getaddrinfo : Non-recoverable failure in name resolution

Ken Brown kbrown@cornell.edu
Mon Jun 23 16:16:00 GMT 2014


On 6/23/2014 9:07 AM, Marco Atzeri wrote:
>
> On 23/06/2014 14:25, Corinna Vinschen wrote:
>> Hi Marco,
>>
>> On Jun 23 13:41, Marco Atzeri wrote:
>
>>>
>>> 64 $ ./getaddrinfo_test-2_64
>>> getaddrinfo: Non-recoverable failure in name resolution
>>>
>>>
>>> Am I missing something ?
>>> The second way is currently used on postgresql in several places,
>>> but it seems to fail only for "127.0.0.1"
>>
>> I don't know why this only fails for "127.0.0.1".  But this is clearly a
>> problem in the 64 bit Cygwin DLL.
>>
>> What happens is that the field ai_addrlen is defined as socklen_t in
>> POSIX, but as size_t in the W32 API.  On 64 bit, socklen_t is 4 bytes
>> while size_t is 8 bytes.  Setting all the hintp members manually (in
>> contrast to calloc'ing it or memset'ing it to 0) leaves the 4 upper
>> bytes of the ai_addrlen untouched.  This in turn leads to a high
>> probability that ai_addrlen has an invalid value when entering Winsock's
>> getsockopt.
>>
>> I'm really surprised this hasn't been hit before.  I'm going to fix that
>> in Cygwin by setting the upper 4 bytes of ai_addrlen to 0 explicitely.
>
> Probably we have seen already but not identified.

I wonder if this could explain the mysterious emacs crashes that have 
been reported on 64-bit Cygwin.  The emacs code does set the hintp 
members manually in some places.

The crashes seem to occur randomly, and the backtraces often don't make 
sense.

Ken

--
Problem reports:       http://cygwin.com/problems.html
FAQ:                   http://cygwin.com/faq/
Documentation:         http://cygwin.com/docs.html
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple



More information about the Cygwin mailing list