Fixing interpose on macOS

include/mimalloc-internal.h

@@ -10,7 +10,7 @@ terms of the MIT license. A copy of the license can be found in the file
 
 #include "mimalloc-types.h"
 
-#if defined(MI_MALLOC_OVERRIDE) && defined(MI_INTERPOSE)
+#if defined(MI_MALLOC_OVERRIDE) && defined(__APPLE__)
 #define MI_TLS_RECURSE_GUARD
 #endif
 

src/alloc-override-osx.c

@@ -21,6 +21,7 @@ terms of the MIT license. A copy of the license can be found in the file
 
 #include <AvailabilityMacros.h>
 #include <malloc/malloc.h>
+#include <string.h>  // memset
 
 #if defined(MAC_OS_X_VERSION_10_6) && \
     MAC_OS_X_VERSION_MAX_ALLOWED >= MAC_OS_X_VERSION_10_6
@@ -65,7 +66,7 @@ static void zone_destroy(malloc_zone_t* zone) {
   // todo: ignore for now?
 }
 
-static size_t zone_batch_malloc(malloc_zone_t* zone, size_t size, void** ps, size_t count) {
+static unsigned zone_batch_malloc(malloc_zone_t* zone, size_t size, void** ps, unsigned count) {
   size_t i;
   for (i = 0; i < count; i++) {
     ps[i] = zone_malloc(zone, size);
@@ -74,7 +75,7 @@ static size_t zone_batch_malloc(malloc_zone_t* zone, size_t size, void** ps, siz
   return i;
 }
 
-static void zone_batch_free(malloc_zone_t* zone, void** ps, size_t count) {
+static void zone_batch_free(malloc_zone_t* zone, void** ps, unsigned count) {
   for(size_t i = 0; i < count; i++) {
     zone_free(zone, ps[i]);
     ps[i] = NULL;
@@ -149,7 +150,7 @@ static malloc_zone_t* mi_get_default_zone()
 {
   // The first returned zone is the real default
   malloc_zone_t** zones = NULL;
-  size_t count = 0;
+  unsigned count = 0;
   kern_return_t ret = malloc_get_all_zones(0, NULL, (vm_address_t**)&zones, &count);
   if (ret == KERN_SUCCESS && count > 0) {
     return zones[0];

src/alloc-override.c

@@ -50,6 +50,9 @@ terms of the MIT license. A copy of the license can be found in the file
     MI_INTERPOSE_MI(malloc),
     MI_INTERPOSE_MI(calloc),
     MI_INTERPOSE_MI(realloc),
+    MI_INTERPOSE_MI(strdup),
+    MI_INTERPOSE_MI(strndup),
+    MI_INTERPOSE_MI(realpath),
     MI_INTERPOSE_MI(free)
   };
 #else
@@ -131,6 +134,12 @@ size_t malloc_size(void* p)              MI_FORWARD1(mi_usable_size,p)
 size_t malloc_usable_size(void *p)       MI_FORWARD1(mi_usable_size,p)
 void   cfree(void* p)                    MI_FORWARD0(mi_free, p)
 
+#ifdef __APPLE__
+char*  strdup(const char* s)             MI_FORWARD1(mi_strdup,s)
+char*  strndup(const char* s, size_t n)  MI_FORWARD2(mi_strndup,s,n)
+char*  realpath(const char* fname, char* resolved_name) MI_FORWARD2(mi_realpath,fname,resolved_name)
+#endif
+
 int posix_memalign(void** p, size_t alignment, size_t size) {
   // TODO: the spec says we should return EINVAL also if alignment is not a power of 2.
   // The spec also dictates we should not modify `*p` on an error. (issue#27)

src/init.c

@@ -102,7 +102,7 @@ mi_heap_t _mi_heap_main = {
   NULL,
   0,
   0,
-  0,
+  0xCDCDCDCDCDCDCDL,
   0,
   false   // can reclaim
 };
@@ -355,11 +355,15 @@ static void mi_process_done(void);
 void mi_process_init(void) mi_attr_noexcept {
   // ensure we are called once
   if (_mi_process_is_initialized) return;
+  // access _mi_heap_default before setting _mi_process_is_initialized to ensure
+  // that the TLS slot is allocated without getting into recursion on macOS
+  // when using dynamic linking with interpose.
+  mi_heap_t* h = _mi_heap_default;
   _mi_process_is_initialized = true;
 
   _mi_heap_main.thread_id = _mi_thread_id();
   _mi_verbose_message("process init: 0x%zx\n", _mi_heap_main.thread_id);
-  uintptr_t random = _mi_random_init(_mi_heap_main.thread_id);
+  uintptr_t random = _mi_random_init(_mi_heap_main.thread_id)  ^ (uintptr_t)h;
   _mi_heap_main.cookie = (uintptr_t)&_mi_heap_main ^ random;
   _mi_heap_main.random = _mi_random_shuffle(random);
   #if (MI_DEBUG)

src/page.c

@@ -98,7 +98,7 @@ bool _mi_page_is_valid(mi_page_t* page) {
   mi_assert_internal(page->cookie != 0);
   if (page->heap!=NULL) {
     mi_segment_t* segment = _mi_page_segment(page);
-    mi_assert_internal(segment->thread_id == page->heap->thread_id);
+    mi_assert_internal(!_mi_process_is_initialized || segment->thread_id == page->heap->thread_id);
     mi_page_queue_t* pq = mi_page_queue_of(page);
     mi_assert_internal(mi_page_queue_contains(pq, page));
     mi_assert_internal(pq->block_size==page->block_size || page->block_size > MI_LARGE_SIZE_MAX || page->flags.in_full);