[RAW]: Introduce raw_hashinfo structure
authorPavel Emelyanov <xemul@openvz.org>
Tue, 20 Nov 2007 06:36:45 +0000 (22:36 -0800)
committerDavid S. Miller <davem@davemloft.net>
Mon, 28 Jan 2008 22:54:30 +0000 (14:54 -0800)
The ipv4/raw.c and ipv6/raw.c contain many common code (most
of which is proc interface) which can be consolidated.

Most of the places to consolidate deal with the raw sockets
hashtable, so introduce a struct raw_hashinfo which describes
the raw sockets hash.

Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
include/net/raw.h
net/ipv4/raw.c
net/ipv6/raw.c

index 7fc3c770f1704e964cb04f629684d76f896ae102..70b27c7c816760c78b9a00f9ddff8650b40282f4 100644 (file)
@@ -27,6 +27,13 @@ int raw_local_deliver(struct sk_buff *, int);
 
 extern int     raw_rcv(struct sock *, struct sk_buff *);
 
+#define RAW_HTABLE_SIZE        MAX_INET_PROTOS
+
+struct raw_hashinfo {
+       rwlock_t lock;
+       struct hlist_head ht[RAW_HTABLE_SIZE];
+};
+
 #ifdef CONFIG_PROC_FS
 extern int  raw_proc_init(void);
 extern void raw_proc_exit(void);
index 8a506618b91258bfad7a3bcec4c0d103857eaf6f..dd9f00b3ab42618553e385afa15a55bfcbe567cf 100644 (file)
 #include <linux/netfilter.h>
 #include <linux/netfilter_ipv4.h>
 
-#define RAWV4_HTABLE_SIZE      MAX_INET_PROTOS
-
-static struct hlist_head raw_v4_htable[RAWV4_HTABLE_SIZE];
-static DEFINE_RWLOCK(raw_v4_lock);
+static struct raw_hashinfo raw_v4_hashinfo = {
+       .lock = __RW_LOCK_UNLOCKED(),
+};
 
 static void raw_v4_hash(struct sock *sk)
 {
-       struct hlist_head *head = &raw_v4_htable[inet_sk(sk)->num &
-                                                (RAWV4_HTABLE_SIZE - 1)];
+       struct hlist_head *head = &raw_v4_hashinfo.ht[inet_sk(sk)->num &
+                                                (RAW_HTABLE_SIZE - 1)];
 
-       write_lock_bh(&raw_v4_lock);
+       write_lock_bh(&raw_v4_hashinfo.lock);
        sk_add_node(sk, head);
        sock_prot_inc_use(sk->sk_prot);
-       write_unlock_bh(&raw_v4_lock);
+       write_unlock_bh(&raw_v4_hashinfo.lock);
 }
 
 static void raw_v4_unhash(struct sock *sk)
 {
-       write_lock_bh(&raw_v4_lock);
+       write_lock_bh(&raw_v4_hashinfo.lock);
        if (sk_del_node_init(sk))
                sock_prot_dec_use(sk->sk_prot);
-       write_unlock_bh(&raw_v4_lock);
+       write_unlock_bh(&raw_v4_hashinfo.lock);
 }
 
 static struct sock *__raw_v4_lookup(struct sock *sk, unsigned short num,
@@ -158,8 +157,8 @@ static int raw_v4_input(struct sk_buff *skb, struct iphdr *iph, int hash)
        struct hlist_head *head;
        int delivered = 0;
 
-       read_lock(&raw_v4_lock);
-       head = &raw_v4_htable[hash];
+       read_lock(&raw_v4_hashinfo.lock);
+       head = &raw_v4_hashinfo.ht[hash];
        if (hlist_empty(head))
                goto out;
        sk = __raw_v4_lookup(__sk_head(head), iph->protocol,
@@ -180,7 +179,7 @@ static int raw_v4_input(struct sk_buff *skb, struct iphdr *iph, int hash)
                                     skb->dev->ifindex);
        }
 out:
-       read_unlock(&raw_v4_lock);
+       read_unlock(&raw_v4_hashinfo.lock);
        return delivered;
 }
 
@@ -189,8 +188,8 @@ int raw_local_deliver(struct sk_buff *skb, int protocol)
        int hash;
        struct sock *raw_sk;
 
-       hash = protocol & (RAWV4_HTABLE_SIZE - 1);
-       raw_sk = sk_head(&raw_v4_htable[hash]);
+       hash = protocol & (RAW_HTABLE_SIZE - 1);
+       raw_sk = sk_head(&raw_v4_hashinfo.ht[hash]);
 
        /* If there maybe a raw socket we must check - if not we
         * don't care less
@@ -262,10 +261,10 @@ void raw_icmp_error(struct sk_buff *skb, int protocol, u32 info)
        struct sock *raw_sk;
        struct iphdr *iph;
 
-       hash = protocol & (RAWV4_HTABLE_SIZE - 1);
+       hash = protocol & (RAW_HTABLE_SIZE - 1);
 
-       read_lock(&raw_v4_lock);
-       raw_sk = sk_head(&raw_v4_htable[hash]);
+       read_lock(&raw_v4_hashinfo.lock);
+       raw_sk = sk_head(&raw_v4_hashinfo.ht[hash]);
        if (raw_sk != NULL) {
                iph = (struct iphdr *)skb->data;
                while ((raw_sk = __raw_v4_lookup(raw_sk, protocol, iph->daddr,
@@ -276,7 +275,7 @@ void raw_icmp_error(struct sk_buff *skb, int protocol, u32 info)
                        iph = (struct iphdr *)skb->data;
                }
        }
-       read_unlock(&raw_v4_lock);
+       read_unlock(&raw_v4_hashinfo.lock);
 }
 
 static int raw_rcv_skb(struct sock * sk, struct sk_buff * skb)
@@ -844,10 +843,11 @@ static struct sock *raw_get_first(struct seq_file *seq)
        struct sock *sk;
        struct raw_iter_state* state = raw_seq_private(seq);
 
-       for (state->bucket = 0; state->bucket < RAWV4_HTABLE_SIZE; ++state->bucket) {
+       for (state->bucket = 0; state->bucket < RAW_HTABLE_SIZE;
+                       ++state->bucket) {
                struct hlist_node *node;
 
-               sk_for_each(sk, node, &raw_v4_htable[state->bucket])
+               sk_for_each(sk, node, &raw_v4_hashinfo.ht[state->bucket])
                        if (sk->sk_family == PF_INET)
                                goto found;
        }
@@ -866,8 +866,8 @@ try_again:
                ;
        } while (sk && sk->sk_family != PF_INET);
 
-       if (!sk && ++state->bucket < RAWV4_HTABLE_SIZE) {
-               sk = sk_head(&raw_v4_htable[state->bucket]);
+       if (!sk && ++state->bucket < RAW_HTABLE_SIZE) {
+               sk = sk_head(&raw_v4_hashinfo.ht[state->bucket]);
                goto try_again;
        }
        return sk;
@@ -885,7 +885,7 @@ static struct sock *raw_get_idx(struct seq_file *seq, loff_t pos)
 
 static void *raw_seq_start(struct seq_file *seq, loff_t *pos)
 {
-       read_lock(&raw_v4_lock);
+       read_lock(&raw_v4_hashinfo.lock);
        return *pos ? raw_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
 }
 
@@ -903,7 +903,7 @@ static void *raw_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 
 static void raw_seq_stop(struct seq_file *seq, void *v)
 {
-       read_unlock(&raw_v4_lock);
+       read_unlock(&raw_v4_hashinfo.lock);
 }
 
 static __inline__ char *get_raw_sock(struct sock *sp, char *tmpbuf, int i)
index 53f01b4982c712395cf1924bb8220d0f4a37f6bd..15c72a6365a667e49420e66aadf0700817948abf 100644 (file)
 #include <net/mip6.h>
 #endif
 
+#include <net/raw.h>
 #include <net/rawv6.h>
 #include <net/xfrm.h>
 
 #include <linux/proc_fs.h>
 #include <linux/seq_file.h>
 
-#define RAWV6_HTABLE_SIZE      MAX_INET_PROTOS
-
-static struct hlist_head raw_v6_htable[RAWV6_HTABLE_SIZE];
-static DEFINE_RWLOCK(raw_v6_lock);
+static struct raw_hashinfo raw_v6_hashinfo = {
+       .lock = __RW_LOCK_UNLOCKED(),
+};
 
 static void raw_v6_hash(struct sock *sk)
 {
-       struct hlist_head *list = &raw_v6_htable[inet_sk(sk)->num &
-                                                (RAWV6_HTABLE_SIZE - 1)];
+       struct hlist_head *list = &raw_v6_hashinfo.ht[inet_sk(sk)->num &
+                                                (RAW_HTABLE_SIZE - 1)];
 
-       write_lock_bh(&raw_v6_lock);
+       write_lock_bh(&raw_v6_hashinfo.lock);
        sk_add_node(sk, list);
        sock_prot_inc_use(sk->sk_prot);
-       write_unlock_bh(&raw_v6_lock);
+       write_unlock_bh(&raw_v6_hashinfo.lock);
 }
 
 static void raw_v6_unhash(struct sock *sk)
 {
-       write_lock_bh(&raw_v6_lock);
+       write_lock_bh(&raw_v6_hashinfo.lock);
        if (sk_del_node_init(sk))
                sock_prot_dec_use(sk->sk_prot);
-       write_unlock_bh(&raw_v6_lock);
+       write_unlock_bh(&raw_v6_hashinfo.lock);
 }
 
 
@@ -180,8 +180,8 @@ static int ipv6_raw_deliver(struct sk_buff *skb, int nexthdr)
 
        hash = nexthdr & (MAX_INET_PROTOS - 1);
 
-       read_lock(&raw_v6_lock);
-       sk = sk_head(&raw_v6_htable[hash]);
+       read_lock(&raw_v6_hashinfo.lock);
+       sk = sk_head(&raw_v6_hashinfo.ht[hash]);
 
        /*
         *      The first socket found will be delivered after
@@ -238,7 +238,7 @@ static int ipv6_raw_deliver(struct sk_buff *skb, int nexthdr)
                                     IP6CB(skb)->iif);
        }
 out:
-       read_unlock(&raw_v6_lock);
+       read_unlock(&raw_v6_hashinfo.lock);
        return delivered;
 }
 
@@ -246,7 +246,7 @@ int raw6_local_deliver(struct sk_buff *skb, int nexthdr)
 {
        struct sock *raw_sk;
 
-       raw_sk = sk_head(&raw_v6_htable[nexthdr & (MAX_INET_PROTOS - 1)]);
+       raw_sk = sk_head(&raw_v6_hashinfo.ht[nexthdr & (MAX_INET_PROTOS - 1)]);
        if (raw_sk && !ipv6_raw_deliver(skb, nexthdr))
                raw_sk = NULL;
 
@@ -368,10 +368,10 @@ void raw6_icmp_error(struct sk_buff *skb, int nexthdr,
        int hash;
        struct in6_addr *saddr, *daddr;
 
-       hash = nexthdr & (RAWV6_HTABLE_SIZE - 1);
+       hash = nexthdr & (RAW_HTABLE_SIZE - 1);
 
-       read_lock(&raw_v6_lock);
-       sk = sk_head(&raw_v6_htable[hash]);
+       read_lock(&raw_v6_hashinfo.lock);
+       sk = sk_head(&raw_v6_hashinfo.ht[hash]);
        if (sk != NULL) {
                saddr = &ipv6_hdr(skb)->saddr;
                daddr = &ipv6_hdr(skb)->daddr;
@@ -383,7 +383,7 @@ void raw6_icmp_error(struct sk_buff *skb, int nexthdr,
                        sk = sk_next(sk);
                }
        }
-       read_unlock(&raw_v6_lock);
+       read_unlock(&raw_v6_hashinfo.lock);
 }
 
 static inline int rawv6_rcv_skb(struct sock * sk, struct sk_buff * skb)
@@ -1221,8 +1221,9 @@ static struct sock *raw6_get_first(struct seq_file *seq)
        struct hlist_node *node;
        struct raw6_iter_state* state = raw6_seq_private(seq);
 
-       for (state->bucket = 0; state->bucket < RAWV6_HTABLE_SIZE; ++state->bucket)
-               sk_for_each(sk, node, &raw_v6_htable[state->bucket])
+       for (state->bucket = 0; state->bucket < RAW_HTABLE_SIZE;
+                       ++state->bucket)
+               sk_for_each(sk, node, &raw_v6_hashinfo.ht[state->bucket])
                        if (sk->sk_family == PF_INET6)
                                goto out;
        sk = NULL;
@@ -1240,8 +1241,8 @@ try_again:
                ;
        } while (sk && sk->sk_family != PF_INET6);
 
-       if (!sk && ++state->bucket < RAWV6_HTABLE_SIZE) {
-               sk = sk_head(&raw_v6_htable[state->bucket]);
+       if (!sk && ++state->bucket < RAW_HTABLE_SIZE) {
+               sk = sk_head(&raw_v6_hashinfo.ht[state->bucket]);
                goto try_again;
        }
        return sk;
@@ -1258,7 +1259,7 @@ static struct sock *raw6_get_idx(struct seq_file *seq, loff_t pos)
 
 static void *raw6_seq_start(struct seq_file *seq, loff_t *pos)
 {
-       read_lock(&raw_v6_lock);
+       read_lock(&raw_v6_hashinfo.lock);
        return *pos ? raw6_get_idx(seq, *pos - 1) : SEQ_START_TOKEN;
 }
 
@@ -1276,7 +1277,7 @@ static void *raw6_seq_next(struct seq_file *seq, void *v, loff_t *pos)
 
 static void raw6_seq_stop(struct seq_file *seq, void *v)
 {
-       read_unlock(&raw_v6_lock);
+       read_unlock(&raw_v6_hashinfo.lock);
 }
 
 static void raw6_sock_seq_show(struct seq_file *seq, struct sock *sp, int i)