Fedora Account System
Red Hat Associate
Red Hat Customer
libfontenc's .enc encoding-file parser contains a heap out-of-bounds write in setCode() (src/encparse.c). The destination index is computed as (from >> 8) * row_size + (from & 0xFF), where row_size is taken straight from the SIZE line of the encoding file and is never bounded, while the destination buffer is allocated at a constant cap of 0x10000 entries. A crafted .enc file makes the index exceed the allocation and writes past it. The Xorg/Xwayland X server links libfontenc via libXfont2 and parses .enc files in its own process during font rasterization. Where the X server runs as root, a local user that places a font file and an encoding file on a server-read font path can trigger the OOB write, triggering potential local privilege escalation. Fixed in: libfontenc-1.1.10 Fixes: - https://gitlab.freedesktop.org/xorg/lib/libfontenc/-/commit/TBD Acknowledgements: Adam Bedard working with TrendAI Zero Day Initiative * Patch From 63bd5096e2f4889fc486a282fd5c1dfb93d559b0 Mon Sep 17 00:00:00 2001 From: Peter Hutterer <peter.hutterer> Date: Wed, 16 Sep 2026 15:24:17 +1000 Subject: [PATCH libFontenc 1/1] encparse: fix heap out-of-bounds write in setCode() setCode() computes a destination index from two attacker-controlled values taken from the .enc encoding file: index = (from >> 8) * row_size + (from & 0xFF); The "from" parameter is <= 0xFFFF (see the check a few lines above), so from >> 8 can reach 0xFF. row_size comes straight from the SIZE line of the encoding file. The destination buffer is allocated with a size capped at 0x10000 entries. At row_size > 256, the index exceeds the allocation, causing a heap out-of-bounds write. For example, SIZE 256 258 and mapping 0xff00 -> 0 gives: index = 255 * 258 + 0 = 65790, allocation = 65536 entries Additionally, the column bounds check incorrectly tested the global variable "value1" instead of the function parameter "from". For CODE_LINE this happens to be the same value, but for CODE_RANGE_LINE and CODE_UNDEFINE_LINE the loop calls setCode() with different values of "from" while value1 remains unchanged. This vulnerability was discovered by: Adam Bedard working with TrendAI Zero Day Initiative CVE-2026-109908, ZDI-CAN-31888 Assisted-by: Claude:claude-opus-4-6 #+begin_src diff --- src/encparse.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) diff --git ./src/encparse.c ../src/encparse.c index 39dbc89862f8..8949fd475b88 100644 --- ./src/encparse.c +++ ../src/encparse.c @@ -450,19 +450,21 @@ setCode(unsigned from, unsigned to, unsigned row_size, unsigned short *newenc; if (from > 0xFFFF) return 0; /* success */ if (row_size == 0) index = from; else { - if ((value1 & 0xFF) >= row_size) + if ((from & 0xFF) >= row_size) return 0; /* ignore out of range mappings */ index = (from >> 8) * row_size + (from & 0xFF); + if (index > 0xFFFF) + return 0; /* ignore mappings that exceed the encoding size */ } /* Optimize away useless identity mappings. This is only expected to be useful with linear encodings. */ if (index == to && (index < *first || index > *last)) return 0; if (*encsize == 0) { *encsize = (index < 256) ? 256 : 0x10000; -- 2.56.0 #+end_src