--- title: "x86-64 NASM using syswrite to print a string" author: "0x3bb" date: M07-27-2024 --- I'm reading a book on x86-64 NASM. One of the exercises involves printing a string to stdout by leveraging a Linux syscall called `sys_write`, but doesn't mention the reasoning behind the `mov` instructions into the registers or how they're used once you make the syscall. An approachable way for beginners to understand how syscalls are made in Linux is referring to the [Linux System Call Table](https://blog.rchapman.org/posts/Linux_System_Call_Table_for_x86_64/). It documents each parameter and in which specific register it expects to find it. Looking at that table, it can be concluded the following registers need to be populated: - `rdi` → the call identifier. for `sys_write`, it's `1` from reading the first column of that table - `rdi` → file descriptor, `1` being for `stdout` - `rsi` → contents of buffer (i.e. the string) - `rdx` → length of the buffer for stdout ## .data The first step is to define the `.data` section and initialize memory for `rsi`. The syntax looks like this: `0xa` is just the ASCII hex representation of a new line. ``` ; sys_write_string.asm section .data s1 db "s1", 0x0a, 0; terminated string with NL ``` Then, for `rdx`, the requirement is the length of the string, minus the terminator. Many standard libraries add the `NULL` terminator to initialized strings. This is so when the string is passed around, its length can be implicitly determined without the caller having to pass an additional parameter (the length of the string). Although, since `sys_write` does not expect a terminated string, this should be stripped. By taking the address of `s1` minus the terminated character (1 byte/8 bits). ``` s1Len equ $-s1-1; offset of s1 - terminator ``` ## .bss - `.bss` stands for _Block Started by Symbol_, and contains uninitialized variables that assign memory at runtime. For this example, it makes sense to define the value in the `.data` block above instead. ``` section .bss ``` ## .text The final section required is to define the entrypoint of the program - `.text` refers to the code segment: the program's virtual address space that contains executable instructions. - `main` is our program's entrypoint. ``` section .text global main ``` `main` must be prefixed with the `global` directive. If the `global` keyword was omitted, then the linker (`ld`) will not see it, since `main` is scoped to the object file: ``` [0x3bb@heimat 2]$ make nasm -f elf64 -g -F dwarf sys_write_string.asm -l sys_write_string.lst sys_write_string.asm:9: warning: label alone on a line without a colon might be in error [-w+label-orphan] gcc -o sys_write_string sys_write_string.o -no-pie /usr/sbin/ld: /usr/lib/gcc/x86_64-pc-linux-gnu/14.1.1/../../../../lib/crt1.o: in function `_start': (.text+0x1b): undefined reference to `main' collect2: error: ld returned 1 exit status ``` This can also be seen using GNU `nm`, a tool that can dump symbols from binaries and object files. Without the `global` directive, the symbol `t` is local. ``` [0x3bb@heimat 2]$ nm sys_write_string.o 0000000000000000 t main 0000000000000000 d s1 ``` With the `global` directive set correctly, the [symbol type identifier](https://sourceware.org/binutils/docs-2.39/binutils/nm.html) will be uppercase and linked correctly. ## label The `main` label serves as an alias to the block of instructions defined below ``` main: push rbp; push base address of stack frame to restore later mov rbp, rsp; copy address of current stack frame mov rax, 1; 1 = sys_write call mov rdi, 1; 1 = file descriptor (stdout) mov rsi, s1; buf mov rdx, s1Len; length of buf syscall ; invoke sys_write to stdout mov rsp, rbp; copy original base address of stack frame pop rbp; restore caller stack frame mov rax, 60; sys_exit mov rdi, 0; = exit code syscall ; invoke sys_exit ``` ### assembler output Looking to see what the assembler did, the label instructions have been assigned a relative address by the assembler ``` [0x3bb@heimat 2]$ cat sys_write_string.lst 1 ; sys_write_string.asm 2 section .data 3 00000000 73310A00 s1 db "s1", 0x0a, 0; terminated string with NL 4 s1Len equ $-s1-1; address of s1 - terminator 5 6 section .bss 7 8 section .text 9 global main 10 11 main: 12 00000000 55 push rbp; push base address of stack frame to restore later 13 00000001 4889E5 mov rbp, rsp; copy address of current stack frame 14 00000004 B801000000 mov rax, 1; 1 = sys_write call ``` ### gdb Debugging the program, the assembler chose to use 32-bit value registers; as it would be wasteful to put these values for `sys_write` in their 64-bit counterparts. ``` [0x3bb@heimat 2]$ gdb sys_write_string GNU gdb (GDB) 15.1 Copyright (C) 2024 Free Software Foundation, Inc. (gdb) disassemble main Dump of assembler code for function main: 0x0000000000401110 <+0>: push %rbp 0x0000000000401111 <+1>: mov %rsp,%rbp 0x0000000000401114 <+4>: mov $0x1,%eax 0x0000000000401119 <+9>: mov $0x1,%edi 0x000000000040111e <+14>: movabs $0x404010,%rsi 0x0000000000401128 <+24>: mov $0x3,%edx 0x000000000040112d <+29>: syscall 0x000000000040112f <+31>: mov %rbp,%rsp 0x0000000000401132 <+34>: pop %rbp 0x0000000000401133 <+35>: mov $0x3c,%eax 0x0000000000401138 <+40>: mov $0x0,%edi 0x000000000040113d <+45>: syscall End of assembler dump. (gdb) break main Breakpoint 1 at 0x401110: file sys_write_string.asm, line 12. (gdb) run Starting program: /mnt/h/src/asm/2/sys_write_string Breakpoint 1, main () at sys_write_string.asm:12 12 push rbp; function prologue (gdb) step 13 mov rbp, rsp; function prologue (gdb) step 14 mov rax, 1; 1 = sys_write call (gdb) step 15 mov rdi, 1; 1 = file descriptor (stdout) (gdb) step 16 mov rsi, s1; buf (gdb) step 17 mov rdx, s1Len; length of buf (gdb) step 18 syscall ; invoke sys_write to stdout (gdb) info registers rax 0x1 1 rbx 0x7fffffffe368 140737488348008 rcx 0x403e30 4210224 rdx 0x3 3 rsi 0x404010 4210704 rdi 0x1 1 rbp 0x7fffffffe240 0x7fffffffe240 rsp 0x7fffffffe240 0x7fffffffe240 r8 0x0 0 r9 0x7ffff7fcb200 140737353921024 r10 0x7fffffffdf70 140737488346992 r11 0x203 515 r12 0x1 1 r13 0x0 0 r14 0x7ffff7ffd000 140737354125312 r15 0x403e30 4210224 rip 0x40112d 0x40112d eflags 0x246 [ PF ZF IF ] ``` The string length from `rdx` passed to `sys_write` contains no `NULL` terminator ``` (gdb) info registers rdx rdx 0x3 3 (gdb) x/3 $rsi 0x404010 : 115 's' 49 '1' 10 '\n' (gdb) x/4 $rsi 0x404010 : 115 's' 49 '1' 10 '\n' 0 '\000' ```