How do I create a file through JNA

Question

In the java 25 move, we're removing some code that uses reflection in a soon-to-be-disallowed way. Instead, we're looking at using JNA. I need to access a file descriptor (the unix kind) and I thought I'd open("some_path",O_CREAT|O_RDWR) , do whatever I need to do with the file descriptor and then open the same path with File instead. Problem is I get errno 2 (no such file or directory).
If I create the file first with through

var f = new File("some_path");
f.createNewFile();
int fd = JNALib.open("some_path", O_CREAT|O_RDWR);

Then that works perfectly. I do not understand why the first option fails. Mac OS X, ARM Mac. JNA 5.18.1

Code:

 @Test
 public void testSOOpenFile() throws Exception {
 String tmpFile = "dummy-file.tst";
 var relPath = Paths.get("").toAbsolutePath();
 Path file = Path.of(relPath + "/" + tmpFile);
 System.out.println("Creating " + file);
 int fd = JNALibMinimal.open(file.toString(), O_CREAT | O_RDWR);
 if (fd < 0 ) {
 System.out.println("JNALib failed to open with error code " + JNALib.getLastError());
 } else {
 System.out.println("JNALib successfully opened " + fd);
 }
 }

public class JNALibMinimal {
 static {
 Native.register(Platform.C_LIBRARY_NAME);
 }
 public static native int open(String pathName, int flags);
 public static native int open(String pathName, int flags, int mode);
}

UPDATE I (finally) figured out the problem and it was (not uncommon) CBtK (crap behind the keyboard). I've been too long away from c programming. The problem was that linux and Darwin actually has different values for the symbols O_RDWR & O_CREAT. Once I used Darwin values, the file was created ok.

• g00se: If I create the file first with through... That's because a File is really more of a wrapper around the business of manipulating paths on the file system. Creating an object of type File doesn't create anything on the file system
• Holger: When you’re migrating to Java 25, you may consider switching to the built-in facility for invoking native functions instead.
• aled: Is java.nio.file.Files.setPosixFilePermissions() not enough for your needs?
• aled: Also please add a minimal reproducible example to the question.
• Holger: So, fileChannel.position(intendedSize - 1).write(ByteBuffer.allocate(1)) does not work for you?

Answer 1

You forget the third argument: the file mode (permissions) are required when creating a file

import java.io.File;
File file = new File("some/complex/path/to/my_file.dat");
File parentDir = file.getParentFile();
if (parentDir != null && !parentDir.exists()) {
 parentDir.mkdirs();
}
int fd = JNALib.INSTANCE.open(
 file.getAbsolutePath(),
 JNALib.O_CREAT | JNALib.O_RDWR, 
 JNALib.DEFAULT_MODE
);

• Erik: For the libc open() call, file permissions are optional.
• Daniel Widdis: The permissions are optional unless you're creating a new file. From the man page: "The oflag argument may indicate that the file is to be created if it does not exist (by specifying the O_CREAT flag). In this case, open requires a third argument mode_t mode". Max is correct here (and this answer could be improved by quoting this).
• Erik: Since the c code works without using perms, no.
• Erik: Want to clarify here, using O_CREAT without mode will grab the next 4 bytes in the frame as mode bytes so you'll get what is a random mode. I didn't use explicit mode because I wanted to simplify things.
• Daniel Widdis: So is it possible the "random" mode was 0 = no file access to even the user, thus resulting in the "no such file" error? In any case, I would expect "working" here means "giving nondeterministic results". :)