diff --git a/Makefile b/Makefile
index 2883b28b..c981f8f2 100644
--- a/Makefile
+++ b/Makefile
@@ -5,16 +5,13 @@ MAKEFLAGS += --warn-undefined-variables
 MAKEFLAGS += --no-builtin-rules
 
 .PHONY: build
-build: filprofiler/_filpreload.so filprofiler/libpymemprofile_api.so build_ext
+build: filprofiler/libpymemprofile_api.so build_ext
 
 .PHONY: build_ext
 build_ext: filprofiler/libpymemprofile_api.so
 	python3.8 setup.py build_ext --inplace
 
-filprofiler/_filpreload.so: filprofiler/_filpreload.c
-	gcc -std=c11 -D_FORTIFY_SOURCE=2 -fasynchronous-unwind-tables -fstack-clash-protection -fstack-protector -Werror=format-security -Werror=implicit-function-declaration -O2 -shared -ldl -g -fPIC -fvisibility=hidden -Wall -o $@ $<
-
-filprofiler/libpymemprofile_api.so: Cargo.lock memapi/Cargo.toml memapi/src/*.rs
+filprofiler/libpymemprofile_api.so: Cargo.lock memapi/Cargo.toml memapi/src/*.rs memapi/src/*.c memapi/build.rs
 	rm -f filprofiler/libymemprofile_api.so
 	cargo build --release
 	cp -f target/release/libpymemprofile_api.so filprofiler/
diff --git a/filprofiler/_script.py b/filprofiler/_script.py
index 5b30b3b8..9d2cf255 100644
--- a/filprofiler/_script.py
+++ b/filprofiler/_script.py
@@ -22,8 +22,8 @@ def stage_1():
     # Route all allocations from Python through malloc() directly:
     environ["PYTHONMALLOC"] = "malloc"
     # Library setup:
-    environ["LD_PRELOAD"] = library_path("_filpreload")
-    environ["FIL_API_LIBRARY"] = library_path("libpymemprofile_api")
+    environ["LD_PRELOAD"] = library_path("libpymemprofile_api")
+
     # Disable multi-threaded backends in various scientific computing libraries
     # (Zarr uses Blosc, NumPy uses BLAS):
     environ["BLOSC_NTHREADS"] = "1"
diff --git a/filprofiler/_tracer.py b/filprofiler/_tracer.py
index f4d4fc7b..0750e882 100644
--- a/filprofiler/_tracer.py
+++ b/filprofiler/_tracer.py
@@ -8,10 +8,9 @@
 
 from ._utils import library_path
 
-# Load with RTLD_GLOBAL so _profiler.so has access to those symbols; explicit
-# linking may be possible but haven't done that yet, oh well.
-pymemprofile = CDLL(library_path("libpymemprofile_api"), mode=RTLD_GLOBAL)
-preload = CDLL(library_path("_filpreload"), mode=RTLD_GLOBAL)
+# Load with RTLD_GLOBAL so _profiler.so has access to those symbols; TODO
+# explicit linking may be possible but haven't done that yet.
+preload = CDLL(library_path("libpymemprofile_api"), mode=RTLD_GLOBAL)
 from . import _profiler
 
 
diff --git a/memapi/build.rs b/memapi/build.rs
index 1e8f9cae..6c6c69d4 100644
--- a/memapi/build.rs
+++ b/memapi/build.rs
@@ -1,6 +1,10 @@
 use cc;
+use std::env;
 
 fn main() {
+    let out_dir = env::var("OUT_DIR").unwrap();
+    println!("cargo:rerun-if-changed=src/filpreload.c");
+    println!("cargo:rustc-link-search=native={}", out_dir);
     cc::Build::new()
         .file("src/filpreload.c")
         .warnings_into_errors(true)
@@ -8,6 +12,9 @@ fn main() {
         .flag("-Wall")
         .flag("-Werror=format-security")
         .flag("-Werror=implicit-function-declaration")
+        .static_flag(true)
+        .shared_flag(false)
+        .flag("-fvisibility=hidden")
+        .cargo_metadata(false)
         .compile("filpreload");
-    println!("cargo:rerun-if-changed=src/filpreload.c");
 }
diff --git a/memapi/src/filpreload.c b/memapi/src/filpreload.c
index ce3d332d..c25b0c91 100644
--- a/memapi/src/filpreload.c
+++ b/memapi/src/filpreload.c
@@ -11,10 +11,6 @@ static void *(*underlying_real_mmap)(void *addr, size_t length, int prot,
                                      int flags, int fd, off_t offset) = 0;
 static void (*underlying_real_free)(void *addr) = 0;
 
-// The internal API we're notifying of allocations:
-static void (*add_allocation_hook)(size_t address, size_t length) = 0;
-static void (*free_allocation_hook)(size_t address) = 0;
-
 // Note whether we've been initialized yet or not:
 static int initialized = 0;
 
@@ -29,23 +25,6 @@ static void __attribute__((constructor)) constructor() {
     fprintf(stderr, "BUG: expected size of size_t and void* to be the same.\n");
     exit(1);
   }
-  void *lib =
-      dlopen(getenv("FIL_API_LIBRARY"), RTLD_NOW | RTLD_DEEPBIND | RTLD_GLOBAL);
-  if (!lib) {
-    fprintf(stderr, "Couldn't load libpymemprofile_api.so library: %s\n",
-            dlerror());
-    exit(1);
-  }
-  add_allocation_hook = dlsym(lib, "pymemprofile_add_allocation");
-  if (!add_allocation_hook) {
-    fprintf(stderr, "Couldn't load pymemprofile API function: %s\n", dlerror());
-    exit(1);
-  }
-  free_allocation_hook = dlsym(lib, "pymemprofile_free_allocation");
-  if (!free_allocation_hook) {
-    fprintf(stderr, "Couldn't load pymemprofile API function: %s\n", dlerror());
-    exit(1);
-  }
   underlying_real_mmap = dlsym(RTLD_NEXT, "mmap");
   if (!underlying_real_mmap) {
     fprintf(stderr, "Couldn't load mmap(): %s\n", dlerror());
@@ -61,10 +40,14 @@ static void __attribute__((constructor)) constructor() {
 
 extern void *__libc_malloc(size_t size);
 extern void *__libc_calloc(size_t nmemb, size_t size);
+
+// The Rust API in lib.rs:
+extern void pymemprofile_add_allocation(size_t address, size_t length);
+extern void pymemprofile_free_allocation(size_t address);
 extern void pymemprofile_start_call(const char *filename, const char *funcname);
 extern void pymemprofile_finish_call();
 extern void pymemprofile_reset();
-extern void pymemprofile_dump_peak_to_flamegraph(const char* path);
+extern void pymemprofile_dump_peak_to_flamegraph(const char *path);
 
 __attribute__((visibility("default"))) void
 fil_start_call(const char *filename, const char *funcname) {
@@ -91,7 +74,8 @@ __attribute__((visibility("default"))) void fil_reset() {
   }
 }
 
-__attribute__((visibility("default"))) void fil_dump_peak_to_flamegraph(const char* path) {
+__attribute__((visibility("default"))) void
+fil_dump_peak_to_flamegraph(const char *path) {
   if (!will_i_be_reentrant) {
     will_i_be_reentrant = 1;
     pymemprofile_dump_peak_to_flamegraph(path);
@@ -104,7 +88,7 @@ __attribute__((visibility("default"))) void *malloc(size_t size) {
   void *result = __libc_malloc(size);
   if (!will_i_be_reentrant && initialized) {
     will_i_be_reentrant = 1;
-    add_allocation_hook((size_t)result, size);
+    pymemprofile_add_allocation((size_t)result, size);
     will_i_be_reentrant = 0;
   }
   return result;
@@ -115,7 +99,7 @@ __attribute__((visibility("default"))) void *calloc(size_t nmemb, size_t size) {
   size_t allocated = nmemb * size;
   if (!will_i_be_reentrant && initialized) {
     will_i_be_reentrant = 1;
-    add_allocation_hook((size_t)result, allocated);
+    pymemprofile_add_allocation((size_t)result, allocated);
     will_i_be_reentrant = 0;
   }
   return result;
@@ -129,7 +113,7 @@ __attribute__((visibility("default"))) void free(void *addr) {
   underlying_real_free(addr);
   if (!will_i_be_reentrant) {
     will_i_be_reentrant = 1;
-    free_allocation_hook((size_t)addr);
+    pymemprofile_free_allocation((size_t)addr);
     will_i_be_reentrant = 0;
   }
 }
diff --git a/memapi/src/lib.rs b/memapi/src/lib.rs
index 13daa75d..1430a30c 100644
--- a/memapi/src/lib.rs
+++ b/memapi/src/lib.rs
@@ -51,5 +51,13 @@ pub unsafe extern "C" fn pymemprofile_dump_peak_to_flamegraph(path: *const c_cha
     memorytracking::dump_peak_to_flamegraph(&path);
 }
 
+// Re-export symbols from filpreload.c
+#[link(name = "filpreload", kind = "static")]
+extern "C" {
+    pub fn malloc(size: libc::size_t) -> *mut libc::c_void;
+    pub fn free(addr: *mut libc::c_void);
+    pub fn fil_dump_peak_to_flamegraph(path: *const c_char);
+}
+
 #[cfg(test)]
 mod tests {}