netfilter: ipset: fix linking with CONFIG_IPV6=n

netfilter: ipset: fix linking with CONFIG_IPV6=n

[Patrick McHardy]

[-- Attachment #1: Type: text/plain, Size: 336 bytes --]

Add some #ifdefs to unconditionally return false in
ip_set_get_ip6_port() when CONFIG_IPV6=n and convert
to ipv6_skip_exthdr() to avoid pulling in the ip6_tables
module when loading ipset.

A slightly more code-saving alternative would be to add
a dummy inline function or ifdef the callsites. Please
let me know which way you prefer.


[-- Attachment #2: 05.diff --]
[-- Type: text/x-patch, Size: 1126 bytes --]

diff --git a/net/netfilter/ipset/ip_set_getport.c b/net/netfilter/ipset/ip_set_getport.c
index 4dd2785..5ea0736 100644
--- a/net/netfilter/ipset/ip_set_getport.c
+++ b/net/netfilter/ipset/ip_set_getport.c
@@ -13,6 +13,7 @@
 #include <linux/icmpv6.h>
 #include <linux/netfilter_ipv6/ip6_tables.h>
 #include <net/ip.h>
+#include <net/ipv6.h>
 
 #include <linux/netfilter/ipset/ip_set_getport.h>
 
@@ -97,15 +98,19 @@ bool
 ip_set_get_ip6_port(const struct sk_buff *skb, bool src,
 		    __be16 *port, u8 *proto)
 {
-	unsigned int protooff = 0;
-	int protocol;
-	unsigned short fragoff;
+#if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
+	int protoff;
+	u8 nexthdr;
 
-	protocol = ipv6_find_hdr(skb, &protooff, -1, &fragoff);
-	if (protocol <= 0 || fragoff)
+	nexthdr = ipv6_hdr(skb)->nexthdr;
+	protoff = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr);
+	if (protoff < 0)
 		return false;
 
-	return get_port(skb, protocol, protooff, src, port, proto);
+	return get_port(skb, nexthdr, protoff, src, port, proto);
+#else
+	return false;
+#endif
 }
 EXPORT_SYMBOL_GPL(ip_set_get_ip6_port);
 

Re: netfilter: ipset: fix linking with CONFIG_IPV6=n

[Patrick McHardy]

[-- Attachment #1: Type: text/plain, Size: 456 bytes --]

[resend with netfilter-devel CCed instead of netfilter]

On 02.02.2011 09:40, Patrick McHardy wrote:
> Add some #ifdefs to unconditionally return false in
> ip_set_get_ip6_port() when CONFIG_IPV6=n and convert
> to ipv6_skip_exthdr() to avoid pulling in the ip6_tables
> module when loading ipset.
> 
> A slightly more code-saving alternative would be to add
> a dummy inline function or ifdef the callsites. Please
> let me know which way you prefer.
> 


[-- Attachment #2: 05.diff --]
[-- Type: text/x-patch, Size: 1126 bytes --]

diff --git a/net/netfilter/ipset/ip_set_getport.c b/net/netfilter/ipset/ip_set_getport.c
index 4dd2785..5ea0736 100644
--- a/net/netfilter/ipset/ip_set_getport.c
+++ b/net/netfilter/ipset/ip_set_getport.c
@@ -13,6 +13,7 @@
 #include <linux/icmpv6.h>
 #include <linux/netfilter_ipv6/ip6_tables.h>
 #include <net/ip.h>
+#include <net/ipv6.h>
 
 #include <linux/netfilter/ipset/ip_set_getport.h>
 
@@ -97,15 +98,19 @@ bool
 ip_set_get_ip6_port(const struct sk_buff *skb, bool src,
 		    __be16 *port, u8 *proto)
 {
-	unsigned int protooff = 0;
-	int protocol;
-	unsigned short fragoff;
+#if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
+	int protoff;
+	u8 nexthdr;
 
-	protocol = ipv6_find_hdr(skb, &protooff, -1, &fragoff);
-	if (protocol <= 0 || fragoff)
+	nexthdr = ipv6_hdr(skb)->nexthdr;
+	protoff = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr);
+	if (protoff < 0)
 		return false;
 
-	return get_port(skb, protocol, protooff, src, port, proto);
+	return get_port(skb, nexthdr, protoff, src, port, proto);
+#else
+	return false;
+#endif
 }
 EXPORT_SYMBOL_GPL(ip_set_get_ip6_port);
 

Re: netfilter: ipset: fix linking with CONFIG_IPV6=n

[Jozsef Kadlecsik]

Hi Patrick,

On Wed, 2 Feb 2011, Patrick McHardy wrote:

> Add some #ifdefs to unconditionally return false in
> ip_set_get_ip6_port() when CONFIG_IPV6=n and convert
> to ipv6_skip_exthdr() to avoid pulling in the ip6_tables
> module when loading ipset.

You left out checking fragmented packets. Is it intentional?
 
> A slightly more code-saving alternative would be to add
> a dummy inline function or ifdef the callsites. Please
> let me know which way you prefer.

A dummy inline looks nicer to me, because that way the IPv6 and non-IPv6 
cases were clearly separated.

Best regards,
Jozsef
-
E-mail  : kadlec@blackhole.kfki.hu, kadlec@mail.kfki.hu
PGP key : http://www.kfki.hu/~kadlec/pgp_public_key.txt
Address : KFKI Research Institute for Particle and Nuclear Physics
          H-1525 Budapest 114, POB. 49, Hungary

Re: netfilter: ipset: fix linking with CONFIG_IPV6=n

[Patrick McHardy]

[-- Attachment #1: Type: text/plain, Size: 851 bytes --]

Am 02.02.2011 20:52, schrieb Jozsef Kadlecsik:
> Hi Patrick,
> 
> On Wed, 2 Feb 2011, Patrick McHardy wrote:
> 
>> Add some #ifdefs to unconditionally return false in
>> ip_set_get_ip6_port() when CONFIG_IPV6=n and convert
>> to ipv6_skip_exthdr() to avoid pulling in the ip6_tables
>> module when loading ipset.
> 
> You left out checking fragmented packets. Is it intentional?

Yes, ipv6_skip_exthdr() returns an error for non-first fragments.
For the first fragment skb_header_pointer() checks that enough
data for the port numbers is present.

>> A slightly more code-saving alternative would be to add
>> a dummy inline function or ifdef the callsites. Please
>> let me know which way you prefer.
> 
> A dummy inline looks nicer to me, because that way the IPv6 and non-IPv6 
> cases were clearly separated.

OK, I'm going to commit this patch.


[-- Attachment #2: 05.diff --]
[-- Type: text/plain, Size: 2089 bytes --]

diff --git a/include/linux/netfilter/ipset/ip_set_getport.h b/include/linux/netfilter/ipset/ip_set_getport.h
index 694c433..3882a81 100644
--- a/include/linux/netfilter/ipset/ip_set_getport.h
+++ b/include/linux/netfilter/ipset/ip_set_getport.h
@@ -3,8 +3,18 @@
 
 extern bool ip_set_get_ip4_port(const struct sk_buff *skb, bool src,
 				__be16 *port, u8 *proto);
+
+#if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
 extern bool ip_set_get_ip6_port(const struct sk_buff *skb, bool src,
 				__be16 *port, u8 *proto);
+#else
+static inline bool ip_set_get_ip6_port(const struct sk_buff *skb, bool src,
+				       __be16 *port, u8 *proto)
+{
+	return false;
+}
+#endif
+
 extern bool ip_set_get_ip_port(const struct sk_buff *skb, u8 pf, bool src,
 				__be16 *port);
 
diff --git a/net/netfilter/ipset/ip_set_getport.c b/net/netfilter/ipset/ip_set_getport.c
index 4dd2785..8d52272 100644
--- a/net/netfilter/ipset/ip_set_getport.c
+++ b/net/netfilter/ipset/ip_set_getport.c
@@ -13,6 +13,7 @@
 #include <linux/icmpv6.h>
 #include <linux/netfilter_ipv6/ip6_tables.h>
 #include <net/ip.h>
+#include <net/ipv6.h>
 
 #include <linux/netfilter/ipset/ip_set_getport.h>
 
@@ -93,21 +94,23 @@ ip_set_get_ip4_port(const struct sk_buff *skb, bool src,
 }
 EXPORT_SYMBOL_GPL(ip_set_get_ip4_port);
 
+#if defined(CONFIG_IP6_NF_IPTABLES) || defined(CONFIG_IP6_NF_IPTABLES_MODULE)
 bool
 ip_set_get_ip6_port(const struct sk_buff *skb, bool src,
 		    __be16 *port, u8 *proto)
 {
-	unsigned int protooff = 0;
-	int protocol;
-	unsigned short fragoff;
+	int protoff;
+	u8 nexthdr;
 
-	protocol = ipv6_find_hdr(skb, &protooff, -1, &fragoff);
-	if (protocol <= 0 || fragoff)
+	nexthdr = ipv6_hdr(skb)->nexthdr;
+	protoff = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr);
+	if (protoff < 0)
 		return false;
 
-	return get_port(skb, protocol, protooff, src, port, proto);
+	return get_port(skb, nexthdr, protoff, src, port, proto);
 }
 EXPORT_SYMBOL_GPL(ip_set_get_ip6_port);
+#endif
 
 bool
 ip_set_get_ip_port(const struct sk_buff *skb, u8 pf, bool src, __be16 *port)

Re: netfilter: ipset: fix linking with CONFIG_IPV6=n

[Jozsef Kadlecsik]

On Wed, 2 Feb 2011, Patrick McHardy wrote:

> Am 02.02.2011 20:52, schrieb Jozsef Kadlecsik:
> > Hi Patrick,
> > 
> > On Wed, 2 Feb 2011, Patrick McHardy wrote:
> > 
> >> Add some #ifdefs to unconditionally return false in
> >> ip_set_get_ip6_port() when CONFIG_IPV6=n and convert
> >> to ipv6_skip_exthdr() to avoid pulling in the ip6_tables
> >> module when loading ipset.
> > 
> > You left out checking fragmented packets. Is it intentional?
> 
> Yes, ipv6_skip_exthdr() returns an error for non-first fragments.
> For the first fragment skb_header_pointer() checks that enough
> data for the port numbers is present.
> 
> >> A slightly more code-saving alternative would be to add
> >> a dummy inline function or ifdef the callsites. Please
> >> let me know which way you prefer.
> > 
> > A dummy inline looks nicer to me, because that way the IPv6 and non-IPv6 
> > cases were clearly separated.
> 
> OK, I'm going to commit this patch.

I committed it in my tree too. Thanks!

Best regards,
Jozsef
-
E-mail  : kadlec@blackhole.kfki.hu, kadlec@mail.kfki.hu
PGP key : http://www.kfki.hu/~kadlec/pgp_public_key.txt
Address : KFKI Research Institute for Particle and Nuclear Physics
          H-1525 Budapest 114, POB. 49, Hungary