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-AF_CAN"      ,
 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-AF_CAN"      ,
 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         if (!sk_rmem_schedule(sk, skb->truesize)) {
 286                 err = -ENOBUFS;
 287                 goto out;
 288         }
 289
 290         skb->dev = NULL;
 291         skb_set_owner_r(skb, sk);
 292
 293         /* Cache the SKB length before we tack it onto the receive
 294          * queue.  Once it is added it no longer belongs to us and
 295          * may be freed by other threads of control pulling packets
 296          * from the queue.
 297          */
 298         skb_len = skb->len;
 299
 300         skb_queue_tail(&sk->sk_receive_queue, skb);
 301
 302         if (!sock_flag(sk, SOCK_DEAD))
 303                 sk->sk_data_ready(sk, skb_len);
 304 out:
 305         return err;
 306 }
 307 EXPORT_SYMBOL(sock_queue_rcv_skb);
 308
 309 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
 310 {
 311         int rc = NET_RX_SUCCESS;
 312
 313         if (sk_filter(sk, skb))
 314                 goto discard_and_relse;
 315
 316         skb->dev = NULL;
 317
 318         if (nested)
 319                 bh_lock_sock_nested(sk);
 320         else
 321                 bh_lock_sock(sk);
 322         if (!sock_owned_by_user(sk)) {
 323                 /*
 324                  * trylock + unlock semantics:
 325                  */
 326                 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
 327
 328                 rc = sk->sk_backlog_rcv(sk, skb);
 329
 330                 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
 331         } else
 332                 sk_add_backlog(sk, skb);
 333         bh_unlock_sock(sk);
 334 out:
 335         sock_put(sk);
 336         return rc;
 337 discard_and_relse:
 338         kfree_skb(skb);
 339         goto out;
 340 }
 341 EXPORT_SYMBOL(sk_receive_skb);
 342
 343 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
 344 {
 345         struct dst_entry *dst = sk->sk_dst_cache;
 346
 347         if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
 348                 sk->sk_dst_cache = NULL;
 349                 dst_release(dst);
 350                 return NULL;
 351         }
 352
 353         return dst;
 354 }
 355 EXPORT_SYMBOL(__sk_dst_check);
 356
 357 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
 358 {
 359         struct dst_entry *dst = sk_dst_get(sk);
 360
 361         if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
 362                 sk_dst_reset(sk);
 363                 dst_release(dst);
 364                 return NULL;
 365         }
 366
 367         return dst;
 368 }
 369 EXPORT_SYMBOL(sk_dst_check);
 370
 371 static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
 372 {
 373         int ret = -ENOPROTOOPT;
 374 #ifdef CONFIG_NETDEVICES
 375         struct net *net = sk->sk_net;
 376         char devname[IFNAMSIZ];
 377         int index;
 378
 379         /* Sorry... */
 380         ret = -EPERM;
 381         if (!capable(CAP_NET_RAW))
 382                 goto out;
 383
 384         ret = -EINVAL;
 385         if (optlen < 0)
 386                 goto out;
 387
 388         /* Bind this socket to a particular device like "eth0",
 389          * as specified in the passed interface name. If the
 390          * name is "" or the option length is zero the socket
 391          * is not bound.
 392          */
 393         if (optlen > IFNAMSIZ - 1)
 394                 optlen = IFNAMSIZ - 1;
 395         memset(devname, 0, sizeof(devname));
 396
 397         ret = -EFAULT;
 398         if (copy_from_user(devname, optval, optlen))
 399                 goto out;
 400
 401         if (devname[0] == '\0') {
 402                 index = 0;
 403         } else {
 404                 struct net_device *dev = dev_get_by_name(net, devname);
 405
 406                 ret = -ENODEV;
 407                 if (!dev)
 408                         goto out;
 409
 410                 index = dev->ifindex;
 411                 dev_put(dev);
 412         }
 413
 414         lock_sock(sk);
 415         sk->sk_bound_dev_if = index;
 416         sk_dst_reset(sk);
 417         release_sock(sk);
 418
 419         ret = 0;
 420
 421 out:
 422 #endif
 423
 424         return ret;
 425 }
 426
 427 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
 428 {
 429         if (valbool)
 430                 sock_set_flag(sk, bit);
 431         else
 432                 sock_reset_flag(sk, bit);
 433 }
 434
 435 /*
 436  *      This is meant for all protocols to use and covers goings on
 437  *      at the socket level. Everything here is generic.
 438  */
 439
 440 int sock_setsockopt(struct socket *sock, int level, int optname,
 441                     char __user *optval, int optlen)
 442 {
 443         struct sock *sk=sock->sk;
 444         int val;
 445         int valbool;
 446         struct linger ling;
 447         int ret = 0;
 448
 449         /*
 450          *      Options without arguments
 451          */
 452
 453 #ifdef SO_DONTLINGER            /* Compatibility item... */
 454         if (optname == SO_DONTLINGER) {
 455                 lock_sock(sk);
 456                 sock_reset_flag(sk, SOCK_LINGER);
 457                 release_sock(sk);
 458                 return 0;
 459         }
 460 #endif
 461
 462         if (optname == SO_BINDTODEVICE)
 463                 return sock_bindtodevice(sk, optval, optlen);
 464
 465         if (optlen < sizeof(int))
 466                 return -EINVAL;
 467
 468         if (get_user(val, (int __user *)optval))
 469                 return -EFAULT;
 470
 471         valbool = val?1:0;
 472
 473         lock_sock(sk);
 474
 475         switch(optname) {
 476         case SO_DEBUG:
 477                 if (val && !capable(CAP_NET_ADMIN)) {
 478                         ret = -EACCES;
 479                 } else
 480                         sock_valbool_flag(sk, SOCK_DBG, valbool);
 481                 break;
 482         case SO_REUSEADDR:
 483                 sk->sk_reuse = valbool;
 484                 break;
 485         case SO_TYPE:
 486         case SO_ERROR:
 487                 ret = -ENOPROTOOPT;
 488                 break;
 489         case SO_DONTROUTE:
 490                 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
 491                 break;
 492         case SO_BROADCAST:
 493                 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
 494                 break;
 495         case SO_SNDBUF:
 496                 /* Don't error on this BSD doesn't and if you think
 497                    about it this is right. Otherwise apps have to
 498                    play 'guess the biggest size' games. RCVBUF/SNDBUF
 499                    are treated in BSD as hints */
 500
 501                 if (val > sysctl_wmem_max)
 502                         val = sysctl_wmem_max;
 503 set_sndbuf:
 504                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
 505                 if ((val * 2) < SOCK_MIN_SNDBUF)
 506                         sk->sk_sndbuf = SOCK_MIN_SNDBUF;
 507                 else
 508                         sk->sk_sndbuf = val * 2;
 509
 510                 /*
 511                  *      Wake up sending tasks if we
 512                  *      upped the value.
 513                  */
 514                 sk->sk_write_space(sk);
 515                 break;
 516
 517         case SO_SNDBUFFORCE:
 518                 if (!capable(CAP_NET_ADMIN)) {
 519                         ret = -EPERM;
 520                         break;
 521                 }
 522                 goto set_sndbuf;
 523
 524         case SO_RCVBUF:
 525                 /* Don't error on this BSD doesn't and if you think
 526                    about it this is right. Otherwise apps have to
 527                    play 'guess the biggest size' games. RCVBUF/SNDBUF
 528                    are treated in BSD as hints */
 529
 530                 if (val > sysctl_rmem_max)
 531                         val = sysctl_rmem_max;
 532 set_rcvbuf:
 533                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
 534                 /*
 535                  * We double it on the way in to account for
 536                  * "struct sk_buff" etc. overhead.   Applications
 537                  * assume that the SO_RCVBUF setting they make will
 538                  * allow that much actual data to be received on that
 539                  * socket.
 540                  *
 541                  * Applications are unaware that "struct sk_buff" and
 542                  * other overheads allocate from the receive buffer
 543                  * during socket buffer allocation.
 544                  *
 545                  * And after considering the possible alternatives,
 546                  * returning the value we actually used in getsockopt
 547                  * is the most desirable behavior.
 548                  */
 549                 if ((val * 2) < SOCK_MIN_RCVBUF)
 550                         sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
 551                 else
 552                         sk->sk_rcvbuf = val * 2;
 553                 break;
 554
 555         case SO_RCVBUFFORCE:
 556                 if (!capable(CAP_NET_ADMIN)) {
 557                         ret = -EPERM;
 558                         break;
 559                 }
 560                 goto set_rcvbuf;
 561
 562         case SO_KEEPALIVE:
 563 #ifdef CONFIG_INET
 564                 if (sk->sk_protocol == IPPROTO_TCP)
 565                         tcp_set_keepalive(sk, valbool);
 566 #endif
 567                 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
 568                 break;
 569
 570         case SO_OOBINLINE:
 571                 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
 572                 break;
 573
 574         case SO_NO_CHECK:
 575                 sk->sk_no_check = valbool;
 576                 break;
 577
 578         case SO_PRIORITY:
 579                 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
 580                         sk->sk_priority = val;
 581                 else
 582                         ret = -EPERM;
 583                 break;
 584
 585         case SO_LINGER:
 586                 if (optlen < sizeof(ling)) {
 587                         ret = -EINVAL;  /* 1003.1g */
 588                         break;
 589                 }
 590                 if (copy_from_user(&ling,optval,sizeof(ling))) {
 591                         ret = -EFAULT;
 592                         break;
 593                 }
 594                 if (!ling.l_onoff)
 595                         sock_reset_flag(sk, SOCK_LINGER);
 596                 else {
 597 #if (BITS_PER_LONG == 32)
 598                         if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
 599                                 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
 600                         else
 601 #endif
 602                                 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
 603                         sock_set_flag(sk, SOCK_LINGER);
 604                 }
 605                 break;
 606
 607         case SO_BSDCOMPAT:
 608                 sock_warn_obsolete_bsdism("setsockopt");
 609                 break;
 610
 611         case SO_PASSCRED:
 612                 if (valbool)
 613                         set_bit(SOCK_PASSCRED, &sock->flags);
 614                 else
 615                         clear_bit(SOCK_PASSCRED, &sock->flags);
 616                 break;
 617
 618         case SO_TIMESTAMP:
 619         case SO_TIMESTAMPNS:
 620                 if (valbool)  {
 621                         if (optname == SO_TIMESTAMP)
 622                                 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
 623                         else
 624                                 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
 625                         sock_set_flag(sk, SOCK_RCVTSTAMP);
 626                         sock_enable_timestamp(sk);
 627                 } else {
 628                         sock_reset_flag(sk, SOCK_RCVTSTAMP);
 629                         sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
 630                 }
 631                 break;
 632
 633         case SO_RCVLOWAT:
 634                 if (val < 0)
 635                         val = INT_MAX;
 636                 sk->sk_rcvlowat = val ? : 1;
 637                 break;
 638
 639         case SO_RCVTIMEO:
 640                 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
 641                 break;
 642
 643         case SO_SNDTIMEO:
 644                 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
 645                 break;
 646
 647         case SO_ATTACH_FILTER:
 648                 ret = -EINVAL;
 649                 if (optlen == sizeof(struct sock_fprog)) {
 650                         struct sock_fprog fprog;
 651
 652                         ret = -EFAULT;
 653                         if (copy_from_user(&fprog, optval, sizeof(fprog)))
 654                                 break;
 655
 656                         ret = sk_attach_filter(&fprog, sk);
 657                 }
 658                 break;
 659
 660         case SO_DETACH_FILTER:
 661                 ret = sk_detach_filter(sk);
 662                 break;
 663
 664         case SO_PASSSEC:
 665                 if (valbool)
 666                         set_bit(SOCK_PASSSEC, &sock->flags);
 667                 else
 668                         clear_bit(SOCK_PASSSEC, &sock->flags);
 669                 break;
 670         case SO_MARK:
 671                 if (!capable(CAP_NET_ADMIN))
 672                         ret = -EPERM;
 673                 else {
 674                         sk->sk_mark = val;
 675                 }
 676                 break;
 677
 678                 /* We implement the SO_SNDLOWAT etc to
 679                    not be settable (1003.1g 5.3) */
 680         default:
 681                 ret = -ENOPROTOOPT;
 682                 break;
 683         }
 684         release_sock(sk);
 685         return ret;
 686 }
 687
 688
 689 int sock_getsockopt(struct socket *sock, int level, int optname,
 690                     char __user *optval, int __user *optlen)
 691 {
 692         struct sock *sk = sock->sk;
 693
 694         union {
 695                 int val;
 696                 struct linger ling;
 697                 struct timeval tm;
 698         } v;
 699
 700         unsigned int lv = sizeof(int);
 701         int len;
 702
 703         if (get_user(len, optlen))
 704                 return -EFAULT;
 705         if (len < 0)
 706                 return -EINVAL;
 707
 708         switch(optname) {
 709         case SO_DEBUG:
 710                 v.val = sock_flag(sk, SOCK_DBG);
 711                 break;
 712
 713         case SO_DONTROUTE:
 714                 v.val = sock_flag(sk, SOCK_LOCALROUTE);
 715                 break;
 716
 717         case SO_BROADCAST:
 718                 v.val = !!sock_flag(sk, SOCK_BROADCAST);
 719                 break;
 720
 721         case SO_SNDBUF:
 722                 v.val = sk->sk_sndbuf;
 723                 break;
 724
 725         case SO_RCVBUF:
 726                 v.val = sk->sk_rcvbuf;
 727                 break;
 728
 729         case SO_REUSEADDR:
 730                 v.val = sk->sk_reuse;
 731                 break;
 732
 733         case SO_KEEPALIVE:
 734                 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
 735                 break;
 736
 737         case SO_TYPE:
 738                 v.val = sk->sk_type;
 739                 break;
 740
 741         case SO_ERROR:
 742                 v.val = -sock_error(sk);
 743                 if (v.val==0)
 744                         v.val = xchg(&sk->sk_err_soft, 0);
 745                 break;
 746
 747         case SO_OOBINLINE:
 748                 v.val = !!sock_flag(sk, SOCK_URGINLINE);
 749                 break;
 750
 751         case SO_NO_CHECK:
 752                 v.val = sk->sk_no_check;
 753                 break;
 754
 755         case SO_PRIORITY:
 756                 v.val = sk->sk_priority;
 757                 break;
 758
 759         case SO_LINGER:
 760                 lv              = sizeof(v.ling);
 761                 v.ling.l_onoff  = !!sock_flag(sk, SOCK_LINGER);
 762                 v.ling.l_linger = sk->sk_lingertime / HZ;
 763                 break;
 764
 765         case SO_BSDCOMPAT:
 766                 sock_warn_obsolete_bsdism("getsockopt");
 767                 break;
 768
 769         case SO_TIMESTAMP:
 770                 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
 771                                 !sock_flag(sk, SOCK_RCVTSTAMPNS);
 772                 break;
 773
 774         case SO_TIMESTAMPNS:
 775                 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
 776                 break;
 777
 778         case SO_RCVTIMEO:
 779                 lv=sizeof(struct timeval);
 780                 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
 781                         v.tm.tv_sec = 0;
 782                         v.tm.tv_usec = 0;
 783                 } else {
 784                         v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
 785                         v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
 786                 }
 787                 break;
 788
 789         case SO_SNDTIMEO:
 790                 lv=sizeof(struct timeval);
 791                 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
 792                         v.tm.tv_sec = 0;
 793                         v.tm.tv_usec = 0;
 794                 } else {
 795                         v.tm.tv_sec = sk->sk_sndtimeo / HZ;
 796                         v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
 797                 }
 798                 break;
 799
 800         case SO_RCVLOWAT:
 801                 v.val = sk->sk_rcvlowat;
 802                 break;
 803
 804         case SO_SNDLOWAT:
 805                 v.val=1;
 806                 break;
 807
 808         case SO_PASSCRED:
 809                 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
 810                 break;
 811
 812         case SO_PEERCRED:
 813                 if (len > sizeof(sk->sk_peercred))
 814                         len = sizeof(sk->sk_peercred);
 815                 if (copy_to_user(optval, &sk->sk_peercred, len))
 816                         return -EFAULT;
 817                 goto lenout;
 818
 819         case SO_PEERNAME:
 820         {
 821                 char address[128];
 822
 823                 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
 824                         return -ENOTCONN;
 825                 if (lv < len)
 826                         return -EINVAL;
 827                 if (copy_to_user(optval, address, len))
 828                         return -EFAULT;
 829                 goto lenout;
 830         }
 831
 832         /* Dubious BSD thing... Probably nobody even uses it, but
 833          * the UNIX standard wants it for whatever reason... -DaveM
 834          */
 835         case SO_ACCEPTCONN:
 836                 v.val = sk->sk_state == TCP_LISTEN;
 837                 break;
 838
 839         case SO_PASSSEC:
 840                 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
 841                 break;
 842
 843         case SO_PEERSEC:
 844                 return security_socket_getpeersec_stream(sock, optval, optlen, len);
 845
 846         case SO_MARK:
 847                 v.val = sk->sk_mark;
 848                 break;
 849
 850         default:
 851                 return -ENOPROTOOPT;
 852         }
 853
 854         if (len > lv)
 855                 len = lv;
 856         if (copy_to_user(optval, &v, len))
 857                 return -EFAULT;
 858 lenout:
 859         if (put_user(len, optlen))
 860                 return -EFAULT;
 861         return 0;
 862 }
 863
 864 /*
 865  * Initialize an sk_lock.
 866  *
 867  * (We also register the sk_lock with the lock validator.)
 868  */
 869 static inline void sock_lock_init(struct sock *sk)
 870 {
 871         sock_lock_init_class_and_name(sk,
 872                         af_family_slock_key_strings[sk->sk_family],
 873                         af_family_slock_keys + sk->sk_family,
 874                         af_family_key_strings[sk->sk_family],
 875                         af_family_keys + sk->sk_family);
 876 }
 877
 878 static void sock_copy(struct sock *nsk, const struct sock *osk)
 879 {
 880 #ifdef CONFIG_SECURITY_NETWORK
 881         void *sptr = nsk->sk_security;
 882 #endif
 883
 884         memcpy(nsk, osk, osk->sk_prot->obj_size);
 885 #ifdef CONFIG_SECURITY_NETWORK
 886         nsk->sk_security = sptr;
 887         security_sk_clone(osk, nsk);
 888 #endif
 889 }
 890
 891 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
 892                 int family)
 893 {
 894         struct sock *sk;
 895         struct kmem_cache *slab;
 896
 897         slab = prot->slab;
 898         if (slab != NULL)
 899                 sk = kmem_cache_alloc(slab, priority);
 900         else
 901                 sk = kmalloc(prot->obj_size, priority);
 902
 903         if (sk != NULL) {
 904                 if (security_sk_alloc(sk, family, priority))
 905                         goto out_free;
 906
 907                 if (!try_module_get(prot->owner))
 908                         goto out_free_sec;
 909         }
 910
 911         return sk;
 912
 913 out_free_sec:
 914         security_sk_free(sk);
 915 out_free:
 916         if (slab != NULL)
 917                 kmem_cache_free(slab, sk);
 918         else
 919                 kfree(sk);
 920         return NULL;
 921 }
 922
 923 static void sk_prot_free(struct proto *prot, struct sock *sk)
 924 {
 925         struct kmem_cache *slab;
 926         struct module *owner;
 927
 928         owner = prot->owner;
 929         slab = prot->slab;
 930
 931         security_sk_free(sk);
 932         if (slab != NULL)
 933                 kmem_cache_free(slab, sk);
 934         else
 935                 kfree(sk);
 936         module_put(owner);
 937 }
 938
 939 /**
 940  *      sk_alloc - All socket objects are allocated here
 941  *      @net: the applicable net namespace
 942  *      @family: protocol family
 943  *      @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
 944  *      @prot: struct proto associated with this new sock instance
 945  *      @zero_it: if we should zero the newly allocated sock
 946  */
 947 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
 948                       struct proto *prot)
 949 {
 950         struct sock *sk;
 951
 952         sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
 953         if (sk) {
 954                 sk->sk_family = family;
 955                 /*
 956                  * See comment in struct sock definition to understand
 957                  * why we need sk_prot_creator -acme
 958                  */
 959                 sk->sk_prot = sk->sk_prot_creator = prot;
 960                 sock_lock_init(sk);
 961                 sk->sk_net = get_net(net);
 962         }
 963
 964         return sk;
 965 }
 966
 967 void sk_free(struct sock *sk)
 968 {
 969         struct sk_filter *filter;
 970
 971         if (sk->sk_destruct)
 972                 sk->sk_destruct(sk);
 973
 974         filter = rcu_dereference(sk->sk_filter);
 975         if (filter) {
 976                 sk_filter_uncharge(sk, filter);
 977                 rcu_assign_pointer(sk->sk_filter, NULL);
 978         }
 979
 980         sock_disable_timestamp(sk);
 981
 982         if (atomic_read(&sk->sk_omem_alloc))
 983                 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
 984                        __FUNCTION__, atomic_read(&sk->sk_omem_alloc));
 985
 986         put_net(sk->sk_net);
 987         sk_prot_free(sk->sk_prot_creator, sk);
 988 }
 989
 990 struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
 991 {
 992         struct sock *newsk;
 993
 994         newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
 995         if (newsk != NULL) {
 996                 struct sk_filter *filter;
 997
 998                 sock_copy(newsk, sk);
 999
1000                 /* SANITY */
1001                 get_net(newsk->sk_net);
1002                 sk_node_init(&newsk->sk_node);
1003                 sock_lock_init(newsk);
1004                 bh_lock_sock(newsk);
1005                 newsk->sk_backlog.head  = newsk->sk_backlog.tail = NULL;
1006
1007                 atomic_set(&newsk->sk_rmem_alloc, 0);
1008                 atomic_set(&newsk->sk_wmem_alloc, 0);
1009                 atomic_set(&newsk->sk_omem_alloc, 0);
1010                 skb_queue_head_init(&newsk->sk_receive_queue);
1011                 skb_queue_head_init(&newsk->sk_write_queue);
1012 #ifdef CONFIG_NET_DMA
1013                 skb_queue_head_init(&newsk->sk_async_wait_queue);
1014 #endif
1015
1016                 rwlock_init(&newsk->sk_dst_lock);
1017                 rwlock_init(&newsk->sk_callback_lock);
1018                 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1019                                 af_callback_keys + newsk->sk_family,
1020                                 af_family_clock_key_strings[newsk->sk_family]);
1021
1022                 newsk->sk_dst_cache     = NULL;
1023                 newsk->sk_wmem_queued   = 0;
1024                 newsk->sk_forward_alloc = 0;
1025                 newsk->sk_send_head     = NULL;
1026                 newsk->sk_userlocks     = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1027
1028                 sock_reset_flag(newsk, SOCK_DONE);
1029                 skb_queue_head_init(&newsk->sk_error_queue);
1030
1031                 filter = newsk->sk_filter;
1032                 if (filter != NULL)
1033                         sk_filter_charge(newsk, filter);
1034
1035                 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1036                         /* It is still raw copy of parent, so invalidate
1037                          * destructor and make plain sk_free() */
1038                         newsk->sk_destruct = NULL;
1039                         sk_free(newsk);
1040                         newsk = NULL;
1041                         goto out;
1042                 }
1043
1044                 newsk->sk_err      = 0;
1045                 newsk->sk_priority = 0;
1046                 atomic_set(&newsk->sk_refcnt, 2);
1047
1048                 /*
1049                  * Increment the counter in the same struct proto as the master
1050                  * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1051                  * is the same as sk->sk_prot->socks, as this field was copied
1052                  * with memcpy).
1053                  *
1054                  * This _changes_ the previous behaviour, where
1055                  * tcp_create_openreq_child always was incrementing the
1056                  * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1057                  * to be taken into account in all callers. -acme
1058                  */
1059                 sk_refcnt_debug_inc(newsk);
1060                 newsk->sk_socket = NULL;
1061                 newsk->sk_sleep  = NULL;
1062
1063                 if (newsk->sk_prot->sockets_allocated)
1064                         atomic_inc(newsk->sk_prot->sockets_allocated);
1065         }
1066 out:
1067         return newsk;
1068 }
1069
1070 EXPORT_SYMBOL_GPL(sk_clone);
1071
1072 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1073 {
1074         __sk_dst_set(sk, dst);
1075         sk->sk_route_caps = dst->dev->features;
1076         if (sk->sk_route_caps & NETIF_F_GSO)
1077                 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1078         if (sk_can_gso(sk)) {
1079                 if (dst->header_len)
1080                         sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1081                 else
1082                         sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1083         }
1084 }
1085 EXPORT_SYMBOL_GPL(sk_setup_caps);
1086
1087 void __init sk_init(void)
1088 {
1089         if (num_physpages <= 4096) {
1090                 sysctl_wmem_max = 32767;
1091                 sysctl_rmem_max = 32767;
1092                 sysctl_wmem_default = 32767;
1093                 sysctl_rmem_default = 32767;
1094         } else if (num_physpages >= 131072) {
1095                 sysctl_wmem_max = 131071;
1096                 sysctl_rmem_max = 131071;
1097         }
1098 }
1099
1100 /*
1101  *      Simple resource managers for sockets.
1102  */
1103
1104
1105 /*
1106  * Write buffer destructor automatically called from kfree_skb.
1107  */
1108 void sock_wfree(struct sk_buff *skb)
1109 {
1110         struct sock *sk = skb->sk;
1111
1112         /* In case it might be waiting for more memory. */
1113         atomic_sub(skb->truesize, &sk->sk_wmem_alloc);
1114         if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE))
1115                 sk->sk_write_space(sk);
1116         sock_put(sk);
1117 }
1118
1119 /*
1120  * Read buffer destructor automatically called from kfree_skb.
1121  */
1122 void sock_rfree(struct sk_buff *skb)
1123 {
1124         struct sock *sk = skb->sk;
1125
1126         skb_truesize_check(skb);
1127         atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1128         sk_mem_uncharge(skb->sk, skb->truesize);
1129 }
1130
1131
1132 int sock_i_uid(struct sock *sk)
1133 {
1134         int uid;
1135
1136         read_lock(&sk->sk_callback_lock);
1137         uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1138         read_unlock(&sk->sk_callback_lock);
1139         return uid;
1140 }
1141
1142 unsigned long sock_i_ino(struct sock *sk)
1143 {
1144         unsigned long ino;
1145
1146         read_lock(&sk->sk_callback_lock);
1147         ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1148         read_unlock(&sk->sk_callback_lock);
1149         return ino;
1150 }
1151
1152 /*
1153  * Allocate a skb from the socket's send buffer.
1154  */
1155 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1156                              gfp_t priority)
1157 {
1158         if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1159                 struct sk_buff * skb = alloc_skb(size, priority);
1160                 if (skb) {
1161                         skb_set_owner_w(skb, sk);
1162                         return skb;
1163                 }
1164         }
1165         return NULL;
1166 }
1167
1168 /*
1169  * Allocate a skb from the socket's receive buffer.
1170  */
1171 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1172                              gfp_t priority)
1173 {
1174         if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1175                 struct sk_buff *skb = alloc_skb(size, priority);
1176                 if (skb) {
1177                         skb_set_owner_r(skb, sk);
1178                         return skb;
1179                 }
1180         }
1181         return NULL;
1182 }
1183
1184 /*
1185  * Allocate a memory block from the socket's option memory buffer.
1186  */
1187 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1188 {
1189         if ((unsigned)size <= sysctl_optmem_max &&
1190             atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1191                 void *mem;
1192                 /* First do the add, to avoid the race if kmalloc
1193                  * might sleep.
1194                  */
1195                 atomic_add(size, &sk->sk_omem_alloc);
1196                 mem = kmalloc(size, priority);
1197                 if (mem)
1198                         return mem;
1199                 atomic_sub(size, &sk->sk_omem_alloc);
1200         }
1201         return NULL;
1202 }
1203
1204 /*
1205  * Free an option memory block.
1206  */
1207 void sock_kfree_s(struct sock *sk, void *mem, int size)
1208 {
1209         kfree(mem);
1210         atomic_sub(size, &sk->sk_omem_alloc);
1211 }
1212
1213 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1214    I think, these locks should be removed for datagram sockets.
1215  */
1216 static long sock_wait_for_wmem(struct sock * sk, long timeo)
1217 {
1218         DEFINE_WAIT(wait);
1219
1220         clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1221         for (;;) {
1222                 if (!timeo)
1223                         break;
1224                 if (signal_pending(current))
1225                         break;
1226                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1227                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1228                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1229                         break;
1230                 if (sk->sk_shutdown & SEND_SHUTDOWN)
1231                         break;
1232                 if (sk->sk_err)
1233                         break;
1234                 timeo = schedule_timeout(timeo);
1235         }
1236         finish_wait(sk->sk_sleep, &wait);
1237         return timeo;
1238 }
1239
1240
1241 /*
1242  *      Generic send/receive buffer handlers
1243  */
1244
1245 static struct sk_buff *sock_alloc_send_pskb(struct sock *sk,
1246                                             unsigned long header_len,
1247                                             unsigned long data_len,
1248                                             int noblock, int *errcode)
1249 {
1250         struct sk_buff *skb;
1251         gfp_t gfp_mask;
1252         long timeo;
1253         int err;
1254
1255         gfp_mask = sk->sk_allocation;
1256         if (gfp_mask & __GFP_WAIT)
1257                 gfp_mask |= __GFP_REPEAT;
1258
1259         timeo = sock_sndtimeo(sk, noblock);
1260         while (1) {
1261                 err = sock_error(sk);
1262                 if (err != 0)
1263                         goto failure;
1264
1265                 err = -EPIPE;
1266                 if (sk->sk_shutdown & SEND_SHUTDOWN)
1267                         goto failure;
1268
1269                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1270                         skb = alloc_skb(header_len, gfp_mask);
1271                         if (skb) {
1272                                 int npages;
1273                                 int i;
1274
1275                                 /* No pages, we're done... */
1276                                 if (!data_len)
1277                                         break;
1278
1279                                 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1280                                 skb->truesize += data_len;
1281                                 skb_shinfo(skb)->nr_frags = npages;
1282                                 for (i = 0; i < npages; i++) {
1283                                         struct page *page;
1284                                         skb_frag_t *frag;
1285
1286                                         page = alloc_pages(sk->sk_allocation, 0);
1287                                         if (!page) {
1288                                                 err = -ENOBUFS;
1289                                                 skb_shinfo(skb)->nr_frags = i;
1290                                                 kfree_skb(skb);
1291                                                 goto failure;
1292                                         }
1293
1294                                         frag = &skb_shinfo(skb)->frags[i];
1295                                         frag->page = page;
1296                                         frag->page_offset = 0;
1297                                         frag->size = (data_len >= PAGE_SIZE ?
1298                                                       PAGE_SIZE :
1299                                                       data_len);
1300                                         data_len -= PAGE_SIZE;
1301                                 }
1302
1303                                 /* Full success... */
1304                                 break;
1305                         }
1306                         err = -ENOBUFS;
1307                         goto failure;
1308                 }
1309                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1310                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1311                 err = -EAGAIN;
1312                 if (!timeo)
1313                         goto failure;
1314                 if (signal_pending(current))
1315                         goto interrupted;
1316                 timeo = sock_wait_for_wmem(sk, timeo);
1317         }
1318
1319         skb_set_owner_w(skb, sk);
1320         return skb;
1321
1322 interrupted:
1323         err = sock_intr_errno(timeo);
1324 failure:
1325         *errcode = err;
1326         return NULL;
1327 }
1328
1329 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1330                                     int noblock, int *errcode)
1331 {
1332         return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1333 }
1334
1335 static void __lock_sock(struct sock *sk)
1336 {
1337         DEFINE_WAIT(wait);
1338
1339         for (;;) {
1340                 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1341                                         TASK_UNINTERRUPTIBLE);
1342                 spin_unlock_bh(&sk->sk_lock.slock);
1343                 schedule();
1344                 spin_lock_bh(&sk->sk_lock.slock);
1345                 if (!sock_owned_by_user(sk))
1346                         break;
1347         }
1348         finish_wait(&sk->sk_lock.wq, &wait);
1349 }
1350
1351 static void __release_sock(struct sock *sk)
1352 {
1353         struct sk_buff *skb = sk->sk_backlog.head;
1354
1355         do {
1356                 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1357                 bh_unlock_sock(sk);
1358
1359                 do {
1360                         struct sk_buff *next = skb->next;
1361
1362                         skb->next = NULL;
1363                         sk->sk_backlog_rcv(sk, skb);
1364
1365                         /*
1366                          * We are in process context here with softirqs
1367                          * disabled, use cond_resched_softirq() to preempt.
1368                          * This is safe to do because we've taken the backlog
1369                          * queue private:
1370                          */
1371                         cond_resched_softirq();
1372
1373                         skb = next;
1374                 } while (skb != NULL);
1375
1376                 bh_lock_sock(sk);
1377         } while ((skb = sk->sk_backlog.head) != NULL);
1378 }
1379
1380 /**
1381  * sk_wait_data - wait for data to arrive at sk_receive_queue
1382  * @sk:    sock to wait on
1383  * @timeo: for how long
1384  *
1385  * Now socket state including sk->sk_err is changed only under lock,
1386  * hence we may omit checks after joining wait queue.
1387  * We check receive queue before schedule() only as optimization;
1388  * it is very likely that release_sock() added new data.
1389  */
1390 int sk_wait_data(struct sock *sk, long *timeo)
1391 {
1392         int rc;
1393         DEFINE_WAIT(wait);
1394
1395         prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1396         set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1397         rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1398         clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1399         finish_wait(sk->sk_sleep, &wait);
1400         return rc;
1401 }
1402
1403 EXPORT_SYMBOL(sk_wait_data);
1404
1405 /**
1406  *      __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1407  *      @sk: socket
1408  *      @size: memory size to allocate
1409  *      @kind: allocation type
1410  *
1411  *      If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1412  *      rmem allocation. This function assumes that protocols which have
1413  *      memory_pressure use sk_wmem_queued as write buffer accounting.
1414  */
1415 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1416 {
1417         struct proto *prot = sk->sk_prot;
1418         int amt = sk_mem_pages(size);
1419         int allocated;
1420
1421         sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1422         allocated = atomic_add_return(amt, prot->memory_allocated);
1423
1424         /* Under limit. */
1425         if (allocated <= prot->sysctl_mem[0]) {
1426                 if (prot->memory_pressure && *prot->memory_pressure)
1427                         *prot->memory_pressure = 0;
1428                 return 1;
1429         }
1430
1431         /* Under pressure. */
1432         if (allocated > prot->sysctl_mem[1])
1433                 if (prot->enter_memory_pressure)
1434                         prot->enter_memory_pressure();
1435
1436         /* Over hard limit. */
1437         if (allocated > prot->sysctl_mem[2])
1438                 goto suppress_allocation;
1439
1440         /* guarantee minimum buffer size under pressure */
1441         if (kind == SK_MEM_RECV) {
1442                 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1443                         return 1;
1444         } else { /* SK_MEM_SEND */
1445                 if (sk->sk_type == SOCK_STREAM) {
1446                         if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1447                                 return 1;
1448                 } else if (atomic_read(&sk->sk_wmem_alloc) <
1449                            prot->sysctl_wmem[0])
1450                                 return 1;
1451         }
1452
1453         if (prot->memory_pressure) {
1454                 if (!*prot->memory_pressure ||
1455                     prot->sysctl_mem[2] > atomic_read(prot->sockets_allocated) *
1456                     sk_mem_pages(sk->sk_wmem_queued +
1457                                  atomic_read(&sk->sk_rmem_alloc) +
1458                                  sk->sk_forward_alloc))
1459                         return 1;
1460         }
1461
1462 suppress_allocation:
1463
1464         if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1465                 sk_stream_moderate_sndbuf(sk);
1466
1467                 /* Fail only if socket is _under_ its sndbuf.
1468                  * In this case we cannot block, so that we have to fail.
1469                  */
1470                 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1471                         return 1;
1472         }
1473
1474         /* Alas. Undo changes. */
1475         sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1476         atomic_sub(amt, prot->memory_allocated);
1477         return 0;
1478 }
1479
1480 EXPORT_SYMBOL(__sk_mem_schedule);
1481
1482 /**
1483  *      __sk_reclaim - reclaim memory_allocated
1484  *      @sk: socket
1485  */
1486 void __sk_mem_reclaim(struct sock *sk)
1487 {
1488         struct proto *prot = sk->sk_prot;
1489
1490         atomic_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
1491                    prot->memory_allocated);
1492         sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1493
1494         if (prot->memory_pressure && *prot->memory_pressure &&
1495             (atomic_read(prot->memory_allocated) < prot->sysctl_mem[0]))
1496                 *prot->memory_pressure = 0;
1497 }
1498
1499 EXPORT_SYMBOL(__sk_mem_reclaim);
1500
1501
1502 /*
1503  * Set of default routines for initialising struct proto_ops when
1504  * the protocol does not support a particular function. In certain
1505  * cases where it makes no sense for a protocol to have a "do nothing"
1506  * function, some default processing is provided.
1507  */
1508
1509 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1510 {
1511         return -EOPNOTSUPP;
1512 }
1513
1514 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1515                     int len, int flags)
1516 {
1517         return -EOPNOTSUPP;
1518 }
1519
1520 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1521 {
1522         return -EOPNOTSUPP;
1523 }
1524
1525 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1526 {
1527         return -EOPNOTSUPP;
1528 }
1529
1530 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1531                     int *len, int peer)
1532 {
1533         return -EOPNOTSUPP;
1534 }
1535
1536 unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt)
1537 {
1538         return 0;
1539 }
1540
1541 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1542 {
1543         return -EOPNOTSUPP;
1544 }
1545
1546 int sock_no_listen(struct socket *sock, int backlog)
1547 {
1548         return -EOPNOTSUPP;
1549 }
1550
1551 int sock_no_shutdown(struct socket *sock, int how)
1552 {
1553         return -EOPNOTSUPP;
1554 }
1555
1556 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1557                     char __user *optval, int optlen)
1558 {
1559         return -EOPNOTSUPP;
1560 }
1561
1562 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1563                     char __user *optval, int __user *optlen)
1564 {
1565         return -EOPNOTSUPP;
1566 }
1567
1568 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1569                     size_t len)
1570 {
1571         return -EOPNOTSUPP;
1572 }
1573
1574 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1575                     size_t len, int flags)
1576 {
1577         return -EOPNOTSUPP;
1578 }
1579
1580 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1581 {
1582         /* Mirror missing mmap method error code */
1583         return -ENODEV;
1584 }
1585
1586 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1587 {
1588         ssize_t res;
1589         struct msghdr msg = {.msg_flags = flags};
1590         struct kvec iov;
1591         char *kaddr = kmap(page);
1592         iov.iov_base = kaddr + offset;
1593         iov.iov_len = size;
1594         res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1595         kunmap(page);
1596         return res;
1597 }
1598
1599 /*
1600  *      Default Socket Callbacks
1601  */
1602
1603 static void sock_def_wakeup(struct sock *sk)
1604 {
1605         read_lock(&sk->sk_callback_lock);
1606         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1607                 wake_up_interruptible_all(sk->sk_sleep);
1608         read_unlock(&sk->sk_callback_lock);
1609 }
1610
1611 static void sock_def_error_report(struct sock *sk)
1612 {
1613         read_lock(&sk->sk_callback_lock);
1614         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1615                 wake_up_interruptible(sk->sk_sleep);
1616         sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
1617         read_unlock(&sk->sk_callback_lock);
1618 }
1619
1620 static void sock_def_readable(struct sock *sk, int len)
1621 {
1622         read_lock(&sk->sk_callback_lock);
1623         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1624                 wake_up_interruptible_sync(sk->sk_sleep);
1625         sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1626         read_unlock(&sk->sk_callback_lock);
1627 }
1628
1629 static void sock_def_write_space(struct sock *sk)
1630 {
1631         read_lock(&sk->sk_callback_lock);
1632
1633         /* Do not wake up a writer until he can make "significant"
1634          * progress.  --DaveM
1635          */
1636         if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1637                 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1638                         wake_up_interruptible_sync(sk->sk_sleep);
1639
1640                 /* Should agree with poll, otherwise some programs break */
1641                 if (sock_writeable(sk))
1642                         sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1643         }
1644
1645         read_unlock(&sk->sk_callback_lock);
1646 }
1647
1648 static void sock_def_destruct(struct sock *sk)
1649 {
1650         kfree(sk->sk_protinfo);
1651 }
1652
1653 void sk_send_sigurg(struct sock *sk)
1654 {
1655         if (sk->sk_socket && sk->sk_socket->file)
1656                 if (send_sigurg(&sk->sk_socket->file->f_owner))
1657                         sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1658 }
1659
1660 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1661                     unsigned long expires)
1662 {
1663         if (!mod_timer(timer, expires))
1664                 sock_hold(sk);
1665 }
1666
1667 EXPORT_SYMBOL(sk_reset_timer);
1668
1669 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1670 {
1671         if (timer_pending(timer) && del_timer(timer))
1672                 __sock_put(sk);
1673 }
1674
1675 EXPORT_SYMBOL(sk_stop_timer);
1676
1677 void sock_init_data(struct socket *sock, struct sock *sk)
1678 {
1679         skb_queue_head_init(&sk->sk_receive_queue);
1680         skb_queue_head_init(&sk->sk_write_queue);
1681         skb_queue_head_init(&sk->sk_error_queue);
1682 #ifdef CONFIG_NET_DMA
1683         skb_queue_head_init(&sk->sk_async_wait_queue);
1684 #endif
1685
1686         sk->sk_send_head        =       NULL;
1687
1688         init_timer(&sk->sk_timer);
1689
1690         sk->sk_allocation       =       GFP_KERNEL;
1691         sk->sk_rcvbuf           =       sysctl_rmem_default;
1692         sk->sk_sndbuf           =       sysctl_wmem_default;
1693         sk->sk_state            =       TCP_CLOSE;
1694         sk->sk_socket           =       sock;
1695
1696         sock_set_flag(sk, SOCK_ZAPPED);
1697
1698         if (sock) {
1699                 sk->sk_type     =       sock->type;
1700                 sk->sk_sleep    =       &sock->wait;
1701                 sock->sk        =       sk;
1702         } else
1703                 sk->sk_sleep    =       NULL;
1704
1705         rwlock_init(&sk->sk_dst_lock);
1706         rwlock_init(&sk->sk_callback_lock);
1707         lockdep_set_class_and_name(&sk->sk_callback_lock,
1708                         af_callback_keys + sk->sk_family,
1709                         af_family_clock_key_strings[sk->sk_family]);
1710
1711         sk->sk_state_change     =       sock_def_wakeup;
1712         sk->sk_data_ready       =       sock_def_readable;
1713         sk->sk_write_space      =       sock_def_write_space;
1714         sk->sk_error_report     =       sock_def_error_report;
1715         sk->sk_destruct         =       sock_def_destruct;
1716
1717         sk->sk_sndmsg_page      =       NULL;
1718         sk->sk_sndmsg_off       =       0;
1719
1720         sk->sk_peercred.pid     =       0;
1721         sk->sk_peercred.uid     =       -1;
1722         sk->sk_peercred.gid     =       -1;
1723         sk->sk_write_pending    =       0;
1724         sk->sk_rcvlowat         =       1;
1725         sk->sk_rcvtimeo         =       MAX_SCHEDULE_TIMEOUT;
1726         sk->sk_sndtimeo         =       MAX_SCHEDULE_TIMEOUT;
1727
1728         sk->sk_stamp = ktime_set(-1L, -1L);
1729
1730         atomic_set(&sk->sk_refcnt, 1);
1731         atomic_set(&sk->sk_drops, 0);
1732 }
1733
1734 void lock_sock_nested(struct sock *sk, int subclass)
1735 {
1736         might_sleep();
1737         spin_lock_bh(&sk->sk_lock.slock);
1738         if (sk->sk_lock.owned)
1739                 __lock_sock(sk);
1740         sk->sk_lock.owned = 1;
1741         spin_unlock(&sk->sk_lock.slock);
1742         /*
1743          * The sk_lock has mutex_lock() semantics here:
1744          */
1745         mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
1746         local_bh_enable();
1747 }
1748
1749 EXPORT_SYMBOL(lock_sock_nested);
1750
1751 void release_sock(struct sock *sk)
1752 {
1753         /*
1754          * The sk_lock has mutex_unlock() semantics:
1755          */
1756         mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1757
1758         spin_lock_bh(&sk->sk_lock.slock);
1759         if (sk->sk_backlog.tail)
1760                 __release_sock(sk);
1761         sk->sk_lock.owned = 0;
1762         if (waitqueue_active(&sk->sk_lock.wq))
1763                 wake_up(&sk->sk_lock.wq);
1764         spin_unlock_bh(&sk->sk_lock.slock);
1765 }
1766 EXPORT_SYMBOL(release_sock);
1767
1768 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1769 {
1770         struct timeval tv;
1771         if (!sock_flag(sk, SOCK_TIMESTAMP))
1772                 sock_enable_timestamp(sk);
1773         tv = ktime_to_timeval(sk->sk_stamp);
1774         if (tv.tv_sec == -1)
1775                 return -ENOENT;
1776         if (tv.tv_sec == 0) {
1777                 sk->sk_stamp = ktime_get_real();
1778                 tv = ktime_to_timeval(sk->sk_stamp);
1779         }
1780         return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
1781 }
1782 EXPORT_SYMBOL(sock_get_timestamp);
1783
1784 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
1785 {
1786         struct timespec ts;
1787         if (!sock_flag(sk, SOCK_TIMESTAMP))
1788                 sock_enable_timestamp(sk);
1789         ts = ktime_to_timespec(sk->sk_stamp);
1790         if (ts.tv_sec == -1)
1791                 return -ENOENT;
1792         if (ts.tv_sec == 0) {
1793                 sk->sk_stamp = ktime_get_real();
1794                 ts = ktime_to_timespec(sk->sk_stamp);
1795         }
1796         return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
1797 }
1798 EXPORT_SYMBOL(sock_get_timestampns);
1799
1800 void sock_enable_timestamp(struct sock *sk)
1801 {
1802         if (!sock_flag(sk, SOCK_TIMESTAMP)) {
1803                 sock_set_flag(sk, SOCK_TIMESTAMP);
1804                 net_enable_timestamp();
1805         }
1806 }
1807
1808 /*
1809  *      Get a socket option on an socket.
1810  *
1811  *      FIX: POSIX 1003.1g is very ambiguous here. It states that
1812  *      asynchronous errors should be reported by getsockopt. We assume
1813  *      this means if you specify SO_ERROR (otherwise whats the point of it).
1814  */
1815 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1816                            char __user *optval, int __user *optlen)
1817 {
1818         struct sock *sk = sock->sk;
1819
1820         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1821 }
1822
1823 EXPORT_SYMBOL(sock_common_getsockopt);
1824
1825 #ifdef CONFIG_COMPAT
1826 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
1827                                   char __user *optval, int __user *optlen)
1828 {
1829         struct sock *sk = sock->sk;
1830
1831         if (sk->sk_prot->compat_getsockopt != NULL)
1832                 return sk->sk_prot->compat_getsockopt(sk, level, optname,
1833                                                       optval, optlen);
1834         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1835 }
1836 EXPORT_SYMBOL(compat_sock_common_getsockopt);
1837 #endif
1838
1839 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1840                         struct msghdr *msg, size_t size, int flags)
1841 {
1842         struct sock *sk = sock->sk;
1843         int addr_len = 0;
1844         int err;
1845
1846         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
1847                                    flags & ~MSG_DONTWAIT, &addr_len);
1848         if (err >= 0)
1849                 msg->msg_namelen = addr_len;
1850         return err;
1851 }
1852
1853 EXPORT_SYMBOL(sock_common_recvmsg);
1854
1855 /*
1856  *      Set socket options on an inet socket.
1857  */
1858 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1859                            char __user *optval, int optlen)
1860 {
1861         struct sock *sk = sock->sk;
1862
1863         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1864 }
1865
1866 EXPORT_SYMBOL(sock_common_setsockopt);
1867
1868 #ifdef CONFIG_COMPAT
1869 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
1870                                   char __user *optval, int optlen)
1871 {
1872         struct sock *sk = sock->sk;
1873
1874         if (sk->sk_prot->compat_setsockopt != NULL)
1875                 return sk->sk_prot->compat_setsockopt(sk, level, optname,
1876                                                       optval, optlen);
1877         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1878 }
1879 EXPORT_SYMBOL(compat_sock_common_setsockopt);
1880 #endif
1881
1882 void sk_common_release(struct sock *sk)
1883 {
1884         if (sk->sk_prot->destroy)
1885                 sk->sk_prot->destroy(sk);
1886
1887         /*
1888          * Observation: when sock_common_release is called, processes have
1889          * no access to socket. But net still has.
1890          * Step one, detach it from networking:
1891          *
1892          * A. Remove from hash tables.
1893          */
1894
1895         sk->sk_prot->unhash(sk);
1896
1897         /*
1898          * In this point socket cannot receive new packets, but it is possible
1899          * that some packets are in flight because some CPU runs receiver and
1900          * did hash table lookup before we unhashed socket. They will achieve
1901          * receive queue and will be purged by socket destructor.
1902          *
1903          * Also we still have packets pending on receive queue and probably,
1904          * our own packets waiting in device queues. sock_destroy will drain
1905          * receive queue, but transmitted packets will delay socket destruction
1906          * until the last reference will be released.
1907          */
1908
1909         sock_orphan(sk);
1910
1911         xfrm_sk_free_policy(sk);
1912
1913         sk_refcnt_debug_release(sk);
1914         sock_put(sk);
1915 }
1916
1917 EXPORT_SYMBOL(sk_common_release);
1918
1919 static DEFINE_RWLOCK(proto_list_lock);
1920 static LIST_HEAD(proto_list);
1921
1922 int proto_register(struct proto *prot, int alloc_slab)
1923 {
1924         char *request_sock_slab_name = NULL;
1925         char *timewait_sock_slab_name;
1926
1927         if (sock_prot_inuse_init(prot) != 0) {
1928                 printk(KERN_CRIT "%s: Can't alloc inuse counters!\n", prot->name);
1929                 goto out;
1930         }
1931
1932         if (alloc_slab) {
1933                 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
1934                                                SLAB_HWCACHE_ALIGN, NULL);
1935
1936                 if (prot->slab == NULL) {
1937                         printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
1938                                prot->name);
1939                         goto out_free_inuse;
1940                 }
1941
1942                 if (prot->rsk_prot != NULL) {
1943                         static const char mask[] = "request_sock_%s";
1944
1945                         request_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
1946                         if (request_sock_slab_name == NULL)
1947                                 goto out_free_sock_slab;
1948
1949                         sprintf(request_sock_slab_name, mask, prot->name);
1950                         prot->rsk_prot->slab = kmem_cache_create(request_sock_slab_name,
1951                                                                  prot->rsk_prot->obj_size, 0,
1952                                                                  SLAB_HWCACHE_ALIGN, NULL);
1953
1954                         if (prot->rsk_prot->slab == NULL) {
1955                                 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
1956                                        prot->name);
1957                                 goto out_free_request_sock_slab_name;
1958                         }
1959                 }
1960
1961                 if (prot->twsk_prot != NULL) {
1962                         static const char mask[] = "tw_sock_%s";
1963
1964                         timewait_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
1965
1966                         if (timewait_sock_slab_name == NULL)
1967                                 goto out_free_request_sock_slab;
1968
1969                         sprintf(timewait_sock_slab_name, mask, prot->name);
1970                         prot->twsk_prot->twsk_slab =
1971                                 kmem_cache_create(timewait_sock_slab_name,
1972                                                   prot->twsk_prot->twsk_obj_size,
1973                                                   0, SLAB_HWCACHE_ALIGN,
1974                                                   NULL);
1975                         if (prot->twsk_prot->twsk_slab == NULL)
1976                                 goto out_free_timewait_sock_slab_name;
1977                 }
1978         }
1979
1980         write_lock(&proto_list_lock);
1981         list_add(&prot->node, &proto_list);
1982         write_unlock(&proto_list_lock);
1983         return 0;
1984
1985 out_free_timewait_sock_slab_name:
1986         kfree(timewait_sock_slab_name);
1987 out_free_request_sock_slab:
1988         if (prot->rsk_prot && prot->rsk_prot->slab) {
1989                 kmem_cache_destroy(prot->rsk_prot->slab);
1990                 prot->rsk_prot->slab = NULL;
1991         }
1992 out_free_request_sock_slab_name:
1993         kfree(request_sock_slab_name);
1994 out_free_sock_slab:
1995         kmem_cache_destroy(prot->slab);
1996         prot->slab = NULL;
1997 out_free_inuse:
1998         sock_prot_inuse_free(prot);
1999 out:
2000         return -ENOBUFS;
2001 }
2002
2003 EXPORT_SYMBOL(proto_register);
2004
2005 void proto_unregister(struct proto *prot)
2006 {
2007         write_lock(&proto_list_lock);
2008         list_del(&prot->node);
2009         write_unlock(&proto_list_lock);
2010
2011         sock_prot_inuse_free(prot);
2012
2013         if (prot->slab != NULL) {
2014                 kmem_cache_destroy(prot->slab);
2015                 prot->slab = NULL;
2016         }
2017
2018         if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2019                 const char *name = kmem_cache_name(prot->rsk_prot->slab);
2020
2021                 kmem_cache_destroy(prot->rsk_prot->slab);
2022                 kfree(name);
2023                 prot->rsk_prot->slab = NULL;
2024         }
2025
2026         if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2027                 const char *name = kmem_cache_name(prot->twsk_prot->twsk_slab);
2028
2029                 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2030                 kfree(name);
2031                 prot->twsk_prot->twsk_slab = NULL;
2032         }
2033 }
2034
2035 EXPORT_SYMBOL(proto_unregister);
2036
2037 #ifdef CONFIG_PROC_FS
2038 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2039         __acquires(proto_list_lock)
2040 {
2041         read_lock(&proto_list_lock);
2042         return seq_list_start_head(&proto_list, *pos);
2043 }
2044
2045 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2046 {
2047         return seq_list_next(v, &proto_list, pos);
2048 }
2049
2050 static void proto_seq_stop(struct seq_file *seq, void *v)
2051         __releases(proto_list_lock)
2052 {
2053         read_unlock(&proto_list_lock);
2054 }
2055
2056 static char proto_method_implemented(const void *method)
2057 {
2058         return method == NULL ? 'n' : 'y';
2059 }
2060
2061 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2062 {
2063         seq_printf(seq, "%-9s %4u %6d  %6d   %-3s %6u   %-3s  %-10s "
2064                         "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2065                    proto->name,
2066                    proto->obj_size,
2067                    proto->sockets_allocated != NULL ? atomic_read(proto->sockets_allocated) : -1,
2068                    proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
2069                    proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2070                    proto->max_header,
2071                    proto->slab == NULL ? "no" : "yes",
2072                    module_name(proto->owner),
2073                    proto_method_implemented(proto->close),
2074                    proto_method_implemented(proto->connect),
2075                    proto_method_implemented(proto->disconnect),
2076                    proto_method_implemented(proto->accept),
2077                    proto_method_implemented(proto->ioctl),
2078                    proto_method_implemented(proto->init),
2079                    proto_method_implemented(proto->destroy),
2080                    proto_method_implemented(proto->shutdown),
2081                    proto_method_implemented(proto->setsockopt),
2082                    proto_method_implemented(proto->getsockopt),
2083                    proto_method_implemented(proto->sendmsg),
2084                    proto_method_implemented(proto->recvmsg),
2085                    proto_method_implemented(proto->sendpage),
2086                    proto_method_implemented(proto->bind),
2087                    proto_method_implemented(proto->backlog_rcv),
2088                    proto_method_implemented(proto->hash),
2089                    proto_method_implemented(proto->unhash),
2090                    proto_method_implemented(proto->get_port),
2091                    proto_method_implemented(proto->enter_memory_pressure));
2092 }
2093
2094 static int proto_seq_show(struct seq_file *seq, void *v)
2095 {
2096         if (v == &proto_list)
2097                 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2098                            "protocol",
2099                            "size",
2100                            "sockets",
2101                            "memory",
2102                            "press",
2103                            "maxhdr",
2104                            "slab",
2105                            "module",
2106                            "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2107         else
2108                 proto_seq_printf(seq, list_entry(v, struct proto, node));
2109         return 0;
2110 }
2111
2112 static const struct seq_operations proto_seq_ops = {
2113         .start  = proto_seq_start,
2114         .next   = proto_seq_next,
2115         .stop   = proto_seq_stop,
2116         .show   = proto_seq_show,
2117 };
2118
2119 static int proto_seq_open(struct inode *inode, struct file *file)
2120 {
2121         return seq_open(file, &proto_seq_ops);
2122 }
2123
2124 static const struct file_operations proto_seq_fops = {
2125         .owner          = THIS_MODULE,
2126         .open           = proto_seq_open,
2127         .read           = seq_read,
2128         .llseek         = seq_lseek,
2129         .release        = seq_release,
2130 };
2131
2132 static int __init proto_init(void)
2133 {
2134         /* register /proc/net/protocols */
2135         return proc_net_fops_create(&init_net, "protocols", S_IRUGO, &proto_seq_fops) == NULL ? -ENOBUFS : 0;
2136 }
2137
2138 subsys_initcall(proto_init);
2139
2140 #endif /* PROC_FS */
2141
2142 EXPORT_SYMBOL(sk_alloc);
2143 EXPORT_SYMBOL(sk_free);
2144 EXPORT_SYMBOL(sk_send_sigurg);
2145 EXPORT_SYMBOL(sock_alloc_send_skb);
2146 EXPORT_SYMBOL(sock_init_data);
2147 EXPORT_SYMBOL(sock_kfree_s);
2148 EXPORT_SYMBOL(sock_kmalloc);
2149 EXPORT_SYMBOL(sock_no_accept);
2150 EXPORT_SYMBOL(sock_no_bind);
2151 EXPORT_SYMBOL(sock_no_connect);
2152 EXPORT_SYMBOL(sock_no_getname);
2153 EXPORT_SYMBOL(sock_no_getsockopt);
2154 EXPORT_SYMBOL(sock_no_ioctl);
2155 EXPORT_SYMBOL(sock_no_listen);
2156 EXPORT_SYMBOL(sock_no_mmap);
2157 EXPORT_SYMBOL(sock_no_poll);
2158 EXPORT_SYMBOL(sock_no_recvmsg);
2159 EXPORT_SYMBOL(sock_no_sendmsg);
2160 EXPORT_SYMBOL(sock_no_sendpage);
2161 EXPORT_SYMBOL(sock_no_setsockopt);
2162 EXPORT_SYMBOL(sock_no_shutdown);
2163 EXPORT_SYMBOL(sock_no_socketpair);
2164 EXPORT_SYMBOL(sock_rfree);
2165 EXPORT_SYMBOL(sock_setsockopt);
2166 EXPORT_SYMBOL(sock_wfree);
2167 EXPORT_SYMBOL(sock_wmalloc);
2168 EXPORT_SYMBOL(sock_i_uid);
2169 EXPORT_SYMBOL(sock_i_ino);
2170 EXPORT_SYMBOL(sysctl_optmem_max);