net/core/sock.c

   1 /*
   2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   3  *              operating system.  INET is implemented using the  BSD Socket
   4  *              interface as the means of communication with the user level.
   5  *
   6  *              Generic socket support routines. Memory allocators, socket lock/release
   7  *              handler for protocols to use and generic option handler.
   8  *
   9  *
  10  * Authors:     Ross Biro
  11  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  12  *              Florian La Roche, <flla@stud.uni-sb.de>
  13  *              Alan Cox, <A.Cox@swansea.ac.uk>
  14  *
  15  * Fixes:
  16  *              Alan Cox        :       Numerous verify_area() problems
  17  *              Alan Cox        :       Connecting on a connecting socket
  18  *                                      now returns an error for tcp.
  19  *              Alan Cox        :       sock->protocol is set correctly.
  20  *                                      and is not sometimes left as 0.
  21  *              Alan Cox        :       connect handles icmp errors on a
  22  *                                      connect properly. Unfortunately there
  23  *                                      is a restart syscall nasty there. I
  24  *                                      can't match BSD without hacking the C
  25  *                                      library. Ideas urgently sought!
  26  *              Alan Cox        :       Disallow bind() to addresses that are
  27  *                                      not ours - especially broadcast ones!!
  28  *              Alan Cox        :       Socket 1024 _IS_ ok for users. (fencepost)
  29  *              Alan Cox        :       sock_wfree/sock_rfree don't destroy sockets,
  30  *                                      instead they leave that for the DESTROY timer.
  31  *              Alan Cox        :       Clean up error flag in accept
  32  *              Alan Cox        :       TCP ack handling is buggy, the DESTROY timer
  33  *                                      was buggy. Put a remove_sock() in the handler
  34  *                                      for memory when we hit 0. Also altered the timer
  35  *                                      code. The ACK stuff can wait and needs major
  36  *                                      TCP layer surgery.
  37  *              Alan Cox        :       Fixed TCP ack bug, removed remove sock
  38  *                                      and fixed timer/inet_bh race.
  39  *              Alan Cox        :       Added zapped flag for TCP
  40  *              Alan Cox        :       Move kfree_skb into skbuff.c and tidied up surplus code
  41  *              Alan Cox        :       for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
  42  *              Alan Cox        :       kfree_s calls now are kfree_skbmem so we can track skb resources
  43  *              Alan Cox        :       Supports socket option broadcast now as does udp. Packet and raw need fixing.
  44  *              Alan Cox        :       Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
  45  *              Rick Sladkey    :       Relaxed UDP rules for matching packets.
  46  *              C.E.Hawkins     :       IFF_PROMISC/SIOCGHWADDR support
  47  *      Pauline Middelink       :       identd support
  48  *              Alan Cox        :       Fixed connect() taking signals I think.
  49  *              Alan Cox        :       SO_LINGER supported
  50  *              Alan Cox        :       Error reporting fixes
  51  *              Anonymous       :       inet_create tidied up (sk->reuse setting)
  52  *              Alan Cox        :       inet sockets don't set sk->type!
  53  *              Alan Cox        :       Split socket option code
  54  *              Alan Cox        :       Callbacks
  55  *              Alan Cox        :       Nagle flag for Charles & Johannes stuff
  56  *              Alex            :       Removed restriction on inet fioctl
  57  *              Alan Cox        :       Splitting INET from NET core
  58  *              Alan Cox        :       Fixed bogus SO_TYPE handling in getsockopt()
  59  *              Adam Caldwell   :       Missing return in SO_DONTROUTE/SO_DEBUG code
  60  *              Alan Cox        :       Split IP from generic code
  61  *              Alan Cox        :       New kfree_skbmem()
  62  *              Alan Cox        :       Make SO_DEBUG superuser only.
  63  *              Alan Cox        :       Allow anyone to clear SO_DEBUG
  64  *                                      (compatibility fix)
  65  *              Alan Cox        :       Added optimistic memory grabbing for AF_UNIX throughput.
  66  *              Alan Cox        :       Allocator for a socket is settable.
  67  *              Alan Cox        :       SO_ERROR includes soft errors.
  68  *              Alan Cox        :       Allow NULL arguments on some SO_ opts
  69  *              Alan Cox        :       Generic socket allocation to make hooks
  70  *                                      easier (suggested by Craig Metz).
  71  *              Michael Pall    :       SO_ERROR returns positive errno again
  72  *              Steve Whitehouse:       Added default destructor to free
  73  *                                      protocol private data.
  74  *              Steve Whitehouse:       Added various other default routines
  75  *                                      common to several socket families.
  76  *              Chris Evans     :       Call suser() check last on F_SETOWN
  77  *              Jay Schulist    :       Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
  78  *              Andi Kleen      :       Add sock_kmalloc()/sock_kfree_s()
  79  *              Andi Kleen      :       Fix write_space callback
  80  *              Chris Evans     :       Security fixes - signedness again
  81  *              Arnaldo C. Melo :       cleanups, use skb_queue_purge
  82  *
  83  * To Fix:
  84  *
  85  *
  86  *              This program is free software; you can redistribute it and/or
  87  *              modify it under the terms of the GNU General Public License
  88  *              as published by the Free Software Foundation; either version
  89  *              2 of the License, or (at your option) any later version.
  90  */
  91
  92 #include <linux/capability.h>
  93 #include <linux/errno.h>
  94 #include <linux/types.h>
  95 #include <linux/socket.h>
  96 #include <linux/in.h>
  97 #include <linux/kernel.h>
  98 #include <linux/module.h>
  99 #include <linux/proc_fs.h>
 100 #include <linux/seq_file.h>
 101 #include <linux/sched.h>
 102 #include <linux/timer.h>
 103 #include <linux/string.h>
 104 #include <linux/sockios.h>
 105 #include <linux/net.h>
 106 #include <linux/mm.h>
 107 #include <linux/slab.h>
 108 #include <linux/interrupt.h>
 109 #include <linux/poll.h>
 110 #include <linux/tcp.h>
 111 #include <linux/init.h>
 112 #include <linux/highmem.h>
 113 #include <linux/user_namespace.h>
 114
 115 #include <asm/uaccess.h>
 116 #include <asm/system.h>
 117
 118 #include <linux/netdevice.h>
 119 #include <net/protocol.h>
 120 #include <linux/skbuff.h>
 121 #include <net/net_namespace.h>
 122 #include <net/request_sock.h>
 123 #include <net/sock.h>
 124 #include <linux/net_tstamp.h>
 125 #include <net/xfrm.h>
 126 #include <linux/ipsec.h>
 127 #include <net/cls_cgroup.h>
 128
 129 #include <linux/filter.h>
 130
 131 #include <trace/events/sock.h>
 132
 133 #ifdef CONFIG_INET
 134 #include <net/tcp.h>
 135 #endif
 136
 137 /*
 138  * Each address family might have different locking rules, so we have
 139  * one slock key per address family:
 140  */
 141 static struct lock_class_key af_family_keys[AF_MAX];
 142 static struct lock_class_key af_family_slock_keys[AF_MAX];
 143
 144 /*
 145  * Make lock validator output more readable. (we pre-construct these
 146  * strings build-time, so that runtime initialization of socket
 147  * locks is fast):
 148  */
 149 static const char *const af_family_key_strings[AF_MAX+1] = {
 150   "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX"     , "sk_lock-AF_INET"     ,
 151   "sk_lock-AF_AX25"  , "sk_lock-AF_IPX"      , "sk_lock-AF_APPLETALK",
 152   "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE"   , "sk_lock-AF_ATMPVC"   ,
 153   "sk_lock-AF_X25"   , "sk_lock-AF_INET6"    , "sk_lock-AF_ROSE"     ,
 154   "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI"  , "sk_lock-AF_SECURITY" ,
 155   "sk_lock-AF_KEY"   , "sk_lock-AF_NETLINK"  , "sk_lock-AF_PACKET"   ,
 156   "sk_lock-AF_ASH"   , "sk_lock-AF_ECONET"   , "sk_lock-AF_ATMSVC"   ,
 157   "sk_lock-AF_RDS"   , "sk_lock-AF_SNA"      , "sk_lock-AF_IRDA"     ,
 158   "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE"  , "sk_lock-AF_LLC"      ,
 159   "sk_lock-27"       , "sk_lock-28"          , "sk_lock-AF_CAN"      ,
 160   "sk_lock-AF_TIPC"  , "sk_lock-AF_BLUETOOTH", "sk_lock-IUCV"        ,
 161   "sk_lock-AF_RXRPC" , "sk_lock-AF_ISDN"     , "sk_lock-AF_PHONET"   ,
 162   "sk_lock-AF_IEEE802154", "sk_lock-AF_CAIF" , "sk_lock-AF_ALG"      ,
 163   "sk_lock-AF_NFC"   , "sk_lock-AF_MAX"
 164 };
 165 static const char *const af_family_slock_key_strings[AF_MAX+1] = {
 166   "slock-AF_UNSPEC", "slock-AF_UNIX"     , "slock-AF_INET"     ,
 167   "slock-AF_AX25"  , "slock-AF_IPX"      , "slock-AF_APPLETALK",
 168   "slock-AF_NETROM", "slock-AF_BRIDGE"   , "slock-AF_ATMPVC"   ,
 169   "slock-AF_X25"   , "slock-AF_INET6"    , "slock-AF_ROSE"     ,
 170   "slock-AF_DECnet", "slock-AF_NETBEUI"  , "slock-AF_SECURITY" ,
 171   "slock-AF_KEY"   , "slock-AF_NETLINK"  , "slock-AF_PACKET"   ,
 172   "slock-AF_ASH"   , "slock-AF_ECONET"   , "slock-AF_ATMSVC"   ,
 173   "slock-AF_RDS"   , "slock-AF_SNA"      , "slock-AF_IRDA"     ,
 174   "slock-AF_PPPOX" , "slock-AF_WANPIPE"  , "slock-AF_LLC"      ,
 175   "slock-27"       , "slock-28"          , "slock-AF_CAN"      ,
 176   "slock-AF_TIPC"  , "slock-AF_BLUETOOTH", "slock-AF_IUCV"     ,
 177   "slock-AF_RXRPC" , "slock-AF_ISDN"     , "slock-AF_PHONET"   ,
 178   "slock-AF_IEEE802154", "slock-AF_CAIF" , "slock-AF_ALG"      ,
 179   "slock-AF_NFC"   , "slock-AF_MAX"
 180 };
 181 static const char *const af_family_clock_key_strings[AF_MAX+1] = {
 182   "clock-AF_UNSPEC", "clock-AF_UNIX"     , "clock-AF_INET"     ,
 183   "clock-AF_AX25"  , "clock-AF_IPX"      , "clock-AF_APPLETALK",
 184   "clock-AF_NETROM", "clock-AF_BRIDGE"   , "clock-AF_ATMPVC"   ,
 185   "clock-AF_X25"   , "clock-AF_INET6"    , "clock-AF_ROSE"     ,
 186   "clock-AF_DECnet", "clock-AF_NETBEUI"  , "clock-AF_SECURITY" ,
 187   "clock-AF_KEY"   , "clock-AF_NETLINK"  , "clock-AF_PACKET"   ,
 188   "clock-AF_ASH"   , "clock-AF_ECONET"   , "clock-AF_ATMSVC"   ,
 189   "clock-AF_RDS"   , "clock-AF_SNA"      , "clock-AF_IRDA"     ,
 190   "clock-AF_PPPOX" , "clock-AF_WANPIPE"  , "clock-AF_LLC"      ,
 191   "clock-27"       , "clock-28"          , "clock-AF_CAN"      ,
 192   "clock-AF_TIPC"  , "clock-AF_BLUETOOTH", "clock-AF_IUCV"     ,
 193   "clock-AF_RXRPC" , "clock-AF_ISDN"     , "clock-AF_PHONET"   ,
 194   "clock-AF_IEEE802154", "clock-AF_CAIF" , "clock-AF_ALG"      ,
 195   "clock-AF_NFC"   , "clock-AF_MAX"
 196 };
 197
 198 /*
 199  * sk_callback_lock locking rules are per-address-family,
 200  * so split the lock classes by using a per-AF key:
 201  */
 202 static struct lock_class_key af_callback_keys[AF_MAX];
 203
 204 /* Take into consideration the size of the struct sk_buff overhead in the
 205  * determination of these values, since that is non-constant across
 206  * platforms.  This makes socket queueing behavior and performance
 207  * not depend upon such differences.
 208  */
 209 #define _SK_MEM_PACKETS         256
 210 #define _SK_MEM_OVERHEAD        SKB_TRUESIZE(256)
 211 #define SK_WMEM_MAX             (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
 212 #define SK_RMEM_MAX             (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
 213
 214 /* Run time adjustable parameters. */
 215 __u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
 216 __u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
 217 __u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
 218 __u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
 219
 220 /* Maximal space eaten by iovec or ancillary data plus some space */
 221 int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
 222 EXPORT_SYMBOL(sysctl_optmem_max);
 223
 224 #if defined(CONFIG_CGROUPS) && !defined(CONFIG_NET_CLS_CGROUP)
 225 int net_cls_subsys_id = -1;
 226 EXPORT_SYMBOL_GPL(net_cls_subsys_id);
 227 #endif
 228
 229 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
 230 {
 231         struct timeval tv;
 232
 233         if (optlen < sizeof(tv))
 234                 return -EINVAL;
 235         if (copy_from_user(&tv, optval, sizeof(tv)))
 236                 return -EFAULT;
 237         if (tv.tv_usec < 0 || tv.tv_usec >= USEC_PER_SEC)
 238                 return -EDOM;
 239
 240         if (tv.tv_sec < 0) {
 241                 static int warned __read_mostly;
 242
 243                 *timeo_p = 0;
 244                 if (warned < 10 && net_ratelimit()) {
 245                         warned++;
 246                         printk(KERN_INFO "sock_set_timeout: `%s' (pid %d) "
 247                                "tries to set negative timeout\n",
 248                                 current->comm, task_pid_nr(current));
 249                 }
 250                 return 0;
 251         }
 252         *timeo_p = MAX_SCHEDULE_TIMEOUT;
 253         if (tv.tv_sec == 0 && tv.tv_usec == 0)
 254                 return 0;
 255         if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
 256                 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
 257         return 0;
 258 }
 259
 260 static void sock_warn_obsolete_bsdism(const char *name)
 261 {
 262         static int warned;
 263         static char warncomm[TASK_COMM_LEN];
 264         if (strcmp(warncomm, current->comm) && warned < 5) {
 265                 strcpy(warncomm,  current->comm);
 266                 printk(KERN_WARNING "process `%s' is using obsolete "
 267                        "%s SO_BSDCOMPAT\n", warncomm, name);
 268                 warned++;
 269         }
 270 }
 271
 272 static void sock_disable_timestamp(struct sock *sk, int flag)
 273 {
 274         if (sock_flag(sk, flag)) {
 275                 sock_reset_flag(sk, flag);
 276                 if (!sock_flag(sk, SOCK_TIMESTAMP) &&
 277                     !sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE)) {
 278                         net_disable_timestamp();
 279                 }
 280         }
 281 }
 282
 283
 284 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
 285 {
 286         int err;
 287         int skb_len;
 288         unsigned long flags;
 289         struct sk_buff_head *list = &sk->sk_receive_queue;
 290
 291         if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
 292                 atomic_inc(&sk->sk_drops);
 293                 trace_sock_rcvqueue_full(sk, skb);
 294                 return -ENOMEM;
 295         }
 296
 297         err = sk_filter(sk, skb);
 298         if (err)
 299                 return err;
 300
 301         if (!sk_rmem_schedule(sk, skb->truesize)) {
 302                 atomic_inc(&sk->sk_drops);
 303                 return -ENOBUFS;
 304         }
 305
 306         skb->dev = NULL;
 307         skb_set_owner_r(skb, sk);
 308
 309         /* Cache the SKB length before we tack it onto the receive
 310          * queue.  Once it is added it no longer belongs to us and
 311          * may be freed by other threads of control pulling packets
 312          * from the queue.
 313          */
 314         skb_len = skb->len;
 315
 316         /* we escape from rcu protected region, make sure we dont leak
 317          * a norefcounted dst
 318          */
 319         skb_dst_force(skb);
 320
 321         spin_lock_irqsave(&list->lock, flags);
 322         skb->dropcount = atomic_read(&sk->sk_drops);
 323         __skb_queue_tail(list, skb);
 324         spin_unlock_irqrestore(&list->lock, flags);
 325
 326         if (!sock_flag(sk, SOCK_DEAD))
 327                 sk->sk_data_ready(sk, skb_len);
 328         return 0;
 329 }
 330 EXPORT_SYMBOL(sock_queue_rcv_skb);
 331
 332 int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
 333 {
 334         int rc = NET_RX_SUCCESS;
 335
 336         if (sk_filter(sk, skb))
 337                 goto discard_and_relse;
 338
 339         skb->dev = NULL;
 340
 341         if (sk_rcvqueues_full(sk, skb)) {
 342                 atomic_inc(&sk->sk_drops);
 343                 goto discard_and_relse;
 344         }
 345         if (nested)
 346                 bh_lock_sock_nested(sk);
 347         else
 348                 bh_lock_sock(sk);
 349         if (!sock_owned_by_user(sk)) {
 350                 /*
 351                  * trylock + unlock semantics:
 352                  */
 353                 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
 354
 355                 rc = sk_backlog_rcv(sk, skb);
 356
 357                 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
 358         } else if (sk_add_backlog(sk, skb)) {
 359                 bh_unlock_sock(sk);
 360                 atomic_inc(&sk->sk_drops);
 361                 goto discard_and_relse;
 362         }
 363
 364         bh_unlock_sock(sk);
 365 out:
 366         sock_put(sk);
 367         return rc;
 368 discard_and_relse:
 369         kfree_skb(skb);
 370         goto out;
 371 }
 372 EXPORT_SYMBOL(sk_receive_skb);
 373
 374 void sk_reset_txq(struct sock *sk)
 375 {
 376         sk_tx_queue_clear(sk);
 377 }
 378 EXPORT_SYMBOL(sk_reset_txq);
 379
 380 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
 381 {
 382         struct dst_entry *dst = __sk_dst_get(sk);
 383
 384         if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
 385                 sk_tx_queue_clear(sk);
 386                 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
 387                 dst_release(dst);
 388                 return NULL;
 389         }
 390
 391         return dst;
 392 }
 393 EXPORT_SYMBOL(__sk_dst_check);
 394
 395 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
 396 {
 397         struct dst_entry *dst = sk_dst_get(sk);
 398
 399         if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
 400                 sk_dst_reset(sk);
 401                 dst_release(dst);
 402                 return NULL;
 403         }
 404
 405         return dst;
 406 }
 407 EXPORT_SYMBOL(sk_dst_check);
 408
 409 static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
 410 {
 411         int ret = -ENOPROTOOPT;
 412 #ifdef CONFIG_NETDEVICES
 413         struct net *net = sock_net(sk);
 414         char devname[IFNAMSIZ];
 415         int index;
 416
 417         /* Sorry... */
 418         ret = -EPERM;
 419         if (!capable(CAP_NET_RAW))
 420                 goto out;
 421
 422         ret = -EINVAL;
 423         if (optlen < 0)
 424                 goto out;
 425
 426         /* Bind this socket to a particular device like "eth0",
 427          * as specified in the passed interface name. If the
 428          * name is "" or the option length is zero the socket
 429          * is not bound.
 430          */
 431         if (optlen > IFNAMSIZ - 1)
 432                 optlen = IFNAMSIZ - 1;
 433         memset(devname, 0, sizeof(devname));
 434
 435         ret = -EFAULT;
 436         if (copy_from_user(devname, optval, optlen))
 437                 goto out;
 438
 439         index = 0;
 440         if (devname[0] != '\0') {
 441                 struct net_device *dev;
 442
 443                 rcu_read_lock();
 444                 dev = dev_get_by_name_rcu(net, devname);
 445                 if (dev)
 446                         index = dev->ifindex;
 447                 rcu_read_unlock();
 448                 ret = -ENODEV;
 449                 if (!dev)
 450                         goto out;
 451         }
 452
 453         lock_sock(sk);
 454         sk->sk_bound_dev_if = index;
 455         sk_dst_reset(sk);
 456         release_sock(sk);
 457
 458         ret = 0;
 459
 460 out:
 461 #endif
 462
 463         return ret;
 464 }
 465
 466 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
 467 {
 468         if (valbool)
 469                 sock_set_flag(sk, bit);
 470         else
 471                 sock_reset_flag(sk, bit);
 472 }
 473
 474 /*
 475  *      This is meant for all protocols to use and covers goings on
 476  *      at the socket level. Everything here is generic.
 477  */
 478
 479 int sock_setsockopt(struct socket *sock, int level, int optname,
 480                     char __user *optval, unsigned int optlen)
 481 {
 482         struct sock *sk = sock->sk;
 483         int val;
 484         int valbool;
 485         struct linger ling;
 486         int ret = 0;
 487
 488         /*
 489          *      Options without arguments
 490          */
 491
 492         if (optname == SO_BINDTODEVICE)
 493                 return sock_bindtodevice(sk, optval, optlen);
 494
 495         if (optlen < sizeof(int))
 496                 return -EINVAL;
 497
 498         if (get_user(val, (int __user *)optval))
 499                 return -EFAULT;
 500
 501         valbool = val ? 1 : 0;
 502
 503         lock_sock(sk);
 504
 505         switch (optname) {
 506         case SO_DEBUG:
 507                 if (val && !capable(CAP_NET_ADMIN))
 508                         ret = -EACCES;
 509                 else
 510                         sock_valbool_flag(sk, SOCK_DBG, valbool);
 511                 break;
 512         case SO_REUSEADDR:
 513                 sk->sk_reuse = valbool;
 514                 break;
 515         case SO_TYPE:
 516         case SO_PROTOCOL:
 517         case SO_DOMAIN:
 518         case SO_ERROR:
 519                 ret = -ENOPROTOOPT;
 520                 break;
 521         case SO_DONTROUTE:
 522                 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
 523                 break;
 524         case SO_BROADCAST:
 525                 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
 526                 break;
 527         case SO_SNDBUF:
 528                 /* Don't error on this BSD doesn't and if you think
 529                  * about it this is right. Otherwise apps have to
 530                  * play 'guess the biggest size' games. RCVBUF/SNDBUF
 531                  * are treated in BSD as hints
 532                  */
 533                 val = min_t(u32, val, sysctl_wmem_max);
 534 set_sndbuf:
 535                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
 536                 sk->sk_sndbuf = max_t(u32, val * 2, SOCK_MIN_SNDBUF);
 537                 /* Wake up sending tasks if we upped the value. */
 538                 sk->sk_write_space(sk);
 539                 break;
 540
 541         case SO_SNDBUFFORCE:
 542                 if (!capable(CAP_NET_ADMIN)) {
 543                         ret = -EPERM;
 544                         break;
 545                 }
 546                 goto set_sndbuf;
 547
 548         case SO_RCVBUF:
 549                 /* Don't error on this BSD doesn't and if you think
 550                  * about it this is right. Otherwise apps have to
 551                  * play 'guess the biggest size' games. RCVBUF/SNDBUF
 552                  * are treated in BSD as hints
 553                  */
 554                 val = min_t(u32, val, sysctl_rmem_max);
 555 set_rcvbuf:
 556                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
 557                 /*
 558                  * We double it on the way in to account for
 559                  * "struct sk_buff" etc. overhead.   Applications
 560                  * assume that the SO_RCVBUF setting they make will
 561                  * allow that much actual data to be received on that
 562                  * socket.
 563                  *
 564                  * Applications are unaware that "struct sk_buff" and
 565                  * other overheads allocate from the receive buffer
 566                  * during socket buffer allocation.
 567                  *
 568                  * And after considering the possible alternatives,
 569                  * returning the value we actually used in getsockopt
 570                  * is the most desirable behavior.
 571                  */
 572                 sk->sk_rcvbuf = max_t(u32, val * 2, SOCK_MIN_RCVBUF);
 573                 break;
 574
 575         case SO_RCVBUFFORCE:
 576                 if (!capable(CAP_NET_ADMIN)) {
 577                         ret = -EPERM;
 578                         break;
 579                 }
 580                 goto set_rcvbuf;
 581
 582         case SO_KEEPALIVE:
 583 #ifdef CONFIG_INET
 584                 if (sk->sk_protocol == IPPROTO_TCP &&
 585                     sk->sk_type == SOCK_STREAM)
 586                         tcp_set_keepalive(sk, valbool);
 587 #endif
 588                 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
 589                 break;
 590
 591         case SO_OOBINLINE:
 592                 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
 593                 break;
 594
 595         case SO_NO_CHECK:
 596                 sk->sk_no_check = valbool;
 597                 break;
 598
 599         case SO_PRIORITY:
 600                 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
 601                         sk->sk_priority = val;
 602                 else
 603                         ret = -EPERM;
 604                 break;
 605
 606         case SO_LINGER:
 607                 if (optlen < sizeof(ling)) {
 608                         ret = -EINVAL;  /* 1003.1g */
 609                         break;
 610                 }
 611                 if (copy_from_user(&ling, optval, sizeof(ling))) {
 612                         ret = -EFAULT;
 613                         break;
 614                 }
 615                 if (!ling.l_onoff)
 616                         sock_reset_flag(sk, SOCK_LINGER);
 617                 else {
 618 #if (BITS_PER_LONG == 32)
 619                         if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
 620                                 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
 621                         else
 622 #endif
 623                                 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
 624                         sock_set_flag(sk, SOCK_LINGER);
 625                 }
 626                 break;
 627
 628         case SO_BSDCOMPAT:
 629                 sock_warn_obsolete_bsdism("setsockopt");
 630                 break;
 631
 632         case SO_PASSCRED:
 633                 if (valbool)
 634                         set_bit(SOCK_PASSCRED, &sock->flags);
 635                 else
 636                         clear_bit(SOCK_PASSCRED, &sock->flags);
 637                 break;
 638
 639         case SO_TIMESTAMP:
 640         case SO_TIMESTAMPNS:
 641                 if (valbool)  {
 642                         if (optname == SO_TIMESTAMP)
 643                                 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
 644                         else
 645                                 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
 646                         sock_set_flag(sk, SOCK_RCVTSTAMP);
 647                         sock_enable_timestamp(sk, SOCK_TIMESTAMP);
 648                 } else {
 649                         sock_reset_flag(sk, SOCK_RCVTSTAMP);
 650                         sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
 651                 }
 652                 break;
 653
 654         case SO_TIMESTAMPING:
 655                 if (val & ~SOF_TIMESTAMPING_MASK) {
 656                         ret = -EINVAL;
 657                         break;
 658                 }
 659                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
 660                                   val & SOF_TIMESTAMPING_TX_HARDWARE);
 661                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
 662                                   val & SOF_TIMESTAMPING_TX_SOFTWARE);
 663                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
 664                                   val & SOF_TIMESTAMPING_RX_HARDWARE);
 665                 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
 666                         sock_enable_timestamp(sk,
 667                                               SOCK_TIMESTAMPING_RX_SOFTWARE);
 668                 else
 669                         sock_disable_timestamp(sk,
 670                                                SOCK_TIMESTAMPING_RX_SOFTWARE);
 671                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
 672                                   val & SOF_TIMESTAMPING_SOFTWARE);
 673                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
 674                                   val & SOF_TIMESTAMPING_SYS_HARDWARE);
 675                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
 676                                   val & SOF_TIMESTAMPING_RAW_HARDWARE);
 677                 break;
 678
 679         case SO_RCVLOWAT:
 680                 if (val < 0)
 681                         val = INT_MAX;
 682                 sk->sk_rcvlowat = val ? : 1;
 683                 break;
 684
 685         case SO_RCVTIMEO:
 686                 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
 687                 break;
 688
 689         case SO_SNDTIMEO:
 690                 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
 691                 break;
 692
 693         case SO_ATTACH_FILTER:
 694                 ret = -EINVAL;
 695                 if (optlen == sizeof(struct sock_fprog)) {
 696                         struct sock_fprog fprog;
 697
 698                         ret = -EFAULT;
 699                         if (copy_from_user(&fprog, optval, sizeof(fprog)))
 700                                 break;
 701
 702                         ret = sk_attach_filter(&fprog, sk);
 703                 }
 704                 break;
 705
 706         case SO_DETACH_FILTER:
 707                 ret = sk_detach_filter(sk);
 708                 break;
 709
 710         case SO_PASSSEC:
 711                 if (valbool)
 712                         set_bit(SOCK_PASSSEC, &sock->flags);
 713                 else
 714                         clear_bit(SOCK_PASSSEC, &sock->flags);
 715                 break;
 716         case SO_MARK:
 717                 if (!capable(CAP_NET_ADMIN))
 718                         ret = -EPERM;
 719                 else
 720                         sk->sk_mark = val;
 721                 break;
 722
 723                 /* We implement the SO_SNDLOWAT etc to
 724                    not be settable (1003.1g 5.3) */
 725         case SO_RXQ_OVFL:
 726                 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
 727                 break;
 728         default:
 729                 ret = -ENOPROTOOPT;
 730                 break;
 731         }
 732         release_sock(sk);
 733         return ret;
 734 }
 735 EXPORT_SYMBOL(sock_setsockopt);
 736
 737
 738 void cred_to_ucred(struct pid *pid, const struct cred *cred,
 739                    struct ucred *ucred)
 740 {
 741         ucred->pid = pid_vnr(pid);
 742         ucred->uid = ucred->gid = -1;
 743         if (cred) {
 744                 struct user_namespace *current_ns = current_user_ns();
 745
 746                 ucred->uid = user_ns_map_uid(current_ns, cred, cred->euid);
 747                 ucred->gid = user_ns_map_gid(current_ns, cred, cred->egid);
 748         }
 749 }
 750 EXPORT_SYMBOL_GPL(cred_to_ucred);
 751
 752 void cred_real_to_ucred(struct pid *pid, const struct cred *cred,
 753                         struct ucred *ucred)
 754 {
 755         ucred->pid = pid_vnr(pid);
 756         ucred->uid = ucred->gid = -1;
 757         if (cred) {
 758                 struct user_namespace *current_ns = current_user_ns();
 759
 760                 ucred->uid = user_ns_map_uid(current_ns, cred, cred->uid);
 761                 ucred->gid = user_ns_map_gid(current_ns, cred, cred->gid);
 762         }
 763 }
 764 EXPORT_SYMBOL_GPL(cred_real_to_ucred);
 765
 766 int sock_getsockopt(struct socket *sock, int level, int optname,
 767                     char __user *optval, int __user *optlen)
 768 {
 769         struct sock *sk = sock->sk;
 770
 771         union {
 772                 int val;
 773                 struct linger ling;
 774                 struct timeval tm;
 775         } v;
 776
 777         int lv = sizeof(int);
 778         int len;
 779
 780         if (get_user(len, optlen))
 781                 return -EFAULT;
 782         if (len < 0)
 783                 return -EINVAL;
 784
 785         memset(&v, 0, sizeof(v));
 786
 787         switch (optname) {
 788         case SO_DEBUG:
 789                 v.val = sock_flag(sk, SOCK_DBG);
 790                 break;
 791
 792         case SO_DONTROUTE:
 793                 v.val = sock_flag(sk, SOCK_LOCALROUTE);
 794                 break;
 795
 796         case SO_BROADCAST:
 797                 v.val = !!sock_flag(sk, SOCK_BROADCAST);
 798                 break;
 799
 800         case SO_SNDBUF:
 801                 v.val = sk->sk_sndbuf;
 802                 break;
 803
 804         case SO_RCVBUF:
 805                 v.val = sk->sk_rcvbuf;
 806                 break;
 807
 808         case SO_REUSEADDR:
 809                 v.val = sk->sk_reuse;
 810                 break;
 811
 812         case SO_KEEPALIVE:
 813                 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
 814                 break;
 815
 816         case SO_TYPE:
 817                 v.val = sk->sk_type;
 818                 break;
 819
 820         case SO_PROTOCOL:
 821                 v.val = sk->sk_protocol;
 822                 break;
 823
 824         case SO_DOMAIN:
 825                 v.val = sk->sk_family;
 826                 break;
 827
 828         case SO_ERROR:
 829                 v.val = -sock_error(sk);
 830                 if (v.val == 0)
 831                         v.val = xchg(&sk->sk_err_soft, 0);
 832                 break;
 833
 834         case SO_OOBINLINE:
 835                 v.val = !!sock_flag(sk, SOCK_URGINLINE);
 836                 break;
 837
 838         case SO_NO_CHECK:
 839                 v.val = sk->sk_no_check;
 840                 break;
 841
 842         case SO_PRIORITY:
 843                 v.val = sk->sk_priority;
 844                 break;
 845
 846         case SO_LINGER:
 847                 lv              = sizeof(v.ling);
 848                 v.ling.l_onoff  = !!sock_flag(sk, SOCK_LINGER);
 849                 v.ling.l_linger = sk->sk_lingertime / HZ;
 850                 break;
 851
 852         case SO_BSDCOMPAT:
 853                 sock_warn_obsolete_bsdism("getsockopt");
 854                 break;
 855
 856         case SO_TIMESTAMP:
 857                 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
 858                                 !sock_flag(sk, SOCK_RCVTSTAMPNS);
 859                 break;
 860
 861         case SO_TIMESTAMPNS:
 862                 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
 863                 break;
 864
 865         case SO_TIMESTAMPING:
 866                 v.val = 0;
 867                 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
 868                         v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
 869                 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
 870                         v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
 871                 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
 872                         v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
 873                 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
 874                         v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
 875                 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
 876                         v.val |= SOF_TIMESTAMPING_SOFTWARE;
 877                 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
 878                         v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
 879                 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
 880                         v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
 881                 break;
 882
 883         case SO_RCVTIMEO:
 884                 lv = sizeof(struct timeval);
 885                 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
 886                         v.tm.tv_sec = 0;
 887                         v.tm.tv_usec = 0;
 888                 } else {
 889                         v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
 890                         v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
 891                 }
 892                 break;
 893
 894         case SO_SNDTIMEO:
 895                 lv = sizeof(struct timeval);
 896                 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
 897                         v.tm.tv_sec = 0;
 898                         v.tm.tv_usec = 0;
 899                 } else {
 900                         v.tm.tv_sec = sk->sk_sndtimeo / HZ;
 901                         v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
 902                 }
 903                 break;
 904
 905         case SO_RCVLOWAT:
 906                 v.val = sk->sk_rcvlowat;
 907                 break;
 908
 909         case SO_SNDLOWAT:
 910                 v.val = 1;
 911                 break;
 912
 913         case SO_PASSCRED:
 914                 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
 915                 break;
 916
 917         case SO_PEERCRED:
 918         {
 919                 struct ucred peercred;
 920                 if (len > sizeof(peercred))
 921                         len = sizeof(peercred);
 922                 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
 923                 if (copy_to_user(optval, &peercred, len))
 924                         return -EFAULT;
 925                 goto lenout;
 926         }
 927
 928         case SO_PEERNAME:
 929         {
 930                 char address[128];
 931
 932                 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
 933                         return -ENOTCONN;
 934                 if (lv < len)
 935                         return -EINVAL;
 936                 if (copy_to_user(optval, address, len))
 937                         return -EFAULT;
 938                 goto lenout;
 939         }
 940
 941         /* Dubious BSD thing... Probably nobody even uses it, but
 942          * the UNIX standard wants it for whatever reason... -DaveM
 943          */
 944         case SO_ACCEPTCONN:
 945                 v.val = sk->sk_state == TCP_LISTEN;
 946                 break;
 947
 948         case SO_PASSSEC:
 949                 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
 950                 break;
 951
 952         case SO_PEERSEC:
 953                 return security_socket_getpeersec_stream(sock, optval, optlen, len);
 954
 955         case SO_MARK:
 956                 v.val = sk->sk_mark;
 957                 break;
 958
 959         case SO_RXQ_OVFL:
 960                 v.val = !!sock_flag(sk, SOCK_RXQ_OVFL);
 961                 break;
 962
 963         default:
 964                 return -ENOPROTOOPT;
 965         }
 966
 967         if (len > lv)
 968                 len = lv;
 969         if (copy_to_user(optval, &v, len))
 970                 return -EFAULT;
 971 lenout:
 972         if (put_user(len, optlen))
 973                 return -EFAULT;
 974         return 0;
 975 }
 976
 977 /*
 978  * Initialize an sk_lock.
 979  *
 980  * (We also register the sk_lock with the lock validator.)
 981  */
 982 static inline void sock_lock_init(struct sock *sk)
 983 {
 984         sock_lock_init_class_and_name(sk,
 985                         af_family_slock_key_strings[sk->sk_family],
 986                         af_family_slock_keys + sk->sk_family,
 987                         af_family_key_strings[sk->sk_family],
 988                         af_family_keys + sk->sk_family);
 989 }
 990
 991 /*
 992  * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
 993  * even temporarly, because of RCU lookups. sk_node should also be left as is.
 994  * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
 995  */
 996 static void sock_copy(struct sock *nsk, const struct sock *osk)
 997 {
 998 #ifdef CONFIG_SECURITY_NETWORK
 999         void *sptr = nsk->sk_security;
1000 #endif
1001         memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1002
1003         memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1004                osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1005
1006 #ifdef CONFIG_SECURITY_NETWORK
1007         nsk->sk_security = sptr;
1008         security_sk_clone(osk, nsk);
1009 #endif
1010 }
1011
1012 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1013 {
1014         unsigned long nulls1, nulls2;
1015
1016         nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1017         nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1018         if (nulls1 > nulls2)
1019                 swap(nulls1, nulls2);
1020
1021         if (nulls1 != 0)
1022                 memset((char *)sk, 0, nulls1);
1023         memset((char *)sk + nulls1 + sizeof(void *), 0,
1024                nulls2 - nulls1 - sizeof(void *));
1025         memset((char *)sk + nulls2 + sizeof(void *), 0,
1026                size - nulls2 - sizeof(void *));
1027 }
1028 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1029
1030 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1031                 int family)
1032 {
1033         struct sock *sk;
1034         struct kmem_cache *slab;
1035
1036         slab = prot->slab;
1037         if (slab != NULL) {
1038                 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1039                 if (!sk)
1040                         return sk;
1041                 if (priority & __GFP_ZERO) {
1042                         if (prot->clear_sk)
1043                                 prot->clear_sk(sk, prot->obj_size);
1044                         else
1045                                 sk_prot_clear_nulls(sk, prot->obj_size);
1046                 }
1047         } else
1048                 sk = kmalloc(prot->obj_size, priority);
1049
1050         if (sk != NULL) {
1051                 kmemcheck_annotate_bitfield(sk, flags);
1052
1053                 if (security_sk_alloc(sk, family, priority))
1054                         goto out_free;
1055
1056                 if (!try_module_get(prot->owner))
1057                         goto out_free_sec;
1058                 sk_tx_queue_clear(sk);
1059         }
1060
1061         return sk;
1062
1063 out_free_sec:
1064         security_sk_free(sk);
1065 out_free:
1066         if (slab != NULL)
1067                 kmem_cache_free(slab, sk);
1068         else
1069                 kfree(sk);
1070         return NULL;
1071 }
1072
1073 static void sk_prot_free(struct proto *prot, struct sock *sk)
1074 {
1075         struct kmem_cache *slab;
1076         struct module *owner;
1077
1078         owner = prot->owner;
1079         slab = prot->slab;
1080
1081         security_sk_free(sk);
1082         if (slab != NULL)
1083                 kmem_cache_free(slab, sk);
1084         else
1085                 kfree(sk);
1086         module_put(owner);
1087 }
1088
1089 #ifdef CONFIG_CGROUPS
1090 void sock_update_classid(struct sock *sk)
1091 {
1092         u32 classid;
1093
1094         rcu_read_lock();  /* doing current task, which cannot vanish. */
1095         classid = task_cls_classid(current);
1096         rcu_read_unlock();
1097         if (classid && classid != sk->sk_classid)
1098                 sk->sk_classid = classid;
1099 }
1100 EXPORT_SYMBOL(sock_update_classid);
1101 #endif
1102
1103 /**
1104  *      sk_alloc - All socket objects are allocated here
1105  *      @net: the applicable net namespace
1106  *      @family: protocol family
1107  *      @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1108  *      @prot: struct proto associated with this new sock instance
1109  */
1110 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1111                       struct proto *prot)
1112 {
1113         struct sock *sk;
1114
1115         sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1116         if (sk) {
1117                 sk->sk_family = family;
1118                 /*
1119                  * See comment in struct sock definition to understand
1120                  * why we need sk_prot_creator -acme
1121                  */
1122                 sk->sk_prot = sk->sk_prot_creator = prot;
1123                 sock_lock_init(sk);
1124                 sock_net_set(sk, get_net(net));
1125                 atomic_set(&sk->sk_wmem_alloc, 1);
1126
1127                 sock_update_classid(sk);
1128         }
1129
1130         return sk;
1131 }
1132 EXPORT_SYMBOL(sk_alloc);
1133
1134 static void __sk_free(struct sock *sk)
1135 {
1136         struct sk_filter *filter;
1137
1138         if (sk->sk_destruct)
1139                 sk->sk_destruct(sk);
1140
1141         filter = rcu_dereference_check(sk->sk_filter,
1142                                        atomic_read(&sk->sk_wmem_alloc) == 0);
1143         if (filter) {
1144                 sk_filter_uncharge(sk, filter);
1145                 RCU_INIT_POINTER(sk->sk_filter, NULL);
1146         }
1147
1148         sock_disable_timestamp(sk, SOCK_TIMESTAMP);
1149         sock_disable_timestamp(sk, SOCK_TIMESTAMPING_RX_SOFTWARE);
1150
1151         if (atomic_read(&sk->sk_omem_alloc))
1152                 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
1153                        __func__, atomic_read(&sk->sk_omem_alloc));
1154
1155         if (sk->sk_peer_cred)
1156                 put_cred(sk->sk_peer_cred);
1157         put_pid(sk->sk_peer_pid);
1158         put_net(sock_net(sk));
1159         sk_prot_free(sk->sk_prot_creator, sk);
1160 }
1161
1162 void sk_free(struct sock *sk)
1163 {
1164         /*
1165          * We subtract one from sk_wmem_alloc and can know if
1166          * some packets are still in some tx queue.
1167          * If not null, sock_wfree() will call __sk_free(sk) later
1168          */
1169         if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1170                 __sk_free(sk);
1171 }
1172 EXPORT_SYMBOL(sk_free);
1173
1174 /*
1175  * Last sock_put should drop reference to sk->sk_net. It has already
1176  * been dropped in sk_change_net. Taking reference to stopping namespace
1177  * is not an option.
1178  * Take reference to a socket to remove it from hash _alive_ and after that
1179  * destroy it in the context of init_net.
1180  */
1181 void sk_release_kernel(struct sock *sk)
1182 {
1183         if (sk == NULL || sk->sk_socket == NULL)
1184                 return;
1185
1186         sock_hold(sk);
1187         sock_release(sk->sk_socket);
1188         release_net(sock_net(sk));
1189         sock_net_set(sk, get_net(&init_net));
1190         sock_put(sk);
1191 }
1192 EXPORT_SYMBOL(sk_release_kernel);
1193
1194 struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
1195 {
1196         struct sock *newsk;
1197
1198         newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1199         if (newsk != NULL) {
1200                 struct sk_filter *filter;
1201
1202                 sock_copy(newsk, sk);
1203
1204                 /* SANITY */
1205                 get_net(sock_net(newsk));
1206                 sk_node_init(&newsk->sk_node);
1207                 sock_lock_init(newsk);
1208                 bh_lock_sock(newsk);
1209                 newsk->sk_backlog.head  = newsk->sk_backlog.tail = NULL;
1210                 newsk->sk_backlog.len = 0;
1211
1212                 atomic_set(&newsk->sk_rmem_alloc, 0);
1213                 /*
1214                  * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1215                  */
1216                 atomic_set(&newsk->sk_wmem_alloc, 1);
1217                 atomic_set(&newsk->sk_omem_alloc, 0);
1218                 skb_queue_head_init(&newsk->sk_receive_queue);
1219                 skb_queue_head_init(&newsk->sk_write_queue);
1220 #ifdef CONFIG_NET_DMA
1221                 skb_queue_head_init(&newsk->sk_async_wait_queue);
1222 #endif
1223
1224                 spin_lock_init(&newsk->sk_dst_lock);
1225                 rwlock_init(&newsk->sk_callback_lock);
1226                 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1227                                 af_callback_keys + newsk->sk_family,
1228                                 af_family_clock_key_strings[newsk->sk_family]);
1229
1230                 newsk->sk_dst_cache     = NULL;
1231                 newsk->sk_wmem_queued   = 0;
1232                 newsk->sk_forward_alloc = 0;
1233                 newsk->sk_send_head     = NULL;
1234                 newsk->sk_userlocks     = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1235
1236                 sock_reset_flag(newsk, SOCK_DONE);
1237                 skb_queue_head_init(&newsk->sk_error_queue);
1238
1239                 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1240                 if (filter != NULL)
1241                         sk_filter_charge(newsk, filter);
1242
1243                 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1244                         /* It is still raw copy of parent, so invalidate
1245                          * destructor and make plain sk_free() */
1246                         newsk->sk_destruct = NULL;
1247                         bh_unlock_sock(newsk);
1248                         sk_free(newsk);
1249                         newsk = NULL;
1250                         goto out;
1251                 }
1252
1253                 newsk->sk_err      = 0;
1254                 newsk->sk_priority = 0;
1255                 /*
1256                  * Before updating sk_refcnt, we must commit prior changes to memory
1257                  * (Documentation/RCU/rculist_nulls.txt for details)
1258                  */
1259                 smp_wmb();
1260                 atomic_set(&newsk->sk_refcnt, 2);
1261
1262                 /*
1263                  * Increment the counter in the same struct proto as the master
1264                  * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1265                  * is the same as sk->sk_prot->socks, as this field was copied
1266                  * with memcpy).
1267                  *
1268                  * This _changes_ the previous behaviour, where
1269                  * tcp_create_openreq_child always was incrementing the
1270                  * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1271                  * to be taken into account in all callers. -acme
1272                  */
1273                 sk_refcnt_debug_inc(newsk);
1274                 sk_set_socket(newsk, NULL);
1275                 newsk->sk_wq = NULL;
1276
1277                 if (newsk->sk_prot->sockets_allocated)
1278                         percpu_counter_inc(newsk->sk_prot->sockets_allocated);
1279
1280                 if (sock_flag(newsk, SOCK_TIMESTAMP) ||
1281                     sock_flag(newsk, SOCK_TIMESTAMPING_RX_SOFTWARE))
1282                         net_enable_timestamp();
1283         }
1284 out:
1285         return newsk;
1286 }
1287 EXPORT_SYMBOL_GPL(sk_clone);
1288
1289 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1290 {
1291         __sk_dst_set(sk, dst);
1292         sk->sk_route_caps = dst->dev->features;
1293         if (sk->sk_route_caps & NETIF_F_GSO)
1294                 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1295         sk->sk_route_caps &= ~sk->sk_route_nocaps;
1296         if (sk_can_gso(sk)) {
1297                 if (dst->header_len) {
1298                         sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1299                 } else {
1300                         sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1301                         sk->sk_gso_max_size = dst->dev->gso_max_size;
1302                 }
1303         }
1304 }
1305 EXPORT_SYMBOL_GPL(sk_setup_caps);
1306
1307 void __init sk_init(void)
1308 {
1309         if (totalram_pages <= 4096) {
1310                 sysctl_wmem_max = 32767;
1311                 sysctl_rmem_max = 32767;
1312                 sysctl_wmem_default = 32767;
1313                 sysctl_rmem_default = 32767;
1314         } else if (totalram_pages >= 131072) {
1315                 sysctl_wmem_max = 131071;
1316                 sysctl_rmem_max = 131071;
1317         }
1318 }
1319
1320 /*
1321  *      Simple resource managers for sockets.
1322  */
1323
1324
1325 /*
1326  * Write buffer destructor automatically called from kfree_skb.
1327  */
1328 void sock_wfree(struct sk_buff *skb)
1329 {
1330         struct sock *sk = skb->sk;
1331         unsigned int len = skb->truesize;
1332
1333         if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1334                 /*
1335                  * Keep a reference on sk_wmem_alloc, this will be released
1336                  * after sk_write_space() call
1337                  */
1338                 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1339                 sk->sk_write_space(sk);
1340                 len = 1;
1341         }
1342         /*
1343          * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1344          * could not do because of in-flight packets
1345          */
1346         if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1347                 __sk_free(sk);
1348 }
1349 EXPORT_SYMBOL(sock_wfree);
1350
1351 /*
1352  * Read buffer destructor automatically called from kfree_skb.
1353  */
1354 void sock_rfree(struct sk_buff *skb)
1355 {
1356         struct sock *sk = skb->sk;
1357         unsigned int len = skb->truesize;
1358
1359         atomic_sub(len, &sk->sk_rmem_alloc);
1360         sk_mem_uncharge(sk, len);
1361 }
1362 EXPORT_SYMBOL(sock_rfree);
1363
1364
1365 int sock_i_uid(struct sock *sk)
1366 {
1367         int uid;
1368
1369         read_lock_bh(&sk->sk_callback_lock);
1370         uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1371         read_unlock_bh(&sk->sk_callback_lock);
1372         return uid;
1373 }
1374 EXPORT_SYMBOL(sock_i_uid);
1375
1376 unsigned long sock_i_ino(struct sock *sk)
1377 {
1378         unsigned long ino;
1379
1380         read_lock_bh(&sk->sk_callback_lock);
1381         ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1382         read_unlock_bh(&sk->sk_callback_lock);
1383         return ino;
1384 }
1385 EXPORT_SYMBOL(sock_i_ino);
1386
1387 /*
1388  * Allocate a skb from the socket's send buffer.
1389  */
1390 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1391                              gfp_t priority)
1392 {
1393         if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1394                 struct sk_buff *skb = alloc_skb(size, priority);
1395                 if (skb) {
1396                         skb_set_owner_w(skb, sk);
1397                         return skb;
1398                 }
1399         }
1400         return NULL;
1401 }
1402 EXPORT_SYMBOL(sock_wmalloc);
1403
1404 /*
1405  * Allocate a skb from the socket's receive buffer.
1406  */
1407 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1408                              gfp_t priority)
1409 {
1410         if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1411                 struct sk_buff *skb = alloc_skb(size, priority);
1412                 if (skb) {
1413                         skb_set_owner_r(skb, sk);
1414                         return skb;
1415                 }
1416         }
1417         return NULL;
1418 }
1419
1420 /*
1421  * Allocate a memory block from the socket's option memory buffer.
1422  */
1423 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1424 {
1425         if ((unsigned)size <= sysctl_optmem_max &&
1426             atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1427                 void *mem;
1428                 /* First do the add, to avoid the race if kmalloc
1429                  * might sleep.
1430                  */
1431                 atomic_add(size, &sk->sk_omem_alloc);
1432                 mem = kmalloc(size, priority);
1433                 if (mem)
1434                         return mem;
1435                 atomic_sub(size, &sk->sk_omem_alloc);
1436         }
1437         return NULL;
1438 }
1439 EXPORT_SYMBOL(sock_kmalloc);
1440
1441 /*
1442  * Free an option memory block.
1443  */
1444 void sock_kfree_s(struct sock *sk, void *mem, int size)
1445 {
1446         kfree(mem);
1447         atomic_sub(size, &sk->sk_omem_alloc);
1448 }
1449 EXPORT_SYMBOL(sock_kfree_s);
1450
1451 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1452    I think, these locks should be removed for datagram sockets.
1453  */
1454 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1455 {
1456         DEFINE_WAIT(wait);
1457
1458         clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1459         for (;;) {
1460                 if (!timeo)
1461                         break;
1462                 if (signal_pending(current))
1463                         break;
1464                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1465                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1466                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1467                         break;
1468                 if (sk->sk_shutdown & SEND_SHUTDOWN)
1469                         break;
1470                 if (sk->sk_err)
1471                         break;
1472                 timeo = schedule_timeout(timeo);
1473         }
1474         finish_wait(sk_sleep(sk), &wait);
1475         return timeo;
1476 }
1477
1478
1479 /*
1480  *      Generic send/receive buffer handlers
1481  */
1482
1483 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1484                                      unsigned long data_len, int noblock,
1485                                      int *errcode)
1486 {
1487         struct sk_buff *skb;
1488         gfp_t gfp_mask;
1489         long timeo;
1490         int err;
1491         int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1492
1493         err = -EMSGSIZE;
1494         if (npages > MAX_SKB_FRAGS)
1495                 goto failure;
1496
1497         gfp_mask = sk->sk_allocation;
1498         if (gfp_mask & __GFP_WAIT)
1499                 gfp_mask |= __GFP_REPEAT;
1500
1501         timeo = sock_sndtimeo(sk, noblock);
1502         while (1) {
1503                 err = sock_error(sk);
1504                 if (err != 0)
1505                         goto failure;
1506
1507                 err = -EPIPE;
1508                 if (sk->sk_shutdown & SEND_SHUTDOWN)
1509                         goto failure;
1510
1511                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1512                         skb = alloc_skb(header_len, gfp_mask);
1513                         if (skb) {
1514                                 int i;
1515
1516                                 /* No pages, we're done... */
1517                                 if (!data_len)
1518                                         break;
1519
1520                                 skb->truesize += data_len;
1521                                 skb_shinfo(skb)->nr_frags = npages;
1522                                 for (i = 0; i < npages; i++) {
1523                                         struct page *page;
1524
1525                                         page = alloc_pages(sk->sk_allocation, 0);
1526                                         if (!page) {
1527                                                 err = -ENOBUFS;
1528                                                 skb_shinfo(skb)->nr_frags = i;
1529                                                 kfree_skb(skb);
1530                                                 goto failure;
1531                                         }
1532
1533                                         __skb_fill_page_desc(skb, i,
1534                                                         page, 0,
1535                                                         (data_len >= PAGE_SIZE ?
1536                                                          PAGE_SIZE :
1537                                                          data_len));
1538                                         data_len -= PAGE_SIZE;
1539                                 }
1540
1541                                 /* Full success... */
1542                                 break;
1543                         }
1544                         err = -ENOBUFS;
1545                         goto failure;
1546                 }
1547                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1548                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1549                 err = -EAGAIN;
1550                 if (!timeo)
1551                         goto failure;
1552                 if (signal_pending(current))
1553                         goto interrupted;
1554                 timeo = sock_wait_for_wmem(sk, timeo);
1555         }
1556
1557         skb_set_owner_w(skb, sk);
1558         return skb;
1559
1560 interrupted:
1561         err = sock_intr_errno(timeo);
1562 failure:
1563         *errcode = err;
1564         return NULL;
1565 }
1566 EXPORT_SYMBOL(sock_alloc_send_pskb);
1567
1568 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1569                                     int noblock, int *errcode)
1570 {
1571         return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1572 }
1573 EXPORT_SYMBOL(sock_alloc_send_skb);
1574
1575 static void __lock_sock(struct sock *sk)
1576         __releases(&sk->sk_lock.slock)
1577         __acquires(&sk->sk_lock.slock)
1578 {
1579         DEFINE_WAIT(wait);
1580
1581         for (;;) {
1582                 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1583                                         TASK_UNINTERRUPTIBLE);
1584                 spin_unlock_bh(&sk->sk_lock.slock);
1585                 schedule();
1586                 spin_lock_bh(&sk->sk_lock.slock);
1587                 if (!sock_owned_by_user(sk))
1588                         break;
1589         }
1590         finish_wait(&sk->sk_lock.wq, &wait);
1591 }
1592
1593 static void __release_sock(struct sock *sk)
1594         __releases(&sk->sk_lock.slock)
1595         __acquires(&sk->sk_lock.slock)
1596 {
1597         struct sk_buff *skb = sk->sk_backlog.head;
1598
1599         do {
1600                 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1601                 bh_unlock_sock(sk);
1602
1603                 do {
1604                         struct sk_buff *next = skb->next;
1605
1606                         WARN_ON_ONCE(skb_dst_is_noref(skb));
1607                         skb->next = NULL;
1608                         sk_backlog_rcv(sk, skb);
1609
1610                         /*
1611                          * We are in process context here with softirqs
1612                          * disabled, use cond_resched_softirq() to preempt.
1613                          * This is safe to do because we've taken the backlog
1614                          * queue private:
1615                          */
1616                         cond_resched_softirq();
1617
1618                         skb = next;
1619                 } while (skb != NULL);
1620
1621                 bh_lock_sock(sk);
1622         } while ((skb = sk->sk_backlog.head) != NULL);
1623
1624         /*
1625          * Doing the zeroing here guarantee we can not loop forever
1626          * while a wild producer attempts to flood us.
1627          */
1628         sk->sk_backlog.len = 0;
1629 }
1630
1631 /**
1632  * sk_wait_data - wait for data to arrive at sk_receive_queue
1633  * @sk:    sock to wait on
1634  * @timeo: for how long
1635  *
1636  * Now socket state including sk->sk_err is changed only under lock,
1637  * hence we may omit checks after joining wait queue.
1638  * We check receive queue before schedule() only as optimization;
1639  * it is very likely that release_sock() added new data.
1640  */
1641 int sk_wait_data(struct sock *sk, long *timeo)
1642 {
1643         int rc;
1644         DEFINE_WAIT(wait);
1645
1646         prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1647         set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1648         rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1649         clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1650         finish_wait(sk_sleep(sk), &wait);
1651         return rc;
1652 }
1653 EXPORT_SYMBOL(sk_wait_data);
1654
1655 /**
1656  *      __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1657  *      @sk: socket
1658  *      @size: memory size to allocate
1659  *      @kind: allocation type
1660  *
1661  *      If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1662  *      rmem allocation. This function assumes that protocols which have
1663  *      memory_pressure use sk_wmem_queued as write buffer accounting.
1664  */
1665 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1666 {
1667         struct proto *prot = sk->sk_prot;
1668         int amt = sk_mem_pages(size);
1669         long allocated;
1670
1671         sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1672         allocated = atomic_long_add_return(amt, prot->memory_allocated);
1673
1674         /* Under limit. */
1675         if (allocated <= prot->sysctl_mem[0]) {
1676                 if (prot->memory_pressure && *prot->memory_pressure)
1677                         *prot->memory_pressure = 0;
1678                 return 1;
1679         }
1680
1681         /* Under pressure. */
1682         if (allocated > prot->sysctl_mem[1])
1683                 if (prot->enter_memory_pressure)
1684                         prot->enter_memory_pressure(sk);
1685
1686         /* Over hard limit. */
1687         if (allocated > prot->sysctl_mem[2])
1688                 goto suppress_allocation;
1689
1690         /* guarantee minimum buffer size under pressure */
1691         if (kind == SK_MEM_RECV) {
1692                 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1693                         return 1;
1694         } else { /* SK_MEM_SEND */
1695                 if (sk->sk_type == SOCK_STREAM) {
1696                         if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1697                                 return 1;
1698                 } else if (atomic_read(&sk->sk_wmem_alloc) <
1699                            prot->sysctl_wmem[0])
1700                                 return 1;
1701         }
1702
1703         if (prot->memory_pressure) {
1704                 int alloc;
1705
1706                 if (!*prot->memory_pressure)
1707                         return 1;
1708                 alloc = percpu_counter_read_positive(prot->sockets_allocated);
1709                 if (prot->sysctl_mem[2] > alloc *
1710                     sk_mem_pages(sk->sk_wmem_queued +
1711                                  atomic_read(&sk->sk_rmem_alloc) +
1712                                  sk->sk_forward_alloc))
1713                         return 1;
1714         }
1715
1716 suppress_allocation:
1717
1718         if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1719                 sk_stream_moderate_sndbuf(sk);
1720
1721                 /* Fail only if socket is _under_ its sndbuf.
1722                  * In this case we cannot block, so that we have to fail.
1723                  */
1724                 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1725                         return 1;
1726         }
1727
1728         trace_sock_exceed_buf_limit(sk, prot, allocated);
1729
1730         /* Alas. Undo changes. */
1731         sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1732         atomic_long_sub(amt, prot->memory_allocated);
1733         return 0;
1734 }
1735 EXPORT_SYMBOL(__sk_mem_schedule);
1736
1737 /**
1738  *      __sk_reclaim - reclaim memory_allocated
1739  *      @sk: socket
1740  */
1741 void __sk_mem_reclaim(struct sock *sk)
1742 {
1743         struct proto *prot = sk->sk_prot;
1744
1745         atomic_long_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
1746                    prot->memory_allocated);
1747         sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1748
1749         if (prot->memory_pressure && *prot->memory_pressure &&
1750             (atomic_long_read(prot->memory_allocated) < prot->sysctl_mem[0]))
1751                 *prot->memory_pressure = 0;
1752 }
1753 EXPORT_SYMBOL(__sk_mem_reclaim);
1754
1755
1756 /*
1757  * Set of default routines for initialising struct proto_ops when
1758  * the protocol does not support a particular function. In certain
1759  * cases where it makes no sense for a protocol to have a "do nothing"
1760  * function, some default processing is provided.
1761  */
1762
1763 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1764 {
1765         return -EOPNOTSUPP;
1766 }
1767 EXPORT_SYMBOL(sock_no_bind);
1768
1769 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1770                     int len, int flags)
1771 {
1772         return -EOPNOTSUPP;
1773 }
1774 EXPORT_SYMBOL(sock_no_connect);
1775
1776 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1777 {
1778         return -EOPNOTSUPP;
1779 }
1780 EXPORT_SYMBOL(sock_no_socketpair);
1781
1782 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1783 {
1784         return -EOPNOTSUPP;
1785 }
1786 EXPORT_SYMBOL(sock_no_accept);
1787
1788 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1789                     int *len, int peer)
1790 {
1791         return -EOPNOTSUPP;
1792 }
1793 EXPORT_SYMBOL(sock_no_getname);
1794
1795 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1796 {
1797         return 0;
1798 }
1799 EXPORT_SYMBOL(sock_no_poll);
1800
1801 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1802 {
1803         return -EOPNOTSUPP;
1804 }
1805 EXPORT_SYMBOL(sock_no_ioctl);
1806
1807 int sock_no_listen(struct socket *sock, int backlog)
1808 {
1809         return -EOPNOTSUPP;
1810 }
1811 EXPORT_SYMBOL(sock_no_listen);
1812
1813 int sock_no_shutdown(struct socket *sock, int how)
1814 {
1815         return -EOPNOTSUPP;
1816 }
1817 EXPORT_SYMBOL(sock_no_shutdown);
1818
1819 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1820                     char __user *optval, unsigned int optlen)
1821 {
1822         return -EOPNOTSUPP;
1823 }
1824 EXPORT_SYMBOL(sock_no_setsockopt);
1825
1826 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1827                     char __user *optval, int __user *optlen)
1828 {
1829         return -EOPNOTSUPP;
1830 }
1831 EXPORT_SYMBOL(sock_no_getsockopt);
1832
1833 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1834                     size_t len)
1835 {
1836         return -EOPNOTSUPP;
1837 }
1838 EXPORT_SYMBOL(sock_no_sendmsg);
1839
1840 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1841                     size_t len, int flags)
1842 {
1843         return -EOPNOTSUPP;
1844 }
1845 EXPORT_SYMBOL(sock_no_recvmsg);
1846
1847 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1848 {
1849         /* Mirror missing mmap method error code */
1850         return -ENODEV;
1851 }
1852 EXPORT_SYMBOL(sock_no_mmap);
1853
1854 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1855 {
1856         ssize_t res;
1857         struct msghdr msg = {.msg_flags = flags};
1858         struct kvec iov;
1859         char *kaddr = kmap(page);
1860         iov.iov_base = kaddr + offset;
1861         iov.iov_len = size;
1862         res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1863         kunmap(page);
1864         return res;
1865 }
1866 EXPORT_SYMBOL(sock_no_sendpage);
1867
1868 /*
1869  *      Default Socket Callbacks
1870  */
1871
1872 static void sock_def_wakeup(struct sock *sk)
1873 {
1874         struct socket_wq *wq;
1875
1876         rcu_read_lock();
1877         wq = rcu_dereference(sk->sk_wq);
1878         if (wq_has_sleeper(wq))
1879                 wake_up_interruptible_all(&wq->wait);
1880         rcu_read_unlock();
1881 }
1882
1883 static void sock_def_error_report(struct sock *sk)
1884 {
1885         struct socket_wq *wq;
1886
1887         rcu_read_lock();
1888         wq = rcu_dereference(sk->sk_wq);
1889         if (wq_has_sleeper(wq))
1890                 wake_up_interruptible_poll(&wq->wait, POLLERR);
1891         sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
1892         rcu_read_unlock();
1893 }
1894
1895 static void sock_def_readable(struct sock *sk, int len)
1896 {
1897         struct socket_wq *wq;
1898
1899         rcu_read_lock();
1900         wq = rcu_dereference(sk->sk_wq);
1901         if (wq_has_sleeper(wq))
1902                 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
1903                                                 POLLRDNORM | POLLRDBAND);
1904         sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1905         rcu_read_unlock();
1906 }
1907
1908 static void sock_def_write_space(struct sock *sk)
1909 {
1910         struct socket_wq *wq;
1911
1912         rcu_read_lock();
1913
1914         /* Do not wake up a writer until he can make "significant"
1915          * progress.  --DaveM
1916          */
1917         if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1918                 wq = rcu_dereference(sk->sk_wq);
1919                 if (wq_has_sleeper(wq))
1920                         wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
1921                                                 POLLWRNORM | POLLWRBAND);
1922
1923                 /* Should agree with poll, otherwise some programs break */
1924                 if (sock_writeable(sk))
1925                         sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1926         }
1927
1928         rcu_read_unlock();
1929 }
1930
1931 static void sock_def_destruct(struct sock *sk)
1932 {
1933         kfree(sk->sk_protinfo);
1934 }
1935
1936 void sk_send_sigurg(struct sock *sk)
1937 {
1938         if (sk->sk_socket && sk->sk_socket->file)
1939                 if (send_sigurg(&sk->sk_socket->file->f_owner))
1940                         sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1941 }
1942 EXPORT_SYMBOL(sk_send_sigurg);
1943
1944 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1945                     unsigned long expires)
1946 {
1947         if (!mod_timer(timer, expires))
1948                 sock_hold(sk);
1949 }
1950 EXPORT_SYMBOL(sk_reset_timer);
1951
1952 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1953 {
1954         if (timer_pending(timer) && del_timer(timer))
1955                 __sock_put(sk);
1956 }
1957 EXPORT_SYMBOL(sk_stop_timer);
1958
1959 void sock_init_data(struct socket *sock, struct sock *sk)
1960 {
1961         skb_queue_head_init(&sk->sk_receive_queue);
1962         skb_queue_head_init(&sk->sk_write_queue);
1963         skb_queue_head_init(&sk->sk_error_queue);
1964 #ifdef CONFIG_NET_DMA
1965         skb_queue_head_init(&sk->sk_async_wait_queue);
1966 #endif
1967
1968         sk->sk_send_head        =       NULL;
1969
1970         init_timer(&sk->sk_timer);
1971
1972         sk->sk_allocation       =       GFP_KERNEL;
1973         sk->sk_rcvbuf           =       sysctl_rmem_default;
1974         sk->sk_sndbuf           =       sysctl_wmem_default;
1975         sk->sk_state            =       TCP_CLOSE;
1976         sk_set_socket(sk, sock);
1977
1978         sock_set_flag(sk, SOCK_ZAPPED);
1979
1980         if (sock) {
1981                 sk->sk_type     =       sock->type;
1982                 sk->sk_wq       =       sock->wq;
1983                 sock->sk        =       sk;
1984         } else
1985                 sk->sk_wq       =       NULL;
1986
1987         spin_lock_init(&sk->sk_dst_lock);
1988         rwlock_init(&sk->sk_callback_lock);
1989         lockdep_set_class_and_name(&sk->sk_callback_lock,
1990                         af_callback_keys + sk->sk_family,
1991                         af_family_clock_key_strings[sk->sk_family]);
1992
1993         sk->sk_state_change     =       sock_def_wakeup;
1994         sk->sk_data_ready       =       sock_def_readable;
1995         sk->sk_write_space      =       sock_def_write_space;
1996         sk->sk_error_report     =       sock_def_error_report;
1997         sk->sk_destruct         =       sock_def_destruct;
1998
1999         sk->sk_sndmsg_page      =       NULL;
2000         sk->sk_sndmsg_off       =       0;
2001
2002         sk->sk_peer_pid         =       NULL;
2003         sk->sk_peer_cred        =       NULL;
2004         sk->sk_write_pending    =       0;
2005         sk->sk_rcvlowat         =       1;
2006         sk->sk_rcvtimeo         =       MAX_SCHEDULE_TIMEOUT;
2007         sk->sk_sndtimeo         =       MAX_SCHEDULE_TIMEOUT;
2008
2009         sk->sk_stamp = ktime_set(-1L, 0);
2010
2011         /*
2012          * Before updating sk_refcnt, we must commit prior changes to memory
2013          * (Documentation/RCU/rculist_nulls.txt for details)
2014          */
2015         smp_wmb();
2016         atomic_set(&sk->sk_refcnt, 1);
2017         atomic_set(&sk->sk_drops, 0);
2018 }
2019 EXPORT_SYMBOL(sock_init_data);
2020
2021 void lock_sock_nested(struct sock *sk, int subclass)
2022 {
2023         might_sleep();
2024         spin_lock_bh(&sk->sk_lock.slock);
2025         if (sk->sk_lock.owned)
2026                 __lock_sock(sk);
2027         sk->sk_lock.owned = 1;
2028         spin_unlock(&sk->sk_lock.slock);
2029         /*
2030          * The sk_lock has mutex_lock() semantics here:
2031          */
2032         mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2033         local_bh_enable();
2034 }
2035 EXPORT_SYMBOL(lock_sock_nested);
2036
2037 void release_sock(struct sock *sk)
2038 {
2039         /*
2040          * The sk_lock has mutex_unlock() semantics:
2041          */
2042         mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2043
2044         spin_lock_bh(&sk->sk_lock.slock);
2045         if (sk->sk_backlog.tail)
2046                 __release_sock(sk);
2047         sk->sk_lock.owned = 0;
2048         if (waitqueue_active(&sk->sk_lock.wq))
2049                 wake_up(&sk->sk_lock.wq);
2050         spin_unlock_bh(&sk->sk_lock.slock);
2051 }
2052 EXPORT_SYMBOL(release_sock);
2053
2054 /**
2055  * lock_sock_fast - fast version of lock_sock
2056  * @sk: socket
2057  *
2058  * This version should be used for very small section, where process wont block
2059  * return false if fast path is taken
2060  *   sk_lock.slock locked, owned = 0, BH disabled
2061  * return true if slow path is taken
2062  *   sk_lock.slock unlocked, owned = 1, BH enabled
2063  */
2064 bool lock_sock_fast(struct sock *sk)
2065 {
2066         might_sleep();
2067         spin_lock_bh(&sk->sk_lock.slock);
2068
2069         if (!sk->sk_lock.owned)
2070                 /*
2071                  * Note : We must disable BH
2072                  */
2073                 return false;
2074
2075         __lock_sock(sk);
2076         sk->sk_lock.owned = 1;
2077         spin_unlock(&sk->sk_lock.slock);
2078         /*
2079          * The sk_lock has mutex_lock() semantics here:
2080          */
2081         mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2082         local_bh_enable();
2083         return true;
2084 }
2085 EXPORT_SYMBOL(lock_sock_fast);
2086
2087 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2088 {
2089         struct timeval tv;
2090         if (!sock_flag(sk, SOCK_TIMESTAMP))
2091                 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2092         tv = ktime_to_timeval(sk->sk_stamp);
2093         if (tv.tv_sec == -1)
2094                 return -ENOENT;
2095         if (tv.tv_sec == 0) {
2096                 sk->sk_stamp = ktime_get_real();
2097                 tv = ktime_to_timeval(sk->sk_stamp);
2098         }
2099         return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2100 }
2101 EXPORT_SYMBOL(sock_get_timestamp);
2102
2103 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2104 {
2105         struct timespec ts;
2106         if (!sock_flag(sk, SOCK_TIMESTAMP))
2107                 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2108         ts = ktime_to_timespec(sk->sk_stamp);
2109         if (ts.tv_sec == -1)
2110                 return -ENOENT;
2111         if (ts.tv_sec == 0) {
2112                 sk->sk_stamp = ktime_get_real();
2113                 ts = ktime_to_timespec(sk->sk_stamp);
2114         }
2115         return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2116 }
2117 EXPORT_SYMBOL(sock_get_timestampns);
2118
2119 void sock_enable_timestamp(struct sock *sk, int flag)
2120 {
2121         if (!sock_flag(sk, flag)) {
2122                 sock_set_flag(sk, flag);
2123                 /*
2124                  * we just set one of the two flags which require net
2125                  * time stamping, but time stamping might have been on
2126                  * already because of the other one
2127                  */
2128                 if (!sock_flag(sk,
2129                                 flag == SOCK_TIMESTAMP ?
2130                                 SOCK_TIMESTAMPING_RX_SOFTWARE :
2131                                 SOCK_TIMESTAMP))
2132                         net_enable_timestamp();
2133         }
2134 }
2135
2136 /*
2137  *      Get a socket option on an socket.
2138  *
2139  *      FIX: POSIX 1003.1g is very ambiguous here. It states that
2140  *      asynchronous errors should be reported by getsockopt. We assume
2141  *      this means if you specify SO_ERROR (otherwise whats the point of it).
2142  */
2143 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2144                            char __user *optval, int __user *optlen)
2145 {
2146         struct sock *sk = sock->sk;
2147
2148         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2149 }
2150 EXPORT_SYMBOL(sock_common_getsockopt);
2151
2152 #ifdef CONFIG_COMPAT
2153 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2154                                   char __user *optval, int __user *optlen)
2155 {
2156         struct sock *sk = sock->sk;
2157
2158         if (sk->sk_prot->compat_getsockopt != NULL)
2159                 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2160                                                       optval, optlen);
2161         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2162 }
2163 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2164 #endif
2165
2166 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2167                         struct msghdr *msg, size_t size, int flags)
2168 {
2169         struct sock *sk = sock->sk;
2170         int addr_len = 0;
2171         int err;
2172
2173         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2174                                    flags & ~MSG_DONTWAIT, &addr_len);
2175         if (err >= 0)
2176                 msg->msg_namelen = addr_len;
2177         return err;
2178 }
2179 EXPORT_SYMBOL(sock_common_recvmsg);
2180
2181 /*
2182  *      Set socket options on an inet socket.
2183  */
2184 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2185                            char __user *optval, unsigned int optlen)
2186 {
2187         struct sock *sk = sock->sk;
2188
2189         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2190 }
2191 EXPORT_SYMBOL(sock_common_setsockopt);
2192
2193 #ifdef CONFIG_COMPAT
2194 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2195                                   char __user *optval, unsigned int optlen)
2196 {
2197         struct sock *sk = sock->sk;
2198
2199         if (sk->sk_prot->compat_setsockopt != NULL)
2200                 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2201                                                       optval, optlen);
2202         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2203 }
2204 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2205 #endif
2206
2207 void sk_common_release(struct sock *sk)
2208 {
2209         if (sk->sk_prot->destroy)
2210                 sk->sk_prot->destroy(sk);
2211
2212         /*
2213          * Observation: when sock_common_release is called, processes have
2214          * no access to socket. But net still has.
2215          * Step one, detach it from networking:
2216          *
2217          * A. Remove from hash tables.
2218          */
2219
2220         sk->sk_prot->unhash(sk);
2221
2222         /*
2223          * In this point socket cannot receive new packets, but it is possible
2224          * that some packets are in flight because some CPU runs receiver and
2225          * did hash table lookup before we unhashed socket. They will achieve
2226          * receive queue and will be purged by socket destructor.
2227          *
2228          * Also we still have packets pending on receive queue and probably,
2229          * our own packets waiting in device queues. sock_destroy will drain
2230          * receive queue, but transmitted packets will delay socket destruction
2231          * until the last reference will be released.
2232          */
2233
2234         sock_orphan(sk);
2235
2236         xfrm_sk_free_policy(sk);
2237
2238         sk_refcnt_debug_release(sk);
2239         sock_put(sk);
2240 }
2241 EXPORT_SYMBOL(sk_common_release);
2242
2243 static DEFINE_RWLOCK(proto_list_lock);
2244 static LIST_HEAD(proto_list);
2245
2246 #ifdef CONFIG_PROC_FS
2247 #define PROTO_INUSE_NR  64      /* should be enough for the first time */
2248 struct prot_inuse {
2249         int val[PROTO_INUSE_NR];
2250 };
2251
2252 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2253
2254 #ifdef CONFIG_NET_NS
2255 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2256 {
2257         __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2258 }
2259 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2260
2261 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2262 {
2263         int cpu, idx = prot->inuse_idx;
2264         int res = 0;
2265
2266         for_each_possible_cpu(cpu)
2267                 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2268
2269         return res >= 0 ? res : 0;
2270 }
2271 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2272
2273 static int __net_init sock_inuse_init_net(struct net *net)
2274 {
2275         net->core.inuse = alloc_percpu(struct prot_inuse);
2276         return net->core.inuse ? 0 : -ENOMEM;
2277 }
2278
2279 static void __net_exit sock_inuse_exit_net(struct net *net)
2280 {
2281         free_percpu(net->core.inuse);
2282 }
2283
2284 static struct pernet_operations net_inuse_ops = {
2285         .init = sock_inuse_init_net,
2286         .exit = sock_inuse_exit_net,
2287 };
2288
2289 static __init int net_inuse_init(void)
2290 {
2291         if (register_pernet_subsys(&net_inuse_ops))
2292                 panic("Cannot initialize net inuse counters");
2293
2294         return 0;
2295 }
2296
2297 core_initcall(net_inuse_init);
2298 #else
2299 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2300
2301 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2302 {
2303         __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2304 }
2305 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2306
2307 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2308 {
2309         int cpu, idx = prot->inuse_idx;
2310         int res = 0;
2311
2312         for_each_possible_cpu(cpu)
2313                 res += per_cpu(prot_inuse, cpu).val[idx];
2314
2315         return res >= 0 ? res : 0;
2316 }
2317 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2318 #endif
2319
2320 static void assign_proto_idx(struct proto *prot)
2321 {
2322         prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2323
2324         if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2325                 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2326                 return;
2327         }
2328
2329         set_bit(prot->inuse_idx, proto_inuse_idx);
2330 }
2331
2332 static void release_proto_idx(struct proto *prot)
2333 {
2334         if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2335                 clear_bit(prot->inuse_idx, proto_inuse_idx);
2336 }
2337 #else
2338 static inline void assign_proto_idx(struct proto *prot)
2339 {
2340 }
2341
2342 static inline void release_proto_idx(struct proto *prot)
2343 {
2344 }
2345 #endif
2346
2347 int proto_register(struct proto *prot, int alloc_slab)
2348 {
2349         if (alloc_slab) {
2350                 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2351                                         SLAB_HWCACHE_ALIGN | prot->slab_flags,
2352                                         NULL);
2353
2354                 if (prot->slab == NULL) {
2355                         printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2356                                prot->name);
2357                         goto out;
2358                 }
2359
2360                 if (prot->rsk_prot != NULL) {
2361                         prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
2362                         if (prot->rsk_prot->slab_name == NULL)
2363                                 goto out_free_sock_slab;
2364
2365                         prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2366                                                                  prot->rsk_prot->obj_size, 0,
2367                                                                  SLAB_HWCACHE_ALIGN, NULL);
2368
2369                         if (prot->rsk_prot->slab == NULL) {
2370                                 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2371                                        prot->name);
2372                                 goto out_free_request_sock_slab_name;
2373                         }
2374                 }
2375
2376                 if (prot->twsk_prot != NULL) {
2377                         prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2378
2379                         if (prot->twsk_prot->twsk_slab_name == NULL)
2380                                 goto out_free_request_sock_slab;
2381
2382                         prot->twsk_prot->twsk_slab =
2383                                 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2384                                                   prot->twsk_prot->twsk_obj_size,
2385                                                   0,
2386                                                   SLAB_HWCACHE_ALIGN |
2387                                                         prot->slab_flags,
2388                                                   NULL);
2389                         if (prot->twsk_prot->twsk_slab == NULL)
2390                                 goto out_free_timewait_sock_slab_name;
2391                 }
2392         }
2393
2394         write_lock(&proto_list_lock);
2395         list_add(&prot->node, &proto_list);
2396         assign_proto_idx(prot);
2397         write_unlock(&proto_list_lock);
2398         return 0;
2399
2400 out_free_timewait_sock_slab_name:
2401         kfree(prot->twsk_prot->twsk_slab_name);
2402 out_free_request_sock_slab:
2403         if (prot->rsk_prot && prot->rsk_prot->slab) {
2404                 kmem_cache_destroy(prot->rsk_prot->slab);
2405                 prot->rsk_prot->slab = NULL;
2406         }
2407 out_free_request_sock_slab_name:
2408         if (prot->rsk_prot)
2409                 kfree(prot->rsk_prot->slab_name);
2410 out_free_sock_slab:
2411         kmem_cache_destroy(prot->slab);
2412         prot->slab = NULL;
2413 out:
2414         return -ENOBUFS;
2415 }
2416 EXPORT_SYMBOL(proto_register);
2417
2418 void proto_unregister(struct proto *prot)
2419 {
2420         write_lock(&proto_list_lock);
2421         release_proto_idx(prot);
2422         list_del(&prot->node);
2423         write_unlock(&proto_list_lock);
2424
2425         if (prot->slab != NULL) {
2426                 kmem_cache_destroy(prot->slab);
2427                 prot->slab = NULL;
2428         }
2429
2430         if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2431                 kmem_cache_destroy(prot->rsk_prot->slab);
2432                 kfree(prot->rsk_prot->slab_name);
2433                 prot->rsk_prot->slab = NULL;
2434         }
2435
2436         if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2437                 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2438                 kfree(prot->twsk_prot->twsk_slab_name);
2439                 prot->twsk_prot->twsk_slab = NULL;
2440         }
2441 }
2442 EXPORT_SYMBOL(proto_unregister);
2443
2444 #ifdef CONFIG_PROC_FS
2445 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2446         __acquires(proto_list_lock)
2447 {
2448         read_lock(&proto_list_lock);
2449         return seq_list_start_head(&proto_list, *pos);
2450 }
2451
2452 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2453 {
2454         return seq_list_next(v, &proto_list, pos);
2455 }
2456
2457 static void proto_seq_stop(struct seq_file *seq, void *v)
2458         __releases(proto_list_lock)
2459 {
2460         read_unlock(&proto_list_lock);
2461 }
2462
2463 static char proto_method_implemented(const void *method)
2464 {
2465         return method == NULL ? 'n' : 'y';
2466 }
2467
2468 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2469 {
2470         seq_printf(seq, "%-9s %4u %6d  %6ld   %-3s %6u   %-3s  %-10s "
2471                         "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2472                    proto->name,
2473                    proto->obj_size,
2474                    sock_prot_inuse_get(seq_file_net(seq), proto),
2475                    proto->memory_allocated != NULL ? atomic_long_read(proto->memory_allocated) : -1L,
2476                    proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2477                    proto->max_header,
2478                    proto->slab == NULL ? "no" : "yes",
2479                    module_name(proto->owner),
2480                    proto_method_implemented(proto->close),
2481                    proto_method_implemented(proto->connect),
2482                    proto_method_implemented(proto->disconnect),
2483                    proto_method_implemented(proto->accept),
2484                    proto_method_implemented(proto->ioctl),
2485                    proto_method_implemented(proto->init),
2486                    proto_method_implemented(proto->destroy),
2487                    proto_method_implemented(proto->shutdown),
2488                    proto_method_implemented(proto->setsockopt),
2489                    proto_method_implemented(proto->getsockopt),
2490                    proto_method_implemented(proto->sendmsg),
2491                    proto_method_implemented(proto->recvmsg),
2492                    proto_method_implemented(proto->sendpage),
2493                    proto_method_implemented(proto->bind),
2494                    proto_method_implemented(proto->backlog_rcv),
2495                    proto_method_implemented(proto->hash),
2496                    proto_method_implemented(proto->unhash),
2497                    proto_method_implemented(proto->get_port),
2498                    proto_method_implemented(proto->enter_memory_pressure));
2499 }
2500
2501 static int proto_seq_show(struct seq_file *seq, void *v)
2502 {
2503         if (v == &proto_list)
2504                 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2505                            "protocol",
2506                            "size",
2507                            "sockets",
2508                            "memory",
2509                            "press",
2510                            "maxhdr",
2511                            "slab",
2512                            "module",
2513                            "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2514         else
2515                 proto_seq_printf(seq, list_entry(v, struct proto, node));
2516         return 0;
2517 }
2518
2519 static const struct seq_operations proto_seq_ops = {
2520         .start  = proto_seq_start,
2521         .next   = proto_seq_next,
2522         .stop   = proto_seq_stop,
2523         .show   = proto_seq_show,
2524 };
2525
2526 static int proto_seq_open(struct inode *inode, struct file *file)
2527 {
2528         return seq_open_net(inode, file, &proto_seq_ops,
2529                             sizeof(struct seq_net_private));
2530 }
2531
2532 static const struct file_operations proto_seq_fops = {
2533         .owner          = THIS_MODULE,
2534         .open           = proto_seq_open,
2535         .read           = seq_read,
2536         .llseek         = seq_lseek,
2537         .release        = seq_release_net,
2538 };
2539
2540 static __net_init int proto_init_net(struct net *net)
2541 {
2542         if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2543                 return -ENOMEM;
2544
2545         return 0;
2546 }
2547
2548 static __net_exit void proto_exit_net(struct net *net)
2549 {
2550         proc_net_remove(net, "protocols");
2551 }
2552
2553
2554 static __net_initdata struct pernet_operations proto_net_ops = {
2555         .init = proto_init_net,
2556         .exit = proto_exit_net,
2557 };
2558
2559 static int __init proto_init(void)
2560 {
2561         return register_pernet_subsys(&proto_net_ops);
2562 }
2563
2564 subsys_initcall(proto_init);
2565
2566 #endif /* PROC_FS */