Bugtraq mailing list archives

PoPToP PPTP server remotely exploitable buffer overflow


From: Timo Sirainen <tss () iki fi>
Date: 09 Apr 2003 19:19:33 +0300

Versions older than 1.1.4-b3 and 1.1.3-20030409 affected. This seems to be
exploitable only with Linux.

PPTP?
-----

PPTP-over-IPSEC is commonly used to create VPNs. Windows plays quite nicely
with it.


problem
-------

PPTP packet header contain 16bit length which specifies the full size of
the packet:

        bytes_this = read(clientFd, packet + bytes_ttl, 2 - bytes_ttl);
        // ...
        bytes_ttl += bytes_this;
        // ...
        length = htons(*(u_int16_t *) packet);
        if (length > PPTP_MAX_CTRL_PCKT_SIZE) {
          // abort
        }

Looks good so far, except:

        bytes_this = read(clientFd, packet + bytes_ttl, length - bytes_ttl);

If given length was 0 or 1, the "length - bytes_ttl" result is -1 or -2,
which means that it reads unlimited amount of data from client into
"packet", which is a buffer located in stack.

The exploitability only depends on if libc allows the size parameter to be
larger than SSIZE_MAX bytes. GLIBC does, Solaris and *BSD don't.


tips of the day
---------------

Don't do arithmetic in parameters specifying buffer size.

Block PPTP port for non-IPSEC connections, if you don't already.


patch
-----

This is the same as in latest versions:

--- ctrlpacket.c.old    1999-12-23 23:43:33.000000000 +0200
+++ ctrlpacket.c        2003-04-09 18:58:21.000000000 +0300
@@ -254,8 +254,8 @@
        }
        /* OK, we have (at least) the first 2 bytes, and there is data waiting */
        length = htons(*(u_int16_t *) packet);
-       if (length > PPTP_MAX_CTRL_PCKT_SIZE) {
-               syslog(LOG_ERR, "CTRL: Control packet > PPTP_MAX_CTRL_PCKT_SIZE (length = %d)", length);
+       if (length <= 10 || length > PPTP_MAX_CTRL_PCKT_SIZE) {
+               syslog(LOG_ERR, "CTRL: 11 < Control packet (length=%d) < ", length);
                /* we loose sync (unless we malloc something big, which isn't a good
                 * idea - potential DoS) so we must close connection (draft states that
                 * if you loose sync you must close the control connection immediately)


Current thread: