JDK 8 RFR 8010371: getaddrinfo can fail with EAI_SYSTEM/EAGAIN, causes UnknownHostException to be thrown

[Brian Burkhalter]

On Oct 10, 2013, at 6:52 AM, Alan Bateman wrote:

> Thanks for persevering with this, we are on the right road now.
> 
> I just looked at JDK-8010371 and the "OS" field is set to "linux". I don't know if this is right but as IPv6 is usually enabled by default these days then I have to guess that the person that submitted the bug has IPv6 disabled or is running with -Djava.net.preferIPv4Stack=true, otherwise it would be a non-issue (as you have discovered).
> 
> The other thing about your observation is that ThrowUnknownHostExceptionWithGaiError is creating the UHE with the both the hostname and the error message. This resolves to the concern in one or two of the mails that the UHE names the exception message "host" and that someone might assume that the exception detail is the hostname.
> 
> So thumbs up on the update to src/solaris/native/java/net/Inet4Address.c.

Thanks.
>> For Windows I added a similar NET_ ThrowUnknownHostExceptionWithGaiError and modified Inet{4,6} to mimic the Unix case. Note that the Windows code has not yet been compiled pending comments.
>> 
> Inet4AddressImpl.lookupAllHostAddr uses gethostbyname and when I look at the MSDN documentation then I don't see EAI_AGAIN as possible error. It does list WSATRY_AGAIN and I'm wondering if they have the same value and whether the WSA* error is only returned by WSAGetLastError?

This symbolic name definition

#define EAI_AGAIN       WSATRY_AGAIN

is at line 99 of net_util_mh. which is included by net_util.h hence the two aforementioned values are the same.

> However I think we have a problem with using gai_strerror here as the MSDN documentation says that it is not thread safe. This means we may have to look for a WSA equivalent or maybe drop the Windows part of the fix (using a mutex of some synchronization might be possible but the scope would require research and of course it wouldn't work with 3rd party native code that is also using it).


Yes, I read about the thread safety. There are other ways to do this but I wanted first to air the general idea using gai_strerror() before doing any further work.

> getaddrinfo could return EAI_AGAIN[1]
> 
> But recomended way is to just check return value of getaddrinfo and then
> use WSA functions especially, FormatMessage[2] instead of gai_strerror.

Thanks, I did read about FormatMessage() but forewent the usage thereof for the just stated reason.

> Also we may consider rewriting this code using GetAddrInfoW sometimes in
> a future to support IDN.
> 
> PS:
> Brian, please notice mixing WSA and EAI at ll. 137.

Agreed, that indeed should be changed as EAI_AGAIN -> WSATRY_AGAIN.

Thanks,

Brian