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 /**
 861  *      sk_alloc - All socket objects are allocated here
 862  *      @net: the applicable net namespace
 863  *      @family: protocol family
 864  *      @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
 865  *      @prot: struct proto associated with this new sock instance
 866  *      @zero_it: if we should zero the newly allocated sock
 867  */
 868 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
 869                       struct proto *prot, int zero_it)
 870 {
 871         struct sock *sk = NULL;
 872         struct kmem_cache *slab = prot->slab;
 873
 874         if (slab != NULL)
 875                 sk = kmem_cache_alloc(slab, priority);
 876         else
 877                 sk = kmalloc(prot->obj_size, priority);
 878
 879         if (sk) {
 880                 if (zero_it) {
 881                         memset(sk, 0, prot->obj_size);
 882                         sk->sk_family = family;
 883                         /*
 884                          * See comment in struct sock definition to understand
 885                          * why we need sk_prot_creator -acme
 886                          */
 887                         sk->sk_prot = sk->sk_prot_creator = prot;
 888                         sock_lock_init(sk);
 889                         sk->sk_net = get_net(net);
 890                 }
 891
 892                 if (security_sk_alloc(sk, family, priority))
 893                         goto out_free;
 894
 895                 if (!try_module_get(prot->owner))
 896                         goto out_free;
 897         }
 898         return sk;
 899
 900 out_free:
 901         if (slab != NULL)
 902                 kmem_cache_free(slab, sk);
 903         else
 904                 kfree(sk);
 905         return NULL;
 906 }
 907
 908 void sk_free(struct sock *sk)
 909 {
 910         struct sk_filter *filter;
 911         struct module *owner = sk->sk_prot_creator->owner;
 912
 913         if (sk->sk_destruct)
 914                 sk->sk_destruct(sk);
 915
 916         filter = rcu_dereference(sk->sk_filter);
 917         if (filter) {
 918                 sk_filter_uncharge(sk, filter);
 919                 rcu_assign_pointer(sk->sk_filter, NULL);
 920         }
 921
 922         sock_disable_timestamp(sk);
 923
 924         if (atomic_read(&sk->sk_omem_alloc))
 925                 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
 926                        __FUNCTION__, atomic_read(&sk->sk_omem_alloc));
 927
 928         security_sk_free(sk);
 929         put_net(sk->sk_net);
 930         if (sk->sk_prot_creator->slab != NULL)
 931                 kmem_cache_free(sk->sk_prot_creator->slab, sk);
 932         else
 933                 kfree(sk);
 934         module_put(owner);
 935 }
 936
 937 struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
 938 {
 939         struct sock *newsk = sk_alloc(sk->sk_net, sk->sk_family, priority, sk->sk_prot, 0);
 940
 941         if (newsk != NULL) {
 942                 struct sk_filter *filter;
 943
 944                 sock_copy(newsk, sk);
 945
 946                 /* SANITY */
 947                 sk_node_init(&newsk->sk_node);
 948                 sock_lock_init(newsk);
 949                 bh_lock_sock(newsk);
 950                 newsk->sk_backlog.head  = newsk->sk_backlog.tail = NULL;
 951
 952                 atomic_set(&newsk->sk_rmem_alloc, 0);
 953                 atomic_set(&newsk->sk_wmem_alloc, 0);
 954                 atomic_set(&newsk->sk_omem_alloc, 0);
 955                 skb_queue_head_init(&newsk->sk_receive_queue);
 956                 skb_queue_head_init(&newsk->sk_write_queue);
 957 #ifdef CONFIG_NET_DMA
 958                 skb_queue_head_init(&newsk->sk_async_wait_queue);
 959 #endif
 960
 961                 rwlock_init(&newsk->sk_dst_lock);
 962                 rwlock_init(&newsk->sk_callback_lock);
 963                 lockdep_set_class_and_name(&newsk->sk_callback_lock,
 964                                 af_callback_keys + newsk->sk_family,
 965                                 af_family_clock_key_strings[newsk->sk_family]);
 966
 967                 newsk->sk_dst_cache     = NULL;
 968                 newsk->sk_wmem_queued   = 0;
 969                 newsk->sk_forward_alloc = 0;
 970                 newsk->sk_send_head     = NULL;
 971                 newsk->sk_userlocks     = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
 972
 973                 sock_reset_flag(newsk, SOCK_DONE);
 974                 skb_queue_head_init(&newsk->sk_error_queue);
 975
 976                 filter = newsk->sk_filter;
 977                 if (filter != NULL)
 978                         sk_filter_charge(newsk, filter);
 979
 980                 if (unlikely(xfrm_sk_clone_policy(newsk))) {
 981                         /* It is still raw copy of parent, so invalidate
 982                          * destructor and make plain sk_free() */
 983                         newsk->sk_destruct = NULL;
 984                         sk_free(newsk);
 985                         newsk = NULL;
 986                         goto out;
 987                 }
 988
 989                 newsk->sk_err      = 0;
 990                 newsk->sk_priority = 0;
 991                 atomic_set(&newsk->sk_refcnt, 2);
 992
 993                 /*
 994                  * Increment the counter in the same struct proto as the master
 995                  * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
 996                  * is the same as sk->sk_prot->socks, as this field was copied
 997                  * with memcpy).
 998                  *
 999                  * This _changes_ the previous behaviour, where
1000                  * tcp_create_openreq_child always was incrementing the
1001                  * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1002                  * to be taken into account in all callers. -acme
1003                  */
1004                 sk_refcnt_debug_inc(newsk);
1005                 newsk->sk_socket = NULL;
1006                 newsk->sk_sleep  = NULL;
1007
1008                 if (newsk->sk_prot->sockets_allocated)
1009                         atomic_inc(newsk->sk_prot->sockets_allocated);
1010         }
1011 out:
1012         return newsk;
1013 }
1014
1015 EXPORT_SYMBOL_GPL(sk_clone);
1016
1017 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1018 {
1019         __sk_dst_set(sk, dst);
1020         sk->sk_route_caps = dst->dev->features;
1021         if (sk->sk_route_caps & NETIF_F_GSO)
1022                 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1023         if (sk_can_gso(sk)) {
1024                 if (dst->header_len)
1025                         sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1026                 else
1027                         sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1028         }
1029 }
1030 EXPORT_SYMBOL_GPL(sk_setup_caps);
1031
1032 void __init sk_init(void)
1033 {
1034         if (num_physpages <= 4096) {
1035                 sysctl_wmem_max = 32767;
1036                 sysctl_rmem_max = 32767;
1037                 sysctl_wmem_default = 32767;
1038                 sysctl_rmem_default = 32767;
1039         } else if (num_physpages >= 131072) {
1040                 sysctl_wmem_max = 131071;
1041                 sysctl_rmem_max = 131071;
1042         }
1043 }
1044
1045 /*
1046  *      Simple resource managers for sockets.
1047  */
1048
1049
1050 /*
1051  * Write buffer destructor automatically called from kfree_skb.
1052  */
1053 void sock_wfree(struct sk_buff *skb)
1054 {
1055         struct sock *sk = skb->sk;
1056
1057         /* In case it might be waiting for more memory. */
1058         atomic_sub(skb->truesize, &sk->sk_wmem_alloc);
1059         if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE))
1060                 sk->sk_write_space(sk);
1061         sock_put(sk);
1062 }
1063
1064 /*
1065  * Read buffer destructor automatically called from kfree_skb.
1066  */
1067 void sock_rfree(struct sk_buff *skb)
1068 {
1069         struct sock *sk = skb->sk;
1070
1071         atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1072 }
1073
1074
1075 int sock_i_uid(struct sock *sk)
1076 {
1077         int uid;
1078
1079         read_lock(&sk->sk_callback_lock);
1080         uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1081         read_unlock(&sk->sk_callback_lock);
1082         return uid;
1083 }
1084
1085 unsigned long sock_i_ino(struct sock *sk)
1086 {
1087         unsigned long ino;
1088
1089         read_lock(&sk->sk_callback_lock);
1090         ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1091         read_unlock(&sk->sk_callback_lock);
1092         return ino;
1093 }
1094
1095 /*
1096  * Allocate a skb from the socket's send buffer.
1097  */
1098 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1099                              gfp_t priority)
1100 {
1101         if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1102                 struct sk_buff * skb = alloc_skb(size, priority);
1103                 if (skb) {
1104                         skb_set_owner_w(skb, sk);
1105                         return skb;
1106                 }
1107         }
1108         return NULL;
1109 }
1110
1111 /*
1112  * Allocate a skb from the socket's receive buffer.
1113  */
1114 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1115                              gfp_t priority)
1116 {
1117         if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1118                 struct sk_buff *skb = alloc_skb(size, priority);
1119                 if (skb) {
1120                         skb_set_owner_r(skb, sk);
1121                         return skb;
1122                 }
1123         }
1124         return NULL;
1125 }
1126
1127 /*
1128  * Allocate a memory block from the socket's option memory buffer.
1129  */
1130 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1131 {
1132         if ((unsigned)size <= sysctl_optmem_max &&
1133             atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1134                 void *mem;
1135                 /* First do the add, to avoid the race if kmalloc
1136                  * might sleep.
1137                  */
1138                 atomic_add(size, &sk->sk_omem_alloc);
1139                 mem = kmalloc(size, priority);
1140                 if (mem)
1141                         return mem;
1142                 atomic_sub(size, &sk->sk_omem_alloc);
1143         }
1144         return NULL;
1145 }
1146
1147 /*
1148  * Free an option memory block.
1149  */
1150 void sock_kfree_s(struct sock *sk, void *mem, int size)
1151 {
1152         kfree(mem);
1153         atomic_sub(size, &sk->sk_omem_alloc);
1154 }
1155
1156 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1157    I think, these locks should be removed for datagram sockets.
1158  */
1159 static long sock_wait_for_wmem(struct sock * sk, long timeo)
1160 {
1161         DEFINE_WAIT(wait);
1162
1163         clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1164         for (;;) {
1165                 if (!timeo)
1166                         break;
1167                 if (signal_pending(current))
1168                         break;
1169                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1170                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1171                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1172                         break;
1173                 if (sk->sk_shutdown & SEND_SHUTDOWN)
1174                         break;
1175                 if (sk->sk_err)
1176                         break;
1177                 timeo = schedule_timeout(timeo);
1178         }
1179         finish_wait(sk->sk_sleep, &wait);
1180         return timeo;
1181 }
1182
1183
1184 /*
1185  *      Generic send/receive buffer handlers
1186  */
1187
1188 static struct sk_buff *sock_alloc_send_pskb(struct sock *sk,
1189                                             unsigned long header_len,
1190                                             unsigned long data_len,
1191                                             int noblock, int *errcode)
1192 {
1193         struct sk_buff *skb;
1194         gfp_t gfp_mask;
1195         long timeo;
1196         int err;
1197
1198         gfp_mask = sk->sk_allocation;
1199         if (gfp_mask & __GFP_WAIT)
1200                 gfp_mask |= __GFP_REPEAT;
1201
1202         timeo = sock_sndtimeo(sk, noblock);
1203         while (1) {
1204                 err = sock_error(sk);
1205                 if (err != 0)
1206                         goto failure;
1207
1208                 err = -EPIPE;
1209                 if (sk->sk_shutdown & SEND_SHUTDOWN)
1210                         goto failure;
1211
1212                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1213                         skb = alloc_skb(header_len, gfp_mask);
1214                         if (skb) {
1215                                 int npages;
1216                                 int i;
1217
1218                                 /* No pages, we're done... */
1219                                 if (!data_len)
1220                                         break;
1221
1222                                 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1223                                 skb->truesize += data_len;
1224                                 skb_shinfo(skb)->nr_frags = npages;
1225                                 for (i = 0; i < npages; i++) {
1226                                         struct page *page;
1227                                         skb_frag_t *frag;
1228
1229                                         page = alloc_pages(sk->sk_allocation, 0);
1230                                         if (!page) {
1231                                                 err = -ENOBUFS;
1232                                                 skb_shinfo(skb)->nr_frags = i;
1233                                                 kfree_skb(skb);
1234                                                 goto failure;
1235                                         }
1236
1237                                         frag = &skb_shinfo(skb)->frags[i];
1238                                         frag->page = page;
1239                                         frag->page_offset = 0;
1240                                         frag->size = (data_len >= PAGE_SIZE ?
1241                                                       PAGE_SIZE :
1242                                                       data_len);
1243                                         data_len -= PAGE_SIZE;
1244                                 }
1245
1246                                 /* Full success... */
1247                                 break;
1248                         }
1249                         err = -ENOBUFS;
1250                         goto failure;
1251                 }
1252                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1253                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1254                 err = -EAGAIN;
1255                 if (!timeo)
1256                         goto failure;
1257                 if (signal_pending(current))
1258                         goto interrupted;
1259                 timeo = sock_wait_for_wmem(sk, timeo);
1260         }
1261
1262         skb_set_owner_w(skb, sk);
1263         return skb;
1264
1265 interrupted:
1266         err = sock_intr_errno(timeo);
1267 failure:
1268         *errcode = err;
1269         return NULL;
1270 }
1271
1272 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1273                                     int noblock, int *errcode)
1274 {
1275         return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1276 }
1277
1278 static void __lock_sock(struct sock *sk)
1279 {
1280         DEFINE_WAIT(wait);
1281
1282         for (;;) {
1283                 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1284                                         TASK_UNINTERRUPTIBLE);
1285                 spin_unlock_bh(&sk->sk_lock.slock);
1286                 schedule();
1287                 spin_lock_bh(&sk->sk_lock.slock);
1288                 if (!sock_owned_by_user(sk))
1289                         break;
1290         }
1291         finish_wait(&sk->sk_lock.wq, &wait);
1292 }
1293
1294 static void __release_sock(struct sock *sk)
1295 {
1296         struct sk_buff *skb = sk->sk_backlog.head;
1297
1298         do {
1299                 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1300                 bh_unlock_sock(sk);
1301
1302                 do {
1303                         struct sk_buff *next = skb->next;
1304
1305                         skb->next = NULL;
1306                         sk->sk_backlog_rcv(sk, skb);
1307
1308                         /*
1309                          * We are in process context here with softirqs
1310                          * disabled, use cond_resched_softirq() to preempt.
1311                          * This is safe to do because we've taken the backlog
1312                          * queue private:
1313                          */
1314                         cond_resched_softirq();
1315
1316                         skb = next;
1317                 } while (skb != NULL);
1318
1319                 bh_lock_sock(sk);
1320         } while ((skb = sk->sk_backlog.head) != NULL);
1321 }
1322
1323 /**
1324  * sk_wait_data - wait for data to arrive at sk_receive_queue
1325  * @sk:    sock to wait on
1326  * @timeo: for how long
1327  *
1328  * Now socket state including sk->sk_err is changed only under lock,
1329  * hence we may omit checks after joining wait queue.
1330  * We check receive queue before schedule() only as optimization;
1331  * it is very likely that release_sock() added new data.
1332  */
1333 int sk_wait_data(struct sock *sk, long *timeo)
1334 {
1335         int rc;
1336         DEFINE_WAIT(wait);
1337
1338         prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1339         set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1340         rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1341         clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1342         finish_wait(sk->sk_sleep, &wait);
1343         return rc;
1344 }
1345
1346 EXPORT_SYMBOL(sk_wait_data);
1347
1348 /*
1349  * Set of default routines for initialising struct proto_ops when
1350  * the protocol does not support a particular function. In certain
1351  * cases where it makes no sense for a protocol to have a "do nothing"
1352  * function, some default processing is provided.
1353  */
1354
1355 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1356 {
1357         return -EOPNOTSUPP;
1358 }
1359
1360 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1361                     int len, int flags)
1362 {
1363         return -EOPNOTSUPP;
1364 }
1365
1366 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1367 {
1368         return -EOPNOTSUPP;
1369 }
1370
1371 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1372 {
1373         return -EOPNOTSUPP;
1374 }
1375
1376 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1377                     int *len, int peer)
1378 {
1379         return -EOPNOTSUPP;
1380 }
1381
1382 unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt)
1383 {
1384         return 0;
1385 }
1386
1387 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1388 {
1389         return -EOPNOTSUPP;
1390 }
1391
1392 int sock_no_listen(struct socket *sock, int backlog)
1393 {
1394         return -EOPNOTSUPP;
1395 }
1396
1397 int sock_no_shutdown(struct socket *sock, int how)
1398 {
1399         return -EOPNOTSUPP;
1400 }
1401
1402 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1403                     char __user *optval, int optlen)
1404 {
1405         return -EOPNOTSUPP;
1406 }
1407
1408 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1409                     char __user *optval, int __user *optlen)
1410 {
1411         return -EOPNOTSUPP;
1412 }
1413
1414 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1415                     size_t len)
1416 {
1417         return -EOPNOTSUPP;
1418 }
1419
1420 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1421                     size_t len, int flags)
1422 {
1423         return -EOPNOTSUPP;
1424 }
1425
1426 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1427 {
1428         /* Mirror missing mmap method error code */
1429         return -ENODEV;
1430 }
1431
1432 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1433 {
1434         ssize_t res;
1435         struct msghdr msg = {.msg_flags = flags};
1436         struct kvec iov;
1437         char *kaddr = kmap(page);
1438         iov.iov_base = kaddr + offset;
1439         iov.iov_len = size;
1440         res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1441         kunmap(page);
1442         return res;
1443 }
1444
1445 /*
1446  *      Default Socket Callbacks
1447  */
1448
1449 static void sock_def_wakeup(struct sock *sk)
1450 {
1451         read_lock(&sk->sk_callback_lock);
1452         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1453                 wake_up_interruptible_all(sk->sk_sleep);
1454         read_unlock(&sk->sk_callback_lock);
1455 }
1456
1457 static void sock_def_error_report(struct sock *sk)
1458 {
1459         read_lock(&sk->sk_callback_lock);
1460         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1461                 wake_up_interruptible(sk->sk_sleep);
1462         sk_wake_async(sk,0,POLL_ERR);
1463         read_unlock(&sk->sk_callback_lock);
1464 }
1465
1466 static void sock_def_readable(struct sock *sk, int len)
1467 {
1468         read_lock(&sk->sk_callback_lock);
1469         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1470                 wake_up_interruptible(sk->sk_sleep);
1471         sk_wake_async(sk,1,POLL_IN);
1472         read_unlock(&sk->sk_callback_lock);
1473 }
1474
1475 static void sock_def_write_space(struct sock *sk)
1476 {
1477         read_lock(&sk->sk_callback_lock);
1478
1479         /* Do not wake up a writer until he can make "significant"
1480          * progress.  --DaveM
1481          */
1482         if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1483                 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1484                         wake_up_interruptible(sk->sk_sleep);
1485
1486                 /* Should agree with poll, otherwise some programs break */
1487                 if (sock_writeable(sk))
1488                         sk_wake_async(sk, 2, POLL_OUT);
1489         }
1490
1491         read_unlock(&sk->sk_callback_lock);
1492 }
1493
1494 static void sock_def_destruct(struct sock *sk)
1495 {
1496         kfree(sk->sk_protinfo);
1497 }
1498
1499 void sk_send_sigurg(struct sock *sk)
1500 {
1501         if (sk->sk_socket && sk->sk_socket->file)
1502                 if (send_sigurg(&sk->sk_socket->file->f_owner))
1503                         sk_wake_async(sk, 3, POLL_PRI);
1504 }
1505
1506 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1507                     unsigned long expires)
1508 {
1509         if (!mod_timer(timer, expires))
1510                 sock_hold(sk);
1511 }
1512
1513 EXPORT_SYMBOL(sk_reset_timer);
1514
1515 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1516 {
1517         if (timer_pending(timer) && del_timer(timer))
1518                 __sock_put(sk);
1519 }
1520
1521 EXPORT_SYMBOL(sk_stop_timer);
1522
1523 void sock_init_data(struct socket *sock, struct sock *sk)
1524 {
1525         skb_queue_head_init(&sk->sk_receive_queue);
1526         skb_queue_head_init(&sk->sk_write_queue);
1527         skb_queue_head_init(&sk->sk_error_queue);
1528 #ifdef CONFIG_NET_DMA
1529         skb_queue_head_init(&sk->sk_async_wait_queue);
1530 #endif
1531
1532         sk->sk_send_head        =       NULL;
1533
1534         init_timer(&sk->sk_timer);
1535
1536         sk->sk_allocation       =       GFP_KERNEL;
1537         sk->sk_rcvbuf           =       sysctl_rmem_default;
1538         sk->sk_sndbuf           =       sysctl_wmem_default;
1539         sk->sk_state            =       TCP_CLOSE;
1540         sk->sk_socket           =       sock;
1541
1542         sock_set_flag(sk, SOCK_ZAPPED);
1543
1544         if (sock) {
1545                 sk->sk_type     =       sock->type;
1546                 sk->sk_sleep    =       &sock->wait;
1547                 sock->sk        =       sk;
1548         } else
1549                 sk->sk_sleep    =       NULL;
1550
1551         rwlock_init(&sk->sk_dst_lock);
1552         rwlock_init(&sk->sk_callback_lock);
1553         lockdep_set_class_and_name(&sk->sk_callback_lock,
1554                         af_callback_keys + sk->sk_family,
1555                         af_family_clock_key_strings[sk->sk_family]);
1556
1557         sk->sk_state_change     =       sock_def_wakeup;
1558         sk->sk_data_ready       =       sock_def_readable;
1559         sk->sk_write_space      =       sock_def_write_space;
1560         sk->sk_error_report     =       sock_def_error_report;
1561         sk->sk_destruct         =       sock_def_destruct;
1562
1563         sk->sk_sndmsg_page      =       NULL;
1564         sk->sk_sndmsg_off       =       0;
1565
1566         sk->sk_peercred.pid     =       0;
1567         sk->sk_peercred.uid     =       -1;
1568         sk->sk_peercred.gid     =       -1;
1569         sk->sk_write_pending    =       0;
1570         sk->sk_rcvlowat         =       1;
1571         sk->sk_rcvtimeo         =       MAX_SCHEDULE_TIMEOUT;
1572         sk->sk_sndtimeo         =       MAX_SCHEDULE_TIMEOUT;
1573
1574         sk->sk_stamp = ktime_set(-1L, -1L);
1575
1576         atomic_set(&sk->sk_refcnt, 1);
1577 }
1578
1579 void fastcall lock_sock_nested(struct sock *sk, int subclass)
1580 {
1581         might_sleep();
1582         spin_lock_bh(&sk->sk_lock.slock);
1583         if (sk->sk_lock.owned)
1584                 __lock_sock(sk);
1585         sk->sk_lock.owned = 1;
1586         spin_unlock(&sk->sk_lock.slock);
1587         /*
1588          * The sk_lock has mutex_lock() semantics here:
1589          */
1590         mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
1591         local_bh_enable();
1592 }
1593
1594 EXPORT_SYMBOL(lock_sock_nested);
1595
1596 void fastcall release_sock(struct sock *sk)
1597 {
1598         /*
1599          * The sk_lock has mutex_unlock() semantics:
1600          */
1601         mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1602
1603         spin_lock_bh(&sk->sk_lock.slock);
1604         if (sk->sk_backlog.tail)
1605                 __release_sock(sk);
1606         sk->sk_lock.owned = 0;
1607         if (waitqueue_active(&sk->sk_lock.wq))
1608                 wake_up(&sk->sk_lock.wq);
1609         spin_unlock_bh(&sk->sk_lock.slock);
1610 }
1611 EXPORT_SYMBOL(release_sock);
1612
1613 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1614 {
1615         struct timeval tv;
1616         if (!sock_flag(sk, SOCK_TIMESTAMP))
1617                 sock_enable_timestamp(sk);
1618         tv = ktime_to_timeval(sk->sk_stamp);
1619         if (tv.tv_sec == -1)
1620                 return -ENOENT;
1621         if (tv.tv_sec == 0) {
1622                 sk->sk_stamp = ktime_get_real();
1623                 tv = ktime_to_timeval(sk->sk_stamp);
1624         }
1625         return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
1626 }
1627 EXPORT_SYMBOL(sock_get_timestamp);
1628
1629 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
1630 {
1631         struct timespec ts;
1632         if (!sock_flag(sk, SOCK_TIMESTAMP))
1633                 sock_enable_timestamp(sk);
1634         ts = ktime_to_timespec(sk->sk_stamp);
1635         if (ts.tv_sec == -1)
1636                 return -ENOENT;
1637         if (ts.tv_sec == 0) {
1638                 sk->sk_stamp = ktime_get_real();
1639                 ts = ktime_to_timespec(sk->sk_stamp);
1640         }
1641         return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
1642 }
1643 EXPORT_SYMBOL(sock_get_timestampns);
1644
1645 void sock_enable_timestamp(struct sock *sk)
1646 {
1647         if (!sock_flag(sk, SOCK_TIMESTAMP)) {
1648                 sock_set_flag(sk, SOCK_TIMESTAMP);
1649                 net_enable_timestamp();
1650         }
1651 }
1652
1653 /*
1654  *      Get a socket option on an socket.
1655  *
1656  *      FIX: POSIX 1003.1g is very ambiguous here. It states that
1657  *      asynchronous errors should be reported by getsockopt. We assume
1658  *      this means if you specify SO_ERROR (otherwise whats the point of it).
1659  */
1660 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1661                            char __user *optval, int __user *optlen)
1662 {
1663         struct sock *sk = sock->sk;
1664
1665         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1666 }
1667
1668 EXPORT_SYMBOL(sock_common_getsockopt);
1669
1670 #ifdef CONFIG_COMPAT
1671 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
1672                                   char __user *optval, int __user *optlen)
1673 {
1674         struct sock *sk = sock->sk;
1675
1676         if (sk->sk_prot->compat_getsockopt != NULL)
1677                 return sk->sk_prot->compat_getsockopt(sk, level, optname,
1678                                                       optval, optlen);
1679         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1680 }
1681 EXPORT_SYMBOL(compat_sock_common_getsockopt);
1682 #endif
1683
1684 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1685                         struct msghdr *msg, size_t size, int flags)
1686 {
1687         struct sock *sk = sock->sk;
1688         int addr_len = 0;
1689         int err;
1690
1691         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
1692                                    flags & ~MSG_DONTWAIT, &addr_len);
1693         if (err >= 0)
1694                 msg->msg_namelen = addr_len;
1695         return err;
1696 }
1697
1698 EXPORT_SYMBOL(sock_common_recvmsg);
1699
1700 /*
1701  *      Set socket options on an inet socket.
1702  */
1703 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1704                            char __user *optval, int optlen)
1705 {
1706         struct sock *sk = sock->sk;
1707
1708         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1709 }
1710
1711 EXPORT_SYMBOL(sock_common_setsockopt);
1712
1713 #ifdef CONFIG_COMPAT
1714 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
1715                                   char __user *optval, int optlen)
1716 {
1717         struct sock *sk = sock->sk;
1718
1719         if (sk->sk_prot->compat_setsockopt != NULL)
1720                 return sk->sk_prot->compat_setsockopt(sk, level, optname,
1721                                                       optval, optlen);
1722         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1723 }
1724 EXPORT_SYMBOL(compat_sock_common_setsockopt);
1725 #endif
1726
1727 void sk_common_release(struct sock *sk)
1728 {
1729         if (sk->sk_prot->destroy)
1730                 sk->sk_prot->destroy(sk);
1731
1732         /*
1733          * Observation: when sock_common_release is called, processes have
1734          * no access to socket. But net still has.
1735          * Step one, detach it from networking:
1736          *
1737          * A. Remove from hash tables.
1738          */
1739
1740         sk->sk_prot->unhash(sk);
1741
1742         /*
1743          * In this point socket cannot receive new packets, but it is possible
1744          * that some packets are in flight because some CPU runs receiver and
1745          * did hash table lookup before we unhashed socket. They will achieve
1746          * receive queue and will be purged by socket destructor.
1747          *
1748          * Also we still have packets pending on receive queue and probably,
1749          * our own packets waiting in device queues. sock_destroy will drain
1750          * receive queue, but transmitted packets will delay socket destruction
1751          * until the last reference will be released.
1752          */
1753
1754         sock_orphan(sk);
1755
1756         xfrm_sk_free_policy(sk);
1757
1758         sk_refcnt_debug_release(sk);
1759         sock_put(sk);
1760 }
1761
1762 EXPORT_SYMBOL(sk_common_release);
1763
1764 static DEFINE_RWLOCK(proto_list_lock);
1765 static LIST_HEAD(proto_list);
1766
1767 int proto_register(struct proto *prot, int alloc_slab)
1768 {
1769         char *request_sock_slab_name = NULL;
1770         char *timewait_sock_slab_name;
1771         int rc = -ENOBUFS;
1772
1773         if (alloc_slab) {
1774                 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
1775                                                SLAB_HWCACHE_ALIGN, NULL);
1776
1777                 if (prot->slab == NULL) {
1778                         printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
1779                                prot->name);
1780                         goto out;
1781                 }
1782
1783                 if (prot->rsk_prot != NULL) {
1784                         static const char mask[] = "request_sock_%s";
1785
1786                         request_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
1787                         if (request_sock_slab_name == NULL)
1788                                 goto out_free_sock_slab;
1789
1790                         sprintf(request_sock_slab_name, mask, prot->name);
1791                         prot->rsk_prot->slab = kmem_cache_create(request_sock_slab_name,
1792                                                                  prot->rsk_prot->obj_size, 0,
1793                                                                  SLAB_HWCACHE_ALIGN, NULL);
1794
1795                         if (prot->rsk_prot->slab == NULL) {
1796                                 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
1797                                        prot->name);
1798                                 goto out_free_request_sock_slab_name;
1799                         }
1800                 }
1801
1802                 if (prot->twsk_prot != NULL) {
1803                         static const char mask[] = "tw_sock_%s";
1804
1805                         timewait_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
1806
1807                         if (timewait_sock_slab_name == NULL)
1808                                 goto out_free_request_sock_slab;
1809
1810                         sprintf(timewait_sock_slab_name, mask, prot->name);
1811                         prot->twsk_prot->twsk_slab =
1812                                 kmem_cache_create(timewait_sock_slab_name,
1813                                                   prot->twsk_prot->twsk_obj_size,
1814                                                   0, SLAB_HWCACHE_ALIGN,
1815                                                   NULL);
1816                         if (prot->twsk_prot->twsk_slab == NULL)
1817                                 goto out_free_timewait_sock_slab_name;
1818                 }
1819         }
1820
1821         write_lock(&proto_list_lock);
1822         list_add(&prot->node, &proto_list);
1823         write_unlock(&proto_list_lock);
1824         rc = 0;
1825 out:
1826         return rc;
1827 out_free_timewait_sock_slab_name:
1828         kfree(timewait_sock_slab_name);
1829 out_free_request_sock_slab:
1830         if (prot->rsk_prot && prot->rsk_prot->slab) {
1831                 kmem_cache_destroy(prot->rsk_prot->slab);
1832                 prot->rsk_prot->slab = NULL;
1833         }
1834 out_free_request_sock_slab_name:
1835         kfree(request_sock_slab_name);
1836 out_free_sock_slab:
1837         kmem_cache_destroy(prot->slab);
1838         prot->slab = NULL;
1839         goto out;
1840 }
1841
1842 EXPORT_SYMBOL(proto_register);
1843
1844 void proto_unregister(struct proto *prot)
1845 {
1846         write_lock(&proto_list_lock);
1847         list_del(&prot->node);
1848         write_unlock(&proto_list_lock);
1849
1850         if (prot->slab != NULL) {
1851                 kmem_cache_destroy(prot->slab);
1852                 prot->slab = NULL;
1853         }
1854
1855         if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
1856                 const char *name = kmem_cache_name(prot->rsk_prot->slab);
1857
1858                 kmem_cache_destroy(prot->rsk_prot->slab);
1859                 kfree(name);
1860                 prot->rsk_prot->slab = NULL;
1861         }
1862
1863         if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
1864                 const char *name = kmem_cache_name(prot->twsk_prot->twsk_slab);
1865
1866                 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
1867                 kfree(name);
1868                 prot->twsk_prot->twsk_slab = NULL;
1869         }
1870 }
1871
1872 EXPORT_SYMBOL(proto_unregister);
1873
1874 #ifdef CONFIG_PROC_FS
1875 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
1876 {
1877         read_lock(&proto_list_lock);
1878         return seq_list_start_head(&proto_list, *pos);
1879 }
1880
1881 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1882 {
1883         return seq_list_next(v, &proto_list, pos);
1884 }
1885
1886 static void proto_seq_stop(struct seq_file *seq, void *v)
1887 {
1888         read_unlock(&proto_list_lock);
1889 }
1890
1891 static char proto_method_implemented(const void *method)
1892 {
1893         return method == NULL ? 'n' : 'y';
1894 }
1895
1896 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
1897 {
1898         seq_printf(seq, "%-9s %4u %6d  %6d   %-3s %6u   %-3s  %-10s "
1899                         "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
1900                    proto->name,
1901                    proto->obj_size,
1902                    proto->sockets_allocated != NULL ? atomic_read(proto->sockets_allocated) : -1,
1903                    proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
1904                    proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
1905                    proto->max_header,
1906                    proto->slab == NULL ? "no" : "yes",
1907                    module_name(proto->owner),
1908                    proto_method_implemented(proto->close),
1909                    proto_method_implemented(proto->connect),
1910                    proto_method_implemented(proto->disconnect),
1911                    proto_method_implemented(proto->accept),
1912                    proto_method_implemented(proto->ioctl),
1913                    proto_method_implemented(proto->init),
1914                    proto_method_implemented(proto->destroy),
1915                    proto_method_implemented(proto->shutdown),
1916                    proto_method_implemented(proto->setsockopt),
1917                    proto_method_implemented(proto->getsockopt),
1918                    proto_method_implemented(proto->sendmsg),
1919                    proto_method_implemented(proto->recvmsg),
1920                    proto_method_implemented(proto->sendpage),
1921                    proto_method_implemented(proto->bind),
1922                    proto_method_implemented(proto->backlog_rcv),
1923                    proto_method_implemented(proto->hash),
1924                    proto_method_implemented(proto->unhash),
1925                    proto_method_implemented(proto->get_port),
1926                    proto_method_implemented(proto->enter_memory_pressure));
1927 }
1928
1929 static int proto_seq_show(struct seq_file *seq, void *v)
1930 {
1931         if (v == &proto_list)
1932                 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
1933                            "protocol",
1934                            "size",
1935                            "sockets",
1936                            "memory",
1937                            "press",
1938                            "maxhdr",
1939                            "slab",
1940                            "module",
1941                            "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
1942         else
1943                 proto_seq_printf(seq, list_entry(v, struct proto, node));
1944         return 0;
1945 }
1946
1947 static const struct seq_operations proto_seq_ops = {
1948         .start  = proto_seq_start,
1949         .next   = proto_seq_next,
1950         .stop   = proto_seq_stop,
1951         .show   = proto_seq_show,
1952 };
1953
1954 static int proto_seq_open(struct inode *inode, struct file *file)
1955 {
1956         return seq_open(file, &proto_seq_ops);
1957 }
1958
1959 static const struct file_operations proto_seq_fops = {
1960         .owner          = THIS_MODULE,
1961         .open           = proto_seq_open,
1962         .read           = seq_read,
1963         .llseek         = seq_lseek,
1964         .release        = seq_release,
1965 };
1966
1967 static int __init proto_init(void)
1968 {
1969         /* register /proc/net/protocols */
1970         return proc_net_fops_create(&init_net, "protocols", S_IRUGO, &proto_seq_fops) == NULL ? -ENOBUFS : 0;
1971 }
1972
1973 subsys_initcall(proto_init);
1974
1975 #endif /* PROC_FS */
1976
1977 EXPORT_SYMBOL(sk_alloc);
1978 EXPORT_SYMBOL(sk_free);
1979 EXPORT_SYMBOL(sk_send_sigurg);
1980 EXPORT_SYMBOL(sock_alloc_send_skb);
1981 EXPORT_SYMBOL(sock_init_data);
1982 EXPORT_SYMBOL(sock_kfree_s);
1983 EXPORT_SYMBOL(sock_kmalloc);
1984 EXPORT_SYMBOL(sock_no_accept);
1985 EXPORT_SYMBOL(sock_no_bind);
1986 EXPORT_SYMBOL(sock_no_connect);
1987 EXPORT_SYMBOL(sock_no_getname);
1988 EXPORT_SYMBOL(sock_no_getsockopt);
1989 EXPORT_SYMBOL(sock_no_ioctl);
1990 EXPORT_SYMBOL(sock_no_listen);
1991 EXPORT_SYMBOL(sock_no_mmap);
1992 EXPORT_SYMBOL(sock_no_poll);
1993 EXPORT_SYMBOL(sock_no_recvmsg);
1994 EXPORT_SYMBOL(sock_no_sendmsg);
1995 EXPORT_SYMBOL(sock_no_sendpage);
1996 EXPORT_SYMBOL(sock_no_setsockopt);
1997 EXPORT_SYMBOL(sock_no_shutdown);
1998 EXPORT_SYMBOL(sock_no_socketpair);
1999 EXPORT_SYMBOL(sock_rfree);
2000 EXPORT_SYMBOL(sock_setsockopt);
2001 EXPORT_SYMBOL(sock_wfree);
2002 EXPORT_SYMBOL(sock_wmalloc);
2003 EXPORT_SYMBOL(sock_i_uid);
2004 EXPORT_SYMBOL(sock_i_ino);
2005 EXPORT_SYMBOL(sysctl_optmem_max);
2006 #ifdef CONFIG_SYSCTL
2007 EXPORT_SYMBOL(sysctl_rmem_max);
2008 EXPORT_SYMBOL(sysctl_wmem_max);
2009 #endif