A linker provides access to foreign functions from Java code, and access to Java code from foreign functions.

Foreign functions typically reside in libraries that can be loaded on demand. Each library conforms to a specific ABI (Application Binary Interface). An ABI is a set of calling conventions and data types associated with the compiler, OS, and processor where the library was built. For example, a C compiler on Linux/x64 usually builds libraries that conform to the SystemV ABI.

A linker has detailed knowledge of the calling conventions and data types used by a specific ABI. For any library that conforms to that ABI, the linker can mediate between Java code running in the JVM and foreign functions in the library. In particular:

• A linker allows Java code to link against foreign functions, via downcall method handles^RESTRICTED; and
• A linker allows foreign functions to call Java method handles, via the generation of upcall stubs^RESTRICTED.

A linker provides a way to look up the canonical layouts associated with the data types used by the ABI. For example, a linker implementing the C ABI might choose to provide a canonical layout for the C size_t type. On 64-bit platforms, this canonical layout might be equal to ValueLayout.JAVA_LONG. The canonical layouts supported by a linker are exposed via the canonicalLayouts() method, which returns a map from type names to canonical layouts.

In addition, a linker provides a way to look up foreign functions in libraries that conform to the ABI. Each linker chooses a set of libraries that are commonly used on the OS and processor combination associated with the ABI. For example, a linker for Linux/x64 might choose two libraries: libc and libm. The functions in these libraries are exposed via a symbol lookup.

Calling native functions

The native linker can be used to link against functions defined in C libraries (native functions). Suppose we wish to downcall from Java to the strlen function defined in the standard C library:

size_t strlen(const char *s);

A downcall method handle that exposes strlen is obtained, using the native linker, as follows:

Linker linker = Linker.nativeLinker();
MethodHandle strlen = linker.downcallHandle(
    linker.defaultLookup().find("strlen").orElseThrow(),
    FunctionDescriptor.of(JAVA_LONG, ADDRESS)
);

Note how the native linker also provides access, via its default lookup, to the native functions defined by the C libraries loaded with the Java runtime. Above, the default lookup is used to search the address of the strlen native function. That address is then passed, along with a platform-dependent description of the signature of the function expressed as a FunctionDescriptor (more on that below) to the native linker's downcallHandle(MemorySegment, FunctionDescriptor, Option...)^RESTRICTED method. The obtained downcall method handle is then invoked as follows:

 try (Arena arena = Arena.ofConfined()) {
     MemorySegment str = arena.allocateFrom("Hello");
     long len = (long) strlen.invokeExact(str);  // 5
 }

Describing C signatures

When interacting with the native linker, clients must provide a platform-dependent description of the signature of the C function they wish to link against. This description, a function descriptor, defines the layouts associated with the parameter types and return type (if any) of the C function.

Scalar C types such as bool, int are modeled as value layouts of a suitable carrier. The mapping between a scalar type and its corresponding canonical layout is dependent on the ABI implemented by the native linker (see below).

Composite types are modeled as group layouts. More specifically, a C struct type maps to a struct layout, whereas a C union type maps to a union layout. When defining a struct or union layout, clients must pay attention to the size and alignment constraint of the corresponding composite type definition in C. For instance, padding between two struct fields must be modeled explicitly, by adding an adequately sized padding layout member to the resulting struct layout.

Finally, pointer types such as int** and int(*)(size_t*, size_t*) are modeled as address layouts. When the spatial bounds of the pointer type are known statically, the address layout can be associated with a target layout. For instance, a pointer that is known to point to a C int[2] array can be modeled as an address layout whose target layout is a sequence layout whose element count is 2, and whose element type is ValueLayout.JAVA_INT.

All native linker implementations are guaranteed to provide canonical layouts for the following set of types:

• bool
• char
• short
• int
• long
• long long
• float
• double
• size_t
• wchar_t
• void*

As noted above, the specific canonical layout associated with each type can vary, depending on the data model supported by a given ABI. For instance, the C type long maps to the layout constant ValueLayout.JAVA_LONG on Linux/x64, but maps to the layout constant ValueLayout.JAVA_INT on Windows/x64. Similarly, the C type size_t maps to the layout constant ValueLayout.JAVA_LONG on 64-bit platforms, but maps to the layout constant ValueLayout.JAVA_INT on 32-bit platforms.

A native linker typically does not provide canonical layouts for C's unsigned integral types. Instead, they are modeled using the canonical layouts associated with their corresponding signed integral types. For instance, the C type unsigned long maps to the layout constant ValueLayout.JAVA_LONG on Linux/x64, but maps to the layout constant ValueLayout.JAVA_INT on Windows/x64.

The following table shows some examples of how C types are modeled in Linux/x64 according to the "System V Application Binary Interface" (all the examples provided here will assume these platform-dependent mappings):

C type	Layout	Java type
bool	ValueLayout.JAVA_BOOLEAN	boolean
char unsigned char	ValueLayout.JAVA_BYTE	byte
short unsigned short	ValueLayout.JAVA_SHORT	short
int unsigned int	ValueLayout.JAVA_INT	int
long unsigned long	ValueLayout.JAVA_LONG	long
long long unsigned long long	ValueLayout.JAVA_LONG	long
float	ValueLayout.JAVA_FLOAT	float
double	ValueLayout.JAVA_DOUBLE	double
size_t	ValueLayout.JAVA_LONG	long
char*, int**, struct Point*	ValueLayout.ADDRESS	MemorySegment
int (*ptr)[10]	ValueLayout.ADDRESS.withTargetLayout( MemoryLayout.sequenceLayout(10, ValueLayout.JAVA_INT) );	MemorySegment
struct Point { int x; long y; };	MemoryLayout.structLayout( ValueLayout.JAVA_INT.withName("x"), MemoryLayout.paddingLayout(32), ValueLayout.JAVA_LONG.withName("y") );	MemorySegment
union Choice { float a; int b; }	MemoryLayout.unionLayout( ValueLayout.JAVA_FLOAT.withName("a"), ValueLayout.JAVA_INT.withName("b") );	MemorySegment

All native linker implementations support a well-defined subset of layouts. More formally, a layout L is supported by a native linker NL if:

• L is a value layout V and V.withoutName() is a canonical layout
• L is a sequence layout S and all the following conditions hold:
  1. the alignment constraint of S is set to its natural alignment, and
  2. S.elementLayout() is a layout supported by NL.
• L is a group layout G and all the following conditions hold:
  1. the alignment constraint of G is set to its natural alignment;
  2. the size of G is a multiple of its alignment constraint;
  3. each member layout in G.memberLayouts() is either a padding layout or a layout supported by NL, and
  4. G does not contain padding other than what is strictly required to align its non-padding layout elements, or to satisfy (2).

Linker implementations may optionally support additional layouts, such as packed struct layouts. A packed struct is a struct in which there is at least one member layout L that has an alignment constraint less strict than its natural alignment. This allows to avoid padding between member layouts, as well as avoiding padding at the end of the struct layout. For example:

// No padding between the 2 element layouts:
MemoryLayout noFieldPadding = MemoryLayout.structLayout(
        ValueLayout.JAVA_INT,
        ValueLayout.JAVA_DOUBLE.withByteAlignment(4));

// No padding at the end of the struct:
MemoryLayout noTrailingPadding = MemoryLayout.structLayout(
        ValueLayout.JAVA_DOUBLE.withByteAlignment(4),
        ValueLayout.JAVA_INT);

A native linker only supports function descriptors whose argument/return layouts are layouts supported by that linker and are not sequence layouts.

Function pointers

Sometimes, it is useful to pass Java code as a function pointer to some native function; this is achieved by using an upcall stub^RESTRICTED. To demonstrate this, let's consider the following function from the C standard library:

void qsort(void *base, size_t nmemb, size_t size,
           int (*compar)(const void *, const void *));

The qsort function can be used to sort the contents of an array, using a custom comparator function which is passed as a function pointer (the compar parameter). To be able to call the qsort function from Java, we must first create a downcall method handle for it, as follows:

Linker linker = Linker.nativeLinker();
MethodHandle qsort = linker.downcallHandle(
    linker.defaultLookup().find("qsort").orElseThrow(),
        FunctionDescriptor.ofVoid(ADDRESS, JAVA_LONG, JAVA_LONG, ADDRESS)
);

As before, we use ValueLayout.JAVA_LONG to map the C type size_t type, and ValueLayout.ADDRESS for both the first pointer parameter (the array pointer) and the last parameter (the function pointer).

To invoke the qsort downcall handle obtained above, we need a function pointer to be passed as the last parameter. That is, we need to create a function pointer out of an existing method handle. First, let's write a Java method that can compare two int elements passed as pointers (i.e. as memory segments):

class Qsort {
    static int qsortCompare(MemorySegment elem1, MemorySegment elem2) {
        return Integer.compare(elem1.get(JAVA_INT, 0), elem2.get(JAVA_INT, 0));
    }
}

Now let's create a method handle for the comparator method defined above:

FunctionDescriptor comparDesc = FunctionDescriptor.of(JAVA_INT,
                                                      ADDRESS.withTargetLayout(JAVA_INT),
                                                      ADDRESS.withTargetLayout(JAVA_INT));
MethodHandle comparHandle = MethodHandles.lookup()
                                         .findStatic(Qsort.class, "qsortCompare",
                                                     comparDesc.toMethodType());

First, we create a function descriptor for the function pointer type. Since we know that the parameters passed to the comparator method will be pointers to elements of a C int[] array, we can specify ValueLayout.JAVA_INT as the target layout for the address layouts of both parameters. This will allow the comparator method to access the contents of the array elements to be compared. We then turn that function descriptor into a suitable method type which we then use to look up the comparator method handle. We can now create an upcall stub that points to that method, and pass it, as a function pointer, to the qsort downcall handle, as follows:

try (Arena arena = Arena.ofConfined()) {
    MemorySegment comparFunc = linker.upcallStub(comparHandle, comparDesc, arena);
    MemorySegment array = arena.allocateFrom(JAVA_INT, 0, 9, 3, 4, 6, 5, 1, 8, 2, 7);
    qsort.invokeExact(array, 10L, 4L, comparFunc);
    int[] sorted = array.toArray(JAVA_INT); // [ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9 ]
}

This code creates an off-heap array, copies the contents of a Java array into it, and then passes the array to the qsort method handle along with the comparator function we obtained from the native linker. After the invocation, the contents of the off-heap array will be sorted according to our comparator function, written in Java. We then extract a new Java array from the segment, which contains the sorted elements.

Functions returning pointers

When interacting with native functions, it is common for those functions to allocate a region of memory and return a pointer to that region. Let's consider the following function from the C standard library:

void *malloc(size_t size);

The malloc function allocates a region of memory with the given size, and returns a pointer to that region of memory, which is later deallocated using another function from the C standard library:

The free function takes a pointer to a region of memory and deallocates that region. In this section we will show how to interact with these native functions, with the aim of providing a safe allocation API (the approach outlined below can of course be generalized to allocation functions other than malloc and free).

First, we need to create the downcall method handles for malloc and free, as follows:

Linker linker = Linker.nativeLinker();

MethodHandle malloc = linker.downcallHandle(
    linker.defaultLookup().find("malloc").orElseThrow(),
    FunctionDescriptor.of(ADDRESS, JAVA_LONG)
);

MethodHandle free = linker.downcallHandle(
    linker.defaultLookup().find("free").orElseThrow(),
    FunctionDescriptor.ofVoid(ADDRESS)
);

When a native function returning a pointer (such as malloc) is invoked using a downcall method handle, the Java runtime has no insight into the size or the lifetime of the returned pointer. Consider the following code:

MemorySegment segment = (MemorySegment)malloc.invokeExact(100);

The size of the segment returned by the malloc downcall method handle is zero. Moreover, the scope of the returned segment is the global scope. To provide safe access to the segment, we must, unsafely, resize the segment to the desired size (100, in this case). It might also be desirable to attach the segment to some existing arena, so that the lifetime of the region of memory backing the segment can be managed automatically, as for any other native segment created directly from Java code. Both of these operations are accomplished using the restricted method MemorySegment.reinterpret(long, Arena, Consumer)^RESTRICTED, as follows:

MemorySegment allocateMemory(long byteSize, Arena arena) throws Throwable {
    MemorySegment segment = (MemorySegment) malloc.invokeExact(byteSize); // size = 0, scope = always alive
    return segment.reinterpret(byteSize, arena, s -> {
        try {
            free.invokeExact(s);
        } catch (Throwable e) {
            throw new RuntimeException(e);
        }
    });  // size = byteSize, scope = arena.scope()
}

The allocateMemory method defined above accepts two parameters: a size and an arena. The method calls the malloc downcall method handle, and unsafely reinterprets the returned segment, by giving it a new size (the size passed to the allocateMemory method) and a new scope (the scope of the provided arena). The method also specifies a cleanup action to be executed when the provided arena is closed. Unsurprisingly, the cleanup action passes the segment to the free downcall method handle, to deallocate the underlying region of memory. We can use the allocateMemory method as follows:

try (Arena arena = Arena.ofConfined()) {
    MemorySegment segment = allocateMemory(100, arena);
} // 'free' called here

Note how the segment obtained from allocateMemory acts as any other segment managed by the confined arena. More specifically, the obtained segment has the desired size, can only be accessed by a single thread (the thread that created the confined arena), and its lifetime is tied to the surrounding try-with-resources block.

Variadic functions

Variadic functions are C functions that can accept a variable number and type of arguments. They are declared with a trailing ellipsis (...) at the end of the formal parameter list, such as: void foo(int x, ...); The arguments passed in place of the ellipsis are called variadic arguments. Variadic functions are, essentially, templates that can be specialized into multiple non-variadic functions by replacing the ... with a list of variadic parameters of a fixed number and type.

It should be noted that values passed as variadic arguments undergo default argument promotion in C. For instance, the following argument promotions are applied:

• _Bool -> unsigned int
• [signed] char -> [signed] int
• [signed] short -> [signed] int
• float -> double

whereby the signed-ness of the source type corresponds to the signed-ness of the promoted type. The complete process of default argument promotion is described in the C specification. In effect, these promotions place limits on the types that can be used to replace the ..., as the variadic parameters of the specialized form of a variadic function will always have a promoted type.

The native linker only supports linking the specialized form of a variadic function. A variadic function in its specialized form can be linked using a function descriptor describing the specialized form. Additionally, the Linker.Option.firstVariadicArg(int) linker option must be provided to indicate the first variadic parameter in the parameter list. The corresponding argument layout (if any), and all following argument layouts in the specialized function descriptor, are called variadic argument layouts.

The native linker does not automatically perform default argument promotions. However, since passing an argument of a non-promoted type as a variadic argument is not supported in C, the native linker will reject an attempt to link a specialized function descriptor with any variadic argument value layouts corresponding to a non-promoted C type. Since the size of the C int type is platform-specific, exactly which layouts will be rejected is platform-specific as well. As an example: on Linux/x64 the layouts corresponding to the C types _Bool, (unsigned) char, (unsigned) short, and float (among others), will be rejected by the linker. The canonicalLayouts() method can be used to find which layout corresponds to a particular C type.

A well-known variadic function is the printf function, defined in the C standard library:

int printf(const char *format, ...);

This function takes a format string, and a number of additional arguments (the number of such arguments is dictated by the format string). Consider the following variadic call:

printf("%d plus %d equals %d", 2, 2, 4);

To perform an equivalent call using a downcall method handle we must create a function descriptor which describes the specialized signature of the C function we want to call. This descriptor must include an additional layout for each variadic argument we intend to provide. In this case, the specialized signature of the C function is (char*, int, int, int) as the format string accepts three integer parameters. We then need to use a linker option to specify the position of the first variadic layout in the provided function descriptor (starting from 0). In this case, since the first parameter is the format string (a non-variadic argument), the first variadic index needs to be set to 1, as follows:

Linker linker = Linker.nativeLinker();
MethodHandle printf = linker.downcallHandle(
    linker.defaultLookup().find("printf").orElseThrow(),
        FunctionDescriptor.of(JAVA_INT, ADDRESS, JAVA_INT, JAVA_INT, JAVA_INT),
        Linker.Option.firstVariadicArg(1) // first int is variadic
);

We can then call the specialized downcall handle as usual:

 try (Arena arena = Arena.ofConfined()) {
     //prints "2 plus 2 equals 4"
     int res = (int)printf.invokeExact(arena.allocateFrom("%d plus %d equals %d"), 2, 2, 4);
 }

Safety considerations

Creating a downcall method handle is intrinsically unsafe. A symbol in a foreign library does not, in general, contain enough signature information (e.g. arity and types of foreign function parameters). As a consequence, the linker runtime cannot validate linkage requests. When a client interacts with a downcall method handle obtained through an invalid linkage request (e.g. by specifying a function descriptor featuring too many argument layouts), the result of such interaction is unspecified and can lead to JVM crashes.

When an upcall stub is passed to a foreign function, a JVM crash might occur, if the foreign code casts the function pointer associated with the upcall stub to a type that is incompatible with the type of the upcall stub, and then attempts to invoke the function through the resulting function pointer. Moreover, if the method handle associated with an upcall stub returns a memory segment, clients must ensure that this address cannot become invalid after the upcall is completed. This can lead to unspecified behavior, and even JVM crashes, since an upcall is typically executed in the context of a downcall method handle invocation.