Returns an error immediately without sending IPC message when a font family n...

Returns an error immediately without sending IPC message when a font family name to resolve is too long.

This change is important when a site has @font-face rule like:

  // http://paulirish.com/webkit-fontface-hang.html
  @font-face{font-family:testfont;src:url('data:font/ttf;base64,AA.....<<looooooooooong base64 data>>.....aQ==')}

In such a case, WebCore first calls SkFontHost::CreateTypeface() with the (possibly very long) data-uri string itself, then calls SkFontHost::CreateTypefaceFromStream() with decoded byte stream. Since render_sandbox_host_linux.cc just ignores too long IPC message, the renderer process could block indefinitely waiting for a reply inside recvmsg() system call called from SkFontHost::CreateTypeface().

I'm not sure if the WebCore behavior (i.e. calling CreateTypeface with data-uris) is reasonable, but I believe the Skia part is better to be fixed anyway. Non data-uri font family names could be very long too:

  @font-face{font-family:testfont;src:local('AA........AA');}

BUG=29861
TEST=First, set up your Linux SUID Sandbox binary: http://code.google.com/p/chromium/wiki/LinuxSUIDSandboxDevelopment. Then start Chromium and visit http://paulirish.com/webkit-fontface-hang.html or http://typekit.com/. Verify that the renderer does not freeze.


Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=34915

[agl, 2009-12-17 18:11:38]

LGTM

http://codereview.chromium.org/507037/diff/1/5
File chrome/browser/renderer_host/render_sandbox_host_linux.cc (right):

http://codereview.chromium.org/507037/diff/1/5#newcode105
chrome/browser/renderer_host/render_sandbox_host_linux.cc:105: // bytes long
(all other messages are smaller than the message).
(all other ...) -> (this is the largest message type)

http://codereview.chromium.org/507037/diff/1/5#newcode106
chrome/browser/renderer_host/render_sandbox_host_linux.cc:106: // 128 byets
padding are necessary so recvmsg() does not return MSG_TRUNC
byets -> bytes.

http://codereview.chromium.org/507037/diff/1/5#newcode113
chrome/browser/renderer_host/render_sandbox_host_linux.cc:113: NOTREACHED() <<
"The message is larger than kMaxFontFamilyLength??";
"The m" -> "Sandbox host m"
"??" -> ""

[Yusuke Sato, 2009-12-18 01:14:09]

Thanks, submitting.

http://codereview.chromium.org/507037/diff/1/5
File chrome/browser/renderer_host/render_sandbox_host_linux.cc (right):

http://codereview.chromium.org/507037/diff/1/5#newcode105
chrome/browser/renderer_host/render_sandbox_host_linux.cc:105: // bytes long
(all other messages are smaller than the message).
On 2009/12/17 18:11:38, agl wrote:
> (all other ...) -> (this is the largest message type)

Done.

http://codereview.chromium.org/507037/diff/1/5#newcode106
chrome/browser/renderer_host/render_sandbox_host_linux.cc:106: // 128 byets
padding are necessary so recvmsg() does not return MSG_TRUNC
On 2009/12/17 18:11:38, agl wrote:
> byets -> bytes.

Done.

http://codereview.chromium.org/507037/diff/1/5#newcode113
chrome/browser/renderer_host/render_sandbox_host_linux.cc:113: NOTREACHED() <<
"The message is larger than kMaxFontFamilyLength??";
On 2009/12/17 18:11:38, agl wrote:
> "The m" -> "Sandbox host m"
> "??" -> ""

Done.