Throw meaningful IOException instances from native code.

The Java side of OSFileSystem was throwing IOExceptions with no detail
message. If we throw from the native side instead, we can supply
meaningful explanations. This turned up a couple of bugs:

* read, readDirect, readv, writev, and ttyRead would only throw IOException
if they returned < -1, which is impossible. (writev was probably a copy & paste
from readv, and the reads were probably confused by the impedence mismatch
between Unix's use of 0 to mean end of file and -1 to mean error, and Java's
use of -1 for end of file.)

* inconsistent checking for null byte[]s passed in.

* read and write would retry on EINTR, but readDirect and writeDirect wouldn't.

* we'd silently truncate seek/lock/truncate offsets that didn't fit in 32 bits;
we now throw an IOException instead.

It also means a few native functions become "void" because errors are now
reported by throwing exceptions, and the Java functions that used to call them
are no longer needed.

Also change ProcessManager to use jniThrowIOException, remove the unused
throwIOExceptionStr from OSNetworkSystem.cpp, and remove the KnownFailure from
FileTest's test_delete, now we have a fixed version of yaffs that won't
rmdir(2) non-empty directories.

Bug: 1542253

1 files changed, 1 insertions, 2 deletions

diff --git a/luni-kernel/src/main/native/java_lang_ProcessManager.c b/luni-kernel/src/main/native/java_lang_ProcessManager.c
index eaefc9f..46e78f5 100644
--- a/luni-kernel/src/main/native/java_lang_ProcessManager.c
+++ b/luni-kernel/src/main/native/java_lang_ProcessManager.c
@@ -64,8 +64,7 @@ static void java_lang_ProcessManager_close(JNIEnv* env,
         jclass clazz, jobject javaDescriptor) {
     int fd = (*env)->GetIntField(env, javaDescriptor, descriptorField);
     if (closeNow(fd) == -1) {
-        jclass ioException = (*env)->FindClass(env, "java/io/IOException");
-        (*env)->ThrowNew(env, ioException, strerror(errno));
+        jniThrowIOException(env, errno);
     }
 }