SAFE public projects git trees. - safe/jmp/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  * Authors:     Ross Biro
  11  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  12  *              Florian La Roche, <flla@stud.uni-sb.de>
  13  *              Alan Cox, <A.Cox@swansea.ac.uk>
  14  *
  15  * Fixes:
  16  *              Alan Cox        :       Numerous verify_area() problems
  17  *              Alan Cox        :       Connecting on a connecting socket
  18  *                                      now returns an error for tcp.
  19  *              Alan Cox        :       sock->protocol is set correctly.
  20  *                                      and is not sometimes left as 0.
  21  *              Alan Cox        :       connect handles icmp errors on a
  22  *                                      connect properly. Unfortunately there
  23  *                                      is a restart syscall nasty there. I
  24  *                                      can't match BSD without hacking the C
  25  *                                      library. Ideas urgently sought!
  26  *              Alan Cox        :       Disallow bind() to addresses that are
  27  *                                      not ours - especially broadcast ones!!
  28  *              Alan Cox        :       Socket 1024 _IS_ ok for users. (fencepost)
  29  *              Alan Cox        :       sock_wfree/sock_rfree don't destroy sockets,
  30  *                                      instead they leave that for the DESTROY timer.
  31  *              Alan Cox        :       Clean up error flag in accept
  32  *              Alan Cox        :       TCP ack handling is buggy, the DESTROY timer
  33  *                                      was buggy. Put a remove_sock() in the handler
  34  *                                      for memory when we hit 0. Also altered the timer
  35  *                                      code. The ACK stuff can wait and needs major
  36  *                                      TCP layer surgery.
  37  *              Alan Cox        :       Fixed TCP ack bug, removed remove sock
  38  *                                      and fixed timer/inet_bh race.
  39  *              Alan Cox        :       Added zapped flag for TCP
  40  *              Alan Cox        :       Move kfree_skb into skbuff.c and tidied up surplus code
  41  *              Alan Cox        :       for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
  42  *              Alan Cox        :       kfree_s calls now are kfree_skbmem so we can track skb resources
  43  *              Alan Cox        :       Supports socket option broadcast now as does udp. Packet and raw need fixing.
  44  *              Alan Cox        :       Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
  45  *              Rick Sladkey    :       Relaxed UDP rules for matching packets.
  46  *              C.E.Hawkins     :       IFF_PROMISC/SIOCGHWADDR support
  47  *      Pauline Middelink       :       identd support
  48  *              Alan Cox        :       Fixed connect() taking signals I think.
  49  *              Alan Cox        :       SO_LINGER supported
  50  *              Alan Cox        :       Error reporting fixes
  51  *              Anonymous       :       inet_create tidied up (sk->reuse setting)
  52  *              Alan Cox        :       inet sockets don't set sk->type!
  53  *              Alan Cox        :       Split socket option code
  54  *              Alan Cox        :       Callbacks
  55  *              Alan Cox        :       Nagle flag for Charles & Johannes stuff
  56  *              Alex            :       Removed restriction on inet fioctl
  57  *              Alan Cox        :       Splitting INET from NET core
  58  *              Alan Cox        :       Fixed bogus SO_TYPE handling in getsockopt()
  59  *              Adam Caldwell   :       Missing return in SO_DONTROUTE/SO_DEBUG code
  60  *              Alan Cox        :       Split IP from generic code
  61  *              Alan Cox        :       New kfree_skbmem()
  62  *              Alan Cox        :       Make SO_DEBUG superuser only.
  63  *              Alan Cox        :       Allow anyone to clear SO_DEBUG
  64  *                                      (compatibility fix)
  65  *              Alan Cox        :       Added optimistic memory grabbing for AF_UNIX throughput.
  66  *              Alan Cox        :       Allocator for a socket is settable.
  67  *              Alan Cox        :       SO_ERROR includes soft errors.
  68  *              Alan Cox        :       Allow NULL arguments on some SO_ opts
  69  *              Alan Cox        :       Generic socket allocation to make hooks
  70  *                                      easier (suggested by Craig Metz).
  71  *              Michael Pall    :       SO_ERROR returns positive errno again
  72  *              Steve Whitehouse:       Added default destructor to free
  73  *                                      protocol private data.
  74  *              Steve Whitehouse:       Added various other default routines
  75  *                                      common to several socket families.
  76  *              Chris Evans     :       Call suser() check last on F_SETOWN
  77  *              Jay Schulist    :       Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
  78  *              Andi Kleen      :       Add sock_kmalloc()/sock_kfree_s()
  79  *              Andi Kleen      :       Fix write_space callback
  80  *              Chris Evans     :       Security fixes - signedness again
  81  *              Arnaldo C. Melo :       cleanups, use skb_queue_purge
  82  *
  83  * To Fix:
  84  *
  85  *
  86  *              This program is free software; you can redistribute it and/or
  87  *              modify it under the terms of the GNU General Public License
  88  *              as published by the Free Software Foundation; either version
  89  *              2 of the License, or (at your option) any later version.
  90  */
  91
  92 #include <linux/capability.h>
  93 #include <linux/errno.h>
  94 #include <linux/types.h>
  95 #include <linux/socket.h>
  96 #include <linux/in.h>
  97 #include <linux/kernel.h>
  98 #include <linux/module.h>
  99 #include <linux/proc_fs.h>
 100 #include <linux/seq_file.h>
 101 #include <linux/sched.h>
 102 #include <linux/timer.h>
 103 #include <linux/string.h>
 104 #include <linux/sockios.h>
 105 #include <linux/net.h>
 106 #include <linux/mm.h>
 107 #include <linux/slab.h>
 108 #include <linux/interrupt.h>
 109 #include <linux/poll.h>
 110 #include <linux/tcp.h>
 111 #include <linux/init.h>
 112 #include <linux/highmem.h>
 113
 114 #include <asm/uaccess.h>
 115 #include <asm/system.h>
 116
 117 #include <linux/netdevice.h>
 118 #include <net/protocol.h>
 119 #include <linux/skbuff.h>
 120 #include <net/net_namespace.h>
 121 #include <net/request_sock.h>
 122 #include <net/sock.h>
 123 #include <linux/net_tstamp.h>
 124 #include <net/xfrm.h>
 125 #include <linux/ipsec.h>
 126
 127 #include <linux/filter.h>
 128
 129 #ifdef CONFIG_INET
 130 #include <net/tcp.h>
 131 #endif
 132
 133 /*
 134  * Each address family might have different locking rules, so we have
 135  * one slock key per address family:
 136  */
 137 static struct lock_class_key af_family_keys[AF_MAX];
 138 static struct lock_class_key af_family_slock_keys[AF_MAX];
 139
 140 /*
 141  * Make lock validator output more readable. (we pre-construct these
 142  * strings build-time, so that runtime initialization of socket
 143  * locks is fast):
 144  */
 145 static const char *const af_family_key_strings[AF_MAX+1] = {
 146   "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX"     , "sk_lock-AF_INET"     ,
 147   "sk_lock-AF_AX25"  , "sk_lock-AF_IPX"      , "sk_lock-AF_APPLETALK",
 148   "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE"   , "sk_lock-AF_ATMPVC"   ,
 149   "sk_lock-AF_X25"   , "sk_lock-AF_INET6"    , "sk_lock-AF_ROSE"     ,
 150   "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI"  , "sk_lock-AF_SECURITY" ,
 151   "sk_lock-AF_KEY"   , "sk_lock-AF_NETLINK"  , "sk_lock-AF_PACKET"   ,
 152   "sk_lock-AF_ASH"   , "sk_lock-AF_ECONET"   , "sk_lock-AF_ATMSVC"   ,
 153   "sk_lock-AF_RDS"   , "sk_lock-AF_SNA"      , "sk_lock-AF_IRDA"     ,
 154   "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE"  , "sk_lock-AF_LLC"      ,
 155   "sk_lock-27"       , "sk_lock-28"          , "sk_lock-AF_CAN"      ,
 156   "sk_lock-AF_TIPC"  , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV"        ,
 157   "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN"     , "sk_lock-AF_PHONET"   ,
 158   "sk_lock-AF_IEEE802154",
 159   "sk_lock-AF_MAX"
 160 };
 161 static const char *const 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-AF_RDS"   , "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_ISDN"     , "slock-AF_PHONET"   ,
 174   "slock-AF_IEEE802154",
 175   "slock-AF_MAX"
 176 };
 177 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
 178   "clock-AF_UNSPEC", "clock-AF_UNIX"     , "clock-AF_INET"     ,
 179   "clock-AF_AX25"  , "clock-AF_IPX"      , "clock-AF_APPLETALK",
 180   "clock-AF_NETROM", "clock-AF_BRIDGE"   , "clock-AF_ATMPVC"   ,
 181   "clock-AF_X25"   , "clock-AF_INET6"    , "clock-AF_ROSE"     ,
 182   "clock-AF_DECnet", "clock-AF_NETBEUI"  , "clock-AF_SECURITY" ,
 183   "clock-AF_KEY"   , "clock-AF_NETLINK"  , "clock-AF_PACKET"   ,
 184   "clock-AF_ASH"   , "clock-AF_ECONET"   , "clock-AF_ATMSVC"   ,
 185   "clock-AF_RDS"   , "clock-AF_SNA"      , "clock-AF_IRDA"     ,
 186   "clock-AF_PPPOX" , "clock-AF_WANPIPE"  , "clock-AF_LLC"      ,
 187   "clock-27"       , "clock-28"          , "clock-AF_CAN"      ,
 188   "clock-AF_TIPC"  , "clock-AF_BLUETOOTH", "clock-AF_IUCV"     ,
 189   "clock-AF_RXRPC" , "clock-AF_ISDN"     , "clock-AF_PHONET"   ,
 190   "clock-AF_IEEE802154",
 191   "clock-AF_MAX"
 192 };
 193
 194 /*
 195  * sk_callback_lock locking rules are per-address-family,
 196  * so split the lock classes by using a per-AF key:
 197  */
 198 static struct lock_class_key af_callback_keys[AF_MAX];
 199
 200 /* Take into consideration the size of the struct sk_buff overhead in the
 201  * determination of these values, since that is non-constant across
 202  * platforms.  This makes socket queueing behavior and performance
 203  * not depend upon such differences.
 204  */
 205 #define _SK_MEM_PACKETS         256
 206 #define _SK_MEM_OVERHEAD        (sizeof(struct sk_buff) + 256)
 207 #define SK_WMEM_MAX             (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
 208 #define SK_RMEM_MAX             (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
 209
 210 /* Run time adjustable parameters. */
 211 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
 212 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
 213 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
 214 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
 215
 216 /* Maximal space eaten by iovec or ancilliary data plus some space */
 217 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
 218 EXPORT_SYMBOL(sysctl_optmem_max);
 219
 220 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
 221 {
 222         struct timeval tv;
 223
 224         if (optlen < sizeof(tv))
 225                 return -EINVAL;
 226         if (copy_from_user(&tv, optval, sizeof(tv)))
 227                 return -EFAULT;
 228         if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
 229                 return -EDOM;
 230
 231         if (tv.tv_sec < 0) {
 232                 static int warned __read_mostly;
 233
 234                 *timeo_p = 0;
 235                 if (warned < 10 && net_ratelimit()) {
 236                         warned++;
 237                         printk(KERN_INFO "sock_set_timeout: `%s' (pid %d) "
 238                                "tries to set negative timeout\n",
 239                                 current->comm, task_pid_nr(current));
 240                 }
 241                 return 0;
 242         }
 243         *timeo_p = MAX_SCHEDULE_TIMEOUT;
 244         if (tv.tv_sec == 0 && tv.tv_usec == 0)
 245                 return 0;
 246         if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
 247                 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
 248         return 0;
 249 }
 250
 251 static void sock_warn_obsolete_bsdism(const char *name)
 252 {
 253         static int warned;
 254         static char warncomm[TASK_COMM_LEN];
 255         if (strcmp(warncomm, current->comm) && warned < 5) {
 256                 strcpy(warncomm,  current->comm);
 257                 printk(KERN_WARNING "process `%s' is using obsolete "
 258                        "%s SO_BSDCOMPAT\n", warncomm, name);
 259                 warned++;
 260         }
 261 }
 262
 263 static void sock_disable_timestamp(struct sock *sk, int flag)
 264 {
 265         if (sock_flag(sk, flag)) {
 266                 sock_reset_flag(sk, flag);
 267                 if (!sock_flag(sk, SOCK_TIMESTAMP) &&
 268                     !sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE)) {
 269                         net_disable_timestamp();
 270                 }
 271         }
 272 }
 273
 274
 275 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
 276 {
 277         int err;
 278         int skb_len;
 279         unsigned long flags;
 280         struct sk_buff_head *list = &sk->sk_receive_queue;
 281
 282         /* Cast sk->rcvbuf to unsigned... It's pointless, but reduces
 283            number of warnings when compiling with -W --ANK
 284          */
 285         if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
 286             (unsigned)sk->sk_rcvbuf) {
 287                 atomic_inc(&sk->sk_drops);
 288                 return -ENOMEM;
 289         }
 290
 291         err = sk_filter(sk, skb);
 292         if (err)
 293                 return err;
 294
 295         if (!sk_rmem_schedule(sk, skb->truesize)) {
 296                 atomic_inc(&sk->sk_drops);
 297                 return -ENOBUFS;
 298         }
 299
 300         skb->dev = NULL;
 301         skb_set_owner_r(skb, sk);
 302
 303         /* Cache the SKB length before we tack it onto the receive
 304          * queue.  Once it is added it no longer belongs to us and
 305          * may be freed by other threads of control pulling packets
 306          * from the queue.
 307          */
 308         skb_len = skb->len;
 309
 310         spin_lock_irqsave(&list->lock, flags);
 311         skb->dropcount = atomic_read(&sk->sk_drops);
 312         __skb_queue_tail(list, skb);
 313         spin_unlock_irqrestore(&list->lock, flags);
 314
 315         if (!sock_flag(sk, SOCK_DEAD))
 316                 sk->sk_data_ready(sk, skb_len);
 317         return 0;
 318 }
 319 EXPORT_SYMBOL(sock_queue_rcv_skb);
 320
 321 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
 322 {
 323         int rc = NET_RX_SUCCESS;
 324
 325         if (sk_filter(sk, skb))
 326                 goto discard_and_relse;
 327
 328         skb->dev = NULL;
 329
 330         if (nested)
 331                 bh_lock_sock_nested(sk);
 332         else
 333                 bh_lock_sock(sk);
 334         if (!sock_owned_by_user(sk)) {
 335                 /*
 336                  * trylock + unlock semantics:
 337                  */
 338                 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
 339
 340                 rc = sk_backlog_rcv(sk, skb);
 341
 342                 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
 343         } else
 344                 sk_add_backlog(sk, skb);
 345         bh_unlock_sock(sk);
 346 out:
 347         sock_put(sk);
 348         return rc;
 349 discard_and_relse:
 350         kfree_skb(skb);
 351         goto out;
 352 }
 353 EXPORT_SYMBOL(sk_receive_skb);
 354
 355 void sk_reset_txq(struct sock *sk)
 356 {
 357         sk_tx_queue_clear(sk);
 358 }
 359 EXPORT_SYMBOL(sk_reset_txq);
 360
 361 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
 362 {
 363         struct dst_entry *dst = sk->sk_dst_cache;
 364
 365         if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
 366                 sk_tx_queue_clear(sk);
 367                 sk->sk_dst_cache = NULL;
 368                 dst_release(dst);
 369                 return NULL;
 370         }
 371
 372         return dst;
 373 }
 374 EXPORT_SYMBOL(__sk_dst_check);
 375
 376 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
 377 {
 378         struct dst_entry *dst = sk_dst_get(sk);
 379
 380         if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
 381                 sk_dst_reset(sk);
 382                 dst_release(dst);
 383                 return NULL;
 384         }
 385
 386         return dst;
 387 }
 388 EXPORT_SYMBOL(sk_dst_check);
 389
 390 static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
 391 {
 392         int ret = -ENOPROTOOPT;
 393 #ifdef CONFIG_NETDEVICES
 394         struct net *net = sock_net(sk);
 395         char devname[IFNAMSIZ];
 396         int index;
 397
 398         /* Sorry... */
 399         ret = -EPERM;
 400         if (!capable(CAP_NET_RAW))
 401                 goto out;
 402
 403         ret = -EINVAL;
 404         if (optlen < 0)
 405                 goto out;
 406
 407         /* Bind this socket to a particular device like "eth0",
 408          * as specified in the passed interface name. If the
 409          * name is "" or the option length is zero the socket
 410          * is not bound.
 411          */
 412         if (optlen > IFNAMSIZ - 1)
 413                 optlen = IFNAMSIZ - 1;
 414         memset(devname, 0, sizeof(devname));
 415
 416         ret = -EFAULT;
 417         if (copy_from_user(devname, optval, optlen))
 418                 goto out;
 419
 420         if (devname[0] == '\0') {
 421                 index = 0;
 422         } else {
 423                 struct net_device *dev;
 424
 425                 rcu_read_lock();
 426                 dev = dev_get_by_name_rcu(net, devname);
 427                 if (dev)
 428                         index = dev->ifindex;
 429                 rcu_read_unlock();
 430                 ret = -ENODEV;
 431                 if (!dev)
 432                         goto out;
 433         }
 434
 435         lock_sock(sk);
 436         sk->sk_bound_dev_if = index;
 437         sk_dst_reset(sk);
 438         release_sock(sk);
 439
 440         ret = 0;
 441
 442 out:
 443 #endif
 444
 445         return ret;
 446 }
 447
 448 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
 449 {
 450         if (valbool)
 451                 sock_set_flag(sk, bit);
 452         else
 453                 sock_reset_flag(sk, bit);
 454 }
 455
 456 /*
 457  *      This is meant for all protocols to use and covers goings on
 458  *      at the socket level. Everything here is generic.
 459  */
 460
 461 int sock_setsockopt(struct socket *sock, int level, int optname,
 462                     char __user *optval, unsigned int optlen)
 463 {
 464         struct sock *sk = sock->sk;
 465         int val;
 466         int valbool;
 467         struct linger ling;
 468         int ret = 0;
 469
 470         /*
 471          *      Options without arguments
 472          */
 473
 474         if (optname == SO_BINDTODEVICE)
 475                 return sock_bindtodevice(sk, optval, optlen);
 476
 477         if (optlen < sizeof(int))
 478                 return -EINVAL;
 479
 480         if (get_user(val, (int __user *)optval))
 481                 return -EFAULT;
 482
 483         valbool = val ? 1 : 0;
 484
 485         lock_sock(sk);
 486
 487         switch (optname) {
 488         case SO_DEBUG:
 489                 if (val && !capable(CAP_NET_ADMIN))
 490                         ret = -EACCES;
 491                 else
 492                         sock_valbool_flag(sk, SOCK_DBG, valbool);
 493                 break;
 494         case SO_REUSEADDR:
 495                 sk->sk_reuse = valbool;
 496                 break;
 497         case SO_TYPE:
 498         case SO_PROTOCOL:
 499         case SO_DOMAIN:
 500         case SO_ERROR:
 501                 ret = -ENOPROTOOPT;
 502                 break;
 503         case SO_DONTROUTE:
 504                 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
 505                 break;
 506         case SO_BROADCAST:
 507                 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
 508                 break;
 509         case SO_SNDBUF:
 510                 /* Don't error on this BSD doesn't and if you think
 511                    about it this is right. Otherwise apps have to
 512                    play 'guess the biggest size' games. RCVBUF/SNDBUF
 513                    are treated in BSD as hints */
 514
 515                 if (val > sysctl_wmem_max)
 516                         val = sysctl_wmem_max;
 517 set_sndbuf:
 518                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
 519                 if ((val * 2) < SOCK_MIN_SNDBUF)
 520                         sk->sk_sndbuf = SOCK_MIN_SNDBUF;
 521                 else
 522                         sk->sk_sndbuf = val * 2;
 523
 524                 /*
 525                  *      Wake up sending tasks if we
 526                  *      upped the value.
 527                  */
 528                 sk->sk_write_space(sk);
 529                 break;
 530
 531         case SO_SNDBUFFORCE:
 532                 if (!capable(CAP_NET_ADMIN)) {
 533                         ret = -EPERM;
 534                         break;
 535                 }
 536                 goto set_sndbuf;
 537
 538         case SO_RCVBUF:
 539                 /* Don't error on this BSD doesn't and if you think
 540                    about it this is right. Otherwise apps have to
 541                    play 'guess the biggest size' games. RCVBUF/SNDBUF
 542                    are treated in BSD as hints */
 543
 544                 if (val > sysctl_rmem_max)
 545                         val = sysctl_rmem_max;
 546 set_rcvbuf:
 547                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
 548                 /*
 549                  * We double it on the way in to account for
 550                  * "struct sk_buff" etc. overhead.   Applications
 551                  * assume that the SO_RCVBUF setting they make will
 552                  * allow that much actual data to be received on that
 553                  * socket.
 554                  *
 555                  * Applications are unaware that "struct sk_buff" and
 556                  * other overheads allocate from the receive buffer
 557                  * during socket buffer allocation.
 558                  *
 559                  * And after considering the possible alternatives,
 560                  * returning the value we actually used in getsockopt
 561                  * is the most desirable behavior.
 562                  */
 563                 if ((val * 2) < SOCK_MIN_RCVBUF)
 564                         sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
 565                 else
 566                         sk->sk_rcvbuf = val * 2;
 567                 break;
 568
 569         case SO_RCVBUFFORCE:
 570                 if (!capable(CAP_NET_ADMIN)) {
 571                         ret = -EPERM;
 572                         break;
 573                 }
 574                 goto set_rcvbuf;
 575
 576         case SO_KEEPALIVE:
 577 #ifdef CONFIG_INET
 578                 if (sk->sk_protocol == IPPROTO_TCP)
 579                         tcp_set_keepalive(sk, valbool);
 580 #endif
 581                 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
 582                 break;
 583
 584         case SO_OOBINLINE:
 585                 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
 586                 break;
 587
 588         case SO_NO_CHECK:
 589                 sk->sk_no_check = valbool;
 590                 break;
 591
 592         case SO_PRIORITY:
 593                 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
 594                         sk->sk_priority = val;
 595                 else
 596                         ret = -EPERM;
 597                 break;
 598
 599         case SO_LINGER:
 600                 if (optlen < sizeof(ling)) {
 601                         ret = -EINVAL;  /* 1003.1g */
 602                         break;
 603                 }
 604                 if (copy_from_user(&ling, optval, sizeof(ling))) {
 605                         ret = -EFAULT;
 606                         break;
 607                 }
 608                 if (!ling.l_onoff)
 609                         sock_reset_flag(sk, SOCK_LINGER);
 610                 else {
 611 #if (BITS_PER_LONG == 32)
 612                         if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
 613                                 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
 614                         else
 615 #endif
 616                                 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
 617                         sock_set_flag(sk, SOCK_LINGER);
 618                 }
 619                 break;
 620
 621         case SO_BSDCOMPAT:
 622                 sock_warn_obsolete_bsdism("setsockopt");
 623                 break;
 624
 625         case SO_PASSCRED:
 626                 if (valbool)
 627                         set_bit(SOCK_PASSCRED, &sock->flags);
 628                 else
 629                         clear_bit(SOCK_PASSCRED, &sock->flags);
 630                 break;
 631
 632         case SO_TIMESTAMP:
 633         case SO_TIMESTAMPNS:
 634                 if (valbool)  {
 635                         if (optname == SO_TIMESTAMP)
 636                                 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
 637                         else
 638                                 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
 639                         sock_set_flag(sk, SOCK_RCVTSTAMP);
 640                         sock_enable_timestamp(sk, SOCK_TIMESTAMP);
 641                 } else {
 642                         sock_reset_flag(sk, SOCK_RCVTSTAMP);
 643                         sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
 644                 }
 645                 break;
 646
 647         case SO_TIMESTAMPING:
 648                 if (val & ~SOF_TIMESTAMPING_MASK) {
 649                         ret = -EINVAL;
 650                         break;
 651                 }
 652                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
 653                                   val & SOF_TIMESTAMPING_TX_HARDWARE);
 654                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
 655                                   val & SOF_TIMESTAMPING_TX_SOFTWARE);
 656                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
 657                                   val & SOF_TIMESTAMPING_RX_HARDWARE);
 658                 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
 659                         sock_enable_timestamp(sk,
 660                                               SOCK_TIMESTAMPING_RX_SOFTWARE);
 661                 else
 662                         sock_disable_timestamp(sk,
 663                                                SOCK_TIMESTAMPING_RX_SOFTWARE);
 664                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
 665                                   val & SOF_TIMESTAMPING_SOFTWARE);
 666                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
 667                                   val & SOF_TIMESTAMPING_SYS_HARDWARE);
 668                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
 669                                   val & SOF_TIMESTAMPING_RAW_HARDWARE);
 670                 break;
 671
 672         case SO_RCVLOWAT:
 673                 if (val < 0)
 674                         val = INT_MAX;
 675                 sk->sk_rcvlowat = val ? : 1;
 676                 break;
 677
 678         case SO_RCVTIMEO:
 679                 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
 680                 break;
 681
 682         case SO_SNDTIMEO:
 683                 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
 684                 break;
 685
 686         case SO_ATTACH_FILTER:
 687                 ret = -EINVAL;
 688                 if (optlen == sizeof(struct sock_fprog)) {
 689                         struct sock_fprog fprog;
 690
 691                         ret = -EFAULT;
 692                         if (copy_from_user(&fprog, optval, sizeof(fprog)))
 693                                 break;
 694
 695                         ret = sk_attach_filter(&fprog, sk);
 696                 }
 697                 break;
 698
 699         case SO_DETACH_FILTER:
 700                 ret = sk_detach_filter(sk);
 701                 break;
 702
 703         case SO_PASSSEC:
 704                 if (valbool)
 705                         set_bit(SOCK_PASSSEC, &sock->flags);
 706                 else
 707                         clear_bit(SOCK_PASSSEC, &sock->flags);
 708                 break;
 709         case SO_MARK:
 710                 if (!capable(CAP_NET_ADMIN))
 711                         ret = -EPERM;
 712                 else
 713                         sk->sk_mark = val;
 714                 break;
 715
 716                 /* We implement the SO_SNDLOWAT etc to
 717                    not be settable (1003.1g 5.3) */
 718         case SO_RXQ_OVFL:
 719                 if (valbool)
 720                         sock_set_flag(sk, SOCK_RXQ_OVFL);
 721                 else
 722                         sock_reset_flag(sk, SOCK_RXQ_OVFL);
 723                 break;
 724         default:
 725                 ret = -ENOPROTOOPT;
 726                 break;
 727         }
 728         release_sock(sk);
 729         return ret;
 730 }
 731 EXPORT_SYMBOL(sock_setsockopt);
 732
 733
 734 int sock_getsockopt(struct socket *sock, int level, int optname,
 735                     char __user *optval, int __user *optlen)
 736 {
 737         struct sock *sk = sock->sk;
 738
 739         union {
 740                 int val;
 741                 struct linger ling;
 742                 struct timeval tm;
 743         } v;
 744
 745         unsigned int lv = sizeof(int);
 746         int len;
 747
 748         if (get_user(len, optlen))
 749                 return -EFAULT;
 750         if (len < 0)
 751                 return -EINVAL;
 752
 753         memset(&v, 0, sizeof(v));
 754
 755         switch (optname) {
 756         case SO_DEBUG:
 757                 v.val = sock_flag(sk, SOCK_DBG);
 758                 break;
 759
 760         case SO_DONTROUTE:
 761                 v.val = sock_flag(sk, SOCK_LOCALROUTE);
 762                 break;
 763
 764         case SO_BROADCAST:
 765                 v.val = !!sock_flag(sk, SOCK_BROADCAST);
 766                 break;
 767
 768         case SO_SNDBUF:
 769                 v.val = sk->sk_sndbuf;
 770                 break;
 771
 772         case SO_RCVBUF:
 773                 v.val = sk->sk_rcvbuf;
 774                 break;
 775
 776         case SO_REUSEADDR:
 777                 v.val = sk->sk_reuse;
 778                 break;
 779
 780         case SO_KEEPALIVE:
 781                 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
 782                 break;
 783
 784         case SO_TYPE:
 785                 v.val = sk->sk_type;
 786                 break;
 787
 788         case SO_PROTOCOL:
 789                 v.val = sk->sk_protocol;
 790                 break;
 791
 792         case SO_DOMAIN:
 793                 v.val = sk->sk_family;
 794                 break;
 795
 796         case SO_ERROR:
 797                 v.val = -sock_error(sk);
 798                 if (v.val == 0)
 799                         v.val = xchg(&sk->sk_err_soft, 0);
 800                 break;
 801
 802         case SO_OOBINLINE:
 803                 v.val = !!sock_flag(sk, SOCK_URGINLINE);
 804                 break;
 805
 806         case SO_NO_CHECK:
 807                 v.val = sk->sk_no_check;
 808                 break;
 809
 810         case SO_PRIORITY:
 811                 v.val = sk->sk_priority;
 812                 break;
 813
 814         case SO_LINGER:
 815                 lv              = sizeof(v.ling);
 816                 v.ling.l_onoff  = !!sock_flag(sk, SOCK_LINGER);
 817                 v.ling.l_linger = sk->sk_lingertime / HZ;
 818                 break;
 819
 820         case SO_BSDCOMPAT:
 821                 sock_warn_obsolete_bsdism("getsockopt");
 822                 break;
 823
 824         case SO_TIMESTAMP:
 825                 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
 826                                 !sock_flag(sk, SOCK_RCVTSTAMPNS);
 827                 break;
 828
 829         case SO_TIMESTAMPNS:
 830                 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
 831                 break;
 832
 833         case SO_TIMESTAMPING:
 834                 v.val = 0;
 835                 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
 836                         v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
 837                 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
 838                         v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
 839                 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
 840                         v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
 841                 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
 842                         v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
 843                 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
 844                         v.val |= SOF_TIMESTAMPING_SOFTWARE;
 845                 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
 846                         v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
 847                 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
 848                         v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
 849                 break;
 850
 851         case SO_RCVTIMEO:
 852                 lv = sizeof(struct timeval);
 853                 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
 854                         v.tm.tv_sec = 0;
 855                         v.tm.tv_usec = 0;
 856                 } else {
 857                         v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
 858                         v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
 859                 }
 860                 break;
 861
 862         case SO_SNDTIMEO:
 863                 lv = sizeof(struct timeval);
 864                 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
 865                         v.tm.tv_sec = 0;
 866                         v.tm.tv_usec = 0;
 867                 } else {
 868                         v.tm.tv_sec = sk->sk_sndtimeo / HZ;
 869                         v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
 870                 }
 871                 break;
 872
 873         case SO_RCVLOWAT:
 874                 v.val = sk->sk_rcvlowat;
 875                 break;
 876
 877         case SO_SNDLOWAT:
 878                 v.val = 1;
 879                 break;
 880
 881         case SO_PASSCRED:
 882                 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
 883                 break;
 884
 885         case SO_PEERCRED:
 886                 if (len > sizeof(sk->sk_peercred))
 887                         len = sizeof(sk->sk_peercred);
 888                 if (copy_to_user(optval, &sk->sk_peercred, len))
 889                         return -EFAULT;
 890                 goto lenout;
 891
 892         case SO_PEERNAME:
 893         {
 894                 char address[128];
 895
 896                 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
 897                         return -ENOTCONN;
 898                 if (lv < len)
 899                         return -EINVAL;
 900                 if (copy_to_user(optval, address, len))
 901                         return -EFAULT;
 902                 goto lenout;
 903         }
 904
 905         /* Dubious BSD thing... Probably nobody even uses it, but
 906          * the UNIX standard wants it for whatever reason... -DaveM
 907          */
 908         case SO_ACCEPTCONN:
 909                 v.val = sk->sk_state == TCP_LISTEN;
 910                 break;
 911
 912         case SO_PASSSEC:
 913                 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
 914                 break;
 915
 916         case SO_PEERSEC:
 917                 return security_socket_getpeersec_stream(sock, optval, optlen, len);
 918
 919         case SO_MARK:
 920                 v.val = sk->sk_mark;
 921                 break;
 922
 923         case SO_RXQ_OVFL:
 924                 v.val = !!sock_flag(sk, SOCK_RXQ_OVFL);
 925                 break;
 926
 927         default:
 928                 return -ENOPROTOOPT;
 929         }
 930
 931         if (len > lv)
 932                 len = lv;
 933         if (copy_to_user(optval, &v, len))
 934                 return -EFAULT;
 935 lenout:
 936         if (put_user(len, optlen))
 937                 return -EFAULT;
 938         return 0;
 939 }
 940
 941 /*
 942  * Initialize an sk_lock.
 943  *
 944  * (We also register the sk_lock with the lock validator.)
 945  */
 946 static inline void sock_lock_init(struct sock *sk)
 947 {
 948         sock_lock_init_class_and_name(sk,
 949                         af_family_slock_key_strings[sk->sk_family],
 950                         af_family_slock_keys + sk->sk_family,
 951                         af_family_key_strings[sk->sk_family],
 952                         af_family_keys + sk->sk_family);
 953 }
 954
 955 /*
 956  * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
 957  * even temporarly, because of RCU lookups. sk_node should also be left as is.
 958  */
 959 static void sock_copy(struct sock *nsk, const struct sock *osk)
 960 {
 961 #ifdef CONFIG_SECURITY_NETWORK
 962         void *sptr = nsk->sk_security;
 963 #endif
 964         BUILD_BUG_ON(offsetof(struct sock, sk_copy_start) !=
 965                      sizeof(osk->sk_node) + sizeof(osk->sk_refcnt) +
 966                      sizeof(osk->sk_tx_queue_mapping));
 967         memcpy(&nsk->sk_copy_start, &osk->sk_copy_start,
 968                osk->sk_prot->obj_size - offsetof(struct sock, sk_copy_start));
 969 #ifdef CONFIG_SECURITY_NETWORK
 970         nsk->sk_security = sptr;
 971         security_sk_clone(osk, nsk);
 972 #endif
 973 }
 974
 975 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
 976                 int family)
 977 {
 978         struct sock *sk;
 979         struct kmem_cache *slab;
 980
 981         slab = prot->slab;
 982         if (slab != NULL) {
 983                 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
 984                 if (!sk)
 985                         return sk;
 986                 if (priority & __GFP_ZERO) {
 987                         /*
 988                          * caches using SLAB_DESTROY_BY_RCU should let
 989                          * sk_node.next un-modified. Special care is taken
 990                          * when initializing object to zero.
 991                          */
 992                         if (offsetof(struct sock, sk_node.next) != 0)
 993                                 memset(sk, 0, offsetof(struct sock, sk_node.next));
 994                         memset(&sk->sk_node.pprev, 0,
 995                                prot->obj_size - offsetof(struct sock,
 996                                                          sk_node.pprev));
 997                 }
 998         }
 999         else
1000                 sk = kmalloc(prot->obj_size, priority);
1001
1002         if (sk != NULL) {
1003                 kmemcheck_annotate_bitfield(sk, flags);
1004
1005                 if (security_sk_alloc(sk, family, priority))
1006                         goto out_free;
1007
1008                 if (!try_module_get(prot->owner))
1009                         goto out_free_sec;
1010                 sk_tx_queue_clear(sk);
1011         }
1012
1013         return sk;
1014
1015 out_free_sec:
1016         security_sk_free(sk);
1017 out_free:
1018         if (slab != NULL)
1019                 kmem_cache_free(slab, sk);
1020         else
1021                 kfree(sk);
1022         return NULL;
1023 }
1024
1025 static void sk_prot_free(struct proto *prot, struct sock *sk)
1026 {
1027         struct kmem_cache *slab;
1028         struct module *owner;
1029
1030         owner = prot->owner;
1031         slab = prot->slab;
1032
1033         security_sk_free(sk);
1034         if (slab != NULL)
1035                 kmem_cache_free(slab, sk);
1036         else
1037                 kfree(sk);
1038         module_put(owner);
1039 }
1040
1041 /**
1042  *      sk_alloc - All socket objects are allocated here
1043  *      @net: the applicable net namespace
1044  *      @family: protocol family
1045  *      @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1046  *      @prot: struct proto associated with this new sock instance
1047  */
1048 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1049                       struct proto *prot)
1050 {
1051         struct sock *sk;
1052
1053         sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1054         if (sk) {
1055                 sk->sk_family = family;
1056                 /*
1057                  * See comment in struct sock definition to understand
1058                  * why we need sk_prot_creator -acme
1059                  */
1060                 sk->sk_prot = sk->sk_prot_creator = prot;
1061                 sock_lock_init(sk);
1062                 sock_net_set(sk, get_net(net));
1063                 atomic_set(&sk->sk_wmem_alloc, 1);
1064         }
1065
1066         return sk;
1067 }
1068 EXPORT_SYMBOL(sk_alloc);
1069
1070 static void __sk_free(struct sock *sk)
1071 {
1072         struct sk_filter *filter;
1073
1074         if (sk->sk_destruct)
1075                 sk->sk_destruct(sk);
1076
1077         filter = rcu_dereference(sk->sk_filter);
1078         if (filter) {
1079                 sk_filter_uncharge(sk, filter);
1080                 rcu_assign_pointer(sk->sk_filter, NULL);
1081         }
1082
1083         sock_disable_timestamp(sk, SOCK_TIMESTAMP);
1084         sock_disable_timestamp(sk, SOCK_TIMESTAMPING_RX_SOFTWARE);
1085
1086         if (atomic_read(&sk->sk_omem_alloc))
1087                 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
1088                        __func__, atomic_read(&sk->sk_omem_alloc));
1089
1090         put_net(sock_net(sk));
1091         sk_prot_free(sk->sk_prot_creator, sk);
1092 }
1093
1094 void sk_free(struct sock *sk)
1095 {
1096         /*
1097          * We substract one from sk_wmem_alloc and can know if
1098          * some packets are still in some tx queue.
1099          * If not null, sock_wfree() will call __sk_free(sk) later
1100          */
1101         if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1102                 __sk_free(sk);
1103 }
1104 EXPORT_SYMBOL(sk_free);
1105
1106 /*
1107  * Last sock_put should drop referrence to sk->sk_net. It has already
1108  * been dropped in sk_change_net. Taking referrence to stopping namespace
1109  * is not an option.
1110  * Take referrence to a socket to remove it from hash _alive_ and after that
1111  * destroy it in the context of init_net.
1112  */
1113 void sk_release_kernel(struct sock *sk)
1114 {
1115         if (sk == NULL || sk->sk_socket == NULL)
1116                 return;
1117
1118         sock_hold(sk);
1119         sock_release(sk->sk_socket);
1120         release_net(sock_net(sk));
1121         sock_net_set(sk, get_net(&init_net));
1122         sock_put(sk);
1123 }
1124 EXPORT_SYMBOL(sk_release_kernel);
1125
1126 struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
1127 {
1128         struct sock *newsk;
1129
1130         newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1131         if (newsk != NULL) {
1132                 struct sk_filter *filter;
1133
1134                 sock_copy(newsk, sk);
1135
1136                 /* SANITY */
1137                 get_net(sock_net(newsk));
1138                 sk_node_init(&newsk->sk_node);
1139                 sock_lock_init(newsk);
1140                 bh_lock_sock(newsk);
1141                 newsk->sk_backlog.head  = newsk->sk_backlog.tail = NULL;
1142
1143                 atomic_set(&newsk->sk_rmem_alloc, 0);
1144                 /*
1145                  * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1146                  */
1147                 atomic_set(&newsk->sk_wmem_alloc, 1);
1148                 atomic_set(&newsk->sk_omem_alloc, 0);
1149                 skb_queue_head_init(&newsk->sk_receive_queue);
1150                 skb_queue_head_init(&newsk->sk_write_queue);
1151 #ifdef CONFIG_NET_DMA
1152                 skb_queue_head_init(&newsk->sk_async_wait_queue);
1153 #endif
1154
1155                 rwlock_init(&newsk->sk_dst_lock);
1156                 rwlock_init(&newsk->sk_callback_lock);
1157                 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1158                                 af_callback_keys + newsk->sk_family,
1159                                 af_family_clock_key_strings[newsk->sk_family]);
1160
1161                 newsk->sk_dst_cache     = NULL;
1162                 newsk->sk_wmem_queued   = 0;
1163                 newsk->sk_forward_alloc = 0;
1164                 newsk->sk_send_head     = NULL;
1165                 newsk->sk_userlocks     = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1166
1167                 sock_reset_flag(newsk, SOCK_DONE);
1168                 skb_queue_head_init(&newsk->sk_error_queue);
1169
1170                 filter = newsk->sk_filter;
1171                 if (filter != NULL)
1172                         sk_filter_charge(newsk, filter);
1173
1174                 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1175                         /* It is still raw copy of parent, so invalidate
1176                          * destructor and make plain sk_free() */
1177                         newsk->sk_destruct = NULL;
1178                         sk_free(newsk);
1179                         newsk = NULL;
1180                         goto out;
1181                 }
1182
1183                 newsk->sk_err      = 0;
1184                 newsk->sk_priority = 0;
1185                 /*
1186                  * Before updating sk_refcnt, we must commit prior changes to memory
1187                  * (Documentation/RCU/rculist_nulls.txt for details)
1188                  */
1189                 smp_wmb();
1190                 atomic_set(&newsk->sk_refcnt, 2);
1191
1192                 /*
1193                  * Increment the counter in the same struct proto as the master
1194                  * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1195                  * is the same as sk->sk_prot->socks, as this field was copied
1196                  * with memcpy).
1197                  *
1198                  * This _changes_ the previous behaviour, where
1199                  * tcp_create_openreq_child always was incrementing the
1200                  * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1201                  * to be taken into account in all callers. -acme
1202                  */
1203                 sk_refcnt_debug_inc(newsk);
1204                 sk_set_socket(newsk, NULL);
1205                 newsk->sk_sleep  = NULL;
1206
1207                 if (newsk->sk_prot->sockets_allocated)
1208                         percpu_counter_inc(newsk->sk_prot->sockets_allocated);
1209         }
1210 out:
1211         return newsk;
1212 }
1213 EXPORT_SYMBOL_GPL(sk_clone);
1214
1215 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1216 {
1217         __sk_dst_set(sk, dst);
1218         sk->sk_route_caps = dst->dev->features;
1219         if (sk->sk_route_caps & NETIF_F_GSO)
1220                 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1221         if (sk_can_gso(sk)) {
1222                 if (dst->header_len) {
1223                         sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1224                 } else {
1225                         sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1226                         sk->sk_gso_max_size = dst->dev->gso_max_size;
1227                 }
1228         }
1229 }
1230 EXPORT_SYMBOL_GPL(sk_setup_caps);
1231
1232 void __init sk_init(void)
1233 {
1234         if (totalram_pages <= 4096) {
1235                 sysctl_wmem_max = 32767;
1236                 sysctl_rmem_max = 32767;
1237                 sysctl_wmem_default = 32767;
1238                 sysctl_rmem_default = 32767;
1239         } else if (totalram_pages >= 131072) {
1240                 sysctl_wmem_max = 131071;
1241                 sysctl_rmem_max = 131071;
1242         }
1243 }
1244
1245 /*
1246  *      Simple resource managers for sockets.
1247  */
1248
1249
1250 /*
1251  * Write buffer destructor automatically called from kfree_skb.
1252  */
1253 void sock_wfree(struct sk_buff *skb)
1254 {
1255         struct sock *sk = skb->sk;
1256         unsigned int len = skb->truesize;
1257
1258         if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1259                 /*
1260                  * Keep a reference on sk_wmem_alloc, this will be released
1261                  * after sk_write_space() call
1262                  */
1263                 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1264                 sk->sk_write_space(sk);
1265                 len = 1;
1266         }
1267         /*
1268          * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1269          * could not do because of in-flight packets
1270          */
1271         if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1272                 __sk_free(sk);
1273 }
1274 EXPORT_SYMBOL(sock_wfree);
1275
1276 /*
1277  * Read buffer destructor automatically called from kfree_skb.
1278  */
1279 void sock_rfree(struct sk_buff *skb)
1280 {
1281         struct sock *sk = skb->sk;
1282
1283         atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1284         sk_mem_uncharge(skb->sk, skb->truesize);
1285 }
1286 EXPORT_SYMBOL(sock_rfree);
1287
1288
1289 int sock_i_uid(struct sock *sk)
1290 {
1291         int uid;
1292
1293         read_lock(&sk->sk_callback_lock);
1294         uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1295         read_unlock(&sk->sk_callback_lock);
1296         return uid;
1297 }
1298 EXPORT_SYMBOL(sock_i_uid);
1299
1300 unsigned long sock_i_ino(struct sock *sk)
1301 {
1302         unsigned long ino;
1303
1304         read_lock(&sk->sk_callback_lock);
1305         ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1306         read_unlock(&sk->sk_callback_lock);
1307         return ino;
1308 }
1309 EXPORT_SYMBOL(sock_i_ino);
1310
1311 /*
1312  * Allocate a skb from the socket's send buffer.
1313  */
1314 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1315                              gfp_t priority)
1316 {
1317         if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1318                 struct sk_buff *skb = alloc_skb(size, priority);
1319                 if (skb) {
1320                         skb_set_owner_w(skb, sk);
1321                         return skb;
1322                 }
1323         }
1324         return NULL;
1325 }
1326 EXPORT_SYMBOL(sock_wmalloc);
1327
1328 /*
1329  * Allocate a skb from the socket's receive buffer.
1330  */
1331 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1332                              gfp_t priority)
1333 {
1334         if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1335                 struct sk_buff *skb = alloc_skb(size, priority);
1336                 if (skb) {
1337                         skb_set_owner_r(skb, sk);
1338                         return skb;
1339                 }
1340         }
1341         return NULL;
1342 }
1343
1344 /*
1345  * Allocate a memory block from the socket's option memory buffer.
1346  */
1347 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1348 {
1349         if ((unsigned)size <= sysctl_optmem_max &&
1350             atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1351                 void *mem;
1352                 /* First do the add, to avoid the race if kmalloc
1353                  * might sleep.
1354                  */
1355                 atomic_add(size, &sk->sk_omem_alloc);
1356                 mem = kmalloc(size, priority);
1357                 if (mem)
1358                         return mem;
1359                 atomic_sub(size, &sk->sk_omem_alloc);
1360         }
1361         return NULL;
1362 }
1363 EXPORT_SYMBOL(sock_kmalloc);
1364
1365 /*
1366  * Free an option memory block.
1367  */
1368 void sock_kfree_s(struct sock *sk, void *mem, int size)
1369 {
1370         kfree(mem);
1371         atomic_sub(size, &sk->sk_omem_alloc);
1372 }
1373 EXPORT_SYMBOL(sock_kfree_s);
1374
1375 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1376    I think, these locks should be removed for datagram sockets.
1377  */
1378 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1379 {
1380         DEFINE_WAIT(wait);
1381
1382         clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1383         for (;;) {
1384                 if (!timeo)
1385                         break;
1386                 if (signal_pending(current))
1387                         break;
1388                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1389                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1390                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1391                         break;
1392                 if (sk->sk_shutdown & SEND_SHUTDOWN)
1393                         break;
1394                 if (sk->sk_err)
1395                         break;
1396                 timeo = schedule_timeout(timeo);
1397         }
1398         finish_wait(sk->sk_sleep, &wait);
1399         return timeo;
1400 }
1401
1402
1403 /*
1404  *      Generic send/receive buffer handlers
1405  */
1406
1407 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1408                                      unsigned long data_len, int noblock,
1409                                      int *errcode)
1410 {
1411         struct sk_buff *skb;
1412         gfp_t gfp_mask;
1413         long timeo;
1414         int err;
1415
1416         gfp_mask = sk->sk_allocation;
1417         if (gfp_mask & __GFP_WAIT)
1418                 gfp_mask |= __GFP_REPEAT;
1419
1420         timeo = sock_sndtimeo(sk, noblock);
1421         while (1) {
1422                 err = sock_error(sk);
1423                 if (err != 0)
1424                         goto failure;
1425
1426                 err = -EPIPE;
1427                 if (sk->sk_shutdown & SEND_SHUTDOWN)
1428                         goto failure;
1429
1430                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1431                         skb = alloc_skb(header_len, gfp_mask);
1432                         if (skb) {
1433                                 int npages;
1434                                 int i;
1435
1436                                 /* No pages, we're done... */
1437                                 if (!data_len)
1438                                         break;
1439
1440                                 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1441                                 skb->truesize += data_len;
1442                                 skb_shinfo(skb)->nr_frags = npages;
1443                                 for (i = 0; i < npages; i++) {
1444                                         struct page *page;
1445                                         skb_frag_t *frag;
1446
1447                                         page = alloc_pages(sk->sk_allocation, 0);
1448                                         if (!page) {
1449                                                 err = -ENOBUFS;
1450                                                 skb_shinfo(skb)->nr_frags = i;
1451                                                 kfree_skb(skb);
1452                                                 goto failure;
1453                                         }
1454
1455                                         frag = &skb_shinfo(skb)->frags[i];
1456                                         frag->page = page;
1457                                         frag->page_offset = 0;
1458                                         frag->size = (data_len >= PAGE_SIZE ?
1459                                                       PAGE_SIZE :
1460                                                       data_len);
1461                                         data_len -= PAGE_SIZE;
1462                                 }
1463
1464                                 /* Full success... */
1465                                 break;
1466                         }
1467                         err = -ENOBUFS;
1468                         goto failure;
1469                 }
1470                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1471                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1472                 err = -EAGAIN;
1473                 if (!timeo)
1474                         goto failure;
1475                 if (signal_pending(current))
1476                         goto interrupted;
1477                 timeo = sock_wait_for_wmem(sk, timeo);
1478         }
1479
1480         skb_set_owner_w(skb, sk);
1481         return skb;
1482
1483 interrupted:
1484         err = sock_intr_errno(timeo);
1485 failure:
1486         *errcode = err;
1487         return NULL;
1488 }
1489 EXPORT_SYMBOL(sock_alloc_send_pskb);
1490
1491 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1492                                     int noblock, int *errcode)
1493 {
1494         return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1495 }
1496 EXPORT_SYMBOL(sock_alloc_send_skb);
1497
1498 static void __lock_sock(struct sock *sk)
1499 {
1500         DEFINE_WAIT(wait);
1501
1502         for (;;) {
1503                 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1504                                         TASK_UNINTERRUPTIBLE);
1505                 spin_unlock_bh(&sk->sk_lock.slock);
1506                 schedule();
1507                 spin_lock_bh(&sk->sk_lock.slock);
1508                 if (!sock_owned_by_user(sk))
1509                         break;
1510         }
1511         finish_wait(&sk->sk_lock.wq, &wait);
1512 }
1513
1514 static void __release_sock(struct sock *sk)
1515 {
1516         struct sk_buff *skb = sk->sk_backlog.head;
1517
1518         do {
1519                 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1520                 bh_unlock_sock(sk);
1521
1522                 do {
1523                         struct sk_buff *next = skb->next;
1524
1525                         skb->next = NULL;
1526                         sk_backlog_rcv(sk, skb);
1527
1528                         /*
1529                          * We are in process context here with softirqs
1530                          * disabled, use cond_resched_softirq() to preempt.
1531                          * This is safe to do because we've taken the backlog
1532                          * queue private:
1533                          */
1534                         cond_resched_softirq();
1535
1536                         skb = next;
1537                 } while (skb != NULL);
1538
1539                 bh_lock_sock(sk);
1540         } while ((skb = sk->sk_backlog.head) != NULL);
1541 }
1542
1543 /**
1544  * sk_wait_data - wait for data to arrive at sk_receive_queue
1545  * @sk:    sock to wait on
1546  * @timeo: for how long
1547  *
1548  * Now socket state including sk->sk_err is changed only under lock,
1549  * hence we may omit checks after joining wait queue.
1550  * We check receive queue before schedule() only as optimization;
1551  * it is very likely that release_sock() added new data.
1552  */
1553 int sk_wait_data(struct sock *sk, long *timeo)
1554 {
1555         int rc;
1556         DEFINE_WAIT(wait);
1557
1558         prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1559         set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1560         rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1561         clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1562         finish_wait(sk->sk_sleep, &wait);
1563         return rc;
1564 }
1565 EXPORT_SYMBOL(sk_wait_data);
1566
1567 /**
1568  *      __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1569  *      @sk: socket
1570  *      @size: memory size to allocate
1571  *      @kind: allocation type
1572  *
1573  *      If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1574  *      rmem allocation. This function assumes that protocols which have
1575  *      memory_pressure use sk_wmem_queued as write buffer accounting.
1576  */
1577 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1578 {
1579         struct proto *prot = sk->sk_prot;
1580         int amt = sk_mem_pages(size);
1581         int allocated;
1582
1583         sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1584         allocated = atomic_add_return(amt, prot->memory_allocated);
1585
1586         /* Under limit. */
1587         if (allocated <= prot->sysctl_mem[0]) {
1588                 if (prot->memory_pressure && *prot->memory_pressure)
1589                         *prot->memory_pressure = 0;
1590                 return 1;
1591         }
1592
1593         /* Under pressure. */
1594         if (allocated > prot->sysctl_mem[1])
1595                 if (prot->enter_memory_pressure)
1596                         prot->enter_memory_pressure(sk);
1597
1598         /* Over hard limit. */
1599         if (allocated > prot->sysctl_mem[2])
1600                 goto suppress_allocation;
1601
1602         /* guarantee minimum buffer size under pressure */
1603         if (kind == SK_MEM_RECV) {
1604                 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1605                         return 1;
1606         } else { /* SK_MEM_SEND */
1607                 if (sk->sk_type == SOCK_STREAM) {
1608                         if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1609                                 return 1;
1610                 } else if (atomic_read(&sk->sk_wmem_alloc) <
1611                            prot->sysctl_wmem[0])
1612                                 return 1;
1613         }
1614
1615         if (prot->memory_pressure) {
1616                 int alloc;
1617
1618                 if (!*prot->memory_pressure)
1619                         return 1;
1620                 alloc = percpu_counter_read_positive(prot->sockets_allocated);
1621                 if (prot->sysctl_mem[2] > alloc *
1622                     sk_mem_pages(sk->sk_wmem_queued +
1623                                  atomic_read(&sk->sk_rmem_alloc) +
1624                                  sk->sk_forward_alloc))
1625                         return 1;
1626         }
1627
1628 suppress_allocation:
1629
1630         if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1631                 sk_stream_moderate_sndbuf(sk);
1632
1633                 /* Fail only if socket is _under_ its sndbuf.
1634                  * In this case we cannot block, so that we have to fail.
1635                  */
1636                 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1637                         return 1;
1638         }
1639
1640         /* Alas. Undo changes. */
1641         sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1642         atomic_sub(amt, prot->memory_allocated);
1643         return 0;
1644 }
1645 EXPORT_SYMBOL(__sk_mem_schedule);
1646
1647 /**
1648  *      __sk_reclaim - reclaim memory_allocated
1649  *      @sk: socket
1650  */
1651 void __sk_mem_reclaim(struct sock *sk)
1652 {
1653         struct proto *prot = sk->sk_prot;
1654
1655         atomic_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
1656                    prot->memory_allocated);
1657         sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1658
1659         if (prot->memory_pressure && *prot->memory_pressure &&
1660             (atomic_read(prot->memory_allocated) < prot->sysctl_mem[0]))
1661                 *prot->memory_pressure = 0;
1662 }
1663 EXPORT_SYMBOL(__sk_mem_reclaim);
1664
1665
1666 /*
1667  * Set of default routines for initialising struct proto_ops when
1668  * the protocol does not support a particular function. In certain
1669  * cases where it makes no sense for a protocol to have a "do nothing"
1670  * function, some default processing is provided.
1671  */
1672
1673 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1674 {
1675         return -EOPNOTSUPP;
1676 }
1677 EXPORT_SYMBOL(sock_no_bind);
1678
1679 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1680                     int len, int flags)
1681 {
1682         return -EOPNOTSUPP;
1683 }
1684 EXPORT_SYMBOL(sock_no_connect);
1685
1686 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1687 {
1688         return -EOPNOTSUPP;
1689 }
1690 EXPORT_SYMBOL(sock_no_socketpair);
1691
1692 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1693 {
1694         return -EOPNOTSUPP;
1695 }
1696 EXPORT_SYMBOL(sock_no_accept);
1697
1698 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1699                     int *len, int peer)
1700 {
1701         return -EOPNOTSUPP;
1702 }
1703 EXPORT_SYMBOL(sock_no_getname);
1704
1705 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1706 {
1707         return 0;
1708 }
1709 EXPORT_SYMBOL(sock_no_poll);
1710
1711 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1712 {
1713         return -EOPNOTSUPP;
1714 }
1715 EXPORT_SYMBOL(sock_no_ioctl);
1716
1717 int sock_no_listen(struct socket *sock, int backlog)
1718 {
1719         return -EOPNOTSUPP;
1720 }
1721 EXPORT_SYMBOL(sock_no_listen);
1722
1723 int sock_no_shutdown(struct socket *sock, int how)
1724 {
1725         return -EOPNOTSUPP;
1726 }
1727 EXPORT_SYMBOL(sock_no_shutdown);
1728
1729 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1730                     char __user *optval, unsigned int optlen)
1731 {
1732         return -EOPNOTSUPP;
1733 }
1734 EXPORT_SYMBOL(sock_no_setsockopt);
1735
1736 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1737                     char __user *optval, int __user *optlen)
1738 {
1739         return -EOPNOTSUPP;
1740 }
1741 EXPORT_SYMBOL(sock_no_getsockopt);
1742
1743 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1744                     size_t len)
1745 {
1746         return -EOPNOTSUPP;
1747 }
1748 EXPORT_SYMBOL(sock_no_sendmsg);
1749
1750 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1751                     size_t len, int flags)
1752 {
1753         return -EOPNOTSUPP;
1754 }
1755 EXPORT_SYMBOL(sock_no_recvmsg);
1756
1757 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1758 {
1759         /* Mirror missing mmap method error code */
1760         return -ENODEV;
1761 }
1762 EXPORT_SYMBOL(sock_no_mmap);
1763
1764 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1765 {
1766         ssize_t res;
1767         struct msghdr msg = {.msg_flags = flags};
1768         struct kvec iov;
1769         char *kaddr = kmap(page);
1770         iov.iov_base = kaddr + offset;
1771         iov.iov_len = size;
1772         res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1773         kunmap(page);
1774         return res;
1775 }
1776 EXPORT_SYMBOL(sock_no_sendpage);
1777
1778 /*
1779  *      Default Socket Callbacks
1780  */
1781
1782 static void sock_def_wakeup(struct sock *sk)
1783 {
1784         read_lock(&sk->sk_callback_lock);
1785         if (sk_has_sleeper(sk))
1786                 wake_up_interruptible_all(sk->sk_sleep);
1787         read_unlock(&sk->sk_callback_lock);
1788 }
1789
1790 static void sock_def_error_report(struct sock *sk)
1791 {
1792         read_lock(&sk->sk_callback_lock);
1793         if (sk_has_sleeper(sk))
1794                 wake_up_interruptible_poll(sk->sk_sleep, POLLERR);
1795         sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
1796         read_unlock(&sk->sk_callback_lock);
1797 }
1798
1799 static void sock_def_readable(struct sock *sk, int len)
1800 {
1801         read_lock(&sk->sk_callback_lock);
1802         if (sk_has_sleeper(sk))
1803                 wake_up_interruptible_sync_poll(sk->sk_sleep, POLLIN |
1804                                                 POLLRDNORM | POLLRDBAND);
1805         sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1806         read_unlock(&sk->sk_callback_lock);
1807 }
1808
1809 static void sock_def_write_space(struct sock *sk)
1810 {
1811         read_lock(&sk->sk_callback_lock);
1812
1813         /* Do not wake up a writer until he can make "significant"
1814          * progress.  --DaveM
1815          */
1816         if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1817                 if (sk_has_sleeper(sk))
1818                         wake_up_interruptible_sync_poll(sk->sk_sleep, POLLOUT |
1819                                                 POLLWRNORM | POLLWRBAND);
1820
1821                 /* Should agree with poll, otherwise some programs break */
1822                 if (sock_writeable(sk))
1823                         sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1824         }
1825
1826         read_unlock(&sk->sk_callback_lock);
1827 }
1828
1829 static void sock_def_destruct(struct sock *sk)
1830 {
1831         kfree(sk->sk_protinfo);
1832 }
1833
1834 void sk_send_sigurg(struct sock *sk)
1835 {
1836         if (sk->sk_socket && sk->sk_socket->file)
1837                 if (send_sigurg(&sk->sk_socket->file->f_owner))
1838                         sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1839 }
1840 EXPORT_SYMBOL(sk_send_sigurg);
1841
1842 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1843                     unsigned long expires)
1844 {
1845         if (!mod_timer(timer, expires))
1846                 sock_hold(sk);
1847 }
1848 EXPORT_SYMBOL(sk_reset_timer);
1849
1850 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1851 {
1852         if (timer_pending(timer) && del_timer(timer))
1853                 __sock_put(sk);
1854 }
1855 EXPORT_SYMBOL(sk_stop_timer);
1856
1857 void sock_init_data(struct socket *sock, struct sock *sk)
1858 {
1859         skb_queue_head_init(&sk->sk_receive_queue);
1860         skb_queue_head_init(&sk->sk_write_queue);
1861         skb_queue_head_init(&sk->sk_error_queue);
1862 #ifdef CONFIG_NET_DMA
1863         skb_queue_head_init(&sk->sk_async_wait_queue);
1864 #endif
1865
1866         sk->sk_send_head        =       NULL;
1867
1868         init_timer(&sk->sk_timer);
1869
1870         sk->sk_allocation       =       GFP_KERNEL;
1871         sk->sk_rcvbuf           =       sysctl_rmem_default;
1872         sk->sk_sndbuf           =       sysctl_wmem_default;
1873         sk->sk_state            =       TCP_CLOSE;
1874         sk_set_socket(sk, sock);
1875
1876         sock_set_flag(sk, SOCK_ZAPPED);
1877
1878         if (sock) {
1879                 sk->sk_type     =       sock->type;
1880                 sk->sk_sleep    =       &sock->wait;
1881                 sock->sk        =       sk;
1882         } else
1883                 sk->sk_sleep    =       NULL;
1884
1885         rwlock_init(&sk->sk_dst_lock);
1886         rwlock_init(&sk->sk_callback_lock);
1887         lockdep_set_class_and_name(&sk->sk_callback_lock,
1888                         af_callback_keys + sk->sk_family,
1889                         af_family_clock_key_strings[sk->sk_family]);
1890
1891         sk->sk_state_change     =       sock_def_wakeup;
1892         sk->sk_data_ready       =       sock_def_readable;
1893         sk->sk_write_space      =       sock_def_write_space;
1894         sk->sk_error_report     =       sock_def_error_report;
1895         sk->sk_destruct         =       sock_def_destruct;
1896
1897         sk->sk_sndmsg_page      =       NULL;
1898         sk->sk_sndmsg_off       =       0;
1899
1900         sk->sk_peercred.pid     =       0;
1901         sk->sk_peercred.uid     =       -1;
1902         sk->sk_peercred.gid     =       -1;
1903         sk->sk_write_pending    =       0;
1904         sk->sk_rcvlowat         =       1;
1905         sk->sk_rcvtimeo         =       MAX_SCHEDULE_TIMEOUT;
1906         sk->sk_sndtimeo         =       MAX_SCHEDULE_TIMEOUT;
1907
1908         sk->sk_stamp = ktime_set(-1L, 0);
1909
1910         /*
1911          * Before updating sk_refcnt, we must commit prior changes to memory
1912          * (Documentation/RCU/rculist_nulls.txt for details)
1913          */
1914         smp_wmb();
1915         atomic_set(&sk->sk_refcnt, 1);
1916         atomic_set(&sk->sk_drops, 0);
1917 }
1918 EXPORT_SYMBOL(sock_init_data);
1919
1920 void lock_sock_nested(struct sock *sk, int subclass)
1921 {
1922         might_sleep();
1923         spin_lock_bh(&sk->sk_lock.slock);
1924         if (sk->sk_lock.owned)
1925                 __lock_sock(sk);
1926         sk->sk_lock.owned = 1;
1927         spin_unlock(&sk->sk_lock.slock);
1928         /*
1929          * The sk_lock has mutex_lock() semantics here:
1930          */
1931         mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
1932         local_bh_enable();
1933 }
1934 EXPORT_SYMBOL(lock_sock_nested);
1935
1936 void release_sock(struct sock *sk)
1937 {
1938         /*
1939          * The sk_lock has mutex_unlock() semantics:
1940          */
1941         mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1942
1943         spin_lock_bh(&sk->sk_lock.slock);
1944         if (sk->sk_backlog.tail)
1945                 __release_sock(sk);
1946         sk->sk_lock.owned = 0;
1947         if (waitqueue_active(&sk->sk_lock.wq))
1948                 wake_up(&sk->sk_lock.wq);
1949         spin_unlock_bh(&sk->sk_lock.slock);
1950 }
1951 EXPORT_SYMBOL(release_sock);
1952
1953 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1954 {
1955         struct timeval tv;
1956         if (!sock_flag(sk, SOCK_TIMESTAMP))
1957                 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
1958         tv = ktime_to_timeval(sk->sk_stamp);
1959         if (tv.tv_sec == -1)
1960                 return -ENOENT;
1961         if (tv.tv_sec == 0) {
1962                 sk->sk_stamp = ktime_get_real();
1963                 tv = ktime_to_timeval(sk->sk_stamp);
1964         }
1965         return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
1966 }
1967 EXPORT_SYMBOL(sock_get_timestamp);
1968
1969 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
1970 {
1971         struct timespec ts;
1972         if (!sock_flag(sk, SOCK_TIMESTAMP))
1973                 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
1974         ts = ktime_to_timespec(sk->sk_stamp);
1975         if (ts.tv_sec == -1)
1976                 return -ENOENT;
1977         if (ts.tv_sec == 0) {
1978                 sk->sk_stamp = ktime_get_real();
1979                 ts = ktime_to_timespec(sk->sk_stamp);
1980         }
1981         return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
1982 }
1983 EXPORT_SYMBOL(sock_get_timestampns);
1984
1985 void sock_enable_timestamp(struct sock *sk, int flag)
1986 {
1987         if (!sock_flag(sk, flag)) {
1988                 sock_set_flag(sk, flag);
1989                 /*
1990                  * we just set one of the two flags which require net
1991                  * time stamping, but time stamping might have been on
1992                  * already because of the other one
1993                  */
1994                 if (!sock_flag(sk,
1995                                 flag == SOCK_TIMESTAMP ?
1996                                 SOCK_TIMESTAMPING_RX_SOFTWARE :
1997                                 SOCK_TIMESTAMP))
1998                         net_enable_timestamp();
1999         }
2000 }
2001
2002 /*
2003  *      Get a socket option on an socket.
2004  *
2005  *      FIX: POSIX 1003.1g is very ambiguous here. It states that
2006  *      asynchronous errors should be reported by getsockopt. We assume
2007  *      this means if you specify SO_ERROR (otherwise whats the point of it).
2008  */
2009 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2010                            char __user *optval, int __user *optlen)
2011 {
2012         struct sock *sk = sock->sk;
2013
2014         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2015 }
2016 EXPORT_SYMBOL(sock_common_getsockopt);
2017
2018 #ifdef CONFIG_COMPAT
2019 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2020                                   char __user *optval, int __user *optlen)
2021 {
2022         struct sock *sk = sock->sk;
2023
2024         if (sk->sk_prot->compat_getsockopt != NULL)
2025                 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2026                                                       optval, optlen);
2027         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2028 }
2029 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2030 #endif
2031
2032 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2033                         struct msghdr *msg, size_t size, int flags)
2034 {
2035         struct sock *sk = sock->sk;
2036         int addr_len = 0;
2037         int err;
2038
2039         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2040                                    flags & ~MSG_DONTWAIT, &addr_len);
2041         if (err >= 0)
2042                 msg->msg_namelen = addr_len;
2043         return err;
2044 }
2045 EXPORT_SYMBOL(sock_common_recvmsg);
2046
2047 /*
2048  *      Set socket options on an inet socket.
2049  */
2050 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2051                            char __user *optval, unsigned int optlen)
2052 {
2053         struct sock *sk = sock->sk;
2054
2055         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2056 }
2057 EXPORT_SYMBOL(sock_common_setsockopt);
2058
2059 #ifdef CONFIG_COMPAT
2060 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2061                                   char __user *optval, unsigned int optlen)
2062 {
2063         struct sock *sk = sock->sk;
2064
2065         if (sk->sk_prot->compat_setsockopt != NULL)
2066                 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2067                                                       optval, optlen);
2068         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2069 }
2070 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2071 #endif
2072
2073 void sk_common_release(struct sock *sk)
2074 {
2075         if (sk->sk_prot->destroy)
2076                 sk->sk_prot->destroy(sk);
2077
2078         /*
2079          * Observation: when sock_common_release is called, processes have
2080          * no access to socket. But net still has.
2081          * Step one, detach it from networking:
2082          *
2083          * A. Remove from hash tables.
2084          */
2085
2086         sk->sk_prot->unhash(sk);
2087
2088         /*
2089          * In this point socket cannot receive new packets, but it is possible
2090          * that some packets are in flight because some CPU runs receiver and
2091          * did hash table lookup before we unhashed socket. They will achieve
2092          * receive queue and will be purged by socket destructor.
2093          *
2094          * Also we still have packets pending on receive queue and probably,
2095          * our own packets waiting in device queues. sock_destroy will drain
2096          * receive queue, but transmitted packets will delay socket destruction
2097          * until the last reference will be released.
2098          */
2099
2100         sock_orphan(sk);
2101
2102         xfrm_sk_free_policy(sk);
2103
2104         sk_refcnt_debug_release(sk);
2105         sock_put(sk);
2106 }
2107 EXPORT_SYMBOL(sk_common_release);
2108
2109 static DEFINE_RWLOCK(proto_list_lock);
2110 static LIST_HEAD(proto_list);
2111
2112 #ifdef CONFIG_PROC_FS
2113 #define PROTO_INUSE_NR  64      /* should be enough for the first time */
2114 struct prot_inuse {
2115         int val[PROTO_INUSE_NR];
2116 };
2117
2118 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2119
2120 #ifdef CONFIG_NET_NS
2121 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2122 {
2123         int cpu = smp_processor_id();
2124         per_cpu_ptr(net->core.inuse, cpu)->val[prot->inuse_idx] += val;
2125 }
2126 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2127
2128 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2129 {
2130         int cpu, idx = prot->inuse_idx;
2131         int res = 0;
2132
2133         for_each_possible_cpu(cpu)
2134                 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2135
2136         return res >= 0 ? res : 0;
2137 }
2138 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2139
2140 static int sock_inuse_init_net(struct net *net)
2141 {
2142         net->core.inuse = alloc_percpu(struct prot_inuse);
2143         return net->core.inuse ? 0 : -ENOMEM;
2144 }
2145
2146 static void sock_inuse_exit_net(struct net *net)
2147 {
2148         free_percpu(net->core.inuse);
2149 }
2150
2151 static struct pernet_operations net_inuse_ops = {
2152         .init = sock_inuse_init_net,
2153         .exit = sock_inuse_exit_net,
2154 };
2155
2156 static __init int net_inuse_init(void)
2157 {
2158         if (register_pernet_subsys(&net_inuse_ops))
2159                 panic("Cannot initialize net inuse counters");
2160
2161         return 0;
2162 }
2163
2164 core_initcall(net_inuse_init);
2165 #else
2166 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2167
2168 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2169 {
2170         __get_cpu_var(prot_inuse).val[prot->inuse_idx] += val;
2171 }
2172 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2173
2174 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2175 {
2176         int cpu, idx = prot->inuse_idx;
2177         int res = 0;
2178
2179         for_each_possible_cpu(cpu)
2180                 res += per_cpu(prot_inuse, cpu).val[idx];
2181
2182         return res >= 0 ? res : 0;
2183 }
2184 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2185 #endif
2186
2187 static void assign_proto_idx(struct proto *prot)
2188 {
2189         prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2190
2191         if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2192                 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2193                 return;
2194         }
2195
2196         set_bit(prot->inuse_idx, proto_inuse_idx);
2197 }
2198
2199 static void release_proto_idx(struct proto *prot)
2200 {
2201         if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2202                 clear_bit(prot->inuse_idx, proto_inuse_idx);
2203 }
2204 #else
2205 static inline void assign_proto_idx(struct proto *prot)
2206 {
2207 }
2208
2209 static inline void release_proto_idx(struct proto *prot)
2210 {
2211 }
2212 #endif
2213
2214 int proto_register(struct proto *prot, int alloc_slab)
2215 {
2216         if (alloc_slab) {
2217                 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2218                                         SLAB_HWCACHE_ALIGN | prot->slab_flags,
2219                                         NULL);
2220
2221                 if (prot->slab == NULL) {
2222                         printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2223                                prot->name);
2224                         goto out;
2225                 }
2226
2227                 if (prot->rsk_prot != NULL) {
2228                         static const char mask[] = "request_sock_%s";
2229
2230                         prot->rsk_prot->slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
2231                         if (prot->rsk_prot->slab_name == NULL)
2232                                 goto out_free_sock_slab;
2233
2234                         sprintf(prot->rsk_prot->slab_name, mask, prot->name);
2235                         prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2236                                                                  prot->rsk_prot->obj_size, 0,
2237                                                                  SLAB_HWCACHE_ALIGN, NULL);
2238
2239                         if (prot->rsk_prot->slab == NULL) {
2240                                 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2241                                        prot->name);
2242                                 goto out_free_request_sock_slab_name;
2243                         }
2244                 }
2245
2246                 if (prot->twsk_prot != NULL) {
2247                         static const char mask[] = "tw_sock_%s";
2248
2249                         prot->twsk_prot->twsk_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
2250
2251                         if (prot->twsk_prot->twsk_slab_name == NULL)
2252                                 goto out_free_request_sock_slab;
2253
2254                         sprintf(prot->twsk_prot->twsk_slab_name, mask, prot->name);
2255                         prot->twsk_prot->twsk_slab =
2256                                 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2257                                                   prot->twsk_prot->twsk_obj_size,
2258                                                   0,
2259                                                   SLAB_HWCACHE_ALIGN |
2260                                                         prot->slab_flags,
2261                                                   NULL);
2262                         if (prot->twsk_prot->twsk_slab == NULL)
2263                                 goto out_free_timewait_sock_slab_name;
2264                 }
2265         }
2266
2267         write_lock(&proto_list_lock);
2268         list_add(&prot->node, &proto_list);
2269         assign_proto_idx(prot);
2270         write_unlock(&proto_list_lock);
2271         return 0;
2272
2273 out_free_timewait_sock_slab_name:
2274         kfree(prot->twsk_prot->twsk_slab_name);
2275 out_free_request_sock_slab:
2276         if (prot->rsk_prot && prot->rsk_prot->slab) {
2277                 kmem_cache_destroy(prot->rsk_prot->slab);
2278                 prot->rsk_prot->slab = NULL;
2279         }
2280 out_free_request_sock_slab_name:
2281         kfree(prot->rsk_prot->slab_name);
2282 out_free_sock_slab:
2283         kmem_cache_destroy(prot->slab);
2284         prot->slab = NULL;
2285 out:
2286         return -ENOBUFS;
2287 }
2288 EXPORT_SYMBOL(proto_register);
2289
2290 void proto_unregister(struct proto *prot)
2291 {
2292         write_lock(&proto_list_lock);
2293         release_proto_idx(prot);
2294         list_del(&prot->node);
2295         write_unlock(&proto_list_lock);
2296
2297         if (prot->slab != NULL) {
2298                 kmem_cache_destroy(prot->slab);
2299                 prot->slab = NULL;
2300         }
2301
2302         if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2303                 kmem_cache_destroy(prot->rsk_prot->slab);
2304                 kfree(prot->rsk_prot->slab_name);
2305                 prot->rsk_prot->slab = NULL;
2306         }
2307
2308         if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2309                 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2310                 kfree(prot->twsk_prot->twsk_slab_name);
2311                 prot->twsk_prot->twsk_slab = NULL;
2312         }
2313 }
2314 EXPORT_SYMBOL(proto_unregister);
2315
2316 #ifdef CONFIG_PROC_FS
2317 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2318         __acquires(proto_list_lock)
2319 {
2320         read_lock(&proto_list_lock);
2321         return seq_list_start_head(&proto_list, *pos);
2322 }
2323
2324 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2325 {
2326         return seq_list_next(v, &proto_list, pos);
2327 }
2328
2329 static void proto_seq_stop(struct seq_file *seq, void *v)
2330         __releases(proto_list_lock)
2331 {
2332         read_unlock(&proto_list_lock);
2333 }
2334
2335 static char proto_method_implemented(const void *method)
2336 {
2337         return method == NULL ? 'n' : 'y';
2338 }
2339
2340 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2341 {
2342         seq_printf(seq, "%-9s %4u %6d  %6d   %-3s %6u   %-3s  %-10s "
2343                         "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2344                    proto->name,
2345                    proto->obj_size,
2346                    sock_prot_inuse_get(seq_file_net(seq), proto),
2347                    proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
2348                    proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2349                    proto->max_header,
2350                    proto->slab == NULL ? "no" : "yes",
2351                    module_name(proto->owner),
2352                    proto_method_implemented(proto->close),
2353                    proto_method_implemented(proto->connect),
2354                    proto_method_implemented(proto->disconnect),
2355                    proto_method_implemented(proto->accept),
2356                    proto_method_implemented(proto->ioctl),
2357                    proto_method_implemented(proto->init),
2358                    proto_method_implemented(proto->destroy),
2359                    proto_method_implemented(proto->shutdown),
2360                    proto_method_implemented(proto->setsockopt),
2361                    proto_method_implemented(proto->getsockopt),
2362                    proto_method_implemented(proto->sendmsg),
2363                    proto_method_implemented(proto->recvmsg),
2364                    proto_method_implemented(proto->sendpage),
2365                    proto_method_implemented(proto->bind),
2366                    proto_method_implemented(proto->backlog_rcv),
2367                    proto_method_implemented(proto->hash),
2368                    proto_method_implemented(proto->unhash),
2369                    proto_method_implemented(proto->get_port),
2370                    proto_method_implemented(proto->enter_memory_pressure));
2371 }
2372
2373 static int proto_seq_show(struct seq_file *seq, void *v)
2374 {
2375         if (v == &proto_list)
2376                 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2377                            "protocol",
2378                            "size",
2379                            "sockets",
2380                            "memory",
2381                            "press",
2382                            "maxhdr",
2383                            "slab",
2384                            "module",
2385                            "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2386         else
2387                 proto_seq_printf(seq, list_entry(v, struct proto, node));
2388         return 0;
2389 }
2390
2391 static const struct seq_operations proto_seq_ops = {
2392         .start  = proto_seq_start,
2393         .next   = proto_seq_next,
2394         .stop   = proto_seq_stop,
2395         .show   = proto_seq_show,
2396 };
2397
2398 static int proto_seq_open(struct inode *inode, struct file *file)
2399 {
2400         return seq_open_net(inode, file, &proto_seq_ops,
2401                             sizeof(struct seq_net_private));
2402 }
2403
2404 static const struct file_operations proto_seq_fops = {
2405         .owner          = THIS_MODULE,
2406         .open           = proto_seq_open,
2407         .read           = seq_read,
2408         .llseek         = seq_lseek,
2409         .release        = seq_release_net,
2410 };
2411
2412 static __net_init int proto_init_net(struct net *net)
2413 {
2414         if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2415                 return -ENOMEM;
2416
2417         return 0;
2418 }
2419
2420 static __net_exit void proto_exit_net(struct net *net)
2421 {
2422         proc_net_remove(net, "protocols");
2423 }
2424
2425
2426 static __net_initdata struct pernet_operations proto_net_ops = {
2427         .init = proto_init_net,
2428         .exit = proto_exit_net,
2429 };
2430
2431 static int __init proto_init(void)
2432 {
2433         return register_pernet_subsys(&proto_net_ops);
2434 }
2435
2436 subsys_initcall(proto_init);
2437
2438 #endif /* PROC_FS */