RFR: 8307185: pkcs11 native libraries make JNI calls into java code while holding GC lock

[Valerie Peng]

On Wed, 19 Jul 2023 12:03:14 GMT, Daniel Jeliński <djelinski at openjdk.org> wrote:

> This patch fixes random deadlocks in PKCS11 decryption and encryption code.
> 
> The deadlocks were caused by object allocation in `ckAssertReturnValueOK` waiting for GC; GC was in turn waiting for `ReleasePrimitiveArrayCritical`, which never happened.
> 
> The fix moves the call to `ckAssertReturnValueOK` after `ReleasePrimitiveArrayCritical`.
> 
> All tests in sun/security/pkcs11 were executed with `JTREG=JAVA_OPTIONS=-Xcheck:jni` with and without this patch.
> Without this patch, 5 tests produced the following warning:
> 
> Warning: Calling other JNI functions in the scope of Get/ReleasePrimitiveArrayCritical or Get/ReleaseStringCritical
> 
> With this patch the warning was not observed.
> 
> Tier2-3 tests are still green.

src/jdk.crypto.cryptoki/share/native/libj2pkcs11/p11_crypt.c line 172:

> 170:                                     &ckEncryptedLen);
> 171: 
> 172:     if (directIn == 0 && inBufP != NULL) {

with this change, inBufP and outBufP should be non-null when directIn ==0 and directOut == 0, right? Perhaps we only need to check for one instead of both.

-------------

PR Review Comment: https://git.openjdk.org/jdk/pull/14931#discussion_r1268400446