[Qemu-devel] [PATCH v7 0/4] Unit test+fix for problem with QEMU handling of multiple bind()s to the same port

Knut Omang posted 4 patches 6 years, 8 months ago
Patches applied successfully (tree, apply log)
git fetch https://github.com/patchew-project/qemu tags/patchew/cover.9aca98788e1a99635ffb85ab9c01cb0497d2b3d8.1502103332.git-series.knut.omang@oracle.com
Test FreeBSD passed
Test checkpatch passed
Test docker passed
Test s390x passed
tests/Makefile.include |   2 +-
tests/test-listen.c    | 253 ++++++++++++++++++++++++++++++++++++++++++-
util/qemu-sockets.c    | 138 +++++++++++++++--------
3 files changed, 345 insertions(+), 48 deletions(-)
create mode 100644 tests/test-listen.c
[Qemu-devel] [PATCH v7 0/4] Unit test+fix for problem with QEMU handling of multiple bind()s to the same port
Posted by Knut Omang 6 years, 8 months ago
This series contains:
* a unit test that exposes a race condition which causes QEMU to fail
  to find a port even when there is plenty of available ports.
* a refactor of the qemu-sockets inet_listen_saddr() function
  to better handle this situation.

Changes from v6:
* Changed license of the (new) test-listen.c source file from
  GNU v2 to GNU v2 and later, according to QEMU standards.

Changes from v5:
* Also move setting of error from socket creation out
  into the main double for loop.
* Simplify and enhance reporting if socket creation fails:
  Only report failure to create sockets if none of the
  provided addrinfo list elements allows creation of a
  socket, otherwise report failure to find an available port.
* Further simplify if's within the for loops and make sure
  failure to recreate a socket gets distinctively reported.
* Rebased to current master.

Changes from v4:
* Move the complexity of recreating a socket and setting the error pointer
  into the main for loop, eliminating the try_bind_listen() function
  again. Cleaning up and improving error handling in the process.

Changes from v3:
* Test changes: Add missing license
  Add subtests for ipv4, ipv6 and both
  Various g_* usage improvements
* Split patch into 3 patches with two refactoring patches ahead
  of the actual fix.

Changes from v2:
* Non-trivial rebase + further abstraction
  on top of 7ad9af343c7f1c70c8015c7c519c312d8c5f9fa1
  'tests: add functional test validating ipv4/ipv6 address flag handling'

Changes from v1:
* Fix potential uninitialized variable only detected by optimize.
* Improve unexpected error detection in test-listen to give more
  details about why the test fails unexpectedly.
* Fix some line length style issues.

Thanks,
Knut

Knut Omang (4):
  tests: Add test-listen - a stress test for QEMU socket listen
  sockets: factor out a new try_bind() function
  sockets: factor out create_fast_reuse_socket
  sockets: Handle race condition between binds to the same port

 tests/Makefile.include |   2 +-
 tests/test-listen.c    | 253 ++++++++++++++++++++++++++++++++++++++++++-
 util/qemu-sockets.c    | 138 +++++++++++++++--------
 3 files changed, 345 insertions(+), 48 deletions(-)
 create mode 100644 tests/test-listen.c

base-commit: a588c4985eff363154d65aee8607d0a4601655f7
-- 
git-series 0.9.1

Re: [Qemu-devel] [PATCH v7 0/4] Unit test+fix for problem with QEMU handling of multiple bind()s to the same port
Posted by Daniel P. Berrange 6 years, 8 months ago
On Mon, Aug 07, 2017 at 12:58:38PM +0200, Knut Omang wrote:
> This series contains:
> * a unit test that exposes a race condition which causes QEMU to fail
>   to find a port even when there is plenty of available ports.
> * a refactor of the qemu-sockets inet_listen_saddr() function
>   to better handle this situation.
> 
> Changes from v6:
> * Changed license of the (new) test-listen.c source file from
>   GNU v2 to GNU v2 and later, according to QEMU standards.
> 
> Changes from v5:
> * Also move setting of error from socket creation out
>   into the main double for loop.
> * Simplify and enhance reporting if socket creation fails:
>   Only report failure to create sockets if none of the
>   provided addrinfo list elements allows creation of a
>   socket, otherwise report failure to find an available port.
> * Further simplify if's within the for loops and make sure
>   failure to recreate a socket gets distinctively reported.
> * Rebased to current master.
> 
> Changes from v4:
> * Move the complexity of recreating a socket and setting the error pointer
>   into the main for loop, eliminating the try_bind_listen() function
>   again. Cleaning up and improving error handling in the process.
> 
> Changes from v3:
> * Test changes: Add missing license
>   Add subtests for ipv4, ipv6 and both
>   Various g_* usage improvements
> * Split patch into 3 patches with two refactoring patches ahead
>   of the actual fix.
> 
> Changes from v2:
> * Non-trivial rebase + further abstraction
>   on top of 7ad9af343c7f1c70c8015c7c519c312d8c5f9fa1
>   'tests: add functional test validating ipv4/ipv6 address flag handling'
> 
> Changes from v1:
> * Fix potential uninitialized variable only detected by optimize.
> * Improve unexpected error detection in test-listen to give more
>   details about why the test fails unexpectedly.
> * Fix some line length style issues.
> 
> Thanks,
> Knut
> 
> Knut Omang (4):
>   tests: Add test-listen - a stress test for QEMU socket listen
>   sockets: factor out a new try_bind() function
>   sockets: factor out create_fast_reuse_socket
>   sockets: Handle race condition between binds to the same port
> 
>  tests/Makefile.include |   2 +-
>  tests/test-listen.c    | 253 ++++++++++++++++++++++++++++++++++++++++++-
>  util/qemu-sockets.c    | 138 +++++++++++++++--------
>  3 files changed, 345 insertions(+), 48 deletions(-)
>  create mode 100644 tests/test-listen.c
> 
> base-commit: a588c4985eff363154d65aee8607d0a4601655f7

Thanks, queued for 2.11

Regards,
Daniel
-- 
|: https://berrange.com      -o-    https://www.flickr.com/photos/dberrange :|
|: https://libvirt.org         -o-            https://fstop138.berrange.com :|
|: https://entangle-photo.org    -o-    https://www.instagram.com/dberrange :|