RFR (XXXS): 8193234: When using -Xcheck:jni an internally allocated buffer can leak

[David Holmes]

Hi Aleksey,

Thanks for looking at this.

On 8/02/2019 1:53 am, Aleksey Shipilev wrote:
> On 2/7/19 11:13 AM, David Holmes wrote:
>> bug: https://bugs.openjdk.java.net/browse/JDK-8193234
>> webrev: http://cr.openjdk.java.net/~dholmes/8193234/webrev
> 
> The patch looks good.
> 
>> Full details in bug report but basically with the use of GuardedMemory in the checked versions of
>> the *ArrayCritical functions we always have to release the GuardedMemory copy.
> 
> Ah, oops. It seems fine to free the copy of guarded memory unconditionally. While you are at it,
> switch statement can also be simplified by coalescing case labels. It also makes the dichotomy
> between "normal" and "abort" paths evident:
> 
>    switch (mode) {
>    case 0:
>    case JNI_COMMIT:
>      memcpy(orig_result, carray, sz);
>      break;
>    case JNI_ABORT:
>      break;
>    default:
>      ...
>    }

The coalescing is only possible because we never uses copies and so 
never have to free them - which is the difference between modes 0 and 
JNI_COMMIT. But I can do that and add a clarifying comment.

Thanks,
David

> -Aleksey
>