[Patch included] nspluginwrapper has a race condition on NPP_Destroy and may crash Flash
Binary package hint: nspluginwrapper
nspluginwrapper has a race condition during NPP_Destroy (called when a tab is closed) that can crash the plugin. The race happens particularly often when another tab has a video playing; I suspect this is because it causes enough traffic over the IPC to delay the processes and trigger the race.
If NPP_Destroy is called by the wrapper process at the same time the viewer (plugin) process makes some call, then, from the plugin's perspective, its call to NPN_InvalidateRect resulted in the plugin instance being destroyed from under its feet. This is, of course, nonsense, so Flash shortly crashes to let us know how silly we are being. :-)
I've written patches for this issue here
Only the second of the two patches is strictly relevant; the other is a separate race I came across in a previous iteration of this patch. It detects when NPP_Destroy is being called at an unsafe point and delays it to another message loop iteration. With the caveat that requests can't be reordered. So, when it must, the patch lies to the wrapper about NPP_Destroy's return values. Any delayed NPSavedData gets discarded. That said, I've never seen Flash use this feature, and the docs do allow the browser to discard them arbitrarily.
The relevant bug in Chromium is here:
(To be thorough, this is on nspluginwrapper 1.2.2-0ubuntu7 on google-chrome-beta 10.0.648.82-r75062 on Ubuntu maverick. I've also reproduced this crash in Firefox, and a cursory look at Debian stable's nspluginwrapper 1.3.0-1 suggests the bug is there too...)
As far as I can tell, nspluginwrapper no longer has an upstream. If that's not the case, this patch should probably be forward.
|Changed in nspluginwrapper (Ubuntu):|
|status:||New → Fix Released|