Re: [tipc-discussion] [PATCH] tipc: remove periodic flush of reassembly buffers

[Ying X. <yin...@wi...>]

eri...@er... wrote:
> From: Erik Hugne <eri...@er...>
>
> There is a periodic per-link job running every [link tolerance] seconds.
> This job performs a check on the active reassembly buffers, and if any
> of these have not received a packet during the last 4 runs it will be
> purged. This causes an inconsistency on the link level, since the
> individual fragments from [1...(n-x)] have already been acked and
> retransmission will only be done for packets [(n-x)...n].
> These "orphaned" fragments will be free'd when we cannot find an active
> reassembly buffer for them. The end result of this is that the message
> is lost silently on the link level.
>
> The selected solution for this is to remove the periodic check for stale
> reassembly buffers. This is safe because the buffers are freed when
> a link is reset and repeated reassembly failures will always cause a link
> to reset. The reassembly timeout mechanism was originally introduced as
> part of the (abandonded) effort to introduce multi-cluster functionality
> where message could be routed between links.
>
> Signed-off-by: Erik Hugne <eri...@er...>
>   

Acked-by: Ying Xue <yin...@wi...>