[PATCH-next] netfilter: don't use module_init/exit in core IPV4 code

[PATCH-next] netfilter: don't use module_init/exit in core IPV4 code

[Paul Gortmaker]

The file net/ipv4/netfilter.o is created based on whether
CONFIG_NETFILTER is set.  However that is defined as a bool, and
hence this file with the core netfilter hooks will never be
modular.  So using module_init as an alias for __initcall can be
somewhat misleading.

Fix this up now, so that we can relocate module_init from
init.h into module.h in the future.  If we don't do this, we'd
have to add module.h to obviously non-modular code, and that
would be a worse thing.  Also add an inclusion of init.h, as
that was previously implicit here in the netfilter.c file.

Note that direct use of __initcall is discouraged, vs. one
of the priority categorized subgroups.  As __initcall gets
mapped onto device_initcall, our use of subsys_initcall (which
seems to make sense for netfilter code) will thus change this
registration from level 6-device to level 4-subsys (i.e. slightly
earlier).  However no observable impact of that small difference
has been observed during testing, or is expected. (i.e. the
location of the netfilter messages in dmesg remains unchanged
with respect to all the other surrounding messages.)

Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>

diff --git a/net/ipv4/netfilter.c b/net/ipv4/netfilter.c
index c3e0adea9c27..e1237a64b14e 100644
--- a/net/ipv4/netfilter.c
+++ b/net/ipv4/netfilter.c
@@ -7,6 +7,7 @@
 #include <linux/kernel.h>
 #include <linux/netfilter.h>
 #include <linux/netfilter_ipv4.h>
+#include <linux/init.h>
 #include <linux/ip.h>
 #include <linux/skbuff.h>
 #include <linux/gfp.h>
@@ -203,5 +204,5 @@ static void __exit ipv4_netfilter_fini(void)
 	nf_unregister_afinfo(&nf_ip_afinfo);
 }
 
-module_init(ipv4_netfilter_init);
-module_exit(ipv4_netfilter_fini);
+subsys_initcall(ipv4_netfilter_init);
+__exitcall(ipv4_netfilter_fini);
-- 
1.8.5.2

[PATCH-next v2] netfilter: don't use module_init/exit in core IPV4 code

[Paul Gortmaker]

The file net/ipv4/netfilter.o is created based on whether
CONFIG_NETFILTER is set.  However that is defined as a bool, and
hence this file with the core netfilter hooks will never be
modular.  So using module_init as an alias for __initcall can be
somewhat misleading.

Fix this up now, so that we can relocate module_init from
init.h into module.h in the future.  If we don't do this, we'd
have to add module.h to obviously non-modular code, and that
would be a worse thing.  Also add an inclusion of init.h, as
that was previously implicit here in the netfilter.c file.

Note that direct use of __initcall is discouraged, vs. one
of the priority categorized subgroups.  As __initcall gets
mapped onto device_initcall, our use of subsys_initcall (which
seems to make sense for netfilter code) will thus change this
registration from level 6-device to level 4-subsys (i.e. slightly
earlier).  However no observable impact of that small difference
has been observed during testing, or is expected. (i.e. the
location of the netfilter messages in dmesg remains unchanged
with respect to all the other surrounding messages.)

As for the module_exit, rather than replace it with __exitcall,
we simply remove it, since it appears only UML does anything
with those, and even for UML, there is no relevant cleanup
to be done here.

Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
---

[v2: Drop __exitcall stuff completely, as per Eric's suggestion
 given for patch at  http://patchwork.ozlabs.org/patch/311164/ ]

 net/ipv4/netfilter.c | 9 +--------
 1 file changed, 1 insertion(+), 8 deletions(-)

diff --git a/net/ipv4/netfilter.c b/net/ipv4/netfilter.c
index c3e0adea9c27..31abf9636ba7 100644
--- a/net/ipv4/netfilter.c
+++ b/net/ipv4/netfilter.c
@@ -197,11 +197,4 @@ static int __init ipv4_netfilter_init(void)
 {
 	return nf_register_afinfo(&nf_ip_afinfo);
 }
-
-static void __exit ipv4_netfilter_fini(void)
-{
-	nf_unregister_afinfo(&nf_ip_afinfo);
-}
-
-module_init(ipv4_netfilter_init);
-module_exit(ipv4_netfilter_fini);
+device_initcall(ipv4_netfilter_init);
-- 
1.8.5.2

Re: [PATCH-next v2] netfilter: don't use module_init/exit in core IPV4 code

[Benjamin Poirier]

On 2014/01/15 15:57, Paul Gortmaker wrote:
> The file net/ipv4/netfilter.o is created based on whether
> CONFIG_NETFILTER is set.  However that is defined as a bool, and
> hence this file with the core netfilter hooks will never be
> modular.  So using module_init as an alias for __initcall can be
> somewhat misleading.
> 
> Fix this up now, so that we can relocate module_init from
> init.h into module.h in the future.  If we don't do this, we'd
> have to add module.h to obviously non-modular code, and that
> would be a worse thing.  Also add an inclusion of init.h, as
> that was previously implicit here in the netfilter.c file.
> 
> Note that direct use of __initcall is discouraged, vs. one
> of the priority categorized subgroups.  As __initcall gets
> mapped onto device_initcall, our use of subsys_initcall (which
> seems to make sense for netfilter code) will thus change this
> registration from level 6-device to level 4-subsys (i.e. slightly
> earlier).  However no observable impact of that small difference
> has been observed during testing, or is expected. (i.e. the
> location of the netfilter messages in dmesg remains unchanged
> with respect to all the other surrounding messages.)
> 
> As for the module_exit, rather than replace it with __exitcall,
> we simply remove it, since it appears only UML does anything
> with those, and even for UML, there is no relevant cleanup
> to be done here.
> 
> Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
> ---
> 
> [v2: Drop __exitcall stuff completely, as per Eric's suggestion
>  given for patch at  http://patchwork.ozlabs.org/patch/311164/ ]
> 
>  net/ipv4/netfilter.c | 9 +--------
>  1 file changed, 1 insertion(+), 8 deletions(-)
> 
> diff --git a/net/ipv4/netfilter.c b/net/ipv4/netfilter.c
> index c3e0adea9c27..31abf9636ba7 100644
> --- a/net/ipv4/netfilter.c
> +++ b/net/ipv4/netfilter.c
> @@ -197,11 +197,4 @@ static int __init ipv4_netfilter_init(void)
>  {
>  	return nf_register_afinfo(&nf_ip_afinfo);
>  }
> -
> -static void __exit ipv4_netfilter_fini(void)
> -{
> -	nf_unregister_afinfo(&nf_ip_afinfo);
> -}
> -
> -module_init(ipv4_netfilter_init);
> -module_exit(ipv4_netfilter_fini);
> +device_initcall(ipv4_netfilter_init);

device_ vs. subsys_initcall changed from v1 of the patch and no longer
matches the description I think.

> -- 
> 1.8.5.2
> 
> --
> To unsubscribe from this list: send the line "unsubscribe netdev" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH-next v2] netfilter: don't use module_init/exit in core IPV4 code

[Paul Gortmaker]

On 14-01-16 02:26 PM, Benjamin Poirier wrote:
> On 2014/01/15 15:57, Paul Gortmaker wrote:
>> The file net/ipv4/netfilter.o is created based on whether
>> CONFIG_NETFILTER is set.  However that is defined as a bool, and
>> hence this file with the core netfilter hooks will never be
>> modular.  So using module_init as an alias for __initcall can be
>> somewhat misleading.
>>
>> Fix this up now, so that we can relocate module_init from
>> init.h into module.h in the future.  If we don't do this, we'd
>> have to add module.h to obviously non-modular code, and that
>> would be a worse thing.  Also add an inclusion of init.h, as
>> that was previously implicit here in the netfilter.c file.
>>
>> Note that direct use of __initcall is discouraged, vs. one
>> of the priority categorized subgroups.  As __initcall gets
>> mapped onto device_initcall, our use of subsys_initcall (which
>> seems to make sense for netfilter code) will thus change this
>> registration from level 6-device to level 4-subsys (i.e. slightly
>> earlier).  However no observable impact of that small difference
>> has been observed during testing, or is expected. (i.e. the
>> location of the netfilter messages in dmesg remains unchanged
>> with respect to all the other surrounding messages.)
>>
>> As for the module_exit, rather than replace it with __exitcall,
>> we simply remove it, since it appears only UML does anything
>> with those, and even for UML, there is no relevant cleanup
>> to be done here.
>>
>> Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
>> ---
>>
>> [v2: Drop __exitcall stuff completely, as per Eric's suggestion
>>  given for patch at  http://patchwork.ozlabs.org/patch/311164/ ]
>>
>>  net/ipv4/netfilter.c | 9 +--------
>>  1 file changed, 1 insertion(+), 8 deletions(-)
>>
>> diff --git a/net/ipv4/netfilter.c b/net/ipv4/netfilter.c
>> index c3e0adea9c27..31abf9636ba7 100644
>> --- a/net/ipv4/netfilter.c
>> +++ b/net/ipv4/netfilter.c
>> @@ -197,11 +197,4 @@ static int __init ipv4_netfilter_init(void)
>>  {
>>  	return nf_register_afinfo(&nf_ip_afinfo);
>>  }
>> -
>> -static void __exit ipv4_netfilter_fini(void)
>> -{
>> -	nf_unregister_afinfo(&nf_ip_afinfo);
>> -}
>> -
>> -module_init(ipv4_netfilter_init);
>> -module_exit(ipv4_netfilter_fini);
>> +device_initcall(ipv4_netfilter_init);
> 
> device_ vs. subsys_initcall changed from v1 of the patch and no longer
> matches the description I think.

Thanks for spotting that ; not sure how I fat fingered that
during the exitcall removal.  Trying to rush things I guess.

P.
--

> 
>> -- 
>> 1.8.5.2
>>
>> --
>> To unsubscribe from this list: send the line "unsubscribe netdev" in
>> the body of a message to majordomo@vger.kernel.org
>> More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH-next v3] netfilter: don't use module_init/exit in core IPV4 code

[Paul Gortmaker]

The file net/ipv4/netfilter.o is created based on whether
CONFIG_NETFILTER is set.  However that is defined as a bool, and
hence this file with the core netfilter hooks will never be
modular.  So using module_init as an alias for __initcall can be
somewhat misleading.

Fix this up now, so that we can relocate module_init from
init.h into module.h in the future.  If we don't do this, we'd
have to add module.h to obviously non-modular code, and that
would be a worse thing.  Also add an inclusion of init.h, as
that was previously implicit here in the netfilter.c file.

Note that direct use of __initcall is discouraged, vs. one
of the priority categorized subgroups.  As __initcall gets
mapped onto device_initcall, our use of subsys_initcall (which
seems to make sense for netfilter code) will thus change this
registration from level 6-device to level 4-subsys (i.e. slightly
earlier).  However no observable impact of that small difference
has been observed during testing, or is expected. (i.e. the
location of the netfilter messages in dmesg remains unchanged
with respect to all the other surrounding messages.)

As for the module_exit, rather than replace it with __exitcall,
we simply remove it, since it appears only UML does anything
with those, and even for UML, there is no relevant cleanup
to be done here.

Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
---

[v3: switch accidental device_initcall back to subsys_initcall
 as spotted by Benjamin Poirier; retest on ppc-sbc8548 net-next]

[v2: Drop __exitcall stuff completely, as per Eric's suggestion
 given for patch at  http://patchwork.ozlabs.org/patch/311164/ ]

 net/ipv4/netfilter.c | 9 +--------
 1 file changed, 1 insertion(+), 8 deletions(-)

diff --git a/net/ipv4/netfilter.c b/net/ipv4/netfilter.c
index c3e0adea9c27..31abf9636ba7 100644
--- a/net/ipv4/netfilter.c
+++ b/net/ipv4/netfilter.c
@@ -197,11 +197,4 @@ static int __init ipv4_netfilter_init(void)
 {
 	return nf_register_afinfo(&nf_ip_afinfo);
 }
-
-static void __exit ipv4_netfilter_fini(void)
-{
-	nf_unregister_afinfo(&nf_ip_afinfo);
-}
-
-module_init(ipv4_netfilter_init);
-module_exit(ipv4_netfilter_fini);
+subsys_initcall(ipv4_netfilter_init);
-- 
1.8.5.2

Re: [PATCH-next v3] netfilter: don't use module_init/exit in core IPV4 code

[Patrick McHardy]

On Thu, Jan 16, 2014 at 03:24:03PM -0500, Paul Gortmaker wrote:
> The file net/ipv4/netfilter.o is created based on whether
> CONFIG_NETFILTER is set.  However that is defined as a bool, and
> hence this file with the core netfilter hooks will never be
> modular.  So using module_init as an alias for __initcall can be
> somewhat misleading.
> 
> Fix this up now, so that we can relocate module_init from
> init.h into module.h in the future.  If we don't do this, we'd
> have to add module.h to obviously non-modular code, and that
> would be a worse thing.  Also add an inclusion of init.h, as
> that was previously implicit here in the netfilter.c file.
> 
> Note that direct use of __initcall is discouraged, vs. one
> of the priority categorized subgroups.  As __initcall gets
> mapped onto device_initcall, our use of subsys_initcall (which
> seems to make sense for netfilter code) will thus change this
> registration from level 6-device to level 4-subsys (i.e. slightly
> earlier).  However no observable impact of that small difference
> has been observed during testing, or is expected. (i.e. the
> location of the netfilter messages in dmesg remains unchanged
> with respect to all the other surrounding messages.)
> 
> As for the module_exit, rather than replace it with __exitcall,
> we simply remove it, since it appears only UML does anything
> with those, and even for UML, there is no relevant cleanup
> to be done here.

Just to bring up the idea: there's no reason why we couldn't make
netfilter fully modular for individual AFs. I've seen quite a lot
of embedded systems that include the IPv6 netfilter core without
using it.

Its rather complicated to get all the dependencies of xtables
modules right though, so this is really just to bring up the
idea.

Your patch looks fine to me.