retrans path update problem

retrans path update problem

[Gui Jianfeng]

hello everybody,
When updating retrans path(asoc->peer.retran_path), should 
we update the retran path to another transport even if only the 
original retrans path is active?
In the other words, i think, retrans path should always be 
changed if there are more than one transport in an association.

Am i right?
-- 
Regards
Gui Jianfeng

Re: retrans path update problem

[Vlad Yasevich]

Gui Jianfeng wrote:
> hello everybody,
> When updating retrans path(asoc->peer.retran_path), should 
> we update the retran path to another transport even if only the 
> original retrans path is active?
> In the other words, i think, retrans path should always be 
> changed if there are more than one transport in an association.
> 
> Am i right?

And we do that with sctp_assoc_update_retran_path().  If we
don't find another active transport, we'll still try set the
retran_path to the next transport in the list.

Are you asking if we should only use ACTIVE transports?

-vlad

Re: retrans path update problem

[Gui Jianfeng]

Vlad Yasevich wrote:
> Gui Jianfeng wrote:
>> hello everybody,
>> When updating retrans path(asoc->peer.retran_path), should we update
>> the retran path to another transport even if only the original retrans
>> path is active?
>> In the other words, i think, retrans path should always be changed if
>> there are more than one transport in an association.
>>
>> Am i right?
> 
> And we do that with sctp_assoc_update_retran_path().  If we
> don't find another active transport, we'll still try set the
> retran_path to the next transport in the list.
> 
> Are you asking if we should only use ACTIVE transports?

Vlad,  
  I'd like to know, if the current asoc->peer.retran_path is the only active
  transport in the transport list, whether we should change asoc->peer.retran_path 
  to another inactive one. The sctp_assoc_update_retran_path() won't change to a 
  inactive transport(next one) in this scenario now, is that ok?
> 
> -vlad
> 
> 
> 

-- 
Regards
Gui Jianfeng

Re: retrans path update problem

[Vlad Yasevich]

Gui Jianfeng wrote:
> Vlad Yasevich wrote:
>> Gui Jianfeng wrote:
>>> hello everybody,
>>> When updating retrans path(asoc->peer.retran_path), should we update
>>> the retran path to another transport even if only the original retrans
>>> path is active?
>>> In the other words, i think, retrans path should always be changed if
>>> there are more than one transport in an association.
>>>
>>> Am i right?
>> And we do that with sctp_assoc_update_retran_path().  If we
>> don't find another active transport, we'll still try set the
>> retran_path to the next transport in the list.
>>
>> Are you asking if we should only use ACTIVE transports?
> 
> Vlad,  
>   I'd like to know, if the current asoc->peer.retran_path is the only active
>   transport in the transport list, whether we should change asoc->peer.retran_path 
>   to another inactive one. The sctp_assoc_update_retran_path() won't change to a 
>   inactive transport(next one) in this scenario now, is that ok?

I see now.  The appears to be a bug in sctp_assoc_update_retran_path().  Even
though it attempts to change to a non-active one based on this comment,

      /* We have exhausted the list, but didn't find any
       * other active transports.  If so, use the next
       * transport.
       */

it never does, since it finds the active retran_path and just uses that again.

Feel free to fix.

-vlad