Parsing Call and Ret with ptrace.

问题:

I try to parse all the Calls and Rets from an executable with ptrace. Conforming the the x64opcode, I found opcodes for Calls: 0xe8 and for Rets: 0xc3, 0xc2, 0xca, 0xcb.

Since I parsed them I found more Rets than Calls.

There is the program I trace:

void func()                                                                                    
{                                                                                              
  write(1, "i", 1);                                                                            
}                                                                                              

int main(int ac)                                                                               
{                                                                                              
  func();                                                                                      
  return(0);                                                                                   
}

There is my tracer:

int                     tracer(t_info *info)                                                   
{                                                                                              
  int                   status;                                                                
  long                  ptr;                                                                   
  int                   ret = 0;                                                                                 
  int                   call = 0;                                                                                


  waitpid(info->pid, &status, 0);                                                              
  while (WIFSTOPPED(status))                                                                   
    {                                                                                          
      ptrace(PTRACE_GETREGS, info->pid, NULL, info->regs);                                     
      ptr = ptrace(PTRACE_PEEKDATA, info->pid, info->regs->rip);                               
      if (((ptr & 0x000000ff) == 0xe8)) // Opcode for call                                                              
        {                                                                                      
          call++;                                                                              
        }                                                                                      
      else if (((ptr & 0x000000ff) == 0xc3) // Opcodes for rets                                
               || ((ptr & 0x000000ff) == 0xc2)                                                 
               || ((ptr & 0x000000ff) == 0xca)                                                 
               || ((ptr & 0x000000ff) == 0xcb))                                                
        {                                                                                      
          ret++;                                                                               
        }                                                                                      
      ptrace(PTRACE_SINGLESTEP, info->pid, 0, 0);                                              
      waitpid(info->pid, &status, 0);                                                          
    }                                                                                          
  printf("Calls: %i\nRets: %i\nDiff: %i\n", call, ret, call - ret);                                             
  return (0);                                                                                  
}

There is my output:

Calls: 656
Rets: 666
Diff: -10

Why is there not the same number of rets and calls ? Do I miss some opcodes ? Is there functions that not return?

回答1:

You for example miss indirect calls like

callq *(<expr>)

which use other opcodes. Libc standard initialization routines make use of these. Depending on the expression several opcodes are possible, two examples:

ff d0                   callq  *%rax
41 ff 14 dc             callq  *(%r12,%rbx,8)

It's probably not easy to get them all. Maybe it would be easier and cleaner to decode the instructions with a library like libbfd and libopcodes