390768 - crash if I open large PNG [XError calls exit on "BadAlloc (insufficient resources for operation)"]

Status: RESOLVED DUPLICATE of bug 424333

(Core :: Widget: Gtk, defect, P2)

Description

[quintenbernaert]

User-Agent:       Mozilla/5.0 (X11; U; Linux i686; nl; rv:1.8.1.6) Gecko/20070725 Firefox/2.0.0.6
Build Identifier: Mozilla/5.0 (X11; U; Linux i686; nl; rv:1.8.1.6) Gecko/20070725 Firefox/2.0.0.6

Firefox crashes when I open a large PNG on the url (http://apollo.sese.asu.edu/METRIC_PREVIEW/AS15-M-0081/AS15-M-0081.html).


Reproducible: Always

Steps to Reproduce:
1) Go to http://apollo.sese.asu.edu/METRIC_PREVIEW/AS15-M-0081/AS15-M-0081.html
2) Click "Large PNG, 8.8 m/p (251 MB)"
Actual Results:  
I see the first line/lines of pixels and then Firefox crashes.

Expected Results:  
Load the image.

Comment 1

[quintenbernaert]

Firefox 3 alpha 6 says the image contains errors and it doesn't crash.

Comment 2

[Adam Guthrie]

Can you get a stacktrace for the crash? http://kb.mozillazine.org/Talkback

Comment 3

[Mats Palmgren (inactive)]

Attachment: stack

Error calls exit in response to an XCreatePixmap error...

Comment 4

[Mats Palmgren (inactive)]

We have bugs filed on this, eg bug 348463 and bug 210931.

Comment 5

[Benjamin]

On a mac with Firefox 2.0.0.9, I get the same behaviour. The two bugs referenced above were reported on Linux with GTK, so I am a bit puzzled. I attach a test-case png which always kills firefox on my mac as well as on Linux when trying to zoom in. A stack trace is also included. The command line spits out this:

/Applications/Firefox.app/Contents/MacOS/run-mozilla.sh: line 424:  9222 Segmentation fault      "$prog" ${1+"$@"}

Is this a bug in Apples rendering? Safari shows the picture nicely...

Comment 6

[Benjamin]

Attachment: Stack trace of Firefox 2.0.0.9 crash on Mac

Comment 7

[Benjamin]

Sorry, just realized the PNG is too large to show. Try this url:

http://www.sanger.ac.uk/cgi-bin/software/analysis/logomat-m.cgi?pfamid=PF04735

Comment 8

[timeless]

benjamin: file a new bug. this bug is *only* for Gtk. Your crash in apple's code could either be a bug in apple's code, or a bug in some of our mac code, but it needs to be tracked differently.

Comment 9

[Daniel Holbert [:dholbert]]

I see this bug using Ubuntu 8.04, using FF3 and FF2, at these places:
 a) the URL in comment 7
 b) http://www.howtoforge.com
 c) attachment 310818 [details]  (a wide PNG taken from howtoforge.com)
 d) http://tech.yahoo.com/blogs/

I initially posted about this in bug 369971 comment 24 through 37, but I'm directing further info to this bug page, because what I'm seeing is Linux-specific and PNG-specific.

If it matters, my libgtk2.0-0 package is version 2.0-0_2.12.9-2ubuntu1

Comment 10

[Mats Palmgren (inactive)]

Daniel, bug 424333 now has a patch for trunk too... does that help?

Comment 11

[Daniel Holbert [:dholbert]]

Nope.  I tested
   attachment 310818 [details]  (evil howtoforge PNG)
along with
   attachment 310964 [details]  (gif that crashes FF, from bug 424333)
and they both still crash (thought the gif takes ~5 seconds)

Comment 12

[Daniel Holbert [:dholbert]]

Oops -- I think I was running the wrong build when I posted comment 11.

I just tested it with the patch correctly applied, and while the patch *does* fix attachment 310964 [details] (gif from bug 424333) on my machine, it doesn't fix attachment 310818 [details] or howtoforge.com or tech.yahoo.com/blogs.

Comment 13

[Daniel Holbert [:dholbert]]

Attachment: Stack trace of Trunk crash on Linux (Ubuntu 8.04)

Here's a stack trace I caught in GDB for the linux crash on attachment 310818 [details].

If I continue from this point, I immediately hit the _Xerror call.

Comment 14

[Daniel Holbert [:dholbert]]

I can actually scale the howtoforge image down to 8189px wide and still produce the crash. (but at 8188px wide, we don't crash)

Then, keeping the 8189px width, I can also scale it vertically down to 2px tall and still produce the crash.  (but at 1px, or at 2px by 8188px, we don't crash)

So what's the significance of 8189px?  It's *almost* 2^13:
    2^13 = 8192 = 8189 + 3

So I think we're hitting some byte-count boundary, or something... 

e.g. if we're hypothetically using 4 bits per pixel of width, that puts us at
    8188 * 4 = 32752 = (2^15 - 16) => no crash
    8189 * 4 = 32756 = (2^15 - 12) => crash
so we could be dealing with a signed 16-bit number (i.e. 15 bits of value) which is offset by 16 for some reason, and which is overflowing & causing the crash.

Comment 15

[Daniel Holbert [:dholbert]]

Attachment: reduced PNG testcase (crashes firefox on some systems)

Here's a black 2px-high 8189px-wide PNG which crashes both Firefox 2 and trunk builds.  (per my last comment)

Comment 16

[Daniel Holbert [:dholbert]]

For convenience, here's that last testcase as a data-url: http://preview.tinyurl.com/398gxr

Comment 17

[Daniel Holbert [:dholbert]]

Requesting blocking1.9, as this affects trunk (along with branch) and seems fairly serious.

Comment 18

[Daniel Holbert [:dholbert]]

Comment on attachment 312172 [details]
reduced PNG testcase (crashes firefox on some systems)

On my laptop, which is also running Ubuntu 8.04, I only get the crash with the URL from comment 7 -- not with the reduced testcase, the howtoforge.com site, or the tech.yahoo.com/blogs site.

Comment 19

[Vladimir Vukicevic [:vlad] [:vladv] (needinfo me, slow to respond)]

Swapping this based on roc's comments -- it's not a regression, but we should fix it at some point.

Comment 22

[Alain Tuor]

same issue with www.directnet.com (quite critical, it's for Credit Suisse
online banking)

Comment 24

[Martin Stránský [:stransky] (ni? me)]

In 1.8, it's handled here:

nsresult nsImageGTK::Init(PRInt32 aWidth, PRInt32 aHeight,
                          PRInt32 aDepth, nsMaskRequirements aMaskRequirements)
{

  [snip]

  // X Protocol limits us to image dimensions less than 32767
  // unless we want to go through lots of pain and suffering.
  if (aWidth > SHRT_MAX || aHeight > SHRT_MAX)
    return NS_ERROR_FAILURE;

  [snip]

}

Comment 26

[Mats Palmgren (inactive)]

The patch in attachment 381814 [details] [diff] [review] in bug 424333 fixes this bug (and duplicates)
for me.

Comment 27

[Johnny Stenback (:jst)]

Should this bug be closed then, or is there something still remaining here?

Comment 28

[Mats Palmgren (inactive)]

Nothing remaining here besides the issues in bug 424333 as far as I know.