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_err_soft = 0;
1263                 newsk->sk_priority = 0;
1264                 /*
1265                  * Before updating sk_refcnt, we must commit prior changes to memory
1266                  * (Documentation/RCU/rculist_nulls.txt for details)
1267                  */
1268                 smp_wmb();
1269                 atomic_set(&newsk->sk_refcnt, 2);
1270
1271                 /*
1272                  * Increment the counter in the same struct proto as the master
1273                  * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
1274                  * is the same as sk->sk_prot->socks, as this field was copied
1275                  * with memcpy).
1276                  *
1277                  * This _changes_ the previous behaviour, where
1278                  * tcp_create_openreq_child always was incrementing the
1279                  * equivalent to tcp_prot->socks (inet_sock_nr), so this have
1280                  * to be taken into account in all callers. -acme
1281                  */
1282                 sk_refcnt_debug_inc(newsk);
1283                 sk_set_socket(newsk, NULL);
1284                 newsk->sk_wq = NULL;
1285
1286                 if (newsk->sk_prot->sockets_allocated)
1287                         percpu_counter_inc(newsk->sk_prot->sockets_allocated);
1288
1289                 if (sock_flag(newsk, SOCK_TIMESTAMP) ||
1290                     sock_flag(newsk, SOCK_TIMESTAMPING_RX_SOFTWARE))
1291                         net_enable_timestamp();
1292         }
1293 out:
1294         return newsk;
1295 }
1296 EXPORT_SYMBOL_GPL(sk_clone);
1297
1298 void sk_setup_caps(struct sock *sk, struct dst_entry *dst)
1299 {
1300         __sk_dst_set(sk, dst);
1301         sk->sk_route_caps = dst->dev->features;
1302         if (sk->sk_route_caps & NETIF_F_GSO)
1303                 sk->sk_route_caps |= NETIF_F_GSO_SOFTWARE;
1304         sk->sk_route_caps &= ~sk->sk_route_nocaps;
1305         if (sk_can_gso(sk)) {
1306                 if (dst->header_len) {
1307                         sk->sk_route_caps &= ~NETIF_F_GSO_MASK;
1308                 } else {
1309                         sk->sk_route_caps |= NETIF_F_SG | NETIF_F_HW_CSUM;
1310                         sk->sk_gso_max_size = dst->dev->gso_max_size;
1311                 }
1312         }
1313 }
1314 EXPORT_SYMBOL_GPL(sk_setup_caps);
1315
1316 void __init sk_init(void)
1317 {
1318         if (totalram_pages <= 4096) {
1319                 sysctl_wmem_max = 32767;
1320                 sysctl_rmem_max = 32767;
1321                 sysctl_wmem_default = 32767;
1322                 sysctl_rmem_default = 32767;
1323         } else if (totalram_pages >= 131072) {
1324                 sysctl_wmem_max = 131071;
1325                 sysctl_rmem_max = 131071;
1326         }
1327 }
1328
1329 /*
1330  *      Simple resource managers for sockets.
1331  */
1332
1333
1334 /*
1335  * Write buffer destructor automatically called from kfree_skb.
1336  */
1337 void sock_wfree(struct sk_buff *skb)
1338 {
1339         struct sock *sk = skb->sk;
1340         unsigned int len = skb->truesize;
1341
1342         if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE)) {
1343                 /*
1344                  * Keep a reference on sk_wmem_alloc, this will be released
1345                  * after sk_write_space() call
1346                  */
1347                 atomic_sub(len - 1, &sk->sk_wmem_alloc);
1348                 sk->sk_write_space(sk);
1349                 len = 1;
1350         }
1351         /*
1352          * if sk_wmem_alloc reaches 0, we must finish what sk_free()
1353          * could not do because of in-flight packets
1354          */
1355         if (atomic_sub_and_test(len, &sk->sk_wmem_alloc))
1356                 __sk_free(sk);
1357 }
1358 EXPORT_SYMBOL(sock_wfree);
1359
1360 /*
1361  * Read buffer destructor automatically called from kfree_skb.
1362  */
1363 void sock_rfree(struct sk_buff *skb)
1364 {
1365         struct sock *sk = skb->sk;
1366         unsigned int len = skb->truesize;
1367
1368         atomic_sub(len, &sk->sk_rmem_alloc);
1369         sk_mem_uncharge(sk, len);
1370 }
1371 EXPORT_SYMBOL(sock_rfree);
1372
1373 void sock_efree(struct sk_buff *skb)
1374 {
1375         sock_put(skb->sk);
1376 }
1377 EXPORT_SYMBOL(sock_efree);
1378
1379 int sock_i_uid(struct sock *sk)
1380 {
1381         int uid;
1382
1383         read_lock_bh(&sk->sk_callback_lock);
1384         uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
1385         read_unlock_bh(&sk->sk_callback_lock);
1386         return uid;
1387 }
1388 EXPORT_SYMBOL(sock_i_uid);
1389
1390 unsigned long sock_i_ino(struct sock *sk)
1391 {
1392         unsigned long ino;
1393
1394         read_lock_bh(&sk->sk_callback_lock);
1395         ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
1396         read_unlock_bh(&sk->sk_callback_lock);
1397         return ino;
1398 }
1399 EXPORT_SYMBOL(sock_i_ino);
1400
1401 /*
1402  * Allocate a skb from the socket's send buffer.
1403  */
1404 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
1405                              gfp_t priority)
1406 {
1407         if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1408                 struct sk_buff *skb = alloc_skb(size, priority);
1409                 if (skb) {
1410                         skb_set_owner_w(skb, sk);
1411                         return skb;
1412                 }
1413         }
1414         return NULL;
1415 }
1416 EXPORT_SYMBOL(sock_wmalloc);
1417
1418 /*
1419  * Allocate a skb from the socket's receive buffer.
1420  */
1421 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
1422                              gfp_t priority)
1423 {
1424         if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
1425                 struct sk_buff *skb = alloc_skb(size, priority);
1426                 if (skb) {
1427                         skb_set_owner_r(skb, sk);
1428                         return skb;
1429                 }
1430         }
1431         return NULL;
1432 }
1433
1434 /*
1435  * Allocate a memory block from the socket's option memory buffer.
1436  */
1437 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
1438 {
1439         if ((unsigned)size <= sysctl_optmem_max &&
1440             atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
1441                 void *mem;
1442                 /* First do the add, to avoid the race if kmalloc
1443                  * might sleep.
1444                  */
1445                 atomic_add(size, &sk->sk_omem_alloc);
1446                 mem = kmalloc(size, priority);
1447                 if (mem)
1448                         return mem;
1449                 atomic_sub(size, &sk->sk_omem_alloc);
1450         }
1451         return NULL;
1452 }
1453 EXPORT_SYMBOL(sock_kmalloc);
1454
1455 /*
1456  * Free an option memory block.
1457  */
1458 void sock_kfree_s(struct sock *sk, void *mem, int size)
1459 {
1460         kfree(mem);
1461         atomic_sub(size, &sk->sk_omem_alloc);
1462 }
1463 EXPORT_SYMBOL(sock_kfree_s);
1464
1465 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
1466    I think, these locks should be removed for datagram sockets.
1467  */
1468 static long sock_wait_for_wmem(struct sock *sk, long timeo)
1469 {
1470         DEFINE_WAIT(wait);
1471
1472         clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1473         for (;;) {
1474                 if (!timeo)
1475                         break;
1476                 if (signal_pending(current))
1477                         break;
1478                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1479                 prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1480                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
1481                         break;
1482                 if (sk->sk_shutdown & SEND_SHUTDOWN)
1483                         break;
1484                 if (sk->sk_err)
1485                         break;
1486                 timeo = schedule_timeout(timeo);
1487         }
1488         finish_wait(sk_sleep(sk), &wait);
1489         return timeo;
1490 }
1491
1492
1493 /*
1494  *      Generic send/receive buffer handlers
1495  */
1496
1497 struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
1498                                      unsigned long data_len, int noblock,
1499                                      int *errcode)
1500 {
1501         struct sk_buff *skb;
1502         gfp_t gfp_mask;
1503         long timeo;
1504         int err;
1505         int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
1506
1507         err = -EMSGSIZE;
1508         if (npages > MAX_SKB_FRAGS)
1509                 goto failure;
1510
1511         gfp_mask = sk->sk_allocation;
1512         if (gfp_mask & __GFP_WAIT)
1513                 gfp_mask |= __GFP_REPEAT;
1514
1515         timeo = sock_sndtimeo(sk, noblock);
1516         while (1) {
1517                 err = sock_error(sk);
1518                 if (err != 0)
1519                         goto failure;
1520
1521                 err = -EPIPE;
1522                 if (sk->sk_shutdown & SEND_SHUTDOWN)
1523                         goto failure;
1524
1525                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
1526                         skb = alloc_skb(header_len, gfp_mask);
1527                         if (skb) {
1528                                 int i;
1529
1530                                 /* No pages, we're done... */
1531                                 if (!data_len)
1532                                         break;
1533
1534                                 skb->truesize += data_len;
1535                                 skb_shinfo(skb)->nr_frags = npages;
1536                                 for (i = 0; i < npages; i++) {
1537                                         struct page *page;
1538
1539                                         page = alloc_pages(sk->sk_allocation, 0);
1540                                         if (!page) {
1541                                                 err = -ENOBUFS;
1542                                                 skb_shinfo(skb)->nr_frags = i;
1543                                                 kfree_skb(skb);
1544                                                 goto failure;
1545                                         }
1546
1547                                         __skb_fill_page_desc(skb, i,
1548                                                         page, 0,
1549                                                         (data_len >= PAGE_SIZE ?
1550                                                          PAGE_SIZE :
1551                                                          data_len));
1552                                         data_len -= PAGE_SIZE;
1553                                 }
1554
1555                                 /* Full success... */
1556                                 break;
1557                         }
1558                         err = -ENOBUFS;
1559                         goto failure;
1560                 }
1561                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1562                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1563                 err = -EAGAIN;
1564                 if (!timeo)
1565                         goto failure;
1566                 if (signal_pending(current))
1567                         goto interrupted;
1568                 timeo = sock_wait_for_wmem(sk, timeo);
1569         }
1570
1571         skb_set_owner_w(skb, sk);
1572         return skb;
1573
1574 interrupted:
1575         err = sock_intr_errno(timeo);
1576 failure:
1577         *errcode = err;
1578         return NULL;
1579 }
1580 EXPORT_SYMBOL(sock_alloc_send_pskb);
1581
1582 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1583                                     int noblock, int *errcode)
1584 {
1585         return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1586 }
1587 EXPORT_SYMBOL(sock_alloc_send_skb);
1588
1589 static void __lock_sock(struct sock *sk)
1590         __releases(&sk->sk_lock.slock)
1591         __acquires(&sk->sk_lock.slock)
1592 {
1593         DEFINE_WAIT(wait);
1594
1595         for (;;) {
1596                 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1597                                         TASK_UNINTERRUPTIBLE);
1598                 spin_unlock_bh(&sk->sk_lock.slock);
1599                 schedule();
1600                 spin_lock_bh(&sk->sk_lock.slock);
1601                 if (!sock_owned_by_user(sk))
1602                         break;
1603         }
1604         finish_wait(&sk->sk_lock.wq, &wait);
1605 }
1606
1607 static void __release_sock(struct sock *sk)
1608         __releases(&sk->sk_lock.slock)
1609         __acquires(&sk->sk_lock.slock)
1610 {
1611         struct sk_buff *skb = sk->sk_backlog.head;
1612
1613         do {
1614                 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1615                 bh_unlock_sock(sk);
1616
1617                 do {
1618                         struct sk_buff *next = skb->next;
1619
1620                         WARN_ON_ONCE(skb_dst_is_noref(skb));
1621                         skb->next = NULL;
1622                         sk_backlog_rcv(sk, skb);
1623
1624                         /*
1625                          * We are in process context here with softirqs
1626                          * disabled, use cond_resched_softirq() to preempt.
1627                          * This is safe to do because we've taken the backlog
1628                          * queue private:
1629                          */
1630                         cond_resched_softirq();
1631
1632                         skb = next;
1633                 } while (skb != NULL);
1634
1635                 bh_lock_sock(sk);
1636         } while ((skb = sk->sk_backlog.head) != NULL);
1637
1638         /*
1639          * Doing the zeroing here guarantee we can not loop forever
1640          * while a wild producer attempts to flood us.
1641          */
1642         sk->sk_backlog.len = 0;
1643 }
1644
1645 /**
1646  * sk_wait_data - wait for data to arrive at sk_receive_queue
1647  * @sk:    sock to wait on
1648  * @timeo: for how long
1649  *
1650  * Now socket state including sk->sk_err is changed only under lock,
1651  * hence we may omit checks after joining wait queue.
1652  * We check receive queue before schedule() only as optimization;
1653  * it is very likely that release_sock() added new data.
1654  */
1655 int sk_wait_data(struct sock *sk, long *timeo)
1656 {
1657         int rc;
1658         DEFINE_WAIT(wait);
1659
1660         prepare_to_wait(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
1661         set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1662         rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1663         clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1664         finish_wait(sk_sleep(sk), &wait);
1665         return rc;
1666 }
1667 EXPORT_SYMBOL(sk_wait_data);
1668
1669 /**
1670  *      __sk_mem_schedule - increase sk_forward_alloc and memory_allocated
1671  *      @sk: socket
1672  *      @size: memory size to allocate
1673  *      @kind: allocation type
1674  *
1675  *      If kind is SK_MEM_SEND, it means wmem allocation. Otherwise it means
1676  *      rmem allocation. This function assumes that protocols which have
1677  *      memory_pressure use sk_wmem_queued as write buffer accounting.
1678  */
1679 int __sk_mem_schedule(struct sock *sk, int size, int kind)
1680 {
1681         struct proto *prot = sk->sk_prot;
1682         int amt = sk_mem_pages(size);
1683         long allocated;
1684
1685         sk->sk_forward_alloc += amt * SK_MEM_QUANTUM;
1686         allocated = atomic_long_add_return(amt, prot->memory_allocated);
1687
1688         /* Under limit. */
1689         if (allocated <= prot->sysctl_mem[0]) {
1690                 if (prot->memory_pressure && *prot->memory_pressure)
1691                         *prot->memory_pressure = 0;
1692                 return 1;
1693         }
1694
1695         /* Under pressure. */
1696         if (allocated > prot->sysctl_mem[1])
1697                 if (prot->enter_memory_pressure)
1698                         prot->enter_memory_pressure(sk);
1699
1700         /* Over hard limit. */
1701         if (allocated > prot->sysctl_mem[2])
1702                 goto suppress_allocation;
1703
1704         /* guarantee minimum buffer size under pressure */
1705         if (kind == SK_MEM_RECV) {
1706                 if (atomic_read(&sk->sk_rmem_alloc) < prot->sysctl_rmem[0])
1707                         return 1;
1708         } else { /* SK_MEM_SEND */
1709                 if (sk->sk_type == SOCK_STREAM) {
1710                         if (sk->sk_wmem_queued < prot->sysctl_wmem[0])
1711                                 return 1;
1712                 } else if (atomic_read(&sk->sk_wmem_alloc) <
1713                            prot->sysctl_wmem[0])
1714                                 return 1;
1715         }
1716
1717         if (prot->memory_pressure) {
1718                 int alloc;
1719
1720                 if (!*prot->memory_pressure)
1721                         return 1;
1722                 alloc = percpu_counter_read_positive(prot->sockets_allocated);
1723                 if (prot->sysctl_mem[2] > alloc *
1724                     sk_mem_pages(sk->sk_wmem_queued +
1725                                  atomic_read(&sk->sk_rmem_alloc) +
1726                                  sk->sk_forward_alloc))
1727                         return 1;
1728         }
1729
1730 suppress_allocation:
1731
1732         if (kind == SK_MEM_SEND && sk->sk_type == SOCK_STREAM) {
1733                 sk_stream_moderate_sndbuf(sk);
1734
1735                 /* Fail only if socket is _under_ its sndbuf.
1736                  * In this case we cannot block, so that we have to fail.
1737                  */
1738                 if (sk->sk_wmem_queued + size >= sk->sk_sndbuf)
1739                         return 1;
1740         }
1741
1742         trace_sock_exceed_buf_limit(sk, prot, allocated);
1743
1744         /* Alas. Undo changes. */
1745         sk->sk_forward_alloc -= amt * SK_MEM_QUANTUM;
1746         atomic_long_sub(amt, prot->memory_allocated);
1747         return 0;
1748 }
1749 EXPORT_SYMBOL(__sk_mem_schedule);
1750
1751 /**
1752  *      __sk_reclaim - reclaim memory_allocated
1753  *      @sk: socket
1754  *      @amount: number of bytes (rounded down to a SK_MEM_QUANTUM multiple)
1755  */
1756 void __sk_mem_reclaim(struct sock *sk, int amount)
1757 {
1758         struct proto *prot = sk->sk_prot;
1759
1760         amount >>= SK_MEM_QUANTUM_SHIFT;
1761         atomic_long_sub(amount, prot->memory_allocated);
1762         sk->sk_forward_alloc -= amount << SK_MEM_QUANTUM_SHIFT;
1763
1764         if (prot->memory_pressure && *prot->memory_pressure &&
1765             (atomic_long_read(prot->memory_allocated) < prot->sysctl_mem[0]))
1766                 *prot->memory_pressure = 0;
1767 }
1768 EXPORT_SYMBOL(__sk_mem_reclaim);
1769
1770
1771 /*
1772  * Set of default routines for initialising struct proto_ops when
1773  * the protocol does not support a particular function. In certain
1774  * cases where it makes no sense for a protocol to have a "do nothing"
1775  * function, some default processing is provided.
1776  */
1777
1778 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1779 {
1780         return -EOPNOTSUPP;
1781 }
1782 EXPORT_SYMBOL(sock_no_bind);
1783
1784 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1785                     int len, int flags)
1786 {
1787         return -EOPNOTSUPP;
1788 }
1789 EXPORT_SYMBOL(sock_no_connect);
1790
1791 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1792 {
1793         return -EOPNOTSUPP;
1794 }
1795 EXPORT_SYMBOL(sock_no_socketpair);
1796
1797 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1798 {
1799         return -EOPNOTSUPP;
1800 }
1801 EXPORT_SYMBOL(sock_no_accept);
1802
1803 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1804                     int *len, int peer)
1805 {
1806         return -EOPNOTSUPP;
1807 }
1808 EXPORT_SYMBOL(sock_no_getname);
1809
1810 unsigned int sock_no_poll(struct file *file, struct socket *sock, poll_table *pt)
1811 {
1812         return 0;
1813 }
1814 EXPORT_SYMBOL(sock_no_poll);
1815
1816 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1817 {
1818         return -EOPNOTSUPP;
1819 }
1820 EXPORT_SYMBOL(sock_no_ioctl);
1821
1822 int sock_no_listen(struct socket *sock, int backlog)
1823 {
1824         return -EOPNOTSUPP;
1825 }
1826 EXPORT_SYMBOL(sock_no_listen);
1827
1828 int sock_no_shutdown(struct socket *sock, int how)
1829 {
1830         return -EOPNOTSUPP;
1831 }
1832 EXPORT_SYMBOL(sock_no_shutdown);
1833
1834 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1835                     char __user *optval, unsigned int optlen)
1836 {
1837         return -EOPNOTSUPP;
1838 }
1839 EXPORT_SYMBOL(sock_no_setsockopt);
1840
1841 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1842                     char __user *optval, int __user *optlen)
1843 {
1844         return -EOPNOTSUPP;
1845 }
1846 EXPORT_SYMBOL(sock_no_getsockopt);
1847
1848 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1849                     size_t len)
1850 {
1851         return -EOPNOTSUPP;
1852 }
1853 EXPORT_SYMBOL(sock_no_sendmsg);
1854
1855 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1856                     size_t len, int flags)
1857 {
1858         return -EOPNOTSUPP;
1859 }
1860 EXPORT_SYMBOL(sock_no_recvmsg);
1861
1862 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1863 {
1864         /* Mirror missing mmap method error code */
1865         return -ENODEV;
1866 }
1867 EXPORT_SYMBOL(sock_no_mmap);
1868
1869 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1870 {
1871         ssize_t res;
1872         struct msghdr msg = {.msg_flags = flags};
1873         struct kvec iov;
1874         char *kaddr = kmap(page);
1875         iov.iov_base = kaddr + offset;
1876         iov.iov_len = size;
1877         res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1878         kunmap(page);
1879         return res;
1880 }
1881 EXPORT_SYMBOL(sock_no_sendpage);
1882
1883 /*
1884  *      Default Socket Callbacks
1885  */
1886
1887 static void sock_def_wakeup(struct sock *sk)
1888 {
1889         struct socket_wq *wq;
1890
1891         rcu_read_lock();
1892         wq = rcu_dereference(sk->sk_wq);
1893         if (wq_has_sleeper(wq))
1894                 wake_up_interruptible_all(&wq->wait);
1895         rcu_read_unlock();
1896 }
1897
1898 static void sock_def_error_report(struct sock *sk)
1899 {
1900         struct socket_wq *wq;
1901
1902         rcu_read_lock();
1903         wq = rcu_dereference(sk->sk_wq);
1904         if (wq_has_sleeper(wq))
1905                 wake_up_interruptible_poll(&wq->wait, POLLERR);
1906         sk_wake_async(sk, SOCK_WAKE_IO, POLL_ERR);
1907         rcu_read_unlock();
1908 }
1909
1910 static void sock_def_readable(struct sock *sk, int len)
1911 {
1912         struct socket_wq *wq;
1913
1914         rcu_read_lock();
1915         wq = rcu_dereference(sk->sk_wq);
1916         if (wq_has_sleeper(wq))
1917                 wake_up_interruptible_sync_poll(&wq->wait, POLLIN | POLLPRI |
1918                                                 POLLRDNORM | POLLRDBAND);
1919         sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1920         rcu_read_unlock();
1921 }
1922
1923 static void sock_def_write_space(struct sock *sk)
1924 {
1925         struct socket_wq *wq;
1926
1927         rcu_read_lock();
1928
1929         /* Do not wake up a writer until he can make "significant"
1930          * progress.  --DaveM
1931          */
1932         if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1933                 wq = rcu_dereference(sk->sk_wq);
1934                 if (wq_has_sleeper(wq))
1935                         wake_up_interruptible_sync_poll(&wq->wait, POLLOUT |
1936                                                 POLLWRNORM | POLLWRBAND);
1937
1938                 /* Should agree with poll, otherwise some programs break */
1939                 if (sock_writeable(sk))
1940                         sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
1941         }
1942
1943         rcu_read_unlock();
1944 }
1945
1946 static void sock_def_destruct(struct sock *sk)
1947 {
1948         kfree(sk->sk_protinfo);
1949 }
1950
1951 void sk_send_sigurg(struct sock *sk)
1952 {
1953         if (sk->sk_socket && sk->sk_socket->file)
1954                 if (send_sigurg(&sk->sk_socket->file->f_owner))
1955                         sk_wake_async(sk, SOCK_WAKE_URG, POLL_PRI);
1956 }
1957 EXPORT_SYMBOL(sk_send_sigurg);
1958
1959 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1960                     unsigned long expires)
1961 {
1962         if (!mod_timer(timer, expires))
1963                 sock_hold(sk);
1964 }
1965 EXPORT_SYMBOL(sk_reset_timer);
1966
1967 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1968 {
1969         if (timer_pending(timer) && del_timer(timer))
1970                 __sock_put(sk);
1971 }
1972 EXPORT_SYMBOL(sk_stop_timer);
1973
1974 void sock_init_data(struct socket *sock, struct sock *sk)
1975 {
1976         skb_queue_head_init(&sk->sk_receive_queue);
1977         skb_queue_head_init(&sk->sk_write_queue);
1978         skb_queue_head_init(&sk->sk_error_queue);
1979 #ifdef CONFIG_NET_DMA
1980         skb_queue_head_init(&sk->sk_async_wait_queue);
1981 #endif
1982
1983         sk->sk_send_head        =       NULL;
1984
1985         init_timer(&sk->sk_timer);
1986
1987         sk->sk_allocation       =       GFP_KERNEL;
1988         sk->sk_rcvbuf           =       sysctl_rmem_default;
1989         sk->sk_sndbuf           =       sysctl_wmem_default;
1990         sk->sk_state            =       TCP_CLOSE;
1991         sk_set_socket(sk, sock);
1992
1993         sock_set_flag(sk, SOCK_ZAPPED);
1994
1995         if (sock) {
1996                 sk->sk_type     =       sock->type;
1997                 sk->sk_wq       =       sock->wq;
1998                 sock->sk        =       sk;
1999         } else
2000                 sk->sk_wq       =       NULL;
2001
2002         spin_lock_init(&sk->sk_dst_lock);
2003         rwlock_init(&sk->sk_callback_lock);
2004         lockdep_set_class_and_name(&sk->sk_callback_lock,
2005                         af_callback_keys + sk->sk_family,
2006                         af_family_clock_key_strings[sk->sk_family]);
2007
2008         sk->sk_state_change     =       sock_def_wakeup;
2009         sk->sk_data_ready       =       sock_def_readable;
2010         sk->sk_write_space      =       sock_def_write_space;
2011         sk->sk_error_report     =       sock_def_error_report;
2012         sk->sk_destruct         =       sock_def_destruct;
2013
2014         sk->sk_sndmsg_page      =       NULL;
2015         sk->sk_sndmsg_off       =       0;
2016
2017         sk->sk_peer_pid         =       NULL;
2018         sk->sk_peer_cred        =       NULL;
2019         sk->sk_write_pending    =       0;
2020         sk->sk_rcvlowat         =       1;
2021         sk->sk_rcvtimeo         =       MAX_SCHEDULE_TIMEOUT;
2022         sk->sk_sndtimeo         =       MAX_SCHEDULE_TIMEOUT;
2023
2024         sk->sk_stamp = ktime_set(-1L, 0);
2025
2026         /*
2027          * Before updating sk_refcnt, we must commit prior changes to memory
2028          * (Documentation/RCU/rculist_nulls.txt for details)
2029          */
2030         smp_wmb();
2031         atomic_set(&sk->sk_refcnt, 1);
2032         atomic_set(&sk->sk_drops, 0);
2033 }
2034 EXPORT_SYMBOL(sock_init_data);
2035
2036 void lock_sock_nested(struct sock *sk, int subclass)
2037 {
2038         might_sleep();
2039         spin_lock_bh(&sk->sk_lock.slock);
2040         if (sk->sk_lock.owned)
2041                 __lock_sock(sk);
2042         sk->sk_lock.owned = 1;
2043         spin_unlock(&sk->sk_lock.slock);
2044         /*
2045          * The sk_lock has mutex_lock() semantics here:
2046          */
2047         mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
2048         local_bh_enable();
2049 }
2050 EXPORT_SYMBOL(lock_sock_nested);
2051
2052 void release_sock(struct sock *sk)
2053 {
2054         /*
2055          * The sk_lock has mutex_unlock() semantics:
2056          */
2057         mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
2058
2059         spin_lock_bh(&sk->sk_lock.slock);
2060         if (sk->sk_backlog.tail)
2061                 __release_sock(sk);
2062         sk->sk_lock.owned = 0;
2063         if (waitqueue_active(&sk->sk_lock.wq))
2064                 wake_up(&sk->sk_lock.wq);
2065         spin_unlock_bh(&sk->sk_lock.slock);
2066 }
2067 EXPORT_SYMBOL(release_sock);
2068
2069 /**
2070  * lock_sock_fast - fast version of lock_sock
2071  * @sk: socket
2072  *
2073  * This version should be used for very small section, where process wont block
2074  * return false if fast path is taken
2075  *   sk_lock.slock locked, owned = 0, BH disabled
2076  * return true if slow path is taken
2077  *   sk_lock.slock unlocked, owned = 1, BH enabled
2078  */
2079 bool lock_sock_fast(struct sock *sk)
2080 {
2081         might_sleep();
2082         spin_lock_bh(&sk->sk_lock.slock);
2083
2084         if (!sk->sk_lock.owned)
2085                 /*
2086                  * Note : We must disable BH
2087                  */
2088                 return false;
2089
2090         __lock_sock(sk);
2091         sk->sk_lock.owned = 1;
2092         spin_unlock(&sk->sk_lock.slock);
2093         /*
2094          * The sk_lock has mutex_lock() semantics here:
2095          */
2096         mutex_acquire(&sk->sk_lock.dep_map, 0, 0, _RET_IP_);
2097         local_bh_enable();
2098         return true;
2099 }
2100 EXPORT_SYMBOL(lock_sock_fast);
2101
2102 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
2103 {
2104         struct timeval tv;
2105         if (!sock_flag(sk, SOCK_TIMESTAMP))
2106                 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2107         tv = ktime_to_timeval(sk->sk_stamp);
2108         if (tv.tv_sec == -1)
2109                 return -ENOENT;
2110         if (tv.tv_sec == 0) {
2111                 sk->sk_stamp = ktime_get_real();
2112                 tv = ktime_to_timeval(sk->sk_stamp);
2113         }
2114         return copy_to_user(userstamp, &tv, sizeof(tv)) ? -EFAULT : 0;
2115 }
2116 EXPORT_SYMBOL(sock_get_timestamp);
2117
2118 int sock_get_timestampns(struct sock *sk, struct timespec __user *userstamp)
2119 {
2120         struct timespec ts;
2121         if (!sock_flag(sk, SOCK_TIMESTAMP))
2122                 sock_enable_timestamp(sk, SOCK_TIMESTAMP);
2123         ts = ktime_to_timespec(sk->sk_stamp);
2124         if (ts.tv_sec == -1)
2125                 return -ENOENT;
2126         if (ts.tv_sec == 0) {
2127                 sk->sk_stamp = ktime_get_real();
2128                 ts = ktime_to_timespec(sk->sk_stamp);
2129         }
2130         return copy_to_user(userstamp, &ts, sizeof(ts)) ? -EFAULT : 0;
2131 }
2132 EXPORT_SYMBOL(sock_get_timestampns);
2133
2134 void sock_enable_timestamp(struct sock *sk, int flag)
2135 {
2136         if (!sock_flag(sk, flag)) {
2137                 sock_set_flag(sk, flag);
2138                 /*
2139                  * we just set one of the two flags which require net
2140                  * time stamping, but time stamping might have been on
2141                  * already because of the other one
2142                  */
2143                 if (!sock_flag(sk,
2144                                 flag == SOCK_TIMESTAMP ?
2145                                 SOCK_TIMESTAMPING_RX_SOFTWARE :
2146                                 SOCK_TIMESTAMP))
2147                         net_enable_timestamp();
2148         }
2149 }
2150
2151 /*
2152  *      Get a socket option on an socket.
2153  *
2154  *      FIX: POSIX 1003.1g is very ambiguous here. It states that
2155  *      asynchronous errors should be reported by getsockopt. We assume
2156  *      this means if you specify SO_ERROR (otherwise whats the point of it).
2157  */
2158 int sock_common_getsockopt(struct socket *sock, int level, int optname,
2159                            char __user *optval, int __user *optlen)
2160 {
2161         struct sock *sk = sock->sk;
2162
2163         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2164 }
2165 EXPORT_SYMBOL(sock_common_getsockopt);
2166
2167 #ifdef CONFIG_COMPAT
2168 int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
2169                                   char __user *optval, int __user *optlen)
2170 {
2171         struct sock *sk = sock->sk;
2172
2173         if (sk->sk_prot->compat_getsockopt != NULL)
2174                 return sk->sk_prot->compat_getsockopt(sk, level, optname,
2175                                                       optval, optlen);
2176         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
2177 }
2178 EXPORT_SYMBOL(compat_sock_common_getsockopt);
2179 #endif
2180
2181 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
2182                         struct msghdr *msg, size_t size, int flags)
2183 {
2184         struct sock *sk = sock->sk;
2185         int addr_len = 0;
2186         int err;
2187
2188         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
2189                                    flags & ~MSG_DONTWAIT, &addr_len);
2190         if (err >= 0)
2191                 msg->msg_namelen = addr_len;
2192         return err;
2193 }
2194 EXPORT_SYMBOL(sock_common_recvmsg);
2195
2196 /*
2197  *      Set socket options on an inet socket.
2198  */
2199 int sock_common_setsockopt(struct socket *sock, int level, int optname,
2200                            char __user *optval, unsigned int optlen)
2201 {
2202         struct sock *sk = sock->sk;
2203
2204         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2205 }
2206 EXPORT_SYMBOL(sock_common_setsockopt);
2207
2208 #ifdef CONFIG_COMPAT
2209 int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
2210                                   char __user *optval, unsigned int optlen)
2211 {
2212         struct sock *sk = sock->sk;
2213
2214         if (sk->sk_prot->compat_setsockopt != NULL)
2215                 return sk->sk_prot->compat_setsockopt(sk, level, optname,
2216                                                       optval, optlen);
2217         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
2218 }
2219 EXPORT_SYMBOL(compat_sock_common_setsockopt);
2220 #endif
2221
2222 void sk_common_release(struct sock *sk)
2223 {
2224         if (sk->sk_prot->destroy)
2225                 sk->sk_prot->destroy(sk);
2226
2227         /*
2228          * Observation: when sock_common_release is called, processes have
2229          * no access to socket. But net still has.
2230          * Step one, detach it from networking:
2231          *
2232          * A. Remove from hash tables.
2233          */
2234
2235         sk->sk_prot->unhash(sk);
2236
2237         /*
2238          * In this point socket cannot receive new packets, but it is possible
2239          * that some packets are in flight because some CPU runs receiver and
2240          * did hash table lookup before we unhashed socket. They will achieve
2241          * receive queue and will be purged by socket destructor.
2242          *
2243          * Also we still have packets pending on receive queue and probably,
2244          * our own packets waiting in device queues. sock_destroy will drain
2245          * receive queue, but transmitted packets will delay socket destruction
2246          * until the last reference will be released.
2247          */
2248
2249         sock_orphan(sk);
2250
2251         xfrm_sk_free_policy(sk);
2252
2253         sk_refcnt_debug_release(sk);
2254         sock_put(sk);
2255 }
2256 EXPORT_SYMBOL(sk_common_release);
2257
2258 static DEFINE_RWLOCK(proto_list_lock);
2259 static LIST_HEAD(proto_list);
2260
2261 #ifdef CONFIG_PROC_FS
2262 #define PROTO_INUSE_NR  64      /* should be enough for the first time */
2263 struct prot_inuse {
2264         int val[PROTO_INUSE_NR];
2265 };
2266
2267 static DECLARE_BITMAP(proto_inuse_idx, PROTO_INUSE_NR);
2268
2269 #ifdef CONFIG_NET_NS
2270 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2271 {
2272         __this_cpu_add(net->core.inuse->val[prot->inuse_idx], val);
2273 }
2274 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2275
2276 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2277 {
2278         int cpu, idx = prot->inuse_idx;
2279         int res = 0;
2280
2281         for_each_possible_cpu(cpu)
2282                 res += per_cpu_ptr(net->core.inuse, cpu)->val[idx];
2283
2284         return res >= 0 ? res : 0;
2285 }
2286 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2287
2288 static int __net_init sock_inuse_init_net(struct net *net)
2289 {
2290         net->core.inuse = alloc_percpu(struct prot_inuse);
2291         return net->core.inuse ? 0 : -ENOMEM;
2292 }
2293
2294 static void __net_exit sock_inuse_exit_net(struct net *net)
2295 {
2296         free_percpu(net->core.inuse);
2297 }
2298
2299 static struct pernet_operations net_inuse_ops = {
2300         .init = sock_inuse_init_net,
2301         .exit = sock_inuse_exit_net,
2302 };
2303
2304 static __init int net_inuse_init(void)
2305 {
2306         if (register_pernet_subsys(&net_inuse_ops))
2307                 panic("Cannot initialize net inuse counters");
2308
2309         return 0;
2310 }
2311
2312 core_initcall(net_inuse_init);
2313 #else
2314 static DEFINE_PER_CPU(struct prot_inuse, prot_inuse);
2315
2316 void sock_prot_inuse_add(struct net *net, struct proto *prot, int val)
2317 {
2318         __this_cpu_add(prot_inuse.val[prot->inuse_idx], val);
2319 }
2320 EXPORT_SYMBOL_GPL(sock_prot_inuse_add);
2321
2322 int sock_prot_inuse_get(struct net *net, struct proto *prot)
2323 {
2324         int cpu, idx = prot->inuse_idx;
2325         int res = 0;
2326
2327         for_each_possible_cpu(cpu)
2328                 res += per_cpu(prot_inuse, cpu).val[idx];
2329
2330         return res >= 0 ? res : 0;
2331 }
2332 EXPORT_SYMBOL_GPL(sock_prot_inuse_get);
2333 #endif
2334
2335 static void assign_proto_idx(struct proto *prot)
2336 {
2337         prot->inuse_idx = find_first_zero_bit(proto_inuse_idx, PROTO_INUSE_NR);
2338
2339         if (unlikely(prot->inuse_idx == PROTO_INUSE_NR - 1)) {
2340                 printk(KERN_ERR "PROTO_INUSE_NR exhausted\n");
2341                 return;
2342         }
2343
2344         set_bit(prot->inuse_idx, proto_inuse_idx);
2345 }
2346
2347 static void release_proto_idx(struct proto *prot)
2348 {
2349         if (prot->inuse_idx != PROTO_INUSE_NR - 1)
2350                 clear_bit(prot->inuse_idx, proto_inuse_idx);
2351 }
2352 #else
2353 static inline void assign_proto_idx(struct proto *prot)
2354 {
2355 }
2356
2357 static inline void release_proto_idx(struct proto *prot)
2358 {
2359 }
2360 #endif
2361
2362 int proto_register(struct proto *prot, int alloc_slab)
2363 {
2364         if (alloc_slab) {
2365                 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
2366                                         SLAB_HWCACHE_ALIGN | prot->slab_flags,
2367                                         NULL);
2368
2369                 if (prot->slab == NULL) {
2370                         printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
2371                                prot->name);
2372                         goto out;
2373                 }
2374
2375                 if (prot->rsk_prot != NULL) {
2376                         prot->rsk_prot->slab_name = kasprintf(GFP_KERNEL, "request_sock_%s", prot->name);
2377                         if (prot->rsk_prot->slab_name == NULL)
2378                                 goto out_free_sock_slab;
2379
2380                         prot->rsk_prot->slab = kmem_cache_create(prot->rsk_prot->slab_name,
2381                                                                  prot->rsk_prot->obj_size, 0,
2382                                                                  SLAB_HWCACHE_ALIGN, NULL);
2383
2384                         if (prot->rsk_prot->slab == NULL) {
2385                                 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
2386                                        prot->name);
2387                                 goto out_free_request_sock_slab_name;
2388                         }
2389                 }
2390
2391                 if (prot->twsk_prot != NULL) {
2392                         prot->twsk_prot->twsk_slab_name = kasprintf(GFP_KERNEL, "tw_sock_%s", prot->name);
2393
2394                         if (prot->twsk_prot->twsk_slab_name == NULL)
2395                                 goto out_free_request_sock_slab;
2396
2397                         prot->twsk_prot->twsk_slab =
2398                                 kmem_cache_create(prot->twsk_prot->twsk_slab_name,
2399                                                   prot->twsk_prot->twsk_obj_size,
2400                                                   0,
2401                                                   SLAB_HWCACHE_ALIGN |
2402                                                         prot->slab_flags,
2403                                                   NULL);
2404                         if (prot->twsk_prot->twsk_slab == NULL)
2405                                 goto out_free_timewait_sock_slab_name;
2406                 }
2407         }
2408
2409         write_lock(&proto_list_lock);
2410         list_add(&prot->node, &proto_list);
2411         assign_proto_idx(prot);
2412         write_unlock(&proto_list_lock);
2413         return 0;
2414
2415 out_free_timewait_sock_slab_name:
2416         kfree(prot->twsk_prot->twsk_slab_name);
2417 out_free_request_sock_slab:
2418         if (prot->rsk_prot && prot->rsk_prot->slab) {
2419                 kmem_cache_destroy(prot->rsk_prot->slab);
2420                 prot->rsk_prot->slab = NULL;
2421         }
2422 out_free_request_sock_slab_name:
2423         if (prot->rsk_prot)
2424                 kfree(prot->rsk_prot->slab_name);
2425 out_free_sock_slab:
2426         kmem_cache_destroy(prot->slab);
2427         prot->slab = NULL;
2428 out:
2429         return -ENOBUFS;
2430 }
2431 EXPORT_SYMBOL(proto_register);
2432
2433 void proto_unregister(struct proto *prot)
2434 {
2435         write_lock(&proto_list_lock);
2436         release_proto_idx(prot);
2437         list_del(&prot->node);
2438         write_unlock(&proto_list_lock);
2439
2440         if (prot->slab != NULL) {
2441                 kmem_cache_destroy(prot->slab);
2442                 prot->slab = NULL;
2443         }
2444
2445         if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
2446                 kmem_cache_destroy(prot->rsk_prot->slab);
2447                 kfree(prot->rsk_prot->slab_name);
2448                 prot->rsk_prot->slab = NULL;
2449         }
2450
2451         if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
2452                 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
2453                 kfree(prot->twsk_prot->twsk_slab_name);
2454                 prot->twsk_prot->twsk_slab = NULL;
2455         }
2456 }
2457 EXPORT_SYMBOL(proto_unregister);
2458
2459 #ifdef CONFIG_PROC_FS
2460 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
2461         __acquires(proto_list_lock)
2462 {
2463         read_lock(&proto_list_lock);
2464         return seq_list_start_head(&proto_list, *pos);
2465 }
2466
2467 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2468 {
2469         return seq_list_next(v, &proto_list, pos);
2470 }
2471
2472 static void proto_seq_stop(struct seq_file *seq, void *v)
2473         __releases(proto_list_lock)
2474 {
2475         read_unlock(&proto_list_lock);
2476 }
2477
2478 static char proto_method_implemented(const void *method)
2479 {
2480         return method == NULL ? 'n' : 'y';
2481 }
2482
2483 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
2484 {
2485         seq_printf(seq, "%-9s %4u %6d  %6ld   %-3s %6u   %-3s  %-10s "
2486                         "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
2487                    proto->name,
2488                    proto->obj_size,
2489                    sock_prot_inuse_get(seq_file_net(seq), proto),
2490                    proto->memory_allocated != NULL ? atomic_long_read(proto->memory_allocated) : -1L,
2491                    proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
2492                    proto->max_header,
2493                    proto->slab == NULL ? "no" : "yes",
2494                    module_name(proto->owner),
2495                    proto_method_implemented(proto->close),
2496                    proto_method_implemented(proto->connect),
2497                    proto_method_implemented(proto->disconnect),
2498                    proto_method_implemented(proto->accept),
2499                    proto_method_implemented(proto->ioctl),
2500                    proto_method_implemented(proto->init),
2501                    proto_method_implemented(proto->destroy),
2502                    proto_method_implemented(proto->shutdown),
2503                    proto_method_implemented(proto->setsockopt),
2504                    proto_method_implemented(proto->getsockopt),
2505                    proto_method_implemented(proto->sendmsg),
2506                    proto_method_implemented(proto->recvmsg),
2507                    proto_method_implemented(proto->sendpage),
2508                    proto_method_implemented(proto->bind),
2509                    proto_method_implemented(proto->backlog_rcv),
2510                    proto_method_implemented(proto->hash),
2511                    proto_method_implemented(proto->unhash),
2512                    proto_method_implemented(proto->get_port),
2513                    proto_method_implemented(proto->enter_memory_pressure));
2514 }
2515
2516 static int proto_seq_show(struct seq_file *seq, void *v)
2517 {
2518         if (v == &proto_list)
2519                 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
2520                            "protocol",
2521                            "size",
2522                            "sockets",
2523                            "memory",
2524                            "press",
2525                            "maxhdr",
2526                            "slab",
2527                            "module",
2528                            "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
2529         else
2530                 proto_seq_printf(seq, list_entry(v, struct proto, node));
2531         return 0;
2532 }
2533
2534 static const struct seq_operations proto_seq_ops = {
2535         .start  = proto_seq_start,
2536         .next   = proto_seq_next,
2537         .stop   = proto_seq_stop,
2538         .show   = proto_seq_show,
2539 };
2540
2541 static int proto_seq_open(struct inode *inode, struct file *file)
2542 {
2543         return seq_open_net(inode, file, &proto_seq_ops,
2544                             sizeof(struct seq_net_private));
2545 }
2546
2547 static const struct file_operations proto_seq_fops = {
2548         .owner          = THIS_MODULE,
2549         .open           = proto_seq_open,
2550         .read           = seq_read,
2551         .llseek         = seq_lseek,
2552         .release        = seq_release_net,
2553 };
2554
2555 static __net_init int proto_init_net(struct net *net)
2556 {
2557         if (!proc_net_fops_create(net, "protocols", S_IRUGO, &proto_seq_fops))
2558                 return -ENOMEM;
2559
2560         return 0;
2561 }
2562
2563 static __net_exit void proto_exit_net(struct net *net)
2564 {
2565         proc_net_remove(net, "protocols");
2566 }
2567
2568
2569 static __net_initdata struct pernet_operations proto_net_ops = {
2570         .init = proto_init_net,
2571         .exit = proto_exit_net,
2572 };
2573
2574 static int __init proto_init(void)
2575 {
2576         return register_pernet_subsys(&proto_net_ops);
2577 }
2578
2579 subsys_initcall(proto_init);
2580
2581 #endif /* PROC_FS */