Package jdk.incubator.foreign

Classes to support low-level and efficient foreign memory/function access, directly from Java.

Foreign memory access

The key abstractions introduced to support foreign memory access are MemorySegment and MemoryAddress. The first models a contiguous memory region, which can reside either inside or outside the Java heap; the latter models an address - which also can reside either inside or outside the Java heap (and can sometimes be expressed as an offset into a given segment). A memory segment represents the main access coordinate of a memory access var handle, which can be obtained using the combinator methods defined in the MemoryHandles class; a set of common dereference operations is provided also by the MemoryAccess class, which can be useful for simple, non-structured access. Finally, the MemoryLayout class hierarchy enables description of memory layouts and basic operations such as computing the size in bytes of a given layout, obtain its alignment requirements, and so on. Memory layouts also provide an alternate, more abstract way, to produce memory access var handles, e.g. using layout paths. For example, to allocate an off-heap memory region big enough to hold 10 values of the primitive type int, and fill it with values ranging from 0 to 9, we can use the following code:

MemorySegment segment = MemorySegment.allocateNative(10 * 4, ResourceScope.newImplicitScope());
for (int i = 0 ; i < 10 ; i++) {
   MemoryAccess.setIntAtIndex(segment, i, 42);
}
 

Here create a native memory segment, that is, a memory segment backed by off-heap memory; the size of the segment is 40 bytes, enough to store 10 values of the primitive type int. Inside a loop, we then initialize the contents of the memory segment using the MemoryAccess.setIntAtIndex(jdk.incubator.foreign.MemorySegment, long, int) helper method; more specifically, if we view the memory segment as a set of 10 adjacent slots, s[i], where 0 <= i < 10, where the size of each slot is exactly 4 bytes, the initialization logic above will set each slot so that s[i] = i, again where 0 <= i < 10.

Deterministic deallocation

When writing code that manipulates memory segments, especially if backed by memory which resides outside the Java heap, it is often crucial that the resources associated with a memory segment are released when the segment is no longer in use, and in a timely fashion. For this reason, there might be cases where waiting for the garbage collector to determine that a segment is unreachable is not optimal. Clients that operate under these assumptions might want to programmatically release the memory associated with a memory segment. This can be done, using the ResourceScope abstraction, as shown below:

try (ResourceScope scope = ResourceScope.newConfinedScope()) {
    MemorySegment segment = MemorySegment.allocateNative(10 * 4, scope);
    for (int i = 0 ; i < 10 ; i++) {
        MemoryAccess.setIntAtIndex(segment, i, 42);
    }
}
 

This example is almost identical to the prior one; this time we first create a so called resource scope, which is used to bind the life-cycle of the segment created immediately afterwards. Note the use of the try-with-resources construct: this idiom ensures that all the memory resources associated with the segment will be released at the end of the block, according to the semantics described in Section 14.20.3 of The Java Language Specification.

Safety

This API provides strong safety guarantees when it comes to memory access. First, when dereferencing a memory segment, the access coordinates are validated (upon access), to make sure that access does not occur at an address which resides outside the boundaries of the memory segment used by the dereference operation. We call this guarantee spatial safety; in other words, access to memory segments is bounds-checked, in the same way as array access is, as described in Section 15.10.4 of The Java Language Specification.

Since memory segments can be closed (see above), segments are also validated (upon access) to make sure that the resource scope associated with the segment being accessed has not been closed prematurely. We call this guarantee temporal safety. Together, spatial and temporal safety ensure that each memory access operation either succeeds - and accesses a valid memory location - or fails.

Foreign function access

The key abstractions introduced to support foreign function access are SymbolLookup and CLinker. The former is used to lookup symbols inside native libraries; the latter provides linking capabilities which allow to model foreign functions as MethodHandle instances, so that clients can perform foreign function calls directly in Java, without the need for intermediate layers of native code (as it's the case with the Java Native Interface (JNI)).

For example, to compute the length of a string using the C standard library function strlen on a Linux x64 platform, we can use the following code:

      MethodHandle strlen = CLinker.getInstance().downcallHandle(
        CLinker.systemLookup().lookup("strlen").get(),
        MethodType.methodType(long.class, MemoryAddress.class),
        FunctionDescriptor.of(CLinker.C_LONG, CLinker.C_POINTER)
      );

      try (var scope = ResourceScope.newConfinedScope()) {
         var cString = CLinker.toCString("Hello", scope);
         long len = (long)strlen.invokeExact(cString.address()); // 5
      }
 

Here, we lookup the strlen symbol in the system lookup. Then, we obtain a linker instance (see CLinker.getInstance()) and we use it to obtain a method handle which targets the strlen library symbol. To complete the linking successfully, we must provide (i) a MethodType instance, describing the type of the resulting method handle and (ii) a FunctionDescriptor instance, describing the signature of the strlen function. From this information, the linker will uniquely determine the sequence of steps which will turn the method handle invocation (here performed using MethodHandle.invokeExact(java.lang.Object...)) into a foreign function call, according to the rules specified by the platform C ABI. The CLinker class also provides many useful methods for interacting with native code, such as converting Java strings into native strings and viceversa (see CLinker.toCString(java.lang.String, ResourceScope) and CLinker.toJavaString(jdk.incubator.foreign.MemorySegment), respectively), as demonstrated in the above example.

Foreign addresses

When a memory segment is created from Java code, the segment properties (spatial bounds, temporal bounds and confinement) are fully known at segment creation. But when interacting with native libraries, clients will often receive raw pointers; such pointers have no spatial bounds (example: does the C type char* refer to a single char value, or an array of char values, of given size?), no notion of temporal bounds, nor thread-confinement.

When clients receive a MemoryAddress instance from a foreign function call, it might be necessary to obtain a MemorySegment instance to dereference the memory pointed to by that address. To do that, clients can proceed in three different ways, described below.

First, if the memory address is known to belong to a segment the client already owns, a rebase operation can be performed; in other words, the client can ask the address what its offset relative to a given segment is, and, then, proceed to dereference the original segment accordingly, as follows:

MemorySegment segment = MemorySegment.allocateNative(100, scope);
...
MemoryAddress addr = ... //obtain address from native code
int x = MemoryAccess.getIntAtOffset(segment, addr.segmentOffset(segment));
 

Secondly, if the client does not have a segment which contains a given memory address, it can create one unsafely, using the MemoryAddress.asSegment(long, ResourceScope) factory. This allows the client to inject extra knowledge about spatial bounds which might, for instance, be available in the documentation of the foreign function which produced the native address. Here is how an unsafe segment can be created from a native address:

ResourceScope scope = ... // initialize a resource scope object
MemoryAddress addr = ... //obtain address from native code
MemorySegment segment = addr.asSegment(4, scope); // segment is 4 bytes long
int x = MemoryAccess.getInt(segment);
 

Alternatively, the client can fall back to use the so called everything segment - that is, a primordial segment which covers the entire native heap. This segment can be obtained by calling the MemorySegment.globalNativeSegment() method, so that dereference can happen without the need of creating any additional segment instances:

MemoryAddress addr = ... //obtain address from native code
int x = MemoryAccess.getIntAtOffset(MemorySegment.globalNativeSegment(), addr.toRawLongValue());
 

Upcalls

The CLinker interface also allows to turn an existing method handle (which might point to a Java method) into a native memory address (see MemoryAddress), so that Java code can effectively be passed to other foreign functions. For instance, we can write a method that compares two integer values, as follows:

class IntComparator {
    static int intCompare(MemoryAddress addr1, MemoryAddress addr2) {
        return MemoryAccess.getIntAtOffset(MemorySegment.globalNativeSegment(), addr1.toRawLongValue()) -
               MemoryAccess.getIntAtOffset(MemorySegment.globalNativeSegment(), addr2.toRawLongValue());
    }
}
 

The above method dereferences two memory addresses containing an integer value, and performs a simple comparison by returning the difference between such values. We can then obtain a method handle which targets the above static method, as follows:

MethodHandle intCompareHandle = MethodHandles.lookup().findStatic(IntComparator.class,
                                                   "intCompare",
                                                   MethodType.methodType(int.class, MemoryAddress.class, MemoryAddress.class));
 

Now that we have a method handle instance, we can link it into a fresh native memory address, using the CLinker interface, as follows:

ResourceScope scope = ...
MemoryAddress comparFunc = CLinker.getInstance().upcallStub(
     intCompareHandle,
     FunctionDescriptor.of(C_INT, C_POINTER, C_POINTER),
     scope
);
 

As before, we need to provide a FunctionDescriptor instance describing the signature of the function pointer we want to create; as before, this, coupled with the method handle type, uniquely determines the sequence of steps which will allow foreign code to call intCompareHandle according to the rules specified by the platform C ABI. The lifecycle of the memory address returned by CLinker.upcallStub(java.lang.invoke.MethodHandle, jdk.incubator.foreign.FunctionDescriptor, jdk.incubator.foreign.ResourceScope) is tied to the resource scope parameter passed to that method.

Restricted methods

Some methods in this package are considered restricted. Restricted methods are typically used to bind native foreign data and/or functions to first-class Java API elements which can then be used directly by clients. For instance the restricted method MemoryAddress.asSegment(long, ResourceScope) can be used to create a fresh segment with given spatial bounds out of a native address.

Binding foreign data and/or functions is generally unsafe and, if done incorrectly, can result in VM crashes, or memory corruption when the bound Java API element is accessed. For instance, in the case of MemoryAddress.asSegment(long, ResourceScope), if the provided spatial bounds are incorrect, a client of the segment returned by that method might crash the VM, or corrupt memory when attempting to dereference said segment. For these reasons, it is crucial for code that calls a restricted method to never pass arguments that might cause incorrect binding of foreign data and/or functions to a Java API.

Access to restricted methods is disabled by default; to enable restricted methods, the command line option --enable-native-access must mention the name of the caller's module.