Module java.base

Interface Linker


public sealed interface Linker
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 which 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; and
  • A linker allows foreign functions to call Java method handles, via the generation of upcall stubs.
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...) 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 modelled 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 modelled 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 modelled 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 modelled 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 modelled 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 modelled 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 modelled in Linux/x64 (all the examples provided here will assume these platform-dependent mappings):

Mapping C types
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 operate on a subset of memory 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).
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. 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 which 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 of given size, and returns a pointer to that region of memory, which is later deallocated using another function from the C standard library:
void free(void *ptr);
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 a fresh scope that is always alive. 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), 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 which created the confined arena), and its lifetime is tied to the surrounding try-with-resources block.

Variadic functions

Variadic functions are C functions which 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()) {
     int res = (int)printf.invokeExact(arena.allocateFrom("%d plus %d equals %d"), 2, 2, 4); //prints "2 plus 2 equals 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 completes. 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.

Since:
22
Implementation Requirements:
Implementations of this interface are immutable, thread-safe and value-based.
  • Method Details

    • nativeLinker

      static Linker nativeLinker()
      Returns a linker for the ABI associated with the underlying native platform. The underlying native platform is the combination of OS and processor where the Java runtime is currently executing.
      Returns:
      a linker for the ABI associated with the underlying native platform
      API Note:
      It is not currently possible to obtain a linker for a different combination of OS and processor.
      Implementation Requirements:
      A native linker implementation is guaranteed to provide canonical layouts for basic C types.
      Implementation Note:
      The libraries exposed by the default lookup associated with the returned linker are the native libraries loaded in the process where the Java runtime is currently executing. For example, on Linux, these libraries typically include libc, libm and libdl.
    • downcallHandle

      MethodHandle downcallHandle(MemorySegment address, FunctionDescriptor function, Linker.Option... options)
      Creates a method handle which is used to call a foreign function with the given signature and address.

      Calling this method is equivalent to the following code:

      linker.downcallHandle(function).bindTo(symbol);
      

      This method is restricted. Restricted methods are unsafe, and, if used incorrectly, their use might crash the JVM or, worse, silently result in memory corruption. Thus, clients should refrain from depending on restricted methods, and use safe and supported functionalities, where possible.

      Parameters:
      address - the native memory segment whose base address is the address of the target foreign function.
      function - the function descriptor of the target foreign function.
      options - the linker options associated with this linkage request.
      Returns:
      a downcall method handle.
      Throws:
      IllegalArgumentException - if the provided function descriptor is not supported by this linker.
      IllegalArgumentException - if !address.isNative(), or if address.equals(MemorySegment.NULL).
      IllegalArgumentException - if an invalid combination of linker options is given.
      IllegalCallerException - If the caller is in a module that does not have native access enabled.
      See Also:
    • downcallHandle

      MethodHandle downcallHandle(FunctionDescriptor function, Linker.Option... options)
      Creates a method handle which is used to call a foreign function with the given signature.

      The Java method type associated with the returned method handle is derived from the argument and return layouts in the function descriptor, but features an additional leading parameter of type MemorySegment, from which the address of the target foreign function is derived. Moreover, if the function descriptor's return layout is a group layout, the resulting downcall method handle accepts an additional leading parameter of type SegmentAllocator, which is used by the linker runtime to allocate the memory region associated with the struct returned by the downcall method handle.

      Upon invoking a downcall method handle, the linker provides the following guarantees for any argument A of type MemorySegment whose corresponding layout is an address layout:

      • A.scope().isAlive() == true. Otherwise, the invocation throws IllegalStateException;
      • The invocation occurs in a thread T such that A.isAccessibleBy(T) == true. Otherwise, the invocation throws WrongThreadException; and
      • A is kept alive during the invocation. For instance, if A has been obtained using a shared arena, any attempt to close the arena while the downcall method handle is still executing will result in an IllegalStateException.

      Moreover, if the provided function descriptor's return layout is an address layout, invoking the returned method handle will return a native segment associated with a fresh scope that is always alive. Under normal conditions, the size of the returned segment is 0. However, if the function descriptor's return layout has a target layout T, then the size of the returned segment is set to T.byteSize().

      The returned method handle will throw an IllegalArgumentException if the MemorySegment representing the target address of the foreign function is the MemorySegment.NULL address. The returned method handle will additionally throw NullPointerException if any argument passed to it is null.

      This method is restricted. Restricted methods are unsafe, and, if used incorrectly, their use might crash the JVM or, worse, silently result in memory corruption. Thus, clients should refrain from depending on restricted methods, and use safe and supported functionalities, where possible.

      Parameters:
      function - the function descriptor of the target foreign function.
      options - the linker options associated with this linkage request.
      Returns:
      a downcall method handle.
      Throws:
      IllegalArgumentException - if the provided function descriptor is not supported by this linker.
      IllegalArgumentException - if an invalid combination of linker options is given.
      IllegalCallerException - If the caller is in a module that does not have native access enabled.
    • upcallStub

      MemorySegment upcallStub(MethodHandle target, FunctionDescriptor function, Arena arena, Linker.Option... options)
      Creates an upcall stub which can be passed to other foreign functions as a function pointer, associated with the given arena. Calling such a function pointer from foreign code will result in the execution of the provided method handle.

      The returned memory segment's address points to the newly allocated upcall stub, and is associated with the provided arena. As such, the lifetime of the returned upcall stub segment is controlled by the provided arena. For instance, if the provided arena is a confined arena, the returned upcall stub segment will be deallocated when the provided confined arena is closed.

      An upcall stub argument whose corresponding layout is an address layout is a native segment associated with a fresh scope that is always alive. Under normal conditions, the size of this segment argument is 0. However, if the address layout has a target layout T, then the size of the segment argument is set to T.byteSize().

      The target method handle should not throw any exceptions. If the target method handle does throw an exception, the JVM will terminate abruptly. To avoid this, clients should wrap the code in the target method handle in a try/catch block to catch any unexpected exceptions. This can be done using the MethodHandles.catchException(MethodHandle, Class, MethodHandle) method handle combinator, and handle exceptions as desired in the corresponding catch block.

      This method is restricted. Restricted methods are unsafe, and, if used incorrectly, their use might crash the JVM or, worse, silently result in memory corruption. Thus, clients should refrain from depending on restricted methods, and use safe and supported functionalities, where possible.

      Parameters:
      target - the target method handle.
      function - the upcall stub function descriptor.
      arena - the arena associated with the returned upcall stub segment.
      options - the linker options associated with this linkage request.
      Returns:
      a zero-length segment whose address is the address of the upcall stub.
      Throws:
      IllegalArgumentException - if the provided function descriptor is not supported by this linker.
      IllegalArgumentException - if the type of target is incompatible with the type derived from function.
      IllegalArgumentException - if it is determined that the target method handle can throw an exception.
      IllegalStateException - if arena.scope().isAlive() == false
      WrongThreadException - if arena is a confined arena, and this method is called from a thread T, other than the arena's owner thread.
      IllegalCallerException - If the caller is in a module that does not have native access enabled.
    • defaultLookup

      SymbolLookup defaultLookup()
      Returns a symbol lookup for symbols in a set of commonly used libraries.

      Each Linker is responsible for choosing libraries that are widely recognized as useful on the OS and processor combination supported by the Linker. Accordingly, the precise set of symbols exposed by the symbol lookup is unspecified; it varies from one Linker to another.

      Returns:
      a symbol lookup for symbols in a set of commonly used libraries.
      Implementation Note:
      It is strongly recommended that the result of defaultLookup() exposes a set of symbols that is stable over time. Clients of defaultLookup() are likely to fail if a symbol that was previously exposed by the symbol lookup is no longer exposed.

      If an implementer provides Linker implementations for multiple OS and processor combinations, then it is strongly recommended that the result of defaultLookup() exposes, as much as possible, a consistent set of symbols across all the OS and processor combinations.

    • canonicalLayouts

      Map<String,MemoryLayout> canonicalLayouts()
      Returns an unmodifiable mapping between the names of data types used by the ABI implemented by this linker and their canonical layouts.

      Each Linker is responsible for choosing the data types that are widely recognized as useful on the OS and processor combination supported by the Linker. Accordingly, the precise set of data type names and canonical layouts exposed by the linker is unspecified; it varies from one Linker to another.

      Returns:
      an unmodifiable mapping between the names of data types used by the ABI implemented by this linker and their canonical layouts
      Implementation Note:
      It is strongly recommended that the result of canonicalLayouts() exposes a set of symbols that is stable over time. Clients of canonicalLayouts() are likely to fail if a data type that was previously exposed by the linker is no longer exposed, or if its canonical layout is updated.

      If an implementer provides Linker implementations for multiple OS and processor combinations, then it is strongly recommended that the result of canonicalLayouts() exposes, as much as possible, a consistent set of symbols across all the OS and processor combinations.