Re: [PATCH net-next v3] ipv6: add IFLA_INET6_RA_MTU to expose mtu value in the RA message

[David Ahern]

On 8/11/21 7:56 AM, David Ahern wrote:
> On 8/10/21 6:33 AM, Rocco Yue wrote:
>> On Mon, 2021-08-09 at 16:43 -0600, David Ahern wrote:
>>> On 8/9/21 8:01 AM, Rocco Yue wrote:
>>
>>> +
>>>>  #ifdef CONFIG_SYSCTL
>>>>  
>>>>  static int addrconf_sysctl_forward(struct ctl_table *ctl, int write,
>>>> diff --git a/net/ipv6/ndisc.c b/net/ipv6/ndisc.c
>>>> index c467c6419893..a04164cbd77f 100644
>>>> --- a/net/ipv6/ndisc.c
>>>> +++ b/net/ipv6/ndisc.c
>>>> @@ -1496,6 +1496,12 @@ static void ndisc_router_discovery(struct sk_buff *skb)
>>>>  		memcpy(&n, ((u8 *)(ndopts.nd_opts_mtu+1))+2, sizeof(mtu));
>>>>  		mtu = ntohl(n);
>>>>  
>>>> +		if (in6_dev->ra_mtu != mtu) {
>>>> +			in6_dev->ra_mtu = mtu;
>>>> +			inet6_iframtu_notify(in6_dev);
>>>> +			ND_PRINTK(2, info, "update ra_mtu to %d\n", in6_dev->ra_mtu);
>>>> +		}
>>>> +
>>>>  		if (mtu < IPV6_MIN_MTU || mtu > skb->dev->mtu) {
>>>>  			ND_PRINTK(2, warn, "RA: invalid mtu: %d\n", mtu);
>>>>  		} else if (in6_dev->cnf.mtu6 != mtu) {
>>>
>>> Since this MTU is getting reported via af_info infrastructure,
>>> rtmsg_ifinfo should be sufficient.
>>>
>>> From there use 'ip monitor' to make sure you are not generating multiple
>>> notifications; you may only need this on the error path.
>>
>> Hi David,
>>
>> To avoid generating multiple notifications, I added a separate ramtu notify
>> function in this patch, and I added RTNLGRP_IPV6_IFINFO nl_mgrp to the ipmonitor.c
>> to verify this patch was as expected.
>>
>> I look at the rtmsg_ifinfo code, it should be appropriate and I will use it and
>> verify it.
>>
>> But there's one thing, I'm sorry I didn't fully understand the meaning of this
>> sentence "you may only need this on the error path". Honestly, I'm not sure what
>> the error patch refers to, do you mean "if (mtu < IPV6_MIN_MTU || mtu > skb->dev->mtu)" ?
>>
> 
> looks like nothing under:
>     if (ndopts.nd_opts_mtu && in6_dev->cnf.accept_ra_mtu) {
> 
>     }
> 
> is going to send a link notification so you can just replace
> inet6_iframtu_notify with rtmsg_ifinfo in your proposed change.
> 

Taking a deeper dive on the code, you do not need to call rtmsg_ifinfo.
Instead, the existing:

        /*
         *      Send a notify if RA changed managed/otherconf flags or
timer settings
         */
        if (send_ifinfo_notify)
                inet6_ifinfo_notify(RTM_NEWLINK, in6_dev);

is called too early. For one the RA can change the MTU and that is done
after this notify.

I think if you moved this down to the out:

out:
        /*
         *      Send a notify if RA changed managed/otherconf flags or
timer settings
         */
        if (send_ifinfo_notify)
                inet6_ifinfo_notify(RTM_NEWLINK, in6_dev);

and then set send_ifinfo_notify when the mtu is *changed* by the RA you
should be good.