Re: [PATCH] devinet: sysctl setup off by one

Re: [PATCH] devinet: sysctl setup off by one

[David Miller]

From: Stephen Hemminger <stephen.hemminger@vyatta.com>
Date: Fri, 23 May 2008 15:35:59 -0700

> It look like the code to setup /proc/sys/net/ipv4/conf/XXX has an
> off by one bug that causes the last entry (arp_accept) not to be
> adjusted properly.  This causes changes to the value to occur in the
> default namespace not the clone? Don't use network namespaces
> myself, but this looks like a bug.
>
> Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>

Yikes, that is a very strange expression to be there by
accident.

Pavel, can you take a look at this?  Thanks.

Re: [PATCH] devinet: sysctl setup off by one

[Pavel Emelyanov]

Stephen Hemminger wrote:
> It look like the code to setup /proc/sys/net/ipv4/conf/XXX has an off by one bug
> that causes the last entry (arp_accept) not to be adjusted properly.  This causes
> changes to the value to occur in the default namespace not the clone? Don't use
> network namespaces myself, but this looks like a bug.
> 
> Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
> 
> --- a/net/ipv4/devinet.c	2008-05-23 15:23:19.000000000 -0700
> +++ b/net/ipv4/devinet.c	2008-05-23 15:29:17.000000000 -0700
> @@ -1466,7 +1466,7 @@ static int __devinet_sysctl_register(str
>  	if (!t)
>  		goto out;
>  
> -	for (i = 0; i < ARRAY_SIZE(t->devinet_vars) - 1; i++) {
> +	for (i = 0; i < ARRAY_SIZE(t->devinet_vars); i++) {
>  		t->devinet_vars[i].data += (char *)p - (char *)&ipv4_devconf;
>  		t->devinet_vars[i].extra1 = p;
>  		t->devinet_vars[i].extra2 = net;
> 

This off-by-one is required, since the very last element of this array
is used as a sentinel, that shows sysctl engine when to stop processing 
the table.

Look, the sizeof(t->devinet_vars) is __NET_IPV4_CONF_MAX == 22, thus
entries 0, 1, ... 20 are tables and the 21-st is this sentinel, so the
initialization of this last entry, whose index is (this_sizeof() - 1),
is not required.

Thanks,
Pavel

Re: [PATCH] devinet: sysctl setup off by one

[Stephen Hemminger]

On Mon, 26 May 2008 11:39:21 +0400
Pavel Emelyanov <xemul@openvz.org> wrote:

> Stephen Hemminger wrote:
> > It look like the code to setup /proc/sys/net/ipv4/conf/XXX has an off by one bug
> > that causes the last entry (arp_accept) not to be adjusted properly.  This causes
> > changes to the value to occur in the default namespace not the clone? Don't use
> > network namespaces myself, but this looks like a bug.
> > 
> > Signed-off-by: Stephen Hemminger <shemminger@vyatta.com>
> > 
> > --- a/net/ipv4/devinet.c	2008-05-23 15:23:19.000000000 -0700
> > +++ b/net/ipv4/devinet.c	2008-05-23 15:29:17.000000000 -0700
> > @@ -1466,7 +1466,7 @@ static int __devinet_sysctl_register(str
> >  	if (!t)
> >  		goto out;
> >  
> > -	for (i = 0; i < ARRAY_SIZE(t->devinet_vars) - 1; i++) {
> > +	for (i = 0; i < ARRAY_SIZE(t->devinet_vars); i++) {
> >  		t->devinet_vars[i].data += (char *)p - (char *)&ipv4_devconf;
> >  		t->devinet_vars[i].extra1 = p;
> >  		t->devinet_vars[i].extra2 = net;
> > 
> 
> This off-by-one is required, since the very last element of this array
> is used as a sentinel, that shows sysctl engine when to stop processing 
> the table.
> 
> Look, the sizeof(t->devinet_vars) is __NET_IPV4_CONF_MAX == 22, thus
> entries 0, 1, ... 20 are tables and the 21-st is this sentinel, so the
> initialization of this last entry, whose index is (this_sizeof() - 1),
> is not required.
> 
> Thanks,
> Pavel

okay, so that means there is one extra entry in the vars than needed? which
is harmless.