[squid-dev] [PATCH] Fixed step3 splicing

Amos Jeffries squid3 at treenet.co.nz
Sun Feb 7 01:52:58 UTC 2016


On 5/02/2016 5:14 a.m., Christos Tsantilas wrote:
> The information about PeekingPeerConnector splicing the connections
> was lost in some cases, resulting in two different bugs:
> 
>  - With a certificate validator, the PeekingPeerConnector class calls
> back FwdState, which calls the ConnStateData class, which then tries
> secure the connection with the already tunneled SSL client and closes
> the connection on negotiating errors.
> 
>  - Without a certificate validator, the PeekingPeerConnector class never
> calls FwdState class, and both PeekingPeerConnector and FwdState objects
> stall until finishing tunnelState closes server and client connections.
> 
> Now, PeerConnector always calls FwdState back, marking spliced
> connections as such. This has the following positive side-effects:
> 
>  - When FwdState learns about spliced connections, it does not call
> ConnStateData back. Instead, it terminates and gets destroyed. The
> tunnel continues uninterrupted.
> 
>  - The PeekingPeerConnector job ends and is destroyed instead of waiting
> to call FwdState.
> 


+1. Looks reasonable. If it has been tested out already I'm okay with it
going in.

Amos



More information about the squid-dev mailing list