Bug 183677

Freetype integer underflow

While fuzzing some pdf files a few weeks back, I ended up finding an
integer underflow in freetype2.  Upstream did commit some fixes, so
this issue is somewhat public.

1.ttf will generate this error.

Patches:
http://cvs.savannah.gnu.org/viewcvs/freetype/freetype2/src/pshinter/pshglob.c?r1=1.30&r2=1.31&makepatch=1&diff_format=h
http://cvs.savannah.gnu.org/viewcvs/freetype/freetype2/src/cff/cffload.c?r1=1.73&r2=1.74&makepatch=1&diff_format=h
http://cvs.savannah.gnu.org/viewcvs/freetype/freetype2/src/type1/t1load.c?r1=1.104&r2=1.105&makepatch=1&diff_format=h

The problem is the number of blue values needs to be even.  If a font
file claims it's odd, freetype2 doesn't handle it well.


The crash is seen in src/pshinter/pshglob.c:psh_blues_set_zones_0()

What was basically happening is since read_count is an unsigned
integer, and is decremented by 2, it is possible to cause an integer
underflow by ensuring the value of read_count is an odd number.  Once
read_count underflows, the loop starts dumping garbage onto the heap.
Normally I wouldn't think this is exploitable as it should crash
before anything exciting can happen, most graphical applications are
multi-threaded, so this does worry me.  At the very least this issue
is a denial of service bug.

The 2.ttf file in the testcase will also trigger a NULL pointer
dereference, which I'm not considering a secuirty issue.  The patch is
here:

http://cvs.savannah.gnu.org/viewcvs/freetype/freetype2/src/base/ftutil.c?r1=1.17&r2=1.18&makepatch=1&diff_format=h