如何实现在Windows上运行Linux程序,附示例代码

微软在去年发布了Bash On Windows, 这项技术允许在Windows上运行Linux程序, 我相信已经有很多文章解释过Bash On Windows的原理,
而今天的这篇文章将会讲解如何自己实现一个简单的原生Linux程序运行器, 这个运行器在用户层实现, 原理和Bash On Windows不完全一样,比较接近Linux上的Wine.

示例程序完整的代码在github上, 地址是 https://github.com/303248153/HelloElfLoader

初步了解ELF格式

首先让我们先了解什么是原生Linux程序, 以下说明摘自维基百科

In computing, the Executable and Linkable Format (ELF, formerly named Extensible Linking Format), is a common standard file format for executable files, object code, shared libraries, and core dumps. First published in the specification for the application binary interface (ABI) of the Unix operating system version named System V Release 4 (SVR4),[2] and later in the Tool Interface Standard,[1] it was quickly accepted among different vendors of Unix systems. In 1999, it was chosen as the standard binary file format for Unix and Unix-like systems on x86 processors by the 86open project.

By design, ELF is flexible, extensible, and cross-platform, not bound to any given central processing unit (CPU) or instruction set architecture. This has allowed it to be adopted by many different operating systems on many different hardware platforms.

Linux的可执行文件格式采用了ELF格式, 而Windows采用了PE格式, 也就是我们经常使用的exe文件的格式.

ELF格式的结构如下

By Surueña - Own work, CC BY-SA 3.0, https://commons.wikimedia.org/w/index.php?curid=2922583

大致上可以分为这些部分

  • ELF头,在文件的最开头,储存了类型和版本等信息
  • 程序头, 供程序运行时解释器(interpreter)使用
  • 节头, 供程序编译时链接器(linker)使用, 运行时不需要读节头
  • 节内容, 不同的节作用都不一样
    • .text 代码节,保存了主要的程序代码
    • .rodata 保存了只读的数据,例如字符串(const char*)
    • .data 保存了可读写的数据,例如全局变量
    • 还有其他各种各样的节

让我们来实际看一下Linux可执行程序的样子
以下的编译环境是Ubuntu 16.04 x64 + gcc 5.4.0, 编译环境不一样可能会得出不同的结果

首先创建hello.c,写入以下的代码

#include <stdio.h>

int max(int x, int y) {
    return x > y ? x : y;
}

int main() {
    printf("max is %d\n", max(123, 321));
    printf("test many arguments %d %d %d %s %s %s %s %s %s\n", 1, 2, 3, "a", "b", "c", "d", "e", "f");
    return 100;
}

然后使用gcc编译这份代码

gcc hello.c

编译完成后你可以看到hello.c旁边多了一个a.out, 这就是linux的可执行文件了, 现在可以在linux上运行它

./a.out

你可以看到以下输出

max is 321
test many arguments 1 2 3 a b c d e f

我们来看看a.out包含了什么,解析ELF文件可以使用readelf命令

readelf -a ./a.out

可以看到输出了以下的信息

ELF 头:
  Magic:   7f 45 4c 46 02 01 01 00 00 00 00 00 00 00 00 00 
  类别:                              ELF64
  数据:                              2 补码,小端序 (little endian)
  版本:                              1 (current)
  OS/ABI:                            UNIX - System V
  ABI 版本:                          0
  类型:                              EXEC (可执行文件)
  系统架构:                          Advanced Micro Devices X86-64
  版本:                              0x1
  入口点地址:               0x400430
  程序头起点:          64 (bytes into file)
  Start of section headers:          6648 (bytes into file)
  标志:             0x0
  本头的大小:       64 (字节)
  程序头大小:       56 (字节)
  Number of program headers:         9
  节头大小:         64 (字节)
  节头数量:         31
  字符串表索引节头: 28

节头:
  [号] 名称              类型             地址              偏移量
       大小              全体大小          旗标   链接   信息   对齐
  [ 0]                   NULL             0000000000000000  00000000
       0000000000000000  0000000000000000           0     0     0
  [ 1] .interp           PROGBITS         0000000000400238  00000238
       000000000000001c  0000000000000000   A       0     0     1
  [ 2] .note.ABI-tag     NOTE             0000000000400254  00000254
       0000000000000020  0000000000000000   A       0     0     4
  [ 3] .note.gnu.build-i NOTE             0000000000400274  00000274
       0000000000000024  0000000000000000   A       0     0     4
  [ 4] .gnu.hash         GNU_HASH         0000000000400298  00000298
       000000000000001c  0000000000000000   A       5     0     8
  [ 5] .dynsym           DYNSYM           00000000004002b8  000002b8
       0000000000000060  0000000000000018   A       6     1     8
  [ 6] .dynstr           STRTAB           0000000000400318  00000318
       000000000000003f  0000000000000000   A       0     0     1
  [ 7] .gnu.version      VERSYM           0000000000400358  00000358
       0000000000000008  0000000000000002   A       5     0     2
  [ 8] .gnu.version_r    VERNEED          0000000000400360  00000360
       0000000000000020  0000000000000000   A       6     1     8
  [ 9] .rela.dyn         RELA             0000000000400380  00000380
       0000000000000018  0000000000000018   A       5     0     8
  [10] .rela.plt         RELA             0000000000400398  00000398
       0000000000000030  0000000000000018  AI       5    24     8
  [11] .init             PROGBITS         00000000004003c8  000003c8
       000000000000001a  0000000000000000  AX       0     0     4
  [12] .plt              PROGBITS         00000000004003f0  000003f0
       0000000000000030  0000000000000010  AX       0     0     16
  [13] .plt.got          PROGBITS         0000000000400420  00000420
       0000000000000008  0000000000000000  AX       0     0     8
  [14] .text             PROGBITS         0000000000400430  00000430
       00000000000001f2  0000000000000000  AX       0     0     16
  [15] .fini             PROGBITS         0000000000400624  00000624
       0000000000000009  0000000000000000  AX       0     0     4
  [16] .rodata           PROGBITS         0000000000400630  00000630
       0000000000000050  0000000000000000   A       0     0     8
  [17] .eh_frame_hdr     PROGBITS         0000000000400680  00000680
       000000000000003c  0000000000000000   A       0     0     4
  [18] .eh_frame         PROGBITS         00000000004006c0  000006c0
       0000000000000114  0000000000000000   A       0     0     8
  [19] .init_array       INIT_ARRAY       0000000000600e10  00000e10
       0000000000000008  0000000000000000  WA       0     0     8
  [20] .fini_array       FINI_ARRAY       0000000000600e18  00000e18
       0000000000000008  0000000000000000  WA       0     0     8
  [21] .jcr              PROGBITS         0000000000600e20  00000e20
       0000000000000008  0000000000000000  WA       0     0     8
  [22] .dynamic          DYNAMIC          0000000000600e28  00000e28
       00000000000001d0  0000000000000010  WA       6     0     8
  [23] .got              PROGBITS         0000000000600ff8  00000ff8
       0000000000000008  0000000000000008  WA       0     0     8
  [24] .got.plt          PROGBITS         0000000000601000  00001000
       0000000000000028  0000000000000008  WA       0     0     8
  [25] .data             PROGBITS         0000000000601028  00001028
       0000000000000010  0000000000000000  WA       0     0     8
  [26] .bss              NOBITS           0000000000601038  00001038
       0000000000000008  0000000000000000  WA       0     0     1
  [27] .comment          PROGBITS         0000000000000000  00001038
       0000000000000034  0000000000000001  MS       0     0     1
  [28] .shstrtab         STRTAB           0000000000000000  000018ea
       000000000000010c  0000000000000000           0     0     1
  [29] .symtab           SYMTAB           0000000000000000  00001070
       0000000000000660  0000000000000018          30    47     8
  [30] .strtab           STRTAB           0000000000000000  000016d0
       000000000000021a  0000000000000000           0     0     1
Key to Flags:
  W (write), A (alloc), X (execute), M (merge), S (strings), l (large)
  I (info), L (link order), G (group), T (TLS), E (exclude), x (unknown)
  O (extra OS processing required) o (OS specific), p (processor specific)

There are no section groups in this file.

程序头:
  Type           Offset             VirtAddr           PhysAddr
                 FileSiz            MemSiz              Flags  Align
  PHDR           0x0000000000000040 0x0000000000400040 0x0000000000400040
                 0x00000000000001f8 0x00000000000001f8  R E    8
  INTERP         0x0000000000000238 0x0000000000400238 0x0000000000400238
                 0x000000000000001c 0x000000000000001c  R      1
      [Requesting program interpreter: /lib64/ld-linux-x86-64.so.2]
  LOAD           0x0000000000000000 0x0000000000400000 0x0000000000400000
                 0x00000000000007d4 0x00000000000007d4  R E    200000
  LOAD           0x0000000000000e10 0x0000000000600e10 0x0000000000600e10
                 0x0000000000000228 0x0000000000000230  RW     200000
  DYNAMIC        0x0000000000000e28 0x0000000000600e28 0x0000000000600e28
                 0x00000000000001d0 0x00000000000001d0  RW     8
  NOTE           0x0000000000000254 0x0000000000400254 0x0000000000400254
                 0x0000000000000044 0x0000000000000044  R      4
  GNU_EH_FRAME   0x0000000000000680 0x0000000000400680 0x0000000000400680
                 0x000000000000003c 0x000000000000003c  R      4
  GNU_STACK      0x0000000000000000 0x0000000000000000 0x0000000000000000
                 0x0000000000000000 0x0000000000000000  RW     10
  GNU_RELRO      0x0000000000000e10 0x0000000000600e10 0x0000000000600e10
                 0x00000000000001f0 0x00000000000001f0  R      1

 Section to Segment mapping:
  段节...
   00     
   01     .interp 
   02     .interp .note.ABI-tag .note.gnu.build-id .gnu.hash .dynsym .dynstr .gnu.version .gnu.version_r .rela.dyn .rela.plt .init .plt .plt.got .text .fini .rodata .eh_frame_hdr .eh_frame 
   03     .init_array .fini_array .jcr .dynamic .got .got.plt .data .bss 
   04     .dynamic 
   05     .note.ABI-tag .note.gnu.build-id 
   06     .eh_frame_hdr 
   07     
   08     .init_array .fini_array .jcr .dynamic .got 

Dynamic section at offset 0xe28 contains 24 entries:
  标记        类型                         名称/值
 0x0000000000000001 (NEEDED)             共享库:[libc.so.6]
 0x000000000000000c (INIT)               0x4003c8
 0x000000000000000d (FINI)               0x400624
 0x0000000000000019 (INIT_ARRAY)         0x600e10
 0x000000000000001b (INIT_ARRAYSZ)       8 (bytes)
 0x000000000000001a (FINI_ARRAY)         0x600e18
 0x000000000000001c (FINI_ARRAYSZ)       8 (bytes)
 0x000000006ffffef5 (GNU_HASH)           0x400298
 0x0000000000000005 (STRTAB)             0x400318
 0x0000000000000006 (SYMTAB)             0x4002b8
 0x000000000000000a (STRSZ)              63 (bytes)
 0x000000000000000b (SYMENT)             24 (bytes)
 0x0000000000000015 (DEBUG)              0x0
 0x0000000000000003 (PLTGOT)             0x601000
 0x0000000000000002 (PLTRELSZ)           48 (bytes)
 0x0000000000000014 (PLTREL)             RELA
 0x0000000000000017 (JMPREL)             0x400398
 0x0000000000000007 (RELA)               0x400380
 0x0000000000000008 (RELASZ)             24 (bytes)
 0x0000000000000009 (RELAENT)            24 (bytes)
 0x000000006ffffffe (VERNEED)            0x400360
 0x000000006fffffff (VERNEEDNUM)         1
 0x000000006ffffff0 (VERSYM)             0x400358
 0x0000000000000000 (NULL)               0x0

重定位节 '.rela.dyn' 位于偏移量 0x380 含有 1 个条目:
  偏移量          信息           类型           符号值        符号名称 + 加数
000000600ff8  000300000006 R_X86_64_GLOB_DAT 0000000000000000 __gmon_start__ + 0

重定位节 '.rela.plt' 位于偏移量 0x398 含有 2 个条目:
  偏移量          信息           类型           符号值        符号名称 + 加数
000000601018  000100000007 R_X86_64_JUMP_SLO 0000000000000000 printf@GLIBC_2.2.5 + 0
000000601020  000200000007 R_X86_64_JUMP_SLO 0000000000000000 __libc_start_main@GLIBC_2.2.5 + 0

The decoding of unwind sections for machine type Advanced Micro Devices X86-64 is not currently supported.

Symbol table '.dynsym' contains 4 entries:
   Num:    Value          Size Type    Bind   Vis      Ndx Name
     0: 0000000000000000     0 NOTYPE  LOCAL  DEFAULT  UND 
     1: 0000000000000000     0 FUNC    GLOBAL DEFAULT  UND printf@GLIBC_2.2.5 (2)
     2: 0000000000000000     0 FUNC    GLOBAL DEFAULT  UND __libc_start_main@GLIBC_2.2.5 (2)
     3: 0000000000000000     0 NOTYPE  WEAK   DEFAULT  UND __gmon_start__

Symbol table '.symtab' contains 68 entries:
   Num:    Value          Size Type    Bind   Vis      Ndx Name
     0: 0000000000000000     0 NOTYPE  LOCAL  DEFAULT  UND 
     1: 0000000000400238     0 SECTION LOCAL  DEFAULT    1 
     2: 0000000000400254     0 SECTION LOCAL  DEFAULT    2 
     3: 0000000000400274     0 SECTION LOCAL  DEFAULT    3 
     4: 0000000000400298     0 SECTION LOCAL  DEFAULT    4 
     5: 00000000004002b8     0 SECTION LOCAL  DEFAULT    5 
     6: 0000000000400318     0 SECTION LOCAL  DEFAULT    6 
     7: 0000000000400358     0 SECTION LOCAL  DEFAULT    7 
     8: 0000000000400360     0 SECTION LOCAL  DEFAULT    8 
     9: 0000000000400380     0 SECTION LOCAL  DEFAULT    9 
    10: 0000000000400398     0 SECTION LOCAL  DEFAULT   10 
    11: 00000000004003c8     0 SECTION LOCAL  DEFAULT   11 
    12: 00000000004003f0     0 SECTION LOCAL  DEFAULT   12 
    13: 0000000000400420     0 SECTION LOCAL  DEFAULT   13 
    14: 0000000000400430     0 SECTION LOCAL  DEFAULT   14 
    15: 0000000000400624     0 SECTION LOCAL  DEFAULT   15 
    16: 0000000000400630     0 SECTION LOCAL  DEFAULT   16 
    17: 0000000000400680     0 SECTION LOCAL  DEFAULT   17 
    18: 00000000004006c0     0 SECTION LOCAL  DEFAULT   18 
    19: 0000000000600e10     0 SECTION LOCAL  DEFAULT   19 
    20: 0000000000600e18     0 SECTION LOCAL  DEFAULT   20 
    21: 0000000000600e20     0 SECTION LOCAL  DEFAULT   21 
    22: 0000000000600e28     0 SECTION LOCAL  DEFAULT   22 
    23: 0000000000600ff8     0 SECTION LOCAL  DEFAULT   23 
    24: 0000000000601000     0 SECTION LOCAL  DEFAULT   24 
    25: 0000000000601028     0 SECTION LOCAL  DEFAULT   25 
    26: 0000000000601038     0 SECTION LOCAL  DEFAULT   26 
    27: 0000000000000000     0 SECTION LOCAL  DEFAULT   27 
    28: 0000000000000000     0 FILE    LOCAL  DEFAULT  ABS crtstuff.c
    29: 0000000000600e20     0 OBJECT  LOCAL  DEFAULT   21 __JCR_LIST__
    30: 0000000000400460     0 FUNC    LOCAL  DEFAULT   14 deregister_tm_clones
    31: 00000000004004a0     0 FUNC    LOCAL  DEFAULT   14 register_tm_clones
    32: 00000000004004e0     0 FUNC    LOCAL  DEFAULT   14 __do_global_dtors_aux
    33: 0000000000601038     1 OBJECT  LOCAL  DEFAULT   26 completed.7585
    34: 0000000000600e18     0 OBJECT  LOCAL  DEFAULT   20 __do_global_dtors_aux_fin
    35: 0000000000400500     0 FUNC    LOCAL  DEFAULT   14 frame_dummy
    36: 0000000000600e10     0 OBJECT  LOCAL  DEFAULT   19 __frame_dummy_init_array_
    37: 0000000000000000     0 FILE    LOCAL  DEFAULT  ABS hello.c
    38: 0000000000000000     0 FILE    LOCAL  DEFAULT  ABS crtstuff.c
    39: 00000000004007d0     0 OBJECT  LOCAL  DEFAULT   18 __FRAME_END__
    40: 0000000000600e20     0 OBJECT  LOCAL  DEFAULT   21 __JCR_END__
    41: 0000000000000000     0 FILE    LOCAL  DEFAULT  ABS 
    42: 0000000000600e18     0 NOTYPE  LOCAL  DEFAULT   19 __init_array_end
    43: 0000000000600e28     0 OBJECT  LOCAL  DEFAULT   22 _DYNAMIC
    44: 0000000000600e10     0 NOTYPE  LOCAL  DEFAULT   19 __init_array_start
    45: 0000000000400680     0 NOTYPE  LOCAL  DEFAULT   17 __GNU_EH_FRAME_HDR
    46: 0000000000601000     0 OBJECT  LOCAL  DEFAULT   24 _GLOBAL_OFFSET_TABLE_
    47: 0000000000400620     2 FUNC    GLOBAL DEFAULT   14 __libc_csu_fini
    48: 0000000000000000     0 NOTYPE  WEAK   DEFAULT  UND _ITM_deregisterTMCloneTab
    49: 0000000000601028     0 NOTYPE  WEAK   DEFAULT   25 data_start
    50: 0000000000601038     0 NOTYPE  GLOBAL DEFAULT   25 _edata
    51: 0000000000400624     0 FUNC    GLOBAL DEFAULT   15 _fini
    52: 0000000000000000     0 FUNC    GLOBAL DEFAULT  UND printf@@GLIBC_2.2.5
    53: 0000000000400526    22 FUNC    GLOBAL DEFAULT   14 max
    54: 0000000000000000     0 FUNC    GLOBAL DEFAULT  UND __libc_start_main@@GLIBC_
    55: 0000000000601028     0 NOTYPE  GLOBAL DEFAULT   25 __data_start
    56: 0000000000000000     0 NOTYPE  WEAK   DEFAULT  UND __gmon_start__
    57: 0000000000601030     0 OBJECT  GLOBAL HIDDEN    25 __dso_handle
    58: 0000000000400630     4 OBJECT  GLOBAL DEFAULT   16 _IO_stdin_used
    59: 00000000004005b0   101 FUNC    GLOBAL DEFAULT   14 __libc_csu_init
    60: 0000000000601040     0 NOTYPE  GLOBAL DEFAULT   26 _end
    61: 0000000000400430    42 FUNC    GLOBAL DEFAULT   14 _start
    62: 0000000000601038     0 NOTYPE  GLOBAL DEFAULT   26 __bss_start
    63: 000000000040053c   109 FUNC    GLOBAL DEFAULT   14 main
    64: 0000000000000000     0 NOTYPE  WEAK   DEFAULT  UND _Jv_RegisterClasses
    65: 0000000000601038     0 OBJECT  GLOBAL HIDDEN    25 __TMC_END__
    66: 0000000000000000     0 NOTYPE  WEAK   DEFAULT  UND _ITM_registerTMCloneTable
    67: 00000000004003c8     0 FUNC    GLOBAL DEFAULT   11 _init

Version symbols section '.gnu.version' contains 4 entries:
 地址: 0000000000400358  Offset: 0x000358  Link: 5 (.dynsym)
  000:   0 (*本地*)       2 (GLIBC_2.2.5)   2 (GLIBC_2.2.5)   0 (*本地*)    

Version needs section '.gnu.version_r' contains 1 entries:
 地址:0x0000000000400360  Offset: 0x000360  Link: 6 (.dynstr)
  000000: 版本: 1  文件:libc.so.6  计数:1
  0x0010:名称:GLIBC_2.2.5  标志:无  版本:2

Displaying notes found at file offset 0x00000254 with length 0x00000020:
  Owner                 Data size   Description
  GNU                  0x00000010   NT_GNU_ABI_TAG (ABI version tag)
    OS: Linux, ABI: 2.6.32

Displaying notes found at file offset 0x00000274 with length 0x00000024:
  Owner                 Data size   Description
  GNU                  0x00000014   NT_GNU_BUILD_ID (unique build ID bitstring)
    Build ID: debd3d7912be860a432b5c685a6cff7fd9418528

从上面的信息中我们可以知道这个文件的类型是ELF64, 也就是64位的可执行程序, 并且有9个程序头和31个节头, 各个节的作用大家可以在网上找到资料, 这篇文章中只涉及到以下的节

  • .init 程序初始化的代码
  • .rela.dyn 需要重定位的变量列表
  • .rela.plt 需要重定位的函数列表
  • .plt 调用动态链接函数的代码
  • .text 保存了主要的程序代码
  • .init 保存了程序的初始化代码, 用于初始化全局变量等
  • .fini 保存了程序的终止代码, 用于析构全局变量等
  • .rodata 保存了只读的数据,例如字符串(const char*)
  • .data 保存了可读写的数据,例如全局变量
  • .dynsym 动态链接的符号表
  • .dynstr 动态链接的符号名称字符串
  • .dynamic 动态链接所需要的信息,供程序运行时使用(不需要访问节头)

什么是动态链接

上面的程序中调用了printf函数, 然而这个函数的实现并不在./a.out中, 那么printf函数在哪里, 又是怎么被调用的?

printf函数的实现在glibc库中, 也就是/lib/x86_64-linux-gnu/libc.so.6中, 在执行./a.out的时候会在glibc库中找到这个函数并进行调用, 我们来看看这段代码

执行以下命令反编译./a.out

objdump -c -S ./a.out

我们可以看到以下的代码

00000000004003f0 <printf@plt-0x10>:
  4003f0:   ff 35 12 0c 20 00       pushq  0x200c12(%rip)        # 601008 <_GLOBAL_OFFSET_TABLE_+0x8>
  4003f6:   ff 25 14 0c 20 00       jmpq   *0x200c14(%rip)        # 601010 <_GLOBAL_OFFSET_TABLE_+0x10>
  4003fc:   0f 1f 40 00             nopl   0x0(%rax)

0000000000400400 <printf@plt>:
  400400:   ff 25 12 0c 20 00       jmpq   *0x200c12(%rip)        # 601018 <_GLOBAL_OFFSET_TABLE_+0x18>
  400406:   68 00 00 00 00          pushq  $0x0
  40040b:   e9 e0 ff ff ff          jmpq   4003f0 <_init+0x28>

000000000040053c <main>:
  40053c:   55                      push   %rbp
  40053d:   48 89 e5                mov    %rsp,%rbp
  400540:   be 41 01 00 00          mov    $0x141,%esi
  400545:   bf 7b 00 00 00          mov    $0x7b,%edi
  40054a:   e8 d7 ff ff ff          callq  400526 <max>
  40054f:   89 c6                   mov    %eax,%esi
  400551:   bf 38 06 40 00          mov    $0x400638,%edi
  400556:   b8 00 00 00 00          mov    $0x0,%eax
  40055b:   e8 a0 fe ff ff          callq  400400 <printf@plt>

在这一段代码中,我们可以看到调用printf会首先调用0x400400printf@plt
printf@plt会负责在运行时找到实际的printf函数并跳转到该函数
在这里实际的printf函数会保存在0x400406 + 0x200c12 = 0x601018

需要注意的是0x601018一开始并不会指向实际的printf函数,而是会指向0x400406, 为什么会这样? 因为Linux的可执行程序为了考虑性能,不会在一开始就解决所有动态连接的函数,而是选择了延迟解决.
在上面第一次jmpq *0x200c12(%rip)会跳转到下一条指令0x400406, 又会继续跳转到0x4003f0, 再跳转到0x601010指向的地址, 0x601010指向的地址就是延迟解决的实现, 第一次延迟解决成功后, 0x601018就会指向实际的printf, 以后调用就会直接跳转到实际的printf上.

程序入口点

Linux程序运行首先会从_start函数开始, 上面readelf中的入口点地址0x400430就是_start函数的地址,

0000000000400430 <_start>:
  400430:   31 ed                   xor    %ebp,%ebp
  400432:   49 89 d1                mov    %rdx,%r9
  400435:   5e                      pop    %rsi
  400436:   48 89 e2                mov    %rsp,%rdx
  400439:   48 83 e4 f0             and    $0xfffffffffffffff0,%rsp
  40043d:   50                      push   %rax
  40043e:   54                      push   %rsp
  40043f:   49 c7 c0 20 06 40 00    mov    $0x400620,%r8
  400446:   48 c7 c1 b0 05 40 00    mov    $0x4005b0,%rcx
  40044d:   48 c7 c7 3c 05 40 00    mov    $0x40053c,%rdi
  400454:   e8 b7 ff ff ff          callq  400410 <__libc_start_main@plt>
  400459:   f4                      hlt    
  40045a:   66 0f 1f 44 00 00       nopw   0x0(%rax,%rax,1)

接下来_start函数会调用__libc_start_main函数, __libc_start_main是libc库中定义的初始化函数, 负责初始化全局变量和调用main函数等工作.

__libc_start_main函数还负责设置返回值和退出进程, 可以看到上面调用__libc_start_main后的指令是hlt, 这个指令永远不会被执行.

实现Linux程序运行器

在拥有以上的知识后我们可以先构想以下的运行器需要做什么.

因为x64的Windows和Linux程序使用的cpu指令集都是一样的,我们可以直接执行汇编而不需要一个指令模拟器,
而且这次我打算在用户层实现, 所以不能像Bash On Windows一样模拟syscall, 这个运行器会像下图一样模拟libc库的函数

这样运行器需要做的事情有:

  • 解析ELF文件
  • 加载程序代码到指定的内存地址
  • 加载数据到指定的内存地址
  • 提供动态链接的函数实现
  • 执行加载的程序代码

这些工作会在以下的示例程序中一一实现, 完整的源代码可以看文章顶部的链接

首先我们需要把ELF文件格式对应的代码从binutils中复制过来, 它包含了ELF头, 程序头和相关的数据结构, 里面用unsigned char[]是为了防止alignment, 这样结构体可以直接从文件内容中转换过来

ELFDefine.h:

#pragma once

namespace HelloElfLoader {
    // 以下内容复制自
    // https://github.com/aeste/binutils/blob/develop/elfcpp/elfcpp.h
    // https://github.com/aeste/binutils/blob/develop/include/elf/external.h

    // e_ident中各项的偏移值
    const int EI_MAG0 = 0;
    const int EI_MAG1 = 1;
    const int EI_MAG2 = 2;
    const int EI_MAG3 = 3;
    const int EI_CLASS = 4;
    const int EI_DATA = 5;
    const int EI_VERSION = 6;
    const int EI_OSABI = 7;
    const int EI_ABIVERSION = 8;
    const int EI_PAD = 9;
    const int EI_NIDENT = 16;

    // ELF文件类型
    enum {
        ELFCLASSNONE = 0,
        ELFCLASS32 = 1,
        ELFCLASS64 = 2
    };

    // ByteOrder
    enum {
        ELFDATANONE = 0,
        ELFDATA2LSB = 1,
        ELFDATA2MSB = 2
    };

    // 程序头类型
    enum PT
    {
        PT_NULL = 0,
        PT_LOAD = 1,
        PT_DYNAMIC = 2,
        PT_INTERP = 3,
        PT_NOTE = 4,
        PT_SHLIB = 5,
        PT_PHDR = 6,
        PT_TLS = 7,
        PT_LOOS = 0x60000000,
        PT_HIOS = 0x6fffffff,
        PT_LOPROC = 0x70000000,
        PT_HIPROC = 0x7fffffff,
        // The remaining values are not in the standard.
        // Frame unwind information.
        PT_GNU_EH_FRAME = 0x6474e550,
        PT_SUNW_EH_FRAME = 0x6474e550,
        // Stack flags.
        PT_GNU_STACK = 0x6474e551,
        // Read only after relocation.
        PT_GNU_RELRO = 0x6474e552,
        // Platform architecture compatibility information
        PT_ARM_ARCHEXT = 0x70000000,
        // Exception unwind tables
        PT_ARM_EXIDX = 0x70000001
    };

    // 动态节类型
    enum DT
    {
        DT_NULL = 0,
        DT_NEEDED = 1,
        DT_PLTRELSZ = 2,
        DT_PLTGOT = 3,
        DT_HASH = 4,
        DT_STRTAB = 5,
        DT_SYMTAB = 6,
        DT_RELA = 7,
        DT_RELASZ = 8,
        DT_RELAENT = 9,
        DT_STRSZ = 10,
        DT_SYMENT = 11,
        DT_INIT = 12,
        DT_FINI = 13,
        DT_SONAME = 14,
        DT_RPATH = 15,
        DT_SYMBOLIC = 16,
        DT_REL = 17,
        DT_RELSZ = 18,
        DT_RELENT = 19,
        DT_PLTREL = 20,
        DT_DEBUG = 21,
        DT_TEXTREL = 22,
        DT_JMPREL = 23,
        DT_BIND_NOW = 24,
        DT_INIT_ARRAY = 25,
        DT_FINI_ARRAY = 26,
        DT_INIT_ARRAYSZ = 27,
        DT_FINI_ARRAYSZ = 28,
        DT_RUNPATH = 29,
        DT_FLAGS = 30,

        // This is used to mark a range of dynamic tags.  It is not really
        // a tag value.
        DT_ENCODING = 32,

        DT_PREINIT_ARRAY = 32,
        DT_PREINIT_ARRAYSZ = 33,
        DT_LOOS = 0x6000000d,
        DT_HIOS = 0x6ffff000,
        DT_LOPROC = 0x70000000,
        DT_HIPROC = 0x7fffffff,

        // The remaining values are extensions used by GNU or Solaris.
        DT_VALRNGLO = 0x6ffffd00,
        DT_GNU_PRELINKED = 0x6ffffdf5,
        DT_GNU_CONFLICTSZ = 0x6ffffdf6,
        DT_GNU_LIBLISTSZ = 0x6ffffdf7,
        DT_CHECKSUM = 0x6ffffdf8,
        DT_PLTPADSZ = 0x6ffffdf9,
        DT_MOVEENT = 0x6ffffdfa,
        DT_MOVESZ = 0x6ffffdfb,
        DT_FEATURE = 0x6ffffdfc,
        DT_POSFLAG_1 = 0x6ffffdfd,
        DT_SYMINSZ = 0x6ffffdfe,
        DT_SYMINENT = 0x6ffffdff,
        DT_VALRNGHI = 0x6ffffdff,

        DT_ADDRRNGLO = 0x6ffffe00,
        DT_GNU_HASH = 0x6ffffef5,
        DT_TLSDESC_PLT = 0x6ffffef6,
        DT_TLSDESC_GOT = 0x6ffffef7,
        DT_GNU_CONFLICT = 0x6ffffef8,
        DT_GNU_LIBLIST = 0x6ffffef9,
        DT_CONFIG = 0x6ffffefa,
        DT_DEPAUDIT = 0x6ffffefb,
        DT_AUDIT = 0x6ffffefc,
        DT_PLTPAD = 0x6ffffefd,
        DT_MOVETAB = 0x6ffffefe,
        DT_SYMINFO = 0x6ffffeff,
        DT_ADDRRNGHI = 0x6ffffeff,

        DT_RELACOUNT = 0x6ffffff9,
        DT_RELCOUNT = 0x6ffffffa,
        DT_FLAGS_1 = 0x6ffffffb,
        DT_VERDEF = 0x6ffffffc,
        DT_VERDEFNUM = 0x6ffffffd,
        DT_VERNEED = 0x6ffffffe,
        DT_VERNEEDNUM = 0x6fffffff,

        DT_VERSYM = 0x6ffffff0,

        // Specify the value of _GLOBAL_OFFSET_TABLE_.
        DT_PPC_GOT = 0x70000000,

        // Specify the start of the .glink section.
        DT_PPC64_GLINK = 0x70000000,

        // Specify the start and size of the .opd section.
        DT_PPC64_OPD = 0x70000001,
        DT_PPC64_OPDSZ = 0x70000002,

        // The index of an STT_SPARC_REGISTER symbol within the DT_SYMTAB
        // symbol table.  One dynamic entry exists for every STT_SPARC_REGISTER
        // symbol in the symbol table.
        DT_SPARC_REGISTER = 0x70000001,

        DT_AUXILIARY = 0x7ffffffd,
        DT_USED = 0x7ffffffe,
        DT_FILTER = 0x7fffffff
    };;

    // ELF头的定义
    typedef struct {
        unsigned char   e_ident[16];        /* ELF "magic number" */
        unsigned char   e_type[2];      /* Identifies object file type */
        unsigned char   e_machine[2];       /* Specifies required architecture */
        unsigned char   e_version[4];       /* Identifies object file version */
        unsigned char   e_entry[8];     /* Entry point virtual address */
        unsigned char   e_phoff[8];     /* Program header table file offset */
        unsigned char   e_shoff[8];     /* Section header table file offset */
        unsigned char   e_flags[4];     /* Processor-specific flags */
        unsigned char   e_ehsize[2];        /* ELF header size in bytes */
        unsigned char   e_phentsize[2];     /* Program header table entry size */
        unsigned char   e_phnum[2];     /* Program header table entry count */
        unsigned char   e_shentsize[2];     /* Section header table entry size */
        unsigned char   e_shnum[2];     /* Section header table entry count */
        unsigned char   e_shstrndx[2];      /* Section header string table index */
    } Elf64_External_Ehdr;

    // 程序头的定义
    typedef struct {
        unsigned char   p_type[4];      /* Identifies program segment type */
        unsigned char   p_flags[4];     /* Segment flags */
        unsigned char   p_offset[8];        /* Segment file offset */
        unsigned char   p_vaddr[8];     /* Segment virtual address */
        unsigned char   p_paddr[8];     /* Segment physical address */
        unsigned char   p_filesz[8];        /* Segment size in file */
        unsigned char   p_memsz[8];     /* Segment size in memory */
        unsigned char   p_align[8];     /* Segment alignment, file & memory */
    } Elf64_External_Phdr;

    // DYNAMIC类型的程序头的内容定义
    typedef struct {
        unsigned char   d_tag[8];       /* entry tag value */
        union {
            unsigned char   d_val[8];
            unsigned char   d_ptr[8];
        } d_un;
    } Elf64_External_Dyn;

    // 动态链接的重定位记录,部分系统会用Elf64_External_Rel
    typedef struct {
        unsigned char r_offset[8];  /* Location at which to apply the action */
        unsigned char   r_info[8];  /* index and type of relocation */
        unsigned char   r_addend[8];    /* Constant addend used to compute value */
    } Elf64_External_Rela;

    // 动态链接的符号信息
    typedef struct {
        unsigned char   st_name[4];     /* Symbol name, index in string tbl */
        unsigned char   st_info[1];     /* Type and binding attributes */
        unsigned char   st_other[1];        /* No defined meaning, 0 */
        unsigned char   st_shndx[2];        /* Associated section index */
        unsigned char   st_value[8];        /* Value of the symbol */
        unsigned char   st_size[8];     /* Associated symbol size */
    } Elf64_External_Sym;
}

接下来我们定义一个读取和执行ELF文件的类, 这个类会在初始化时把文件加载到fileStream_, execute函数会负责执行

HelloElfLoader.h:

#pragma once
#include <string>
#include <fstream>

namespace HelloElfLoader {
    class Loader {
        std::ifstream fileStream_;

    public:
        Loader(const std::string& path);
        Loader(std::ifstream&& fileStream);
        void execute();
    };
}

构造函数如下, 也就是标准的c++打开文件的代码

HelloElfLoader.cpp:

Loader::Loader(const std::string& path) :
    Loader(std::ifstream(path, std::ios::in | std::ios::binary)) {}

Loader::Loader(std::ifstream&& fileStream) :
    fileStream_(std::move(fileStream)) {
    if (!fileStream_) {
        throw std::runtime_error("open file failed");
    }
}

接下来将实现上面所说的步骤, 首先是解析ELF文件

void Loader::execute() {
    std::cout << "====== start loading elf ======" << std::endl;

    // 检查当前运行程序是否64位
    if (sizeof(intptr_t) != sizeof(std::int64_t)) {
        throw std::runtime_error("please use x64 compile and run this program");
    }

    // 读取ELF头
    Elf64_External_Ehdr elfHeader = {};
    fileStream_.seekg(0);
    fileStream_.read(reinterpret_cast<char*>(&elfHeader), sizeof(elfHeader));

    // 检查ELF头,只支持64位且byte order是little endian的程序
    if (std::string(reinterpret_cast<char*>(elfHeader.e_ident), 4) != "\x7f\x45\x4c\x46") {
        throw std::runtime_error("magic not match");
    }
    else if (elfHeader.e_ident[EI_CLASS] != ELFCLASS64) {
        throw std::runtime_error("only support ELF64");
    }
    else if (elfHeader.e_ident[EI_DATA] != ELFDATA2LSB) {
        throw std::runtime_error("only support little endian");
    }

    // 获取program table的信息
    std::uint32_t programTableOffset = *reinterpret_cast<std::uint32_t*>(elfHeader.e_phoff);
    std::uint16_t programTableEntrySize = *reinterpret_cast<std::uint16_t*>(elfHeader.e_phentsize);
    std::uint16_t programTableEntryNum = *reinterpret_cast<std::uint16_t*>(elfHeader.e_phnum);
    std::cout << "program table at: " << programTableOffset << ", "
        << programTableEntryNum << " x " << programTableEntrySize << std::endl;

    // 获取section table的信息
    // section table只给linker用,loader中其实不需要访问section table
    std::uint32_t sectionTableOffset = *reinterpret_cast<std::uint32_t*>(elfHeader.e_shoff);
    std::uint16_t sectionTableEntrySize = *reinterpret_cast<std::uint16_t*>(elfHeader.e_shentsize);
    std::uint16_t sectionTableEntryNum = *reinterpret_cast<std::uint16_t*>(elfHeader.e_shentsize);
    std::cout << "section table at: " << sectionTableOffset << ", "
        << sectionTableEntryNum << " x " << sectionTableEntrySize << std::endl;

ELF文件的的开始部分就是ELF头,和Elf64_External_Ehdr结构体的结构相同, 我们可以读到Elf64_External_Ehdr结构体中,
然后ELF头包含了程序头和节头的偏移值, 我们可以预先获取到这些参数

节头在运行时不需要使用, 运行时需要遍历程序头

    // 准备动态链接的信息
    std::uint64_t jmpRelAddr = 0; // 重定位记录的开始地址
    std::uint64_t pltRelType = 0; // 重定位记录的类型 RELA或REL
    std::uint64_t pltRelSize = 0; // 重定位记录的总大小
    std::uint64_t symTabAddr = 0; // 动态符号表的开始地址
    std::uint64_t strTabAddr = 0; // 动态符号名称表的开始地址
    std::uint64_t strTabSize = 0; // 动态符号名称表的总大小

    // 遍历program hedaer
    std::vector<Elf64_External_Phdr> programHeaders;
    programHeaders.resize(programTableEntryNum);
    fileStream_.read(reinterpret_cast<char*>(programHeaders.data()), programTableEntryNum * programTableEntrySize);
    std::vector<std::shared_ptr<void>> loadedSegments;
    for (const auto& programHeader : programHeaders) {
        std::uint32_t type = *reinterpret_cast<const std::uint32_t*>(programHeader.p_type);
        if (type == PT_LOAD) {
            // 把文件内容(包含程序代码和数据)加载到虚拟内存,这个示例不考虑地址冲突
            std::uint64_t fileOffset = *reinterpret_cast<const std::uint64_t*>(programHeader.p_offset);
            std::uint64_t fileSize = *reinterpret_cast<const std::uint64_t*>(programHeader.p_filesz);
            std::uint64_t virtAddr = *reinterpret_cast<const std::uint64_t*>(programHeader.p_vaddr);
            std::uint64_t memSize = *reinterpret_cast<const std::uint64_t*>(programHeader.p_memsz);
            if (memSize < fileSize) {
                throw std::runtime_error("invalid memsz in program header, it shouldn't less than filesz");
            }
            // 在指定的虚拟地址分配内存
            std::cout << std::hex << "allocate address at: 0x" << virtAddr <<
                " size: 0x" << memSize << std::dec << std::endl;
            void* addr = ::VirtualAlloc((void*)virtAddr, memSize, MEM_COMMIT | MEM_RESERVE, PAGE_EXECUTE_READWRITE);
            if (addr == nullptr) {
                throw std::runtime_error("allocate memory at specific address failed");
            }
            loadedSegments.emplace_back(addr, [](void* ptr) { ::VirtualFree(ptr, 0, MEM_RELEASE); });
            // 复制文件内容到虚拟内存
            fileStream_.seekg(fileOffset);
            if (!fileStream_.read(reinterpret_cast<char*>(addr), fileSize)) {
                throw std::runtime_error("read contents into memory from LOAD program header failed");
            }
        }
        else if (type == PT_DYNAMIC) {
            // 遍历动态节
            std::uint64_t fileOffset = *reinterpret_cast<const std::uint64_t*>(programHeader.p_offset);
            fileStream_.seekg(fileOffset);
            Elf64_External_Dyn dynSection = {};
            std::uint64_t dynSectionTag = 0;
            std::uint64_t dynSectionVal = 0;
            do {
                if (!fileStream_.read(reinterpret_cast<char*>(&dynSection), sizeof(dynSection))) {
                    throw std::runtime_error("read dynamic section failed");
                }
                dynSectionTag = *reinterpret_cast<const std::uint64_t*>(dynSection.d_tag);
                dynSectionVal = *reinterpret_cast<const std::uint64_t*>(dynSection.d_un.d_val);
                if (dynSectionTag == DT_JMPREL) {
                    jmpRelAddr = dynSectionVal;
                }
                else if (dynSectionTag == DT_PLTREL) {
                    pltRelType = dynSectionVal;
                }
                else if (dynSectionTag == DT_PLTRELSZ) {
                    pltRelSize = dynSectionVal;
                }
                else if (dynSectionTag == DT_SYMTAB) {
                    symTabAddr = dynSectionVal;
                }
                else if (dynSectionTag == DT_STRTAB) {
                    strTabAddr = dynSectionVal;
                }
                else if (dynSectionTag == DT_STRSZ) {
                    strTabSize = dynSectionVal;
                }
            } while (dynSectionTag != 0);
        }
    }

还记得我们上面使用readelf读取到的信息吗?

程序头:
  Type           Offset             VirtAddr           PhysAddr
                 FileSiz            MemSiz              Flags  Align
  PHDR           0x0000000000000040 0x0000000000400040 0x0000000000400040
                 0x00000000000001f8 0x00000000000001f8  R E    8
  INTERP         0x0000000000000238 0x0000000000400238 0x0000000000400238
                 0x000000000000001c 0x000000000000001c  R      1
      [Requesting program interpreter: /lib64/ld-linux-x86-64.so.2]
  LOAD           0x0000000000000000 0x0000000000400000 0x0000000000400000
                 0x00000000000007d4 0x00000000000007d4  R E    200000
  LOAD           0x0000000000000e10 0x0000000000600e10 0x0000000000600e10
                 0x0000000000000228 0x0000000000000230  RW     200000
  DYNAMIC        0x0000000000000e28 0x0000000000600e28 0x0000000000600e28
                 0x00000000000001d0 0x00000000000001d0  RW     8
  NOTE           0x0000000000000254 0x0000000000400254 0x0000000000400254
                 0x0000000000000044 0x0000000000000044  R      4
  GNU_EH_FRAME   0x0000000000000680 0x0000000000400680 0x0000000000400680
                 0x000000000000003c 0x000000000000003c  R      4
  GNU_STACK      0x0000000000000000 0x0000000000000000 0x0000000000000000
                 0x0000000000000000 0x0000000000000000  RW     10
  GNU_RELRO      0x0000000000000e10 0x0000000000600e10 0x0000000000600e10
                 0x00000000000001f0 0x00000000000001f0  R      1

这里面类型是LOAD的头代表需要加载文件的内容到内存,
Offset是文件的偏移值, VirtAddr是虚拟内存地址, FileSiz是需要加载的文件大小, MemSiz是需要分配的内存大小, Flags是内存的访问权限,
这个示例不考虑访问权限(统一使用PAGE_EXECUTE_READWRITE).

这个程序有两个LOAD头, 第一个包含了代码和只读数据(.data, .init, .rodata等节的内容), 第二个包含了可写数据(.init_array, .fini_array等节的内容).

LOAD头对应的内容加载到指定的内存地址后我们就完成了构想中的第2个第3个步骤, 现在代码和数据都在内存中了.

接下来我们还需要处理动态链接的函数, 处理所需的信息可以从DYNAMIC头得到
DYNAMIC头包含的信息有

Dynamic section at offset 0xe28 contains 24 entries:
  标记        类型                         名称/值
 0x0000000000000001 (NEEDED)             共享库:[libc.so.6]
 0x000000000000000c (INIT)               0x4003c8
 0x000000000000000d (FINI)               0x400624
 0x0000000000000019 (INIT_ARRAY)         0x600e10
 0x000000000000001b (INIT_ARRAYSZ)       8 (bytes)
 0x000000000000001a (FINI_ARRAY)         0x600e18
 0x000000000000001c (FINI_ARRAYSZ)       8 (bytes)
 0x000000006ffffef5 (GNU_HASH)           0x400298
 0x0000000000000005 (STRTAB)             0x400318
 0x0000000000000006 (SYMTAB)             0x4002b8
 0x000000000000000a (STRSZ)              63 (bytes)
 0x000000000000000b (SYMENT)             24 (bytes)
 0x0000000000000015 (DEBUG)              0x0
 0x0000000000000003 (PLTGOT)             0x601000
 0x0000000000000002 (PLTRELSZ)           48 (bytes)
 0x0000000000000014 (PLTREL)             RELA
 0x0000000000000017 (JMPREL)             0x400398
 0x0000000000000007 (RELA)               0x400380
 0x0000000000000008 (RELASZ)             24 (bytes)
 0x0000000000000009 (RELAENT)            24 (bytes)
 0x000000006ffffffe (VERNEED)            0x400360
 0x000000006fffffff (VERNEEDNUM)         1
 0x000000006ffffff0 (VERSYM)             0x400358
 0x0000000000000000 (NULL)               0x0

一个个看上面代码中涉及到的类型

  • DT_JMPREL: 重定位记录的开始地址, 指向.rela.plt节在内存中保存的地址
  • DT_PLTREL: 重定位记录的类型 RELA或RE, 这里是RELAL
  • DT_PLTRELSZ: 重定位记录的总大小, 这里是24 * 2 = 48
重定位节 '.rela.plt' 位于偏移量 0x398 含有 2 个条目:
  偏移量          信息           类型           符号值        符号名称 + 加数
000000601018  000100000007 R_X86_64_JUMP_SLO 0000000000000000 printf@GLIBC_2.2.5 + 0
000000601020  000200000007 R_X86_64_JUMP_SLO 0000000000000000 __libc_start_main@GLIBC_2.2.5 + 0
  • DT_SYMTAB: 动态符号表的开始地址, 指向.dynsym节在内存中保存的地址
  • DT_STRTAB: 动态符号名称表的开始地址, 指向.dynstr节在内存中保存的地址
  • DT_STRSZ: 动态符号名称表的总大小
Symbol table '.dynsym' contains 4 entries:
   Num:    Value          Size Type    Bind   Vis      Ndx Name
     0: 0000000000000000     0 NOTYPE  LOCAL  DEFAULT  UND 
     1: 0000000000000000     0 FUNC    GLOBAL DEFAULT  UND printf@GLIBC_2.2.5 (2)
     2: 0000000000000000     0 FUNC    GLOBAL DEFAULT  UND __libc_start_main@GLIBC_2.2.5 (2)
     3: 0000000000000000     0 NOTYPE  WEAK   DEFAULT  UND __gmon_start__

在遍历完程序头以后, 我们可以知道有两个动态链接的函数需要重定位, 它们分别是__libc_start_mainprintf, 其中__libc_start_main负责调用main函数
接下来让我们需要设置这些函数的地址

    // 读取动态链接符号表
    std::string dynamicSymbolNames(reinterpret_cast<char*>(strTabAddr), strTabSize);
    Elf64_External_Sym* dynamicSymbols = reinterpret_cast<Elf64_External_Sym*>(symTabAddr);

    // 设置动态链接的函数地址
    std::cout << std::hex << "read dynamic entires at: 0x" << jmpRelAddr <<
        " size: 0x" << pltRelSize << std::dec << std::endl;
    if (jmpRelAddr == 0 || pltRelType != DT_RELA || pltRelSize % sizeof(Elf64_External_Rela) != 0) {
        throw std::runtime_error("invalid dynamic entry info, rel type should be rela");
    }
    std::vector<std::shared_ptr<void>> libraryFuncs;
    for (std::uint64_t offset = 0; offset < pltRelSize; offset += sizeof(Elf64_External_Rela)) {
        Elf64_External_Rela* rela = (Elf64_External_Rela*)(jmpRelAddr + offset);
        std::uint64_t relaOffset = *reinterpret_cast<const std::uint64_t*>(rela->r_offset);
        std::uint64_t relaInfo = *reinterpret_cast<const std::uint64_t*>(rela->r_info);
        std::uint64_t relaSym = relaInfo >> 32; // ELF64_R_SYM
        std::uint64_t relaType = relaInfo & 0xffffffff; // ELF64_R_TYPE
        // 获取符号
        Elf64_External_Sym* symbol = dynamicSymbols + relaSym;
        std::uint32_t symbolNameOffset = *reinterpret_cast<std::uint32_t*>(symbol->st_name);
        std::string symbolName(dynamicSymbolNames.data() + symbolNameOffset);
        std::cout << "relocate symbol: " << symbolName << std::endl;
        // 替换函数地址
        // 原本应该延迟解决,这里图简单就直接覆盖了
        void** relaPtr = reinterpret_cast<void**>(relaOffset);
        std::shared_ptr<void> func = resolveLibraryFunc(symbolName);
        if (func == nullptr) {
            throw std::runtime_error("unsupport symbol name");
        }
        libraryFuncs.emplace_back(func);
        *relaPtr = func.get();
    }

上面的代码遍历了DT_JMPREL重定位记录, 并且在加载时设置了这些函数的地址,
其实应该通过延迟解决实现的, 但是这里为了简单就直接替换成最终的地址了.

上面获取函数实际地址的逻辑我写到了resolveLibraryFunc中,这个函数的实现在另外一个文件, 如下

namespace HelloElfLoader {
    namespace {
        // 原始的返回地址
        thread_local void* originalReturnAddress = nullptr;

        void* getOriginalReturnAddress() {
            return originalReturnAddress;
        }

        void setOriginalReturnAddress(void* address) {
            originalReturnAddress = address;
        }

        // 模拟libc调用main的函数,目前不支持传入argc和argv
        void __libc_start_main(int(*main)()) {
            std::cout << "call main: " << main << std::endl;
            int ret = main();
            std::cout << "result: " << ret << std::endl;
            std::exit(0);
        }

        // 模拟printf函数
        int printf(const char* fmt, ...) {
            int ret;
            va_list myargs;
            va_start(myargs, fmt);
            ret = ::vprintf(fmt, myargs);
            va_end(myargs);
            return ret;
        }

        // 把System V AMD64 ABI转换为Microsoft x64 calling convention
        // 因为vc++不支持inline asm,只能直接写hex
        // 这个函数支持任意长度的参数,但是性能会有损耗,如果参数数量已知可以编写更快的loader代码   
        const char generic_func_loader[]{
            // 让参数连续排列在栈上
            // [第一个参数] [第二个参数] [第三个参数] ...
            0x58, // pop %rax 暂存原返回地址
            0x41, 0x51, // push %r9 入栈第六个参数,之后的参数都在后续的栈上
            0x41, 0x50, // push %r8 入栈第五个参数
            0x51, // push %rcx 入栈第四个参数
            0x52, // push %rdx 入栈第三个参数
            0x56, // push %rsi 入栈第二个参数
            0x57, // push %rdi 入栈第一个参数

            // 调用setOriginalReturnAddress保存原返回地址
            0x48, 0x89, 0xc1, // mov %rax, %rcx 第一个参数是原返回地址
            0x48, 0x83, 0xec, 0x20, // sub $0x20, %rsp 预留32位的影子空间
            0x48, 0xb8, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, // movabs $0, %rax
            0xff, 0xd0, // callq *%rax 调用setOriginalReturnAddress
            0x48, 0x83, 0xc4, 0x20, // add %0x20, %rsp 释放影子空间

            // 转换到Microsoft x64 calling convention
            0x59, // pop %rcx 出栈第一个参数
            0x5a, // pop %rdx 出栈第二个参数
            0x41, 0x58, // pop %r8 // 出栈第三个参数
            0x41, 0x59, // pop %r9 // 出栈第四个参数

            // 调用目标函数
            0x48, 0x83, 0xec, 0x20, // sub $0x20, %esp 预留32位的影子空间
            0x48, 0xb8, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, // movabs 0, %rax
            0xff, 0xd0, // callq *%rax 调用模拟的函数
            0x48, 0x83, 0xc4, 0x30, // add $0x30, %rsp 释放影子空间和参数(影子空间32 + 参数8*2)
            0x50, // push %rax 保存返回值

            // 调用getOriginalReturnAddress获取原返回地址
            0x48, 0xb8, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, // movabs $0, %rax
            0xff, 0xd0, // callq *%rax 调用getOriginalReturnAddress
            0x48, 0x89, 0xc1, // mov %rax, %rcx 原返回地址存到rcx
            0x58, // 恢复返回值
            0x51, // 原返回地址入栈顶
            0xc3 // 返回
        };
        const int generic_func_loader_set_addr_offset = 18;
        const int generic_func_loader_target_offset = 44;
        const int generic_func_loader_get_addr_offset = 61;
    }

    // 获取动态链接函数的调用地址
    std::shared_ptr<void> resolveLibraryFunc(const std::string& name) {
        void* funcPtr = nullptr;
        if (name == "__libc_start_main") {
            funcPtr = __libc_start_main;
        }
        else if (name == "printf") {
            funcPtr = printf;
        }
        else {
            return nullptr;
        }
        void* addr = ::VirtualAlloc(nullptr,
            sizeof(generic_func_loader), MEM_COMMIT | MEM_RESERVE, PAGE_EXECUTE_READWRITE);
        if (addr == nullptr) {
            throw std::runtime_error("allocate memory for _libc_start_main_loader failed");
        }
        std::shared_ptr<void> result(addr, [](void* ptr) { ::VirtualFree(ptr, 0, MEM_RELEASE); });
        std::memcpy(addr, generic_func_loader, sizeof(generic_func_loader));
        char* addr_c = reinterpret_cast<char*>(addr);
        *reinterpret_cast<void**>(addr_c + generic_func_loader_set_addr_offset) = setOriginalReturnAddress;
        *reinterpret_cast<void**>(addr_c + generic_func_loader_target_offset) = funcPtr;
        *reinterpret_cast<void**>(addr_c + generic_func_loader_get_addr_offset) = getOriginalReturnAddress;
        return result;
    }
}

理解这段代码需要先了解什么是x86 calling conventions, 在汇编中传递函数参数的办法由很多种, 像cdecl是把所有参数都放在栈中从低到高排列, 而fastcall是把第一个参数放ecx, 第二个参数放edx, 其余参数放栈中.

我们需要模拟的64位Linux程序,它传参使用了System V AMD64 ABI标准, 先把参数按RDI, RSI, RDX, RCX, R8, R9的顺序设置,如果有再多参数就放在栈中.
而64位的Windows传参使用了Microsoft x64 calling convention标准, 先把参数按RCX, RDX, R8, R9的顺序设置,如果有再多参数就放在栈中, 除此之外还需要预留一个32字节的影子空间.
如果我们需要让Linux程序调用Windows程序中的函数, 需要对参数的顺序进行转换, 这就是上面的汇编代码所做的事情.

转换前的栈结构如下

[原返回地址 8bytes] [第七个参数] [第八个参数] ...

转换后的栈结构如下

[返回地址 8bytes] [影子空间 32 bytes] [第五个参数] [第六个参数] [第七个参数] ...

因为需要支持不定个数的参数, 上面的代码用了一个thread local变量来保存原返回地址, 这样的处理会影响性能, 如果函数的参数个数已知可以换成更高效的转换代码.

在设置好动态链接的函数地址后, 我们完成了构想中的第4步, 接下来就可以运行主程序了

    // 获取入口点
    std::uint64_t entryPointAddress = *reinterpret_cast<const std::uint64_t*>(elfHeader.e_entry);
    void(*entryPointFunc)() = reinterpret_cast<void(*)()>(entryPointAddress);
    std::cout << "entry point: " << entryPointFunc << std::endl;
    std::cout << "====== finish loading elf ======" << std::endl;

    // 执行主程序
    // 会先调用__libc_start_main, 然后再调用main
    // 调用__libc_start_main后的指令是hlt,所以必须在__libc_start_main中退出执行
    entryPointFunc();

入口点的地址在ELF头中可以获取到,这个地址就是_start函数的地址, 我们把它转换成一个void()类型的函数指针再执行即可,
至此示例程序完成了构想中的所有功能.

执行效果如下图

这份示例程序还有很多不足, 例如未支持32位Linux程序, 不支持加载其他Linux动态链接库(so), 不支持命令行参数等等.
而且这份示例程序和Bash On Windows的原理有所出入, 因为在用户层是无法模拟syscall.
我希望它可以让你对如何运行其他系统的可执行文件有一个初步的了解, 如果你希望更深入的了解如何模拟syscall, 可以查找rdmsrwrmsr指令相关的资料.

最后附上我在编写这份示例程序中查阅的链接:

posted on 2017-05-15 16:16 q303248153 阅读(...) 评论(...) 编辑 收藏

导航

公告