While clang's sanitizer recommends a cast to unsigned for safe
negation (to handle -INT_MIN), both MSVC and Visualc emit warning
C4146 if an unsigned value gets negated.
* include/freetype/internal/ftcalc.h (NEG_LONG, NEG_INT32),
src/base/ftcalc.c (FT_MOVE_SIGN): Replace negation with a
subtraction.
* builds/unix/unix-def.in (PKG_CONFIG): New variable.
(freetype-config): Use it in sed expression.
* builds/unix/freetype-config.in: s/pkg-config/%PKG_CONFIG%/.
Reported by Nikolaus Waxweiler <madigens@gmail.com>.
* src/truetype/ttgxvar.c (TT_Set_Var_Design): Correctly handle the
case where we have less input coordinates than axes.
These versions don't produce run-time errors due to integer
overflow.
* include/freetype/internal/ftobjs.h: Include FT_INTERNAL_CALC_H.
(FT_PAD_ROUND_LONG, FT_PAD_CEIL_LONG, FT_PIX_ROUND_LONG,
FT_PIX_CEIL_LONG): New macros.
(FT_PAD_ROUND_INT32, FT_PAD_CEIL_INT32, FT_PIX_ROUND_INT32,
FT_PIX_CEIL_INT32): New macros.
* src/smooth/ftgrays.c [STANDALONE] (OVERFLOW_ADD_LONG,
OVERFLOW_SUB_LONG, OVERFLOW_MUL_LONG, NEG_LONG): New macros.
[!STANDALONE]: Include FT_INTERNAL_CALC_H.
(gray_render_cubic): Use those macros where appropriate.
Problem reported by Marek Kašík <mkasik@redhat.com>.
The problematic font that exceeds the old limit is Padauk-Bold,
version 3.002, containing bytecode generated by a buggy version of
ttfautohint.
This commit handles the new engine.
* include/freetype/internal/ftcalc.h (OVERFLOW_ADD_INT32,
OVERFLOW_SUB_INT32, OVERFLOW_MUL_INT32, NEG_INT, NEG_LONG,
NEG_INT32): New macros.
* src/cff/cf2ft.c (cf2_getScaleAndHintFlag): Use OVERFLOW_ADD_INT32.
* src/cff/cf2hints.c (cf2_getWindingMomentum, cf2_hint_init,
cf2_hintmap_map, cf2_glyphpath_hintPoint,
cf2_glyphpath_computeIntersection, cf2_glyphpath_computeOffset,
cf2_glyphpath_lineTo, cf2_glyphpath_curveTo): Use
OVERFLOW_ADD_INT32, OVERFLOW_SUB_INT32, OVERFLOW_MUL_INT32, and
NEG_INT32 where appropriate.
* src/cff/cf2intrp.c (cf2_doFlex, cf2_doBlend,
cf2_interpT2CharString): Ditto.
Also add some other code where needed to avoid overflow.
This commit handles the old engine.
* src/cff/cffgload.c: Include FT_INTERNAL_CALC_H.
(cff_decoder_parse_charstrings): Use OVERFLOW_ADD_LONG and
OVERFLOW_SUB_LONG where needed.
* src/cff/cffparse.c: Include FT_INTERNAL_CALC_H.
(power_ten_limits): New static array.
(do_fixed): Use it to prevent multiplication overflow.
(cff_parser_run): Use OVERFLOW_ADD_LONG.
* src/pcf/pcfread.c (pcf_get_accel): Add sanity checks for
`fontAscent' and `fontDescent'.
(pcf_load_font): Add sanity checks for global height.
Add sanity checks for AVERAGE_WIDTH, POINT_SIZE, PIXEL_SIZE,
RESOLUTION_X, and RESOLUTION_Y properties.
This commit (mainly for 32bit CPUs) is the first of a series of
similar commits to handle known integer overflows. Basically, all
of them are harmless, since they affect rendering of glyphs only,
not posing security threats. It is expected that fuzzying will show
up more overflows, to be fixed in due course.
The idea is to mark places where overflows can occur, using macros
that simply cast to unsigned integers, because overflow arithmetic
is well defined in this case. Doing so suppresses run-time errors
of sanitizers without adding computational overhead.
* include/freetype/internal/ftcalc.h (OVERFLOW_ADD_INT,
OVERFLOW_SUB_INT, OVERFLOW_MUL_INT, OVERFLOW_ADD_LONG,
OVERFLOW_SUB_LONG, OVERFLOW_MUL_LONG): New macros.
* src/base/ftcalc.c (FT_RoundFix, FT_CeilFix, FT_Matrix_Multiply,
FT_Matrix_Multiply_Scaled, FT_Vector_Transform_Scaled,
ft_corner_orientation): Use new macros.
* src/base/ftoutln.c (FT_Outline_Get_Orientation): Use new macros.
The new name better describes what the macro actually does;
additionally, we don't need a trailing `f' for literals (there was
only a single such instance in the code, but this caused a clang
warning because the macro itself uses `double' literals).
* src/cff/cf2blues.c, src/cff/cf2blues.h, src/cff/cf2fixed.h,
src/cff/cf2font.c, src/cff/cf2hints.c: Updated.
* src/base/ftcalc.c (FT_MOVE_SIGN): Add argument to pass unsigned
value, to be used as the result.
(FT_MulDiv, FT_MulDiv_No_Round, FT_DivFix, FT_MulFix,
FT_Vector_NormLen): Updated.
Werner Lemberg