[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <AANLkTi=Yoi-gCfCimpyxjrE9idwqcYNGM5Fn22gEaorg@mail.gmail.com>
Date: Mon, 26 Jul 2010 16:37:37 -0400
From: Mahesh Kelkar <maheshkelkar@...il.com>
To: Brian Haley <brian.haley@...com>
Cc: netdev@...r.kernel.org, David Miller <davem@...emloft.net>
Subject: Re: Last Ipv6 address removal causes "addrconf_sysctl_unregister",
which inihibits from changing disable_ipv6 setting
Thanks. Please see comments inline [mahesh]
On Mon, Jul 26, 2010 at 4:17 PM, Brian Haley <brian.haley@...com> wrote:
> On 07/26/2010 10:03 AM, Mahesh Kelkar wrote:
>> Odd behavior associated with the patch:
>> **Last address removal causes "addrconf_sysctl_unregister", which
>> inihibits from changing disable_ipv6 setting
>> (connected issue: With disable_ipv6 set to 1 on an interface, ff00:/8
>> and fe80::/64 are still added on device UP)
>
> This behavior doesn't seem specific to the patch that was accepted last
> week to fix the disable_ipv6 behavior (64e724f6) as all of the sysctl
> values are gone for eth0.
>
[mahesh] Your are right. The behavior is not specific to the patch.
Actually I meant to say that the behavior can be observed with the
disable_ipv6 in general.
>> Current sysctl config:
>> net.ipv6.conf.all.disable_ipv6 = 1
>> net.ipv6.conf.default.disable_ipv6 = 1
>>
>> Steps:
>> - Remove last IPv6 address assigned to the "eth0" interface
>> - inet6_addr_del => addrconf_ifdown(idev->dev, 1) => does the device
>> sysctl unregister
>> ******Not sure why the addrconf_sysctl_unregister is necessary on last
>> address removal*******
>> - Now, "sysctl -a" does not show "net.ipv6.conf.eth0.disable_ipv6"
>
> Do they get restored if you bring the link down and back up?
>
[mahesh] Yes. It does get restored on link down and back up.
>> Problem:
>> - If you WANT to assign IPv6 address to eth0,
>> -> Do it once, which fails due to "disable_ipv6" check in
>> addrconf_add_dev OR ipv6_add_addr
>> -> But, this process does "addrconf_sysctl_register" (addrconf_add_dev
>> => ipv6_find_idev => ipv6_add_dev)
>> -> set net.ipv6.conf.eth0.disable_ipv6=0 and then successfully assign
>> ipv6-address to the eth0
>> (Another alternative is to change all or default to 1; but I wanted to
>> disable ipv6 by default)
>>
>> ===============
>> Probable Solution:
>> ===============
>> @@ -1948,7 +1959,7 @@ static int inet6_addr_del(int ifindex, s
>> disable IPv6 on this interface.
>> */
>> if (idev->addr_list == NULL)
>> - addrconf_ifdown(idev->dev, 1);
>> + addrconf_ifdown(idev->dev, 0);
>> return 0;
>> }
>> }
>> I have tested the above solution and it seems to work fine - so far.
>
> This code has been this way forever in 2.6 (< 2005), changing it could break
> existing users. I'm not sure why it was decided to do it this way as it
> was before my time.
>
[mahesh] Can anyone on netdev list sheds light on this?
> Also, the full comment here is:
>
> /* If the last address is deleted administratively,
> disable IPv6 on this interface.
> */
>
> Which says that you're removing all the IPv6 addresses by hand using
> /sbin/ip, so you're forcing the removal. If you want to disable IPv6
> on it and remove all the addresses, why don't you just set disable_ipv6=1?
>
[mahesh] I did not add the comment.
[mahesh] But, you are right. One of the alternative is to disable_ipv6
instead of removing last IP address.
[mahesh] But, if you or somebody else happen to touch this code in
future, please see if you could avoid
addrconf_sysctl_register/unregister.
> -Brian
>
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Powered by blists - more mailing lists