Implement resources as references

Following OTP, make resources appear as references to Erlang code.
Also following OTP, make resources comparable and serializable, with the same
properties. For a resource to be properly unserialized, it must still exist
in the VM.

Also fix a bug with reference ticks where the top 32 bits were incorrectly
handled on 32 bits platforms, yielding collisions of `make_ref/0`.

Signed-off-by: Paul Guyot <[email protected]>

Author: pguyot

CHANGELOG.md

@@ -81,6 +81,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - `binary_to_integer/1` no longer accepts binaries such as `<<"0xFF">>` or `<<"  123">>`
 - `binary_to_integer` and `list_to_integer` do not raise anymore `overflow` error, they raise
 instead `badarg`.
+- Resources are now references instead of empty binaries.
 
 ### Fixed
 
@@ -95,6 +96,7 @@ instead `badarg`.
 - Added missing support for supervisor `one_for_all` strategy.
 - Supervisor now honors period and intensity options.
 - Fix supervisor crash if a `one_for_one` child fails to restart.
+- Fix collision in references created with `make_ref/0` on 32 bits platforms.
 
 ## [0.6.7] - Unreleased
 

doc/src/differences-with-beam.md

@@ -217,12 +217,6 @@ locking the VM.
 
 Ports are also executed by the schedulers and should return quickly.
 
-### Resources
-
-AtomVM supports resources but for historical reasons these appear as zero-length binaries as they
-used to before OTP20, and not as references as they currently do with recent versions of the BEAM.
-This has some consequences on matching and serialization.
-
 ### BEAM file compatibility
 
 AtomVM can run BEAM files generated by `erlc` compiler from OTP21 to the latest master version,

src/libAtomVM/context.c

@@ -550,7 +550,7 @@ bool context_get_process_info(Context *ctx, term *out, size_t *term_size, term a
                 if (monitor->monitor_type == CONTEXT_MONITOR_MONITORED_LOCAL) {
                     ret_size += CONS_SIZE;
                 } else if (monitor->monitor_type == CONTEXT_MONITOR_RESOURCE) {
-                    ret_size += CONS_SIZE + TERM_BOXED_RESOURCE_SIZE;
+                    ret_size += CONS_SIZE + TERM_BOXED_REFERENCE_RESOURCE_SIZE;
                 }
             }
             break;

src/libAtomVM/dist_nifs.c

@@ -349,7 +349,7 @@ static term nif_erlang_setnode_3(Context *ctx, int argc, term argv[])
     }
     synclist_unlock(&ctx->global->dist_connections);
 
-    if (UNLIKELY(memory_ensure_free_opt(ctx, TERM_BOXED_RESOURCE_SIZE, MEMORY_CAN_SHRINK) != MEMORY_GC_OK)) {
+    if (UNLIKELY(memory_ensure_free_opt(ctx, TERM_BOXED_REFERENCE_RESOURCE_SIZE, MEMORY_CAN_SHRINK) != MEMORY_GC_OK)) {
         RAISE_ERROR(OUT_OF_MEMORY_ATOM);
     }
     term obj = term_from_resource(conn_obj, &ctx->heap);
@@ -705,7 +705,7 @@ static term dist_get_net_kernel_and_create_connection(struct DistConnection **co
 
 static void dist_net_kernel_send_connect(term net_kernel_proc, struct DistConnection *new_conn_obj, int node_atom_index, Context *ctx)
 {
-    BEGIN_WITH_STACK_HEAP(TUPLE_SIZE(3) + TERM_BOXED_RESOURCE_SIZE, heap)
+    BEGIN_WITH_STACK_HEAP(TUPLE_SIZE(3) + TERM_BOXED_REFERENCE_RESOURCE_SIZE, heap)
     term autoconnect_message = term_alloc_tuple(3, &heap);
     term_put_tuple_element(autoconnect_message, 0, CONNECT_ATOM);
     term_put_tuple_element(autoconnect_message, 1, term_from_atom_index(node_atom_index));

src/libAtomVM/externalterm.c

@@ -32,6 +32,7 @@
 #include "defaultatoms.h"
 #include "memory.h"
 #include "module.h"
+#include "resources.h"
 #include "term.h"
 #include "unicode.h"
 #include "utils.h"
@@ -488,20 +489,38 @@ static int serialize_term(uint8_t *buf, term t, GlobalContext *glb)
             buf[0] = NEWER_REFERENCE_EXT;
         }
         size_t k = 1;
-        uint32_t len = 2;
+        uint32_t len;
+        if (term_is_resource_reference(t)) {
+            len = 4;
+        } else {
+            len = 2;
+        }
         if (!IS_NULL_PTR(buf)) {
             WRITE_16_UNALIGNED(buf + k, len);
         }
         k += 2;
         term node_name = glb->node_name;
         uint32_t creation = node_name == NONODE_AT_NOHOST_ATOM ? 0 : glb->creation;
         k += serialize_term(IS_NULL_PTR(buf) ? NULL : buf + k, node_name, glb);
-        if (!IS_NULL_PTR(buf)) {
-            uint64_t ticks = term_to_ref_ticks(t);
-            WRITE_32_UNALIGNED(buf + k, creation);
-            WRITE_64_UNALIGNED(buf + k + 4, ticks);
+        if (term_is_resource_reference(t)) {
+            if (!IS_NULL_PTR(buf)) {
+                WRITE_32_UNALIGNED(buf + k, creation);
+                struct RefcBinary *refc_binary = term_resource_refc_binary_ptr(t);
+                struct ResourceType *resource_type = refc_binary->resource_type;
+                void *resource = refc_binary->data;
+                uint64_t serialize_ref = resource_serialize(resource, resource_type);
+                WRITE_64_UNALIGNED(buf + k + 4, ((uintptr_t) resource_type));
+                WRITE_64_UNALIGNED(buf + k + 12, ((uintptr_t) serialize_ref));
+            }
+            return k + 20;
+        } else {
+            if (!IS_NULL_PTR(buf)) {
+                uint64_t ticks = term_to_ref_ticks(t);
+                WRITE_32_UNALIGNED(buf + k, creation);
+                WRITE_64_UNALIGNED(buf + k + 4, ticks);
+            }
+            return k + 12;
         }
-        return k + 12;
     } else if (term_is_external_reference(t)) {
         if (!IS_NULL_PTR(buf)) {
             buf[0] = NEWER_REFERENCE_EXT;
@@ -882,6 +901,11 @@ static term parse_external_terms(const uint8_t *external_term_buf, size_t *eterm
                 if (len == 2 && node == this_node && creation == this_creation) {
                     uint64_t ticks = ((uint64_t) data[0]) << 32 | data[1];
                     return term_from_ref_ticks(ticks, heap);
+                } else if (len == 4 && node == this_node && creation == this_creation) {
+                    // This is a resource
+                    uint64_t resource_type_ptr = ((uint64_t) data[0]) << 32 | data[1];
+                    uint64_t resource_serialize_ref = ((uint64_t) data[2]) << 32 | data[3];
+                    return term_from_resource_type_and_serialize_ref(resource_type_ptr, resource_serialize_ref, heap, glb);
                 } else {
                     return term_make_external_reference(node, len, data, creation, heap);
                 }
@@ -1294,8 +1318,12 @@ static int calculate_heap_usage(const uint8_t *external_term_buf, size_t remaini
             }
             if (external_term_buf[3] == SMALL_ATOM_UTF8_EXT) {
                 // Check if it's non-distributed node, in which case it's always a local ref
-                if (len == 2 && external_term_buf[4] == strlen("nonode@nohost") && memcmp(external_term_buf + 5, "nonode@nohost", strlen("nonode@nohost")) == 0) {
-                    heap_size = REF_SIZE;
+                if (external_term_buf[4] == strlen("nonode@nohost") && memcmp(external_term_buf + 5, "nonode@nohost", strlen("nonode@nohost")) == 0) {
+                    if (len == 2) {
+                        heap_size = REF_SIZE;
+                    } else if (len == 4) {
+                        heap_size = TERM_BOXED_REFERENCE_RESOURCE_SIZE;
+                    }
                 }
                 // See above for pids
             } else if (UNLIKELY(external_term_buf[3] != ATOM_EXT)) {

src/libAtomVM/memory.c

@@ -616,6 +616,11 @@ static void memory_scan_and_copy(HeapFragment *old_fragment, term *mem_start, co
 
                     case TERM_BOXED_REF:
                         TRACE("- Found ref.\n");
+                        term ref = ((term) ptr) | TERM_PRIMARY_BOXED;
+                        if (term_is_resource_reference(ref)) {
+                            *mso_list = term_list_init_prepend(ptr + REFERENCE_RESOURCE_CONS_OFFSET, ref, *mso_list);
+                            refc_binary_increment_refcount((struct RefcBinary *) term_resource_refc_binary_ptr(ref));
+                        }
                         break;
 
                     case TERM_BOXED_EXTERNAL_PID:
@@ -953,6 +958,18 @@ void memory_sweep_mso_list(term mso_list, GlobalContext *global, bool from_task)
                 refc_binary_decrement_refcount(term_refc_binary_ptr(h), global);
 #ifdef AVM_TASK_DRIVER_ENABLED
             }
+#endif
+        }
+        if (term_is_resource_reference(h)) {
+            // unreferenced binary; decrement reference count
+#ifdef AVM_TASK_DRIVER_ENABLED
+            if (from_task) {
+                globalcontext_refc_decrement_refcount_from_task(global, term_resource_refc_binary_ptr(h));
+            } else {
+#endif
+                refc_binary_decrement_refcount(term_resource_refc_binary_ptr(h), global);
+#ifdef AVM_TASK_DRIVER_ENABLED
+            }
 #endif
         }
         l = term_get_list_tail(l);

src/libAtomVM/otp_socket.c

@@ -245,7 +245,7 @@ static const AtomStringIntPair otp_socket_setopt_level_table[] = {
 
 static ErlNifResourceType *socket_resource_type;
 
-#define SOCKET_MAKE_SELECT_NOTIFICATION_SIZE (TUPLE_SIZE(4) + REF_SIZE + TUPLE_SIZE(2) + REF_SIZE + TERM_BOXED_RESOURCE_SIZE)
+#define SOCKET_MAKE_SELECT_NOTIFICATION_SIZE (TUPLE_SIZE(4) + REF_SIZE + TUPLE_SIZE(2) + REF_SIZE + TERM_BOXED_REFERENCE_RESOURCE_SIZE)
 static term socket_make_select_notification(struct SocketResource *rsrc_obj, Heap *heap);
 
 //
@@ -632,7 +632,7 @@ static term nif_socket_open(Context *ctx, int argc, term argv[])
 #endif
         rsrc_obj->buf_size = DEFAULT_BUFFER_SIZE;
 
-        if (UNLIKELY(memory_ensure_free(ctx, TERM_BOXED_RESOURCE_SIZE) != MEMORY_GC_OK)) {
+        if (UNLIKELY(memory_ensure_free(ctx, TERM_BOXED_REFERENCE_RESOURCE_SIZE) != MEMORY_GC_OK)) {
             AVM_LOGW(TAG, "Failed to allocate memory: %s:%i.", __FILE__, __LINE__);
             RAISE_ERROR(OUT_OF_MEMORY_ATOM);
         }
@@ -675,7 +675,7 @@ bool term_is_otp_socket(term socket_term)
 {
     bool ret = term_is_tuple(socket_term)
         && term_get_tuple_arity(socket_term) == 2
-        && term_is_binary(term_get_tuple_element(socket_term, 0))
+        && term_is_resource_reference(term_get_tuple_element(socket_term, 0))
         && term_is_local_reference(term_get_tuple_element(socket_term, 1));
 
     TRACE("term is a socket: %i\n", ret);
@@ -1929,7 +1929,7 @@ static term nif_socket_accept(Context *ctx, int argc, term argv[])
             SMP_RWLOCK_UNLOCK(rsrc_obj->socket_lock);
             RAISE_ERROR(OUT_OF_MEMORY_ATOM);
         }
-        size_t requested_size = TERM_BOXED_RESOURCE_SIZE + TUPLE_SIZE(2) + TUPLE_SIZE(2) + REF_SIZE;
+        size_t requested_size = TERM_BOXED_REFERENCE_RESOURCE_SIZE + TUPLE_SIZE(2) + TUPLE_SIZE(2) + REF_SIZE;
         if (UNLIKELY(memory_ensure_free_with_roots(ctx, requested_size, 1, argv, MEMORY_CAN_SHRINK) != MEMORY_GC_OK)) {
             AVM_LOGW(TAG, "Failed to allocate memory: %s:%i.", __FILE__, __LINE__);
             LWIP_END();

src/libAtomVM/otp_ssl.c

@@ -226,7 +226,7 @@ static term nif_ssl_entropy_init(Context *ctx, int argc, term argv[])
     UNUSED(argc);
     UNUSED(argv);
 
-    if (UNLIKELY(memory_ensure_free(ctx, TERM_BOXED_RESOURCE_SIZE) != MEMORY_GC_OK)) {
+    if (UNLIKELY(memory_ensure_free(ctx, TERM_BOXED_REFERENCE_RESOURCE_SIZE) != MEMORY_GC_OK)) {
         AVM_LOGW(TAG, "Failed to allocate memory: %s:%i.", __FILE__, __LINE__);
         RAISE_ERROR(OUT_OF_MEMORY_ATOM);
     }
@@ -249,7 +249,7 @@ static term nif_ssl_ctr_drbg_init(Context *ctx, int argc, term argv[])
     UNUSED(argc);
     UNUSED(argv);
 
-    if (UNLIKELY(memory_ensure_free(ctx, TERM_BOXED_RESOURCE_SIZE) != MEMORY_GC_OK)) {
+    if (UNLIKELY(memory_ensure_free(ctx, TERM_BOXED_REFERENCE_RESOURCE_SIZE) != MEMORY_GC_OK)) {
         AVM_LOGW(TAG, "Failed to allocate memory: %s:%i.", __FILE__, __LINE__);
         RAISE_ERROR(OUT_OF_MEMORY_ATOM);
     }
@@ -301,7 +301,7 @@ static term nif_ssl_init(Context *ctx, int argc, term argv[])
     UNUSED(argc);
     UNUSED(argv);
 
-    if (UNLIKELY(memory_ensure_free(ctx, TERM_BOXED_RESOURCE_SIZE) != MEMORY_GC_OK)) {
+    if (UNLIKELY(memory_ensure_free(ctx, TERM_BOXED_REFERENCE_RESOURCE_SIZE) != MEMORY_GC_OK)) {
         AVM_LOGW(TAG, "Failed to allocate memory: %s:%i.", __FILE__, __LINE__);
         RAISE_ERROR(OUT_OF_MEMORY_ATOM);
     }
@@ -354,7 +354,7 @@ static term nif_ssl_config_init(Context *ctx, int argc, term argv[])
     UNUSED(argc);
     UNUSED(argv);
 
-    if (UNLIKELY(memory_ensure_free(ctx, TERM_BOXED_RESOURCE_SIZE) != MEMORY_GC_OK)) {
+    if (UNLIKELY(memory_ensure_free(ctx, TERM_BOXED_REFERENCE_RESOURCE_SIZE) != MEMORY_GC_OK)) {
         AVM_LOGW(TAG, "Failed to allocate memory: %s:%i.", __FILE__, __LINE__);
         RAISE_ERROR(OUT_OF_MEMORY_ATOM);
     }

src/libAtomVM/posix_nifs.c

@@ -348,7 +348,7 @@ static term nif_atomvm_posix_open(Context *ctx, int argc, term argv[])
         term_put_tuple_element(result, 0, ERROR_ATOM);
         term_put_tuple_element(result, 1, posix_errno_to_term(errno, glb));
     } else {
-        if (UNLIKELY(memory_ensure_free_opt(ctx, TUPLE_SIZE(2) + TERM_BOXED_RESOURCE_SIZE, MEMORY_CAN_SHRINK) != MEMORY_GC_OK)) {
+        if (UNLIKELY(memory_ensure_free_opt(ctx, TUPLE_SIZE(2) + TERM_BOXED_REFERENCE_RESOURCE_SIZE, MEMORY_CAN_SHRINK) != MEMORY_GC_OK)) {
             RAISE_ERROR(OUT_OF_MEMORY_ATOM);
         }
         term obj = make_posix_fd_resource(ctx, fd);
@@ -672,7 +672,7 @@ static term nif_atomvm_subprocess(Context *ctx, int argc, term argv[])
     free_string_list(args);
     free_string_list(envp_array);
 
-    if (UNLIKELY(memory_ensure_free_opt(ctx, TUPLE_SIZE(3) + TERM_BOXED_RESOURCE_SIZE, MEMORY_CAN_SHRINK) != MEMORY_GC_OK)) {
+    if (UNLIKELY(memory_ensure_free_opt(ctx, TUPLE_SIZE(3) + TERM_BOXED_REFERENCE_RESOURCE_SIZE, MEMORY_CAN_SHRINK) != MEMORY_GC_OK)) {
         RAISE_ERROR(OUT_OF_MEMORY_ATOM);
     }
 
@@ -846,7 +846,7 @@ static term nif_atomvm_posix_opendir(Context *ctx, int argc, term argv[])
         }
         dir_obj->dir = dir;
         if (UNLIKELY(memory_ensure_free_opt(
-                         ctx, TUPLE_SIZE(2) + TERM_BOXED_RESOURCE_SIZE, MEMORY_CAN_SHRINK)
+                         ctx, TUPLE_SIZE(2) + TERM_BOXED_REFERENCE_RESOURCE_SIZE, MEMORY_CAN_SHRINK)
                 != MEMORY_GC_OK)) {
             RAISE_ERROR(OUT_OF_MEMORY_ATOM);
         }

src/libAtomVM/refc_binary.c

@@ -66,19 +66,24 @@ void refc_binary_increment_refcount(struct RefcBinary *refc)
 bool refc_binary_decrement_refcount(struct RefcBinary *refc, struct GlobalContext *global)
 {
     if (--refc->ref_count == 0) {
-        if (refc->resource_type && refc->resource_type->down) {
-            // There may be monitors associated with this resource.
-            destroy_resource_monitors(refc, global);
-            // After this point, the resource can no longer be found by
-            // resource_type_fire_monitor
-            // However, resource_type_fire_monitor may have incremented ref_count
-            // to call the monitor handler.
-            // So we check ref_count again. We're not affected by the ABA problem
-            // here as the resource cannot (should not) be monitoring while it is
-            // being destroyed, i.e. no resource monitor will be created now
-            if (UNLIKELY(refc->ref_count != 0)) {
-                return false;
+        if (refc->resource_type) {
+            if (refc->resource_type->down) {
+                // There may be monitors associated with this resource.
+                destroy_resource_monitors(refc, global);
+                // After this point, the resource can no longer be found by
+                // resource_type_fire_monitor
+                // However, resource_type_fire_monitor may have incremented ref_count
+                // to call the monitor handler.
+                // So we check ref_count again. We're not affected by the ABA problem
+                // here as the resource cannot (should not) be monitoring while it is
+                // being destroyed, i.e. no resource monitor will be created now
+                if (UNLIKELY(refc->ref_count != 0)) {
+                    return false;
+                }
             }
+            // Remove the resource from the list of serialized resources
+            // so it no longer can be unserialized.
+            resource_unmark_serialized(refc->data, refc->resource_type);
         }
         synclist_remove(&global->refc_binaries, &refc->head);
         refc_binary_destroy(refc, global);