Data of chunks which are malloc'd are not zeroed out, giving us free heap and libc leaks (will be elaborated on later)
We can only call scanf and puts on the last chunk we allocated since the program uses local_1a4 to track our most recently malloc'd chunk
We can overflow local_198 with our malloc'd chunk pointers; This is because local_1a4 can increment until it is ≥ 15, allowing us to malloc a pointer into local_198[16] and beyond.
Notice that if our malloc pointers overflow into the command buffer, when we are writing the command we can modify those pointers as well! For example, if I have already malloc'd 18 times, then local_118[0] would have a malloc pointer and so would local_118[1]. Then, if I input a command like b'puts\x00\x00\x00\x00\xAA\xAA\xAA\xAA\xAA\xAA\xAA\xAA', then my command would be correctly interpreted to be puts (since my command ends with a null terminator) and the program would attempt to read whatever is at address 0xAAAAAAAAAAAAAAAA. This means that we have an arbitrary read.
If we use a command like b'scanf\x00\x00\x00\xAA\xAA\xAA\xAA\xAA\xAA\xAA\xAA' then we could also write data into 0xAAAAAAAAAAAAAAAA . This means we also have an arbitrary write!
So what should we overwrite in order to get a shell?
Some of the popular options would include:
Overwriting a global offset table entry, e.g. overwrite GOT entry for free to system to potentially call system('/bin/sh') when we free a chunk. This won't work since both libc and the binary have full RELRO enabled.
Overwriting __malloc_hook or __free_hook. These won't work since it has been patched in this libc version.
Overwriting the return address of main. This won't work since main doesn't return. Rather, main calls exit to exit the program.
Note: The challenge author's solution was actually to get a stack leak, then overwrite the return address of scanf. However, I adopted a different solution.
It turns out that when a program exits, it calls a list of exit functions. Furthermore, the structs which contain these exit functions are in the DATA region of libc, which is readable and writeable. Hence, if we can overwrite these exit functions correctly to functions that can give us a shell, when the program exits we could get a shell!
TLDR:
When exit() is called, it calls __run_exit_handlers
__run_exit_handlers takes in a struct exit_function_list ** , then for each entry it demangles the function pointer and calls the function:
/* Call all functions registered with `atexit' and `on_exit',
in the reverse of the order in which they were registered
perform stdio cleanup, and terminate program execution with STATUS. */
void
attribute_hidden
__run_exit_handlers (int status, struct exit_function_list **listp,
bool run_list_atexit, bool run_dtors)
{
/* First, call the TLS destructors. */
#ifndef SHARED
if (&__call_tls_dtors != NULL)
#endif
if (run_dtors)
__call_tls_dtors ();
/* We do it this way to handle recursive calls to exit () made by
the functions registered with `atexit' and `on_exit'. We call
everyone on the list and use the status value in the last
exit (). */
while (*listp != NULL)
{
struct exit_function_list *cur = *listp;
while (cur->idx > 0)
{
const struct exit_function *const f =
&cur->fns[--cur->idx];
switch (f->flavor)
{
void (*atfct) (void);
void (*onfct) (int status, void *arg);
void (*cxafct) (void *arg, int status);
case ef_free:
case ef_us:
break;
case ef_on:
onfct = f->func.on.fn;
#ifdef PTR_DEMANGLE
PTR_DEMANGLE (onfct);
#endif
onfct (status, f->func.on.arg);
break;
case ef_at:
atfct = f->func.at;
#ifdef PTR_DEMANGLE
PTR_DEMANGLE (atfct);
#endif
atfct ();
break;
case ef_cxa:
cxafct = f->func.cxa.fn;
#ifdef PTR_DEMANGLE
PTR_DEMANGLE (cxafct);
#endif
cxafct (f->func.cxa.arg, status);
break;
}
}
*listp = cur->next;
if (*listp != NULL)
/* Don't free the last element in the chain, this is the statically
allocate element. */
free (cur);
}
if (run_list_atexit)
RUN_HOOK (__libc_atexit, ());
_exit (status);
}
The struct exit_function_list and the other relevant structs look like this:
enum
{
ef_free, /* `ef_free' MUST be zero! */
ef_us,
ef_on,
ef_at,
ef_cxa
};
struct exit_function
{
/* `flavour' should be of type of the `enum' above but since we need
this element in an atomic operation we have to use `long int'. */
long int flavor;
union
{
void (*at) (void);
struct
{
void (*fn) (int status, void *arg);
void *arg;
} on;
struct
{
void (*fn) (void *arg, int status);
void *arg;
void *dso_handle;
} cxa;
} func;
};
struct exit_function_list
{
struct exit_function_list *next;
size_t idx;
struct exit_function fns[32];
};
There are 5 different types of exit handlers: ef_free, ef_us, ef_on, ef_at and ef_cxa.
During demangling, the following assembly in __run_exit_handlers is applied:
Usually, the struct exit_function_list ** passed to __run_exit_handlers has at least one entry in it, and the first entry is usually the _dl_fini function which is in ld code. Therefore to find the pointer guard value, there are 2 ways to do so:
Leak the value from fs:0x30
Calculate the value from the address of _dl_fini and the mangled pointer
For my solution I adopted the second approach.
First I needed to find out the location of __exit_funcs (which is the struct exit_function_list ** ) so in GDB I can do something like this:
So we can see that it contains two pointers, the first of which is a pointer to the exit_function_list struct for _dl_fini.
Now that I know the address of exit_function_list * and I know that it is in a fixed position within the libc data section, I can calculate its offset.
Alternatively if I wanted to find the position of fs:0x30 I could also have done something like this:
Notes:
TCB region is always located directly above libc
The 0x6b66c574a13346de we see is the pointer guard value
Note that the TCB is readable and writeable, and the exit function structs are also readable and writeable. Therefore for exploitation, we can either:
Rewrite pointer guard value to a value of our choosing and change the exit_function_list struct for _dl_fini (arb write + arb write), or
Read pointer guard value, then change exit_function_list struct for _dl_fini (arb read + arb write)
I also went with the second approach for my solution.
A brief outline of my solution is written below, including all the leaks I got, how I got them, and how I finally exploited the program:
Malloc, free, malloc then read a chunk to get a heap leak (this is because the heap metadata is not cleared when mallocing a chunk)
Malloc a bunch of times so our malloc pointers overflow into the command buffer
for _ in range(20):
r.sendline(b"malloc")
At this point the program has malloc'd 21 times, the last 5 of which overflow into the command buffer.
Since the program forces all chunks malloc'd to be of size 0x40, we cannot directly free a chunk into the unsorted bin and get a libc leak. However, I used arbitrary write to change the size field of a chunk to be a large value such as 0x800 then freed it. The freed chunk would go into the unsorted bin. I then used the arbitrary read to read its metadata, which points to main arena in libc which is at a constant offset, thereby getting the libc leak.
Note:
To prevent coalescing of this big chunk with the wilderness, and to prevent double free or corruptions from being detected, I had 1 fake 0x800 size chunk followed by 2 fake 0x30 size chunks. I free the first 0x30 chunk to act as a guard chunk to prevent the 0x800 chunk from coalescing, then free the 0x800 chunk. The third chunk is necessary to prevent corruption errors.
Since I have a libc leak, and ld is adjacent to libc, and _dl_fini is at a constant offset in ld, I know the address of _dl_fini. Furthermore, since the exit_function_list struct of _dl_fini is at a constant offset in libc, I also know its address. Thus I know both the unmangled and mangled values of the _dl_fini function pointer and can calculate the pointer guard value:
Note: For some reason, on remote the offset between ld and libc is 0x2000 more than the local offset. I got this value by bruteforcing.
Now that I have the key (aka pointer guard) value, I can mangle a function pointer to system in libc. I then overwrite the exit_function_list struct for _dl_fini with my own. In order for it to run system("/bin/sh") on exit, the exit function list has to be of type ex_cfa (which is represented by 4 in the enum). We keep the id field the same (which is 1), and we add a pointer to /bin/sh in libc as *arg.
