Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <BANLkTimkv0n5sTqZrH3m71JvD1XdRYfZzA@mail.gmail.com>
Date: Tue, 5 Apr 2011 11:37:37 -0400
From: Dan Rosenberg <dan.j.rosenberg@...il.com>
To: oss-security@...ts.openwall.com
Cc: "Steven M. Christey" <coley@...-smtp.mitre.org>, Josh Bressers <bressers@...hat.com>, 
	Eugene Teo <eugene@...hat.com>
Subject: Re: CVE request: kernel: multiple issues in ROSE

Hi,

This breakdown seems to make sense.  I'll do my best to break up the
issues below.

>
> Dan, could you confirm that this breakdown makes sense?
>
> 1) buffer overflows (not validating length is <= the maximum)
>

1) When parsing the FAC_NATIONAL_DIGIS facilities field, it's possible
for a remote host to provide more digipeaters than expected, resulting
in heap corruption.  Check against ROSE_MAX_DIGIS to prevent
overflows, and abort facilities parsing on failure.  It looks like
this will be CVE-2011-1493.

2) When parsing the FAC_CCITT_DEST_NSAP and FAC_CCITT_SRC_NSAP
facilities fields, a remote host can provide a length of greater than
20, resulting in a stack overflow of the callsign array.

> 2) use of negative signed integers in memcpy() and other operations where
>   conversion creates a large unsigned integer, referred to as
>   "underflow"
>

3) When parsing the FAC_CCITT_DEST_NSAP and FAC_CCITT_SRC_NSAP
facilities fields, a remote host can provide a length
of less than 10, resulting in an underflow in a memcpy size, causing a
kernel panic due to massive heap corruption.

Note that 2) and 3) are solved by validating a single length field, so
maybe they should be grouped together?  The above three issues were
all found by me.

> 3) any other types of problems that aren't covered by those two?  (The
>   length validation checks don't always have enough context in the source
>   code).
>

4) Ben Hutchings' fixes addressed multiple cases where the ROSE
protocol did not ensure that socket data being parsed wasn't being
read in from beyond the boundaries of the incoming socket buffer.  For
example, a received packet might provide a length field longer than
the amount of remaining data in the socket buffer.

Looking at the patch, it doesn't appear that any memory corruption
would be caused by this, since the out-of-bounds data is still
validated by the parsing code.  I'd say the impact is likely limited
to possible information disclosure, if the contents of the
out-of-bounds memory could be inferred by the behavior of the protocol
during parsing.  It's theoretically possible (but very unlikely) that
this could cause read accesses to unmapped memory, which would cause a
DOS.

-Dan

> We would need separate CVE's for the issues found by Dan versus the issues
> found by Ben Hutchings.
>
> Arguably, #2 could probably be broken down further, but without enough
> source code context in the patches, it's not immediately clear.
>
> - Steve
>

Powered by blists - more mailing lists

Please check out the Open Source Software Security Wiki, which is counterpart to this mailing list.

Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.