git.oblomov.eu Git - linux-2.6/blob - net/core/sock.c

   1 /*
   2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   3  *              operating system.  INET is implemented using the  BSD Socket
   4  *              interface as the means of communication with the user level.
   5  *
   6  *              Generic socket support routines. Memory allocators, socket lock/release
   7  *              handler for protocols to use and generic option handler.
   8  *
   9  *
  10  * Version:     $Id: sock.c,v 1.117 2002/02/01 22:01:03 davem Exp $
  11  *
  12  * Authors:     Ross Biro
  13  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  14  *              Florian La Roche, <flla@stud.uni-sb.de>
  15  *              Alan Cox, <A.Cox@swansea.ac.uk>
  16  *
  17  * Fixes:
  18  *              Alan Cox        :       Numerous verify_area() problems
  19  *              Alan Cox        :       Connecting on a connecting socket
  20  *                                      now returns an error for tcp.
  21  *              Alan Cox        :       sock->protocol is set correctly.
  22  *                                      and is not sometimes left as 0.
  23  *              Alan Cox        :       connect handles icmp errors on a
  24  *                                      connect properly. Unfortunately there
  25  *                                      is a restart syscall nasty there. I
  26  *                                      can't match BSD without hacking the C
  27  *                                      library. Ideas urgently sought!
  28  *              Alan Cox        :       Disallow bind() to addresses that are
  29  *                                      not ours - especially broadcast ones!!
  30  *              Alan Cox        :       Socket 1024 _IS_ ok for users. (fencepost)
  31  *              Alan Cox        :       sock_wfree/sock_rfree don't destroy sockets,
  32  *                                      instead they leave that for the DESTROY timer.
  33  *              Alan Cox        :       Clean up error flag in accept
  34  *              Alan Cox        :       TCP ack handling is buggy, the DESTROY timer
  35  *                                      was buggy. Put a remove_sock() in the handler
  36  *                                      for memory when we hit 0. Also altered the timer
  37  *                                      code. The ACK stuff can wait and needs major
  38  *                                      TCP layer surgery.
  39  *              Alan Cox        :       Fixed TCP ack bug, removed remove sock
  40  *                                      and fixed timer/inet_bh race.
  41  *              Alan Cox        :       Added zapped flag for TCP
  42  *              Alan Cox        :       Move kfree_skb into skbuff.c and tidied up surplus code
  43  *              Alan Cox        :       for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
  44  *              Alan Cox        :       kfree_s calls now are kfree_skbmem so we can track skb resources
  45  *              Alan Cox        :       Supports socket option broadcast now as does udp. Packet and raw need fixing.
  46  *              Alan Cox        :       Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
  47  *              Rick Sladkey    :       Relaxed UDP rules for matching packets.
  48  *              C.E.Hawkins     :       IFF_PROMISC/SIOCGHWADDR support
  49  *      Pauline Middelink       :       identd support
  50  *              Alan Cox        :       Fixed connect() taking signals I think.
  51  *              Alan Cox        :       SO_LINGER supported
  52  *              Alan Cox        :       Error reporting fixes
  53  *              Anonymous       :       inet_create tidied up (sk->reuse setting)
  54  *              Alan Cox        :       inet sockets don't set sk->type!
  55  *              Alan Cox        :       Split socket option code
  56  *              Alan Cox        :       Callbacks
  57  *              Alan Cox        :       Nagle flag for Charles & Johannes stuff
  58  *              Alex            :       Removed restriction on inet fioctl
  59  *              Alan Cox        :       Splitting INET from NET core
  60  *              Alan Cox        :       Fixed bogus SO_TYPE handling in getsockopt()
  61  *              Adam Caldwell   :       Missing return in SO_DONTROUTE/SO_DEBUG code
  62  *              Alan Cox        :       Split IP from generic code
  63  *              Alan Cox        :       New kfree_skbmem()
  64  *              Alan Cox        :       Make SO_DEBUG superuser only.
  65  *              Alan Cox        :       Allow anyone to clear SO_DEBUG
  66  *                                      (compatibility fix)
  67  *              Alan Cox        :       Added optimistic memory grabbing for AF_UNIX throughput.
  68  *              Alan Cox        :       Allocator for a socket is settable.
  69  *              Alan Cox        :       SO_ERROR includes soft errors.
  70  *              Alan Cox        :       Allow NULL arguments on some SO_ opts
  71  *              Alan Cox        :       Generic socket allocation to make hooks
  72  *                                      easier (suggested by Craig Metz).
  73  *              Michael Pall    :       SO_ERROR returns positive errno again
  74  *              Steve Whitehouse:       Added default destructor to free
  75  *                                      protocol private data.
  76  *              Steve Whitehouse:       Added various other default routines
  77  *                                      common to several socket families.
  78  *              Chris Evans     :       Call suser() check last on F_SETOWN
  79  *              Jay Schulist    :       Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
  80  *              Andi Kleen      :       Add sock_kmalloc()/sock_kfree_s()
  81  *              Andi Kleen      :       Fix write_space callback
  82  *              Chris Evans     :       Security fixes - signedness again
  83  *              Arnaldo C. Melo :       cleanups, use skb_queue_purge
  84  *
  85  * To Fix:
  86  *
  87  *
  88  *              This program is free software; you can redistribute it and/or
  89  *              modify it under the terms of the GNU General Public License
  90  *              as published by the Free Software Foundation; either version
  91  *              2 of the License, or (at your option) any later version.
  92  */
  93
  94 #include <linux/capability.h>
  95 #include <linux/errno.h>
  96 #include <linux/types.h>
  97 #include <linux/socket.h>
  98 #include <linux/in.h>
  99 #include <linux/kernel.h>
 100 #include <linux/module.h>
 101 #include <linux/proc_fs.h>
 102 #include <linux/seq_file.h>
 103 #include <linux/sched.h>
 104 #include <linux/timer.h>
 105 #include <linux/string.h>
 106 #include <linux/sockios.h>
 107 #include <linux/net.h>
 108 #include <linux/mm.h>
 109 #include <linux/slab.h>
 110 #include <linux/interrupt.h>
 111 #include <linux/poll.h>
 112 #include <linux/tcp.h>
 113 #include <linux/init.h>
 114 #include <linux/highmem.h>
 115
 116 #include <asm/uaccess.h>
 117 #include <asm/system.h>
 118
 119 #include <linux/netdevice.h>
 120 #include <net/protocol.h>
 121 #include <linux/skbuff.h>
 122 #include <net/net_namespace.h>
 123 #include <net/request_sock.h>
 124 #include <net/sock.h>
 125 #include <net/xfrm.h>
 126 #include <linux/ipsec.h>
 127
 128 #include <linux/filter.h>
 129
 130 #ifdef CONFIG_INET
 131 #include <net/tcp.h>
 132 #endif
 133
 134 /*
 135  * Each address family might have different locking rules, so we have
 136  * one slock key per address family:
 137  */
 138 static struct lock_class_key af_family_keys[AF_MAX];
 139 static struct lock_class_key af_family_slock_keys[AF_MAX];
 140
 141 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 142 /*
 143  * Make lock validator output more readable. (we pre-construct these
 144  * strings build-time, so that runtime initialization of socket
 145  * locks is fast):
 146  */
 147 static const char *af_family_key_strings[AF_MAX+1] = {
 148   "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX"     , "sk_lock-AF_INET"     ,
 149   "sk_lock-AF_AX25"  , "sk_lock-AF_IPX"      , "sk_lock-AF_APPLETALK",
 150   "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE"   , "sk_lock-AF_ATMPVC"   ,
 151   "sk_lock-AF_X25"   , "sk_lock-AF_INET6"    , "sk_lock-AF_ROSE"     ,
 152   "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI"  , "sk_lock-AF_SECURITY" ,
 153   "sk_lock-AF_KEY"   , "sk_lock-AF_NETLINK"  , "sk_lock-AF_PACKET"   ,
 154   "sk_lock-AF_ASH"   , "sk_lock-AF_ECONET"   , "sk_lock-AF_ATMSVC"   ,
 155   "sk_lock-21"       , "sk_lock-AF_SNA"      , "sk_lock-AF_IRDA"     ,
 156   "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE"  , "sk_lock-AF_LLC"      ,
 157   "sk_lock-27"       , "sk_lock-28"          , "sk_lock-29"          ,
 158   "sk_lock-AF_TIPC"  , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV"        ,
 159   "sk_lock-AF_RXRPC" , "sk_lock-AF_MAX"
 160 };
 161 static const char *af_family_slock_key_strings[AF_MAX+1] = {
 162   "slock-AF_UNSPEC", "slock-AF_UNIX"     , "slock-AF_INET"     ,
 163   "slock-AF_AX25"  , "slock-AF_IPX"      , "slock-AF_APPLETALK",
 164   "slock-AF_NETROM", "slock-AF_BRIDGE"   , "slock-AF_ATMPVC"   ,
 165   "slock-AF_X25"   , "slock-AF_INET6"    , "slock-AF_ROSE"     ,
 166   "slock-AF_DECnet", "slock-AF_NETBEUI"  , "slock-AF_SECURITY" ,
 167   "slock-AF_KEY"   , "slock-AF_NETLINK"  , "slock-AF_PACKET"   ,
 168   "slock-AF_ASH"   , "slock-AF_ECONET"   , "slock-AF_ATMSVC"   ,
 169   "slock-21"       , "slock-AF_SNA"      , "slock-AF_IRDA"     ,
 170   "slock-AF_PPPOX" , "slock-AF_WANPIPE"  , "slock-AF_LLC"      ,
 171   "slock-27"       , "slock-28"          , "slock-29"          ,
 172   "slock-AF_TIPC"  , "slock-AF_BLUETOOTH", "slock-AF_IUCV"     ,
 173   "slock-AF_RXRPC" , "slock-AF_MAX"
 174 };
 175 static const char *af_family_clock_key_strings[AF_MAX+1] = {
 176   "clock-AF_UNSPEC", "clock-AF_UNIX"     , "clock-AF_INET"     ,
 177   "clock-AF_AX25"  , "clock-AF_IPX"      , "clock-AF_APPLETALK",
 178   "clock-AF_NETROM", "clock-AF_BRIDGE"   , "clock-AF_ATMPVC"   ,
 179   "clock-AF_X25"   , "clock-AF_INET6"    , "clock-AF_ROSE"     ,
 180   "clock-AF_DECnet", "clock-AF_NETBEUI"  , "clock-AF_SECURITY" ,
 181   "clock-AF_KEY"   , "clock-AF_NETLINK"  , "clock-AF_PACKET"   ,
 182   "clock-AF_ASH"   , "clock-AF_ECONET"   , "clock-AF_ATMSVC"   ,
 183   "clock-21"       , "clock-AF_SNA"      , "clock-AF_IRDA"     ,
 184   "clock-AF_PPPOX" , "clock-AF_WANPIPE"  , "clock-AF_LLC"      ,
 185   "clock-27"       , "clock-28"          , "clock-29"          ,
 186   "clock-AF_TIPC"  , "clock-AF_BLUETOOTH", "clock-AF_IUCV"     ,
 187   "clock-AF_RXRPC" , "clock-AF_MAX"
 188 };
 189 #endif
 190
 191 /*
 192  * sk_callback_lock locking rules are per-address-family,
 193  * so split the lock classes by using a per-AF key:
 194  */
 195 static struct lock_class_key af_callback_keys[AF_MAX];
 196
 197 /* Take into consideration the size of the struct sk_buff overhead in the
 198  * determination of these values, since that is non-constant across
 199  * platforms.  This makes socket queueing behavior and performance
 200  * not depend upon such differences.
 201  */
 202 #define _SK_MEM_PACKETS         256
 203 #define _SK_MEM_OVERHEAD        (sizeof(struct sk_buff) + 256)
 204 #define SK_WMEM_MAX             (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
 205 #define SK_RMEM_MAX             (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
 206
 207 /* Run time adjustable parameters. */
 208 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
 209 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
 210 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
 211 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
 212
 213 /* Maximal space eaten by iovec or ancilliary data plus some space */
 214 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
 215
 216 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
 217 {
 218         struct timeval tv;
 219
 220         if (optlen < sizeof(tv))
 221                 return -EINVAL;
 222         if (copy_from_user(&tv, optval, sizeof(tv)))
 223                 return -EFAULT;
 224         if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
 225                 return -EDOM;
 226
 227         if (tv.tv_sec < 0) {
 228                 static int warned __read_mostly;
 229
 230                 *timeo_p = 0;
 231                 if (warned < 10 && net_ratelimit())
 232                         warned++;
 233                         printk(KERN_INFO "sock_set_timeout: `%s' (pid %d) "
 234                                "tries to set negative timeout\n",
 235                                 current->comm, task_pid_nr(current));
 236                 return 0;
 237         }
 238         *timeo_p = MAX_SCHEDULE_TIMEOUT;
 239         if (tv.tv_sec == 0 && tv.tv_usec == 0)
 240                 return 0;
 241         if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
 242                 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
 243         return 0;
 244 }
 245
 246 static void sock_warn_obsolete_bsdism(const char *name)
 247 {
 248         static int warned;
 249         static char warncomm[TASK_COMM_LEN];
 250         if (strcmp(warncomm, current->comm) && warned < 5) {
 251                 strcpy(warncomm,  current->comm);
 252                 printk(KERN_WARNING "process `%s' is using obsolete "
 253                        "%s SO_BSDCOMPAT\n", warncomm, name);
 254                 warned++;
 255         }
 256 }
 257
 258 static void sock_disable_timestamp(struct sock *sk)
 259 {
 260         if (sock_flag(sk, SOCK_TIMESTAMP)) {
 261                 sock_reset_flag(sk, SOCK_TIMESTAMP);
 262                 net_disable_timestamp();
 263         }
 264 }
 265
 266
 267 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
 268 {
 269         int err = 0;
 270         int skb_len;
 271
 272         /* Cast skb->rcvbuf to unsigned... It's pointless, but reduces
 273            number of warnings when compiling with -W --ANK
 274          */
 275         if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
 276             (unsigned)sk->sk_rcvbuf) {
 277                 err = -ENOMEM;
 278                 goto out;
 279         }
 280
 281         err = sk_filter(sk, skb);
 282         if (err)
 283                 goto out;
 284
 285         skb->dev = NULL;
 286         skb_set_owner_r(skb, sk);
 287
 288         /* Cache the SKB length before we tack it onto the receive
 289          * queue.  Once it is added it no longer belongs to us and
 290          * may be freed by other threads of control pulling packets
 291          * from the queue.
 292          */
 293         skb_len = skb->len;
 294
 295         skb_queue_tail(&sk->sk_receive_queue, skb);
 296
 297         if (!sock_flag(sk, SOCK_DEAD))
 298                 sk->sk_data_ready(sk, skb_len);
 299 out:
 300         return err;
 301 }
 302 EXPORT_SYMBOL(sock_queue_rcv_skb);
 303
 304 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
 305 {
 306         int rc = NET_RX_SUCCESS;
 307
 308         if (sk_filter(sk, skb))
 309                 goto discard_and_relse;
 310
 311         skb->dev = NULL;
 312
 313         if (nested)
 314                 bh_lock_sock_nested(sk);
 315         else
 316                 bh_lock_sock(sk);
 317         if (!sock_owned_by_user(sk)) {
 318                 /*
 319                  * trylock + unlock semantics:
 320                  */
 321                 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
 322
 323                 rc = sk->sk_backlog_rcv(sk, skb);
 324
 325                 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
 326         } else
 327                 sk_add_backlog(sk, skb);
 328         bh_unlock_sock(sk);
 329 out:
 330         sock_put(sk);
 331         return rc;
 332 discard_and_relse:
 333         kfree_skb(skb);
 334         goto out;
 335 }
 336 EXPORT_SYMBOL(sk_receive_skb);
 337
 338 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
 339 {
 340         struct dst_entry *dst = sk->sk_dst_cache;
 341
 342         if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
 343                 sk->sk_dst_cache = NULL;
 344                 dst_release(dst);
 345                 return NULL;
 346         }
 347
 348         return dst;
 349 }
 350 EXPORT_SYMBOL(__sk_dst_check);
 351
 352 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
 353 {
 354         struct dst_entry *dst = sk_dst_get(sk);
 355
 356         if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
 357                 sk_dst_reset(sk);
 358                 dst_release(dst);
 359                 return NULL;
 360         }
 361
 362         return dst;
 363 }
 364 EXPORT_SYMBOL(sk_dst_check);
 365
 366 static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
 367 {
 368         int ret = -ENOPROTOOPT;
 369 #ifdef CONFIG_NETDEVICES
 370         struct net *net = sk->sk_net;
 371         char devname[IFNAMSIZ];
 372         int index;
 373
 374         /* Sorry... */
 375         ret = -EPERM;
 376         if (!capable(CAP_NET_RAW))
 377                 goto out;
 378
 379         ret = -EINVAL;
 380         if (optlen < 0)
 381                 goto out;
 382
 383         /* Bind this socket to a particular device like "eth0",
 384          * as specified in the passed interface name. If the
 385          * name is "" or the option length is zero the socket
 386          * is not bound.
 387          */
 388         if (optlen > IFNAMSIZ - 1)
 389                 optlen = IFNAMSIZ - 1;
 390         memset(devname, 0, sizeof(devname));
 391
 392         ret = -EFAULT;
 393         if (copy_from_user(devname, optval, optlen))
 394                 goto out;
 395
 396         if (devname[0] == '\0') {
 397                 index = 0;
 398         } else {
 399                 struct net_device *dev = dev_get_by_name(net, devname);
 400
 401                 ret = -ENODEV;
 402                 if (!dev)
 403                         goto out;
 404
 405                 index = dev->ifindex;
 406                 dev_put(dev);
 407         }
 408
 409         lock_sock(sk);
 410         sk->sk_bound_dev_if = index;
 411         sk_dst_reset(sk);
 412         release_sock(sk);
 413
 414         ret = 0;
 415
 416 out:
 417 #endif
 418
 419         return ret;
 420 }
 421
 422 /*
 423  *      This is meant for all protocols to use and covers goings on
 424  *      at the socket level. Everything here is generic.
 425  */
 426
 427 int sock_setsockopt(struct socket *sock, int level, int optname,
 428                     char __user *optval, int optlen)
 429 {
 430         struct sock *sk=sock->sk;
 431         int val;
 432         int valbool;
 433         struct linger ling;
 434         int ret = 0;
 435
 436         /*
 437          *      Options without arguments
 438          */
 439
 440 #ifdef SO_DONTLINGER            /* Compatibility item... */
 441         if (optname == SO_DONTLINGER) {
 442                 lock_sock(sk);
 443                 sock_reset_flag(sk, SOCK_LINGER);
 444                 release_sock(sk);
 445                 return 0;
 446         }
 447 #endif
 448
 449         if (optname == SO_BINDTODEVICE)
 450                 return sock_bindtodevice(sk, optval, optlen);
 451
 452         if (optlen < sizeof(int))
 453                 return -EINVAL;
 454
 455         if (get_user(val, (int __user *)optval))
 456                 return -EFAULT;
 457
 458         valbool = val?1:0;
 459
 460         lock_sock(sk);
 461
 462         switch(optname) {
 463         case SO_DEBUG:
 464                 if (val && !capable(CAP_NET_ADMIN)) {
 465                         ret = -EACCES;
 466                 }
 467                 else if (valbool)
 468                         sock_set_flag(sk, SOCK_DBG);
 469                 else
 470                         sock_reset_flag(sk, SOCK_DBG);
 471                 break;
 472         case SO_REUSEADDR:
 473                 sk->sk_reuse = valbool;
 474                 break;
 475         case SO_TYPE:
 476         case SO_ERROR:
 477                 ret = -ENOPROTOOPT;
 478                 break;
 479         case SO_DONTROUTE:
 480                 if (valbool)
 481                         sock_set_flag(sk, SOCK_LOCALROUTE);
 482                 else
 483                         sock_reset_flag(sk, SOCK_LOCALROUTE);
 484                 break;
 485         case SO_BROADCAST:
 486                 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
 487                 break;
 488         case SO_SNDBUF:
 489                 /* Don't error on this BSD doesn't and if you think
 490                    about it this is right. Otherwise apps have to
 491                    play 'guess the biggest size' games. RCVBUF/SNDBUF
 492                    are treated in BSD as hints */
 493
 494                 if (val > sysctl_wmem_max)
 495                         val = sysctl_wmem_max;
 496 set_sndbuf:
 497                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
 498                 if ((val * 2) < SOCK_MIN_SNDBUF)
 499                         sk->sk_sndbuf = SOCK_MIN_SNDBUF;
 500                 else
 501                         sk->sk_sndbuf = val * 2;
 502
 503                 /*
 504                  *      Wake up sending tasks if we
 505                  *      upped the value.
 506                  */
 507                 sk->sk_write_space(sk);
 508                 break;
 509
 510         case SO_SNDBUFFORCE:
 511                 if (!capable(CAP_NET_ADMIN)) {
 512                         ret = -EPERM;
 513                         break;
 514                 }
 515                 goto set_sndbuf;
 516
 517         case SO_RCVBUF:
 518                 /* Don't error on this BSD doesn't and if you think
 519                    about it this is right. Otherwise apps have to
 520                    play 'guess the biggest size' games. RCVBUF/SNDBUF
 521                    are treated in BSD as hints */
 522
 523                 if (val > sysctl_rmem_max)
 524                         val = sysctl_rmem_max;
 525 set_rcvbuf:
 526                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
 527                 /*
 528                  * We double it on the way in to account for
 529                  * "struct sk_buff" etc. overhead.   Applications
 530                  * assume that the SO_RCVBUF setting they make will
 531                  * allow that much actual data to be received on that
 532                  * socket.
 533                  *
 534                  * Applications are unaware that "struct sk_buff" and
 535                  * other overheads allocate from the receive buffer
 536                  * during socket buffer allocation.
 537                  *
 538                  * And after considering the possible alternatives,
 539                  * returning the value we actually used in getsockopt
 540                  * is the most desirable behavior.
 541                  */
 542                 if ((val * 2) < SOCK_MIN_RCVBUF)
 543                         sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
 544                 else
 545                         sk->sk_rcvbuf = val * 2;
 546                 break;
 547
 548         case SO_RCVBUFFORCE:
 549                 if (!capable(CAP_NET_ADMIN)) {
 550                         ret = -EPERM;
 551                         break;
 552                 }
 553                 goto set_rcvbuf;
 554
 555         case SO_KEEPALIVE:
 556 #ifdef CONFIG_INET
 557                 if (sk->sk_protocol == IPPROTO_TCP)
 558                         tcp_set_keepalive(sk, valbool);
 559 #endif
 560                 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
 561                 break;
 562
 563         case SO_OOBINLINE:
 564                 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
 565                 break;
 566
 567         case SO_NO_CHECK:
 568                 sk->sk_no_check = valbool;
 569                 break;
 570
 571         case SO_PRIORITY:
 572                 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
 573                         sk->sk_priority = val;
 574                 else
 575                         ret = -EPERM;
 576                 break;
 577
 578         case SO_LINGER:
 579                 if (optlen < sizeof(ling)) {
 580                         ret = -EINVAL;  /* 1003.1g */
 581                         break;
 582                 }
 583                 if (copy_from_user(&ling,optval,sizeof(ling))) {
 584                         ret = -EFAULT;
 585                         break;
 586                 }
 587                 if (!ling.l_onoff)
 588                         sock_reset_flag(sk, SOCK_LINGER);
 589                 else {
 590 #if (BITS_PER_LONG == 32)
 591                         if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
 592                                 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
 593                         else
 594 #endif
 595                                 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
 596                         sock_set_flag(sk, SOCK_LINGER);
 597                 }
 598                 break;
 599
 600         case SO_BSDCOMPAT:
 601                 sock_warn_obsolete_bsdism("setsockopt");
 602                 break;
 603
 604         case SO_PASSCRED:
 605                 if (valbool)
 606                         set_bit(SOCK_PASSCRED, &sock->flags);
 607                 else
 608                         clear_bit(SOCK_PASSCRED, &sock->flags);
 609                 break;
 610
 611         case SO_TIMESTAMP:
 612         case SO_TIMESTAMPNS:
 613                 if (valbool)  {
 614                         if (optname == SO_TIMESTAMP)
 615                                 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
 616                         else
 617                                 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
 618                         sock_set_flag(sk, SOCK_RCVTSTAMP);
 619                         sock_enable_timestamp(sk);
 620                 } else {
 621                         sock_reset_flag(sk, SOCK_RCVTSTAMP);
 622                         sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
 623                 }
 624                 break;
 625
 626         case SO_RCVLOWAT:
 627                 if (val < 0)
 628                         val = INT_MAX;
 629                 sk->sk_rcvlowat = val ? : 1;
 630                 break;
 631
 632         case SO_RCVTIMEO:
 633                 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
 634                 break;
 635
 636         case SO_SNDTIMEO:
 637                 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
 638                 break;
 639
 640         case SO_ATTACH_FILTER:
 641                 ret = -EINVAL;
 642                 if (optlen == sizeof(struct sock_fprog)) {
 643                         struct sock_fprog fprog;
 644
 645                         ret = -EFAULT;
 646                         if (copy_from_user(&fprog, optval, sizeof(fprog)))
 647                                 break;
 648
 649                         ret = sk_attach_filter(&fprog, sk);
 650                 }
 651                 break;
 652
 653         case SO_DETACH_FILTER:
 654                 ret = sk_detach_filter(sk);
 655                 break;
 656
 657         case SO_PASSSEC:
 658                 if (valbool)
 659                         set_bit(SOCK_PASSSEC, &sock->flags);
 660                 else
 661                         clear_bit(SOCK_PASSSEC, &sock->flags);
 662                 break;
 663
 664                 /* We implement the SO_SNDLOWAT etc to
 665                    not be settable (1003.1g 5.3) */
 666         default:
 667                 ret = -ENOPROTOOPT;
 668                 break;
 669         }
 670         release_sock(sk);
 671         return ret;
 672 }
 673
 674
 675 int sock_getsockopt(struct socket *sock, int level, int optname,
 676                     char __user *optval, int __user *optlen)
 677 {
 678         struct sock *sk = sock->sk;
 679
 680         union {
 681                 int val;
 682                 struct linger ling;
 683                 struct timeval tm;
 684         } v;
 685
 686         unsigned int lv = sizeof(int);
 687         int len;
 688
 689         if (get_user(len, optlen))
 690                 return -EFAULT;
 691         if (len < 0)
 692                 return -EINVAL;
 693
 694         switch(optname) {
 695         case SO_DEBUG:
 696                 v.val = sock_flag(sk, SOCK_DBG);
 697                 break;
 698
 699         case SO_DONTROUTE:
 700                 v.val = sock_flag(sk, SOCK_LOCALROUTE);
 701                 break;
 702
 703         case SO_BROADCAST:
 704                 v.val = !!sock_flag(sk, SOCK_BROADCAST);
 705                 break;
 706
 707         case SO_SNDBUF:
 708                 v.val = sk->sk_sndbuf;
 709                 break;
 710
 711         case SO_RCVBUF:
 712                 v.val = sk->sk_rcvbuf;
 713                 break;
 714
 715         case SO_REUSEADDR:
 716                 v.val = sk->sk_reuse;
 717                 break;
 718
 719         case SO_KEEPALIVE:
 720                 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
 721                 break;
 722
 723         case SO_TYPE:
 724                 v.val = sk->sk_type;
 725                 break;
 726
 727         case SO_ERROR:
 728                 v.val = -sock_error(sk);
 729                 if (v.val==0)
 730                         v.val = xchg(&sk->sk_err_soft, 0);
 731                 break;
 732
 733         case SO_OOBINLINE:
 734                 v.val = !!sock_flag(sk, SOCK_URGINLINE);
 735                 break;
 736
 737         case SO_NO_CHECK:
 738                 v.val = sk->sk_no_check;
 739                 break;
 740
 741         case SO_PRIORITY:
 742                 v.val = sk->sk_priority;
 743                 break;
 744
 745         case SO_LINGER:
 746                 lv              = sizeof(v.ling);
 747                 v.ling.l_onoff  = !!sock_flag(sk, SOCK_LINGER);
 748                 v.ling.l_linger = sk->sk_lingertime / HZ;
 749                 break;
 750
 751         case SO_BSDCOMPAT:
 752                 sock_warn_obsolete_bsdism("getsockopt");
 753                 break;
 754
 755         case SO_TIMESTAMP:
 756                 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
 757                                 !sock_flag(sk, SOCK_RCVTSTAMPNS);
 758                 break;
 759
 760         case SO_TIMESTAMPNS:
 761                 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
 762                 break;
 763
 764         case SO_RCVTIMEO:
 765                 lv=sizeof(struct timeval);
 766                 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
 767                         v.tm.tv_sec = 0;
 768                         v.tm.tv_usec = 0;
 769                 } else {
 770                         v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
 771                         v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
 772                 }
 773                 break;
 774
 775         case SO_SNDTIMEO:
 776                 lv=sizeof(struct timeval);
 777                 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
 778                         v.tm.tv_sec = 0;
 779                         v.tm.tv_usec = 0;
 780                 } else {
 781                         v.tm.tv_sec = sk->sk_sndtimeo / HZ;
 782                         v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
 783                 }
 784                 break;
 785
 786         case SO_RCVLOWAT:
 787                 v.val = sk->sk_rcvlowat;
 788                 break;
 789
 790         case SO_SNDLOWAT:
 791                 v.val=1;
 792                 break;
 793
 794         case SO_PASSCRED:
 795                 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
 796                 break;
 797
 798         case SO_PEERCRED:
 799                 if (len > sizeof(sk->sk_peercred))
 800                         len = sizeof(sk->sk_peercred);
 801                 if (copy_to_user(optval, &sk->sk_peercred, len))
 802                         return -EFAULT;
 803                 goto lenout;
 804
 805         case SO_PEERNAME:
 806         {
 807                 char address[128];
 808
 809                 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
 810                         return -ENOTCONN;
 811                 if (lv < len)
 812                         return -EINVAL;
 813                 if (copy_to_user(optval, address, len))
 814                         return -EFAULT;
 815                 goto lenout;
 816         }
 817
 818         /* Dubious BSD thing... Probably nobody even uses it, but
 819          * the UNIX standard wants it for whatever reason... -DaveM
 820          */
 821         case SO_ACCEPTCONN:
 822                 v.val = sk->sk_state == TCP_LISTEN;
 823                 break;
 824
 825         case SO_PASSSEC:
 826                 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
 827                 break;
 828
 829         case SO_PEERSEC:
 830                 return security_socket_getpeersec_stream(sock, optval, optlen, len);
 831
 832         default:
 833                 return -ENOPROTOOPT;
 834         }
 835
 836         if (len > lv)
 837                 len = lv;
 838         if (copy_to_user(optval, &v, len))
 839                 return -EFAULT;
 840 lenout:
 841         if (put_user(len, optlen))
 842                 return -EFAULT;
 843         return 0;
 844 }
 845
 846 /*
 847  * Initialize an sk_lock.
 848  *
 849  * (We also register the sk_lock with the lock validator.)
 850  */
 851 static inline void sock_lock_init(struct sock *sk)
 852 {
 853         sock_lock_init_class_and_name(sk,
 854                         af_family_slock_key_strings[sk->sk_family],
 855                         af_family_slock_keys + sk->sk_family,
 856                         af_family_key_strings[sk->sk_family],
 857                         af_family_keys + sk->sk_family);
 858 }
 859
 860 static void sock_copy(struct sock *nsk, const struct sock *osk)
 861 {
 862 #ifdef CONFIG_SECURITY_NETWORK
 863         void *sptr = nsk->sk_security;
 864 #endif
 865
 866         memcpy(nsk, osk, osk->sk_prot->obj_size);
 867 #ifdef CONFIG_SECURITY_NETWORK
 868         nsk->sk_security = sptr;
 869         security_sk_clone(osk, nsk);
 870 #endif
 871 }
 872
 873 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
 874                 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)
 931 {
 932         struct sock *sk;
 933
 934         sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
 935         if (sk) {
 936                 sk->sk_family = family;
 937                 /*
 938                  * See comment in struct sock definition to understand
 939                  * why we need sk_prot_creator -acme
 940                  */
 941                 sk->sk_prot = sk->sk_prot_creator = prot;
 942                 sock_lock_init(sk);
 943                 sk->sk_net = get_net(net);
 944         }
 945
 946         return sk;
 947 }
 948
 949 void sk_free(struct sock *sk)
 950 {
 951         struct sk_filter *filter;
 952
 953         if (sk->sk_destruct)
 954                 sk->sk_destruct(sk);
 955
 956         filter = rcu_dereference(sk->sk_filter);
 957         if (filter) {
 958                 sk_filter_uncharge(sk, filter);
 959                 rcu_assign_pointer(sk->sk_filter, NULL);
 960         }
 961
 962         sock_disable_timestamp(sk);
 963
 964         if (atomic_read(&sk->sk_omem_alloc))
 965                 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
 966                        __FUNCTION__, atomic_read(&sk->sk_omem_alloc));
 967
 968         put_net(sk->sk_net);
 969         sk_prot_free(sk->sk_prot_creator, sk);
 970 }
 971
 972 struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
 973 {
 974         struct sock *newsk;
 975
 976         newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
 977         if (newsk != NULL) {
 978                 struct sk_filter *filter;
 979
 980                 sock_copy(newsk, sk);
 981
 982                 /* SANITY */
 983                 get_net(newsk->sk_net);
 984                 sk_node_init(&newsk->sk_node);
 985                 sock_lock_init(newsk);
 986                 bh_lock_sock(newsk);
 987                 newsk->sk_backlog.head  = newsk->sk_backlog.tail = NULL;
 988
 989                 atomic_set(&newsk->sk_rmem_alloc, 0);
 990                 atomic_set(&newsk->sk_wmem_alloc, 0);
 991                 atomic_set(&newsk->sk_omem_alloc, 0);
 992                 skb_queue_head_init(&newsk->sk_receive_queue);
 993                 skb_queue_head_init(&newsk->sk_write_queue);
 994 #ifdef CONFIG_NET_DMA
 995                 skb_queue_head_init(&newsk->sk_async_wait_queue);
 996 #endif
 997
 998                 rwlock_init(&newsk->sk_dst_lock);
 999                 rwlock_init(&newsk->sk_callback_lock);
1000                 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1001                                 af_callback_keys + newsk->sk_family,
1002                                 af_family_clock_key_strings[newsk->sk_family]);
1003
1004                 newsk->sk_dst_cache     = NULL;
1005                 newsk->sk_wmem_queued   = 0;
1006                 newsk->sk_forward_alloc = 0;
1007                 newsk->sk_send_head     = NULL;
1008                 newsk->sk_userlocks     = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1009
1010                 sock_reset_flag(newsk, SOCK_DONE);
1011                 skb_queue_head_init(&newsk->sk_error_queue);
1012
1013                 filter = newsk->sk_filter;
1014                 if (filter != NULL)
1015                         sk_filter_charge(newsk, filter);
1016
1017                 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1018                         /* It is still raw copy of parent, so invalidate
1019                          * destructor and make plain sk_free() */
1020                         newsk->sk_destruct = NULL;
1021                         sk_free(newsk);
1022                         newsk = NULL;
1023                         goto out;
1024                 }
1025
1026                 newsk->sk_err      = 0;
1027                 newsk->sk_priority = 0;
1028                 atomic_set(&newsk->sk_refcnt, 2);
1029
1030                 /*
1031                  * Increment the counter in the same struct proto as the master
1032                  * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1033                  * is the same as sk->sk_prot->socks, as this field was copied
1034                  * with memcpy).
1035                  *
1036                  * This _changes_ the previous behaviour, where
1037                  * tcp_create_openreq_child always was incrementing the
1038                  * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1039                  * to be taken into account in all callers. -acme
1040                  */
1041                 sk_refcnt_debug_inc(newsk);
1042                 newsk->sk_socket = NULL;
1043                 newsk->sk_sleep  = NULL;
1044
1045                 if (newsk->sk_prot->sockets_allocated)
1046                         atomic_inc(newsk->sk_prot->sockets_allocated);
1047         }
1048 out:
1049         return newsk;
1050 }
1051
1052 EXPORT_SYMBOL_GPL(sk_clone);
1053
1054 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1055 {
1056         __sk_dst_set(sk, dst);
1057         sk->sk_route_caps = dst->dev->features;
1058         if (sk->sk_route_caps & NETIF_F_GSO)
1059                 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1060         if (sk_can_gso(sk)) {
1061                 if (dst->header_len)
1062                         sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1063                 else
1064                         sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1065         }
1066 }
1067 EXPORT_SYMBOL_GPL(sk_setup_caps);
1068
1069 void __init sk_init(void)
1070 {
1071         if (num_physpages <= 4096) {
1072                 sysctl_wmem_max = 32767;
1073                 sysctl_rmem_max = 32767;
1074                 sysctl_wmem_default = 32767;
1075                 sysctl_rmem_default = 32767;
1076         } else if (num_physpages >= 131072) {
1077                 sysctl_wmem_max = 131071;
1078                 sysctl_rmem_max = 131071;
1079         }
1080 }
1081
1082 /*
1083  *      Simple resource managers for sockets.
1084  */
1085
1086
1087 /*
1088  * Write buffer destructor automatically called from kfree_skb.
1089  */
1090 void sock_wfree(struct sk_buff *skb)
1091 {
1092         struct sock *sk = skb->sk;
1093
1094         /* In case it might be waiting for more memory. */
1095         atomic_sub(skb->truesize, &sk->sk_wmem_alloc);
1096         if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE))
1097                 sk->sk_write_space(sk);
1098         sock_put(sk);
1099 }
1100
1101 /*
1102  * Read buffer destructor automatically called from kfree_skb.
1103  */
1104 void sock_rfree(struct sk_buff *skb)
1105 {
1106         struct sock *sk = skb->sk;
1107
1108         atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1109 }
1110
1111
1112 int sock_i_uid(struct sock *sk)
1113 {
1114         int uid;
1115
1116         read_lock(&sk->sk_callback_lock);
1117         uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1118         read_unlock(&sk->sk_callback_lock);
1119         return uid;
1120 }
1121
1122 unsigned long sock_i_ino(struct sock *sk)
1123 {
1124         unsigned long ino;
1125
1126         read_lock(&sk->sk_callback_lock);
1127         ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1128         read_unlock(&sk->sk_callback_lock);
1129         return ino;
1130 }
1131
1132 /*
1133  * Allocate a skb from the socket's send buffer.
1134  */
1135 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1136                              gfp_t priority)
1137 {
1138         if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1139                 struct sk_buff * skb = alloc_skb(size, priority);
1140                 if (skb) {
1141                         skb_set_owner_w(skb, sk);
1142                         return skb;
1143                 }
1144         }
1145         return NULL;
1146 }
1147
1148 /*
1149  * Allocate a skb from the socket's receive buffer.
1150  */
1151 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1152                              gfp_t priority)
1153 {
1154         if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1155                 struct sk_buff *skb = alloc_skb(size, priority);
1156                 if (skb) {
1157                         skb_set_owner_r(skb, sk);
1158                         return skb;
1159                 }
1160         }
1161         return NULL;
1162 }
1163
1164 /*
1165  * Allocate a memory block from the socket's option memory buffer.
1166  */
1167 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1168 {
1169         if ((unsigned)size <= sysctl_optmem_max &&
1170             atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1171                 void *mem;
1172                 /* First do the add, to avoid the race if kmalloc
1173                  * might sleep.
1174                  */
1175                 atomic_add(size, &sk->sk_omem_alloc);
1176                 mem = kmalloc(size, priority);
1177                 if (mem)
1178                         return mem;
1179                 atomic_sub(size, &sk->sk_omem_alloc);
1180         }
1181         return NULL;
1182 }
1183
1184 /*
1185  * Free an option memory block.
1186  */
1187 void sock_kfree_s(struct sock *sk, void *mem, int size)
1188 {
1189         kfree(mem);
1190         atomic_sub(size, &sk->sk_omem_alloc);
1191 }
1192
1193 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1194    I think, these locks should be removed for datagram sockets.
1195  */
1196 static long sock_wait_for_wmem(struct sock * sk, long timeo)
1197 {
1198         DEFINE_WAIT(wait);
1199
1200         clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1201         for (;;) {
1202                 if (!timeo)
1203                         break;
1204                 if (signal_pending(current))
1205                         break;
1206                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1207                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1208                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1209                         break;
1210                 if (sk->sk_shutdown & SEND_SHUTDOWN)
1211                         break;
1212                 if (sk->sk_err)
1213                         break;
1214                 timeo = schedule_timeout(timeo);
1215         }
1216         finish_wait(sk->sk_sleep, &wait);
1217         return timeo;
1218 }
1219
1220
1221 /*
1222  *      Generic send/receive buffer handlers
1223  */
1224
1225 static struct sk_buff *sock_alloc_send_pskb(struct sock *sk,
1226                                             unsigned long header_len,
1227                                             unsigned long data_len,
1228                                             int noblock, int *errcode)
1229 {
1230         struct sk_buff *skb;
1231         gfp_t gfp_mask;
1232         long timeo;
1233         int err;
1234
1235         gfp_mask = sk->sk_allocation;
1236         if (gfp_mask & __GFP_WAIT)
1237                 gfp_mask |= __GFP_REPEAT;
1238
1239         timeo = sock_sndtimeo(sk, noblock);
1240         while (1) {
1241                 err = sock_error(sk);
1242                 if (err != 0)
1243                         goto failure;
1244
1245                 err = -EPIPE;
1246                 if (sk->sk_shutdown & SEND_SHUTDOWN)
1247                         goto failure;
1248
1249                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1250                         skb = alloc_skb(header_len, gfp_mask);
1251                         if (skb) {
1252                                 int npages;
1253                                 int i;
1254
1255                                 /* No pages, we're done... */
1256                                 if (!data_len)
1257                                         break;
1258
1259                                 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1260                                 skb->truesize += data_len;
1261                                 skb_shinfo(skb)->nr_frags = npages;
1262                                 for (i = 0; i < npages; i++) {
1263                                         struct page *page;
1264                                         skb_frag_t *frag;
1265
1266                                         page = alloc_pages(sk->sk_allocation, 0);
1267                                         if (!page) {
1268                                                 err = -ENOBUFS;
1269                                                 skb_shinfo(skb)->nr_frags = i;
1270                                                 kfree_skb(skb);
1271                                                 goto failure;
1272                                         }
1273
1274                                         frag = &skb_shinfo(skb)->frags[i];
1275                                         frag->page = page;
1276                                         frag->page_offset = 0;
1277                                         frag->size = (data_len >= PAGE_SIZE ?
1278                                                       PAGE_SIZE :
1279                                                       data_len);
1280                                         data_len -= PAGE_SIZE;
1281                                 }
1282
1283                                 /* Full success... */
1284                                 break;
1285                         }
1286                         err = -ENOBUFS;
1287                         goto failure;
1288                 }
1289                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1290                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1291                 err = -EAGAIN;
1292                 if (!timeo)
1293                         goto failure;
1294                 if (signal_pending(current))
1295                         goto interrupted;
1296                 timeo = sock_wait_for_wmem(sk, timeo);
1297         }
1298
1299         skb_set_owner_w(skb, sk);
1300         return skb;
1301
1302 interrupted:
1303         err = sock_intr_errno(timeo);
1304 failure:
1305         *errcode = err;
1306         return NULL;
1307 }
1308
1309 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1310                                     int noblock, int *errcode)
1311 {
1312         return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1313 }
1314
1315 static void __lock_sock(struct sock *sk)
1316 {
1317         DEFINE_WAIT(wait);
1318
1319         for (;;) {
1320                 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1321                                         TASK_UNINTERRUPTIBLE);
1322                 spin_unlock_bh(&sk->sk_lock.slock);
1323                 schedule();
1324                 spin_lock_bh(&sk->sk_lock.slock);
1325                 if (!sock_owned_by_user(sk))
1326                         break;
1327         }
1328         finish_wait(&sk->sk_lock.wq, &wait);
1329 }
1330
1331 static void __release_sock(struct sock *sk)
1332 {
1333         struct sk_buff *skb = sk->sk_backlog.head;
1334
1335         do {
1336                 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1337                 bh_unlock_sock(sk);
1338
1339                 do {
1340                         struct sk_buff *next = skb->next;
1341
1342                         skb->next = NULL;
1343                         sk->sk_backlog_rcv(sk, skb);
1344
1345                         /*
1346                          * We are in process context here with softirqs
1347                          * disabled, use cond_resched_softirq() to preempt.
1348                          * This is safe to do because we've taken the backlog
1349                          * queue private:
1350                          */
1351                         cond_resched_softirq();
1352
1353                         skb = next;
1354                 } while (skb != NULL);
1355
1356                 bh_lock_sock(sk);
1357         } while ((skb = sk->sk_backlog.head) != NULL);
1358 }
1359
1360 /**
1361  * sk_wait_data - wait for data to arrive at sk_receive_queue
1362  * @sk:    sock to wait on
1363  * @timeo: for how long
1364  *
1365  * Now socket state including sk->sk_err is changed only under lock,
1366  * hence we may omit checks after joining wait queue.
1367  * We check receive queue before schedule() only as optimization;
1368  * it is very likely that release_sock() added new data.
1369  */
1370 int sk_wait_data(struct sock *sk, long *timeo)
1371 {
1372         int rc;
1373         DEFINE_WAIT(wait);
1374
1375         prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1376         set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1377         rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1378         clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1379         finish_wait(sk->sk_sleep, &wait);
1380         return rc;
1381 }
1382
1383 EXPORT_SYMBOL(sk_wait_data);
1384
1385 /*
1386  * Set of default routines for initialising struct proto_ops when
1387  * the protocol does not support a particular function. In certain
1388  * cases where it makes no sense for a protocol to have a "do nothing"
1389  * function, some default processing is provided.
1390  */
1391
1392 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1393 {
1394         return -EOPNOTSUPP;
1395 }
1396
1397 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1398                     int len, int flags)
1399 {
1400         return -EOPNOTSUPP;
1401 }
1402
1403 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1404 {
1405         return -EOPNOTSUPP;
1406 }
1407
1408 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1409 {
1410         return -EOPNOTSUPP;
1411 }
1412
1413 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1414                     int *len, int peer)
1415 {
1416         return -EOPNOTSUPP;
1417 }
1418
1419 unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt)
1420 {
1421         return 0;
1422 }
1423
1424 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1425 {
1426         return -EOPNOTSUPP;
1427 }
1428
1429 int sock_no_listen(struct socket *sock, int backlog)
1430 {
1431         return -EOPNOTSUPP;
1432 }
1433
1434 int sock_no_shutdown(struct socket *sock, int how)
1435 {
1436         return -EOPNOTSUPP;
1437 }
1438
1439 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1440                     char __user *optval, int optlen)
1441 {
1442         return -EOPNOTSUPP;
1443 }
1444
1445 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1446                     char __user *optval, int __user *optlen)
1447 {
1448         return -EOPNOTSUPP;
1449 }
1450
1451 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1452                     size_t len)
1453 {
1454         return -EOPNOTSUPP;
1455 }
1456
1457 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1458                     size_t len, int flags)
1459 {
1460         return -EOPNOTSUPP;
1461 }
1462
1463 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1464 {
1465         /* Mirror missing mmap method error code */
1466         return -ENODEV;
1467 }
1468
1469 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1470 {
1471         ssize_t res;
1472         struct msghdr msg = {.msg_flags = flags};
1473         struct kvec iov;
1474         char *kaddr = kmap(page);
1475         iov.iov_base = kaddr + offset;
1476         iov.iov_len = size;
1477         res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1478         kunmap(page);
1479         return res;
1480 }
1481
1482 /*
1483  *      Default Socket Callbacks
1484  */
1485
1486 static void sock_def_wakeup(struct sock *sk)
1487 {
1488         read_lock(&sk->sk_callback_lock);
1489         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1490                 wake_up_interruptible_all(sk->sk_sleep);
1491         read_unlock(&sk->sk_callback_lock);
1492 }
1493
1494 static void sock_def_error_report(struct sock *sk)
1495 {
1496         read_lock(&sk->sk_callback_lock);
1497         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1498                 wake_up_interruptible(sk->sk_sleep);
1499         sk_wake_async(sk,0,POLL_ERR);
1500         read_unlock(&sk->sk_callback_lock);
1501 }
1502
1503 static void sock_def_readable(struct sock *sk, int len)
1504 {
1505         read_lock(&sk->sk_callback_lock);
1506         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1507                 wake_up_interruptible(sk->sk_sleep);
1508         sk_wake_async(sk,1,POLL_IN);
1509         read_unlock(&sk->sk_callback_lock);
1510 }
1511
1512 static void sock_def_write_space(struct sock *sk)
1513 {
1514         read_lock(&sk->sk_callback_lock);
1515
1516         /* Do not wake up a writer until he can make "significant"
1517          * progress.  --DaveM
1518          */
1519         if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1520                 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1521                         wake_up_interruptible(sk->sk_sleep);
1522
1523                 /* Should agree with poll, otherwise some programs break */
1524                 if (sock_writeable(sk))
1525                         sk_wake_async(sk, 2, POLL_OUT);
1526         }
1527
1528         read_unlock(&sk->sk_callback_lock);
1529 }
1530
1531 static void sock_def_destruct(struct sock *sk)
1532 {
1533         kfree(sk->sk_protinfo);
1534 }
1535
1536 void sk_send_sigurg(struct sock *sk)
1537 {
1538         if (sk->sk_socket && sk->sk_socket->file)
1539                 if (send_sigurg(&sk->sk_socket->file->f_owner))
1540                         sk_wake_async(sk, 3, POLL_PRI);
1541 }
1542
1543 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1544                     unsigned long expires)
1545 {
1546         if (!mod_timer(timer, expires))
1547                 sock_hold(sk);
1548 }
1549
1550 EXPORT_SYMBOL(sk_reset_timer);
1551
1552 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1553 {
1554         if (timer_pending(timer) && del_timer(timer))
1555                 __sock_put(sk);
1556 }
1557
1558 EXPORT_SYMBOL(sk_stop_timer);
1559
1560 void sock_init_data(struct socket *sock, struct sock *sk)
1561 {
1562         skb_queue_head_init(&sk->sk_receive_queue);
1563         skb_queue_head_init(&sk->sk_write_queue);
1564         skb_queue_head_init(&sk->sk_error_queue);
1565 #ifdef CONFIG_NET_DMA
1566         skb_queue_head_init(&sk->sk_async_wait_queue);
1567 #endif
1568
1569         sk->sk_send_head        =       NULL;
1570
1571         init_timer(&sk->sk_timer);
1572
1573         sk->sk_allocation       =       GFP_KERNEL;
1574         sk->sk_rcvbuf           =       sysctl_rmem_default;
1575         sk->sk_sndbuf           =       sysctl_wmem_default;
1576         sk->sk_state            =       TCP_CLOSE;
1577         sk->sk_socket           =       sock;
1578
1579         sock_set_flag(sk, SOCK_ZAPPED);
1580
1581         if (sock) {
1582                 sk->sk_type     =       sock->type;
1583                 sk->sk_sleep    =       &sock->wait;
1584                 sock->sk        =       sk;
1585         } else
1586                 sk->sk_sleep    =       NULL;
1587
1588         rwlock_init(&sk->sk_dst_lock);
1589         rwlock_init(&sk->sk_callback_lock);
1590         lockdep_set_class_and_name(&sk->sk_callback_lock,
1591                         af_callback_keys + sk->sk_family,
1592                         af_family_clock_key_strings[sk->sk_family]);
1593
1594         sk->sk_state_change     =       sock_def_wakeup;
1595         sk->sk_data_ready       =       sock_def_readable;
1596         sk->sk_write_space      =       sock_def_write_space;
1597         sk->sk_error_report     =       sock_def_error_report;
1598         sk->sk_destruct         =       sock_def_destruct;
1599
1600         sk->sk_sndmsg_page      =       NULL;
1601         sk->sk_sndmsg_off       =       0;
1602
1603         sk->sk_peercred.pid     =       0;
1604         sk->sk_peercred.uid     =       -1;
1605         sk->sk_peercred.gid     =       -1;
1606         sk->sk_write_pending    =       0;
1607         sk->sk_rcvlowat         =       1;
1608         sk->sk_rcvtimeo         =       MAX_SCHEDULE_TIMEOUT;
1609         sk->sk_sndtimeo         =       MAX_SCHEDULE_TIMEOUT;
1610
1611         sk->sk_stamp = ktime_set(-1L, -1L);
1612
1613         atomic_set(&sk->sk_refcnt, 1);
1614 }
1615
1616 void fastcall lock_sock_nested(struct sock *sk, int subclass)
1617 {
1618         might_sleep();
1619         spin_lock_bh(&sk->sk_lock.slock);
1620         if (sk->sk_lock.owned)
1621                 __lock_sock(sk);
1622         sk->sk_lock.owned = 1;
1623         spin_unlock(&sk->sk_lock.slock);
1624         /*
1625          * The sk_lock has mutex_lock() semantics here:
1626          */
1627         mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
1628         local_bh_enable();
1629 }
1630
1631 EXPORT_SYMBOL(lock_sock_nested);
1632
1633 void fastcall release_sock(struct sock *sk)
1634 {
1635         /*
1636          * The sk_lock has mutex_unlock() semantics:
1637          */
1638         mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1639
1640         spin_lock_bh(&sk->sk_lock.slock);
1641         if (sk->sk_backlog.tail)
1642                 __release_sock(sk);
1643         sk->sk_lock.owned = 0;
1644         if (waitqueue_active(&sk->sk_lock.wq))
1645                 wake_up(&sk->sk_lock.wq);
1646         spin_unlock_bh(&sk->sk_lock.slock);
1647 }
1648 EXPORT_SYMBOL(release_sock);
1649
1650 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1651 {
1652         struct timeval tv;
1653         if (!sock_flag(sk, SOCK_TIMESTAMP))
1654                 sock_enable_timestamp(sk);
1655         tv = ktime_to_timeval(sk->sk_stamp);
1656         if (tv.tv_sec == -1)
1657                 return -ENOENT;
1658         if (tv.tv_sec == 0) {
1659                 sk->sk_stamp = ktime_get_real();
1660                 tv = ktime_to_timeval(sk->sk_stamp);
1661         }
1662         return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
1663 }
1664 EXPORT_SYMBOL(sock_get_timestamp);
1665
1666 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
1667 {
1668         struct timespec ts;
1669         if (!sock_flag(sk, SOCK_TIMESTAMP))
1670                 sock_enable_timestamp(sk);
1671         ts = ktime_to_timespec(sk->sk_stamp);
1672         if (ts.tv_sec == -1)
1673                 return -ENOENT;
1674         if (ts.tv_sec == 0) {
1675                 sk->sk_stamp = ktime_get_real();
1676                 ts = ktime_to_timespec(sk->sk_stamp);
1677         }
1678         return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
1679 }
1680 EXPORT_SYMBOL(sock_get_timestampns);
1681
1682 void sock_enable_timestamp(struct sock *sk)
1683 {
1684         if (!sock_flag(sk, SOCK_TIMESTAMP)) {
1685                 sock_set_flag(sk, SOCK_TIMESTAMP);
1686                 net_enable_timestamp();
1687         }
1688 }
1689
1690 /*
1691  *      Get a socket option on an socket.
1692  *
1693  *      FIX: POSIX 1003.1g is very ambiguous here. It states that
1694  *      asynchronous errors should be reported by getsockopt. We assume
1695  *      this means if you specify SO_ERROR (otherwise whats the point of it).
1696  */
1697 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1698                            char __user *optval, int __user *optlen)
1699 {
1700         struct sock *sk = sock->sk;
1701
1702         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1703 }
1704
1705 EXPORT_SYMBOL(sock_common_getsockopt);
1706
1707 #ifdef CONFIG_COMPAT
1708 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
1709                                   char __user *optval, int __user *optlen)
1710 {
1711         struct sock *sk = sock->sk;
1712
1713         if (sk->sk_prot->compat_getsockopt != NULL)
1714                 return sk->sk_prot->compat_getsockopt(sk, level, optname,
1715                                                       optval, optlen);
1716         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1717 }
1718 EXPORT_SYMBOL(compat_sock_common_getsockopt);
1719 #endif
1720
1721 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1722                         struct msghdr *msg, size_t size, int flags)
1723 {
1724         struct sock *sk = sock->sk;
1725         int addr_len = 0;
1726         int err;
1727
1728         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
1729                                    flags & ~MSG_DONTWAIT, &addr_len);
1730         if (err >= 0)
1731                 msg->msg_namelen = addr_len;
1732         return err;
1733 }
1734
1735 EXPORT_SYMBOL(sock_common_recvmsg);
1736
1737 /*
1738  *      Set socket options on an inet socket.
1739  */
1740 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1741                            char __user *optval, int optlen)
1742 {
1743         struct sock *sk = sock->sk;
1744
1745         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1746 }
1747
1748 EXPORT_SYMBOL(sock_common_setsockopt);
1749
1750 #ifdef CONFIG_COMPAT
1751 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
1752                                   char __user *optval, int optlen)
1753 {
1754         struct sock *sk = sock->sk;
1755
1756         if (sk->sk_prot->compat_setsockopt != NULL)
1757                 return sk->sk_prot->compat_setsockopt(sk, level, optname,
1758                                                       optval, optlen);
1759         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1760 }
1761 EXPORT_SYMBOL(compat_sock_common_setsockopt);
1762 #endif
1763
1764 void sk_common_release(struct sock *sk)
1765 {
1766         if (sk->sk_prot->destroy)
1767                 sk->sk_prot->destroy(sk);
1768
1769         /*
1770          * Observation: when sock_common_release is called, processes have
1771          * no access to socket. But net still has.
1772          * Step one, detach it from networking:
1773          *
1774          * A. Remove from hash tables.
1775          */
1776
1777         sk->sk_prot->unhash(sk);
1778
1779         /*
1780          * In this point socket cannot receive new packets, but it is possible
1781          * that some packets are in flight because some CPU runs receiver and
1782          * did hash table lookup before we unhashed socket. They will achieve
1783          * receive queue and will be purged by socket destructor.
1784          *
1785          * Also we still have packets pending on receive queue and probably,
1786          * our own packets waiting in device queues. sock_destroy will drain
1787          * receive queue, but transmitted packets will delay socket destruction
1788          * until the last reference will be released.
1789          */
1790
1791         sock_orphan(sk);
1792
1793         xfrm_sk_free_policy(sk);
1794
1795         sk_refcnt_debug_release(sk);
1796         sock_put(sk);
1797 }
1798
1799 EXPORT_SYMBOL(sk_common_release);
1800
1801 static DEFINE_RWLOCK(proto_list_lock);
1802 static LIST_HEAD(proto_list);
1803
1804 #ifdef CONFIG_SMP
1805 /*
1806  * Define default functions to keep track of inuse sockets per protocol
1807  * Note that often used protocols use dedicated functions to get a speed increase.
1808  * (see DEFINE_PROTO_INUSE/REF_PROTO_INUSE)
1809  */
1810 static void inuse_add(struct proto *prot, int inc)
1811 {
1812         per_cpu_ptr(prot->inuse_ptr, smp_processor_id())[0] += inc;
1813 }
1814
1815 static int inuse_get(const struct proto *prot)
1816 {
1817         int res = 0, cpu;
1818         for_each_possible_cpu(cpu)
1819                 res += per_cpu_ptr(prot->inuse_ptr, cpu)[0];
1820         return res;
1821 }
1822
1823 static int inuse_init(struct proto *prot)
1824 {
1825         if (!prot->inuse_getval || !prot->inuse_add) {
1826                 prot->inuse_ptr = alloc_percpu(int);
1827                 if (prot->inuse_ptr == NULL)
1828                         return -ENOBUFS;
1829
1830                 prot->inuse_getval = inuse_get;
1831                 prot->inuse_add = inuse_add;
1832         }
1833         return 0;
1834 }
1835
1836 static void inuse_fini(struct proto *prot)
1837 {
1838         if (prot->inuse_ptr != NULL) {
1839                 free_percpu(prot->inuse_ptr);
1840                 prot->inuse_ptr = NULL;
1841                 prot->inuse_getval = NULL;
1842                 prot->inuse_add = NULL;
1843         }
1844 }
1845 #else
1846 static inline int inuse_init(struct proto *prot)
1847 {
1848         return 0;
1849 }
1850
1851 static inline void inuse_fini(struct proto *prot)
1852 {
1853 }
1854 #endif
1855
1856 int proto_register(struct proto *prot, int alloc_slab)
1857 {
1858         char *request_sock_slab_name = NULL;
1859         char *timewait_sock_slab_name;
1860
1861         if (inuse_init(prot))
1862                 goto out;
1863
1864         if (alloc_slab) {
1865                 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
1866                                                SLAB_HWCACHE_ALIGN, NULL);
1867
1868                 if (prot->slab == NULL) {
1869                         printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
1870                                prot->name);
1871                         goto out_free_inuse;
1872                 }
1873
1874                 if (prot->rsk_prot != NULL) {
1875                         static const char mask[] = "request_sock_%s";
1876
1877                         request_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
1878                         if (request_sock_slab_name == NULL)
1879                                 goto out_free_sock_slab;
1880
1881                         sprintf(request_sock_slab_name, mask, prot->name);
1882                         prot->rsk_prot->slab = kmem_cache_create(request_sock_slab_name,
1883                                                                  prot->rsk_prot->obj_size, 0,
1884                                                                  SLAB_HWCACHE_ALIGN, NULL);
1885
1886                         if (prot->rsk_prot->slab == NULL) {
1887                                 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
1888                                        prot->name);
1889                                 goto out_free_request_sock_slab_name;
1890                         }
1891                 }
1892
1893                 if (prot->twsk_prot != NULL) {
1894                         static const char mask[] = "tw_sock_%s";
1895
1896                         timewait_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
1897
1898                         if (timewait_sock_slab_name == NULL)
1899                                 goto out_free_request_sock_slab;
1900
1901                         sprintf(timewait_sock_slab_name, mask, prot->name);
1902                         prot->twsk_prot->twsk_slab =
1903                                 kmem_cache_create(timewait_sock_slab_name,
1904                                                   prot->twsk_prot->twsk_obj_size,
1905                                                   0, SLAB_HWCACHE_ALIGN,
1906                                                   NULL);
1907                         if (prot->twsk_prot->twsk_slab == NULL)
1908                                 goto out_free_timewait_sock_slab_name;
1909                 }
1910         }
1911
1912         write_lock(&proto_list_lock);
1913         list_add(&prot->node, &proto_list);
1914         write_unlock(&proto_list_lock);
1915         return 0;
1916
1917 out_free_timewait_sock_slab_name:
1918         kfree(timewait_sock_slab_name);
1919 out_free_request_sock_slab:
1920         if (prot->rsk_prot && prot->rsk_prot->slab) {
1921                 kmem_cache_destroy(prot->rsk_prot->slab);
1922                 prot->rsk_prot->slab = NULL;
1923         }
1924 out_free_request_sock_slab_name:
1925         kfree(request_sock_slab_name);
1926 out_free_sock_slab:
1927         kmem_cache_destroy(prot->slab);
1928         prot->slab = NULL;
1929 out_free_inuse:
1930         inuse_fini(prot);
1931 out:
1932         return -ENOBUFS;
1933 }
1934
1935 EXPORT_SYMBOL(proto_register);
1936
1937 void proto_unregister(struct proto *prot)
1938 {
1939         write_lock(&proto_list_lock);
1940         list_del(&prot->node);
1941         write_unlock(&proto_list_lock);
1942
1943         inuse_fini(prot);
1944         if (prot->slab != NULL) {
1945                 kmem_cache_destroy(prot->slab);
1946                 prot->slab = NULL;
1947         }
1948
1949         if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
1950                 const char *name = kmem_cache_name(prot->rsk_prot->slab);
1951
1952                 kmem_cache_destroy(prot->rsk_prot->slab);
1953                 kfree(name);
1954                 prot->rsk_prot->slab = NULL;
1955         }
1956
1957         if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
1958                 const char *name = kmem_cache_name(prot->twsk_prot->twsk_slab);
1959
1960                 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
1961                 kfree(name);
1962                 prot->twsk_prot->twsk_slab = NULL;
1963         }
1964 }
1965
1966 EXPORT_SYMBOL(proto_unregister);
1967
1968 #ifdef CONFIG_PROC_FS
1969 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
1970 {
1971         read_lock(&proto_list_lock);
1972         return seq_list_start_head(&proto_list, *pos);
1973 }
1974
1975 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1976 {
1977         return seq_list_next(v, &proto_list, pos);
1978 }
1979
1980 static void proto_seq_stop(struct seq_file *seq, void *v)
1981 {
1982         read_unlock(&proto_list_lock);
1983 }
1984
1985 static char proto_method_implemented(const void *method)
1986 {
1987         return method == NULL ? 'n' : 'y';
1988 }
1989
1990 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
1991 {
1992         seq_printf(seq, "%-9s %4u %6d  %6d   %-3s %6u   %-3s  %-10s "
1993                         "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
1994                    proto->name,
1995                    proto->obj_size,
1996                    proto->sockets_allocated != NULL ? atomic_read(proto->sockets_allocated) : -1,
1997                    proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
1998                    proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
1999                    proto->max_header,
2000                    proto->slab == NULL ? "no" : "yes",
2001                    module_name(proto->owner),
2002                    proto_method_implemented(proto->close),
2003                    proto_method_implemented(proto->connect),
2004                    proto_method_implemented(proto->disconnect),
2005                    proto_method_implemented(proto->accept),
2006                    proto_method_implemented(proto->ioctl),
2007                    proto_method_implemented(proto->init),
2008                    proto_method_implemented(proto->destroy),
2009                    proto_method_implemented(proto->shutdown),
2010                    proto_method_implemented(proto->setsockopt),
2011                    proto_method_implemented(proto->getsockopt),
2012                    proto_method_implemented(proto->sendmsg),
2013                    proto_method_implemented(proto->recvmsg),
2014                    proto_method_implemented(proto->sendpage),
2015                    proto_method_implemented(proto->bind),
2016                    proto_method_implemented(proto->backlog_rcv),
2017                    proto_method_implemented(proto->hash),
2018                    proto_method_implemented(proto->unhash),
2019                    proto_method_implemented(proto->get_port),
2020                    proto_method_implemented(proto->enter_memory_pressure));
2021 }
2022
2023 static int proto_seq_show(struct seq_file *seq, void *v)
2024 {
2025         if (v == &proto_list)
2026                 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2027                            "protocol",
2028                            "size",
2029                            "sockets",
2030                            "memory",
2031                            "press",
2032                            "maxhdr",
2033                            "slab",
2034                            "module",
2035                            "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2036         else
2037                 proto_seq_printf(seq, list_entry(v, struct proto, node));
2038         return 0;
2039 }
2040
2041 static const struct seq_operations proto_seq_ops = {
2042         .start  = proto_seq_start,
2043         .next   = proto_seq_next,
2044         .stop   = proto_seq_stop,
2045         .show   = proto_seq_show,
2046 };
2047
2048 static int proto_seq_open(struct inode *inode, struct file *file)
2049 {
2050         return seq_open(file, &proto_seq_ops);
2051 }
2052
2053 static const struct file_operations proto_seq_fops = {
2054         .owner          = THIS_MODULE,
2055         .open           = proto_seq_open,
2056         .read           = seq_read,
2057         .llseek         = seq_lseek,
2058         .release        = seq_release,
2059 };
2060
2061 static int __init proto_init(void)
2062 {
2063         /* register /proc/net/protocols */
2064         return proc_net_fops_create(&init_net, "protocols", S_IRUGO, &proto_seq_fops) == NULL ? -ENOBUFS : 0;
2065 }
2066
2067 subsys_initcall(proto_init);
2068
2069 #endif /* PROC_FS */
2070
2071 EXPORT_SYMBOL(sk_alloc);
2072 EXPORT_SYMBOL(sk_free);
2073 EXPORT_SYMBOL(sk_send_sigurg);
2074 EXPORT_SYMBOL(sock_alloc_send_skb);
2075 EXPORT_SYMBOL(sock_init_data);
2076 EXPORT_SYMBOL(sock_kfree_s);
2077 EXPORT_SYMBOL(sock_kmalloc);
2078 EXPORT_SYMBOL(sock_no_accept);
2079 EXPORT_SYMBOL(sock_no_bind);
2080 EXPORT_SYMBOL(sock_no_connect);
2081 EXPORT_SYMBOL(sock_no_getname);
2082 EXPORT_SYMBOL(sock_no_getsockopt);
2083 EXPORT_SYMBOL(sock_no_ioctl);
2084 EXPORT_SYMBOL(sock_no_listen);
2085 EXPORT_SYMBOL(sock_no_mmap);
2086 EXPORT_SYMBOL(sock_no_poll);
2087 EXPORT_SYMBOL(sock_no_recvmsg);
2088 EXPORT_SYMBOL(sock_no_sendmsg);
2089 EXPORT_SYMBOL(sock_no_sendpage);
2090 EXPORT_SYMBOL(sock_no_setsockopt);
2091 EXPORT_SYMBOL(sock_no_shutdown);
2092 EXPORT_SYMBOL(sock_no_socketpair);
2093 EXPORT_SYMBOL(sock_rfree);
2094 EXPORT_SYMBOL(sock_setsockopt);
2095 EXPORT_SYMBOL(sock_wfree);
2096 EXPORT_SYMBOL(sock_wmalloc);
2097 EXPORT_SYMBOL(sock_i_uid);
2098 EXPORT_SYMBOL(sock_i_ino);
2099 EXPORT_SYMBOL(sysctl_optmem_max);
2100 #ifdef CONFIG_SYSCTL
2101 EXPORT_SYMBOL(sysctl_rmem_max);
2102 EXPORT_SYMBOL(sysctl_wmem_max);
2103 #endif