Chinaunix首页 | 论坛 | 博客
  • 博客访问: 1880824
  • 博文数量: 376
  • 博客积分: 2147
  • 博客等级: 大尉
  • 技术积分: 3642
  • 用 户 组: 普通用户
  • 注册时间: 2012-02-06 10:47
文章分类

全部博文(376)

文章存档

2019年(3)

2017年(28)

2016年(15)

2015年(17)

2014年(182)

2013年(16)

2012年(115)

我的朋友

分类: 嵌入式

2015-12-11 15:36:27

创建时间:2001-09-25
文章属性:原创
文章来源:http://www.xfocus.org/
文章提交:alert7 (sztcww_at_sina.com)

Before main() 分析


作者:alert7
                     alert7@xfocus.org
             >

主页: http://www.xfocus.org
时间: 2001-9-25


★ 前言

本文分析了在main()之前的ELF程序流程,试图让您更清楚的把握程序的流程的脉络走向。
从而更深入的了解ELF。不正确之处,还请斧正。


★ 综述

ELF的可执行文件与共享库在结构上非常类似,它们具有一张程序段表,用来描述这些段如何映射到进程空间.
对于可执行文件来说,段的加载位置是固定的,程序段表中如实反映了段的加载地址.对于共享库来说,段的加
载位置是浮动的,位置无关的,程序段表反映的是以0作为基准地址的相对加载地址.尽管共享库的连接是不
充分的,为了便于测试动态链接器,Linux允许直接加载共享库运行.如果应用程序具有动态链接器的描述段,
内核在完成程序段加载后,紧接着加载动态链接器,并且启动动态链接器的入口.如果没有动态链接器的描述段,
就直接交给用户程序入口。
上述这部分请参考:linuxforum论坛上opera写的《分析ELF的加载过程》

在控制权交给动态链接器的入口后,首先调用_dl_start函数获得真实的程序入口(注:该入口地址
不是main的地址,也就是说一般程序的入口不是main),然后循环调用每个共享object的初始化函数,
接着跳转到真实的程序入口,一般为_start(程序中的_start)的一个例程,该例程压入一些参数到堆栈,
就直接调用__libc_start_main函数。在__libc_start_main函数中替动态连接器和自己程序安排
destructor,并运行程序的初始化函数。然后才把控制权交给main()函数。



★ main()之前流程

下面就是动态链接器的入口。
/* Initial entry point code for the dynamic linker.
   The C function `_dl_start' is the real entry point;
   its return value is the user program's entry point.  */

#define RTLD_START asm ("\
.text\n\
.globl _start\n\
.globl _dl_start_user\n\
_start:\n\
    pushl %esp\n\
    call _dl_start\n\/*该函数返回时候,%eax中存放着user entry point address*/
    popl %ebx\n\/*%ebx放着是esp的内容*/
_dl_start_user:\n\
    # Save the user entry point address in %edi.\n\
    movl %eax, %edi\n\/*入口地址放在%edi*/

    # Point %ebx at the GOT.
    call 0f\n\
0:    popl %ebx\n\
    addl $_GLOBAL_OFFSET_TABLE_+[.-0b], %ebx\n\

    # Store the highest stack address\n\
    movl __libc_stack_end@GOT(%ebx), %eax\n\
    movl %esp, (%eax)\n\/*把栈顶%esp放到GOT的__libc_stack_end中*/

    # See if we were run as a command with the executable file\n\
    # name as an extra leading argument.\n\
    movl _dl_skip_args@GOT(%ebx), %eax\n\
    movl (%eax), %eax\n\

    # Pop the original argument count.\n\
    popl %ecx\n\

    # Subtract _dl_skip_args from it.\n\
    subl %eax, %ecx\n\
    
    # Adjust the stack pointer to skip _dl_skip_args words.\n\
    leal (%esp,%eax,4), %esp\n\
    
    # Push back the modified argument count.\n\
    pushl %ecx\n\
    
    # Push the searchlist of the main object as argument in\n\
    # _dl_init_next call below.\n\
    movl _dl_main_searchlist@GOT(%ebx), %eax\n\
    movl (%eax), %esi\n\
0:    movl %esi,%eax\n\
    
    # Call _dl_init_next to return the address of an initializer\n\
    # function to run.\n\
    call _dl_init_next@PLT\n\/*该函数返回初始化函数的地址,返回地址放在%eax中*/
    
    # Check for zero return, when out of initializers.\n\
    testl %eax, %eax\n\
    jz 1f\n\
    
    # Call the shared object initializer function.\n\
    # NOTE: We depend only on the registers (%ebx, %esi and %edi)\n\
    # and the return address pushed by this call;\n\
    # the initializer is called with the stack just\n\
    # as it appears on entry, and it is free to move\n\
    # the stack around, as long as it winds up jumping to\n\
    # the return address on the top of the stack.\n\
    call *%eax\n\/*调用共享object初始化函数*/
    
    # Loop to call _dl_init_next for the next initializer.\n\
    jmp 0b\n\

1:    # Clear the startup flag.\n\
    movl _dl_starting_up@GOT(%ebx), %eax\n\
    movl $0, (%eax)\n\
    
    # Pass our finalizer function to the user in %edx, as per ELF ABI.\n\
    movl _dl_fini@GOT(%ebx), %edx\n\
    
    # Jump to the user's entry point.\n\
    jmp *%edi\n\
.previous\n\
");


sysdeps\i386\start.s中
user's entry也就是下面的_start例程

/* This is the canonical entry point, usually the first thing in the text
   segment.  The SVR4/i386 ABI (pages 3-31, 3-32) says that when the entry
   point runs, most registers' values are unspecified, except for:

   %edx        Contains a function pointer to be registered with `atexit'.
           This is how the dynamic linker arranges to have DT_FINI
        functions called for shared libraries that have been loaded
        before this code runs.

   %esp        The stack contains the arguments and environment:
           0(%esp)            argc
        4(%esp)            argv[0]
        ...
        (4*argc)(%esp)        NULL
        (4*(argc+1))(%esp)    envp[0]
        ...
                    NULL
*/

    .text
    .globl _start
_start:
    /* Clear the frame pointer.  The ABI suggests this be done, to mark
       the outermost frame obviously.  */
    xorl %ebp, %ebp

    /* Extract the arguments as encoded on the stack and set up
       the arguments for `main': argc, argv.  envp will be determined
       later in __libc_start_main.  */
    popl %esi        /* Pop the argument count.  */
    movl %esp, %ecx        /* argv starts just at the current stack top.*/

    /* Before pushing the arguments align the stack to a double word
       boundary to avoid penalties from misaligned accesses.  Thanks
       to Edward Seidl for pointing this out.  */
    andl $0xfffffff8, %esp
    pushl %eax        /* Push garbage because we allocate
                   28 more bytes.  */

    /* Provide the highest stack address to the user code (for stacks
       which grow downwards).  */
    pushl %esp

    pushl %edx        /* Push address of the shared library
                   termination function.  */

    /* Push address of our own entry points to .fini and .init.  */
    pushl $_fini
    pushl $_init

    pushl %ecx        /* Push second argument: argv.  */
    pushl %esi        /* Push first argument: argc.  */

    pushl $main

    /* Call the user's main function, and exit with its value.
       But let the libc call main.    */
    call __libc_start_main

    hlt            /* Crash if somehow `exit' does return.  */



__libc_start_main在sysdeps\generic\libc_start.c中
假设定义的是PIC的代码。
struct startup_info
{
  void *sda_base;
  int (*main) (int, char **, char **, void *);
  int (*init) (int, char **, char **, void *);
  void (*fini) (void);
};

int
__libc_start_main (int argc, char **argv, char **envp,
           void *auxvec, void (*rtld_fini) (void),
           struct startup_info *stinfo,
           char **stack_on_entry)
{

  /* the PPC SVR4 ABI says that the top thing on the stack will
     be a NULL pointer, so if not we assume that we're being called
     as a statically-linked program by Linux...     */
  if (*stack_on_entry != NULL)
    {
      /* ...in which case, we have argc as the top thing on the
     stack, followed by argv (NULL-terminated), envp (likewise),
     and the auxilary vector.  */
      argc = *(int *) stack_on_entry;
      argv = stack_on_entry + 1;
      envp = argv + argc + 1;
      auxvec = envp;
      while (*(char **) auxvec != NULL)
    ++auxvec;
      ++auxvec;
      rtld_fini = NULL;
    }

  /* Store something that has some relationship to the end of the
     stack, for backtraces.  This variable should be thread-specific.  */
  __libc_stack_end = stack_on_entry + 4;

  /* Set the global _environ variable correctly.  */
  __environ = envp;

  /* Register the destructor of the dynamic linker if there is any.  */
  if (rtld_fini != NULL)
    atexit (rtld_fini);/*替动态连接器安排destructor*/

  /* Call the initializer of the libc.  */

  __libc_init_first (argc, argv, envp);/*一个空函数*/

  /* Register the destructor of the program, if any.  */
  if (stinfo->fini)
    atexit (stinfo->fini);/*安排程序自己的destructor*/

  /* Call the initializer of the program, if any.  */

  /*运行程序的初始化函数*/
  if (stinfo->init)
    stinfo->init (argc, argv, __environ, auxvec);

/*运行程序main函数,到此,控制权才交给我们一般所说的程序入口*/
  exit (stinfo->main (argc, argv, __environ, auxvec));

}



void
__libc_init_first (int argc __attribute__ ((unused)), ...)
{
}

int
atexit (void (*func) (void))
{
  struct exit_function *new = __new_exitfn ();

  if (new == NULL)
    return -1;

  new->flavor = ef_at;
  new->func.at = func;
  return 0;
}


/* Run initializers for MAP and its dependencies, in inverse dependency
   order (that is, leaf nodes first).  */

ElfW(Addr)
internal_function
_dl_init_next (struct r_scope_elem *searchlist)
{
  unsigned int i;

  /* The search list for symbol lookup is a flat list in top-down
     dependency order, so processing that list from back to front gets us
     breadth-first leaf-to-root order.  */

  i = searchlist->r_nlist;
  while (i-- > 0)
    {
      struct link_map *l = searchlist->r_list[i];

      if (l->l_init_called)
    /* This object is all done.  */
    continue;

      if (l->l_init_running)
    {
      /* This object's initializer was just running.
         Now mark it as having run, so this object
         will be skipped in the future.  */
      l->l_init_running = 0;
      l->l_init_called = 1;
      continue;
    }

      if (l->l_info[DT_INIT]
      && (l->l_name[0] != '\0' || l->l_type != lt_executable))
    {
      /* Run this object's initializer.  */
      l->l_init_running = 1;

      /* Print a debug message if wanted.  */
      if (_dl_debug_impcalls)
        _dl_debug_message (1, "\ncalling init: ",
                l->l_name[0] ? l->l_name : _dl_argv[0],
                "\n\n", NULL);

      /*共享库的基地址+init在基地址中的偏移量*/
      return l->l_addr + l->l_info[DT_INIT]->d_un.d_ptr;
      
    }

      /* No initializer for this object.
     Mark it so we will skip it in the future.  */
      l->l_init_called = 1;
    }


  /* Notify the debugger all new objects are now ready to go.  */
  _r_debug.r_state = RT_CONSISTENT;
  _dl_debug_state ();

  return 0;
}
在main()之前的程序流程看试有点简单,但正在运行的时候还是比较复杂的
(自己用GBD跟踪下就知道了),因为一般的程序都需要涉及到PLT,GOT标号的
重定位。弄清楚这个对ELF由为重要,以后有机会再补上一篇吧。


★ 手动确定程序和动态连接器的入口

[alert7@redhat62 alert7]$ cat helo.c
#include
int main(int argc,char **argv)
{
printf("hello\n");
return 0;
}

[alert7@redhat62 alert7]$ gcc -o helo helo.c
[alert7@redhat62 alert7]$ readelf -h helo
ELF Header:
  Magic:   7f 45 4c 46 01 01 01 00 00 00 00 00 00 00 00 00
  Class:                             ELF32
  Data:                              2's complement, little endian
  Version:                           1 (current)
  OS/ABI:                            UNIX - System V
  ABI Version:                       0
  Type:                              EXEC (Executable file)
  Machine:                           Intel 80386
  Version:                           0x1
  Entry point address:               0x8048320
  Start of program headers:          52 (bytes into file)
  Start of section headers:          8848 (bytes into file)
  Flags:                             0x0
  Size of this header:               52 (bytes)
  Size of program headers:           32 (bytes)
  Number of program headers:         6
  Size of section headers:           40 (bytes)
  Number of section headers:         29
  Section header string table index: 26
在这里我们看到程序的入口为0x8048320,可以看看是否为main函数。

[alert7@redhat62 alert7]$ gdb -q helo
(gdb) disass 0x8048320
Dump of assembler code for function _start:
0x8048320 <_start>:     xor    %ebp,%ebp
0x8048322 <_start+2>:   pop    %esi
0x8048323 <_start+3>:   mov    %esp,%ecx
0x8048325 <_start+5>:   and    $0xfffffff8,%esp
0x8048328 <_start+8>:   push   %eax
0x8048329 <_start+9>:   push   %esp
0x804832a <_start+10>:  push   %edx
0x804832b <_start+11>:  push   $0x804841c
0x8048330 <_start+16>:  push   $0x8048298
0x8048335 <_start+21>:  push   %ecx
0x8048336 <_start+22>:  push   %esi
0x8048337 <_start+23>:  push   $0x80483d0
0x804833c <_start+28>:  call   0x80482f8 <__libc_start_main>
0x8048341 <_start+33>:  hlt
0x8048342 <_start+34>:  nop
End of assembler dump.
呵呵,不是main吧,程序的入口是个_start例程。

再来看动态连接器的入口是多少
[alert7@redhat62 alert7]$ ldd helo
        libc.so.6 => /lib/libc.so.6 (0x40018000)
        /lib/ld-linux.so.2 => /lib/ld-linux.so.2 (0x40000000)
动态连接器ld-linux.so.2加载到进程地址空间0x40000000。

[alert7@redhat62 alert7]$ readelf -h /lib/ld-linux.so.2
ELF Header:
  Magic:   7f 45 4c 46 01 01 01 00 00 00 00 00 00 00 00 00
  Class:                             ELF32
  Data:                              2's complement, little endian
  Version:                           1 (current)
  OS/ABI:                            UNIX - System V
  ABI Version:                       0
  Type:                              DYN (Shared object file)
  Machine:                           Intel 80386
  Version:                           0x1
  Entry point address:               0x1990
  Start of program headers:          52 (bytes into file)
  Start of section headers:          328916 (bytes into file)
  Flags:                             0x0
  Size of this header:               52 (bytes)
  Size of program headers:           32 (bytes)
  Number of program headers:         3
  Size of section headers:           40 (bytes)
  Number of section headers:         23
  Section header string table index: 20
共享object入口地址为0x1990。加上整个ld-linux.so.2被加载到进程地址空间0x40000000。
那么动态连接器的入口地址为0x1990+0x40000000=0x40001990。

用户空间执行的第一条指令地址就是0x40001990,既上面#define RTLD_START的开始。
阅读(1353) | 评论(0) | 转发(0) |
给主人留下些什么吧!~~