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);
 944         else
 945                 sk = kmalloc(prot->obj_size, priority);
 946
 947         if (sk != NULL) {
 948                 kmemcheck_annotate_bitfield(sk, flags);
 949
 950                 if (security_sk_alloc(sk, family, priority))
 951                         goto out_free;
 952
 953                 if (!try_module_get(prot->owner))
 954                         goto out_free_sec;
 955         }
 956
 957         return sk;
 958
 959 out_free_sec:
 960         security_sk_free(sk);
 961 out_free:
 962         if (slab != NULL)
 963                 kmem_cache_free(slab, sk);
 964         else
 965                 kfree(sk);
 966         return NULL;
 967 }
 968
 969 static void sk_prot_free(struct proto *prot, struct sock *sk)
 970 {
 971         struct kmem_cache *slab;
 972         struct module *owner;
 973
 974         owner = prot->owner;
 975         slab = prot->slab;
 976
 977         security_sk_free(sk);
 978         if (slab != NULL)
 979                 kmem_cache_free(slab, sk);
 980         else
 981                 kfree(sk);
 982         module_put(owner);
 983 }
 984
 985 /**
 986  *      sk_alloc - All socket objects are allocated here
 987  *      @net: the applicable net namespace
 988  *      @family: protocol family
 989  *      @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
 990  *      @prot: struct proto associated with this new sock instance
 991  */
 992 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
 993                       struct proto *prot)
 994 {
 995         struct sock *sk;
 996
 997         sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
 998         if (sk) {
 999                 sk->sk_family = family;
1000                 /*
1001                  * See comment in struct sock definition to understand
1002                  * why we need sk_prot_creator -acme
1003                  */
1004                 sk->sk_prot = sk->sk_prot_creator = prot;
1005                 sock_lock_init(sk);
1006                 sock_net_set(sk, get_net(net));
1007         }
1008
1009         return sk;
1010 }
1011 EXPORT_SYMBOL(sk_alloc);
1012
1013 static void __sk_free(struct sock *sk)
1014 {
1015         struct sk_filter *filter;
1016
1017         if (sk->sk_destruct)
1018                 sk->sk_destruct(sk);
1019
1020         filter = rcu_dereference(sk->sk_filter);
1021         if (filter) {
1022                 sk_filter_uncharge(sk, filter);
1023                 rcu_assign_pointer(sk->sk_filter, NULL);
1024         }
1025
1026         sock_disable_timestamp(sk, SOCK_TIMESTAMP);
1027         sock_disable_timestamp(sk, SOCK_TIMESTAMPING_RX_SOFTWARE);
1028
1029         if (atomic_read(&sk->sk_omem_alloc))
1030                 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
1031                        __func__, atomic_read(&sk->sk_omem_alloc));
1032
1033         put_net(sock_net(sk));
1034         sk_prot_free(sk->sk_prot_creator, sk);
1035 }
1036
1037 void sk_free(struct sock *sk)
1038 {
1039         /*
1040          * We substract one from sk_wmem_alloc and can know if
1041          * some packets are still in some tx queue.
1042          * If not null, sock_wfree() will call __sk_free(sk) later
1043          */
1044         if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1045                 __sk_free(sk);
1046 }
1047 EXPORT_SYMBOL(sk_free);
1048
1049 /*
1050  * Last sock_put should drop referrence to sk->sk_net. It has already
1051  * been dropped in sk_change_net. Taking referrence to stopping namespace
1052  * is not an option.
1053  * Take referrence to a socket to remove it from hash _alive_ and after that
1054  * destroy it in the context of init_net.
1055  */
1056 void sk_release_kernel(struct sock *sk)
1057 {
1058         if (sk == NULL || sk->sk_socket == NULL)
1059                 return;
1060
1061         sock_hold(sk);
1062         sock_release(sk->sk_socket);
1063         release_net(sock_net(sk));
1064         sock_net_set(sk, get_net(&init_net));
1065         sock_put(sk);
1066 }
1067 EXPORT_SYMBOL(sk_release_kernel);
1068
1069 struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
1070 {
1071         struct sock *newsk;
1072
1073         newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1074         if (newsk != NULL) {
1075                 struct sk_filter *filter;
1076
1077                 sock_copy(newsk, sk);
1078
1079                 /* SANITY */
1080                 get_net(sock_net(newsk));
1081                 sk_node_init(&newsk->sk_node);
1082                 sock_lock_init(newsk);
1083                 bh_lock_sock(newsk);
1084                 newsk->sk_backlog.head  = newsk->sk_backlog.tail = NULL;
1085
1086                 atomic_set(&newsk->sk_rmem_alloc, 0);
1087                 /*
1088                  * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1089                  */
1090                 atomic_set(&newsk->sk_wmem_alloc, 1);
1091                 atomic_set(&newsk->sk_omem_alloc, 0);
1092                 skb_queue_head_init(&newsk->sk_receive_queue);
1093                 skb_queue_head_init(&newsk->sk_write_queue);
1094 #ifdef CONFIG_NET_DMA
1095                 skb_queue_head_init(&newsk->sk_async_wait_queue);
1096 #endif
1097
1098                 rwlock_init(&newsk->sk_dst_lock);
1099                 rwlock_init(&newsk->sk_callback_lock);
1100                 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1101                                 af_callback_keys + newsk->sk_family,
1102                                 af_family_clock_key_strings[newsk->sk_family]);
1103
1104                 newsk->sk_dst_cache     = NULL;
1105                 newsk->sk_wmem_queued   = 0;
1106                 newsk->sk_forward_alloc = 0;
1107                 newsk->sk_send_head     = NULL;
1108                 newsk->sk_userlocks     = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1109
1110                 sock_reset_flag(newsk, SOCK_DONE);
1111                 skb_queue_head_init(&newsk->sk_error_queue);
1112
1113                 filter = newsk->sk_filter;
1114                 if (filter != NULL)
1115                         sk_filter_charge(newsk, filter);
1116
1117                 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1118                         /* It is still raw copy of parent, so invalidate
1119                          * destructor and make plain sk_free() */
1120                         newsk->sk_destruct = NULL;
1121                         sk_free(newsk);
1122                         newsk = NULL;
1123                         goto out;
1124                 }
1125
1126                 newsk->sk_err      = 0;
1127                 newsk->sk_priority = 0;
1128                 atomic_set(&newsk->sk_refcnt, 2);
1129
1130                 /*
1131                  * Increment the counter in the same struct proto as the master
1132                  * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1133                  * is the same as sk->sk_prot->socks, as this field was copied
1134                  * with memcpy).
1135                  *
1136                  * This _changes_ the previous behaviour, where
1137                  * tcp_create_openreq_child always was incrementing the
1138                  * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1139                  * to be taken into account in all callers. -acme
1140                  */
1141                 sk_refcnt_debug_inc(newsk);
1142                 sk_set_socket(newsk, NULL);
1143                 newsk->sk_sleep  = NULL;
1144
1145                 if (newsk->sk_prot->sockets_allocated)
1146                         percpu_counter_inc(newsk->sk_prot->sockets_allocated);
1147         }
1148 out:
1149         return newsk;
1150 }
1151 EXPORT_SYMBOL_GPL(sk_clone);
1152
1153 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1154 {
1155         __sk_dst_set(sk, dst);
1156         sk->sk_route_caps = dst->dev->features;
1157         if (sk->sk_route_caps & NETIF_F_GSO)
1158                 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1159         if (sk_can_gso(sk)) {
1160                 if (dst->header_len) {
1161                         sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1162                 } else {
1163                         sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1164                         sk->sk_gso_max_size = dst->dev->gso_max_size;
1165                 }
1166         }
1167 }
1168 EXPORT_SYMBOL_GPL(sk_setup_caps);
1169
1170 void __init sk_init(void)
1171 {
1172         if (num_physpages <= 4096) {
1173                 sysctl_wmem_max = 32767;
1174                 sysctl_rmem_max = 32767;
1175                 sysctl_wmem_default = 32767;
1176                 sysctl_rmem_default = 32767;
1177         } else if (num_physpages >= 131072) {
1178                 sysctl_wmem_max = 131071;
1179                 sysctl_rmem_max = 131071;
1180         }
1181 }
1182
1183 /*
1184  *      Simple resource managers for sockets.
1185  */
1186
1187
1188 /*
1189  * Write buffer destructor automatically called from kfree_skb.
1190  */
1191 void sock_wfree(struct sk_buff *skb)
1192 {
1193         struct sock *sk = skb->sk;
1194         int res;
1195
1196         /* In case it might be waiting for more memory. */
1197         res = atomic_sub_return(skb->truesize, &sk->sk_wmem_alloc);
1198         if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE))
1199                 sk->sk_write_space(sk);
1200         /*
1201          * if sk_wmem_alloc reached 0, we are last user and should
1202          * free this sock, as sk_free() call could not do it.
1203          */
1204         if (res == 0)
1205                 __sk_free(sk);
1206 }
1207 EXPORT_SYMBOL(sock_wfree);
1208
1209 /*
1210  * Read buffer destructor automatically called from kfree_skb.
1211  */
1212 void sock_rfree(struct sk_buff *skb)
1213 {
1214         struct sock *sk = skb->sk;
1215
1216         atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1217         sk_mem_uncharge(skb->sk, skb->truesize);
1218 }
1219 EXPORT_SYMBOL(sock_rfree);
1220
1221
1222 int sock_i_uid(struct sock *sk)
1223 {
1224         int uid;
1225
1226         read_lock(&sk->sk_callback_lock);
1227         uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1228         read_unlock(&sk->sk_callback_lock);
1229         return uid;
1230 }
1231 EXPORT_SYMBOL(sock_i_uid);
1232
1233 unsigned long sock_i_ino(struct sock *sk)
1234 {
1235         unsigned long ino;
1236
1237         read_lock(&sk->sk_callback_lock);
1238         ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1239         read_unlock(&sk->sk_callback_lock);
1240         return ino;
1241 }
1242 EXPORT_SYMBOL(sock_i_ino);
1243
1244 /*
1245  * Allocate a skb from the socket's send buffer.
1246  */
1247 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1248                              gfp_t priority)
1249 {
1250         if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1251                 struct sk_buff *skb = alloc_skb(size, priority);
1252                 if (skb) {
1253                         skb_set_owner_w(skb, sk);
1254                         return skb;
1255                 }
1256         }
1257         return NULL;
1258 }
1259 EXPORT_SYMBOL(sock_wmalloc);
1260
1261 /*
1262  * Allocate a skb from the socket's receive buffer.
1263  */
1264 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1265                              gfp_t priority)
1266 {
1267         if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1268                 struct sk_buff *skb = alloc_skb(size, priority);
1269                 if (skb) {
1270                         skb_set_owner_r(skb, sk);
1271                         return skb;
1272                 }
1273         }
1274         return NULL;
1275 }
1276
1277 /*
1278  * Allocate a memory block from the socket's option memory buffer.
1279  */
1280 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1281 {
1282         if ((unsigned)size <= sysctl_optmem_max &&
1283             atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1284                 void *mem;
1285                 /* First do the add, to avoid the race if kmalloc
1286                  * might sleep.
1287                  */
1288                 atomic_add(size, &sk->sk_omem_alloc);
1289                 mem = kmalloc(size, priority);
1290                 if (mem)
1291                         return mem;
1292                 atomic_sub(size, &sk->sk_omem_alloc);
1293         }
1294         return NULL;
1295 }
1296 EXPORT_SYMBOL(sock_kmalloc);
1297
1298 /*
1299  * Free an option memory block.
1300  */
1301 void sock_kfree_s(struct sock *sk, void *mem, int size)
1302 {
1303         kfree(mem);
1304         atomic_sub(size, &sk->sk_omem_alloc);
1305 }
1306 EXPORT_SYMBOL(sock_kfree_s);
1307
1308 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1309    I think, these locks should be removed for datagram sockets.
1310  */
1311 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1312 {
1313         DEFINE_WAIT(wait);
1314
1315         clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1316         for (;;) {
1317                 if (!timeo)
1318                         break;
1319                 if (signal_pending(current))
1320                         break;
1321                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1322                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1323                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1324                         break;
1325                 if (sk->sk_shutdown & SEND_SHUTDOWN)
1326                         break;
1327                 if (sk->sk_err)
1328                         break;
1329                 timeo = schedule_timeout(timeo);
1330         }
1331         finish_wait(sk->sk_sleep, &wait);
1332         return timeo;
1333 }
1334
1335
1336 /*
1337  *      Generic send/receive buffer handlers
1338  */
1339
1340 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1341                                      unsigned long data_len, int noblock,
1342                                      int *errcode)
1343 {
1344         struct sk_buff *skb;
1345         gfp_t gfp_mask;
1346         long timeo;
1347         int err;
1348
1349         gfp_mask = sk->sk_allocation;
1350         if (gfp_mask & __GFP_WAIT)
1351                 gfp_mask |= __GFP_REPEAT;
1352
1353         timeo = sock_sndtimeo(sk, noblock);
1354         while (1) {
1355                 err = sock_error(sk);
1356                 if (err != 0)
1357                         goto failure;
1358
1359                 err = -EPIPE;
1360                 if (sk->sk_shutdown & SEND_SHUTDOWN)
1361                         goto failure;
1362
1363                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1364                         skb = alloc_skb(header_len, gfp_mask);
1365                         if (skb) {
1366                                 int npages;
1367                                 int i;
1368
1369                                 /* No pages, we're done... */
1370                                 if (!data_len)
1371                                         break;
1372
1373                                 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1374                                 skb->truesize += data_len;
1375                                 skb_shinfo(skb)->nr_frags = npages;
1376                                 for (i = 0; i < npages; i++) {
1377                                         struct page *page;
1378                                         skb_frag_t *frag;
1379
1380                                         page = alloc_pages(sk->sk_allocation, 0);
1381                                         if (!page) {
1382                                                 err = -ENOBUFS;
1383                                                 skb_shinfo(skb)->nr_frags = i;
1384                                                 kfree_skb(skb);
1385                                                 goto failure;
1386                                         }
1387
1388                                         frag = &skb_shinfo(skb)->frags[i];
1389                                         frag->page = page;
1390                                         frag->page_offset = 0;
1391                                         frag->size = (data_len >= PAGE_SIZE ?
1392                                                       PAGE_SIZE :
1393                                                       data_len);
1394                                         data_len -= PAGE_SIZE;
1395                                 }
1396
1397                                 /* Full success... */
1398                                 break;
1399                         }
1400                         err = -ENOBUFS;
1401                         goto failure;
1402                 }
1403                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1404                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1405                 err = -EAGAIN;
1406                 if (!timeo)
1407                         goto failure;
1408                 if (signal_pending(current))
1409                         goto interrupted;
1410                 timeo = sock_wait_for_wmem(sk, timeo);
1411         }
1412
1413         skb_set_owner_w(skb, sk);
1414         return skb;
1415
1416 interrupted:
1417         err = sock_intr_errno(timeo);
1418 failure:
1419         *errcode = err;
1420         return NULL;
1421 }
1422 EXPORT_SYMBOL(sock_alloc_send_pskb);
1423
1424 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1425                                     int noblock, int *errcode)
1426 {
1427         return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1428 }
1429 EXPORT_SYMBOL(sock_alloc_send_skb);
1430
1431 static void __lock_sock(struct sock *sk)
1432 {
1433         DEFINE_WAIT(wait);
1434
1435         for (;;) {
1436                 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1437                                         TASK_UNINTERRUPTIBLE);
1438                 spin_unlock_bh(&sk->sk_lock.slock);
1439                 schedule();
1440                 spin_lock_bh(&sk->sk_lock.slock);
1441                 if (!sock_owned_by_user(sk))
1442                         break;
1443         }
1444         finish_wait(&sk->sk_lock.wq, &wait);
1445 }
1446
1447 static void __release_sock(struct sock *sk)
1448 {
1449         struct sk_buff *skb = sk->sk_backlog.head;
1450
1451         do {
1452                 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1453                 bh_unlock_sock(sk);
1454
1455                 do {
1456                         struct sk_buff *next = skb->next;
1457
1458                         skb->next = NULL;
1459                         sk_backlog_rcv(sk, skb);
1460
1461                         /*
1462                          * We are in process context here with softirqs
1463                          * disabled, use cond_resched_softirq() to preempt.
1464                          * This is safe to do because we've taken the backlog
1465                          * queue private:
1466                          */
1467                         cond_resched_softirq();
1468
1469                         skb = next;
1470                 } while (skb != NULL);
1471
1472                 bh_lock_sock(sk);
1473         } while ((skb = sk->sk_backlog.head) != NULL);
1474 }
1475
1476 /**
1477  * sk_wait_data - wait for data to arrive at sk_receive_queue
1478  * @sk:    sock to wait on
1479  * @timeo: for how long
1480  *
1481  * Now socket state including sk->sk_err is changed only under lock,
1482  * hence we may omit checks after joining wait queue.
1483  * We check receive queue before schedule() only as optimization;
1484  * it is very likely that release_sock() added new data.
1485  */
1486 int sk_wait_data(struct sock *sk, long *timeo)
1487 {
1488         int rc;
1489         DEFINE_WAIT(wait);
1490
1491         prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1492         set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1493         rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1494         clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1495         finish_wait(sk->sk_sleep, &wait);
1496         return rc;
1497 }
1498 EXPORT_SYMBOL(sk_wait_data);
1499
1500 /**
1501  *      __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1502  *      @sk: socket
1503  *      @size: memory size to allocate
1504  *      @kind: allocation type
1505  *
1506  *      If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1507  *      rmem allocation. This function assumes that protocols which have
1508  *      memory_pressure use sk_wmem_queued as write buffer accounting.
1509  */
1510 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1511 {
1512         struct proto *prot = sk->sk_prot;
1513         int amt = sk_mem_pages(size);
1514         int allocated;
1515
1516         sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1517         allocated = atomic_add_return(amt, prot->memory_allocated);
1518
1519         /* Under limit. */
1520         if (allocated <= prot->sysctl_mem[0]) {
1521                 if (prot->memory_pressure && *prot->memory_pressure)
1522                         *prot->memory_pressure = 0;
1523                 return 1;
1524         }
1525
1526         /* Under pressure. */
1527         if (allocated > prot->sysctl_mem[1])
1528                 if (prot->enter_memory_pressure)
1529                         prot->enter_memory_pressure(sk);
1530
1531         /* Over hard limit. */
1532         if (allocated > prot->sysctl_mem[2])
1533                 goto suppress_allocation;
1534
1535         /* guarantee minimum buffer size under pressure */
1536         if (kind == SK_MEM_RECV) {
1537                 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1538                         return 1;
1539         } else { /* SK_MEM_SEND */
1540                 if (sk->sk_type == SOCK_STREAM) {
1541                         if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1542                                 return 1;
1543                 } else if (atomic_read(&sk->sk_wmem_alloc) <
1544                            prot->sysctl_wmem[0])
1545                                 return 1;
1546         }
1547
1548         if (prot->memory_pressure) {
1549                 int alloc;
1550
1551                 if (!*prot->memory_pressure)
1552                         return 1;
1553                 alloc = percpu_counter_read_positive(prot->sockets_allocated);
1554                 if (prot->sysctl_mem[2] > alloc *
1555                     sk_mem_pages(sk->sk_wmem_queued +
1556                                  atomic_read(&sk->sk_rmem_alloc) +
1557                                  sk->sk_forward_alloc))
1558                         return 1;
1559         }
1560
1561 suppress_allocation:
1562
1563         if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1564                 sk_stream_moderate_sndbuf(sk);
1565
1566                 /* Fail only if socket is _under_ its sndbuf.
1567                  * In this case we cannot block, so that we have to fail.
1568                  */
1569                 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1570                         return 1;
1571         }
1572
1573         /* Alas. Undo changes. */
1574         sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1575         atomic_sub(amt, prot->memory_allocated);
1576         return 0;
1577 }
1578 EXPORT_SYMBOL(__sk_mem_schedule);
1579
1580 /**
1581  *      __sk_reclaim - reclaim memory_allocated
1582  *      @sk: socket
1583  */
1584 void __sk_mem_reclaim(struct sock *sk)
1585 {
1586         struct proto *prot = sk->sk_prot;
1587
1588         atomic_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
1589                    prot->memory_allocated);
1590         sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1591
1592         if (prot->memory_pressure && *prot->memory_pressure &&
1593             (atomic_read(prot->memory_allocated) < prot->sysctl_mem[0]))
1594                 *prot->memory_pressure = 0;
1595 }
1596 EXPORT_SYMBOL(__sk_mem_reclaim);
1597
1598
1599 /*
1600  * Set of default routines for initialising struct proto_ops when
1601  * the protocol does not support a particular function. In certain
1602  * cases where it makes no sense for a protocol to have a "do nothing"
1603  * function, some default processing is provided.
1604  */
1605
1606 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1607 {
1608         return -EOPNOTSUPP;
1609 }
1610 EXPORT_SYMBOL(sock_no_bind);
1611
1612 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1613                     int len, int flags)
1614 {
1615         return -EOPNOTSUPP;
1616 }
1617 EXPORT_SYMBOL(sock_no_connect);
1618
1619 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1620 {
1621         return -EOPNOTSUPP;
1622 }
1623 EXPORT_SYMBOL(sock_no_socketpair);
1624
1625 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1626 {
1627         return -EOPNOTSUPP;
1628 }
1629 EXPORT_SYMBOL(sock_no_accept);
1630
1631 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1632                     int *len, int peer)
1633 {
1634         return -EOPNOTSUPP;
1635 }
1636 EXPORT_SYMBOL(sock_no_getname);
1637
1638 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1639 {
1640         return 0;
1641 }
1642 EXPORT_SYMBOL(sock_no_poll);
1643
1644 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1645 {
1646         return -EOPNOTSUPP;
1647 }
1648 EXPORT_SYMBOL(sock_no_ioctl);
1649
1650 int sock_no_listen(struct socket *sock, int backlog)
1651 {
1652         return -EOPNOTSUPP;
1653 }
1654 EXPORT_SYMBOL(sock_no_listen);
1655
1656 int sock_no_shutdown(struct socket *sock, int how)
1657 {
1658         return -EOPNOTSUPP;
1659 }
1660 EXPORT_SYMBOL(sock_no_shutdown);
1661
1662 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1663                     char __user *optval, int optlen)
1664 {
1665         return -EOPNOTSUPP;
1666 }
1667 EXPORT_SYMBOL(sock_no_setsockopt);
1668
1669 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1670                     char __user *optval, int __user *optlen)
1671 {
1672         return -EOPNOTSUPP;
1673 }
1674 EXPORT_SYMBOL(sock_no_getsockopt);
1675
1676 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1677                     size_t len)
1678 {
1679         return -EOPNOTSUPP;
1680 }
1681 EXPORT_SYMBOL(sock_no_sendmsg);
1682
1683 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1684                     size_t len, int flags)
1685 {
1686         return -EOPNOTSUPP;
1687 }
1688 EXPORT_SYMBOL(sock_no_recvmsg);
1689
1690 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1691 {
1692         /* Mirror missing mmap method error code */
1693         return -ENODEV;
1694 }
1695 EXPORT_SYMBOL(sock_no_mmap);
1696
1697 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1698 {
1699         ssize_t res;
1700         struct msghdr msg = {.msg_flags = flags};
1701         struct kvec iov;
1702         char *kaddr = kmap(page);
1703         iov.iov_base = kaddr + offset;
1704         iov.iov_len = size;
1705         res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1706         kunmap(page);
1707         return res;
1708 }
1709 EXPORT_SYMBOL(sock_no_sendpage);
1710
1711 /*
1712  *      Default Socket Callbacks
1713  */
1714
1715 static void sock_def_wakeup(struct sock *sk)
1716 {
1717         read_lock(&sk->sk_callback_lock);
1718         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1719                 wake_up_interruptible_all(sk->sk_sleep);
1720         read_unlock(&sk->sk_callback_lock);
1721 }
1722
1723 static void sock_def_error_report(struct sock *sk)
1724 {
1725         read_lock(&sk->sk_callback_lock);
1726         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1727                 wake_up_interruptible_poll(sk->sk_sleep, POLLERR);
1728         sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
1729         read_unlock(&sk->sk_callback_lock);
1730 }
1731
1732 static void sock_def_readable(struct sock *sk, int len)
1733 {
1734         read_lock(&sk->sk_callback_lock);
1735         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1736                 wake_up_interruptible_sync_poll(sk->sk_sleep, POLLIN |
1737                                                 POLLRDNORM | POLLRDBAND);
1738         sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1739         read_unlock(&sk->sk_callback_lock);
1740 }
1741
1742 static void sock_def_write_space(struct sock *sk)
1743 {
1744         read_lock(&sk->sk_callback_lock);
1745
1746         /* Do not wake up a writer until he can make "significant"
1747          * progress.  --DaveM
1748          */
1749         if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1750                 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1751                         wake_up_interruptible_sync_poll(sk->sk_sleep, POLLOUT |
1752                                                 POLLWRNORM | POLLWRBAND);
1753
1754                 /* Should agree with poll, otherwise some programs break */
1755                 if (sock_writeable(sk))
1756                         sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1757         }
1758
1759         read_unlock(&sk->sk_callback_lock);
1760 }
1761
1762 static void sock_def_destruct(struct sock *sk)
1763 {
1764         kfree(sk->sk_protinfo);
1765 }
1766
1767 void sk_send_sigurg(struct sock *sk)
1768 {
1769         if (sk->sk_socket && sk->sk_socket->file)
1770                 if (send_sigurg(&sk->sk_socket->file->f_owner))
1771                         sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1772 }
1773 EXPORT_SYMBOL(sk_send_sigurg);
1774
1775 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1776                     unsigned long expires)
1777 {
1778         if (!mod_timer(timer, expires))
1779                 sock_hold(sk);
1780 }
1781 EXPORT_SYMBOL(sk_reset_timer);
1782
1783 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1784 {
1785         if (timer_pending(timer) && del_timer(timer))
1786                 __sock_put(sk);
1787 }
1788 EXPORT_SYMBOL(sk_stop_timer);
1789
1790 void sock_init_data(struct socket *sock, struct sock *sk)
1791 {
1792         skb_queue_head_init(&sk->sk_receive_queue);
1793         skb_queue_head_init(&sk->sk_write_queue);
1794         skb_queue_head_init(&sk->sk_error_queue);
1795 #ifdef CONFIG_NET_DMA
1796         skb_queue_head_init(&sk->sk_async_wait_queue);
1797 #endif
1798
1799         sk->sk_send_head        =       NULL;
1800
1801         init_timer(&sk->sk_timer);
1802
1803         sk->sk_allocation       =       GFP_KERNEL;
1804         sk->sk_rcvbuf           =       sysctl_rmem_default;
1805         sk->sk_sndbuf           =       sysctl_wmem_default;
1806         sk->sk_state            =       TCP_CLOSE;
1807         sk_set_socket(sk, sock);
1808
1809         sock_set_flag(sk, SOCK_ZAPPED);
1810
1811         if (sock) {
1812                 sk->sk_type     =       sock->type;
1813                 sk->sk_sleep    =       &sock->wait;
1814                 sock->sk        =       sk;
1815         } else
1816                 sk->sk_sleep    =       NULL;
1817
1818         rwlock_init(&sk->sk_dst_lock);
1819         rwlock_init(&sk->sk_callback_lock);
1820         lockdep_set_class_and_name(&sk->sk_callback_lock,
1821                         af_callback_keys + sk->sk_family,
1822                         af_family_clock_key_strings[sk->sk_family]);
1823
1824         sk->sk_state_change     =       sock_def_wakeup;
1825         sk->sk_data_ready       =       sock_def_readable;
1826         sk->sk_write_space      =       sock_def_write_space;
1827         sk->sk_error_report     =       sock_def_error_report;
1828         sk->sk_destruct         =       sock_def_destruct;
1829
1830         sk->sk_sndmsg_page      =       NULL;
1831         sk->sk_sndmsg_off       =       0;
1832
1833         sk->sk_peercred.pid     =       0;
1834         sk->sk_peercred.uid     =       -1;
1835         sk->sk_peercred.gid     =       -1;
1836         sk->sk_write_pending    =       0;
1837         sk->sk_rcvlowat         =       1;
1838         sk->sk_rcvtimeo         =       MAX_SCHEDULE_TIMEOUT;
1839         sk->sk_sndtimeo         =       MAX_SCHEDULE_TIMEOUT;
1840
1841         sk->sk_stamp = ktime_set(-1L, 0);
1842
1843         atomic_set(&sk->sk_refcnt, 1);
1844         atomic_set(&sk->sk_wmem_alloc, 1);
1845         atomic_set(&sk->sk_drops, 0);
1846 }
1847 EXPORT_SYMBOL(sock_init_data);
1848
1849 void lock_sock_nested(struct sock *sk, int subclass)
1850 {
1851         might_sleep();
1852         spin_lock_bh(&sk->sk_lock.slock);
1853         if (sk->sk_lock.owned)
1854                 __lock_sock(sk);
1855         sk->sk_lock.owned = 1;
1856         spin_unlock(&sk->sk_lock.slock);
1857         /*
1858          * The sk_lock has mutex_lock() semantics here:
1859          */
1860         mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
1861         local_bh_enable();
1862 }
1863 EXPORT_SYMBOL(lock_sock_nested);
1864
1865 void release_sock(struct sock *sk)
1866 {
1867         /*
1868          * The sk_lock has mutex_unlock() semantics:
1869          */
1870         mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1871
1872         spin_lock_bh(&sk->sk_lock.slock);
1873         if (sk->sk_backlog.tail)
1874                 __release_sock(sk);
1875         sk->sk_lock.owned = 0;
1876         if (waitqueue_active(&sk->sk_lock.wq))
1877                 wake_up(&sk->sk_lock.wq);
1878         spin_unlock_bh(&sk->sk_lock.slock);
1879 }
1880 EXPORT_SYMBOL(release_sock);
1881
1882 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1883 {
1884         struct timeval tv;
1885         if (!sock_flag(sk, SOCK_TIMESTAMP))
1886                 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
1887         tv = ktime_to_timeval(sk->sk_stamp);
1888         if (tv.tv_sec == -1)
1889                 return -ENOENT;
1890         if (tv.tv_sec == 0) {
1891                 sk->sk_stamp = ktime_get_real();
1892                 tv = ktime_to_timeval(sk->sk_stamp);
1893         }
1894         return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
1895 }
1896 EXPORT_SYMBOL(sock_get_timestamp);
1897
1898 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
1899 {
1900         struct timespec ts;
1901         if (!sock_flag(sk, SOCK_TIMESTAMP))
1902                 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
1903         ts = ktime_to_timespec(sk->sk_stamp);
1904         if (ts.tv_sec == -1)
1905                 return -ENOENT;
1906         if (ts.tv_sec == 0) {
1907                 sk->sk_stamp = ktime_get_real();
1908                 ts = ktime_to_timespec(sk->sk_stamp);
1909         }
1910         return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
1911 }
1912 EXPORT_SYMBOL(sock_get_timestampns);
1913
1914 void sock_enable_timestamp(struct sock *sk, int flag)
1915 {
1916         if (!sock_flag(sk, flag)) {
1917                 sock_set_flag(sk, flag);
1918                 /*
1919                  * we just set one of the two flags which require net
1920                  * time stamping, but time stamping might have been on
1921                  * already because of the other one
1922                  */
1923                 if (!sock_flag(sk,
1924                                 flag == SOCK_TIMESTAMP ?
1925                                 SOCK_TIMESTAMPING_RX_SOFTWARE :
1926                                 SOCK_TIMESTAMP))
1927                         net_enable_timestamp();
1928         }
1929 }
1930
1931 /*
1932  *      Get a socket option on an socket.
1933  *
1934  *      FIX: POSIX 1003.1g is very ambiguous here. It states that
1935  *      asynchronous errors should be reported by getsockopt. We assume
1936  *      this means if you specify SO_ERROR (otherwise whats the point of it).
1937  */
1938 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1939                            char __user *optval, int __user *optlen)
1940 {
1941         struct sock *sk = sock->sk;
1942
1943         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1944 }
1945 EXPORT_SYMBOL(sock_common_getsockopt);
1946
1947 #ifdef CONFIG_COMPAT
1948 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
1949                                   char __user *optval, int __user *optlen)
1950 {
1951         struct sock *sk = sock->sk;
1952
1953         if (sk->sk_prot->compat_getsockopt != NULL)
1954                 return sk->sk_prot->compat_getsockopt(sk, level, optname,
1955                                                       optval, optlen);
1956         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1957 }
1958 EXPORT_SYMBOL(compat_sock_common_getsockopt);
1959 #endif
1960
1961 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1962                         struct msghdr *msg, size_t size, int flags)
1963 {
1964         struct sock *sk = sock->sk;
1965         int addr_len = 0;
1966         int err;
1967
1968         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
1969                                    flags & ~MSG_DONTWAIT, &addr_len);
1970         if (err >= 0)
1971                 msg->msg_namelen = addr_len;
1972         return err;
1973 }
1974 EXPORT_SYMBOL(sock_common_recvmsg);
1975
1976 /*
1977  *      Set socket options on an inet socket.
1978  */
1979 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1980                            char __user *optval, int optlen)
1981 {
1982         struct sock *sk = sock->sk;
1983
1984         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1985 }
1986 EXPORT_SYMBOL(sock_common_setsockopt);
1987
1988 #ifdef CONFIG_COMPAT
1989 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
1990                                   char __user *optval, int optlen)
1991 {
1992         struct sock *sk = sock->sk;
1993
1994         if (sk->sk_prot->compat_setsockopt != NULL)
1995                 return sk->sk_prot->compat_setsockopt(sk, level, optname,
1996                                                       optval, optlen);
1997         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1998 }
1999 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2000 #endif
2001
2002 void sk_common_release(struct sock *sk)
2003 {
2004         if (sk->sk_prot->destroy)
2005                 sk->sk_prot->destroy(sk);
2006
2007         /*
2008          * Observation: when sock_common_release is called, processes have
2009          * no access to socket. But net still has.
2010          * Step one, detach it from networking:
2011          *
2012          * A. Remove from hash tables.
2013          */
2014
2015         sk->sk_prot->unhash(sk);
2016
2017         /*
2018          * In this point socket cannot receive new packets, but it is possible
2019          * that some packets are in flight because some CPU runs receiver and
2020          * did hash table lookup before we unhashed socket. They will achieve
2021          * receive queue and will be purged by socket destructor.
2022          *
2023          * Also we still have packets pending on receive queue and probably,
2024          * our own packets waiting in device queues. sock_destroy will drain
2025          * receive queue, but transmitted packets will delay socket destruction
2026          * until the last reference will be released.
2027          */
2028
2029         sock_orphan(sk);
2030
2031         xfrm_sk_free_policy(sk);
2032
2033         sk_refcnt_debug_release(sk);
2034         sock_put(sk);
2035 }
2036 EXPORT_SYMBOL(sk_common_release);
2037
2038 static DEFINE_RWLOCK(proto_list_lock);
2039 static LIST_HEAD(proto_list);
2040
2041 #ifdef CONFIG_PROC_FS
2042 #define PROTO_INUSE_NR  64      /* should be enough for the first time */
2043 struct prot_inuse {
2044         int val[PROTO_INUSE_NR];
2045 };
2046
2047 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2048
2049 #ifdef CONFIG_NET_NS
2050 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2051 {
2052         int cpu = smp_processor_id();
2053         per_cpu_ptr(net->core.inuse, cpu)->val[prot->inuse_idx] += val;
2054 }
2055 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2056
2057 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2058 {
2059         int cpu, idx = prot->inuse_idx;
2060         int res = 0;
2061
2062         for_each_possible_cpu(cpu)
2063                 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2064
2065         return res >= 0 ? res : 0;
2066 }
2067 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2068
2069 static int sock_inuse_init_net(struct net *net)
2070 {
2071         net->core.inuse = alloc_percpu(struct prot_inuse);
2072         return net->core.inuse ? 0 : -ENOMEM;
2073 }
2074
2075 static void sock_inuse_exit_net(struct net *net)
2076 {
2077         free_percpu(net->core.inuse);
2078 }
2079
2080 static struct pernet_operations net_inuse_ops = {
2081         .init = sock_inuse_init_net,
2082         .exit = sock_inuse_exit_net,
2083 };
2084
2085 static __init int net_inuse_init(void)
2086 {
2087         if (register_pernet_subsys(&net_inuse_ops))
2088                 panic("Cannot initialize net inuse counters");
2089
2090         return 0;
2091 }
2092
2093 core_initcall(net_inuse_init);
2094 #else
2095 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2096
2097 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2098 {
2099         __get_cpu_var(prot_inuse).val[prot->inuse_idx] += val;
2100 }
2101 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2102
2103 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2104 {
2105         int cpu, idx = prot->inuse_idx;
2106         int res = 0;
2107
2108         for_each_possible_cpu(cpu)
2109                 res += per_cpu(prot_inuse, cpu).val[idx];
2110
2111         return res >= 0 ? res : 0;
2112 }
2113 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2114 #endif
2115
2116 static void assign_proto_idx(struct proto *prot)
2117 {
2118         prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2119
2120         if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2121                 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2122                 return;
2123         }
2124
2125         set_bit(prot->inuse_idx, proto_inuse_idx);
2126 }
2127
2128 static void release_proto_idx(struct proto *prot)
2129 {
2130         if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2131                 clear_bit(prot->inuse_idx, proto_inuse_idx);
2132 }
2133 #else
2134 static inline void assign_proto_idx(struct proto *prot)
2135 {
2136 }
2137
2138 static inline void release_proto_idx(struct proto *prot)
2139 {
2140 }
2141 #endif
2142
2143 int proto_register(struct proto *prot, int alloc_slab)
2144 {
2145         if (alloc_slab) {
2146                 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2147                                         SLAB_HWCACHE_ALIGN | prot->slab_flags,
2148                                         NULL);
2149
2150                 if (prot->slab == NULL) {
2151                         printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2152                                prot->name);
2153                         goto out;
2154                 }
2155
2156                 if (prot->rsk_prot != NULL) {
2157                         static const char mask[] = "request_sock_%s";
2158
2159                         prot->rsk_prot->slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
2160                         if (prot->rsk_prot->slab_name == NULL)
2161                                 goto out_free_sock_slab;
2162
2163                         sprintf(prot->rsk_prot->slab_name, mask, prot->name);
2164                         prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2165                                                                  prot->rsk_prot->obj_size, 0,
2166                                                                  SLAB_HWCACHE_ALIGN, NULL);
2167
2168                         if (prot->rsk_prot->slab == NULL) {
2169                                 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2170                                        prot->name);
2171                                 goto out_free_request_sock_slab_name;
2172                         }
2173                 }
2174
2175                 if (prot->twsk_prot != NULL) {
2176                         static const char mask[] = "tw_sock_%s";
2177
2178                         prot->twsk_prot->twsk_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
2179
2180                         if (prot->twsk_prot->twsk_slab_name == NULL)
2181                                 goto out_free_request_sock_slab;
2182
2183                         sprintf(prot->twsk_prot->twsk_slab_name, mask, prot->name);
2184                         prot->twsk_prot->twsk_slab =
2185                                 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2186                                                   prot->twsk_prot->twsk_obj_size,
2187                                                   0,
2188                                                   SLAB_HWCACHE_ALIGN |
2189                                                         prot->slab_flags,
2190                                                   NULL);
2191                         if (prot->twsk_prot->twsk_slab == NULL)
2192                                 goto out_free_timewait_sock_slab_name;
2193                 }
2194         }
2195
2196         write_lock(&proto_list_lock);
2197         list_add(&prot->node, &proto_list);
2198         assign_proto_idx(prot);
2199         write_unlock(&proto_list_lock);
2200         return 0;
2201
2202 out_free_timewait_sock_slab_name:
2203         kfree(prot->twsk_prot->twsk_slab_name);
2204 out_free_request_sock_slab:
2205         if (prot->rsk_prot && prot->rsk_prot->slab) {
2206                 kmem_cache_destroy(prot->rsk_prot->slab);
2207                 prot->rsk_prot->slab = NULL;
2208         }
2209 out_free_request_sock_slab_name:
2210         kfree(prot->rsk_prot->slab_name);
2211 out_free_sock_slab:
2212         kmem_cache_destroy(prot->slab);
2213         prot->slab = NULL;
2214 out:
2215         return -ENOBUFS;
2216 }
2217 EXPORT_SYMBOL(proto_register);
2218
2219 void proto_unregister(struct proto *prot)
2220 {
2221         write_lock(&proto_list_lock);
2222         release_proto_idx(prot);
2223         list_del(&prot->node);
2224         write_unlock(&proto_list_lock);
2225
2226         if (prot->slab != NULL) {
2227                 kmem_cache_destroy(prot->slab);
2228                 prot->slab = NULL;
2229         }
2230
2231         if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2232                 kmem_cache_destroy(prot->rsk_prot->slab);
2233                 kfree(prot->rsk_prot->slab_name);
2234                 prot->rsk_prot->slab = NULL;
2235         }
2236
2237         if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2238                 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2239                 kfree(prot->twsk_prot->twsk_slab_name);
2240                 prot->twsk_prot->twsk_slab = NULL;
2241         }
2242 }
2243 EXPORT_SYMBOL(proto_unregister);
2244
2245 #ifdef CONFIG_PROC_FS
2246 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2247         __acquires(proto_list_lock)
2248 {
2249         read_lock(&proto_list_lock);
2250         return seq_list_start_head(&proto_list, *pos);
2251 }
2252
2253 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2254 {
2255         return seq_list_next(v, &proto_list, pos);
2256 }
2257
2258 static void proto_seq_stop(struct seq_file *seq, void *v)
2259         __releases(proto_list_lock)
2260 {
2261         read_unlock(&proto_list_lock);
2262 }
2263
2264 static char proto_method_implemented(const void *method)
2265 {
2266         return method == NULL ? 'n' : 'y';
2267 }
2268
2269 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2270 {
2271         seq_printf(seq, "%-9s %4u %6d  %6d   %-3s %6u   %-3s  %-10s "
2272                         "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2273                    proto->name,
2274                    proto->obj_size,
2275                    sock_prot_inuse_get(seq_file_net(seq), proto),
2276                    proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
2277                    proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2278                    proto->max_header,
2279                    proto->slab == NULL ? "no" : "yes",
2280                    module_name(proto->owner),
2281                    proto_method_implemented(proto->close),
2282                    proto_method_implemented(proto->connect),
2283                    proto_method_implemented(proto->disconnect),
2284                    proto_method_implemented(proto->accept),
2285                    proto_method_implemented(proto->ioctl),
2286                    proto_method_implemented(proto->init),
2287                    proto_method_implemented(proto->destroy),
2288                    proto_method_implemented(proto->shutdown),
2289                    proto_method_implemented(proto->setsockopt),
2290                    proto_method_implemented(proto->getsockopt),
2291                    proto_method_implemented(proto->sendmsg),
2292                    proto_method_implemented(proto->recvmsg),
2293                    proto_method_implemented(proto->sendpage),
2294                    proto_method_implemented(proto->bind),
2295                    proto_method_implemented(proto->backlog_rcv),
2296                    proto_method_implemented(proto->hash),
2297                    proto_method_implemented(proto->unhash),
2298                    proto_method_implemented(proto->get_port),
2299                    proto_method_implemented(proto->enter_memory_pressure));
2300 }
2301
2302 static int proto_seq_show(struct seq_file *seq, void *v)
2303 {
2304         if (v == &proto_list)
2305                 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2306                            "protocol",
2307                            "size",
2308                            "sockets",
2309                            "memory",
2310                            "press",
2311                            "maxhdr",
2312                            "slab",
2313                            "module",
2314                            "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2315         else
2316                 proto_seq_printf(seq, list_entry(v, struct proto, node));
2317         return 0;
2318 }
2319
2320 static const struct seq_operations proto_seq_ops = {
2321         .start  = proto_seq_start,
2322         .next   = proto_seq_next,
2323         .stop   = proto_seq_stop,
2324         .show   = proto_seq_show,
2325 };
2326
2327 static int proto_seq_open(struct inode *inode, struct file *file)
2328 {
2329         return seq_open_net(inode, file, &proto_seq_ops,
2330                             sizeof(struct seq_net_private));
2331 }
2332
2333 static const struct file_operations proto_seq_fops = {
2334         .owner          = THIS_MODULE,
2335         .open           = proto_seq_open,
2336         .read           = seq_read,
2337         .llseek         = seq_lseek,
2338         .release        = seq_release_net,
2339 };
2340
2341 static __net_init int proto_init_net(struct net *net)
2342 {
2343         if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2344                 return -ENOMEM;
2345
2346         return 0;
2347 }
2348
2349 static __net_exit void proto_exit_net(struct net *net)
2350 {
2351         proc_net_remove(net, "protocols");
2352 }
2353
2354
2355 static __net_initdata struct pernet_operations proto_net_ops = {
2356         .init = proto_init_net,
2357         .exit = proto_exit_net,
2358 };
2359
2360 static int __init proto_init(void)
2361 {
2362         return register_pernet_subsys(&proto_net_ops);
2363 }
2364
2365 subsys_initcall(proto_init);
2366
2367 #endif /* PROC_FS */