Merge master.kernel.org:/pub/scm/linux/kernel/git/sam/kbuild
[pandora-kernel.git] / net / core / sock.c
index a6ec3ad..c135945 100644 (file)
@@ -206,13 +206,14 @@ 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);
        
@@ -259,7 +260,7 @@ 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;
@@ -273,6 +274,13 @@ 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
@@ -281,7 +289,7 @@ 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? */
                        if ((val * 2) < SOCK_MIN_RCVBUF)
@@ -290,6 +298,13 @@ int sock_setsockopt(struct socket *sock, int level, int optname,
                                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)
@@ -622,7 +637,8 @@ lenout:
  *     @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, struct proto *prot, int zero_it)
+struct sock *sk_alloc(int family, unsigned int __nocast priority,
+                     struct proto *prot, int zero_it)
 {
        struct sock *sk = NULL;
        kmem_cache_t *slab = prot->slab;
@@ -684,6 +700,80 @@ void sk_free(struct sock *sk)
        module_put(owner);
 }
 
+struct sock *sk_clone(const struct sock *sk, const unsigned int __nocast priority)
+{
+       struct sock *newsk = sk_alloc(sk->sk_family, priority, sk->sk_prot, 0);
+
+       if (newsk != NULL) {
+               struct sk_filter *filter;
+
+               memcpy(newsk, sk, sk->sk_prot->obj_size);
+
+               /* SANITY */
+               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);
+
+               rwlock_init(&newsk->sk_dst_lock);
+               rwlock_init(&newsk->sk_callback_lock);
+
+               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);
+
+               /*
+                * 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) {
@@ -750,7 +840,8 @@ unsigned long sock_i_ino(struct sock *sk)
 /*
  * 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,
+                            unsigned int __nocast priority)
 {
        if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
                struct sk_buff * skb = alloc_skb(size, priority);
@@ -765,7 +856,8 @@ 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,
+                            unsigned int __nocast priority)
 {
        if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
                struct sk_buff *skb = alloc_skb(size, priority);
@@ -780,7 +872,7 @@ 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, unsigned int __nocast priority)
 {
        if ((unsigned)size <= sysctl_optmem_max &&
            atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
@@ -1349,11 +1441,7 @@ void sk_common_release(struct sock *sk)
 
        xfrm_sk_free_policy(sk);
 
-#ifdef INET_REFCNT_DEBUG
-       if (atomic_read(&sk->sk_refcnt) != 1)
-               printk(KERN_DEBUG "Destruction of the socket %p delayed, c=%d\n",
-                      sk, atomic_read(&sk->sk_refcnt));
-#endif
+       sk_refcnt_debug_release(sk);
        sock_put(sk);
 }
 
@@ -1364,7 +1452,8 @@ static LIST_HEAD(proto_list);
 
 int proto_register(struct proto *prot, int alloc_slab)
 {
-       char *request_sock_slab_name;
+       char *request_sock_slab_name = NULL;
+       char *timewait_sock_slab_name;
        int rc = -ENOBUFS;
 
        if (alloc_slab) {
@@ -1395,6 +1484,23 @@ int proto_register(struct proto *prot, int alloc_slab)
                                goto out_free_request_sock_slab_name;
                        }
                }
+
+               if (prot->twsk_obj_size) {
+                       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_slab = kmem_cache_create(timewait_sock_slab_name,
+                                                           prot->twsk_obj_size,
+                                                           0, SLAB_HWCACHE_ALIGN,
+                                                           NULL, NULL);
+                       if (prot->twsk_slab == NULL)
+                               goto out_free_timewait_sock_slab_name;
+               }
        }
 
        write_lock(&proto_list_lock);
@@ -1403,6 +1509,13 @@ int proto_register(struct proto *prot, int alloc_slab)
        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:
@@ -1430,6 +1543,14 @@ void proto_unregister(struct proto *prot)
                prot->rsk_prot->slab = NULL;
        }
 
+       if (prot->twsk_slab != NULL) {
+               const char *name = kmem_cache_name(prot->twsk_slab);
+
+               kmem_cache_destroy(prot->twsk_slab);
+               kfree(name);
+               prot->twsk_slab = NULL;
+       }
+
        list_del(&prot->node);
        write_unlock(&proto_list_lock);
 }
@@ -1598,8 +1719,8 @@ EXPORT_SYMBOL(sock_wfree);
 EXPORT_SYMBOL(sock_wmalloc);
 EXPORT_SYMBOL(sock_i_uid);
 EXPORT_SYMBOL(sock_i_ino);
-#ifdef CONFIG_SYSCTL
 EXPORT_SYMBOL(sysctl_optmem_max);
+#ifdef CONFIG_SYSCTL
 EXPORT_SYMBOL(sysctl_rmem_max);
 EXPORT_SYMBOL(sysctl_wmem_max);
 #endif