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 skb_len;
 287         unsigned long flags;
 288         struct sk_buff_head *list = &sk->sk_receive_queue;
 289
 290         if (atomic_read(&sk->sk_rmem_alloc) >= sk->sk_rcvbuf) {
 291                 atomic_inc(&sk->sk_drops);
 292                 trace_sock_rcvqueue_full(sk, skb);
 293                 return -ENOMEM;
 294         }
 295
 296         if (!sk_rmem_schedule(sk, skb->truesize)) {
 297                 atomic_inc(&sk->sk_drops);
 298                 return -ENOBUFS;
 299         }
 300
 301         skb->dev = NULL;
 302         skb_set_owner_r(skb, sk);
 303
 304         /* Cache the SKB length before we tack it onto the receive
 305          * queue.  Once it is added it no longer belongs to us and
 306          * may be freed by other threads of control pulling packets
 307          * from the queue.
 308          */
 309         skb_len = skb->len;
 310
 311         /* we escape from rcu protected region, make sure we dont leak
 312          * a norefcounted dst
 313          */
 314         skb_dst_force(skb);
 315
 316         spin_lock_irqsave(&list->lock, flags);
 317         skb->dropcount = atomic_read(&sk->sk_drops);
 318         __skb_queue_tail(list, skb);
 319         spin_unlock_irqrestore(&list->lock, flags);
 320
 321         if (!sock_flag(sk, SOCK_DEAD))
 322                 sk->sk_data_ready(sk, skb_len);
 323         return 0;
 324 }
 325 EXPORT_SYMBOL(__sock_queue_rcv_skb);
 326
 327 int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
 328 {
 329         int err;
 330
 331         err = sk_filter(sk, skb);
 332         if (err)
 333                 return err;
 334
 335         return __sock_queue_rcv_skb(sk, skb);
 336 }
 337 EXPORT_SYMBOL(sock_queue_rcv_skb);
 338
 339 int __sk_receive_skb(struct sock *sk, struct sk_buff *skb,
 340                      const int nested, unsigned int trim_cap)
 341 {
 342         int rc = NET_RX_SUCCESS;
 343
 344         if (sk_filter_trim_cap(sk, skb, trim_cap))
 345                 goto discard_and_relse;
 346
 347         skb->dev = NULL;
 348
 349         if (sk_rcvqueues_full(sk, skb)) {
 350                 atomic_inc(&sk->sk_drops);
 351                 goto discard_and_relse;
 352         }
 353         if (nested)
 354                 bh_lock_sock_nested(sk);
 355         else
 356                 bh_lock_sock(sk);
 357         if (!sock_owned_by_user(sk)) {
 358                 /*
 359                  * trylock + unlock semantics:
 360                  */
 361                 mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
 362
 363                 rc = sk_backlog_rcv(sk, skb);
 364
 365                 mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
 366         } else if (sk_add_backlog(sk, skb)) {
 367                 bh_unlock_sock(sk);
 368                 atomic_inc(&sk->sk_drops);
 369                 goto discard_and_relse;
 370         }
 371
 372         bh_unlock_sock(sk);
 373 out:
 374         sock_put(sk);
 375         return rc;
 376 discard_and_relse:
 377         kfree_skb(skb);
 378         goto out;
 379 }
 380 EXPORT_SYMBOL(__sk_receive_skb);
 381
 382 void sk_reset_txq(struct sock *sk)
 383 {
 384         sk_tx_queue_clear(sk);
 385 }
 386 EXPORT_SYMBOL(sk_reset_txq);
 387
 388 struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
 389 {
 390         struct dst_entry *dst = __sk_dst_get(sk);
 391
 392         if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
 393                 sk_tx_queue_clear(sk);
 394                 RCU_INIT_POINTER(sk->sk_dst_cache, NULL);
 395                 dst_release(dst);
 396                 return NULL;
 397         }
 398
 399         return dst;
 400 }
 401 EXPORT_SYMBOL(__sk_dst_check);
 402
 403 struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
 404 {
 405         struct dst_entry *dst = sk_dst_get(sk);
 406
 407         if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
 408                 sk_dst_reset(sk);
 409                 dst_release(dst);
 410                 return NULL;
 411         }
 412
 413         return dst;
 414 }
 415 EXPORT_SYMBOL(sk_dst_check);
 416
 417 static int sock_bindtodevice(struct sock *sk, char __user *optval, int optlen)
 418 {
 419         int ret = -ENOPROTOOPT;
 420 #ifdef CONFIG_NETDEVICES
 421         struct net *net = sock_net(sk);
 422         char devname[IFNAMSIZ];
 423         int index;
 424
 425         /* Sorry... */
 426         ret = -EPERM;
 427         if (!capable(CAP_NET_RAW))
 428                 goto out;
 429
 430         ret = -EINVAL;
 431         if (optlen < 0)
 432                 goto out;
 433
 434         /* Bind this socket to a particular device like "eth0",
 435          * as specified in the passed interface name. If the
 436          * name is "" or the option length is zero the socket
 437          * is not bound.
 438          */
 439         if (optlen > IFNAMSIZ - 1)
 440                 optlen = IFNAMSIZ - 1;
 441         memset(devname, 0, sizeof(devname));
 442
 443         ret = -EFAULT;
 444         if (copy_from_user(devname, optval, optlen))
 445                 goto out;
 446
 447         index = 0;
 448         if (devname[0] != '\0') {
 449                 struct net_device *dev;
 450
 451                 rcu_read_lock();
 452                 dev = dev_get_by_name_rcu(net, devname);
 453                 if (dev)
 454                         index = dev->ifindex;
 455                 rcu_read_unlock();
 456                 ret = -ENODEV;
 457                 if (!dev)
 458                         goto out;
 459         }
 460
 461         lock_sock(sk);
 462         sk->sk_bound_dev_if = index;
 463         sk_dst_reset(sk);
 464         release_sock(sk);
 465
 466         ret = 0;
 467
 468 out:
 469 #endif
 470
 471         return ret;
 472 }
 473
 474 static inline void sock_valbool_flag(struct sock *sk, int bit, int valbool)
 475 {
 476         if (valbool)
 477                 sock_set_flag(sk, bit);
 478         else
 479                 sock_reset_flag(sk, bit);
 480 }
 481
 482 /*
 483  *      This is meant for all protocols to use and covers goings on
 484  *      at the socket level. Everything here is generic.
 485  */
 486
 487 int sock_setsockopt(struct socket *sock, int level, int optname,
 488                     char __user *optval, unsigned int optlen)
 489 {
 490         struct sock *sk = sock->sk;
 491         int val;
 492         int valbool;
 493         struct linger ling;
 494         int ret = 0;
 495
 496         /*
 497          *      Options without arguments
 498          */
 499
 500         if (optname == SO_BINDTODEVICE)
 501                 return sock_bindtodevice(sk, optval, optlen);
 502
 503         if (optlen < sizeof(int))
 504                 return -EINVAL;
 505
 506         if (get_user(val, (int __user *)optval))
 507                 return -EFAULT;
 508
 509         valbool = val ? 1 : 0;
 510
 511         lock_sock(sk);
 512
 513         switch (optname) {
 514         case SO_DEBUG:
 515                 if (val && !capable(CAP_NET_ADMIN))
 516                         ret = -EACCES;
 517                 else
 518                         sock_valbool_flag(sk, SOCK_DBG, valbool);
 519                 break;
 520         case SO_REUSEADDR:
 521                 sk->sk_reuse = valbool;
 522                 break;
 523         case SO_TYPE:
 524         case SO_PROTOCOL:
 525         case SO_DOMAIN:
 526         case SO_ERROR:
 527                 ret = -ENOPROTOOPT;
 528                 break;
 529         case SO_DONTROUTE:
 530                 sock_valbool_flag(sk, SOCK_LOCALROUTE, valbool);
 531                 break;
 532         case SO_BROADCAST:
 533                 sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
 534                 break;
 535         case SO_SNDBUF:
 536                 /* Don't error on this BSD doesn't and if you think
 537                  * about it this is right. Otherwise apps have to
 538                  * play 'guess the biggest size' games. RCVBUF/SNDBUF
 539                  * are treated in BSD as hints
 540                  */
 541                 val = min_t(u32, val, sysctl_wmem_max);
 542 set_sndbuf:
 543                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
 544                 sk->sk_sndbuf = max_t(int, val * 2, SOCK_MIN_SNDBUF);
 545                 /* Wake up sending tasks if we upped the value. */
 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                 val = min_t(u32, val, sysctl_rmem_max);
 563 set_rcvbuf:
 564                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
 565                 /*
 566                  * We double it on the way in to account for
 567                  * "struct sk_buff" etc. overhead.   Applications
 568                  * assume that the SO_RCVBUF setting they make will
 569                  * allow that much actual data to be received on that
 570                  * socket.
 571                  *
 572                  * Applications are unaware that "struct sk_buff" and
 573                  * other overheads allocate from the receive buffer
 574                  * during socket buffer allocation.
 575                  *
 576                  * And after considering the possible alternatives,
 577                  * returning the value we actually used in getsockopt
 578                  * is the most desirable behavior.
 579                  */
 580                 sk->sk_rcvbuf = max_t(int, val * 2, SOCK_MIN_RCVBUF);
 581                 break;
 582
 583         case SO_RCVBUFFORCE:
 584                 if (!capable(CAP_NET_ADMIN)) {
 585                         ret = -EPERM;
 586                         break;
 587                 }
 588                 goto set_rcvbuf;
 589
 590         case SO_KEEPALIVE:
 591 #ifdef CONFIG_INET
 592                 if (sk->sk_protocol == IPPROTO_TCP &&
 593                     sk->sk_type == SOCK_STREAM)
 594                         tcp_set_keepalive(sk, valbool);
 595 #endif
 596                 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
 597                 break;
 598
 599         case SO_OOBINLINE:
 600                 sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
 601                 break;
 602
 603         case SO_NO_CHECK:
 604                 sk->sk_no_check = valbool;
 605                 break;
 606
 607         case SO_PRIORITY:
 608                 if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
 609                         sk->sk_priority = val;
 610                 else
 611                         ret = -EPERM;
 612                 break;
 613
 614         case SO_LINGER:
 615                 if (optlen < sizeof(ling)) {
 616                         ret = -EINVAL;  /* 1003.1g */
 617                         break;
 618                 }
 619                 if (copy_from_user(&ling, optval, sizeof(ling))) {
 620                         ret = -EFAULT;
 621                         break;
 622                 }
 623                 if (!ling.l_onoff)
 624                         sock_reset_flag(sk, SOCK_LINGER);
 625                 else {
 626 #if (BITS_PER_LONG == 32)
 627                         if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
 628                                 sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
 629                         else
 630 #endif
 631                                 sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
 632                         sock_set_flag(sk, SOCK_LINGER);
 633                 }
 634                 break;
 635
 636         case SO_BSDCOMPAT:
 637                 sock_warn_obsolete_bsdism("setsockopt");
 638                 break;
 639
 640         case SO_PASSCRED:
 641                 if (valbool)
 642                         set_bit(SOCK_PASSCRED, &sock->flags);
 643                 else
 644                         clear_bit(SOCK_PASSCRED, &sock->flags);
 645                 break;
 646
 647         case SO_TIMESTAMP:
 648         case SO_TIMESTAMPNS:
 649                 if (valbool)  {
 650                         if (optname == SO_TIMESTAMP)
 651                                 sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
 652                         else
 653                                 sock_set_flag(sk, SOCK_RCVTSTAMPNS);
 654                         sock_set_flag(sk, SOCK_RCVTSTAMP);
 655                         sock_enable_timestamp(sk, SOCK_TIMESTAMP);
 656                 } else {
 657                         sock_reset_flag(sk, SOCK_RCVTSTAMP);
 658                         sock_reset_flag(sk, SOCK_RCVTSTAMPNS);
 659                 }
 660                 break;
 661
 662         case SO_TIMESTAMPING:
 663                 if (val & ~SOF_TIMESTAMPING_MASK) {
 664                         ret = -EINVAL;
 665                         break;
 666                 }
 667                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE,
 668                                   val & SOF_TIMESTAMPING_TX_HARDWARE);
 669                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE,
 670                                   val & SOF_TIMESTAMPING_TX_SOFTWARE);
 671                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE,
 672                                   val & SOF_TIMESTAMPING_RX_HARDWARE);
 673                 if (val & SOF_TIMESTAMPING_RX_SOFTWARE)
 674                         sock_enable_timestamp(sk,
 675                                               SOCK_TIMESTAMPING_RX_SOFTWARE);
 676                 else
 677                         sock_disable_timestamp(sk,
 678                                                SOCK_TIMESTAMPING_RX_SOFTWARE);
 679                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SOFTWARE,
 680                                   val & SOF_TIMESTAMPING_SOFTWARE);
 681                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE,
 682                                   val & SOF_TIMESTAMPING_SYS_HARDWARE);
 683                 sock_valbool_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE,
 684                                   val & SOF_TIMESTAMPING_RAW_HARDWARE);
 685                 break;
 686
 687         case SO_RCVLOWAT:
 688                 if (val < 0)
 689                         val = INT_MAX;
 690                 sk->sk_rcvlowat = val ? : 1;
 691                 break;
 692
 693         case SO_RCVTIMEO:
 694                 ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
 695                 break;
 696
 697         case SO_SNDTIMEO:
 698                 ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
 699                 break;
 700
 701         case SO_ATTACH_FILTER:
 702                 ret = -EINVAL;
 703                 if (optlen == sizeof(struct sock_fprog)) {
 704                         struct sock_fprog fprog;
 705
 706                         ret = -EFAULT;
 707                         if (copy_from_user(&fprog, optval, sizeof(fprog)))
 708                                 break;
 709
 710                         ret = sk_attach_filter(&fprog, sk);
 711                 }
 712                 break;
 713
 714         case SO_DETACH_FILTER:
 715                 ret = sk_detach_filter(sk);
 716                 break;
 717
 718         case SO_PASSSEC:
 719                 if (valbool)
 720                         set_bit(SOCK_PASSSEC, &sock->flags);
 721                 else
 722                         clear_bit(SOCK_PASSSEC, &sock->flags);
 723                 break;
 724         case SO_MARK:
 725                 if (!capable(CAP_NET_ADMIN))
 726                         ret = -EPERM;
 727                 else
 728                         sk->sk_mark = val;
 729                 break;
 730
 731                 /* We implement the SO_SNDLOWAT etc to
 732                    not be settable (1003.1g 5.3) */
 733         case SO_RXQ_OVFL:
 734                 sock_valbool_flag(sk, SOCK_RXQ_OVFL, valbool);
 735                 break;
 736         default:
 737                 ret = -ENOPROTOOPT;
 738                 break;
 739         }
 740         release_sock(sk);
 741         return ret;
 742 }
 743 EXPORT_SYMBOL(sock_setsockopt);
 744
 745
 746 void cred_to_ucred(struct pid *pid, const struct cred *cred,
 747                    struct ucred *ucred)
 748 {
 749         ucred->pid = pid_vnr(pid);
 750         ucred->uid = ucred->gid = -1;
 751         if (cred) {
 752                 struct user_namespace *current_ns = current_user_ns();
 753
 754                 ucred->uid = user_ns_map_uid(current_ns, cred, cred->euid);
 755                 ucred->gid = user_ns_map_gid(current_ns, cred, cred->egid);
 756         }
 757 }
 758 EXPORT_SYMBOL_GPL(cred_to_ucred);
 759
 760 void cred_real_to_ucred(struct pid *pid, const struct cred *cred,
 761                         struct ucred *ucred)
 762 {
 763         ucred->pid = pid_vnr(pid);
 764         ucred->uid = ucred->gid = -1;
 765         if (cred) {
 766                 struct user_namespace *current_ns = current_user_ns();
 767
 768                 ucred->uid = user_ns_map_uid(current_ns, cred, cred->uid);
 769                 ucred->gid = user_ns_map_gid(current_ns, cred, cred->gid);
 770         }
 771 }
 772 EXPORT_SYMBOL_GPL(cred_real_to_ucred);
 773
 774 int sock_getsockopt(struct socket *sock, int level, int optname,
 775                     char __user *optval, int __user *optlen)
 776 {
 777         struct sock *sk = sock->sk;
 778
 779         union {
 780                 int val;
 781                 struct linger ling;
 782                 struct timeval tm;
 783         } v;
 784
 785         int lv = sizeof(int);
 786         int len;
 787
 788         if (get_user(len, optlen))
 789                 return -EFAULT;
 790         if (len < 0)
 791                 return -EINVAL;
 792
 793         memset(&v, 0, sizeof(v));
 794
 795         switch (optname) {
 796         case SO_DEBUG:
 797                 v.val = sock_flag(sk, SOCK_DBG);
 798                 break;
 799
 800         case SO_DONTROUTE:
 801                 v.val = sock_flag(sk, SOCK_LOCALROUTE);
 802                 break;
 803
 804         case SO_BROADCAST:
 805                 v.val = !!sock_flag(sk, SOCK_BROADCAST);
 806                 break;
 807
 808         case SO_SNDBUF:
 809                 v.val = sk->sk_sndbuf;
 810                 break;
 811
 812         case SO_RCVBUF:
 813                 v.val = sk->sk_rcvbuf;
 814                 break;
 815
 816         case SO_REUSEADDR:
 817                 v.val = sk->sk_reuse;
 818                 break;
 819
 820         case SO_KEEPALIVE:
 821                 v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
 822                 break;
 823
 824         case SO_TYPE:
 825                 v.val = sk->sk_type;
 826                 break;
 827
 828         case SO_PROTOCOL:
 829                 v.val = sk->sk_protocol;
 830                 break;
 831
 832         case SO_DOMAIN:
 833                 v.val = sk->sk_family;
 834                 break;
 835
 836         case SO_ERROR:
 837                 v.val = -sock_error(sk);
 838                 if (v.val == 0)
 839                         v.val = xchg(&sk->sk_err_soft, 0);
 840                 break;
 841
 842         case SO_OOBINLINE:
 843                 v.val = !!sock_flag(sk, SOCK_URGINLINE);
 844                 break;
 845
 846         case SO_NO_CHECK:
 847                 v.val = sk->sk_no_check;
 848                 break;
 849
 850         case SO_PRIORITY:
 851                 v.val = sk->sk_priority;
 852                 break;
 853
 854         case SO_LINGER:
 855                 lv              = sizeof(v.ling);
 856                 v.ling.l_onoff  = !!sock_flag(sk, SOCK_LINGER);
 857                 v.ling.l_linger = sk->sk_lingertime / HZ;
 858                 break;
 859
 860         case SO_BSDCOMPAT:
 861                 sock_warn_obsolete_bsdism("getsockopt");
 862                 break;
 863
 864         case SO_TIMESTAMP:
 865                 v.val = sock_flag(sk, SOCK_RCVTSTAMP) &&
 866                                 !sock_flag(sk, SOCK_RCVTSTAMPNS);
 867                 break;
 868
 869         case SO_TIMESTAMPNS:
 870                 v.val = sock_flag(sk, SOCK_RCVTSTAMPNS);
 871                 break;
 872
 873         case SO_TIMESTAMPING:
 874                 v.val = 0;
 875                 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_HARDWARE))
 876                         v.val |= SOF_TIMESTAMPING_TX_HARDWARE;
 877                 if (sock_flag(sk, SOCK_TIMESTAMPING_TX_SOFTWARE))
 878                         v.val |= SOF_TIMESTAMPING_TX_SOFTWARE;
 879                 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_HARDWARE))
 880                         v.val |= SOF_TIMESTAMPING_RX_HARDWARE;
 881                 if (sock_flag(sk, SOCK_TIMESTAMPING_RX_SOFTWARE))
 882                         v.val |= SOF_TIMESTAMPING_RX_SOFTWARE;
 883                 if (sock_flag(sk, SOCK_TIMESTAMPING_SOFTWARE))
 884                         v.val |= SOF_TIMESTAMPING_SOFTWARE;
 885                 if (sock_flag(sk, SOCK_TIMESTAMPING_SYS_HARDWARE))
 886                         v.val |= SOF_TIMESTAMPING_SYS_HARDWARE;
 887                 if (sock_flag(sk, SOCK_TIMESTAMPING_RAW_HARDWARE))
 888                         v.val |= SOF_TIMESTAMPING_RAW_HARDWARE;
 889                 break;
 890
 891         case SO_RCVTIMEO:
 892                 lv = sizeof(struct timeval);
 893                 if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
 894                         v.tm.tv_sec = 0;
 895                         v.tm.tv_usec = 0;
 896                 } else {
 897                         v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
 898                         v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
 899                 }
 900                 break;
 901
 902         case SO_SNDTIMEO:
 903                 lv = sizeof(struct timeval);
 904                 if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
 905                         v.tm.tv_sec = 0;
 906                         v.tm.tv_usec = 0;
 907                 } else {
 908                         v.tm.tv_sec = sk->sk_sndtimeo / HZ;
 909                         v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
 910                 }
 911                 break;
 912
 913         case SO_RCVLOWAT:
 914                 v.val = sk->sk_rcvlowat;
 915                 break;
 916
 917         case SO_SNDLOWAT:
 918                 v.val = 1;
 919                 break;
 920
 921         case SO_PASSCRED:
 922                 v.val = !!test_bit(SOCK_PASSCRED, &sock->flags);
 923                 break;
 924
 925         case SO_PEERCRED:
 926         {
 927                 struct ucred peercred;
 928                 if (len > sizeof(peercred))
 929                         len = sizeof(peercred);
 930                 cred_to_ucred(sk->sk_peer_pid, sk->sk_peer_cred, &peercred);
 931                 if (copy_to_user(optval, &peercred, len))
 932                         return -EFAULT;
 933                 goto lenout;
 934         }
 935
 936         case SO_PEERNAME:
 937         {
 938                 char address[128];
 939
 940                 if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
 941                         return -ENOTCONN;
 942                 if (lv < len)
 943                         return -EINVAL;
 944                 if (copy_to_user(optval, address, len))
 945                         return -EFAULT;
 946                 goto lenout;
 947         }
 948
 949         /* Dubious BSD thing... Probably nobody even uses it, but
 950          * the UNIX standard wants it for whatever reason... -DaveM
 951          */
 952         case SO_ACCEPTCONN:
 953                 v.val = sk->sk_state == TCP_LISTEN;
 954                 break;
 955
 956         case SO_PASSSEC:
 957                 v.val = !!test_bit(SOCK_PASSSEC, &sock->flags);
 958                 break;
 959
 960         case SO_PEERSEC:
 961                 return security_socket_getpeersec_stream(sock, optval, optlen, len);
 962
 963         case SO_MARK:
 964                 v.val = sk->sk_mark;
 965                 break;
 966
 967         case SO_RXQ_OVFL:
 968                 v.val = !!sock_flag(sk, SOCK_RXQ_OVFL);
 969                 break;
 970
 971         default:
 972                 return -ENOPROTOOPT;
 973         }
 974
 975         if (len > lv)
 976                 len = lv;
 977         if (copy_to_user(optval, &v, len))
 978                 return -EFAULT;
 979 lenout:
 980         if (put_user(len, optlen))
 981                 return -EFAULT;
 982         return 0;
 983 }
 984
 985 /*
 986  * Initialize an sk_lock.
 987  *
 988  * (We also register the sk_lock with the lock validator.)
 989  */
 990 static inline void sock_lock_init(struct sock *sk)
 991 {
 992         sock_lock_init_class_and_name(sk,
 993                         af_family_slock_key_strings[sk->sk_family],
 994                         af_family_slock_keys + sk->sk_family,
 995                         af_family_key_strings[sk->sk_family],
 996                         af_family_keys + sk->sk_family);
 997 }
 998
 999 /*
1000  * Copy all fields from osk to nsk but nsk->sk_refcnt must not change yet,
1001  * even temporarly, because of RCU lookups. sk_node should also be left as is.
1002  * We must not copy fields between sk_dontcopy_begin and sk_dontcopy_end
1003  */
1004 static void sock_copy(struct sock *nsk, const struct sock *osk)
1005 {
1006 #ifdef CONFIG_SECURITY_NETWORK
1007         void *sptr = nsk->sk_security;
1008 #endif
1009         memcpy(nsk, osk, offsetof(struct sock, sk_dontcopy_begin));
1010
1011         memcpy(&nsk->sk_dontcopy_end, &osk->sk_dontcopy_end,
1012                osk->sk_prot->obj_size - offsetof(struct sock, sk_dontcopy_end));
1013
1014 #ifdef CONFIG_SECURITY_NETWORK
1015         nsk->sk_security = sptr;
1016         security_sk_clone(osk, nsk);
1017 #endif
1018 }
1019
1020 void sk_prot_clear_portaddr_nulls(struct sock *sk, int size)
1021 {
1022         unsigned long nulls1, nulls2;
1023
1024         nulls1 = offsetof(struct sock, __sk_common.skc_node.next);
1025         nulls2 = offsetof(struct sock, __sk_common.skc_portaddr_node.next);
1026         if (nulls1 > nulls2)
1027                 swap(nulls1, nulls2);
1028
1029         if (nulls1 != 0)
1030                 memset((char *)sk, 0, nulls1);
1031         memset((char *)sk + nulls1 + sizeof(void *), 0,
1032                nulls2 - nulls1 - sizeof(void *));
1033         memset((char *)sk + nulls2 + sizeof(void *), 0,
1034                size - nulls2 - sizeof(void *));
1035 }
1036 EXPORT_SYMBOL(sk_prot_clear_portaddr_nulls);
1037
1038 static struct sock *sk_prot_alloc(struct proto *prot, gfp_t priority,
1039                 int family)
1040 {
1041         struct sock *sk;
1042         struct kmem_cache *slab;
1043
1044         slab = prot->slab;
1045         if (slab != NULL) {
1046                 sk = kmem_cache_alloc(slab, priority & ~__GFP_ZERO);
1047                 if (!sk)
1048                         return sk;
1049                 if (priority & __GFP_ZERO) {
1050                         if (prot->clear_sk)
1051                                 prot->clear_sk(sk, prot->obj_size);
1052                         else
1053                                 sk_prot_clear_nulls(sk, prot->obj_size);
1054                 }
1055         } else
1056                 sk = kmalloc(prot->obj_size, priority);
1057
1058         if (sk != NULL) {
1059                 kmemcheck_annotate_bitfield(sk, flags);
1060
1061                 if (security_sk_alloc(sk, family, priority))
1062                         goto out_free;
1063
1064                 if (!try_module_get(prot->owner))
1065                         goto out_free_sec;
1066                 sk_tx_queue_clear(sk);
1067         }
1068
1069         return sk;
1070
1071 out_free_sec:
1072         security_sk_free(sk);
1073 out_free:
1074         if (slab != NULL)
1075                 kmem_cache_free(slab, sk);
1076         else
1077                 kfree(sk);
1078         return NULL;
1079 }
1080
1081 static void sk_prot_free(struct proto *prot, struct sock *sk)
1082 {
1083         struct kmem_cache *slab;
1084         struct module *owner;
1085
1086         owner = prot->owner;
1087         slab = prot->slab;
1088
1089         security_sk_free(sk);
1090         if (slab != NULL)
1091                 kmem_cache_free(slab, sk);
1092         else
1093                 kfree(sk);
1094         module_put(owner);
1095 }
1096
1097 #ifdef CONFIG_CGROUPS
1098 void sock_update_classid(struct sock *sk)
1099 {
1100         u32 classid;
1101
1102         rcu_read_lock();  /* doing current task, which cannot vanish. */
1103         classid = task_cls_classid(current);
1104         rcu_read_unlock();
1105         if (classid && classid != sk->sk_classid)
1106                 sk->sk_classid = classid;
1107 }
1108 EXPORT_SYMBOL(sock_update_classid);
1109 #endif
1110
1111 /**
1112  *      sk_alloc - All socket objects are allocated here
1113  *      @net: the applicable net namespace
1114  *      @family: protocol family
1115  *      @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
1116  *      @prot: struct proto associated with this new sock instance
1117  */
1118 struct sock *sk_alloc(struct net *net, int family, gfp_t priority,
1119                       struct proto *prot)
1120 {
1121         struct sock *sk;
1122
1123         sk = sk_prot_alloc(prot, priority | __GFP_ZERO, family);
1124         if (sk) {
1125                 sk->sk_family = family;
1126                 /*
1127                  * See comment in struct sock definition to understand
1128                  * why we need sk_prot_creator -acme
1129                  */
1130                 sk->sk_prot = sk->sk_prot_creator = prot;
1131                 sock_lock_init(sk);
1132                 sock_net_set(sk, get_net(net));
1133                 atomic_set(&sk->sk_wmem_alloc, 1);
1134
1135                 sock_update_classid(sk);
1136         }
1137
1138         return sk;
1139 }
1140 EXPORT_SYMBOL(sk_alloc);
1141
1142 static void __sk_free(struct sock *sk)
1143 {
1144         struct sk_filter *filter;
1145
1146         if (sk->sk_destruct)
1147                 sk->sk_destruct(sk);
1148
1149         filter = rcu_dereference_check(sk->sk_filter,
1150                                        atomic_read(&sk->sk_wmem_alloc) == 0);
1151         if (filter) {
1152                 sk_filter_uncharge(sk, filter);
1153                 RCU_INIT_POINTER(sk->sk_filter, NULL);
1154         }
1155
1156         sock_disable_timestamp(sk, SOCK_TIMESTAMP);
1157         sock_disable_timestamp(sk, SOCK_TIMESTAMPING_RX_SOFTWARE);
1158
1159         if (atomic_read(&sk->sk_omem_alloc))
1160                 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
1161                        __func__, atomic_read(&sk->sk_omem_alloc));
1162
1163         if (sk->sk_peer_cred)
1164                 put_cred(sk->sk_peer_cred);
1165         put_pid(sk->sk_peer_pid);
1166         put_net(sock_net(sk));
1167         sk_prot_free(sk->sk_prot_creator, sk);
1168 }
1169
1170 void sk_free(struct sock *sk)
1171 {
1172         /*
1173          * We subtract one from sk_wmem_alloc and can know if
1174          * some packets are still in some tx queue.
1175          * If not null, sock_wfree() will call __sk_free(sk) later
1176          */
1177         if (atomic_dec_and_test(&sk->sk_wmem_alloc))
1178                 __sk_free(sk);
1179 }
1180 EXPORT_SYMBOL(sk_free);
1181
1182 /*
1183  * Last sock_put should drop reference to sk->sk_net. It has already
1184  * been dropped in sk_change_net. Taking reference to stopping namespace
1185  * is not an option.
1186  * Take reference to a socket to remove it from hash _alive_ and after that
1187  * destroy it in the context of init_net.
1188  */
1189 void sk_release_kernel(struct sock *sk)
1190 {
1191         if (sk == NULL || sk->sk_socket == NULL)
1192                 return;
1193
1194         sock_hold(sk);
1195         sock_release(sk->sk_socket);
1196         release_net(sock_net(sk));
1197         sock_net_set(sk, get_net(&init_net));
1198         sock_put(sk);
1199 }
1200 EXPORT_SYMBOL(sk_release_kernel);
1201
1202 struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
1203 {
1204         struct sock *newsk;
1205
1206         newsk = sk_prot_alloc(sk->sk_prot, priority, sk->sk_family);
1207         if (newsk != NULL) {
1208                 struct sk_filter *filter;
1209
1210                 sock_copy(newsk, sk);
1211
1212                 /* SANITY */
1213                 get_net(sock_net(newsk));
1214                 sk_node_init(&newsk->sk_node);
1215                 sock_lock_init(newsk);
1216                 bh_lock_sock(newsk);
1217                 newsk->sk_backlog.head  = newsk->sk_backlog.tail = NULL;
1218                 newsk->sk_backlog.len = 0;
1219
1220                 atomic_set(&newsk->sk_rmem_alloc, 0);
1221                 /*
1222                  * sk_wmem_alloc set to one (see sk_free() and sock_wfree())
1223                  */
1224                 atomic_set(&newsk->sk_wmem_alloc, 1);
1225                 atomic_set(&newsk->sk_omem_alloc, 0);
1226                 skb_queue_head_init(&newsk->sk_receive_queue);
1227                 skb_queue_head_init(&newsk->sk_write_queue);
1228 #ifdef CONFIG_NET_DMA
1229                 skb_queue_head_init(&newsk->sk_async_wait_queue);
1230 #endif
1231
1232                 spin_lock_init(&newsk->sk_dst_lock);
1233                 rwlock_init(&newsk->sk_callback_lock);
1234                 lockdep_set_class_and_name(&newsk->sk_callback_lock,
1235                                 af_callback_keys + newsk->sk_family,
1236                                 af_family_clock_key_strings[newsk->sk_family]);
1237
1238                 newsk->sk_dst_cache     = NULL;
1239                 newsk->sk_wmem_queued   = 0;
1240                 newsk->sk_forward_alloc = 0;
1241                 newsk->sk_send_head     = NULL;
1242                 newsk->sk_userlocks     = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
1243
1244                 sock_reset_flag(newsk, SOCK_DONE);
1245                 skb_queue_head_init(&newsk->sk_error_queue);
1246
1247                 filter = rcu_dereference_protected(newsk->sk_filter, 1);
1248                 if (filter != NULL)
1249                         sk_filter_charge(newsk, filter);
1250
1251                 if (unlikely(xfrm_sk_clone_policy(newsk))) {
1252                         /* It is still raw copy of parent, so invalidate
1253                          * destructor and make plain sk_free() */
1254                         newsk->sk_destruct = NULL;
1255                         bh_unlock_sock(newsk);
1256                         sk_free(newsk);
1257                         newsk = NULL;
1258                         goto out;
1259                 }
1260
1261                 newsk->sk_err      = 0;
1262                 newsk->sk_priority = 0;
1263                 /*
1264                  * Before updating sk_refcnt, we must commit prior changes to memory
1265                  * (Documentation/RCU/rculist_nulls.txt for details)
1266                  */
1267                 smp_wmb();
1268                 atomic_set(&newsk->sk_refcnt, 2);
1269
1270                 /*
1271                  * Increment the counter in the same struct proto as the master
1272                  * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1273                  * is the same as sk->sk_prot->socks, as this field was copied
1274                  * with memcpy).
1275                  *
1276                  * This _changes_ the previous behaviour, where
1277                  * tcp_create_openreq_child always was incrementing the
1278                  * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1279                  * to be taken into account in all callers. -acme
1280                  */
1281                 sk_refcnt_debug_inc(newsk);
1282                 sk_set_socket(newsk, NULL);
1283                 newsk->sk_wq = NULL;
1284
1285                 if (newsk->sk_prot->sockets_allocated)
1286                         percpu_counter_inc(newsk->sk_prot->sockets_allocated);
1287
1288                 if (sock_flag(newsk, SOCK_TIMESTAMP) ||
1289                     sock_flag(newsk, SOCK_TIMESTAMPING_RX_SOFTWARE))
1290                         net_enable_timestamp();
1291         }
1292 out:
1293         return newsk;
1294 }
1295 EXPORT_SYMBOL_GPL(sk_clone);
1296
1297 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1298 {
1299         __sk_dst_set(sk, dst);
1300         sk->sk_route_caps = dst->dev->features;
1301         if (sk->sk_route_caps & NETIF_F_GSO)
1302                 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1303         sk->sk_route_caps &= ~sk->sk_route_nocaps;
1304         if (sk_can_gso(sk)) {
1305                 if (dst->header_len) {
1306                         sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1307                 } else {
1308                         sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1309                         sk->sk_gso_max_size = dst->dev->gso_max_size;
1310                 }
1311         }
1312 }
1313 EXPORT_SYMBOL_GPL(sk_setup_caps);
1314
1315 void __init sk_init(void)
1316 {
1317         if (totalram_pages <= 4096) {
1318                 sysctl_wmem_max = 32767;
1319                 sysctl_rmem_max = 32767;
1320                 sysctl_wmem_default = 32767;
1321                 sysctl_rmem_default = 32767;
1322         } else if (totalram_pages >= 131072) {
1323                 sysctl_wmem_max = 131071;
1324                 sysctl_rmem_max = 131071;
1325         }
1326 }
1327
1328 /*
1329  *      Simple resource managers for sockets.
1330  */
1331
1332
1333 /*
1334  * Write buffer destructor automatically called from kfree_skb.
1335  */
1336 void sock_wfree(struct sk_buff *skb)
1337 {
1338         struct sock *sk = skb->sk;
1339         unsigned int len = skb->truesize;
1340
1341         if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1342                 /*
1343                  * Keep a reference on sk_wmem_alloc, this will be released
1344                  * after sk_write_space() call
1345                  */
1346                 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1347                 sk->sk_write_space(sk);
1348                 len = 1;
1349         }
1350         /*
1351          * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1352          * could not do because of in-flight packets
1353          */
1354         if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1355                 __sk_free(sk);
1356 }
1357 EXPORT_SYMBOL(sock_wfree);
1358
1359 /*
1360  * Read buffer destructor automatically called from kfree_skb.
1361  */
1362 void sock_rfree(struct sk_buff *skb)
1363 {
1364         struct sock *sk = skb->sk;
1365         unsigned int len = skb->truesize;
1366
1367         atomic_sub(len, &sk->sk_rmem_alloc);
1368         sk_mem_uncharge(sk, len);
1369 }
1370 EXPORT_SYMBOL(sock_rfree);
1371
1372 void sock_efree(struct sk_buff *skb)
1373 {
1374         sock_put(skb->sk);
1375 }
1376 EXPORT_SYMBOL(sock_efree);
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  *      @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
1754  */
1755 void __sk_mem_reclaim(struct sock *sk, int amount)
1756 {
1757         struct proto *prot = sk->sk_prot;
1758
1759         amount >>= SK_MEM_QUANTUM_SHIFT;
1760         atomic_long_sub(amount, prot->memory_allocated);
1761         sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
1762
1763         if (prot->memory_pressure && *prot->memory_pressure &&
1764             (atomic_long_read(prot->memory_allocated) < prot->sysctl_mem[0]))
1765                 *prot->memory_pressure = 0;
1766 }
1767 EXPORT_SYMBOL(__sk_mem_reclaim);
1768
1769
1770 /*
1771  * Set of default routines for initialising struct proto_ops when
1772  * the protocol does not support a particular function. In certain
1773  * cases where it makes no sense for a protocol to have a "do nothing"
1774  * function, some default processing is provided.
1775  */
1776
1777 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1778 {
1779         return -EOPNOTSUPP;
1780 }
1781 EXPORT_SYMBOL(sock_no_bind);
1782
1783 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1784                     int len, int flags)
1785 {
1786         return -EOPNOTSUPP;
1787 }
1788 EXPORT_SYMBOL(sock_no_connect);
1789
1790 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1791 {
1792         return -EOPNOTSUPP;
1793 }
1794 EXPORT_SYMBOL(sock_no_socketpair);
1795
1796 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1797 {
1798         return -EOPNOTSUPP;
1799 }
1800 EXPORT_SYMBOL(sock_no_accept);
1801
1802 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1803                     int *len, int peer)
1804 {
1805         return -EOPNOTSUPP;
1806 }
1807 EXPORT_SYMBOL(sock_no_getname);
1808
1809 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1810 {
1811         return 0;
1812 }
1813 EXPORT_SYMBOL(sock_no_poll);
1814
1815 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1816 {
1817         return -EOPNOTSUPP;
1818 }
1819 EXPORT_SYMBOL(sock_no_ioctl);
1820
1821 int sock_no_listen(struct socket *sock, int backlog)
1822 {
1823         return -EOPNOTSUPP;
1824 }
1825 EXPORT_SYMBOL(sock_no_listen);
1826
1827 int sock_no_shutdown(struct socket *sock, int how)
1828 {
1829         return -EOPNOTSUPP;
1830 }
1831 EXPORT_SYMBOL(sock_no_shutdown);
1832
1833 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1834                     char __user *optval, unsigned int optlen)
1835 {
1836         return -EOPNOTSUPP;
1837 }
1838 EXPORT_SYMBOL(sock_no_setsockopt);
1839
1840 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1841                     char __user *optval, int __user *optlen)
1842 {
1843         return -EOPNOTSUPP;
1844 }
1845 EXPORT_SYMBOL(sock_no_getsockopt);
1846
1847 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1848                     size_t len)
1849 {
1850         return -EOPNOTSUPP;
1851 }
1852 EXPORT_SYMBOL(sock_no_sendmsg);
1853
1854 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1855                     size_t len, int flags)
1856 {
1857         return -EOPNOTSUPP;
1858 }
1859 EXPORT_SYMBOL(sock_no_recvmsg);
1860
1861 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1862 {
1863         /* Mirror missing mmap method error code */
1864         return -ENODEV;
1865 }
1866 EXPORT_SYMBOL(sock_no_mmap);
1867
1868 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1869 {
1870         ssize_t res;
1871         struct msghdr msg = {.msg_flags = flags};
1872         struct kvec iov;
1873         char *kaddr = kmap(page);
1874         iov.iov_base = kaddr + offset;
1875         iov.iov_len = size;
1876         res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1877         kunmap(page);
1878         return res;
1879 }
1880 EXPORT_SYMBOL(sock_no_sendpage);
1881
1882 /*
1883  *      Default Socket Callbacks
1884  */
1885
1886 static void sock_def_wakeup(struct sock *sk)
1887 {
1888         struct socket_wq *wq;
1889
1890         rcu_read_lock();
1891         wq = rcu_dereference(sk->sk_wq);
1892         if (wq_has_sleeper(wq))
1893                 wake_up_interruptible_all(&wq->wait);
1894         rcu_read_unlock();
1895 }
1896
1897 static void sock_def_error_report(struct sock *sk)
1898 {
1899         struct socket_wq *wq;
1900
1901         rcu_read_lock();
1902         wq = rcu_dereference(sk->sk_wq);
1903         if (wq_has_sleeper(wq))
1904                 wake_up_interruptible_poll(&wq->wait, POLLERR);
1905         sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
1906         rcu_read_unlock();
1907 }
1908
1909 static void sock_def_readable(struct sock *sk, int len)
1910 {
1911         struct socket_wq *wq;
1912
1913         rcu_read_lock();
1914         wq = rcu_dereference(sk->sk_wq);
1915         if (wq_has_sleeper(wq))
1916                 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
1917                                                 POLLRDNORM | POLLRDBAND);
1918         sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1919         rcu_read_unlock();
1920 }
1921
1922 static void sock_def_write_space(struct sock *sk)
1923 {
1924         struct socket_wq *wq;
1925
1926         rcu_read_lock();
1927
1928         /* Do not wake up a writer until he can make "significant"
1929          * progress.  --DaveM
1930          */
1931         if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1932                 wq = rcu_dereference(sk->sk_wq);
1933                 if (wq_has_sleeper(wq))
1934                         wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
1935                                                 POLLWRNORM | POLLWRBAND);
1936
1937                 /* Should agree with poll, otherwise some programs break */
1938                 if (sock_writeable(sk))
1939                         sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1940         }
1941
1942         rcu_read_unlock();
1943 }
1944
1945 static void sock_def_destruct(struct sock *sk)
1946 {
1947         kfree(sk->sk_protinfo);
1948 }
1949
1950 void sk_send_sigurg(struct sock *sk)
1951 {
1952         if (sk->sk_socket && sk->sk_socket->file)
1953                 if (send_sigurg(&sk->sk_socket->file->f_owner))
1954                         sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1955 }
1956 EXPORT_SYMBOL(sk_send_sigurg);
1957
1958 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1959                     unsigned long expires)
1960 {
1961         if (!mod_timer(timer, expires))
1962                 sock_hold(sk);
1963 }
1964 EXPORT_SYMBOL(sk_reset_timer);
1965
1966 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1967 {
1968         if (timer_pending(timer) && del_timer(timer))
1969                 __sock_put(sk);
1970 }
1971 EXPORT_SYMBOL(sk_stop_timer);
1972
1973 void sock_init_data(struct socket *sock, struct sock *sk)
1974 {
1975         skb_queue_head_init(&sk->sk_receive_queue);
1976         skb_queue_head_init(&sk->sk_write_queue);
1977         skb_queue_head_init(&sk->sk_error_queue);
1978 #ifdef CONFIG_NET_DMA
1979         skb_queue_head_init(&sk->sk_async_wait_queue);
1980 #endif
1981
1982         sk->sk_send_head        =       NULL;
1983
1984         init_timer(&sk->sk_timer);
1985
1986         sk->sk_allocation       =       GFP_KERNEL;
1987         sk->sk_rcvbuf           =       sysctl_rmem_default;
1988         sk->sk_sndbuf           =       sysctl_wmem_default;
1989         sk->sk_state            =       TCP_CLOSE;
1990         sk_set_socket(sk, sock);
1991
1992         sock_set_flag(sk, SOCK_ZAPPED);
1993
1994         if (sock) {
1995                 sk->sk_type     =       sock->type;
1996                 sk->sk_wq       =       sock->wq;
1997                 sock->sk        =       sk;
1998         } else
1999                 sk->sk_wq       =       NULL;
2000
2001         spin_lock_init(&sk->sk_dst_lock);
2002         rwlock_init(&sk->sk_callback_lock);
2003         lockdep_set_class_and_name(&sk->sk_callback_lock,
2004                         af_callback_keys + sk->sk_family,
2005                         af_family_clock_key_strings[sk->sk_family]);
2006
2007         sk->sk_state_change     =       sock_def_wakeup;
2008         sk->sk_data_ready       =       sock_def_readable;
2009         sk->sk_write_space      =       sock_def_write_space;
2010         sk->sk_error_report     =       sock_def_error_report;
2011         sk->sk_destruct         =       sock_def_destruct;
2012
2013         sk->sk_sndmsg_page      =       NULL;
2014         sk->sk_sndmsg_off       =       0;
2015
2016         sk->sk_peer_pid         =       NULL;
2017         sk->sk_peer_cred        =       NULL;
2018         sk->sk_write_pending    =       0;
2019         sk->sk_rcvlowat         =       1;
2020         sk->sk_rcvtimeo         =       MAX_SCHEDULE_TIMEOUT;
2021         sk->sk_sndtimeo         =       MAX_SCHEDULE_TIMEOUT;
2022
2023         sk->sk_stamp = ktime_set(-1L, 0);
2024
2025         /*
2026          * Before updating sk_refcnt, we must commit prior changes to memory
2027          * (Documentation/RCU/rculist_nulls.txt for details)
2028          */
2029         smp_wmb();
2030         atomic_set(&sk->sk_refcnt, 1);
2031         atomic_set(&sk->sk_drops, 0);
2032 }
2033 EXPORT_SYMBOL(sock_init_data);
2034
2035 void lock_sock_nested(struct sock *sk, int subclass)
2036 {
2037         might_sleep();
2038         spin_lock_bh(&sk->sk_lock.slock);
2039         if (sk->sk_lock.owned)
2040                 __lock_sock(sk);
2041         sk->sk_lock.owned = 1;
2042         spin_unlock(&sk->sk_lock.slock);
2043         /*
2044          * The sk_lock has mutex_lock() semantics here:
2045          */
2046         mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2047         local_bh_enable();
2048 }
2049 EXPORT_SYMBOL(lock_sock_nested);
2050
2051 void release_sock(struct sock *sk)
2052 {
2053         /*
2054          * The sk_lock has mutex_unlock() semantics:
2055          */
2056         mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2057
2058         spin_lock_bh(&sk->sk_lock.slock);
2059         if (sk->sk_backlog.tail)
2060                 __release_sock(sk);
2061         sk->sk_lock.owned = 0;
2062         if (waitqueue_active(&sk->sk_lock.wq))
2063                 wake_up(&sk->sk_lock.wq);
2064         spin_unlock_bh(&sk->sk_lock.slock);
2065 }
2066 EXPORT_SYMBOL(release_sock);
2067
2068 /**
2069  * lock_sock_fast - fast version of lock_sock
2070  * @sk: socket
2071  *
2072  * This version should be used for very small section, where process wont block
2073  * return false if fast path is taken
2074  *   sk_lock.slock locked, owned = 0, BH disabled
2075  * return true if slow path is taken
2076  *   sk_lock.slock unlocked, owned = 1, BH enabled
2077  */
2078 bool lock_sock_fast(struct sock *sk)
2079 {
2080         might_sleep();
2081         spin_lock_bh(&sk->sk_lock.slock);
2082
2083         if (!sk->sk_lock.owned)
2084                 /*
2085                  * Note : We must disable BH
2086                  */
2087                 return false;
2088
2089         __lock_sock(sk);
2090         sk->sk_lock.owned = 1;
2091         spin_unlock(&sk->sk_lock.slock);
2092         /*
2093          * The sk_lock has mutex_lock() semantics here:
2094          */
2095         mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2096         local_bh_enable();
2097         return true;
2098 }
2099 EXPORT_SYMBOL(lock_sock_fast);
2100
2101 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2102 {
2103         struct timeval tv;
2104         if (!sock_flag(sk, SOCK_TIMESTAMP))
2105                 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2106         tv = ktime_to_timeval(sk->sk_stamp);
2107         if (tv.tv_sec == -1)
2108                 return -ENOENT;
2109         if (tv.tv_sec == 0) {
2110                 sk->sk_stamp = ktime_get_real();
2111                 tv = ktime_to_timeval(sk->sk_stamp);
2112         }
2113         return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2114 }
2115 EXPORT_SYMBOL(sock_get_timestamp);
2116
2117 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2118 {
2119         struct timespec ts;
2120         if (!sock_flag(sk, SOCK_TIMESTAMP))
2121                 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2122         ts = ktime_to_timespec(sk->sk_stamp);
2123         if (ts.tv_sec == -1)
2124                 return -ENOENT;
2125         if (ts.tv_sec == 0) {
2126                 sk->sk_stamp = ktime_get_real();
2127                 ts = ktime_to_timespec(sk->sk_stamp);
2128         }
2129         return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2130 }
2131 EXPORT_SYMBOL(sock_get_timestampns);
2132
2133 void sock_enable_timestamp(struct sock *sk, int flag)
2134 {
2135         if (!sock_flag(sk, flag)) {
2136                 sock_set_flag(sk, flag);
2137                 /*
2138                  * we just set one of the two flags which require net
2139                  * time stamping, but time stamping might have been on
2140                  * already because of the other one
2141                  */
2142                 if (!sock_flag(sk,
2143                                 flag == SOCK_TIMESTAMP ?
2144                                 SOCK_TIMESTAMPING_RX_SOFTWARE :
2145                                 SOCK_TIMESTAMP))
2146                         net_enable_timestamp();
2147         }
2148 }
2149
2150 /*
2151  *      Get a socket option on an socket.
2152  *
2153  *      FIX: POSIX 1003.1g is very ambiguous here. It states that
2154  *      asynchronous errors should be reported by getsockopt. We assume
2155  *      this means if you specify SO_ERROR (otherwise whats the point of it).
2156  */
2157 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2158                            char __user *optval, int __user *optlen)
2159 {
2160         struct sock *sk = sock->sk;
2161
2162         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2163 }
2164 EXPORT_SYMBOL(sock_common_getsockopt);
2165
2166 #ifdef CONFIG_COMPAT
2167 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2168                                   char __user *optval, int __user *optlen)
2169 {
2170         struct sock *sk = sock->sk;
2171
2172         if (sk->sk_prot->compat_getsockopt != NULL)
2173                 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2174                                                       optval, optlen);
2175         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2176 }
2177 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2178 #endif
2179
2180 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2181                         struct msghdr *msg, size_t size, int flags)
2182 {
2183         struct sock *sk = sock->sk;
2184         int addr_len = 0;
2185         int err;
2186
2187         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2188                                    flags & ~MSG_DONTWAIT, &addr_len);
2189         if (err >= 0)
2190                 msg->msg_namelen = addr_len;
2191         return err;
2192 }
2193 EXPORT_SYMBOL(sock_common_recvmsg);
2194
2195 /*
2196  *      Set socket options on an inet socket.
2197  */
2198 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2199                            char __user *optval, unsigned int optlen)
2200 {
2201         struct sock *sk = sock->sk;
2202
2203         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2204 }
2205 EXPORT_SYMBOL(sock_common_setsockopt);
2206
2207 #ifdef CONFIG_COMPAT
2208 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2209                                   char __user *optval, unsigned int optlen)
2210 {
2211         struct sock *sk = sock->sk;
2212
2213         if (sk->sk_prot->compat_setsockopt != NULL)
2214                 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2215                                                       optval, optlen);
2216         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2217 }
2218 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2219 #endif
2220
2221 void sk_common_release(struct sock *sk)
2222 {
2223         if (sk->sk_prot->destroy)
2224                 sk->sk_prot->destroy(sk);
2225
2226         /*
2227          * Observation: when sock_common_release is called, processes have
2228          * no access to socket. But net still has.
2229          * Step one, detach it from networking:
2230          *
2231          * A. Remove from hash tables.
2232          */
2233
2234         sk->sk_prot->unhash(sk);
2235
2236         /*
2237          * In this point socket cannot receive new packets, but it is possible
2238          * that some packets are in flight because some CPU runs receiver and
2239          * did hash table lookup before we unhashed socket. They will achieve
2240          * receive queue and will be purged by socket destructor.
2241          *
2242          * Also we still have packets pending on receive queue and probably,
2243          * our own packets waiting in device queues. sock_destroy will drain
2244          * receive queue, but transmitted packets will delay socket destruction
2245          * until the last reference will be released.
2246          */
2247
2248         sock_orphan(sk);
2249
2250         xfrm_sk_free_policy(sk);
2251
2252         sk_refcnt_debug_release(sk);
2253         sock_put(sk);
2254 }
2255 EXPORT_SYMBOL(sk_common_release);
2256
2257 static DEFINE_RWLOCK(proto_list_lock);
2258 static LIST_HEAD(proto_list);
2259
2260 #ifdef CONFIG_PROC_FS
2261 #define PROTO_INUSE_NR  64      /* should be enough for the first time */
2262 struct prot_inuse {
2263         int val[PROTO_INUSE_NR];
2264 };
2265
2266 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2267
2268 #ifdef CONFIG_NET_NS
2269 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2270 {
2271         __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2272 }
2273 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2274
2275 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2276 {
2277         int cpu, idx = prot->inuse_idx;
2278         int res = 0;
2279
2280         for_each_possible_cpu(cpu)
2281                 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2282
2283         return res >= 0 ? res : 0;
2284 }
2285 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2286
2287 static int __net_init sock_inuse_init_net(struct net *net)
2288 {
2289         net->core.inuse = alloc_percpu(struct prot_inuse);
2290         return net->core.inuse ? 0 : -ENOMEM;
2291 }
2292
2293 static void __net_exit sock_inuse_exit_net(struct net *net)
2294 {
2295         free_percpu(net->core.inuse);
2296 }
2297
2298 static struct pernet_operations net_inuse_ops = {
2299         .init = sock_inuse_init_net,
2300         .exit = sock_inuse_exit_net,
2301 };
2302
2303 static __init int net_inuse_init(void)
2304 {
2305         if (register_pernet_subsys(&net_inuse_ops))
2306                 panic("Cannot initialize net inuse counters");
2307
2308         return 0;
2309 }
2310
2311 core_initcall(net_inuse_init);
2312 #else
2313 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2314
2315 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2316 {
2317         __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2318 }
2319 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2320
2321 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2322 {
2323         int cpu, idx = prot->inuse_idx;
2324         int res = 0;
2325
2326         for_each_possible_cpu(cpu)
2327                 res += per_cpu(prot_inuse, cpu).val[idx];
2328
2329         return res >= 0 ? res : 0;
2330 }
2331 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2332 #endif
2333
2334 static void assign_proto_idx(struct proto *prot)
2335 {
2336         prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2337
2338         if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2339                 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2340                 return;
2341         }
2342
2343         set_bit(prot->inuse_idx, proto_inuse_idx);
2344 }
2345
2346 static void release_proto_idx(struct proto *prot)
2347 {
2348         if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2349                 clear_bit(prot->inuse_idx, proto_inuse_idx);
2350 }
2351 #else
2352 static inline void assign_proto_idx(struct proto *prot)
2353 {
2354 }
2355
2356 static inline void release_proto_idx(struct proto *prot)
2357 {
2358 }
2359 #endif
2360
2361 int proto_register(struct proto *prot, int alloc_slab)
2362 {
2363         if (alloc_slab) {
2364                 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2365                                         SLAB_HWCACHE_ALIGN | prot->slab_flags,
2366                                         NULL);
2367
2368                 if (prot->slab == NULL) {
2369                         printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2370                                prot->name);
2371                         goto out;
2372                 }
2373
2374                 if (prot->rsk_prot != NULL) {
2375                         prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
2376                         if (prot->rsk_prot->slab_name == NULL)
2377                                 goto out_free_sock_slab;
2378
2379                         prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2380                                                                  prot->rsk_prot->obj_size, 0,
2381                                                                  SLAB_HWCACHE_ALIGN, NULL);
2382
2383                         if (prot->rsk_prot->slab == NULL) {
2384                                 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2385                                        prot->name);
2386                                 goto out_free_request_sock_slab_name;
2387                         }
2388                 }
2389
2390                 if (prot->twsk_prot != NULL) {
2391                         prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2392
2393                         if (prot->twsk_prot->twsk_slab_name == NULL)
2394                                 goto out_free_request_sock_slab;
2395
2396                         prot->twsk_prot->twsk_slab =
2397                                 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2398                                                   prot->twsk_prot->twsk_obj_size,
2399                                                   0,
2400                                                   SLAB_HWCACHE_ALIGN |
2401                                                         prot->slab_flags,
2402                                                   NULL);
2403                         if (prot->twsk_prot->twsk_slab == NULL)
2404                                 goto out_free_timewait_sock_slab_name;
2405                 }
2406         }
2407
2408         write_lock(&proto_list_lock);
2409         list_add(&prot->node, &proto_list);
2410         assign_proto_idx(prot);
2411         write_unlock(&proto_list_lock);
2412         return 0;
2413
2414 out_free_timewait_sock_slab_name:
2415         kfree(prot->twsk_prot->twsk_slab_name);
2416 out_free_request_sock_slab:
2417         if (prot->rsk_prot && prot->rsk_prot->slab) {
2418                 kmem_cache_destroy(prot->rsk_prot->slab);
2419                 prot->rsk_prot->slab = NULL;
2420         }
2421 out_free_request_sock_slab_name:
2422         if (prot->rsk_prot)
2423                 kfree(prot->rsk_prot->slab_name);
2424 out_free_sock_slab:
2425         kmem_cache_destroy(prot->slab);
2426         prot->slab = NULL;
2427 out:
2428         return -ENOBUFS;
2429 }
2430 EXPORT_SYMBOL(proto_register);
2431
2432 void proto_unregister(struct proto *prot)
2433 {
2434         write_lock(&proto_list_lock);
2435         release_proto_idx(prot);
2436         list_del(&prot->node);
2437         write_unlock(&proto_list_lock);
2438
2439         if (prot->slab != NULL) {
2440                 kmem_cache_destroy(prot->slab);
2441                 prot->slab = NULL;
2442         }
2443
2444         if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2445                 kmem_cache_destroy(prot->rsk_prot->slab);
2446                 kfree(prot->rsk_prot->slab_name);
2447                 prot->rsk_prot->slab = NULL;
2448         }
2449
2450         if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2451                 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2452                 kfree(prot->twsk_prot->twsk_slab_name);
2453                 prot->twsk_prot->twsk_slab = NULL;
2454         }
2455 }
2456 EXPORT_SYMBOL(proto_unregister);
2457
2458 #ifdef CONFIG_PROC_FS
2459 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2460         __acquires(proto_list_lock)
2461 {
2462         read_lock(&proto_list_lock);
2463         return seq_list_start_head(&proto_list, *pos);
2464 }
2465
2466 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2467 {
2468         return seq_list_next(v, &proto_list, pos);
2469 }
2470
2471 static void proto_seq_stop(struct seq_file *seq, void *v)
2472         __releases(proto_list_lock)
2473 {
2474         read_unlock(&proto_list_lock);
2475 }
2476
2477 static char proto_method_implemented(const void *method)
2478 {
2479         return method == NULL ? 'n' : 'y';
2480 }
2481
2482 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2483 {
2484         seq_printf(seq, "%-9s %4u %6d  %6ld   %-3s %6u   %-3s  %-10s "
2485                         "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2486                    proto->name,
2487                    proto->obj_size,
2488                    sock_prot_inuse_get(seq_file_net(seq), proto),
2489                    proto->memory_allocated != NULL ? atomic_long_read(proto->memory_allocated) : -1L,
2490                    proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2491                    proto->max_header,
2492                    proto->slab == NULL ? "no" : "yes",
2493                    module_name(proto->owner),
2494                    proto_method_implemented(proto->close),
2495                    proto_method_implemented(proto->connect),
2496                    proto_method_implemented(proto->disconnect),
2497                    proto_method_implemented(proto->accept),
2498                    proto_method_implemented(proto->ioctl),
2499                    proto_method_implemented(proto->init),
2500                    proto_method_implemented(proto->destroy),
2501                    proto_method_implemented(proto->shutdown),
2502                    proto_method_implemented(proto->setsockopt),
2503                    proto_method_implemented(proto->getsockopt),
2504                    proto_method_implemented(proto->sendmsg),
2505                    proto_method_implemented(proto->recvmsg),
2506                    proto_method_implemented(proto->sendpage),
2507                    proto_method_implemented(proto->bind),
2508                    proto_method_implemented(proto->backlog_rcv),
2509                    proto_method_implemented(proto->hash),
2510                    proto_method_implemented(proto->unhash),
2511                    proto_method_implemented(proto->get_port),
2512                    proto_method_implemented(proto->enter_memory_pressure));
2513 }
2514
2515 static int proto_seq_show(struct seq_file *seq, void *v)
2516 {
2517         if (v == &proto_list)
2518                 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2519                            "protocol",
2520                            "size",
2521                            "sockets",
2522                            "memory",
2523                            "press",
2524                            "maxhdr",
2525                            "slab",
2526                            "module",
2527                            "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2528         else
2529                 proto_seq_printf(seq, list_entry(v, struct proto, node));
2530         return 0;
2531 }
2532
2533 static const struct seq_operations proto_seq_ops = {
2534         .start  = proto_seq_start,
2535         .next   = proto_seq_next,
2536         .stop   = proto_seq_stop,
2537         .show   = proto_seq_show,
2538 };
2539
2540 static int proto_seq_open(struct inode *inode, struct file *file)
2541 {
2542         return seq_open_net(inode, file, &proto_seq_ops,
2543                             sizeof(struct seq_net_private));
2544 }
2545
2546 static const struct file_operations proto_seq_fops = {
2547         .owner          = THIS_MODULE,
2548         .open           = proto_seq_open,
2549         .read           = seq_read,
2550         .llseek         = seq_lseek,
2551         .release        = seq_release_net,
2552 };
2553
2554 static __net_init int proto_init_net(struct net *net)
2555 {
2556         if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2557                 return -ENOMEM;
2558
2559         return 0;
2560 }
2561
2562 static __net_exit void proto_exit_net(struct net *net)
2563 {
2564         proc_net_remove(net, "protocols");
2565 }
2566
2567
2568 static __net_initdata struct pernet_operations proto_net_ops = {
2569         .init = proto_init_net,
2570         .exit = proto_exit_net,
2571 };
2572
2573 static int __init proto_init(void)
2574 {
2575         return register_pernet_subsys(&proto_net_ops);
2576 }
2577
2578 subsys_initcall(proto_init);
2579
2580 #endif /* PROC_FS */