Symbolic Debugging with DWARF

Debugging  Debuggers!! Samy Al Bahra / @0xF390

Symbolic Debugging The wc program tells you the number of ‘w’ characters in a ﬁle, terribly. static unsigned int count(char *buffer, size_t n) { unsigned int sum = 0; size_t i; for (i = 0; i < n; i++) sum += buffer[i] == 'w'; return sum; }

Symbolic Debugging The C program is compiled and linked into an executable ﬁle in a format the underlying operating system understands. $ gcc -o wc wc.c -O2 -ggdb $ ./wc /etc/passwd 16 The executable format consists of various sections that tell the kernel how the executable should be loaded as well as executable code.

Symbolic Debugging ELF for Executable and Linking Format sbahra@broadwell:~/projects/backtrace/bangbangcon$ readelf -l ./wc Elf file type is EXEC (Executable file) Entry point 0x400800 There are 9 program headers, starting at offset 64 Program Headers: 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 0x0000000000000b24 0x0000000000000b24 R E 200000 LOAD 0x0000000000000e08 0x0000000000600e08 0x0000000000600e08 0x0000000000000270 0x0000000000000288 RW 200000 DYNAMIC 0x0000000000000e28 0x0000000000600e28 0x0000000000600e28 0x00000000000001d0 0x00000000000001d0 RW 8 NOTE 0x0000000000000254 0x0000000000400254 0x0000000000400254 0x0000000000000044 0x0000000000000044 R 4 GNU_EH_FRAME 0x00000000000009b8 0x00000000004009b8 0x00000000004009b8 0x000000000000003c 0x000000000000003c R 4 GNU_STACK 0x0000000000000000 0x0000000000000000 0x0000000000000000 0x0000000000000000 0x0000000000000000 RW 10 GNU_RELRO 0x0000000000000e08 0x0000000000600e08 0x0000000000600e08 0x00000000000001f8 0x00000000000001f8 R 1

Symbolic Debugging static unsigned int count(char *buffer, size_t n) { unsigned int sum = 0; size_t i; for (i = 0; i < n; i++) sum += buffer[i] == ‘w'; return sum; } Dump of assembler code for function count: 0x00000000004008b0 <+0>: push %rbx 0x00000000004008b1 <+1>: lea 0x1000(%rdi),%rdx 0x00000000004008b8 <+8>: xor %ebx,%ebx 0x00000000004008ba <+10>: nopw 0x0(%rax,%rax,1) 0x00000000004008c0 <+16>: xor %eax,%eax 0x00000000004008c2 <+18>: cmpb $0x77,(%rdi) 0x00000000004008c5 <+21>: sete %al 0x00000000004008c8 <+24>: add $0x1,%rdi 0x00000000004008cc <+28>: add %eax,%ebx 0x00000000004008ce <+30>: cmp %rdx,%rdi 0x00000000004008d1 <+33>: jne 0x4008c0 <count+16> … 0x00000000004008d8 <+40>: mov %ebx,%eax 0x00000000004008da <+42>: pop %rbx 0x00000000004008db <+43>: retq End of assembler dump. C is compiled into executable code that interacts with memory and a limited set of registers.

Symbolic Debugging A symbolic debugger is able to map the state of registers and memory to a backtrace with variables and type information. sbahra$ bt `pgrep wc` PID: 8870 -------------------------------------------------------------------------------- Thread 8870 [ 0] libc-2.21.so pause (../sysdeps/unix/syscall-template.S:81)  … | 1| wc count (wc.c:19) buffer = (parameter) reference(0, 0x7ffce7b64e10) {char} -><> = string(0x7ffce7b64e10, 32, [root:x:0:0:root:/root:/bin/bash?]) sum = 0 i = -- n = (parameter) 4096 [ 2] wc main (wc.c:42)  … [ 3] libc-2.21.so __libc_start_main …

Symbolic Debugging The ﬁrst step is to extract register state to determine the current state of the process. NAME ptrace -- process tracing and debugging SYNOPSIS #include <sys/types.h> #include <sys/ptrace.h> int ptrace(int request, pid_t pid, caddr_t addr, int data); Available in various broken ways on Linux and the BSD family of operating systems.

Symbolic Debugging The %rip register tells us the currently executing instruction in a program. Thread 6870 -------------------------------------------------------------------------------- [ 0] libc-2.21.so pause (../sysdeps/unix/syscall-template.S:81) -------------------------------------------------------------------------------- rax=0xfffffffffffffdfe rdx=0x7ffce7b65e10 rcx=0x7ff9b15b5b70 rbx=0 rsi=0x7ff9b1ad5020 rdi=0x7ffce7b65e10 rbp=0x10b6010 rsp=0x7ffce7b64df8 r8=0x7ff9b1ad5020 r9=0 r10=0x3c1 r11=0x246 r12=0x4007b0 r13=0x7ffce7b65f10 r14=0 r15=0 rip=0x7ff9b15b5b70 cs=0x33 ss=0x2b ds=0 es=0 fs=0 gs=0 eflags=0x246

Symbolic Debugging The kernel provides a mechanism to determine what the process address space looks like. sbahra@broadwell:~/projects/backtrace/bangbangcon$ cat /proc/`pgrep wc`/maps 00400000-00401000 r-xp 00000000 08:02 6034972 /home/sbahra/projects/backtrace/bangbangcon/wc 00600000-00601000 r--p 00000000 08:02 6034972 /home/sbahra/projects/backtrace/bangbangcon/wc 00601000-00602000 rw-p 00001000 08:02 6034972 /home/sbahra/projects/backtrace/bangbangcon/wc 010b6000-010d7000 rw-p 00000000 00:00 0 [heap] 7ff9b14eb000-7ff9b16ab000 r-xp 00000000 08:02 5246298 /lib/x86_64-linux-gnu/libc-2.21.so 7ff9b16ab000-7ff9b18ab000 ---p 001c0000 08:02 5246298 /lib/x86_64-linux-gnu/libc-2.21.so 7ff9b18ab000-7ff9b18af000 r--p 001c0000 08:02 5246298 /lib/x86_64-linux-gnu/libc-2.21.so 7ff9b18af000-7ff9b18b1000 rw-p 001c4000 08:02 5246298 /lib/x86_64-linux-gnu/libc-2.21.so 7ff9b18b1000-7ff9b18b5000 rw-p 00000000 00:00 0 7ff9b18b5000-7ff9b18d9000 r-xp 00000000 08:02 5246292 /lib/x86_64-linux-gnu/ld-2.21.so 7ff9b1ab1000-7ff9b1ab4000 rw-p 00000000 00:00 0 7ff9b1ad5000-7ff9b1ad8000 rw-p 00000000 00:00 0 7ff9b1ad8000-7ff9b1ad9000 r--p 00023000 08:02 5246292 /lib/x86_64-linux-gnu/ld-2.21.so 7ff9b1ad9000-7ff9b1ada000 rw-p 00024000 08:02 5246292 /lib/x86_64-linux-gnu/ld-2.21.so 7ff9b1ada000-7ff9b1adb000 rw-p 00000000 00:00 0 7ffce7b48000-7ffce7b69000 rw-p 00000000 00:00 0 [stack] 7ffce7bf9000-7ffce7bfb000 r--p 00000000 00:00 0 [vvar] 7ffce7bfb000-7ffce7bfd000 r-xp 00000000 00:00 0 [vdso] ffffffffff600000-ffffffffff601000 r-xp 00000000 00:00 0 [vsyscall] rip=0x7ff9b15b5b70

Symbolic Debugging ELF objects contain various debug-related sections. Section Description .debug_line Maps memory addresses to line numbers. .debug_info Type, variable and function information. .debug_frame Unwinding information by memory address. .eh_frame Unwinding information by memory address. These sections and their contents are speciﬁed by the DWARF format.

[0x000000dc] Extended opcode 2: set Address to 0x4008b0 [0x000000e7] Special opcode 12: advance Address by 0 to 0x4008b0 and Line by 7 to 8 [0x000000e8] Special opcode 119: advance Address by 8 to 0x4008b8 and Line by 2 to 10 [0x000000e9] Special opcode 121: advance Address by 8 to 0x4008c0 and Line by 4 to 14 [0x000000ea] Special opcode 200: advance Address by 14 to 0x4008ce and Line by -1 to 13 [0x000000eb] Special opcode 78: advance Address by 5 to 0x4008d3 and Line by 3 to 16 [0x000000ec] Special opcode 77: advance Address by 5 to 0x4008d8 and Line by 2 to 18 [0x000000ed] Advance PC by 4 to 0x4008dc [0x000000ef] Extended opcode 1: End of Sequence [0x000000f2] Extended opcode 2: set Address to 0x4006c0 [0x000000fd] Advance Line by 21 to 22 [0x000000ff] Copy [0x00000100] Special opcode 131: advance Address by 9 to 0x4006c9 and Line by 0 to 22 [0x00000101] Advance PC by constant 17 to 0x4006da [0x00000102] Special opcode 38: advance Address by 2 to 0x4006dc and Line by 5 to 27 [0x00000103] Extended opcode 4: set Discriminator to 1 [0x00000107] Set is_stmt to 0 [0x00000108] Special opcode 75: advance Address by 5 to 0x4006e1 and Line by 0 to 27 Symbolic Debugging .debug_line contains a sequence of operations that are executed by a state machine to form a giant matrix.

[0x00000102] Special opcode 38: advance Address by 2 to 0x4006dc and Line by 5 to 27 [0x00000103] Extended opcode 4: set Discriminator to 1 [0x00000107] Set is_stmt to 0 [0x00000108] Special opcode 75: advance Address by 5 to 0x4006e1 and Line by 0 to 27 [0x00000109] Set is_stmt to 1 [0x0000010a] Special opcode 136: advance Address by 9 to 0x4006ea and Line by 5 to 32 [0x0000010b] Special opcode 188: advance Address by 13 to 0x4006f7 and Line by 1 to 33 [0x0000010c] Special opcode 137: advance Address by 9 to 0x400700 and Line by 6 to 39 [0x0000010d] Set File Name to entry 2 in the File Name Table [0x0000010f] Advance Line by 224 to 263 [0x00000112] Special opcode 145: advance Address by 10 to 0x40070a and Line by 0 to 263 [0x00000113] Set File Name to entry 1 in the File Name Table [0x00000115] Advance Line by -225 to 38 [0x00000118] Special opcode 229: advance Address by 16 to 0x40071a and Line by 0 to 38 [0x00000119] Set File Name to entry 2 in the File Name Table [0x0000011b] Advance Line by 66 to 104 [0x0000011e] Special opcode 75: advance Address by 5 to 0x40071f and Line by 0 to 104 [0x0000011f] Set File Name to entry 1 in the File Name Table [0x00000121] Advance Line by -61 to 43 [0x00000123] Advance PC by constant 17 to 0x400730 [0x00000124] Special opcode 5: advance Address by 0 to 0x400730 and Line by 0 to 43 [0x00000125] Set File Name to entry 2 in the File Name Table [0x00000127] Advance Line by 54 to 97 [0x00000129] Advance PC by constant 17 to 0x400741 [0x0000012a] Special opcode 201: advance Address by 14 to 0x40074f and Line by 0 to 97 [0x0000012b] Set File Name to entry 1 in the File Name Table [0x0000012d] Advance Line by -68 to 29 [0x00000130] Advance PC by constant 17 to 0x400760 [0x00000131] Special opcode 145: advance Address by 10 to 0x40076a and Line by 0 to 29 [0x00000132] Advance Line by 14 to 43 [0x00000134] Special opcode 145: advance Address by 10 to 0x400774 and Line by 0 to 43 [0x00000135] Advance Line by -9 to 34 [0x00000137] Special opcode 75: advance Address by 5 to 0x400779 and Line by 0 to 34 [0x00000138] Special opcode 146: advance Address by 10 to 0x400783 and Line by 1 to 35 [0x00000139] Advance PC by 10 to 0x40078d Symbolic Debugging .debug_line contains a sequence of operations that are executed by a state machine to form a giant matrix.

Symbolic Debugging .debug_line contains a sequence of operations that are executed by a state machine to form a giant matrix. Address Source 0x4000 wc.c:80 0x4008 wc.c:20 0x400d wc.c:19 0x401a wc.c:10

00000090 000000000000002c 00000064 FDE cie=00000030 pc=00000000004006c0..000000000040078d DW_CFA_advance_loc: 1 to 00000000004006c1 DW_CFA_def_cfa_offset: 16 DW_CFA_offset: r6 (rbp) at cfa-16 DW_CFA_advance_loc: 1 to 00000000004006c2 DW_CFA_def_cfa_offset: 24 DW_CFA_offset: r3 (rbx) at cfa-24 DW_CFA_advance_loc: 7 to 00000000004006c9 DW_CFA_def_cfa_offset: 4144 DW_CFA_advance_loc1: 131 to 000000000040074c DW_CFA_remember_state DW_CFA_def_cfa_offset: 24 DW_CFA_advance_loc: 1 to 000000000040074d DW_CFA_def_cfa_offset: 16 DW_CFA_advance_loc: 1 to 000000000040074e DW_CFA_def_cfa_offset: 8 DW_CFA_advance_loc: 1 to 000000000040074f DW_CFA_restore_state Symbolic Debugging .debug_frame or .eh_frame contain a sequence of operations that are executed by a state machine to form a giant matrix of register states.

DW_CFA_advance_loc: 1 to 000000000040074e DW_CFA_def_cfa_offset: 8 DW_CFA_advance_loc: 1 to 000000000040074f DW_CFA_restore_state DW_CFA_nop DW_CFA_nop DW_CFA_nop DW_CFA_nop 000000c0 0000000000000044 00000094 FDE cie=00000030 pc=00000000004008e0..0000000000400945 DW_CFA_advance_loc: 2 to 00000000004008e2 DW_CFA_def_cfa_offset: 16 DW_CFA_offset: r15 (r15) at cfa-16 DW_CFA_advance_loc: 2 to 00000000004008e4 DW_CFA_def_cfa_offset: 24 DW_CFA_offset: r14 (r14) at cfa-24 DW_CFA_advance_loc: 5 to 00000000004008e9 DW_CFA_def_cfa_offset: 32 DW_CFA_offset: r13 (r13) at cfa-32 DW_CFA_advance_loc: 2 to 00000000004008eb DW_CFA_def_cfa_offset: 40 DW_CFA_offset: r12 (r12) at cfa-40 DW_CFA_advance_loc: 8 to 00000000004008f3 DW_CFA_def_cfa_offset: 48 DW_CFA_offset: r6 (rbp) at cfa-48 DW_CFA_advance_loc: 8 to 00000000004008fb DW_CFA_def_cfa_offset: 56 DW_CFA_offset: r3 (rbx) at cfa-56 DW_CFA_advance_loc: 15 to 000000000040090a DW_CFA_def_cfa_offset: 64 DW_CFA_advance_loc: 48 to 000000000040093a DW_CFA_def_cfa_offset: 56 DW_CFA_advance_loc: 1 to 000000000040093b DW_CFA_def_cfa_offset: 48 DW_CFA_advance_loc: 1 to 000000000040093c DW_CFA_def_cfa_offset: 40 Symbolic Debugging .debug_frame or .eh_frame contain a sequence of operations that are executed by a state machine to form a giant matrix of register states.

Symbolic Debugging .debug_frame or .eh_frame contain a sequence of operations that are executed by a state machine to form a giant matrix of register states. 000000d0 0000000000000044 000000a4 FDE cie=00000030 pc=4005e0..400645 LOC CFA rbx rbp r12 r13 r14 r15 ra 00000000004005e0 rsp+8 u u u u u u c-8 00000000004005e2 rsp+16 u u u u u c-16 c-8 00000000004005e4 rsp+24 u u u u c-24 c-16 c-8 00000000004005e9 rsp+32 u u u c-32 c-24 c-16 c-8 00000000004005eb rsp+40 u u c-40 c-32 c-24 c-16 c-8 00000000004005f3 rsp+48 u c-48 c-40 c-32 c-24 c-16 c-8 00000000004005fb rsp+56 c-56 c-48 c-40 c-32 c-24 c-16 c-8 000000000040060a rsp+64 c-56 c-48 c-40 c-32 c-24 c-16 c-8 000000000040063a rsp+56 c-56 c-48 c-40 c-32 c-24 c-16 c-8 000000000040063b rsp+48 c-56 c-48 c-40 c-32 c-24 c-16 c-8 000000000040063c rsp+40 c-56 c-48 c-40 c-32 c-24 c-16 c-8 000000000040063e rsp+32 c-56 c-48 c-40 c-32 c-24 c-16 c-8 0000000000400640 rsp+24 c-56 c-48 c-40 c-32 c-24 c-16 c-8 0000000000400642 rsp+16 c-56 c-48 c-40 c-32 c-24 c-16 c-8

Symbolic Debugging .debug_info speciﬁes how the program is structured. It includes type, variable and other structural information.

Symbolic Debugging DW_TAG_compile_unit DW_AT_name = wc.c DW_AT_ranges = A .. B DW_TAG_subprogram DW_AT_name = count DW_AT_type = DW_TAG_formal_parameter DW_AT_name = buffer DW_AT_type = DW_TAG_variable DW_AT_name = sum DW_AT_type = DW_TAG_variable DW_AT_name = i DW_AT_type = DW_TAG_typedef DW_AT_name = size_t DW_AT_type = DW_TAG_base_type DW_AT_byte_size = 8 DW_AT_encoding = unsigned DW_AT_name = “long unsigned”

Symbolic Debugging DWARF must support all kinds of aggressive compiler optimizations and has a complex expression state machine for expressing variable values. static unsigned int count(char *buffer, size_t n) { unsigned int sum = 0; size_t i; for (i = 0; i < n; i++) sum += buffer[i] == ‘w'; return sum; } Dump of assembler code for function count: 0x00000000004008b0 <+0>: push %rbx 0x00000000004008b1 <+1>: lea 0x1000(%rdi),%rdx 0x00000000004008b8 <+8>: xor %ebx,%ebx 0x00000000004008ba <+10>: nopw 0x0(%rax,%rax,1) 0x00000000004008c0 <+16>: xor %eax,%eax 0x00000000004008c2 <+18>: cmpb $0x77,(%rdi) 0x00000000004008c5 <+21>: sete %al 0x00000000004008c8 <+24>: add $0x1,%rdi 0x00000000004008cc <+28>: add %eax,%ebx 0x00000000004008ce <+30>: cmp %rdx,%rdi 0x00000000004008d1 <+33>: jne 0x4008c0 <count+16> … 0x00000000004008d8 <+40>: mov %ebx,%eax 0x00000000004008da <+42>: pop %rbx 0x00000000004008db <+43>: retq End of assembler dump.

Symbolic Debugging < 2><0x00000361> DW_TAG_variable DW_AT_name "i" DW_AT_decl_file 0x00000001 /home/sbahra/projects/backtrace/bangbangcon/wc.c DW_AT_decl_line 0x0000000b DW_AT_type <0x00000029> … < 1><0x0000036b> DW_TAG_subprogram DW_AT_abstract_origin <0x00000332> DW_AT_low_pc 0x004008b0 DW_AT_high_pc <offset-from-lowpc>44 … < 2><0x0000038f> DW_TAG_variable DW_AT_abstract_origin <0x00000356> DW_AT_location <loclist with 3 entries follows> 4008b0 4008ba (DW_OP_lit0; DW_OP_stack_value) 4008ba 4008cc (DW_OP_breg5 (rdi): 0; DW_OP_breg1 (rdx): 0; DW_OP_minus; DW_OP_plus_uconst: 4096; DW_OP_stack_value) 4008cc 4008ce (DW_OP_breg5 (rdi): 0; DW_OP_breg1 (rdx): 0; DW_OP_minus; DW_OP_plus_uconst: 4095; DW_OP_stack_value) Dump of assembler code for function count: 0x00000000004008b0 <+0>: push %rbx 0x00000000004008b1 <+1>: lea 0x1000(%rdi),%rdx 0x00000000004008b8 <+8>: xor %ebx,%ebx 0x00000000004008ba <+10>: nopw 0x0(%rax,%rax,1) 0x00000000004008c0 <+16>: xor %eax,%eax 0x00000000004008c2 <+18>: cmpb $0x77,(%rdi) 0x00000000004008c5 <+21>: sete %al 0x00000000004008c8 <+24>: add $0x1,%rdi 0x00000000004008cc <+28>: add %eax,%ebx 0x00000000004008ce <+30>: cmp %rdx,%rdi 0x00000000004008d1 <+33>: jne 0x4008c0 <count+16> … 0x00000000004008d8 <+40>: mov %ebx,%eax 0x00000000004008da <+42>: pop %rbx 0x00000000004008db <+43>: retq End of assembler dump.

Symbolic Debugging < 2><0x00000361> DW_TAG_variable DW_AT_name "i" DW_AT_decl_file 0x00000001 /home/sbahra/projects/backtrace/bangbangcon/wc.c DW_AT_decl_line 0x0000000b DW_AT_type <0x00000029> … < 1><0x0000036b> DW_TAG_subprogram DW_AT_abstract_origin <0x00000332> DW_AT_low_pc 0x004008b0 DW_AT_high_pc <offset-from-lowpc>44 … < 2><0x0000038f> DW_TAG_variable DW_AT_abstract_origin <0x00000356> DW_AT_location <loclist with 3 entries follows> 4008b0 4008ba (DW_OP_lit0; DW_OP_stack_value) 4008ba 4008cc (DW_OP_breg5 (rdi): 0; DW_OP_breg1 (rdx): 0; DW_OP_minus; DW_OP_plus_uconst: 4096; DW_OP_stack_value) 4008cc 4008ce (DW_OP_breg5 (rdi): 0; DW_OP_breg1 (rdx): 0; DW_OP_minus; DW_OP_plus_uconst: 4095; DW_OP_stack_value) Dump of assembler code for function count: 0x00000000004008b0 <+0>: push %rbx 0x00000000004008b1 <+1>: lea 0x1000(%rdi),%rdx 0x00000000004008b8 <+8>: xor %ebx,%ebx 0x00000000004008ba <+10>: nopw 0x0(%rax,%rax,1) 0x00000000004008c0 <+16>: xor %eax,%eax 0x00000000004008c2 <+18>: cmpb $0x77,(%rdi) 0x00000000004008c5 <+21>: sete %al 0x00000000004008c8 <+24>: add $0x1,%rdi 0x00000000004008cc <+28>: add %eax,%ebx 0x00000000004008ce <+30>: cmp %rdx,%rdi 0x00000000004008d1 <+33>: jne 0x4008c0 <count+16> … 0x00000000004008d8 <+40>: mov %ebx,%eax 0x00000000004008da <+42>: pop %rbx 0x00000000004008db <+43>: retq End of assembler dump. buffer + 1

Symbolic Debugging < 2><0x00000361> DW_TAG_variable DW_AT_name "i" DW_AT_decl_file 0x00000001 /home/sbahra/projects/backtrace/bangbangcon/wc.c DW_AT_decl_line 0x0000000b DW_AT_type <0x00000029> … < 1><0x0000036b> DW_TAG_subprogram DW_AT_abstract_origin <0x00000332> DW_AT_low_pc 0x004008b0 DW_AT_high_pc <offset-from-lowpc>44 … < 2><0x0000038f> DW_TAG_variable DW_AT_abstract_origin <0x00000356> DW_AT_location <loclist with 3 entries follows> 4008b0 4008ba (DW_OP_lit0; DW_OP_stack_value) 4008ba 4008cc (DW_OP_breg5 (rdi): 0; DW_OP_breg1 (rdx): 0; DW_OP_minus; DW_OP_plus_uconst: 4096; DW_OP_stack_value) 4008cc 4008ce (DW_OP_breg5 (rdi): 0; DW_OP_breg1 (rdx): 0; DW_OP_minus; DW_OP_plus_uconst: 4095; DW_OP_stack_value) Dump of assembler code for function count: 0x00000000004008b0 <+0>: push %rbx 0x00000000004008b1 <+1>: lea 0x1000(%rdi),%rdx 0x00000000004008b8 <+8>: xor %ebx,%ebx 0x00000000004008ba <+10>: nopw 0x0(%rax,%rax,1) 0x00000000004008c0 <+16>: xor %eax,%eax 0x00000000004008c2 <+18>: cmpb $0x77,(%rdi) 0x00000000004008c5 <+21>: sete %al 0x00000000004008c8 <+24>: add $0x1,%rdi 0x00000000004008cc <+28>: add %eax,%ebx 0x00000000004008ce <+30>: cmp %rdx,%rdi 0x00000000004008d1 <+33>: jne 0x4008c0 <count+16> … 0x00000000004008d8 <+40>: mov %ebx,%eax 0x00000000004008da <+42>: pop %rbx 0x00000000004008db <+43>: retq End of assembler dump. buffer + 1 buffer + 4096

Symbolic Debugging < 2><0x00000361> DW_TAG_variable DW_AT_name "i" DW_AT_decl_file 0x00000001 /home/sbahra/projects/backtrace/bangbangcon/wc.c DW_AT_decl_line 0x0000000b DW_AT_type <0x00000029> … < 1><0x0000036b> DW_TAG_subprogram DW_AT_abstract_origin <0x00000332> DW_AT_low_pc 0x004008b0 DW_AT_high_pc <offset-from-lowpc>44 … < 2><0x0000038f> DW_TAG_variable DW_AT_abstract_origin <0x00000356> DW_AT_location <loclist with 3 entries follows> 4008b0 4008ba (DW_OP_lit0; DW_OP_stack_value) 4008ba 4008cc (DW_OP_breg5 (rdi): 0; DW_OP_breg1 (rdx): 0; DW_OP_minus; DW_OP_plus_uconst: 4096; DW_OP_stack_value) 4008cc 4008ce (DW_OP_breg5 (rdi): 0; DW_OP_breg1 (rdx): 0; DW_OP_minus; DW_OP_plus_uconst: 4095; DW_OP_stack_value) Dump of assembler code for function count: 0x00000000004008b0 <+0>: push %rbx 0x00000000004008b1 <+1>: lea 0x1000(%rdi),%rdx 0x00000000004008b8 <+8>: xor %ebx,%ebx 0x00000000004008ba <+10>: nopw 0x0(%rax,%rax,1) 0x00000000004008c0 <+16>: xor %eax,%eax 0x00000000004008c2 <+18>: cmpb $0x77,(%rdi) 0x00000000004008c5 <+21>: sete %al 0x00000000004008c8 <+24>: add $0x1,%rdi 0x00000000004008cc <+28>: add %eax,%ebx 0x00000000004008ce <+30>: cmp %rdx,%rdi 0x00000000004008d1 <+33>: jne 0x4008c0 <count+16> … 0x00000000004008d8 <+40>: mov %ebx,%eax 0x00000000004008da <+42>: pop %rbx 0x00000000004008db <+43>: retq End of assembler dump. buffer + 1 buffer + 4096 -

Symbolic Debugging < 2><0x00000361> DW_TAG_variable DW_AT_name "i" DW_AT_decl_file 0x00000001 /home/sbahra/projects/backtrace/bangbangcon/wc.c DW_AT_decl_line 0x0000000b DW_AT_type <0x00000029> … < 1><0x0000036b> DW_TAG_subprogram DW_AT_abstract_origin <0x00000332> DW_AT_low_pc 0x004008b0 DW_AT_high_pc <offset-from-lowpc>44 … < 2><0x0000038f> DW_TAG_variable DW_AT_abstract_origin <0x00000356> DW_AT_location <loclist with 3 entries follows> 4008b0 4008ba (DW_OP_lit0; DW_OP_stack_value) 4008ba 4008cc (DW_OP_breg5 (rdi): 0; DW_OP_breg1 (rdx): 0; DW_OP_minus; DW_OP_plus_uconst: 4096; DW_OP_stack_value) 4008cc 4008ce (DW_OP_breg5 (rdi): 0; DW_OP_breg1 (rdx): 0; DW_OP_minus; DW_OP_plus_uconst: 4095; DW_OP_stack_value) Dump of assembler code for function count: 0x00000000004008b0 <+0>: push %rbx 0x00000000004008b1 <+1>: lea 0x1000(%rdi),%rdx 0x00000000004008b8 <+8>: xor %ebx,%ebx 0x00000000004008ba <+10>: nopw 0x0(%rax,%rax,1) 0x00000000004008c0 <+16>: xor %eax,%eax 0x00000000004008c2 <+18>: cmpb $0x77,(%rdi) 0x00000000004008c5 <+21>: sete %al 0x00000000004008c8 <+24>: add $0x1,%rdi 0x00000000004008cc <+28>: add %eax,%ebx 0x00000000004008ce <+30>: cmp %rdx,%rdi 0x00000000004008d1 <+33>: jne 0x4008c0 <count+16> … 0x00000000004008d8 <+40>: mov %ebx,%eax 0x00000000004008da <+42>: pop %rbx 0x00000000004008db <+43>: retq End of assembler dump. -4095

Symbolic Debugging < 2><0x00000361> DW_TAG_variable DW_AT_name "i" DW_AT_decl_file 0x00000001 /home/sbahra/projects/backtrace/bangbangcon/wc.c DW_AT_decl_line 0x0000000b DW_AT_type <0x00000029> … < 1><0x0000036b> DW_TAG_subprogram DW_AT_abstract_origin <0x00000332> DW_AT_low_pc 0x004008b0 DW_AT_high_pc <offset-from-lowpc>44 … < 2><0x0000038f> DW_TAG_variable DW_AT_abstract_origin <0x00000356> DW_AT_location <loclist with 3 entries follows> 4008b0 4008ba (DW_OP_lit0; DW_OP_stack_value) 4008ba 4008cc (DW_OP_breg5 (rdi): 0; DW_OP_breg1 (rdx): 0; DW_OP_minus; DW_OP_plus_uconst: 4096; DW_OP_stack_value) 4008cc 4008ce (DW_OP_breg5 (rdi): 0; DW_OP_breg1 (rdx): 0; DW_OP_minus; DW_OP_plus_uconst: 4095; DW_OP_stack_value) Dump of assembler code for function count: 0x00000000004008b0 <+0>: push %rbx 0x00000000004008b1 <+1>: lea 0x1000(%rdi),%rdx 0x00000000004008b8 <+8>: xor %ebx,%ebx 0x00000000004008ba <+10>: nopw 0x0(%rax,%rax,1) 0x00000000004008c0 <+16>: xor %eax,%eax 0x00000000004008c2 <+18>: cmpb $0x77,(%rdi) 0x00000000004008c5 <+21>: sete %al 0x00000000004008c8 <+24>: add $0x1,%rdi 0x00000000004008cc <+28>: add %eax,%ebx 0x00000000004008ce <+30>: cmp %rdx,%rdi 0x00000000004008d1 <+33>: jne 0x4008c0 <count+16> … 0x00000000004008d8 <+40>: mov %ebx,%eax 0x00000000004008da <+42>: pop %rbx 0x00000000004008db <+43>: retq End of assembler dump. -4095 +4096

Symbolic Debugging < 2><0x00000361> DW_TAG_variable DW_AT_name "i" DW_AT_decl_file 0x00000001 /home/sbahra/projects/backtrace/bangbangcon/wc.c DW_AT_decl_line 0x0000000b DW_AT_type <0x00000029> … < 1><0x0000036b> DW_TAG_subprogram DW_AT_abstract_origin <0x00000332> DW_AT_low_pc 0x004008b0 DW_AT_high_pc <offset-from-lowpc>44 … < 2><0x0000038f> DW_TAG_variable DW_AT_abstract_origin <0x00000356> DW_AT_location <loclist with 3 entries follows> 4008b0 4008ba (DW_OP_lit0; DW_OP_stack_value) 4008ba 4008cc (DW_OP_breg5 (rdi): 0; DW_OP_breg1 (rdx): 0; DW_OP_minus; DW_OP_plus_uconst: 4096; DW_OP_stack_value) 4008cc 4008ce (DW_OP_breg5 (rdi): 0; DW_OP_breg1 (rdx): 0; DW_OP_minus; DW_OP_plus_uconst: 4095; DW_OP_stack_value) Dump of assembler code for function count: 0x00000000004008b0 <+0>: push %rbx 0x00000000004008b1 <+1>: lea 0x1000(%rdi),%rdx 0x00000000004008b8 <+8>: xor %ebx,%ebx 0x00000000004008ba <+10>: nopw 0x0(%rax,%rax,1) 0x00000000004008c0 <+16>: xor %eax,%eax 0x00000000004008c2 <+18>: cmpb $0x77,(%rdi) 0x00000000004008c5 <+21>: sete %al 0x00000000004008c8 <+24>: add $0x1,%rdi 0x00000000004008cc <+28>: add %eax,%ebx 0x00000000004008ce <+30>: cmp %rdx,%rdi 0x00000000004008d1 <+33>: jne 0x4008c0 <count+16> … 0x00000000004008d8 <+40>: mov %ebx,%eax 0x00000000004008da <+42>: pop %rbx 0x00000000004008db <+43>: retq End of assembler dump. -40951

Symbolic Debugging < 2><0x00000361> DW_TAG_variable DW_AT_name "i" DW_AT_decl_file 0x00000001 /home/sbahra/projects/backtrace/bangbangcon/wc.c DW_AT_decl_line 0x0000000b DW_AT_type <0x00000029> … < 1><0x0000036b> DW_TAG_subprogram DW_AT_abstract_origin <0x00000332> DW_AT_low_pc 0x004008b0 DW_AT_high_pc <offset-from-lowpc>44 … < 2><0x0000038f> DW_TAG_variable DW_AT_abstract_origin <0x00000356> DW_AT_location <loclist with 3 entries follows> 4008b0 4008ba (DW_OP_lit0; DW_OP_stack_value) 4008ba 4008cc (DW_OP_breg5 (rdi): 0; DW_OP_breg1 (rdx): 0; DW_OP_minus; DW_OP_plus_uconst: 4096; DW_OP_stack_value) 4008cc 4008ce (DW_OP_breg5 (rdi): 0; DW_OP_breg1 (rdx): 0; DW_OP_minus; DW_OP_plus_uconst: 4095; DW_OP_stack_value) Dump of assembler code for function count: 0x00000000004008b0 <+0>: push %rbx 0x00000000004008b1 <+1>: lea 0x1000(%rdi),%rdx 0x00000000004008b8 <+8>: xor %ebx,%ebx 0x00000000004008ba <+10>: nopw 0x0(%rax,%rax,1) 0x00000000004008c0 <+16>: xor %eax,%eax 0x00000000004008c2 <+18>: cmpb $0x77,(%rdi) 0x00000000004008c5 <+21>: sete %al 0x00000000004008c8 <+24>: add $0x1,%rdi 0x00000000004008cc <+28>: add %eax,%ebx 0x00000000004008ce <+30>: cmp %rdx,%rdi 0x00000000004008d1 <+33>: jne 0x4008c0 <count+16> … 0x00000000004008d8 <+40>: mov %ebx,%eax 0x00000000004008da <+42>: pop %rbx 0x00000000004008db <+43>: retq End of assembler dump. (gdb) info locals sum = 0 i = <optimized out> (gdb) break *0x4008cc Breakpoint 2 at 0x4008cc: file wc.c, line 14. (gdb) r 14 sum += buffer[i] == 'w'; (gdb) info locals sum = 0 i = 0

Symbolic Debugging Dump of assembler code for function count: 0x00000000004008b0 <+0>: push %rbx 0x00000000004008b1 <+1>: lea 0x1000(%rdi),%rdx 0x00000000004008b8 <+8>: xor %ebx,%ebx 0x00000000004008ba <+10>: nopw 0x0(%rax,%rax,1) 0x00000000004008c0 <+16>: xor %eax,%eax 0x00000000004008c2 <+18>: cmpb $0x77,(%rdi) 0x00000000004008c5 <+21>: sete %al 0x00000000004008c8 <+24>: add $0x1,%rdi 0x00000000004008cc <+28>: add %eax,%ebx 0x00000000004008ce <+30>: cmp %rdx,%rdi 0x00000000004008d1 <+33>: jne 0x4008c0 <count+16> … 0x00000000004008d8 <+40>: mov %ebx,%eax 0x00000000004008da <+42>: pop %rbx 0x00000000004008db <+43>: retq End of assembler dump. (gdb) info locals sum = 0 i = <optimized out> (gdb) break *0x4008cc Breakpoint 2 at 0x4008cc: file wc.c, line 14. (gdb) r 14 sum += buffer[i] == 'w'; (gdb) info locals sum = 0 i = 0 Though a debugger may show a value as optimized output, it may just be a side-effect of bad DWARF being emitted.

Symbolic Debugging Different compilers emit DWARF at varying levels of quality and accuracy. [ 1] wc.gcc count.constprop.0 buffer = (parameter) reference(-, 0x7ffe288e3990) {char} -><> = “wwwwww…” sum = 39 i = -- n = (parameter) 4096 [ 1] wc.clang count buffer = -- n = (parameter) 4096 sum = 0 i = 0 GCC 5.2.1 clang 3.6.2-1 WRONGMISSING DATA

Symbolic Debugging Different compilers emit DWARF at varying levels of quality and accuracy.

Symbolic Debugging Some languages are still early in their support for DWARF and emit no or wild debug information. if depth == 0 { trace_me() } a := 10 b := "foo" c := []int{3, 4, 5} var d [3]int e := map[string]int{"a": 10, "b": 5} f := make(chan bool, 5) g := [3]int{7, 8, 9}

Symbolic Debugging Some languages are still early in their support for DWARF and emit no or wild debug information. if depth == 0 { trace_me() } a := 10 b := "foo" c := []int{3, 4, 5} var d [3]int e := map[string]int{"a": 10, "b": 5} f := make(chan bool, 5) g := [3]int{7, 8, 9} a = 0 b = (unreadable could not read string at 0xa due to input/output error) c = []int len: 859531071680, cap: 859531071576, [int nil,(unreadable inpu (unreadable input/output error),(unreadable input/output error),(unreadab error),...+859531071675 more] d = [3]int [859531071680,859530965504,8] e = map[string]int [,...+4939120891025722184 more] f = chan bool 199/10 g = [3]int [0,0,4263753] h = "" i = "" j = map[string]int [,...+199 more] k = *struct main.poop {a: 8, b: 0, c: -9076996248722771462, d: "@EVx00x x00x00x00x00x00FVx00x00x00x00x00FV x00x00x00x00x00x00x00x00x00x00x00x00x00x00x00x00x00x00 00x00x00x00x00x00x00x00x00x00x00x00x00x00x00...+5650518 mor z = 0

Symbolic Debugging Some debuggers rely on system-facilities such as thread_db to crawl internal thread data structures. These data structures are easily corruptible. PID: 11756 -------------------------------------------------------------------------------- Thread 11756 [ 0] libpthread-2.21.so pthread_join (pthread_join.c:90) [ 1] crash crash_environment_create (crash.c:374) [ 2] crash main (crash.c:570) [ 3] libc-2.21.so __libc_start_main

Symbolic Debugging Some debuggers rely on system-facilities such as thread_db to crawl internal thread data structures. These data structures are easily corruptible. PID: 11756 -------------------------------------------------------------------------------- Thread 11756 [ 0] libpthread-2.21.so pthread_join (pthread_join.c:90) [ 1] crash crash_environment_create (crash.c:374) [ 2] crash main (crash.c:570) [ 3] libc-2.21.so __libc_start_main Thread 11757 [ 0] libpthread-2.21.so __libc_pause (../sysdeps/unix/syscall-template.S:81) | 1| crash inlined (crash.c:104) [ 2] crash crash_signal (crash.c:433) [ 3] libpthread-2.21.so 0x7f527f8f7d10 (../sysdeps/pthread/funlockfile.c:29) * [ 4] crash scenario_smash_fault (smash.c:16) ----------------------------------------------------------------------- Signal: [11] SIGSEGV (segmentation violation) Reason: SEGV_ACCERR (invalid permissions for object) Faulting address: 0x7f527f215000 ----------------------------------------------------------------------- [ 5] 0x4141414141414141

Symbolic Debugging Unfortunately, things do actually get optimized out. static unsigned int count(char *buffer, size_t n) { unsigned int sum = 0; size_t i; for (i = 0; i < n; i++) sum += buffer[i] == 'w'; return sum; } [ 1] wc.gcc count.constprop.0 buffer = (parameter) reference(0, 0x7ffe8ef83ac0) {char} -><> = string(0x7ffe8ef83ac0, 32, [root:x:0:0:root:/root:/bin/ bash?]) sum = 0 i = -- n = (parameter) 4096

Symbolic Debugging Unfortunately, things do actually get optimized out. static unsigned int count(char *buffer, size_t n) { unsigned int sum = 0; size_t i; for (i = 0; i < n; i++) sum += buffer[i] == 'w'; return sum; } static unsigned int count(char *buffer, size_t n) { unsigned int sum = 0; char *end = buffer + n; do { sum += *buffer == ‘w'; } while (++buffer != end); return sum; } i is optimized out by induction variable elimination.

Symbolic Debugging Unfortunately, things are actually non- recoverable. [ 0] libc-2.21.so pause (../sysdeps/unix/syscall-template.S:81) [ 1] wc.gcc count.constprop.0 buffer = (parameter) reference(0, 0x7ffe111f2310) {char} -><> = string(0x7ffe111f2310, 32, [root:x:0:0:root:/root:/bin/ bash?]) sum = 0 i = -- n = (parameter) 4096 [ 2] wc.gcc main (wc.c:43) argc = -- argv = -- Registers are extremely fast. The compiler’s register allocator tries to optimally allocate registers for variables.

Symbolic Debugging Since argc is never used after the call into count, the compiler doesn’t spill it from the register it lived in. [ 0] libc-2.21.so pause (../sysdeps/unix/syscall-template.S:81) [ 1] wc.gcc count.constprop.0 buffer = (parameter) reference(0, 0x7ffe111f2310) {char} -><> = string(0x7ffe111f2310, 32, [root:x:0:0:root:/root:/bin/ bash?]) sum = 0 i = -- n = (parameter) 4096 [ 2] wc.gcc main (wc.c:43) argc = -- argv = --

Symbolic Debugging The platform ABI determines which registers must be saved across function calls, this mandates spilling said register if used. 000000d0 0000000000000044 000000a4 FDE cie=00000030 pc=4005e0..400645 LOC CFA rbx rbp r12 r13 r14 r15 ra 00000000004005e0 rsp+8 u u u u u u c-8 00000000004005e2 rsp+16 u u u u u c-16 c-8 00000000004005e4 rsp+24 u u u u c-24 c-16 c-8 00000000004005e9 rsp+32 u u u c-32 c-24 c-16 c-8 00000000004005eb rsp+40 u u c-40 c-32 c-24 c-16 c-8 00000000004005f3 rsp+48 u c-48 c-40 c-32 c-24 c-16 c-8 00000000004005fb rsp+56 c-56 c-48 c-40 c-32 c-24 c-16 c-8 000000000040060a rsp+64 c-56 c-48 c-40 c-32 c-24 c-16 c-8 000000000040063a rsp+56 c-56 c-48 c-40 c-32 c-24 c-16 c-8 000000000040063b rsp+48 c-56 c-48 c-40 c-32 c-24 c-16 c-8 000000000040063c rsp+40 c-56 c-48 c-40 c-32 c-24 c-16 c-8 000000000040063e rsp+32 c-56 c-48 c-40 c-32 c-24 c-16 c-8 0000000000400640 rsp+24 c-56 c-48 c-40 c-32 c-24 c-16 c-8 0000000000400642 rsp+16 c-56 c-48 c-40 c-32 c-24 c-16 c-8

Symbolic Debugging Frames might disappear in the presence of tail call optimization. main: callq one … one: callq two … two: callq three … three: retq main: callq one … one: jmp two … two: jmp three … three: retq

Symbolic Debugging Frames might disappear in the presence of tail call optimization. Thread 8163 [ 0] libc-2.21.so pause [ 1] chain three [ 2] chain two [ 3] chain one [ 4] chain main Thread 8224 [ 0] libc-2.21.so pause [ 1] chain three [ 2] chain main Tail calls are disambiguated by call sites and some times a debugger will only be able to extract the inner-most tail call in a chain.

I’m probably over my time How do we measure debuggability?  How do we defeat the optimizer?  How do we marry DWARF with tracing?  …or horror stories so you can avoid creating your own. @0xF390

Symbolic Debugging with DWARF

More Related Content

What's hot

Viewers also liked

Similar to Symbolic Debugging with DWARF

Recently uploaded

Symbolic Debugging with DWARF