RFR (S): 8218939: vm/mlvm/anonloader/stress/byteMutation crashed on windows

[Igor Ignatyev]

> On Feb 14, 2019, at 5:26 PM, David Holmes <david.holmes at oracle.com> wrote:
> 
> On 15/02/2019 10:45 am, Igor Ignatyev wrote:
>> I'm sorry I wasn't clear enough, I'm suggesting to harden runtime/classFileParserBug/TestBadClassName.java test, not vm/mlvm/anonloader/stress/byteMutation.
>> in TestBadClassName.java, we should always get CFE, so we can assert that it has the expected message.
> 
> Ah! Sure. That would have caught this bug much earlier.
> 
> http://cr.openjdk.java.net/~dholmes/8218939/webrev.v2/ <http://cr.openjdk.java.net/~dholmes/8218939/webrev.v2/>

LGTM.

-- Igor
> 
> Thanks,
> David
> 
>> -- Igor
>>> On Feb 14, 2019, at 4:36 PM, David Holmes <david.holmes at oracle.com> wrote:
>>> 
>>> Hi Igor,
>>> 
>>> Thanks for looking at this.
>>> 
>>> On 15/02/2019 10:20 am, Igor Ignatyev wrote:
>>>> Hi David,
>>>> the fix looks good to me. does it make sense to harden the test and assert the message?
>>> 
>>> In what way would you harden the test? It does random byte writes which could possibly still result in a legal classfile. You'd actually have to control the modifications to trigger specific errors that could be asserted. And I would expect such tests live somewhere in the JCK already.
>>> 
>>> Thanks,
>>> David
>>> -----
>>> 
>>>> -- Igor
>>>>> On Feb 14, 2019, at 2:58 PM, David Holmes <david.holmes at oracle.com> wrote:
>>>>> 
>>>>> webrev: http://cr.openjdk.java.net/~dholmes/8218939/webrev
>>>>> bug: https://bugs.openjdk.java.net/browse/JDK-8218939
>>>>> 
>>>>> The bytes() in a Symbol are UTF8 but do not form a NUL-terminated C-string. In a couple of places in the classfile parser we are treating it as NUL-terminated:
>>>>> 
>>>>> - in name validation we were calling strchr to find the semi-colon, but this could run off through memory if there was no semi-colon (as per the testcase). This is replaced with memchr which takes the expected length.
>>>>> 
>>>>> - in formatting the exception message we used %s but instead we need %.*s and pass the length
>>>>> 
>>>>> There's a minor change to a test to print the exception information as that exposes the fact we were doing things incorrectly e.g:
>>>>> 
>>>>> java.lang.ClassFormatError: Illegal class name "p1//BadInterface1\u00f1\u00ab\u00ab\u00ab\u00ab\u00ab\u00ab\u00ab\u00ab\u00ab\u00ab\u00ab\u00ab\u00ab\u00ab\u00ab\u00abU" in class file
>>>>> 
>>>>> versus (after bug fix)
>>>>> 
>>>>> java.lang.ClassFormatError: Illegal class name "p1//BadInterface1" in class file UseBadInterface1
>>>>> 
>>>>> Testing:
>>>>> - re-ran failing test on Windows with seeds known to have caused failures
>>>>> - ran ran failing test on Windows 150 times with random seeds
>>>>> - tiers 1-3
>>>>> 
>>>>> Thanks,
>>>>> David
>>>>> -----