RFR [8055421]: Unknown error in UnixNativeDispatcher

[Martin Buchholz]

Of course Ivan's change is a good bug fix.

Meta comments:

Y'all really need to work on common C-level infrastructure.  E.g. I see 15
definitions of RESTARTABLE in the jdk sources.  Maybe the modularization
initiative has made that worse?

Because RESTARTABLE is already a do ... while, it doesn't need a do while
(0) wrapper.
Extra parens around macro variables wouldn't hurt either.

#define RESTARTABLE(_cmd, _result) do {         \
    (_result) = (_cmd);                         \
} while (((_result) == -1) && (errno == EINTR))

It seems worthwhile to do an audit of all the other calls to RESTARTABLE.
 Bugs like this often come in flocks, especially given the copy/paste
nature of these files.  A quick look suggests that there is a serious bug
here:

    RESTARTABLE(dup((int)fd), res);
    if (fd == -1) {
        throwUnixException(env, errno);
    }



On Wed, Aug 20, 2014 at 11:59 PM, Florian Weimer <fweimer at redhat.com> wrote:

> On 08/20/2014 11:24 AM, Ivan Gerasimov wrote:
>
>> BUGURL: https://bugs.openjdk.java.net/browse/JDK-8055421
>>>>>> WEBREV: http://cr.openjdk.java.net/~igerasim/8055421/0/webrev/
>>>>>>
>>>>>
>  It's not clear from the report what are the reason of the observed
>> failure of closedir.
>>
>
> I think the most likely explanation is that closedir succeeded, but errno
> was -1 for some random reason, and the exception was thrown without without
> an actual error.  It's still odd that errno was -1, though, but it can
> obviously happen, it's just some thread-local variable, after all.
>
>
>  However, the bug and fix are quite obvious I think.
>>
>
> Yes, it's clear that the new code is more correct. :-)
>
>
> --
> Florian Weimer / Red Hat Product Security
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.openjdk.java.net/pipermail/nio-dev/attachments/20140821/bfd43ada/attachment-0001.html>