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                 int family)
 875 {
 876         struct sock *sk;
 877         struct kmem_cache *slab;
 878
 879         slab = prot->slab;
 880         if (slab != NULL)
 881                 sk = kmem_cache_alloc(slab, priority);
 882         else
 883                 sk = kmalloc(prot->obj_size, priority);
 884
 885         if (sk != NULL) {
 886                 if (security_sk_alloc(sk, family, priority))
 887                         goto out_free;
 888
 889                 if (!try_module_get(prot->owner))
 890                         goto out_free_sec;
 891         }
 892
 893         return sk;
 894
 895 out_free_sec:
 896         security_sk_free(sk);
 897 out_free:
 898         if (slab != NULL)
 899                 kmem_cache_free(slab, sk);
 900         else
 901                 kfree(sk);
 902         return NULL;
 903 }
 904
 905 static void sk_prot_free(struct proto *prot, struct sock *sk)
 906 {
 907         struct kmem_cache *slab;
 908         struct module *owner;
 909
 910         owner = prot->owner;
 911         slab = prot->slab;
 912
 913         security_sk_free(sk);
 914         if (slab != NULL)
 915                 kmem_cache_free(slab, sk);
 916         else
 917                 kfree(sk);
 918         module_put(owner);
 919 }
 920
 921 /**
 922  *      sk_alloc - All socket objects are allocated here
 923  *      @net: the applicable net namespace
 924  *      @family: protocol family
 925  *      @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
 926  *      @prot: struct proto associated with this new sock instance
 927  *      @zero_it: if we should zero the newly allocated sock
 928  */
 929 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
 930                       struct proto *prot, int zero_it)
 931 {
 932         struct sock *sk;
 933
 934         if (zero_it)
 935                 priority |= __GFP_ZERO;
 936
 937         sk = sk_prot_alloc(prot, priority, family);
 938         if (sk) {
 939                 if (zero_it) {
 940                         sk->sk_family = family;
 941                         /*
 942                          * See comment in struct sock definition to understand
 943                          * why we need sk_prot_creator -acme
 944                          */
 945                         sk->sk_prot = sk->sk_prot_creator = prot;
 946                         sock_lock_init(sk);
 947                         sk->sk_net = get_net(net);
 948                 }
 949         }
 950
 951         return sk;
 952 }
 953
 954 void sk_free(struct sock *sk)
 955 {
 956         struct sk_filter *filter;
 957
 958         if (sk->sk_destruct)
 959                 sk->sk_destruct(sk);
 960
 961         filter = rcu_dereference(sk->sk_filter);
 962         if (filter) {
 963                 sk_filter_uncharge(sk, filter);
 964                 rcu_assign_pointer(sk->sk_filter, NULL);
 965         }
 966
 967         sock_disable_timestamp(sk);
 968
 969         if (atomic_read(&sk->sk_omem_alloc))
 970                 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
 971                        __FUNCTION__, atomic_read(&sk->sk_omem_alloc));
 972
 973         put_net(sk->sk_net);
 974         sk_prot_free(sk->sk_prot_creator, sk);
 975 }
 976
 977 struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
 978 {
 979         struct sock *newsk;
 980
 981         newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
 982         if (newsk != NULL) {
 983                 struct sk_filter *filter;
 984
 985                 sock_copy(newsk, sk);
 986
 987                 /* SANITY */
 988                 get_net(newsk->sk_net);
 989                 sk_node_init(&newsk->sk_node);
 990                 sock_lock_init(newsk);
 991                 bh_lock_sock(newsk);
 992                 newsk->sk_backlog.head  = newsk->sk_backlog.tail = NULL;
 993
 994                 atomic_set(&newsk->sk_rmem_alloc, 0);
 995                 atomic_set(&newsk->sk_wmem_alloc, 0);
 996                 atomic_set(&newsk->sk_omem_alloc, 0);
 997                 skb_queue_head_init(&newsk->sk_receive_queue);
 998                 skb_queue_head_init(&newsk->sk_write_queue);
 999 #ifdef CONFIG_NET_DMA
1000                 skb_queue_head_init(&newsk->sk_async_wait_queue);
1001 #endif
1002
1003                 rwlock_init(&newsk->sk_dst_lock);
1004                 rwlock_init(&newsk->sk_callback_lock);
1005                 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1006                                 af_callback_keys + newsk->sk_family,
1007                                 af_family_clock_key_strings[newsk->sk_family]);
1008
1009                 newsk->sk_dst_cache     = NULL;
1010                 newsk->sk_wmem_queued   = 0;
1011                 newsk->sk_forward_alloc = 0;
1012                 newsk->sk_send_head     = NULL;
1013                 newsk->sk_userlocks     = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1014
1015                 sock_reset_flag(newsk, SOCK_DONE);
1016                 skb_queue_head_init(&newsk->sk_error_queue);
1017
1018                 filter = newsk->sk_filter;
1019                 if (filter != NULL)
1020                         sk_filter_charge(newsk, filter);
1021
1022                 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1023                         /* It is still raw copy of parent, so invalidate
1024                          * destructor and make plain sk_free() */
1025                         newsk->sk_destruct = NULL;
1026                         sk_free(newsk);
1027                         newsk = NULL;
1028                         goto out;
1029                 }
1030
1031                 newsk->sk_err      = 0;
1032                 newsk->sk_priority = 0;
1033                 atomic_set(&newsk->sk_refcnt, 2);
1034
1035                 /*
1036                  * Increment the counter in the same struct proto as the master
1037                  * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1038                  * is the same as sk->sk_prot->socks, as this field was copied
1039                  * with memcpy).
1040                  *
1041                  * This _changes_ the previous behaviour, where
1042                  * tcp_create_openreq_child always was incrementing the
1043                  * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1044                  * to be taken into account in all callers. -acme
1045                  */
1046                 sk_refcnt_debug_inc(newsk);
1047                 newsk->sk_socket = NULL;
1048                 newsk->sk_sleep  = NULL;
1049
1050                 if (newsk->sk_prot->sockets_allocated)
1051                         atomic_inc(newsk->sk_prot->sockets_allocated);
1052         }
1053 out:
1054         return newsk;
1055 }
1056
1057 EXPORT_SYMBOL_GPL(sk_clone);
1058
1059 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1060 {
1061         __sk_dst_set(sk, dst);
1062         sk->sk_route_caps = dst->dev->features;
1063         if (sk->sk_route_caps & NETIF_F_GSO)
1064                 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1065         if (sk_can_gso(sk)) {
1066                 if (dst->header_len)
1067                         sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1068                 else
1069                         sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1070         }
1071 }
1072 EXPORT_SYMBOL_GPL(sk_setup_caps);
1073
1074 void __init sk_init(void)
1075 {
1076         if (num_physpages <= 4096) {
1077                 sysctl_wmem_max = 32767;
1078                 sysctl_rmem_max = 32767;
1079                 sysctl_wmem_default = 32767;
1080                 sysctl_rmem_default = 32767;
1081         } else if (num_physpages >= 131072) {
1082                 sysctl_wmem_max = 131071;
1083                 sysctl_rmem_max = 131071;
1084         }
1085 }
1086
1087 /*
1088  *      Simple resource managers for sockets.
1089  */
1090
1091
1092 /*
1093  * Write buffer destructor automatically called from kfree_skb.
1094  */
1095 void sock_wfree(struct sk_buff *skb)
1096 {
1097         struct sock *sk = skb->sk;
1098
1099         /* In case it might be waiting for more memory. */
1100         atomic_sub(skb->truesize, &sk->sk_wmem_alloc);
1101         if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE))
1102                 sk->sk_write_space(sk);
1103         sock_put(sk);
1104 }
1105
1106 /*
1107  * Read buffer destructor automatically called from kfree_skb.
1108  */
1109 void sock_rfree(struct sk_buff *skb)
1110 {
1111         struct sock *sk = skb->sk;
1112
1113         atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1114 }
1115
1116
1117 int sock_i_uid(struct sock *sk)
1118 {
1119         int uid;
1120
1121         read_lock(&sk->sk_callback_lock);
1122         uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1123         read_unlock(&sk->sk_callback_lock);
1124         return uid;
1125 }
1126
1127 unsigned long sock_i_ino(struct sock *sk)
1128 {
1129         unsigned long ino;
1130
1131         read_lock(&sk->sk_callback_lock);
1132         ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1133         read_unlock(&sk->sk_callback_lock);
1134         return ino;
1135 }
1136
1137 /*
1138  * Allocate a skb from the socket's send buffer.
1139  */
1140 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1141                              gfp_t priority)
1142 {
1143         if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1144                 struct sk_buff * skb = alloc_skb(size, priority);
1145                 if (skb) {
1146                         skb_set_owner_w(skb, sk);
1147                         return skb;
1148                 }
1149         }
1150         return NULL;
1151 }
1152
1153 /*
1154  * Allocate a skb from the socket's receive buffer.
1155  */
1156 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1157                              gfp_t priority)
1158 {
1159         if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1160                 struct sk_buff *skb = alloc_skb(size, priority);
1161                 if (skb) {
1162                         skb_set_owner_r(skb, sk);
1163                         return skb;
1164                 }
1165         }
1166         return NULL;
1167 }
1168
1169 /*
1170  * Allocate a memory block from the socket's option memory buffer.
1171  */
1172 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1173 {
1174         if ((unsigned)size <= sysctl_optmem_max &&
1175             atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1176                 void *mem;
1177                 /* First do the add, to avoid the race if kmalloc
1178                  * might sleep.
1179                  */
1180                 atomic_add(size, &sk->sk_omem_alloc);
1181                 mem = kmalloc(size, priority);
1182                 if (mem)
1183                         return mem;
1184                 atomic_sub(size, &sk->sk_omem_alloc);
1185         }
1186         return NULL;
1187 }
1188
1189 /*
1190  * Free an option memory block.
1191  */
1192 void sock_kfree_s(struct sock *sk, void *mem, int size)
1193 {
1194         kfree(mem);
1195         atomic_sub(size, &sk->sk_omem_alloc);
1196 }
1197
1198 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1199    I think, these locks should be removed for datagram sockets.
1200  */
1201 static long sock_wait_for_wmem(struct sock * sk, long timeo)
1202 {
1203         DEFINE_WAIT(wait);
1204
1205         clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1206         for (;;) {
1207                 if (!timeo)
1208                         break;
1209                 if (signal_pending(current))
1210                         break;
1211                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1212                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1213                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1214                         break;
1215                 if (sk->sk_shutdown & SEND_SHUTDOWN)
1216                         break;
1217                 if (sk->sk_err)
1218                         break;
1219                 timeo = schedule_timeout(timeo);
1220         }
1221         finish_wait(sk->sk_sleep, &wait);
1222         return timeo;
1223 }
1224
1225
1226 /*
1227  *      Generic send/receive buffer handlers
1228  */
1229
1230 static struct sk_buff *sock_alloc_send_pskb(struct sock *sk,
1231                                             unsigned long header_len,
1232                                             unsigned long data_len,
1233                                             int noblock, int *errcode)
1234 {
1235         struct sk_buff *skb;
1236         gfp_t gfp_mask;
1237         long timeo;
1238         int err;
1239
1240         gfp_mask = sk->sk_allocation;
1241         if (gfp_mask & __GFP_WAIT)
1242                 gfp_mask |= __GFP_REPEAT;
1243
1244         timeo = sock_sndtimeo(sk, noblock);
1245         while (1) {
1246                 err = sock_error(sk);
1247                 if (err != 0)
1248                         goto failure;
1249
1250                 err = -EPIPE;
1251                 if (sk->sk_shutdown & SEND_SHUTDOWN)
1252                         goto failure;
1253
1254                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1255                         skb = alloc_skb(header_len, gfp_mask);
1256                         if (skb) {
1257                                 int npages;
1258                                 int i;
1259
1260                                 /* No pages, we're done... */
1261                                 if (!data_len)
1262                                         break;
1263
1264                                 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1265                                 skb->truesize += data_len;
1266                                 skb_shinfo(skb)->nr_frags = npages;
1267                                 for (i = 0; i < npages; i++) {
1268                                         struct page *page;
1269                                         skb_frag_t *frag;
1270
1271                                         page = alloc_pages(sk->sk_allocation, 0);
1272                                         if (!page) {
1273                                                 err = -ENOBUFS;
1274                                                 skb_shinfo(skb)->nr_frags = i;
1275                                                 kfree_skb(skb);
1276                                                 goto failure;
1277                                         }
1278
1279                                         frag = &skb_shinfo(skb)->frags[i];
1280                                         frag->page = page;
1281                                         frag->page_offset = 0;
1282                                         frag->size = (data_len >= PAGE_SIZE ?
1283                                                       PAGE_SIZE :
1284                                                       data_len);
1285                                         data_len -= PAGE_SIZE;
1286                                 }
1287
1288                                 /* Full success... */
1289                                 break;
1290                         }
1291                         err = -ENOBUFS;
1292                         goto failure;
1293                 }
1294                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1295                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1296                 err = -EAGAIN;
1297                 if (!timeo)
1298                         goto failure;
1299                 if (signal_pending(current))
1300                         goto interrupted;
1301                 timeo = sock_wait_for_wmem(sk, timeo);
1302         }
1303
1304         skb_set_owner_w(skb, sk);
1305         return skb;
1306
1307 interrupted:
1308         err = sock_intr_errno(timeo);
1309 failure:
1310         *errcode = err;
1311         return NULL;
1312 }
1313
1314 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1315                                     int noblock, int *errcode)
1316 {
1317         return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1318 }
1319
1320 static void __lock_sock(struct sock *sk)
1321 {
1322         DEFINE_WAIT(wait);
1323
1324         for (;;) {
1325                 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1326                                         TASK_UNINTERRUPTIBLE);
1327                 spin_unlock_bh(&sk->sk_lock.slock);
1328                 schedule();
1329                 spin_lock_bh(&sk->sk_lock.slock);
1330                 if (!sock_owned_by_user(sk))
1331                         break;
1332         }
1333         finish_wait(&sk->sk_lock.wq, &wait);
1334 }
1335
1336 static void __release_sock(struct sock *sk)
1337 {
1338         struct sk_buff *skb = sk->sk_backlog.head;
1339
1340         do {
1341                 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1342                 bh_unlock_sock(sk);
1343
1344                 do {
1345                         struct sk_buff *next = skb->next;
1346
1347                         skb->next = NULL;
1348                         sk->sk_backlog_rcv(sk, skb);
1349
1350                         /*
1351                          * We are in process context here with softirqs
1352                          * disabled, use cond_resched_softirq() to preempt.
1353                          * This is safe to do because we've taken the backlog
1354                          * queue private:
1355                          */
1356                         cond_resched_softirq();
1357
1358                         skb = next;
1359                 } while (skb != NULL);
1360
1361                 bh_lock_sock(sk);
1362         } while ((skb = sk->sk_backlog.head) != NULL);
1363 }
1364
1365 /**
1366  * sk_wait_data - wait for data to arrive at sk_receive_queue
1367  * @sk:    sock to wait on
1368  * @timeo: for how long
1369  *
1370  * Now socket state including sk->sk_err is changed only under lock,
1371  * hence we may omit checks after joining wait queue.
1372  * We check receive queue before schedule() only as optimization;
1373  * it is very likely that release_sock() added new data.
1374  */
1375 int sk_wait_data(struct sock *sk, long *timeo)
1376 {
1377         int rc;
1378         DEFINE_WAIT(wait);
1379
1380         prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1381         set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1382         rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1383         clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1384         finish_wait(sk->sk_sleep, &wait);
1385         return rc;
1386 }
1387
1388 EXPORT_SYMBOL(sk_wait_data);
1389
1390 /*
1391  * Set of default routines for initialising struct proto_ops when
1392  * the protocol does not support a particular function. In certain
1393  * cases where it makes no sense for a protocol to have a "do nothing"
1394  * function, some default processing is provided.
1395  */
1396
1397 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1398 {
1399         return -EOPNOTSUPP;
1400 }
1401
1402 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1403                     int len, int flags)
1404 {
1405         return -EOPNOTSUPP;
1406 }
1407
1408 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1409 {
1410         return -EOPNOTSUPP;
1411 }
1412
1413 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1414 {
1415         return -EOPNOTSUPP;
1416 }
1417
1418 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1419                     int *len, int peer)
1420 {
1421         return -EOPNOTSUPP;
1422 }
1423
1424 unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt)
1425 {
1426         return 0;
1427 }
1428
1429 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1430 {
1431         return -EOPNOTSUPP;
1432 }
1433
1434 int sock_no_listen(struct socket *sock, int backlog)
1435 {
1436         return -EOPNOTSUPP;
1437 }
1438
1439 int sock_no_shutdown(struct socket *sock, int how)
1440 {
1441         return -EOPNOTSUPP;
1442 }
1443
1444 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1445                     char __user *optval, int optlen)
1446 {
1447         return -EOPNOTSUPP;
1448 }
1449
1450 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1451                     char __user *optval, int __user *optlen)
1452 {
1453         return -EOPNOTSUPP;
1454 }
1455
1456 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1457                     size_t len)
1458 {
1459         return -EOPNOTSUPP;
1460 }
1461
1462 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1463                     size_t len, int flags)
1464 {
1465         return -EOPNOTSUPP;
1466 }
1467
1468 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1469 {
1470         /* Mirror missing mmap method error code */
1471         return -ENODEV;
1472 }
1473
1474 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1475 {
1476         ssize_t res;
1477         struct msghdr msg = {.msg_flags = flags};
1478         struct kvec iov;
1479         char *kaddr = kmap(page);
1480         iov.iov_base = kaddr + offset;
1481         iov.iov_len = size;
1482         res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1483         kunmap(page);
1484         return res;
1485 }
1486
1487 /*
1488  *      Default Socket Callbacks
1489  */
1490
1491 static void sock_def_wakeup(struct sock *sk)
1492 {
1493         read_lock(&sk->sk_callback_lock);
1494         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1495                 wake_up_interruptible_all(sk->sk_sleep);
1496         read_unlock(&sk->sk_callback_lock);
1497 }
1498
1499 static void sock_def_error_report(struct sock *sk)
1500 {
1501         read_lock(&sk->sk_callback_lock);
1502         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1503                 wake_up_interruptible(sk->sk_sleep);
1504         sk_wake_async(sk,0,POLL_ERR);
1505         read_unlock(&sk->sk_callback_lock);
1506 }
1507
1508 static void sock_def_readable(struct sock *sk, int len)
1509 {
1510         read_lock(&sk->sk_callback_lock);
1511         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1512                 wake_up_interruptible(sk->sk_sleep);
1513         sk_wake_async(sk,1,POLL_IN);
1514         read_unlock(&sk->sk_callback_lock);
1515 }
1516
1517 static void sock_def_write_space(struct sock *sk)
1518 {
1519         read_lock(&sk->sk_callback_lock);
1520
1521         /* Do not wake up a writer until he can make "significant"
1522          * progress.  --DaveM
1523          */
1524         if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1525                 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1526                         wake_up_interruptible(sk->sk_sleep);
1527
1528                 /* Should agree with poll, otherwise some programs break */
1529                 if (sock_writeable(sk))
1530                         sk_wake_async(sk, 2, POLL_OUT);
1531         }
1532
1533         read_unlock(&sk->sk_callback_lock);
1534 }
1535
1536 static void sock_def_destruct(struct sock *sk)
1537 {
1538         kfree(sk->sk_protinfo);
1539 }
1540
1541 void sk_send_sigurg(struct sock *sk)
1542 {
1543         if (sk->sk_socket && sk->sk_socket->file)
1544                 if (send_sigurg(&sk->sk_socket->file->f_owner))
1545                         sk_wake_async(sk, 3, POLL_PRI);
1546 }
1547
1548 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1549                     unsigned long expires)
1550 {
1551         if (!mod_timer(timer, expires))
1552                 sock_hold(sk);
1553 }
1554
1555 EXPORT_SYMBOL(sk_reset_timer);
1556
1557 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1558 {
1559         if (timer_pending(timer) && del_timer(timer))
1560                 __sock_put(sk);
1561 }
1562
1563 EXPORT_SYMBOL(sk_stop_timer);
1564
1565 void sock_init_data(struct socket *sock, struct sock *sk)
1566 {
1567         skb_queue_head_init(&sk->sk_receive_queue);
1568         skb_queue_head_init(&sk->sk_write_queue);
1569         skb_queue_head_init(&sk->sk_error_queue);
1570 #ifdef CONFIG_NET_DMA
1571         skb_queue_head_init(&sk->sk_async_wait_queue);
1572 #endif
1573
1574         sk->sk_send_head        =       NULL;
1575
1576         init_timer(&sk->sk_timer);
1577
1578         sk->sk_allocation       =       GFP_KERNEL;
1579         sk->sk_rcvbuf           =       sysctl_rmem_default;
1580         sk->sk_sndbuf           =       sysctl_wmem_default;
1581         sk->sk_state            =       TCP_CLOSE;
1582         sk->sk_socket           =       sock;
1583
1584         sock_set_flag(sk, SOCK_ZAPPED);
1585
1586         if (sock) {
1587                 sk->sk_type     =       sock->type;
1588                 sk->sk_sleep    =       &sock->wait;
1589                 sock->sk        =       sk;
1590         } else
1591                 sk->sk_sleep    =       NULL;
1592
1593         rwlock_init(&sk->sk_dst_lock);
1594         rwlock_init(&sk->sk_callback_lock);
1595         lockdep_set_class_and_name(&sk->sk_callback_lock,
1596                         af_callback_keys + sk->sk_family,
1597                         af_family_clock_key_strings[sk->sk_family]);
1598
1599         sk->sk_state_change     =       sock_def_wakeup;
1600         sk->sk_data_ready       =       sock_def_readable;
1601         sk->sk_write_space      =       sock_def_write_space;
1602         sk->sk_error_report     =       sock_def_error_report;
1603         sk->sk_destruct         =       sock_def_destruct;
1604
1605         sk->sk_sndmsg_page      =       NULL;
1606         sk->sk_sndmsg_off       =       0;
1607
1608         sk->sk_peercred.pid     =       0;
1609         sk->sk_peercred.uid     =       -1;
1610         sk->sk_peercred.gid     =       -1;
1611         sk->sk_write_pending    =       0;
1612         sk->sk_rcvlowat         =       1;
1613         sk->sk_rcvtimeo         =       MAX_SCHEDULE_TIMEOUT;
1614         sk->sk_sndtimeo         =       MAX_SCHEDULE_TIMEOUT;
1615
1616         sk->sk_stamp = ktime_set(-1L, -1L);
1617
1618         atomic_set(&sk->sk_refcnt, 1);
1619 }
1620
1621 void fastcall lock_sock_nested(struct sock *sk, int subclass)
1622 {
1623         might_sleep();
1624         spin_lock_bh(&sk->sk_lock.slock);
1625         if (sk->sk_lock.owned)
1626                 __lock_sock(sk);
1627         sk->sk_lock.owned = 1;
1628         spin_unlock(&sk->sk_lock.slock);
1629         /*
1630          * The sk_lock has mutex_lock() semantics here:
1631          */
1632         mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
1633         local_bh_enable();
1634 }
1635
1636 EXPORT_SYMBOL(lock_sock_nested);
1637
1638 void fastcall release_sock(struct sock *sk)
1639 {
1640         /*
1641          * The sk_lock has mutex_unlock() semantics:
1642          */
1643         mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1644
1645         spin_lock_bh(&sk->sk_lock.slock);
1646         if (sk->sk_backlog.tail)
1647                 __release_sock(sk);
1648         sk->sk_lock.owned = 0;
1649         if (waitqueue_active(&sk->sk_lock.wq))
1650                 wake_up(&sk->sk_lock.wq);
1651         spin_unlock_bh(&sk->sk_lock.slock);
1652 }
1653 EXPORT_SYMBOL(release_sock);
1654
1655 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1656 {
1657         struct timeval tv;
1658         if (!sock_flag(sk, SOCK_TIMESTAMP))
1659                 sock_enable_timestamp(sk);
1660         tv = ktime_to_timeval(sk->sk_stamp);
1661         if (tv.tv_sec == -1)
1662                 return -ENOENT;
1663         if (tv.tv_sec == 0) {
1664                 sk->sk_stamp = ktime_get_real();
1665                 tv = ktime_to_timeval(sk->sk_stamp);
1666         }
1667         return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
1668 }
1669 EXPORT_SYMBOL(sock_get_timestamp);
1670
1671 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
1672 {
1673         struct timespec ts;
1674         if (!sock_flag(sk, SOCK_TIMESTAMP))
1675                 sock_enable_timestamp(sk);
1676         ts = ktime_to_timespec(sk->sk_stamp);
1677         if (ts.tv_sec == -1)
1678                 return -ENOENT;
1679         if (ts.tv_sec == 0) {
1680                 sk->sk_stamp = ktime_get_real();
1681                 ts = ktime_to_timespec(sk->sk_stamp);
1682         }
1683         return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
1684 }
1685 EXPORT_SYMBOL(sock_get_timestampns);
1686
1687 void sock_enable_timestamp(struct sock *sk)
1688 {
1689         if (!sock_flag(sk, SOCK_TIMESTAMP)) {
1690                 sock_set_flag(sk, SOCK_TIMESTAMP);
1691                 net_enable_timestamp();
1692         }
1693 }
1694
1695 /*
1696  *      Get a socket option on an socket.
1697  *
1698  *      FIX: POSIX 1003.1g is very ambiguous here. It states that
1699  *      asynchronous errors should be reported by getsockopt. We assume
1700  *      this means if you specify SO_ERROR (otherwise whats the point of it).
1701  */
1702 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1703                            char __user *optval, int __user *optlen)
1704 {
1705         struct sock *sk = sock->sk;
1706
1707         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1708 }
1709
1710 EXPORT_SYMBOL(sock_common_getsockopt);
1711
1712 #ifdef CONFIG_COMPAT
1713 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
1714                                   char __user *optval, int __user *optlen)
1715 {
1716         struct sock *sk = sock->sk;
1717
1718         if (sk->sk_prot->compat_getsockopt != NULL)
1719                 return sk->sk_prot->compat_getsockopt(sk, level, optname,
1720                                                       optval, optlen);
1721         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1722 }
1723 EXPORT_SYMBOL(compat_sock_common_getsockopt);
1724 #endif
1725
1726 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1727                         struct msghdr *msg, size_t size, int flags)
1728 {
1729         struct sock *sk = sock->sk;
1730         int addr_len = 0;
1731         int err;
1732
1733         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
1734                                    flags & ~MSG_DONTWAIT, &addr_len);
1735         if (err >= 0)
1736                 msg->msg_namelen = addr_len;
1737         return err;
1738 }
1739
1740 EXPORT_SYMBOL(sock_common_recvmsg);
1741
1742 /*
1743  *      Set socket options on an inet socket.
1744  */
1745 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1746                            char __user *optval, int optlen)
1747 {
1748         struct sock *sk = sock->sk;
1749
1750         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1751 }
1752
1753 EXPORT_SYMBOL(sock_common_setsockopt);
1754
1755 #ifdef CONFIG_COMPAT
1756 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
1757                                   char __user *optval, int optlen)
1758 {
1759         struct sock *sk = sock->sk;
1760
1761         if (sk->sk_prot->compat_setsockopt != NULL)
1762                 return sk->sk_prot->compat_setsockopt(sk, level, optname,
1763                                                       optval, optlen);
1764         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1765 }
1766 EXPORT_SYMBOL(compat_sock_common_setsockopt);
1767 #endif
1768
1769 void sk_common_release(struct sock *sk)
1770 {
1771         if (sk->sk_prot->destroy)
1772                 sk->sk_prot->destroy(sk);
1773
1774         /*
1775          * Observation: when sock_common_release is called, processes have
1776          * no access to socket. But net still has.
1777          * Step one, detach it from networking:
1778          *
1779          * A. Remove from hash tables.
1780          */
1781
1782         sk->sk_prot->unhash(sk);
1783
1784         /*
1785          * In this point socket cannot receive new packets, but it is possible
1786          * that some packets are in flight because some CPU runs receiver and
1787          * did hash table lookup before we unhashed socket. They will achieve
1788          * receive queue and will be purged by socket destructor.
1789          *
1790          * Also we still have packets pending on receive queue and probably,
1791          * our own packets waiting in device queues. sock_destroy will drain
1792          * receive queue, but transmitted packets will delay socket destruction
1793          * until the last reference will be released.
1794          */
1795
1796         sock_orphan(sk);
1797
1798         xfrm_sk_free_policy(sk);
1799
1800         sk_refcnt_debug_release(sk);
1801         sock_put(sk);
1802 }
1803
1804 EXPORT_SYMBOL(sk_common_release);
1805
1806 static DEFINE_RWLOCK(proto_list_lock);
1807 static LIST_HEAD(proto_list);
1808
1809 int proto_register(struct proto *prot, int alloc_slab)
1810 {
1811         char *request_sock_slab_name = NULL;
1812         char *timewait_sock_slab_name;
1813         int rc = -ENOBUFS;
1814
1815         if (alloc_slab) {
1816                 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
1817                                                SLAB_HWCACHE_ALIGN, NULL);
1818
1819                 if (prot->slab == NULL) {
1820                         printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
1821                                prot->name);
1822                         goto out;
1823                 }
1824
1825                 if (prot->rsk_prot != NULL) {
1826                         static const char mask[] = "request_sock_%s";
1827
1828                         request_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
1829                         if (request_sock_slab_name == NULL)
1830                                 goto out_free_sock_slab;
1831
1832                         sprintf(request_sock_slab_name, mask, prot->name);
1833                         prot->rsk_prot->slab = kmem_cache_create(request_sock_slab_name,
1834                                                                  prot->rsk_prot->obj_size, 0,
1835                                                                  SLAB_HWCACHE_ALIGN, NULL);
1836
1837                         if (prot->rsk_prot->slab == NULL) {
1838                                 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
1839                                        prot->name);
1840                                 goto out_free_request_sock_slab_name;
1841                         }
1842                 }
1843
1844                 if (prot->twsk_prot != NULL) {
1845                         static const char mask[] = "tw_sock_%s";
1846
1847                         timewait_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
1848
1849                         if (timewait_sock_slab_name == NULL)
1850                                 goto out_free_request_sock_slab;
1851
1852                         sprintf(timewait_sock_slab_name, mask, prot->name);
1853                         prot->twsk_prot->twsk_slab =
1854                                 kmem_cache_create(timewait_sock_slab_name,
1855                                                   prot->twsk_prot->twsk_obj_size,
1856                                                   0, SLAB_HWCACHE_ALIGN,
1857                                                   NULL);
1858                         if (prot->twsk_prot->twsk_slab == NULL)
1859                                 goto out_free_timewait_sock_slab_name;
1860                 }
1861         }
1862
1863         write_lock(&proto_list_lock);
1864         list_add(&prot->node, &proto_list);
1865         write_unlock(&proto_list_lock);
1866         rc = 0;
1867 out:
1868         return rc;
1869 out_free_timewait_sock_slab_name:
1870         kfree(timewait_sock_slab_name);
1871 out_free_request_sock_slab:
1872         if (prot->rsk_prot && prot->rsk_prot->slab) {
1873                 kmem_cache_destroy(prot->rsk_prot->slab);
1874                 prot->rsk_prot->slab = NULL;
1875         }
1876 out_free_request_sock_slab_name:
1877         kfree(request_sock_slab_name);
1878 out_free_sock_slab:
1879         kmem_cache_destroy(prot->slab);
1880         prot->slab = NULL;
1881         goto out;
1882 }
1883
1884 EXPORT_SYMBOL(proto_register);
1885
1886 void proto_unregister(struct proto *prot)
1887 {
1888         write_lock(&proto_list_lock);
1889         list_del(&prot->node);
1890         write_unlock(&proto_list_lock);
1891
1892         if (prot->slab != NULL) {
1893                 kmem_cache_destroy(prot->slab);
1894                 prot->slab = NULL;
1895         }
1896
1897         if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
1898                 const char *name = kmem_cache_name(prot->rsk_prot->slab);
1899
1900                 kmem_cache_destroy(prot->rsk_prot->slab);
1901                 kfree(name);
1902                 prot->rsk_prot->slab = NULL;
1903         }
1904
1905         if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
1906                 const char *name = kmem_cache_name(prot->twsk_prot->twsk_slab);
1907
1908                 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
1909                 kfree(name);
1910                 prot->twsk_prot->twsk_slab = NULL;
1911         }
1912 }
1913
1914 EXPORT_SYMBOL(proto_unregister);
1915
1916 #ifdef CONFIG_PROC_FS
1917 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
1918 {
1919         read_lock(&proto_list_lock);
1920         return seq_list_start_head(&proto_list, *pos);
1921 }
1922
1923 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1924 {
1925         return seq_list_next(v, &proto_list, pos);
1926 }
1927
1928 static void proto_seq_stop(struct seq_file *seq, void *v)
1929 {
1930         read_unlock(&proto_list_lock);
1931 }
1932
1933 static char proto_method_implemented(const void *method)
1934 {
1935         return method == NULL ? 'n' : 'y';
1936 }
1937
1938 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
1939 {
1940         seq_printf(seq, "%-9s %4u %6d  %6d   %-3s %6u   %-3s  %-10s "
1941                         "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
1942                    proto->name,
1943                    proto->obj_size,
1944                    proto->sockets_allocated != NULL ? atomic_read(proto->sockets_allocated) : -1,
1945                    proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
1946                    proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
1947                    proto->max_header,
1948                    proto->slab == NULL ? "no" : "yes",
1949                    module_name(proto->owner),
1950                    proto_method_implemented(proto->close),
1951                    proto_method_implemented(proto->connect),
1952                    proto_method_implemented(proto->disconnect),
1953                    proto_method_implemented(proto->accept),
1954                    proto_method_implemented(proto->ioctl),
1955                    proto_method_implemented(proto->init),
1956                    proto_method_implemented(proto->destroy),
1957                    proto_method_implemented(proto->shutdown),
1958                    proto_method_implemented(proto->setsockopt),
1959                    proto_method_implemented(proto->getsockopt),
1960                    proto_method_implemented(proto->sendmsg),
1961                    proto_method_implemented(proto->recvmsg),
1962                    proto_method_implemented(proto->sendpage),
1963                    proto_method_implemented(proto->bind),
1964                    proto_method_implemented(proto->backlog_rcv),
1965                    proto_method_implemented(proto->hash),
1966                    proto_method_implemented(proto->unhash),
1967                    proto_method_implemented(proto->get_port),
1968                    proto_method_implemented(proto->enter_memory_pressure));
1969 }
1970
1971 static int proto_seq_show(struct seq_file *seq, void *v)
1972 {
1973         if (v == &proto_list)
1974                 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
1975                            "protocol",
1976                            "size",
1977                            "sockets",
1978                            "memory",
1979                            "press",
1980                            "maxhdr",
1981                            "slab",
1982                            "module",
1983                            "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
1984         else
1985                 proto_seq_printf(seq, list_entry(v, struct proto, node));
1986         return 0;
1987 }
1988
1989 static const struct seq_operations proto_seq_ops = {
1990         .start  = proto_seq_start,
1991         .next   = proto_seq_next,
1992         .stop   = proto_seq_stop,
1993         .show   = proto_seq_show,
1994 };
1995
1996 static int proto_seq_open(struct inode *inode, struct file *file)
1997 {
1998         return seq_open(file, &proto_seq_ops);
1999 }
2000
2001 static const struct file_operations proto_seq_fops = {
2002         .owner          = THIS_MODULE,
2003         .open           = proto_seq_open,
2004         .read           = seq_read,
2005         .llseek         = seq_lseek,
2006         .release        = seq_release,
2007 };
2008
2009 static int __init proto_init(void)
2010 {
2011         /* register /proc/net/protocols */
2012         return proc_net_fops_create(&init_net, "protocols", S_IRUGO, &proto_seq_fops) == NULL ? -ENOBUFS : 0;
2013 }
2014
2015 subsys_initcall(proto_init);
2016
2017 #endif /* PROC_FS */
2018
2019 EXPORT_SYMBOL(sk_alloc);
2020 EXPORT_SYMBOL(sk_free);
2021 EXPORT_SYMBOL(sk_send_sigurg);
2022 EXPORT_SYMBOL(sock_alloc_send_skb);
2023 EXPORT_SYMBOL(sock_init_data);
2024 EXPORT_SYMBOL(sock_kfree_s);
2025 EXPORT_SYMBOL(sock_kmalloc);
2026 EXPORT_SYMBOL(sock_no_accept);
2027 EXPORT_SYMBOL(sock_no_bind);
2028 EXPORT_SYMBOL(sock_no_connect);
2029 EXPORT_SYMBOL(sock_no_getname);
2030 EXPORT_SYMBOL(sock_no_getsockopt);
2031 EXPORT_SYMBOL(sock_no_ioctl);
2032 EXPORT_SYMBOL(sock_no_listen);
2033 EXPORT_SYMBOL(sock_no_mmap);
2034 EXPORT_SYMBOL(sock_no_poll);
2035 EXPORT_SYMBOL(sock_no_recvmsg);
2036 EXPORT_SYMBOL(sock_no_sendmsg);
2037 EXPORT_SYMBOL(sock_no_sendpage);
2038 EXPORT_SYMBOL(sock_no_setsockopt);
2039 EXPORT_SYMBOL(sock_no_shutdown);
2040 EXPORT_SYMBOL(sock_no_socketpair);
2041 EXPORT_SYMBOL(sock_rfree);
2042 EXPORT_SYMBOL(sock_setsockopt);
2043 EXPORT_SYMBOL(sock_wfree);
2044 EXPORT_SYMBOL(sock_wmalloc);
2045 EXPORT_SYMBOL(sock_i_uid);
2046 EXPORT_SYMBOL(sock_i_ino);
2047 EXPORT_SYMBOL(sysctl_optmem_max);
2048 #ifdef CONFIG_SYSCTL
2049 EXPORT_SYMBOL(sysctl_rmem_max);
2050 EXPORT_SYMBOL(sysctl_wmem_max);
2051 #endif