[v3] Fix handling of IPv4/IPv6 dual stack

[Qemu-devel] [PATCH v3 0/5] Fix handling of IPv4/IPv6 dual stack

Posted by Daniel P. Berrange 8 years, 5 months ago

This is a followup to:

  v1: https://lists.nongnu.org/archive/html/qemu-devel/2017-04/msg05659.html
  v2: https://lists.nongnu.org/archive/html/qemu-devel/2017-05/msg04706.html

This series aims to fix a lot of bugs related to handling of IPv4 / IPv6
dual stack.

 - The VNC server mistakenly listened on two separate ports 5900+5901
   when the to= parameter was given
 - IPv6 sockets are accepting IPv4 clients even when IPv4 is set to
   be disabled
 - IPv6 sockets are failing to accept IPv4 clients when IPv4 is not set
   to be disabled
 - The VNC server was loosing the ipv4=/ipv6= settings due to a bug
   in the DNS resolver

The behaviour of all this is really subtle and hard to get working correctly
across all the different network backends. Thus, the most important part of
this patch series is the last patch which adds a test case covering the
backends for -vnc, -chardev tcp, -net socket, and -incoming socket, with
a 120 entry matrix.

IOW, if you think any of the first 4 patches are applying the wrong logic,
then take a look at the last patch and indicate which test matrix entries
are believed to be defining wrong behaviour :-)

Changed in v2:

 - Clarified error message in commit message
 - Replace assert with exit (Eric)
 - Fix typo in test comment (Eric)
 - Fix wierd copyright line (Eric)

Daniel P. Berrange (5):
  sockets: ensure we can bind to both ipv4 & ipv6 separately
  sockets: don't block IPv4 clients when listening on "::"
  sockets: ensure we don't accept IPv4 clients when IPv4 is disabled
  io: preserve ipv4/ipv6 flags when resolving InetSocketAddress
  tests: add functional test validating ipv4/ipv6 address flag handling

 io/dns-resolver.c          |   6 +-
 tests/.gitignore           |   1 +
 tests/Makefile.include     |   4 +
 tests/test-sockets-proto.c | 855 +++++++++++++++++++++++++++++++++++++++++++++
 util/qemu-sockets.c        |  71 +++-
 5 files changed, 916 insertions(+), 21 deletions(-)
 create mode 100644 tests/test-sockets-proto.c

-- 
2.9.3

Re: [Qemu-devel] [PATCH v3 0/5] Fix handling of IPv4/IPv6 dual stack

Posted by Gerd Hoffmann 8 years, 5 months ago

On Thu, 2017-06-01 at 09:29 +0100, Daniel P. Berrange wrote:
> This series aims to fix a lot of bugs related to handling of IPv4 /
> IPv6
> dual stack.
> 
>  - The VNC server mistakenly listened on two separate ports 5900+5901
>    when the to= parameter was given
>  - IPv6 sockets are accepting IPv4 clients even when IPv4 is set to
>    be disabled
>  - IPv6 sockets are failing to accept IPv4 clients when IPv4 is not
> set
>    to be disabled
>  - The VNC server was loosing the ipv4=/ipv6= settings due to a bug
>    in the DNS resolver
> 
> The behaviour of all this is really subtle and hard to get working
> correctly
> across all the different network backends. Thus, the most important
> part of
> this patch series is the last patch which adds a test case covering
> the
> backends for -vnc, -chardev tcp, -net socket, and -incoming socket,
> with
> a 120 entry matrix.

Acked-by: Gerd Hoffmann <kraxel@gmail.com>

Re: [Qemu-devel] [PATCH v3 0/5] Fix handling of IPv4/IPv6 dual stack

Posted by Vlad Yasevich 8 years, 4 months ago

On 06/01/2017 04:29 AM, Daniel P. Berrange wrote:
> This is a followup to:
> 
>   v1: https://lists.nongnu.org/archive/html/qemu-devel/2017-04/msg05659.html
>   v2: https://lists.nongnu.org/archive/html/qemu-devel/2017-05/msg04706.html
> 
> This series aims to fix a lot of bugs related to handling of IPv4 / IPv6
> dual stack.
> 
>  - The VNC server mistakenly listened on two separate ports 5900+5901
>    when the to= parameter was given
>  - IPv6 sockets are accepting IPv4 clients even when IPv4 is set to
>    be disabled
>  - IPv6 sockets are failing to accept IPv4 clients when IPv4 is not set
>    to be disabled
>  - The VNC server was loosing the ipv4=/ipv6= settings due to a bug
>    in the DNS resolver
> 
> The behaviour of all this is really subtle and hard to get working correctly
> across all the different network backends. Thus, the most important part of
> this patch series is the last patch which adds a test case covering the
> backends for -vnc, -chardev tcp, -net socket, and -incoming socket, with
> a 120 entry matrix.
> 
> IOW, if you think any of the first 4 patches are applying the wrong logic,
> then take a look at the last patch and indicate which test matrix entries
> are believed to be defining wrong behaviour :-)
> 
> Changed in v2:
> 
>  - Clarified error message in commit message
>  - Replace assert with exit (Eric)
>  - Fix typo in test comment (Eric)
>  - Fix wierd copyright line (Eric)
> 
> Daniel P. Berrange (5):
>   sockets: ensure we can bind to both ipv4 & ipv6 separately
>   sockets: don't block IPv4 clients when listening on "::"
>   sockets: ensure we don't accept IPv4 clients when IPv4 is disabled
>   io: preserve ipv4/ipv6 flags when resolving InetSocketAddress
>   tests: add functional test validating ipv4/ipv6 address flag handling
> 
>  io/dns-resolver.c          |   6 +-
>  tests/.gitignore           |   1 +
>  tests/Makefile.include     |   4 +
>  tests/test-sockets-proto.c | 855 +++++++++++++++++++++++++++++++++++++++++++++
>  util/qemu-sockets.c        |  71 +++-
>  5 files changed, 916 insertions(+), 21 deletions(-)
>  create mode 100644 tests/test-sockets-proto.c
> 

Series Reviewed-by: Vlad Yasevich <vyasevic@redhat.com>

-vlad