fedora core 6 1.2949 + vserver 2.2.0
[linux-2.6.git] / net / core / sock.c
index 5cc0268..3ca5f8d 100644 (file)
@@ -9,7 +9,7 @@
  *
  * Version:    $Id: sock.c,v 1.117 2002/02/01 22:01:03 davem Exp $
  *
- * Authors:    Ross Biro, <bir7@leland.Stanford.Edu>
+ * Authors:    Ross Biro
  *             Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  *             Florian La Roche, <flla@stud.uni-sb.de>
  *             Alan Cox, <A.Cox@swansea.ac.uk>
  *             2 of the License, or (at your option) any later version.
  */
 
-#include <linux/config.h>
+#include <linux/capability.h>
 #include <linux/errno.h>
 #include <linux/types.h>
 #include <linux/socket.h>
 #include <linux/in.h>
 #include <linux/kernel.h>
-#include <linux/major.h>
 #include <linux/module.h>
+#include <linux/proc_fs.h>
+#include <linux/seq_file.h>
 #include <linux/sched.h>
 #include <linux/timer.h>
 #include <linux/string.h>
 #include <linux/poll.h>
 #include <linux/tcp.h>
 #include <linux/init.h>
+#include <linux/highmem.h>
 
 #include <asm/uaccess.h>
 #include <asm/system.h>
 #include <linux/netdevice.h>
 #include <net/protocol.h>
 #include <linux/skbuff.h>
+#include <net/request_sock.h>
 #include <net/sock.h>
+#include <net/xfrm.h>
 #include <linux/ipsec.h>
 
 #include <linux/filter.h>
 #include <linux/vs_socket.h>
+#include <linux/vs_limit.h>
+#include <linux/vs_context.h>
 
 #ifdef CONFIG_INET
 #include <net/tcp.h>
 #endif
 
+/*
+ * Each address family might have different locking rules, so we have
+ * one slock key per address family:
+ */
+static struct lock_class_key af_family_keys[AF_MAX];
+static struct lock_class_key af_family_slock_keys[AF_MAX];
+
+#ifdef CONFIG_DEBUG_LOCK_ALLOC
+/*
+ * Make lock validator output more readable. (we pre-construct these
+ * strings build-time, so that runtime initialization of socket
+ * locks is fast):
+ */
+static const char *af_family_key_strings[AF_MAX+1] = {
+  "sk_lock-AF_UNSPEC", "sk_lock-AF_UNIX"     , "sk_lock-AF_INET"     ,
+  "sk_lock-AF_AX25"  , "sk_lock-AF_IPX"      , "sk_lock-AF_APPLETALK",
+  "sk_lock-AF_NETROM", "sk_lock-AF_BRIDGE"   , "sk_lock-AF_ATMPVC"   ,
+  "sk_lock-AF_X25"   , "sk_lock-AF_INET6"    , "sk_lock-AF_ROSE"     ,
+  "sk_lock-AF_DECnet", "sk_lock-AF_NETBEUI"  , "sk_lock-AF_SECURITY" ,
+  "sk_lock-AF_KEY"   , "sk_lock-AF_NETLINK"  , "sk_lock-AF_PACKET"   ,
+  "sk_lock-AF_ASH"   , "sk_lock-AF_ECONET"   , "sk_lock-AF_ATMSVC"   ,
+  "sk_lock-21"       , "sk_lock-AF_SNA"      , "sk_lock-AF_IRDA"     ,
+  "sk_lock-AF_PPPOX" , "sk_lock-AF_WANPIPE"  , "sk_lock-AF_LLC"      ,
+  "sk_lock-27"       , "sk_lock-28"          , "sk_lock-29"          ,
+  "sk_lock-AF_TIPC"  , "sk_lock-AF_BLUETOOTH", "sk_lock-AF_MAX"
+};
+static const char *af_family_slock_key_strings[AF_MAX+1] = {
+  "slock-AF_UNSPEC", "slock-AF_UNIX"     , "slock-AF_INET"     ,
+  "slock-AF_AX25"  , "slock-AF_IPX"      , "slock-AF_APPLETALK",
+  "slock-AF_NETROM", "slock-AF_BRIDGE"   , "slock-AF_ATMPVC"   ,
+  "slock-AF_X25"   , "slock-AF_INET6"    , "slock-AF_ROSE"     ,
+  "slock-AF_DECnet", "slock-AF_NETBEUI"  , "slock-AF_SECURITY" ,
+  "slock-AF_KEY"   , "slock-AF_NETLINK"  , "slock-AF_PACKET"   ,
+  "slock-AF_ASH"   , "slock-AF_ECONET"   , "slock-AF_ATMSVC"   ,
+  "slock-21"       , "slock-AF_SNA"      , "slock-AF_IRDA"     ,
+  "slock-AF_PPPOX" , "slock-AF_WANPIPE"  , "slock-AF_LLC"      ,
+  "slock-27"       , "slock-28"          , "slock-29"          ,
+  "slock-AF_TIPC"  , "slock-AF_BLUETOOTH", "slock-AF_MAX"
+};
+#endif
+
+/*
+ * sk_callback_lock locking rules are per-address-family,
+ * so split the lock classes by using a per-AF key:
+ */
+static struct lock_class_key af_callback_keys[AF_MAX];
+
 /* Take into consideration the size of the struct sk_buff overhead in the
  * determination of these values, since that is non-constant across
  * platforms.  This makes socket queueing behavior and performance
 #define SK_RMEM_MAX            (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
 
 /* Run time adjustable parameters. */
-__u32 sysctl_wmem_max = SK_WMEM_MAX;
-__u32 sysctl_rmem_max = SK_RMEM_MAX;
-__u32 sysctl_wmem_default = SK_WMEM_MAX;
-__u32 sysctl_rmem_default = SK_RMEM_MAX;
+__u32 sysctl_wmem_max __read_mostly = SK_WMEM_MAX;
+__u32 sysctl_rmem_max __read_mostly = SK_RMEM_MAX;
+__u32 sysctl_wmem_default __read_mostly = SK_WMEM_MAX;
+__u32 sysctl_rmem_default __read_mostly = SK_RMEM_MAX;
 
 /* Maximal space eaten by iovec or ancilliary data plus some space */
-int sysctl_optmem_max = sizeof(unsigned long)*(2*UIO_MAXIOV + 512);
+int sysctl_optmem_max __read_mostly = sizeof(unsigned long)*(2*UIO_MAXIOV+512);
 
 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
 {
@@ -166,7 +219,7 @@ static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
 static void sock_warn_obsolete_bsdism(const char *name)
 {
        static int warned;
-       static char warncomm[16];
+       static char warncomm[TASK_COMM_LEN];
        if (strcmp(warncomm, current->comm) && warned < 5) { 
                strcpy(warncomm,  current->comm); 
                printk(KERN_WARNING "process `%s' is using obsolete "
@@ -175,6 +228,114 @@ static void sock_warn_obsolete_bsdism(const char *name)
        }
 }
 
+static void sock_disable_timestamp(struct sock *sk)
+{      
+       if (sock_flag(sk, SOCK_TIMESTAMP)) { 
+               sock_reset_flag(sk, SOCK_TIMESTAMP);
+               net_disable_timestamp();
+       }
+}
+
+
+int sock_queue_rcv_skb(struct sock *sk, struct sk_buff *skb)
+{
+       int err = 0;
+       int skb_len;
+
+       /* Cast skb->rcvbuf to unsigned... It's pointless, but reduces
+          number of warnings when compiling with -W --ANK
+        */
+       if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
+           (unsigned)sk->sk_rcvbuf) {
+               err = -ENOMEM;
+               goto out;
+       }
+
+       err = sk_filter(sk, skb);
+       if (err)
+               goto out;
+
+       skb->dev = NULL;
+       skb_set_owner_r(skb, sk);
+
+       /* Cache the SKB length before we tack it onto the receive
+        * queue.  Once it is added it no longer belongs to us and
+        * may be freed by other threads of control pulling packets
+        * from the queue.
+        */
+       skb_len = skb->len;
+
+       skb_queue_tail(&sk->sk_receive_queue, skb);
+
+       if (!sock_flag(sk, SOCK_DEAD))
+               sk->sk_data_ready(sk, skb_len);
+out:
+       return err;
+}
+EXPORT_SYMBOL(sock_queue_rcv_skb);
+
+int sk_receive_skb(struct sock *sk, struct sk_buff *skb, const int nested)
+{
+       int rc = NET_RX_SUCCESS;
+
+       if (sk_filter(sk, skb))
+               goto discard_and_relse;
+
+       skb->dev = NULL;
+
+       if (nested)
+               bh_lock_sock_nested(sk);
+       else
+               bh_lock_sock(sk);
+       if (!sock_owned_by_user(sk)) {
+               /*
+                * trylock + unlock semantics:
+                */
+               mutex_acquire(&sk->sk_lock.dep_map, 0, 1, _RET_IP_);
+
+               rc = sk->sk_backlog_rcv(sk, skb);
+
+               mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
+       } else
+               sk_add_backlog(sk, skb);
+       bh_unlock_sock(sk);
+out:
+       sock_put(sk);
+       return rc;
+discard_and_relse:
+       kfree_skb(skb);
+       goto out;
+}
+EXPORT_SYMBOL(sk_receive_skb);
+
+struct dst_entry *__sk_dst_check(struct sock *sk, u32 cookie)
+{
+       struct dst_entry *dst = sk->sk_dst_cache;
+
+       if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
+               sk->sk_dst_cache = NULL;
+               dst_release(dst);
+               return NULL;
+       }
+
+       return dst;
+}
+EXPORT_SYMBOL(__sk_dst_check);
+
+struct dst_entry *sk_dst_check(struct sock *sk, u32 cookie)
+{
+       struct dst_entry *dst = sk_dst_get(sk);
+
+       if (dst && dst->obsolete && dst->ops->check(dst, cookie) == NULL) {
+               sk_dst_reset(sk);
+               dst_release(dst);
+               return NULL;
+       }
+
+       return dst;
+}
+EXPORT_SYMBOL(sk_dst_check);
+
 /*
  *     This is meant for all protocols to use and covers goings on
  *     at the socket level. Everything here is generic.
@@ -195,13 +356,14 @@ int sock_setsockopt(struct socket *sock, int level, int optname,
         */
 
 #ifdef SO_DONTLINGER           /* Compatibility item... */
-       switch (optname) {
-               case SO_DONTLINGER:
-                       sock_reset_flag(sk, SOCK_LINGER);
-                       return 0;
+       if (optname == SO_DONTLINGER) {
+               lock_sock(sk);
+               sock_reset_flag(sk, SOCK_LINGER);
+               release_sock(sk);
+               return 0;
        }
-#endif 
-               
+#endif
+       
        if(optlen<sizeof(int))
                return(-EINVAL);
        
@@ -219,8 +381,10 @@ int sock_setsockopt(struct socket *sock, int level, int optname,
                        {
                                ret = -EACCES;
                        }
+                       else if (valbool)
+                               sock_set_flag(sk, SOCK_DBG);
                        else
-                               sk->sk_debug = valbool;
+                               sock_reset_flag(sk, SOCK_DBG);
                        break;
                case SO_REUSEADDR:
                        sk->sk_reuse = valbool;
@@ -230,7 +394,10 @@ int sock_setsockopt(struct socket *sock, int level, int optname,
                        ret = -ENOPROTOOPT;
                        break;
                case SO_DONTROUTE:
-                       sk->sk_localroute = valbool;
+                       if (valbool)
+                               sock_set_flag(sk, SOCK_LOCALROUTE);
+                       else
+                               sock_reset_flag(sk, SOCK_LOCALROUTE);
                        break;
                case SO_BROADCAST:
                        sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
@@ -243,7 +410,7 @@ int sock_setsockopt(struct socket *sock, int level, int optname,
                           
                        if (val > sysctl_wmem_max)
                                val = sysctl_wmem_max;
-
+set_sndbuf:
                        sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
                        if ((val * 2) < SOCK_MIN_SNDBUF)
                                sk->sk_sndbuf = SOCK_MIN_SNDBUF;
@@ -257,6 +424,13 @@ int sock_setsockopt(struct socket *sock, int level, int optname,
                        sk->sk_write_space(sk);
                        break;
 
+               case SO_SNDBUFFORCE:
+                       if (!capable(CAP_NET_ADMIN)) {
+                               ret = -EPERM;
+                               break;
+                       }
+                       goto set_sndbuf;
+
                case SO_RCVBUF:
                        /* Don't error on this BSD doesn't and if you think
                           about it this is right. Otherwise apps have to
@@ -265,15 +439,36 @@ int sock_setsockopt(struct socket *sock, int level, int optname,
                          
                        if (val > sysctl_rmem_max)
                                val = sysctl_rmem_max;
-
+set_rcvbuf:
                        sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
-                       /* FIXME: is this lower bound the right one? */
+                       /*
+                        * We double it on the way in to account for
+                        * "struct sk_buff" etc. overhead.   Applications
+                        * assume that the SO_RCVBUF setting they make will
+                        * allow that much actual data to be received on that
+                        * socket.
+                        *
+                        * Applications are unaware that "struct sk_buff" and
+                        * other overheads allocate from the receive buffer
+                        * during socket buffer allocation.
+                        *
+                        * And after considering the possible alternatives,
+                        * returning the value we actually used in getsockopt
+                        * is the most desirable behavior.
+                        */
                        if ((val * 2) < SOCK_MIN_RCVBUF)
                                sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
                        else
                                sk->sk_rcvbuf = val * 2;
                        break;
 
+               case SO_RCVBUFFORCE:
+                       if (!capable(CAP_NET_ADMIN)) {
+                               ret = -EPERM;
+                               break;
+                       }
+                       goto set_rcvbuf;
+
                case SO_KEEPALIVE:
 #ifdef CONFIG_INET
                        if (sk->sk_protocol == IPPROTO_TCP)
@@ -310,11 +505,11 @@ int sock_setsockopt(struct socket *sock, int level, int optname,
                                sock_reset_flag(sk, SOCK_LINGER);
                        else {
 #if (BITS_PER_LONG == 32)
-                               if (ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
+                               if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
                                        sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
                                else
 #endif
-                                       sk->sk_lingertime = ling.l_linger * HZ;
+                                       sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
                                sock_set_flag(sk, SOCK_LINGER);
                        }
                        break;
@@ -325,15 +520,17 @@ int sock_setsockopt(struct socket *sock, int level, int optname,
 
                case SO_PASSCRED:
                        if (valbool)
-                               set_bit(SOCK_PASS_CRED, &sock->flags);
+                               set_bit(SOCK_PASSCRED, &sock->flags);
                        else
-                               clear_bit(SOCK_PASS_CRED, &sock->flags);
+                               clear_bit(SOCK_PASSCRED, &sock->flags);
                        break;
 
                case SO_TIMESTAMP:
-                       sk->sk_rcvtstamp = valbool;
-                       if (valbool) 
+                       if (valbool)  {
+                               sock_set_flag(sk, SOCK_RCVTSTAMP);
                                sock_enable_timestamp(sk);
+                       } else
+                               sock_reset_flag(sk, SOCK_RCVTSTAMP);
                        break;
 
                case SO_RCVLOWAT:
@@ -370,8 +567,9 @@ int sock_setsockopt(struct socket *sock, int level, int optname,
                        if (!valbool) {
                                sk->sk_bound_dev_if = 0;
                        } else {
-                               if (optlen > IFNAMSIZ) 
-                                       optlen = IFNAMSIZ; 
+                               if (optlen > IFNAMSIZ - 1)
+                                       optlen = IFNAMSIZ - 1;
+                               memset(devname, 0, sizeof(devname));
                                if (copy_from_user(devname, optval, optlen)) {
                                        ret = -EFAULT;
                                        break;
@@ -411,18 +609,25 @@ int sock_setsockopt(struct socket *sock, int level, int optname,
                        break;
 
                case SO_DETACH_FILTER:
-                       spin_lock_bh(&sk->sk_lock.slock);
-                       filter = sk->sk_filter;
+                       rcu_read_lock_bh();
+                       filter = rcu_dereference(sk->sk_filter);
                         if (filter) {
-                               sk->sk_filter = NULL;
-                               spin_unlock_bh(&sk->sk_lock.slock);
+                               rcu_assign_pointer(sk->sk_filter, NULL);
                                sk_filter_release(sk, filter);
+                               rcu_read_unlock_bh();
                                break;
                        }
-                       spin_unlock_bh(&sk->sk_lock.slock);
+                       rcu_read_unlock_bh();
                        ret = -ENONET;
                        break;
 
+               case SO_PASSSEC:
+                       if (valbool)
+                               set_bit(SOCK_PASSSEC, &sock->flags);
+                       else
+                               clear_bit(SOCK_PASSSEC, &sock->flags);
+                       break;
+
                /* We implement the SO_SNDLOWAT etc to
                   not be settable (1003.1g 5.3) */
                default:
@@ -446,7 +651,8 @@ int sock_getsockopt(struct socket *sock, int level, int optname,
                struct timeval tm;
        } v;
        
-       unsigned int lv=sizeof(int),len;
+       unsigned int lv = sizeof(int);
+       int len;
        
        if(get_user(len,optlen))
                return -EFAULT;
@@ -456,11 +662,11 @@ int sock_getsockopt(struct socket *sock, int level, int optname,
        switch(optname) 
        {
                case SO_DEBUG:          
-                       v.val = sk->sk_debug;
+                       v.val = sock_flag(sk, SOCK_DBG);
                        break;
                
                case SO_DONTROUTE:
-                       v.val = sk->sk_localroute;
+                       v.val = sock_flag(sk, SOCK_LOCALROUTE);
                        break;
                
                case SO_BROADCAST:
@@ -516,7 +722,7 @@ int sock_getsockopt(struct socket *sock, int level, int optname,
                        break;
 
                case SO_TIMESTAMP:
-                       v.val = sk->sk_rcvtstamp;
+                       v.val = sock_flag(sk, SOCK_RCVTSTAMP);
                        break;
 
                case SO_RCVTIMEO:
@@ -526,7 +732,7 @@ int sock_getsockopt(struct socket *sock, int level, int optname,
                                v.tm.tv_usec = 0;
                        } else {
                                v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
-                               v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000) / HZ;
+                               v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
                        }
                        break;
 
@@ -537,7 +743,7 @@ int sock_getsockopt(struct socket *sock, int level, int optname,
                                v.tm.tv_usec = 0;
                        } else {
                                v.tm.tv_sec = sk->sk_sndtimeo / HZ;
-                               v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000) / HZ;
+                               v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
                        }
                        break;
 
@@ -550,7 +756,7 @@ int sock_getsockopt(struct socket *sock, int level, int optname,
                        break; 
 
                case SO_PASSCRED:
-                       v.val = test_bit(SOCK_PASS_CRED, &sock->flags)?1:0;
+                       v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
                        break;
 
                case SO_PEERCRED:
@@ -580,8 +786,12 @@ int sock_getsockopt(struct socket *sock, int level, int optname,
                        v.val = sk->sk_state == TCP_LISTEN;
                        break;
 
+               case SO_PASSSEC:
+                       v.val = test_bit(SOCK_PASSSEC, &sock->flags) ? 1 : 0;
+                       break;
+
                case SO_PEERSEC:
-                       return security_socket_getpeersec(sock, optval, optlen, len);
+                       return security_socket_getpeersec_stream(sock, optval, optlen, len);
 
                default:
                        return(-ENOPROTOOPT);
@@ -596,58 +806,80 @@ lenout:
        return 0;
 }
 
-static kmem_cache_t *sk_cachep;
+/*
+ * Initialize an sk_lock.
+ *
+ * (We also register the sk_lock with the lock validator.)
+ */
+static void inline sock_lock_init(struct sock *sk)
+{
+       sock_lock_init_class_and_name(sk,
+                       af_family_slock_key_strings[sk->sk_family],
+                       af_family_slock_keys + sk->sk_family,
+                       af_family_key_strings[sk->sk_family],
+                       af_family_keys + sk->sk_family);
+}
 
 /**
  *     sk_alloc - All socket objects are allocated here
- *     @family - protocol family
- *     @priority - for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
- *     @zero_it - zeroes the allocated sock
- *     @slab - alternate slab
- *
- *     All socket objects are allocated here. If @zero_it is non-zero
- *     it should have the size of the are to be zeroed, because the
- *     private slabcaches have different sizes of the generic struct sock.
- *     1 has been kept as a way to say sizeof(struct sock).
+ *     @family: protocol family
+ *     @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
+ *     @prot: struct proto associated with this new sock instance
+ *     @zero_it: if we should zero the newly allocated sock
  */
-struct sock *sk_alloc(int family, int priority, int zero_it, kmem_cache_t *slab)
+struct sock *sk_alloc(int family, gfp_t priority,
+                     struct proto *prot, int zero_it)
 {
        struct sock *sk = NULL;
+       struct kmem_cache *slab = prot->slab;
+
+       if (slab != NULL)
+               sk = kmem_cache_alloc(slab, priority);
+       else
+               sk = kmalloc(prot->obj_size, priority);
 
-       if (!slab)
-               slab = sk_cachep;
-       sk = kmem_cache_alloc(slab, priority);
        if (sk) {
                if (zero_it) {
-                       memset(sk, 0,
-                              zero_it == 1 ? sizeof(struct sock) : zero_it);
+                       memset(sk, 0, prot->obj_size);
                        sk->sk_family = family;
+                       /*
+                        * See comment in struct sock definition to understand
+                        * why we need sk_prot_creator -acme
+                        */
+                       sk->sk_prot = sk->sk_prot_creator = prot;
                        sock_lock_init(sk);
                }
-               sk->sk_slab = slab;
                sock_vx_init(sk);
                sock_nx_init(sk);
                
-               if (security_sk_alloc(sk, family, priority)) {
-                       kmem_cache_free(slab, sk);
-                       sk = NULL;
-               }
+               if (security_sk_alloc(sk, family, priority))
+                       goto out_free;
+
+               if (!try_module_get(prot->owner))
+                       goto out_free;
        }
        return sk;
+
+out_free:
+       if (slab != NULL)
+               kmem_cache_free(slab, sk);
+       else
+               kfree(sk);
+       return NULL;
 }
 
 void sk_free(struct sock *sk)
 {
        struct sk_filter *filter;
-       struct module *owner = sk->sk_owner;
+       struct module *owner = sk->sk_prot_creator->owner;
 
        if (sk->sk_destruct)
                sk->sk_destruct(sk);
 
-       filter = sk->sk_filter;
+       filter = rcu_dereference(sk->sk_filter);
        if (filter) {
                sk_filter_release(sk, filter);
-               sk->sk_filter = NULL;
+               rcu_assign_pointer(sk->sk_filter, NULL);
        }
 
        sock_disable_timestamp(sk);
@@ -657,19 +889,107 @@ void sk_free(struct sock *sk)
                       __FUNCTION__, atomic_read(&sk->sk_omem_alloc));
 
        security_sk_free(sk);
-       BUG_ON(sk->sk_vx_info);
-       BUG_ON(sk->sk_nx_info);
-       kmem_cache_free(sk->sk_slab, sk);
+       vx_sock_dec(sk);
+       clr_vx_info(&sk->sk_vx_info);
+       sk->sk_xid = -1;
+       clr_nx_info(&sk->sk_nx_info);
+       sk->sk_nid = -1;
+       if (sk->sk_prot_creator->slab != NULL)
+               kmem_cache_free(sk->sk_prot_creator->slab, sk);
+       else
+               kfree(sk);
        module_put(owner);
 }
 
-void __init sk_init(void)
+struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
 {
-       sk_cachep = kmem_cache_create("sock", sizeof(struct sock), 0,
-                                     SLAB_HWCACHE_ALIGN, 0, 0);
-       if (!sk_cachep)
-               printk(KERN_CRIT "sk_init: Cannot create sock SLAB cache!");
+       struct sock *newsk = sk_alloc(sk->sk_family, priority, sk->sk_prot, 0);
+
+       if (newsk != NULL) {
+               struct sk_filter *filter;
+
+               sock_copy(newsk, sk);
+
+               /* SANITY */
+               sock_vx_init(newsk);
+               sock_nx_init(newsk);
+               sk_node_init(&newsk->sk_node);
+               sock_lock_init(newsk);
+               bh_lock_sock(newsk);
+
+               atomic_set(&newsk->sk_rmem_alloc, 0);
+               atomic_set(&newsk->sk_wmem_alloc, 0);
+               atomic_set(&newsk->sk_omem_alloc, 0);
+               skb_queue_head_init(&newsk->sk_receive_queue);
+               skb_queue_head_init(&newsk->sk_write_queue);
+#ifdef CONFIG_NET_DMA
+               skb_queue_head_init(&newsk->sk_async_wait_queue);
+#endif
+
+               rwlock_init(&newsk->sk_dst_lock);
+               rwlock_init(&newsk->sk_callback_lock);
+               lockdep_set_class(&newsk->sk_callback_lock,
+                                  af_callback_keys + newsk->sk_family);
+
+               newsk->sk_dst_cache     = NULL;
+               newsk->sk_wmem_queued   = 0;
+               newsk->sk_forward_alloc = 0;
+               newsk->sk_send_head     = NULL;
+               newsk->sk_backlog.head  = newsk->sk_backlog.tail = NULL;
+               newsk->sk_userlocks     = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
+
+               sock_reset_flag(newsk, SOCK_DONE);
+               skb_queue_head_init(&newsk->sk_error_queue);
+
+               filter = newsk->sk_filter;
+               if (filter != NULL)
+                       sk_filter_charge(newsk, filter);
+
+               if (unlikely(xfrm_sk_clone_policy(newsk))) {
+                       /* It is still raw copy of parent, so invalidate
+                        * destructor and make plain sk_free() */
+                       newsk->sk_destruct = NULL;
+                       sk_free(newsk);
+                       newsk = NULL;
+                       goto out;
+               }
+
+               newsk->sk_err      = 0;
+               newsk->sk_priority = 0;
+               atomic_set(&newsk->sk_refcnt, 2);
+
+               set_vx_info(&newsk->sk_vx_info, sk->sk_vx_info);
+               newsk->sk_xid = sk->sk_xid;
+               vx_sock_inc(newsk);
+               set_nx_info(&newsk->sk_nx_info, sk->sk_nx_info);
+               newsk->sk_nid = sk->sk_nid;
+
+               /*
+                * Increment the counter in the same struct proto as the master
+                * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
+                * is the same as sk->sk_prot->socks, as this field was copied
+                * with memcpy).
+                *
+                * This _changes_ the previous behaviour, where
+                * tcp_create_openreq_child always was incrementing the
+                * equivalent to tcp_prot->socks (inet_sock_nr), so this have
+                * to be taken into account in all callers. -acme
+                */
+               sk_refcnt_debug_inc(newsk);
+               newsk->sk_socket = NULL;
+               newsk->sk_sleep  = NULL;
+
+               if (newsk->sk_prot->sockets_allocated)
+                       atomic_inc(newsk->sk_prot->sockets_allocated);
+       }
+out:
+       return newsk;
+}
 
+EXPORT_SYMBOL_GPL(sk_clone);
+
+void __init sk_init(void)
+{
        if (num_physpages <= 4096) {
                sysctl_wmem_max = 32767;
                sysctl_rmem_max = 32767;
@@ -695,7 +1015,7 @@ void sock_wfree(struct sk_buff *skb)
 
        /* In case it might be waiting for more memory. */
        atomic_sub(skb->truesize, &sk->sk_wmem_alloc);
-       if (!sk->sk_use_write_queue)
+       if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE))
                sk->sk_write_space(sk);
        sock_put(sk);
 }
@@ -710,10 +1030,32 @@ void sock_rfree(struct sk_buff *skb)
        atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
 }
 
+
+int sock_i_uid(struct sock *sk)
+{
+       int uid;
+
+       read_lock(&sk->sk_callback_lock);
+       uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
+       read_unlock(&sk->sk_callback_lock);
+       return uid;
+}
+
+unsigned long sock_i_ino(struct sock *sk)
+{
+       unsigned long ino;
+
+       read_lock(&sk->sk_callback_lock);
+       ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
+       read_unlock(&sk->sk_callback_lock);
+       return ino;
+}
+
 /*
  * Allocate a skb from the socket's send buffer.
  */
-struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, int priority)
+struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
+                            gfp_t priority)
 {
        if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
                struct sk_buff * skb = alloc_skb(size, priority);
@@ -728,7 +1070,8 @@ struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force, int
 /*
  * Allocate a skb from the socket's receive buffer.
  */ 
-struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force, int priority)
+struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
+                            gfp_t priority)
 {
        if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
                struct sk_buff *skb = alloc_skb(size, priority);
@@ -743,7 +1086,7 @@ struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force, int
 /* 
  * Allocate a memory block from the socket's option memory buffer.
  */ 
-void *sock_kmalloc(struct sock *sk, int size, int priority)
+void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
 {
        if ((unsigned)size <= sysctl_optmem_max &&
            atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
@@ -801,11 +1144,13 @@ static long sock_wait_for_wmem(struct sock * sk, long timeo)
  *     Generic send/receive buffer handlers
  */
 
-struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
-                                    unsigned long data_len, int noblock, int *errcode)
+static struct sk_buff *sock_alloc_send_pskb(struct sock *sk,
+                                           unsigned long header_len,
+                                           unsigned long data_len,
+                                           int noblock, int *errcode)
 {
        struct sk_buff *skb;
-       unsigned int gfp_mask;
+       gfp_t gfp_mask;
        long timeo;
        int err;
 
@@ -824,7 +1169,7 @@ struct sk_buff *sock_alloc_send_pskb(struct sock *sk, unsigned long header_len,
                        goto failure;
 
                if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
-                       skb = alloc_skb(header_len, sk->sk_allocation);
+                       skb = alloc_skb(header_len, gfp_mask);
                        if (skb) {
                                int npages;
                                int i;
@@ -889,7 +1234,7 @@ struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
        return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
 }
 
-void __lock_sock(struct sock *sk)
+static void __lock_sock(struct sock *sk)
 {
        DEFINE_WAIT(wait);
 
@@ -905,7 +1250,7 @@ void __lock_sock(struct sock *sk)
        finish_wait(&sk->sk_lock.wq, &wait);
 }
 
-void __release_sock(struct sock *sk)
+static void __release_sock(struct sock *sk)
 {
        struct sk_buff *skb = sk->sk_backlog.head;
 
@@ -918,6 +1263,15 @@ void __release_sock(struct sock *sk)
 
                        skb->next = NULL;
                        sk->sk_backlog_rcv(sk, skb);
+
+                       /*
+                        * We are in process context here with softirqs
+                        * disabled, use cond_resched_softirq() to preempt.
+                        * This is safe to do because we've taken the backlog
+                        * queue private:
+                        */
+                       cond_resched_softirq();
+
                        skb = next;
                } while (skb != NULL);
 
@@ -927,8 +1281,8 @@ void __release_sock(struct sock *sk)
 
 /**
  * sk_wait_data - wait for data to arrive at sk_receive_queue
- * sk - sock to wait on
- * timeo - for how long
+ * @sk:    sock to wait on
+ * @timeo: for how long
  *
  * Now socket state including sk->sk_err is changed only under lock,
  * hence we may omit checks after joining wait queue.
@@ -957,11 +1311,6 @@ EXPORT_SYMBOL(sk_wait_data);
  * function, some default processing is provided.
  */
 
-int sock_no_release(struct socket *sock)
-{
-       return 0;
-}
-
 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
 {
        return -EOPNOTSUPP;
@@ -1042,30 +1391,12 @@ int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *
 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
 {
        ssize_t res;
-       struct msghdr msg;
-       struct iovec iov;
-       mm_segment_t old_fs;
-       char *kaddr;
-
-       kaddr = kmap(page);
-
-       msg.msg_name = NULL;
-       msg.msg_namelen = 0;
-       msg.msg_iov = &iov;
-       msg.msg_iovlen = 1;
-       msg.msg_control = NULL;
-       msg.msg_controllen = 0;
-       msg.msg_flags = flags;
-
-       /* This cast is ok because of the "set_fs(KERNEL_DS)" */
-       iov.iov_base = (void __user *) (kaddr + offset);
+       struct msghdr msg = {.msg_flags = flags};
+       struct kvec iov;
+       char *kaddr = kmap(page);
+       iov.iov_base = kaddr + offset;
        iov.iov_len = size;
-
-       old_fs = get_fs();
-       set_fs(KERNEL_DS);
-       res = sock_sendmsg(sock, &msg, size);
-       set_fs(old_fs);
-
+       res = kernel_sendmsg(sock, &msg, &iov, 1, size);
        kunmap(page);
        return res;
 }
@@ -1074,7 +1405,7 @@ ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, siz
  *     Default Socket Callbacks
  */
 
-void sock_def_wakeup(struct sock *sk)
+static void sock_def_wakeup(struct sock *sk)
 {
        read_lock(&sk->sk_callback_lock);
        if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
@@ -1082,7 +1413,7 @@ void sock_def_wakeup(struct sock *sk)
        read_unlock(&sk->sk_callback_lock);
 }
 
-void sock_def_error_report(struct sock *sk)
+static void sock_def_error_report(struct sock *sk)
 {
        read_lock(&sk->sk_callback_lock);
        if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
@@ -1091,7 +1422,7 @@ void sock_def_error_report(struct sock *sk)
        read_unlock(&sk->sk_callback_lock);
 }
 
-void sock_def_readable(struct sock *sk, int len)
+static void sock_def_readable(struct sock *sk, int len)
 {
        read_lock(&sk->sk_callback_lock);
        if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
@@ -1100,7 +1431,7 @@ void sock_def_readable(struct sock *sk, int len)
        read_unlock(&sk->sk_callback_lock);
 }
 
-void sock_def_write_space(struct sock *sk)
+static void sock_def_write_space(struct sock *sk)
 {
        read_lock(&sk->sk_callback_lock);
 
@@ -1119,10 +1450,9 @@ void sock_def_write_space(struct sock *sk)
        read_unlock(&sk->sk_callback_lock);
 }
 
-void sock_def_destruct(struct sock *sk)
+static void sock_def_destruct(struct sock *sk)
 {
-       if (sk->sk_protinfo)
-               kfree(sk->sk_protinfo);
+       kfree(sk->sk_protinfo);
 }
 
 void sk_send_sigurg(struct sock *sk)
@@ -1154,6 +1484,11 @@ void sock_init_data(struct socket *sock, struct sock *sk)
        skb_queue_head_init(&sk->sk_receive_queue);
        skb_queue_head_init(&sk->sk_write_queue);
        skb_queue_head_init(&sk->sk_error_queue);
+#ifdef CONFIG_NET_DMA
+       skb_queue_head_init(&sk->sk_async_wait_queue);
+#endif
+
+       sk->sk_send_head        =       NULL;
 
        init_timer(&sk->sk_timer);
        
@@ -1161,9 +1496,10 @@ void sock_init_data(struct socket *sock, struct sock *sk)
        sk->sk_rcvbuf           =       sysctl_rmem_default;
        sk->sk_sndbuf           =       sysctl_wmem_default;
        sk->sk_state            =       TCP_CLOSE;
-       sk->sk_zapped           =       1;
        sk->sk_socket           =       sock;
 
+       sock_set_flag(sk, SOCK_ZAPPED);
+
        if(sock)
        {
                sk->sk_type     =       sock->type;
@@ -1172,8 +1508,10 @@ void sock_init_data(struct socket *sock, struct sock *sk)
        } else
                sk->sk_sleep    =       NULL;
 
-       sk->sk_dst_lock         =       RW_LOCK_UNLOCKED;
-       sk->sk_callback_lock    =       RW_LOCK_UNLOCKED;
+       rwlock_init(&sk->sk_dst_lock);
+       rwlock_init(&sk->sk_callback_lock);
+       lockdep_set_class(&sk->sk_callback_lock,
+                          af_callback_keys + sk->sk_family);
 
        sk->sk_state_change     =       sock_def_wakeup;
        sk->sk_data_ready       =       sock_def_readable;
@@ -1181,52 +1519,62 @@ void sock_init_data(struct socket *sock, struct sock *sk)
        sk->sk_error_report     =       sock_def_error_report;
        sk->sk_destruct         =       sock_def_destruct;
 
+       sk->sk_sndmsg_page      =       NULL;
+       sk->sk_sndmsg_off       =       0;
+
        sk->sk_peercred.pid     =       0;
        sk->sk_peercred.uid     =       -1;
        sk->sk_peercred.gid     =       -1;
+       sk->sk_write_pending    =       0;
        sk->sk_rcvlowat         =       1;
        sk->sk_rcvtimeo         =       MAX_SCHEDULE_TIMEOUT;
        sk->sk_sndtimeo         =       MAX_SCHEDULE_TIMEOUT;
-       sk->sk_owner            =       NULL;
 
        sk->sk_stamp.tv_sec     = -1L;
        sk->sk_stamp.tv_usec    = -1L;
 
-       sk->sk_vx_info          =       NULL;
-       sk->sk_xid              =       0;
-       sk->sk_nx_info          =       NULL;
-       sk->sk_nid              =       0;
-
+       set_vx_info(&sk->sk_vx_info, current->vx_info);
+       sk->sk_xid = vx_current_xid();
+       vx_sock_inc(sk);
+       set_nx_info(&sk->sk_nx_info, current->nx_info);
+       sk->sk_nid = nx_current_nid();
        atomic_set(&sk->sk_refcnt, 1);
 }
 
-void fastcall lock_sock(struct sock *sk)
+void fastcall lock_sock_nested(struct sock *sk, int subclass)
 {
        might_sleep();
-       spin_lock_bh(&(sk->sk_lock.slock));
+       spin_lock_bh(&sk->sk_lock.slock);
        if (sk->sk_lock.owner)
                __lock_sock(sk);
        sk->sk_lock.owner = (void *)1;
-       spin_unlock_bh(&(sk->sk_lock.slock));
+       spin_unlock(&sk->sk_lock.slock);
+       /*
+        * The sk_lock has mutex_lock() semantics here:
+        */
+       mutex_acquire(&sk->sk_lock.dep_map, subclass, 0, _RET_IP_);
+       local_bh_enable();
 }
 
-EXPORT_SYMBOL(lock_sock);
+EXPORT_SYMBOL(lock_sock_nested);
 
 void fastcall release_sock(struct sock *sk)
 {
-       spin_lock_bh(&(sk->sk_lock.slock));
+       /*
+        * The sk_lock has mutex_unlock() semantics:
+        */
+       mutex_release(&sk->sk_lock.dep_map, 1, _RET_IP_);
+
+       spin_lock_bh(&sk->sk_lock.slock);
        if (sk->sk_backlog.tail)
                __release_sock(sk);
        sk->sk_lock.owner = NULL;
-        if (waitqueue_active(&(sk->sk_lock.wq)))
-               wake_up(&(sk->sk_lock.wq));
-       spin_unlock_bh(&(sk->sk_lock.slock));
+       if (waitqueue_active(&sk->sk_lock.wq))
+               wake_up(&sk->sk_lock.wq);
+       spin_unlock_bh(&sk->sk_lock.slock);
 }
 EXPORT_SYMBOL(release_sock);
 
-/* When > 0 there are consumers of rx skb time stamps */
-atomic_t netstamp_needed = ATOMIC_INIT(0); 
-
 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
 { 
        if (!sock_flag(sk, SOCK_TIMESTAMP))
@@ -1244,28 +1592,370 @@ void sock_enable_timestamp(struct sock *sk)
 {      
        if (!sock_flag(sk, SOCK_TIMESTAMP)) { 
                sock_set_flag(sk, SOCK_TIMESTAMP);
-               atomic_inc(&netstamp_needed);
+               net_enable_timestamp();
        }
 }
 EXPORT_SYMBOL(sock_enable_timestamp); 
 
-void sock_disable_timestamp(struct sock *sk)
-{      
-       if (sock_flag(sk, SOCK_TIMESTAMP)) { 
-               sock_reset_flag(sk, SOCK_TIMESTAMP);
-               atomic_dec(&netstamp_needed);
+/*
+ *     Get a socket option on an socket.
+ *
+ *     FIX: POSIX 1003.1g is very ambiguous here. It states that
+ *     asynchronous errors should be reported by getsockopt. We assume
+ *     this means if you specify SO_ERROR (otherwise whats the point of it).
+ */
+int sock_common_getsockopt(struct socket *sock, int level, int optname,
+                          char __user *optval, int __user *optlen)
+{
+       struct sock *sk = sock->sk;
+
+       return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
+}
+
+EXPORT_SYMBOL(sock_common_getsockopt);
+
+#ifdef CONFIG_COMPAT
+int compat_sock_common_getsockopt(struct socket *sock, int level, int optname,
+                                 char __user *optval, int __user *optlen)
+{
+       struct sock *sk = sock->sk;
+
+       if (sk->sk_prot->compat_getsockopt != NULL)
+               return sk->sk_prot->compat_getsockopt(sk, level, optname,
+                                                     optval, optlen);
+       return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
+}
+EXPORT_SYMBOL(compat_sock_common_getsockopt);
+#endif
+
+int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
+                       struct msghdr *msg, size_t size, int flags)
+{
+       struct sock *sk = sock->sk;
+       int addr_len = 0;
+       int err;
+
+       err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
+                                  flags & ~MSG_DONTWAIT, &addr_len);
+       if (err >= 0)
+               msg->msg_namelen = addr_len;
+       return err;
+}
+
+EXPORT_SYMBOL(sock_common_recvmsg);
+
+/*
+ *     Set socket options on an inet socket.
+ */
+int sock_common_setsockopt(struct socket *sock, int level, int optname,
+                          char __user *optval, int optlen)
+{
+       struct sock *sk = sock->sk;
+
+       return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
+}
+
+EXPORT_SYMBOL(sock_common_setsockopt);
+
+#ifdef CONFIG_COMPAT
+int compat_sock_common_setsockopt(struct socket *sock, int level, int optname,
+                                 char __user *optval, int optlen)
+{
+       struct sock *sk = sock->sk;
+
+       if (sk->sk_prot->compat_setsockopt != NULL)
+               return sk->sk_prot->compat_setsockopt(sk, level, optname,
+                                                     optval, optlen);
+       return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
+}
+EXPORT_SYMBOL(compat_sock_common_setsockopt);
+#endif
+
+void sk_common_release(struct sock *sk)
+{
+       if (sk->sk_prot->destroy)
+               sk->sk_prot->destroy(sk);
+
+       /*
+        * Observation: when sock_common_release is called, processes have
+        * no access to socket. But net still has.
+        * Step one, detach it from networking:
+        *
+        * A. Remove from hash tables.
+        */
+
+       sk->sk_prot->unhash(sk);
+
+       /*
+        * In this point socket cannot receive new packets, but it is possible
+        * that some packets are in flight because some CPU runs receiver and
+        * did hash table lookup before we unhashed socket. They will achieve
+        * receive queue and will be purged by socket destructor.
+        *
+        * Also we still have packets pending on receive queue and probably,
+        * our own packets waiting in device queues. sock_destroy will drain
+        * receive queue, but transmitted packets will delay socket destruction
+        * until the last reference will be released.
+        */
+
+       sock_orphan(sk);
+
+       xfrm_sk_free_policy(sk);
+
+       sk_refcnt_debug_release(sk);
+       sock_put(sk);
+}
+
+EXPORT_SYMBOL(sk_common_release);
+
+static DEFINE_RWLOCK(proto_list_lock);
+static LIST_HEAD(proto_list);
+
+int proto_register(struct proto *prot, int alloc_slab)
+{
+       char *request_sock_slab_name = NULL;
+       char *timewait_sock_slab_name;
+       int rc = -ENOBUFS;
+
+       if (alloc_slab) {
+               prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
+                                              SLAB_HWCACHE_ALIGN, NULL, NULL);
+
+               if (prot->slab == NULL) {
+                       printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
+                              prot->name);
+                       goto out;
+               }
+
+               if (prot->rsk_prot != NULL) {
+                       static const char mask[] = "request_sock_%s";
+
+                       request_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
+                       if (request_sock_slab_name == NULL)
+                               goto out_free_sock_slab;
+
+                       sprintf(request_sock_slab_name, mask, prot->name);
+                       prot->rsk_prot->slab = kmem_cache_create(request_sock_slab_name,
+                                                                prot->rsk_prot->obj_size, 0,
+                                                                SLAB_HWCACHE_ALIGN, NULL, NULL);
+
+                       if (prot->rsk_prot->slab == NULL) {
+                               printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
+                                      prot->name);
+                               goto out_free_request_sock_slab_name;
+                       }
+               }
+
+               if (prot->twsk_prot != NULL) {
+                       static const char mask[] = "tw_sock_%s";
+
+                       timewait_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
+
+                       if (timewait_sock_slab_name == NULL)
+                               goto out_free_request_sock_slab;
+
+                       sprintf(timewait_sock_slab_name, mask, prot->name);
+                       prot->twsk_prot->twsk_slab =
+                               kmem_cache_create(timewait_sock_slab_name,
+                                                 prot->twsk_prot->twsk_obj_size,
+                                                 0, SLAB_HWCACHE_ALIGN,
+                                                 NULL, NULL);
+                       if (prot->twsk_prot->twsk_slab == NULL)
+                               goto out_free_timewait_sock_slab_name;
+               }
+       }
+
+       write_lock(&proto_list_lock);
+       list_add(&prot->node, &proto_list);
+       write_unlock(&proto_list_lock);
+       rc = 0;
+out:
+       return rc;
+out_free_timewait_sock_slab_name:
+       kfree(timewait_sock_slab_name);
+out_free_request_sock_slab:
+       if (prot->rsk_prot && prot->rsk_prot->slab) {
+               kmem_cache_destroy(prot->rsk_prot->slab);
+               prot->rsk_prot->slab = NULL;
+       }
+out_free_request_sock_slab_name:
+       kfree(request_sock_slab_name);
+out_free_sock_slab:
+       kmem_cache_destroy(prot->slab);
+       prot->slab = NULL;
+       goto out;
+}
+
+EXPORT_SYMBOL(proto_register);
+
+void proto_unregister(struct proto *prot)
+{
+       write_lock(&proto_list_lock);
+       list_del(&prot->node);
+       write_unlock(&proto_list_lock);
+
+       if (prot->slab != NULL) {
+               kmem_cache_destroy(prot->slab);
+               prot->slab = NULL;
+       }
+
+       if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
+               const char *name = kmem_cache_name(prot->rsk_prot->slab);
+
+               kmem_cache_destroy(prot->rsk_prot->slab);
+               kfree(name);
+               prot->rsk_prot->slab = NULL;
        }
+
+       if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
+               const char *name = kmem_cache_name(prot->twsk_prot->twsk_slab);
+
+               kmem_cache_destroy(prot->twsk_prot->twsk_slab);
+               kfree(name);
+               prot->twsk_prot->twsk_slab = NULL;
+       }
+}
+
+EXPORT_SYMBOL(proto_unregister);
+
+#ifdef CONFIG_PROC_FS
+static inline struct proto *__proto_head(void)
+{
+       return list_entry(proto_list.next, struct proto, node);
+}
+
+static inline struct proto *proto_head(void)
+{
+       return list_empty(&proto_list) ? NULL : __proto_head();
+}
+
+static inline struct proto *proto_next(struct proto *proto)
+{
+       return proto->node.next == &proto_list ? NULL :
+               list_entry(proto->node.next, struct proto, node);
+}
+
+static inline struct proto *proto_get_idx(loff_t pos)
+{
+       struct proto *proto;
+       loff_t i = 0;
+
+       list_for_each_entry(proto, &proto_list, node)
+               if (i++ == pos)
+                       goto out;
+
+       proto = NULL;
+out:
+       return proto;
+}
+
+static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
+{
+       read_lock(&proto_list_lock);
+       return *pos ? proto_get_idx(*pos - 1) : SEQ_START_TOKEN;
+}
+
+static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
+{
+       ++*pos;
+       return v == SEQ_START_TOKEN ? proto_head() : proto_next(v);
+}
+
+static void proto_seq_stop(struct seq_file *seq, void *v)
+{
+       read_unlock(&proto_list_lock);
+}
+
+static char proto_method_implemented(const void *method)
+{
+       return method == NULL ? 'n' : 'y';
+}
+
+static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
+{
+       seq_printf(seq, "%-9s %4u %6d  %6d   %-3s %6u   %-3s  %-10s "
+                       "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
+                  proto->name,
+                  proto->obj_size,
+                  proto->sockets_allocated != NULL ? atomic_read(proto->sockets_allocated) : -1,
+                  proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
+                  proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
+                  proto->max_header,
+                  proto->slab == NULL ? "no" : "yes",
+                  module_name(proto->owner),
+                  proto_method_implemented(proto->close),
+                  proto_method_implemented(proto->connect),
+                  proto_method_implemented(proto->disconnect),
+                  proto_method_implemented(proto->accept),
+                  proto_method_implemented(proto->ioctl),
+                  proto_method_implemented(proto->init),
+                  proto_method_implemented(proto->destroy),
+                  proto_method_implemented(proto->shutdown),
+                  proto_method_implemented(proto->setsockopt),
+                  proto_method_implemented(proto->getsockopt),
+                  proto_method_implemented(proto->sendmsg),
+                  proto_method_implemented(proto->recvmsg),
+                  proto_method_implemented(proto->sendpage),
+                  proto_method_implemented(proto->bind),
+                  proto_method_implemented(proto->backlog_rcv),
+                  proto_method_implemented(proto->hash),
+                  proto_method_implemented(proto->unhash),
+                  proto_method_implemented(proto->get_port),
+                  proto_method_implemented(proto->enter_memory_pressure));
 }
-EXPORT_SYMBOL(sock_disable_timestamp);
 
-EXPORT_SYMBOL(__lock_sock);
-EXPORT_SYMBOL(__release_sock);
+static int proto_seq_show(struct seq_file *seq, void *v)
+{
+       if (v == SEQ_START_TOKEN)
+               seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
+                          "protocol",
+                          "size",
+                          "sockets",
+                          "memory",
+                          "press",
+                          "maxhdr",
+                          "slab",
+                          "module",
+                          "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
+       else
+               proto_seq_printf(seq, v);
+       return 0;
+}
+
+static struct seq_operations proto_seq_ops = {
+       .start  = proto_seq_start,
+       .next   = proto_seq_next,
+       .stop   = proto_seq_stop,
+       .show   = proto_seq_show,
+};
+
+static int proto_seq_open(struct inode *inode, struct file *file)
+{
+       return seq_open(file, &proto_seq_ops);
+}
+
+static struct file_operations proto_seq_fops = {
+       .owner          = THIS_MODULE,
+       .open           = proto_seq_open,
+       .read           = seq_read,
+       .llseek         = seq_lseek,
+       .release        = seq_release,
+};
+
+static int __init proto_init(void)
+{
+       /* register /proc/net/protocols */
+       return proc_net_fops_create("protocols", S_IRUGO, &proto_seq_fops) == NULL ? -ENOBUFS : 0;
+}
+
+subsys_initcall(proto_init);
+
+#endif /* PROC_FS */
+
 EXPORT_SYMBOL(sk_alloc);
 EXPORT_SYMBOL(sk_free);
 EXPORT_SYMBOL(sk_send_sigurg);
-EXPORT_SYMBOL(sock_alloc_send_pskb);
 EXPORT_SYMBOL(sock_alloc_send_skb);
-EXPORT_SYMBOL(sock_getsockopt);
 EXPORT_SYMBOL(sock_init_data);
 EXPORT_SYMBOL(sock_kfree_s);
 EXPORT_SYMBOL(sock_kmalloc);
@@ -1279,19 +1969,19 @@ EXPORT_SYMBOL(sock_no_listen);
 EXPORT_SYMBOL(sock_no_mmap);
 EXPORT_SYMBOL(sock_no_poll);
 EXPORT_SYMBOL(sock_no_recvmsg);
-EXPORT_SYMBOL(sock_no_release);
 EXPORT_SYMBOL(sock_no_sendmsg);
 EXPORT_SYMBOL(sock_no_sendpage);
 EXPORT_SYMBOL(sock_no_setsockopt);
 EXPORT_SYMBOL(sock_no_shutdown);
 EXPORT_SYMBOL(sock_no_socketpair);
 EXPORT_SYMBOL(sock_rfree);
-EXPORT_SYMBOL(sock_rmalloc);
 EXPORT_SYMBOL(sock_setsockopt);
 EXPORT_SYMBOL(sock_wfree);
 EXPORT_SYMBOL(sock_wmalloc);
-#ifdef CONFIG_SYSCTL
+EXPORT_SYMBOL(sock_i_uid);
+EXPORT_SYMBOL(sock_i_ino);
 EXPORT_SYMBOL(sysctl_optmem_max);
+#ifdef CONFIG_SYSCTL
 EXPORT_SYMBOL(sysctl_rmem_max);
 EXPORT_SYMBOL(sysctl_wmem_max);
 #endif