SAFE public projects git trees. - safe/jmp/linux-2.6/blob - net/core/sock.c

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