Upgrade to Pro — share decks privately, control downloads, hide ads and more …

Secrets Behind HelloWorld.exe — Compilation, Li...

Dan Chen
November 13, 2015

Secrets Behind HelloWorld.exe — Compilation, Linking, Execution

November, 13, 2015. “Secrets Behind HelloWorld.exe — Compilation, Linking, Execution” at Trend Micro internal sharing session.

Target audience: Intermediate and experienced software developers.

CC-BY-SA 4.0

Dan Chen

November 13, 2015
Tweet

More Decks by Dan Chen

Other Decks in Programming

Transcript

  1. Goal 1. Write better programs 2. Build large programs 3.

    Avoid dangerous programming errors 4. Understand how language rules are implemented 5. Understand important system concepts
  2. Topics 1. Compilation 2. Static linking 3. Dynamic linking 4.

    Process launch [Note] Unless stated, all experiments are conducted with Visual Studio 2012 on Windows 7 SP1 x86
  3. #include <stdio.h> int main(int argc, char* argv[]) { char const

    greeting[] = "Hello, world!"; printf("%s\n", greeting); return 0; } See anything?
  4. #include <stdio.h> int main(int argc, char* argv[]) { char const

    greeting[] = "Hello, world!"; printf("%s\n", greeting); return 0; } Huh?
  5. #include <stdio.h> int main(int argc, char* argv[]) { char const

    greeting[] = "Hello, world!"; printf("%s\n", greeting); return 0; } So many questions! What is stdio.h? What is #include? Why? Who invokes main()? Where’re the arguments from? When’s const-ness checked? Where’s and How’s printf() implemented? Why does main() return an int? What’ll happen after it returns? How does printf() format the output string? How does it take variable-length arguments? Where is the string literal stored? Can we modify the content? Are char[] and char* different? What’s the value of greetings? When’s the value assigned? Is it always the same value?
  6. Hello.c Write Hello.asm Hello.obj Hello.exe Hello, World! Compile Assemble Link

    Execute msvcrt.lib ntdll.dll kernel32.dll msvcr110.dll (Static Libraries) (Dynamic Libraries) Lifecycle Overview of Windows Executables Hello.i Preprocess stdio.h (Header Files)
  7. #include <stdio.h> // Hello! int main(int argc, char* argv[]) {

    char const greeting[] = "Hello, world!"; printf("%s\n", greeting); return 0; } Hello.c #line 1 "Hello.c" #line 1 "C:\\Program Files\\Microsoft Visual Studio 11.0\\VC\\INCLUDE\\stdio.h" #pragma once // …5700 lines removed… #line 271 "C:\\Program Files\\Microsoft Visual Studio 11.0\\VC\\INCLUDE\\stdio.h" int __cdecl printf( const char * _Format, ...); // …600 lines removed… #line 2 "Hello.c" int main(int argc, char* argv[]) { char const greeting[] = "Hello, world!"; printf("%s\n", greeting); return 0; } Hello.i > CL.exe /P Hello.c Preprocessing 1. Apply #defines, and expand all macros 2. Handle all conditional compilation directives (e.g., #if) 3. Insert all #include’d files in place, recursively 4. Remove all comments (i.e., //… and /*…*/) 5. Insert line numbers and file paths for error handling 6. Preserve all #pragma directives for compiler
  8. // … #line 2 "Hello.c" int main(int argc, char* argv[])

    { // … } Hello.i .686P INCLUDELIB MSVCRT INCLUDELIB OLDNAMES _DATA SEGMENT $SG2940 DB! 'Hello, world!', 00H $SG2941 DB! '%s', 0aH, 00H _DATA ENDS PUBLIC _main EXTERN _printf:PROC _TEXT SEGMENT _main PROC push ebp mov ebp, esp push OFFSET $SG2940 push OFFSET $SG2941 call _printf mov esp, ebp pop ebp ret 0 _main! ENDP _TEXT! ENDS END Hello.asm > CL.exe /MD /Fa /Tc Hello.i Compilation Source Code Tokens Abstract Syntax Tree Tagged Abstract Syntax Tree Intermediate Representation Optimized I.R. Target Code Optimized Target Code (Hello.i) (Hello.asm) Lexical Analysis Syntactic Analysis IR Code Gen. IR Optimization Target Code Optimization Semantic Analysis Target Code Generation
  9. Token Type array Identifier [ TK_LBracket index Identifier ] TK_RBracket

    = Tk_Assign ( TK_LParen index Identifier + TK_Plus 4 Number ) TK_RParen * TK_Asterisk ( TK_LParen 2 Number + TK_Plus 6 Number ) TK_RParen ; TK_Semicolon // … array[index] = (index + 4) * (2 + 6); // … Stmt.c Compilation: Lexical Analysis Character Stream Token Stream Application: Syntax highlighting.
  10. Token Type array Identifier [ TK_LBracket index Identifier ] TK_RBracket

    = Tk_Assign ( TK_LParen index Identifier + TK_Plus 4 Number ) TK_RParen * TK_Asterisk ( TK_LParen 2 Number + TK_Plus 6 Number ) TK_RParen ; TK_Semicolon Compilation: Syntactic Analysis Token Stream Abstract Syntax Tree Application: Source code formatting. Assignment = Subscription [] Multiplication * Identifier array Identifier index Addition + Addition + Identifier index Number 4 Number 2 Number 6
  11. Compilation: Semantic Analysis Abstract Syntax Tree Tagged AST Application: IntelliSense.

    Assignment = int Subscription [] int Multiplication * int Identifier array int[] Identifier index int Addition + int Addition + int Identifier index int Number 4 int Number 2 int Number 6 int 1. Check for static semantics (e.g., type information) 2. Insert nodes for implicit type conversion
  12. Compilation: IR Code Generation Tagged AST Intermediate Representation This experiment

    is conducted with LLVM-3.4 on Ubuntu 14.04.3 amd64 int main() { int a = 55, b = 66; return (a + b); } Simple.c TranslationUnitDecl  0x272b990  <<invalid  sloc>> |-­‐TypedefDecl  0x272be90  <<invalid  sloc>>  __int128_t  '__int128' |-­‐TypedefDecl  0x272bef0  <<invalid  sloc>>  __uint128_t  'unsigned  __int128' |-­‐TypedefDecl  0x272c240  <<invalid  sloc>>  __builtin_va_list  '__va_list_tag  [1]' `-­‐FunctionDecl  0x272c2e0  <Simple.c:1:1,  line:4:1>  main  'int  ()'    `-­‐CompoundStmt  0x272c5b0  <line:1:12,  line:4:1>        |-­‐DeclStmt  0x272c4b0  <line:2:3,  col:21>        |  |-­‐VarDecl  0x272c390  <col:3,  col:11>  a  'int'        |  |  `-­‐IntegerLiteral  0x272c3e8  <col:11>  'int'  55        |  `-­‐VarDecl  0x272c420  <col:3,  col:19>  b  'int'        |      `-­‐IntegerLiteral  0x272c478  <col:19>  'int'  66        `-­‐ReturnStmt  0x272c590  <line:3:3,  col:16>            `-­‐ParenExpr  0x272c570  <col:10,  col:16>  'int'                `-­‐BinaryOperator  0x272c548  <col:11,  col:15>  'int'  '+'                    |-­‐ImplicitCastExpr  0x272c518  <col:11>  'int'  <LValueToRValue>                    |  `-­‐DeclRefExpr  0x272c4c8  <col:11>  'int'  lvalue  Var  0x272c390  'a'  'int'                    `-­‐ImplicitCastExpr  0x272c530  <col:15>  'int'  <LValueToRValue>                        `-­‐DeclRefExpr  0x272c4f0  <col:15>  'int'  lvalue  Var  0x272c420  'b'  'int' Hello.ast $ clang -cc1 -ast-dump Simple.c ;  ModuleID  =  'Simple.c' target  datalayout  =  "e-­‐p:64:64:64-­‐i1:8:8-­‐i8:8:8-­‐ i16:16:16-­‐i32:32:32-­‐i64:64:64-­‐f32:32:32-­‐f64:64:64-­‐ v64:64:64-­‐v128:128:128-­‐a0:0:64-­‐s0:64:64-­‐ f80:128:128-­‐n8:16:32:64-­‐S128" target  triple  =  "x86_64-­‐pc-­‐linux-­‐gnu" ;  Function  Attrs:  nounwind define  i32  @main()  #0  {    %1  =  alloca  i32,  align  4    %a  =  alloca  i32,  align  4    %b  =  alloca  i32,  align  4    store  i32  0,  i32*  %1    store  i32  55,  i32*  %a,  align  4    store  i32  66,  i32*  %b,  align  4    %2  =  load  i32*  %a,  align  4    %3  =  load  i32*  %b,  align  4    %4  =  add  nsw  i32  %2,  %3    ret  i32  %4 } attributes  #0  =  {  nounwind  "less-­‐precise-­‐ fpmad"="false"  "no-­‐frame-­‐pointer-­‐elim"="false"   "no-­‐infs-­‐fp-­‐math"="false"  "no-­‐nans-­‐fp-­‐ math"="false"  "no-­‐realign-­‐stack"  "stack-­‐protector-­‐ buffer-­‐size"="8"  "unsafe-­‐fp-­‐math"="false"  "use-­‐ soft-­‐float"="false"  } !llvm.ident  =  !{!0} Simple.ll $ clang -S -emit-llvm Simple.c
  13. Who’s next? Intermediate Representation Optimized I.R. Target Code Optimized Target

    Code IR Optimization Target Code Optimization Target Code Generation … … (Hello.asm) Hello.c
  14. .686P INCLUDELIB MSVCRT INCLUDELIB OLDNAMES _DATA SEGMENT $SG2940 DB! 'Hello,

    world!', 00H $SG2941 DB! '%s', 0aH, 00H _DATA ENDS PUBLIC _main EXTERN _printf:PROC _TEXT SEGMENT _main PROC push ebp mov ebp, esp push OFFSET $SG2940 push OFFSET $SG2941 call _printf mov esp, ebp pop ebp ret 0 _main! ENDP _TEXT! ENDS END Hello.asm Assembly Target Code PE Object File ?
  15. Segmentation MZ Header .text .data PE Header … .idata …

    reserved … .text .data heap stack … kernel space Executable Image File Process Virtual Memory Space Disk Storage Hello.exe Page #1 Page #2 Page #3 Page #4 Page #220 … Physical Memory Pages …
  16. .686P INCLUDELIB MSVCRT INCLUDELIB OLDNAMES _DATA SEGMENT $SG2940 DB! 'Hello,

    world!', 00H $SG2941 DB! '%s', 0aH, 00H _DATA ENDS PUBLIC _main EXTERN _printf:PROC _TEXT SEGMENT _main PROC push ebp mov ebp, esp push OFFSET $SG2940 push OFFSET $SG2941 call _printf mov esp, ebp pop ebp ret 0 _main! ENDP _TEXT! ENDS END Hello.asm [.text] _main: 00000000: 55 push ebp 00000001: 8B EC mov ebp,esp 00000003: 68 00 00 00 00 push offset $SG2940 00000008: 68 00 00 00 00 push offset $SG2941 0000000D: E8 00 00 00 00 call _printf 00000012: 8B E5 mov esp,ebp 00000014: 5D pop ebp 00000015: C3 ret Symbol Symbol Offset Type Applied To Index Name -------- ---------------- ----------------- -------- ------ 00000004 DIR32 00000000 B $SG2940 00000009 DIR32 00000000 C $SG2941 0000000E REL32 00000000 A _printf [.data] 00000000: 4865 6C6C 6F2C 2077 6F72 6C64 2100 2573 Hello, world!.%s 00000010: 0A00 .. [.debug$S] … [.drectve] /DEFAULTLIB:MSVCRT /DEFAULTLIB:OLDNAMES [COFF SYMBOL TABLE] 00000000 SECT1 notype Static | .data 00000000 SECT2 notype Static | .text 00000000 SECT3 notype Static | .debug$S 00000000 SECT4 notype Static | .drectve 00000000 UNDEF notype External | _printf 00000000 SECT1 notype Static | $SG2940 0000000E SECT1 notype Static | $SG2941 Hello.obj Assembly Target Code PE Object File COFF Header .text .data … COFF Symtab .drectve
  17. Static Linking 1. Address & storage allocation 2. Symbol resolution

    3. Relocation COFF Header .text .data … COFF Symtab .drectve Relocatable Object File A COFF Header .text .data … COFF Symtab .drectve Relocatable Object File B COFF Header .text .data … COFF Symtab .drectve Relocatable Object File C MZ Header .text .data PE Header … .idata … Executable Object File + + = a.obj b.obj c.obj Hello.exe In contrast to the compiler and assembler, Linkers should have minimal understanding of the target machine.
  18. Symbols 1. Global symbols — can be referenced by other

    modules 2. External global symbols — defined by some other module 3. Local symbols — defined and referenced exclusively extern int shared; int add(int, int); int a = 0x689; int main() { return add(a, shared); } a.c int shared = 0x92; static int private; int add(int x, int y) { return x + y; } b.c External global Global Global Global Local Global
  19. extern int shared; int add(int, int); int a = 0x689;

    int main() { return add(a, shared); } a.c [.text] _main: 00000000: 55 push ebp 00000001: 8B EC mov ebp,esp 00000003: A1 00 00 00 00 mov eax,dword ptr [_shared] 00000008: 50 push eax 00000009: 8B 0D 00 00 00 00 mov ecx,dword ptr [_a] 0000000F: 51 push ecx 00000010: E8 00 00 00 00 call _add 00000015: 83 C4 08 add esp,8 00000018: 5D pop ebp 00000019: C3 ret Symbol Symbol Offset Type Applied To Index Name -------- -------- ------------ -------- ------ 00000004 DIR32 00000000 D _shared 0000000B DIR32 00000000 8 _a 00000011 REL32 00000000 B _add [.data] 00000000: 89 06 00 00 [COFF SYMBOL TABLE] 02 00000000 SECT1 notype Static | .drectve 06 00000000 SECT3 notype Static | .data 08 00000000 SECT3 notype External | _a 09 00000000 SECT4 notype Static | .text 0B 00000000 UNDEF notype () External | _add 0C 00000000 SECT4 notype () External | _main 0D 00000000 UNDEF notype External | _shared a.obj int shared = 0x92; static int private; int add(int x, int y) { return x + y; } b.c [.text] _add: 00000000: 55 push ebp 00000001: 8B EC mov ebp,esp 00000003: 8B 45 08 mov eax,dword ptr [ebp+8] 00000006: 8B 00 mov eax,dword ptr [eax] 00000008: 8B 4D 0C mov ecx,dword ptr [ebp+0Ch] 0000000B: 03 01 add eax,dword ptr [ecx] 0000000D: 5D pop ebp 0000000E: C3 ret [.data] 00000000: 92 00 00 00 [COFF SYMBOL TABLE] 02 00000000 SECT1 notype Static | .drectve 04 00000000 SECT2 notype Static | .debug$S 06 00000000 SECT3 notype Static | .data 08 00000000 SECT3 notype External | _shared 09 00000000 SECT4 notype Static | .text 0B 00000000 SECT4 notype () External | _add b.obj
  20. [OPTIONAL HEADER VALUES] … 400000 image base (00400000 to 00404FFF)

    1000 section alignment 1298 entry point (00401298) … [.text : 00401000–00401869] _main: 00401000: 55 push ebp 00401001: 8B EC mov ebp,esp 00401003: A1 04 30 40 00 mov eax,dword ptr ds:[00403004h] 00401008: 50 push eax 00401009: 8B 0D 00 30 40 00 mov ecx,dword ptr ds:[00403000h] 0040100F: 51 push ecx 00401010: E8 0B 00 00 00 call 00401020 00401015: 83 C4 08 add esp,8 00401018: 5D pop ebp 00401019: C3 ret … _add: 00401020: 55 push ebp 00401021: 8B EC mov ebp,esp 00401023: 8B 45 08 mov eax,dword ptr [ebp+8] 00401026: 8B 00 mov eax,dword ptr [eax] 00401028: 8B 4D 0C mov ecx,dword ptr [ebp+0Ch] 0040102B: 03 01 add eax,dword ptr [ecx] 0040102D: 5D pop ebp 0040102E: C3 ret … 00401298: E8 B4 01 00 00 call 00401451 0040129D: E9 91 FE FF FF jmp 00401133 … [.data : 00403000–00403383] 00403000: 89 06 00 00 92 00 00 00 01 00 00 00 00 00 00 00 00403010: FE FF FF FF FF FF FF FF 4E E6 40 BB B1 19 BF 44 00403020: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 … ab.exe int shared = 0x92; int add(int x, int y) { return x + y; } b.c [.text] _add: 00000000: 55 push ebp 00000001: 8B EC mov ebp,esp 00000003: 8B 45 08 mov eax,dword ptr [ebp+8] 00000006: 8B 00 mov eax,dword ptr [eax] 00000008: 8B 4D 0C mov ecx,dword ptr [ebp+0Ch] 0000000B: 03 01 add eax,dword ptr [ecx] 0000000D: 5D pop ebp 0000000E: C3 ret b.obj extern int shared; int add(int, int); int a = 0x689; int main() { return add(a, shared); } a.c [.text] _main: 00000000: 55 push ebp 00000001: 8B EC mov ebp,esp 00000003: A1 00 00 00 00 mov eax,dword ptr [_shared] 00000008: 50 push eax 00000009: 8B 0D 00 00 00 00 mov ecx,dword ptr [_a] 0000000F: 51 push ecx 00000010: E8 00 00 00 00 call _add 00000015: 83 C4 08 add esp,8 00000018: 5D pop ebp 00000019: C3 ret a.obj + CL.exe
  21. MSVC Runtime Static Library Dynamic Library Category Compiler Arg. libcmt.lib

    - Multi-Thread, Static /MT msvcrt.lib msvcr110.dll Multi-Thread, Dynamic /MD libcmtd.lib - Multi-Thread, Static, Debug /MTd msvcrtd.lib msvcr110d.dll Multi-Thread, Dynamic, Debug /MDd msvcmrt.lib msvcm110.dll Managed / Unmanaged Hybrid /CLR msvcurt.lib msvcm110.dll Managed /CLR:PURE … ntdll!NtRequestWaitReplyPort kernel32!ConsoleClientCallServer kernel32!WriteConsoleInternal kernel32!WriteFileImplementation Hello!_write_no_lock Hello!_write Hello!_flush Hello!_ftbuf Hello!printf Hello!main Hello!__tmainCRTStartup kernel32!BaseThreadInitThunk ntdll!__RtlUserThreadStart ntdll!_RtlUserThreadStart … Windows Windows CRT CRT Hello.c
  22. 00fd1016 ff259020fd00 jmp dword ptr [Hello+0x2090 (00fd2090)] Hello.exe msvcr110.dll …

    … ntdll.dll … … … Hello.exe Process Virtual Memory Space 00fd1000 55 push ebp 00fd1001 8bec mov ebp,esp 00fd1003 680030fd00 push offset Hello+0x3000 (00fd3000) 00fd1008 680e30fd00 push offset Hello+0x300e (00fd300e) 00fd100d e804000000 call Hello+0x1016 (00fd1016) 00fd1012 8be5 mov esp,ebp 00fd1014 5d pop ebp 00fd1015 c3 ret … 00fd2090 00002248 00fd2248 20067072 696e7466 00004d53 56435231 .printf..MSVCR1 00fd2258 31302e64 6c6c00 10.dll. 00fd2090 6af8d1d5 MSVCR110!printf: 6af8d1d5 6a0c push 0Ch 6af8d1d7 6888d2f86a push 6af8d288 (offset MSVCR110!_CT??_R0?AV…) 6af8d1dc e8371af9ff call MSVCR110!_SEH_prolog4 (6af1ec18) 6af8d1e1 33c0 xor eax,eax Dynamic Linking: MSVCR110!printf Before Relocation After Relocation 00fd0000 00fd5000 6af10000 6afe2000 76dc0000 76f01000 (RVA) (Relative CALL)
  23. Double-Click Hello.exe SHELL32! CExecuteAssociation::Execute kernel32!CreateProcessW ntdll!NtCreateUserProcess nt!ZwOpenFile nt!ZwCreateSection nt!PspAllocateProcess nt!PspAllocateThread

    nt!PspUserThreadStartup nt!NtTerminateProcess ntdll!LoadDll ntdll!LdrpProcessStaticImports Hello!mainCRTStartup msvcrt11d!exit kernel32!ExitProcessStub ntdll!RtlExitUserProcess Open & validate executable image file Allocate process resources Launch loader Handle dynamic linking Queue user-mode APC Initialize C runtime (msvcrtd.lib) Hello!main Clean up process resources Explorer.exe Hello.exe Windows Kernel
  24. • jserv, “How a Compiler Works: GNU Toolchain” • ຂ৴੒

    “Compiling a Compiler” • Kito Cheng, ʬᕆஊฤᩄث࠷ՂԽٕज़ʭ • MSDN, “Peering Inside the PE: A Tour of the Win32 Portable Executable File Format” • Alexander Sotirov, “Tiny PE: Creating the smallest possible PE executable” • Randal E. Bryant & David R. O'Hallaron, “Computer Systems: A Programmer's Perspective” w 俞ߕࢠɺੴຌɺᖊѪຽ ʬఔࣜઃܭࢣతࣗզमཆɿ࿈݁ɺࡌೖɺఔࣜݿʭ http://csapp.cs.cmu.edu/ http://www.tenlong.com.tw/items/9861818286?item_id=53897 http://www.phreedom.org/research/tinype/ https://msdn.microsoft.com/en-us/library/ms809762.aspx http://www.slideshare.net/kitocheng/ss-42438227 https://wiki.sars.tw/doku.php?id=programming:compiling_a_compiler http://www.slideshare.net/jserv/how-a-compiler-works-gnu-toolchain Further Reading