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

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