偶然中发现,下面的两端代码表现不一样
void main(){
void* p1 = malloc(32);
free(p1);
free(p1); // 这里会报double free 错误,程序退出
}
void main(){
void* p1 = malloc(32);
void* p2 = malloc(32);
free(p1);
free(p2);
free(p1); // 正常没有报错
free(p2);//正常,没有报错
...
}
我就开始疑惑,难道glibc malloc库对doublefree错误的检测那么傻B,只有连续两次free的指针一样才能检测出来?然后又尝试了memset溢出的情况,如下
void main(){
void* p1 = malloc(32);
memset(p1,1,64); // 这里溢出了p1
void* p2 = malloc(32); //
printf("p1=%p,p2=%p\n",p1,p2); // 打印发现,malloc没有检测 memset的溢出影响
}
经过搜索,发现了,原来glibc malloc 库有一个环境变量MALLOC_CHECK_,当其值是0的时候,对上述两种情况(不连续地free同一段虚拟内存,memset溢出)是没有检测的,当我将其值设置为1/2/3时,对上述情况都会正常报错。经过这次测试,建议各位在开发过程中,最好是将MALLOC_CHECK_设置为非0,等到发布的时候,才将其值设置为0提升速度。
下面是glibc malloc调试相关的环境变量或工具说明,是比较好的参考
(转)
一)MALLOC_CHECK_
GNU的标准库(glibc)可以通过内置的调试特性对动态内存进行调试,它就是MALLOC_CHECK_环境变量,
它在默认情况下是不设定的,在老的版本默认这个值为0,新的版本默认值为2,但有一个矛盾,如果设定为空,它将会打印出长长的跟踪信息,这比设为2更详细.
MALLOC_CHECK_有三种设定,即:
MALLOC_CHECK_=0 ----- 关闭所有检查.
MALLOC_CHECK_=1 ----- 当有错误被探测到时,在标准错误输出(stderr)上打印错误信息.
MALLOC_CHECK_=2 ----- 当有错误被探测到时,不显示错误信息,直接进行中断.
我们用下面的小程序做一下测试,源程序如下:
#include <stdio.h>
#include <stdlib.h>
int main (int argc,char *argv[])
{
int i;
char* p = (char *)malloc(10);
char* pt = p;
for (i = 0;i < 10;i++)
{
p[i] = ‘z‘;
}
free (p);
free(pt);
return 0;
}
gcc double-free.c -o double-free
注:这个程序会释放两次指针.
echo $MALLOC_CHECK_
我们在MALLOC_CHECK_默认设定的情况下,执行test程序,输出如下的信息:
./test
*** glibc detected *** ./test: double free or corruption (fasttop): 0x0890f008 ***
======= Backtrace: =========
/lib/libc.so.6[0x175f7d]
/lib/libc.so.6(cfree+0x90)[0x1795d0]
./test[0x80483dc]
/lib/libc.so.6(__libc_start_main+0xdc)[0x125dec]
./test[0x8048301]
======= Memory map: ========
00110000-00247000 r-xp 00000000 08:01 3704502 /lib/libc-2.5.so
00247000-00249000 r-xp 00137000 08:01 3704502 /lib/libc-2.5.so
00249000-0024a000 rwxp 00139000 08:01 3704502 /lib/libc-2.5.so
0024a000-0024d000 rwxp 0024a000 00:00 0
00b51000-00b6a000 r-xp 00000000 08:01 3704501 /lib/ld-2.5.so
00b6a000-00b6b000 r-xp 00018000 08:01 3704501 /lib/ld-2.5.so
00b6b000-00b6c000 rwxp 00019000 08:01 3704501 /lib/ld-2.5.so
00bf3000-00bf4000 r-xp 00bf3000 00:00 0 [vdso]
00dab000-00db6000 r-xp 00000000 08:01 3704511 /lib/libgcc_s-4.1.1-20070105.so.1
00db6000-00db7000 rwxp 0000a000 08:01 3704511 /lib/libgcc_s-4.1.1-20070105.so.1
08048000-08049000 r-xp 00000000 08:01 327681 /root/test
08049000-0804a000 rw-p 00000000 08:01 327681 /root/test
0890f000-08930000 rw-p 0890f000 00:00 0
b7e00000-b7e21000 rw-p b7e00000 00:00 0
b7e21000-b7f00000 ---p b7e21000 00:00 0
b7f26000-b7f27000 rw-p b7f26000 00:00 0
b7f3b000-b7f3c000 rw-p b7f3b000 00:00 0
bfdcf000-bfde4000 rw-p bfdcf000 00:00 0 [stack]
Aborted
这里我们调整MALLOC_CHECK_为0,再次运行程序,如下:
export MALLOC_CHECK_=0
./test
注:我们看到程序没有任何输出.
我们将MALLOC_CHECK_调整为1,再次运行程序,如下:
export MALLOC_CHECK_=1
./test
malloc: using debugging hooks
*** glibc detected *** ./test: free(): invalid pointer: 0x0811e008 ***
注:我们看到每次运行程序都会有malloc: using debugging hooks的输出,同时程序检测到free()两次释放的问题.
我们将MALLOC_CHECK_调整为2,再次运行程序,如下:
export MALLOC_CHECK_=2
./test
Aborted
注:我们看到程序只输出了Aborted,并中断了程序的运行.
二)用mtrace查找内存泄露
mtrace是由glibc提供的一个工具,在Redhat中将它打包在glibc-utils包中.
我们安装此包,如下:
rpm -ivh /mnt/Server/glibc-utils-2.5-12.i386.rpm
mtrace的主要作用是查找内存泄露,为了应用mtrace程序,必须在代码中使用glibc提供的函数mtrace和muntrace.另外,必须设置一个文件的名字给环境变量MALLOC_TRACE,因为glibc利用它为mtrace程序存储数据.
当执行完代码后,数据将会存在这个确认的文件中,每执行一次程序,这个文件的内容都会被重写.
我们用下面的代码进行测试,如下:
#include <stdio.h>
#include <stdlib.h>
#include <errno.h>
#include <mcheck.h>
int main (int argc,char *argv[])
{
setenv("MALLOC_TRACE","output",1);
mtrace();
int i;
char* p = (char *)malloc(10);
char* pt = p;
for (i = 0;i < 10;i++)
{
p[i] = ‘z‘;
}
return 0;
}
编译:
gcc test.c -o test
注:程序用setenv函数设定环境变量MALLOC_TRACE
运行程序:
./test
这时在当前目录下生成了一个名为output的文件,如下:
cat output
= Start
@ ./test:[0x80483f2] + 0x82ba438 0xa
@ /lib/libc.so.6:(clearenv+0x7c)[0xb9910c] - 0x82ba008
@ /lib/libc.so.6:(tdestroy+0x47)[0xc39b77] - 0x82ba090
@ /lib/libc.so.6:(tdestroy+0x4f)[0xc39b7f] - 0x82ba0b0
用mtrace查找内存泄露,它告诉我们memory not freed
mtrace output
- 0x082ba008 Free 3 was never alloc‘d 0xb9910c
- 0x082ba090 Free 4 was never alloc‘d 0xc39b77
- 0x082ba0b0 Free 5 was never alloc‘d 0xc39b7f
Memory not freed:
-----------------
Address Size Caller
0x082ba438 0xa at 0x80483f2
三)使用memusage收集内存统计数据
memusage不需要在代码中做出任何指示.这个工具也来自由glibc-utils包.它以柱形显示程序占用了多少内存.它默认输出到标准输出中,用ASCII文本显示一个绘成图画似的柱形.如下:
memusage awk ‘BEGIN{print "hello world"}‘
hello world
Memory usage summary: heap total: 7487, heap peak: 6891, stack peak: 8624
total calls total memory failed calls
malloc| 58 7487 0
realloc| 0 0 0 (nomove:0, dec:0, free:0)
calloc| 0 0 0
free| 15 797
Histogram for block sizes:
0-15 27 46% ==================================================
16-31 7 12% ============
32-47 2 3% ===
48-63 6 10% ===========
64-79 1 1% =
80-95 1 1% =
96-111 1 1% =
112-127 4 6% =======
160-175 1 1% =
176-191 2 3% ===
192-207 1 1% =
208-223 2 3% ===
384-399 1 1% =
480-495 1 1% =
4000-4015 1 1% =
四)使用Electric Fence检测内存泄漏
Electric Fence用一些巧妙的技术来检测程序在堆内存区上的溢出,不需要用Electric Fence来修改代码,相反,它提供一个动态库,这个库有多个动态分配函数.
一个名为ef的脚本被用来处理环境变量LD_PRELOAD的设置,我们可以用ef命令来调用程序.
下面是安装Electric Fence,如下:
rpm -ivh /mnt/Server/ElectricFence-2.2.2-20.2.2.i386.rpm
我们下面用一个小程序做测试,源代码如下:
#include <string.h>
int
main (int argc, char *argv[])
{
int *ptr = new int;
memset(ptr, 0, sizeof(int) + 1);
delete ptr;
}
编译:
g++ new-corrupt.cpp -o new-corrupt
注:这个小程序会导致边界溢出.
执行程序:
./new-corrupt
注:程序没有任何指示.
我们用ef执行这个程序,如下:
ef ./new-corrupt
Electric Fence 2.2.0 Copyright (C) 1987-1999 Bruce Perens <bruce@perens.com>
/usr/bin/ef: line 20: 4148 Segmentation fault ( export LD_PRELOAD=libefence.so.0.0; exec $* )
注:此时有输出信息,它告诉我们出现了Segmentation {敏感词}t,并且指明在哪行出现的问题.
我们也可以将electric fence和gdb联用,如下:
编译程序,同时指定-g选项
g++ -g new-corrupt.cpp -o new-corrupt
用gdb打开程序,如下:
gdb ./new-corrupt
GNU gdb Red Hat Linux (6.5-16.el5rh)
Copyright (C) 2006 Free Software Foundation, Inc.
GDB is free software, covered by the GNU General Public License, and you are
welcome to change it and/or distribute copies of it under certain conditions.
Type "show copying" to see the conditions.
There is absolutely no warranty for GDB. Type "show warranty" for details.
This GDB was configured as "i386-redhat-linux-gnu"...Using host libthread_db library "/lib/libthread_db.so.1".
(gdb) set environment LD_PRELOAD libefence.so.0.0 /*设置环境变量LD_PRELOAD为libefence.so.0.0*/
(gdb) run /*运行程序*/
Starting program: /root/new-corrupt
Electric Fence 2.2.0 Copyright (C) 1987-1999 Bruce Perens <bruce@perens.com>
Electric Fence 2.2.0 Copyright (C) 1987-1999 Bruce Perens <bruce@perens.com>
Program received signal SIGSEGV, Segmentation fault.
0x0804849d in main () at new-corrupt.cpp:7 /*检查出在调用memset函数时导致越界*/
7 memset(ptr, 0, sizeof(int) + 1);
(gdb) quit
下面我们在gdb中不指定环境变量,我们看到gdb没有打印出相关的错误信息.
gdb ./new-corrupt
GNU gdb Red Hat Linux (6.5-16.el5rh)
Copyright (C) 2006 Free Software Foundation, Inc.
GDB is free software, covered by the GNU General Public License, and you are
welcome to change it and/or distribute copies of it under certain conditions.
Type "show copying" to see the conditions.
There is absolutely no warranty for GDB. Type "show warranty" for details.
This GDB was configured as "i386-redhat-linux-gnu"...Using host libthread_db library "/lib/libthread_db.so.1".
(gdb) run
Starting program: /root/new-corrupt
Program exited normally.