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                 if (val > sysctl_wmem_max)
 534                         val = sysctl_wmem_max;
 535 set_sndbuf:
 536                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
 537                 if ((val * 2) < SOCK_MIN_SNDBUF)
 538                         sk->sk_sndbuf = SOCK_MIN_SNDBUF;
 539                 else
 540                         sk->sk_sndbuf = val * 2;
 541
 542                 /*
 543                  *      Wake up sending tasks if we
 544                  *      upped the value.
 545                  */
 546                 sk->sk_write_space(sk);
 547                 break;
 548
 549         case SO_SNDBUFFORCE:
 550                 if (!capable(CAP_NET_ADMIN)) {
 551                         ret = -EPERM;
 552                         break;
 553                 }
 554                 goto set_sndbuf;
 555
 556         case SO_RCVBUF:
 557                 /* Don't error on this BSD doesn't and if you think
 558                    about it this is right. Otherwise apps have to
 559                    play 'guess the biggest size' games. RCVBUF/SNDBUF
 560                    are treated in BSD as hints */
 561
 562                 if (val > sysctl_rmem_max)
 563                         val = sysctl_rmem_max;
 564 set_rcvbuf:
 565                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
 566                 /*
 567                  * We double it on the way in to account for
 568                  * "struct sk_buff" etc. overhead.   Applications
 569                  * assume that the SO_RCVBUF setting they make will
 570                  * allow that much actual data to be received on that
 571                  * socket.
 572                  *
 573                  * Applications are unaware that "struct sk_buff" and
 574                  * other overheads allocate from the receive buffer
 575                  * during socket buffer allocation.
 576                  *
 577                  * And after considering the possible alternatives,
 578                  * returning the value we actually used in getsockopt
 579                  * is the most desirable behavior.
 580                  */
 581                 if ((val * 2) < SOCK_MIN_RCVBUF)
 582                         sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
 583                 else
 584                         sk->sk_rcvbuf = val * 2;
 585                 break;
 586
 587         case SO_RCVBUFFORCE:
 588                 if (!capable(CAP_NET_ADMIN)) {
 589                         ret = -EPERM;
 590                         break;
 591                 }
 592                 goto set_rcvbuf;
 593
 594         case SO_KEEPALIVE:
 595 #ifdef CONFIG_INET
 596                 if (sk->sk_protocol == IPPROTO_TCP &&
 597                     sk->sk_type == SOCK_STREAM)
 598                         tcp_set_keepalive(sk, valbool);
 599 #endif
 600                 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
 601                 break;
 602
 603         case SO_OOBINLINE:
 604                 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
 605                 break;
 606
 607         case SO_NO_CHECK:
 608                 sk->sk_no_check = valbool;
 609                 break;
 610
 611         case SO_PRIORITY:
 612                 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
 613                         sk->sk_priority = val;
 614                 else
 615                         ret = -EPERM;
 616                 break;
 617
 618         case SO_LINGER:
 619                 if (optlen < sizeof(ling)) {
 620                         ret = -EINVAL;  /* 1003.1g */
 621                         break;
 622                 }
 623                 if (copy_from_user(&ling, optval, sizeof(ling))) {
 624                         ret = -EFAULT;
 625                         break;
 626                 }
 627                 if (!ling.l_onoff)
 628                         sock_reset_flag(sk, SOCK_LINGER);
 629                 else {
 630 #if (BITS_PER_LONG == 32)
 631                         if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
 632                                 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
 633                         else
 634 #endif
 635                                 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
 636                         sock_set_flag(sk, SOCK_LINGER);
 637                 }
 638                 break;
 639
 640         case SO_BSDCOMPAT:
 641                 sock_warn_obsolete_bsdism("setsockopt");
 642                 break;
 643
 644         case SO_PASSCRED:
 645                 if (valbool)
 646                         set_bit(SOCK_PASSCRED, &sock->flags);
 647                 else
 648                         clear_bit(SOCK_PASSCRED, &sock->flags);
 649                 break;
 650
 651         case SO_TIMESTAMP:
 652         case SO_TIMESTAMPNS:
 653                 if (valbool)  {
 654                         if (optname == SO_TIMESTAMP)
 655                                 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
 656                         else
 657                                 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
 658                         sock_set_flag(sk, SOCK_RCVTSTAMP);
 659                         sock_enable_timestamp(sk, SOCK_TIMESTAMP);
 660                 } else {
 661                         sock_reset_flag(sk, SOCK_RCVTSTAMP);
 662                         sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
 663                 }
 664                 break;
 665
 666         case SO_TIMESTAMPING:
 667                 if (val & ~SOF_TIMESTAMPING_MASK) {
 668                         ret = -EINVAL;
 669                         break;
 670                 }
 671                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
 672                                   val & SOF_TIMESTAMPING_TX_HARDWARE);
 673                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
 674                                   val & SOF_TIMESTAMPING_TX_SOFTWARE);
 675                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
 676                                   val & SOF_TIMESTAMPING_RX_HARDWARE);
 677                 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
 678                         sock_enable_timestamp(sk,
 679                                               SOCK_TIMESTAMPING_RX_SOFTWARE);
 680                 else
 681                         sock_disable_timestamp(sk,
 682                                                SOCK_TIMESTAMPING_RX_SOFTWARE);
 683                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
 684                                   val & SOF_TIMESTAMPING_SOFTWARE);
 685                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
 686                                   val & SOF_TIMESTAMPING_SYS_HARDWARE);
 687                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
 688                                   val & SOF_TIMESTAMPING_RAW_HARDWARE);
 689                 break;
 690
 691         case SO_RCVLOWAT:
 692                 if (val < 0)
 693                         val = INT_MAX;
 694                 sk->sk_rcvlowat = val ? : 1;
 695                 break;
 696
 697         case SO_RCVTIMEO:
 698                 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
 699                 break;
 700
 701         case SO_SNDTIMEO:
 702                 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
 703                 break;
 704
 705         case SO_ATTACH_FILTER:
 706                 ret = -EINVAL;
 707                 if (optlen == sizeof(struct sock_fprog)) {
 708                         struct sock_fprog fprog;
 709
 710                         ret = -EFAULT;
 711                         if (copy_from_user(&fprog, optval, sizeof(fprog)))
 712                                 break;
 713
 714                         ret = sk_attach_filter(&fprog, sk);
 715                 }
 716                 break;
 717
 718         case SO_DETACH_FILTER:
 719                 ret = sk_detach_filter(sk);
 720                 break;
 721
 722         case SO_PASSSEC:
 723                 if (valbool)
 724                         set_bit(SOCK_PASSSEC, &sock->flags);
 725                 else
 726                         clear_bit(SOCK_PASSSEC, &sock->flags);
 727                 break;
 728         case SO_MARK:
 729                 if (!capable(CAP_NET_ADMIN))
 730                         ret = -EPERM;
 731                 else
 732                         sk->sk_mark = val;
 733                 break;
 734
 735                 /* We implement the SO_SNDLOWAT etc to
 736                    not be settable (1003.1g 5.3) */
 737         case SO_RXQ_OVFL:
 738                 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
 739                 break;
 740         default:
 741                 ret = -ENOPROTOOPT;
 742                 break;
 743         }
 744         release_sock(sk);
 745         return ret;
 746 }
 747 EXPORT_SYMBOL(sock_setsockopt);
 748
 749
 750 void cred_to_ucred(struct pid *pid, const struct cred *cred,
 751                    struct ucred *ucred)
 752 {
 753         ucred->pid = pid_vnr(pid);
 754         ucred->uid = ucred->gid = -1;
 755         if (cred) {
 756                 struct user_namespace *current_ns = current_user_ns();
 757
 758                 ucred->uid = user_ns_map_uid(current_ns, cred, cred->euid);
 759                 ucred->gid = user_ns_map_gid(current_ns, cred, cred->egid);
 760         }
 761 }
 762 EXPORT_SYMBOL_GPL(cred_to_ucred);
 763
 764 void cred_real_to_ucred(struct pid *pid, const struct cred *cred,
 765                         struct ucred *ucred)
 766 {
 767         ucred->pid = pid_vnr(pid);
 768         ucred->uid = ucred->gid = -1;
 769         if (cred) {
 770                 struct user_namespace *current_ns = current_user_ns();
 771
 772                 ucred->uid = user_ns_map_uid(current_ns, cred, cred->uid);
 773                 ucred->gid = user_ns_map_gid(current_ns, cred, cred->gid);
 774         }
 775 }
 776 EXPORT_SYMBOL_GPL(cred_real_to_ucred);
 777
 778 int sock_getsockopt(struct socket *sock, int level, int optname,
 779                     char __user *optval, int __user *optlen)
 780 {
 781         struct sock *sk = sock->sk;
 782
 783         union {
 784                 int val;
 785                 struct linger ling;
 786                 struct timeval tm;
 787         } v;
 788
 789         int lv = sizeof(int);
 790         int len;
 791
 792         if (get_user(len, optlen))
 793                 return -EFAULT;
 794         if (len < 0)
 795                 return -EINVAL;
 796
 797         memset(&v, 0, sizeof(v));
 798
 799         switch (optname) {
 800         case SO_DEBUG:
 801                 v.val = sock_flag(sk, SOCK_DBG);
 802                 break;
 803
 804         case SO_DONTROUTE:
 805                 v.val = sock_flag(sk, SOCK_LOCALROUTE);
 806                 break;
 807
 808         case SO_BROADCAST:
 809                 v.val = !!sock_flag(sk, SOCK_BROADCAST);
 810                 break;
 811
 812         case SO_SNDBUF:
 813                 v.val = sk->sk_sndbuf;
 814                 break;
 815
 816         case SO_RCVBUF:
 817                 v.val = sk->sk_rcvbuf;
 818                 break;
 819
 820         case SO_REUSEADDR:
 821                 v.val = sk->sk_reuse;
 822                 break;
 823
 824         case SO_KEEPALIVE:
 825                 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
 826                 break;
 827
 828         case SO_TYPE:
 829                 v.val = sk->sk_type;
 830                 break;
 831
 832         case SO_PROTOCOL:
 833                 v.val = sk->sk_protocol;
 834                 break;
 835
 836         case SO_DOMAIN:
 837                 v.val = sk->sk_family;
 838                 break;
 839
 840         case SO_ERROR:
 841                 v.val = -sock_error(sk);
 842                 if (v.val == 0)
 843                         v.val = xchg(&sk->sk_err_soft, 0);
 844                 break;
 845
 846         case SO_OOBINLINE:
 847                 v.val = !!sock_flag(sk, SOCK_URGINLINE);
 848                 break;
 849
 850         case SO_NO_CHECK:
 851                 v.val = sk->sk_no_check;
 852                 break;
 853
 854         case SO_PRIORITY:
 855                 v.val = sk->sk_priority;
 856                 break;
 857
 858         case SO_LINGER:
 859                 lv              = sizeof(v.ling);
 860                 v.ling.l_onoff  = !!sock_flag(sk, SOCK_LINGER);
 861                 v.ling.l_linger = sk->sk_lingertime / HZ;
 862                 break;
 863
 864         case SO_BSDCOMPAT:
 865                 sock_warn_obsolete_bsdism("getsockopt");
 866                 break;
 867
 868         case SO_TIMESTAMP:
 869                 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
 870                                 !sock_flag(sk, SOCK_RCVTSTAMPNS);
 871                 break;
 872
 873         case SO_TIMESTAMPNS:
 874                 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
 875                 break;
 876
 877         case SO_TIMESTAMPING:
 878                 v.val = 0;
 879                 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
 880                         v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
 881                 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
 882                         v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
 883                 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
 884                         v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
 885                 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
 886                         v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
 887                 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
 888                         v.val |= SOF_TIMESTAMPING_SOFTWARE;
 889                 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
 890                         v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
 891                 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
 892                         v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
 893                 break;
 894
 895         case SO_RCVTIMEO:
 896                 lv = sizeof(struct timeval);
 897                 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
 898                         v.tm.tv_sec = 0;
 899                         v.tm.tv_usec = 0;
 900                 } else {
 901                         v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
 902                         v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
 903                 }
 904                 break;
 905
 906         case SO_SNDTIMEO:
 907                 lv = sizeof(struct timeval);
 908                 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
 909                         v.tm.tv_sec = 0;
 910                         v.tm.tv_usec = 0;
 911                 } else {
 912                         v.tm.tv_sec = sk->sk_sndtimeo / HZ;
 913                         v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
 914                 }
 915                 break;
 916
 917         case SO_RCVLOWAT:
 918                 v.val = sk->sk_rcvlowat;
 919                 break;
 920
 921         case SO_SNDLOWAT:
 922                 v.val = 1;
 923                 break;
 924
 925         case SO_PASSCRED:
 926                 v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
 927                 break;
 928
 929         case SO_PEERCRED:
 930         {
 931                 struct ucred peercred;
 932                 if (len > sizeof(peercred))
 933                         len = sizeof(peercred);
 934                 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
 935                 if (copy_to_user(optval, &peercred, len))
 936                         return -EFAULT;
 937                 goto lenout;
 938         }
 939
 940         case SO_PEERNAME:
 941         {
 942                 char address[128];
 943
 944                 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
 945                         return -ENOTCONN;
 946                 if (lv < len)
 947                         return -EINVAL;
 948                 if (copy_to_user(optval, address, len))
 949                         return -EFAULT;
 950                 goto lenout;
 951         }
 952
 953         /* Dubious BSD thing... Probably nobody even uses it, but
 954          * the UNIX standard wants it for whatever reason... -DaveM
 955          */
 956         case SO_ACCEPTCONN:
 957                 v.val = sk->sk_state == TCP_LISTEN;
 958                 break;
 959
 960         case SO_PASSSEC:
 961                 v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
 962                 break;
 963
 964         case SO_PEERSEC:
 965                 return security_socket_getpeersec_stream(sock, optval, optlen, len);
 966
 967         case SO_MARK:
 968                 v.val = sk->sk_mark;
 969                 break;
 970
 971         case SO_RXQ_OVFL:
 972                 v.val = !!sock_flag(sk, SOCK_RXQ_OVFL);
 973                 break;
 974
 975         default:
 976                 return -ENOPROTOOPT;
 977         }
 978
 979         if (len > lv)
 980                 len = lv;
 981         if (copy_to_user(optval, &v, len))
 982                 return -EFAULT;
 983 lenout:
 984         if (put_user(len, optlen))
 985                 return -EFAULT;
 986         return 0;
 987 }
 988
 989 /*
 990  * Initialize an sk_lock.
 991  *
 992  * (We also register the sk_lock with the lock validator.)
 993  */
 994 static inline void sock_lock_init(struct sock *sk)
 995 {
 996         sock_lock_init_class_and_name(sk,
 997                         af_family_slock_key_strings[sk->sk_family],
 998                         af_family_slock_keys + sk->sk_family,
 999                         af_family_key_strings[sk->sk_family],
1000                         af_family_keys + sk->sk_family);
1001 }
1002
1003 /*
1004  * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1005  * even temporarly, because of RCU lookups. sk_node should also be left as is.
1006  * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1007  */
1008 static void sock_copy(struct sock *nsk, const struct sock *osk)
1009 {
1010 #ifdef CONFIG_SECURITY_NETWORK
1011         void *sptr = nsk->sk_security;
1012 #endif
1013         memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1014
1015         memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1016                osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1017
1018 #ifdef CONFIG_SECURITY_NETWORK
1019         nsk->sk_security = sptr;
1020         security_sk_clone(osk, nsk);
1021 #endif
1022 }
1023
1024 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1025 {
1026         unsigned long nulls1, nulls2;
1027
1028         nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1029         nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1030         if (nulls1 > nulls2)
1031                 swap(nulls1, nulls2);
1032
1033         if (nulls1 != 0)
1034                 memset((char *)sk, 0, nulls1);
1035         memset((char *)sk + nulls1 + sizeof(void *), 0,
1036                nulls2 - nulls1 - sizeof(void *));
1037         memset((char *)sk + nulls2 + sizeof(void *), 0,
1038                size - nulls2 - sizeof(void *));
1039 }
1040 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1041
1042 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1043                 int family)
1044 {
1045         struct sock *sk;
1046         struct kmem_cache *slab;
1047
1048         slab = prot->slab;
1049         if (slab != NULL) {
1050                 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1051                 if (!sk)
1052                         return sk;
1053                 if (priority & __GFP_ZERO) {
1054                         if (prot->clear_sk)
1055                                 prot->clear_sk(sk, prot->obj_size);
1056                         else
1057                                 sk_prot_clear_nulls(sk, prot->obj_size);
1058                 }
1059         } else
1060                 sk = kmalloc(prot->obj_size, priority);
1061
1062         if (sk != NULL) {
1063                 kmemcheck_annotate_bitfield(sk, flags);
1064
1065                 if (security_sk_alloc(sk, family, priority))
1066                         goto out_free;
1067
1068                 if (!try_module_get(prot->owner))
1069                         goto out_free_sec;
1070                 sk_tx_queue_clear(sk);
1071         }
1072
1073         return sk;
1074
1075 out_free_sec:
1076         security_sk_free(sk);
1077 out_free:
1078         if (slab != NULL)
1079                 kmem_cache_free(slab, sk);
1080         else
1081                 kfree(sk);
1082         return NULL;
1083 }
1084
1085 static void sk_prot_free(struct proto *prot, struct sock *sk)
1086 {
1087         struct kmem_cache *slab;
1088         struct module *owner;
1089
1090         owner = prot->owner;
1091         slab = prot->slab;
1092
1093         security_sk_free(sk);
1094         if (slab != NULL)
1095                 kmem_cache_free(slab, sk);
1096         else
1097                 kfree(sk);
1098         module_put(owner);
1099 }
1100
1101 #ifdef CONFIG_CGROUPS
1102 void sock_update_classid(struct sock *sk)
1103 {
1104         u32 classid;
1105
1106         rcu_read_lock();  /* doing current task, which cannot vanish. */
1107         classid = task_cls_classid(current);
1108         rcu_read_unlock();
1109         if (classid && classid != sk->sk_classid)
1110                 sk->sk_classid = classid;
1111 }
1112 EXPORT_SYMBOL(sock_update_classid);
1113 #endif
1114
1115 /**
1116  *      sk_alloc - All socket objects are allocated here
1117  *      @net: the applicable net namespace
1118  *      @family: protocol family
1119  *      @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1120  *      @prot: struct proto associated with this new sock instance
1121  */
1122 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1123                       struct proto *prot)
1124 {
1125         struct sock *sk;
1126
1127         sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1128         if (sk) {
1129                 sk->sk_family = family;
1130                 /*
1131                  * See comment in struct sock definition to understand
1132                  * why we need sk_prot_creator -acme
1133                  */
1134                 sk->sk_prot = sk->sk_prot_creator = prot;
1135                 sock_lock_init(sk);
1136                 sock_net_set(sk, get_net(net));
1137                 atomic_set(&sk->sk_wmem_alloc, 1);
1138
1139                 sock_update_classid(sk);
1140         }
1141
1142         return sk;
1143 }
1144 EXPORT_SYMBOL(sk_alloc);
1145
1146 static void __sk_free(struct sock *sk)
1147 {
1148         struct sk_filter *filter;
1149
1150         if (sk->sk_destruct)
1151                 sk->sk_destruct(sk);
1152
1153         filter = rcu_dereference_check(sk->sk_filter,
1154                                        atomic_read(&sk->sk_wmem_alloc) == 0);
1155         if (filter) {
1156                 sk_filter_uncharge(sk, filter);
1157                 RCU_INIT_POINTER(sk->sk_filter, NULL);
1158         }
1159
1160         sock_disable_timestamp(sk, SOCK_TIMESTAMP);
1161         sock_disable_timestamp(sk, SOCK_TIMESTAMPING_RX_SOFTWARE);
1162
1163         if (atomic_read(&sk->sk_omem_alloc))
1164                 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
1165                        __func__, atomic_read(&sk->sk_omem_alloc));
1166
1167         if (sk->sk_peer_cred)
1168                 put_cred(sk->sk_peer_cred);
1169         put_pid(sk->sk_peer_pid);
1170         put_net(sock_net(sk));
1171         sk_prot_free(sk->sk_prot_creator, sk);
1172 }
1173
1174 void sk_free(struct sock *sk)
1175 {
1176         /*
1177          * We subtract one from sk_wmem_alloc and can know if
1178          * some packets are still in some tx queue.
1179          * If not null, sock_wfree() will call __sk_free(sk) later
1180          */
1181         if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1182                 __sk_free(sk);
1183 }
1184 EXPORT_SYMBOL(sk_free);
1185
1186 /*
1187  * Last sock_put should drop reference to sk->sk_net. It has already
1188  * been dropped in sk_change_net. Taking reference to stopping namespace
1189  * is not an option.
1190  * Take reference to a socket to remove it from hash _alive_ and after that
1191  * destroy it in the context of init_net.
1192  */
1193 void sk_release_kernel(struct sock *sk)
1194 {
1195         if (sk == NULL || sk->sk_socket == NULL)
1196                 return;
1197
1198         sock_hold(sk);
1199         sock_release(sk->sk_socket);
1200         release_net(sock_net(sk));
1201         sock_net_set(sk, get_net(&init_net));
1202         sock_put(sk);
1203 }
1204 EXPORT_SYMBOL(sk_release_kernel);
1205
1206 struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
1207 {
1208         struct sock *newsk;
1209
1210         newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1211         if (newsk != NULL) {
1212                 struct sk_filter *filter;
1213
1214                 sock_copy(newsk, sk);
1215
1216                 /* SANITY */
1217                 get_net(sock_net(newsk));
1218                 sk_node_init(&newsk->sk_node);
1219                 sock_lock_init(newsk);
1220                 bh_lock_sock(newsk);
1221                 newsk->sk_backlog.head  = newsk->sk_backlog.tail = NULL;
1222                 newsk->sk_backlog.len = 0;
1223
1224                 atomic_set(&newsk->sk_rmem_alloc, 0);
1225                 /*
1226                  * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1227                  */
1228                 atomic_set(&newsk->sk_wmem_alloc, 1);
1229                 atomic_set(&newsk->sk_omem_alloc, 0);
1230                 skb_queue_head_init(&newsk->sk_receive_queue);
1231                 skb_queue_head_init(&newsk->sk_write_queue);
1232 #ifdef CONFIG_NET_DMA
1233                 skb_queue_head_init(&newsk->sk_async_wait_queue);
1234 #endif
1235
1236                 spin_lock_init(&newsk->sk_dst_lock);
1237                 rwlock_init(&newsk->sk_callback_lock);
1238                 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1239                                 af_callback_keys + newsk->sk_family,
1240                                 af_family_clock_key_strings[newsk->sk_family]);
1241
1242                 newsk->sk_dst_cache     = NULL;
1243                 newsk->sk_wmem_queued   = 0;
1244                 newsk->sk_forward_alloc = 0;
1245                 newsk->sk_send_head     = NULL;
1246                 newsk->sk_userlocks     = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1247
1248                 sock_reset_flag(newsk, SOCK_DONE);
1249                 skb_queue_head_init(&newsk->sk_error_queue);
1250
1251                 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1252                 if (filter != NULL)
1253                         sk_filter_charge(newsk, filter);
1254
1255                 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1256                         /* It is still raw copy of parent, so invalidate
1257                          * destructor and make plain sk_free() */
1258                         newsk->sk_destruct = NULL;
1259                         bh_unlock_sock(newsk);
1260                         sk_free(newsk);
1261                         newsk = NULL;
1262                         goto out;
1263                 }
1264
1265                 newsk->sk_err      = 0;
1266                 newsk->sk_priority = 0;
1267                 /*
1268                  * Before updating sk_refcnt, we must commit prior changes to memory
1269                  * (Documentation/RCU/rculist_nulls.txt for details)
1270                  */
1271                 smp_wmb();
1272                 atomic_set(&newsk->sk_refcnt, 2);
1273
1274                 /*
1275                  * Increment the counter in the same struct proto as the master
1276                  * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1277                  * is the same as sk->sk_prot->socks, as this field was copied
1278                  * with memcpy).
1279                  *
1280                  * This _changes_ the previous behaviour, where
1281                  * tcp_create_openreq_child always was incrementing the
1282                  * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1283                  * to be taken into account in all callers. -acme
1284                  */
1285                 sk_refcnt_debug_inc(newsk);
1286                 sk_set_socket(newsk, NULL);
1287                 newsk->sk_wq = NULL;
1288
1289                 if (newsk->sk_prot->sockets_allocated)
1290                         percpu_counter_inc(newsk->sk_prot->sockets_allocated);
1291
1292                 if (sock_flag(newsk, SOCK_TIMESTAMP) ||
1293                     sock_flag(newsk, SOCK_TIMESTAMPING_RX_SOFTWARE))
1294                         net_enable_timestamp();
1295         }
1296 out:
1297         return newsk;
1298 }
1299 EXPORT_SYMBOL_GPL(sk_clone);
1300
1301 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1302 {
1303         __sk_dst_set(sk, dst);
1304         sk->sk_route_caps = dst->dev->features;
1305         if (sk->sk_route_caps & NETIF_F_GSO)
1306                 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1307         sk->sk_route_caps &= ~sk->sk_route_nocaps;
1308         if (sk_can_gso(sk)) {
1309                 if (dst->header_len) {
1310                         sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1311                 } else {
1312                         sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1313                         sk->sk_gso_max_size = dst->dev->gso_max_size;
1314                         sk->sk_gso_max_segs = dst->dev->gso_max_segs;
1315                 }
1316         }
1317 }
1318 EXPORT_SYMBOL_GPL(sk_setup_caps);
1319
1320 void __init sk_init(void)
1321 {
1322         if (totalram_pages <= 4096) {
1323                 sysctl_wmem_max = 32767;
1324                 sysctl_rmem_max = 32767;
1325                 sysctl_wmem_default = 32767;
1326                 sysctl_rmem_default = 32767;
1327         } else if (totalram_pages >= 131072) {
1328                 sysctl_wmem_max = 131071;
1329                 sysctl_rmem_max = 131071;
1330         }
1331 }
1332
1333 /*
1334  *      Simple resource managers for sockets.
1335  */
1336
1337
1338 /*
1339  * Write buffer destructor automatically called from kfree_skb.
1340  */
1341 void sock_wfree(struct sk_buff *skb)
1342 {
1343         struct sock *sk = skb->sk;
1344         unsigned int len = skb->truesize;
1345
1346         if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1347                 /*
1348                  * Keep a reference on sk_wmem_alloc, this will be released
1349                  * after sk_write_space() call
1350                  */
1351                 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1352                 sk->sk_write_space(sk);
1353                 len = 1;
1354         }
1355         /*
1356          * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1357          * could not do because of in-flight packets
1358          */
1359         if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1360                 __sk_free(sk);
1361 }
1362 EXPORT_SYMBOL(sock_wfree);
1363
1364 /*
1365  * Read buffer destructor automatically called from kfree_skb.
1366  */
1367 void sock_rfree(struct sk_buff *skb)
1368 {
1369         struct sock *sk = skb->sk;
1370         unsigned int len = skb->truesize;
1371
1372         atomic_sub(len, &sk->sk_rmem_alloc);
1373         sk_mem_uncharge(sk, len);
1374 }
1375 EXPORT_SYMBOL(sock_rfree);
1376
1377
1378 int sock_i_uid(struct sock *sk)
1379 {
1380         int uid;
1381
1382         read_lock_bh(&sk->sk_callback_lock);
1383         uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1384         read_unlock_bh(&sk->sk_callback_lock);
1385         return uid;
1386 }
1387 EXPORT_SYMBOL(sock_i_uid);
1388
1389 unsigned long sock_i_ino(struct sock *sk)
1390 {
1391         unsigned long ino;
1392
1393         read_lock_bh(&sk->sk_callback_lock);
1394         ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1395         read_unlock_bh(&sk->sk_callback_lock);
1396         return ino;
1397 }
1398 EXPORT_SYMBOL(sock_i_ino);
1399
1400 /*
1401  * Allocate a skb from the socket's send buffer.
1402  */
1403 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1404                              gfp_t priority)
1405 {
1406         if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1407                 struct sk_buff *skb = alloc_skb(size, priority);
1408                 if (skb) {
1409                         skb_set_owner_w(skb, sk);
1410                         return skb;
1411                 }
1412         }
1413         return NULL;
1414 }
1415 EXPORT_SYMBOL(sock_wmalloc);
1416
1417 /*
1418  * Allocate a skb from the socket's receive buffer.
1419  */
1420 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1421                              gfp_t priority)
1422 {
1423         if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1424                 struct sk_buff *skb = alloc_skb(size, priority);
1425                 if (skb) {
1426                         skb_set_owner_r(skb, sk);
1427                         return skb;
1428                 }
1429         }
1430         return NULL;
1431 }
1432
1433 /*
1434  * Allocate a memory block from the socket's option memory buffer.
1435  */
1436 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1437 {
1438         if ((unsigned)size <= sysctl_optmem_max &&
1439             atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1440                 void *mem;
1441                 /* First do the add, to avoid the race if kmalloc
1442                  * might sleep.
1443                  */
1444                 atomic_add(size, &sk->sk_omem_alloc);
1445                 mem = kmalloc(size, priority);
1446                 if (mem)
1447                         return mem;
1448                 atomic_sub(size, &sk->sk_omem_alloc);
1449         }
1450         return NULL;
1451 }
1452 EXPORT_SYMBOL(sock_kmalloc);
1453
1454 /*
1455  * Free an option memory block.
1456  */
1457 void sock_kfree_s(struct sock *sk, void *mem, int size)
1458 {
1459         kfree(mem);
1460         atomic_sub(size, &sk->sk_omem_alloc);
1461 }
1462 EXPORT_SYMBOL(sock_kfree_s);
1463
1464 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1465    I think, these locks should be removed for datagram sockets.
1466  */
1467 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1468 {
1469         DEFINE_WAIT(wait);
1470
1471         clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1472         for (;;) {
1473                 if (!timeo)
1474                         break;
1475                 if (signal_pending(current))
1476                         break;
1477                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1478                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1479                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1480                         break;
1481                 if (sk->sk_shutdown & SEND_SHUTDOWN)
1482                         break;
1483                 if (sk->sk_err)
1484                         break;
1485                 timeo = schedule_timeout(timeo);
1486         }
1487         finish_wait(sk_sleep(sk), &wait);
1488         return timeo;
1489 }
1490
1491
1492 /*
1493  *      Generic send/receive buffer handlers
1494  */
1495
1496 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1497                                      unsigned long data_len, int noblock,
1498                                      int *errcode)
1499 {
1500         struct sk_buff *skb;
1501         gfp_t gfp_mask;
1502         long timeo;
1503         int err;
1504         int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1505
1506         err = -EMSGSIZE;
1507         if (npages > MAX_SKB_FRAGS)
1508                 goto failure;
1509
1510         gfp_mask = sk->sk_allocation;
1511         if (gfp_mask & __GFP_WAIT)
1512                 gfp_mask |= __GFP_REPEAT;
1513
1514         timeo = sock_sndtimeo(sk, noblock);
1515         while (1) {
1516                 err = sock_error(sk);
1517                 if (err != 0)
1518                         goto failure;
1519
1520                 err = -EPIPE;
1521                 if (sk->sk_shutdown & SEND_SHUTDOWN)
1522                         goto failure;
1523
1524                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1525                         skb = alloc_skb(header_len, gfp_mask);
1526                         if (skb) {
1527                                 int i;
1528
1529                                 /* No pages, we're done... */
1530                                 if (!data_len)
1531                                         break;
1532
1533                                 skb->truesize += data_len;
1534                                 skb_shinfo(skb)->nr_frags = npages;
1535                                 for (i = 0; i < npages; i++) {
1536                                         struct page *page;
1537
1538                                         page = alloc_pages(sk->sk_allocation, 0);
1539                                         if (!page) {
1540                                                 err = -ENOBUFS;
1541                                                 skb_shinfo(skb)->nr_frags = i;
1542                                                 kfree_skb(skb);
1543                                                 goto failure;
1544                                         }
1545
1546                                         __skb_fill_page_desc(skb, i,
1547                                                         page, 0,
1548                                                         (data_len >= PAGE_SIZE ?
1549                                                          PAGE_SIZE :
1550                                                          data_len));
1551                                         data_len -= PAGE_SIZE;
1552                                 }
1553
1554                                 /* Full success... */
1555                                 break;
1556                         }
1557                         err = -ENOBUFS;
1558                         goto failure;
1559                 }
1560                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1561                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1562                 err = -EAGAIN;
1563                 if (!timeo)
1564                         goto failure;
1565                 if (signal_pending(current))
1566                         goto interrupted;
1567                 timeo = sock_wait_for_wmem(sk, timeo);
1568         }
1569
1570         skb_set_owner_w(skb, sk);
1571         return skb;
1572
1573 interrupted:
1574         err = sock_intr_errno(timeo);
1575 failure:
1576         *errcode = err;
1577         return NULL;
1578 }
1579 EXPORT_SYMBOL(sock_alloc_send_pskb);
1580
1581 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1582                                     int noblock, int *errcode)
1583 {
1584         return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1585 }
1586 EXPORT_SYMBOL(sock_alloc_send_skb);
1587
1588 static void __lock_sock(struct sock *sk)
1589         __releases(&sk->sk_lock.slock)
1590         __acquires(&sk->sk_lock.slock)
1591 {
1592         DEFINE_WAIT(wait);
1593
1594         for (;;) {
1595                 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1596                                         TASK_UNINTERRUPTIBLE);
1597                 spin_unlock_bh(&sk->sk_lock.slock);
1598                 schedule();
1599                 spin_lock_bh(&sk->sk_lock.slock);
1600                 if (!sock_owned_by_user(sk))
1601                         break;
1602         }
1603         finish_wait(&sk->sk_lock.wq, &wait);
1604 }
1605
1606 static void __release_sock(struct sock *sk)
1607         __releases(&sk->sk_lock.slock)
1608         __acquires(&sk->sk_lock.slock)
1609 {
1610         struct sk_buff *skb = sk->sk_backlog.head;
1611
1612         do {
1613                 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1614                 bh_unlock_sock(sk);
1615
1616                 do {
1617                         struct sk_buff *next = skb->next;
1618
1619                         WARN_ON_ONCE(skb_dst_is_noref(skb));
1620                         skb->next = NULL;
1621                         sk_backlog_rcv(sk, skb);
1622
1623                         /*
1624                          * We are in process context here with softirqs
1625                          * disabled, use cond_resched_softirq() to preempt.
1626                          * This is safe to do because we've taken the backlog
1627                          * queue private:
1628                          */
1629                         cond_resched_softirq();
1630
1631                         skb = next;
1632                 } while (skb != NULL);
1633
1634                 bh_lock_sock(sk);
1635         } while ((skb = sk->sk_backlog.head) != NULL);
1636
1637         /*
1638          * Doing the zeroing here guarantee we can not loop forever
1639          * while a wild producer attempts to flood us.
1640          */
1641         sk->sk_backlog.len = 0;
1642 }
1643
1644 /**
1645  * sk_wait_data - wait for data to arrive at sk_receive_queue
1646  * @sk:    sock to wait on
1647  * @timeo: for how long
1648  *
1649  * Now socket state including sk->sk_err is changed only under lock,
1650  * hence we may omit checks after joining wait queue.
1651  * We check receive queue before schedule() only as optimization;
1652  * it is very likely that release_sock() added new data.
1653  */
1654 int sk_wait_data(struct sock *sk, long *timeo)
1655 {
1656         int rc;
1657         DEFINE_WAIT(wait);
1658
1659         prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1660         set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1661         rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1662         clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1663         finish_wait(sk_sleep(sk), &wait);
1664         return rc;
1665 }
1666 EXPORT_SYMBOL(sk_wait_data);
1667
1668 /**
1669  *      __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1670  *      @sk: socket
1671  *      @size: memory size to allocate
1672  *      @kind: allocation type
1673  *
1674  *      If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1675  *      rmem allocation. This function assumes that protocols which have
1676  *      memory_pressure use sk_wmem_queued as write buffer accounting.
1677  */
1678 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1679 {
1680         struct proto *prot = sk->sk_prot;
1681         int amt = sk_mem_pages(size);
1682         long allocated;
1683
1684         sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1685         allocated = atomic_long_add_return(amt, prot->memory_allocated);
1686
1687         /* Under limit. */
1688         if (allocated <= prot->sysctl_mem[0]) {
1689                 if (prot->memory_pressure && *prot->memory_pressure)
1690                         *prot->memory_pressure = 0;
1691                 return 1;
1692         }
1693
1694         /* Under pressure. */
1695         if (allocated > prot->sysctl_mem[1])
1696                 if (prot->enter_memory_pressure)
1697                         prot->enter_memory_pressure(sk);
1698
1699         /* Over hard limit. */
1700         if (allocated > prot->sysctl_mem[2])
1701                 goto suppress_allocation;
1702
1703         /* guarantee minimum buffer size under pressure */
1704         if (kind == SK_MEM_RECV) {
1705                 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1706                         return 1;
1707         } else { /* SK_MEM_SEND */
1708                 if (sk->sk_type == SOCK_STREAM) {
1709                         if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1710                                 return 1;
1711                 } else if (atomic_read(&sk->sk_wmem_alloc) <
1712                            prot->sysctl_wmem[0])
1713                                 return 1;
1714         }
1715
1716         if (prot->memory_pressure) {
1717                 int alloc;
1718
1719                 if (!*prot->memory_pressure)
1720                         return 1;
1721                 alloc = percpu_counter_read_positive(prot->sockets_allocated);
1722                 if (prot->sysctl_mem[2] > alloc *
1723                     sk_mem_pages(sk->sk_wmem_queued +
1724                                  atomic_read(&sk->sk_rmem_alloc) +
1725                                  sk->sk_forward_alloc))
1726                         return 1;
1727         }
1728
1729 suppress_allocation:
1730
1731         if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1732                 sk_stream_moderate_sndbuf(sk);
1733
1734                 /* Fail only if socket is _under_ its sndbuf.
1735                  * In this case we cannot block, so that we have to fail.
1736                  */
1737                 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1738                         return 1;
1739         }
1740
1741         trace_sock_exceed_buf_limit(sk, prot, allocated);
1742
1743         /* Alas. Undo changes. */
1744         sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1745         atomic_long_sub(amt, prot->memory_allocated);
1746         return 0;
1747 }
1748 EXPORT_SYMBOL(__sk_mem_schedule);
1749
1750 /**
1751  *      __sk_reclaim - reclaim memory_allocated
1752  *      @sk: socket
1753  */
1754 void __sk_mem_reclaim(struct sock *sk)
1755 {
1756         struct proto *prot = sk->sk_prot;
1757
1758         atomic_long_sub(sk->sk_forward_alloc >> SK_MEM_QUANTUM_SHIFT,
1759                    prot->memory_allocated);
1760         sk->sk_forward_alloc &= SK_MEM_QUANTUM - 1;
1761
1762         if (prot->memory_pressure && *prot->memory_pressure &&
1763             (atomic_long_read(prot->memory_allocated) < prot->sysctl_mem[0]))
1764                 *prot->memory_pressure = 0;
1765 }
1766 EXPORT_SYMBOL(__sk_mem_reclaim);
1767
1768
1769 /*
1770  * Set of default routines for initialising struct proto_ops when
1771  * the protocol does not support a particular function. In certain
1772  * cases where it makes no sense for a protocol to have a "do nothing"
1773  * function, some default processing is provided.
1774  */
1775
1776 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1777 {
1778         return -EOPNOTSUPP;
1779 }
1780 EXPORT_SYMBOL(sock_no_bind);
1781
1782 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1783                     int len, int flags)
1784 {
1785         return -EOPNOTSUPP;
1786 }
1787 EXPORT_SYMBOL(sock_no_connect);
1788
1789 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1790 {
1791         return -EOPNOTSUPP;
1792 }
1793 EXPORT_SYMBOL(sock_no_socketpair);
1794
1795 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1796 {
1797         return -EOPNOTSUPP;
1798 }
1799 EXPORT_SYMBOL(sock_no_accept);
1800
1801 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1802                     int *len, int peer)
1803 {
1804         return -EOPNOTSUPP;
1805 }
1806 EXPORT_SYMBOL(sock_no_getname);
1807
1808 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1809 {
1810         return 0;
1811 }
1812 EXPORT_SYMBOL(sock_no_poll);
1813
1814 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1815 {
1816         return -EOPNOTSUPP;
1817 }
1818 EXPORT_SYMBOL(sock_no_ioctl);
1819
1820 int sock_no_listen(struct socket *sock, int backlog)
1821 {
1822         return -EOPNOTSUPP;
1823 }
1824 EXPORT_SYMBOL(sock_no_listen);
1825
1826 int sock_no_shutdown(struct socket *sock, int how)
1827 {
1828         return -EOPNOTSUPP;
1829 }
1830 EXPORT_SYMBOL(sock_no_shutdown);
1831
1832 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1833                     char __user *optval, unsigned int optlen)
1834 {
1835         return -EOPNOTSUPP;
1836 }
1837 EXPORT_SYMBOL(sock_no_setsockopt);
1838
1839 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1840                     char __user *optval, int __user *optlen)
1841 {
1842         return -EOPNOTSUPP;
1843 }
1844 EXPORT_SYMBOL(sock_no_getsockopt);
1845
1846 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1847                     size_t len)
1848 {
1849         return -EOPNOTSUPP;
1850 }
1851 EXPORT_SYMBOL(sock_no_sendmsg);
1852
1853 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1854                     size_t len, int flags)
1855 {
1856         return -EOPNOTSUPP;
1857 }
1858 EXPORT_SYMBOL(sock_no_recvmsg);
1859
1860 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1861 {
1862         /* Mirror missing mmap method error code */
1863         return -ENODEV;
1864 }
1865 EXPORT_SYMBOL(sock_no_mmap);
1866
1867 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1868 {
1869         ssize_t res;
1870         struct msghdr msg = {.msg_flags = flags};
1871         struct kvec iov;
1872         char *kaddr = kmap(page);
1873         iov.iov_base = kaddr + offset;
1874         iov.iov_len = size;
1875         res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1876         kunmap(page);
1877         return res;
1878 }
1879 EXPORT_SYMBOL(sock_no_sendpage);
1880
1881 /*
1882  *      Default Socket Callbacks
1883  */
1884
1885 static void sock_def_wakeup(struct sock *sk)
1886 {
1887         struct socket_wq *wq;
1888
1889         rcu_read_lock();
1890         wq = rcu_dereference(sk->sk_wq);
1891         if (wq_has_sleeper(wq))
1892                 wake_up_interruptible_all(&wq->wait);
1893         rcu_read_unlock();
1894 }
1895
1896 static void sock_def_error_report(struct sock *sk)
1897 {
1898         struct socket_wq *wq;
1899
1900         rcu_read_lock();
1901         wq = rcu_dereference(sk->sk_wq);
1902         if (wq_has_sleeper(wq))
1903                 wake_up_interruptible_poll(&wq->wait, POLLERR);
1904         sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
1905         rcu_read_unlock();
1906 }
1907
1908 static void sock_def_readable(struct sock *sk, int len)
1909 {
1910         struct socket_wq *wq;
1911
1912         rcu_read_lock();
1913         wq = rcu_dereference(sk->sk_wq);
1914         if (wq_has_sleeper(wq))
1915                 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
1916                                                 POLLRDNORM | POLLRDBAND);
1917         sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1918         rcu_read_unlock();
1919 }
1920
1921 static void sock_def_write_space(struct sock *sk)
1922 {
1923         struct socket_wq *wq;
1924
1925         rcu_read_lock();
1926
1927         /* Do not wake up a writer until he can make "significant"
1928          * progress.  --DaveM
1929          */
1930         if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1931                 wq = rcu_dereference(sk->sk_wq);
1932                 if (wq_has_sleeper(wq))
1933                         wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
1934                                                 POLLWRNORM | POLLWRBAND);
1935
1936                 /* Should agree with poll, otherwise some programs break */
1937                 if (sock_writeable(sk))
1938                         sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1939         }
1940
1941         rcu_read_unlock();
1942 }
1943
1944 static void sock_def_destruct(struct sock *sk)
1945 {
1946         kfree(sk->sk_protinfo);
1947 }
1948
1949 void sk_send_sigurg(struct sock *sk)
1950 {
1951         if (sk->sk_socket && sk->sk_socket->file)
1952                 if (send_sigurg(&sk->sk_socket->file->f_owner))
1953                         sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1954 }
1955 EXPORT_SYMBOL(sk_send_sigurg);
1956
1957 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1958                     unsigned long expires)
1959 {
1960         if (!mod_timer(timer, expires))
1961                 sock_hold(sk);
1962 }
1963 EXPORT_SYMBOL(sk_reset_timer);
1964
1965 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1966 {
1967         if (timer_pending(timer) && del_timer(timer))
1968                 __sock_put(sk);
1969 }
1970 EXPORT_SYMBOL(sk_stop_timer);
1971
1972 void sock_init_data(struct socket *sock, struct sock *sk)
1973 {
1974         skb_queue_head_init(&sk->sk_receive_queue);
1975         skb_queue_head_init(&sk->sk_write_queue);
1976         skb_queue_head_init(&sk->sk_error_queue);
1977 #ifdef CONFIG_NET_DMA
1978         skb_queue_head_init(&sk->sk_async_wait_queue);
1979 #endif
1980
1981         sk->sk_send_head        =       NULL;
1982
1983         init_timer(&sk->sk_timer);
1984
1985         sk->sk_allocation       =       GFP_KERNEL;
1986         sk->sk_rcvbuf           =       sysctl_rmem_default;
1987         sk->sk_sndbuf           =       sysctl_wmem_default;
1988         sk->sk_state            =       TCP_CLOSE;
1989         sk_set_socket(sk, sock);
1990
1991         sock_set_flag(sk, SOCK_ZAPPED);
1992
1993         if (sock) {
1994                 sk->sk_type     =       sock->type;
1995                 sk->sk_wq       =       sock->wq;
1996                 sock->sk        =       sk;
1997         } else
1998                 sk->sk_wq       =       NULL;
1999
2000         spin_lock_init(&sk->sk_dst_lock);
2001         rwlock_init(&sk->sk_callback_lock);
2002         lockdep_set_class_and_name(&sk->sk_callback_lock,
2003                         af_callback_keys + sk->sk_family,
2004                         af_family_clock_key_strings[sk->sk_family]);
2005
2006         sk->sk_state_change     =       sock_def_wakeup;
2007         sk->sk_data_ready       =       sock_def_readable;
2008         sk->sk_write_space      =       sock_def_write_space;
2009         sk->sk_error_report     =       sock_def_error_report;
2010         sk->sk_destruct         =       sock_def_destruct;
2011
2012         sk->sk_sndmsg_page      =       NULL;
2013         sk->sk_sndmsg_off       =       0;
2014
2015         sk->sk_peer_pid         =       NULL;
2016         sk->sk_peer_cred        =       NULL;
2017         sk->sk_write_pending    =       0;
2018         sk->sk_rcvlowat         =       1;
2019         sk->sk_rcvtimeo         =       MAX_SCHEDULE_TIMEOUT;
2020         sk->sk_sndtimeo         =       MAX_SCHEDULE_TIMEOUT;
2021
2022         sk->sk_stamp = ktime_set(-1L, 0);
2023
2024         /*
2025          * Before updating sk_refcnt, we must commit prior changes to memory
2026          * (Documentation/RCU/rculist_nulls.txt for details)
2027          */
2028         smp_wmb();
2029         atomic_set(&sk->sk_refcnt, 1);
2030         atomic_set(&sk->sk_drops, 0);
2031 }
2032 EXPORT_SYMBOL(sock_init_data);
2033
2034 void lock_sock_nested(struct sock *sk, int subclass)
2035 {
2036         might_sleep();
2037         spin_lock_bh(&sk->sk_lock.slock);
2038         if (sk->sk_lock.owned)
2039                 __lock_sock(sk);
2040         sk->sk_lock.owned = 1;
2041         spin_unlock(&sk->sk_lock.slock);
2042         /*
2043          * The sk_lock has mutex_lock() semantics here:
2044          */
2045         mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2046         local_bh_enable();
2047 }
2048 EXPORT_SYMBOL(lock_sock_nested);
2049
2050 void release_sock(struct sock *sk)
2051 {
2052         /*
2053          * The sk_lock has mutex_unlock() semantics:
2054          */
2055         mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2056
2057         spin_lock_bh(&sk->sk_lock.slock);
2058         if (sk->sk_backlog.tail)
2059                 __release_sock(sk);
2060         sk->sk_lock.owned = 0;
2061         if (waitqueue_active(&sk->sk_lock.wq))
2062                 wake_up(&sk->sk_lock.wq);
2063         spin_unlock_bh(&sk->sk_lock.slock);
2064 }
2065 EXPORT_SYMBOL(release_sock);
2066
2067 /**
2068  * lock_sock_fast - fast version of lock_sock
2069  * @sk: socket
2070  *
2071  * This version should be used for very small section, where process wont block
2072  * return false if fast path is taken
2073  *   sk_lock.slock locked, owned = 0, BH disabled
2074  * return true if slow path is taken
2075  *   sk_lock.slock unlocked, owned = 1, BH enabled
2076  */
2077 bool lock_sock_fast(struct sock *sk)
2078 {
2079         might_sleep();
2080         spin_lock_bh(&sk->sk_lock.slock);
2081
2082         if (!sk->sk_lock.owned)
2083                 /*
2084                  * Note : We must disable BH
2085                  */
2086                 return false;
2087
2088         __lock_sock(sk);
2089         sk->sk_lock.owned = 1;
2090         spin_unlock(&sk->sk_lock.slock);
2091         /*
2092          * The sk_lock has mutex_lock() semantics here:
2093          */
2094         mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2095         local_bh_enable();
2096         return true;
2097 }
2098 EXPORT_SYMBOL(lock_sock_fast);
2099
2100 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2101 {
2102         struct timeval tv;
2103         if (!sock_flag(sk, SOCK_TIMESTAMP))
2104                 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2105         tv = ktime_to_timeval(sk->sk_stamp);
2106         if (tv.tv_sec == -1)
2107                 return -ENOENT;
2108         if (tv.tv_sec == 0) {
2109                 sk->sk_stamp = ktime_get_real();
2110                 tv = ktime_to_timeval(sk->sk_stamp);
2111         }
2112         return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2113 }
2114 EXPORT_SYMBOL(sock_get_timestamp);
2115
2116 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2117 {
2118         struct timespec ts;
2119         if (!sock_flag(sk, SOCK_TIMESTAMP))
2120                 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2121         ts = ktime_to_timespec(sk->sk_stamp);
2122         if (ts.tv_sec == -1)
2123                 return -ENOENT;
2124         if (ts.tv_sec == 0) {
2125                 sk->sk_stamp = ktime_get_real();
2126                 ts = ktime_to_timespec(sk->sk_stamp);
2127         }
2128         return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2129 }
2130 EXPORT_SYMBOL(sock_get_timestampns);
2131
2132 void sock_enable_timestamp(struct sock *sk, int flag)
2133 {
2134         if (!sock_flag(sk, flag)) {
2135                 sock_set_flag(sk, flag);
2136                 /*
2137                  * we just set one of the two flags which require net
2138                  * time stamping, but time stamping might have been on
2139                  * already because of the other one
2140                  */
2141                 if (!sock_flag(sk,
2142                                 flag == SOCK_TIMESTAMP ?
2143                                 SOCK_TIMESTAMPING_RX_SOFTWARE :
2144                                 SOCK_TIMESTAMP))
2145                         net_enable_timestamp();
2146         }
2147 }
2148
2149 /*
2150  *      Get a socket option on an socket.
2151  *
2152  *      FIX: POSIX 1003.1g is very ambiguous here. It states that
2153  *      asynchronous errors should be reported by getsockopt. We assume
2154  *      this means if you specify SO_ERROR (otherwise whats the point of it).
2155  */
2156 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2157                            char __user *optval, int __user *optlen)
2158 {
2159         struct sock *sk = sock->sk;
2160
2161         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2162 }
2163 EXPORT_SYMBOL(sock_common_getsockopt);
2164
2165 #ifdef CONFIG_COMPAT
2166 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2167                                   char __user *optval, int __user *optlen)
2168 {
2169         struct sock *sk = sock->sk;
2170
2171         if (sk->sk_prot->compat_getsockopt != NULL)
2172                 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2173                                                       optval, optlen);
2174         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2175 }
2176 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2177 #endif
2178
2179 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2180                         struct msghdr *msg, size_t size, int flags)
2181 {
2182         struct sock *sk = sock->sk;
2183         int addr_len = 0;
2184         int err;
2185
2186         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2187                                    flags & ~MSG_DONTWAIT, &addr_len);
2188         if (err >= 0)
2189                 msg->msg_namelen = addr_len;
2190         return err;
2191 }
2192 EXPORT_SYMBOL(sock_common_recvmsg);
2193
2194 /*
2195  *      Set socket options on an inet socket.
2196  */
2197 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2198                            char __user *optval, unsigned int optlen)
2199 {
2200         struct sock *sk = sock->sk;
2201
2202         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2203 }
2204 EXPORT_SYMBOL(sock_common_setsockopt);
2205
2206 #ifdef CONFIG_COMPAT
2207 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2208                                   char __user *optval, unsigned int optlen)
2209 {
2210         struct sock *sk = sock->sk;
2211
2212         if (sk->sk_prot->compat_setsockopt != NULL)
2213                 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2214                                                       optval, optlen);
2215         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2216 }
2217 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2218 #endif
2219
2220 void sk_common_release(struct sock *sk)
2221 {
2222         if (sk->sk_prot->destroy)
2223                 sk->sk_prot->destroy(sk);
2224
2225         /*
2226          * Observation: when sock_common_release is called, processes have
2227          * no access to socket. But net still has.
2228          * Step one, detach it from networking:
2229          *
2230          * A. Remove from hash tables.
2231          */
2232
2233         sk->sk_prot->unhash(sk);
2234
2235         /*
2236          * In this point socket cannot receive new packets, but it is possible
2237          * that some packets are in flight because some CPU runs receiver and
2238          * did hash table lookup before we unhashed socket. They will achieve
2239          * receive queue and will be purged by socket destructor.
2240          *
2241          * Also we still have packets pending on receive queue and probably,
2242          * our own packets waiting in device queues. sock_destroy will drain
2243          * receive queue, but transmitted packets will delay socket destruction
2244          * until the last reference will be released.
2245          */
2246
2247         sock_orphan(sk);
2248
2249         xfrm_sk_free_policy(sk);
2250
2251         sk_refcnt_debug_release(sk);
2252         sock_put(sk);
2253 }
2254 EXPORT_SYMBOL(sk_common_release);
2255
2256 static DEFINE_RWLOCK(proto_list_lock);
2257 static LIST_HEAD(proto_list);
2258
2259 #ifdef CONFIG_PROC_FS
2260 #define PROTO_INUSE_NR  64      /* should be enough for the first time */
2261 struct prot_inuse {
2262         int val[PROTO_INUSE_NR];
2263 };
2264
2265 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2266
2267 #ifdef CONFIG_NET_NS
2268 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2269 {
2270         __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2271 }
2272 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2273
2274 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2275 {
2276         int cpu, idx = prot->inuse_idx;
2277         int res = 0;
2278
2279         for_each_possible_cpu(cpu)
2280                 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2281
2282         return res >= 0 ? res : 0;
2283 }
2284 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2285
2286 static int __net_init sock_inuse_init_net(struct net *net)
2287 {
2288         net->core.inuse = alloc_percpu(struct prot_inuse);
2289         return net->core.inuse ? 0 : -ENOMEM;
2290 }
2291
2292 static void __net_exit sock_inuse_exit_net(struct net *net)
2293 {
2294         free_percpu(net->core.inuse);
2295 }
2296
2297 static struct pernet_operations net_inuse_ops = {
2298         .init = sock_inuse_init_net,
2299         .exit = sock_inuse_exit_net,
2300 };
2301
2302 static __init int net_inuse_init(void)
2303 {
2304         if (register_pernet_subsys(&net_inuse_ops))
2305                 panic("Cannot initialize net inuse counters");
2306
2307         return 0;
2308 }
2309
2310 core_initcall(net_inuse_init);
2311 #else
2312 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2313
2314 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2315 {
2316         __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2317 }
2318 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2319
2320 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2321 {
2322         int cpu, idx = prot->inuse_idx;
2323         int res = 0;
2324
2325         for_each_possible_cpu(cpu)
2326                 res += per_cpu(prot_inuse, cpu).val[idx];
2327
2328         return res >= 0 ? res : 0;
2329 }
2330 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2331 #endif
2332
2333 static void assign_proto_idx(struct proto *prot)
2334 {
2335         prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2336
2337         if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2338                 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2339                 return;
2340         }
2341
2342         set_bit(prot->inuse_idx, proto_inuse_idx);
2343 }
2344
2345 static void release_proto_idx(struct proto *prot)
2346 {
2347         if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2348                 clear_bit(prot->inuse_idx, proto_inuse_idx);
2349 }
2350 #else
2351 static inline void assign_proto_idx(struct proto *prot)
2352 {
2353 }
2354
2355 static inline void release_proto_idx(struct proto *prot)
2356 {
2357 }
2358 #endif
2359
2360 int proto_register(struct proto *prot, int alloc_slab)
2361 {
2362         if (alloc_slab) {
2363                 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2364                                         SLAB_HWCACHE_ALIGN | prot->slab_flags,
2365                                         NULL);
2366
2367                 if (prot->slab == NULL) {
2368                         printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2369                                prot->name);
2370                         goto out;
2371                 }
2372
2373                 if (prot->rsk_prot != NULL) {
2374                         prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
2375                         if (prot->rsk_prot->slab_name == NULL)
2376                                 goto out_free_sock_slab;
2377
2378                         prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2379                                                                  prot->rsk_prot->obj_size, 0,
2380                                                                  SLAB_HWCACHE_ALIGN, NULL);
2381
2382                         if (prot->rsk_prot->slab == NULL) {
2383                                 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2384                                        prot->name);
2385                                 goto out_free_request_sock_slab_name;
2386                         }
2387                 }
2388
2389                 if (prot->twsk_prot != NULL) {
2390                         prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2391
2392                         if (prot->twsk_prot->twsk_slab_name == NULL)
2393                                 goto out_free_request_sock_slab;
2394
2395                         prot->twsk_prot->twsk_slab =
2396                                 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2397                                                   prot->twsk_prot->twsk_obj_size,
2398                                                   0,
2399                                                   SLAB_HWCACHE_ALIGN |
2400                                                         prot->slab_flags,
2401                                                   NULL);
2402                         if (prot->twsk_prot->twsk_slab == NULL)
2403                                 goto out_free_timewait_sock_slab_name;
2404                 }
2405         }
2406
2407         write_lock(&proto_list_lock);
2408         list_add(&prot->node, &proto_list);
2409         assign_proto_idx(prot);
2410         write_unlock(&proto_list_lock);
2411         return 0;
2412
2413 out_free_timewait_sock_slab_name:
2414         kfree(prot->twsk_prot->twsk_slab_name);
2415 out_free_request_sock_slab:
2416         if (prot->rsk_prot && prot->rsk_prot->slab) {
2417                 kmem_cache_destroy(prot->rsk_prot->slab);
2418                 prot->rsk_prot->slab = NULL;
2419         }
2420 out_free_request_sock_slab_name:
2421         if (prot->rsk_prot)
2422                 kfree(prot->rsk_prot->slab_name);
2423 out_free_sock_slab:
2424         kmem_cache_destroy(prot->slab);
2425         prot->slab = NULL;
2426 out:
2427         return -ENOBUFS;
2428 }
2429 EXPORT_SYMBOL(proto_register);
2430
2431 void proto_unregister(struct proto *prot)
2432 {
2433         write_lock(&proto_list_lock);
2434         release_proto_idx(prot);
2435         list_del(&prot->node);
2436         write_unlock(&proto_list_lock);
2437
2438         if (prot->slab != NULL) {
2439                 kmem_cache_destroy(prot->slab);
2440                 prot->slab = NULL;
2441         }
2442
2443         if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2444                 kmem_cache_destroy(prot->rsk_prot->slab);
2445                 kfree(prot->rsk_prot->slab_name);
2446                 prot->rsk_prot->slab = NULL;
2447         }
2448
2449         if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2450                 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2451                 kfree(prot->twsk_prot->twsk_slab_name);
2452                 prot->twsk_prot->twsk_slab = NULL;
2453         }
2454 }
2455 EXPORT_SYMBOL(proto_unregister);
2456
2457 #ifdef CONFIG_PROC_FS
2458 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2459         __acquires(proto_list_lock)
2460 {
2461         read_lock(&proto_list_lock);
2462         return seq_list_start_head(&proto_list, *pos);
2463 }
2464
2465 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2466 {
2467         return seq_list_next(v, &proto_list, pos);
2468 }
2469
2470 static void proto_seq_stop(struct seq_file *seq, void *v)
2471         __releases(proto_list_lock)
2472 {
2473         read_unlock(&proto_list_lock);
2474 }
2475
2476 static char proto_method_implemented(const void *method)
2477 {
2478         return method == NULL ? 'n' : 'y';
2479 }
2480
2481 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2482 {
2483         seq_printf(seq, "%-9s %4u %6d  %6ld   %-3s %6u   %-3s  %-10s "
2484                         "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2485                    proto->name,
2486                    proto->obj_size,
2487                    sock_prot_inuse_get(seq_file_net(seq), proto),
2488                    proto->memory_allocated != NULL ? atomic_long_read(proto->memory_allocated) : -1L,
2489                    proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2490                    proto->max_header,
2491                    proto->slab == NULL ? "no" : "yes",
2492                    module_name(proto->owner),
2493                    proto_method_implemented(proto->close),
2494                    proto_method_implemented(proto->connect),
2495                    proto_method_implemented(proto->disconnect),
2496                    proto_method_implemented(proto->accept),
2497                    proto_method_implemented(proto->ioctl),
2498                    proto_method_implemented(proto->init),
2499                    proto_method_implemented(proto->destroy),
2500                    proto_method_implemented(proto->shutdown),
2501                    proto_method_implemented(proto->setsockopt),
2502                    proto_method_implemented(proto->getsockopt),
2503                    proto_method_implemented(proto->sendmsg),
2504                    proto_method_implemented(proto->recvmsg),
2505                    proto_method_implemented(proto->sendpage),
2506                    proto_method_implemented(proto->bind),
2507                    proto_method_implemented(proto->backlog_rcv),
2508                    proto_method_implemented(proto->hash),
2509                    proto_method_implemented(proto->unhash),
2510                    proto_method_implemented(proto->get_port),
2511                    proto_method_implemented(proto->enter_memory_pressure));
2512 }
2513
2514 static int proto_seq_show(struct seq_file *seq, void *v)
2515 {
2516         if (v == &proto_list)
2517                 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2518                            "protocol",
2519                            "size",
2520                            "sockets",
2521                            "memory",
2522                            "press",
2523                            "maxhdr",
2524                            "slab",
2525                            "module",
2526                            "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2527         else
2528                 proto_seq_printf(seq, list_entry(v, struct proto, node));
2529         return 0;
2530 }
2531
2532 static const struct seq_operations proto_seq_ops = {
2533         .start  = proto_seq_start,
2534         .next   = proto_seq_next,
2535         .stop   = proto_seq_stop,
2536         .show   = proto_seq_show,
2537 };
2538
2539 static int proto_seq_open(struct inode *inode, struct file *file)
2540 {
2541         return seq_open_net(inode, file, &proto_seq_ops,
2542                             sizeof(struct seq_net_private));
2543 }
2544
2545 static const struct file_operations proto_seq_fops = {
2546         .owner          = THIS_MODULE,
2547         .open           = proto_seq_open,
2548         .read           = seq_read,
2549         .llseek         = seq_lseek,
2550         .release        = seq_release_net,
2551 };
2552
2553 static __net_init int proto_init_net(struct net *net)
2554 {
2555         if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2556                 return -ENOMEM;
2557
2558         return 0;
2559 }
2560
2561 static __net_exit void proto_exit_net(struct net *net)
2562 {
2563         proc_net_remove(net, "protocols");
2564 }
2565
2566
2567 static __net_initdata struct pernet_operations proto_net_ops = {
2568         .init = proto_init_net,
2569         .exit = proto_exit_net,
2570 };
2571
2572 static int __init proto_init(void)
2573 {
2574         return register_pernet_subsys(&proto_net_ops);
2575 }
2576
2577 subsys_initcall(proto_init);
2578
2579 #endif /* PROC_FS */