Bug 191805 – Segfaults on https://terminalizer.com/

• Status: RESOLVED
• Resolution: FIXED
• Component: WebKitGTK

[Cédric Bellegarde]

When using last WebKitGTK release on ArchLinux and Fedora 29. I guess you should be able to reproduce.

#0  0x00007f29fb3bece9 in  () at /lib64/libwebkit2gtk-4.0.so.37
#1  0x00007f29f5fa25df in _ZN9hb_font_t17get_nominal_glyphEjPj (glyph=0x7ffd2579a6c8, unicode=0, this=<optimized out>) at hb-font-private.hh:211
#2  0x00007f29f5fa25df in decompose_current_character (shortest=true, c=0x7ffd2579a6d0) at hb-ot-shape-normalize.cc:169
#3  0x00007f29f5fa25df in decompose_cluster (always_short_circuit=<optimized out>, might_short_circuit=true, end=<optimized out>, c=0x7ffd2579a6d0) at hb-ot-shape-normalize.cc:271
#4  0x00007f29f5fa25df in _Z22_hb_ot_shape_normalizePK18hb_ot_shape_plan_tP11hb_buffer_tP9hb_font_t (plan=plan@entry=0x560db2f332f0, buffer=buffer@entry=0x560db2d81c00, font=font@entry=0x560db379cd70) at hb-ot-shape-normalize.cc:330
#5  0x00007f29f5f907d6 in hb_ot_substitute_default (c=<synthetic pointer>) at hb-ot-shape.cc:604
#6  0x00007f29f5f907d6 in hb_ot_substitute (c=<synthetic pointer>) at hb-ot-shape.cc:636
#7  0x00007f29f5f907d6 in hb_ot_shape_internal (c=<synthetic pointer>) at hb-ot-shape.cc:870
#8  0x00007f29f5f907d6 in _hb_ot_shape(hb_shape_plan_t*, hb_font_t*, hb_buffer_t*, hb_feature_t const*, unsigned int) (shape_plan=shape_plan@entry=0x560db36f3250, font=font@entry=0x560db379cd70, buffer=buffer@entry=0x560db2d81c00, features=features@entry=0x7ffd2579a990, num_features=num_features@entry=1) at hb-ot-shape.cc:898
#9  0x00007f29f5f551ff in hb_shape_plan_execute(hb_shape_plan_t*, hb_font_t*, hb_buffer_t*, hb_feature_t const*, unsigned int) (shape_plan=shape_plan@entry=0x560db36f3250, font=font@entry=0x560db379cd70, buffer=buffer@entry=0x560db2d81c00, features=features@entry=0x7ffd2579a990, num_features=num_features@entry=1) at hb-shaper-list.hh:43
#10 0x00007f29f5f5478a in hb_shape_full(hb_font_t*, hb_buffer_t*, hb_feature_t const*, unsigned int, char const* const*) (font=0x560db379cd70, buffer=0x560db2d81c00, features=0x7ffd2579a990, num_features=1, shaper_list=<optimized out>) at hb-shape.cc:137
#11 0x00007f29fb3bca0d in  () at /lib64/libwebkit2gtk-4.0.so.37
#12 0x00007f29fae2b338 in  () at /lib64/libwebkit2gtk-4.0.so.37
#13 0x00007f29fae2d269 in  () at /lib64/libwebkit2gtk-4.0.so.37
#14 0x00007f29fae438a8 in  () at /lib64/libwebkit2gtk-4.0.so.37
#15 0x00007f29fae43f45 in  () at /lib64/libwebkit2gtk-4.0.so.37
#16 0x00007f29fae63065 in  () at /lib64/libwebkit2gtk-4.0.so.37
#17 0x00007f29fab249e9 in  () at /lib64/libwebkit2gtk-4.0.so.37
#18 0x00007f29fab24da2 in  () at /lib64/libwebkit2gtk-4.0.so.37
#19 0x00007f29fb485012 in  () at /lib64/libwebkit2gtk-4.0.so.37
#20 0x00007f2997fff177 in  ()
#21 0x00007ffd257ae6a0 in  ()
#22 0x00007f29f8c33172 in  () at /lib64/libjavascriptcoregtk-4.0.so.18
#23 0x0000000000000000 in  ()

[Michael Catanzaro]

The crash is deep inside harfbuzz, so you need to report it to harfbuzz: https://github.com/harfbuzz/harfbuzz/issues/

And you've been reporting good bugs for a long time now, so you should know better that to report a crash where ithout debug symbols for WebKit....

[Cédric Bellegarde]

I know, I rebooted on Fedora 29 to get symbols but gdb was unable to load them:
- gdb told me to install them with dnf install
- dnf install was saying the package was installed

[Cédric Bellegarde]

I know, I rebooted on Fedora 29 to get symbols but gdb was unable to load them:
- gdb told me to install them with dnf install
- dnf install was saying the package is installed

[Ebrahim Byagowi]

Hey there, per this comment, https://github.com/harfbuzz/harfbuzz/issues/1390#issuecomment-439682578 the stack-trace refers back to libwebkit2gtk itself, maybe you could see why WebKit callback is acting faulty on the case? We couldn't reproduce the issue here but I guess you can help more on finding what is happening from WebKit side. Any help on the case will be nice. Thanks

[Cédric Bellegarde]

Here the full backtrace (dnf update does not update debug symbols :-/)

#0  0x00007eff0d1a1ce9 in WebCore::harfBuzzGetGlyph(hb_font_t*, void*, hb_codepoint_t, hb_codepoint_t, hb_codepoint_t*, void*) (fontData=<optimized out>, unicode=<optimized out>, glyph=0x7fffd9067d08) at /usr/src/debug/webkit2gtk3-2.22.3-1.fc29.x86_64/Source/WebCore/platform/graphics/harfbuzz/HarfBuzzFaceCairo.cpp:107
#1  0x00007eff07d855df in _ZN9hb_font_t17get_nominal_glyphEjPj (glyph=0x7fffd9067d08, unicode=0, this=<optimized out>) at hb-font-private.hh:211
#2  0x00007eff07d855df in decompose_current_character (shortest=true, c=0x7fffd9067d10) at hb-ot-shape-normalize.cc:169
#3  0x00007eff07d855df in decompose_cluster (always_short_circuit=<optimized out>, might_short_circuit=true, end=<optimized out>, c=0x7fffd9067d10) at hb-ot-shape-normalize.cc:271
#4  0x00007eff07d855df in _Z22_hb_ot_shape_normalizePK18hb_ot_shape_plan_tP11hb_buffer_tP9hb_font_t (plan=plan@entry=0x55b1e7556160, buffer=buffer@entry=0x55b1e75b4f70, font=font@entry=0x55b1e75b5060) at hb-ot-shape-normalize.cc:330
#5  0x00007eff07d737d6 in hb_ot_substitute_default (c=<synthetic pointer>) at hb-ot-shape.cc:604
#6  0x00007eff07d737d6 in hb_ot_substitute (c=<synthetic pointer>) at hb-ot-shape.cc:636
#7  0x00007eff07d737d6 in hb_ot_shape_internal (c=<synthetic pointer>) at hb-ot-shape.cc:870
#8  0x00007eff07d737d6 in _hb_ot_shape(hb_shape_plan_t*, hb_font_t*, hb_buffer_t*, hb_feature_t const*, unsigned int) (shape_plan=shape_plan@entry=0x55b1e7548920, font=font@entry=0x55b1e75b5060, buffer=buffer@entry=0x55b1e75b4f70, features=features@entry=0x7fffd9067fd0, num_features=num_features@entry=1) at hb-ot-shape.cc:898
#9  0x00007eff07d381ff in hb_shape_plan_execute(hb_shape_plan_t*, hb_font_t*, hb_buffer_t*, hb_feature_t const*, unsigned int) (shape_plan=shape_plan@entry=0x55b1e7548920, font=font@entry=0x55b1e75b5060, buffer=buffer@entry=0x55b1e75b4f70, features=features@entry=0x7fffd9067fd0, num_features=num_features@entry=1) at hb-shaper-list.hh:43
#10 0x00007eff07d3778a in hb_shape_full(hb_font_t*, hb_buffer_t*, hb_feature_t const*, unsigned int, char const* const*) (font=0x55b1e75b5060, buffer=0x55b1e75b4f70, features=0x7fffd9067fd0, num_features=1, shaper_list=<optimized out>) at hb-shape.cc:137
#11 0x00007eff0d19fa0d in _ZN7WebCore21ComplexTextController35collectComplexTextRunsForCharactersEPKDsjjPKNS_4FontE (this=0x7fffd9068150, characters=0x7efe8f49a68c u"", length=1, stringLocation=0, font=0x7efe63cc4b58) at /usr/src/debug/webkit2gtk3-2.22.3-1.fc29.x86_64/x86_64-redhat-linux-gnu/DerivedSources/ForwardingHeaders/wtf/Vector.h:724
#12 0x00007eff0cc0e338 in _ZN7WebCore21ComplexTextController22collectComplexTextRunsEv (this=this@entry=0x7fffd9068150) at /usr/src/debug/webkit2gtk3-2.22.3-1.fc29.x86_64/Source/WebCore/platform/graphics/ComplexTextController.cpp:468
#13 0x00007eff0cc10269 in _ZN7WebCore21ComplexTextControllerC2ERKNS_11FontCascadeERKNS_7TextRunEbPN3WTF7HashSetIPKNS_4FontENS7_7PtrHashISB_EENS7_10HashTraitsISB_EEEEb (this=0x7fffd9068150, font=..., run=..., mayUseNaturalWritingDirection=<optimized out>, fallbackFonts=<optimized out>, forTextEmphasis=<optimized out>) at /usr/src/debug/webkit2gtk3-2.22.3-1.fc29.x86_64/Source/WebCore/platform/graphics/ComplexTextController.cpp:155
#14 0x00007eff0cc268a8 in _ZNK7WebCore11FontCascade34getGlyphsAndAdvancesForComplexTextERKNS_7TextRunEjjRNS_11GlyphBufferENS0_20ForTextEmphasisOrNotE (this=<optimized out>, run=..., from=0, to=1, glyphBuffer=..., forTextEmphasis=<optimized out>) at /usr/src/debug/webkit2gtk3-2.22.3-1.fc29.x86_64/Source/WebCore/platform/graphics/FontCascade.cpp:1391
#15 0x00007eff0cc26f45 in _ZNK7WebCore11FontCascade8drawTextERNS_15GraphicsContextERKNS_7TextRunERKNS_10FloatPointEjSt8optionalIjENS0_24CustomFontNotReadyActionE (this=this@entry=0x7efea5c73be8, context=..., run=..., point=..., from=from@entry=0, to=Python Exception <class 'gdb.error'> There is no member or method named _M_payload.: 
..., customFontNotReadyAction=WebCore::FontCascade::UseFallbackIfFontNotReady) at /usr/src/debug/webkit2gtk3-2.22.3-1.fc29.x86_64/x86_64-redhat-linux-gnu/DerivedSources/ForwardingHeaders/wtf/Optional.h:312
#16 0x00007eff0cc46065 in _ZN7WebCore15GraphicsContext12drawBidiTextERKNS_11FontCascadeERKNS_7TextRunERKNS_10FloatPointENS1_24CustomFontNotReadyActionE (this=this@entry=0x7efe63cc4948, font=..., run=..., point=..., customFontNotReadyAction=customFontNotReadyAction@entry=WebCore::FontCascade::UseFallbackIfFontNotReady) at /usr/src/debug/webkit2gtk3-2.22.3-1.fc29.x86_64/x86_64-redhat-linux-gnu/DerivedSources/ForwardingHeaders/wtf/Optional.h:418
#17 0x00007eff0c9079e9 in _ZNK7WebCore28CanvasRenderingContext2DBase9FontProxy12drawBidiTextERNS_15GraphicsContextERKNS_7TextRunERKNS_10FloatPointENS_11FontCascade24CustomFontNotReadyActionE (action=WebCore::FontCascade::UseFallbackIfFontNotReady, point=..., run=..., context=..., this=0x7efea5c73be0) at /usr/src/debug/webkit2gtk3-2.22.3-1.fc29.x86_64/Source/WebCore/html/canvas/CanvasRenderingContext2DBase.cpp:348
#18 0x00007eff0c9079e9 in _ZN7WebCore24CanvasRenderingContext2D16drawTextInternalERKN3WTF6StringEffbSt8optionalIfE (this=0x7efe63c84000, text=..., x=<optimized out>, y=<optimized out>, fill=fill@entry=true, maxWidth=Python Exception <class 'gdb.error'> There is no member or method named _M_payload.: 
...) at /usr/src/debug/webkit2gtk3-2.22.3-1.fc29.x86_64/Source/WebCore/html/canvas/CanvasRenderingContext2D.cpp:586
#19 0x00007eff0c907da2 in _ZN7WebCore24CanvasRenderingContext2D8fillTextERKN3WTF6StringEffSt8optionalIfE (this=this@entry=0x7efe63c84000, text=..., x=<optimized out>, y=<optimized out>, maxWidth=Python Exception <class 'gdb.error'> There is no member or method named _M_payload.: 
...) at /usr/include/c++/8/new:169
#20 0x00007eff0d268012 in WebCore::jsCanvasRenderingContext2DPrototypeFunctionFillTextBody (throwScope=..., castedThis=<optimized out>, state=<optimized out>) at /usr/include/c++/8/new:169
#21 0x00007eff0d268012 in WebCore::IDLOperation<WebCore::JSCanvasRenderingContext2D>::call<WebCore::jsCanvasRenderingContext2DPrototypeFunctionFillTextBody> (operationName=0x7eff0d42c208 "fillText", state=...) at /usr/src/debug/webkit2gtk3-2.22.3-1.fc29.x86_64/Source/WebCore/bindings/js/JSDOMOperation.h:53
#22 0x00007eff0d268012 in _ZN7WebCore51jsCanvasRenderingContext2DPrototypeFunctionFillTextEPN3JSC9ExecStateE (state=<optimized out>) at /usr/src/debug/webkit2gtk3-2.22.3-1.fc29.x86_64/x86_64-redhat-linux-gnu/DerivedSources/WebCore/JSCanvasRenderingContext2D.cpp:3142
#23 0x00007efea7fff177 in  ()
#24 0x00007fffd907bce0 in  ()
#25 0x00007eff0aa16172 in llint_entry () at /lib64/libjavascriptcoregtk-4.0.so.18

[Michael Catanzaro]

My guess is bug #191825. Zan would know for sure.

[Michael Catanzaro]

(In reply to Michael Catanzaro from comment #6)
> My guess is bug #191825. Zan would know for sure.

Well it was a good guess since that patch fixed a serious bug with the glyph cache, and this function is crashing while modifying the glyph cache. But no, it's still broken even with that patch.

[Zan Dobersek]

Can you post output of the `disassemble` operation in gdb, along with output of `info registers`?

[Michael Catanzaro]

Note it's 100% reproducible with the master runtime.

I assumed it was a use after free, but actually the cache entries are NEVER freed (not even when quitting the UI process!), neither the face cache nor the glyph cache, so this is a tough one.

[Michael Catanzaro]

Created attachment 355291 [details]
disassembly

[Michael Catanzaro]

Created attachment 355292 [details]
registers

[Michael Catanzaro]

First two frames with an -Og build:

#0  0x00007fc93ea70d61 in WebCore::harfBuzzGetGlyph (fontData=<optimized out>, unicode=<optimized out>, 
    glyph=0x7ffccfbdf288)
    at /home/mcatanzaro/Projects/WebKit/Source/WebCore/platform/graphics/harfbuzz/HarfBuzzFaceCairo.cpp:107
        glyphs = 0x7416370
        numGlyphs = 1
        buffer = "​"
        bufferLength = <optimized out>
        hbFontData = <optimized out>
        scaledFont = 0x73b3590
        result = {
          iterator = {<std::iterator<std::forward_iterator_tag, WTF::KeyValuePair<unsigned int, unsigned int>, long, WTF::KeyValuePair<unsigned int, unsigned int>*, WTF::KeyValuePair<unsigned int, unsigned int>&>> = {<No data fields>}, 
            m_iterator = {<std::iterator<std::forward_iterator_tag, WTF::KeyValuePair<unsigned int, unsigned int>, long, WTF::KeyValuePair<unsigned int, unsigned int> const*, WTF::KeyValuePair<unsigned int, unsigned int> const&>> = {<No data fields>}, m_position = 0x0, m_endPosition = <optimized out>}}, 
          isNewEntry = <optimized out>}
#1  0x00007fc9397645df in hb_font_t::get_nominal_glyph (glyph=0x7ffccfbdf288, unicode=0, 
    this=<optimized out>) at hb-font-private.hh:211
No locals.


m_position = 0x0 looks suspicious to me.

[Zan Dobersek]

Created attachment 355352 [details]
Patch

[Michael Catanzaro]

Comment on attachment 355352 [details]
Patch

Ah amazing! I would never have guessed that zero was not a valid key. I wonder how many similar bugs exist due to this.

[WebKit Commit Bot]

Comment on attachment 355352 [details]
Patch

Clearing flags on attachment: 355352

Committed r238405: <https://trac.webkit.org/changeset/238405>

[WebKit Commit Bot]

All reviewed patches have been landed.  Closing bug.