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