[Qemu-devel] [slirp] guest program's tcp connection hang on close_wait.

[Sam Liao]

Hi All,

    I am using qemu-arm running linux (guest), I start a server
program on the linux(guest); then I start a program on host machine to
connect to the server(on qemu linux guest) to a specified tcp port.
(the qemu is started with tcp redir configuration).

    Then I got a problem when the client program on host closed the
tcp connection when the server is busy sending data to client. The
server program will hang on close_wait state.  I captured the tcp
packets between the host <-> qemu and qemu<->guest linux. It seems
like the connection is closed like:

1. host <-> qemu

   FIN -->
       <-- ACK
       <-- More data with PUSH
   RST -->

2. qemu <-> guest

   FIN -->
       <--  ACK
       <-- More data
   ACK -->
       <-- More data
   ACK -->
       ...
   ACK, zero window->
        <- Keep-Alive
   ACK, zero window->
        <- keep alive

So, here the connection between the qemu and guest did not closed
correctly. After the qemu's FIN is acked by guest os, the qemu still
receives data until it's
window came to be zero(which I think maybe the receive buf is filled),
then it came to a zerowindow-keepalive loop, and the connection is
hanging there. I checked the slirp's source code, from the packet
between host and qemu, the host connection is reset, and the problem
is the connection between qemu and guest is not correctly closed by
slirp.

I'm not sure this is a slirp bug or something else, any suggestion
would be appreciated.

Thanks.

-Sam