Merge pull request #91 from cloudwu/jemalloc

Use macro for memory hook

Makefile

@@ -20,7 +20,6 @@ $(LUA_STATICLIB) :
 
 JEMALLOC_STATICLIB := 3rd/jemalloc/lib/libjemalloc_pic.a
 JEMALLOC_INC := 3rd/jemalloc/include/jemalloc
-NOBUILDIN_MALLOC := -fno-builtin-malloc -fno-builtin-calloc -fno-builtin-realloc -fno-builtin-free -fno-builtin-memalign
 
 all : jemalloc
 .PHONY : jemalloc
@@ -28,10 +27,10 @@ all : jemalloc
 $(JEMALLOC_STATICLIB) : 3rd/jemalloc/Makefile
 	cd 3rd/jemalloc && $(MAKE) CC=$(CC) 
 
-3rd/jemalloc :
+3rd/jemalloc/autogen.sh :
 	git submodule update --init
 
-3rd/jemalloc/Makefile : | 3rd/jemalloc
+3rd/jemalloc/Makefile : | 3rd/jemalloc/autogen.sh
 	cd 3rd/jemalloc && ./autogen.sh --with-jemalloc-prefix=je_ --disable-valgrind
 
 jemalloc : $(JEMALLOC_STATICLIB)
@@ -52,7 +51,7 @@ all : \
   $(foreach v, $(LUA_CLIB), $(LUA_CLIB_PATH)/$(v).so) 
 
 $(SKYNET_BUILD_PATH)/skynet : $(foreach v, $(SKYNET_SRC), skynet-src/$(v)) $(LUA_LIB) $(JEMALLOC_STATICLIB)
-	$(CC) $(CFLAGS) $(NOBUILDIN_MALLOC) -o $@ $^ -Iskynet-src -I$(JEMALLOC_INC) $(LDFLAGS) $(EXPORT) $(LIBS)
+	$(CC) $(CFLAGS) -o $@ $^ -Iskynet-src -I$(JEMALLOC_INC) $(LDFLAGS) $(EXPORT) $(LIBS)
 
 $(LUA_CLIB_PATH) :
 	mkdir $(LUA_CLIB_PATH)
@@ -77,10 +76,10 @@ $(LUA_CLIB_PATH)/int64.so : 3rd/lua-int64/int64.c | $(LUA_CLIB_PATH)
 	$(CC) $(CFLAGS) $(SHARED) $^ -o $@ 
 
 $(LUA_CLIB_PATH)/bson.so : lualib-src/lua-bson.c | $(LUA_CLIB_PATH)
-	$(CC) $(CFLAGS) $(SHARED) $^ -o $@ 
+	$(CC) $(CFLAGS) $(SHARED) $^ -o $@ -Iskynet-src
 
 $(LUA_CLIB_PATH)/mongo.so : lualib-src/lua-mongo.c | $(LUA_CLIB_PATH)
-	$(CC) $(CFLAGS) $(SHARED) $^ -o $@ 
+	$(CC) $(CFLAGS) $(SHARED) $^ -o $@ -Iskynet-src
 
 $(LUA_CLIB_PATH)/md5.so : 3rd/lua-md5/md5.c 3rd/lua-md5/md5lib.c 3rd/lua-md5/compat-5.2.c | $(LUA_CLIB_PATH)
 	$(CC) $(CFLAGS) $(SHARED) -I3rd/lua-md5 $^ -o $@ 

lualib-src/lua-bson.c

@@ -1,3 +1,6 @@
+// include skynet.h first for malloc hook
+#include "skynet.h"
+
 #include <lua.h>
 #include <lauxlib.h>
 

lualib-src/lua-memory.c

@@ -1,3 +1,6 @@
+// include skynet.h first for malloc hook
+#include "skynet.h"
+
 #include <lua.h>
 #include <lauxlib.h>
 

lualib-src/lua-mongo.c

@@ -1,3 +1,6 @@
+// include skynet.h first for malloc hook
+#include "skynet.h"
+
 #include <lua.h>
 #include <lauxlib.h>
 

lualib-src/lua-netpack.c

@@ -1,3 +1,6 @@
+// include skynet.h first for malloc hook
+#include "skynet.h"
+
 #include "skynet_socket.h"
 
 #include <lua.h>

lualib-src/lua-seri.c

@@ -2,6 +2,9 @@
   https://github.com/cloudwu/lua-serialize
  */
 
+// include skynet.h first for malloc hook
+#include "skynet.h"
+
 #include <lua.h>
 #include <lauxlib.h>
 #include <stdlib.h>

lualib-src/lua-socket.c

@@ -1,3 +1,6 @@
+// include skynet.h first for malloc hook
+#include "skynet.h"
+
 #include <stdlib.h>
 #include <string.h>
 #include <stdbool.h>

service-src/service_harbor.c

@@ -586,14 +586,14 @@ harbor_init(struct harbor *h, struct skynet_context *ctx, const char * args) {
 	char local_addr[sz];
 	int harbor_id = 0;
 	sscanf(args,"%s %s %d",master_addr, local_addr, &harbor_id);
-	h->master_addr = strdup(master_addr);
+	h->master_addr = skynet_strdup(master_addr);
 	h->id = harbor_id;
 	h->master_fd = _connect_to(h, master_addr, true);
 	if (h->master_fd == -1) {
 		fprintf(stderr, "Harbor: Connect to master failed\n");
 		exit(1);
 	}
-	h->local_addr = strdup(local_addr);
+	h->local_addr = skynet_strdup(local_addr);
 
 	_launch_gate(ctx, local_addr);
 	skynet_callback(ctx, h, _mainloop);

service-src/service_snlua.c

@@ -267,7 +267,7 @@ struct snlua *
 snlua_create(void) {
 	struct snlua * l = malloc(sizeof(*l));
 	memset(l,0,sizeof(*l));
-	l->L = luaL_newstate();
+	l->L = lua_newstate(skynet_lalloc, NULL);
 	l->init = _init;
 	return l;
 }

skynet-src/malloc_hook.c

@@ -3,36 +3,37 @@
 #include <assert.h>
 
 #include "malloc_hook.h"
-
 #include "jemalloc.h"
 
 #include "skynet.h"
 
+
 static size_t _used_memory = 0;
 static size_t _memory_block = 0;
 typedef struct _mem_data {
     uint32_t handle;
-    size_t   allocated;
+    ssize_t allocated;
 } mem_data;
 
-#define SLOT_SIZE 0xffff
+#define SLOT_SIZE 0x10000
 #define PREFIX_SIZE sizeof(uint32_t)
 
 static mem_data mem_stats[SLOT_SIZE];
 
-static void _init() {
-    memset(mem_stats, 0, sizeof(mem_stats));
-}
-
 static size_t*
-get_allocated_field(handle) {
-    int h = (handle & SLOT_SIZE) - 1;
+get_allocated_field(uint32_t handle) {
+    int h = (int)(handle & (SLOT_SIZE - 1));
     mem_data *data = &mem_stats[h];
-    if(data->handle == 0 || data->allocated == 0) {
-        __sync_synchronize();
-        if(!__sync_bool_compare_and_swap(&data->handle, 0, handle)) {
-            return 0;
-        }
+	uint32_t old_handle = data->handle;
+	ssize_t old_alloc = data->allocated;
+    if(old_handle == 0 || old_alloc <= 0) {
+		// data->allocated may less than zero, because it may not count at start.
+		if(!__sync_bool_compare_and_swap(&data->handle, old_handle, handle)) {
+			return 0;
+		}
+		if (old_alloc < 0) {
+			__sync_bool_compare_and_swap(&data->allocated, old_alloc, 0);
+		}
     }
     if(data->handle != handle) {
         return 0;
@@ -73,8 +74,7 @@ fill_prefix(char* ptr) {
 
 inline static void*
 clean_prefix(char* ptr) {
-    uint32_t* rawptr = (uint32_t*)ptr - 1;
-    size_t size = je_malloc_usable_size(rawptr);
+    size_t size = je_malloc_usable_size(ptr);
     uint32_t *p = (uint32_t *)(ptr + size - sizeof(uint32_t));
     uint32_t handle;
     memcpy(&handle, p, sizeof(handle));
@@ -89,50 +89,6 @@ static void malloc_oom(size_t size) {
     abort();
 }
 
-// hook : malloc, realloc, memalign, free, calloc
-
-void *
-malloc(size_t size) {
-    void* ptr = je_malloc(size + PREFIX_SIZE);
-    if(!ptr) malloc_oom(size);
-    return fill_prefix(ptr);
-}
-
-void *
-realloc(void *ptr, size_t size) {
-    if (ptr == NULL) return malloc(size);
-
-    void* rawptr = clean_prefix(ptr);
-    void *newptr = je_realloc(rawptr, size+PREFIX_SIZE);
-    if(!newptr) malloc_oom(size);
-    return fill_prefix(newptr);
-}
-
-#ifdef JEMALLOC_OVERRIDE_MEMALIGN
-
-void *
-memalign(size_t alignment, size_t size) {
-    void *ptr = je_memalign(alignment, size+PREFIX_SIZE);
-    if(!ptr) malloc_oom(size);
-    return fill_prefix(ptr);
-}
-
-#endif
-
-void
-free(void *ptr) {
-    if (ptr == NULL) return;
-    void* rawptr = clean_prefix(ptr);
-    je_free(rawptr);
-}
-
-void *
-calloc(size_t nmemb,size_t size) {
-    void* ptr = je_calloc(nmemb + ((PREFIX_SIZE+size-1)/size), size );
-    if(!ptr) malloc_oom(size);
-    return fill_prefix(ptr);
-}
-
 size_t
 malloc_used_memory(void) {
     return _used_memory;
@@ -191,10 +147,54 @@ dump_c_mem() {
     printf("+total: %zdkb\n",total >> 10);
 }
 
-/*
- * init 
- */
-void __attribute__ ((constructor)) malloc_hook_init(void){
-    _init();
+// hook : malloc, realloc, memalign, free, calloc
+
+void *
+skynet_malloc(size_t size) {
+    void* ptr = je_malloc(size + PREFIX_SIZE);
+    if(!ptr) malloc_oom(size);
+    return fill_prefix(ptr);
+}
+
+void *
+skynet_realloc(void *ptr, size_t size) {
+    if (ptr == NULL) return malloc(size);
+
+    void* rawptr = clean_prefix(ptr);
+    void *newptr = je_realloc(rawptr, size+PREFIX_SIZE);
+    if(!newptr) malloc_oom(size);
+    return fill_prefix(newptr);
+}
+
+void
+skynet_free(void *ptr) {
+    if (ptr == NULL) return;
+    void* rawptr = clean_prefix(ptr);
+    je_free(rawptr);
+}
+
+void *
+skynet_calloc(size_t nmemb,size_t size) {
+    void* ptr = je_calloc(nmemb + ((PREFIX_SIZE+size-1)/size), size );
+    if(!ptr) malloc_oom(size);
+    return fill_prefix(ptr);
+}
+
+char *
+skynet_strdup(const char *str) {
+	size_t sz = strlen(str);
+	char * ret = skynet_malloc(sz+1);
+	memcpy(ret, str, sz+1);
+	return ret;
+}
+
+void * 
+skynet_lalloc(void *ud, void *ptr, size_t osize, size_t nsize) {
+	if (nsize == 0) {
+		skynet_free(ptr);
+		return NULL;
+	} else {
+		return skynet_realloc(ptr, nsize);
+	}
 }
 

skynet-src/malloc_hook.h

@@ -1,5 +1,6 @@
 #ifndef __MALLOC_HOOK_H
 #define __MALLOC_HOOK_H
+
 #include <stdlib.h>
 
 extern size_t malloc_used_memory(void);

skynet-src/skynet.h

@@ -1,6 +1,8 @@
 #ifndef SKYNET_H
 #define SKYNET_H
 
+#include "skynet_malloc.h"
+
 #include <stddef.h>
 #include <stdint.h>
 

skynet-src/skynet_env.c

@@ -1,3 +1,5 @@
+// include skynet.h first for malloc hook
+#include "skynet.h"
 #include "skynet_env.h"
 
 #include <lua.h>

skynet-src/skynet_error.c

@@ -29,7 +29,7 @@ skynet_error(struct skynet_context * context, const char *msg, ...) {
 	int len = vsnprintf(tmp, LOG_MESSAGE_SIZE, msg, ap);
 	va_end(ap);
 	if (len < LOG_MESSAGE_SIZE) {
-		data = strdup(tmp);
+		data = skynet_strdup(tmp);
 	} else {
 		int max_size = LOG_MESSAGE_SIZE;
 		for (;;) {

skynet-src/skynet_handle.c

@@ -1,3 +1,6 @@
+// include skynet.h first for malloc hook
+#include "skynet.h"
+
 #include "skynet_handle.h"
 #include "skynet_server.h"
 #include "rwlock.h"
@@ -204,7 +207,7 @@ _insert_name(struct handle_storage *s, const char * name, uint32_t handle) {
 			end = mid - 1;
 		}
 	}
-	char * result = strdup(name);
+	char * result = skynet_strdup(name);
 
 	_insert_name_before(s, result, handle, begin);
 

skynet-src/skynet_main.c

@@ -1,3 +1,6 @@
+// include skynet.h first for malloc hook
+#include "skynet.h"
+
 #include "skynet_imp.h"
 #include "skynet_env.h"
 

skynet-src/skynet_malloc.h

@@ -0,0 +1,18 @@
+#ifndef skynet_malloc_h
+#define skynet_malloc_h
+
+#include <stddef.h>
+
+#define malloc skynet_malloc
+#define calloc skynet_calloc
+#define realloc skynet_realloc
+#define free skynet_free
+
+void * skynet_malloc(size_t sz);
+void * skynet_calloc(size_t nmemb,size_t size);
+void * skynet_realloc(void *ptr, size_t size);
+void skynet_free(void *ptr);
+char * skynet_strdup(const char *str);
+void * skynet_lalloc(void *ud, void *ptr, size_t osize, size_t nsize);	// use for lua
+
+#endif

skynet-src/skynet_module.c

@@ -1,3 +1,6 @@
+// include skynet.h first for malloc hook
+#include "skynet.h"
+
 #include "skynet_module.h"
 
 #include <assert.h>
@@ -103,7 +106,7 @@ skynet_module_query(const char * name) {
 			M->m[index].module = dl;
 
 			if (_open_sym(&M->m[index]) == 0) {
-				M->m[index].name = strdup(name);
+				M->m[index].name = skynet_strdup(name);
 				M->count ++;
 				result = &M->m[index];
 			}
@@ -152,7 +155,7 @@ void
 skynet_module_init(const char *path) {
 	struct modules *m = malloc(sizeof(*m));
 	m->count = 0;
-	m->path = strdup(path);
+	m->path = skynet_strdup(path);
 	m->lock = 0;
 
 	M = m;

skynet-src/skynet_monitor.c

@@ -1,3 +1,6 @@
+// include skynet.h first for malloc hook
+#include "skynet.h"
+
 #include "skynet_monitor.h"
 #include "skynet_server.h"
 #include "skynet.h"

skynet-src/skynet_server.c

@@ -1,3 +1,6 @@
+// include skynet.h first for malloc hook
+#include "skynet.h"
+
 #include "skynet_server.h"
 #include "skynet_module.h"
 #include "skynet_handle.h"
@@ -5,7 +8,6 @@
 #include "skynet_timer.h"
 #include "skynet_harbor.h"
 #include "skynet_env.h"
-#include "skynet.h"
 #include "skynet_monitor.h"
 
 #include <string.h>

skynet-src/skynet_socket.c

@@ -1,9 +1,11 @@
+// include skynet.h first for malloc hook
+#include "skynet.h"
+
 #include "skynet_socket.h"
 #include "socket_server.h"
 #include "skynet_server.h"
 #include "skynet_mq.h"
 #include "skynet_harbor.h"
-#include "skynet.h"
 
 #include <assert.h>
 #include <stdlib.h>

skynet-src/skynet_timer.c

@@ -1,8 +1,10 @@
+// include skynet.h first for malloc hook
+#include "skynet.h"
+
 #include "skynet_timer.h"
 #include "skynet_mq.h"
 #include "skynet_server.h"
 #include "skynet_handle.h"
-#include "skynet.h"
 
 #include <time.h>
 #include <assert.h>

skynet-src/socket_server.c

@@ -1,3 +1,6 @@
+// include skynet.h first for malloc hook
+#include "skynet.h"
+
 #include "socket_server.h"
 #include "socket_poll.h"