Bug 105359 – REGRESSION(r137607): PluginDocument loads consume huge amounts of memory

• Status: RESOLVED
• Resolution: FIXED
• Component: WebCore Misc.

[Nate Chapin]

Original report: https://code.google.com/p/chromium/issues/detail?id=166374

To reproduce, go to a plugin that loads as a main resource in a chromium canary (e.g., http://www.scb.se/statistik/_publikationer/NR0001_2012K02_TI_A28TI1203.pdf). That 4MB pdf can easily consume multiple GBs of memory.

The problem is that CachedRawResource implicitly assumes that, if setShouldBufferData() will be called, it will be called before the load starts. PluginDocument calls it after receiving the first chunk of data, which leaves CachedRawResource in an inconsistent state> It ends up sending the entirety of the data received so far in each callback instead of only the incremental data.

[Nate Chapin]

Created attachment 180047 [details]
patch

[Alexey Proskuryakov]

Comment on attachment 180047 [details]
patch

View in context: https://bugs.webkit.org/attachment.cgi?id=180047&action=review

> Source/WebCore/ChangeLog:12
> +        (WebCore::ResourceLoader::setShouldBufferData): shouldBufferData is an enum, not a boolean, so this is reversed.

It's named as a boolean, so it should be renamed.

[Nate Chapin]

(In reply to comment #2)
> (From update of attachment 180047 [details])
> View in context: https://bugs.webkit.org/attachment.cgi?id=180047&action=review
> 
> > Source/WebCore/ChangeLog:12
> > +        (WebCore::ResourceLoader::setShouldBufferData): shouldBufferData is an enum, not a boolean, so this is reversed.
> 
> It's named as a boolean, so it should be renamed.

true

I'll do that in a separate patch.

[WebKit Review Bot]

Comment on attachment 180047 [details]
patch

Clearing flags on attachment: 180047

Committed r138174: <http://trac.webkit.org/changeset/138174>

[WebKit Review Bot]

All reviewed patches have been landed.  Closing bug.