asan was complaining that the limits struct goes out
of scope before it is used via the pointer in the polygon struct,
and it is right:
==386746==ERROR: AddressSanitizer: stack-use-after-scope on address 0x7ffd3ccebdfc at pc 0x7f783d5eaaee bp 0x7ffd3cceba80 sp 0x7ffd3cceba70
READ of size 4 at 0x7ffd3ccebdfc thread T0
#0 0x7f783d5eaaed in _add_clipped_edge ../src/cairo-polygon.c:351
#1 0x7f783d5ebba3 in _cairo_polygon_add_edge ../src/cairo-polygon.c:520
#2 0x7f783d5ebc82 in _cairo_polygon_add_external_edge ../src/cairo-polygon.c:530
#3 0x7f783d582149 in _cairo_filler_line_to ../src/cairo-path-fill.c:63
#4 0x7f783d588d9c in _cairo_path_fixed_interpret ../src/cairo-path-fixed.c:831
#5 0x7f783d582a44 in _cairo_path_fixed_fill_to_polygon ../src/cairo-path-fill.c:147
#6 0x7f783d6204fe in _cairo_spans_compositor_fill ../src/cairo-spans-compositor.c:1151
#7 0x7f783d5126de in _cairo_compositor_fill ../src/cairo-compositor.c:203
#8 0x7f783d5571f9 in _cairo_image_surface_fill ../src/cairo-image-surface.c:1003
#9 0x7f783d647f2f in _cairo_surface_fill ../src/cairo-surface.c:2424
#10 0x7f783d52ebea in _cairo_gstate_fill ../src/cairo-gstate.c:1312
#11 0x7f783d51cca4 in _cairo_default_context_fill ../src/cairo-default-context.c:1057
#12 0x7f783d6812d6 in cairo_fill ../src/cairo.c:2421
GTK has a testcase that tests the error when creating
an oversize image, and asan tells me that it triggers
a memory leak in cairo:
Direct leak of 160 byte(s) in 1 object(s) allocated from:
#0 0x7f1122755667 in __interceptor_malloc (/lib64/libasan.so.6+0xb0667)
#1 0x7f1120cc83e8 in _cairo_pattern_create_solid ../src/cairo-pattern.c:607
#2 0x7f1120cc8487 in _cairo_pattern_create_in_error ../src/cairo-pattern.c:630
#3 0x7f1120cc87cb in INT_cairo_pattern_create_for_surface ../src/cairo-pattern.c:736
#4 0x7f1120c1f1c7 in _cairo_default_context_set_source_surface ../src/cairo-default-context.c:327
#5 0x7f1120d8386a in INT_cairo_set_source_surface ../src/cairo.c:982
#6 0x7f1121d005a2 in gdk_cairo_set_source_pixbuf ../gdk/gdkcairo.c:234
#7 0x401427 in test_set_source_big_pixbuf ../testsuite/gdk/cairo.c:23
The expression &image_surface->base basically just casts the
cairo_image_surface_t* to cairo_surface_t*. However, technically it is a
NULL pointer dereference and UndefinedBehaviorSanitizer flags it as
such:
runtime error: member access within null pointer of type 'cairo_image_surface_t' (aka 'struct _cairo_image_surface')
This commit fixes this by adding a NULL check.
Signed-off-by: Uli Schlachter <psychon@znc.in>
-----BEGIN PGP SIGNATURE-----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=jna2
-----END PGP SIGNATURE-----
Merge tag '1.17.4'
cairo 1.17.4 release
We are not quite sure what happened, but the 1.17.4 tag and the master
branch have different Git histories, but the same content. The merge
commits are missing from the history of the tag. This merge is simply
the result of "git merge 1.17.4".
This fixes: https://gitlab.freedesktop.org/cairo/cairo/-/issues/446
X11 use uint16_t for the width/height of things. Anything too large will
be truncated when sending the request to the X11 server. This commit
adds a size check to a function that did not check things and then later
caused a segmentation fault.
Not adding a test case because the test case from the below bug report
allocates 3,5 GiB of memory, which I find too much for a test.
Fixes: https://gitlab.freedesktop.org/cairo/cairo/-/issues/414
Signed-off-by: Uli Schlachter <psychon@znc.in>
Surfaces from _cairo_surface_create_in_error() have no backend. This
commit fixes a NULL pointer dereference in cairo_win32_surface_get_dc().
Fixes: https://gitlab.freedesktop.org/cairo/cairo/-/issues/405
Signed-off-by: Uli Schlachter <psychon@znc.in>
In commit 10e58a4a I changed the code in cairo-surface.c to avoid
setting surface->is_clear = FALSE; in some situations where it was not
necessary, because the operation did not actually modify anything (it
returned CAIRO_INT_STATUS_NOTHING_TO_DO). However, that change
accidentally also caused _cairo_surface_paint() not to set
surface->is_clear = TRUE; in similar cases. That was unintended.
This commit fixes that by always setting is_clear = TRUE when necessary,
but keeps the optimisation of not setting is_clear = FALSE when not
necessary.
The connection to the below issue is that the issue happened with
surfaces with width=0. Clearing such a surface with CAIRO_OPERATOR_CLEAR
causes CAIRO_INT_STATUS_NOTHING_TO_DO and thus is_clear = TRUE was not
set. This error was later caught by a failed assertion.
Fixes: https://gitlab.freedesktop.org/cairo/cairo/-/issues/283
Signed-off-by: Uli Schlachter <psychon@znc.in>
An error in _cairo_surface_snapshot_copy_on_write() results in a
snapshot in an error state and the snapshot's ->target could now point
to a surface from _cairo_surface_create_in_error(). These surfaces e.g.
have ->backend == NULL. Thus, anything looking at ->backend->type now
explodes. This commit deals with two places which caused segfaults in
this situation.
There is no test case for this, because
_cairo_surface_snapshot_copy_on_write() really is not supposed to fail.
Found-while-investigating: https://gitlab.freedesktop.org/cairo/cairo/-/issues/448
Signed-off-by: Uli Schlachter <psychon@znc.in>
The code in cairo-cff-subset.c parses a binary format without seeming to
bother much with verifying the data. The result is that poppler can be
used to cause an out-of-bounds write in cairo_cff_font_read_fdselect()
via a crafted font file. Fix this by adding the needed length check.
The other code in the file also contains lots of similar things. Since I
cannot really fix everything properly, I'll just fix the one instance
that was found by a fuzzer.
No testcase is added, because this depends on a broken font that is
quite large. Adding something this big to the test suite does not seem
sensible.
Fixes: https://gitlab.freedesktop.org/cairo/cairo/-/issues/451
Signed-off-by: Uli Schlachter <psychon@znc.in>
The code was copying from the wrong member of an union. This caused a
huge num_dashes value to be read, which then caused a so large memory
allocation that malloc returned an error.
Fixes: https://gitlab.freedesktop.org/cairo/cairo/-/issues/448
Signed-off-by: Uli Schlachter <psychon@znc.in>
The code in cairo-cff-subset.c parses a binary font format without
seeming to bother much verifying the data. The result is that poppler
can be used to cause an out-of-bounds access in
cairo_cff_parse_charstring() via a crafted font file. Fix this by adding
the needed length check.
The other code in the file also contains lots of similar things. Since I
cannot really fix everything properly, I'll just fix the one instance
that was found by a fuzzer.
No testcase is added, because this depends on a broken font that is
quite large. Adding something this big to the test suite does not seem
sensible.
Fixes: https://gitlab.freedesktop.org/cairo/cairo/-/issues/444
Signed-off-by: Uli Schlachter <psychon@znc.in>
The code here temporary replaces extents->clip with another clip to call
a function. Afterwards, it restores the previous copy. The temporary
clip is only freed when it still is pointed to by extents->clip.
This logic is wrong. It is indeed possible that the clip is simplified
and changes. In this case, the original clip is also correctly freed.
However, this still means that we have to clean up and destroy the new
clip. The previous code just leaked it.
This was originally identified by Massimo in [1]. I am just committing
his patch.
[1]: https://bugs.freedesktop.org/show_bug.cgi?id=91267
Fixes: https://gitlab.freedesktop.org/cairo/cairo/-/issues/24
Signed-off-by: Uli Schlachter <psychon@znc.in>
Since we now copy the data that CGImage needs we don't need to
keep the surface around anymore, nor release it or the image in the
DataProviderReleaseCallback.
Operators BlueValues, OtherBlues, FamilyBlues, FamilyOtherBlues,
StemSnapH and StemSnapV have operands of type delta which can be
a number or an array of delta-encoded numbers. This array can be
empty according to freetype developers.
This commit checks whether current operator is among those listed
and permits empty operand in such case.
Operators BlueValues, OtherBlues, FamilyBlues, FamilyOtherBlues,
StemSnapH and StemSnapV have operands of type delta which can be
a number or an array of delta-encoded numbers. This array can be
empty according to freetype developers.
This commit checks whether current operator is among those listed
and permits empty operand in such case.
That way when other projects consume our declared dep, they get
transitive dependencies too based on what features cairo was built
with. Without this, projects that build cairo as a subproject and also
build, say, fontconfig as a subproject will fail to find cairo-ft.h
(etc).
It was originally added to make bisecting easier,
but has outlived its usefuleness now.
Going forward we'll have just a single cairo-version.h
header file, the one with the real version numbers.
This is needed to fix the case where cairo is being
built as a Meson subproject, but also simplifies
things in general.
Fixes#421
'make check' currently fails with the following in src/test-suite.log
(how can one tell automake not to hide the error message in a file?!?):
FAIL: check-doc-syntax.sh
=========================
Checking documentation for incorrect syntax
./cairo-cogl-surface.c: * cairo_cogl_pipeline_ts. On failure, both pointers will be set to
./cairo-cogl-surface.c:/* Mostly taken from cairo_vg_surface.c */
Error: some function names in the docs are not followed by parentheses.
Fix this by searching for the following regexp in the above files:
'^[^:]*:[/ ][*]\(\|[ ].*\)\([^#']\|^\)\<\(cairo_[][<>/0-9a-z_]*\>[^][<>(]\)'
FAIL check-doc-syntax.sh (exit status: 1)
This commit silences that. The fix is likely not correct, but since
these are normal comments and not actual doc comments, it does not
matter much.
Signed-off-by: Uli Schlachter <psychon@znc.in>
This constitutes few fixes that are necessary to compile correctly
and reduce errors when using musl libc.
Signed-off-by: George Matsumura <gmmatsumura01@bvsd.org>
This removes code that uses the cogl-path library, which was not
used except when manually modifying a preprocessor flag. It could
not use path caching, was slightly broken, and all of its
functionality was provided better by different code paths.
Signed-off-by: George Matsumura <gmmatsumura01@bvsd.org>
