pcp
[Top] [All Lists]

pcp updates: qa fallout (bug fixes)

To: PCP <pcp@xxxxxxxxxxx>
Subject: pcp updates: qa fallout (bug fixes)
From: Nathan Scott <nathans@xxxxxxxxxx>
Date: Wed, 6 Mar 2013 23:16:27 -0500 (EST)
Delivered-to: pcp@xxxxxxxxxxx
Reply-to: Nathan Scott <nathans@xxxxxxxxxx>
Changes committed to git://oss.sgi.com/pcp/pcp.git dev

 src/include/pcp/impl.h       |    6 +-
 src/libpcp/src/auxconnect.c  |   48 ++++++++++++--------
 src/libpcp/src/auxserver.c   |   99 +++++++++++++++++++++----------------------
 src/libpcp/src/context.c     |    5 ++
 src/libpcp/src/nss_connect.c |   83 +++++++++++++++++++++++++-----------
 src/libpcp/src/pdu.c         |    2 
 6 files changed, 148 insertions(+), 95 deletions(-)

commit 3b9872454486cf0caa7cf4ca3047778e9c4e4264
Author: Nathan Scott <nathans@xxxxxxxxxx>
Date:   Thu Mar 7 15:11:14 2013 +1100

    Resolve compiler warning in ipv6 socket options

commit 75ab98afb0aa4a0a6dddc1bd7b84b9522bb43343
Author: Nathan Scott <nathans@xxxxxxxxxx>
Date:   Thu Mar 7 14:56:00 2013 +1100

    Fix timeout handling for NSS-enabled server builds
    
    Test 131 fails on NSS-enabled builds because we've asked
    for no timeouts on connect/accept.  Requests were being
    handled correctly but not the initial connection setup.
    
    This extends the __pmConvertTimeout implementation to be
    aware of these other timeout settings as well, and makes
    the NSS implementation use it.

commit 10022f3120e2f989e8eeaa9adbc34dc63c5fc373
Author: Nathan Scott <nathans@xxxxxxxxxx>
Date:   Thu Mar 7 14:21:50 2013 +1100

    Fix further issues to get IPv4-only hosts working again
    
    This code has been tested on a kernel booted with the kernel
    flag ipv6.enabled cleared (bootup option), which picked up a
    series of errors from coding assumptions.
    
    The most problematic is the fact that the setsockopt call to
    request a socket be made IPv6-only fails silently in kernels
    booted this way.  This is the cause of the strange diagnostic
    from a (later) bind call, as we will incorrectly attempt to
    bind the same IPv4 port twice then.  This also resulted in a
    cascading error (now fixed too):
    
    Only add the listen file descriptor into the select mask once
    we are certain its setup will succeed (previously we added it
    before all possible error paths were evaluated).  This fixes
    a sigsegv in the server code after the error paths have been
    exercised.
    
    Other small issues resolved in the process:
    - Double-reporting of error message between OpenRequestPorts
    and its helper OpenRequestSocket.
    - For any setsockopts calls that are non-fatal, we now report
    warning and not error level diagnostics.
    - Improved error checking in NSS-enabled socket creation paths
    which lacked an error check on __pmSetDataIPC (although the
    __pmSetSocketIPC following was checking - should be the other
    way around to correctly detect the malloc failure case).

commit 89695674083809ee72e04d6af11066b71084b947
Author: Dave Brolley <brolley@xxxxxxxxxx>
Date:   Thu Mar 7 11:09:47 2013 +1100

    Handle server port setup case where IPv6 is totally disabled

<Prev in Thread] Current Thread [Next in Thread>
  • pcp updates: qa fallout (bug fixes), Nathan Scott <=