605347 - @font-face SIGFPE crash with Google Web Font directory in libpangooft2 [@ libpangoft2-1.0.so.0.2800.1@0x1c6c8 ]

Status: RESOLVED FIXED

Description

[Chris Morgan]

User-Agent:       Mozilla/5.0 (X11; Linux x86_64; rv:2.0b8pre) Gecko/20101012 Firefox-4.0/4.0b8pre
Build Identifier: Mozilla/5.0 (X11; Linux x86_64; rv:2.0b8pre) Gecko/20101012 Firefox-4.0/4.0b8pre

Loading the Google Web Font directory in Firefox 4 in Linux, the page starts loading the fonts and then the browser crashes.

Reproducible: Always

Steps to Reproduce:
1. Use Firefox 4 in Linux
2. Load http://code.google.com/webfonts
Actual Results:  
Crash.

Expected Results:  
Not crash.

Talkback crash ID: 766d4e60-bcb7-4c6c-b168-d9c062101018

http://crash-stats.mozilla.com/report/index/766d4e60-bcb7-4c6c-b168-d9c062101018

Seems to be experienced by a number of people.

Comment 1

[John Daggett (:jtd)]

Pango integration code lives in gfx, switching back to gfx.

Comment 2

[Karl Tomlinson (:karlt)]

Here, with Firefox 3.6.9, it appears as a hang, as if our fpehandler is trying to continue after the exceptions.  (I don't have crashreporter enabled.)

#0  fpehandler (signum=8, si=0x7fff3d0a0f30, context=0x7fff3d0a0e00)
    at nsSigHandlers.cpp:256
#1  <signal handler called>
#2  0x00007fe31280e1c3 in _hb_sanitize_array (this=0x40de5cc, 
    context=0x7fff3d0a13c0) at hb-open-type-private.hh:213
#3  PairPosFormat2::sanitize (this=0x40de5cc, context=0x7fff3d0a13c0)
    at hb-ot-layout-gpos-private.hh:711
#4  PairPos::sanitize (this=0x40de5cc, context=0x7fff3d0a13c0)
    at hb-ot-layout-gpos-private.hh:765
#5  0x00007fe31280fa0c in GenericOffsetTo<USHORT, PosLookupSubTable>::sanitize (
    this=<value optimized out>, context=0x7fff3d0a13c0)
    at hb-open-type-private.hh:479
#6  GenericArrayOf<USHORT, OffsetTo<PosLookupSubTable> >::sanitize (
    this=<value optimized out>, context=0x7fff3d0a13c0)
    at hb-open-type-private.hh:550
#7  PosLookup::sanitize (this=<value optimized out>, context=0x7fff3d0a13c0)
    at hb-ot-layout-gpos-private.hh:1538
#8  GenericOffsetTo<USHORT, PosLookup>::sanitize (this=<value optimized out>, 
    context=0x7fff3d0a13c0) at hb-open-type-private.hh:465
#9  GenericArrayOf<USHORT, OffsetTo<PosLookup> >::sanitize (
    this=<value optimized out>, context=0x7fff3d0a13c0)
    at hb-open-type-private.hh:532
#10 OffsetListOf<PosLookup>::sanitize (this=<value optimized out>, 
    context=0x7fff3d0a13c0) at hb-open-type-private.hh:591
#11 GenericOffsetTo<USHORT, OffsetListOf<PosLookup> >::sanitize (
    this=<value optimized out>, context=0x7fff3d0a13c0)
    at hb-open-type-private.hh:465
#12 GPOS::sanitize (this=<value optimized out>, context=0x7fff3d0a13c0)
    at hb-ot-layout-gpos-private.hh:1569
#13 0x00007fe3128055ce in Sanitizer<GPOS>::sanitize (face=0x30fb090)
    at hb-open-type-private.hh:279
#14 _hb_ot_layout_init (face=0x30fb090) at hb-ot-layout.cc:55
#15 0x00007fe312801c2d in hb_face_create_for_data (blob=<value optimized out>, 
    index=<value optimized out>) at hb-font.cc:182
#16 0x00007fe3127ff21d in pango_ot_info_get (face=0x4107190)
    at pango-ot-info.c:154
#17 0x00007fe2f0cc1549 in basic_engine_shape (engine=<value optimized out>, 
    font=<value optimized out>, text=<value optimized out>, 
    length=<value optimized out>, analysis=<value optimized out>, 
    glyphs=<value optimized out>) at basic-fc.c:209
#18 0x00007fe3125d21dd in pango_shape (text=0x7fff3d0a17e3 "Philosopher", 
    length=11, analysis=0x40a6db0, glyphs=0x3f5eba0) at shape.c:55
#19 0x00007fe316a1e022 in gfxPangoFontGroup::CreateGlyphRunsItemizing (
    this=0x3f265d0, aTextRun=<value optimized out>, 
    aUTF8=<value optimized out>, aUTF8Length=<value optimized out>, 
    aUTF8HeaderLen=<value optimized out>) at gfxPangoFonts.cpp:3088
#20 0x00007fe316a1ee8c in gfxPangoFontGroup::MakeTextRun (this=0x3f265d0, 
    aString=<value optimized out>, aLength=<value optimized out>, 
    aParams=<value optimized out>, aFlags=<value optimized out>)
    at gfxPangoFonts.cpp:2373
#21 0x00007fe316a18d37 in TextRunWordCache::MakeTextRun (
    this=<value optimized out>, aText=<value optimized out>, 
    aLength=<value optimized out>, aFontGroup=0x3f265d0, 
    aParams=0x7fff3d0a3370, aFlags=<value optimized out>)
    at gfxTextRunWordCache.cpp:811

Works fine in my m-c build including patches for bug 569770, which doesn't use this path for the scripts on this page.

Comment 3

[Behdad Esfahbod]

I fixed this in pango and it's in the latest releases, but apparently not all distros have picked it up.

Can you guys make your fpehandler more lenient?

Comment 4

[Karl Tomlinson (:karlt)]

(In reply to comment #2)
> Here, with Firefox 3.6.9, it appears as a hang, as if our fpehandler is trying
> to continue after the exceptions.  (I don't have crashreporter enabled.)

That's because we don't have bug 538338 fixed on 1.9.2.
But, with release builds, I get the crashreporter anyway.

Comment 5

[Karl Tomlinson (:karlt)]

(In reply to comment #3)
> I fixed this in pango and it's in the latest releases, but apparently not all
> distros have picked it up.
> 
> Can you guys make your fpehandler more lenient?

Thanks, Behdad for
http://git.gnome.org/browse/pango/commit/?id=152e0aab5bb29d691e5e69e2f375b3b42e15e48e
(which is also in 1.28.2).

I'm not sure that we want to pick any sort of result for integer divide by zero.
I think it's probably safer from a security perspective to abort than to continue with an inappropriate integer result for an unknown purpose.

Comment 7

[Karl Tomlinson (:karlt)]

Bug 569770 has now landed.  I can't reproduce now even with 3.6.12, so I suspect the Philosopher font may have been updated.  The font says it was last modified 2010-09-10 16:08:43 UTC.

The http response header for
http://themes.googleusercontent.com/font?kit=OttjxgcoEsufOGSINYBGLWeudeTO44zf-ht3k-KNzwg says it was last modified:
Thu, 04 Nov 2010 18:05:56 GMT