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, <[email protected]>
  12  *              Florian La Roche, <[email protected]>
  13  *              Alan Cox, <[email protected]>
  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;
 278         int skb_len;
 279         unsigned long flags;
 280         struct sk_buff_head *list = &sk->sk_receive_queue;
 281
 282         /* Cast sk->rcvbuf to unsigned... It's pointless, but reduces
 283            number of warnings when compiling with -W --ANK
 284          */
 285         if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
 286             (unsigned)sk->sk_rcvbuf) {
 287                 atomic_inc(&sk->sk_drops);
 288                 return -ENOMEM;
 289         }
 290
 291         err = sk_filter(sk, skb);
 292         if (err)
 293                 return err;
 294
 295         if (!sk_rmem_schedule(sk, skb->truesize)) {
 296                 atomic_inc(&sk->sk_drops);
 297                 return -ENOBUFS;
 298         }
 299
 300         skb->dev = NULL;
 301         skb_set_owner_r(skb, sk);
 302
 303         /* Cache the SKB length before we tack it onto the receive
 304          * queue.  Once it is added it no longer belongs to us and
 305          * may be freed by other threads of control pulling packets
 306          * from the queue.
 307          */
 308         skb_len = skb->len;
 309
 310         spin_lock_irqsave(&list->lock, flags);
 311         skb->dropcount = atomic_read(&sk->sk_drops);
 312         __skb_queue_tail(list, skb);
 313         spin_unlock_irqrestore(&list->lock, flags);
 314
 315         if (!sock_flag(sk, SOCK_DEAD))
 316                 sk->sk_data_ready(sk, skb_len);
 317         return 0;
 318 }
 319 EXPORT_SYMBOL(sock_queue_rcv_skb);
 320
 321 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
 322 {
 323         int rc = NET_RX_SUCCESS;
 324
 325         if (sk_filter(sk, skb))
 326                 goto discard_and_relse;
 327
 328         skb->dev = NULL;
 329
 330         if (nested)
 331                 bh_lock_sock_nested(sk);
 332         else
 333                 bh_lock_sock(sk);
 334         if (!sock_owned_by_user(sk)) {
 335                 /*
 336                  * trylock + unlock semantics:
 337                  */
 338                 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
 339
 340                 rc = sk_backlog_rcv(sk, skb);
 341
 342                 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
 343         } else
 344                 sk_add_backlog(sk, skb);
 345         bh_unlock_sock(sk);
 346 out:
 347         sock_put(sk);
 348         return rc;
 349 discard_and_relse:
 350         kfree_skb(skb);
 351         goto out;
 352 }
 353 EXPORT_SYMBOL(sk_receive_skb);
 354
 355 void sk_reset_txq(struct sock *sk)
 356 {
 357         sk_tx_queue_clear(sk);
 358 }
 359 EXPORT_SYMBOL(sk_reset_txq);
 360
 361 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
 362 {
 363         struct dst_entry *dst = sk->sk_dst_cache;
 364
 365         if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
 366                 sk_tx_queue_clear(sk);
 367                 sk->sk_dst_cache = NULL;
 368                 dst_release(dst);
 369                 return NULL;
 370         }
 371
 372         return dst;
 373 }
 374 EXPORT_SYMBOL(__sk_dst_check);
 375
 376 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
 377 {
 378         struct dst_entry *dst = sk_dst_get(sk);
 379
 380         if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
 381                 sk_dst_reset(sk);
 382                 dst_release(dst);
 383                 return NULL;
 384         }
 385
 386         return dst;
 387 }
 388 EXPORT_SYMBOL(sk_dst_check);
 389
 390 static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
 391 {
 392         int ret = -ENOPROTOOPT;
 393 #ifdef CONFIG_NETDEVICES
 394         struct net *net = sock_net(sk);
 395         char devname[IFNAMSIZ];
 396         int index;
 397
 398         /* Sorry... */
 399         ret = -EPERM;
 400         if (!capable(CAP_NET_RAW))
 401                 goto out;
 402
 403         ret = -EINVAL;
 404         if (optlen < 0)
 405                 goto out;
 406
 407         /* Bind this socket to a particular device like "eth0",
 408          * as specified in the passed interface name. If the
 409          * name is "" or the option length is zero the socket
 410          * is not bound.
 411          */
 412         if (optlen > IFNAMSIZ - 1)
 413                 optlen = IFNAMSIZ - 1;
 414         memset(devname, 0, sizeof(devname));
 415
 416         ret = -EFAULT;
 417         if (copy_from_user(devname, optval, optlen))
 418                 goto out;
 419
 420         index = 0;
 421         if (devname[0] != '\0') {
 422                 struct net_device *dev;
 423
 424                 rcu_read_lock();
 425                 dev = dev_get_by_name_rcu(net, devname);
 426                 if (dev)
 427                         index = dev->ifindex;
 428                 rcu_read_unlock();
 429                 ret = -ENODEV;
 430                 if (!dev)
 431                         goto out;
 432         }
 433
 434         lock_sock(sk);
 435         sk->sk_bound_dev_if = index;
 436         sk_dst_reset(sk);
 437         release_sock(sk);
 438
 439         ret = 0;
 440
 441 out:
 442 #endif
 443
 444         return ret;
 445 }
 446
 447 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
 448 {
 449         if (valbool)
 450                 sock_set_flag(sk, bit);
 451         else
 452                 sock_reset_flag(sk, bit);
 453 }
 454
 455 /*
 456  *      This is meant for all protocols to use and covers goings on
 457  *      at the socket level. Everything here is generic.
 458  */
 459
 460 int sock_setsockopt(struct socket *sock, int level, int optname,
 461                     char __user *optval, unsigned int optlen)
 462 {
 463         struct sock *sk = sock->sk;
 464         int val;
 465         int valbool;
 466         struct linger ling;
 467         int ret = 0;
 468
 469         /*
 470          *      Options without arguments
 471          */
 472
 473         if (optname == SO_BINDTODEVICE)
 474                 return sock_bindtodevice(sk, optval, optlen);
 475
 476         if (optlen < sizeof(int))
 477                 return -EINVAL;
 478
 479         if (get_user(val, (int __user *)optval))
 480                 return -EFAULT;
 481
 482         valbool = val ? 1 : 0;
 483
 484         lock_sock(sk);
 485
 486         switch (optname) {
 487         case SO_DEBUG:
 488                 if (val && !capable(CAP_NET_ADMIN))
 489                         ret = -EACCES;
 490                 else
 491                         sock_valbool_flag(sk, SOCK_DBG, valbool);
 492                 break;
 493         case SO_REUSEADDR:
 494                 sk->sk_reuse = valbool;
 495                 break;
 496         case SO_TYPE:
 497         case SO_PROTOCOL:
 498         case SO_DOMAIN:
 499         case SO_ERROR:
 500                 ret = -ENOPROTOOPT;
 501                 break;
 502         case SO_DONTROUTE:
 503                 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
 504                 break;
 505         case SO_BROADCAST:
 506                 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
 507                 break;
 508         case SO_SNDBUF:
 509                 /* Don't error on this BSD doesn't and if you think
 510                    about it this is right. Otherwise apps have to
 511                    play 'guess the biggest size' games. RCVBUF/SNDBUF
 512                    are treated in BSD as hints */
 513
 514                 if (val > sysctl_wmem_max)
 515                         val = sysctl_wmem_max;
 516 set_sndbuf:
 517                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
 518                 if ((val * 2) < SOCK_MIN_SNDBUF)
 519                         sk->sk_sndbuf = SOCK_MIN_SNDBUF;
 520                 else
 521                         sk->sk_sndbuf = val * 2;
 522
 523                 /*
 524                  *      Wake up sending tasks if we
 525                  *      upped the value.
 526                  */
 527                 sk->sk_write_space(sk);
 528                 break;
 529
 530         case SO_SNDBUFFORCE:
 531                 if (!capable(CAP_NET_ADMIN)) {
 532                         ret = -EPERM;
 533                         break;
 534                 }
 535                 goto set_sndbuf;
 536
 537         case SO_RCVBUF:
 538                 /* Don't error on this BSD doesn't and if you think
 539                    about it this is right. Otherwise apps have to
 540                    play 'guess the biggest size' games. RCVBUF/SNDBUF
 541                    are treated in BSD as hints */
 542
 543                 if (val > sysctl_rmem_max)
 544                         val = sysctl_rmem_max;
 545 set_rcvbuf:
 546                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
 547                 /*
 548                  * We double it on the way in to account for
 549                  * "struct sk_buff" etc. overhead.   Applications
 550                  * assume that the SO_RCVBUF setting they make will
 551                  * allow that much actual data to be received on that
 552                  * socket.
 553                  *
 554                  * Applications are unaware that "struct sk_buff" and
 555                  * other overheads allocate from the receive buffer
 556                  * during socket buffer allocation.
 557                  *
 558                  * And after considering the possible alternatives,
 559                  * returning the value we actually used in getsockopt
 560                  * is the most desirable behavior.
 561                  */
 562                 if ((val * 2) < SOCK_MIN_RCVBUF)
 563                         sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
 564                 else
 565                         sk->sk_rcvbuf = val * 2;
 566                 break;
 567
 568         case SO_RCVBUFFORCE:
 569                 if (!capable(CAP_NET_ADMIN)) {
 570                         ret = -EPERM;
 571                         break;
 572                 }
 573                 goto set_rcvbuf;
 574
 575         case SO_KEEPALIVE:
 576 #ifdef CONFIG_INET
 577                 if (sk->sk_protocol == IPPROTO_TCP)
 578                         tcp_set_keepalive(sk, valbool);
 579 #endif
 580                 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
 581                 break;
 582
 583         case SO_OOBINLINE:
 584                 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
 585                 break;
 586
 587         case SO_NO_CHECK:
 588                 sk->sk_no_check = valbool;
 589                 break;
 590
 591         case SO_PRIORITY:
 592                 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
 593                         sk->sk_priority = val;
 594                 else
 595                         ret = -EPERM;
 596                 break;
 597
 598         case SO_LINGER:
 599                 if (optlen < sizeof(ling)) {
 600                         ret = -EINVAL;  /* 1003.1g */
 601                         break;
 602                 }
 603                 if (copy_from_user(&ling, optval, sizeof(ling))) {
 604                         ret = -EFAULT;
 605                         break;
 606                 }
 607                 if (!ling.l_onoff)
 608                         sock_reset_flag(sk, SOCK_LINGER);
 609                 else {
 610 #if (BITS_PER_LONG == 32)
 611                         if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
 612                                 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
 613                         else
 614 #endif
 615                                 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
 616                         sock_set_flag(sk, SOCK_LINGER);
 617                 }
 618                 break;
 619
 620         case SO_BSDCOMPAT:
 621                 sock_warn_obsolete_bsdism("setsockopt");
 622                 break;
 623
 624         case SO_PASSCRED:
 625                 if (valbool)
 626                         set_bit(SOCK_PASSCRED, &sock->flags);
 627                 else
 628                         clear_bit(SOCK_PASSCRED, &sock->flags);
 629                 break;
 630
 631         case SO_TIMESTAMP:
 632         case SO_TIMESTAMPNS:
 633                 if (valbool)  {
 634                         if (optname == SO_TIMESTAMP)
 635                                 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
 636                         else
 637                                 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
 638                         sock_set_flag(sk, SOCK_RCVTSTAMP);
 639                         sock_enable_timestamp(sk, SOCK_TIMESTAMP);
 640                 } else {
 641                         sock_reset_flag(sk, SOCK_RCVTSTAMP);
 642                         sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
 643                 }
 644                 break;
 645
 646         case SO_TIMESTAMPING:
 647                 if (val & ~SOF_TIMESTAMPING_MASK) {
 648                         ret = -EINVAL;
 649                         break;
 650                 }
 651                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
 652                                   val & SOF_TIMESTAMPING_TX_HARDWARE);
 653                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
 654                                   val & SOF_TIMESTAMPING_TX_SOFTWARE);
 655                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
 656                                   val & SOF_TIMESTAMPING_RX_HARDWARE);
 657                 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
 658                         sock_enable_timestamp(sk,
 659                                               SOCK_TIMESTAMPING_RX_SOFTWARE);
 660                 else
 661                         sock_disable_timestamp(sk,
 662                                                SOCK_TIMESTAMPING_RX_SOFTWARE);
 663                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
 664                                   val & SOF_TIMESTAMPING_SOFTWARE);
 665                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
 666                                   val & SOF_TIMESTAMPING_SYS_HARDWARE);
 667                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
 668                                   val & SOF_TIMESTAMPING_RAW_HARDWARE);
 669                 break;
 670
 671         case SO_RCVLOWAT:
 672                 if (val < 0)
 673                         val = INT_MAX;
 674                 sk->sk_rcvlowat = val ? : 1;
 675                 break;
 676
 677         case SO_RCVTIMEO:
 678                 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
 679                 break;
 680
 681         case SO_SNDTIMEO:
 682                 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
 683                 break;
 684
 685         case SO_ATTACH_FILTER:
 686                 ret = -EINVAL;
 687                 if (optlen == sizeof(struct sock_fprog)) {
 688                         struct sock_fprog fprog;
 689
 690                         ret = -EFAULT;
 691                         if (copy_from_user(&fprog, optval, sizeof(fprog)))
 692                                 break;
 693
 694                         ret = sk_attach_filter(&fprog, sk);
 695                 }
 696                 break;
 697
 698         case SO_DETACH_FILTER:
 699                 ret = sk_detach_filter(sk);
 700                 break;
 701
 702         case SO_PASSSEC:
 703                 if (valbool)
 704                         set_bit(SOCK_PASSSEC, &sock->flags);
 705                 else
 706                         clear_bit(SOCK_PASSSEC, &sock->flags);
 707                 break;
 708         case SO_MARK:
 709                 if (!capable(CAP_NET_ADMIN))
 710                         ret = -EPERM;
 711                 else
 712                         sk->sk_mark = val;
 713                 break;
 714
 715                 /* We implement the SO_SNDLOWAT etc to
 716                    not be settable (1003.1g 5.3) */
 717         case SO_RXQ_OVFL:
 718                 if (valbool)
 719                         sock_set_flag(sk, SOCK_RXQ_OVFL);
 720                 else
 721                         sock_reset_flag(sk, SOCK_RXQ_OVFL);
 722                 break;
 723         default:
 724                 ret = -ENOPROTOOPT;
 725                 break;
 726         }
 727         release_sock(sk);
 728         return ret;
 729 }
 730 EXPORT_SYMBOL(sock_setsockopt);
 731
 732
 733 int sock_getsockopt(struct socket *sock, int level, int optname,
 734                     char __user *optval, int __user *optlen)
 735 {
 736         struct sock *sk = sock->sk;
 737
 738         union {
 739                 int val;
 740                 struct linger ling;
 741                 struct timeval tm;
 742         } v;
 743
 744         unsigned int lv = sizeof(int);
 745         int len;
 746
 747         if (get_user(len, optlen))
 748                 return -EFAULT;
 749         if (len < 0)
 750                 return -EINVAL;
 751
 752         memset(&v, 0, sizeof(v));
 753
 754         switch (optname) {
 755         case SO_DEBUG:
 756                 v.val = sock_flag(sk, SOCK_DBG);
 757                 break;
 758
 759         case SO_DONTROUTE:
 760                 v.val = sock_flag(sk, SOCK_LOCALROUTE);
 761                 break;
 762
 763         case SO_BROADCAST:
 764                 v.val = !!sock_flag(sk, SOCK_BROADCAST);
 765                 break;
 766
 767         case SO_SNDBUF:
 768                 v.val = sk->sk_sndbuf;
 769                 break;
 770
 771         case SO_RCVBUF:
 772                 v.val = sk->sk_rcvbuf;
 773                 break;
 774
 775         case SO_REUSEADDR:
 776                 v.val = sk->sk_reuse;
 777                 break;
 778
 779         case SO_KEEPALIVE:
 780                 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
 781                 break;
 782
 783         case SO_TYPE:
 784                 v.val = sk->sk_type;
 785                 break;
 786
 787         case SO_PROTOCOL:
 788                 v.val = sk->sk_protocol;
 789                 break;
 790
 791         case SO_DOMAIN:
 792                 v.val = sk->sk_family;
 793                 break;
 794
 795         case SO_ERROR:
 796                 v.val = -sock_error(sk);
 797                 if (v.val == 0)
 798                         v.val = xchg(&sk->sk_err_soft, 0);
 799                 break;
 800
 801         case SO_OOBINLINE:
 802                 v.val = !!sock_flag(sk, SOCK_URGINLINE);
 803                 break;
 804
 805         case SO_NO_CHECK:
 806                 v.val = sk->sk_no_check;
 807                 break;
 808
 809         case SO_PRIORITY:
 810                 v.val = sk->sk_priority;
 811                 break;
 812
 813         case SO_LINGER:
 814                 lv              = sizeof(v.ling);
 815                 v.ling.l_onoff  = !!sock_flag(sk, SOCK_LINGER);
 816                 v.ling.l_linger = sk->sk_lingertime / HZ;
 817                 break;
 818
 819         case SO_BSDCOMPAT:
 820                 sock_warn_obsolete_bsdism("getsockopt");
 821                 break;
 822
 823         case SO_TIMESTAMP:
 824                 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
 825                                 !sock_flag(sk, SOCK_RCVTSTAMPNS);
 826                 break;
 827
 828         case SO_TIMESTAMPNS:
 829                 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
 830                 break;
 831
 832         case SO_TIMESTAMPING:
 833                 v.val = 0;
 834                 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
 835                         v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
 836                 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
 837                         v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
 838                 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
 839                         v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
 840                 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
 841                         v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
 842                 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
 843                         v.val |= SOF_TIMESTAMPING_SOFTWARE;
 844                 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
 845                         v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
 846                 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
 847                         v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
 848                 break;
 849
 850         case SO_RCVTIMEO:
 851                 lv = sizeof(struct timeval);
 852                 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
 853                         v.tm.tv_sec = 0;
 854                         v.tm.tv_usec = 0;
 855                 } else {
 856                         v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
 857                         v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
 858                 }
 859                 break;
 860
 861         case SO_SNDTIMEO:
 862                 lv = sizeof(struct timeval);
 863                 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
 864                         v.tm.tv_sec = 0;
 865                         v.tm.tv_usec = 0;
 866                 } else {
 867                         v.tm.tv_sec = sk->sk_sndtimeo / HZ;
 868                         v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
 869                 }
 870                 break;
 871
 872         case SO_RCVLOWAT:
 873                 v.val = sk->sk_rcvlowat;
 874                 break;
 875
 876         case SO_SNDLOWAT:
 877                 v.val = 1;
 878                 break;
 879
 880         case SO_PASSCRED:
 881                 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
 882                 break;
 883
 884         case SO_PEERCRED:
 885                 if (len > sizeof(sk->sk_peercred))
 886                         len = sizeof(sk->sk_peercred);
 887                 if (copy_to_user(optval, &sk->sk_peercred, len))
 888                         return -EFAULT;
 889                 goto lenout;
 890
 891         case SO_PEERNAME:
 892         {
 893                 char address[128];
 894
 895                 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
 896                         return -ENOTCONN;
 897                 if (lv < len)
 898                         return -EINVAL;
 899                 if (copy_to_user(optval, address, len))
 900                         return -EFAULT;
 901                 goto lenout;
 902         }
 903
 904         /* Dubious BSD thing... Probably nobody even uses it, but
 905          * the UNIX standard wants it for whatever reason... -DaveM
 906          */
 907         case SO_ACCEPTCONN:
 908                 v.val = sk->sk_state == TCP_LISTEN;
 909                 break;
 910
 911         case SO_PASSSEC:
 912                 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
 913                 break;
 914
 915         case SO_PEERSEC:
 916                 return security_socket_getpeersec_stream(sock, optval, optlen, len);
 917
 918         case SO_MARK:
 919                 v.val = sk->sk_mark;
 920                 break;
 921
 922         case SO_RXQ_OVFL:
 923                 v.val = !!sock_flag(sk, SOCK_RXQ_OVFL);
 924                 break;
 925
 926         default:
 927                 return -ENOPROTOOPT;
 928         }
 929
 930         if (len > lv)
 931                 len = lv;
 932         if (copy_to_user(optval, &v, len))
 933                 return -EFAULT;
 934 lenout:
 935         if (put_user(len, optlen))
 936                 return -EFAULT;
 937         return 0;
 938 }
 939
 940 /*
 941  * Initialize an sk_lock.
 942  *
 943  * (We also register the sk_lock with the lock validator.)
 944  */
 945 static inline void sock_lock_init(struct sock *sk)
 946 {
 947         sock_lock_init_class_and_name(sk,
 948                         af_family_slock_key_strings[sk->sk_family],
 949                         af_family_slock_keys + sk->sk_family,
 950                         af_family_key_strings[sk->sk_family],
 951                         af_family_keys + sk->sk_family);
 952 }
 953
 954 /*
 955  * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
 956  * even temporarly, because of RCU lookups. sk_node should also be left as is.
 957  */
 958 static void sock_copy(struct sock *nsk, const struct sock *osk)
 959 {
 960 #ifdef CONFIG_SECURITY_NETWORK
 961         void *sptr = nsk->sk_security;
 962 #endif
 963         BUILD_BUG_ON(offsetof(struct sock, sk_copy_start) !=
 964                      sizeof(osk->sk_node) + sizeof(osk->sk_refcnt) +
 965                      sizeof(osk->sk_tx_queue_mapping));
 966         memcpy(&nsk->sk_copy_start, &osk->sk_copy_start,
 967                osk->sk_prot->obj_size - offsetof(struct sock, sk_copy_start));
 968 #ifdef CONFIG_SECURITY_NETWORK
 969         nsk->sk_security = sptr;
 970         security_sk_clone(osk, nsk);
 971 #endif
 972 }
 973
 974 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
 975                 int family)
 976 {
 977         struct sock *sk;
 978         struct kmem_cache *slab;
 979
 980         slab = prot->slab;
 981         if (slab != NULL) {
 982                 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
 983                 if (!sk)
 984                         return sk;
 985                 if (priority & __GFP_ZERO) {
 986                         /*
 987                          * caches using SLAB_DESTROY_BY_RCU should let
 988                          * sk_node.next un-modified. Special care is taken
 989                          * when initializing object to zero.
 990                          */
 991                         if (offsetof(struct sock, sk_node.next) != 0)
 992                                 memset(sk, 0, offsetof(struct sock, sk_node.next));
 993                         memset(&sk->sk_node.pprev, 0,
 994                                prot->obj_size - offsetof(struct sock,
 995                                                          sk_node.pprev));
 996                 }
 997         }
 998         else
 999                 sk = kmalloc(prot->obj_size, priority);
1000
1001         if (sk != NULL) {
1002                 kmemcheck_annotate_bitfield(sk, flags);
1003
1004                 if (security_sk_alloc(sk, family, priority))
1005                         goto out_free;
1006
1007                 if (!try_module_get(prot->owner))
1008                         goto out_free_sec;
1009                 sk_tx_queue_clear(sk);
1010         }
1011
1012         return sk;
1013
1014 out_free_sec:
1015         security_sk_free(sk);
1016 out_free:
1017         if (slab != NULL)
1018                 kmem_cache_free(slab, sk);
1019         else
1020                 kfree(sk);
1021         return NULL;
1022 }
1023
1024 static void sk_prot_free(struct proto *prot, struct sock *sk)
1025 {
1026         struct kmem_cache *slab;
1027         struct module *owner;
1028
1029         owner = prot->owner;
1030         slab = prot->slab;
1031
1032         security_sk_free(sk);
1033         if (slab != NULL)
1034                 kmem_cache_free(slab, sk);
1035         else
1036                 kfree(sk);
1037         module_put(owner);
1038 }
1039
1040 /**
1041  *      sk_alloc - All socket objects are allocated here
1042  *      @net: the applicable net namespace
1043  *      @family: protocol family
1044  *      @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1045  *      @prot: struct proto associated with this new sock instance
1046  */
1047 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1048                       struct proto *prot)
1049 {
1050         struct sock *sk;
1051
1052         sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1053         if (sk) {
1054                 sk->sk_family = family;
1055                 /*
1056                  * See comment in struct sock definition to understand
1057                  * why we need sk_prot_creator -acme
1058                  */
1059                 sk->sk_prot = sk->sk_prot_creator = prot;
1060                 sock_lock_init(sk);
1061                 sock_net_set(sk, get_net(net));
1062                 atomic_set(&sk->sk_wmem_alloc, 1);
1063         }
1064
1065         return sk;
1066 }
1067 EXPORT_SYMBOL(sk_alloc);
1068
1069 static void __sk_free(struct sock *sk)
1070 {
1071         struct sk_filter *filter;
1072
1073         if (sk->sk_destruct)
1074                 sk->sk_destruct(sk);
1075
1076         filter = rcu_dereference(sk->sk_filter);
1077         if (filter) {
1078                 sk_filter_uncharge(sk, filter);
1079                 rcu_assign_pointer(sk->sk_filter, NULL);
1080         }
1081
1082         sock_disable_timestamp(sk, SOCK_TIMESTAMP);
1083         sock_disable_timestamp(sk, SOCK_TIMESTAMPING_RX_SOFTWARE);
1084
1085         if (atomic_read(&sk->sk_omem_alloc))
1086                 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
1087                        __func__, atomic_read(&sk->sk_omem_alloc));
1088
1089         put_net(sock_net(sk));
1090         sk_prot_free(sk->sk_prot_creator, sk);
1091 }
1092
1093 void sk_free(struct sock *sk)
1094 {
1095         /*
1096          * We substract one from sk_wmem_alloc and can know if
1097          * some packets are still in some tx queue.
1098          * If not null, sock_wfree() will call __sk_free(sk) later
1099          */
1100         if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1101                 __sk_free(sk);
1102 }
1103 EXPORT_SYMBOL(sk_free);
1104
1105 /*
1106  * Last sock_put should drop referrence to sk->sk_net. It has already
1107  * been dropped in sk_change_net. Taking referrence to stopping namespace
1108  * is not an option.
1109  * Take referrence to a socket to remove it from hash _alive_ and after that
1110  * destroy it in the context of init_net.
1111  */
1112 void sk_release_kernel(struct sock *sk)
1113 {
1114         if (sk == NULL || sk->sk_socket == NULL)
1115                 return;
1116
1117         sock_hold(sk);
1118         sock_release(sk->sk_socket);
1119         release_net(sock_net(sk));
1120         sock_net_set(sk, get_net(&init_net));
1121         sock_put(sk);
1122 }
1123 EXPORT_SYMBOL(sk_release_kernel);
1124
1125 struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
1126 {
1127         struct sock *newsk;
1128
1129         newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1130         if (newsk != NULL) {
1131                 struct sk_filter *filter;
1132
1133                 sock_copy(newsk, sk);
1134
1135                 /* SANITY */
1136                 get_net(sock_net(newsk));
1137                 sk_node_init(&newsk->sk_node);
1138                 sock_lock_init(newsk);
1139                 bh_lock_sock(newsk);
1140                 newsk->sk_backlog.head  = newsk->sk_backlog.tail = NULL;
1141
1142                 atomic_set(&newsk->sk_rmem_alloc, 0);
1143                 /*
1144                  * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1145                  */
1146                 atomic_set(&newsk->sk_wmem_alloc, 1);
1147                 atomic_set(&newsk->sk_omem_alloc, 0);
1148                 skb_queue_head_init(&newsk->sk_receive_queue);
1149                 skb_queue_head_init(&newsk->sk_write_queue);
1150 #ifdef CONFIG_NET_DMA
1151                 skb_queue_head_init(&newsk->sk_async_wait_queue);
1152 #endif
1153
1154                 rwlock_init(&newsk->sk_dst_lock);
1155                 rwlock_init(&newsk->sk_callback_lock);
1156                 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1157                                 af_callback_keys + newsk->sk_family,
1158                                 af_family_clock_key_strings[newsk->sk_family]);
1159
1160                 newsk->sk_dst_cache     = NULL;
1161                 newsk->sk_wmem_queued   = 0;
1162                 newsk->sk_forward_alloc = 0;
1163                 newsk->sk_send_head     = NULL;
1164                 newsk->sk_userlocks     = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1165
1166                 sock_reset_flag(newsk, SOCK_DONE);
1167                 skb_queue_head_init(&newsk->sk_error_queue);
1168
1169                 filter = newsk->sk_filter;
1170                 if (filter != NULL)
1171                         sk_filter_charge(newsk, filter);
1172
1173                 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1174                         /* It is still raw copy of parent, so invalidate
1175                          * destructor and make plain sk_free() */
1176                         newsk->sk_destruct = NULL;
1177                         sk_free(newsk);
1178                         newsk = NULL;
1179                         goto out;
1180                 }
1181
1182                 newsk->sk_err      = 0;
1183                 newsk->sk_priority = 0;
1184                 /*
1185                  * Before updating sk_refcnt, we must commit prior changes to memory
1186                  * (Documentation/RCU/rculist_nulls.txt for details)
1187                  */
1188                 smp_wmb();
1189                 atomic_set(&newsk->sk_refcnt, 2);
1190
1191                 /*
1192                  * Increment the counter in the same struct proto as the master
1193                  * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1194                  * is the same as sk->sk_prot->socks, as this field was copied
1195                  * with memcpy).
1196                  *
1197                  * This _changes_ the previous behaviour, where
1198                  * tcp_create_openreq_child always was incrementing the
1199                  * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1200                  * to be taken into account in all callers. -acme
1201                  */
1202                 sk_refcnt_debug_inc(newsk);
1203                 sk_set_socket(newsk, NULL);
1204                 newsk->sk_sleep  = NULL;
1205
1206                 if (newsk->sk_prot->sockets_allocated)
1207                         percpu_counter_inc(newsk->sk_prot->sockets_allocated);
1208         }
1209 out:
1210         return newsk;
1211 }
1212 EXPORT_SYMBOL_GPL(sk_clone);
1213
1214 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1215 {
1216         __sk_dst_set(sk, dst);
1217         sk->sk_route_caps = dst->dev->features;
1218         if (sk->sk_route_caps & NETIF_F_GSO)
1219                 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1220         if (sk_can_gso(sk)) {
1221                 if (dst->header_len) {
1222                         sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1223                 } else {
1224                         sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1225                         sk->sk_gso_max_size = dst->dev->gso_max_size;
1226                 }
1227         }
1228 }
1229 EXPORT_SYMBOL_GPL(sk_setup_caps);
1230
1231 void __init sk_init(void)
1232 {
1233         if (totalram_pages <= 4096) {
1234                 sysctl_wmem_max = 32767;
1235                 sysctl_rmem_max = 32767;
1236                 sysctl_wmem_default = 32767;
1237                 sysctl_rmem_default = 32767;
1238         } else if (totalram_pages >= 131072) {
1239                 sysctl_wmem_max = 131071;
1240                 sysctl_rmem_max = 131071;
1241         }
1242 }
1243
1244 /*
1245  *      Simple resource managers for sockets.
1246  */
1247
1248
1249 /*
1250  * Write buffer destructor automatically called from kfree_skb.
1251  */
1252 void sock_wfree(struct sk_buff *skb)
1253 {
1254         struct sock *sk = skb->sk;
1255         unsigned int len = skb->truesize;
1256
1257         if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1258                 /*
1259                  * Keep a reference on sk_wmem_alloc, this will be released
1260                  * after sk_write_space() call
1261                  */
1262                 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1263                 sk->sk_write_space(sk);
1264                 len = 1;
1265         }
1266         /*
1267          * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1268          * could not do because of in-flight packets
1269          */
1270         if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1271                 __sk_free(sk);
1272 }
1273 EXPORT_SYMBOL(sock_wfree);
1274
1275 /*
1276  * Read buffer destructor automatically called from kfree_skb.
1277  */
1278 void sock_rfree(struct sk_buff *skb)
1279 {
1280         struct sock *sk = skb->sk;
1281
1282         atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
1283         sk_mem_uncharge(skb->sk, skb->truesize);
1284 }
1285 EXPORT_SYMBOL(sock_rfree);
1286
1287
1288 int sock_i_uid(struct sock *sk)
1289 {
1290         int uid;
1291
1292         read_lock(&sk->sk_callback_lock);
1293         uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1294         read_unlock(&sk->sk_callback_lock);
1295         return uid;
1296 }
1297 EXPORT_SYMBOL(sock_i_uid);
1298
1299 unsigned long sock_i_ino(struct sock *sk)
1300 {
1301         unsigned long ino;
1302
1303         read_lock(&sk->sk_callback_lock);
1304         ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1305         read_unlock(&sk->sk_callback_lock);
1306         return ino;
1307 }
1308 EXPORT_SYMBOL(sock_i_ino);
1309
1310 /*
1311  * Allocate a skb from the socket's send buffer.
1312  */
1313 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1314                              gfp_t priority)
1315 {
1316         if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1317                 struct sk_buff *skb = alloc_skb(size, priority);
1318                 if (skb) {
1319                         skb_set_owner_w(skb, sk);
1320                         return skb;
1321                 }
1322         }
1323         return NULL;
1324 }
1325 EXPORT_SYMBOL(sock_wmalloc);
1326
1327 /*
1328  * Allocate a skb from the socket's receive buffer.
1329  */
1330 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1331                              gfp_t priority)
1332 {
1333         if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1334                 struct sk_buff *skb = alloc_skb(size, priority);
1335                 if (skb) {
1336                         skb_set_owner_r(skb, sk);
1337                         return skb;
1338                 }
1339         }
1340         return NULL;
1341 }
1342
1343 /*
1344  * Allocate a memory block from the socket's option memory buffer.
1345  */
1346 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1347 {
1348         if ((unsigned)size <= sysctl_optmem_max &&
1349             atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1350                 void *mem;
1351                 /* First do the add, to avoid the race if kmalloc
1352                  * might sleep.
1353                  */
1354                 atomic_add(size, &sk->sk_omem_alloc);
1355                 mem = kmalloc(size, priority);
1356                 if (mem)
1357                         return mem;
1358                 atomic_sub(size, &sk->sk_omem_alloc);
1359         }
1360         return NULL;
1361 }
1362 EXPORT_SYMBOL(sock_kmalloc);
1363
1364 /*
1365  * Free an option memory block.
1366  */
1367 void sock_kfree_s(struct sock *sk, void *mem, int size)
1368 {
1369         kfree(mem);
1370         atomic_sub(size, &sk->sk_omem_alloc);
1371 }
1372 EXPORT_SYMBOL(sock_kfree_s);
1373
1374 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1375    I think, these locks should be removed for datagram sockets.
1376  */
1377 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1378 {
1379         DEFINE_WAIT(wait);
1380
1381         clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1382         for (;;) {
1383                 if (!timeo)
1384                         break;
1385                 if (signal_pending(current))
1386                         break;
1387                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1388                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1389                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1390                         break;
1391                 if (sk->sk_shutdown & SEND_SHUTDOWN)
1392                         break;
1393                 if (sk->sk_err)
1394                         break;
1395                 timeo = schedule_timeout(timeo);
1396         }
1397         finish_wait(sk->sk_sleep, &wait);
1398         return timeo;
1399 }
1400
1401
1402 /*
1403  *      Generic send/receive buffer handlers
1404  */
1405
1406 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1407                                      unsigned long data_len, int noblock,
1408                                      int *errcode)
1409 {
1410         struct sk_buff *skb;
1411         gfp_t gfp_mask;
1412         long timeo;
1413         int err;
1414
1415         gfp_mask = sk->sk_allocation;
1416         if (gfp_mask & __GFP_WAIT)
1417                 gfp_mask |= __GFP_REPEAT;
1418
1419         timeo = sock_sndtimeo(sk, noblock);
1420         while (1) {
1421                 err = sock_error(sk);
1422                 if (err != 0)
1423                         goto failure;
1424
1425                 err = -EPIPE;
1426                 if (sk->sk_shutdown & SEND_SHUTDOWN)
1427                         goto failure;
1428
1429                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1430                         skb = alloc_skb(header_len, gfp_mask);
1431                         if (skb) {
1432                                 int npages;
1433                                 int i;
1434
1435                                 /* No pages, we're done... */
1436                                 if (!data_len)
1437                                         break;
1438
1439                                 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1440                                 skb->truesize += data_len;
1441                                 skb_shinfo(skb)->nr_frags = npages;
1442                                 for (i = 0; i < npages; i++) {
1443                                         struct page *page;
1444                                         skb_frag_t *frag;
1445
1446                                         page = alloc_pages(sk->sk_allocation, 0);
1447                                         if (!page) {
1448                                                 err = -ENOBUFS;
1449                                                 skb_shinfo(skb)->nr_frags = i;
1450                                                 kfree_skb(skb);
1451                                                 goto failure;
1452                                         }
1453
1454                                         frag = &skb_shinfo(skb)->frags[i];
1455                                         frag->page = page;
1456                                         frag->page_offset = 0;
1457                                         frag->size = (data_len >= PAGE_SIZE ?
1458                                                       PAGE_SIZE :
1459                                                       data_len);
1460                                         data_len -= PAGE_SIZE;
1461                                 }
1462
1463                                 /* Full success... */
1464                                 break;
1465                         }
1466                         err = -ENOBUFS;
1467                         goto failure;
1468                 }
1469                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1470                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1471                 err = -EAGAIN;
1472                 if (!timeo)
1473                         goto failure;
1474                 if (signal_pending(current))
1475                         goto interrupted;
1476                 timeo = sock_wait_for_wmem(sk, timeo);
1477         }
1478
1479         skb_set_owner_w(skb, sk);
1480         return skb;
1481
1482 interrupted:
1483         err = sock_intr_errno(timeo);
1484 failure:
1485         *errcode = err;
1486         return NULL;
1487 }
1488 EXPORT_SYMBOL(sock_alloc_send_pskb);
1489
1490 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1491                                     int noblock, int *errcode)
1492 {
1493         return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1494 }
1495 EXPORT_SYMBOL(sock_alloc_send_skb);
1496
1497 static void __lock_sock(struct sock *sk)
1498 {
1499         DEFINE_WAIT(wait);
1500
1501         for (;;) {
1502                 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1503                                         TASK_UNINTERRUPTIBLE);
1504                 spin_unlock_bh(&sk->sk_lock.slock);
1505                 schedule();
1506                 spin_lock_bh(&sk->sk_lock.slock);
1507                 if (!sock_owned_by_user(sk))
1508                         break;
1509         }
1510         finish_wait(&sk->sk_lock.wq, &wait);
1511 }
1512
1513 static void __release_sock(struct sock *sk)
1514 {
1515         struct sk_buff *skb = sk->sk_backlog.head;
1516
1517         do {
1518                 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1519                 bh_unlock_sock(sk);
1520
1521                 do {
1522                         struct sk_buff *next = skb->next;
1523
1524                         skb->next = NULL;
1525                         sk_backlog_rcv(sk, skb);
1526
1527                         /*
1528                          * We are in process context here with softirqs
1529                          * disabled, use cond_resched_softirq() to preempt.
1530                          * This is safe to do because we've taken the backlog
1531                          * queue private:
1532                          */
1533                         cond_resched_softirq();
1534
1535                         skb = next;
1536                 } while (skb != NULL);
1537
1538                 bh_lock_sock(sk);
1539         } while ((skb = sk->sk_backlog.head) != NULL);
1540 }
1541
1542 /**
1543  * sk_wait_data - wait for data to arrive at sk_receive_queue
1544  * @sk:    sock to wait on
1545  * @timeo: for how long
1546  *
1547  * Now socket state including sk->sk_err is changed only under lock,
1548  * hence we may omit checks after joining wait queue.
1549  * We check receive queue before schedule() only as optimization;
1550  * it is very likely that release_sock() added new data.
1551  */
1552 int sk_wait_data(struct sock *sk, long *timeo)
1553 {
1554         int rc;
1555         DEFINE_WAIT(wait);
1556
1557         prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1558         set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1559         rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1560         clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1561         finish_wait(sk->sk_sleep, &wait);
1562         return rc;
1563 }
1564 EXPORT_SYMBOL(sk_wait_data);
1565
1566 /**
1567  *      __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1568  *      @sk: socket
1569  *      @size: memory size to allocate
1570  *      @kind: allocation type
1571  *
1572  *      If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1573  *      rmem allocation. This function assumes that protocols which have
1574  *      memory_pressure use sk_wmem_queued as write buffer accounting.
1575  */
1576 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1577 {
1578         struct proto *prot = sk->sk_prot;
1579         int amt = sk_mem_pages(size);
1580         int allocated;
1581
1582         sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1583         allocated = atomic_add_return(amt, prot->memory_allocated);
1584
1585         /* Under limit. */
1586         if (allocated <= prot->sysctl_mem[0]) {
1587                 if (prot->memory_pressure && *prot->memory_pressure)
1588                         *prot->memory_pressure = 0;
1589                 return 1;
1590         }
1591
1592         /* Under pressure. */
1593         if (allocated > prot->sysctl_mem[1])
1594                 if (prot->enter_memory_pressure)
1595                         prot->enter_memory_pressure(sk);
1596
1597         /* Over hard limit. */
1598         if (allocated > prot->sysctl_mem[2])
1599                 goto suppress_allocation;
1600
1601         /* guarantee minimum buffer size under pressure */
1602         if (kind == SK_MEM_RECV) {
1603                 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1604                         return 1;
1605         } else { /* SK_MEM_SEND */
1606                 if (sk->sk_type == SOCK_STREAM) {
1607                         if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1608                                 return 1;
1609                 } else if (atomic_read(&sk->sk_wmem_alloc) <
1610                            prot->sysctl_wmem[0])
1611                                 return 1;
1612         }
1613
1614         if (prot->memory_pressure) {
1615                 int alloc;
1616
1617                 if (!*prot->memory_pressure)
1618                         return 1;
1619                 alloc = percpu_counter_read_positive(prot->sockets_allocated);
1620                 if (prot->sysctl_mem[2] > alloc *
1621                     sk_mem_pages(sk->sk_wmem_queued +
1622                                  atomic_read(&sk->sk_rmem_alloc) +
1623                                  sk->sk_forward_alloc))
1624                         return 1;
1625         }
1626
1627 suppress_allocation:
1628
1629         if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1630                 sk_stream_moderate_sndbuf(sk);
1631
1632                 /* Fail only if socket is _under_ its sndbuf.
1633                  * In this case we cannot block, so that we have to fail.
1634                  */
1635                 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1636                         return 1;
1637         }
1638
1639         /* Alas. Undo changes. */
1640         sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1641         atomic_sub(amt, prot->memory_allocated);
1642         return 0;
1643 }
1644 EXPORT_SYMBOL(__sk_mem_schedule);
1645
1646 /**
1647  *      __sk_reclaim - reclaim memory_allocated
1648  *      @sk: socket
1649  */
1650 void __sk_mem_reclaim(struct sock *sk)
1651 {
1652         struct proto *prot = sk->sk_prot;
1653
1654         atomic_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
1655                    prot->memory_allocated);
1656         sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1657
1658         if (prot->memory_pressure && *prot->memory_pressure &&
1659             (atomic_read(prot->memory_allocated) < prot->sysctl_mem[0]))
1660                 *prot->memory_pressure = 0;
1661 }
1662 EXPORT_SYMBOL(__sk_mem_reclaim);
1663
1664
1665 /*
1666  * Set of default routines for initialising struct proto_ops when
1667  * the protocol does not support a particular function. In certain
1668  * cases where it makes no sense for a protocol to have a "do nothing"
1669  * function, some default processing is provided.
1670  */
1671
1672 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1673 {
1674         return -EOPNOTSUPP;
1675 }
1676 EXPORT_SYMBOL(sock_no_bind);
1677
1678 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1679                     int len, int flags)
1680 {
1681         return -EOPNOTSUPP;
1682 }
1683 EXPORT_SYMBOL(sock_no_connect);
1684
1685 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1686 {
1687         return -EOPNOTSUPP;
1688 }
1689 EXPORT_SYMBOL(sock_no_socketpair);
1690
1691 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1692 {
1693         return -EOPNOTSUPP;
1694 }
1695 EXPORT_SYMBOL(sock_no_accept);
1696
1697 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1698                     int *len, int peer)
1699 {
1700         return -EOPNOTSUPP;
1701 }
1702 EXPORT_SYMBOL(sock_no_getname);
1703
1704 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1705 {
1706         return 0;
1707 }
1708 EXPORT_SYMBOL(sock_no_poll);
1709
1710 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1711 {
1712         return -EOPNOTSUPP;
1713 }
1714 EXPORT_SYMBOL(sock_no_ioctl);
1715
1716 int sock_no_listen(struct socket *sock, int backlog)
1717 {
1718         return -EOPNOTSUPP;
1719 }
1720 EXPORT_SYMBOL(sock_no_listen);
1721
1722 int sock_no_shutdown(struct socket *sock, int how)
1723 {
1724         return -EOPNOTSUPP;
1725 }
1726 EXPORT_SYMBOL(sock_no_shutdown);
1727
1728 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1729                     char __user *optval, unsigned int optlen)
1730 {
1731         return -EOPNOTSUPP;
1732 }
1733 EXPORT_SYMBOL(sock_no_setsockopt);
1734
1735 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1736                     char __user *optval, int __user *optlen)
1737 {
1738         return -EOPNOTSUPP;
1739 }
1740 EXPORT_SYMBOL(sock_no_getsockopt);
1741
1742 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1743                     size_t len)
1744 {
1745         return -EOPNOTSUPP;
1746 }
1747 EXPORT_SYMBOL(sock_no_sendmsg);
1748
1749 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1750                     size_t len, int flags)
1751 {
1752         return -EOPNOTSUPP;
1753 }
1754 EXPORT_SYMBOL(sock_no_recvmsg);
1755
1756 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1757 {
1758         /* Mirror missing mmap method error code */
1759         return -ENODEV;
1760 }
1761 EXPORT_SYMBOL(sock_no_mmap);
1762
1763 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1764 {
1765         ssize_t res;
1766         struct msghdr msg = {.msg_flags = flags};
1767         struct kvec iov;
1768         char *kaddr = kmap(page);
1769         iov.iov_base = kaddr + offset;
1770         iov.iov_len = size;
1771         res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1772         kunmap(page);
1773         return res;
1774 }
1775 EXPORT_SYMBOL(sock_no_sendpage);
1776
1777 /*
1778  *      Default Socket Callbacks
1779  */
1780
1781 static void sock_def_wakeup(struct sock *sk)
1782 {
1783         read_lock(&sk->sk_callback_lock);
1784         if (sk_has_sleeper(sk))
1785                 wake_up_interruptible_all(sk->sk_sleep);
1786         read_unlock(&sk->sk_callback_lock);
1787 }
1788
1789 static void sock_def_error_report(struct sock *sk)
1790 {
1791         read_lock(&sk->sk_callback_lock);
1792         if (sk_has_sleeper(sk))
1793                 wake_up_interruptible_poll(sk->sk_sleep, POLLERR);
1794         sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
1795         read_unlock(&sk->sk_callback_lock);
1796 }
1797
1798 static void sock_def_readable(struct sock *sk, int len)
1799 {
1800         read_lock(&sk->sk_callback_lock);
1801         if (sk_has_sleeper(sk))
1802                 wake_up_interruptible_sync_poll(sk->sk_sleep, POLLIN |
1803                                                 POLLRDNORM | POLLRDBAND);
1804         sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1805         read_unlock(&sk->sk_callback_lock);
1806 }
1807
1808 static void sock_def_write_space(struct sock *sk)
1809 {
1810         read_lock(&sk->sk_callback_lock);
1811
1812         /* Do not wake up a writer until he can make "significant"
1813          * progress.  --DaveM
1814          */
1815         if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1816                 if (sk_has_sleeper(sk))
1817                         wake_up_interruptible_sync_poll(sk->sk_sleep, POLLOUT |
1818                                                 POLLWRNORM | POLLWRBAND);
1819
1820                 /* Should agree with poll, otherwise some programs break */
1821                 if (sock_writeable(sk))
1822                         sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1823         }
1824
1825         read_unlock(&sk->sk_callback_lock);
1826 }
1827
1828 static void sock_def_destruct(struct sock *sk)
1829 {
1830         kfree(sk->sk_protinfo);
1831 }
1832
1833 void sk_send_sigurg(struct sock *sk)
1834 {
1835         if (sk->sk_socket && sk->sk_socket->file)
1836                 if (send_sigurg(&sk->sk_socket->file->f_owner))
1837                         sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1838 }
1839 EXPORT_SYMBOL(sk_send_sigurg);
1840
1841 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1842                     unsigned long expires)
1843 {
1844         if (!mod_timer(timer, expires))
1845                 sock_hold(sk);
1846 }
1847 EXPORT_SYMBOL(sk_reset_timer);
1848
1849 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1850 {
1851         if (timer_pending(timer) && del_timer(timer))
1852                 __sock_put(sk);
1853 }
1854 EXPORT_SYMBOL(sk_stop_timer);
1855
1856 void sock_init_data(struct socket *sock, struct sock *sk)
1857 {
1858         skb_queue_head_init(&sk->sk_receive_queue);
1859         skb_queue_head_init(&sk->sk_write_queue);
1860         skb_queue_head_init(&sk->sk_error_queue);
1861 #ifdef CONFIG_NET_DMA
1862         skb_queue_head_init(&sk->sk_async_wait_queue);
1863 #endif
1864
1865         sk->sk_send_head        =       NULL;
1866
1867         init_timer(&sk->sk_timer);
1868
1869         sk->sk_allocation       =       GFP_KERNEL;
1870         sk->sk_rcvbuf           =       sysctl_rmem_default;
1871         sk->sk_sndbuf           =       sysctl_wmem_default;
1872         sk->sk_state            =       TCP_CLOSE;
1873         sk_set_socket(sk, sock);
1874
1875         sock_set_flag(sk, SOCK_ZAPPED);
1876
1877         if (sock) {
1878                 sk->sk_type     =       sock->type;
1879                 sk->sk_sleep    =       &sock->wait;
1880                 sock->sk        =       sk;
1881         } else
1882                 sk->sk_sleep    =       NULL;
1883
1884         rwlock_init(&sk->sk_dst_lock);
1885         rwlock_init(&sk->sk_callback_lock);
1886         lockdep_set_class_and_name(&sk->sk_callback_lock,
1887                         af_callback_keys + sk->sk_family,
1888                         af_family_clock_key_strings[sk->sk_family]);
1889
1890         sk->sk_state_change     =       sock_def_wakeup;
1891         sk->sk_data_ready       =       sock_def_readable;
1892         sk->sk_write_space      =       sock_def_write_space;
1893         sk->sk_error_report     =       sock_def_error_report;
1894         sk->sk_destruct         =       sock_def_destruct;
1895
1896         sk->sk_sndmsg_page      =       NULL;
1897         sk->sk_sndmsg_off       =       0;
1898
1899         sk->sk_peercred.pid     =       0;
1900         sk->sk_peercred.uid     =       -1;
1901         sk->sk_peercred.gid     =       -1;
1902         sk->sk_write_pending    =       0;
1903         sk->sk_rcvlowat         =       1;
1904         sk->sk_rcvtimeo         =       MAX_SCHEDULE_TIMEOUT;
1905         sk->sk_sndtimeo         =       MAX_SCHEDULE_TIMEOUT;
1906
1907         sk->sk_stamp = ktime_set(-1L, 0);
1908
1909         /*
1910          * Before updating sk_refcnt, we must commit prior changes to memory
1911          * (Documentation/RCU/rculist_nulls.txt for details)
1912          */
1913         smp_wmb();
1914         atomic_set(&sk->sk_refcnt, 1);
1915         atomic_set(&sk->sk_drops, 0);
1916 }
1917 EXPORT_SYMBOL(sock_init_data);
1918
1919 void lock_sock_nested(struct sock *sk, int subclass)
1920 {
1921         might_sleep();
1922         spin_lock_bh(&sk->sk_lock.slock);
1923         if (sk->sk_lock.owned)
1924                 __lock_sock(sk);
1925         sk->sk_lock.owned = 1;
1926         spin_unlock(&sk->sk_lock.slock);
1927         /*
1928          * The sk_lock has mutex_lock() semantics here:
1929          */
1930         mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
1931         local_bh_enable();
1932 }
1933 EXPORT_SYMBOL(lock_sock_nested);
1934
1935 void release_sock(struct sock *sk)
1936 {
1937         /*
1938          * The sk_lock has mutex_unlock() semantics:
1939          */
1940         mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
1941
1942         spin_lock_bh(&sk->sk_lock.slock);
1943         if (sk->sk_backlog.tail)
1944                 __release_sock(sk);
1945         sk->sk_lock.owned = 0;
1946         if (waitqueue_active(&sk->sk_lock.wq))
1947                 wake_up(&sk->sk_lock.wq);
1948         spin_unlock_bh(&sk->sk_lock.slock);
1949 }
1950 EXPORT_SYMBOL(release_sock);
1951
1952 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1953 {
1954         struct timeval tv;
1955         if (!sock_flag(sk, SOCK_TIMESTAMP))
1956                 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
1957         tv = ktime_to_timeval(sk->sk_stamp);
1958         if (tv.tv_sec == -1)
1959                 return -ENOENT;
1960         if (tv.tv_sec == 0) {
1961                 sk->sk_stamp = ktime_get_real();
1962                 tv = ktime_to_timeval(sk->sk_stamp);
1963         }
1964         return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
1965 }
1966 EXPORT_SYMBOL(sock_get_timestamp);
1967
1968 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
1969 {
1970         struct timespec ts;
1971         if (!sock_flag(sk, SOCK_TIMESTAMP))
1972                 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
1973         ts = ktime_to_timespec(sk->sk_stamp);
1974         if (ts.tv_sec == -1)
1975                 return -ENOENT;
1976         if (ts.tv_sec == 0) {
1977                 sk->sk_stamp = ktime_get_real();
1978                 ts = ktime_to_timespec(sk->sk_stamp);
1979         }
1980         return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
1981 }
1982 EXPORT_SYMBOL(sock_get_timestampns);
1983
1984 void sock_enable_timestamp(struct sock *sk, int flag)
1985 {
1986         if (!sock_flag(sk, flag)) {
1987                 sock_set_flag(sk, flag);
1988                 /*
1989                  * we just set one of the two flags which require net
1990                  * time stamping, but time stamping might have been on
1991                  * already because of the other one
1992                  */
1993                 if (!sock_flag(sk,
1994                                 flag == SOCK_TIMESTAMP ?
1995                                 SOCK_TIMESTAMPING_RX_SOFTWARE :
1996                                 SOCK_TIMESTAMP))
1997                         net_enable_timestamp();
1998         }
1999 }
2000
2001 /*
2002  *      Get a socket option on an socket.
2003  *
2004  *      FIX: POSIX 1003.1g is very ambiguous here. It states that
2005  *      asynchronous errors should be reported by getsockopt. We assume
2006  *      this means if you specify SO_ERROR (otherwise whats the point of it).
2007  */
2008 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2009                            char __user *optval, int __user *optlen)
2010 {
2011         struct sock *sk = sock->sk;
2012
2013         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2014 }
2015 EXPORT_SYMBOL(sock_common_getsockopt);
2016
2017 #ifdef CONFIG_COMPAT
2018 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2019                                   char __user *optval, int __user *optlen)
2020 {
2021         struct sock *sk = sock->sk;
2022
2023         if (sk->sk_prot->compat_getsockopt != NULL)
2024                 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2025                                                       optval, optlen);
2026         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2027 }
2028 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2029 #endif
2030
2031 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2032                         struct msghdr *msg, size_t size, int flags)
2033 {
2034         struct sock *sk = sock->sk;
2035         int addr_len = 0;
2036         int err;
2037
2038         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2039                                    flags & ~MSG_DONTWAIT, &addr_len);
2040         if (err >= 0)
2041                 msg->msg_namelen = addr_len;
2042         return err;
2043 }
2044 EXPORT_SYMBOL(sock_common_recvmsg);
2045
2046 /*
2047  *      Set socket options on an inet socket.
2048  */
2049 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2050                            char __user *optval, unsigned int optlen)
2051 {
2052         struct sock *sk = sock->sk;
2053
2054         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2055 }
2056 EXPORT_SYMBOL(sock_common_setsockopt);
2057
2058 #ifdef CONFIG_COMPAT
2059 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2060                                   char __user *optval, unsigned int optlen)
2061 {
2062         struct sock *sk = sock->sk;
2063
2064         if (sk->sk_prot->compat_setsockopt != NULL)
2065                 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2066                                                       optval, optlen);
2067         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2068 }
2069 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2070 #endif
2071
2072 void sk_common_release(struct sock *sk)
2073 {
2074         if (sk->sk_prot->destroy)
2075                 sk->sk_prot->destroy(sk);
2076
2077         /*
2078          * Observation: when sock_common_release is called, processes have
2079          * no access to socket. But net still has.
2080          * Step one, detach it from networking:
2081          *
2082          * A. Remove from hash tables.
2083          */
2084
2085         sk->sk_prot->unhash(sk);
2086
2087         /*
2088          * In this point socket cannot receive new packets, but it is possible
2089          * that some packets are in flight because some CPU runs receiver and
2090          * did hash table lookup before we unhashed socket. They will achieve
2091          * receive queue and will be purged by socket destructor.
2092          *
2093          * Also we still have packets pending on receive queue and probably,
2094          * our own packets waiting in device queues. sock_destroy will drain
2095          * receive queue, but transmitted packets will delay socket destruction
2096          * until the last reference will be released.
2097          */
2098
2099         sock_orphan(sk);
2100
2101         xfrm_sk_free_policy(sk);
2102
2103         sk_refcnt_debug_release(sk);
2104         sock_put(sk);
2105 }
2106 EXPORT_SYMBOL(sk_common_release);
2107
2108 static DEFINE_RWLOCK(proto_list_lock);
2109 static LIST_HEAD(proto_list);
2110
2111 #ifdef CONFIG_PROC_FS
2112 #define PROTO_INUSE_NR  64      /* should be enough for the first time */
2113 struct prot_inuse {
2114         int val[PROTO_INUSE_NR];
2115 };
2116
2117 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2118
2119 #ifdef CONFIG_NET_NS
2120 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2121 {
2122         int cpu = smp_processor_id();
2123         per_cpu_ptr(net->core.inuse, cpu)->val[prot->inuse_idx] += val;
2124 }
2125 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2126
2127 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2128 {
2129         int cpu, idx = prot->inuse_idx;
2130         int res = 0;
2131
2132         for_each_possible_cpu(cpu)
2133                 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2134
2135         return res >= 0 ? res : 0;
2136 }
2137 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2138
2139 static int sock_inuse_init_net(struct net *net)
2140 {
2141         net->core.inuse = alloc_percpu(struct prot_inuse);
2142         return net->core.inuse ? 0 : -ENOMEM;
2143 }
2144
2145 static void sock_inuse_exit_net(struct net *net)
2146 {
2147         free_percpu(net->core.inuse);
2148 }
2149
2150 static struct pernet_operations net_inuse_ops = {
2151         .init = sock_inuse_init_net,
2152         .exit = sock_inuse_exit_net,
2153 };
2154
2155 static __init int net_inuse_init(void)
2156 {
2157         if (register_pernet_subsys(&net_inuse_ops))
2158                 panic("Cannot initialize net inuse counters");
2159
2160         return 0;
2161 }
2162
2163 core_initcall(net_inuse_init);
2164 #else
2165 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2166
2167 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2168 {
2169         __get_cpu_var(prot_inuse).val[prot->inuse_idx] += val;
2170 }
2171 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2172
2173 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2174 {
2175         int cpu, idx = prot->inuse_idx;
2176         int res = 0;
2177
2178         for_each_possible_cpu(cpu)
2179                 res += per_cpu(prot_inuse, cpu).val[idx];
2180
2181         return res >= 0 ? res : 0;
2182 }
2183 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2184 #endif
2185
2186 static void assign_proto_idx(struct proto *prot)
2187 {
2188         prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2189
2190         if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2191                 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2192                 return;
2193         }
2194
2195         set_bit(prot->inuse_idx, proto_inuse_idx);
2196 }
2197
2198 static void release_proto_idx(struct proto *prot)
2199 {
2200         if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2201                 clear_bit(prot->inuse_idx, proto_inuse_idx);
2202 }
2203 #else
2204 static inline void assign_proto_idx(struct proto *prot)
2205 {
2206 }
2207
2208 static inline void release_proto_idx(struct proto *prot)
2209 {
2210 }
2211 #endif
2212
2213 int proto_register(struct proto *prot, int alloc_slab)
2214 {
2215         if (alloc_slab) {
2216                 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2217                                         SLAB_HWCACHE_ALIGN | prot->slab_flags,
2218                                         NULL);
2219
2220                 if (prot->slab == NULL) {
2221                         printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2222                                prot->name);
2223                         goto out;
2224                 }
2225
2226                 if (prot->rsk_prot != NULL) {
2227                         static const char mask[] = "request_sock_%s";
2228
2229                         prot->rsk_prot->slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
2230                         if (prot->rsk_prot->slab_name == NULL)
2231                                 goto out_free_sock_slab;
2232
2233                         sprintf(prot->rsk_prot->slab_name, mask, prot->name);
2234                         prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2235                                                                  prot->rsk_prot->obj_size, 0,
2236                                                                  SLAB_HWCACHE_ALIGN, NULL);
2237
2238                         if (prot->rsk_prot->slab == NULL) {
2239                                 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2240                                        prot->name);
2241                                 goto out_free_request_sock_slab_name;
2242                         }
2243                 }
2244
2245                 if (prot->twsk_prot != NULL) {
2246                         static const char mask[] = "tw_sock_%s";
2247
2248                         prot->twsk_prot->twsk_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
2249
2250                         if (prot->twsk_prot->twsk_slab_name == NULL)
2251                                 goto out_free_request_sock_slab;
2252
2253                         sprintf(prot->twsk_prot->twsk_slab_name, mask, prot->name);
2254                         prot->twsk_prot->twsk_slab =
2255                                 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2256                                                   prot->twsk_prot->twsk_obj_size,
2257                                                   0,
2258                                                   SLAB_HWCACHE_ALIGN |
2259                                                         prot->slab_flags,
2260                                                   NULL);
2261                         if (prot->twsk_prot->twsk_slab == NULL)
2262                                 goto out_free_timewait_sock_slab_name;
2263                 }
2264         }
2265
2266         write_lock(&proto_list_lock);
2267         list_add(&prot->node, &proto_list);
2268         assign_proto_idx(prot);
2269         write_unlock(&proto_list_lock);
2270         return 0;
2271
2272 out_free_timewait_sock_slab_name:
2273         kfree(prot->twsk_prot->twsk_slab_name);
2274 out_free_request_sock_slab:
2275         if (prot->rsk_prot && prot->rsk_prot->slab) {
2276                 kmem_cache_destroy(prot->rsk_prot->slab);
2277                 prot->rsk_prot->slab = NULL;
2278         }
2279 out_free_request_sock_slab_name:
2280         kfree(prot->rsk_prot->slab_name);
2281 out_free_sock_slab:
2282         kmem_cache_destroy(prot->slab);
2283         prot->slab = NULL;
2284 out:
2285         return -ENOBUFS;
2286 }
2287 EXPORT_SYMBOL(proto_register);
2288
2289 void proto_unregister(struct proto *prot)
2290 {
2291         write_lock(&proto_list_lock);
2292         release_proto_idx(prot);
2293         list_del(&prot->node);
2294         write_unlock(&proto_list_lock);
2295
2296         if (prot->slab != NULL) {
2297                 kmem_cache_destroy(prot->slab);
2298                 prot->slab = NULL;
2299         }
2300
2301         if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2302                 kmem_cache_destroy(prot->rsk_prot->slab);
2303                 kfree(prot->rsk_prot->slab_name);
2304                 prot->rsk_prot->slab = NULL;
2305         }
2306
2307         if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2308                 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2309                 kfree(prot->twsk_prot->twsk_slab_name);
2310                 prot->twsk_prot->twsk_slab = NULL;
2311         }
2312 }
2313 EXPORT_SYMBOL(proto_unregister);
2314
2315 #ifdef CONFIG_PROC_FS
2316 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2317         __acquires(proto_list_lock)
2318 {
2319         read_lock(&proto_list_lock);
2320         return seq_list_start_head(&proto_list, *pos);
2321 }
2322
2323 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2324 {
2325         return seq_list_next(v, &proto_list, pos);
2326 }
2327
2328 static void proto_seq_stop(struct seq_file *seq, void *v)
2329         __releases(proto_list_lock)
2330 {
2331         read_unlock(&proto_list_lock);
2332 }
2333
2334 static char proto_method_implemented(const void *method)
2335 {
2336         return method == NULL ? 'n' : 'y';
2337 }
2338
2339 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2340 {
2341         seq_printf(seq, "%-9s %4u %6d  %6d   %-3s %6u   %-3s  %-10s "
2342                         "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2343                    proto->name,
2344                    proto->obj_size,
2345                    sock_prot_inuse_get(seq_file_net(seq), proto),
2346                    proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
2347                    proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2348                    proto->max_header,
2349                    proto->slab == NULL ? "no" : "yes",
2350                    module_name(proto->owner),
2351                    proto_method_implemented(proto->close),
2352                    proto_method_implemented(proto->connect),
2353                    proto_method_implemented(proto->disconnect),
2354                    proto_method_implemented(proto->accept),
2355                    proto_method_implemented(proto->ioctl),
2356                    proto_method_implemented(proto->init),
2357                    proto_method_implemented(proto->destroy),
2358                    proto_method_implemented(proto->shutdown),
2359                    proto_method_implemented(proto->setsockopt),
2360                    proto_method_implemented(proto->getsockopt),
2361                    proto_method_implemented(proto->sendmsg),
2362                    proto_method_implemented(proto->recvmsg),
2363                    proto_method_implemented(proto->sendpage),
2364                    proto_method_implemented(proto->bind),
2365                    proto_method_implemented(proto->backlog_rcv),
2366                    proto_method_implemented(proto->hash),
2367                    proto_method_implemented(proto->unhash),
2368                    proto_method_implemented(proto->get_port),
2369                    proto_method_implemented(proto->enter_memory_pressure));
2370 }
2371
2372 static int proto_seq_show(struct seq_file *seq, void *v)
2373 {
2374         if (v == &proto_list)
2375                 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2376                            "protocol",
2377                            "size",
2378                            "sockets",
2379                            "memory",
2380                            "press",
2381                            "maxhdr",
2382                            "slab",
2383                            "module",
2384                            "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2385         else
2386                 proto_seq_printf(seq, list_entry(v, struct proto, node));
2387         return 0;
2388 }
2389
2390 static const struct seq_operations proto_seq_ops = {
2391         .start  = proto_seq_start,
2392         .next   = proto_seq_next,
2393         .stop   = proto_seq_stop,
2394         .show   = proto_seq_show,
2395 };
2396
2397 static int proto_seq_open(struct inode *inode, struct file *file)
2398 {
2399         return seq_open_net(inode, file, &proto_seq_ops,
2400                             sizeof(struct seq_net_private));
2401 }
2402
2403 static const struct file_operations proto_seq_fops = {
2404         .owner          = THIS_MODULE,
2405         .open           = proto_seq_open,
2406         .read           = seq_read,
2407         .llseek         = seq_lseek,
2408         .release        = seq_release_net,
2409 };
2410
2411 static __net_init int proto_init_net(struct net *net)
2412 {
2413         if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2414                 return -ENOMEM;
2415
2416         return 0;
2417 }
2418
2419 static __net_exit void proto_exit_net(struct net *net)
2420 {
2421         proc_net_remove(net, "protocols");
2422 }
2423
2424
2425 static __net_initdata struct pernet_operations proto_net_ops = {
2426         .init = proto_init_net,
2427         .exit = proto_exit_net,
2428 };
2429
2430 static int __init proto_init(void)
2431 {
2432         return register_pernet_subsys(&proto_net_ops);
2433 }
2434
2435 subsys_initcall(proto_init);
2436
2437 #endif /* PROC_FS */