net/core/sock.c

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