记录一次gdb debug经历

问题描述

今天在写代码时,运行时奔溃了。segment fault,而且是在程序退出main()函数后,才报的。
唯一的信息是:Segmentation fault (core dumped)
简直是一头雾水。

查看core文件

系统默认是不会生成core文件的,ulimit -c unlimited把core文件设为无限大。

使用gdb查看core文件

gdb ./example/sudoku_batch_test core
提示如下:

Program terminated with signal SIGSEGV, Segmentation fault.
#0  __GI___libc_free (mem=0x313030303030300a) at malloc.c:2951
2951	malloc.c: No such file or directory.
(gdb) 

可以确定崩溃发生在malloc.c中。但是提示没有malloc.c的源码。

首先安装glibc的符号表,命令如下:
sudo apt-get install libc6-dbg

再来是安装glibc的源文件,命令如下:
sudo apt-get source libc6-dev
安装完毕后在当前目录下会多出一个glibc-2.23文件夹,该文件夹包含了glibc的源码。

源码准备就绪后,接着上面,在gdb命令提示符下输入:
directory glibc-2.23/malloc/将glibc-2.23/malloc/设为gdb源码搜索目录。结果如下:

warning: core file may not match specified executable file.
[New LWP 24491]
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".
Core was generated by `./example/sudoku_batch_test ../example/test1000 127.0.0.1 1'.
Program terminated with signal SIGSEGV, Segmentation fault.
#0  __GI___libc_free (mem=0x313030303030300a) at malloc.c:2951
2951	malloc.c: No such file or directory.
(gdb) directory glibc-2.23/malloc/
Source directories searched: /root/work/melon/build/glibc-2.23/malloc:$cdir:$cwd
(gdb) 

现在我们就可以在gdb中查看崩溃处的源码了,执行list

(gdb) l
warning: Source file is more recent than executable.
2946	  if (mem == 0)                              /* free(0) has no effect */
2947	    return;
2948	
2949	  p = mem2chunk (mem);
2950	
2951	  if (chunk_is_mmapped (p))                       /* release mmapped memory. */
2952	    {
2953	      /* see if the dynamic brk/mmap threshold needs adjusting */
2954	      if (!mp_.no_dyn_threshold
2955	          && p->size > mp_.mmap_threshold
(gdb) 

虽然知道了崩溃发生在2951行,但是貌似没有更多有效的信息。这时我想到了是不是可以看下函数的调用栈,或许会有信息。
接着执行backtrace(或者bt):

(gdb) bt
#0  __GI___libc_free (mem=0x313030303030300a) at malloc.c:2951
#1  0x000000000048bc9d in melon::Coroutine::~Coroutine (this=0x1fc9120, __in_chrg=<optimized out>)
    at /root/work/melon/src/Coroutine.cpp:56
#2  0x000000000048d099 in std::_Sp_counted_ptr<melon::Coroutine*, (__gnu_cxx::_Lock_policy)2>::_M_dispose (
    this=0x1fc8190) at /usr/include/c++/5/bits/shared_ptr_base.h:374
#3  0x00000000004630f1 in std::_Sp_counted_base<(__gnu_cxx::_Lock_policy)2>::_M_release (this=0x1fc8190)
    at /usr/include/c++/5/bits/shared_ptr_base.h:150
#4  0x0000000000461f32 in std::__shared_count<(__gnu_cxx::_Lock_policy)2>::~__shared_count (this=0x7f07f4ff1770, 
    __in_chrg=<optimized out>) at /usr/include/c++/5/bits/shared_ptr_base.h:659
#5  0x00000000004749ed in std::__shared_ptr<melon::Coroutine, (__gnu_cxx::_Lock_policy)2>::~__shared_ptr (
    this=0x7f07f4ff1768, __in_chrg=<optimized out>) at /usr/include/c++/5/bits/shared_ptr_base.h:925
#6  0x0000000000474a39 in std::shared_ptr<melon::Coroutine>::~shared_ptr (this=0x7f07f4ff1768, 
    __in_chrg=<optimized out>) at /usr/include/c++/5/bits/shared_ptr.h:93
#7  0x00007f07f40915ff in __GI___call_tls_dtors () at cxa_thread_atexit_impl.c:155
#8  0x00007f07f4090f27 in __run_exit_handlers (status=0, listp=0x7f07f441b5f8 <__exit_funcs>, 
    run_list_atexit=run_list_atexit@entry=true) at exit.c:40
#9  0x00007f07f4091045 in __GI_exit (status=<optimized out>) at exit.c:104
#10 0x00007f07f4077837 in __libc_start_main (main=0x45f1c4 <main(int, char**)>, argc=4, argv=0x7ffcfb2ab218, 
    init=<optimized out>, fini=<optimized out>, rtld_fini=<optimized out>, stack_end=0x7ffcfb2ab208)
    at ../csu/libc-start.c:325
#11 0x000000000045ec89 in _start ()

这下问题找到了,首先在线程结束或者程序运行结束会调用__GI___call_tls_dtors函数来析构线程本地存储。我确实用了thread_local关键字修饰Coroutine::Ptr变量。
#1 0x000000000048bc9d in melon::Coroutine::~Coroutine可知在melon::Coroutine类的析构函数中调用了free()导致奔溃。
这下问题基本明确了,我在Coroutine析构函数中会释放stack_这个指针,

 53 Coroutine::~Coroutine() {
 54     LOG_DEBUG << "destroy coroutine:" << name_;
 55     if (stack_) {
 56         free(stack_);
 57     }
 58 }

有两个构造函数,其中一个如下:

 39 Coroutine::Coroutine()
 40     :c_id_(++t_coroutine_id),
 41     name_("Main-" + std::to_string(c_id_)),
 42     cb_(nullptr),
 43     state_(CoroutineState::INIT) {
 44 
 45     if (getcontext(&context_)) {
 46         LOG_ERROR << "getcontext: errno=" << errno
 47                 << " error string:" << strerror(errno);
 58     }
 59 }

因为大意犯了个非常低级的错误,这个构造函数没有正确初始化statck_指针,将statck_初始化为nullptr后,问题就解决了。

update:2019-10-31
其实不用这么麻烦,gdb有个where命令,能直接打印出函数栈信息。

总结

遇到这类问题,一般用gdb查看core文件都能定位到崩溃的位置,如果不是直接引发的,可以查看函数调用栈,一般都能找到问题原因。

posted @ 2019-10-28 20:29  gatsby123  阅读(6240)  评论(0编辑  收藏  举报