EHOSTUNREACH should not be considered as an exception when send on a loopback network interface.

[Chris Hegarty]

On 07/12/2011 11:12, Alan Bateman wrote:
> On 07/12/2011 08:11, Charles Lee wrote:
>> :
>>
>> I'd like to raise this issue again. The patch is on the [1]:
>>
>> When a loopback network interface is bound to sendto and connect, in
>> some linuxes it will throw an EINVAL errno, in other linuxes (AIX,
>> iSeries) it will throw an EHOSTUNREACH errno. The man page of sendto
>> on Aix and solaris is [2][3].
>> In such situation, EHOSTUNREACH should treat as the same as EINVAL:
>> not throw an exception but return false.
>>
>> Below is the simple test case can test this situation in specific
>> platforms:
>>
>> /public class IsReachableTest {
>> public static void main(String[] args) throws Exception{
>> InetAddress testHost1 = InetAddress.getByName("openjdk.java.net");
>> NetworkInterface loopbackNi = null;
>> Enumeration<NetworkInterface> list =
>> NetworkInterface.getNetworkInterfaces();
>> boolean loopbackFound = false;
>> while (list.hasMoreElements() && !loopbackFound){
>> loopbackNi = list.nextElement();
>> if (loopbackNi.isLoopback() && loopbackNi.isUp()) {
>> loopbackFound = true;
>> }
>> }
>>
>> if (!loopbackFound)
>> return;
>> if (testHost1.isReachable(loopbackNi, 64, 1000))
>> System.out.println(String.format("Fail: external host '%s' is
>> reachable via loopback '%s'!",
>> testHost1, loopbackNi));
>> }
>> }/
>>
>> Any thoughts, guys?
> APIs that return a boolean to indicate success/failure and can also
> throw IOException when there is a failure can be awkward as there will
> be cases where IOException is thrown when false might seem to be more
> correct. In this case, if sendto can fail with EHOSTUNREACH on Linux
> then the proposed patch looks okay to me. I'm kinda surprised that
> EINVAL is returned in some cases, that seems misleading to me.

Ugh, I remember this one. Not pretty, but we found not better 
alternative, see
   http://bugs.sun.com/bugdatabase/view_bug.do?bug_id=6947677

I filed a new CR for this issue:
   7118907: "InetAddress.isReachable() should return false if sendto 
fails with EHOSTUNREACH"

I approved the proposed changes.

Trivially can you put a new line in the comment after EHOSTUNREACH.

   /*
   * On some Linuxes, when bound to the loopback interface, sendto
   * will fail and errno will be set to EINVAL or EHOSTUNREACH.
   * When that happens, don't throw an exception, just return false.
   */

I assume Neil can push this for you? If not, I can help with the push.

-Chris.

>
> -Alan
>
>
>