zoukankan      html  css  js  c++  java
  • 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。

  • 相关阅读:
    容器常用命令
    镜像常用命令
    Docker安装
    Jenkins部署
    IIS配置伪静态 集成模式 样式丢失
    centos7 apache 配置ssl
    centOS7 关闭swap
    Presto集群部署和配置
    HDFS中将普通用户增加到超级用户组supergroup
    superset在 centos 7安装运行
  • 原文地址:https://www.cnblogs.com/gatsby123/p/11755320.html
Copyright © 2011-2022 走看看