609386 - Infallible malloc for 1.9.2 branch

Status: RESOLVED WONTFIX

(Core :: XPCOM, defect)

Description

[Jesse Ruderman]

dveditz says we've been hit with lots of OOM bug reports in the last few months, including several exploitable bugs.  Can we backport infallible malloc to the branch?

Comment 1

[Chris Jones [:cjones] inactive; ni?/f?/r? if you need me]

The difficulties I know of are
 1. memory/mozalloc has been hacked up a bit for android, might depend on some build changes external to that dir
 2. Mozalloc changed configure.in and some Makefiles.  These changes would need to be backported not only to m-c 1.9.2 but also every other almost-but-not-quite-copy of our configure/build system in gecko-1.9.2-using products, thunderbird and seamonkey et al.
 3. We would need to freeze the 3.6 SDK to a version before a mozalloc backport to avoid compat problems with binary extensions.  This probably isn't a big deal though.
 4. Some code may have been switched to explicitly-fallible allocation after mozalloc landed on trunk, to avoid OOM aborts seen in the wild.  We would need to backport those changes too or live with an increased volume of non-exploitable crashes.

I may be missing some stuff, not sure.

This is nontrivial but not impossible AFAICT.  I'm not in a good position to comment on the cost/benefit though.

Comment 2

[Jesse Ruderman]

1.9.2 is nearing EOL.