network: 0.930
device: 0.926
user-level: 0.908
risc-v: 0.898
PID: 0.887
vnc: 0.884
graphic: 0.883
peripherals: 0.879
VMM: 0.861
socket: 0.861
mistranslation: 0.855
semantic: 0.852
debug: 0.842
architecture: 0.836
TCG: 0.832
x86: 0.827
permissions: 0.827
ppc: 0.822
hypervisor: 0.821
performance: 0.795
register: 0.787
files: 0.765
arm: 0.744
boot: 0.743
assembly: 0.707
virtual: 0.682
i386: 0.673
KVM: 0.647
kernel: 0.597

[2.6] user network broken reaching foreign servers on Win64

Both on master and, starting with 2016-03-22 builds from http://qemu.weilnetz.de/w64/, user mode network can't reach foreign servers. For example, wget http://mifritscher resolves the DNS, but then the message "network target couldn't be reached" occures. 2016-03-03 works fine. I suspect the IPv6 changes. My connection is IPv4 only.

I tested via knoppix 7.6. The command line is

qemu-system-x86_64.exe -m 512 -k de --cdrom c:\Users\michaelfritscher\Downloads\linux\KNOPPIX_V7.4.2DVD-2014-09-28-DE.iso -netdev user,id=mynet0,restrict=n -device e1000,netdev=mynet0

Under Linux, it is working fine.

Another thing: if I disable ipv6, ipv4 runs still in sort of a timeout before writing  Network is unreachable (ENETUNREACH), while ipv6 says this without delay (which is ok/good)


It looks like this commit caused the regression:

$ git bisect bad
c619644067f98098dcdbc951e2dda79e97560afa is the first bad commit
commit c619644067f98098dcdbc951e2dda79e97560afa
Author: Daniel P. Berrange <email address hidden>
Date:   Mon Mar 7 11:19:18 2016 +0000

    osdep: fix socket_error() to work with Mingw64


Here is a patch which fixes the problem.

diff --git a/slirp/tcp_input.c b/slirp/tcp_input.c
index 2027a75..696867f 100644
--- a/slirp/tcp_input.c
+++ b/slirp/tcp_input.c
@@ -587,7 +587,7 @@ findso:
 
          if ((tcp_fconnect(so, so->so_ffamily) == -1) &&
 #if defined(_WIN32)
-              socket_error() != WSAEWOULDBLOCK
+              socket_error() != EAGAIN
 #else
               (errno != EINPROGRESS) && (errno != EWOULDBLOCK)
 #endif


Latest QEMU needs a slightly different patch:

diff --git a/slirp/tcp_input.c b/slirp/tcp_input.c
index 5433e7f..3be2d2f 100644
--- a/slirp/tcp_input.c
+++ b/slirp/tcp_input.c
@@ -659,6 +659,7 @@ findso:
          }
 
          if ((tcp_fconnect(so, so->so_ffamily) == -1) &&
+              socket_error() != EAGAIN &&
               (errno != EINPROGRESS) && (errno != EWOULDBLOCK)
           ) {
            uint8_t code;

Fixed since v2.6.0-rc3 release.