ret = -ENOPROTOOPT;
 478                 break;
 479         case SO_DONTROUTE:
 480                 if (valbool)
 481                         sock_set_flag(sk, SOCK_LOCALROUTE);
 482                 else
 483                         sock_reset_flag(sk, SOCK_LOCALROUTE);
 484                 break;
 485         case SO_BROADCAST:
 486                 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
 487                 break;
 488         case SO_SNDBUF:
 489                 /* Don't error on this BSD doesn't and if you think
 490                    about it this is right. Otherwise apps have to
 491                    play 'guess the biggest size' games. RCVBUF/SNDBUF
 492                    are treated in BSD as hints */
 493
 494                 if (val > sysctl_wmem_max)
 495                         val = sysctl_wmem_max;
 496 set_sndbuf:
 497                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
 498                 if ((val * 2) < SOCK_MIN_SNDBUF)
 499                         sk->sk_sndbuf = SOCK_MIN_SNDBUF;
 500                 else
 501                         sk->sk_sndbuf = val * 2;
 502
 503                 /*
 504                  *      Wake up sending tasks if we
 505                  *      upped the value.
 506                  */
 507                 sk->sk_write_space(sk);
 508                 break;
 509
 510         case SO_SNDBUFFORCE:
 511                 if (!capable(CAP_NET_ADMIN)) {
 512                         ret = -EPERM;
 513                         break;
 514                 }
 515                 goto set_sndbuf;
 516
 517         case SO_RCVBUF:
 518                 /* Don't error on this BSD doesn't and if you think
 519                    about it this is right. Otherwise apps have to
 520                    play 'guess the biggest size' games. RCVBUF/SNDBUF
 521                    are treated in BSD as hints */
 522
 523                 if (val > sysctl_rmem_max)
 524                         val = sysctl_rmem_max;
 525 set_rcvbuf:
 526                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
 527                 /*
 528                  * We double it on the way in to account for
 529                  * "struct sk_buff" etc. overhead.   Applications
 530                  * assume that the SO_RCVBUF setting they make will
 531                  * allow that much actual data to be received on that
 532                  * socket.
 533                  *
 534                  * Applications are unaware that "struct sk_buff" and
 535                  * other overheads allocate from the receive buffer
 536                  * during socket buffer allocation.
 537                  *
 538                  * And after considering the possible alternatives,
 539                  * returning the value we actually used in getsockopt
 540                  * is the most desirable behavior.
 541                  */
 542                 if ((val * 2) < SOCK_MIN_RCVBUF)
 543                         sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
 544                 else
 545                         sk->sk_rcvbuf = val * 2;
 546                 break;
 547
 548         case SO_RCVBUFFORCE:
 549                 if (!capable(CAP_NET_ADMIN)) {
 550                         ret = -EPERM;
 551                         break;
 552                 }
 553                 goto set_rcvbuf;
 554
 555         case SO_KEEPALIVE:
 556 #ifdef CONFIG_INET
 557                 if (sk->sk_protocol == IPPROTO_TCP)
 558                         tcp_set_keepalive(sk, valbool);
 559 #endif
 560                 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
 561                 break;
 562
 563         case SO_OOBINLINE:
 564                 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
 565                 break;
 566
 567         case SO_NO_CHECK:
 568                 sk->sk_no_check = valbool;
 569                 break;
 570
 571         case SO_PRIORITY:
 572                 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
 573                         sk->sk_priority = val;
 574                 else
 575                         ret = -EPERM;
 576                 break;
 577
 578         case SO_LINGER:
 579                 if (optlen < sizeof(ling)) {
 580                         ret = -EINVAL;  /* 1003.1g */
 581                         break;
 582                 }
 583                 if (copy_from_user(&ling,optval,sizeof(ling))) {
 584                         ret = -EFAULT;
 585                         break;
 586                 }
 587                 if (!ling.l_onoff)
 588                         sock_reset_flag(sk, SOCK_LINGER);
 589                 else {
 590 #if (BITS_PER_LONG == 32)
 591                         if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
 592                                 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
 593                         else
 594 #endif
 595                                 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
 596                         sock_set_flag(sk, SOCK_LINGER);
 597                 }
 598                 break;
 599
 600         case SO_BSDCOMPAT:
 601                 sock_warn_obsolete_bsdism("setsockopt");
 602                 break;
 603
 604         case SO_PASSCRED:
 605                 if (valbool)
 606                         set_bit(SOCK_PASSCRED, &sock->flags);
 607                 else
 608                         clear_bit(SOCK_PASSCRED, &sock->flags);
 609                 break;
 610
 611         case SO_TIMESTAMP:
 612         case SO_TIMESTAMPNS:
 613                 if (valbool)  {
 614                         if (optname == SO_TIMESTAMP)
 615                                 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
 616                         else
 617                                 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
 618                         sock_set_flag(sk, SOCK_RCVTSTAMP);
 619                         sock_enable_timestamp(sk);
 620                 } else {
 621                         sock_reset_flag(sk, SOCK_RCVTSTAMP);
 622                         sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
 623                 }
 624                 break;
 625
 626         case SO_RCVLOWAT:
 627                 if (val < 0)
 628                         val = INT_MAX;
 629                 sk->sk_rcvlowat = val ? : 1;
 630                 break;
 631
 632         case SO_RCVTIMEO:
 633                 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
 634                 break;
 635
 636         case SO_SNDTIMEO:
 637                 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
 638                 break;
 639
 640         case SO_ATTACH_FILTER:
 641                 ret = -EINVAL;
 642                 if (optlen == sizeof(struct sock_fprog)) {
 643                         struct sock_fprog fprog;
 644
 645                         ret = -EFAULT;
 646                         if (copy_from_user(&fprog, optval, sizeof(fprog)))
 647                                 break;
 648
 649                         ret = sk_attach_filter(&fprog, sk);
 650                 }
 651                 break;
 652
 653         case SO_DETACH_FILTER:
 654                 ret = sk_detach_filter(sk);
 655                 break;
 656
 657         case SO_PASSSEC:
 658                 if (valbool)
 659                         set_bit(SOCK_PASSSEC, &sock->flags);
 660                 else
 661                         clear_bit(SOCK_PASSSEC, &sock->flags);
 662                 break;
 663
 664                 /* We implement the SO_SNDLOWAT etc to
 665                    not be settable (1003.1g 5.3) */
 666         default:
 667                 ret = -ENOPROTOOPT;
 668                 break;
 669         }
 670         release_sock(sk);
 671         return ret;
 672 }
 673
 674
 675 int sock_getsockopt(struct socket *sock, int level, int optname,
 676                     char __user *optval, int __user *optlen)
 677 {
 678         struct sock *sk = sock->sk;
 679
 680         union {
 681                 int val;
 682                 struct linger ling;
 683                 struct timeval tm;
 684         } v;
 685
 686         unsigned int lv = sizeof(int);
 687         int len;
 688
 689         if (get_user(len, optlen))
 690                 return -EFAULT;
 691         if (len < 0)
 692                 return -EINVAL;
 693
 694         switch(optname) {
 695         case SO_DEBUG:
 696                 v.val = sock_flag(sk, SOCK_DBG);
 697                 break;
 698
 699         case SO_DONTROUTE:
 700                 v.val = sock_flag(sk, SOCK_LOCALROUTE);
 701                 break;
 702
 703         case SO_BROADCAST:
 704                 v.val = !!sock_flag(sk, SOCK_BROADCAST);
 705                 break;
 706
 707         case SO_SNDBUF:
 708                 v.val = sk->sk_sndbuf;
 709                 break;
 710
 711         case SO_RCVBUF:
 712                 v.val = sk->sk_rcvbuf;
 713                 break;
 714
 715         case SO_REUSEADDR:
 716                 v.val = sk->sk_reuse;
 717                 break;
 718
 719         case SO_KEEPALIVE:
 720                 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
 721                 break;
 722
 723         case SO_TYPE:
 724                 v.val = sk->sk_type;
 725                 break;
 726
 727         case SO_ERROR:
 728                 v.val = -sock_error(sk);
 729                 if (v.val==0)
 730                         v.val = xchg(&sk->sk_err_soft, 0);
 731                 break;
 732
 733         case SO_OOBINLINE:
 734                 v.val = !!sock_flag(sk, SOCK_URGINLINE);
 735                 break;
 736
 737         case SO_NO_CHECK:
 738                 v.val = sk->sk_no_check;
 739                 break;
 740
 741         case SO_PRIORITY:
 742                 v.val = sk->sk_priority;
 743                 break;
 744
 745         case SO_LINGER:
 746                 lv              = sizeof(v.ling);
 747                 v.ling.l_onoff  = !!sock_flag(sk, SOCK_LINGER);
 748                 v.ling.l_linger = sk->sk_lingertime / HZ;
 749                 break;
 750
 751         case SO_BSDCOMPAT:
 752                 sock_warn_obsolete_bsdism("getsockopt");
 753                 break;
 754
 755         case SO_TIMESTAMP:
 756                 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
 757                                 !sock_flag(sk, SOCK_RCVTSTAMPNS);
 758                 break;
 759
 760         case SO_TIMESTAMPNS:
 761                 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
 762                 break;
 763
 764         case SO_RCVTIMEO:
 765                 lv=sizeof(struct timeval);
 766                 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
 767                         v.tm.tv_sec = 0;
 768                         v.tm.tv_usec = 0;
 769                 } else {
 770                         v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
 771                         v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
 772                 }
 773                 break;
 774
 775         case SO_SNDTIMEO:
 776                 lv=sizeof(struct timeval);
 777                 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
 778                         v.tm.tv_sec = 0;
 779                         v.tm.tv_usec = 0;
 780                 } else {
 781                         v.tm.tv_sec = sk->sk_sndtimeo / HZ;
 782                         v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
 783                 }
 784                 break;
 785
 786         case SO_RCVLOWAT:
 787                 v.val = sk->sk_rcvlowat;
 788                 break;
 789
 790         case SO_SNDLOWAT:
 791                 v.val=1;
 792                 break;
 793
 794         case SO_PASSCRED:
 795                 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
 796                 break;
 797
 798         case SO_PEERCRED:
 799                 if (len > sizeof(sk->sk_peercred))
 800                         len = sizeof(sk->sk_peercred);
 801                 if (copy_to_user(optval, &sk->sk_peercred, len))
 802                         return -EFAULT;
 803                 goto lenout;
 804
 805         case SO_PEERNAME:
 806         {
 807                 char address[128];
 808
 809                 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
 810                         return -ENOTCONN;
 811                 if (lv < len)
 812                         return -EINVAL;
 813                 if (copy_to_user(optval, address, len))
 814                         return -EFAULT;
 815                 goto lenout;
 816         }
 817
 818         /* Dubious BSD thing... Probably nobody even uses it, but
 819          * the UNIX standard wants it for whatever reason... -DaveM
 820          */
 821         case SO_ACCEPTCONN:
 822                 v.val = sk->sk_state == TCP_LISTEN;
 823                 break;
 824
 825         case SO_PASSSEC:
 826                 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
 827                 break;
 828
 829         case SO_PEERSEC:
 830                 return security_socket_getpeersec_stream(sock, optval, optlen, len);
 831
 832         default:
 833                 return -ENOPROTOOPT;
 834         }
 835
 836         if (len > lv)
 837                 len = lv;
 838         if (copy_to_user(optval, &v, len))
 839                 return -EFAULT;
 840 lenout:
 841         if (put_user(len, optlen))
 842                 return -EFAULT;
 843         return 0;
 844 }
 845
 846 /*
 847  * Initialize an sk_lock.
 848  *
 849  * (We also register the sk_lock with the lock validator.)
 850  */
 851 static inline void sock_lock_init(struct sock *sk)
 852 {
 853         sock_lock_init_class_and_name(sk,
 854                         af_family_slock_key_strings[sk->sk_family],
 855                         af_family_slock_keys + sk->sk_family,
 856                         af_family_key_strings[sk->sk_family],
 857                         af_family_keys + sk->sk_family);
 858 }
 859
 860 static void sock_copy(struct sock *nsk, const struct sock *osk)
 861 {
 862 #ifdef CONFIG_SECURITY_NETWORK
 863         void *sptr = nsk->sk_security;
 864 #endif
 865
 866         memcpy(nsk, osk, osk->sk_prot->obj_size);
 867 #ifdef CONFIG_SECURITY_NETWORK
 868         nsk->sk_security = sptr;
 869         security_sk_clone(osk, nsk);
 870 #endif
 871 }
 872
 873 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority)
 874 {
 875         struct sock *sk;
 876         struct kmem_cache *slab;
 877
 878         slab = prot->slab;
 879         if (slab != NULL)
 880                 sk = kmem_cache_alloc(slab, priority);
 881         else
 882                 sk = kmalloc(prot->obj_size, priority);
 883
 884         return sk;
 885 }
 886
 887 static void sk_prot_free(struct proto *prot, struct sock *sk)
 888 {
 889         struct kmem_cache *slab;
 890
 891         slab = prot->slab;
 892         if (slab != NULL)
 893                 kmem_cache_free(slab, sk);
 894         else
 895                 kfree(sk);
 896 }
 897
 898 /**
 899  *      sk_alloc - All socket objects are allocated here
 900  *      @net: the applicable net namespace
 901  *      @family: protocol family
 902  *      @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
 903  *      @prot: struct proto associated with this new sock instance
 904  *      @zero_it: if we should zero the newly allocated sock
 905  */
 906 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
 907                       struct proto *prot, int zero_it)
 908 {
 909         struct sock *sk;
 910
 911         sk = sk_prot_alloc(prot, priority);
 912         if (sk) {
 913                 if (zero_it) {
 914                         memset(sk, 0, prot->obj_size);
 915                         sk->sk_family = family;
 916                         /*
 917                          * See comment in struct sock definition to understand
 918                          * why we need sk_prot_creator -acme
 919                          */
 920                         sk->sk_prot = sk->sk_prot_creator = prot;
 921                         sock_lock_init(sk);
 922                         sk->sk_net = get_net(net);
 923                 }
 924
 925                 if (security_sk_alloc(sk, family, priority))
 926                         goto out_free;
 927
 928                 if (!try_module_get(prot->owner))
 929                         goto out_free;
 930         }
 931         return sk;
 932
 933 out_free:
 934         sk_prot_free(prot, sk);
 935         return NULL;
 936 }
 937
 938 void sk_free(struct sock *sk)
 939 {
 940         struct sk_filter *filter;
 941         struct module *owner = sk->sk_prot_creator->owner;
 942
 943         if (sk->sk_destruct)
 944                 sk->sk_destruct(sk);
 945
 946         filter = rcu_dereference(sk->sk_filter);
 947         if (filter) {
 948                 sk_filter_uncharge(sk, filter);
 949                 rcu_assign_pointer(sk->sk_filter, NULL);
 950         }
 951
 952         sock_disable_timestamp(sk);
 953
 954         if (atomic_read(&sk->sk_omem_alloc))
 955                 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
 956                        __FUNCTION__, atomic_read(&sk->sk_omem_alloc));
 957
 958         security_sk_free(sk);
 959         put_net(sk->sk_net);
 960         sk_prot_free(sk->sk_prot_creator, sk);
 961         module_put(owner);
 962 }
 963
 964 struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
 965 {
 966         struct sock *newsk = sk_alloc(sk->sk_net, sk->sk_family, priority, sk->sk_prot, 0);
 967
 968         if (newsk != NULL) {
 969                 struct sk_filter *filter;
 970
 971                 sock_copy(newsk, sk);
 972
 973                 /* SANITY */
 974                 get_net(newsk->sk_net);
 975                 sk_node_init(&newsk->sk_node);
 976                 sock_lock_init(newsk);
 977                 bh_lock_sock(newsk);
 978                 newsk->sk_backlog.head  = newsk->sk_backlog.tail = NULL;
 979
 980                 atomic_set(&newsk->sk_rmem_alloc, 0);
 981                 atomic_set(&newsk->sk_wmem_alloc, 0);
 982                 atomic_set(&newsk->sk_omem_alloc, 0);
 983                 skb_queue_head_init(&newsk->sk_receive_queue);
 984                 skb_queue_head_init(&newsk->sk_write_queue);
 985 #ifdef CONFIG_NET_DMA
 986                 skb_queue_head_init(&newsk->sk_async_wait_queue);
 987 #endif
 988
 989                 rwlock_init(&newsk->sk_dst_lock);
 990                 rwlock_init(&newsk->sk_callback_lock);
 991                 lockdep_set_class_and_name(&newsk->sk_callback_lock,
 992                                 af_callback_keys + newsk->sk_family,
 993                                 af_family_clock_key_strings[newsk->sk_family]);
 994
 995                 newsk->sk_dst_cache     = NULL;
 996                 newsk->sk_wmem_queued   = 0;
 997                 newsk->sk_forward_alloc = 0;
 998                 newsk->sk_send_head     = NULL;
 999                 newsk->sk_userlocks     = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1000
1001                 sock_reset_flag(newsk, SOCK_DONE);
1002                 skb_queue_head_init(&newsk->sk_error_queue);
1003
1004                 filter = newsk->sk_filter;
1005                 if (filter != NULL)
1006                         sk_filter_charge(newsk, filter);
1007
1008                 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1009                         /* It is still raw copy of parent, so invalidate
1010                          * destructor and make plain sk_free() */
1011                         newsk->sk_destruct = NULL;
1012                         sk_free(newsk);
1013                         newsk = NULL;
1014                         goto out;
1015                 }
1016
1017                 newsk->sk_err      = 0;
1018                 newsk->sk_priority = 0;
1019                 atomic_set(&newsk->sk_refcnt, 2);
1020
1021                 /*
1022                  * Increment the counter in the same struct proto as the master
1023                  * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1024                  * is the same as sk->sk_prot->socks, as this field was copied
1025                  * with memcpy).
1026                  *
1027                  * This _changes_ the previous behaviour, where
1028                  * tcp_create_openreq_child always was incrementing the
1029                  * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1030                  * to be taken into account in all callers. -acme
1031                  */
1032                 sk_refcnt_debug_inc(newsk);
1033                 newsk->sk_socket = NULL;
1034                 newsk->sk_sleep  = NULL;
1035
1036                 if (newsk->sk_prot->sockets_allocated)
1037                         atomic_inc(newsk->sk_prot->sockets_allocated);
1038         }
1039 out:
1040         return newsk;
1041 }
1042
1043 EXPORT_SYMBOL_GPL(sk_clone);
1044
1045 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1046 {
1047         __sk_dst_set(sk, dst);
1048         sk->sk_route_caps = dst->dev->features;
1049         if (sk->sk_route_caps & NETIF_F_GSO)
1050                 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1051         if (sk_can_gso(sk)) {
1052                 if (dst->header_len)
1053                         sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1054                 else
1055                         sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1056         }
1057 }
1058 EXPORT_SYMBOL_GPL(sk_setup_caps);
1059
1060 void __init sk_init(void)
1061 {
1062         if (num_physpages <= 4096) {
1063                 sysctl_wmem_max = 32767;
1064                 sysctl_rmem_max = 32767;
1065                 sysctl_wmem_default = 32767;
1066                 sysctl_rmem_default = 32767;
1067         } else if (num_physpages >= 131072) {
1068                 sysctl_wmem_max = 131071;
1069                 sysctl_rmem_max = 131071;
1070         }
1071 }
1072
1073 /*
1074  *      Simple resource managers for sockets.
1075  */
1076
1077
1078 /*
1079  * Write buffer destructor automatically called from kfree_skb.
1080  */
1081 void sock_wfree(struct sk_buff *skb)
1082 {
1083         struct sock *sk = skb->sk;
1084
1085         /* In case it might be waiting for more memory. */
1086         atomic_sub(skb->truesize, &sk->sk_wmem_alloc);
1087         if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE))
1088                 sk->sk_write_space(sk);
1089         sock_put(sk);
1090 }
1091
1092 /*
1093  * Read buffer destructor automatically called from kfree_skb.
1094  */
1095 void sock_rfree(struct sk_buff *skb)
1096 {
1097         struct sock *sk = skb->sk;
1098
1099         atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1100 }
1101
1102
1103 int sock_i_uid(struct sock *sk)
1104 {
1105         int uid;
1106
1107         read_lock(&sk->sk_callback_lock);
1108         uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1109         read_unlock(&sk->sk_callback_lock);
1110         return uid;
1111 }
1112
1113 unsigned long sock_i_ino(struct sock *sk)
1114 {
1115         unsigned long ino;
1116
1117         read_lock(&sk->sk_callback_lock);
1118         ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1119         read_unlock(&sk->sk_callback_lock);
1120         return ino;
1121 }
1122
1123 /*
1124  * Allocate a skb from the socket's send buffer.
1125  */
1126 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1127                              gfp_t priority)
1128 {
1129         if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1130                 struct sk_buff * skb = alloc_skb(size, priority);
1131                 if (skb) {
1132                         skb_set_owner_w(skb, sk);
1133                         return skb;
1134                 }
1135         }
1136         return NULL;
1137 }
1138
1139 /*
1140  * Allocate a skb from the socket's receive buffer.
1141  */
1142 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1143                              gfp_t priority)
1144 {
1145         if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1146                 struct sk_buff *skb = alloc_skb(size, priority);
1147                 if (skb) {
1148                         skb_set_owner_r(skb, sk);
1149                         return skb;
1150                 }
1151         }
1152         return NULL;
1153 }
1154
1155 /*
1156  * Allocate a memory block from the socket's option memory buffer.
1157  */
1158 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1159 {
1160         if ((unsigned)size <= sysctl_optmem_max &&
1161             atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1162                 void *mem;
1163                 /* First do the add, to avoid the race if kmalloc
1164                  * might sleep.
1165                  */
1166                 atomic_add(size, &sk->sk_omem_alloc);
1167                 mem = kmalloc(size, priority);
1168                 if (mem)
1169                         return mem;
1170                 atomic_sub(size, &sk->sk_omem_alloc);
1171         }
1172         return NULL;
1173 }
1174
1175 /*
1176  * Free an option memory block.
1177  */
1178 void sock_kfree_s(struct sock *sk, void *mem, int size)
1179 {
1180         kfree(mem);
1181         atomic_sub(size, &sk->sk_omem_alloc);
1182 }
1183
1184 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1185    I think, these locks should be removed for datagram sockets.
1186  */
1187 static long sock_wait_for_wmem(struct sock * sk, long timeo)
1188 {
1189         DEFINE_WAIT(wait);
1190
1191         clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1192         for (;;) {
1193                 if (!timeo)
1194                         break;
1195                 if (signal_pending(current))
1196                         break;
1197                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1198                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1199                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1200                         break;
1201                 if (sk->sk_shutdown & SEND_SHUTDOWN)
1202                         break;
1203                 if (sk->sk_err)
1204                         break;
1205                 timeo = schedule_timeout(timeo);
1206         }
1207         finish_wait(sk->sk_sleep, &wait);
1208         return timeo;
1209 }
1210
1211
1212 /*
1213  *      Generic send/receive buffer handlers
1214  */
1215
1216 static struct sk_buff *sock_alloc_send_pskb(struct sock *sk,
1217                                             unsigned long header_len,
1218                                             unsigned long data_len,
1219                                             int noblock, int *errcode)
1220 {
1221         struct sk_buff *skb;
1222         gfp_t gfp_mask;
1223         long timeo;
1224         int err;
1225
1226         gfp_mask = sk->sk_allocation;
1227         if (gfp_mask & __GFP_WAIT)
1228                 gfp_mask |= __GFP_REPEAT;
1229
1230         timeo = sock_sndtimeo(sk, noblock);
1231         while (1) {
1232                 err = sock_error(sk);
1233                 if (err != 0)
1234                         goto failure;
1235
1236                 err = -EPIPE;
1237                 if (sk->sk_shutdown & SEND_SHUTDOWN)
1238                         goto failure;
1239
1240                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1241                         skb = alloc_skb(header_len, gfp_mask);
1242                         if (skb) {
1243                                 int npages;
1244                                 int i;
1245
1246                                 /* No pages, we're done... */
1247                                 if (!data_len)
1248                                         break;
1249
1250                                 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1251                                 skb->truesize += data_len;
1252                                 skb_shinfo(skb)->nr_frags = npages;
1253                                 for (i = 0; i < npages; i++) {
1254                                         struct page *page;
1255                                         skb_frag_t *frag;
1256
1257                                         page = alloc_pages(sk->sk_allocation, 0);
1258                                         if (!page) {
1259                                                 err = -ENOBUFS;
1260                                                 skb_shinfo(skb)->nr_frags = i;
1261                                                 kfree_skb(skb);
1262                                                 goto failure;
1263                                         }
1264
1265                                         frag = &skb_shinfo(skb)->frags[i];
1266                                         frag->page = page;
1267                                         frag->page_offset = 0;
1268                                         frag->size = (data_len >= PAGE_SIZE ?
1269                                                       PAGE_SIZE :
1270                                                       data_len);
1271                                         data_len -= PAGE_SIZE;
1272                                 }
1273
1274                                 /* Full success... */
1275                                 break;
1276                         }
1277                         err = -ENOBUFS;
1278                         goto failure;
1279                 }
1280                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1281                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1282                 err = -EAGAIN;
1283                 if (!timeo)
1284                         goto failure;
1285                 if (signal_pending(current))
1286                         goto interrupted;
1287                 timeo = sock_wait_for_wmem(sk, timeo);
1288         }
1289
1290         skb_set_owner_w(skb, sk);
1291         return skb;
1292
1293 interrupted:
1294         err = sock_intr_errno(timeo);
1295 failure:
1296         *errcode = err;
1297         return NULL;
1298 }
1299
1300 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1301                                     int noblock, int *errcode)
1302 {
1303         return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1304 }
1305
1306 static void __lock_sock(struct sock *sk)
1307 {
1308         DEFINE_WAIT(wait);
1309
1310         for (;;) {
1311                 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1312                                         TASK_UNINTERRUPTIBLE);
1313                 spin_unlock_bh(&sk->sk_lock.slock);
1314                 schedule();
1315                 spin_lock_bh(&sk->sk_lock.slock);
1316                 if (!sock_owned_by_user(sk))
1317                         break;
1318         }
1319         finish_wait(&sk->sk_lock.wq, &wait);
1320 }
1321
1322 static void __release_sock(struct sock *sk)
1323 {
1324         struct sk_buff *skb = sk->sk_backlog.head;
1325
1326         do {
1327                 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1328                 bh_unlock_sock(sk);
1329
1330                 do {
1331                         struct sk_buff *next = skb->next;
1332
1333                         skb->next = NULL;
1334                         sk->sk_backlog_rcv(sk, skb);
1335
1336                         /*
1337                          * We are in process context here with softirqs
1338                          * disabled, use cond_resched_softirq() to preempt.
1339                          * This is safe to do because we've taken the backlog
1340                          * queue private:
1341                          */
1342                         cond_resched_softirq();
1343
1344                         skb = next;
1345                 } while (skb != NULL);
1346
1347                 bh_lock_sock(sk);
1348         } while ((skb = sk->sk_backlog.head) != NULL);
1349 }
1350
1351 /**
1352  * sk_wait_data - wait for data to arrive at sk_receive_queue
1353  * @sk:    sock to wait on
1354  * @timeo: for how long
1355  *
1356  * Now socket state including sk->sk_err is changed only under lock,
1357  * hence we may omit checks after joining wait queue.
1358  * We check receive queue before schedule() only as optimization;
1359  * it is very likely that release_sock() added new data.
1360  */
1361 int sk_wait_data(struct sock *sk, long *timeo)
1362 {
1363         int rc;
1364         DEFINE_WAIT(wait);
1365
1366         prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1367         set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1368         rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1369         clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1370         finish_wait(sk->sk_sleep, &wait);
1371         return rc;
1372 }
1373
1374 EXPORT_SYMBOL(sk_wait_data);
1375
1376 /*
1377  * Set of default routines for initialising struct proto_ops when
1378  * the protocol does not support a particular function. In certain
1379  * cases where it makes no sense for a protocol to have a "do nothing"
1380  * function, some default processing is provided.
1381  */
1382
1383 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1384 {
1385         return -EOPNOTSUPP;
1386 }
1387
1388 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1389                     int len, int flags)
1390 {
1391         return -EOPNOTSUPP;
1392 }
1393
1394 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1395 {
1396         return -EOPNOTSUPP;
1397 }
1398
1399 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1400 {
1401         return -EOPNOTSUPP;
1402 }
1403
1404 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1405                     int *len, int peer)
1406 {
1407         return -EOPNOTSUPP;
1408 }
1409
1410 unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt)
1411 {
1412         return 0;
1413 }
1414
1415 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1416 {
1417         return -EOPNOTSUPP;
1418 }
1419
1420 int sock_no_listen(struct socket *sock, int backlog)
1421 {
1422         return -EOPNOTSUPP;
1423 }
1424
1425 int sock_no_shutdown(struct socket *sock, int how)
1426 {
1427         return -EOPNOTSUPP;
1428 }
1429
1430 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1431                     char __user *optval, int optlen)
1432 {
1433         return -EOPNOTSUPP;
1434 }
1435
1436 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1437                     char __user *optval, int __user *optlen)
1438 {
1439         return -EOPNOTSUPP;
1440 }
1441
1442 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1443                     size_t len)
1444 {
1445         return -EOPNOTSUPP;
1446 }
1447
1448 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1449                     size_t len, int flags)
1450 {
1451         return -EOPNOTSUPP;
1452 }
1453
1454 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1455 {
1456         /* Mirror missing mmap method error code */
1457         return -ENODEV;
1458 }
1459
1460 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1461 {
1462         ssize_t res;
1463         struct msghdr msg = {.msg_flags = flags};
1464         struct kvec iov;
1465         char *kaddr = kmap(page);
1466         iov.iov_base = kaddr + offset;
1467         iov.iov_len = size;
1468         res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1469         kunmap(page);
1470         return res;
1471 }
1472
1473 /*
1474  *      Default Socket Callbacks
1475  */
1476
1477 static void sock_def_wakeup(struct sock *sk)
1478 {
1479         read_lock(&sk->sk_callback_lock);
1480         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1481                 wake_up_interruptible_all(sk->sk_sleep);
1482         read_unlock(&sk->sk_callback_lock);
1483 }
1484
1485 static void sock_def_error_report(struct sock *sk)
1486 {
1487         read_lock(&sk->sk_callback_lock);
1488         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1489                 wake_up_interruptible(sk->sk_sleep);
1490         sk_wake_async(sk,0,POLL_ERR);
1491         read_unlock(&sk->sk_callback_lock);
1492 }
1493
1494 static void sock_def_readable(struct sock *sk, int len)
1495 {
1496         read_lock(&sk->sk_callback_lock);
1497         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1498                 wake_up_interruptible(sk->sk_sleep);
1499         sk_wake_async(sk,1,POLL_IN);
1500         read_unlock(&sk->sk_callback_lock);
1501 }
1502
1503 static void sock_def_write_space(struct sock *sk)
1504 {
1505         read_lock(&sk->sk_callback_lock);
1506
1507         /* Do not wake up a writer until he can make "significant"
1508          * progress.  --DaveM
1509          */
1510         if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1511                 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1512                         wake_up_interruptible(sk->sk_sleep);
1513
1514                 /* Should agree with poll, otherwise some programs break */
1515                 if (sock_writeable(sk))
1516                         sk_wake_async(sk, 2, POLL_OUT);
1517         }
1518
1519         read_unlock(&sk->sk_callback_lock);
1520 }
1521
1522 static void sock_def_destruct(struct sock *sk)
1523 {
1524         kfree(sk->sk_protinfo);
1525 }
1526
1527 void sk_send_sigurg(struct sock *sk)
1528 {
1529         if (sk->sk_socket && sk->sk_socket->file)
1530                 if (send_sigurg(&sk->sk_socket->file->f_owner))
1531                         sk_wake_async(sk, 3, POLL_PRI);
1532 }
1533
1534 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1535                     unsigned long expires)
1536 {
1537         if (!mod_timer(timer, expires))
1538                 sock_hold(sk);
1539 }
1540
1541 EXPORT_SYMBOL(sk_reset_timer);
1542
1543 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1544 {
1545         if (timer_pending(timer) && del_timer(timer))
1546                 __sock_put(sk);
1547 }
1548
1549 EXPORT_SYMBOL(sk_stop_timer);
1550
1551 void sock_init_data(struct socket *sock, struct sock *sk)
1552 {
1553         skb_queue_head_init(&sk->sk_receive_queue);
1554         skb_queue_head_init(&sk->sk_write_queue);
1555         skb_queue_head_init(&sk->sk_error_queue);
1556 #ifdef CONFIG_NET_DMA
1557         skb_queue_head_init(&sk->sk_async_wait_queue);
1558 #endif
1559
1560         sk->sk_send_head        =       NULL;
1561
1562         init_timer(&sk->sk_timer);
1563
1564         sk->sk_allocation       =       GFP_KERNEL;
1565         sk->sk_rcvbuf           =       sysctl_rmem_default;
1566         sk->sk_sndbuf           =       sysctl_wmem_default;
1567         sk->sk_state            =       TCP_CLOSE;
1568         sk->sk_socket           =       sock;
1569
1570         sock_set_flag(sk, SOCK_ZAPPED);
1571
1572         if (sock) {
1573                 sk->sk_type     =       sock->type;
1574                 sk->sk_sleep    =       &sock->wait;
1575                 sock->sk        =       sk;
1576         } else
1577                 sk->sk_sleep    =       NULL;
1578
1579         rwlock_init(&sk->sk_dst_lock);
1580         rwlock_init(&sk->sk_callback_lock);
1581         lockdep_set_class_and_name(&sk->sk_callback_lock,
1582                         af_callback_keys + sk->sk_family,
1583                         af_family_clock_key_strings[sk->sk_family]);
1584
1585         sk->sk_state_change     =       sock_def_wakeup;
1586         sk->sk_data_ready       =       sock_def_readable;
1587         sk->sk_write_space      =       sock_def_write_space;
1588         sk->sk_error_report     =       sock_def_error_report;
1589         sk->sk_destruct         =       sock_def_destruct;
1590
1591         sk->sk_sndmsg_page      =       NULL;
1592         sk->sk_sndmsg_off       =       0;
1593
1594         sk->sk_peercred.pid     =       0;
1595         sk->sk_peercred.uid     =       -1;
1596         sk->sk_peercred.gid     =       -1;
1597         sk->sk_write_pending    =       0;
1598         sk->sk_rcvlowat         =       1;
1599         sk->sk_rcvtimeo         =       MAX_SCHEDULE_TIMEOUT;
1600         sk->sk_sndtimeo         =       MAX_SCHEDULE_TIMEOUT;
1601
1602         sk->sk_stamp = ktime_set(-1L, -1L);
1603
1604         atomic_set(&sk->sk_refcnt, 1);
1605 }
1606
1607 void fastcall lock_sock_nested(struct sock *sk, int subclass)
1608 {
1609         might_sleep();
1610         spin_lock_bh(&sk->sk_lock.slock);
1611         if (sk->sk_lock.owned)
1612                 __lock_sock(sk);
1613         sk->sk_lock.owned = 1;
1614         spin_unlock(&sk->sk_lock.slock);
1615         /*
1616          * The sk_lock has mutex_lock() semantics here:
1617          */
1618         mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
1619         local_bh_enable();
1620 }
1621
1622 EXPORT_SYMBOL(lock_sock_nested);
1623
1624 void fastcall release_sock(struct sock *sk)
1625 {
1626         /*
1627          * The sk_lock has mutex_unlock() semantics:
1628          */
1629         mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1630
1631         spin_lock_bh(&sk->sk_lock.slock);
1632         if (sk->sk_backlog.tail)
1633                 __release_sock(sk);
1634         sk->sk_lock.owned = 0;
1635         if (waitqueue_active(&sk->sk_lock.wq))
1636                 wake_up(&sk->sk_lock.wq);
1637         spin_unlock_bh(&sk->sk_lock.slock);
1638 }
1639 EXPORT_SYMBOL(release_sock);
1640
1641 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1642 {
1643         struct timeval tv;
1644         if (!sock_flag(sk, SOCK_TIMESTAMP))
1645                 sock_enable_timestamp(sk);
1646         tv = ktime_to_timeval(sk->sk_stamp);
1647         if (tv.tv_sec == -1)
1648                 return -ENOENT;
1649         if (tv.tv_sec == 0) {
1650                 sk->sk_stamp = ktime_get_real();
1651                 tv = ktime_to_timeval(sk->sk_stamp);
1652         }
1653         return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
1654 }
1655 EXPORT_SYMBOL(sock_get_timestamp);
1656
1657 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
1658 {
1659         struct timespec ts;
1660         if (!sock_flag(sk, SOCK_TIMESTAMP))
1661                 sock_enable_timestamp(sk);
1662         ts = ktime_to_timespec(sk->sk_stamp);
1663         if (ts.tv_sec == -1)
1664                 return -ENOENT;
1665         if (ts.tv_sec == 0) {
1666                 sk->sk_stamp = ktime_get_real();
1667                 ts = ktime_to_timespec(sk->sk_stamp);
1668         }
1669         return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
1670 }
1671 EXPORT_SYMBOL(sock_get_timestampns);
1672
1673 void sock_enable_timestamp(struct sock *sk)
1674 {
1675         if (!sock_flag(sk, SOCK_TIMESTAMP)) {
1676                 sock_set_flag(sk, SOCK_TIMESTAMP);
1677                 net_enable_timestamp();
1678         }
1679 }
1680
1681 /*
1682  *      Get a socket option on an socket.
1683  *
1684  *      FIX: POSIX 1003.1g is very ambiguous here. It states that
1685  *      asynchronous errors should be reported by getsockopt. We assume
1686  *      this means if you specify SO_ERROR (otherwise whats the point of it).
1687  */
1688 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1689                            char __user *optval, int __user *optlen)
1690 {
1691         struct sock *sk = sock->sk;
1692
1693         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1694 }
1695
1696 EXPORT_SYMBOL(sock_common_getsockopt);
1697
1698 #ifdef CONFIG_COMPAT
1699 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
1700                                   char __user *optval, int __user *optlen)
1701 {
1702         struct sock *sk = sock->sk;
1703
1704         if (sk->sk_prot->compat_getsockopt != NULL)
1705                 return sk->sk_prot->compat_getsockopt(sk, level, optname,
1706                                                       optval, optlen);
1707         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1708 }
1709 EXPORT_SYMBOL(compat_sock_common_getsockopt);
1710 #endif
1711
1712 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1713                         struct msghdr *msg, size_t size, int flags)
1714 {
1715         struct sock *sk = sock->sk;
1716         int addr_len = 0;
1717         int err;
1718
1719         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
1720                                    flags & ~MSG_DONTWAIT, &addr_len);
1721         if (err >= 0)
1722                 msg->msg_namelen = addr_len;
1723         return err;
1724 }
1725
1726 EXPORT_SYMBOL(sock_common_recvmsg);
1727
1728 /*
1729  *      Set socket options on an inet socket.
1730  */
1731 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1732                            char __user *optval, int optlen)
1733 {
1734         struct sock *sk = sock->sk;
1735
1736         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1737 }
1738
1739 EXPORT_SYMBOL(sock_common_setsockopt);
1740
1741 #ifdef CONFIG_COMPAT
1742 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
1743                                   char __user *optval, int optlen)
1744 {
1745         struct sock *sk = sock->sk;
1746
1747         if (sk->sk_prot->compat_setsockopt != NULL)
1748                 return sk->sk_prot->compat_setsockopt(sk, level, optname,
1749                                                       optval, optlen);
1750         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1751 }
1752 EXPORT_SYMBOL(compat_sock_common_setsockopt);
1753 #endif
1754
1755 void sk_common_release(struct sock *sk)
1756 {
1757         if (sk->sk_prot->destroy)
1758                 sk->sk_prot->destroy(sk);
1759
1760         /*
1761          * Observation: when sock_common_release is called, processes have
1762          * no access to socket. But net still has.
1763          * Step one, detach it from networking:
1764          *
1765          * A. Remove from hash tables.
1766          */
1767
1768         sk->sk_prot->unhash(sk);
1769
1770         /*
1771          * In this point socket cannot receive new packets, but it is possible
1772          * that some packets are in flight because some CPU runs receiver and
1773          * did hash table lookup before we unhashed socket. They will achieve
1774          * receive queue and will be purged by socket destructor.
1775          *
1776          * Also we still have packets pending on receive queue and probably,
1777          * our own packets waiting in device queues. sock_destroy will drain
1778          * receive queue, but transmitted packets will delay socket destruction
1779          * until the last reference will be released.
1780          */
1781
1782         sock_orphan(sk);
1783
1784         xfrm_sk_free_policy(sk);
1785
1786         sk_refcnt_debug_release(sk);
1787         sock_put(sk);
1788 }
1789
1790 EXPORT_SYMBOL(sk_common_release);
1791
1792 static DEFINE_RWLOCK(proto_list_lock);
1793 static LIST_HEAD(proto_list);
1794
1795 int proto_register(struct proto *prot, int alloc_slab)
1796 {
1797         char *request_sock_slab_name = NULL;
1798         char *timewait_sock_slab_name;
1799         int rc = -ENOBUFS;
1800
1801         if (alloc_slab) {
1802                 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
1803                                                SLAB_HWCACHE_ALIGN, NULL);
1804
1805                 if (prot->slab == NULL) {
1806                         printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
1807                                prot->name);
1808                         goto out;
1809                 }
1810
1811                 if (prot->rsk_prot != NULL) {
1812                         static const char mask[] = "request_sock_%s";
1813
1814                         request_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
1815                         if (request_sock_slab_name == NULL)
1816                                 goto out_free_sock_slab;
1817
1818                         sprintf(request_sock_slab_name, mask, prot->name);
1819                         prot->rsk_prot->slab = kmem_cache_create(request_sock_slab_name,
1820                                                                  prot->rsk_prot->obj_size, 0,
1821                                                                  SLAB_HWCACHE_ALIGN, NULL);
1822
1823                         if (prot->rsk_prot->slab == NULL) {
1824                                 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
1825                                        prot->name);
1826                                 goto out_free_request_sock_slab_name;
1827                         }
1828                 }
1829
1830                 if (prot->twsk_prot != NULL) {
1831                         static const char mask[] = "tw_sock_%s";
1832
1833                         timewait_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
1834
1835                         if (timewait_sock_slab_name == NULL)
1836                                 goto out_free_request_sock_slab;
1837
1838                         sprintf(timewait_sock_slab_name, mask, prot->name);
1839                         prot->twsk_prot->twsk_slab =
1840                                 kmem_cache_create(timewait_sock_slab_name,
1841                                                   prot->twsk_prot->twsk_obj_size,
1842                                                   0, SLAB_HWCACHE_ALIGN,
1843                                                   NULL);
1844                         if (prot->twsk_prot->twsk_slab == NULL)
1845                                 goto out_free_timewait_sock_slab_name;
1846                 }
1847         }
1848
1849         write_lock(&proto_list_lock);
1850         list_add(&prot->node, &proto_list);
1851         write_unlock(&proto_list_lock);
1852         rc = 0;
1853 out:
1854         return rc;
1855 out_free_timewait_sock_slab_name:
1856         kfree(timewait_sock_slab_name);
1857 out_free_request_sock_slab:
1858         if (prot->rsk_prot && prot->rsk_prot->slab) {
1859                 kmem_cache_destroy(prot->rsk_prot->slab);
1860                 prot->rsk_prot->slab = NULL;
1861         }
1862 out_free_request_sock_slab_name:
1863         kfree(request_sock_slab_name);
1864 out_free_sock_slab:
1865         kmem_cache_destroy(prot->slab);
1866         prot->slab = NULL;
1867         goto out;
1868 }
1869
1870 EXPORT_SYMBOL(proto_register);
1871
1872 void proto_unregister(struct proto *prot)
1873 {
1874         write_lock(&proto_list_lock);
1875         list_del(&prot->node);
1876         write_unlock(&proto_list_lock);
1877
1878         if (prot->slab != NULL) {
1879                 kmem_cache_destroy(prot->slab);
1880                 prot->slab = NULL;
1881         }
1882
1883         if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
1884                 const char *name = kmem_cache_name(prot->rsk_prot->slab);
1885
1886                 kmem_cache_destroy(prot->rsk_prot->slab);
1887                 kfree(name);
1888                 prot->rsk_prot->slab = NULL;
1889         }
1890
1891         if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
1892                 const char *name = kmem_cache_name(prot->twsk_prot->twsk_slab);
1893
1894                 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
1895                 kfree(name);
1896                 prot->twsk_prot->twsk_slab = NULL;
1897         }
1898 }
1899
1900 EXPORT_SYMBOL(proto_unregister);
1901
1902 #ifdef CONFIG_PROC_FS
1903 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
1904 {
1905         read_lock(&proto_list_lock);
1906         return seq_list_start_head(&proto_list, *pos);
1907 }
1908
1909 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1910 {
1911         return seq_list_next(v, &proto_list, pos);
1912 }
1913
1914 static void proto_seq_stop(struct seq_file *seq, void *v)
1915 {
1916         read_unlock(&proto_list_lock);
1917 }
1918
1919 static char proto_method_implemented(const void *method)
1920 {
1921         return method == NULL ? 'n' : 'y';
1922 }
1923
1924 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
1925 {
1926         seq_printf(seq, "%-9s %4u %6d  %6d   %-3s %6u   %-3s  %-10s "
1927                         "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
1928                    proto->name,
1929                    proto->obj_size,
1930                    proto->sockets_allocated != NULL ? atomic_read(proto->sockets_allocated) : -1,
1931                    proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
1932                    proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
1933                    proto->max_header,
1934                    proto->slab == NULL ? "no" : "yes",
1935                    module_name(proto->owner),
1936                    proto_method_implemented(proto->close),
1937                    proto_method_implemented(proto->connect),
1938                    proto_method_implemented(proto->disconnect),
1939                    proto_method_implemented(proto->accept),
1940                    proto_method_implemented(proto->ioctl),
1941                    proto_method_implemented(proto->init),
1942                    proto_method_implemented(proto->destroy),
1943                    proto_method_implemented(proto->shutdown),
1944                    proto_method_implemented(proto->setsockopt),
1945                    proto_method_implemented(proto->getsockopt),
1946                    proto_method_implemented(proto->sendmsg),
1947                    proto_method_implemented(proto->recvmsg),
1948                    proto_method_implemented(proto->sendpage),
1949                    proto_method_implemented(proto->bind),
1950                    proto_method_implemented(proto->backlog_rcv),
1951                    proto_method_implemented(proto->hash),
1952                    proto_method_implemented(proto->unhash),
1953                    proto_method_implemented(proto->get_port),
1954                    proto_method_implemented(proto->enter_memory_pressure));
1955 }
1956
1957 static int proto_seq_show(struct seq_file *seq, void *v)
1958 {
1959         if (v == &proto_list)
1960                 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
1961                            "protocol",
1962                            "size",
1963                            "sockets",
1964                            "memory",
1965                            "press",
1966                            "maxhdr",
1967                            "slab",
1968                            "module",
1969                            "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
1970         else
1971                 proto_seq_printf(seq, list_entry(v, struct proto, node));
1972         return 0;
1973 }
1974
1975 static const struct seq_operations proto_seq_ops = {
1976         .start  = proto_seq_start,
1977         .next   = proto_seq_next,
1978         .stop   = proto_seq_stop,
1979         .show   = proto_seq_show,
1980 };
1981
1982 static int proto_seq_open(struct inode *inode, struct file *file)
1983 {
1984         return seq_open(file, &proto_seq_ops);
1985 }
1986
1987 static const struct file_operations proto_seq_fops = {
1988         .owner          = THIS_MODULE,
1989         .open           = proto_seq_open,
1990         .read           = seq_read,
1991         .llseek         = seq_lseek,
1992         .release        = seq_release,
1993 };
1994
1995 static int __init proto_init(void)
1996 {
1997         /* register /proc/net/protocols */
1998         return proc_net_fops_create(&init_net, "protocols", S_IRUGO, &proto_seq_fops) == NULL ? -ENOBUFS : 0;
1999 }
2000
2001 subsys_initcall(proto_init);
2002
2003 #endif /* PROC_FS */
2004
2005 EXPORT_SYMBOL(sk_alloc);
2006 EXPORT_SYMBOL(sk_free);
2007 EXPORT_SYMBOL(sk_send_sigurg);
2008 EXPORT_SYMBOL(sock_alloc_send_skb);
2009 EXPORT_SYMBOL(sock_init_data);
2010 EXPORT_SYMBOL(sock_kfree_s);
2011 EXPORT_SYMBOL(sock_kmalloc);
2012 EXPORT_SYMBOL(sock_no_accept);
2013 EXPORT_SYMBOL(sock_no_bind);
2014 EXPORT_SYMBOL(sock_no_connect);
2015 EXPORT_SYMBOL(sock_no_getname);
2016 EXPORT_SYMBOL(sock_no_getsockopt);
2017 EXPORT_SYMBOL(sock_no_ioctl);
2018 EXPORT_SYMBOL(sock_no_listen);
2019 EXPORT_SYMBOL(sock_no_mmap);
2020 EXPORT_SYMBOL(sock_no_poll);
2021 EXPORT_SYMBOL(sock_no_recvmsg);
2022 EXPORT_SYMBOL(sock_no_sendmsg);
2023 EXPORT_SYMBOL(sock_no_sendpage);
2024 EXPORT_SYMBOL(sock_no_setsockopt);
2025 EXPORT_SYMBOL(sock_no_shutdown);
2026 EXPORT_SYMBOL(sock_no_socketpair);
2027 EXPORT_SYMBOL(sock_rfree);
2028 EXPORT_SYMBOL(sock_setsockopt);
2029 EXPORT_SYMBOL(sock_wfree);
2030 EXPORT_SYMBOL(sock_wmalloc);
2031 EXPORT_SYMBOL(sock_i_uid);
2032 EXPORT_SYMBOL(sock_i_ino);
2033 EXPORT_SYMBOL(sysctl_optmem_max);
2034 #ifdef CONFIG_SYSCTL
2035 EXPORT_SYMBOL(sysctl_rmem_max);
2036 EXPORT_SYMBOL(sysctl_wmem_max);
2037 #endif