Ilja van Sprundel, a security researcher with IOActive, has discovered an issue in the parsing of BDF font files by libXfont. Additional testing by Alan Coopersmith and William Robinet with the American Fuzzy Lop (afl) tool uncovered two more issues in the parsing of BDF font files.
As libXfont is used by the X server to read font files, and an unprivileged user with access to the X server can tell the X server to read a given font file from a path of their choosing, these vulnerabilities have the potential to allow unprivileged users to run code with the privileges of the X server (often root access).
The vulnerabilities are:
The bdf parser reads a count for the number of properties defined in a font from the font file, and allocates arrays with entries for each property based on that count. It never checked to see if that count was negative, or large enough to overflow when multiplied by the size of the structures being allocated, and could thus allocate the wrong buffer size, leading to out of bounds writes.
If the bdf parser failed to parse the data for the bitmap for any character, it would proceed with an invalid pointer to the bitmap data and later crash when trying to read the bitmap from that pointer.
The bdf parser read metrics values as 32-bit integers, but stored them into 16-bit integers. Overflows could occur in various operations leading to out-of-bounds memory access.
X.Org believes all prior versions of this library contain these flaws, dating back to its introduction in X11R5.
Fixes are available in these libXfont git commits:
Fixes are also included in the libXfont 1.5.1 & 1.4.9 module releases from X.Org.
X.Org thanks Ilja van Sprundel of IOActive, Alan Coopersmith of Oracle, and William Robinet of Conostix for reporting these issues to our security team and helping evaluate and test the fixes; and thanks Michal Zalewski and the American Fuzzy Lop community for providing their fuzz testing tool as an open source project we can all benefit from at http://lcamtuf.coredump.cx/afl/.